xref: /linux/include/net/udp.h (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
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/gso.h>
25 #include <net/sock.h>
26 #include <net/snmp.h>
27 #include <net/ip.h>
28 #include <linux/ipv6.h>
29 #include <linux/seq_file.h>
30 #include <linux/poll.h>
31 #include <linux/indirect_call_wrapper.h>
32 
33 /**
34  *	struct udp_skb_cb  -  UDP(-Lite) private variables
35  *
36  *	@header:      private variables used by IPv4/IPv6
37  *	@cscov:       checksum coverage length (UDP-Lite only)
38  *	@partial_cov: if set indicates partial csum coverage
39  */
40 struct udp_skb_cb {
41 	union {
42 		struct inet_skb_parm	h4;
43 #if IS_ENABLED(CONFIG_IPV6)
44 		struct inet6_skb_parm	h6;
45 #endif
46 	} header;
47 	__u16		cscov;
48 	__u8		partial_cov;
49 };
50 #define UDP_SKB_CB(__skb)	((struct udp_skb_cb *)((__skb)->cb))
51 
52 /**
53  *	struct udp_hslot - UDP hash slot
54  *
55  *	@head:	head of list of sockets
56  *	@count:	number of sockets in 'head' list
57  *	@lock:	spinlock protecting changes to head/count
58  */
59 struct udp_hslot {
60 	struct hlist_head	head;
61 	int			count;
62 	spinlock_t		lock;
63 } __attribute__((aligned(2 * sizeof(long))));
64 
65 /**
66  *	struct udp_table - UDP table
67  *
68  *	@hash:	hash table, sockets are hashed on (local port)
69  *	@hash2:	hash table, sockets are hashed on (local port, local address)
70  *	@mask:	number of slots in hash tables, minus 1
71  *	@log:	log2(number of slots in hash table)
72  */
73 struct udp_table {
74 	struct udp_hslot	*hash;
75 	struct udp_hslot	*hash2;
76 	unsigned int		mask;
77 	unsigned int		log;
78 };
79 extern struct udp_table udp_table;
80 void udp_table_init(struct udp_table *, const char *);
81 static inline struct udp_hslot *udp_hashslot(struct udp_table *table,
82 					     const struct net *net,
83 					     unsigned int num)
84 {
85 	return &table->hash[udp_hashfn(net, num, table->mask)];
86 }
87 /*
88  * For secondary hash, net_hash_mix() is performed before calling
89  * udp_hashslot2(), this explains difference with udp_hashslot()
90  */
91 static inline struct udp_hslot *udp_hashslot2(struct udp_table *table,
92 					      unsigned int hash)
93 {
94 	return &table->hash2[hash & table->mask];
95 }
96 
97 extern struct proto udp_prot;
98 
99 extern atomic_long_t udp_memory_allocated;
100 DECLARE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
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)(const struct sk_buff *skb, __be16 sport,
171 				     __be16 dport);
172 
173 void udp_v6_early_demux(struct sk_buff *skb);
174 INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
175 
176 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
177 				  netdev_features_t features, bool is_ipv6);
178 
179 static inline void udp_lib_init_sock(struct sock *sk)
180 {
181 	struct udp_sock *up = udp_sk(sk);
182 
183 	skb_queue_head_init(&up->reader_queue);
184 	up->forward_threshold = sk->sk_rcvbuf >> 2;
185 	set_bit(SOCK_CUSTOM_SOCKOPT, &sk->sk_socket->flags);
186 }
187 
188 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
189 static inline int udp_lib_hash(struct sock *sk)
190 {
191 	BUG();
192 	return 0;
193 }
194 
195 void udp_lib_unhash(struct sock *sk);
196 void udp_lib_rehash(struct sock *sk, u16 new_hash);
197 
198 static inline void udp_lib_close(struct sock *sk, long timeout)
199 {
200 	sk_common_release(sk);
201 }
202 
203 int udp_lib_get_port(struct sock *sk, unsigned short snum,
204 		     unsigned int hash2_nulladdr);
205 
206 u32 udp_flow_hashrnd(void);
207 
208 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
209 				       int min, int max, bool use_eth)
210 {
211 	u32 hash;
212 
213 	if (min >= max) {
214 		/* Use default range */
215 		inet_get_local_port_range(net, &min, &max);
216 	}
217 
218 	hash = skb_get_hash(skb);
219 	if (unlikely(!hash)) {
220 		if (use_eth) {
221 			/* Can't find a normal hash, caller has indicated an
222 			 * Ethernet packet so use that to compute a hash.
223 			 */
224 			hash = jhash(skb->data, 2 * ETH_ALEN,
225 				     (__force u32) skb->protocol);
226 		} else {
227 			/* Can't derive any sort of hash for the packet, set
228 			 * to some consistent random value.
229 			 */
230 			hash = udp_flow_hashrnd();
231 		}
232 	}
233 
234 	/* Since this is being sent on the wire obfuscate hash a bit
235 	 * to minimize possibility that any useful information to an
236 	 * attacker is leaked. Only upper 16 bits are relevant in the
237 	 * computation for 16 bit port value.
238 	 */
239 	hash ^= hash << 16;
240 
241 	return htons((((u64) hash * (max - min)) >> 32) + min);
242 }
243 
244 static inline int udp_rqueue_get(struct sock *sk)
245 {
246 	return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
247 }
248 
249 static inline bool udp_sk_bound_dev_eq(const struct net *net, int bound_dev_if,
250 				       int dif, int sdif)
251 {
252 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
253 	return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_udp_l3mdev_accept),
254 				 bound_dev_if, dif, sdif);
255 #else
256 	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
257 #endif
258 }
259 
260 /* net/ipv4/udp.c */
261 void udp_destruct_common(struct sock *sk);
262 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
263 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
264 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
265 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, int *off,
266 			       int *err);
267 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
268 					   int *err)
269 {
270 	int off = 0;
271 
272 	return __skb_recv_udp(sk, flags, &off, err);
273 }
274 
275 int udp_v4_early_demux(struct sk_buff *skb);
276 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
277 int udp_err(struct sk_buff *, u32);
278 int udp_abort(struct sock *sk, int err);
279 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
280 void udp_splice_eof(struct socket *sock);
281 int udp_push_pending_frames(struct sock *sk);
282 void udp_flush_pending_frames(struct sock *sk);
283 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
284 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
285 int udp_rcv(struct sk_buff *skb);
286 int udp_ioctl(struct sock *sk, int cmd, int *karg);
287 int udp_init_sock(struct sock *sk);
288 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
289 int __udp_disconnect(struct sock *sk, int flags);
290 int udp_disconnect(struct sock *sk, int flags);
291 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
292 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
293 				       netdev_features_t features,
294 				       bool is_ipv6);
295 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
296 		       char __user *optval, int __user *optlen);
297 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
298 		       sockptr_t optval, unsigned int optlen,
299 		       int (*push_pending_frames)(struct sock *));
300 struct sock *udp4_lib_lookup(const struct net *net, __be32 saddr, __be16 sport,
301 			     __be32 daddr, __be16 dport, int dif);
302 struct sock *__udp4_lib_lookup(const struct net *net, __be32 saddr,
303 			       __be16 sport,
304 			       __be32 daddr, __be16 dport, int dif, int sdif,
305 			       struct udp_table *tbl, struct sk_buff *skb);
306 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
307 				 __be16 sport, __be16 dport);
308 struct sock *udp6_lib_lookup(const struct net *net,
309 			     const struct in6_addr *saddr, __be16 sport,
310 			     const struct in6_addr *daddr, __be16 dport,
311 			     int dif);
312 struct sock *__udp6_lib_lookup(const struct net *net,
313 			       const struct in6_addr *saddr, __be16 sport,
314 			       const struct in6_addr *daddr, __be16 dport,
315 			       int dif, int sdif, struct udp_table *tbl,
316 			       struct sk_buff *skb);
317 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
318 				 __be16 sport, __be16 dport);
319 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
320 
321 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
322  * possibly multiple cache miss on dequeue()
323  */
324 struct udp_dev_scratch {
325 	/* skb->truesize and the stateless bit are embedded in a single field;
326 	 * do not use a bitfield since the compiler emits better/smaller code
327 	 * this way
328 	 */
329 	u32 _tsize_state;
330 
331 #if BITS_PER_LONG == 64
332 	/* len and the bit needed to compute skb_csum_unnecessary
333 	 * will be on cold cache lines at recvmsg time.
334 	 * skb->len can be stored on 16 bits since the udp header has been
335 	 * already validated and pulled.
336 	 */
337 	u16 len;
338 	bool is_linear;
339 	bool csum_unnecessary;
340 #endif
341 };
342 
343 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
344 {
345 	return (struct udp_dev_scratch *)&skb->dev_scratch;
346 }
347 
348 #if BITS_PER_LONG == 64
349 static inline unsigned int udp_skb_len(struct sk_buff *skb)
350 {
351 	return udp_skb_scratch(skb)->len;
352 }
353 
354 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
355 {
356 	return udp_skb_scratch(skb)->csum_unnecessary;
357 }
358 
359 static inline bool udp_skb_is_linear(struct sk_buff *skb)
360 {
361 	return udp_skb_scratch(skb)->is_linear;
362 }
363 
364 #else
365 static inline unsigned int udp_skb_len(struct sk_buff *skb)
366 {
367 	return skb->len;
368 }
369 
370 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
371 {
372 	return skb_csum_unnecessary(skb);
373 }
374 
375 static inline bool udp_skb_is_linear(struct sk_buff *skb)
376 {
377 	return !skb_is_nonlinear(skb);
378 }
379 #endif
380 
381 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
382 				  struct iov_iter *to)
383 {
384 	return copy_to_iter_full(skb->data + off, len, to) ? 0 : -EFAULT;
385 }
386 
387 /*
388  * 	SNMP statistics for UDP and UDP-Lite
389  */
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)
393 #define __UDP_INC_STATS(net, field, is_udplite) 	      do { \
394 	if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field);         \
395 	else		__SNMP_INC_STATS((net)->mib.udp_statistics, field);    }  while(0)
396 
397 #define __UDP6_INC_STATS(net, field, is_udplite)	    do { \
398 	if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
399 	else		__SNMP_INC_STATS((net)->mib.udp_stats_in6, field);  \
400 } while(0)
401 #define UDP6_INC_STATS(net, field, __lite)		    do { \
402 	if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);  \
403 	else	    SNMP_INC_STATS((net)->mib.udp_stats_in6, field);      \
404 } while(0)
405 
406 #if IS_ENABLED(CONFIG_IPV6)
407 #define __UDPX_MIB(sk, ipv4)						\
408 ({									\
409 	ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :	\
410 				 sock_net(sk)->mib.udp_statistics) :	\
411 		(IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 :	\
412 				 sock_net(sk)->mib.udp_stats_in6);	\
413 })
414 #else
415 #define __UDPX_MIB(sk, ipv4)						\
416 ({									\
417 	IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :		\
418 			 sock_net(sk)->mib.udp_statistics;		\
419 })
420 #endif
421 
422 #define __UDPX_INC_STATS(sk, field) \
423 	__SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
424 
425 #ifdef CONFIG_PROC_FS
426 struct udp_seq_afinfo {
427 	sa_family_t			family;
428 	struct udp_table		*udp_table;
429 };
430 
431 struct udp_iter_state {
432 	struct seq_net_private  p;
433 	int			bucket;
434 };
435 
436 void *udp_seq_start(struct seq_file *seq, loff_t *pos);
437 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
438 void udp_seq_stop(struct seq_file *seq, void *v);
439 
440 extern const struct seq_operations udp_seq_ops;
441 extern const struct seq_operations udp6_seq_ops;
442 
443 int udp4_proc_init(void);
444 void udp4_proc_exit(void);
445 #endif /* CONFIG_PROC_FS */
446 
447 int udpv4_offload_init(void);
448 
449 void udp_init(void);
450 
451 DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
452 void udp_encap_enable(void);
453 void udp_encap_disable(void);
454 #if IS_ENABLED(CONFIG_IPV6)
455 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
456 void udpv6_encap_enable(void);
457 #endif
458 
459 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
460 					      struct sk_buff *skb, bool ipv4)
461 {
462 	netdev_features_t features = NETIF_F_SG;
463 	struct sk_buff *segs;
464 
465 	/* Avoid csum recalculation by skb_segment unless userspace explicitly
466 	 * asks for the final checksum values
467 	 */
468 	if (!inet_get_convert_csum(sk))
469 		features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
470 
471 	/* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
472 	 * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
473 	 * packets in udp_gro_complete_segment. As does UDP GSO, verified by
474 	 * udp_send_skb. But when those packets are looped in dev_loopback_xmit
475 	 * their ip_summed CHECKSUM_NONE is changed to CHECKSUM_UNNECESSARY.
476 	 * Reset in this specific case, where PARTIAL is both correct and
477 	 * required.
478 	 */
479 	if (skb->pkt_type == PACKET_LOOPBACK)
480 		skb->ip_summed = CHECKSUM_PARTIAL;
481 
482 	/* the GSO CB lays after the UDP one, no need to save and restore any
483 	 * CB fragment
484 	 */
485 	segs = __skb_gso_segment(skb, features, false);
486 	if (IS_ERR_OR_NULL(segs)) {
487 		int segs_nr = skb_shinfo(skb)->gso_segs;
488 
489 		atomic_add(segs_nr, &sk->sk_drops);
490 		SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
491 		kfree_skb(skb);
492 		return NULL;
493 	}
494 
495 	consume_skb(skb);
496 	return segs;
497 }
498 
499 static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
500 {
501 	/* UDP-lite can't land here - no GRO */
502 	WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
503 
504 	/* UDP packets generated with UDP_SEGMENT and traversing:
505 	 *
506 	 * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
507 	 *
508 	 * can reach an UDP socket with CHECKSUM_NONE, because
509 	 * __iptunnel_pull_header() converts CHECKSUM_PARTIAL into NONE.
510 	 * SKB_GSO_UDP_L4 or SKB_GSO_FRAGLIST packets with no UDP tunnel will
511 	 * have a valid checksum, as the GRO engine validates the UDP csum
512 	 * before the aggregation and nobody strips such info in between.
513 	 * Instead of adding another check in the tunnel fastpath, we can force
514 	 * a valid csum after the segmentation.
515 	 * Additionally fixup the UDP CB.
516 	 */
517 	UDP_SKB_CB(skb)->cscov = skb->len;
518 	if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
519 		skb->csum_valid = 1;
520 }
521 
522 #ifdef CONFIG_BPF_SYSCALL
523 struct sk_psock;
524 int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
525 #endif
526 
527 #endif	/* _UDP_H */
528