1 /* SPDX-License-Identifier: GPL-2.0-or-later */
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.
7 * Definitions for the UDP module.
9 * Version: @(#)udp.h 1.0.2 05/07/93
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
15 * Alan Cox : Turned on udp checksums. I don't want to
16 * chase 'memory corruption' bugs that aren't!
21 #include <linux/list.h>
22 #include <linux/bug.h>
23 #include <net/inet_sock.h>
27 #include <linux/ipv6.h>
28 #include <linux/seq_file.h>
29 #include <linux/poll.h>
30 #include <linux/indirect_call_wrapper.h>
33 * struct udp_skb_cb - UDP(-Lite) private variables
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
41 struct inet_skb_parm h4;
42 #if IS_ENABLED(CONFIG_IPV6)
43 struct inet6_skb_parm h6;
49 #define UDP_SKB_CB(__skb) ((struct udp_skb_cb *)((__skb)->cb))
52 * struct udp_hslot - UDP hash slot
54 * @head: head of list of sockets
55 * @count: number of sockets in 'head' list
56 * @lock: spinlock protecting changes to head/count
59 struct hlist_head head;
62 } __attribute__((aligned(2 * sizeof(long))));
65 * struct udp_table - UDP table
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)
73 struct udp_hslot *hash;
74 struct udp_hslot *hash2;
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)
83 return &table->hash[udp_hashfn(net, num, table->mask)];
86 * For secondary hash, net_hash_mix() is performed before calling
87 * udp_hashslot2(), this explains difference with udp_hashslot()
89 static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
92 return &table->hash2[hash & table->mask];
95 extern struct proto udp_prot;
97 extern atomic_long_t udp_memory_allocated;
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;
107 * Generic checksumming routines for UDP(-Lite) v4 and v6
109 static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
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));
116 static inline int udp_lib_checksum_complete(struct sk_buff *skb)
118 return !skb_csum_unnecessary(skb) &&
119 __udp_lib_checksum_complete(skb);
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)
128 static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
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);
138 static inline __wsum udp_csum(struct sk_buff *skb)
140 __wsum csum = csum_partial(skb_transport_header(skb),
141 sizeof(struct udphdr), skb->csum);
143 for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
144 csum = csum_add(csum, skb->csum);
149 static inline __sum16 udp_v4_check(int len, __be32 saddr,
150 __be32 daddr, __wsum base)
152 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
155 void udp_set_csum(bool nocheck, struct sk_buff *skb,
156 __be32 saddr, __be32 daddr, int len);
158 static inline void udp_csum_pull_header(struct sk_buff *skb)
160 if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
161 skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
163 skb_pull_rcsum(skb, sizeof(struct udphdr));
164 UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
167 typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport,
170 void udp_v6_early_demux(struct sk_buff *skb);
171 INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
173 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
174 netdev_features_t features, bool is_ipv6);
176 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
177 static inline int udp_lib_hash(struct sock *sk)
183 void udp_lib_unhash(struct sock *sk);
184 void udp_lib_rehash(struct sock *sk, u16 new_hash);
186 static inline void udp_lib_close(struct sock *sk, long timeout)
188 sk_common_release(sk);
191 int udp_lib_get_port(struct sock *sk, unsigned short snum,
192 unsigned int hash2_nulladdr);
194 u32 udp_flow_hashrnd(void);
196 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
197 int min, int max, bool use_eth)
202 /* Use default range */
203 inet_get_local_port_range(net, &min, &max);
206 hash = skb_get_hash(skb);
207 if (unlikely(!hash)) {
209 /* Can't find a normal hash, caller has indicated an
210 * Ethernet packet so use that to compute a hash.
212 hash = jhash(skb->data, 2 * ETH_ALEN,
213 (__force u32) skb->protocol);
215 /* Can't derive any sort of hash for the packet, set
216 * to some consistent random value.
218 hash = udp_flow_hashrnd();
222 /* Since this is being sent on the wire obfuscate hash a bit
223 * to minimize possbility that any useful information to an
224 * attacker is leaked. Only upper 16 bits are relevant in the
225 * computation for 16 bit port value.
229 return htons((((u64) hash * (max - min)) >> 32) + min);
232 static inline int udp_rqueue_get(struct sock *sk)
234 return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
237 static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
240 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
241 return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_udp_l3mdev_accept),
242 bound_dev_if, dif, sdif);
244 return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
249 void udp_destruct_sock(struct sock *sk);
250 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
251 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
252 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
253 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, int *off,
255 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
260 return __skb_recv_udp(sk, flags, &off, err);
263 int udp_v4_early_demux(struct sk_buff *skb);
264 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
265 int udp_get_port(struct sock *sk, unsigned short snum,
266 int (*saddr_cmp)(const struct sock *,
267 const struct sock *));
268 int udp_err(struct sk_buff *, u32);
269 int udp_abort(struct sock *sk, int err);
270 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
271 int udp_push_pending_frames(struct sock *sk);
272 void udp_flush_pending_frames(struct sock *sk);
273 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
274 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
275 int udp_rcv(struct sk_buff *skb);
276 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
277 int udp_init_sock(struct sock *sk);
278 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
279 int __udp_disconnect(struct sock *sk, int flags);
280 int udp_disconnect(struct sock *sk, int flags);
281 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
282 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
283 netdev_features_t features,
285 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
286 char __user *optval, int __user *optlen);
287 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
288 sockptr_t optval, unsigned int optlen,
289 int (*push_pending_frames)(struct sock *));
290 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
291 __be32 daddr, __be16 dport, int dif);
292 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
293 __be32 daddr, __be16 dport, int dif, int sdif,
294 struct udp_table *tbl, struct sk_buff *skb);
295 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
296 __be16 sport, __be16 dport);
297 struct sock *udp6_lib_lookup(struct net *net,
298 const struct in6_addr *saddr, __be16 sport,
299 const struct in6_addr *daddr, __be16 dport,
301 struct sock *__udp6_lib_lookup(struct net *net,
302 const struct in6_addr *saddr, __be16 sport,
303 const struct in6_addr *daddr, __be16 dport,
304 int dif, int sdif, struct udp_table *tbl,
305 struct sk_buff *skb);
306 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
307 __be16 sport, __be16 dport);
308 int udp_read_sock(struct sock *sk, read_descriptor_t *desc,
309 sk_read_actor_t recv_actor);
311 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
312 * possibly multiple cache miss on dequeue()
314 struct udp_dev_scratch {
315 /* skb->truesize and the stateless bit are embedded in a single field;
316 * do not use a bitfield since the compiler emits better/smaller code
321 #if BITS_PER_LONG == 64
322 /* len and the bit needed to compute skb_csum_unnecessary
323 * will be on cold cache lines at recvmsg time.
324 * skb->len can be stored on 16 bits since the udp header has been
325 * already validated and pulled.
329 bool csum_unnecessary;
333 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
335 return (struct udp_dev_scratch *)&skb->dev_scratch;
338 #if BITS_PER_LONG == 64
339 static inline unsigned int udp_skb_len(struct sk_buff *skb)
341 return udp_skb_scratch(skb)->len;
344 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
346 return udp_skb_scratch(skb)->csum_unnecessary;
349 static inline bool udp_skb_is_linear(struct sk_buff *skb)
351 return udp_skb_scratch(skb)->is_linear;
355 static inline unsigned int udp_skb_len(struct sk_buff *skb)
360 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
362 return skb_csum_unnecessary(skb);
365 static inline bool udp_skb_is_linear(struct sk_buff *skb)
367 return !skb_is_nonlinear(skb);
371 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
376 n = copy_to_iter(skb->data + off, len, to);
380 iov_iter_revert(to, n);
385 * SNMP statistics for UDP and UDP-Lite
387 #define UDP_INC_STATS(net, field, is_udplite) do { \
388 if (is_udplite) SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
389 else SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
390 #define __UDP_INC_STATS(net, field, is_udplite) do { \
391 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field); \
392 else __SNMP_INC_STATS((net)->mib.udp_statistics, field); } while(0)
394 #define __UDP6_INC_STATS(net, field, is_udplite) do { \
395 if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
396 else __SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
398 #define UDP6_INC_STATS(net, field, __lite) do { \
399 if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field); \
400 else SNMP_INC_STATS((net)->mib.udp_stats_in6, field); \
403 #if IS_ENABLED(CONFIG_IPV6)
404 #define __UDPX_MIB(sk, ipv4) \
406 ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
407 sock_net(sk)->mib.udp_statistics) : \
408 (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 : \
409 sock_net(sk)->mib.udp_stats_in6); \
412 #define __UDPX_MIB(sk, ipv4) \
414 IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics : \
415 sock_net(sk)->mib.udp_statistics; \
419 #define __UDPX_INC_STATS(sk, field) \
420 __SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
422 #ifdef CONFIG_PROC_FS
423 struct udp_seq_afinfo {
425 struct udp_table *udp_table;
428 struct udp_iter_state {
429 struct seq_net_private p;
431 struct udp_seq_afinfo *bpf_seq_afinfo;
434 void *udp_seq_start(struct seq_file *seq, loff_t *pos);
435 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
436 void udp_seq_stop(struct seq_file *seq, void *v);
438 extern const struct seq_operations udp_seq_ops;
439 extern const struct seq_operations udp6_seq_ops;
441 int udp4_proc_init(void);
442 void udp4_proc_exit(void);
443 #endif /* CONFIG_PROC_FS */
445 int udpv4_offload_init(void);
449 DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
450 void udp_encap_enable(void);
451 void udp_encap_disable(void);
452 #if IS_ENABLED(CONFIG_IPV6)
453 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
454 void udpv6_encap_enable(void);
457 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
458 struct sk_buff *skb, bool ipv4)
460 netdev_features_t features = NETIF_F_SG;
461 struct sk_buff *segs;
463 /* Avoid csum recalculation by skb_segment unless userspace explicitly
464 * asks for the final checksum values
466 if (!inet_get_convert_csum(sk))
467 features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
469 /* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
470 * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
471 * packets in udp_gro_complete_segment. As does UDP GSO, verified by
472 * udp_send_skb. But when those packets are looped in dev_loopback_xmit
473 * their ip_summed CHECKSUM_NONE is changed to CHECKSUM_UNNECESSARY.
474 * Reset in this specific case, where PARTIAL is both correct and
477 if (skb->pkt_type == PACKET_LOOPBACK)
478 skb->ip_summed = CHECKSUM_PARTIAL;
480 /* the GSO CB lays after the UDP one, no need to save and restore any
483 segs = __skb_gso_segment(skb, features, false);
484 if (IS_ERR_OR_NULL(segs)) {
485 int segs_nr = skb_shinfo(skb)->gso_segs;
487 atomic_add(segs_nr, &sk->sk_drops);
488 SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
497 static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
499 /* UDP-lite can't land here - no GRO */
500 WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
502 /* UDP packets generated with UDP_SEGMENT and traversing:
504 * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
506 * can reach an UDP socket with CHECKSUM_NONE, because
507 * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE.
508 * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will
509 * have a valid checksum, as the GRO engine validates the UDP csum
510 * before the aggregation and nobody strips such info in between.
511 * Instead of adding another check in the tunnel fastpath, we can force
512 * a valid csum after the segmentation.
513 * Additionally fixup the UDP CB.
515 UDP_SKB_CB(skb)->cscov = skb->len;
516 if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
520 #ifdef CONFIG_BPF_SYSCALL
522 struct proto *udp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
523 int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);