include/net/udp.h

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Definitions for the UDP module.
   7  *
   8  * Version:     @(#)udp.h       1.0.2   05/07/93
   9  *
  10  * Authors:     Ross Biro
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *
  13  * Fixes:
  14  *              Alan Cox        : Turned on udp checksums. I don't want to
  15  *                                chase 'memory corruption' bugs that aren't!
  16  *
  17  *              This program is free software; you can redistribute it and/or
  18  *              modify it under the terms of the GNU General Public License
  19  *              as published by the Free Software Foundation; either version
  20  *              2 of the License, or (at your option) any later version.
  21  */
  22 #ifndef _UDP_H
  23 #define _UDP_H
  24
  25 #include <linux/list.h>
  26 #include <linux/bug.h>
  27 #include <net/inet_sock.h>
  28 #include <net/sock.h>
  29 #include <net/snmp.h>
  30 #include <net/ip.h>
  31 #include <linux/ipv6.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/poll.h>
  34
  35 /**
  36  *      struct udp_skb_cb  -  UDP(-Lite) private variables
  37  *
  38  *      @header:      private variables used by IPv4/IPv6
  39  *      @cscov:       checksum coverage length (UDP-Lite only)
  40  *      @partial_cov: if set indicates partial csum coverage
  41  */
  42 struct udp_skb_cb {
  43         union {
  44                 struct inet_skb_parm    h4;
  45 #if IS_ENABLED(CONFIG_IPV6)
  46                 struct inet6_skb_parm   h6;
  47 #endif
  48         } header;
  49         __u16           cscov;
  50         __u8            partial_cov;
  51 };
  52 #define UDP_SKB_CB(__skb)       ((struct udp_skb_cb *)((__skb)->cb))
  53
  54 /**
  55  *      struct udp_hslot - UDP hash slot
  56  *
  57  *      @head:  head of list of sockets
  58  *      @count: number of sockets in 'head' list
  59  *      @lock:  spinlock protecting changes to head/count
  60  */
  61 struct udp_hslot {
  62         struct hlist_head       head;
  63         int                     count;
  64         spinlock_t              lock;
  65 } __attribute__((aligned(2 * sizeof(long))));
  66
  67 /**
  68  *      struct udp_table - UDP table
  69  *
  70  *      @hash:  hash table, sockets are hashed on (local port)
  71  *      @hash2: hash table, sockets are hashed on (local port, local address)
  72  *      @mask:  number of slots in hash tables, minus 1
  73  *      @log:   log2(number of slots in hash table)
  74  */
  75 struct udp_table {
  76         struct udp_hslot        *hash;
  77         struct udp_hslot        *hash2;
  78         unsigned int            mask;
  79         unsigned int            log;
  80 };
  81 extern struct udp_table udp_table;
  82 void udp_table_init(struct udp_table *, const char *);
  83 static inline struct udp_hslot *udp_hashslot(struct udp_table *table,
  84                                              struct net *net, unsigned int num)
  85 {
  86         return &table->hash[udp_hashfn(net, num, table->mask)];
  87 }
  88 /*
  89  * For secondary hash, net_hash_mix() is performed before calling
  90  * udp_hashslot2(), this explains difference with udp_hashslot()
  91  */
  92 static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
  93                                               unsigned int hash)
  94 {
  95         return &table->hash2[hash & table->mask];
  96 }
  97
  98 extern struct proto udp_prot;
  99
 100 extern atomic_long_t udp_memory_allocated;
 101
 102 /* sysctl variables for udp */
 103 extern long sysctl_udp_mem[3];
 104 extern int sysctl_udp_rmem_min;
 105 extern int sysctl_udp_wmem_min;
 106
 107 struct sk_buff;
 108
 109 /*
 110  *      Generic checksumming routines for UDP(-Lite) v4 and v6
 111  */
 112 static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
 113 {
 114         return (UDP_SKB_CB(skb)->cscov == skb->len ?
 115                 __skb_checksum_complete(skb) :
 116                 __skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov));
 117 }
 118
 119 static inline int udp_lib_checksum_complete(struct sk_buff *skb)
 120 {
 121         return !skb_csum_unnecessary(skb) &&
 122                 __udp_lib_checksum_complete(skb);
 123 }
 124
 125 /**
 126  *      udp_csum_outgoing  -  compute UDPv4/v6 checksum over fragments
 127  *      @sk:    socket we are writing to
 128  *      @skb:   sk_buff containing the filled-in UDP header
 129  *              (checksum field must be zeroed out)
 130  */
 131 static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
 132 {
 133         __wsum csum = csum_partial(skb_transport_header(skb),
 134                                    sizeof(struct udphdr), 0);
 135         skb_queue_walk(&sk->sk_write_queue, skb) {
 136                 csum = csum_add(csum, skb->csum);
 137         }
 138         return csum;
 139 }
 140
 141 static inline __wsum udp_csum(struct sk_buff *skb)
 142 {
 143         __wsum csum = csum_partial(skb_transport_header(skb),
 144                                    sizeof(struct udphdr), skb->csum);
 145
 146         for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
 147                 csum = csum_add(csum, skb->csum);
 148         }
 149         return csum;
 150 }
 151
 152 static inline __sum16 udp_v4_check(int len, __be32 saddr,
 153                                    __be32 daddr, __wsum base)
 154 {
 155         return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
 156 }
 157
 158 void udp_set_csum(bool nocheck, struct sk_buff *skb,
 159                   __be32 saddr, __be32 daddr, int len);
 160
 161 static inline void udp_csum_pull_header(struct sk_buff *skb)
 162 {
 163         if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
 164                 skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
 165                                          skb->csum);
 166         skb_pull_rcsum(skb, sizeof(struct udphdr));
 167         UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
 168 }
 169
 170 typedef struct sock *(*udp_lookup_t)(struct sk_buff *skb, __be16 sport,
 171                                      __be16 dport);
 172
 173 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
 174                                 struct udphdr *uh, udp_lookup_t lookup);
 175 int udp_gro_complete(struct sk_buff *skb, int nhoff, udp_lookup_t lookup);
 176
 177 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
 178                                   netdev_features_t features);
 179
 180 static inline struct udphdr *udp_gro_udphdr(struct sk_buff *skb)
 181 {
 182         struct udphdr *uh;
 183         unsigned int hlen, off;
 184
 185         off  = skb_gro_offset(skb);
 186         hlen = off + sizeof(*uh);
 187         uh   = skb_gro_header_fast(skb, off);
 188         if (skb_gro_header_hard(skb, hlen))
 189                 uh = skb_gro_header_slow(skb, hlen, off);
 190
 191         return uh;
 192 }
 193
 194 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
 195 static inline int udp_lib_hash(struct sock *sk)
 196 {
 197         BUG();
 198         return 0;
 199 }
 200
 201 void udp_lib_unhash(struct sock *sk);
 202 void udp_lib_rehash(struct sock *sk, u16 new_hash);
 203
 204 static inline void udp_lib_close(struct sock *sk, long timeout)
 205 {
 206         sk_common_release(sk);
 207 }
 208
 209 int udp_lib_get_port(struct sock *sk, unsigned short snum,
 210                      unsigned int hash2_nulladdr);
 211
 212 u32 udp_flow_hashrnd(void);
 213
 214 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
 215                                        int min, int max, bool use_eth)
 216 {
 217         u32 hash;
 218
 219         if (min >= max) {
 220                 /* Use default range */
 221                 inet_get_local_port_range(net, &min, &max);
 222         }
 223
 224         hash = skb_get_hash(skb);
 225         if (unlikely(!hash)) {
 226                 if (use_eth) {
 227                         /* Can't find a normal hash, caller has indicated an
 228                          * Ethernet packet so use that to compute a hash.
 229                          */
 230                         hash = jhash(skb->data, 2 * ETH_ALEN,
 231                                      (__force u32) skb->protocol);
 232                 } else {
 233                         /* Can't derive any sort of hash for the packet, set
 234                          * to some consistent random value.
 235                          */
 236                         hash = udp_flow_hashrnd();
 237                 }
 238         }
 239
 240         /* Since this is being sent on the wire obfuscate hash a bit
 241          * to minimize possbility that any useful information to an
 242          * attacker is leaked. Only upper 16 bits are relevant in the
 243          * computation for 16 bit port value.
 244          */
 245         hash ^= hash << 16;
 246
 247         return htons((((u64) hash * (max - min)) >> 32) + min);
 248 }
 249
 250 static inline int udp_rqueue_get(struct sock *sk)
 251 {
 252         return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
 253 }
 254
 255 static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
 256                                        int dif, int sdif)
 257 {
 258 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
 259         return inet_bound_dev_eq(!!net->ipv4.sysctl_udp_l3mdev_accept,
 260                                  bound_dev_if, dif, sdif);
 261 #else
 262         return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
 263 #endif
 264 }
 265
 266 /* net/ipv4/udp.c */
 267 void udp_destruct_sock(struct sock *sk);
 268 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
 269 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
 270 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
 271 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
 272                                int noblock, int *peeked, int *off, int *err);
 273 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
 274                                            int noblock, int *err)
 275 {
 276         int peeked, off = 0;
 277
 278         return __skb_recv_udp(sk, flags, noblock, &peeked, &off, err);
 279 }
 280
 281 int udp_v4_early_demux(struct sk_buff *skb);
 282 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
 283 int udp_get_port(struct sock *sk, unsigned short snum,
 284                  int (*saddr_cmp)(const struct sock *,
 285                                   const struct sock *));
 286 int udp_err(struct sk_buff *, u32);
 287 int udp_abort(struct sock *sk, int err);
 288 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
 289 int udp_push_pending_frames(struct sock *sk);
 290 void udp_flush_pending_frames(struct sock *sk);
 291 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
 292 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
 293 int udp_rcv(struct sk_buff *skb);
 294 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
 295 int udp_init_sock(struct sock *sk);
 296 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
 297 int __udp_disconnect(struct sock *sk, int flags);
 298 int udp_disconnect(struct sock *sk, int flags);
 299 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
 300 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
 301                                        netdev_features_t features,
 302                                        bool is_ipv6);
 303 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
 304                        char __user *optval, int __user *optlen);
 305 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
 306                        char __user *optval, unsigned int optlen,
 307                        int (*push_pending_frames)(struct sock *));
 308 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 309                              __be32 daddr, __be16 dport, int dif);
 310 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 311                                __be32 daddr, __be16 dport, int dif, int sdif,
 312                                struct udp_table *tbl, struct sk_buff *skb);
 313 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
 314                                  __be16 sport, __be16 dport);
 315 struct sock *udp6_lib_lookup(struct net *net,
 316                              const struct in6_addr *saddr, __be16 sport,
 317                              const struct in6_addr *daddr, __be16 dport,
 318                              int dif);
 319 struct sock *__udp6_lib_lookup(struct net *net,
 320                                const struct in6_addr *saddr, __be16 sport,
 321                                const struct in6_addr *daddr, __be16 dport,
 322                                int dif, int sdif, struct udp_table *tbl,
 323                                struct sk_buff *skb);
 324 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
 325                                  __be16 sport, __be16 dport);
 326
 327 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
 328  * possibly multiple cache miss on dequeue()
 329  */
 330 struct udp_dev_scratch {
 331         /* skb->truesize and the stateless bit are embedded in a single field;
 332          * do not use a bitfield since the compiler emits better/smaller code
 333          * this way
 334          */
 335         u32 _tsize_state;
 336
 337 #if BITS_PER_LONG == 64
 338         /* len and the bit needed to compute skb_csum_unnecessary
 339          * will be on cold cache lines at recvmsg time.
 340          * skb->len can be stored on 16 bits since the udp header has been
 341          * already validated and pulled.
 342          */
 343         u16 len;
 344         bool is_linear;
 345         bool csum_unnecessary;
 346 #endif
 347 };
 348
 349 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
 350 {
 351         return (struct udp_dev_scratch *)&skb->dev_scratch;
 352 }
 353
 354 #if BITS_PER_LONG == 64
 355 static inline unsigned int udp_skb_len(struct sk_buff *skb)
 356 {
 357         return udp_skb_scratch(skb)->len;
 358 }
 359
 360 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
 361 {
 362         return udp_skb_scratch(skb)->csum_unnecessary;
 363 }
 364
 365 static inline bool udp_skb_is_linear(struct sk_buff *skb)
 366 {
 367         return udp_skb_scratch(skb)->is_linear;
 368 }
 369
 370 #else
 371 static inline unsigned int udp_skb_len(struct sk_buff *skb)
 372 {
 373         return skb->len;
 374 }
 375
 376 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
 377 {
 378         return skb_csum_unnecessary(skb);
 379 }
 380
 381 static inline bool udp_skb_is_linear(struct sk_buff *skb)
 382 {
 383         return !skb_is_nonlinear(skb);
 384 }
 385 #endif
 386
 387 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
 388                                   struct iov_iter *to)
 389 {
 390         int n;
 391
 392         n = copy_to_iter(skb->data + off, len, to);
 393         if (n == len)
 394                 return 0;
 395
 396         iov_iter_revert(to, n);
 397         return -EFAULT;
 398 }
 399
 400 /*
 401  *      SNMP statistics for UDP and UDP-Lite
 402  */
 403 #define UDP_INC_STATS(net, field, is_udplite)                 do { \
 404         if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field);       \
 405         else            SNMP_INC_STATS((net)->mib.udp_statistics, field);  }  while(0)
 406 #define __UDP_INC_STATS(net, field, is_udplite)               do { \
 407         if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field);         \
 408         else            __SNMP_INC_STATS((net)->mib.udp_statistics, field);    }  while(0)
 409
 410 #define __UDP6_INC_STATS(net, field, is_udplite)            do { \
 411         if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
 412         else            __SNMP_INC_STATS((net)->mib.udp_stats_in6, field);  \
 413 } while(0)
 414 #define UDP6_INC_STATS(net, field, __lite)                  do { \
 415         if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);  \
 416         else        SNMP_INC_STATS((net)->mib.udp_stats_in6, field);      \
 417 } while(0)
 418
 419 #if IS_ENABLED(CONFIG_IPV6)
 420 #define __UDPX_MIB(sk, ipv4)                                            \
 421 ({                                                                      \
 422         ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
 423                                  sock_net(sk)->mib.udp_statistics) :    \
 424                 (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \
 425                                  sock_net(sk)->mib.udp_stats_in6);      \
 426 })
 427 #else
 428 #define __UDPX_MIB(sk, ipv4)                                            \
 429 ({                                                                      \
 430         IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :         \
 431                          sock_net(sk)->mib.udp_statistics;              \
 432 })
 433 #endif
 434
 435 #define __UDPX_INC_STATS(sk, field) \
 436         __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
 437
 438 #ifdef CONFIG_PROC_FS
 439 struct udp_seq_afinfo {
 440         sa_family_t                     family;
 441         struct udp_table                *udp_table;
 442 };
 443
 444 struct udp_iter_state {
 445         struct seq_net_private  p;
 446         int                     bucket;
 447 };
 448
 449 void *udp_seq_start(struct seq_file *seq, loff_t *pos);
 450 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
 451 void udp_seq_stop(struct seq_file *seq, void *v);
 452
 453 extern const struct seq_operations udp_seq_ops;
 454 extern const struct seq_operations udp6_seq_ops;
 455
 456 int udp4_proc_init(void);
 457 void udp4_proc_exit(void);
 458 #endif /* CONFIG_PROC_FS */
 459
 460 int udpv4_offload_init(void);
 461
 462 void udp_init(void);
 463
 464 DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
 465 void udp_encap_enable(void);
 466 #if IS_ENABLED(CONFIG_IPV6)
 467 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
 468 void udpv6_encap_enable(void);
 469 #endif
 470
 471 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
 472                                               struct sk_buff *skb, bool ipv4)
 473 {
 474         struct sk_buff *segs;
 475
 476         /* the GSO CB lays after the UDP one, no need to save and restore any
 477          * CB fragment
 478          */
 479         segs = __skb_gso_segment(skb, NETIF_F_SG, false);
 480         if (unlikely(IS_ERR_OR_NULL(segs))) {
 481                 int segs_nr = skb_shinfo(skb)->gso_segs;
 482
 483                 atomic_add(segs_nr, &sk->sk_drops);
 484                 SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
 485                 kfree_skb(skb);
 486                 return NULL;
 487         }
 488
 489         consume_skb(skb);
 490         return segs;
 491 }
 492
 493 #endif  /* _UDP_H */