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

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

Version: ~ [ linux-5.18-rc6 ] ~ [ linux-5.17.6 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.38 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.114 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.192 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.241 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.277 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.312 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
  3  *              operating system.  INET is implemented using the  BSD Socket
  4  *              interface as the means of communication with the user level.
  5  *
  6  *              The User Datagram Protocol (UDP).
  7  *
  8  * Authors:     Ross Biro
  9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 10  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 11  *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
 12  *              Hirokazu Takahashi, <taka@valinux.co.jp>
 13  *
 14  * Fixes:
 15  *              Alan Cox        :       verify_area() calls
 16  *              Alan Cox        :       stopped close while in use off icmp
 17  *                                      messages. Not a fix but a botch that
 18  *                                      for udp at least is 'valid'.
 19  *              Alan Cox        :       Fixed icmp handling properly
 20  *              Alan Cox        :       Correct error for oversized datagrams
 21  *              Alan Cox        :       Tidied select() semantics.
 22  *              Alan Cox        :       udp_err() fixed properly, also now
 23  *                                      select and read wake correctly on errors
 24  *              Alan Cox        :       udp_send verify_area moved to avoid mem leak
 25  *              Alan Cox        :       UDP can count its memory
 26  *              Alan Cox        :       send to an unknown connection causes
 27  *                                      an ECONNREFUSED off the icmp, but
 28  *                                      does NOT close.
 29  *              Alan Cox        :       Switched to new sk_buff handlers. No more backlog!
 30  *              Alan Cox        :       Using generic datagram code. Even smaller and the PEEK
 31  *                                      bug no longer crashes it.
 32  *              Fred Van Kempen :       Net2e support for sk->broadcast.
 33  *              Alan Cox        :       Uses skb_free_datagram
 34  *              Alan Cox        :       Added get/set sockopt support.
 35  *              Alan Cox        :       Broadcasting without option set returns EACCES.
 36  *              Alan Cox        :       No wakeup calls. Instead we now use the callbacks.
 37  *              Alan Cox        :       Use ip_tos and ip_ttl
 38  *              Alan Cox        :       SNMP Mibs
 39  *              Alan Cox        :       MSG_DONTROUTE, and 0.0.0.0 support.
 40  *              Matt Dillon     :       UDP length checks.
 41  *              Alan Cox        :       Smarter af_inet used properly.
 42  *              Alan Cox        :       Use new kernel side addressing.
 43  *              Alan Cox        :       Incorrect return on truncated datagram receive.
 44  *      Arnt Gulbrandsen        :       New udp_send and stuff
 45  *              Alan Cox        :       Cache last socket
 46  *              Alan Cox        :       Route cache
 47  *              Jon Peatfield   :       Minor efficiency fix to sendto().
 48  *              Mike Shaver     :       RFC1122 checks.
 49  *              Alan Cox        :       Nonblocking error fix.
 50  *      Willy Konynenberg       :       Transparent proxying support.
 51  *              Mike McLagan    :       Routing by source
 52  *              David S. Miller :       New socket lookup architecture.
 53  *                                      Last socket cache retained as it
 54  *                                      does have a high hit rate.
 55  *              Olaf Kirch      :       Don't linearise iovec on sendmsg.
 56  *              Andi Kleen      :       Some cleanups, cache destination entry
 57  *                                      for connect.
 58  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
 59  *              Melvin Smith    :       Check msg_name not msg_namelen in sendto(),
 60  *                                      return ENOTCONN for unconnected sockets (POSIX)
 61  *              Janos Farkas    :       don't deliver multi/broadcasts to a different
 62  *                                      bound-to-device socket
 63  *      Hirokazu Takahashi      :       HW checksumming for outgoing UDP
 64  *                                      datagrams.
 65  *      Hirokazu Takahashi      :       sendfile() on UDP works now.
 66  *              Arnaldo C. Melo :       convert /proc/net/udp to seq_file
 67  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
 68  *      Alexey Kuznetsov:               allow both IPv4 and IPv6 sockets to bind
 69  *                                      a single port at the same time.
 70  *      Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
 71  *      James Chapman           :       Add L2TP encapsulation type.
 72  *
 73  *
 74  *              This program is free software; you can redistribute it and/or
 75  *              modify it under the terms of the GNU General Public License
 76  *              as published by the Free Software Foundation; either version
 77  *              2 of the License, or (at your option) any later version.
 78  */
 79 
 80 #define pr_fmt(fmt) "UDP: " fmt
 81 
 82 #include <linux/uaccess.h>
 83 #include <asm/ioctls.h>
 84 #include <linux/bootmem.h>
 85 #include <linux/highmem.h>
 86 #include <linux/swap.h>
 87 #include <linux/types.h>
 88 #include <linux/fcntl.h>
 89 #include <linux/module.h>
 90 #include <linux/socket.h>
 91 #include <linux/sockios.h>
 92 #include <linux/igmp.h>
 93 #include <linux/inetdevice.h>
 94 #include <linux/in.h>
 95 #include <linux/errno.h>
 96 #include <linux/timer.h>
 97 #include <linux/mm.h>
 98 #include <linux/inet.h>
 99 #include <linux/netdevice.h>
100 #include <linux/slab.h>
101 #include <net/tcp_states.h>
102 #include <linux/skbuff.h>
103 #include <linux/proc_fs.h>
104 #include <linux/seq_file.h>
105 #include <net/net_namespace.h>
106 #include <net/icmp.h>
107 #include <net/inet_hashtables.h>
108 #include <net/route.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <trace/events/udp.h>
112 #include <linux/static_key.h>
113 #include <trace/events/skb.h>
114 #include <net/busy_poll.h>
115 #include "udp_impl.h"
116 #include <net/sock_reuseport.h>
117 #include <net/addrconf.h>
118 
119 struct udp_table udp_table __read_mostly;
120 EXPORT_SYMBOL(udp_table);
121 
122 long sysctl_udp_mem[3] __read_mostly;
123 EXPORT_SYMBOL(sysctl_udp_mem);
124 
125 int sysctl_udp_rmem_min __read_mostly;
126 EXPORT_SYMBOL(sysctl_udp_rmem_min);
127 
128 int sysctl_udp_wmem_min __read_mostly;
129 EXPORT_SYMBOL(sysctl_udp_wmem_min);
130 
131 atomic_long_t udp_memory_allocated;
132 EXPORT_SYMBOL(udp_memory_allocated);
133 
134 #define MAX_UDP_PORTS 65536
135 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
136 
137 static int udp_lib_lport_inuse(struct net *net, __u16 num,
138                                const struct udp_hslot *hslot,
139                                unsigned long *bitmap,
140                                struct sock *sk,
141                                int (*saddr_comp)(const struct sock *sk1,
142                                                  const struct sock *sk2,
143                                                  bool match_wildcard),
144                                unsigned int log)
145 {
146         struct sock *sk2;
147         kuid_t uid = sock_i_uid(sk);
148 
149         sk_for_each(sk2, &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         kuid_t uid = sock_i_uid(sk);
181         int res = 0;
182 
183         spin_lock(&hslot2->lock);
184         udp_portaddr_for_each_entry(sk2, &hslot2->head) {
185                 if (net_eq(sock_net(sk2), net) &&
186                     sk2 != sk &&
187                     (udp_sk(sk2)->udp_port_hash == num) &&
188                     (!sk2->sk_reuse || !sk->sk_reuse) &&
189                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
190                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
191                     (!sk2->sk_reuseport || !sk->sk_reuseport ||
192                      rcu_access_pointer(sk->sk_reuseport_cb) ||
193                      !uid_eq(uid, sock_i_uid(sk2))) &&
194                     saddr_comp(sk, sk2, true)) {
195                         res = 1;
196                         break;
197                 }
198         }
199         spin_unlock(&hslot2->lock);
200         return res;
201 }
202 
203 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot,
204                                   int (*saddr_same)(const struct sock *sk1,
205                                                     const struct sock *sk2,
206                                                     bool match_wildcard))
207 {
208         struct net *net = sock_net(sk);
209         kuid_t uid = sock_i_uid(sk);
210         struct sock *sk2;
211 
212         sk_for_each(sk2, &hslot->head) {
213                 if (net_eq(sock_net(sk2), net) &&
214                     sk2 != sk &&
215                     sk2->sk_family == sk->sk_family &&
216                     ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
217                     (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
218                     (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
219                     sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
220                     (*saddr_same)(sk, sk2, false)) {
221                         return reuseport_add_sock(sk, sk2);
222                 }
223         }
224 
225         /* Initial allocation may have already happened via setsockopt */
226         if (!rcu_access_pointer(sk->sk_reuseport_cb))
227                 return reuseport_alloc(sk);
228         return 0;
229 }
230 
231 /**
232  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
233  *
234  *  @sk:          socket struct in question
235  *  @snum:        port number to look up
236  *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
237  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
238  *                   with NULL address
239  */
240 int udp_lib_get_port(struct sock *sk, unsigned short snum,
241                      int (*saddr_comp)(const struct sock *sk1,
242                                        const struct sock *sk2,
243                                        bool match_wildcard),
244                      unsigned int hash2_nulladdr)
245 {
246         struct udp_hslot *hslot, *hslot2;
247         struct udp_table *udptable = sk->sk_prot->h.udp_table;
248         int    error = 1;
249         struct net *net = sock_net(sk);
250 
251         if (!snum) {
252                 int low, high, remaining;
253                 unsigned int rand;
254                 unsigned short first, last;
255                 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
256 
257                 inet_get_local_port_range(net, &low, &high);
258                 remaining = (high - low) + 1;
259 
260                 rand = prandom_u32();
261                 first = reciprocal_scale(rand, remaining) + low;
262                 /*
263                  * force rand to be an odd multiple of UDP_HTABLE_SIZE
264                  */
265                 rand = (rand | 1) * (udptable->mask + 1);
266                 last = first + udptable->mask + 1;
267                 do {
268                         hslot = udp_hashslot(udptable, net, first);
269                         bitmap_zero(bitmap, PORTS_PER_CHAIN);
270                         spin_lock_bh(&hslot->lock);
271                         udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
272                                             saddr_comp, udptable->log);
273 
274                         snum = first;
275                         /*
276                          * Iterate on all possible values of snum for this hash.
277                          * Using steps of an odd multiple of UDP_HTABLE_SIZE
278                          * give us randomization and full range coverage.
279                          */
280                         do {
281                                 if (low <= snum && snum <= high &&
282                                     !test_bit(snum >> udptable->log, bitmap) &&
283                                     !inet_is_local_reserved_port(net, snum))
284                                         goto found;
285                                 snum += rand;
286                         } while (snum != first);
287                         spin_unlock_bh(&hslot->lock);
288                 } while (++first != last);
289                 goto fail;
290         } else {
291                 hslot = udp_hashslot(udptable, net, snum);
292                 spin_lock_bh(&hslot->lock);
293                 if (hslot->count > 10) {
294                         int exist;
295                         unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
296 
297                         slot2          &= udptable->mask;
298                         hash2_nulladdr &= udptable->mask;
299 
300                         hslot2 = udp_hashslot2(udptable, slot2);
301                         if (hslot->count < hslot2->count)
302                                 goto scan_primary_hash;
303 
304                         exist = udp_lib_lport_inuse2(net, snum, hslot2,
305                                                      sk, saddr_comp);
306                         if (!exist && (hash2_nulladdr != slot2)) {
307                                 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
308                                 exist = udp_lib_lport_inuse2(net, snum, hslot2,
309                                                              sk, saddr_comp);
310                         }
311                         if (exist)
312                                 goto fail_unlock;
313                         else
314                                 goto found;
315                 }
316 scan_primary_hash:
317                 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
318                                         saddr_comp, 0))
319                         goto fail_unlock;
320         }
321 found:
322         inet_sk(sk)->inet_num = snum;
323         udp_sk(sk)->udp_port_hash = snum;
324         udp_sk(sk)->udp_portaddr_hash ^= snum;
325         if (sk_unhashed(sk)) {
326                 if (sk->sk_reuseport &&
327                     udp_reuseport_add_sock(sk, hslot, saddr_comp)) {
328                         inet_sk(sk)->inet_num = 0;
329                         udp_sk(sk)->udp_port_hash = 0;
330                         udp_sk(sk)->udp_portaddr_hash ^= snum;
331                         goto fail_unlock;
332                 }
333 
334                 sk_add_node_rcu(sk, &hslot->head);
335                 hslot->count++;
336                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
337 
338                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
339                 spin_lock(&hslot2->lock);
340                 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
341                     sk->sk_family == AF_INET6)
342                         hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
343                                            &hslot2->head);
344                 else
345                         hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
346                                            &hslot2->head);
347                 hslot2->count++;
348                 spin_unlock(&hslot2->lock);
349         }
350         sock_set_flag(sk, SOCK_RCU_FREE);
351         error = 0;
352 fail_unlock:
353         spin_unlock_bh(&hslot->lock);
354 fail:
355         return error;
356 }
357 EXPORT_SYMBOL(udp_lib_get_port);
358 
359 /* match_wildcard == true:  0.0.0.0 equals to any IPv4 addresses
360  * match_wildcard == false: addresses must be exactly the same, i.e.
361  *                          0.0.0.0 only equals to 0.0.0.0
362  */
363 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2,
364                          bool match_wildcard)
365 {
366         struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
367 
368         if (!ipv6_only_sock(sk2)) {
369                 if (inet1->inet_rcv_saddr == inet2->inet_rcv_saddr)
370                         return 1;
371                 if (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr)
372                         return match_wildcard;
373         }
374         return 0;
375 }
376 
377 static u32 udp4_portaddr_hash(const struct net *net, __be32 saddr,
378                               unsigned int port)
379 {
380         return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
381 }
382 
383 int udp_v4_get_port(struct sock *sk, unsigned short snum)
384 {
385         unsigned int hash2_nulladdr =
386                 udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
387         unsigned int hash2_partial =
388                 udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
389 
390         /* precompute partial secondary hash */
391         udp_sk(sk)->udp_portaddr_hash = hash2_partial;
392         return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
393 }
394 
395 static int compute_score(struct sock *sk, struct net *net,
396                          __be32 saddr, __be16 sport,
397                          __be32 daddr, unsigned short hnum, int dif)
398 {
399         int score;
400         struct inet_sock *inet;
401 
402         if (!net_eq(sock_net(sk), net) ||
403             udp_sk(sk)->udp_port_hash != hnum ||
404             ipv6_only_sock(sk))
405                 return -1;
406 
407         score = (sk->sk_family == PF_INET) ? 2 : 1;
408         inet = inet_sk(sk);
409 
410         if (inet->inet_rcv_saddr) {
411                 if (inet->inet_rcv_saddr != daddr)
412                         return -1;
413                 score += 4;
414         }
415 
416         if (inet->inet_daddr) {
417                 if (inet->inet_daddr != saddr)
418                         return -1;
419                 score += 4;
420         }
421 
422         if (inet->inet_dport) {
423                 if (inet->inet_dport != sport)
424                         return -1;
425                 score += 4;
426         }
427 
428         if (sk->sk_bound_dev_if) {
429                 if (sk->sk_bound_dev_if != dif)
430                         return -1;
431                 score += 4;
432         }
433         if (sk->sk_incoming_cpu == raw_smp_processor_id())
434                 score++;
435         return score;
436 }
437 
438 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
439                        const __u16 lport, const __be32 faddr,
440                        const __be16 fport)
441 {
442         static u32 udp_ehash_secret __read_mostly;
443 
444         net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
445 
446         return __inet_ehashfn(laddr, lport, faddr, fport,
447                               udp_ehash_secret + net_hash_mix(net));
448 }
449 
450 /* called with rcu_read_lock() */
451 static struct sock *udp4_lib_lookup2(struct net *net,
452                 __be32 saddr, __be16 sport,
453                 __be32 daddr, unsigned int hnum, int dif,
454                 struct udp_hslot *hslot2,
455                 struct sk_buff *skb)
456 {
457         struct sock *sk, *result;
458         int score, badness, matches = 0, reuseport = 0;
459         u32 hash = 0;
460 
461         result = NULL;
462         badness = 0;
463         udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
464                 score = compute_score(sk, net, saddr, sport,
465                                       daddr, hnum, dif);
466                 if (score > badness) {
467                         reuseport = sk->sk_reuseport;
468                         if (reuseport) {
469                                 hash = udp_ehashfn(net, daddr, hnum,
470                                                    saddr, sport);
471                                 result = reuseport_select_sock(sk, hash, skb,
472                                                         sizeof(struct udphdr));
473                                 if (result)
474                                         return result;
475                                 matches = 1;
476                         }
477                         badness = score;
478                         result = sk;
479                 } else if (score == badness && reuseport) {
480                         matches++;
481                         if (reciprocal_scale(hash, matches) == 0)
482                                 result = sk;
483                         hash = next_pseudo_random32(hash);
484                 }
485         }
486         return result;
487 }
488 
489 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
490  * harder than this. -DaveM
491  */
492 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
493                 __be16 sport, __be32 daddr, __be16 dport,
494                 int dif, struct udp_table *udptable, struct sk_buff *skb)
495 {
496         struct sock *sk, *result;
497         unsigned short hnum = ntohs(dport);
498         unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
499         struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
500         int score, badness, matches = 0, reuseport = 0;
501         u32 hash = 0;
502 
503         if (hslot->count > 10) {
504                 hash2 = udp4_portaddr_hash(net, daddr, hnum);
505                 slot2 = hash2 & udptable->mask;
506                 hslot2 = &udptable->hash2[slot2];
507                 if (hslot->count < hslot2->count)
508                         goto begin;
509 
510                 result = udp4_lib_lookup2(net, saddr, sport,
511                                           daddr, hnum, dif,
512                                           hslot2, skb);
513                 if (!result) {
514                         unsigned int old_slot2 = slot2;
515                         hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
516                         slot2 = hash2 & udptable->mask;
517                         /* avoid searching the same slot again. */
518                         if (unlikely(slot2 == old_slot2))
519                                 return result;
520 
521                         hslot2 = &udptable->hash2[slot2];
522                         if (hslot->count < hslot2->count)
523                                 goto begin;
524 
525                         result = udp4_lib_lookup2(net, saddr, sport,
526                                                   daddr, hnum, dif,
527                                                   hslot2, skb);
528                 }
529                 return result;
530         }
531 begin:
532         result = NULL;
533         badness = 0;
534         sk_for_each_rcu(sk, &hslot->head) {
535                 score = compute_score(sk, net, saddr, sport,
536                                       daddr, hnum, dif);
537                 if (score > badness) {
538                         reuseport = sk->sk_reuseport;
539                         if (reuseport) {
540                                 hash = udp_ehashfn(net, daddr, hnum,
541                                                    saddr, sport);
542                                 result = reuseport_select_sock(sk, hash, skb,
543                                                         sizeof(struct udphdr));
544                                 if (result)
545                                         return result;
546                                 matches = 1;
547                         }
548                         result = sk;
549                         badness = score;
550                 } else if (score == badness && reuseport) {
551                         matches++;
552                         if (reciprocal_scale(hash, matches) == 0)
553                                 result = sk;
554                         hash = next_pseudo_random32(hash);
555                 }
556         }
557         return result;
558 }
559 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
560 
561 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
562                                                  __be16 sport, __be16 dport,
563                                                  struct udp_table *udptable)
564 {
565         const struct iphdr *iph = ip_hdr(skb);
566 
567         return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
568                                  iph->daddr, dport, inet_iif(skb),
569                                  udptable, skb);
570 }
571 
572 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
573                                  __be16 sport, __be16 dport)
574 {
575         return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
576 }
577 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
578 
579 /* Must be called under rcu_read_lock().
580  * Does increment socket refcount.
581  */
582 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
583     IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
584     IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
585 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
586                              __be32 daddr, __be16 dport, int dif)
587 {
588         struct sock *sk;
589 
590         sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
591                                dif, &udp_table, NULL);
592         if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
593                 sk = NULL;
594         return sk;
595 }
596 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
597 #endif
598 
599 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
600                                        __be16 loc_port, __be32 loc_addr,
601                                        __be16 rmt_port, __be32 rmt_addr,
602                                        int dif, unsigned short hnum)
603 {
604         struct inet_sock *inet = inet_sk(sk);
605 
606         if (!net_eq(sock_net(sk), net) ||
607             udp_sk(sk)->udp_port_hash != hnum ||
608             (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
609             (inet->inet_dport != rmt_port && inet->inet_dport) ||
610             (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
611             ipv6_only_sock(sk) ||
612             (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif))
613                 return false;
614         if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif))
615                 return false;
616         return true;
617 }
618 
619 /*
620  * This routine is called by the ICMP module when it gets some
621  * sort of error condition.  If err < 0 then the socket should
622  * be closed and the error returned to the user.  If err > 0
623  * it's just the icmp type << 8 | icmp code.
624  * Header points to the ip header of the error packet. We move
625  * on past this. Then (as it used to claim before adjustment)
626  * header points to the first 8 bytes of the udp header.  We need
627  * to find the appropriate port.
628  */
629 
630 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
631 {
632         struct inet_sock *inet;
633         const struct iphdr *iph = (const struct iphdr *)skb->data;
634         struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
635         const int type = icmp_hdr(skb)->type;
636         const int code = icmp_hdr(skb)->code;
637         struct sock *sk;
638         int harderr;
639         int err;
640         struct net *net = dev_net(skb->dev);
641 
642         sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
643                         iph->saddr, uh->source, skb->dev->ifindex, udptable,
644                         NULL);
645         if (!sk) {
646                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
647                 return; /* No socket for error */
648         }
649 
650         err = 0;
651         harderr = 0;
652         inet = inet_sk(sk);
653 
654         switch (type) {
655         default:
656         case ICMP_TIME_EXCEEDED:
657                 err = EHOSTUNREACH;
658                 break;
659         case ICMP_SOURCE_QUENCH:
660                 goto out;
661         case ICMP_PARAMETERPROB:
662                 err = EPROTO;
663                 harderr = 1;
664                 break;
665         case ICMP_DEST_UNREACH:
666                 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
667                         ipv4_sk_update_pmtu(skb, sk, info);
668                         if (inet->pmtudisc != IP_PMTUDISC_DONT) {
669                                 err = EMSGSIZE;
670                                 harderr = 1;
671                                 break;
672                         }
673                         goto out;
674                 }
675                 err = EHOSTUNREACH;
676                 if (code <= NR_ICMP_UNREACH) {
677                         harderr = icmp_err_convert[code].fatal;
678                         err = icmp_err_convert[code].errno;
679                 }
680                 break;
681         case ICMP_REDIRECT:
682                 ipv4_sk_redirect(skb, sk);
683                 goto out;
684         }
685 
686         /*
687          *      RFC1122: OK.  Passes ICMP errors back to application, as per
688          *      4.1.3.3.
689          */
690         if (!inet->recverr) {
691                 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
692                         goto out;
693         } else
694                 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
695 
696         sk->sk_err = err;
697         sk->sk_error_report(sk);
698 out:
699         return;
700 }
701 
702 void udp_err(struct sk_buff *skb, u32 info)
703 {
704         __udp4_lib_err(skb, info, &udp_table);
705 }
706 
707 /*
708  * Throw away all pending data and cancel the corking. Socket is locked.
709  */
710 void udp_flush_pending_frames(struct sock *sk)
711 {
712         struct udp_sock *up = udp_sk(sk);
713 
714         if (up->pending) {
715                 up->len = 0;
716                 up->pending = 0;
717                 ip_flush_pending_frames(sk);
718         }
719 }
720 EXPORT_SYMBOL(udp_flush_pending_frames);
721 
722 /**
723  *      udp4_hwcsum  -  handle outgoing HW checksumming
724  *      @skb:   sk_buff containing the filled-in UDP header
725  *              (checksum field must be zeroed out)
726  *      @src:   source IP address
727  *      @dst:   destination IP address
728  */
729 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
730 {
731         struct udphdr *uh = udp_hdr(skb);
732         int offset = skb_transport_offset(skb);
733         int len = skb->len - offset;
734         int hlen = len;
735         __wsum csum = 0;
736 
737         if (!skb_has_frag_list(skb)) {
738                 /*
739                  * Only one fragment on the socket.
740                  */
741                 skb->csum_start = skb_transport_header(skb) - skb->head;
742                 skb->csum_offset = offsetof(struct udphdr, check);
743                 uh->check = ~csum_tcpudp_magic(src, dst, len,
744                                                IPPROTO_UDP, 0);
745         } else {
746                 struct sk_buff *frags;
747 
748                 /*
749                  * HW-checksum won't work as there are two or more
750                  * fragments on the socket so that all csums of sk_buffs
751                  * should be together
752                  */
753                 skb_walk_frags(skb, frags) {
754                         csum = csum_add(csum, frags->csum);
755                         hlen -= frags->len;
756                 }
757 
758                 csum = skb_checksum(skb, offset, hlen, csum);
759                 skb->ip_summed = CHECKSUM_NONE;
760 
761                 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
762                 if (uh->check == 0)
763                         uh->check = CSUM_MANGLED_0;
764         }
765 }
766 EXPORT_SYMBOL_GPL(udp4_hwcsum);
767 
768 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
769  * for the simple case like when setting the checksum for a UDP tunnel.
770  */
771 void udp_set_csum(bool nocheck, struct sk_buff *skb,
772                   __be32 saddr, __be32 daddr, int len)
773 {
774         struct udphdr *uh = udp_hdr(skb);
775 
776         if (nocheck) {
777                 uh->check = 0;
778         } else if (skb_is_gso(skb)) {
779                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
780         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
781                 uh->check = 0;
782                 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
783                 if (uh->check == 0)
784                         uh->check = CSUM_MANGLED_0;
785         } else {
786                 skb->ip_summed = CHECKSUM_PARTIAL;
787                 skb->csum_start = skb_transport_header(skb) - skb->head;
788                 skb->csum_offset = offsetof(struct udphdr, check);
789                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
790         }
791 }
792 EXPORT_SYMBOL(udp_set_csum);
793 
794 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
795 {
796         struct sock *sk = skb->sk;
797         struct inet_sock *inet = inet_sk(sk);
798         struct udphdr *uh;
799         int err = 0;
800         int is_udplite = IS_UDPLITE(sk);
801         int offset = skb_transport_offset(skb);
802         int len = skb->len - offset;
803         __wsum csum = 0;
804 
805         /*
806          * Create a UDP header
807          */
808         uh = udp_hdr(skb);
809         uh->source = inet->inet_sport;
810         uh->dest = fl4->fl4_dport;
811         uh->len = htons(len);
812         uh->check = 0;
813 
814         if (is_udplite)                                  /*     UDP-Lite      */
815                 csum = udplite_csum(skb);
816 
817         else if (sk->sk_no_check_tx) {   /* UDP csum disabled */
818 
819                 skb->ip_summed = CHECKSUM_NONE;
820                 goto send;
821 
822         } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
823 
824                 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
825                 goto send;
826 
827         } else
828                 csum = udp_csum(skb);
829 
830         /* add protocol-dependent pseudo-header */
831         uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
832                                       sk->sk_protocol, csum);
833         if (uh->check == 0)
834                 uh->check = CSUM_MANGLED_0;
835 
836 send:
837         err = ip_send_skb(sock_net(sk), skb);
838         if (err) {
839                 if (err == -ENOBUFS && !inet->recverr) {
840                         UDP_INC_STATS(sock_net(sk),
841                                       UDP_MIB_SNDBUFERRORS, is_udplite);
842                         err = 0;
843                 }
844         } else
845                 UDP_INC_STATS(sock_net(sk),
846                               UDP_MIB_OUTDATAGRAMS, is_udplite);
847         return err;
848 }
849 
850 /*
851  * Push out all pending data as one UDP datagram. Socket is locked.
852  */
853 int udp_push_pending_frames(struct sock *sk)
854 {
855         struct udp_sock  *up = udp_sk(sk);
856         struct inet_sock *inet = inet_sk(sk);
857         struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
858         struct sk_buff *skb;
859         int err = 0;
860 
861         skb = ip_finish_skb(sk, fl4);
862         if (!skb)
863                 goto out;
864 
865         err = udp_send_skb(skb, fl4);
866 
867 out:
868         up->len = 0;
869         up->pending = 0;
870         return err;
871 }
872 EXPORT_SYMBOL(udp_push_pending_frames);
873 
874 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
875 {
876         struct inet_sock *inet = inet_sk(sk);
877         struct udp_sock *up = udp_sk(sk);
878         struct flowi4 fl4_stack;
879         struct flowi4 *fl4;
880         int ulen = len;
881         struct ipcm_cookie ipc;
882         struct rtable *rt = NULL;
883         int free = 0;
884         int connected = 0;
885         __be32 daddr, faddr, saddr;
886         __be16 dport;
887         u8  tos;
888         int err, is_udplite = IS_UDPLITE(sk);
889         int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
890         int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
891         struct sk_buff *skb;
892         struct ip_options_data opt_copy;
893 
894         if (len > 0xFFFF)
895                 return -EMSGSIZE;
896 
897         /*
898          *      Check the flags.
899          */
900 
901         if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
902                 return -EOPNOTSUPP;
903 
904         ipc.opt = NULL;
905         ipc.tx_flags = 0;
906         ipc.ttl = 0;
907         ipc.tos = -1;
908 
909         getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
910 
911         fl4 = &inet->cork.fl.u.ip4;
912         if (up->pending) {
913                 /*
914                  * There are pending frames.
915                  * The socket lock must be held while it's corked.
916                  */
917                 lock_sock(sk);
918                 if (likely(up->pending)) {
919                         if (unlikely(up->pending != AF_INET)) {
920                                 release_sock(sk);
921                                 return -EINVAL;
922                         }
923                         goto do_append_data;
924                 }
925                 release_sock(sk);
926         }
927         ulen += sizeof(struct udphdr);
928 
929         /*
930          *      Get and verify the address.
931          */
932         if (msg->msg_name) {
933                 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
934                 if (msg->msg_namelen < sizeof(*usin))
935                         return -EINVAL;
936                 if (usin->sin_family != AF_INET) {
937                         if (usin->sin_family != AF_UNSPEC)
938                                 return -EAFNOSUPPORT;
939                 }
940 
941                 daddr = usin->sin_addr.s_addr;
942                 dport = usin->sin_port;
943                 if (dport == 0)
944                         return -EINVAL;
945         } else {
946                 if (sk->sk_state != TCP_ESTABLISHED)
947                         return -EDESTADDRREQ;
948                 daddr = inet->inet_daddr;
949                 dport = inet->inet_dport;
950                 /* Open fast path for connected socket.
951                    Route will not be used, if at least one option is set.
952                  */
953                 connected = 1;
954         }
955 
956         ipc.sockc.tsflags = sk->sk_tsflags;
957         ipc.addr = inet->inet_saddr;
958         ipc.oif = sk->sk_bound_dev_if;
959 
960         if (msg->msg_controllen) {
961                 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
962                 if (unlikely(err)) {
963                         kfree(ipc.opt);
964                         return err;
965                 }
966                 if (ipc.opt)
967                         free = 1;
968                 connected = 0;
969         }
970         if (!ipc.opt) {
971                 struct ip_options_rcu *inet_opt;
972 
973                 rcu_read_lock();
974                 inet_opt = rcu_dereference(inet->inet_opt);
975                 if (inet_opt) {
976                         memcpy(&opt_copy, inet_opt,
977                                sizeof(*inet_opt) + inet_opt->opt.optlen);
978                         ipc.opt = &opt_copy.opt;
979                 }
980                 rcu_read_unlock();
981         }
982 
983         saddr = ipc.addr;
984         ipc.addr = faddr = daddr;
985 
986         sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
987 
988         if (ipc.opt && ipc.opt->opt.srr) {
989                 if (!daddr)
990                         return -EINVAL;
991                 faddr = ipc.opt->opt.faddr;
992                 connected = 0;
993         }
994         tos = get_rttos(&ipc, inet);
995         if (sock_flag(sk, SOCK_LOCALROUTE) ||
996             (msg->msg_flags & MSG_DONTROUTE) ||
997             (ipc.opt && ipc.opt->opt.is_strictroute)) {
998                 tos |= RTO_ONLINK;
999                 connected = 0;
1000         }
1001 
1002         if (ipv4_is_multicast(daddr)) {
1003                 if (!ipc.oif)
1004                         ipc.oif = inet->mc_index;
1005                 if (!saddr)
1006                         saddr = inet->mc_addr;
1007                 connected = 0;
1008         } else if (!ipc.oif)
1009                 ipc.oif = inet->uc_index;
1010 
1011         if (connected)
1012                 rt = (struct rtable *)sk_dst_check(sk, 0);
1013 
1014         if (!rt) {
1015                 struct net *net = sock_net(sk);
1016                 __u8 flow_flags = inet_sk_flowi_flags(sk);
1017 
1018                 fl4 = &fl4_stack;
1019 
1020                 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1021                                    RT_SCOPE_UNIVERSE, sk->sk_protocol,
1022                                    flow_flags,
1023                                    faddr, saddr, dport, inet->inet_sport,
1024                                    sk->sk_uid);
1025 
1026                 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1027                 rt = ip_route_output_flow(net, fl4, sk);
1028                 if (IS_ERR(rt)) {
1029                         err = PTR_ERR(rt);
1030                         rt = NULL;
1031                         if (err == -ENETUNREACH)
1032                                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1033                         goto out;
1034                 }
1035 
1036                 err = -EACCES;
1037                 if ((rt->rt_flags & RTCF_BROADCAST) &&
1038                     !sock_flag(sk, SOCK_BROADCAST))
1039                         goto out;
1040                 if (connected)
1041                         sk_dst_set(sk, dst_clone(&rt->dst));
1042         }
1043 
1044         if (msg->msg_flags&MSG_CONFIRM)
1045                 goto do_confirm;
1046 back_from_confirm:
1047 
1048         saddr = fl4->saddr;
1049         if (!ipc.addr)
1050                 daddr = ipc.addr = fl4->daddr;
1051 
1052         /* Lockless fast path for the non-corking case. */
1053         if (!corkreq) {
1054                 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1055                                   sizeof(struct udphdr), &ipc, &rt,
1056                                   msg->msg_flags);
1057                 err = PTR_ERR(skb);
1058                 if (!IS_ERR_OR_NULL(skb))
1059                         err = udp_send_skb(skb, fl4);
1060                 goto out;
1061         }
1062 
1063         lock_sock(sk);
1064         if (unlikely(up->pending)) {
1065                 /* The socket is already corked while preparing it. */
1066                 /* ... which is an evident application bug. --ANK */
1067                 release_sock(sk);
1068 
1069                 net_dbg_ratelimited("cork app bug 2\n");
1070                 err = -EINVAL;
1071                 goto out;
1072         }
1073         /*
1074          *      Now cork the socket to pend data.
1075          */
1076         fl4 = &inet->cork.fl.u.ip4;
1077         fl4->daddr = daddr;
1078         fl4->saddr = saddr;
1079         fl4->fl4_dport = dport;
1080         fl4->fl4_sport = inet->inet_sport;
1081         up->pending = AF_INET;
1082 
1083 do_append_data:
1084         up->len += ulen;
1085         err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1086                              sizeof(struct udphdr), &ipc, &rt,
1087                              corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1088         if (err)
1089                 udp_flush_pending_frames(sk);
1090         else if (!corkreq)
1091                 err = udp_push_pending_frames(sk);
1092         else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1093                 up->pending = 0;
1094         release_sock(sk);
1095 
1096 out:
1097         ip_rt_put(rt);
1098         if (free)
1099                 kfree(ipc.opt);
1100         if (!err)
1101                 return len;
1102         /*
1103          * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1104          * ENOBUFS might not be good (it's not tunable per se), but otherwise
1105          * we don't have a good statistic (IpOutDiscards but it can be too many
1106          * things).  We could add another new stat but at least for now that
1107          * seems like overkill.
1108          */
1109         if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1110                 UDP_INC_STATS(sock_net(sk),
1111                               UDP_MIB_SNDBUFERRORS, is_udplite);
1112         }
1113         return err;
1114 
1115 do_confirm:
1116         dst_confirm(&rt->dst);
1117         if (!(msg->msg_flags&MSG_PROBE) || len)
1118                 goto back_from_confirm;
1119         err = 0;
1120         goto out;
1121 }
1122 EXPORT_SYMBOL(udp_sendmsg);
1123 
1124 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1125                  size_t size, int flags)
1126 {
1127         struct inet_sock *inet = inet_sk(sk);
1128         struct udp_sock *up = udp_sk(sk);
1129         int ret;
1130 
1131         if (flags & MSG_SENDPAGE_NOTLAST)
1132                 flags |= MSG_MORE;
1133 
1134         if (!up->pending) {
1135                 struct msghdr msg = {   .msg_flags = flags|MSG_MORE };
1136 
1137                 /* Call udp_sendmsg to specify destination address which
1138                  * sendpage interface can't pass.
1139                  * This will succeed only when the socket is connected.
1140                  */
1141                 ret = udp_sendmsg(sk, &msg, 0);
1142                 if (ret < 0)
1143                         return ret;
1144         }
1145 
1146         lock_sock(sk);
1147 
1148         if (unlikely(!up->pending)) {
1149                 release_sock(sk);
1150 
1151                 net_dbg_ratelimited("udp cork app bug 3\n");
1152                 return -EINVAL;
1153         }
1154 
1155         ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1156                              page, offset, size, flags);
1157         if (ret == -EOPNOTSUPP) {
1158                 release_sock(sk);
1159                 return sock_no_sendpage(sk->sk_socket, page, offset,
1160                                         size, flags);
1161         }
1162         if (ret < 0) {
1163                 udp_flush_pending_frames(sk);
1164                 goto out;
1165         }
1166 
1167         up->len += size;
1168         if (!(up->corkflag || (flags&MSG_MORE)))
1169                 ret = udp_push_pending_frames(sk);
1170         if (!ret)
1171                 ret = size;
1172 out:
1173         release_sock(sk);
1174         return ret;
1175 }
1176 
1177 /* fully reclaim rmem/fwd memory allocated for skb */
1178 static void udp_rmem_release(struct sock *sk, int size, int partial)
1179 {
1180         struct udp_sock *up = udp_sk(sk);
1181         int amt;
1182 
1183         if (likely(partial)) {
1184                 up->forward_deficit += size;
1185                 size = up->forward_deficit;
1186                 if (size < (sk->sk_rcvbuf >> 2) &&
1187                     !skb_queue_empty(&sk->sk_receive_queue))
1188                         return;
1189         } else {
1190                 size += up->forward_deficit;
1191         }
1192         up->forward_deficit = 0;
1193 
1194         sk->sk_forward_alloc += size;
1195         amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1196         sk->sk_forward_alloc -= amt;
1197 
1198         if (amt)
1199                 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1200 
1201         atomic_sub(size, &sk->sk_rmem_alloc);
1202 }
1203 
1204 /* Note: called with sk_receive_queue.lock held.
1205  * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1206  * This avoids a cache line miss while receive_queue lock is held.
1207  * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1208  */
1209 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1210 {
1211         udp_rmem_release(sk, skb->dev_scratch, 1);
1212 }
1213 EXPORT_SYMBOL(udp_skb_destructor);
1214 
1215 /* Idea of busylocks is to let producers grab an extra spinlock
1216  * to relieve pressure on the receive_queue spinlock shared by consumer.
1217  * Under flood, this means that only one producer can be in line
1218  * trying to acquire the receive_queue spinlock.
1219  * These busylock can be allocated on a per cpu manner, instead of a
1220  * per socket one (that would consume a cache line per socket)
1221  */
1222 static int udp_busylocks_log __read_mostly;
1223 static spinlock_t *udp_busylocks __read_mostly;
1224 
1225 static spinlock_t *busylock_acquire(void *ptr)
1226 {
1227         spinlock_t *busy;
1228 
1229         busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1230         spin_lock(busy);
1231         return busy;
1232 }
1233 
1234 static void busylock_release(spinlock_t *busy)
1235 {
1236         if (busy)
1237                 spin_unlock(busy);
1238 }
1239 
1240 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1241 {
1242         struct sk_buff_head *list = &sk->sk_receive_queue;
1243         int rmem, delta, amt, err = -ENOMEM;
1244         spinlock_t *busy = NULL;
1245         int size;
1246 
1247         /* try to avoid the costly atomic add/sub pair when the receive
1248          * queue is full; always allow at least a packet
1249          */
1250         rmem = atomic_read(&sk->sk_rmem_alloc);
1251         if (rmem > sk->sk_rcvbuf)
1252                 goto drop;
1253 
1254         /* Under mem pressure, it might be helpful to help udp_recvmsg()
1255          * having linear skbs :
1256          * - Reduce memory overhead and thus increase receive queue capacity
1257          * - Less cache line misses at copyout() time
1258          * - Less work at consume_skb() (less alien page frag freeing)
1259          */
1260         if (rmem > (sk->sk_rcvbuf >> 1)) {
1261                 skb_condense(skb);
1262 
1263                 busy = busylock_acquire(sk);
1264         }
1265         size = skb->truesize;
1266         /* Copy skb->truesize into skb->dev_scratch to avoid a cache line miss
1267          * in udp_skb_destructor()
1268          */
1269         skb->dev_scratch = size;
1270 
1271         /* we drop only if the receive buf is full and the receive
1272          * queue contains some other skb
1273          */
1274         rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1275         if (rmem > (size + sk->sk_rcvbuf))
1276                 goto uncharge_drop;
1277 
1278         spin_lock(&list->lock);
1279         if (size >= sk->sk_forward_alloc) {
1280                 amt = sk_mem_pages(size);
1281                 delta = amt << SK_MEM_QUANTUM_SHIFT;
1282                 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1283                         err = -ENOBUFS;
1284                         spin_unlock(&list->lock);
1285                         goto uncharge_drop;
1286                 }
1287 
1288                 sk->sk_forward_alloc += delta;
1289         }
1290 
1291         sk->sk_forward_alloc -= size;
1292 
1293         /* no need to setup a destructor, we will explicitly release the
1294          * forward allocated memory on dequeue
1295          */
1296         sock_skb_set_dropcount(sk, skb);
1297 
1298         __skb_queue_tail(list, skb);
1299         spin_unlock(&list->lock);
1300 
1301         if (!sock_flag(sk, SOCK_DEAD))
1302                 sk->sk_data_ready(sk);
1303 
1304         busylock_release(busy);
1305         return 0;
1306 
1307 uncharge_drop:
1308         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1309 
1310 drop:
1311         atomic_inc(&sk->sk_drops);
1312         busylock_release(busy);
1313         return err;
1314 }
1315 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1316 
1317 void udp_destruct_sock(struct sock *sk)
1318 {
1319         /* reclaim completely the forward allocated memory */
1320         unsigned int total = 0;
1321         struct sk_buff *skb;
1322 
1323         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
1324                 total += skb->truesize;
1325                 kfree_skb(skb);
1326         }
1327         udp_rmem_release(sk, total, 0);
1328 
1329         inet_sock_destruct(sk);
1330 }
1331 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1332 
1333 int udp_init_sock(struct sock *sk)
1334 {
1335         sk->sk_destruct = udp_destruct_sock;
1336         return 0;
1337 }
1338 EXPORT_SYMBOL_GPL(udp_init_sock);
1339 
1340 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1341 {
1342         if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1343                 bool slow = lock_sock_fast(sk);
1344 
1345                 sk_peek_offset_bwd(sk, len);
1346                 unlock_sock_fast(sk, slow);
1347         }
1348         consume_skb(skb);
1349 }
1350 EXPORT_SYMBOL_GPL(skb_consume_udp);
1351 
1352 /**
1353  *      first_packet_length     - return length of first packet in receive queue
1354  *      @sk: socket
1355  *
1356  *      Drops all bad checksum frames, until a valid one is found.
1357  *      Returns the length of found skb, or -1 if none is found.
1358  */
1359 static int first_packet_length(struct sock *sk)
1360 {
1361         struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1362         struct sk_buff *skb;
1363         int total = 0;
1364         int res;
1365 
1366         spin_lock_bh(&rcvq->lock);
1367         while ((skb = skb_peek(rcvq)) != NULL &&
1368                 udp_lib_checksum_complete(skb)) {
1369                 __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1370                                 IS_UDPLITE(sk));
1371                 __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1372                                 IS_UDPLITE(sk));
1373                 atomic_inc(&sk->sk_drops);
1374                 __skb_unlink(skb, rcvq);
1375                 total += skb->truesize;
1376                 kfree_skb(skb);
1377         }
1378         res = skb ? skb->len : -1;
1379         if (total)
1380                 udp_rmem_release(sk, total, 1);
1381         spin_unlock_bh(&rcvq->lock);
1382         return res;
1383 }
1384 
1385 /*
1386  *      IOCTL requests applicable to the UDP protocol
1387  */
1388 
1389 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1390 {
1391         switch (cmd) {
1392         case SIOCOUTQ:
1393         {
1394                 int amount = sk_wmem_alloc_get(sk);
1395 
1396                 return put_user(amount, (int __user *)arg);
1397         }
1398 
1399         case SIOCINQ:
1400         {
1401                 int amount = max_t(int, 0, first_packet_length(sk));
1402 
1403                 return put_user(amount, (int __user *)arg);
1404         }
1405 
1406         default:
1407                 return -ENOIOCTLCMD;
1408         }
1409 
1410         return 0;
1411 }
1412 EXPORT_SYMBOL(udp_ioctl);
1413 
1414 /*
1415  *      This should be easy, if there is something there we
1416  *      return it, otherwise we block.
1417  */
1418 
1419 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1420                 int flags, int *addr_len)
1421 {
1422         struct inet_sock *inet = inet_sk(sk);
1423         DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1424         struct sk_buff *skb;
1425         unsigned int ulen, copied;
1426         int peeked, peeking, off;
1427         int err;
1428         int is_udplite = IS_UDPLITE(sk);
1429         bool checksum_valid = false;
1430 
1431         if (flags & MSG_ERRQUEUE)
1432                 return ip_recv_error(sk, msg, len, addr_len);
1433 
1434 try_again:
1435         peeking = off = sk_peek_offset(sk, flags);
1436         skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1437         if (!skb)
1438                 return err;
1439         if (ccs_socket_post_recvmsg_permission(sk, skb, flags))
1440                 return -EAGAIN; /* Hope less harmful than -EPERM. */
1441 
1442         ulen = skb->len;
1443         copied = len;
1444         if (copied > ulen - off)
1445                 copied = ulen - off;
1446         else if (copied < ulen)
1447                 msg->msg_flags |= MSG_TRUNC;
1448 
1449         /*
1450          * If checksum is needed at all, try to do it while copying the
1451          * data.  If the data is truncated, or if we only want a partial
1452          * coverage checksum (UDP-Lite), do it before the copy.
1453          */
1454 
1455         if (copied < ulen || peeking ||
1456             (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1457                 checksum_valid = !udp_lib_checksum_complete(skb);
1458                 if (!checksum_valid)
1459                         goto csum_copy_err;
1460         }
1461 
1462         if (checksum_valid || skb_csum_unnecessary(skb))
1463                 err = skb_copy_datagram_msg(skb, off, msg, copied);
1464         else {
1465                 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1466 
1467                 if (err == -EINVAL)
1468                         goto csum_copy_err;
1469         }
1470 
1471         if (unlikely(err)) {
1472                 if (!peeked) {
1473                         atomic_inc(&sk->sk_drops);
1474                         UDP_INC_STATS(sock_net(sk),
1475                                       UDP_MIB_INERRORS, is_udplite);
1476                 }
1477                 kfree_skb(skb);
1478                 return err;
1479         }
1480 
1481         if (!peeked)
1482                 UDP_INC_STATS(sock_net(sk),
1483                               UDP_MIB_INDATAGRAMS, is_udplite);
1484 
1485         sock_recv_ts_and_drops(msg, sk, skb);
1486 
1487         /* Copy the address. */
1488         if (sin) {
1489                 sin->sin_family = AF_INET;
1490                 sin->sin_port = udp_hdr(skb)->source;
1491                 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1492                 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1493                 *addr_len = sizeof(*sin);
1494         }
1495         if (inet->cmsg_flags)
1496                 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1497 
1498         err = copied;
1499         if (flags & MSG_TRUNC)
1500                 err = ulen;
1501 
1502         skb_consume_udp(sk, skb, peeking ? -err : err);
1503         return err;
1504 
1505 csum_copy_err:
1506         if (!__sk_queue_drop_skb(sk, skb, flags, udp_skb_destructor)) {
1507                 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1508                 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1509         }
1510         kfree_skb(skb);
1511 
1512         /* starting over for a new packet, but check if we need to yield */
1513         cond_resched();
1514         msg->msg_flags &= ~MSG_TRUNC;
1515         goto try_again;
1516 }
1517 
1518 int __udp_disconnect(struct sock *sk, int flags)
1519 {
1520         struct inet_sock *inet = inet_sk(sk);
1521         /*
1522          *      1003.1g - break association.
1523          */
1524 
1525         sk->sk_state = TCP_CLOSE;
1526         inet->inet_daddr = 0;
1527         inet->inet_dport = 0;
1528         sock_rps_reset_rxhash(sk);
1529         sk->sk_bound_dev_if = 0;
1530         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1531                 inet_reset_saddr(sk);
1532 
1533         if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1534                 sk->sk_prot->unhash(sk);
1535                 inet->inet_sport = 0;
1536         }
1537         sk_dst_reset(sk);
1538         return 0;
1539 }
1540 EXPORT_SYMBOL(__udp_disconnect);
1541 
1542 int udp_disconnect(struct sock *sk, int flags)
1543 {
1544         lock_sock(sk);
1545         __udp_disconnect(sk, flags);
1546         release_sock(sk);
1547         return 0;
1548 }
1549 EXPORT_SYMBOL(udp_disconnect);
1550 
1551 void udp_lib_unhash(struct sock *sk)
1552 {
1553         if (sk_hashed(sk)) {
1554                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1555                 struct udp_hslot *hslot, *hslot2;
1556 
1557                 hslot  = udp_hashslot(udptable, sock_net(sk),
1558                                       udp_sk(sk)->udp_port_hash);
1559                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1560 
1561                 spin_lock_bh(&hslot->lock);
1562                 if (rcu_access_pointer(sk->sk_reuseport_cb))
1563                         reuseport_detach_sock(sk);
1564                 if (sk_del_node_init_rcu(sk)) {
1565                         hslot->count--;
1566                         inet_sk(sk)->inet_num = 0;
1567                         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1568 
1569                         spin_lock(&hslot2->lock);
1570                         hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1571                         hslot2->count--;
1572                         spin_unlock(&hslot2->lock);
1573                 }
1574                 spin_unlock_bh(&hslot->lock);
1575         }
1576 }
1577 EXPORT_SYMBOL(udp_lib_unhash);
1578 
1579 /*
1580  * inet_rcv_saddr was changed, we must rehash secondary hash
1581  */
1582 void udp_lib_rehash(struct sock *sk, u16 newhash)
1583 {
1584         if (sk_hashed(sk)) {
1585                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1586                 struct udp_hslot *hslot, *hslot2, *nhslot2;
1587 
1588                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1589                 nhslot2 = udp_hashslot2(udptable, newhash);
1590                 udp_sk(sk)->udp_portaddr_hash = newhash;
1591 
1592                 if (hslot2 != nhslot2 ||
1593                     rcu_access_pointer(sk->sk_reuseport_cb)) {
1594                         hslot = udp_hashslot(udptable, sock_net(sk),
1595                                              udp_sk(sk)->udp_port_hash);
1596                         /* we must lock primary chain too */
1597                         spin_lock_bh(&hslot->lock);
1598                         if (rcu_access_pointer(sk->sk_reuseport_cb))
1599                                 reuseport_detach_sock(sk);
1600 
1601                         if (hslot2 != nhslot2) {
1602                                 spin_lock(&hslot2->lock);
1603                                 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1604                                 hslot2->count--;
1605                                 spin_unlock(&hslot2->lock);
1606 
1607                                 spin_lock(&nhslot2->lock);
1608                                 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1609                                                          &nhslot2->head);
1610                                 nhslot2->count++;
1611                                 spin_unlock(&nhslot2->lock);
1612                         }
1613 
1614                         spin_unlock_bh(&hslot->lock);
1615                 }
1616         }
1617 }
1618 EXPORT_SYMBOL(udp_lib_rehash);
1619 
1620 static void udp_v4_rehash(struct sock *sk)
1621 {
1622         u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1623                                           inet_sk(sk)->inet_rcv_saddr,
1624                                           inet_sk(sk)->inet_num);
1625         udp_lib_rehash(sk, new_hash);
1626 }
1627 
1628 int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1629 {
1630         int rc;
1631 
1632         if (inet_sk(sk)->inet_daddr) {
1633                 sock_rps_save_rxhash(sk, skb);
1634                 sk_mark_napi_id(sk, skb);
1635                 sk_incoming_cpu_update(sk);
1636         } else {
1637                 sk_mark_napi_id_once(sk, skb);
1638         }
1639 
1640         rc = __udp_enqueue_schedule_skb(sk, skb);
1641         if (rc < 0) {
1642                 int is_udplite = IS_UDPLITE(sk);
1643 
1644                 /* Note that an ENOMEM error is charged twice */
1645                 if (rc == -ENOMEM)
1646                         UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1647                                         is_udplite);
1648                 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1649                 kfree_skb(skb);
1650                 trace_udp_fail_queue_rcv_skb(rc, sk);
1651                 return -1;
1652         }
1653 
1654         return 0;
1655 }
1656 
1657 static struct static_key udp_encap_needed __read_mostly;
1658 void udp_encap_enable(void)
1659 {
1660         if (!static_key_enabled(&udp_encap_needed))
1661                 static_key_slow_inc(&udp_encap_needed);
1662 }
1663 EXPORT_SYMBOL(udp_encap_enable);
1664 
1665 /* returns:
1666  *  -1: error
1667  *   0: success
1668  *  >0: "udp encap" protocol resubmission
1669  *
1670  * Note that in the success and error cases, the skb is assumed to
1671  * have either been requeued or freed.
1672  */
1673 int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1674 {
1675         struct udp_sock *up = udp_sk(sk);
1676         int is_udplite = IS_UDPLITE(sk);
1677 
1678         /*
1679          *      Charge it to the socket, dropping if the queue is full.
1680          */
1681         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1682                 goto drop;
1683         nf_reset(skb);
1684 
1685         if (static_key_false(&udp_encap_needed) && up->encap_type) {
1686                 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1687 
1688                 /*
1689                  * This is an encapsulation socket so pass the skb to
1690                  * the socket's udp_encap_rcv() hook. Otherwise, just
1691                  * fall through and pass this up the UDP socket.
1692                  * up->encap_rcv() returns the following value:
1693                  * =0 if skb was successfully passed to the encap
1694                  *    handler or was discarded by it.
1695                  * >0 if skb should be passed on to UDP.
1696                  * <0 if skb should be resubmitted as proto -N
1697                  */
1698 
1699                 /* if we're overly short, let UDP handle it */
1700                 encap_rcv = ACCESS_ONCE(up->encap_rcv);
1701                 if (encap_rcv) {
1702                         int ret;
1703 
1704                         /* Verify checksum before giving to encap */
1705                         if (udp_lib_checksum_complete(skb))
1706                                 goto csum_error;
1707 
1708                         ret = encap_rcv(sk, skb);
1709                         if (ret <= 0) {
1710                                 __UDP_INC_STATS(sock_net(sk),
1711                                                 UDP_MIB_INDATAGRAMS,
1712                                                 is_udplite);
1713                                 return -ret;
1714                         }
1715                 }
1716 
1717                 /* FALLTHROUGH -- it's a UDP Packet */
1718         }
1719 
1720         /*
1721          *      UDP-Lite specific tests, ignored on UDP sockets
1722          */
1723         if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1724 
1725                 /*
1726                  * MIB statistics other than incrementing the error count are
1727                  * disabled for the following two types of errors: these depend
1728                  * on the application settings, not on the functioning of the
1729                  * protocol stack as such.
1730                  *
1731                  * RFC 3828 here recommends (sec 3.3): "There should also be a
1732                  * way ... to ... at least let the receiving application block
1733                  * delivery of packets with coverage values less than a value
1734                  * provided by the application."
1735                  */
1736                 if (up->pcrlen == 0) {          /* full coverage was set  */
1737                         net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1738                                             UDP_SKB_CB(skb)->cscov, skb->len);
1739                         goto drop;
1740                 }
1741                 /* The next case involves violating the min. coverage requested
1742                  * by the receiver. This is subtle: if receiver wants x and x is
1743                  * greater than the buffersize/MTU then receiver will complain
1744                  * that it wants x while sender emits packets of smaller size y.
1745                  * Therefore the above ...()->partial_cov statement is essential.
1746                  */
1747                 if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1748                         net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1749                                             UDP_SKB_CB(skb)->cscov, up->pcrlen);
1750                         goto drop;
1751                 }
1752         }
1753 
1754         if (rcu_access_pointer(sk->sk_filter) &&
1755             udp_lib_checksum_complete(skb))
1756                         goto csum_error;
1757 
1758         if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1759                 goto drop;
1760 
1761         udp_csum_pull_header(skb);
1762 
1763         ipv4_pktinfo_prepare(sk, skb);
1764         return __udp_queue_rcv_skb(sk, skb);
1765 
1766 csum_error:
1767         __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1768 drop:
1769         __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1770         atomic_inc(&sk->sk_drops);
1771         kfree_skb(skb);
1772         return -1;
1773 }
1774 
1775 /* For TCP sockets, sk_rx_dst is protected by socket lock
1776  * For UDP, we use xchg() to guard against concurrent changes.
1777  */
1778 static void udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1779 {
1780         struct dst_entry *old;
1781 
1782         dst_hold(dst);
1783         old = xchg(&sk->sk_rx_dst, dst);
1784         dst_release(old);
1785 }
1786 
1787 /*
1788  *      Multicasts and broadcasts go to each listener.
1789  *
1790  *      Note: called only from the BH handler context.
1791  */
1792 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1793                                     struct udphdr  *uh,
1794                                     __be32 saddr, __be32 daddr,
1795                                     struct udp_table *udptable,
1796                                     int proto)
1797 {
1798         struct sock *sk, *first = NULL;
1799         unsigned short hnum = ntohs(uh->dest);
1800         struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1801         unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1802         unsigned int offset = offsetof(typeof(*sk), sk_node);
1803         int dif = skb->dev->ifindex;
1804         struct hlist_node *node;
1805         struct sk_buff *nskb;
1806 
1807         if (use_hash2) {
1808                 hash2_any = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1809                             udptable->mask;
1810                 hash2 = udp4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1811 start_lookup:
1812                 hslot = &udptable->hash2[hash2];
1813                 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1814         }
1815 
1816         sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1817                 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1818                                          uh->source, saddr, dif, hnum))
1819                         continue;
1820 
1821                 if (!first) {
1822                         first = sk;
1823                         continue;
1824                 }
1825                 nskb = skb_clone(skb, GFP_ATOMIC);
1826 
1827                 if (unlikely(!nskb)) {
1828                         atomic_inc(&sk->sk_drops);
1829                         __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1830                                         IS_UDPLITE(sk));
1831                         __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1832                                         IS_UDPLITE(sk));
1833                         continue;
1834                 }
1835                 if (udp_queue_rcv_skb(sk, nskb) > 0)
1836                         consume_skb(nskb);
1837         }
1838 
1839         /* Also lookup *:port if we are using hash2 and haven't done so yet. */
1840         if (use_hash2 && hash2 != hash2_any) {
1841                 hash2 = hash2_any;
1842                 goto start_lookup;
1843         }
1844 
1845         if (first) {
1846                 if (udp_queue_rcv_skb(first, skb) > 0)
1847                         consume_skb(skb);
1848         } else {
1849                 kfree_skb(skb);
1850                 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
1851                                 proto == IPPROTO_UDPLITE);
1852         }
1853         return 0;
1854 }
1855 
1856 /* Initialize UDP checksum. If exited with zero value (success),
1857  * CHECKSUM_UNNECESSARY means, that no more checks are required.
1858  * Otherwise, csum completion requires chacksumming packet body,
1859  * including udp header and folding it to skb->csum.
1860  */
1861 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1862                                  int proto)
1863 {
1864         int err;
1865 
1866         UDP_SKB_CB(skb)->partial_cov = 0;
1867         UDP_SKB_CB(skb)->cscov = skb->len;
1868 
1869         if (proto == IPPROTO_UDPLITE) {
1870                 err = udplite_checksum_init(skb, uh);
1871                 if (err)
1872                         return err;
1873         }
1874 
1875         /* Note, we are only interested in != 0 or == 0, thus the
1876          * force to int.
1877          */
1878         return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
1879                                                          inet_compute_pseudo);
1880 }
1881 
1882 /*
1883  *      All we need to do is get the socket, and then do a checksum.
1884  */
1885 
1886 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1887                    int proto)
1888 {
1889         struct sock *sk;
1890         struct udphdr *uh;
1891         unsigned short ulen;
1892         struct rtable *rt = skb_rtable(skb);
1893         __be32 saddr, daddr;
1894         struct net *net = dev_net(skb->dev);
1895 
1896         /*
1897          *  Validate the packet.
1898          */
1899         if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1900                 goto drop;              /* No space for header. */
1901 
1902         uh   = udp_hdr(skb);
1903         ulen = ntohs(uh->len);
1904         saddr = ip_hdr(skb)->saddr;
1905         daddr = ip_hdr(skb)->daddr;
1906 
1907         if (ulen > skb->len)
1908                 goto short_packet;
1909 
1910         if (proto == IPPROTO_UDP) {
1911                 /* UDP validates ulen. */
1912                 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1913                         goto short_packet;
1914                 uh = udp_hdr(skb);
1915         }
1916 
1917         if (udp4_csum_init(skb, uh, proto))
1918                 goto csum_error;
1919 
1920         sk = skb_steal_sock(skb);
1921         if (sk) {
1922                 struct dst_entry *dst = skb_dst(skb);
1923                 int ret;
1924 
1925                 if (unlikely(sk->sk_rx_dst != dst))
1926                         udp_sk_rx_dst_set(sk, dst);
1927 
1928                 ret = udp_queue_rcv_skb(sk, skb);
1929                 sock_put(sk);
1930                 /* a return value > 0 means to resubmit the input, but
1931                  * it wants the return to be -protocol, or 0
1932                  */
1933                 if (ret > 0)
1934                         return -ret;
1935                 return 0;
1936         }
1937 
1938         if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1939                 return __udp4_lib_mcast_deliver(net, skb, uh,
1940                                                 saddr, daddr, udptable, proto);
1941 
1942         sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1943         if (sk) {
1944                 int ret;
1945 
1946                 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
1947                         skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
1948                                                  inet_compute_pseudo);
1949 
1950                 ret = udp_queue_rcv_skb(sk, skb);
1951 
1952                 /* a return value > 0 means to resubmit the input, but
1953                  * it wants the return to be -protocol, or 0
1954                  */
1955                 if (ret > 0)
1956                         return -ret;
1957                 return 0;
1958         }
1959 
1960         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1961                 goto drop;
1962         nf_reset(skb);
1963 
1964         /* No socket. Drop packet silently, if checksum is wrong */
1965         if (udp_lib_checksum_complete(skb))
1966                 goto csum_error;
1967 
1968         __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1969         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1970 
1971         /*
1972          * Hmm.  We got an UDP packet to a port to which we
1973          * don't wanna listen.  Ignore it.
1974          */
1975         kfree_skb(skb);
1976         return 0;
1977 
1978 short_packet:
1979         net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1980                             proto == IPPROTO_UDPLITE ? "Lite" : "",
1981                             &saddr, ntohs(uh->source),
1982                             ulen, skb->len,
1983                             &daddr, ntohs(uh->dest));
1984         goto drop;
1985 
1986 csum_error:
1987         /*
1988          * RFC1122: OK.  Discards the bad packet silently (as far as
1989          * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1990          */
1991         net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1992                             proto == IPPROTO_UDPLITE ? "Lite" : "",
1993                             &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1994                             ulen);
1995         __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
1996 drop:
1997         __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1998         kfree_skb(skb);
1999         return 0;
2000 }
2001 
2002 /* We can only early demux multicast if there is a single matching socket.
2003  * If more than one socket found returns NULL
2004  */
2005 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2006                                                   __be16 loc_port, __be32 loc_addr,
2007                                                   __be16 rmt_port, __be32 rmt_addr,
2008                                                   int dif)
2009 {
2010         struct sock *sk, *result;
2011         unsigned short hnum = ntohs(loc_port);
2012         unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2013         struct udp_hslot *hslot = &udp_table.hash[slot];
2014 
2015         /* Do not bother scanning a too big list */
2016         if (hslot->count > 10)
2017                 return NULL;
2018 
2019         result = NULL;
2020         sk_for_each_rcu(sk, &hslot->head) {
2021                 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2022                                         rmt_port, rmt_addr, dif, hnum)) {
2023                         if (result)
2024                                 return NULL;
2025                         result = sk;
2026                 }
2027         }
2028 
2029         return result;
2030 }
2031 
2032 /* For unicast we should only early demux connected sockets or we can
2033  * break forwarding setups.  The chains here can be long so only check
2034  * if the first socket is an exact match and if not move on.
2035  */
2036 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2037                                             __be16 loc_port, __be32 loc_addr,
2038                                             __be16 rmt_port, __be32 rmt_addr,
2039                                             int dif)
2040 {
2041         unsigned short hnum = ntohs(loc_port);
2042         unsigned int hash2 = udp4_portaddr_hash(net, loc_addr, hnum);
2043         unsigned int slot2 = hash2 & udp_table.mask;
2044         struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2045         INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2046         const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2047         struct sock *sk;
2048 
2049         udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2050                 if (INET_MATCH(sk, net, acookie, rmt_addr,
2051                                loc_addr, ports, dif))
2052                         return sk;
2053                 /* Only check first socket in chain */
2054                 break;
2055         }
2056         return NULL;
2057 }
2058 
2059 void udp_v4_early_demux(struct sk_buff *skb)
2060 {
2061         struct net *net = dev_net(skb->dev);
2062         const struct iphdr *iph;
2063         const struct udphdr *uh;
2064         struct sock *sk = NULL;
2065         struct dst_entry *dst;
2066         int dif = skb->dev->ifindex;
2067         int ours;
2068 
2069         /* validate the packet */
2070         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2071                 return;
2072 
2073         iph = ip_hdr(skb);
2074         uh = udp_hdr(skb);
2075 
2076         if (skb->pkt_type == PACKET_BROADCAST ||
2077             skb->pkt_type == PACKET_MULTICAST) {
2078                 struct in_device *in_dev = __in_dev_get_rcu(skb->dev);
2079 
2080                 if (!in_dev)
2081                         return;
2082 
2083                 /* we are supposed to accept bcast packets */
2084                 if (skb->pkt_type == PACKET_MULTICAST) {
2085                         ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2086                                                iph->protocol);
2087                         if (!ours)
2088                                 return;
2089                 }
2090 
2091                 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2092                                                    uh->source, iph->saddr, dif);
2093         } else if (skb->pkt_type == PACKET_HOST) {
2094                 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2095                                              uh->source, iph->saddr, dif);
2096         }
2097 
2098         if (!sk || !atomic_inc_not_zero_hint(&sk->sk_refcnt, 2))
2099                 return;
2100 
2101         skb->sk = sk;
2102         skb->destructor = sock_efree;
2103         dst = READ_ONCE(sk->sk_rx_dst);
2104 
2105         if (dst)
2106                 dst = dst_check(dst, 0);
2107         if (dst) {
2108                 /* DST_NOCACHE can not be used without taking a reference */
2109                 if (dst->flags & DST_NOCACHE) {
2110                         if (likely(atomic_inc_not_zero(&dst->__refcnt)))
2111                                 skb_dst_set(skb, dst);
2112                 } else {
2113                         skb_dst_set_noref(skb, dst);
2114                 }
2115         }
2116 }
2117 
2118 int udp_rcv(struct sk_buff *skb)
2119 {
2120         return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2121 }
2122 
2123 void udp_destroy_sock(struct sock *sk)
2124 {
2125         struct udp_sock *up = udp_sk(sk);
2126         bool slow = lock_sock_fast(sk);
2127         udp_flush_pending_frames(sk);
2128         unlock_sock_fast(sk, slow);
2129         if (static_key_false(&udp_encap_needed) && up->encap_type) {
2130                 void (*encap_destroy)(struct sock *sk);
2131                 encap_destroy = ACCESS_ONCE(up->encap_destroy);
2132                 if (encap_destroy)
2133                         encap_destroy(sk);
2134         }
2135 }
2136 
2137 /*
2138  *      Socket option code for UDP
2139  */
2140 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2141                        char __user *optval, unsigned int optlen,
2142                        int (*push_pending_frames)(struct sock *))
2143 {
2144         struct udp_sock *up = udp_sk(sk);
2145         int val, valbool;
2146         int err = 0;
2147         int is_udplite = IS_UDPLITE(sk);
2148 
2149         if (optlen < sizeof(int))
2150                 return -EINVAL;
2151 
2152         if (get_user(val, (int __user *)optval))
2153                 return -EFAULT;
2154 
2155         valbool = val ? 1 : 0;
2156 
2157         switch (optname) {
2158         case UDP_CORK:
2159                 if (val != 0) {
2160                         up->corkflag = 1;
2161                 } else {
2162                         up->corkflag = 0;
2163                         lock_sock(sk);
2164                         push_pending_frames(sk);
2165                         release_sock(sk);
2166                 }
2167                 break;
2168 
2169         case UDP_ENCAP:
2170                 switch (val) {
2171                 case 0:
2172                 case UDP_ENCAP_ESPINUDP:
2173                 case UDP_ENCAP_ESPINUDP_NON_IKE:
2174                         up->encap_rcv = xfrm4_udp_encap_rcv;
2175                         /* FALLTHROUGH */
2176                 case UDP_ENCAP_L2TPINUDP:
2177                         up->encap_type = val;
2178                         udp_encap_enable();
2179                         break;
2180                 default:
2181                         err = -ENOPROTOOPT;
2182                         break;
2183                 }
2184                 break;
2185 
2186         case UDP_NO_CHECK6_TX:
2187                 up->no_check6_tx = valbool;
2188                 break;
2189 
2190         case UDP_NO_CHECK6_RX:
2191                 up->no_check6_rx = valbool;
2192                 break;
2193 
2194         /*
2195          *      UDP-Lite's partial checksum coverage (RFC 3828).
2196          */
2197         /* The sender sets actual checksum coverage length via this option.
2198          * The case coverage > packet length is handled by send module. */
2199         case UDPLITE_SEND_CSCOV:
2200                 if (!is_udplite)         /* Disable the option on UDP sockets */
2201                         return -ENOPROTOOPT;
2202                 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2203                         val = 8;
2204                 else if (val > USHRT_MAX)
2205                         val = USHRT_MAX;
2206                 up->pcslen = val;
2207                 up->pcflag |= UDPLITE_SEND_CC;
2208                 break;
2209 
2210         /* The receiver specifies a minimum checksum coverage value. To make
2211          * sense, this should be set to at least 8 (as done below). If zero is
2212          * used, this again means full checksum coverage.                     */
2213         case UDPLITE_RECV_CSCOV:
2214                 if (!is_udplite)         /* Disable the option on UDP sockets */
2215                         return -ENOPROTOOPT;
2216                 if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2217                         val = 8;
2218                 else if (val > USHRT_MAX)
2219                         val = USHRT_MAX;
2220                 up->pcrlen = val;
2221                 up->pcflag |= UDPLITE_RECV_CC;
2222                 break;
2223 
2224         default:
2225                 err = -ENOPROTOOPT;
2226                 break;
2227         }
2228 
2229         return err;
2230 }
2231 EXPORT_SYMBOL(udp_lib_setsockopt);
2232 
2233 int udp_setsockopt(struct sock *sk, int level, int optname,
2234                    char __user *optval, unsigned int optlen)
2235 {
2236         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2237                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2238                                           udp_push_pending_frames);
2239         return ip_setsockopt(sk, level, optname, optval, optlen);
2240 }
2241 
2242 #ifdef CONFIG_COMPAT
2243 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2244                           char __user *optval, unsigned int optlen)
2245 {
2246         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2247                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2248                                           udp_push_pending_frames);
2249         return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2250 }
2251 #endif
2252 
2253 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2254                        char __user *optval, int __user *optlen)
2255 {
2256         struct udp_sock *up = udp_sk(sk);
2257         int val, len;
2258 
2259         if (get_user(len, optlen))
2260                 return -EFAULT;
2261 
2262         len = min_t(unsigned int, len, sizeof(int));
2263 
2264         if (len < 0)
2265                 return -EINVAL;
2266 
2267         switch (optname) {
2268         case UDP_CORK:
2269                 val = up->corkflag;
2270                 break;
2271 
2272         case UDP_ENCAP:
2273                 val = up->encap_type;
2274                 break;
2275 
2276         case UDP_NO_CHECK6_TX:
2277                 val = up->no_check6_tx;
2278                 break;
2279 
2280         case UDP_NO_CHECK6_RX:
2281                 val = up->no_check6_rx;
2282                 break;
2283 
2284         /* The following two cannot be changed on UDP sockets, the return is
2285          * always 0 (which corresponds to the full checksum coverage of UDP). */
2286         case UDPLITE_SEND_CSCOV:
2287                 val = up->pcslen;
2288                 break;
2289 
2290         case UDPLITE_RECV_CSCOV:
2291                 val = up->pcrlen;
2292                 break;
2293 
2294         default:
2295                 return -ENOPROTOOPT;
2296         }
2297 
2298         if (put_user(len, optlen))
2299                 return -EFAULT;
2300         if (copy_to_user(optval, &val, len))
2301                 return -EFAULT;
2302         return 0;
2303 }
2304 EXPORT_SYMBOL(udp_lib_getsockopt);
2305 
2306 int udp_getsockopt(struct sock *sk, int level, int optname,
2307                    char __user *optval, int __user *optlen)
2308 {
2309         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2310                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2311         return ip_getsockopt(sk, level, optname, optval, optlen);
2312 }
2313 
2314 #ifdef CONFIG_COMPAT
2315 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2316                                  char __user *optval, int __user *optlen)
2317 {
2318         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2319                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2320         return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2321 }
2322 #endif
2323 /**
2324  *      udp_poll - wait for a UDP event.
2325  *      @file - file struct
2326  *      @sock - socket
2327  *      @wait - poll table
2328  *
2329  *      This is same as datagram poll, except for the special case of
2330  *      blocking sockets. If application is using a blocking fd
2331  *      and a packet with checksum error is in the queue;
2332  *      then it could get return from select indicating data available
2333  *      but then block when reading it. Add special case code
2334  *      to work around these arguably broken applications.
2335  */
2336 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2337 {
2338         unsigned int mask = datagram_poll(file, sock, wait);
2339         struct sock *sk = sock->sk;
2340 
2341         sock_rps_record_flow(sk);
2342 
2343         /* Check for false positives due to checksum errors */
2344         if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2345             !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2346                 mask &= ~(POLLIN | POLLRDNORM);
2347 
2348         return mask;
2349 
2350 }
2351 EXPORT_SYMBOL(udp_poll);
2352 
2353 int udp_abort(struct sock *sk, int err)
2354 {
2355         lock_sock(sk);
2356 
2357         sk->sk_err = err;
2358         sk->sk_error_report(sk);
2359         __udp_disconnect(sk, 0);
2360 
2361         release_sock(sk);
2362 
2363         return 0;
2364 }
2365 EXPORT_SYMBOL_GPL(udp_abort);
2366 
2367 struct proto udp_prot = {
2368         .name              = "UDP",
2369         .owner             = THIS_MODULE,
2370         .close             = udp_lib_close,
2371         .connect           = ip4_datagram_connect,
2372         .disconnect        = udp_disconnect,
2373         .ioctl             = udp_ioctl,
2374         .init              = udp_init_sock,
2375         .destroy           = udp_destroy_sock,
2376         .setsockopt        = udp_setsockopt,
2377         .getsockopt        = udp_getsockopt,
2378         .sendmsg           = udp_sendmsg,
2379         .recvmsg           = udp_recvmsg,
2380         .sendpage          = udp_sendpage,
2381         .release_cb        = ip4_datagram_release_cb,
2382         .hash              = udp_lib_hash,
2383         .unhash            = udp_lib_unhash,
2384         .rehash            = udp_v4_rehash,
2385         .get_port          = udp_v4_get_port,
2386         .memory_allocated  = &udp_memory_allocated,
2387         .sysctl_mem        = sysctl_udp_mem,
2388         .sysctl_wmem       = &sysctl_udp_wmem_min,
2389         .sysctl_rmem       = &sysctl_udp_rmem_min,
2390         .obj_size          = sizeof(struct udp_sock),
2391         .h.udp_table       = &udp_table,
2392 #ifdef CONFIG_COMPAT
2393         .compat_setsockopt = compat_udp_setsockopt,
2394         .compat_getsockopt = compat_udp_getsockopt,
2395 #endif
2396         .diag_destroy      = udp_abort,
2397 };
2398 EXPORT_SYMBOL(udp_prot);
2399 
2400 /* ------------------------------------------------------------------------ */
2401 #ifdef CONFIG_PROC_FS
2402 
2403 static struct sock *udp_get_first(struct seq_file *seq, int start)
2404 {
2405         struct sock *sk;
2406         struct udp_iter_state *state = seq->private;
2407         struct net *net = seq_file_net(seq);
2408 
2409         for (state->bucket = start; state->bucket <= state->udp_table->mask;
2410              ++state->bucket) {
2411                 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2412 
2413                 if (hlist_empty(&hslot->head))
2414                         continue;
2415 
2416                 spin_lock_bh(&hslot->lock);
2417                 sk_for_each(sk, &hslot->head) {
2418                         if (!net_eq(sock_net(sk), net))
2419                                 continue;
2420                         if (sk->sk_family == state->family)
2421                                 goto found;
2422                 }
2423                 spin_unlock_bh(&hslot->lock);
2424         }
2425         sk = NULL;
2426 found:
2427         return sk;
2428 }
2429 
2430 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2431 {
2432         struct udp_iter_state *state = seq->private;
2433         struct net *net = seq_file_net(seq);
2434 
2435         do {
2436                 sk = sk_next(sk);
2437         } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2438 
2439         if (!sk) {
2440                 if (state->bucket <= state->udp_table->mask)
2441                         spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2442                 return udp_get_first(seq, state->bucket + 1);
2443         }
2444         return sk;
2445 }
2446 
2447 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2448 {
2449         struct sock *sk = udp_get_first(seq, 0);
2450 
2451         if (sk)
2452                 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2453                         --pos;
2454         return pos ? NULL : sk;
2455 }
2456 
2457 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2458 {
2459         struct udp_iter_state *state = seq->private;
2460         state->bucket = MAX_UDP_PORTS;
2461 
2462         return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2463 }
2464 
2465 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2466 {
2467         struct sock *sk;
2468 
2469         if (v == SEQ_START_TOKEN)
2470                 sk = udp_get_idx(seq, 0);
2471         else
2472                 sk = udp_get_next(seq, v);
2473 
2474         ++*pos;
2475         return sk;
2476 }
2477 
2478 static void udp_seq_stop(struct seq_file *seq, void *v)
2479 {
2480         struct udp_iter_state *state = seq->private;
2481 
2482         if (state->bucket <= state->udp_table->mask)
2483                 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2484 }
2485 
2486 int udp_seq_open(struct inode *inode, struct file *file)
2487 {
2488         struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2489         struct udp_iter_state *s;
2490         int err;
2491 
2492         err = seq_open_net(inode, file, &afinfo->seq_ops,
2493                            sizeof(struct udp_iter_state));
2494         if (err < 0)
2495                 return err;
2496 
2497         s = ((struct seq_file *)file->private_data)->private;
2498         s->family               = afinfo->family;
2499         s->udp_table            = afinfo->udp_table;
2500         return err;
2501 }
2502 EXPORT_SYMBOL(udp_seq_open);
2503 
2504 /* ------------------------------------------------------------------------ */
2505 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2506 {
2507         struct proc_dir_entry *p;
2508         int rc = 0;
2509 
2510         afinfo->seq_ops.start           = udp_seq_start;
2511         afinfo->seq_ops.next            = udp_seq_next;
2512         afinfo->seq_ops.stop            = udp_seq_stop;
2513 
2514         p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2515                              afinfo->seq_fops, afinfo);
2516         if (!p)
2517                 rc = -ENOMEM;
2518         return rc;
2519 }
2520 EXPORT_SYMBOL(udp_proc_register);
2521 
2522 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2523 {
2524         remove_proc_entry(afinfo->name, net->proc_net);
2525 }
2526 EXPORT_SYMBOL(udp_proc_unregister);
2527 
2528 /* ------------------------------------------------------------------------ */
2529 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2530                 int bucket)
2531 {
2532         struct inet_sock *inet = inet_sk(sp);
2533         __be32 dest = inet->inet_daddr;
2534         __be32 src  = inet->inet_rcv_saddr;
2535         __u16 destp       = ntohs(inet->inet_dport);
2536         __u16 srcp        = ntohs(inet->inet_sport);
2537 
2538         seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2539                 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2540                 bucket, src, srcp, dest, destp, sp->sk_state,
2541                 sk_wmem_alloc_get(sp),
2542                 sk_rmem_alloc_get(sp),
2543                 0, 0L, 0,
2544                 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2545                 0, sock_i_ino(sp),
2546                 atomic_read(&sp->sk_refcnt), sp,
2547                 atomic_read(&sp->sk_drops));
2548 }
2549 
2550 int udp4_seq_show(struct seq_file *seq, void *v)
2551 {
2552         seq_setwidth(seq, 127);
2553         if (v == SEQ_START_TOKEN)
2554                 seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2555                            "rx_queue tr tm->when retrnsmt   uid  timeout "
2556                            "inode ref pointer drops");
2557         else {
2558                 struct udp_iter_state *state = seq->private;
2559 
2560                 udp4_format_sock(v, seq, state->bucket);
2561         }
2562         seq_pad(seq, '\n');
2563         return 0;
2564 }
2565 
2566 static const struct file_operations udp_afinfo_seq_fops = {
2567         .owner    = THIS_MODULE,
2568         .open     = udp_seq_open,
2569         .read     = seq_read,
2570         .llseek   = seq_lseek,
2571         .release  = seq_release_net
2572 };
2573 
2574 /* ------------------------------------------------------------------------ */
2575 static struct udp_seq_afinfo udp4_seq_afinfo = {
2576         .name           = "udp",
2577         .family         = AF_INET,
2578         .udp_table      = &udp_table,
2579         .seq_fops       = &udp_afinfo_seq_fops,
2580         .seq_ops        = {
2581                 .show           = udp4_seq_show,
2582         },
2583 };
2584 
2585 static int __net_init udp4_proc_init_net(struct net *net)
2586 {
2587         return udp_proc_register(net, &udp4_seq_afinfo);
2588 }
2589 
2590 static void __net_exit udp4_proc_exit_net(struct net *net)
2591 {
2592         udp_proc_unregister(net, &udp4_seq_afinfo);
2593 }
2594 
2595 static struct pernet_operations udp4_net_ops = {
2596         .init = udp4_proc_init_net,
2597         .exit = udp4_proc_exit_net,
2598 };
2599 
2600 int __init udp4_proc_init(void)
2601 {
2602         return register_pernet_subsys(&udp4_net_ops);
2603 }
2604 
2605 void udp4_proc_exit(void)
2606 {
2607         unregister_pernet_subsys(&udp4_net_ops);
2608 }
2609 #endif /* CONFIG_PROC_FS */
2610 
2611 static __initdata unsigned long uhash_entries;
2612 static int __init set_uhash_entries(char *str)
2613 {
2614         ssize_t ret;
2615 
2616         if (!str)
2617                 return 0;
2618 
2619         ret = kstrtoul(str, 0, &uhash_entries);
2620         if (ret)
2621                 return 0;
2622 
2623         if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2624                 uhash_entries = UDP_HTABLE_SIZE_MIN;
2625         return 1;
2626 }
2627 __setup("uhash_entries=", set_uhash_entries);
2628 
2629 void __init udp_table_init(struct udp_table *table, const char *name)
2630 {
2631         unsigned int i;
2632 
2633         table->hash = alloc_large_system_hash(name,
2634                                               2 * sizeof(struct udp_hslot),
2635                                               uhash_entries,
2636                                               21, /* one slot per 2 MB */
2637                                               0,
2638                                               &table->log,
2639                                               &table->mask,
2640                                               UDP_HTABLE_SIZE_MIN,
2641                                               64 * 1024);
2642 
2643         table->hash2 = table->hash + (table->mask + 1);
2644         for (i = 0; i <= table->mask; i++) {
2645                 INIT_HLIST_HEAD(&table->hash[i].head);
2646                 table->hash[i].count = 0;
2647                 spin_lock_init(&table->hash[i].lock);
2648         }
2649         for (i = 0; i <= table->mask; i++) {
2650                 INIT_HLIST_HEAD(&table->hash2[i].head);
2651                 table->hash2[i].count = 0;
2652                 spin_lock_init(&table->hash2[i].lock);
2653         }
2654 }
2655 
2656 u32 udp_flow_hashrnd(void)
2657 {
2658         static u32 hashrnd __read_mostly;
2659 
2660         net_get_random_once(&hashrnd, sizeof(hashrnd));
2661 
2662         return hashrnd;
2663 }
2664 EXPORT_SYMBOL(udp_flow_hashrnd);
2665 
2666 void __init udp_init(void)
2667 {
2668         unsigned long limit;
2669         unsigned int i;
2670 
2671         udp_table_init(&udp_table, "UDP");
2672         limit = nr_free_buffer_pages() / 8;
2673         limit = max(limit, 128UL);
2674         sysctl_udp_mem[0] = limit / 4 * 3;
2675         sysctl_udp_mem[1] = limit;
2676         sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2677 
2678         sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2679         sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2680 
2681         /* 16 spinlocks per cpu */
2682         udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2683         udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2684                                 GFP_KERNEL);
2685         if (!udp_busylocks)
2686                 panic("UDP: failed to alloc udp_busylocks\n");
2687         for (i = 0; i < (1U << udp_busylocks_log); i++)
2688                 spin_lock_init(udp_busylocks + i);
2689 }
2690 

~ [ 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