include/net/udp.h

   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  *              Definitions for the UDP module.
   8  *
   9  * Version:     @(#)udp.h       1.0.2   05/07/93
  10  *
  11  * Authors:     Ross Biro
  12  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  13  *
  14  * Fixes:
  15  *              Alan Cox        : Turned on udp checksums. I don't want to
  16  *                                chase 'memory corruption' bugs that aren't!
  17  */
  18 #ifndef _UDP_H
  19 #define _UDP_H
  20
  21 #include <linux/list.h>
  22 #include <linux/bug.h>
  23 #include <net/inet_sock.h>
  24 #include <net/sock.h>
  25 #include <net/snmp.h>
  26 #include <net/ip.h>
  27 #include <linux/ipv6.h>
  28 #include <linux/seq_file.h>
  29 #include <linux/poll.h>
  30 #include <linux/indirect_call_wrapper.h>
  31
  32 /**
  33  *      struct udp_skb_cb  -  UDP(-Lite) private variables
  34  *
  35  *      @header:      private variables used by IPv4/IPv6
  36  *      @cscov:       checksum coverage length (UDP-Lite only)
  37  *      @partial_cov: if set indicates partial csum coverage
  38  */
  39 struct udp_skb_cb {
  40         union {
  41                 struct inet_skb_parm    h4;
  42 #if IS_ENABLED(CONFIG_IPV6)
  43                 struct inet6_skb_parm   h6;
  44 #endif
  45         } header;
  46         __u16           cscov;
  47         __u8            partial_cov;
  48 };
  49 #define UDP_SKB_CB(__skb)       ((struct udp_skb_cb *)((__skb)->cb))
  50
  51 /**
  52  *      struct udp_hslot - UDP hash slot
  53  *
  54  *      @head:  head of list of sockets
  55  *      @count: number of sockets in 'head' list
  56  *      @lock:  spinlock protecting changes to head/count
  57  */
  58 struct udp_hslot {
  59         struct hlist_head       head;
  60         int                     count;
  61         spinlock_t              lock;
  62 } __attribute__((aligned(2 * sizeof(long))));
  63
  64 /**
  65  *      struct udp_table - UDP table
  66  *
  67  *      @hash:  hash table, sockets are hashed on (local port)
  68  *      @hash2: hash table, sockets are hashed on (local port, local address)
  69  *      @mask:  number of slots in hash tables, minus 1
  70  *      @log:   log2(number of slots in hash table)
  71  */
  72 struct udp_table {
  73         struct udp_hslot        *hash;
  74         struct udp_hslot        *hash2;
  75         unsigned int            mask;
  76         unsigned int            log;
  77 };
  78 extern struct udp_table udp_table;
  79 void udp_table_init(struct udp_table *, const char *);
  80 static inline struct udp_hslot *udp_hashslot(struct udp_table *table,
  81                                              struct net *net, unsigned int num)
  82 {
  83         return &table->hash[udp_hashfn(net, num, table->mask)];
  84 }
  85 /*
  86  * For secondary hash, net_hash_mix() is performed before calling
  87  * udp_hashslot2(), this explains difference with udp_hashslot()
  88  */
  89 static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
  90                                               unsigned int hash)
  91 {
  92         return &table->hash2[hash & table->mask];
  93 }
  94
  95 extern struct proto udp_prot;
  96
  97 extern atomic_long_t udp_memory_allocated;
  98
  99 /* sysctl variables for udp */
 100 extern long sysctl_udp_mem[3];
 101 extern int sysctl_udp_rmem_min;
 102 extern int sysctl_udp_wmem_min;
 103
 104 struct sk_buff;
 105
 106 /*
 107  *      Generic checksumming routines for UDP(-Lite) v4 and v6
 108  */
 109 static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
 110 {
 111         return (UDP_SKB_CB(skb)->cscov == skb->len ?
 112                 __skb_checksum_complete(skb) :
 113                 __skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov));
 114 }
 115
 116 static inline int udp_lib_checksum_complete(struct sk_buff *skb)
 117 {
 118         return !skb_csum_unnecessary(skb) &&
 119                 __udp_lib_checksum_complete(skb);
 120 }
 121
 122 /**
 123  *      udp_csum_outgoing  -  compute UDPv4/v6 checksum over fragments
 124  *      @sk:    socket we are writing to
 125  *      @skb:   sk_buff containing the filled-in UDP header
 126  *              (checksum field must be zeroed out)
 127  */
 128 static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
 129 {
 130         __wsum csum = csum_partial(skb_transport_header(skb),
 131                                    sizeof(struct udphdr), 0);
 132         skb_queue_walk(&sk->sk_write_queue, skb) {
 133                 csum = csum_add(csum, skb->csum);
 134         }
 135         return csum;
 136 }
 137
 138 static inline __wsum udp_csum(struct sk_buff *skb)
 139 {
 140         __wsum csum = csum_partial(skb_transport_header(skb),
 141                                    sizeof(struct udphdr), skb->csum);
 142
 143         for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
 144                 csum = csum_add(csum, skb->csum);
 145         }
 146         return csum;
 147 }
 148
 149 static inline __sum16 udp_v4_check(int len, __be32 saddr,
 150                                    __be32 daddr, __wsum base)
 151 {
 152         return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
 153 }
 154
 155 void udp_set_csum(bool nocheck, struct sk_buff *skb,
 156                   __be32 saddr, __be32 daddr, int len);
 157
 158 static inline void udp_csum_pull_header(struct sk_buff *skb)
 159 {
 160         if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
 161                 skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
 162                                          skb->csum);
 163         skb_pull_rcsum(skb, sizeof(struct udphdr));
 164         UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
 165 }
 166
 167 typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport,
 168                                      __be16 dport);
 169
 170 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp4_gro_receive(struct list_head *,
 171                                                            struct sk_buff *));
 172 INDIRECT_CALLABLE_DECLARE(int udp4_gro_complete(struct sk_buff *, int));
 173 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp6_gro_receive(struct list_head *,
 174                                                            struct sk_buff *));
 175 INDIRECT_CALLABLE_DECLARE(int udp6_gro_complete(struct sk_buff *, int));
 176 INDIRECT_CALLABLE_DECLARE(void udp_v6_early_demux(struct sk_buff *));
 177 INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
 178
 179 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
 180                                 struct udphdr *uh, struct sock *sk);
 181 int udp_gro_complete(struct sk_buff *skb, int nhoff, udp_lookup_t lookup);
 182
 183 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
 184                                   netdev_features_t features, bool is_ipv6);
 185
 186 static inline struct udphdr *udp_gro_udphdr(struct sk_buff *skb)
 187 {
 188         struct udphdr *uh;
 189         unsigned int hlen, off;
 190
 191         off  = skb_gro_offset(skb);
 192         hlen = off + sizeof(*uh);
 193         uh   = skb_gro_header_fast(skb, off);
 194         if (skb_gro_header_hard(skb, hlen))
 195                 uh = skb_gro_header_slow(skb, hlen, off);
 196
 197         return uh;
 198 }
 199
 200 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
 201 static inline int udp_lib_hash(struct sock *sk)
 202 {
 203         BUG();
 204         return 0;
 205 }
 206
 207 void udp_lib_unhash(struct sock *sk);
 208 void udp_lib_rehash(struct sock *sk, u16 new_hash);
 209
 210 static inline void udp_lib_close(struct sock *sk, long timeout)
 211 {
 212         sk_common_release(sk);
 213 }
 214
 215 int udp_lib_get_port(struct sock *sk, unsigned short snum,
 216                      unsigned int hash2_nulladdr);
 217
 218 u32 udp_flow_hashrnd(void);
 219
 220 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
 221                                        int min, int max, bool use_eth)
 222 {
 223         u32 hash;
 224
 225         if (min >= max) {
 226                 /* Use default range */
 227                 inet_get_local_port_range(net, &min, &max);
 228         }
 229
 230         hash = skb_get_hash(skb);
 231         if (unlikely(!hash)) {
 232                 if (use_eth) {
 233                         /* Can't find a normal hash, caller has indicated an
 234                          * Ethernet packet so use that to compute a hash.
 235                          */
 236                         hash = jhash(skb->data, 2 * ETH_ALEN,
 237                                      (__force u32) skb->protocol);
 238                 } else {
 239                         /* Can't derive any sort of hash for the packet, set
 240                          * to some consistent random value.
 241                          */
 242                         hash = udp_flow_hashrnd();
 243                 }
 244         }
 245
 246         /* Since this is being sent on the wire obfuscate hash a bit
 247          * to minimize possbility that any useful information to an
 248          * attacker is leaked. Only upper 16 bits are relevant in the
 249          * computation for 16 bit port value.
 250          */
 251         hash ^= hash << 16;
 252
 253         return htons((((u64) hash * (max - min)) >> 32) + min);
 254 }
 255
 256 static inline int udp_rqueue_get(struct sock *sk)
 257 {
 258         return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
 259 }
 260
 261 static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
 262                                        int dif, int sdif)
 263 {
 264 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
 265         return inet_bound_dev_eq(!!net->ipv4.sysctl_udp_l3mdev_accept,
 266                                  bound_dev_if, dif, sdif);
 267 #else
 268         return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
 269 #endif
 270 }
 271
 272 /* net/ipv4/udp.c */
 273 void udp_destruct_sock(struct sock *sk);
 274 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
 275 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
 276 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
 277 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
 278                                int noblock, int *off, int *err);
 279 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
 280                                            int noblock, int *err)
 281 {
 282         int off = 0;
 283
 284         return __skb_recv_udp(sk, flags, noblock, &off, err);
 285 }
 286
 287 int udp_v4_early_demux(struct sk_buff *skb);
 288 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
 289 int udp_get_port(struct sock *sk, unsigned short snum,
 290                  int (*saddr_cmp)(const struct sock *,
 291                                   const struct sock *));
 292 int udp_err(struct sk_buff *, u32);
 293 int udp_abort(struct sock *sk, int err);
 294 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
 295 int udp_push_pending_frames(struct sock *sk);
 296 void udp_flush_pending_frames(struct sock *sk);
 297 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
 298 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
 299 int udp_rcv(struct sk_buff *skb);
 300 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
 301 int udp_init_sock(struct sock *sk);
 302 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
 303 int __udp_disconnect(struct sock *sk, int flags);
 304 int udp_disconnect(struct sock *sk, int flags);
 305 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
 306 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
 307                                        netdev_features_t features,
 308                                        bool is_ipv6);
 309 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
 310                        char __user *optval, int __user *optlen);
 311 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
 312                        sockptr_t optval, unsigned int optlen,
 313                        int (*push_pending_frames)(struct sock *));
 314 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 315                              __be32 daddr, __be16 dport, int dif);
 316 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 317                                __be32 daddr, __be16 dport, int dif, int sdif,
 318                                struct udp_table *tbl, struct sk_buff *skb);
 319 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
 320                                  __be16 sport, __be16 dport);
 321 struct sock *udp6_lib_lookup(struct net *net,
 322                              const struct in6_addr *saddr, __be16 sport,
 323                              const struct in6_addr *daddr, __be16 dport,
 324                              int dif);
 325 struct sock *__udp6_lib_lookup(struct net *net,
 326                                const struct in6_addr *saddr, __be16 sport,
 327                                const struct in6_addr *daddr, __be16 dport,
 328                                int dif, int sdif, struct udp_table *tbl,
 329                                struct sk_buff *skb);
 330 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
 331                                  __be16 sport, __be16 dport);
 332 int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
 333                   sk_read_actor_t recv_actor);
 334
 335 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
 336  * possibly multiple cache miss on dequeue()
 337  */
 338 struct udp_dev_scratch {
 339         /* skb->truesize and the stateless bit are embedded in a single field;
 340          * do not use a bitfield since the compiler emits better/smaller code
 341          * this way
 342          */
 343         u32 _tsize_state;
 344
 345 #if BITS_PER_LONG == 64
 346         /* len and the bit needed to compute skb_csum_unnecessary
 347          * will be on cold cache lines at recvmsg time.
 348          * skb->len can be stored on 16 bits since the udp header has been
 349          * already validated and pulled.
 350          */
 351         u16 len;
 352         bool is_linear;
 353         bool csum_unnecessary;
 354 #endif
 355 };
 356
 357 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
 358 {
 359         return (struct udp_dev_scratch *)&skb->dev_scratch;
 360 }
 361
 362 #if BITS_PER_LONG == 64
 363 static inline unsigned int udp_skb_len(struct sk_buff *skb)
 364 {
 365         return udp_skb_scratch(skb)->len;
 366 }
 367
 368 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
 369 {
 370         return udp_skb_scratch(skb)->csum_unnecessary;
 371 }
 372
 373 static inline bool udp_skb_is_linear(struct sk_buff *skb)
 374 {
 375         return udp_skb_scratch(skb)->is_linear;
 376 }
 377
 378 #else
 379 static inline unsigned int udp_skb_len(struct sk_buff *skb)
 380 {
 381         return skb->len;
 382 }
 383
 384 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
 385 {
 386         return skb_csum_unnecessary(skb);
 387 }
 388
 389 static inline bool udp_skb_is_linear(struct sk_buff *skb)
 390 {
 391         return !skb_is_nonlinear(skb);
 392 }
 393 #endif
 394
 395 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
 396                                   struct iov_iter *to)
 397 {
 398         int n;
 399
 400         n = copy_to_iter(skb->data + off, len, to);
 401         if (n == len)
 402                 return 0;
 403
 404         iov_iter_revert(to, n);
 405         return -EFAULT;
 406 }
 407
 408 /*
 409  *      SNMP statistics for UDP and UDP-Lite
 410  */
 411 #define UDP_INC_STATS(net, field, is_udplite)                 do { \
 412         if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field);       \
 413         else            SNMP_INC_STATS((net)->mib.udp_statistics, field);  }  while(0)
 414 #define __UDP_INC_STATS(net, field, is_udplite)               do { \
 415         if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field);         \
 416         else            __SNMP_INC_STATS((net)->mib.udp_statistics, field);    }  while(0)
 417
 418 #define __UDP6_INC_STATS(net, field, is_udplite)            do { \
 419         if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
 420         else            __SNMP_INC_STATS((net)->mib.udp_stats_in6, field);  \
 421 } while(0)
 422 #define UDP6_INC_STATS(net, field, __lite)                  do { \
 423         if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);  \
 424         else        SNMP_INC_STATS((net)->mib.udp_stats_in6, field);      \
 425 } while(0)
 426
 427 #if IS_ENABLED(CONFIG_IPV6)
 428 #define __UDPX_MIB(sk, ipv4)                                            \
 429 ({                                                                      \
 430         ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
 431                                  sock_net(sk)->mib.udp_statistics) :    \
 432                 (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \
 433                                  sock_net(sk)->mib.udp_stats_in6);      \
 434 })
 435 #else
 436 #define __UDPX_MIB(sk, ipv4)                                            \
 437 ({                                                                      \
 438         IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :         \
 439                          sock_net(sk)->mib.udp_statistics;              \
 440 })
 441 #endif
 442
 443 #define __UDPX_INC_STATS(sk, field) \
 444         __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
 445
 446 #ifdef CONFIG_PROC_FS
 447 struct udp_seq_afinfo {
 448         sa_family_t                     family;
 449         struct udp_table                *udp_table;
 450 };
 451
 452 struct udp_iter_state {
 453         struct seq_net_private  p;
 454         int                     bucket;
 455         struct udp_seq_afinfo   *bpf_seq_afinfo;
 456 };
 457
 458 void *udp_seq_start(struct seq_file *seq, loff_t *pos);
 459 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
 460 void udp_seq_stop(struct seq_file *seq, void *v);
 461
 462 extern const struct seq_operations udp_seq_ops;
 463 extern const struct seq_operations udp6_seq_ops;
 464
 465 int udp4_proc_init(void);
 466 void udp4_proc_exit(void);
 467 #endif /* CONFIG_PROC_FS */
 468
 469 int udpv4_offload_init(void);
 470
 471 void udp_init(void);
 472
 473 DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
 474 void udp_encap_enable(void);
 475 void udp_encap_disable(void);
 476 #if IS_ENABLED(CONFIG_IPV6)
 477 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
 478 void udpv6_encap_enable(void);
 479 #endif
 480
 481 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
 482                                               struct sk_buff *skb, bool ipv4)
 483 {
 484         netdev_features_t features = NETIF_F_SG;
 485         struct sk_buff *segs;
 486
 487         /* Avoid csum recalculation by skb_segment unless userspace explicitly
 488          * asks for the final checksum values
 489          */
 490         if (!inet_get_convert_csum(sk))
 491                 features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
 492
 493         /* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
 494          * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
 495          * packets in udp_gro_complete_segment. As does UDP GSO, verified by
 496          * udp_send_skb. But when those packets are looped in dev_loopback_xmit
 497          * their ip_summed is set to CHECKSUM_UNNECESSARY. Reset in this
 498          * specific case, where PARTIAL is both correct and required.
 499          */
 500         if (skb->pkt_type == PACKET_LOOPBACK)
 501                 skb->ip_summed = CHECKSUM_PARTIAL;
 502
 503         /* the GSO CB lays after the UDP one, no need to save and restore any
 504          * CB fragment
 505          */
 506         segs = __skb_gso_segment(skb, features, false);
 507         if (IS_ERR_OR_NULL(segs)) {
 508                 int segs_nr = skb_shinfo(skb)->gso_segs;
 509
 510                 atomic_add(segs_nr, &sk->sk_drops);
 511                 SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
 512                 kfree_skb(skb);
 513                 return NULL;
 514         }
 515
 516         consume_skb(skb);
 517         return segs;
 518 }
 519
 520 static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
 521 {
 522         /* UDP-lite can't land here - no GRO */
 523         WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
 524
 525         /* UDP packets generated with UDP_SEGMENT and traversing:
 526          *
 527          * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
 528          *
 529          * can reach an UDP socket with CHECKSUM_NONE, because
 530          * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE.
 531          * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will
 532          * have a valid checksum, as the GRO engine validates the UDP csum
 533          * before the aggregation and nobody strips such info in between.
 534          * Instead of adding another check in the tunnel fastpath, we can force
 535          * a valid csum after the segmentation.
 536          * Additionally fixup the UDP CB.
 537          */
 538         UDP_SKB_CB(skb)->cscov = skb->len;
 539         if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
 540                 skb->csum_valid = 1;
 541 }
 542
 543 #ifdef CONFIG_BPF_SYSCALL
 544 struct sk_psock;
 545 struct proto *udp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
 546 int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
 547 #endif
 548
 549 #endif  /* _UDP_H */