root/net/ipv4/udp.c

/* [<][>][^][v][top][bottom][index][help] */

DEFINITIONS

This source file includes following definitions.
  1. udp_lib_lport_inuse
  2. udp_lib_lport_inuse2
  3. udp_reuseport_add_sock
  4. udp_lib_get_port
  5. udp_v4_get_port
  6. compute_score
  7. udp_ehashfn
  8. udp4_lib_lookup2
  9. __udp4_lib_lookup
  10. __udp4_lib_lookup_skb
  11. udp4_lib_lookup_skb
  12. udp4_lib_lookup
  13. __udp_is_mcast_sock
  14. udp_encap_enable
  15. __udp4_lib_err_encap_no_sk
  16. __udp4_lib_err_encap
  17. __udp4_lib_err
  18. udp_err
  19. udp_flush_pending_frames
  20. udp4_hwcsum
  21. udp_set_csum
  22. udp_send_skb
  23. udp_push_pending_frames
  24. __udp_cmsg_send
  25. udp_cmsg_send
  26. udp_sendmsg
  27. udp_sendpage
  28. udp_try_make_stateless
  29. udp_set_dev_scratch
  30. udp_skb_csum_unnecessary_set
  31. udp_skb_truesize
  32. udp_skb_has_head_state
  33. udp_rmem_release
  34. udp_skb_destructor
  35. udp_skb_dtor_locked
  36. busylock_acquire
  37. busylock_release
  38. __udp_enqueue_schedule_skb
  39. udp_destruct_sock
  40. udp_init_sock
  41. skb_consume_udp
  42. __first_packet_length
  43. first_packet_length
  44. udp_ioctl
  45. __skb_recv_udp
  46. udp_recvmsg
  47. udp_pre_connect
  48. __udp_disconnect
  49. udp_disconnect
  50. udp_lib_unhash
  51. udp_lib_rehash
  52. udp_v4_rehash
  53. __udp_queue_rcv_skb
  54. udp_queue_rcv_one_skb
  55. udp_queue_rcv_skb
  56. udp_sk_rx_dst_set
  57. __udp4_lib_mcast_deliver
  58. udp4_csum_init
  59. udp_unicast_rcv_skb
  60. __udp4_lib_rcv
  61. __udp4_lib_mcast_demux_lookup
  62. __udp4_lib_demux_lookup
  63. udp_v4_early_demux
  64. udp_rcv
  65. udp_destroy_sock
  66. udp_lib_setsockopt
  67. udp_setsockopt
  68. compat_udp_setsockopt
  69. udp_lib_getsockopt
  70. udp_getsockopt
  71. compat_udp_getsockopt
  72. udp_poll
  73. udp_abort
  74. udp_get_first
  75. udp_get_next
  76. udp_get_idx
  77. udp_seq_start
  78. udp_seq_next
  79. udp_seq_stop
  80. udp4_format_sock
  81. udp4_seq_show
  82. udp4_proc_init_net
  83. udp4_proc_exit_net
  84. udp4_proc_init
  85. udp4_proc_exit
  86. set_uhash_entries
  87. udp_table_init
  88. udp_flow_hashrnd
  89. __udp_sysctl_init
  90. udp_sysctl_init
  91. udp_init

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   4  *              operating system.  INET is implemented using the  BSD Socket
   5  *              interface as the means of communication with the user level.
   6  *
   7  *              The User Datagram Protocol (UDP).
   8  *
   9  * Authors:     Ross Biro
  10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  11  *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  12  *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
  13  *              Hirokazu Takahashi, <taka@valinux.co.jp>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       verify_area() calls
  17  *              Alan Cox        :       stopped close while in use off icmp
  18  *                                      messages. Not a fix but a botch that
  19  *                                      for udp at least is 'valid'.
  20  *              Alan Cox        :       Fixed icmp handling properly
  21  *              Alan Cox        :       Correct error for oversized datagrams
  22  *              Alan Cox        :       Tidied select() semantics.
  23  *              Alan Cox        :       udp_err() fixed properly, also now
  24  *                                      select and read wake correctly on errors
  25  *              Alan Cox        :       udp_send verify_area moved to avoid mem leak
  26  *              Alan Cox        :       UDP can count its memory
  27  *              Alan Cox        :       send to an unknown connection causes
  28  *                                      an ECONNREFUSED off the icmp, but
  29  *                                      does NOT close.
  30  *              Alan Cox        :       Switched to new sk_buff handlers. No more backlog!
  31  *              Alan Cox        :       Using generic datagram code. Even smaller and the PEEK
  32  *                                      bug no longer crashes it.
  33  *              Fred Van Kempen :       Net2e support for sk->broadcast.
  34  *              Alan Cox        :       Uses skb_free_datagram
  35  *              Alan Cox        :       Added get/set sockopt support.
  36  *              Alan Cox        :       Broadcasting without option set returns EACCES.
  37  *              Alan Cox        :       No wakeup calls. Instead we now use the callbacks.
  38  *              Alan Cox        :       Use ip_tos and ip_ttl
  39  *              Alan Cox        :       SNMP Mibs
  40  *              Alan Cox        :       MSG_DONTROUTE, and 0.0.0.0 support.
  41  *              Matt Dillon     :       UDP length checks.
  42  *              Alan Cox        :       Smarter af_inet used properly.
  43  *              Alan Cox        :       Use new kernel side addressing.
  44  *              Alan Cox        :       Incorrect return on truncated datagram receive.
  45  *      Arnt Gulbrandsen        :       New udp_send and stuff
  46  *              Alan Cox        :       Cache last socket
  47  *              Alan Cox        :       Route cache
  48  *              Jon Peatfield   :       Minor efficiency fix to sendto().
  49  *              Mike Shaver     :       RFC1122 checks.
  50  *              Alan Cox        :       Nonblocking error fix.
  51  *      Willy Konynenberg       :       Transparent proxying support.
  52  *              Mike McLagan    :       Routing by source
  53  *              David S. Miller :       New socket lookup architecture.
  54  *                                      Last socket cache retained as it
  55  *                                      does have a high hit rate.
  56  *              Olaf Kirch      :       Don't linearise iovec on sendmsg.
  57  *              Andi Kleen      :       Some cleanups, cache destination entry
  58  *                                      for connect.
  59  *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  60  *              Melvin Smith    :       Check msg_name not msg_namelen in sendto(),
  61  *                                      return ENOTCONN for unconnected sockets (POSIX)
  62  *              Janos Farkas    :       don't deliver multi/broadcasts to a different
  63  *                                      bound-to-device socket
  64  *      Hirokazu Takahashi      :       HW checksumming for outgoing UDP
  65  *                                      datagrams.
  66  *      Hirokazu Takahashi      :       sendfile() on UDP works now.
  67  *              Arnaldo C. Melo :       convert /proc/net/udp to seq_file
  68  *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
  69  *      Alexey Kuznetsov:               allow both IPv4 and IPv6 sockets to bind
  70  *                                      a single port at the same time.
  71  *      Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  72  *      James Chapman           :       Add L2TP encapsulation type.
  73  */
  74 
  75 #define pr_fmt(fmt) "UDP: " fmt
  76 
  77 #include <linux/uaccess.h>
  78 #include <asm/ioctls.h>
  79 #include <linux/memblock.h>
  80 #include <linux/highmem.h>
  81 #include <linux/swap.h>
  82 #include <linux/types.h>
  83 #include <linux/fcntl.h>
  84 #include <linux/module.h>
  85 #include <linux/socket.h>
  86 #include <linux/sockios.h>
  87 #include <linux/igmp.h>
  88 #include <linux/inetdevice.h>
  89 #include <linux/in.h>
  90 #include <linux/errno.h>
  91 #include <linux/timer.h>
  92 #include <linux/mm.h>
  93 #include <linux/inet.h>
  94 #include <linux/netdevice.h>
  95 #include <linux/slab.h>
  96 #include <net/tcp_states.h>
  97 #include <linux/skbuff.h>
  98 #include <linux/proc_fs.h>
  99 #include <linux/seq_file.h>
 100 #include <net/net_namespace.h>
 101 #include <net/icmp.h>
 102 #include <net/inet_hashtables.h>
 103 #include <net/ip_tunnels.h>
 104 #include <net/route.h>
 105 #include <net/checksum.h>
 106 #include <net/xfrm.h>
 107 #include <trace/events/udp.h>
 108 #include <linux/static_key.h>
 109 #include <trace/events/skb.h>
 110 #include <net/busy_poll.h>
 111 #include "udp_impl.h"
 112 #include <net/sock_reuseport.h>
 113 #include <net/addrconf.h>
 114 #include <net/udp_tunnel.h>
 115 
 116 struct udp_table udp_table __read_mostly;
 117 EXPORT_SYMBOL(udp_table);
 118 
 119 long sysctl_udp_mem[3] __read_mostly;
 120 EXPORT_SYMBOL(sysctl_udp_mem);
 121 
 122 atomic_long_t udp_memory_allocated;
 123 EXPORT_SYMBOL(udp_memory_allocated);
 124 
 125 #define MAX_UDP_PORTS 65536
 126 #define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
 127 
 128 static int udp_lib_lport_inuse(struct net *net, __u16 num,
 129                                const struct udp_hslot *hslot,
 130                                unsigned long *bitmap,
 131                                struct sock *sk, unsigned int log)
 132 {
 133         struct sock *sk2;
 134         kuid_t uid = sock_i_uid(sk);
 135 
 136         sk_for_each(sk2, &hslot->head) {
 137                 if (net_eq(sock_net(sk2), net) &&
 138                     sk2 != sk &&
 139                     (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 140                     (!sk2->sk_reuse || !sk->sk_reuse) &&
 141                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 142                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 143                     inet_rcv_saddr_equal(sk, sk2, true)) {
 144                         if (sk2->sk_reuseport && sk->sk_reuseport &&
 145                             !rcu_access_pointer(sk->sk_reuseport_cb) &&
 146                             uid_eq(uid, sock_i_uid(sk2))) {
 147                                 if (!bitmap)
 148                                         return 0;
 149                         } else {
 150                                 if (!bitmap)
 151                                         return 1;
 152                                 __set_bit(udp_sk(sk2)->udp_port_hash >> log,
 153                                           bitmap);
 154                         }
 155                 }
 156         }
 157         return 0;
 158 }
 159 
 160 /*
 161  * Note: we still hold spinlock of primary hash chain, so no other writer
 162  * can insert/delete a socket with local_port == num
 163  */
 164 static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 165                                 struct udp_hslot *hslot2,
 166                                 struct sock *sk)
 167 {
 168         struct sock *sk2;
 169         kuid_t uid = sock_i_uid(sk);
 170         int res = 0;
 171 
 172         spin_lock(&hslot2->lock);
 173         udp_portaddr_for_each_entry(sk2, &hslot2->head) {
 174                 if (net_eq(sock_net(sk2), net) &&
 175                     sk2 != sk &&
 176                     (udp_sk(sk2)->udp_port_hash == num) &&
 177                     (!sk2->sk_reuse || !sk->sk_reuse) &&
 178                     (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 179                      sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 180                     inet_rcv_saddr_equal(sk, sk2, true)) {
 181                         if (sk2->sk_reuseport && sk->sk_reuseport &&
 182                             !rcu_access_pointer(sk->sk_reuseport_cb) &&
 183                             uid_eq(uid, sock_i_uid(sk2))) {
 184                                 res = 0;
 185                         } else {
 186                                 res = 1;
 187                         }
 188                         break;
 189                 }
 190         }
 191         spin_unlock(&hslot2->lock);
 192         return res;
 193 }
 194 
 195 static int udp_reuseport_add_sock(struct sock *sk, struct udp_hslot *hslot)
 196 {
 197         struct net *net = sock_net(sk);
 198         kuid_t uid = sock_i_uid(sk);
 199         struct sock *sk2;
 200 
 201         sk_for_each(sk2, &hslot->head) {
 202                 if (net_eq(sock_net(sk2), net) &&
 203                     sk2 != sk &&
 204                     sk2->sk_family == sk->sk_family &&
 205                     ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
 206                     (udp_sk(sk2)->udp_port_hash == udp_sk(sk)->udp_port_hash) &&
 207                     (sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 208                     sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
 209                     inet_rcv_saddr_equal(sk, sk2, false)) {
 210                         return reuseport_add_sock(sk, sk2,
 211                                                   inet_rcv_saddr_any(sk));
 212                 }
 213         }
 214 
 215         return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
 216 }
 217 
 218 /**
 219  *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 220  *
 221  *  @sk:          socket struct in question
 222  *  @snum:        port number to look up
 223  *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 224  *                   with NULL address
 225  */
 226 int udp_lib_get_port(struct sock *sk, unsigned short snum,
 227                      unsigned int hash2_nulladdr)
 228 {
 229         struct udp_hslot *hslot, *hslot2;
 230         struct udp_table *udptable = sk->sk_prot->h.udp_table;
 231         int    error = 1;
 232         struct net *net = sock_net(sk);
 233 
 234         if (!snum) {
 235                 int low, high, remaining;
 236                 unsigned int rand;
 237                 unsigned short first, last;
 238                 DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 239 
 240                 inet_get_local_port_range(net, &low, &high);
 241                 remaining = (high - low) + 1;
 242 
 243                 rand = prandom_u32();
 244                 first = reciprocal_scale(rand, remaining) + low;
 245                 /*
 246                  * force rand to be an odd multiple of UDP_HTABLE_SIZE
 247                  */
 248                 rand = (rand | 1) * (udptable->mask + 1);
 249                 last = first + udptable->mask + 1;
 250                 do {
 251                         hslot = udp_hashslot(udptable, net, first);
 252                         bitmap_zero(bitmap, PORTS_PER_CHAIN);
 253                         spin_lock_bh(&hslot->lock);
 254                         udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 255                                             udptable->log);
 256 
 257                         snum = first;
 258                         /*
 259                          * Iterate on all possible values of snum for this hash.
 260                          * Using steps of an odd multiple of UDP_HTABLE_SIZE
 261                          * give us randomization and full range coverage.
 262                          */
 263                         do {
 264                                 if (low <= snum && snum <= high &&
 265                                     !test_bit(snum >> udptable->log, bitmap) &&
 266                                     !inet_is_local_reserved_port(net, snum))
 267                                         goto found;
 268                                 snum += rand;
 269                         } while (snum != first);
 270                         spin_unlock_bh(&hslot->lock);
 271                         cond_resched();
 272                 } while (++first != last);
 273                 goto fail;
 274         } else {
 275                 hslot = udp_hashslot(udptable, net, snum);
 276                 spin_lock_bh(&hslot->lock);
 277                 if (hslot->count > 10) {
 278                         int exist;
 279                         unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 280 
 281                         slot2          &= udptable->mask;
 282                         hash2_nulladdr &= udptable->mask;
 283 
 284                         hslot2 = udp_hashslot2(udptable, slot2);
 285                         if (hslot->count < hslot2->count)
 286                                 goto scan_primary_hash;
 287 
 288                         exist = udp_lib_lport_inuse2(net, snum, hslot2, sk);
 289                         if (!exist && (hash2_nulladdr != slot2)) {
 290                                 hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 291                                 exist = udp_lib_lport_inuse2(net, snum, hslot2,
 292                                                              sk);
 293                         }
 294                         if (exist)
 295                                 goto fail_unlock;
 296                         else
 297                                 goto found;
 298                 }
 299 scan_primary_hash:
 300                 if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk, 0))
 301                         goto fail_unlock;
 302         }
 303 found:
 304         inet_sk(sk)->inet_num = snum;
 305         udp_sk(sk)->udp_port_hash = snum;
 306         udp_sk(sk)->udp_portaddr_hash ^= snum;
 307         if (sk_unhashed(sk)) {
 308                 if (sk->sk_reuseport &&
 309                     udp_reuseport_add_sock(sk, hslot)) {
 310                         inet_sk(sk)->inet_num = 0;
 311                         udp_sk(sk)->udp_port_hash = 0;
 312                         udp_sk(sk)->udp_portaddr_hash ^= snum;
 313                         goto fail_unlock;
 314                 }
 315 
 316                 sk_add_node_rcu(sk, &hslot->head);
 317                 hslot->count++;
 318                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 319 
 320                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 321                 spin_lock(&hslot2->lock);
 322                 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 323                     sk->sk_family == AF_INET6)
 324                         hlist_add_tail_rcu(&udp_sk(sk)->udp_portaddr_node,
 325                                            &hslot2->head);
 326                 else
 327                         hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 328                                            &hslot2->head);
 329                 hslot2->count++;
 330                 spin_unlock(&hslot2->lock);
 331         }
 332         sock_set_flag(sk, SOCK_RCU_FREE);
 333         error = 0;
 334 fail_unlock:
 335         spin_unlock_bh(&hslot->lock);
 336 fail:
 337         return error;
 338 }
 339 EXPORT_SYMBOL(udp_lib_get_port);
 340 
 341 int udp_v4_get_port(struct sock *sk, unsigned short snum)
 342 {
 343         unsigned int hash2_nulladdr =
 344                 ipv4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 345         unsigned int hash2_partial =
 346                 ipv4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 347 
 348         /* precompute partial secondary hash */
 349         udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 350         return udp_lib_get_port(sk, snum, hash2_nulladdr);
 351 }
 352 
 353 static int compute_score(struct sock *sk, struct net *net,
 354                          __be32 saddr, __be16 sport,
 355                          __be32 daddr, unsigned short hnum,
 356                          int dif, int sdif)
 357 {
 358         int score;
 359         struct inet_sock *inet;
 360         bool dev_match;
 361 
 362         if (!net_eq(sock_net(sk), net) ||
 363             udp_sk(sk)->udp_port_hash != hnum ||
 364             ipv6_only_sock(sk))
 365                 return -1;
 366 
 367         if (sk->sk_rcv_saddr != daddr)
 368                 return -1;
 369 
 370         score = (sk->sk_family == PF_INET) ? 2 : 1;
 371 
 372         inet = inet_sk(sk);
 373         if (inet->inet_daddr) {
 374                 if (inet->inet_daddr != saddr)
 375                         return -1;
 376                 score += 4;
 377         }
 378 
 379         if (inet->inet_dport) {
 380                 if (inet->inet_dport != sport)
 381                         return -1;
 382                 score += 4;
 383         }
 384 
 385         dev_match = udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if,
 386                                         dif, sdif);
 387         if (!dev_match)
 388                 return -1;
 389         score += 4;
 390 
 391         if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
 392                 score++;
 393         return score;
 394 }
 395 
 396 static u32 udp_ehashfn(const struct net *net, const __be32 laddr,
 397                        const __u16 lport, const __be32 faddr,
 398                        const __be16 fport)
 399 {
 400         static u32 udp_ehash_secret __read_mostly;
 401 
 402         net_get_random_once(&udp_ehash_secret, sizeof(udp_ehash_secret));
 403 
 404         return __inet_ehashfn(laddr, lport, faddr, fport,
 405                               udp_ehash_secret + net_hash_mix(net));
 406 }
 407 
 408 /* called with rcu_read_lock() */
 409 static struct sock *udp4_lib_lookup2(struct net *net,
 410                                      __be32 saddr, __be16 sport,
 411                                      __be32 daddr, unsigned int hnum,
 412                                      int dif, int sdif,
 413                                      struct udp_hslot *hslot2,
 414                                      struct sk_buff *skb)
 415 {
 416         struct sock *sk, *result;
 417         int score, badness;
 418         u32 hash = 0;
 419 
 420         result = NULL;
 421         badness = 0;
 422         udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
 423                 score = compute_score(sk, net, saddr, sport,
 424                                       daddr, hnum, dif, sdif);
 425                 if (score > badness) {
 426                         if (sk->sk_reuseport &&
 427                             sk->sk_state != TCP_ESTABLISHED) {
 428                                 hash = udp_ehashfn(net, daddr, hnum,
 429                                                    saddr, sport);
 430                                 result = reuseport_select_sock(sk, hash, skb,
 431                                                         sizeof(struct udphdr));
 432                                 if (result && !reuseport_has_conns(sk, false))
 433                                         return result;
 434                         }
 435                         badness = score;
 436                         result = sk;
 437                 }
 438         }
 439         return result;
 440 }
 441 
 442 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 443  * harder than this. -DaveM
 444  */
 445 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 446                 __be16 sport, __be32 daddr, __be16 dport, int dif,
 447                 int sdif, struct udp_table *udptable, struct sk_buff *skb)
 448 {
 449         struct sock *result;
 450         unsigned short hnum = ntohs(dport);
 451         unsigned int hash2, slot2;
 452         struct udp_hslot *hslot2;
 453 
 454         hash2 = ipv4_portaddr_hash(net, daddr, hnum);
 455         slot2 = hash2 & udptable->mask;
 456         hslot2 = &udptable->hash2[slot2];
 457 
 458         result = udp4_lib_lookup2(net, saddr, sport,
 459                                   daddr, hnum, dif, sdif,
 460                                   hslot2, skb);
 461         if (!result) {
 462                 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 463                 slot2 = hash2 & udptable->mask;
 464                 hslot2 = &udptable->hash2[slot2];
 465 
 466                 result = udp4_lib_lookup2(net, saddr, sport,
 467                                           htonl(INADDR_ANY), hnum, dif, sdif,
 468                                           hslot2, skb);
 469         }
 470         if (IS_ERR(result))
 471                 return NULL;
 472         return result;
 473 }
 474 EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 475 
 476 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 477                                                  __be16 sport, __be16 dport,
 478                                                  struct udp_table *udptable)
 479 {
 480         const struct iphdr *iph = ip_hdr(skb);
 481 
 482         return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 483                                  iph->daddr, dport, inet_iif(skb),
 484                                  inet_sdif(skb), udptable, skb);
 485 }
 486 
 487 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
 488                                  __be16 sport, __be16 dport)
 489 {
 490         const struct iphdr *iph = ip_hdr(skb);
 491 
 492         return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
 493                                  iph->daddr, dport, inet_iif(skb),
 494                                  inet_sdif(skb), &udp_table, NULL);
 495 }
 496 EXPORT_SYMBOL_GPL(udp4_lib_lookup_skb);
 497 
 498 /* Must be called under rcu_read_lock().
 499  * Does increment socket refcount.
 500  */
 501 #if IS_ENABLED(CONFIG_NF_TPROXY_IPV4) || IS_ENABLED(CONFIG_NF_SOCKET_IPV4)
 502 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 503                              __be32 daddr, __be16 dport, int dif)
 504 {
 505         struct sock *sk;
 506 
 507         sk = __udp4_lib_lookup(net, saddr, sport, daddr, dport,
 508                                dif, 0, &udp_table, NULL);
 509         if (sk && !refcount_inc_not_zero(&sk->sk_refcnt))
 510                 sk = NULL;
 511         return sk;
 512 }
 513 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 514 #endif
 515 
 516 static inline bool __udp_is_mcast_sock(struct net *net, struct sock *sk,
 517                                        __be16 loc_port, __be32 loc_addr,
 518                                        __be16 rmt_port, __be32 rmt_addr,
 519                                        int dif, int sdif, unsigned short hnum)
 520 {
 521         struct inet_sock *inet = inet_sk(sk);
 522 
 523         if (!net_eq(sock_net(sk), net) ||
 524             udp_sk(sk)->udp_port_hash != hnum ||
 525             (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 526             (inet->inet_dport != rmt_port && inet->inet_dport) ||
 527             (inet->inet_rcv_saddr && inet->inet_rcv_saddr != loc_addr) ||
 528             ipv6_only_sock(sk) ||
 529             !udp_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
 530                 return false;
 531         if (!ip_mc_sf_allow(sk, loc_addr, rmt_addr, dif, sdif))
 532                 return false;
 533         return true;
 534 }
 535 
 536 DEFINE_STATIC_KEY_FALSE(udp_encap_needed_key);
 537 void udp_encap_enable(void)
 538 {
 539         static_branch_inc(&udp_encap_needed_key);
 540 }
 541 EXPORT_SYMBOL(udp_encap_enable);
 542 
 543 /* Handler for tunnels with arbitrary destination ports: no socket lookup, go
 544  * through error handlers in encapsulations looking for a match.
 545  */
 546 static int __udp4_lib_err_encap_no_sk(struct sk_buff *skb, u32 info)
 547 {
 548         int i;
 549 
 550         for (i = 0; i < MAX_IPTUN_ENCAP_OPS; i++) {
 551                 int (*handler)(struct sk_buff *skb, u32 info);
 552                 const struct ip_tunnel_encap_ops *encap;
 553 
 554                 encap = rcu_dereference(iptun_encaps[i]);
 555                 if (!encap)
 556                         continue;
 557                 handler = encap->err_handler;
 558                 if (handler && !handler(skb, info))
 559                         return 0;
 560         }
 561 
 562         return -ENOENT;
 563 }
 564 
 565 /* Try to match ICMP errors to UDP tunnels by looking up a socket without
 566  * reversing source and destination port: this will match tunnels that force the
 567  * same destination port on both endpoints (e.g. VXLAN, GENEVE). Note that
 568  * lwtunnels might actually break this assumption by being configured with
 569  * different destination ports on endpoints, in this case we won't be able to
 570  * trace ICMP messages back to them.
 571  *
 572  * If this doesn't match any socket, probe tunnels with arbitrary destination
 573  * ports (e.g. FoU, GUE): there, the receiving socket is useless, as the port
 574  * we've sent packets to won't necessarily match the local destination port.
 575  *
 576  * Then ask the tunnel implementation to match the error against a valid
 577  * association.
 578  *
 579  * Return an error if we can't find a match, the socket if we need further
 580  * processing, zero otherwise.
 581  */
 582 static struct sock *__udp4_lib_err_encap(struct net *net,
 583                                          const struct iphdr *iph,
 584                                          struct udphdr *uh,
 585                                          struct udp_table *udptable,
 586                                          struct sk_buff *skb, u32 info)
 587 {
 588         int network_offset, transport_offset;
 589         struct sock *sk;
 590 
 591         network_offset = skb_network_offset(skb);
 592         transport_offset = skb_transport_offset(skb);
 593 
 594         /* Network header needs to point to the outer IPv4 header inside ICMP */
 595         skb_reset_network_header(skb);
 596 
 597         /* Transport header needs to point to the UDP header */
 598         skb_set_transport_header(skb, iph->ihl << 2);
 599 
 600         sk = __udp4_lib_lookup(net, iph->daddr, uh->source,
 601                                iph->saddr, uh->dest, skb->dev->ifindex, 0,
 602                                udptable, NULL);
 603         if (sk) {
 604                 int (*lookup)(struct sock *sk, struct sk_buff *skb);
 605                 struct udp_sock *up = udp_sk(sk);
 606 
 607                 lookup = READ_ONCE(up->encap_err_lookup);
 608                 if (!lookup || lookup(sk, skb))
 609                         sk = NULL;
 610         }
 611 
 612         if (!sk)
 613                 sk = ERR_PTR(__udp4_lib_err_encap_no_sk(skb, info));
 614 
 615         skb_set_transport_header(skb, transport_offset);
 616         skb_set_network_header(skb, network_offset);
 617 
 618         return sk;
 619 }
 620 
 621 /*
 622  * This routine is called by the ICMP module when it gets some
 623  * sort of error condition.  If err < 0 then the socket should
 624  * be closed and the error returned to the user.  If err > 0
 625  * it's just the icmp type << 8 | icmp code.
 626  * Header points to the ip header of the error packet. We move
 627  * on past this. Then (as it used to claim before adjustment)
 628  * header points to the first 8 bytes of the udp header.  We need
 629  * to find the appropriate port.
 630  */
 631 
 632 int __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 633 {
 634         struct inet_sock *inet;
 635         const struct iphdr *iph = (const struct iphdr *)skb->data;
 636         struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 637         const int type = icmp_hdr(skb)->type;
 638         const int code = icmp_hdr(skb)->code;
 639         bool tunnel = false;
 640         struct sock *sk;
 641         int harderr;
 642         int err;
 643         struct net *net = dev_net(skb->dev);
 644 
 645         sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 646                                iph->saddr, uh->source, skb->dev->ifindex,
 647                                inet_sdif(skb), udptable, NULL);
 648         if (!sk) {
 649                 /* No socket for error: try tunnels before discarding */
 650                 sk = ERR_PTR(-ENOENT);
 651                 if (static_branch_unlikely(&udp_encap_needed_key)) {
 652                         sk = __udp4_lib_err_encap(net, iph, uh, udptable, skb,
 653                                                   info);
 654                         if (!sk)
 655                                 return 0;
 656                 }
 657 
 658                 if (IS_ERR(sk)) {
 659                         __ICMP_INC_STATS(net, ICMP_MIB_INERRORS);
 660                         return PTR_ERR(sk);
 661                 }
 662 
 663                 tunnel = true;
 664         }
 665 
 666         err = 0;
 667         harderr = 0;
 668         inet = inet_sk(sk);
 669 
 670         switch (type) {
 671         default:
 672         case ICMP_TIME_EXCEEDED:
 673                 err = EHOSTUNREACH;
 674                 break;
 675         case ICMP_SOURCE_QUENCH:
 676                 goto out;
 677         case ICMP_PARAMETERPROB:
 678                 err = EPROTO;
 679                 harderr = 1;
 680                 break;
 681         case ICMP_DEST_UNREACH:
 682                 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 683                         ipv4_sk_update_pmtu(skb, sk, info);
 684                         if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 685                                 err = EMSGSIZE;
 686                                 harderr = 1;
 687                                 break;
 688                         }
 689                         goto out;
 690                 }
 691                 err = EHOSTUNREACH;
 692                 if (code <= NR_ICMP_UNREACH) {
 693                         harderr = icmp_err_convert[code].fatal;
 694                         err = icmp_err_convert[code].errno;
 695                 }
 696                 break;
 697         case ICMP_REDIRECT:
 698                 ipv4_sk_redirect(skb, sk);
 699                 goto out;
 700         }
 701 
 702         /*
 703          *      RFC1122: OK.  Passes ICMP errors back to application, as per
 704          *      4.1.3.3.
 705          */
 706         if (tunnel) {
 707                 /* ...not for tunnels though: we don't have a sending socket */
 708                 goto out;
 709         }
 710         if (!inet->recverr) {
 711                 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 712                         goto out;
 713         } else
 714                 ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 715 
 716         sk->sk_err = err;
 717         sk->sk_error_report(sk);
 718 out:
 719         return 0;
 720 }
 721 
 722 int udp_err(struct sk_buff *skb, u32 info)
 723 {
 724         return __udp4_lib_err(skb, info, &udp_table);
 725 }
 726 
 727 /*
 728  * Throw away all pending data and cancel the corking. Socket is locked.
 729  */
 730 void udp_flush_pending_frames(struct sock *sk)
 731 {
 732         struct udp_sock *up = udp_sk(sk);
 733 
 734         if (up->pending) {
 735                 up->len = 0;
 736                 up->pending = 0;
 737                 ip_flush_pending_frames(sk);
 738         }
 739 }
 740 EXPORT_SYMBOL(udp_flush_pending_frames);
 741 
 742 /**
 743  *      udp4_hwcsum  -  handle outgoing HW checksumming
 744  *      @skb:   sk_buff containing the filled-in UDP header
 745  *              (checksum field must be zeroed out)
 746  *      @src:   source IP address
 747  *      @dst:   destination IP address
 748  */
 749 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 750 {
 751         struct udphdr *uh = udp_hdr(skb);
 752         int offset = skb_transport_offset(skb);
 753         int len = skb->len - offset;
 754         int hlen = len;
 755         __wsum csum = 0;
 756 
 757         if (!skb_has_frag_list(skb)) {
 758                 /*
 759                  * Only one fragment on the socket.
 760                  */
 761                 skb->csum_start = skb_transport_header(skb) - skb->head;
 762                 skb->csum_offset = offsetof(struct udphdr, check);
 763                 uh->check = ~csum_tcpudp_magic(src, dst, len,
 764                                                IPPROTO_UDP, 0);
 765         } else {
 766                 struct sk_buff *frags;
 767 
 768                 /*
 769                  * HW-checksum won't work as there are two or more
 770                  * fragments on the socket so that all csums of sk_buffs
 771                  * should be together
 772                  */
 773                 skb_walk_frags(skb, frags) {
 774                         csum = csum_add(csum, frags->csum);
 775                         hlen -= frags->len;
 776                 }
 777 
 778                 csum = skb_checksum(skb, offset, hlen, csum);
 779                 skb->ip_summed = CHECKSUM_NONE;
 780 
 781                 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 782                 if (uh->check == 0)
 783                         uh->check = CSUM_MANGLED_0;
 784         }
 785 }
 786 EXPORT_SYMBOL_GPL(udp4_hwcsum);
 787 
 788 /* Function to set UDP checksum for an IPv4 UDP packet. This is intended
 789  * for the simple case like when setting the checksum for a UDP tunnel.
 790  */
 791 void udp_set_csum(bool nocheck, struct sk_buff *skb,
 792                   __be32 saddr, __be32 daddr, int len)
 793 {
 794         struct udphdr *uh = udp_hdr(skb);
 795 
 796         if (nocheck) {
 797                 uh->check = 0;
 798         } else if (skb_is_gso(skb)) {
 799                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 800         } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
 801                 uh->check = 0;
 802                 uh->check = udp_v4_check(len, saddr, daddr, lco_csum(skb));
 803                 if (uh->check == 0)
 804                         uh->check = CSUM_MANGLED_0;
 805         } else {
 806                 skb->ip_summed = CHECKSUM_PARTIAL;
 807                 skb->csum_start = skb_transport_header(skb) - skb->head;
 808                 skb->csum_offset = offsetof(struct udphdr, check);
 809                 uh->check = ~udp_v4_check(len, saddr, daddr, 0);
 810         }
 811 }
 812 EXPORT_SYMBOL(udp_set_csum);
 813 
 814 static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4,
 815                         struct inet_cork *cork)
 816 {
 817         struct sock *sk = skb->sk;
 818         struct inet_sock *inet = inet_sk(sk);
 819         struct udphdr *uh;
 820         int err = 0;
 821         int is_udplite = IS_UDPLITE(sk);
 822         int offset = skb_transport_offset(skb);
 823         int len = skb->len - offset;
 824         int datalen = len - sizeof(*uh);
 825         __wsum csum = 0;
 826 
 827         /*
 828          * Create a UDP header
 829          */
 830         uh = udp_hdr(skb);
 831         uh->source = inet->inet_sport;
 832         uh->dest = fl4->fl4_dport;
 833         uh->len = htons(len);
 834         uh->check = 0;
 835 
 836         if (cork->gso_size) {
 837                 const int hlen = skb_network_header_len(skb) +
 838                                  sizeof(struct udphdr);
 839 
 840                 if (hlen + cork->gso_size > cork->fragsize) {
 841                         kfree_skb(skb);
 842                         return -EINVAL;
 843                 }
 844                 if (skb->len > cork->gso_size * UDP_MAX_SEGMENTS) {
 845                         kfree_skb(skb);
 846                         return -EINVAL;
 847                 }
 848                 if (sk->sk_no_check_tx) {
 849                         kfree_skb(skb);
 850                         return -EINVAL;
 851                 }
 852                 if (skb->ip_summed != CHECKSUM_PARTIAL || is_udplite ||
 853                     dst_xfrm(skb_dst(skb))) {
 854                         kfree_skb(skb);
 855                         return -EIO;
 856                 }
 857 
 858                 if (datalen > cork->gso_size) {
 859                         skb_shinfo(skb)->gso_size = cork->gso_size;
 860                         skb_shinfo(skb)->gso_type = SKB_GSO_UDP_L4;
 861                         skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(datalen,
 862                                                                  cork->gso_size);
 863                 }
 864                 goto csum_partial;
 865         }
 866 
 867         if (is_udplite)                                  /*     UDP-Lite      */
 868                 csum = udplite_csum(skb);
 869 
 870         else if (sk->sk_no_check_tx) {                   /* UDP csum off */
 871 
 872                 skb->ip_summed = CHECKSUM_NONE;
 873                 goto send;
 874 
 875         } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 876 csum_partial:
 877 
 878                 udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 879                 goto send;
 880 
 881         } else
 882                 csum = udp_csum(skb);
 883 
 884         /* add protocol-dependent pseudo-header */
 885         uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 886                                       sk->sk_protocol, csum);
 887         if (uh->check == 0)
 888                 uh->check = CSUM_MANGLED_0;
 889 
 890 send:
 891         err = ip_send_skb(sock_net(sk), skb);
 892         if (err) {
 893                 if (err == -ENOBUFS && !inet->recverr) {
 894                         UDP_INC_STATS(sock_net(sk),
 895                                       UDP_MIB_SNDBUFERRORS, is_udplite);
 896                         err = 0;
 897                 }
 898         } else
 899                 UDP_INC_STATS(sock_net(sk),
 900                               UDP_MIB_OUTDATAGRAMS, is_udplite);
 901         return err;
 902 }
 903 
 904 /*
 905  * Push out all pending data as one UDP datagram. Socket is locked.
 906  */
 907 int udp_push_pending_frames(struct sock *sk)
 908 {
 909         struct udp_sock  *up = udp_sk(sk);
 910         struct inet_sock *inet = inet_sk(sk);
 911         struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 912         struct sk_buff *skb;
 913         int err = 0;
 914 
 915         skb = ip_finish_skb(sk, fl4);
 916         if (!skb)
 917                 goto out;
 918 
 919         err = udp_send_skb(skb, fl4, &inet->cork.base);
 920 
 921 out:
 922         up->len = 0;
 923         up->pending = 0;
 924         return err;
 925 }
 926 EXPORT_SYMBOL(udp_push_pending_frames);
 927 
 928 static int __udp_cmsg_send(struct cmsghdr *cmsg, u16 *gso_size)
 929 {
 930         switch (cmsg->cmsg_type) {
 931         case UDP_SEGMENT:
 932                 if (cmsg->cmsg_len != CMSG_LEN(sizeof(__u16)))
 933                         return -EINVAL;
 934                 *gso_size = *(__u16 *)CMSG_DATA(cmsg);
 935                 return 0;
 936         default:
 937                 return -EINVAL;
 938         }
 939 }
 940 
 941 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size)
 942 {
 943         struct cmsghdr *cmsg;
 944         bool need_ip = false;
 945         int err;
 946 
 947         for_each_cmsghdr(cmsg, msg) {
 948                 if (!CMSG_OK(msg, cmsg))
 949                         return -EINVAL;
 950 
 951                 if (cmsg->cmsg_level != SOL_UDP) {
 952                         need_ip = true;
 953                         continue;
 954                 }
 955 
 956                 err = __udp_cmsg_send(cmsg, gso_size);
 957                 if (err)
 958                         return err;
 959         }
 960 
 961         return need_ip;
 962 }
 963 EXPORT_SYMBOL_GPL(udp_cmsg_send);
 964 
 965 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len)
 966 {
 967         struct inet_sock *inet = inet_sk(sk);
 968         struct udp_sock *up = udp_sk(sk);
 969         DECLARE_SOCKADDR(struct sockaddr_in *, usin, msg->msg_name);
 970         struct flowi4 fl4_stack;
 971         struct flowi4 *fl4;
 972         int ulen = len;
 973         struct ipcm_cookie ipc;
 974         struct rtable *rt = NULL;
 975         int free = 0;
 976         int connected = 0;
 977         __be32 daddr, faddr, saddr;
 978         __be16 dport;
 979         u8  tos;
 980         int err, is_udplite = IS_UDPLITE(sk);
 981         int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
 982         int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
 983         struct sk_buff *skb;
 984         struct ip_options_data opt_copy;
 985 
 986         if (len > 0xFFFF)
 987                 return -EMSGSIZE;
 988 
 989         /*
 990          *      Check the flags.
 991          */
 992 
 993         if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
 994                 return -EOPNOTSUPP;
 995 
 996         getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
 997 
 998         fl4 = &inet->cork.fl.u.ip4;
 999         if (up->pending) {
1000                 /*
1001                  * There are pending frames.
1002                  * The socket lock must be held while it's corked.
1003                  */
1004                 lock_sock(sk);
1005                 if (likely(up->pending)) {
1006                         if (unlikely(up->pending != AF_INET)) {
1007                                 release_sock(sk);
1008                                 return -EINVAL;
1009                         }
1010                         goto do_append_data;
1011                 }
1012                 release_sock(sk);
1013         }
1014         ulen += sizeof(struct udphdr);
1015 
1016         /*
1017          *      Get and verify the address.
1018          */
1019         if (usin) {
1020                 if (msg->msg_namelen < sizeof(*usin))
1021                         return -EINVAL;
1022                 if (usin->sin_family != AF_INET) {
1023                         if (usin->sin_family != AF_UNSPEC)
1024                                 return -EAFNOSUPPORT;
1025                 }
1026 
1027                 daddr = usin->sin_addr.s_addr;
1028                 dport = usin->sin_port;
1029                 if (dport == 0)
1030                         return -EINVAL;
1031         } else {
1032                 if (sk->sk_state != TCP_ESTABLISHED)
1033                         return -EDESTADDRREQ;
1034                 daddr = inet->inet_daddr;
1035                 dport = inet->inet_dport;
1036                 /* Open fast path for connected socket.
1037                    Route will not be used, if at least one option is set.
1038                  */
1039                 connected = 1;
1040         }
1041 
1042         ipcm_init_sk(&ipc, inet);
1043         ipc.gso_size = up->gso_size;
1044 
1045         if (msg->msg_controllen) {
1046                 err = udp_cmsg_send(sk, msg, &ipc.gso_size);
1047                 if (err > 0)
1048                         err = ip_cmsg_send(sk, msg, &ipc,
1049                                            sk->sk_family == AF_INET6);
1050                 if (unlikely(err < 0)) {
1051                         kfree(ipc.opt);
1052                         return err;
1053                 }
1054                 if (ipc.opt)
1055                         free = 1;
1056                 connected = 0;
1057         }
1058         if (!ipc.opt) {
1059                 struct ip_options_rcu *inet_opt;
1060 
1061                 rcu_read_lock();
1062                 inet_opt = rcu_dereference(inet->inet_opt);
1063                 if (inet_opt) {
1064                         memcpy(&opt_copy, inet_opt,
1065                                sizeof(*inet_opt) + inet_opt->opt.optlen);
1066                         ipc.opt = &opt_copy.opt;
1067                 }
1068                 rcu_read_unlock();
1069         }
1070 
1071         if (cgroup_bpf_enabled && !connected) {
1072                 err = BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk,
1073                                             (struct sockaddr *)usin, &ipc.addr);
1074                 if (err)
1075                         goto out_free;
1076                 if (usin) {
1077                         if (usin->sin_port == 0) {
1078                                 /* BPF program set invalid port. Reject it. */
1079                                 err = -EINVAL;
1080                                 goto out_free;
1081                         }
1082                         daddr = usin->sin_addr.s_addr;
1083                         dport = usin->sin_port;
1084                 }
1085         }
1086 
1087         saddr = ipc.addr;
1088         ipc.addr = faddr = daddr;
1089 
1090         if (ipc.opt && ipc.opt->opt.srr) {
1091                 if (!daddr) {
1092                         err = -EINVAL;
1093                         goto out_free;
1094                 }
1095                 faddr = ipc.opt->opt.faddr;
1096                 connected = 0;
1097         }
1098         tos = get_rttos(&ipc, inet);
1099         if (sock_flag(sk, SOCK_LOCALROUTE) ||
1100             (msg->msg_flags & MSG_DONTROUTE) ||
1101             (ipc.opt && ipc.opt->opt.is_strictroute)) {
1102                 tos |= RTO_ONLINK;
1103                 connected = 0;
1104         }
1105 
1106         if (ipv4_is_multicast(daddr)) {
1107                 if (!ipc.oif || netif_index_is_l3_master(sock_net(sk), ipc.oif))
1108                         ipc.oif = inet->mc_index;
1109                 if (!saddr)
1110                         saddr = inet->mc_addr;
1111                 connected = 0;
1112         } else if (!ipc.oif) {
1113                 ipc.oif = inet->uc_index;
1114         } else if (ipv4_is_lbcast(daddr) && inet->uc_index) {
1115                 /* oif is set, packet is to local broadcast and
1116                  * and uc_index is set. oif is most likely set
1117                  * by sk_bound_dev_if. If uc_index != oif check if the
1118                  * oif is an L3 master and uc_index is an L3 slave.
1119                  * If so, we want to allow the send using the uc_index.
1120                  */
1121                 if (ipc.oif != inet->uc_index &&
1122                     ipc.oif == l3mdev_master_ifindex_by_index(sock_net(sk),
1123                                                               inet->uc_index)) {
1124                         ipc.oif = inet->uc_index;
1125                 }
1126         }
1127 
1128         if (connected)
1129                 rt = (struct rtable *)sk_dst_check(sk, 0);
1130 
1131         if (!rt) {
1132                 struct net *net = sock_net(sk);
1133                 __u8 flow_flags = inet_sk_flowi_flags(sk);
1134 
1135                 fl4 = &fl4_stack;
1136 
1137                 flowi4_init_output(fl4, ipc.oif, ipc.sockc.mark, tos,
1138                                    RT_SCOPE_UNIVERSE, sk->sk_protocol,
1139                                    flow_flags,
1140                                    faddr, saddr, dport, inet->inet_sport,
1141                                    sk->sk_uid);
1142 
1143                 security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
1144                 rt = ip_route_output_flow(net, fl4, sk);
1145                 if (IS_ERR(rt)) {
1146                         err = PTR_ERR(rt);
1147                         rt = NULL;
1148                         if (err == -ENETUNREACH)
1149                                 IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
1150                         goto out;
1151                 }
1152 
1153                 err = -EACCES;
1154                 if ((rt->rt_flags & RTCF_BROADCAST) &&
1155                     !sock_flag(sk, SOCK_BROADCAST))
1156                         goto out;
1157                 if (connected)
1158                         sk_dst_set(sk, dst_clone(&rt->dst));
1159         }
1160 
1161         if (msg->msg_flags&MSG_CONFIRM)
1162                 goto do_confirm;
1163 back_from_confirm:
1164 
1165         saddr = fl4->saddr;
1166         if (!ipc.addr)
1167                 daddr = ipc.addr = fl4->daddr;
1168 
1169         /* Lockless fast path for the non-corking case. */
1170         if (!corkreq) {
1171                 struct inet_cork cork;
1172 
1173                 skb = ip_make_skb(sk, fl4, getfrag, msg, ulen,
1174                                   sizeof(struct udphdr), &ipc, &rt,
1175                                   &cork, msg->msg_flags);
1176                 err = PTR_ERR(skb);
1177                 if (!IS_ERR_OR_NULL(skb))
1178                         err = udp_send_skb(skb, fl4, &cork);
1179                 goto out;
1180         }
1181 
1182         lock_sock(sk);
1183         if (unlikely(up->pending)) {
1184                 /* The socket is already corked while preparing it. */
1185                 /* ... which is an evident application bug. --ANK */
1186                 release_sock(sk);
1187 
1188                 net_dbg_ratelimited("socket already corked\n");
1189                 err = -EINVAL;
1190                 goto out;
1191         }
1192         /*
1193          *      Now cork the socket to pend data.
1194          */
1195         fl4 = &inet->cork.fl.u.ip4;
1196         fl4->daddr = daddr;
1197         fl4->saddr = saddr;
1198         fl4->fl4_dport = dport;
1199         fl4->fl4_sport = inet->inet_sport;
1200         up->pending = AF_INET;
1201 
1202 do_append_data:
1203         up->len += ulen;
1204         err = ip_append_data(sk, fl4, getfrag, msg, ulen,
1205                              sizeof(struct udphdr), &ipc, &rt,
1206                              corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1207         if (err)
1208                 udp_flush_pending_frames(sk);
1209         else if (!corkreq)
1210                 err = udp_push_pending_frames(sk);
1211         else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1212                 up->pending = 0;
1213         release_sock(sk);
1214 
1215 out:
1216         ip_rt_put(rt);
1217 out_free:
1218         if (free)
1219                 kfree(ipc.opt);
1220         if (!err)
1221                 return len;
1222         /*
1223          * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1224          * ENOBUFS might not be good (it's not tunable per se), but otherwise
1225          * we don't have a good statistic (IpOutDiscards but it can be too many
1226          * things).  We could add another new stat but at least for now that
1227          * seems like overkill.
1228          */
1229         if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1230                 UDP_INC_STATS(sock_net(sk),
1231                               UDP_MIB_SNDBUFERRORS, is_udplite);
1232         }
1233         return err;
1234 
1235 do_confirm:
1236         if (msg->msg_flags & MSG_PROBE)
1237                 dst_confirm_neigh(&rt->dst, &fl4->daddr);
1238         if (!(msg->msg_flags&MSG_PROBE) || len)
1239                 goto back_from_confirm;
1240         err = 0;
1241         goto out;
1242 }
1243 EXPORT_SYMBOL(udp_sendmsg);
1244 
1245 int udp_sendpage(struct sock *sk, struct page *page, int offset,
1246                  size_t size, int flags)
1247 {
1248         struct inet_sock *inet = inet_sk(sk);
1249         struct udp_sock *up = udp_sk(sk);
1250         int ret;
1251 
1252         if (flags & MSG_SENDPAGE_NOTLAST)
1253                 flags |= MSG_MORE;
1254 
1255         if (!up->pending) {
1256                 struct msghdr msg = {   .msg_flags = flags|MSG_MORE };
1257 
1258                 /* Call udp_sendmsg to specify destination address which
1259                  * sendpage interface can't pass.
1260                  * This will succeed only when the socket is connected.
1261                  */
1262                 ret = udp_sendmsg(sk, &msg, 0);
1263                 if (ret < 0)
1264                         return ret;
1265         }
1266 
1267         lock_sock(sk);
1268 
1269         if (unlikely(!up->pending)) {
1270                 release_sock(sk);
1271 
1272                 net_dbg_ratelimited("cork failed\n");
1273                 return -EINVAL;
1274         }
1275 
1276         ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1277                              page, offset, size, flags);
1278         if (ret == -EOPNOTSUPP) {
1279                 release_sock(sk);
1280                 return sock_no_sendpage(sk->sk_socket, page, offset,
1281                                         size, flags);
1282         }
1283         if (ret < 0) {
1284                 udp_flush_pending_frames(sk);
1285                 goto out;
1286         }
1287 
1288         up->len += size;
1289         if (!(up->corkflag || (flags&MSG_MORE)))
1290                 ret = udp_push_pending_frames(sk);
1291         if (!ret)
1292                 ret = size;
1293 out:
1294         release_sock(sk);
1295         return ret;
1296 }
1297 
1298 #define UDP_SKB_IS_STATELESS 0x80000000
1299 
1300 /* all head states (dst, sk, nf conntrack) except skb extensions are
1301  * cleared by udp_rcv().
1302  *
1303  * We need to preserve secpath, if present, to eventually process
1304  * IP_CMSG_PASSSEC at recvmsg() time.
1305  *
1306  * Other extensions can be cleared.
1307  */
1308 static bool udp_try_make_stateless(struct sk_buff *skb)
1309 {
1310         if (!skb_has_extensions(skb))
1311                 return true;
1312 
1313         if (!secpath_exists(skb)) {
1314                 skb_ext_reset(skb);
1315                 return true;
1316         }
1317 
1318         return false;
1319 }
1320 
1321 static void udp_set_dev_scratch(struct sk_buff *skb)
1322 {
1323         struct udp_dev_scratch *scratch = udp_skb_scratch(skb);
1324 
1325         BUILD_BUG_ON(sizeof(struct udp_dev_scratch) > sizeof(long));
1326         scratch->_tsize_state = skb->truesize;
1327 #if BITS_PER_LONG == 64
1328         scratch->len = skb->len;
1329         scratch->csum_unnecessary = !!skb_csum_unnecessary(skb);
1330         scratch->is_linear = !skb_is_nonlinear(skb);
1331 #endif
1332         if (udp_try_make_stateless(skb))
1333                 scratch->_tsize_state |= UDP_SKB_IS_STATELESS;
1334 }
1335 
1336 static void udp_skb_csum_unnecessary_set(struct sk_buff *skb)
1337 {
1338         /* We come here after udp_lib_checksum_complete() returned 0.
1339          * This means that __skb_checksum_complete() might have
1340          * set skb->csum_valid to 1.
1341          * On 64bit platforms, we can set csum_unnecessary
1342          * to true, but only if the skb is not shared.
1343          */
1344 #if BITS_PER_LONG == 64
1345         if (!skb_shared(skb))
1346                 udp_skb_scratch(skb)->csum_unnecessary = true;
1347 #endif
1348 }
1349 
1350 static int udp_skb_truesize(struct sk_buff *skb)
1351 {
1352         return udp_skb_scratch(skb)->_tsize_state & ~UDP_SKB_IS_STATELESS;
1353 }
1354 
1355 static bool udp_skb_has_head_state(struct sk_buff *skb)
1356 {
1357         return !(udp_skb_scratch(skb)->_tsize_state & UDP_SKB_IS_STATELESS);
1358 }
1359 
1360 /* fully reclaim rmem/fwd memory allocated for skb */
1361 static void udp_rmem_release(struct sock *sk, int size, int partial,
1362                              bool rx_queue_lock_held)
1363 {
1364         struct udp_sock *up = udp_sk(sk);
1365         struct sk_buff_head *sk_queue;
1366         int amt;
1367 
1368         if (likely(partial)) {
1369                 up->forward_deficit += size;
1370                 size = up->forward_deficit;
1371                 if (size < (sk->sk_rcvbuf >> 2) &&
1372                     !skb_queue_empty(&up->reader_queue))
1373                         return;
1374         } else {
1375                 size += up->forward_deficit;
1376         }
1377         up->forward_deficit = 0;
1378 
1379         /* acquire the sk_receive_queue for fwd allocated memory scheduling,
1380          * if the called don't held it already
1381          */
1382         sk_queue = &sk->sk_receive_queue;
1383         if (!rx_queue_lock_held)
1384                 spin_lock(&sk_queue->lock);
1385 
1386 
1387         sk->sk_forward_alloc += size;
1388         amt = (sk->sk_forward_alloc - partial) & ~(SK_MEM_QUANTUM - 1);
1389         sk->sk_forward_alloc -= amt;
1390 
1391         if (amt)
1392                 __sk_mem_reduce_allocated(sk, amt >> SK_MEM_QUANTUM_SHIFT);
1393 
1394         atomic_sub(size, &sk->sk_rmem_alloc);
1395 
1396         /* this can save us from acquiring the rx queue lock on next receive */
1397         skb_queue_splice_tail_init(sk_queue, &up->reader_queue);
1398 
1399         if (!rx_queue_lock_held)
1400                 spin_unlock(&sk_queue->lock);
1401 }
1402 
1403 /* Note: called with reader_queue.lock held.
1404  * Instead of using skb->truesize here, find a copy of it in skb->dev_scratch
1405  * This avoids a cache line miss while receive_queue lock is held.
1406  * Look at __udp_enqueue_schedule_skb() to find where this copy is done.
1407  */
1408 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb)
1409 {
1410         prefetch(&skb->data);
1411         udp_rmem_release(sk, udp_skb_truesize(skb), 1, false);
1412 }
1413 EXPORT_SYMBOL(udp_skb_destructor);
1414 
1415 /* as above, but the caller held the rx queue lock, too */
1416 static void udp_skb_dtor_locked(struct sock *sk, struct sk_buff *skb)
1417 {
1418         prefetch(&skb->data);
1419         udp_rmem_release(sk, udp_skb_truesize(skb), 1, true);
1420 }
1421 
1422 /* Idea of busylocks is to let producers grab an extra spinlock
1423  * to relieve pressure on the receive_queue spinlock shared by consumer.
1424  * Under flood, this means that only one producer can be in line
1425  * trying to acquire the receive_queue spinlock.
1426  * These busylock can be allocated on a per cpu manner, instead of a
1427  * per socket one (that would consume a cache line per socket)
1428  */
1429 static int udp_busylocks_log __read_mostly;
1430 static spinlock_t *udp_busylocks __read_mostly;
1431 
1432 static spinlock_t *busylock_acquire(void *ptr)
1433 {
1434         spinlock_t *busy;
1435 
1436         busy = udp_busylocks + hash_ptr(ptr, udp_busylocks_log);
1437         spin_lock(busy);
1438         return busy;
1439 }
1440 
1441 static void busylock_release(spinlock_t *busy)
1442 {
1443         if (busy)
1444                 spin_unlock(busy);
1445 }
1446 
1447 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb)
1448 {
1449         struct sk_buff_head *list = &sk->sk_receive_queue;
1450         int rmem, delta, amt, err = -ENOMEM;
1451         spinlock_t *busy = NULL;
1452         int size;
1453 
1454         /* try to avoid the costly atomic add/sub pair when the receive
1455          * queue is full; always allow at least a packet
1456          */
1457         rmem = atomic_read(&sk->sk_rmem_alloc);
1458         if (rmem > sk->sk_rcvbuf)
1459                 goto drop;
1460 
1461         /* Under mem pressure, it might be helpful to help udp_recvmsg()
1462          * having linear skbs :
1463          * - Reduce memory overhead and thus increase receive queue capacity
1464          * - Less cache line misses at copyout() time
1465          * - Less work at consume_skb() (less alien page frag freeing)
1466          */
1467         if (rmem > (sk->sk_rcvbuf >> 1)) {
1468                 skb_condense(skb);
1469 
1470                 busy = busylock_acquire(sk);
1471         }
1472         size = skb->truesize;
1473         udp_set_dev_scratch(skb);
1474 
1475         /* we drop only if the receive buf is full and the receive
1476          * queue contains some other skb
1477          */
1478         rmem = atomic_add_return(size, &sk->sk_rmem_alloc);
1479         if (rmem > (size + (unsigned int)sk->sk_rcvbuf))
1480                 goto uncharge_drop;
1481 
1482         spin_lock(&list->lock);
1483         if (size >= sk->sk_forward_alloc) {
1484                 amt = sk_mem_pages(size);
1485                 delta = amt << SK_MEM_QUANTUM_SHIFT;
1486                 if (!__sk_mem_raise_allocated(sk, delta, amt, SK_MEM_RECV)) {
1487                         err = -ENOBUFS;
1488                         spin_unlock(&list->lock);
1489                         goto uncharge_drop;
1490                 }
1491 
1492                 sk->sk_forward_alloc += delta;
1493         }
1494 
1495         sk->sk_forward_alloc -= size;
1496 
1497         /* no need to setup a destructor, we will explicitly release the
1498          * forward allocated memory on dequeue
1499          */
1500         sock_skb_set_dropcount(sk, skb);
1501 
1502         __skb_queue_tail(list, skb);
1503         spin_unlock(&list->lock);
1504 
1505         if (!sock_flag(sk, SOCK_DEAD))
1506                 sk->sk_data_ready(sk);
1507 
1508         busylock_release(busy);
1509         return 0;
1510 
1511 uncharge_drop:
1512         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1513 
1514 drop:
1515         atomic_inc(&sk->sk_drops);
1516         busylock_release(busy);
1517         return err;
1518 }
1519 EXPORT_SYMBOL_GPL(__udp_enqueue_schedule_skb);
1520 
1521 void udp_destruct_sock(struct sock *sk)
1522 {
1523         /* reclaim completely the forward allocated memory */
1524         struct udp_sock *up = udp_sk(sk);
1525         unsigned int total = 0;
1526         struct sk_buff *skb;
1527 
1528         skb_queue_splice_tail_init(&sk->sk_receive_queue, &up->reader_queue);
1529         while ((skb = __skb_dequeue(&up->reader_queue)) != NULL) {
1530                 total += skb->truesize;
1531                 kfree_skb(skb);
1532         }
1533         udp_rmem_release(sk, total, 0, true);
1534 
1535         inet_sock_destruct(sk);
1536 }
1537 EXPORT_SYMBOL_GPL(udp_destruct_sock);
1538 
1539 int udp_init_sock(struct sock *sk)
1540 {
1541         skb_queue_head_init(&udp_sk(sk)->reader_queue);
1542         sk->sk_destruct = udp_destruct_sock;
1543         return 0;
1544 }
1545 EXPORT_SYMBOL_GPL(udp_init_sock);
1546 
1547 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len)
1548 {
1549         if (unlikely(READ_ONCE(sk->sk_peek_off) >= 0)) {
1550                 bool slow = lock_sock_fast(sk);
1551 
1552                 sk_peek_offset_bwd(sk, len);
1553                 unlock_sock_fast(sk, slow);
1554         }
1555 
1556         if (!skb_unref(skb))
1557                 return;
1558 
1559         /* In the more common cases we cleared the head states previously,
1560          * see __udp_queue_rcv_skb().
1561          */
1562         if (unlikely(udp_skb_has_head_state(skb)))
1563                 skb_release_head_state(skb);
1564         __consume_stateless_skb(skb);
1565 }
1566 EXPORT_SYMBOL_GPL(skb_consume_udp);
1567 
1568 static struct sk_buff *__first_packet_length(struct sock *sk,
1569                                              struct sk_buff_head *rcvq,
1570                                              int *total)
1571 {
1572         struct sk_buff *skb;
1573 
1574         while ((skb = skb_peek(rcvq)) != NULL) {
1575                 if (udp_lib_checksum_complete(skb)) {
1576                         __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS,
1577                                         IS_UDPLITE(sk));
1578                         __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS,
1579                                         IS_UDPLITE(sk));
1580                         atomic_inc(&sk->sk_drops);
1581                         __skb_unlink(skb, rcvq);
1582                         *total += skb->truesize;
1583                         kfree_skb(skb);
1584                 } else {
1585                         udp_skb_csum_unnecessary_set(skb);
1586                         break;
1587                 }
1588         }
1589         return skb;
1590 }
1591 
1592 /**
1593  *      first_packet_length     - return length of first packet in receive queue
1594  *      @sk: socket
1595  *
1596  *      Drops all bad checksum frames, until a valid one is found.
1597  *      Returns the length of found skb, or -1 if none is found.
1598  */
1599 static int first_packet_length(struct sock *sk)
1600 {
1601         struct sk_buff_head *rcvq = &udp_sk(sk)->reader_queue;
1602         struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1603         struct sk_buff *skb;
1604         int total = 0;
1605         int res;
1606 
1607         spin_lock_bh(&rcvq->lock);
1608         skb = __first_packet_length(sk, rcvq, &total);
1609         if (!skb && !skb_queue_empty_lockless(sk_queue)) {
1610                 spin_lock(&sk_queue->lock);
1611                 skb_queue_splice_tail_init(sk_queue, rcvq);
1612                 spin_unlock(&sk_queue->lock);
1613 
1614                 skb = __first_packet_length(sk, rcvq, &total);
1615         }
1616         res = skb ? skb->len : -1;
1617         if (total)
1618                 udp_rmem_release(sk, total, 1, false);
1619         spin_unlock_bh(&rcvq->lock);
1620         return res;
1621 }
1622 
1623 /*
1624  *      IOCTL requests applicable to the UDP protocol
1625  */
1626 
1627 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1628 {
1629         switch (cmd) {
1630         case SIOCOUTQ:
1631         {
1632                 int amount = sk_wmem_alloc_get(sk);
1633 
1634                 return put_user(amount, (int __user *)arg);
1635         }
1636 
1637         case SIOCINQ:
1638         {
1639                 int amount = max_t(int, 0, first_packet_length(sk));
1640 
1641                 return put_user(amount, (int __user *)arg);
1642         }
1643 
1644         default:
1645                 return -ENOIOCTLCMD;
1646         }
1647 
1648         return 0;
1649 }
1650 EXPORT_SYMBOL(udp_ioctl);
1651 
1652 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
1653                                int noblock, int *off, int *err)
1654 {
1655         struct sk_buff_head *sk_queue = &sk->sk_receive_queue;
1656         struct sk_buff_head *queue;
1657         struct sk_buff *last;
1658         long timeo;
1659         int error;
1660 
1661         queue = &udp_sk(sk)->reader_queue;
1662         flags |= noblock ? MSG_DONTWAIT : 0;
1663         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1664         do {
1665                 struct sk_buff *skb;
1666 
1667                 error = sock_error(sk);
1668                 if (error)
1669                         break;
1670 
1671                 error = -EAGAIN;
1672                 do {
1673                         spin_lock_bh(&queue->lock);
1674                         skb = __skb_try_recv_from_queue(sk, queue, flags,
1675                                                         udp_skb_destructor,
1676                                                         off, err, &last);
1677                         if (skb) {
1678                                 spin_unlock_bh(&queue->lock);
1679                                 return skb;
1680                         }
1681 
1682                         if (skb_queue_empty_lockless(sk_queue)) {
1683                                 spin_unlock_bh(&queue->lock);
1684                                 goto busy_check;
1685                         }
1686 
1687                         /* refill the reader queue and walk it again
1688                          * keep both queues locked to avoid re-acquiring
1689                          * the sk_receive_queue lock if fwd memory scheduling
1690                          * is needed.
1691                          */
1692                         spin_lock(&sk_queue->lock);
1693                         skb_queue_splice_tail_init(sk_queue, queue);
1694 
1695                         skb = __skb_try_recv_from_queue(sk, queue, flags,
1696                                                         udp_skb_dtor_locked,
1697                                                         off, err, &last);
1698                         spin_unlock(&sk_queue->lock);
1699                         spin_unlock_bh(&queue->lock);
1700                         if (skb)
1701                                 return skb;
1702 
1703 busy_check:
1704                         if (!sk_can_busy_loop(sk))
1705                                 break;
1706 
1707                         sk_busy_loop(sk, flags & MSG_DONTWAIT);
1708                 } while (!skb_queue_empty_lockless(sk_queue));
1709 
1710                 /* sk_queue is empty, reader_queue may contain peeked packets */
1711         } while (timeo &&
1712                  !__skb_wait_for_more_packets(sk, &error, &timeo,
1713                                               (struct sk_buff *)sk_queue));
1714 
1715         *err = error;
1716         return NULL;
1717 }
1718 EXPORT_SYMBOL(__skb_recv_udp);
1719 
1720 /*
1721  *      This should be easy, if there is something there we
1722  *      return it, otherwise we block.
1723  */
1724 
1725 int udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int noblock,
1726                 int flags, int *addr_len)
1727 {
1728         struct inet_sock *inet = inet_sk(sk);
1729         DECLARE_SOCKADDR(struct sockaddr_in *, sin, msg->msg_name);
1730         struct sk_buff *skb;
1731         unsigned int ulen, copied;
1732         int off, err, peeking = flags & MSG_PEEK;
1733         int is_udplite = IS_UDPLITE(sk);
1734         bool checksum_valid = false;
1735 
1736         if (flags & MSG_ERRQUEUE)
1737                 return ip_recv_error(sk, msg, len, addr_len);
1738 
1739 try_again:
1740         off = sk_peek_offset(sk, flags);
1741         skb = __skb_recv_udp(sk, flags, noblock, &off, &err);
1742         if (!skb)
1743                 return err;
1744 
1745         ulen = udp_skb_len(skb);
1746         copied = len;
1747         if (copied > ulen - off)
1748                 copied = ulen - off;
1749         else if (copied < ulen)
1750                 msg->msg_flags |= MSG_TRUNC;
1751 
1752         /*
1753          * If checksum is needed at all, try to do it while copying the
1754          * data.  If the data is truncated, or if we only want a partial
1755          * coverage checksum (UDP-Lite), do it before the copy.
1756          */
1757 
1758         if (copied < ulen || peeking ||
1759             (is_udplite && UDP_SKB_CB(skb)->partial_cov)) {
1760                 checksum_valid = udp_skb_csum_unnecessary(skb) ||
1761                                 !__udp_lib_checksum_complete(skb);
1762                 if (!checksum_valid)
1763                         goto csum_copy_err;
1764         }
1765 
1766         if (checksum_valid || udp_skb_csum_unnecessary(skb)) {
1767                 if (udp_skb_is_linear(skb))
1768                         err = copy_linear_skb(skb, copied, off, &msg->msg_iter);
1769                 else
1770                         err = skb_copy_datagram_msg(skb, off, msg, copied);
1771         } else {
1772                 err = skb_copy_and_csum_datagram_msg(skb, off, msg);
1773 
1774                 if (err == -EINVAL)
1775                         goto csum_copy_err;
1776         }
1777 
1778         if (unlikely(err)) {
1779                 if (!peeking) {
1780                         atomic_inc(&sk->sk_drops);
1781                         UDP_INC_STATS(sock_net(sk),
1782                                       UDP_MIB_INERRORS, is_udplite);
1783                 }
1784                 kfree_skb(skb);
1785                 return err;
1786         }
1787 
1788         if (!peeking)
1789                 UDP_INC_STATS(sock_net(sk),
1790                               UDP_MIB_INDATAGRAMS, is_udplite);
1791 
1792         sock_recv_ts_and_drops(msg, sk, skb);
1793 
1794         /* Copy the address. */
1795         if (sin) {
1796                 sin->sin_family = AF_INET;
1797                 sin->sin_port = udp_hdr(skb)->source;
1798                 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1799                 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1800                 *addr_len = sizeof(*sin);
1801 
1802                 if (cgroup_bpf_enabled)
1803                         BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk,
1804                                                         (struct sockaddr *)sin);
1805         }
1806 
1807         if (udp_sk(sk)->gro_enabled)
1808                 udp_cmsg_recv(msg, sk, skb);
1809 
1810         if (inet->cmsg_flags)
1811                 ip_cmsg_recv_offset(msg, sk, skb, sizeof(struct udphdr), off);
1812 
1813         err = copied;
1814         if (flags & MSG_TRUNC)
1815                 err = ulen;
1816 
1817         skb_consume_udp(sk, skb, peeking ? -err : err);
1818         return err;
1819 
1820 csum_copy_err:
1821         if (!__sk_queue_drop_skb(sk, &udp_sk(sk)->reader_queue, skb, flags,
1822                                  udp_skb_destructor)) {
1823                 UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
1824                 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1825         }
1826         kfree_skb(skb);
1827 
1828         /* starting over for a new packet, but check if we need to yield */
1829         cond_resched();
1830         msg->msg_flags &= ~MSG_TRUNC;
1831         goto try_again;
1832 }
1833 
1834 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
1835 {
1836         /* This check is replicated from __ip4_datagram_connect() and
1837          * intended to prevent BPF program called below from accessing bytes
1838          * that are out of the bound specified by user in addr_len.
1839          */
1840         if (addr_len < sizeof(struct sockaddr_in))
1841                 return -EINVAL;
1842 
1843         return BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr);
1844 }
1845 EXPORT_SYMBOL(udp_pre_connect);
1846 
1847 int __udp_disconnect(struct sock *sk, int flags)
1848 {
1849         struct inet_sock *inet = inet_sk(sk);
1850         /*
1851          *      1003.1g - break association.
1852          */
1853 
1854         sk->sk_state = TCP_CLOSE;
1855         inet->inet_daddr = 0;
1856         inet->inet_dport = 0;
1857         sock_rps_reset_rxhash(sk);
1858         sk->sk_bound_dev_if = 0;
1859         if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) {
1860                 inet_reset_saddr(sk);
1861                 if (sk->sk_prot->rehash &&
1862                     (sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1863                         sk->sk_prot->rehash(sk);
1864         }
1865 
1866         if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1867                 sk->sk_prot->unhash(sk);
1868                 inet->inet_sport = 0;
1869         }
1870         sk_dst_reset(sk);
1871         return 0;
1872 }
1873 EXPORT_SYMBOL(__udp_disconnect);
1874 
1875 int udp_disconnect(struct sock *sk, int flags)
1876 {
1877         lock_sock(sk);
1878         __udp_disconnect(sk, flags);
1879         release_sock(sk);
1880         return 0;
1881 }
1882 EXPORT_SYMBOL(udp_disconnect);
1883 
1884 void udp_lib_unhash(struct sock *sk)
1885 {
1886         if (sk_hashed(sk)) {
1887                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1888                 struct udp_hslot *hslot, *hslot2;
1889 
1890                 hslot  = udp_hashslot(udptable, sock_net(sk),
1891                                       udp_sk(sk)->udp_port_hash);
1892                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1893 
1894                 spin_lock_bh(&hslot->lock);
1895                 if (rcu_access_pointer(sk->sk_reuseport_cb))
1896                         reuseport_detach_sock(sk);
1897                 if (sk_del_node_init_rcu(sk)) {
1898                         hslot->count--;
1899                         inet_sk(sk)->inet_num = 0;
1900                         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1901 
1902                         spin_lock(&hslot2->lock);
1903                         hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1904                         hslot2->count--;
1905                         spin_unlock(&hslot2->lock);
1906                 }
1907                 spin_unlock_bh(&hslot->lock);
1908         }
1909 }
1910 EXPORT_SYMBOL(udp_lib_unhash);
1911 
1912 /*
1913  * inet_rcv_saddr was changed, we must rehash secondary hash
1914  */
1915 void udp_lib_rehash(struct sock *sk, u16 newhash)
1916 {
1917         if (sk_hashed(sk)) {
1918                 struct udp_table *udptable = sk->sk_prot->h.udp_table;
1919                 struct udp_hslot *hslot, *hslot2, *nhslot2;
1920 
1921                 hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1922                 nhslot2 = udp_hashslot2(udptable, newhash);
1923                 udp_sk(sk)->udp_portaddr_hash = newhash;
1924 
1925                 if (hslot2 != nhslot2 ||
1926                     rcu_access_pointer(sk->sk_reuseport_cb)) {
1927                         hslot = udp_hashslot(udptable, sock_net(sk),
1928                                              udp_sk(sk)->udp_port_hash);
1929                         /* we must lock primary chain too */
1930                         spin_lock_bh(&hslot->lock);
1931                         if (rcu_access_pointer(sk->sk_reuseport_cb))
1932                                 reuseport_detach_sock(sk);
1933 
1934                         if (hslot2 != nhslot2) {
1935                                 spin_lock(&hslot2->lock);
1936                                 hlist_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1937                                 hslot2->count--;
1938                                 spin_unlock(&hslot2->lock);
1939 
1940                                 spin_lock(&nhslot2->lock);
1941                                 hlist_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1942                                                          &nhslot2->head);
1943                                 nhslot2->count++;
1944                                 spin_unlock(&nhslot2->lock);
1945                         }
1946 
1947                         spin_unlock_bh(&hslot->lock);
1948                 }
1949         }
1950 }
1951 EXPORT_SYMBOL(udp_lib_rehash);
1952 
1953 void udp_v4_rehash(struct sock *sk)
1954 {
1955         u16 new_hash = ipv4_portaddr_hash(sock_net(sk),
1956                                           inet_sk(sk)->inet_rcv_saddr,
1957                                           inet_sk(sk)->inet_num);
1958         udp_lib_rehash(sk, new_hash);
1959 }
1960 
1961 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1962 {
1963         int rc;
1964 
1965         if (inet_sk(sk)->inet_daddr) {
1966                 sock_rps_save_rxhash(sk, skb);
1967                 sk_mark_napi_id(sk, skb);
1968                 sk_incoming_cpu_update(sk);
1969         } else {
1970                 sk_mark_napi_id_once(sk, skb);
1971         }
1972 
1973         rc = __udp_enqueue_schedule_skb(sk, skb);
1974         if (rc < 0) {
1975                 int is_udplite = IS_UDPLITE(sk);
1976 
1977                 /* Note that an ENOMEM error is charged twice */
1978                 if (rc == -ENOMEM)
1979                         UDP_INC_STATS(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1980                                         is_udplite);
1981                 UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1982                 kfree_skb(skb);
1983                 trace_udp_fail_queue_rcv_skb(rc, sk);
1984                 return -1;
1985         }
1986 
1987         return 0;
1988 }
1989 
1990 /* returns:
1991  *  -1: error
1992  *   0: success
1993  *  >0: "udp encap" protocol resubmission
1994  *
1995  * Note that in the success and error cases, the skb is assumed to
1996  * have either been requeued or freed.
1997  */
1998 static int udp_queue_rcv_one_skb(struct sock *sk, struct sk_buff *skb)
1999 {
2000         struct udp_sock *up = udp_sk(sk);
2001         int is_udplite = IS_UDPLITE(sk);
2002 
2003         /*
2004          *      Charge it to the socket, dropping if the queue is full.
2005          */
2006         if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
2007                 goto drop;
2008         nf_reset_ct(skb);
2009 
2010         if (static_branch_unlikely(&udp_encap_needed_key) && up->encap_type) {
2011                 int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
2012 
2013                 /*
2014                  * This is an encapsulation socket so pass the skb to
2015                  * the socket's udp_encap_rcv() hook. Otherwise, just
2016                  * fall through and pass this up the UDP socket.
2017                  * up->encap_rcv() returns the following value:
2018                  * =0 if skb was successfully passed to the encap
2019                  *    handler or was discarded by it.
2020                  * >0 if skb should be passed on to UDP.
2021                  * <0 if skb should be resubmitted as proto -N
2022                  */
2023 
2024                 /* if we're overly short, let UDP handle it */
2025                 encap_rcv = READ_ONCE(up->encap_rcv);
2026                 if (encap_rcv) {
2027                         int ret;
2028 
2029                         /* Verify checksum before giving to encap */
2030                         if (udp_lib_checksum_complete(skb))
2031                                 goto csum_error;
2032 
2033                         ret = encap_rcv(sk, skb);
2034                         if (ret <= 0) {
2035                                 __UDP_INC_STATS(sock_net(sk),
2036                                                 UDP_MIB_INDATAGRAMS,
2037                                                 is_udplite);
2038                                 return -ret;
2039                         }
2040                 }
2041 
2042                 /* FALLTHROUGH -- it's a UDP Packet */
2043         }
2044 
2045         /*
2046          *      UDP-Lite specific tests, ignored on UDP sockets
2047          */
2048         if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
2049 
2050                 /*
2051                  * MIB statistics other than incrementing the error count are
2052                  * disabled for the following two types of errors: these depend
2053                  * on the application settings, not on the functioning of the
2054                  * protocol stack as such.
2055                  *
2056                  * RFC 3828 here recommends (sec 3.3): "There should also be a
2057                  * way ... to ... at least let the receiving application block
2058                  * delivery of packets with coverage values less than a value
2059                  * provided by the application."
2060                  */
2061                 if (up->pcrlen == 0) {          /* full coverage was set  */
2062                         net_dbg_ratelimited("UDPLite: partial coverage %d while full coverage %d requested\n",
2063                                             UDP_SKB_CB(skb)->cscov, skb->len);
2064                         goto drop;
2065                 }
2066                 /* The next case involves violating the min. coverage requested
2067                  * by the receiver. This is subtle: if receiver wants x and x is
2068                  * greater than the buffersize/MTU then receiver will complain
2069                  * that it wants x while sender emits packets of smaller size y.
2070                  * Therefore the above ...()->partial_cov statement is essential.
2071                  */
2072                 if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
2073                         net_dbg_ratelimited("UDPLite: coverage %d too small, need min %d\n",
2074                                             UDP_SKB_CB(skb)->cscov, up->pcrlen);
2075                         goto drop;
2076                 }
2077         }
2078 
2079         prefetch(&sk->sk_rmem_alloc);
2080         if (rcu_access_pointer(sk->sk_filter) &&
2081             udp_lib_checksum_complete(skb))
2082                         goto csum_error;
2083 
2084         if (sk_filter_trim_cap(sk, skb, sizeof(struct udphdr)))
2085                 goto drop;
2086 
2087         udp_csum_pull_header(skb);
2088 
2089         ipv4_pktinfo_prepare(sk, skb);
2090         return __udp_queue_rcv_skb(sk, skb);
2091 
2092 csum_error:
2093         __UDP_INC_STATS(sock_net(sk), UDP_MIB_CSUMERRORS, is_udplite);
2094 drop:
2095         __UDP_INC_STATS(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
2096         atomic_inc(&sk->sk_drops);
2097         kfree_skb(skb);
2098         return -1;
2099 }
2100 
2101 static int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2102 {
2103         struct sk_buff *next, *segs;
2104         int ret;
2105 
2106         if (likely(!udp_unexpected_gso(sk, skb)))
2107                 return udp_queue_rcv_one_skb(sk, skb);
2108 
2109         BUILD_BUG_ON(sizeof(struct udp_skb_cb) > SKB_SGO_CB_OFFSET);
2110         __skb_push(skb, -skb_mac_offset(skb));
2111         segs = udp_rcv_segment(sk, skb, true);
2112         for (skb = segs; skb; skb = next) {
2113                 next = skb->next;
2114                 __skb_pull(skb, skb_transport_offset(skb));
2115                 ret = udp_queue_rcv_one_skb(sk, skb);
2116                 if (ret > 0)
2117                         ip_protocol_deliver_rcu(dev_net(skb->dev), skb, -ret);
2118         }
2119         return 0;
2120 }
2121 
2122 /* For TCP sockets, sk_rx_dst is protected by socket lock
2123  * For UDP, we use xchg() to guard against concurrent changes.
2124  */
2125 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst)
2126 {
2127         struct dst_entry *old;
2128 
2129         if (dst_hold_safe(dst)) {
2130                 old = xchg(&sk->sk_rx_dst, dst);
2131                 dst_release(old);
2132                 return old != dst;
2133         }
2134         return false;
2135 }
2136 EXPORT_SYMBOL(udp_sk_rx_dst_set);
2137 
2138 /*
2139  *      Multicasts and broadcasts go to each listener.
2140  *
2141  *      Note: called only from the BH handler context.
2142  */
2143 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
2144                                     struct udphdr  *uh,
2145                                     __be32 saddr, __be32 daddr,
2146                                     struct udp_table *udptable,
2147                                     int proto)
2148 {
2149         struct sock *sk, *first = NULL;
2150         unsigned short hnum = ntohs(uh->dest);
2151         struct udp_hslot *hslot = udp_hashslot(udptable, net, hnum);
2152         unsigned int hash2 = 0, hash2_any = 0, use_hash2 = (hslot->count > 10);
2153         unsigned int offset = offsetof(typeof(*sk), sk_node);
2154         int dif = skb->dev->ifindex;
2155         int sdif = inet_sdif(skb);
2156         struct hlist_node *node;
2157         struct sk_buff *nskb;
2158 
2159         if (use_hash2) {
2160                 hash2_any = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum) &
2161                             udptable->mask;
2162                 hash2 = ipv4_portaddr_hash(net, daddr, hnum) & udptable->mask;
2163 start_lookup:
2164                 hslot = &udptable->hash2[hash2];
2165                 offset = offsetof(typeof(*sk), __sk_common.skc_portaddr_node);
2166         }
2167 
2168         sk_for_each_entry_offset_rcu(sk, node, &hslot->head, offset) {
2169                 if (!__udp_is_mcast_sock(net, sk, uh->dest, daddr,
2170                                          uh->source, saddr, dif, sdif, hnum))
2171                         continue;
2172 
2173                 if (!first) {
2174                         first = sk;
2175                         continue;
2176                 }
2177                 nskb = skb_clone(skb, GFP_ATOMIC);
2178 
2179                 if (unlikely(!nskb)) {
2180                         atomic_inc(&sk->sk_drops);
2181                         __UDP_INC_STATS(net, UDP_MIB_RCVBUFERRORS,
2182                                         IS_UDPLITE(sk));
2183                         __UDP_INC_STATS(net, UDP_MIB_INERRORS,
2184                                         IS_UDPLITE(sk));
2185                         continue;
2186                 }
2187                 if (udp_queue_rcv_skb(sk, nskb) > 0)
2188                         consume_skb(nskb);
2189         }
2190 
2191         /* Also lookup *:port if we are using hash2 and haven't done so yet. */
2192         if (use_hash2 && hash2 != hash2_any) {
2193                 hash2 = hash2_any;
2194                 goto start_lookup;
2195         }
2196 
2197         if (first) {
2198                 if (udp_queue_rcv_skb(first, skb) > 0)
2199                         consume_skb(skb);
2200         } else {
2201                 kfree_skb(skb);
2202                 __UDP_INC_STATS(net, UDP_MIB_IGNOREDMULTI,
2203                                 proto == IPPROTO_UDPLITE);
2204         }
2205         return 0;
2206 }
2207 
2208 /* Initialize UDP checksum. If exited with zero value (success),
2209  * CHECKSUM_UNNECESSARY means, that no more checks are required.
2210  * Otherwise, csum completion requires checksumming packet body,
2211  * including udp header and folding it to skb->csum.
2212  */
2213 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
2214                                  int proto)
2215 {
2216         int err;
2217 
2218         UDP_SKB_CB(skb)->partial_cov = 0;
2219         UDP_SKB_CB(skb)->cscov = skb->len;
2220 
2221         if (proto == IPPROTO_UDPLITE) {
2222                 err = udplite_checksum_init(skb, uh);
2223                 if (err)
2224                         return err;
2225 
2226                 if (UDP_SKB_CB(skb)->partial_cov) {
2227                         skb->csum = inet_compute_pseudo(skb, proto);
2228                         return 0;
2229                 }
2230         }
2231 
2232         /* Note, we are only interested in != 0 or == 0, thus the
2233          * force to int.
2234          */
2235         err = (__force int)skb_checksum_init_zero_check(skb, proto, uh->check,
2236                                                         inet_compute_pseudo);
2237         if (err)
2238                 return err;
2239 
2240         if (skb->ip_summed == CHECKSUM_COMPLETE && !skb->csum_valid) {
2241                 /* If SW calculated the value, we know it's bad */
2242                 if (skb->csum_complete_sw)
2243                         return 1;
2244 
2245                 /* HW says the value is bad. Let's validate that.
2246                  * skb->csum is no longer the full packet checksum,
2247                  * so don't treat it as such.
2248                  */
2249                 skb_checksum_complete_unset(skb);
2250         }
2251 
2252         return 0;
2253 }
2254 
2255 /* wrapper for udp_queue_rcv_skb tacking care of csum conversion and
2256  * return code conversion for ip layer consumption
2257  */
2258 static int udp_unicast_rcv_skb(struct sock *sk, struct sk_buff *skb,
2259                                struct udphdr *uh)
2260 {
2261         int ret;
2262 
2263         if (inet_get_convert_csum(sk) && uh->check && !IS_UDPLITE(sk))
2264                 skb_checksum_try_convert(skb, IPPROTO_UDP, inet_compute_pseudo);
2265 
2266         ret = udp_queue_rcv_skb(sk, skb);
2267 
2268         /* a return value > 0 means to resubmit the input, but
2269          * it wants the return to be -protocol, or 0
2270          */
2271         if (ret > 0)
2272                 return -ret;
2273         return 0;
2274 }
2275 
2276 /*
2277  *      All we need to do is get the socket, and then do a checksum.
2278  */
2279 
2280 int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
2281                    int proto)
2282 {
2283         struct sock *sk;
2284         struct udphdr *uh;
2285         unsigned short ulen;
2286         struct rtable *rt = skb_rtable(skb);
2287         __be32 saddr, daddr;
2288         struct net *net = dev_net(skb->dev);
2289 
2290         /*
2291          *  Validate the packet.
2292          */
2293         if (!pskb_may_pull(skb, sizeof(struct udphdr)))
2294                 goto drop;              /* No space for header. */
2295 
2296         uh   = udp_hdr(skb);
2297         ulen = ntohs(uh->len);
2298         saddr = ip_hdr(skb)->saddr;
2299         daddr = ip_hdr(skb)->daddr;
2300 
2301         if (ulen > skb->len)
2302                 goto short_packet;
2303 
2304         if (proto == IPPROTO_UDP) {
2305                 /* UDP validates ulen. */
2306                 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
2307                         goto short_packet;
2308                 uh = udp_hdr(skb);
2309         }
2310 
2311         if (udp4_csum_init(skb, uh, proto))
2312                 goto csum_error;
2313 
2314         sk = skb_steal_sock(skb);
2315         if (sk) {
2316                 struct dst_entry *dst = skb_dst(skb);
2317                 int ret;
2318 
2319                 if (unlikely(sk->sk_rx_dst != dst))
2320                         udp_sk_rx_dst_set(sk, dst);
2321 
2322                 ret = udp_unicast_rcv_skb(sk, skb, uh);
2323                 sock_put(sk);
2324                 return ret;
2325         }
2326 
2327         if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
2328                 return __udp4_lib_mcast_deliver(net, skb, uh,
2329                                                 saddr, daddr, udptable, proto);
2330 
2331         sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
2332         if (sk)
2333                 return udp_unicast_rcv_skb(sk, skb, uh);
2334 
2335         if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
2336                 goto drop;
2337         nf_reset_ct(skb);
2338 
2339         /* No socket. Drop packet silently, if checksum is wrong */
2340         if (udp_lib_checksum_complete(skb))
2341                 goto csum_error;
2342 
2343         __UDP_INC_STATS(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
2344         icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
2345 
2346         /*
2347          * Hmm.  We got an UDP packet to a port to which we
2348          * don't wanna listen.  Ignore it.
2349          */
2350         kfree_skb(skb);
2351         return 0;
2352 
2353 short_packet:
2354         net_dbg_ratelimited("UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
2355                             proto == IPPROTO_UDPLITE ? "Lite" : "",
2356                             &saddr, ntohs(uh->source),
2357                             ulen, skb->len,
2358                             &daddr, ntohs(uh->dest));
2359         goto drop;
2360 
2361 csum_error:
2362         /*
2363          * RFC1122: OK.  Discards the bad packet silently (as far as
2364          * the network is concerned, anyway) as per 4.1.3.4 (MUST).
2365          */
2366         net_dbg_ratelimited("UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
2367                             proto == IPPROTO_UDPLITE ? "Lite" : "",
2368                             &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
2369                             ulen);
2370         __UDP_INC_STATS(net, UDP_MIB_CSUMERRORS, proto == IPPROTO_UDPLITE);
2371 drop:
2372         __UDP_INC_STATS(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
2373         kfree_skb(skb);
2374         return 0;
2375 }
2376 
2377 /* We can only early demux multicast if there is a single matching socket.
2378  * If more than one socket found returns NULL
2379  */
2380 static struct sock *__udp4_lib_mcast_demux_lookup(struct net *net,
2381                                                   __be16 loc_port, __be32 loc_addr,
2382                                                   __be16 rmt_port, __be32 rmt_addr,
2383                                                   int dif, int sdif)
2384 {
2385         struct sock *sk, *result;
2386         unsigned short hnum = ntohs(loc_port);
2387         unsigned int slot = udp_hashfn(net, hnum, udp_table.mask);
2388         struct udp_hslot *hslot = &udp_table.hash[slot];
2389 
2390         /* Do not bother scanning a too big list */
2391         if (hslot->count > 10)
2392                 return NULL;
2393 
2394         result = NULL;
2395         sk_for_each_rcu(sk, &hslot->head) {
2396                 if (__udp_is_mcast_sock(net, sk, loc_port, loc_addr,
2397                                         rmt_port, rmt_addr, dif, sdif, hnum)) {
2398                         if (result)
2399                                 return NULL;
2400                         result = sk;
2401                 }
2402         }
2403 
2404         return result;
2405 }
2406 
2407 /* For unicast we should only early demux connected sockets or we can
2408  * break forwarding setups.  The chains here can be long so only check
2409  * if the first socket is an exact match and if not move on.
2410  */
2411 static struct sock *__udp4_lib_demux_lookup(struct net *net,
2412                                             __be16 loc_port, __be32 loc_addr,
2413                                             __be16 rmt_port, __be32 rmt_addr,
2414                                             int dif, int sdif)
2415 {
2416         unsigned short hnum = ntohs(loc_port);
2417         unsigned int hash2 = ipv4_portaddr_hash(net, loc_addr, hnum);
2418         unsigned int slot2 = hash2 & udp_table.mask;
2419         struct udp_hslot *hslot2 = &udp_table.hash2[slot2];
2420         INET_ADDR_COOKIE(acookie, rmt_addr, loc_addr);
2421         const __portpair ports = INET_COMBINED_PORTS(rmt_port, hnum);
2422         struct sock *sk;
2423 
2424         udp_portaddr_for_each_entry_rcu(sk, &hslot2->head) {
2425                 if (INET_MATCH(sk, net, acookie, rmt_addr,
2426                                loc_addr, ports, dif, sdif))
2427                         return sk;
2428                 /* Only check first socket in chain */
2429                 break;
2430         }
2431         return NULL;
2432 }
2433 
2434 int udp_v4_early_demux(struct sk_buff *skb)
2435 {
2436         struct net *net = dev_net(skb->dev);
2437         struct in_device *in_dev = NULL;
2438         const struct iphdr *iph;
2439         const struct udphdr *uh;
2440         struct sock *sk = NULL;
2441         struct dst_entry *dst;
2442         int dif = skb->dev->ifindex;
2443         int sdif = inet_sdif(skb);
2444         int ours;
2445 
2446         /* validate the packet */
2447         if (!pskb_may_pull(skb, skb_transport_offset(skb) + sizeof(struct udphdr)))
2448                 return 0;
2449 
2450         iph = ip_hdr(skb);
2451         uh = udp_hdr(skb);
2452 
2453         if (skb->pkt_type == PACKET_MULTICAST) {
2454                 in_dev = __in_dev_get_rcu(skb->dev);
2455 
2456                 if (!in_dev)
2457                         return 0;
2458 
2459                 ours = ip_check_mc_rcu(in_dev, iph->daddr, iph->saddr,
2460                                        iph->protocol);
2461                 if (!ours)
2462                         return 0;
2463 
2464                 sk = __udp4_lib_mcast_demux_lookup(net, uh->dest, iph->daddr,
2465                                                    uh->source, iph->saddr,
2466                                                    dif, sdif);
2467         } else if (skb->pkt_type == PACKET_HOST) {
2468                 sk = __udp4_lib_demux_lookup(net, uh->dest, iph->daddr,
2469                                              uh->source, iph->saddr, dif, sdif);
2470         }
2471 
2472         if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
2473                 return 0;
2474 
2475         skb->sk = sk;
2476         skb->destructor = sock_efree;
2477         dst = READ_ONCE(sk->sk_rx_dst);
2478 
2479         if (dst)
2480                 dst = dst_check(dst, 0);
2481         if (dst) {
2482                 u32 itag = 0;
2483 
2484                 /* set noref for now.
2485                  * any place which wants to hold dst has to call
2486                  * dst_hold_safe()
2487                  */
2488                 skb_dst_set_noref(skb, dst);
2489 
2490                 /* for unconnected multicast sockets we need to validate
2491                  * the source on each packet
2492                  */
2493                 if (!inet_sk(sk)->inet_daddr && in_dev)
2494                         return ip_mc_validate_source(skb, iph->daddr,
2495                                                      iph->saddr, iph->tos,
2496                                                      skb->dev, in_dev, &itag);
2497         }
2498         return 0;
2499 }
2500 
2501 int udp_rcv(struct sk_buff *skb)
2502 {
2503         return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
2504 }
2505 
2506 void udp_destroy_sock(struct sock *sk)
2507 {
2508         struct udp_sock *up = udp_sk(sk);
2509         bool slow = lock_sock_fast(sk);
2510         udp_flush_pending_frames(sk);
2511         unlock_sock_fast(sk, slow);
2512         if (static_branch_unlikely(&udp_encap_needed_key)) {
2513                 if (up->encap_type) {
2514                         void (*encap_destroy)(struct sock *sk);
2515                         encap_destroy = READ_ONCE(up->encap_destroy);
2516                         if (encap_destroy)
2517                                 encap_destroy(sk);
2518                 }
2519                 if (up->encap_enabled)
2520                         static_branch_dec(&udp_encap_needed_key);
2521         }
2522 }
2523 
2524 /*
2525  *      Socket option code for UDP
2526  */
2527 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
2528                        char __user *optval, unsigned int optlen,
2529                        int (*push_pending_frames)(struct sock *))
2530 {
2531         struct udp_sock *up = udp_sk(sk);
2532         int val, valbool;
2533         int err = 0;
2534         int is_udplite = IS_UDPLITE(sk);
2535 
2536         if (optlen < sizeof(int))
2537                 return -EINVAL;
2538 
2539         if (get_user(val, (int __user *)optval))
2540                 return -EFAULT;
2541 
2542         valbool = val ? 1 : 0;
2543 
2544         switch (optname) {
2545         case UDP_CORK:
2546                 if (val != 0) {
2547                         up->corkflag = 1;
2548                 } else {
2549                         up->corkflag = 0;
2550                         lock_sock(sk);
2551                         push_pending_frames(sk);
2552                         release_sock(sk);
2553                 }
2554                 break;
2555 
2556         case UDP_ENCAP:
2557                 switch (val) {
2558                 case 0:
2559                 case UDP_ENCAP_ESPINUDP:
2560                 case UDP_ENCAP_ESPINUDP_NON_IKE:
2561                         up->encap_rcv = xfrm4_udp_encap_rcv;
2562                         /* FALLTHROUGH */
2563                 case UDP_ENCAP_L2TPINUDP:
2564                         up->encap_type = val;
2565                         lock_sock(sk);
2566                         udp_tunnel_encap_enable(sk->sk_socket);
2567                         release_sock(sk);
2568                         break;
2569                 default:
2570                         err = -ENOPROTOOPT;
2571                         break;
2572                 }
2573                 break;
2574 
2575         case UDP_NO_CHECK6_TX:
2576                 up->no_check6_tx = valbool;
2577                 break;
2578 
2579         case UDP_NO_CHECK6_RX:
2580                 up->no_check6_rx = valbool;
2581                 break;
2582 
2583         case UDP_SEGMENT:
2584                 if (val < 0 || val > USHRT_MAX)
2585                         return -EINVAL;
2586                 up->gso_size = val;
2587                 break;
2588 
2589         case UDP_GRO:
2590                 lock_sock(sk);
2591                 if (valbool)
2592                         udp_tunnel_encap_enable(sk->sk_socket);
2593                 up->gro_enabled = valbool;
2594                 release_sock(sk);
2595                 break;
2596 
2597         /*
2598          *      UDP-Lite's partial checksum coverage (RFC 3828).
2599          */
2600         /* The sender sets actual checksum coverage length via this option.
2601          * The case coverage > packet length is handled by send module. */
2602         case UDPLITE_SEND_CSCOV:
2603                 if (!is_udplite)         /* Disable the option on UDP sockets */
2604                         return -ENOPROTOOPT;
2605                 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
2606                         val = 8;
2607                 else if (val > USHRT_MAX)
2608                         val = USHRT_MAX;
2609                 up->pcslen = val;
2610                 up->pcflag |= UDPLITE_SEND_CC;
2611                 break;
2612 
2613         /* The receiver specifies a minimum checksum coverage value. To make
2614          * sense, this should be set to at least 8 (as done below). If zero is
2615          * used, this again means full checksum coverage.                     */
2616         case UDPLITE_RECV_CSCOV:
2617                 if (!is_udplite)         /* Disable the option on UDP sockets */
2618                         return -ENOPROTOOPT;
2619                 if (val != 0 && val < 8) /* Avoid silly minimal values.       */
2620                         val = 8;
2621                 else if (val > USHRT_MAX)
2622                         val = USHRT_MAX;
2623                 up->pcrlen = val;
2624                 up->pcflag |= UDPLITE_RECV_CC;
2625                 break;
2626 
2627         default:
2628                 err = -ENOPROTOOPT;
2629                 break;
2630         }
2631 
2632         return err;
2633 }
2634 EXPORT_SYMBOL(udp_lib_setsockopt);
2635 
2636 int udp_setsockopt(struct sock *sk, int level, int optname,
2637                    char __user *optval, unsigned int optlen)
2638 {
2639         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2640                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2641                                           udp_push_pending_frames);
2642         return ip_setsockopt(sk, level, optname, optval, optlen);
2643 }
2644 
2645 #ifdef CONFIG_COMPAT
2646 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
2647                           char __user *optval, unsigned int optlen)
2648 {
2649         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2650                 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
2651                                           udp_push_pending_frames);
2652         return compat_ip_setsockopt(sk, level, optname, optval, optlen);
2653 }
2654 #endif
2655 
2656 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
2657                        char __user *optval, int __user *optlen)
2658 {
2659         struct udp_sock *up = udp_sk(sk);
2660         int val, len;
2661 
2662         if (get_user(len, optlen))
2663                 return -EFAULT;
2664 
2665         len = min_t(unsigned int, len, sizeof(int));
2666 
2667         if (len < 0)
2668                 return -EINVAL;
2669 
2670         switch (optname) {
2671         case UDP_CORK:
2672                 val = up->corkflag;
2673                 break;
2674 
2675         case UDP_ENCAP:
2676                 val = up->encap_type;
2677                 break;
2678 
2679         case UDP_NO_CHECK6_TX:
2680                 val = up->no_check6_tx;
2681                 break;
2682 
2683         case UDP_NO_CHECK6_RX:
2684                 val = up->no_check6_rx;
2685                 break;
2686 
2687         case UDP_SEGMENT:
2688                 val = up->gso_size;
2689                 break;
2690 
2691         /* The following two cannot be changed on UDP sockets, the return is
2692          * always 0 (which corresponds to the full checksum coverage of UDP). */
2693         case UDPLITE_SEND_CSCOV:
2694                 val = up->pcslen;
2695                 break;
2696 
2697         case UDPLITE_RECV_CSCOV:
2698                 val = up->pcrlen;
2699                 break;
2700 
2701         default:
2702                 return -ENOPROTOOPT;
2703         }
2704 
2705         if (put_user(len, optlen))
2706                 return -EFAULT;
2707         if (copy_to_user(optval, &val, len))
2708                 return -EFAULT;
2709         return 0;
2710 }
2711 EXPORT_SYMBOL(udp_lib_getsockopt);
2712 
2713 int udp_getsockopt(struct sock *sk, int level, int optname,
2714                    char __user *optval, int __user *optlen)
2715 {
2716         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2717                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2718         return ip_getsockopt(sk, level, optname, optval, optlen);
2719 }
2720 
2721 #ifdef CONFIG_COMPAT
2722 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
2723                                  char __user *optval, int __user *optlen)
2724 {
2725         if (level == SOL_UDP  ||  level == SOL_UDPLITE)
2726                 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
2727         return compat_ip_getsockopt(sk, level, optname, optval, optlen);
2728 }
2729 #endif
2730 /**
2731  *      udp_poll - wait for a UDP event.
2732  *      @file - file struct
2733  *      @sock - socket
2734  *      @wait - poll table
2735  *
2736  *      This is same as datagram poll, except for the special case of
2737  *      blocking sockets. If application is using a blocking fd
2738  *      and a packet with checksum error is in the queue;
2739  *      then it could get return from select indicating data available
2740  *      but then block when reading it. Add special case code
2741  *      to work around these arguably broken applications.
2742  */
2743 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
2744 {
2745         __poll_t mask = datagram_poll(file, sock, wait);
2746         struct sock *sk = sock->sk;
2747 
2748         if (!skb_queue_empty_lockless(&udp_sk(sk)->reader_queue))
2749                 mask |= EPOLLIN | EPOLLRDNORM;
2750 
2751         /* Check for false positives due to checksum errors */
2752         if ((mask & EPOLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
2753             !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
2754                 mask &= ~(EPOLLIN | EPOLLRDNORM);
2755 
2756         return mask;
2757 
2758 }
2759 EXPORT_SYMBOL(udp_poll);
2760 
2761 int udp_abort(struct sock *sk, int err)
2762 {
2763         lock_sock(sk);
2764 
2765         sk->sk_err = err;
2766         sk->sk_error_report(sk);
2767         __udp_disconnect(sk, 0);
2768 
2769         release_sock(sk);
2770 
2771         return 0;
2772 }
2773 EXPORT_SYMBOL_GPL(udp_abort);
2774 
2775 struct proto udp_prot = {
2776         .name                   = "UDP",
2777         .owner                  = THIS_MODULE,
2778         .close                  = udp_lib_close,
2779         .pre_connect            = udp_pre_connect,
2780         .connect                = ip4_datagram_connect,
2781         .disconnect             = udp_disconnect,
2782         .ioctl                  = udp_ioctl,
2783         .init                   = udp_init_sock,
2784         .destroy                = udp_destroy_sock,
2785         .setsockopt             = udp_setsockopt,
2786         .getsockopt             = udp_getsockopt,
2787         .sendmsg                = udp_sendmsg,
2788         .recvmsg                = udp_recvmsg,
2789         .sendpage               = udp_sendpage,
2790         .release_cb             = ip4_datagram_release_cb,
2791         .hash                   = udp_lib_hash,
2792         .unhash                 = udp_lib_unhash,
2793         .rehash                 = udp_v4_rehash,
2794         .get_port               = udp_v4_get_port,
2795         .memory_allocated       = &udp_memory_allocated,
2796         .sysctl_mem             = sysctl_udp_mem,
2797         .sysctl_wmem_offset     = offsetof(struct net, ipv4.sysctl_udp_wmem_min),
2798         .sysctl_rmem_offset     = offsetof(struct net, ipv4.sysctl_udp_rmem_min),
2799         .obj_size               = sizeof(struct udp_sock),
2800         .h.udp_table            = &udp_table,
2801 #ifdef CONFIG_COMPAT
2802         .compat_setsockopt      = compat_udp_setsockopt,
2803         .compat_getsockopt      = compat_udp_getsockopt,
2804 #endif
2805         .diag_destroy           = udp_abort,
2806 };
2807 EXPORT_SYMBOL(udp_prot);
2808 
2809 /* ------------------------------------------------------------------------ */
2810 #ifdef CONFIG_PROC_FS
2811 
2812 static struct sock *udp_get_first(struct seq_file *seq, int start)
2813 {
2814         struct sock *sk;
2815         struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2816         struct udp_iter_state *state = seq->private;
2817         struct net *net = seq_file_net(seq);
2818 
2819         for (state->bucket = start; state->bucket <= afinfo->udp_table->mask;
2820              ++state->bucket) {
2821                 struct udp_hslot *hslot = &afinfo->udp_table->hash[state->bucket];
2822 
2823                 if (hlist_empty(&hslot->head))
2824                         continue;
2825 
2826                 spin_lock_bh(&hslot->lock);
2827                 sk_for_each(sk, &hslot->head) {
2828                         if (!net_eq(sock_net(sk), net))
2829                                 continue;
2830                         if (sk->sk_family == afinfo->family)
2831                                 goto found;
2832                 }
2833                 spin_unlock_bh(&hslot->lock);
2834         }
2835         sk = NULL;
2836 found:
2837         return sk;
2838 }
2839 
2840 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2841 {
2842         struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2843         struct udp_iter_state *state = seq->private;
2844         struct net *net = seq_file_net(seq);
2845 
2846         do {
2847                 sk = sk_next(sk);
2848         } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != afinfo->family));
2849 
2850         if (!sk) {
2851                 if (state->bucket <= afinfo->udp_table->mask)
2852                         spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2853                 return udp_get_first(seq, state->bucket + 1);
2854         }
2855         return sk;
2856 }
2857 
2858 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2859 {
2860         struct sock *sk = udp_get_first(seq, 0);
2861 
2862         if (sk)
2863                 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2864                         --pos;
2865         return pos ? NULL : sk;
2866 }
2867 
2868 void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2869 {
2870         struct udp_iter_state *state = seq->private;
2871         state->bucket = MAX_UDP_PORTS;
2872 
2873         return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2874 }
2875 EXPORT_SYMBOL(udp_seq_start);
2876 
2877 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2878 {
2879         struct sock *sk;
2880 
2881         if (v == SEQ_START_TOKEN)
2882                 sk = udp_get_idx(seq, 0);
2883         else
2884                 sk = udp_get_next(seq, v);
2885 
2886         ++*pos;
2887         return sk;
2888 }
2889 EXPORT_SYMBOL(udp_seq_next);
2890 
2891 void udp_seq_stop(struct seq_file *seq, void *v)
2892 {
2893         struct udp_seq_afinfo *afinfo = PDE_DATA(file_inode(seq->file));
2894         struct udp_iter_state *state = seq->private;
2895 
2896         if (state->bucket <= afinfo->udp_table->mask)
2897                 spin_unlock_bh(&afinfo->udp_table->hash[state->bucket].lock);
2898 }
2899 EXPORT_SYMBOL(udp_seq_stop);
2900 
2901 /* ------------------------------------------------------------------------ */
2902 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2903                 int bucket)
2904 {
2905         struct inet_sock *inet = inet_sk(sp);
2906         __be32 dest = inet->inet_daddr;
2907         __be32 src  = inet->inet_rcv_saddr;
2908         __u16 destp       = ntohs(inet->inet_dport);
2909         __u16 srcp        = ntohs(inet->inet_sport);
2910 
2911         seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2912                 " %02X %08X:%08X %02X:%08lX %08X %5u %8d %lu %d %pK %u",
2913                 bucket, src, srcp, dest, destp, sp->sk_state,
2914                 sk_wmem_alloc_get(sp),
2915                 udp_rqueue_get(sp),
2916                 0, 0L, 0,
2917                 from_kuid_munged(seq_user_ns(f), sock_i_uid(sp)),
2918                 0, sock_i_ino(sp),
2919                 refcount_read(&sp->sk_refcnt), sp,
2920                 atomic_read(&sp->sk_drops));
2921 }
2922 
2923 int udp4_seq_show(struct seq_file *seq, void *v)
2924 {
2925         seq_setwidth(seq, 127);
2926         if (v == SEQ_START_TOKEN)
2927                 seq_puts(seq, "  sl  local_address rem_address   st tx_queue "
2928                            "rx_queue tr tm->when retrnsmt   uid  timeout "
2929                            "inode ref pointer drops");
2930         else {
2931                 struct udp_iter_state *state = seq->private;
2932 
2933                 udp4_format_sock(v, seq, state->bucket);
2934         }
2935         seq_pad(seq, '\n');
2936         return 0;
2937 }
2938 
2939 const struct seq_operations udp_seq_ops = {
2940         .start          = udp_seq_start,
2941         .next           = udp_seq_next,
2942         .stop           = udp_seq_stop,
2943         .show           = udp4_seq_show,
2944 };
2945 EXPORT_SYMBOL(udp_seq_ops);
2946 
2947 static struct udp_seq_afinfo udp4_seq_afinfo = {
2948         .family         = AF_INET,
2949         .udp_table      = &udp_table,
2950 };
2951 
2952 static int __net_init udp4_proc_init_net(struct net *net)
2953 {
2954         if (!proc_create_net_data("udp", 0444, net->proc_net, &udp_seq_ops,
2955                         sizeof(struct udp_iter_state), &udp4_seq_afinfo))
2956                 return -ENOMEM;
2957         return 0;
2958 }
2959 
2960 static void __net_exit udp4_proc_exit_net(struct net *net)
2961 {
2962         remove_proc_entry("udp", net->proc_net);
2963 }
2964 
2965 static struct pernet_operations udp4_net_ops = {
2966         .init = udp4_proc_init_net,
2967         .exit = udp4_proc_exit_net,
2968 };
2969 
2970 int __init udp4_proc_init(void)
2971 {
2972         return register_pernet_subsys(&udp4_net_ops);
2973 }
2974 
2975 void udp4_proc_exit(void)
2976 {
2977         unregister_pernet_subsys(&udp4_net_ops);
2978 }
2979 #endif /* CONFIG_PROC_FS */
2980 
2981 static __initdata unsigned long uhash_entries;
2982 static int __init set_uhash_entries(char *str)
2983 {
2984         ssize_t ret;
2985 
2986         if (!str)
2987                 return 0;
2988 
2989         ret = kstrtoul(str, 0, &uhash_entries);
2990         if (ret)
2991                 return 0;
2992 
2993         if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2994                 uhash_entries = UDP_HTABLE_SIZE_MIN;
2995         return 1;
2996 }
2997 __setup("uhash_entries=", set_uhash_entries);
2998 
2999 void __init udp_table_init(struct udp_table *table, const char *name)
3000 {
3001         unsigned int i;
3002 
3003         table->hash = alloc_large_system_hash(name,
3004                                               2 * sizeof(struct udp_hslot),
3005                                               uhash_entries,
3006                                               21, /* one slot per 2 MB */
3007                                               0,
3008                                               &table->log,
3009                                               &table->mask,
3010                                               UDP_HTABLE_SIZE_MIN,
3011                                               64 * 1024);
3012 
3013         table->hash2 = table->hash + (table->mask + 1);
3014         for (i = 0; i <= table->mask; i++) {
3015                 INIT_HLIST_HEAD(&table->hash[i].head);
3016                 table->hash[i].count = 0;
3017                 spin_lock_init(&table->hash[i].lock);
3018         }
3019         for (i = 0; i <= table->mask; i++) {
3020                 INIT_HLIST_HEAD(&table->hash2[i].head);
3021                 table->hash2[i].count = 0;
3022                 spin_lock_init(&table->hash2[i].lock);
3023         }
3024 }
3025 
3026 u32 udp_flow_hashrnd(void)
3027 {
3028         static u32 hashrnd __read_mostly;
3029 
3030         net_get_random_once(&hashrnd, sizeof(hashrnd));
3031 
3032         return hashrnd;
3033 }
3034 EXPORT_SYMBOL(udp_flow_hashrnd);
3035 
3036 static void __udp_sysctl_init(struct net *net)
3037 {
3038         net->ipv4.sysctl_udp_rmem_min = SK_MEM_QUANTUM;
3039         net->ipv4.sysctl_udp_wmem_min = SK_MEM_QUANTUM;
3040 
3041 #ifdef CONFIG_NET_L3_MASTER_DEV
3042         net->ipv4.sysctl_udp_l3mdev_accept = 0;
3043 #endif
3044 }
3045 
3046 static int __net_init udp_sysctl_init(struct net *net)
3047 {
3048         __udp_sysctl_init(net);
3049         return 0;
3050 }
3051 
3052 static struct pernet_operations __net_initdata udp_sysctl_ops = {
3053         .init   = udp_sysctl_init,
3054 };
3055 
3056 void __init udp_init(void)
3057 {
3058         unsigned long limit;
3059         unsigned int i;
3060 
3061         udp_table_init(&udp_table, "UDP");
3062         limit = nr_free_buffer_pages() / 8;
3063         limit = max(limit, 128UL);
3064         sysctl_udp_mem[0] = limit / 4 * 3;
3065         sysctl_udp_mem[1] = limit;
3066         sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
3067 
3068         __udp_sysctl_init(&init_net);
3069 
3070         /* 16 spinlocks per cpu */
3071         udp_busylocks_log = ilog2(nr_cpu_ids) + 4;
3072         udp_busylocks = kmalloc(sizeof(spinlock_t) << udp_busylocks_log,
3073                                 GFP_KERNEL);
3074         if (!udp_busylocks)
3075                 panic("UDP: failed to alloc udp_busylocks\n");
3076         for (i = 0; i < (1U << udp_busylocks_log); i++)
3077                 spin_lock_init(udp_busylocks + i);
3078 
3079         if (register_pernet_subsys(&udp_sysctl_ops))
3080                 panic("UDP: failed to init sysctl parameters.\n");
3081 }

/* [<][>][^][v][top][bottom][index][help] */