xref: /linux/include/net/udp.h (revision e1c4c5436b4ad579762fbe78bfabc8aef59bd5b1)
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 DECLARE_PER_CPU(int, udp_memory_per_cpu_fw_alloc);
99 
100 /* sysctl variables for udp */
101 extern long sysctl_udp_mem[3];
102 extern int sysctl_udp_rmem_min;
103 extern int sysctl_udp_wmem_min;
104 
105 struct sk_buff;
106 
107 /*
108  *	Generic checksumming routines for UDP(-Lite) v4 and v6
109  */
110 static inline __sum16 __udp_lib_checksum_complete(struct sk_buff *skb)
111 {
112 	return (UDP_SKB_CB(skb)->cscov == skb->len ?
113 		__skb_checksum_complete(skb) :
114 		__skb_checksum_complete_head(skb, UDP_SKB_CB(skb)->cscov));
115 }
116 
117 static inline int udp_lib_checksum_complete(struct sk_buff *skb)
118 {
119 	return !skb_csum_unnecessary(skb) &&
120 		__udp_lib_checksum_complete(skb);
121 }
122 
123 /**
124  * 	udp_csum_outgoing  -  compute UDPv4/v6 checksum over fragments
125  * 	@sk: 	socket we are writing to
126  * 	@skb: 	sk_buff containing the filled-in UDP header
127  * 	        (checksum field must be zeroed out)
128  */
129 static inline __wsum udp_csum_outgoing(struct sock *sk, struct sk_buff *skb)
130 {
131 	__wsum csum = csum_partial(skb_transport_header(skb),
132 				   sizeof(struct udphdr), 0);
133 	skb_queue_walk(&sk->sk_write_queue, skb) {
134 		csum = csum_add(csum, skb->csum);
135 	}
136 	return csum;
137 }
138 
139 static inline __wsum udp_csum(struct sk_buff *skb)
140 {
141 	__wsum csum = csum_partial(skb_transport_header(skb),
142 				   sizeof(struct udphdr), skb->csum);
143 
144 	for (skb = skb_shinfo(skb)->frag_list; skb; skb = skb->next) {
145 		csum = csum_add(csum, skb->csum);
146 	}
147 	return csum;
148 }
149 
150 static inline __sum16 udp_v4_check(int len, __be32 saddr,
151 				   __be32 daddr, __wsum base)
152 {
153 	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base);
154 }
155 
156 void udp_set_csum(bool nocheck, struct sk_buff *skb,
157 		  __be32 saddr, __be32 daddr, int len);
158 
159 static inline void udp_csum_pull_header(struct sk_buff *skb)
160 {
161 	if (!skb->csum_valid && skb->ip_summed == CHECKSUM_NONE)
162 		skb->csum = csum_partial(skb->data, sizeof(struct udphdr),
163 					 skb->csum);
164 	skb_pull_rcsum(skb, sizeof(struct udphdr));
165 	UDP_SKB_CB(skb)->cscov -= sizeof(struct udphdr);
166 }
167 
168 typedef struct sock *(*udp_lookup_t)(const struct sk_buff *skb, __be16 sport,
169 				     __be16 dport);
170 
171 void udp_v6_early_demux(struct sk_buff *skb);
172 INDIRECT_CALLABLE_DECLARE(int udpv6_rcv(struct sk_buff *));
173 
174 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
175 				  netdev_features_t features, bool is_ipv6);
176 
177 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
178 static inline int udp_lib_hash(struct sock *sk)
179 {
180 	BUG();
181 	return 0;
182 }
183 
184 void udp_lib_unhash(struct sock *sk);
185 void udp_lib_rehash(struct sock *sk, u16 new_hash);
186 
187 static inline void udp_lib_close(struct sock *sk, long timeout)
188 {
189 	sk_common_release(sk);
190 }
191 
192 int udp_lib_get_port(struct sock *sk, unsigned short snum,
193 		     unsigned int hash2_nulladdr);
194 
195 u32 udp_flow_hashrnd(void);
196 
197 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
198 				       int min, int max, bool use_eth)
199 {
200 	u32 hash;
201 
202 	if (min >= max) {
203 		/* Use default range */
204 		inet_get_local_port_range(net, &min, &max);
205 	}
206 
207 	hash = skb_get_hash(skb);
208 	if (unlikely(!hash)) {
209 		if (use_eth) {
210 			/* Can't find a normal hash, caller has indicated an
211 			 * Ethernet packet so use that to compute a hash.
212 			 */
213 			hash = jhash(skb->data, 2 * ETH_ALEN,
214 				     (__force u32) skb->protocol);
215 		} else {
216 			/* Can't derive any sort of hash for the packet, set
217 			 * to some consistent random value.
218 			 */
219 			hash = udp_flow_hashrnd();
220 		}
221 	}
222 
223 	/* Since this is being sent on the wire obfuscate hash a bit
224 	 * to minimize possbility that any useful information to an
225 	 * attacker is leaked. Only upper 16 bits are relevant in the
226 	 * computation for 16 bit port value.
227 	 */
228 	hash ^= hash << 16;
229 
230 	return htons((((u64) hash * (max - min)) >> 32) + min);
231 }
232 
233 static inline int udp_rqueue_get(struct sock *sk)
234 {
235 	return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
236 }
237 
238 static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
239 				       int dif, int sdif)
240 {
241 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
242 	return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_udp_l3mdev_accept),
243 				 bound_dev_if, dif, sdif);
244 #else
245 	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
246 #endif
247 }
248 
249 /* net/ipv4/udp.c */
250 void udp_destruct_common(struct sock *sk);
251 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
252 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
253 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
254 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags, int *off,
255 			       int *err);
256 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
257 					   int *err)
258 {
259 	int off = 0;
260 
261 	return __skb_recv_udp(sk, flags, &off, err);
262 }
263 
264 int udp_v4_early_demux(struct sk_buff *skb);
265 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
266 int udp_get_port(struct sock *sk, unsigned short snum,
267 		 int (*saddr_cmp)(const struct sock *,
268 				  const struct sock *));
269 int udp_err(struct sk_buff *, u32);
270 int udp_abort(struct sock *sk, int err);
271 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
272 int udp_push_pending_frames(struct sock *sk);
273 void udp_flush_pending_frames(struct sock *sk);
274 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
275 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
276 int udp_rcv(struct sk_buff *skb);
277 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
278 int udp_init_sock(struct sock *sk);
279 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
280 int __udp_disconnect(struct sock *sk, int flags);
281 int udp_disconnect(struct sock *sk, int flags);
282 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
283 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
284 				       netdev_features_t features,
285 				       bool is_ipv6);
286 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
287 		       char __user *optval, int __user *optlen);
288 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
289 		       sockptr_t optval, unsigned int optlen,
290 		       int (*push_pending_frames)(struct sock *));
291 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
292 			     __be32 daddr, __be16 dport, int dif);
293 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
294 			       __be32 daddr, __be16 dport, int dif, int sdif,
295 			       struct udp_table *tbl, struct sk_buff *skb);
296 struct sock *udp4_lib_lookup_skb(const struct sk_buff *skb,
297 				 __be16 sport, __be16 dport);
298 struct sock *udp6_lib_lookup(struct net *net,
299 			     const struct in6_addr *saddr, __be16 sport,
300 			     const struct in6_addr *daddr, __be16 dport,
301 			     int dif);
302 struct sock *__udp6_lib_lookup(struct net *net,
303 			       const struct in6_addr *saddr, __be16 sport,
304 			       const struct in6_addr *daddr, __be16 dport,
305 			       int dif, int sdif, struct udp_table *tbl,
306 			       struct sk_buff *skb);
307 struct sock *udp6_lib_lookup_skb(const struct sk_buff *skb,
308 				 __be16 sport, __be16 dport);
309 int udp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
310 
311 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
312  * possibly multiple cache miss on dequeue()
313  */
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
317 	 * this way
318 	 */
319 	u32 _tsize_state;
320 
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.
326 	 */
327 	u16 len;
328 	bool is_linear;
329 	bool csum_unnecessary;
330 #endif
331 };
332 
333 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
334 {
335 	return (struct udp_dev_scratch *)&skb->dev_scratch;
336 }
337 
338 #if BITS_PER_LONG == 64
339 static inline unsigned int udp_skb_len(struct sk_buff *skb)
340 {
341 	return udp_skb_scratch(skb)->len;
342 }
343 
344 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
345 {
346 	return udp_skb_scratch(skb)->csum_unnecessary;
347 }
348 
349 static inline bool udp_skb_is_linear(struct sk_buff *skb)
350 {
351 	return udp_skb_scratch(skb)->is_linear;
352 }
353 
354 #else
355 static inline unsigned int udp_skb_len(struct sk_buff *skb)
356 {
357 	return skb->len;
358 }
359 
360 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
361 {
362 	return skb_csum_unnecessary(skb);
363 }
364 
365 static inline bool udp_skb_is_linear(struct sk_buff *skb)
366 {
367 	return !skb_is_nonlinear(skb);
368 }
369 #endif
370 
371 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
372 				  struct iov_iter *to)
373 {
374 	int n;
375 
376 	n = copy_to_iter(skb->data + off, len, to);
377 	if (n == len)
378 		return 0;
379 
380 	iov_iter_revert(to, n);
381 	return -EFAULT;
382 }
383 
384 /*
385  * 	SNMP statistics for UDP and UDP-Lite
386  */
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)
393 
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);  \
397 } while(0)
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);      \
401 } while(0)
402 
403 #if IS_ENABLED(CONFIG_IPV6)
404 #define __UDPX_MIB(sk, ipv4)						\
405 ({									\
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);	\
410 })
411 #else
412 #define __UDPX_MIB(sk, ipv4)						\
413 ({									\
414 	IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :		\
415 			 sock_net(sk)->mib.udp_statistics;		\
416 })
417 #endif
418 
419 #define __UDPX_INC_STATS(sk, field) \
420 	__SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
421 
422 #ifdef CONFIG_PROC_FS
423 struct udp_seq_afinfo {
424 	sa_family_t			family;
425 	struct udp_table		*udp_table;
426 };
427 
428 struct udp_iter_state {
429 	struct seq_net_private  p;
430 	int			bucket;
431 	struct udp_seq_afinfo	*bpf_seq_afinfo;
432 };
433 
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);
437 
438 extern const struct seq_operations udp_seq_ops;
439 extern const struct seq_operations udp6_seq_ops;
440 
441 int udp4_proc_init(void);
442 void udp4_proc_exit(void);
443 #endif /* CONFIG_PROC_FS */
444 
445 int udpv4_offload_init(void);
446 
447 void udp_init(void);
448 
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);
455 #endif
456 
457 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
458 					      struct sk_buff *skb, bool ipv4)
459 {
460 	netdev_features_t features = NETIF_F_SG;
461 	struct sk_buff *segs;
462 
463 	/* Avoid csum recalculation by skb_segment unless userspace explicitly
464 	 * asks for the final checksum values
465 	 */
466 	if (!inet_get_convert_csum(sk))
467 		features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
468 
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
475 	 * required.
476 	 */
477 	if (skb->pkt_type == PACKET_LOOPBACK)
478 		skb->ip_summed = CHECKSUM_PARTIAL;
479 
480 	/* the GSO CB lays after the UDP one, no need to save and restore any
481 	 * CB fragment
482 	 */
483 	segs = __skb_gso_segment(skb, features, false);
484 	if (IS_ERR_OR_NULL(segs)) {
485 		int segs_nr = skb_shinfo(skb)->gso_segs;
486 
487 		atomic_add(segs_nr, &sk->sk_drops);
488 		SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
489 		kfree_skb(skb);
490 		return NULL;
491 	}
492 
493 	consume_skb(skb);
494 	return segs;
495 }
496 
497 static inline void udp_post_segment_fix_csum(struct sk_buff *skb)
498 {
499 	/* UDP-lite can't land here - no GRO */
500 	WARN_ON_ONCE(UDP_SKB_CB(skb)->partial_cov);
501 
502 	/* UDP packets generated with UDP_SEGMENT and traversing:
503 	 *
504 	 * UDP tunnel(xmit) -> veth (segmentation) -> veth (gro) -> UDP tunnel (rx)
505 	 *
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.
514 	 */
515 	UDP_SKB_CB(skb)->cscov = skb->len;
516 	if (skb->ip_summed == CHECKSUM_NONE && !skb->csum_valid)
517 		skb->csum_valid = 1;
518 }
519 
520 #ifdef CONFIG_BPF_SYSCALL
521 struct sk_psock;
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);
524 #endif
525 
526 #endif	/* _UDP_H */
527