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

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

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
  3  *              operating system.  INET is implemented using the  BSD Socket
  4  *              interface as the means of communication with the user level.
  5  *
  6  *              The User Datagram Protocol (UDP).
  7  *
  8  * Authors:     Ross Biro
  9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 10  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
 11  *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
 12  *              Hirokazu Takahashi, <taka@valinux.co.jp>
 13  *
 14  * Fixes:
 15  *              Alan Cox        :       verify_area() calls
 16  *              Alan Cox        :       stopped close while in use off icmp
 17  *                                      messages. Not a fix but a botch that
 18  *                                      for udp at least is 'valid'.
 19  *              Alan Cox        :       Fixed icmp handling properly
 20  *              Alan Cox        :       Correct error for oversized datagrams
 21  *              Alan Cox        :       Tidied select() semantics.
 22  *              Alan Cox        :       udp_err() fixed properly, also now
 23  *                                      select and read wake correctly on errors
 24  *              Alan Cox        :       udp_send verify_area moved to avoid mem leak
 25  *              Alan Cox        :       UDP can count its memory
 26  *              Alan Cox        :       send to an unknown connection causes
 27  *                                      an ECONNREFUSED off the icmp, but
 28  *                                      does NOT close.
 29  *              Alan Cox        :       Switched to new sk_buff handlers. No more backlog!
 30  *              Alan Cox        :       Using generic datagram code. Even smaller and the PEEK
 31  *                                      bug no longer crashes it.
 32  *              Fred Van Kempen :       Net2e support for sk->broadcast.
 33  *              Alan Cox        :       Uses skb_free_datagram
 34  *              Alan Cox        :       Added get/set sockopt support.
 35  *              Alan Cox        :       Broadcasting without option set returns EACCES.
 36  *              Alan Cox        :       No wakeup calls. Instead we now use the callbacks.
 37  *              Alan Cox        :       Use ip_tos and ip_ttl
 38  *              Alan Cox        :       SNMP Mibs
 39  *              Alan Cox        :       MSG_DONTROUTE, and 0.0.0.0 support.
 40  *              Matt Dillon     :       UDP length checks.
 41  *              Alan Cox        :       Smarter af_inet used properly.
 42  *              Alan Cox        :       Use new kernel side addressing.
 43  *              Alan Cox        :       Incorrect return on truncated datagram receive.
 44  *      Arnt Gulbrandsen        :       New udp_send and stuff
 45  *              Alan Cox        :       Cache last socket
 46  *              Alan Cox        :       Route cache
 47  *              Jon Peatfield   :       Minor efficiency fix to sendto().
 48  *              Mike Shaver     :       RFC1122 checks.
 49  *              Alan Cox        :       Nonblocking error fix.
 50  *      Willy Konynenberg       :       Transparent proxying support.
 51  *              Mike McLagan    :       Routing by source
 52  *              David S. Miller :       New socket lookup architecture.
 53  *                                      Last socket cache retained as it
 54  *                                      does have a high hit rate.
 55  *              Olaf Kirch      :       Don't linearise iovec on sendmsg.
 56  *              Andi Kleen      :       Some cleanups, cache destination entry
 57  *                                      for connect.
 58  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
 59  *              Melvin Smith    :       Check msg_name not msg_namelen in sendto(),
 60  *                                      return ENOTCONN for unconnected sockets (POSIX)
 61  *              Janos Farkas    :       don't deliver multi/broadcasts to a different
 62  *                                      bound-to-device socket
 63  *      Hirokazu Takahashi      :       HW checksumming for outgoing UDP
 64  *                                      datagrams.
 65  *      Hirokazu Takahashi      :       sendfile() on UDP works now.
 66  *              Arnaldo C. Melo :       convert /proc/net/udp to seq_file
 67  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
 68  *      Alexey Kuznetsov:               allow both IPv4 and IPv6 sockets to bind
 69  *                                      a single port at the same time.
 70  *      Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
 71  *      James Chapman           :       Add L2TP encapsulation type.
 72  *
 73  *
 74  *              This program is free software; you can redistribute it and/or
 75  *              modify it under the terms of the GNU General Public License
 76  *              as published by the Free Software Foundation; either version
 77  *              2 of the License, or (at your option) any later version.
 78  */
 79 
 80 #define pr_fmt(fmt) "UDP: " fmt
 81 
 82 #include <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 atomic_long_t udp_memory_allocated;
126 EXPORT_SYMBOL(udp_memory_allocated);
127 
128 #define MAX_UDP_PORTS 65536
129 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
130 
131 /* IPCB reference means this can not be used from early demux */
132 static bool udp_lib_exact_dif_match(struct net *net, struct sk_buff *skb)
133 {
134 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
135         if (!net->ipv4.sysctl_udp_l3mdev_accept &&
136             skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
137                 return true;
138 #endif
139         return false;
140 }
141 
142 static int udp_lib_lport_inuse(struct net *net, __u16 num,
143                                const struct udp_hslot *hslot,
144                                unsigned long *bitmap,
145                                struct sock *sk, unsigned int log)
146 {
147         struct sock *sk2;
148         kuid_t uid = sock_i_uid(sk);
149 
150         sk_for_each(sk2, &hslot->head) {
151                 if (net_eq(sock_net(sk2), net) &&
152                     sk2 != sk &&
153                     (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
154                     (!sk2->sk_reuse || !sk->sk_reuse) &&
155                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
156                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
157                     inet_rcv_saddr_equal(sk, sk2, true)) {
158                         if (sk2->sk_reuseport && sk->sk_reuseport &&
159                             !rcu_access_pointer(sk->sk_reuseport_cb) &&
160                             uid_eq(uid, sock_i_uid(sk2))) {
161                                 if (!bitmap)
162                                         return 0;
163                         } else {
164                                 if (!bitmap)
165                                         return 1;
166                                 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
167                                           bitmap);
168                         }
169                 }
170         }
171         return 0;
172 }
173 
174 /*
175  * Note: we still hold spinlock of primary hash chain, so no other writer
176  * can insert/delete a socket with local_port == num
177  */
178 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
179                                 struct udp_hslot *hslot2,
180                                 struct sock *sk)
181 {
182         struct sock *sk2;
183         kuid_t uid = sock_i_uid(sk);
184         int res = 0;
185 
186         spin_lock(&hslot2->lock);
187         udp_portaddr_for_each_entry(sk2, &hslot2->head) {
188                 if (net_eq(sock_net(sk2), net) &&
189                     sk2 != sk &&
190                     (udp_sk(sk2)->udp_port_hash == num) &&
191                     (!sk2->sk_reuse || !sk->sk_reuse) &&
192                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
193                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
194                     inet_rcv_saddr_equal(sk, sk2, true)) {
195                         if (sk2->sk_reuseport && sk->sk_reuseport &&
196                             !rcu_access_pointer(sk->sk_reuseport_cb) &&
197                             uid_eq(uid, sock_i_uid(sk2))) {
198                                 res = 0;
199                         } else {
200                                 res = 1;
201                         }
202                         break;
203                 }
204         }
205         spin_unlock(&hslot2->lock);
206         return res;
207 }
208 
209 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
210 {
211         struct net *net = sock_net(sk);
212         kuid_t uid = sock_i_uid(sk);
213         struct sock *sk2;
214 
215         sk_for_each(sk2, &hslot->head) {
216                 if (net_eq(sock_net(sk2), net) &&
217                     sk2 != sk &&
218                     sk2->sk_family == sk->sk_family &&
219                     ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
220                     (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
221                     (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
222                     sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
223                     inet_rcv_saddr_equal(sk, sk2, false)) {
224                         return reuseport_add_sock(sk, sk2);
225                 }
226         }
227 
228         return reuseport_alloc(sk);
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  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
237  *                   with NULL address
238  */
239 int udp_lib_get_port(struct sock *sk, unsigned short snum,
240                      unsigned int hash2_nulladdr)
241 {
242         struct udp_hslot *hslot, *hslot2;
243         struct udp_table *udptable = sk->sk_prot->h.udp_table;
244         int    error = 1;
245         struct net *net = sock_net(sk);
246 
247         if (!snum) {
248                 int low, high, remaining;
249                 unsigned int rand;
250                 unsigned short first, last;
251                 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
252 
253                 inet_get_local_port_range(net, &low, &high);
254                 remaining = (high - low) + 1;
255 
256                 rand = prandom_u32();
257                 first = reciprocal_scale(rand, remaining) + low;
258                 /*
259                  * force rand to be an odd multiple of UDP_HTABLE_SIZE
260                  */
261                 rand = (rand | 1) * (udptable->mask + 1);
262                 last = first + udptable->mask + 1;
263                 do {
264                         hslot = udp_hashslot(udptable, net, first);
265                         bitmap_zero(bitmap, PORTS_PER_CHAIN);
266                         spin_lock_bh(&hslot->lock);
267                         udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
268                                             udptable->log);
269 
270                         snum = first;
271                         /*
272                          * Iterate on all possible values of snum for this hash.
273                          * Using steps of an odd multiple of UDP_HTABLE_SIZE
274                          * give us randomization and full range coverage.
275                          */
276                         do {
277                                 if (low <= snum && snum <= high &&
278                                     !test_bit(snum >> udptable->log, bitmap) &&
279                                     !inet_is_local_reserved_port(net, snum))
280                                         goto found;
281                                 snum += rand;
282                         } while (snum != first);
283                         spin_unlock_bh(&hslot->lock);
284                         cond_resched();
285                 } while (++first != last);
286                 goto fail;
287         } else {
288                 hslot = udp_hashslot(udptable, net, snum);
289                 spin_lock_bh(&hslot->lock);
290                 if (hslot->count > 10) {
291                         int exist;
292                         unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
293 
294                         slot2          &= udptable->mask;
295                         hash2_nulladdr &= udptable->mask;
296 
297                         hslot2 = udp_hashslot2(udptable, slot2);
298                         if (hslot->count < hslot2->count)
299                                 goto scan_primary_hash;
300 
301                         exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
302                         if (!exist && (hash2_nulladdr != slot2)) {
303                                 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
304                                 exist = udp_lib_lport_inuse2(net, snum, hslot2,
305                                                              sk);
306                         }
307                         if (exist)
308                                 goto fail_unlock;
309                         else
310                                 goto found;
311                 }
312 scan_primary_hash:
313                 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
314                         goto fail_unlock;
315         }
316 found:
317         inet_sk(sk)->inet_num = snum;
318         udp_sk(sk)->udp_port_hash = snum;
319         udp_sk(sk)->udp_portaddr_hash ^= snum;
320         if (sk_unhashed(sk)) {
321                 if (sk->sk_reuseport &&
322                     udp_reuseport_add_sock(sk, hslot)) {
323                         inet_sk(sk)->inet_num = 0;
324                         udp_sk(sk)->udp_port_hash = 0;
325                         udp_sk(sk)->udp_portaddr_hash ^= snum;
326                         goto fail_unlock;
327                 }
328 
329                 sk_add_node_rcu(sk, &hslot->head);
330                 hslot->count++;
331                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
332 
333                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
334                 spin_lock(&hslot2->lock);
335                 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
336                     sk->sk_family == AF_INET6)
337                         hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
338                                            &hslot2->head);
339                 else
340                         hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
341                                            &hslot2->head);
342                 hslot2->count++;
343                 spin_unlock(&hslot2->lock);
344         }
345         sock_set_flag(sk, SOCK_RCU_FREE);
346         error = 0;
347 fail_unlock:
348         spin_unlock_bh(&hslot->lock);
349 fail:
350         return error;
351 }
352 EXPORT_SYMBOL(udp_lib_get_port);
353 
354 int udp_v4_get_port(struct sock *sk, unsigned short snum)
355 {
356         unsigned int hash2_nulladdr =
357                 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
358         unsigned int hash2_partial =
359                 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
360 
361         /* precompute partial secondary hash */
362         udp_sk(sk)->udp_portaddr_hash = hash2_partial;
363         return udp_lib_get_port(sk, snum, hash2_nulladdr);
364 }
365 
366 static int compute_score(struct sock *sk, struct net *net,
367                          __be32 saddr, __be16 sport,
368                          __be32 daddr, unsigned short hnum,
369                          int dif, int sdif, bool exact_dif)
370 {
371         int score;
372         struct inet_sock *inet;
373 
374         if (!net_eq(sock_net(sk), net) ||
375             udp_sk(sk)->udp_port_hash != hnum ||
376             ipv6_only_sock(sk))
377                 return -1;
378 
379         score = (sk->sk_family == PF_INET) ? 2 : 1;
380         inet = inet_sk(sk);
381 
382         if (inet->inet_rcv_saddr) {
383                 if (inet->inet_rcv_saddr != daddr)
384                         return -1;
385                 score += 4;
386         }
387 
388         if (inet->inet_daddr) {
389                 if (inet->inet_daddr != saddr)
390                         return -1;
391                 score += 4;
392         }
393 
394         if (inet->inet_dport) {
395                 if (inet->inet_dport != sport)
396                         return -1;
397                 score += 4;
398         }
399 
400         if (sk->sk_bound_dev_if || exact_dif) {
401                 bool dev_match = (sk->sk_bound_dev_if == dif ||
402                                   sk->sk_bound_dev_if == sdif);
403 
404                 if (!dev_match)
405                         return -1;
406                 if (sk->sk_bound_dev_if)
407                         score += 4;
408         }
409 
410         if (sk->sk_incoming_cpu == raw_smp_processor_id())
411                 score++;
412         return score;
413 }
414 
415 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
416                        const __u16 lport, const __be32 faddr,
417                        const __be16 fport)
418 {
419         static u32 udp_ehash_secret __read_mostly;
420 
421         net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
422 
423         return __inet_ehashfn(laddr, lport, faddr, fport,
424                               udp_ehash_secret + net_hash_mix(net));
425 }
426 
427 /* called with rcu_read_lock() */
428 static struct sock *udp4_lib_lookup2(struct net *net,
429                                      __be32 saddr, __be16 sport,
430                                      __be32 daddr, unsigned int hnum,
431                                      int dif, int sdif, bool exact_dif,
432                                      struct udp_hslot *hslot2,
433                                      struct sk_buff *skb)
434 {
435         struct sock *sk, *result;
436         int score, badness;
437         u32 hash = 0;
438 
439         result = NULL;
440         badness = 0;
441         udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
442                 score = compute_score(sk, net, saddr, sport,
443                                       daddr, hnum, dif, sdif, exact_dif);
444                 if (score > badness) {
445                         if (sk->sk_reuseport) {
446                                 hash = udp_ehashfn(net, daddr, hnum,
447                                                    saddr, sport);
448                                 result = reuseport_select_sock(sk, hash, skb,
449                                                         sizeof(struct udphdr));
450                                 if (result)
451                                         return result;
452                         }
453                         badness = score;
454                         result = sk;
455                 }
456         }
457         return result;
458 }
459 
460 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
461  * harder than this. -DaveM
462  */
463 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
464                 __be16 sport, __be32 daddr, __be16 dport, int dif,
465                 int sdif, struct udp_table *udptable, struct sk_buff *skb)
466 {
467         struct sock *sk, *result;
468         unsigned short hnum = ntohs(dport);
469         unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
470         struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
471         bool exact_dif = udp_lib_exact_dif_match(net, skb);
472         int score, badness;
473         u32 hash = 0;
474 
475         if (hslot->count > 10) {
476                 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
477                 slot2 = hash2 & udptable->mask;
478                 hslot2 = &udptable->hash2[slot2];
479                 if (hslot->count < hslot2->count)
480                         goto begin;
481 
482                 result = udp4_lib_lookup2(net, saddr, sport,
483                                           daddr, hnum, dif, sdif,
484                                           exact_dif, hslot2, skb);
485                 if (!result) {
486                         unsigned int old_slot2 = slot2;
487                         hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
488                         slot2 = hash2 & udptable->mask;
489                         /* avoid searching the same slot again. */
490                         if (unlikely(slot2 == old_slot2))
491                                 return result;
492 
493                         hslot2 = &udptable->hash2[slot2];
494                         if (hslot->count < hslot2->count)
495                                 goto begin;
496 
497                         result = udp4_lib_lookup2(net, saddr, sport,
498                                                   daddr, hnum, dif, sdif,
499                                                   exact_dif, hslot2, skb);
500                 }
501                 return result;
502         }
503 begin:
504         result = NULL;
505         badness = 0;
506         sk_for_each_rcu(sk, &hslot->head) {
507                 score = compute_score(sk, net, saddr, sport,
508                                       daddr, hnum, dif, sdif, exact_dif);
509                 if (score > badness) {
510                         if (sk->sk_reuseport) {
511                                 hash = udp_ehashfn(net, daddr, hnum,
512                                                    saddr, sport);
513                                 result = reuseport_select_sock(sk, hash, skb,
514                                                         sizeof(struct udphdr));
515                                 if (result)
516                                         return result;
517                         }
518                         result = sk;
519                         badness = score;
520                 }
521         }
522         return result;
523 }
524 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
525 
526 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
527                                                  __be16 sport, __be16 dport,
528                                                  struct udp_table *udptable)
529 {
530         const struct iphdr *iph = ip_hdr(skb);
531 
532         return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
533                                  iph->daddr, dport, inet_iif(skb),
534                                  inet_sdif(skb), udptable, skb);
535 }
536 
537 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
538                                  __be16 sport, __be16 dport)
539 {
540         return __udp4_lib_lookup_skb(skb, sport, dport, &udp_table);
541 }
542 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
543 
544 /* Must be called under rcu_read_lock().
545  * Does increment socket refcount.
546  */
547 #if IS_ENABLED(CONFIG_NETFILTER_XT_MATCH_SOCKET) || \
548     IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TPROXY) || \
549     IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
550 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
551                              __be32 daddr, __be16 dport, int dif)
552 {
553         struct sock *sk;
554 
555         sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
556                                dif, 0, &udp_table, NULL);
557         if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
558                 sk = NULL;
559         return sk;
560 }
561 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
562 #endif
563 
564 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
565                                        __be16 loc_port, __be32 loc_addr,
566                                        __be16 rmt_port, __be32 rmt_addr,
567                                        int dif, int sdif, unsigned short hnum)
568 {
569         struct inet_sock *inet = inet_sk(sk);
570 
571         if (!net_eq(sock_net(sk), net) ||
572             udp_sk(sk)->udp_port_hash != hnum ||
573             (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
574             (inet->inet_dport != rmt_port && inet->inet_dport) ||
575             (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
576             ipv6_only_sock(sk) ||
577             (sk->sk_bound_dev_if && sk->sk_bound_dev_if != dif &&
578              sk->sk_bound_dev_if != sdif))
579                 return false;
580         if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
581                 return false;
582         return true;
583 }
584 
585 /*
586  * This routine is called by the ICMP module when it gets some
587  * sort of error condition.  If err < 0 then the socket should
588  * be closed and the error returned to the user.  If err > 0
589  * it's just the icmp type << 8 | icmp code.
590  * Header points to the ip header of the error packet. We move
591  * on past this. Then (as it used to claim before adjustment)
592  * header points to the first 8 bytes of the udp header.  We need
593  * to find the appropriate port.
594  */
595 
596 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
597 {
598         struct inet_sock *inet;
599         const struct iphdr *iph = (const struct iphdr *)skb->data;
600         struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
601         const int type = icmp_hdr(skb)->type;
602         const int code = icmp_hdr(skb)->code;
603         struct sock *sk;
604         int harderr;
605         int err;
606         struct net *net = dev_net(skb->dev);
607 
608         sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
609                                iph->saddr, uh->source, skb->dev->ifindex, 0,
610                                udptable, NULL);
611         if (!sk) {
612                 __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
613                 return; /* No socket for error */
614         }
615 
616         err = 0;
617         harderr = 0;
618         inet = inet_sk(sk);
619 
620         switch (type) {
621         default:
622         case ICMP_TIME_EXCEEDED:
623                 err = EHOSTUNREACH;
624                 break;
625         case ICMP_SOURCE_QUENCH:
626                 goto out;
627         case ICMP_PARAMETERPROB:
628                 err = EPROTO;
629                 harderr = 1;
630                 break;
631         case ICMP_DEST_UNREACH:
632                 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
633                         ipv4_sk_update_pmtu(skb, sk, info);
634                         if (inet->pmtudisc != IP_PMTUDISC_DONT) {
635                                 err = EMSGSIZE;
636                                 harderr = 1;
637                                 break;
638                         }
639                         goto out;
640                 }
641                 err = EHOSTUNREACH;
642                 if (code <= NR_ICMP_UNREACH) {
643                         harderr = icmp_err_convert[code].fatal;
644                         err = icmp_err_convert[code].errno;
645                 }
646                 break;
647         case ICMP_REDIRECT:
648                 ipv4_sk_redirect(skb, sk);
649                 goto out;
650         }
651 
652         /*
653          *      RFC1122: OK.  Passes ICMP errors back to application, as per
654          *      4.1.3.3.
655          */
656         if (!inet->recverr) {
657                 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
658                         goto out;
659         } else
660                 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
661 
662         sk->sk_err = err;
663         sk->sk_error_report(sk);
664 out:
665         return;
666 }
667 
668 void udp_err(struct sk_buff *skb, u32 info)
669 {
670         __udp4_lib_err(skb, info, &udp_table);
671 }
672 
673 /*
674  * Throw away all pending data and cancel the corking. Socket is locked.
675  */
676 void udp_flush_pending_frames(struct sock *sk)
677 {
678         struct udp_sock *up = udp_sk(sk);
679 
680         if (up->pending) {
681                 up->len = 0;
682                 up->pending = 0;
683                 ip_flush_pending_frames(sk);
684         }
685 }
686 EXPORT_SYMBOL(udp_flush_pending_frames);
687 
688 /**
689  *      udp4_hwcsum  -  handle outgoing HW checksumming
690  *      @skb:   sk_buff containing the filled-in UDP header
691  *              (checksum field must be zeroed out)
692  *      @src:   source IP address
693  *      @dst:   destination IP address
694  */
695 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
696 {
697         struct udphdr *uh = udp_hdr(skb);
698         int offset = skb_transport_offset(skb);
699         int len = skb->len - offset;
700         int hlen = len;
701         __wsum csum = 0;
702 
703         if (!skb_has_frag_list(skb)) {
704                 /*
705                  * Only one fragment on the socket.
706                  */
707                 skb->csum_start = skb_transport_header(skb) - skb->head;
708                 skb->csum_offset = offsetof(struct udphdr, check);
709                 uh->check = ~csum_tcpudp_magic(src, dst, len,
710                                                IPPROTO_UDP, 0);
711         } else {
712                 struct sk_buff *frags;
713 
714                 /*
715                  * HW-checksum won't work as there are two or more
716                  * fragments on the socket so that all csums of sk_buffs
717                  * should be together
718                  */
719                 skb_walk_frags(skb, frags) {
720                         csum = csum_add(csum, frags->csum);
721                         hlen -= frags->len;
722                 }
723 
724                 csum = skb_checksum(skb, offset, hlen, csum);
725                 skb->ip_summed = CHECKSUM_NONE;
726 
727                 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
728                 if (uh->check == 0)
729                         uh->check = CSUM_MANGLED_0;
730         }
731 }
732 EXPORT_SYMBOL_GPL(udp4_hwcsum);
733 
734 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
735  * for the simple case like when setting the checksum for a UDP tunnel.
736  */
737 void udp_set_csum(bool nocheck, struct sk_buff *skb,
738                   __be32 saddr, __be32 daddr, int len)
739 {
740         struct udphdr *uh = udp_hdr(skb);
741 
742         if (nocheck) {
743                 uh->check = 0;
744         } else if (skb_is_gso(skb)) {
745                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
746         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
747                 uh->check = 0;
748                 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
749                 if (uh->check == 0)
750                         uh->check = CSUM_MANGLED_0;
751         } else {
752                 skb->ip_summed = CHECKSUM_PARTIAL;
753                 skb->csum_start = skb_transport_header(skb) - skb->head;
754                 skb->csum_offset = offsetof(struct udphdr, check);
755                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
756         }
757 }
758 EXPORT_SYMBOL(udp_set_csum);
759 
760 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
761 {
762         struct sock *sk = skb->sk;
763         struct inet_sock *inet = inet_sk(sk);
764         struct udphdr *uh;
765         int err = 0;
766         int is_udplite = IS_UDPLITE(sk);
767         int offset = skb_transport_offset(skb);
768         int len = skb->len - offset;
769         __wsum csum = 0;
770 
771         /*
772          * Create a UDP header
773          */
774         uh = udp_hdr(skb);
775         uh->source = inet->inet_sport;
776         uh->dest = fl4->fl4_dport;
777         uh->len = htons(len);
778         uh->check = 0;
779 
780         if (is_udplite)                                  /*     UDP-Lite      */
781                 csum = udplite_csum(skb);
782 
783         else if (sk->sk_no_check_tx) {                   /* UDP csum off */
784 
785                 skb->ip_summed = CHECKSUM_NONE;
786                 goto send;
787 
788         } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
789 
790                 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
791                 goto send;
792 
793         } else
794                 csum = udp_csum(skb);
795 
796         /* add protocol-dependent pseudo-header */
797         uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
798                                       sk->sk_protocol, csum);
799         if (uh->check == 0)
800                 uh->check = CSUM_MANGLED_0;
801 
802 send:
803         err = ip_send_skb(sock_net(sk), skb);
804         if (err) {
805                 if (err == -ENOBUFS && !inet->recverr) {
806                         UDP_INC_STATS(sock_net(sk),
807                                       UDP_MIB_SNDBUFERRORS, is_udplite);
808                         err = 0;
809                 }
810         } else
811                 UDP_INC_STATS(sock_net(sk),
812                               UDP_MIB_OUTDATAGRAMS, is_udplite);
813         return err;
814 }
815 
816 /*
817  * Push out all pending data as one UDP datagram. Socket is locked.
818  */
819 int udp_push_pending_frames(struct sock *sk)
820 {
821         struct udp_sock  *up = udp_sk(sk);
822         struct inet_sock *inet = inet_sk(sk);
823         struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
824         struct sk_buff *skb;
825         int err = 0;
826 
827         skb = ip_finish_skb(sk, fl4);
828         if (!skb)
829                 goto out;
830 
831         err = udp_send_skb(skb, fl4);
832 
833 out:
834         up->len = 0;
835         up->pending = 0;
836         return err;
837 }
838 EXPORT_SYMBOL(udp_push_pending_frames);
839 
840 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
841 {
842         struct inet_sock *inet = inet_sk(sk);
843         struct udp_sock *up = udp_sk(sk);
844         struct flowi4 fl4_stack;
845         struct flowi4 *fl4;
846         int ulen = len;
847         struct ipcm_cookie ipc;
848         struct rtable *rt = NULL;
849         int free = 0;
850         int connected = 0;
851         __be32 daddr, faddr, saddr;
852         __be16 dport;
853         u8  tos;
854         int err, is_udplite = IS_UDPLITE(sk);
855         int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
856         int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
857         struct sk_buff *skb;
858         struct ip_options_data opt_copy;
859 
860         if (len > 0xFFFF)
861                 return -EMSGSIZE;
862 
863         /*
864          *      Check the flags.
865          */
866 
867         if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
868                 return -EOPNOTSUPP;
869 
870         ipc.opt = NULL;
871         ipc.tx_flags = 0;
872         ipc.ttl = 0;
873         ipc.tos = -1;
874 
875         getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
876 
877         fl4 = &inet->cork.fl.u.ip4;
878         if (up->pending) {
879                 /*
880                  * There are pending frames.
881                  * The socket lock must be held while it's corked.
882                  */
883                 lock_sock(sk);
884                 if (likely(up->pending)) {
885                         if (unlikely(up->pending != AF_INET)) {
886                                 release_sock(sk);
887                                 return -EINVAL;
888                         }
889                         goto do_append_data;
890                 }
891                 release_sock(sk);
892         }
893         ulen += sizeof(struct udphdr);
894 
895         /*
896          *      Get and verify the address.
897          */
898         if (msg->msg_name) {
899                 DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
900                 if (msg->msg_namelen < sizeof(*usin))
901                         return -EINVAL;
902                 if (usin->sin_family != AF_INET) {
903                         if (usin->sin_family != AF_UNSPEC)
904                                 return -EAFNOSUPPORT;
905                 }
906 
907                 daddr = usin->sin_addr.s_addr;
908                 dport = usin->sin_port;
909                 if (dport == 0)
910                         return -EINVAL;
911         } else {
912                 if (sk->sk_state != TCP_ESTABLISHED)
913                         return -EDESTADDRREQ;
914                 daddr = inet->inet_daddr;
915                 dport = inet->inet_dport;
916                 /* Open fast path for connected socket.
917                    Route will not be used, if at least one option is set.
918                  */
919                 connected = 1;
920         }
921 
922         ipc.sockc.tsflags = sk->sk_tsflags;
923         ipc.addr = inet->inet_saddr;
924         ipc.oif = sk->sk_bound_dev_if;
925 
926         if (msg->msg_controllen) {
927                 err = ip_cmsg_send(sk, msg, &ipc, sk->sk_family == AF_INET6);
928                 if (unlikely(err)) {
929                         kfree(ipc.opt);
930                         return err;
931                 }
932                 if (ipc.opt)
933                         free = 1;
934                 connected = 0;
935         }
936         if (!ipc.opt) {
937                 struct ip_options_rcu *inet_opt;
938 
939                 rcu_read_lock();
940                 inet_opt = rcu_dereference(inet->inet_opt);
941                 if (inet_opt) {
942                         memcpy(&opt_copy, inet_opt,
943                                sizeof(*inet_opt) + inet_opt->opt.optlen);
944                         ipc.opt = &opt_copy.opt;
945                 }
946                 rcu_read_unlock();
947         }
948 
949         saddr = ipc.addr;
950         ipc.addr = faddr = daddr;
951 
952         sock_tx_timestamp(sk, ipc.sockc.tsflags, &ipc.tx_flags);
953 
954         if (ipc.opt && ipc.opt->opt.srr) {
955                 if (!daddr) {
956                         err = -EINVAL;
957                         goto out_free;
958                 }
959                 faddr = ipc.opt->opt.faddr;
960                 connected = 0;
961         }
962         tos = get_rttos(&ipc, inet);
963         if (sock_flag(sk, SOCK_LOCALROUTE) ||
964             (msg->msg_flags & MSG_DONTROUTE) ||
965             (ipc.opt && ipc.opt->opt.is_strictroute)) {
966                 tos |= RTO_ONLINK;
967                 connected = 0;
968         }
969 
970         if (ipv4_is_multicast(daddr)) {
971                 if (!ipc.oif)
972                         ipc.oif = inet->mc_index;
973                 if (!saddr)
974                         saddr = inet->mc_addr;
975                 connected = 0;
976         } else if (!ipc.oif) {
977                 ipc.oif = inet->uc_index;
978         } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
979                 /* oif is set, packet is to local broadcast and
980                  * and uc_index is set. oif is most likely set
981                  * by sk_bound_dev_if. If uc_index != oif check if the
982                  * oif is an L3 master and uc_index is an L3 slave.
983                  * If so, we want to allow the send using the uc_index.
984                  */
985                 if (ipc.oif != inet->uc_index &&
986                     ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
987                                                               inet->uc_index)) {
988                         ipc.oif = inet->uc_index;
989                 }
990         }
991 
992         if (connected)
993                 rt = (struct rtable *)sk_dst_check(sk, 0);
994 
995         if (!rt) {
996                 struct net *net = sock_net(sk);
997                 __u8 flow_flags = inet_sk_flowi_flags(sk);
998 
999                 fl4 = &fl4_stack;
1000 
1001                 flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
1002                                    RT_SCOPE_UNIVERSE, sk->sk_protocol,
1003                                    flow_flags,
1004                                    faddr, saddr, dport, inet->inet_sport,
1005                                    sk->sk_uid);
1006 
1007                 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1008                 rt = ip_route_output_flow(net, fl4, sk);
1009                 if (IS_ERR(rt)) {
1010                         err = PTR_ERR(rt);
1011                         rt = NULL;
1012                         if (err == -ENETUNREACH)
1013                                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1014                         goto out;
1015                 }
1016 
1017                 err = -EACCES;
1018                 if ((rt->rt_flags & RTCF_BROADCAST) &&
1019                     !sock_flag(sk, SOCK_BROADCAST))
1020                         goto out;
1021                 if (connected)
1022                         sk_dst_set(sk, dst_clone(&rt->dst));
1023         }
1024 
1025         if (msg->msg_flags&MSG_CONFIRM)
1026                 goto do_confirm;
1027 back_from_confirm:
1028 
1029         saddr = fl4->saddr;
1030         if (!ipc.addr)
1031                 daddr = ipc.addr = fl4->daddr;
1032 
1033         /* Lockless fast path for the non-corking case. */
1034         if (!corkreq) {
1035                 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1036                                   sizeof(struct udphdr), &ipc, &rt,
1037                                   msg->msg_flags);
1038                 err = PTR_ERR(skb);
1039                 if (!IS_ERR_OR_NULL(skb))
1040                         err = udp_send_skb(skb, fl4);
1041                 goto out;
1042         }
1043 
1044         lock_sock(sk);
1045         if (unlikely(up->pending)) {
1046                 /* The socket is already corked while preparing it. */
1047                 /* ... which is an evident application bug. --ANK */
1048                 release_sock(sk);
1049 
1050                 net_dbg_ratelimited("socket already corked\n");
1051                 err = -EINVAL;
1052                 goto out;
1053         }
1054         /*
1055          *      Now cork the socket to pend data.
1056          */
1057         fl4 = &inet->cork.fl.u.ip4;
1058         fl4->daddr = daddr;
1059         fl4->saddr = saddr;
1060         fl4->fl4_dport = dport;
1061         fl4->fl4_sport = inet->inet_sport;
1062         up->pending = AF_INET;
1063 
1064 do_append_data:
1065         up->len += ulen;
1066         err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1067                              sizeof(struct udphdr), &ipc, &rt,
1068                              corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1069         if (err)
1070                 udp_flush_pending_frames(sk);
1071         else if (!corkreq)
1072                 err = udp_push_pending_frames(sk);
1073         else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1074                 up->pending = 0;
1075         release_sock(sk);
1076 
1077 out:
1078         ip_rt_put(rt);
1079 out_free:
1080         if (free)
1081                 kfree(ipc.opt);
1082         if (!err)
1083                 return len;
1084         /*
1085          * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1086          * ENOBUFS might not be good (it's not tunable per se), but otherwise
1087          * we don't have a good statistic (IpOutDiscards but it can be too many
1088          * things).  We could add another new stat but at least for now that
1089          * seems like overkill.
1090          */
1091         if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1092                 UDP_INC_STATS(sock_net(sk),
1093                               UDP_MIB_SNDBUFERRORS, is_udplite);
1094         }
1095         return err;
1096 
1097 do_confirm:
1098         if (msg->msg_flags & MSG_PROBE)
1099                 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1100         if (!(msg->msg_flags&MSG_PROBE) || len)
1101                 goto back_from_confirm;
1102         err = 0;
1103         goto out;
1104 }
1105 EXPORT_SYMBOL(udp_sendmsg);
1106 
1107 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1108                  size_t size, int flags)
1109 {
1110         struct inet_sock *inet = inet_sk(sk);
1111         struct udp_sock *up = udp_sk(sk);
1112         int ret;
1113 
1114         if (flags & MSG_SENDPAGE_NOTLAST)
1115                 flags |= MSG_MORE;
1116 
1117         if (!up->pending) {
1118                 struct msghdr msg = {   .msg_flags = flags|MSG_MORE };
1119 
1120                 /* Call udp_sendmsg to specify destination address which
1121                  * sendpage interface can't pass.
1122                  * This will succeed only when the socket is connected.
1123                  */
1124                 ret = udp_sendmsg(sk, &msg, 0);
1125                 if (ret < 0)
1126                         return ret;
1127         }
1128 
1129         lock_sock(sk);
1130 
1131         if (unlikely(!up->pending)) {
1132                 release_sock(sk);
1133 
1134                 net_dbg_ratelimited("cork failed\n");
1135                 return -EINVAL;
1136         }
1137 
1138         ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1139                              page, offset, size, flags);
1140         if (ret == -EOPNOTSUPP) {
1141                 release_sock(sk);
1142                 return sock_no_sendpage(sk->sk_socket, page, offset,
1143                                         size, flags);
1144         }
1145         if (ret < 0) {
1146                 udp_flush_pending_frames(sk);
1147                 goto out;
1148         }
1149 
1150         up->len += size;
1151         if (!(up->corkflag || (flags&MSG_MORE)))
1152                 ret = udp_push_pending_frames(sk);
1153         if (!ret)
1154                 ret = size;
1155 out:
1156         release_sock(sk);
1157         return ret;
1158 }
1159 
1160 #define UDP_SKB_IS_STATELESS 0x80000000
1161 
1162 static void udp_set_dev_scratch(struct sk_buff *skb)
1163 {
1164         struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1165 
1166         BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1167         scratch->_tsize_state = skb->truesize;
1168 #if BITS_PER_LONG == 64
1169         scratch->len = skb->len;
1170         scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1171         scratch->is_linear = !skb_is_nonlinear(skb);
1172 #endif
1173         /* all head states execept sp (dst, sk, nf) are always cleared by
1174          * udp_rcv() and we need to preserve secpath, if present, to eventually
1175          * process IP_CMSG_PASSSEC at recvmsg() time
1176          */
1177         if (likely(!skb_sec_path(skb)))
1178                 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1179 }
1180 
1181 static int udp_skb_truesize(struct sk_buff *skb)
1182 {
1183         return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1184 }
1185 
1186 static bool udp_skb_has_head_state(struct sk_buff *skb)
1187 {
1188         return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1189 }
1190 
1191 /* fully reclaim rmem/fwd memory allocated for skb */
1192 static void udp_rmem_release(struct sock *sk, int size, int partial,
1193                              bool rx_queue_lock_held)
1194 {
1195         struct udp_sock *up = udp_sk(sk);
1196         struct sk_buff_head *sk_queue;
1197         int amt;
1198 
1199         if (likely(partial)) {
1200                 up->forward_deficit += size;
1201                 size = up->forward_deficit;
1202                 if (size < (sk->sk_rcvbuf >> 2))
1203                         return;
1204         } else {
1205                 size += up->forward_deficit;
1206         }
1207         up->forward_deficit = 0;
1208 
1209         /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1210          * if the called don't held it already
1211          */
1212         sk_queue = &sk->sk_receive_queue;
1213         if (!rx_queue_lock_held)
1214                 spin_lock(&sk_queue->lock);
1215 
1216 
1217         sk->sk_forward_alloc += size;
1218         amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1219         sk->sk_forward_alloc -= amt;
1220 
1221         if (amt)
1222                 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1223 
1224         atomic_sub(size, &sk->sk_rmem_alloc);
1225 
1226         /* this can save us from acquiring the rx queue lock on next receive */
1227         skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1228 
1229         if (!rx_queue_lock_held)
1230                 spin_unlock(&sk_queue->lock);
1231 }
1232 
1233 /* Note: called with reader_queue.lock held.
1234  * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1235  * This avoids a cache line miss while receive_queue lock is held.
1236  * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1237  */
1238 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1239 {
1240         prefetch(&skb->data);
1241         udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1242 }
1243 EXPORT_SYMBOL(udp_skb_destructor);
1244 
1245 /* as above, but the caller held the rx queue lock, too */
1246 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1247 {
1248         prefetch(&skb->data);
1249         udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1250 }
1251 
1252 /* Idea of busylocks is to let producers grab an extra spinlock
1253  * to relieve pressure on the receive_queue spinlock shared by consumer.
1254  * Under flood, this means that only one producer can be in line
1255  * trying to acquire the receive_queue spinlock.
1256  * These busylock can be allocated on a per cpu manner, instead of a
1257  * per socket one (that would consume a cache line per socket)
1258  */
1259 static int udp_busylocks_log __read_mostly;
1260 static spinlock_t *udp_busylocks __read_mostly;
1261 
1262 static spinlock_t *busylock_acquire(void *ptr)
1263 {
1264         spinlock_t *busy;
1265 
1266         busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1267         spin_lock(busy);
1268         return busy;
1269 }
1270 
1271 static void busylock_release(spinlock_t *busy)
1272 {
1273         if (busy)
1274                 spin_unlock(busy);
1275 }
1276 
1277 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1278 {
1279         struct sk_buff_head *list = &sk->sk_receive_queue;
1280         int rmem, delta, amt, err = -ENOMEM;
1281         spinlock_t *busy = NULL;
1282         int size;
1283 
1284         /* try to avoid the costly atomic add/sub pair when the receive
1285          * queue is full; always allow at least a packet
1286          */
1287         rmem = atomic_read(&sk->sk_rmem_alloc);
1288         if (rmem > sk->sk_rcvbuf)
1289                 goto drop;
1290 
1291         /* Under mem pressure, it might be helpful to help udp_recvmsg()
1292          * having linear skbs :
1293          * - Reduce memory overhead and thus increase receive queue capacity
1294          * - Less cache line misses at copyout() time
1295          * - Less work at consume_skb() (less alien page frag freeing)
1296          */
1297         if (rmem > (sk->sk_rcvbuf >> 1)) {
1298                 skb_condense(skb);
1299 
1300                 busy = busylock_acquire(sk);
1301         }
1302         size = skb->truesize;
1303         udp_set_dev_scratch(skb);
1304 
1305         /* we drop only if the receive buf is full and the receive
1306          * queue contains some other skb
1307          */
1308         rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1309         if (rmem > (size + sk->sk_rcvbuf))
1310                 goto uncharge_drop;
1311 
1312         spin_lock(&list->lock);
1313         if (size >= sk->sk_forward_alloc) {
1314                 amt = sk_mem_pages(size);
1315                 delta = amt << SK_MEM_QUANTUM_SHIFT;
1316                 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1317                         err = -ENOBUFS;
1318                         spin_unlock(&list->lock);
1319                         goto uncharge_drop;
1320                 }
1321 
1322                 sk->sk_forward_alloc += delta;
1323         }
1324 
1325         sk->sk_forward_alloc -= size;
1326 
1327         /* no need to setup a destructor, we will explicitly release the
1328          * forward allocated memory on dequeue
1329          */
1330         sock_skb_set_dropcount(sk, skb);
1331 
1332         __skb_queue_tail(list, skb);
1333         spin_unlock(&list->lock);
1334 
1335         if (!sock_flag(sk, SOCK_DEAD))
1336                 sk->sk_data_ready(sk);
1337 
1338         busylock_release(busy);
1339         return 0;
1340 
1341 uncharge_drop:
1342         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1343 
1344 drop:
1345         atomic_inc(&sk->sk_drops);
1346         busylock_release(busy);
1347         return err;
1348 }
1349 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1350 
1351 void udp_destruct_sock(struct sock *sk)
1352 {
1353         /* reclaim completely the forward allocated memory */
1354         struct udp_sock *up = udp_sk(sk);
1355         unsigned int total = 0;
1356         struct sk_buff *skb;
1357 
1358         skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1359         while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1360                 total += skb->truesize;
1361                 kfree_skb(skb);
1362         }
1363         udp_rmem_release(sk, total, 0, true);
1364 
1365         inet_sock_destruct(sk);
1366 }
1367 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1368 
1369 int udp_init_sock(struct sock *sk)
1370 {
1371         skb_queue_head_init(&udp_sk(sk)->reader_queue);
1372         sk->sk_destruct = udp_destruct_sock;
1373         return 0;
1374 }
1375 EXPORT_SYMBOL_GPL(udp_init_sock);
1376 
1377 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1378 {
1379         if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1380                 bool slow = lock_sock_fast(sk);
1381 
1382                 sk_peek_offset_bwd(sk, len);
1383                 unlock_sock_fast(sk, slow);
1384         }
1385 
1386         if (!skb_unref(skb))
1387                 return;
1388 
1389         /* In the more common cases we cleared the head states previously,
1390          * see __udp_queue_rcv_skb().
1391          */
1392         if (unlikely(udp_skb_has_head_state(skb)))
1393                 skb_release_head_state(skb);
1394         __consume_stateless_skb(skb);
1395 }
1396 EXPORT_SYMBOL_GPL(skb_consume_udp);
1397 
1398 static struct sk_buff *__first_packet_length(struct sock *sk,
1399                                              struct sk_buff_head *rcvq,
1400                                              int *total)
1401 {
1402         struct sk_buff *skb;
1403 
1404         while ((skb = skb_peek(rcvq)) != NULL) {
1405                 if (udp_lib_checksum_complete(skb)) {
1406                         __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1407                                         IS_UDPLITE(sk));
1408                         __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1409                                         IS_UDPLITE(sk));
1410                         atomic_inc(&sk->sk_drops);
1411                         __skb_unlink(skb, rcvq);
1412                         *total += skb->truesize;
1413                         kfree_skb(skb);
1414                 } else {
1415                         /* the csum related bits could be changed, refresh
1416                          * the scratch area
1417                          */
1418                         udp_set_dev_scratch(skb);
1419                         break;
1420                 }
1421         }
1422         return skb;
1423 }
1424 
1425 /**
1426  *      first_packet_length     - return length of first packet in receive queue
1427  *      @sk: socket
1428  *
1429  *      Drops all bad checksum frames, until a valid one is found.
1430  *      Returns the length of found skb, or -1 if none is found.
1431  */
1432 static int first_packet_length(struct sock *sk)
1433 {
1434         struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1435         struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1436         struct sk_buff *skb;
1437         int total = 0;
1438         int res;
1439 
1440         spin_lock_bh(&rcvq->lock);
1441         skb = __first_packet_length(sk, rcvq, &total);
1442         if (!skb && !skb_queue_empty(sk_queue)) {
1443                 spin_lock(&sk_queue->lock);
1444                 skb_queue_splice_tail_init(sk_queue, rcvq);
1445                 spin_unlock(&sk_queue->lock);
1446 
1447                 skb = __first_packet_length(sk, rcvq, &total);
1448         }
1449         res = skb ? skb->len : -1;
1450         if (total)
1451                 udp_rmem_release(sk, total, 1, false);
1452         spin_unlock_bh(&rcvq->lock);
1453         return res;
1454 }
1455 
1456 /*
1457  *      IOCTL requests applicable to the UDP protocol
1458  */
1459 
1460 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1461 {
1462         switch (cmd) {
1463         case SIOCOUTQ:
1464         {
1465                 int amount = sk_wmem_alloc_get(sk);
1466 
1467                 return put_user(amount, (int __user *)arg);
1468         }
1469 
1470         case SIOCINQ:
1471         {
1472                 int amount = max_t(int, 0, first_packet_length(sk));
1473 
1474                 return put_user(amount, (int __user *)arg);
1475         }
1476 
1477         default:
1478                 return -ENOIOCTLCMD;
1479         }
1480 
1481         return 0;
1482 }
1483 EXPORT_SYMBOL(udp_ioctl);
1484 
1485 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1486                                int noblock, int *peeked, int *off, int *err)
1487 {
1488         struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1489         struct sk_buff_head *queue;
1490         struct sk_buff *last;
1491         long timeo;
1492         int error;
1493 
1494         queue = &udp_sk(sk)->reader_queue;
1495         flags |= noblock ? MSG_DONTWAIT : 0;
1496         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1497         do {
1498                 struct sk_buff *skb;
1499 
1500                 error = sock_error(sk);
1501                 if (error)
1502                         break;
1503 
1504                 error = -EAGAIN;
1505                 *peeked = 0;
1506                 do {
1507                         spin_lock_bh(&queue->lock);
1508                         skb = __skb_try_recv_from_queue(sk, queue, flags,
1509                                                         udp_skb_destructor,
1510                                                         peeked, off, err,
1511                                                         &last);
1512                         if (skb) {
1513                                 spin_unlock_bh(&queue->lock);
1514                                 return skb;
1515                         }
1516 
1517                         if (skb_queue_empty(sk_queue)) {
1518                                 spin_unlock_bh(&queue->lock);
1519                                 goto busy_check;
1520                         }
1521 
1522                         /* refill the reader queue and walk it again
1523                          * keep both queues locked to avoid re-acquiring
1524                          * the sk_receive_queue lock if fwd memory scheduling
1525                          * is needed.
1526                          */
1527                         spin_lock(&sk_queue->lock);
1528                         skb_queue_splice_tail_init(sk_queue, queue);
1529 
1530                         skb = __skb_try_recv_from_queue(sk, queue, flags,
1531                                                         udp_skb_dtor_locked,
1532                                                         peeked, off, err,
1533                                                         &last);
1534                         spin_unlock(&sk_queue->lock);
1535                         spin_unlock_bh(&queue->lock);
1536                         if (skb)
1537                                 return skb;
1538 
1539 busy_check:
1540                         if (!sk_can_busy_loop(sk))
1541                                 break;
1542 
1543                         sk_busy_loop(sk, flags & MSG_DONTWAIT);
1544                 } while (!skb_queue_empty(sk_queue));
1545 
1546                 /* sk_queue is empty, reader_queue may contain peeked packets */
1547         } while (timeo &&
1548                  !__skb_wait_for_more_packets(sk, &error, &timeo,
1549                                               (struct sk_buff *)sk_queue));
1550 
1551         *err = error;
1552         return NULL;
1553 }
1554 EXPORT_SYMBOL_GPL(__skb_recv_udp);
1555 
1556 /*
1557  *      This should be easy, if there is something there we
1558  *      return it, otherwise we block.
1559  */
1560 
1561 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1562                 int flags, int *addr_len)
1563 {
1564         struct inet_sock *inet = inet_sk(sk);
1565         DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1566         struct sk_buff *skb;
1567         unsigned int ulen, copied;
1568         int peeked, peeking, off;
1569         int err;
1570         int is_udplite = IS_UDPLITE(sk);
1571         bool checksum_valid = false;
1572 
1573         if (flags & MSG_ERRQUEUE)
1574                 return ip_recv_error(sk, msg, len, addr_len);
1575 
1576 try_again:
1577         peeking = flags & MSG_PEEK;
1578         off = sk_peek_offset(sk, flags);
1579         skb = __skb_recv_udp(sk, flags, noblock, &peeked, &off, &err);
1580         if (!skb)
1581                 return err;
1582         if (ccs_socket_post_recvmsg_permission(sk, skb, flags))
1583                 return -EAGAIN; /* Hope less harmful than -EPERM. */
1584 
1585         ulen = udp_skb_len(skb);
1586         copied = len;
1587         if (copied > ulen - off)
1588                 copied = ulen - off;
1589         else if (copied < ulen)
1590                 msg->msg_flags |= MSG_TRUNC;
1591 
1592         /*
1593          * If checksum is needed at all, try to do it while copying the
1594          * data.  If the data is truncated, or if we only want a partial
1595          * coverage checksum (UDP-Lite), do it before the copy.
1596          */
1597 
1598         if (copied < ulen || peeking ||
1599             (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1600                 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1601                                 !__udp_lib_checksum_complete(skb);
1602                 if (!checksum_valid)
1603                         goto csum_copy_err;
1604         }
1605 
1606         if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1607                 if (udp_skb_is_linear(skb))
1608                         err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1609                 else
1610                         err = skb_copy_datagram_msg(skb, off, msg, copied);
1611         } else {
1612                 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1613 
1614                 if (err == -EINVAL)
1615                         goto csum_copy_err;
1616         }
1617 
1618         if (unlikely(err)) {
1619                 if (!peeked) {
1620                         atomic_inc(&sk->sk_drops);
1621                         UDP_INC_STATS(sock_net(sk),
1622                                       UDP_MIB_INERRORS, is_udplite);
1623                 }
1624                 kfree_skb(skb);
1625                 return err;
1626         }
1627 
1628         if (!peeked)
1629                 UDP_INC_STATS(sock_net(sk),
1630                               UDP_MIB_INDATAGRAMS, is_udplite);
1631 
1632         sock_recv_ts_and_drops(msg, sk, skb);
1633 
1634         /* Copy the address. */
1635         if (sin) {
1636                 sin->sin_family = AF_INET;
1637                 sin->sin_port = udp_hdr(skb)->source;
1638                 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1639                 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1640                 *addr_len = sizeof(*sin);
1641         }
1642         if (inet->cmsg_flags)
1643                 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1644 
1645         err = copied;
1646         if (flags & MSG_TRUNC)
1647                 err = ulen;
1648 
1649         skb_consume_udp(sk, skb, peeking ? -err : err);
1650         return err;
1651 
1652 csum_copy_err:
1653         if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1654                                  udp_skb_destructor)) {
1655                 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1656                 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1657         }
1658         kfree_skb(skb);
1659 
1660         /* starting over for a new packet, but check if we need to yield */
1661         cond_resched();
1662         msg->msg_flags &= ~MSG_TRUNC;
1663         goto try_again;
1664 }
1665 
1666 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1667 {
1668         /* This check is replicated from __ip4_datagram_connect() and
1669          * intended to prevent BPF program called below from accessing bytes
1670          * that are out of the bound specified by user in addr_len.
1671          */
1672         if (addr_len < sizeof(struct sockaddr_in))
1673                 return -EINVAL;
1674 
1675         return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1676 }
1677 EXPORT_SYMBOL(udp_pre_connect);
1678 
1679 int __udp_disconnect(struct sock *sk, int flags)
1680 {
1681         struct inet_sock *inet = inet_sk(sk);
1682         /*
1683          *      1003.1g - break association.
1684          */
1685 
1686         sk->sk_state = TCP_CLOSE;
1687         inet->inet_daddr = 0;
1688         inet->inet_dport = 0;
1689         sock_rps_reset_rxhash(sk);
1690         sk->sk_bound_dev_if = 0;
1691         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1692                 inet_reset_saddr(sk);
1693 
1694         if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1695                 sk->sk_prot->unhash(sk);
1696                 inet->inet_sport = 0;
1697         }
1698         sk_dst_reset(sk);
1699         return 0;
1700 }
1701 EXPORT_SYMBOL(__udp_disconnect);
1702 
1703 int udp_disconnect(struct sock *sk, int flags)
1704 {
1705         lock_sock(sk);
1706         __udp_disconnect(sk, flags);
1707         release_sock(sk);
1708         return 0;
1709 }
1710 EXPORT_SYMBOL(udp_disconnect);
1711 
1712 void udp_lib_unhash(struct sock *sk)
1713 {
1714         if (sk_hashed(sk)) {
1715                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1716                 struct udp_hslot *hslot, *hslot2;
1717 
1718                 hslot  = udp_hashslot(udptable, sock_net(sk),
1719                                       udp_sk(sk)->udp_port_hash);
1720                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1721 
1722                 spin_lock_bh(&hslot->lock);
1723                 if (rcu_access_pointer(sk->sk_reuseport_cb))
1724                         reuseport_detach_sock(sk);
1725                 if (sk_del_node_init_rcu(sk)) {
1726                         hslot->count--;
1727                         inet_sk(sk)->inet_num = 0;
1728                         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1729 
1730                         spin_lock(&hslot2->lock);
1731                         hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1732                         hslot2->count--;
1733                         spin_unlock(&hslot2->lock);
1734                 }
1735                 spin_unlock_bh(&hslot->lock);
1736         }
1737 }
1738 EXPORT_SYMBOL(udp_lib_unhash);
1739 
1740 /*
1741  * inet_rcv_saddr was changed, we must rehash secondary hash
1742  */
1743 void udp_lib_rehash(struct sock *sk, u16 newhash)
1744 {
1745         if (sk_hashed(sk)) {
1746                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1747                 struct udp_hslot *hslot, *hslot2, *nhslot2;
1748 
1749                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1750                 nhslot2 = udp_hashslot2(udptable, newhash);
1751                 udp_sk(sk)->udp_portaddr_hash = newhash;
1752 
1753                 if (hslot2 != nhslot2 ||
1754                     rcu_access_pointer(sk->sk_reuseport_cb)) {
1755                         hslot = udp_hashslot(udptable, sock_net(sk),
1756                                              udp_sk(sk)->udp_port_hash);
1757                         /* we must lock primary chain too */
1758                         spin_lock_bh(&hslot->lock);
1759                         if (rcu_access_pointer(sk->sk_reuseport_cb))
1760                                 reuseport_detach_sock(sk);
1761 
1762                         if (hslot2 != nhslot2) {
1763                                 spin_lock(&hslot2->lock);
1764                                 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1765                                 hslot2->count--;
1766                                 spin_unlock(&hslot2->lock);
1767 
1768                                 spin_lock(&nhslot2->lock);
1769                                 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1770                                                          &nhslot2->head);
1771                                 nhslot2->count++;
1772                                 spin_unlock(&nhslot2->lock);
1773                         }
1774 
1775                         spin_unlock_bh(&hslot->lock);
1776                 }
1777         }
1778 }
1779 EXPORT_SYMBOL(udp_lib_rehash);
1780 
1781 static void udp_v4_rehash(struct sock *sk)
1782 {
1783         u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1784                                           inet_sk(sk)->inet_rcv_saddr,
1785                                           inet_sk(sk)->inet_num);
1786         udp_lib_rehash(sk, new_hash);
1787 }
1788 
1789 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1790 {
1791         int rc;
1792 
1793         if (inet_sk(sk)->inet_daddr) {
1794                 sock_rps_save_rxhash(sk, skb);
1795                 sk_mark_napi_id(sk, skb);
1796                 sk_incoming_cpu_update(sk);
1797         } else {
1798                 sk_mark_napi_id_once(sk, skb);
1799         }
1800 
1801         rc = __udp_enqueue_schedule_skb(sk, skb);
1802         if (rc < 0) {
1803                 int is_udplite = IS_UDPLITE(sk);
1804 
1805                 /* Note that an ENOMEM error is charged twice */
1806                 if (rc == -ENOMEM)
1807                         UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1808                                         is_udplite);
1809                 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1810                 kfree_skb(skb);
1811                 trace_udp_fail_queue_rcv_skb(rc, sk);
1812                 return -1;
1813         }
1814 
1815         return 0;
1816 }
1817 
1818 static struct static_key udp_encap_needed __read_mostly;
1819 void udp_encap_enable(void)
1820 {
1821         static_key_enable(&udp_encap_needed);
1822 }
1823 EXPORT_SYMBOL(udp_encap_enable);
1824 
1825 /* returns:
1826  *  -1: error
1827  *   0: success
1828  *  >0: "udp encap" protocol resubmission
1829  *
1830  * Note that in the success and error cases, the skb is assumed to
1831  * have either been requeued or freed.
1832  */
1833 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1834 {
1835         struct udp_sock *up = udp_sk(sk);
1836         int is_udplite = IS_UDPLITE(sk);
1837 
1838         /*
1839          *      Charge it to the socket, dropping if the queue is full.
1840          */
1841         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1842                 goto drop;
1843         nf_reset(skb);
1844 
1845         if (static_key_false(&udp_encap_needed) && up->encap_type) {
1846                 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1847 
1848                 /*
1849                  * This is an encapsulation socket so pass the skb to
1850                  * the socket's udp_encap_rcv() hook. Otherwise, just
1851                  * fall through and pass this up the UDP socket.
1852                  * up->encap_rcv() returns the following value:
1853                  * =0 if skb was successfully passed to the encap
1854                  *    handler or was discarded by it.
1855                  * >0 if skb should be passed on to UDP.
1856                  * <0 if skb should be resubmitted as proto -N
1857                  */
1858 
1859                 /* if we're overly short, let UDP handle it */
1860                 encap_rcv = READ_ONCE(up->encap_rcv);
1861                 if (encap_rcv) {
1862                         int ret;
1863 
1864                         /* Verify checksum before giving to encap */
1865                         if (udp_lib_checksum_complete(skb))
1866                                 goto csum_error;
1867 
1868                         ret = encap_rcv(sk, skb);
1869                         if (ret <= 0) {
1870                                 __UDP_INC_STATS(sock_net(sk),
1871                                                 UDP_MIB_INDATAGRAMS,
1872                                                 is_udplite);
1873                                 return -ret;
1874                         }
1875                 }
1876 
1877                 /* FALLTHROUGH -- it's a UDP Packet */
1878         }
1879 
1880         /*
1881          *      UDP-Lite specific tests, ignored on UDP sockets
1882          */
1883         if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1884 
1885                 /*
1886                  * MIB statistics other than incrementing the error count are
1887                  * disabled for the following two types of errors: these depend
1888                  * on the application settings, not on the functioning of the
1889                  * protocol stack as such.
1890                  *
1891                  * RFC 3828 here recommends (sec 3.3): "There should also be a
1892                  * way ... to ... at least let the receiving application block
1893                  * delivery of packets with coverage values less than a value
1894                  * provided by the application."
1895                  */
1896                 if (up->pcrlen == 0) {          /* full coverage was set  */
1897                         net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
1898                                             UDP_SKB_CB(skb)->cscov, skb->len);
1899                         goto drop;
1900                 }
1901                 /* The next case involves violating the min. coverage requested
1902                  * by the receiver. This is subtle: if receiver wants x and x is
1903                  * greater than the buffersize/MTU then receiver will complain
1904                  * that it wants x while sender emits packets of smaller size y.
1905                  * Therefore the above ...()->partial_cov statement is essential.
1906                  */
1907                 if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1908                         net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
1909                                             UDP_SKB_CB(skb)->cscov, up->pcrlen);
1910                         goto drop;
1911                 }
1912         }
1913 
1914         prefetch(&sk->sk_rmem_alloc);
1915         if (rcu_access_pointer(sk->sk_filter) &&
1916             udp_lib_checksum_complete(skb))
1917                         goto csum_error;
1918 
1919         if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
1920                 goto drop;
1921 
1922         udp_csum_pull_header(skb);
1923 
1924         ipv4_pktinfo_prepare(sk, skb);
1925         return __udp_queue_rcv_skb(sk, skb);
1926 
1927 csum_error:
1928         __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1929 drop:
1930         __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1931         atomic_inc(&sk->sk_drops);
1932         kfree_skb(skb);
1933         return -1;
1934 }
1935 
1936 /* For TCP sockets, sk_rx_dst is protected by socket lock
1937  * For UDP, we use xchg() to guard against concurrent changes.
1938  */
1939 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
1940 {
1941         struct dst_entry *old;
1942 
1943         if (dst_hold_safe(dst)) {
1944                 old = xchg(&sk->sk_rx_dst, dst);
1945                 dst_release(old);
1946                 return old != dst;
1947         }
1948         return false;
1949 }
1950 EXPORT_SYMBOL(udp_sk_rx_dst_set);
1951 
1952 /*
1953  *      Multicasts and broadcasts go to each listener.
1954  *
1955  *      Note: called only from the BH handler context.
1956  */
1957 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1958                                     struct udphdr  *uh,
1959                                     __be32 saddr, __be32 daddr,
1960                                     struct udp_table *udptable,
1961                                     int proto)
1962 {
1963         struct sock *sk, *first = NULL;
1964         unsigned short hnum = ntohs(uh->dest);
1965         struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
1966         unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
1967         unsigned int offset = offsetof(typeof(*sk), sk_node);
1968         int dif = skb->dev->ifindex;
1969         int sdif = inet_sdif(skb);
1970         struct hlist_node *node;
1971         struct sk_buff *nskb;
1972 
1973         if (use_hash2) {
1974                 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
1975                             udptable->mask;
1976                 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
1977 start_lookup:
1978                 hslot = &udptable->hash2[hash2];
1979                 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
1980         }
1981 
1982         sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
1983                 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
1984                                          uh->source, saddr, dif, sdif, hnum))
1985                         continue;
1986 
1987                 if (!first) {
1988                         first = sk;
1989                         continue;
1990                 }
1991                 nskb = skb_clone(skb, GFP_ATOMIC);
1992 
1993                 if (unlikely(!nskb)) {
1994                         atomic_inc(&sk->sk_drops);
1995                         __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
1996                                         IS_UDPLITE(sk));
1997                         __UDP_INC_STATS(net, UDP_MIB_INERRORS,
1998                                         IS_UDPLITE(sk));
1999                         continue;
2000                 }
2001                 if (udp_queue_rcv_skb(sk, nskb) > 0)
2002                         consume_skb(nskb);
2003         }
2004 
2005         /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2006         if (use_hash2 && hash2 != hash2_any) {
2007                 hash2 = hash2_any;
2008                 goto start_lookup;
2009         }
2010 
2011         if (first) {
2012                 if (udp_queue_rcv_skb(first, skb) > 0)
2013                         consume_skb(skb);
2014         } else {
2015                 kfree_skb(skb);
2016                 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2017                                 proto == IPPROTO_UDPLITE);
2018         }
2019         return 0;
2020 }
2021 
2022 /* Initialize UDP checksum. If exited with zero value (success),
2023  * CHECKSUM_UNNECESSARY means, that no more checks are required.
2024  * Otherwise, csum completion requires chacksumming packet body,
2025  * including udp header and folding it to skb->csum.
2026  */
2027 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2028                                  int proto)
2029 {
2030         int err;
2031 
2032         UDP_SKB_CB(skb)->partial_cov = 0;
2033         UDP_SKB_CB(skb)->cscov = skb->len;
2034 
2035         if (proto == IPPROTO_UDPLITE) {
2036                 err = udplite_checksum_init(skb, uh);
2037                 if (err)
2038                         return err;
2039 
2040                 if (UDP_SKB_CB(skb)->partial_cov) {
2041                         skb->csum = inet_compute_pseudo(skb, proto);
2042                         return 0;
2043                 }
2044         }
2045 
2046         /* Note, we are only interested in != 0 or == 0, thus the
2047          * force to int.
2048          */
2049         return (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2050                                                          inet_compute_pseudo);
2051 }
2052 
2053 /*
2054  *      All we need to do is get the socket, and then do a checksum.
2055  */
2056 
2057 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2058                    int proto)
2059 {
2060         struct sock *sk;
2061         struct udphdr *uh;
2062         unsigned short ulen;
2063         struct rtable *rt = skb_rtable(skb);
2064         __be32 saddr, daddr;
2065         struct net *net = dev_net(skb->dev);
2066 
2067         /*
2068          *  Validate the packet.
2069          */
2070         if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2071                 goto drop;              /* No space for header. */
2072 
2073         uh   = udp_hdr(skb);
2074         ulen = ntohs(uh->len);
2075         saddr = ip_hdr(skb)->saddr;
2076         daddr = ip_hdr(skb)->daddr;
2077 
2078         if (ulen > skb->len)
2079                 goto short_packet;
2080 
2081         if (proto == IPPROTO_UDP) {
2082                 /* UDP validates ulen. */
2083                 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2084                         goto short_packet;
2085                 uh = udp_hdr(skb);
2086         }
2087 
2088         if (udp4_csum_init(skb, uh, proto))
2089                 goto csum_error;
2090 
2091         sk = skb_steal_sock(skb);
2092         if (sk) {
2093                 struct dst_entry *dst = skb_dst(skb);
2094                 int ret;
2095 
2096                 if (unlikely(sk->sk_rx_dst != dst))
2097                         udp_sk_rx_dst_set(sk, dst);
2098 
2099                 ret = udp_queue_rcv_skb(sk, skb);
2100                 sock_put(sk);
2101                 /* a return value > 0 means to resubmit the input, but
2102                  * it wants the return to be -protocol, or 0
2103                  */
2104                 if (ret > 0)
2105                         return -ret;
2106                 return 0;
2107         }
2108 
2109         if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2110                 return __udp4_lib_mcast_deliver(net, skb, uh,
2111                                                 saddr, daddr, udptable, proto);
2112 
2113         sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2114         if (sk) {
2115                 int ret;
2116 
2117                 if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2118                         skb_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
2119                                                  inet_compute_pseudo);
2120 
2121                 ret = udp_queue_rcv_skb(sk, skb);
2122 
2123                 /* a return value > 0 means to resubmit the input, but
2124                  * it wants the return to be -protocol, or 0
2125                  */
2126                 if (ret > 0)
2127                         return -ret;
2128                 return 0;
2129         }
2130 
2131         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2132                 goto drop;
2133         nf_reset(skb);
2134 
2135         /* No socket. Drop packet silently, if checksum is wrong */
2136         if (udp_lib_checksum_complete(skb))
2137                 goto csum_error;
2138 
2139         __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2140         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2141 
2142         /*
2143          * Hmm.  We got an UDP packet to a port to which we
2144          * don't wanna listen.  Ignore it.
2145          */
2146         kfree_skb(skb);
2147         return 0;
2148 
2149 short_packet:
2150         net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2151                             proto == IPPROTO_UDPLITE ? "Lite" : "",
2152                             &saddr, ntohs(uh->source),
2153                             ulen, skb->len,
2154                             &daddr, ntohs(uh->dest));
2155         goto drop;
2156 
2157 csum_error:
2158         /*
2159          * RFC1122: OK.  Discards the bad packet silently (as far as
2160          * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2161          */
2162         net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2163                             proto == IPPROTO_UDPLITE ? "Lite" : "",
2164                             &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2165                             ulen);
2166         __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2167 drop:
2168         __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2169         kfree_skb(skb);
2170         return 0;
2171 }
2172 
2173 /* We can only early demux multicast if there is a single matching socket.
2174  * If more than one socket found returns NULL
2175  */
2176 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2177                                                   __be16 loc_port, __be32 loc_addr,
2178                                                   __be16 rmt_port, __be32 rmt_addr,
2179                                                   int dif, int sdif)
2180 {
2181         struct sock *sk, *result;
2182         unsigned short hnum = ntohs(loc_port);
2183         unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2184         struct udp_hslot *hslot = &udp_table.hash[slot];
2185 
2186         /* Do not bother scanning a too big list */
2187         if (hslot->count > 10)
2188                 return NULL;
2189 
2190         result = NULL;
2191         sk_for_each_rcu(sk, &hslot->head) {
2192                 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2193                                         rmt_port, rmt_addr, dif, sdif, hnum)) {
2194                         if (result)
2195                                 return NULL;
2196                         result = sk;
2197                 }
2198         }
2199 
2200         return result;
2201 }
2202 
2203 /* For unicast we should only early demux connected sockets or we can
2204  * break forwarding setups.  The chains here can be long so only check
2205  * if the first socket is an exact match and if not move on.
2206  */
2207 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2208                                             __be16 loc_port, __be32 loc_addr,
2209                                             __be16 rmt_port, __be32 rmt_addr,
2210                                             int dif, int sdif)
2211 {
2212         unsigned short hnum = ntohs(loc_port);
2213         unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2214         unsigned int slot2 = hash2 & udp_table.mask;
2215         struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2216         INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2217         const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2218         struct sock *sk;
2219 
2220         udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2221                 if (INET_MATCH(sk, net, acookie, rmt_addr,
2222                                loc_addr, ports, dif, sdif))
2223                         return sk;
2224                 /* Only check first socket in chain */
2225                 break;
2226         }
2227         return NULL;
2228 }
2229 
2230 int udp_v4_early_demux(struct sk_buff *skb)
2231 {
2232         struct net *net = dev_net(skb->dev);
2233         struct in_device *in_dev = NULL;
2234         const struct iphdr *iph;
2235         const struct udphdr *uh;
2236         struct sock *sk = NULL;
2237         struct dst_entry *dst;
2238         int dif = skb->dev->ifindex;
2239         int sdif = inet_sdif(skb);
2240         int ours;
2241 
2242         /* validate the packet */
2243         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2244                 return 0;
2245 
2246         iph = ip_hdr(skb);
2247         uh = udp_hdr(skb);
2248 
2249         if (skb->pkt_type == PACKET_MULTICAST) {
2250                 in_dev = __in_dev_get_rcu(skb->dev);
2251 
2252                 if (!in_dev)
2253                         return 0;
2254 
2255                 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2256                                        iph->protocol);
2257                 if (!ours)
2258                         return 0;
2259 
2260                 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2261                                                    uh->source, iph->saddr,
2262                                                    dif, sdif);
2263         } else if (skb->pkt_type == PACKET_HOST) {
2264                 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2265                                              uh->source, iph->saddr, dif, sdif);
2266         }
2267 
2268         if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2269                 return 0;
2270 
2271         skb->sk = sk;
2272         skb->destructor = sock_efree;
2273         dst = READ_ONCE(sk->sk_rx_dst);
2274 
2275         if (dst)
2276                 dst = dst_check(dst, 0);
2277         if (dst) {
2278                 u32 itag = 0;
2279 
2280                 /* set noref for now.
2281                  * any place which wants to hold dst has to call
2282                  * dst_hold_safe()
2283                  */
2284                 skb_dst_set_noref(skb, dst);
2285 
2286                 /* for unconnected multicast sockets we need to validate
2287                  * the source on each packet
2288                  */
2289                 if (!inet_sk(sk)->inet_daddr && in_dev)
2290                         return ip_mc_validate_source(skb, iph->daddr,
2291                                                      iph->saddr, iph->tos,
2292                                                      skb->dev, in_dev, &itag);
2293         }
2294         return 0;
2295 }
2296 
2297 int udp_rcv(struct sk_buff *skb)
2298 {
2299         return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2300 }
2301 
2302 void udp_destroy_sock(struct sock *sk)
2303 {
2304         struct udp_sock *up = udp_sk(sk);
2305         bool slow = lock_sock_fast(sk);
2306         udp_flush_pending_frames(sk);
2307         unlock_sock_fast(sk, slow);
2308         if (static_key_false(&udp_encap_needed) && up->encap_type) {
2309                 void (*encap_destroy)(struct sock *sk);
2310                 encap_destroy = READ_ONCE(up->encap_destroy);
2311                 if (encap_destroy)
2312                         encap_destroy(sk);
2313         }
2314 }
2315 
2316 /*
2317  *      Socket option code for UDP
2318  */
2319 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2320                        char __user *optval, unsigned int optlen,
2321                        int (*push_pending_frames)(struct sock *))
2322 {
2323         struct udp_sock *up = udp_sk(sk);
2324         int val, valbool;
2325         int err = 0;
2326         int is_udplite = IS_UDPLITE(sk);
2327 
2328         if (optlen < sizeof(int))
2329                 return -EINVAL;
2330 
2331         if (get_user(val, (int __user *)optval))
2332                 return -EFAULT;
2333 
2334         valbool = val ? 1 : 0;
2335 
2336         switch (optname) {
2337         case UDP_CORK:
2338                 if (val != 0) {
2339                         up->corkflag = 1;
2340                 } else {
2341                         up->corkflag = 0;
2342                         lock_sock(sk);
2343                         push_pending_frames(sk);
2344                         release_sock(sk);
2345                 }
2346                 break;
2347 
2348         case UDP_ENCAP:
2349                 switch (val) {
2350                 case 0:
2351                 case UDP_ENCAP_ESPINUDP:
2352                 case UDP_ENCAP_ESPINUDP_NON_IKE:
2353                         up->encap_rcv = xfrm4_udp_encap_rcv;
2354                         /* FALLTHROUGH */
2355                 case UDP_ENCAP_L2TPINUDP:
2356                         up->encap_type = val;
2357                         udp_encap_enable();
2358                         break;
2359                 default:
2360                         err = -ENOPROTOOPT;
2361                         break;
2362                 }
2363                 break;
2364 
2365         case UDP_NO_CHECK6_TX:
2366                 up->no_check6_tx = valbool;
2367                 break;
2368 
2369         case UDP_NO_CHECK6_RX:
2370                 up->no_check6_rx = valbool;
2371                 break;
2372 
2373         /*
2374          *      UDP-Lite's partial checksum coverage (RFC 3828).
2375          */
2376         /* The sender sets actual checksum coverage length via this option.
2377          * The case coverage > packet length is handled by send module. */
2378         case UDPLITE_SEND_CSCOV:
2379                 if (!is_udplite)         /* Disable the option on UDP sockets */
2380                         return -ENOPROTOOPT;
2381                 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2382                         val = 8;
2383                 else if (val > USHRT_MAX)
2384                         val = USHRT_MAX;
2385                 up->pcslen = val;
2386                 up->pcflag |= UDPLITE_SEND_CC;
2387                 break;
2388 
2389         /* The receiver specifies a minimum checksum coverage value. To make
2390          * sense, this should be set to at least 8 (as done below). If zero is
2391          * used, this again means full checksum coverage.                     */
2392         case UDPLITE_RECV_CSCOV:
2393                 if (!is_udplite)         /* Disable the option on UDP sockets */
2394                         return -ENOPROTOOPT;
2395                 if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2396                         val = 8;
2397                 else if (val > USHRT_MAX)
2398                         val = USHRT_MAX;
2399                 up->pcrlen = val;
2400                 up->pcflag |= UDPLITE_RECV_CC;
2401                 break;
2402 
2403         default:
2404                 err = -ENOPROTOOPT;
2405                 break;
2406         }
2407 
2408         return err;
2409 }
2410 EXPORT_SYMBOL(udp_lib_setsockopt);
2411 
2412 int udp_setsockopt(struct sock *sk, int level, int optname,
2413                    char __user *optval, unsigned int optlen)
2414 {
2415         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2416                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2417                                           udp_push_pending_frames);
2418         return ip_setsockopt(sk, level, optname, optval, optlen);
2419 }
2420 
2421 #ifdef CONFIG_COMPAT
2422 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2423                           char __user *optval, unsigned int optlen)
2424 {
2425         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2426                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2427                                           udp_push_pending_frames);
2428         return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2429 }
2430 #endif
2431 
2432 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2433                        char __user *optval, int __user *optlen)
2434 {
2435         struct udp_sock *up = udp_sk(sk);
2436         int val, len;
2437 
2438         if (get_user(len, optlen))
2439                 return -EFAULT;
2440 
2441         len = min_t(unsigned int, len, sizeof(int));
2442 
2443         if (len < 0)
2444                 return -EINVAL;
2445 
2446         switch (optname) {
2447         case UDP_CORK:
2448                 val = up->corkflag;
2449                 break;
2450 
2451         case UDP_ENCAP:
2452                 val = up->encap_type;
2453                 break;
2454 
2455         case UDP_NO_CHECK6_TX:
2456                 val = up->no_check6_tx;
2457                 break;
2458 
2459         case UDP_NO_CHECK6_RX:
2460                 val = up->no_check6_rx;
2461                 break;
2462 
2463         /* The following two cannot be changed on UDP sockets, the return is
2464          * always 0 (which corresponds to the full checksum coverage of UDP). */
2465         case UDPLITE_SEND_CSCOV:
2466                 val = up->pcslen;
2467                 break;
2468 
2469         case UDPLITE_RECV_CSCOV:
2470                 val = up->pcrlen;
2471                 break;
2472 
2473         default:
2474                 return -ENOPROTOOPT;
2475         }
2476 
2477         if (put_user(len, optlen))
2478                 return -EFAULT;
2479         if (copy_to_user(optval, &val, len))
2480                 return -EFAULT;
2481         return 0;
2482 }
2483 EXPORT_SYMBOL(udp_lib_getsockopt);
2484 
2485 int udp_getsockopt(struct sock *sk, int level, int optname,
2486                    char __user *optval, int __user *optlen)
2487 {
2488         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2489                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2490         return ip_getsockopt(sk, level, optname, optval, optlen);
2491 }
2492 
2493 #ifdef CONFIG_COMPAT
2494 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2495                                  char __user *optval, int __user *optlen)
2496 {
2497         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2498                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2499         return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2500 }
2501 #endif
2502 /**
2503  *      udp_poll - wait for a UDP event.
2504  *      @file - file struct
2505  *      @sock - socket
2506  *      @wait - poll table
2507  *
2508  *      This is same as datagram poll, except for the special case of
2509  *      blocking sockets. If application is using a blocking fd
2510  *      and a packet with checksum error is in the queue;
2511  *      then it could get return from select indicating data available
2512  *      but then block when reading it. Add special case code
2513  *      to work around these arguably broken applications.
2514  */
2515 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2516 {
2517         __poll_t mask = datagram_poll(file, sock, wait);
2518         struct sock *sk = sock->sk;
2519 
2520         if (!skb_queue_empty(&udp_sk(sk)->reader_queue))
2521                 mask |= EPOLLIN | EPOLLRDNORM;
2522 
2523         /* Check for false positives due to checksum errors */
2524         if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2525             !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2526                 mask &= ~(EPOLLIN | EPOLLRDNORM);
2527 
2528         return mask;
2529 
2530 }
2531 EXPORT_SYMBOL(udp_poll);
2532 
2533 int udp_abort(struct sock *sk, int err)
2534 {
2535         lock_sock(sk);
2536 
2537         sk->sk_err = err;
2538         sk->sk_error_report(sk);
2539         __udp_disconnect(sk, 0);
2540 
2541         release_sock(sk);
2542 
2543         return 0;
2544 }
2545 EXPORT_SYMBOL_GPL(udp_abort);
2546 
2547 struct proto udp_prot = {
2548         .name                   = "UDP",
2549         .owner                  = THIS_MODULE,
2550         .close                  = udp_lib_close,
2551         .pre_connect            = udp_pre_connect,
2552         .connect                = ip4_datagram_connect,
2553         .disconnect             = udp_disconnect,
2554         .ioctl                  = udp_ioctl,
2555         .init                   = udp_init_sock,
2556         .destroy                = udp_destroy_sock,
2557         .setsockopt             = udp_setsockopt,
2558         .getsockopt             = udp_getsockopt,
2559         .sendmsg                = udp_sendmsg,
2560         .recvmsg                = udp_recvmsg,
2561         .sendpage               = udp_sendpage,
2562         .release_cb             = ip4_datagram_release_cb,
2563         .hash                   = udp_lib_hash,
2564         .unhash                 = udp_lib_unhash,
2565         .rehash                 = udp_v4_rehash,
2566         .get_port               = udp_v4_get_port,
2567         .memory_allocated       = &udp_memory_allocated,
2568         .sysctl_mem             = sysctl_udp_mem,
2569         .sysctl_wmem_offset     = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2570         .sysctl_rmem_offset     = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2571         .obj_size               = sizeof(struct udp_sock),
2572         .h.udp_table            = &udp_table,
2573 #ifdef CONFIG_COMPAT
2574         .compat_setsockopt      = compat_udp_setsockopt,
2575         .compat_getsockopt      = compat_udp_getsockopt,
2576 #endif
2577         .diag_destroy           = udp_abort,
2578 };
2579 EXPORT_SYMBOL(udp_prot);
2580 
2581 /* ------------------------------------------------------------------------ */
2582 #ifdef CONFIG_PROC_FS
2583 
2584 static struct sock *udp_get_first(struct seq_file *seq, int start)
2585 {
2586         struct sock *sk;
2587         struct udp_iter_state *state = seq->private;
2588         struct net *net = seq_file_net(seq);
2589 
2590         for (state->bucket = start; state->bucket <= state->udp_table->mask;
2591              ++state->bucket) {
2592                 struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
2593 
2594                 if (hlist_empty(&hslot->head))
2595                         continue;
2596 
2597                 spin_lock_bh(&hslot->lock);
2598                 sk_for_each(sk, &hslot->head) {
2599                         if (!net_eq(sock_net(sk), net))
2600                                 continue;
2601                         if (sk->sk_family == state->family)
2602                                 goto found;
2603                 }
2604                 spin_unlock_bh(&hslot->lock);
2605         }
2606         sk = NULL;
2607 found:
2608         return sk;
2609 }
2610 
2611 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2612 {
2613         struct udp_iter_state *state = seq->private;
2614         struct net *net = seq_file_net(seq);
2615 
2616         do {
2617                 sk = sk_next(sk);
2618         } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2619 
2620         if (!sk) {
2621                 if (state->bucket <= state->udp_table->mask)
2622                         spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2623                 return udp_get_first(seq, state->bucket + 1);
2624         }
2625         return sk;
2626 }
2627 
2628 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2629 {
2630         struct sock *sk = udp_get_first(seq, 0);
2631 
2632         if (sk)
2633                 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2634                         --pos;
2635         return pos ? NULL : sk;
2636 }
2637 
2638 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2639 {
2640         struct udp_iter_state *state = seq->private;
2641         state->bucket = MAX_UDP_PORTS;
2642 
2643         return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2644 }
2645 
2646 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2647 {
2648         struct sock *sk;
2649 
2650         if (v == SEQ_START_TOKEN)
2651                 sk = udp_get_idx(seq, 0);
2652         else
2653                 sk = udp_get_next(seq, v);
2654 
2655         ++*pos;
2656         return sk;
2657 }
2658 
2659 static void udp_seq_stop(struct seq_file *seq, void *v)
2660 {
2661         struct udp_iter_state *state = seq->private;
2662 
2663         if (state->bucket <= state->udp_table->mask)
2664                 spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2665 }
2666 
2667 int udp_seq_open(struct inode *inode, struct file *file)
2668 {
2669         struct udp_seq_afinfo *afinfo = PDE_DATA(inode);
2670         struct udp_iter_state *s;
2671         int err;
2672 
2673         err = seq_open_net(inode, file, &afinfo->seq_ops,
2674                            sizeof(struct udp_iter_state));
2675         if (err < 0)
2676                 return err;
2677 
2678         s = ((struct seq_file *)file->private_data)->private;
2679         s->family               = afinfo->family;
2680         s->udp_table            = afinfo->udp_table;
2681         return err;
2682 }
2683 EXPORT_SYMBOL(udp_seq_open);
2684 
2685 /* ------------------------------------------------------------------------ */
2686 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2687 {
2688         struct proc_dir_entry *p;
2689         int rc = 0;
2690 
2691         afinfo->seq_ops.start           = udp_seq_start;
2692         afinfo->seq_ops.next            = udp_seq_next;
2693         afinfo->seq_ops.stop            = udp_seq_stop;
2694 
2695         p = proc_create_data(afinfo->name, 0444, net->proc_net,
2696                              afinfo->seq_fops, afinfo);
2697         if (!p)
2698                 rc = -ENOMEM;
2699         return rc;
2700 }
2701 EXPORT_SYMBOL(udp_proc_register);
2702 
2703 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2704 {
2705         remove_proc_entry(afinfo->name, net->proc_net);
2706 }
2707 EXPORT_SYMBOL(udp_proc_unregister);
2708 
2709 /* ------------------------------------------------------------------------ */
2710 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2711                 int bucket)
2712 {
2713         struct inet_sock *inet = inet_sk(sp);
2714         __be32 dest = inet->inet_daddr;
2715         __be32 src  = inet->inet_rcv_saddr;
2716         __u16 destp       = ntohs(inet->inet_dport);
2717         __u16 srcp        = ntohs(inet->inet_sport);
2718 
2719         seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2720                 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %d",
2721                 bucket, src, srcp, dest, destp, sp->sk_state,
2722                 sk_wmem_alloc_get(sp),
2723                 udp_rqueue_get(sp),
2724                 0, 0L, 0,
2725                 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2726                 0, sock_i_ino(sp),
2727                 refcount_read(&sp->sk_refcnt), sp,
2728                 atomic_read(&sp->sk_drops));
2729 }
2730 
2731 int udp4_seq_show(struct seq_file *seq, void *v)
2732 {
2733         seq_setwidth(seq, 127);
2734         if (v == SEQ_START_TOKEN)
2735                 seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2736                            "rx_queue tr tm->when retrnsmt   uid  timeout "
2737                            "inode ref pointer drops");
2738         else {
2739                 struct udp_iter_state *state = seq->private;
2740 
2741                 udp4_format_sock(v, seq, state->bucket);
2742         }
2743         seq_pad(seq, '\n');
2744         return 0;
2745 }
2746 
2747 static const struct file_operations udp_afinfo_seq_fops = {
2748         .open     = udp_seq_open,
2749         .read     = seq_read,
2750         .llseek   = seq_lseek,
2751         .release  = seq_release_net
2752 };
2753 
2754 /* ------------------------------------------------------------------------ */
2755 static struct udp_seq_afinfo udp4_seq_afinfo = {
2756         .name           = "udp",
2757         .family         = AF_INET,
2758         .udp_table      = &udp_table,
2759         .seq_fops       = &udp_afinfo_seq_fops,
2760         .seq_ops        = {
2761                 .show           = udp4_seq_show,
2762         },
2763 };
2764 
2765 static int __net_init udp4_proc_init_net(struct net *net)
2766 {
2767         return udp_proc_register(net, &udp4_seq_afinfo);
2768 }
2769 
2770 static void __net_exit udp4_proc_exit_net(struct net *net)
2771 {
2772         udp_proc_unregister(net, &udp4_seq_afinfo);
2773 }
2774 
2775 static struct pernet_operations udp4_net_ops = {
2776         .init = udp4_proc_init_net,
2777         .exit = udp4_proc_exit_net,
2778 };
2779 
2780 int __init udp4_proc_init(void)
2781 {
2782         return register_pernet_subsys(&udp4_net_ops);
2783 }
2784 
2785 void udp4_proc_exit(void)
2786 {
2787         unregister_pernet_subsys(&udp4_net_ops);
2788 }
2789 #endif /* CONFIG_PROC_FS */
2790 
2791 static __initdata unsigned long uhash_entries;
2792 static int __init set_uhash_entries(char *str)
2793 {
2794         ssize_t ret;
2795 
2796         if (!str)
2797                 return 0;
2798 
2799         ret = kstrtoul(str, 0, &uhash_entries);
2800         if (ret)
2801                 return 0;
2802 
2803         if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2804                 uhash_entries = UDP_HTABLE_SIZE_MIN;
2805         return 1;
2806 }
2807 __setup("uhash_entries=", set_uhash_entries);
2808 
2809 void __init udp_table_init(struct udp_table *table, const char *name)
2810 {
2811         unsigned int i;
2812 
2813         table->hash = alloc_large_system_hash(name,
2814                                               2 * sizeof(struct udp_hslot),
2815                                               uhash_entries,
2816                                               21, /* one slot per 2 MB */
2817                                               0,
2818                                               &table->log,
2819                                               &table->mask,
2820                                               UDP_HTABLE_SIZE_MIN,
2821                                               64 * 1024);
2822 
2823         table->hash2 = table->hash + (table->mask + 1);
2824         for (i = 0; i <= table->mask; i++) {
2825                 INIT_HLIST_HEAD(&table->hash[i].head);
2826                 table->hash[i].count = 0;
2827                 spin_lock_init(&table->hash[i].lock);
2828         }
2829         for (i = 0; i <= table->mask; i++) {
2830                 INIT_HLIST_HEAD(&table->hash2[i].head);
2831                 table->hash2[i].count = 0;
2832                 spin_lock_init(&table->hash2[i].lock);
2833         }
2834 }
2835 
2836 u32 udp_flow_hashrnd(void)
2837 {
2838         static u32 hashrnd __read_mostly;
2839 
2840         net_get_random_once(&hashrnd, sizeof(hashrnd));
2841 
2842         return hashrnd;
2843 }
2844 EXPORT_SYMBOL(udp_flow_hashrnd);
2845 
2846 static void __udp_sysctl_init(struct net *net)
2847 {
2848         net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2849         net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2850 
2851 #ifdef CONFIG_NET_L3_MASTER_DEV
2852         net->ipv4.sysctl_udp_l3mdev_accept = 0;
2853 #endif
2854 }
2855 
2856 static int __net_init udp_sysctl_init(struct net *net)
2857 {
2858         __udp_sysctl_init(net);
2859         return 0;
2860 }
2861 
2862 static struct pernet_operations __net_initdata udp_sysctl_ops = {
2863         .init   = udp_sysctl_init,
2864 };
2865 
2866 void __init udp_init(void)
2867 {
2868         unsigned long limit;
2869         unsigned int i;
2870 
2871         udp_table_init(&udp_table, "UDP");
2872         limit = nr_free_buffer_pages() / 8;
2873         limit = max(limit, 128UL);
2874         sysctl_udp_mem[0] = limit / 4 * 3;
2875         sysctl_udp_mem[1] = limit;
2876         sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2877 
2878         __udp_sysctl_init(&init_net);
2879 
2880         /* 16 spinlocks per cpu */
2881         udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
2882         udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
2883                                 GFP_KERNEL);
2884         if (!udp_busylocks)
2885                 panic("UDP: failed to alloc udp_busylocks\n");
2886         for (i = 0; i < (1U << udp_busylocks_log); i++)
2887                 spin_lock_init(udp_busylocks + i);
2888 
2889         if (register_pernet_subsys(&udp_sysctl_ops))
2890                 panic("UDP: failed to init sysctl parameters.\n");
2891 }
2892 

~ [ 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