xref: /openbmc/linux/include/net/udp.h (revision b03afaa8)
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)(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 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
177 				struct udphdr *uh, struct sock *sk);
178 int udp_gro_complete(struct sk_buff *skb, int nhoff, udp_lookup_t lookup);
179 
180 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
181 				  netdev_features_t features);
182 
183 static inline struct udphdr *udp_gro_udphdr(struct sk_buff *skb)
184 {
185 	struct udphdr *uh;
186 	unsigned int hlen, off;
187 
188 	off  = skb_gro_offset(skb);
189 	hlen = off + sizeof(*uh);
190 	uh   = skb_gro_header_fast(skb, off);
191 	if (skb_gro_header_hard(skb, hlen))
192 		uh = skb_gro_header_slow(skb, hlen, off);
193 
194 	return uh;
195 }
196 
197 /* hash routines shared between UDPv4/6 and UDP-Litev4/6 */
198 static inline int udp_lib_hash(struct sock *sk)
199 {
200 	BUG();
201 	return 0;
202 }
203 
204 void udp_lib_unhash(struct sock *sk);
205 void udp_lib_rehash(struct sock *sk, u16 new_hash);
206 
207 static inline void udp_lib_close(struct sock *sk, long timeout)
208 {
209 	sk_common_release(sk);
210 }
211 
212 int udp_lib_get_port(struct sock *sk, unsigned short snum,
213 		     unsigned int hash2_nulladdr);
214 
215 u32 udp_flow_hashrnd(void);
216 
217 static inline __be16 udp_flow_src_port(struct net *net, struct sk_buff *skb,
218 				       int min, int max, bool use_eth)
219 {
220 	u32 hash;
221 
222 	if (min >= max) {
223 		/* Use default range */
224 		inet_get_local_port_range(net, &min, &max);
225 	}
226 
227 	hash = skb_get_hash(skb);
228 	if (unlikely(!hash)) {
229 		if (use_eth) {
230 			/* Can't find a normal hash, caller has indicated an
231 			 * Ethernet packet so use that to compute a hash.
232 			 */
233 			hash = jhash(skb->data, 2 * ETH_ALEN,
234 				     (__force u32) skb->protocol);
235 		} else {
236 			/* Can't derive any sort of hash for the packet, set
237 			 * to some consistent random value.
238 			 */
239 			hash = udp_flow_hashrnd();
240 		}
241 	}
242 
243 	/* Since this is being sent on the wire obfuscate hash a bit
244 	 * to minimize possbility that any useful information to an
245 	 * attacker is leaked. Only upper 16 bits are relevant in the
246 	 * computation for 16 bit port value.
247 	 */
248 	hash ^= hash << 16;
249 
250 	return htons((((u64) hash * (max - min)) >> 32) + min);
251 }
252 
253 static inline int udp_rqueue_get(struct sock *sk)
254 {
255 	return sk_rmem_alloc_get(sk) - READ_ONCE(udp_sk(sk)->forward_deficit);
256 }
257 
258 static inline bool udp_sk_bound_dev_eq(struct net *net, int bound_dev_if,
259 				       int dif, int sdif)
260 {
261 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
262 	return inet_bound_dev_eq(!!net->ipv4.sysctl_udp_l3mdev_accept,
263 				 bound_dev_if, dif, sdif);
264 #else
265 	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
266 #endif
267 }
268 
269 /* net/ipv4/udp.c */
270 void udp_destruct_sock(struct sock *sk);
271 void skb_consume_udp(struct sock *sk, struct sk_buff *skb, int len);
272 int __udp_enqueue_schedule_skb(struct sock *sk, struct sk_buff *skb);
273 void udp_skb_destructor(struct sock *sk, struct sk_buff *skb);
274 struct sk_buff *__skb_recv_udp(struct sock *sk, unsigned int flags,
275 			       int noblock, int *off, int *err);
276 static inline struct sk_buff *skb_recv_udp(struct sock *sk, unsigned int flags,
277 					   int noblock, int *err)
278 {
279 	int off = 0;
280 
281 	return __skb_recv_udp(sk, flags, noblock, &off, err);
282 }
283 
284 int udp_v4_early_demux(struct sk_buff *skb);
285 bool udp_sk_rx_dst_set(struct sock *sk, struct dst_entry *dst);
286 int udp_get_port(struct sock *sk, unsigned short snum,
287 		 int (*saddr_cmp)(const struct sock *,
288 				  const struct sock *));
289 int udp_err(struct sk_buff *, u32);
290 int udp_abort(struct sock *sk, int err);
291 int udp_sendmsg(struct sock *sk, struct msghdr *msg, size_t len);
292 int udp_push_pending_frames(struct sock *sk);
293 void udp_flush_pending_frames(struct sock *sk);
294 int udp_cmsg_send(struct sock *sk, struct msghdr *msg, u16 *gso_size);
295 void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst);
296 int udp_rcv(struct sk_buff *skb);
297 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg);
298 int udp_init_sock(struct sock *sk);
299 int udp_pre_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
300 int __udp_disconnect(struct sock *sk, int flags);
301 int udp_disconnect(struct sock *sk, int flags);
302 __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait);
303 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
304 				       netdev_features_t features,
305 				       bool is_ipv6);
306 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
307 		       char __user *optval, int __user *optlen);
308 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
309 		       char __user *optval, unsigned int optlen,
310 		       int (*push_pending_frames)(struct sock *));
311 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
312 			     __be32 daddr, __be16 dport, int dif);
313 struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
314 			       __be32 daddr, __be16 dport, int dif, int sdif,
315 			       struct udp_table *tbl, struct sk_buff *skb);
316 struct sock *udp4_lib_lookup_skb(struct sk_buff *skb,
317 				 __be16 sport, __be16 dport);
318 struct sock *udp6_lib_lookup(struct net *net,
319 			     const struct in6_addr *saddr, __be16 sport,
320 			     const struct in6_addr *daddr, __be16 dport,
321 			     int dif);
322 struct sock *__udp6_lib_lookup(struct net *net,
323 			       const struct in6_addr *saddr, __be16 sport,
324 			       const struct in6_addr *daddr, __be16 dport,
325 			       int dif, int sdif, struct udp_table *tbl,
326 			       struct sk_buff *skb);
327 struct sock *udp6_lib_lookup_skb(struct sk_buff *skb,
328 				 __be16 sport, __be16 dport);
329 
330 /* UDP uses skb->dev_scratch to cache as much information as possible and avoid
331  * possibly multiple cache miss on dequeue()
332  */
333 struct udp_dev_scratch {
334 	/* skb->truesize and the stateless bit are embedded in a single field;
335 	 * do not use a bitfield since the compiler emits better/smaller code
336 	 * this way
337 	 */
338 	u32 _tsize_state;
339 
340 #if BITS_PER_LONG == 64
341 	/* len and the bit needed to compute skb_csum_unnecessary
342 	 * will be on cold cache lines at recvmsg time.
343 	 * skb->len can be stored on 16 bits since the udp header has been
344 	 * already validated and pulled.
345 	 */
346 	u16 len;
347 	bool is_linear;
348 	bool csum_unnecessary;
349 #endif
350 };
351 
352 static inline struct udp_dev_scratch *udp_skb_scratch(struct sk_buff *skb)
353 {
354 	return (struct udp_dev_scratch *)&skb->dev_scratch;
355 }
356 
357 #if BITS_PER_LONG == 64
358 static inline unsigned int udp_skb_len(struct sk_buff *skb)
359 {
360 	return udp_skb_scratch(skb)->len;
361 }
362 
363 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
364 {
365 	return udp_skb_scratch(skb)->csum_unnecessary;
366 }
367 
368 static inline bool udp_skb_is_linear(struct sk_buff *skb)
369 {
370 	return udp_skb_scratch(skb)->is_linear;
371 }
372 
373 #else
374 static inline unsigned int udp_skb_len(struct sk_buff *skb)
375 {
376 	return skb->len;
377 }
378 
379 static inline bool udp_skb_csum_unnecessary(struct sk_buff *skb)
380 {
381 	return skb_csum_unnecessary(skb);
382 }
383 
384 static inline bool udp_skb_is_linear(struct sk_buff *skb)
385 {
386 	return !skb_is_nonlinear(skb);
387 }
388 #endif
389 
390 static inline int copy_linear_skb(struct sk_buff *skb, int len, int off,
391 				  struct iov_iter *to)
392 {
393 	int n;
394 
395 	n = copy_to_iter(skb->data + off, len, to);
396 	if (n == len)
397 		return 0;
398 
399 	iov_iter_revert(to, n);
400 	return -EFAULT;
401 }
402 
403 /*
404  * 	SNMP statistics for UDP and UDP-Lite
405  */
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 #define __UDP_INC_STATS(net, field, is_udplite) 	      do { \
410 	if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_statistics, field);         \
411 	else		__SNMP_INC_STATS((net)->mib.udp_statistics, field);    }  while(0)
412 
413 #define __UDP6_INC_STATS(net, field, is_udplite)	    do { \
414 	if (is_udplite) __SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);\
415 	else		__SNMP_INC_STATS((net)->mib.udp_stats_in6, field);  \
416 } while(0)
417 #define UDP6_INC_STATS(net, field, __lite)		    do { \
418 	if (__lite) SNMP_INC_STATS((net)->mib.udplite_stats_in6, field);  \
419 	else	    SNMP_INC_STATS((net)->mib.udp_stats_in6, field);      \
420 } while(0)
421 
422 #if IS_ENABLED(CONFIG_IPV6)
423 #define __UDPX_MIB(sk, ipv4)						\
424 ({									\
425 	ipv4 ? (IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :	\
426 				 sock_net(sk)->mib.udp_statistics) :	\
427 		(IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_stats_in6 :	\
428 				 sock_net(sk)->mib.udp_stats_in6);	\
429 })
430 #else
431 #define __UDPX_MIB(sk, ipv4)						\
432 ({									\
433 	IS_UDPLITE(sk) ? sock_net(sk)->mib.udplite_statistics :		\
434 			 sock_net(sk)->mib.udp_statistics;		\
435 })
436 #endif
437 
438 #define __UDPX_INC_STATS(sk, field) \
439 	__SNMP_INC_STATS(__UDPX_MIB(sk, (sk)->sk_family == AF_INET), field)
440 
441 #ifdef CONFIG_PROC_FS
442 struct udp_seq_afinfo {
443 	sa_family_t			family;
444 	struct udp_table		*udp_table;
445 };
446 
447 struct udp_iter_state {
448 	struct seq_net_private  p;
449 	int			bucket;
450 	struct udp_seq_afinfo	*bpf_seq_afinfo;
451 };
452 
453 void *udp_seq_start(struct seq_file *seq, loff_t *pos);
454 void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
455 void udp_seq_stop(struct seq_file *seq, void *v);
456 
457 extern const struct seq_operations udp_seq_ops;
458 extern const struct seq_operations udp6_seq_ops;
459 
460 int udp4_proc_init(void);
461 void udp4_proc_exit(void);
462 #endif /* CONFIG_PROC_FS */
463 
464 int udpv4_offload_init(void);
465 
466 void udp_init(void);
467 
468 DECLARE_STATIC_KEY_FALSE(udp_encap_needed_key);
469 void udp_encap_enable(void);
470 #if IS_ENABLED(CONFIG_IPV6)
471 DECLARE_STATIC_KEY_FALSE(udpv6_encap_needed_key);
472 void udpv6_encap_enable(void);
473 #endif
474 
475 static inline struct sk_buff *udp_rcv_segment(struct sock *sk,
476 					      struct sk_buff *skb, bool ipv4)
477 {
478 	netdev_features_t features = NETIF_F_SG;
479 	struct sk_buff *segs;
480 
481 	/* Avoid csum recalculation by skb_segment unless userspace explicitly
482 	 * asks for the final checksum values
483 	 */
484 	if (!inet_get_convert_csum(sk))
485 		features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
486 
487 	/* UDP segmentation expects packets of type CHECKSUM_PARTIAL or
488 	 * CHECKSUM_NONE in __udp_gso_segment. UDP GRO indeed builds partial
489 	 * packets in udp_gro_complete_segment. As does UDP GSO, verified by
490 	 * udp_send_skb. But when those packets are looped in dev_loopback_xmit
491 	 * their ip_summed is set to CHECKSUM_UNNECESSARY. Reset in this
492 	 * specific case, where PARTIAL is both correct and required.
493 	 */
494 	if (skb->pkt_type == PACKET_LOOPBACK)
495 		skb->ip_summed = CHECKSUM_PARTIAL;
496 
497 	/* the GSO CB lays after the UDP one, no need to save and restore any
498 	 * CB fragment
499 	 */
500 	segs = __skb_gso_segment(skb, features, false);
501 	if (IS_ERR_OR_NULL(segs)) {
502 		int segs_nr = skb_shinfo(skb)->gso_segs;
503 
504 		atomic_add(segs_nr, &sk->sk_drops);
505 		SNMP_ADD_STATS(__UDPX_MIB(sk, ipv4), UDP_MIB_INERRORS, segs_nr);
506 		kfree_skb(skb);
507 		return NULL;
508 	}
509 
510 	consume_skb(skb);
511 	return segs;
512 }
513 
514 #ifdef CONFIG_BPF_STREAM_PARSER
515 struct sk_psock;
516 struct proto *udp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
517 #endif /* BPF_STREAM_PARSER */
518 
519 #endif	/* _UDP_H */
520