xref: /linux/include/net/inet_sock.h (revision b3827c91cc9979fe04d99e016fb9c5f6260f29a0)
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 inet_sock
8  *
9  * Authors:	Many, reorganised here by
10  * 		Arnaldo Carvalho de Melo <acme@mandriva.com>
11  */
12 #ifndef _INET_SOCK_H
13 #define _INET_SOCK_H
14 
15 #include <linux/bitops.h>
16 #include <linux/string.h>
17 #include <linux/types.h>
18 #include <linux/jhash.h>
19 #include <linux/netdevice.h>
20 
21 #include <net/flow.h>
22 #include <net/inet_dscp.h>
23 #include <net/sock.h>
24 #include <net/request_sock.h>
25 #include <net/netns/hash.h>
26 #include <net/tcp_states.h>
27 #include <net/l3mdev.h>
28 
29 #define IP_OPTIONS_DATA_FIXED_SIZE 40
30 
31 /** struct ip_options - IP Options
32  *
33  * @faddr - Saved first hop address
34  * @nexthop - Saved nexthop address in LSRR and SSRR
35  * @is_strictroute - Strict source route
36  * @srr_is_hit - Packet destination addr was our one
37  * @is_changed - IP checksum more not valid
38  * @rr_needaddr - Need to record addr of outgoing dev
39  * @ts_needtime - Need to record timestamp
40  * @ts_needaddr - Need to record addr of outgoing dev
41  */
42 struct ip_options {
43 	__be32		faddr;
44 	__be32		nexthop;
45 	unsigned char	optlen;
46 	unsigned char	srr;
47 	unsigned char	rr;
48 	unsigned char	ts;
49 	unsigned char	is_strictroute:1,
50 			srr_is_hit:1,
51 			is_changed:1,
52 			rr_needaddr:1,
53 			ts_needtime:1,
54 			ts_needaddr:1;
55 	unsigned char	router_alert;
56 	unsigned char	cipso;
57 	unsigned char	__pad2;
58 	unsigned char	__data[];
59 };
60 
61 struct ip_options_rcu {
62 	struct rcu_head rcu;
63 
64 	/* Must be last as it ends in a flexible-array member. */
65 	struct ip_options opt;
66 };
67 
68 struct inet_request_sock {
69 	struct request_sock	req;
70 #define ir_loc_addr		req.__req_common.skc_rcv_saddr
71 #define ir_rmt_addr		req.__req_common.skc_daddr
72 #define ir_num			req.__req_common.skc_num
73 #define ir_rmt_port		req.__req_common.skc_dport
74 #define ir_v6_rmt_addr		req.__req_common.skc_v6_daddr
75 #define ir_v6_loc_addr		req.__req_common.skc_v6_rcv_saddr
76 #define ir_iif			req.__req_common.skc_bound_dev_if
77 #define ir_cookie		req.__req_common.skc_cookie
78 #define ireq_net		req.__req_common.skc_net
79 #define ireq_state		req.__req_common.skc_state
80 #define ireq_family		req.__req_common.skc_family
81 
82 	u16			snd_wscale : 4,
83 				rcv_wscale : 4,
84 				tstamp_ok  : 1,
85 				sack_ok	   : 1,
86 				wscale_ok  : 1,
87 				ecn_ok	   : 1,
88 				acked	   : 1,
89 				no_srccheck: 1,
90 				smc_ok	   : 1;
91 	u32                     ir_mark;
92 	union {
93 		struct ip_options_rcu __rcu	*ireq_opt;
94 #if IS_ENABLED(CONFIG_IPV6)
95 		struct {
96 			struct ipv6_txoptions	*ipv6_opt;
97 			struct sk_buff		*pktopts;
98 		};
99 #endif
100 	};
101 };
102 
103 #define inet_rsk(ptr) container_of_const(ptr, struct inet_request_sock, req)
104 
105 static inline u32 inet_request_mark(const struct sock *sk, struct sk_buff *skb)
106 {
107 	u32 mark = READ_ONCE(sk->sk_mark);
108 
109 	if (!mark && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fwmark_accept))
110 		return skb->mark;
111 
112 	return mark;
113 }
114 
115 static inline int inet_request_bound_dev_if(const struct sock *sk,
116 					    struct sk_buff *skb)
117 {
118 	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
119 #ifdef CONFIG_NET_L3_MASTER_DEV
120 	struct net *net = sock_net(sk);
121 
122 	if (!bound_dev_if && READ_ONCE(net->ipv4.sysctl_tcp_l3mdev_accept))
123 		return l3mdev_master_ifindex_by_index(net, skb->skb_iif);
124 #endif
125 
126 	return bound_dev_if;
127 }
128 
129 static inline int inet_sk_bound_l3mdev(const struct sock *sk)
130 {
131 #ifdef CONFIG_NET_L3_MASTER_DEV
132 	struct net *net = sock_net(sk);
133 
134 	if (!READ_ONCE(net->ipv4.sysctl_tcp_l3mdev_accept))
135 		return l3mdev_master_ifindex_by_index(net,
136 						      sk->sk_bound_dev_if);
137 #endif
138 
139 	return 0;
140 }
141 
142 static inline bool inet_bound_dev_eq(bool l3mdev_accept, int bound_dev_if,
143 				     int dif, int sdif)
144 {
145 	if (!bound_dev_if)
146 		return !sdif || l3mdev_accept;
147 	return bound_dev_if == dif || bound_dev_if == sdif;
148 }
149 
150 static inline bool inet_sk_bound_dev_eq(const struct net *net,
151 					int bound_dev_if,
152 					int dif, int sdif)
153 {
154 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
155 	return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_tcp_l3mdev_accept),
156 				 bound_dev_if, dif, sdif);
157 #else
158 	return inet_bound_dev_eq(true, bound_dev_if, dif, sdif);
159 #endif
160 }
161 
162 struct inet_cork {
163 	unsigned int		flags;
164 	__be32			addr;
165 	struct ip_options	*opt;
166 	unsigned int		fragsize;
167 	int			length; /* Total length of all frames */
168 	struct dst_entry	*dst;
169 	u8			tx_flags;
170 	__u8			ttl;
171 	__s16			tos;
172 	u32			priority;
173 	__u16			gso_size;
174 	u32			ts_opt_id;
175 	u64			transmit_time;
176 	u32			mark;
177 };
178 
179 struct inet_cork_full {
180 	struct inet_cork	base;
181 	struct flowi		fl;
182 };
183 
184 struct ip_mc_socklist;
185 struct ipv6_pinfo;
186 struct rtable;
187 
188 /** struct inet_sock - representation of INET sockets
189  *
190  * @sk - ancestor class
191  * @pinet6 - pointer to IPv6 control block
192  * @inet_daddr - Foreign IPv4 addr
193  * @inet_rcv_saddr - Bound local IPv4 addr
194  * @inet_dport - Destination port
195  * @inet_num - Local port
196  * @inet_flags - various atomic flags
197  * @inet_saddr - Sending source
198  * @uc_ttl - Unicast TTL
199  * @inet_sport - Source port
200  * @inet_id - ID counter for DF pkts
201  * @tos - TOS
202  * @mc_ttl - Multicasting TTL
203  * @uc_index - Unicast outgoing device index
204  * @mc_index - Multicast device index
205  * @mc_list - Group array
206  * @cork - info to build ip hdr on each ip frag while socket is corked
207  */
208 struct inet_sock {
209 	/* sk and pinet6 has to be the first two members of inet_sock */
210 	struct sock		sk;
211 #if IS_ENABLED(CONFIG_IPV6)
212 	struct ipv6_pinfo	*pinet6;
213 	struct ipv6_fl_socklist __rcu *ipv6_fl_list;
214 #endif
215 	/* Socket demultiplex comparisons on incoming packets. */
216 #define inet_daddr		sk.__sk_common.skc_daddr
217 #define inet_rcv_saddr		sk.__sk_common.skc_rcv_saddr
218 #define inet_dport		sk.__sk_common.skc_dport
219 #define inet_num		sk.__sk_common.skc_num
220 
221 	unsigned long		inet_flags;
222 	__be32			inet_saddr;
223 	__s16			uc_ttl;
224 	__be16			inet_sport;
225 	struct ip_options_rcu __rcu	*inet_opt;
226 	atomic_t		inet_id;
227 
228 	__u8			tos;
229 	__u8			min_ttl;
230 	__u8			mc_ttl;
231 	__u8			pmtudisc;
232 	__u8			rcv_tos;
233 	__u8			convert_csum;
234 	int			uc_index;
235 	int			mc_index;
236 	__be32			mc_addr;
237 	u32			local_port_range;	/* high << 16 | low */
238 
239 	struct ip_mc_socklist __rcu	*mc_list;
240 	struct inet_cork_full	cork;
241 };
242 
243 #define IPCORK_OPT		1	/* ip-options has been held in ipcork.opt */
244 #define IPCORK_TS_OPT_ID	2	/* ts_opt_id field is valid, overriding sk_tskey */
245 
246 enum {
247 	INET_FLAGS_PKTINFO	= 0,
248 	INET_FLAGS_TTL		= 1,
249 	INET_FLAGS_TOS		= 2,
250 	INET_FLAGS_RECVOPTS	= 3,
251 	INET_FLAGS_RETOPTS	= 4,
252 	INET_FLAGS_PASSSEC	= 5,
253 	INET_FLAGS_ORIGDSTADDR	= 6,
254 	INET_FLAGS_CHECKSUM	= 7,
255 	INET_FLAGS_RECVFRAGSIZE	= 8,
256 
257 	INET_FLAGS_RECVERR	= 9,
258 	INET_FLAGS_RECVERR_RFC4884 = 10,
259 	INET_FLAGS_FREEBIND	= 11,
260 	INET_FLAGS_HDRINCL	= 12,
261 	INET_FLAGS_MC_LOOP	= 13,
262 	INET_FLAGS_MC_ALL	= 14,
263 	INET_FLAGS_TRANSPARENT	= 15,
264 	INET_FLAGS_IS_ICSK	= 16,
265 	INET_FLAGS_NODEFRAG	= 17,
266 	INET_FLAGS_BIND_ADDRESS_NO_PORT = 18,
267 	INET_FLAGS_DEFER_CONNECT = 19,
268 	INET_FLAGS_MC6_LOOP	= 20,
269 	INET_FLAGS_RECVERR6_RFC4884 = 21,
270 	INET_FLAGS_MC6_ALL	= 22,
271 	INET_FLAGS_AUTOFLOWLABEL_SET = 23,
272 	INET_FLAGS_AUTOFLOWLABEL = 24,
273 	INET_FLAGS_DONTFRAG	= 25,
274 	INET_FLAGS_RECVERR6	= 26,
275 	INET_FLAGS_REPFLOW	= 27,
276 	INET_FLAGS_RTALERT_ISOLATE = 28,
277 	INET_FLAGS_SNDFLOW	= 29,
278 	INET_FLAGS_RTALERT	= 30,
279 };
280 
281 /* cmsg flags for inet */
282 #define IP_CMSG_PKTINFO		BIT(INET_FLAGS_PKTINFO)
283 #define IP_CMSG_TTL		BIT(INET_FLAGS_TTL)
284 #define IP_CMSG_TOS		BIT(INET_FLAGS_TOS)
285 #define IP_CMSG_RECVOPTS	BIT(INET_FLAGS_RECVOPTS)
286 #define IP_CMSG_RETOPTS		BIT(INET_FLAGS_RETOPTS)
287 #define IP_CMSG_PASSSEC		BIT(INET_FLAGS_PASSSEC)
288 #define IP_CMSG_ORIGDSTADDR	BIT(INET_FLAGS_ORIGDSTADDR)
289 #define IP_CMSG_CHECKSUM	BIT(INET_FLAGS_CHECKSUM)
290 #define IP_CMSG_RECVFRAGSIZE	BIT(INET_FLAGS_RECVFRAGSIZE)
291 
292 #define IP_CMSG_ALL	(IP_CMSG_PKTINFO | IP_CMSG_TTL |		\
293 			 IP_CMSG_TOS | IP_CMSG_RECVOPTS |		\
294 			 IP_CMSG_RETOPTS | IP_CMSG_PASSSEC |		\
295 			 IP_CMSG_ORIGDSTADDR | IP_CMSG_CHECKSUM |	\
296 			 IP_CMSG_RECVFRAGSIZE)
297 
298 static inline unsigned long inet_cmsg_flags(const struct inet_sock *inet)
299 {
300 	return READ_ONCE(inet->inet_flags) & IP_CMSG_ALL;
301 }
302 
303 static inline dscp_t inet_sk_dscp(const struct inet_sock *inet)
304 {
305 	return inet_dsfield_to_dscp(READ_ONCE(inet->tos));
306 }
307 
308 #define inet_test_bit(nr, sk)			\
309 	test_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags)
310 #define inet_set_bit(nr, sk)			\
311 	set_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags)
312 #define inet_clear_bit(nr, sk)			\
313 	clear_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags)
314 #define inet_assign_bit(nr, sk, val)		\
315 	assign_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags, val)
316 
317 /**
318  * sk_to_full_sk - Access to a full socket
319  * @sk: pointer to a socket
320  *
321  * SYNACK messages might be attached to request sockets.
322  * Some places want to reach the listener in this case.
323  */
324 static inline struct sock *sk_to_full_sk(struct sock *sk)
325 {
326 #ifdef CONFIG_INET
327 	if (sk && READ_ONCE(sk->sk_state) == TCP_NEW_SYN_RECV)
328 		sk = inet_reqsk(sk)->rsk_listener;
329 	if (sk && READ_ONCE(sk->sk_state) == TCP_TIME_WAIT)
330 		sk = NULL;
331 #endif
332 	return sk;
333 }
334 
335 /* sk_to_full_sk() variant with a const argument */
336 static inline const struct sock *sk_const_to_full_sk(const struct sock *sk)
337 {
338 #ifdef CONFIG_INET
339 	if (sk && READ_ONCE(sk->sk_state) == TCP_NEW_SYN_RECV)
340 		sk = ((const struct request_sock *)sk)->rsk_listener;
341 	if (sk && READ_ONCE(sk->sk_state) == TCP_TIME_WAIT)
342 		sk = NULL;
343 #endif
344 	return sk;
345 }
346 
347 static inline struct sock *skb_to_full_sk(const struct sk_buff *skb)
348 {
349 	return sk_to_full_sk(skb->sk);
350 }
351 
352 #define inet_sk(ptr) container_of_const(ptr, struct inet_sock, sk)
353 
354 int inet_sk_rebuild_header(struct sock *sk);
355 
356 /**
357  * inet_sk_state_load - read sk->sk_state for lockless contexts
358  * @sk: socket pointer
359  *
360  * Paired with inet_sk_state_store(). Used in places we don't hold socket lock:
361  * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
362  */
363 static inline int inet_sk_state_load(const struct sock *sk)
364 {
365 	/* state change might impact lockless readers. */
366 	return smp_load_acquire(&sk->sk_state);
367 }
368 
369 /**
370  * inet_sk_state_store - update sk->sk_state
371  * @sk: socket pointer
372  * @newstate: new state
373  *
374  * Paired with inet_sk_state_load(). Should be used in contexts where
375  * state change might impact lockless readers.
376  */
377 void inet_sk_state_store(struct sock *sk, int newstate);
378 
379 void inet_sk_set_state(struct sock *sk, int state);
380 
381 static inline unsigned int __inet_ehashfn(const __be32 laddr,
382 					  const __u16 lport,
383 					  const __be32 faddr,
384 					  const __be16 fport,
385 					  u32 initval)
386 {
387 	return jhash_3words((__force __u32) laddr,
388 			    (__force __u32) faddr,
389 			    ((__u32) lport) << 16 | (__force __u32)fport,
390 			    initval);
391 }
392 
393 struct request_sock *inet_reqsk_alloc(const struct request_sock_ops *ops,
394 				      struct sock *sk_listener,
395 				      bool attach_listener);
396 
397 static inline __u8 inet_sk_flowi_flags(const struct sock *sk)
398 {
399 	__u8 flags = 0;
400 
401 	if (inet_test_bit(TRANSPARENT, sk) || inet_test_bit(HDRINCL, sk))
402 		flags |= FLOWI_FLAG_ANYSRC;
403 	return flags;
404 }
405 
406 static inline void inet_inc_convert_csum(struct sock *sk)
407 {
408 	inet_sk(sk)->convert_csum++;
409 }
410 
411 static inline void inet_dec_convert_csum(struct sock *sk)
412 {
413 	if (inet_sk(sk)->convert_csum > 0)
414 		inet_sk(sk)->convert_csum--;
415 }
416 
417 static inline bool inet_get_convert_csum(struct sock *sk)
418 {
419 	return !!inet_sk(sk)->convert_csum;
420 }
421 
422 
423 static inline bool inet_can_nonlocal_bind(struct net *net,
424 					  struct inet_sock *inet)
425 {
426 	return READ_ONCE(net->ipv4.sysctl_ip_nonlocal_bind) ||
427 		test_bit(INET_FLAGS_FREEBIND, &inet->inet_flags) ||
428 		test_bit(INET_FLAGS_TRANSPARENT, &inet->inet_flags);
429 }
430 
431 static inline bool inet_addr_valid_or_nonlocal(struct net *net,
432 					       struct inet_sock *inet,
433 					       __be32 addr,
434 					       int addr_type)
435 {
436 	return inet_can_nonlocal_bind(net, inet) ||
437 		addr == htonl(INADDR_ANY) ||
438 		addr_type == RTN_LOCAL ||
439 		addr_type == RTN_MULTICAST ||
440 		addr_type == RTN_BROADCAST;
441 }
442 
443 #endif	/* _INET_SOCK_H */
444