Commit | Line | Data |
---|---|---|
2874c5fd | 1 | /* SPDX-License-Identifier: GPL-2.0-or-later */ |
14c85021 ACM |
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> | |
14c85021 ACM |
11 | */ |
12 | #ifndef _INET_SOCK_H | |
13 | #define _INET_SOCK_H | |
14 | ||
c44d13d6 | 15 | #include <linux/bitops.h> |
14c85021 ACM |
16 | #include <linux/string.h> |
17 | #include <linux/types.h> | |
b3da2cf3 | 18 | #include <linux/jhash.h> |
fec5e652 | 19 | #include <linux/netdevice.h> |
14c85021 ACM |
20 | |
21 | #include <net/flow.h> | |
22 | #include <net/sock.h> | |
23 | #include <net/request_sock.h> | |
0b441916 | 24 | #include <net/netns/hash.h> |
d34ac51b | 25 | #include <net/tcp_states.h> |
6dd9a14e | 26 | #include <net/l3mdev.h> |
14c85021 ACM |
27 | |
28 | /** struct ip_options - IP Options | |
29 | * | |
30 | * @faddr - Saved first hop address | |
ac8a4810 | 31 | * @nexthop - Saved nexthop address in LSRR and SSRR |
14c85021 ACM |
32 | * @is_strictroute - Strict source route |
33 | * @srr_is_hit - Packet destination addr was our one | |
34 | * @is_changed - IP checksum more not valid | |
35 | * @rr_needaddr - Need to record addr of outgoing dev | |
36 | * @ts_needtime - Need to record timestamp | |
37 | * @ts_needaddr - Need to record addr of outgoing dev | |
38 | */ | |
39 | struct ip_options { | |
3ca3c68e | 40 | __be32 faddr; |
ac8a4810 | 41 | __be32 nexthop; |
14c85021 ACM |
42 | unsigned char optlen; |
43 | unsigned char srr; | |
44 | unsigned char rr; | |
45 | unsigned char ts; | |
ef722495 | 46 | unsigned char is_strictroute:1, |
14c85021 ACM |
47 | srr_is_hit:1, |
48 | is_changed:1, | |
49 | rr_needaddr:1, | |
50 | ts_needtime:1, | |
51 | ts_needaddr:1; | |
52 | unsigned char router_alert; | |
11a03f78 | 53 | unsigned char cipso; |
14c85021 | 54 | unsigned char __pad2; |
48b77df6 | 55 | unsigned char __data[]; |
14c85021 ACM |
56 | }; |
57 | ||
f6d8bd05 ED |
58 | struct ip_options_rcu { |
59 | struct rcu_head rcu; | |
60 | struct ip_options opt; | |
61 | }; | |
62 | ||
63 | struct ip_options_data { | |
64 | struct ip_options_rcu opt; | |
65 | char data[40]; | |
66 | }; | |
14c85021 ACM |
67 | |
68 | struct inet_request_sock { | |
69 | struct request_sock req; | |
634fb979 ED |
70 | #define ir_loc_addr req.__req_common.skc_rcv_saddr |
71 | #define ir_rmt_addr req.__req_common.skc_daddr | |
b44084c2 | 72 | #define ir_num req.__req_common.skc_num |
634fb979 ED |
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 | |
33cf7c90 ED |
77 | #define ir_cookie req.__req_common.skc_cookie |
78 | #define ireq_net req.__req_common.skc_net | |
d34ac51b | 79 | #define ireq_state req.__req_common.skc_state |
3f66b083 | 80 | #define ireq_family req.__req_common.skc_family |
634fb979 | 81 | |
45e3ff82 VN |
82 | u16 snd_wscale : 4, |
83 | rcv_wscale : 4, | |
14c85021 ACM |
84 | tstamp_ok : 1, |
85 | sack_ok : 1, | |
86 | wscale_ok : 1, | |
87 | ecn_ok : 1, | |
88ef4a5a | 88 | acked : 1, |
60e2a778 UB |
89 | no_srccheck: 1, |
90 | smc_ok : 1; | |
adc17d6a | 91 | u32 ir_mark; |
476eab82 | 92 | union { |
c92e8c02 | 93 | struct ip_options_rcu __rcu *ireq_opt; |
56ac42bc HD |
94 | #if IS_ENABLED(CONFIG_IPV6) |
95 | struct { | |
96 | struct ipv6_txoptions *ipv6_opt; | |
97 | struct sk_buff *pktopts; | |
98 | }; | |
99 | #endif | |
476eab82 | 100 | }; |
14c85021 ACM |
101 | }; |
102 | ||
103 | static inline struct inet_request_sock *inet_rsk(const struct request_sock *sk) | |
104 | { | |
105 | return (struct inet_request_sock *)sk; | |
106 | } | |
107 | ||
adc17d6a | 108 | static inline u32 inet_request_mark(const struct sock *sk, struct sk_buff *skb) |
84f39b08 | 109 | { |
3c5b4d69 ED |
110 | u32 mark = READ_ONCE(sk->sk_mark); |
111 | ||
112 | if (!mark && READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fwmark_accept)) | |
84f39b08 | 113 | return skb->mark; |
adc17d6a | 114 | |
3c5b4d69 | 115 | return mark; |
84f39b08 LC |
116 | } |
117 | ||
6dd9a14e DA |
118 | static inline int inet_request_bound_dev_if(const struct sock *sk, |
119 | struct sk_buff *skb) | |
120 | { | |
fdb5fd7f | 121 | int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if); |
6dd9a14e DA |
122 | #ifdef CONFIG_NET_L3_MASTER_DEV |
123 | struct net *net = sock_net(sk); | |
124 | ||
08a75f10 | 125 | if (!bound_dev_if && READ_ONCE(net->ipv4.sysctl_tcp_l3mdev_accept)) |
6dd9a14e DA |
126 | return l3mdev_master_ifindex_by_index(net, skb->skb_iif); |
127 | #endif | |
128 | ||
fdb5fd7f | 129 | return bound_dev_if; |
6dd9a14e DA |
130 | } |
131 | ||
3c82a21f RS |
132 | static inline int inet_sk_bound_l3mdev(const struct sock *sk) |
133 | { | |
134 | #ifdef CONFIG_NET_L3_MASTER_DEV | |
135 | struct net *net = sock_net(sk); | |
136 | ||
08a75f10 | 137 | if (!READ_ONCE(net->ipv4.sysctl_tcp_l3mdev_accept)) |
3c82a21f RS |
138 | return l3mdev_master_ifindex_by_index(net, |
139 | sk->sk_bound_dev_if); | |
140 | #endif | |
141 | ||
142 | return 0; | |
143 | } | |
144 | ||
e7819058 MM |
145 | static inline bool inet_bound_dev_eq(bool l3mdev_accept, int bound_dev_if, |
146 | int dif, int sdif) | |
147 | { | |
148 | if (!bound_dev_if) | |
149 | return !sdif || l3mdev_accept; | |
150 | return bound_dev_if == dif || bound_dev_if == sdif; | |
151 | } | |
152 | ||
944fd1ae MM |
153 | static inline bool inet_sk_bound_dev_eq(struct net *net, int bound_dev_if, |
154 | int dif, int sdif) | |
155 | { | |
156 | #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV) | |
157 | return inet_bound_dev_eq(!!READ_ONCE(net->ipv4.sysctl_tcp_l3mdev_accept), | |
158 | bound_dev_if, dif, sdif); | |
159 | #else | |
160 | return inet_bound_dev_eq(true, bound_dev_if, dif, sdif); | |
161 | #endif | |
162 | } | |
163 | ||
1470ddf7 HX |
164 | struct inet_cork { |
165 | unsigned int flags; | |
bdc712b4 | 166 | __be32 addr; |
1470ddf7 | 167 | struct ip_options *opt; |
bdc712b4 | 168 | unsigned int fragsize; |
1470ddf7 | 169 | int length; /* Total length of all frames */ |
5640f768 | 170 | struct dst_entry *dst; |
1470ddf7 | 171 | u8 tx_flags; |
aa661581 FF |
172 | __u8 ttl; |
173 | __s16 tos; | |
174 | char priority; | |
bec1f6f6 | 175 | __u16 gso_size; |
bc969a97 | 176 | u64 transmit_time; |
c6af0c22 | 177 | u32 mark; |
1470ddf7 HX |
178 | }; |
179 | ||
bdc712b4 DM |
180 | struct inet_cork_full { |
181 | struct inet_cork base; | |
182 | struct flowi fl; | |
183 | }; | |
184 | ||
14c85021 ACM |
185 | struct ip_mc_socklist; |
186 | struct ipv6_pinfo; | |
187 | struct rtable; | |
188 | ||
189 | /** struct inet_sock - representation of INET sockets | |
190 | * | |
191 | * @sk - ancestor class | |
192 | * @pinet6 - pointer to IPv6 control block | |
c720c7e8 ED |
193 | * @inet_daddr - Foreign IPv4 addr |
194 | * @inet_rcv_saddr - Bound local IPv4 addr | |
195 | * @inet_dport - Destination port | |
196 | * @inet_num - Local port | |
c274af22 | 197 | * @inet_flags - various atomic flags |
c720c7e8 | 198 | * @inet_saddr - Sending source |
14c85021 | 199 | * @uc_ttl - Unicast TTL |
c720c7e8 ED |
200 | * @inet_sport - Source port |
201 | * @inet_id - ID counter for DF pkts | |
14c85021 ACM |
202 | * @tos - TOS |
203 | * @mc_ttl - Multicasting TTL | |
76e21053 | 204 | * @uc_index - Unicast outgoing device index |
14c85021 ACM |
205 | * @mc_index - Multicast device index |
206 | * @mc_list - Group array | |
207 | * @cork - info to build ip hdr on each ip frag while socket is corked | |
208 | */ | |
209 | struct inet_sock { | |
210 | /* sk and pinet6 has to be the first two members of inet_sock */ | |
211 | struct sock sk; | |
dfd56b8b | 212 | #if IS_ENABLED(CONFIG_IPV6) |
14c85021 ACM |
213 | struct ipv6_pinfo *pinet6; |
214 | #endif | |
215 | /* Socket demultiplex comparisons on incoming packets. */ | |
68835aba ED |
216 | #define inet_daddr sk.__sk_common.skc_daddr |
217 | #define inet_rcv_saddr sk.__sk_common.skc_rcv_saddr | |
ce43b03e ED |
218 | #define inet_dport sk.__sk_common.skc_dport |
219 | #define inet_num sk.__sk_common.skc_num | |
68835aba | 220 | |
c274af22 | 221 | unsigned long inet_flags; |
c720c7e8 | 222 | __be32 inet_saddr; |
14c85021 | 223 | __s16 uc_ttl; |
c720c7e8 | 224 | __be16 inet_sport; |
c274af22 | 225 | struct ip_options_rcu __rcu *inet_opt; |
f866fbc8 | 226 | atomic_t inet_id; |
d218d111 | 227 | |
14c85021 | 228 | __u8 tos; |
d218d111 | 229 | __u8 min_ttl; |
14c85021 ACM |
230 | __u8 mc_ttl; |
231 | __u8 pmtudisc; | |
4c507d28 | 232 | __u8 rcv_tos; |
224d019c | 233 | __u8 convert_csum; |
76e21053 | 234 | int uc_index; |
14c85021 | 235 | int mc_index; |
011a9261 | 236 | __be32 mc_addr; |
d9f28735 | 237 | u32 local_port_range; /* high << 16 | low */ |
08e39c0d ED |
238 | |
239 | struct ip_mc_socklist __rcu *mc_list; | |
240 | struct inet_cork_full cork; | |
14c85021 ACM |
241 | }; |
242 | ||
243 | #define IPCORK_OPT 1 /* ip-options has been held in ipcork.opt */ | |
14c85021 | 244 | |
c274af22 ED |
245 | enum { |
246 | INET_FLAGS_PKTINFO = 0, | |
247 | INET_FLAGS_TTL = 1, | |
248 | INET_FLAGS_TOS = 2, | |
249 | INET_FLAGS_RECVOPTS = 3, | |
250 | INET_FLAGS_RETOPTS = 4, | |
251 | INET_FLAGS_PASSSEC = 5, | |
252 | INET_FLAGS_ORIGDSTADDR = 6, | |
253 | INET_FLAGS_CHECKSUM = 7, | |
254 | INET_FLAGS_RECVFRAGSIZE = 8, | |
6b5f43ea ED |
255 | |
256 | INET_FLAGS_RECVERR = 9, | |
8e8cfb11 | 257 | INET_FLAGS_RECVERR_RFC4884 = 10, |
3f7e7532 | 258 | INET_FLAGS_FREEBIND = 11, |
cafbe182 | 259 | INET_FLAGS_HDRINCL = 12, |
b09bde5c | 260 | INET_FLAGS_MC_LOOP = 13, |
307b4ac6 | 261 | INET_FLAGS_MC_ALL = 14, |
4bd0623f | 262 | INET_FLAGS_TRANSPARENT = 15, |
b1c0356a | 263 | INET_FLAGS_IS_ICSK = 16, |
f04b8d34 | 264 | INET_FLAGS_NODEFRAG = 17, |
ca571e2e | 265 | INET_FLAGS_BIND_ADDRESS_NO_PORT = 18, |
08e39c0d | 266 | INET_FLAGS_DEFER_CONNECT = 19, |
d986f521 | 267 | INET_FLAGS_MC6_LOOP = 20, |
dcae7462 | 268 | INET_FLAGS_RECVERR6_RFC4884 = 21, |
6559c0ff | 269 | INET_FLAGS_MC6_ALL = 22, |
5121516b ED |
270 | INET_FLAGS_AUTOFLOWLABEL_SET = 23, |
271 | INET_FLAGS_AUTOFLOWLABEL = 24, | |
1086ca7c | 272 | INET_FLAGS_DONTFRAG = 25, |
3fa29971 | 273 | INET_FLAGS_RECVERR6 = 26, |
3cccda8d | 274 | INET_FLAGS_REPFLOW = 27, |
83cd5eb6 | 275 | INET_FLAGS_RTALERT_ISOLATE = 28, |
859f8b26 | 276 | INET_FLAGS_SNDFLOW = 29, |
eeb78df4 | 277 | INET_FLAGS_RTALERT = 30, |
c274af22 ED |
278 | }; |
279 | ||
c44d13d6 | 280 | /* cmsg flags for inet */ |
c274af22 ED |
281 | #define IP_CMSG_PKTINFO BIT(INET_FLAGS_PKTINFO) |
282 | #define IP_CMSG_TTL BIT(INET_FLAGS_TTL) | |
283 | #define IP_CMSG_TOS BIT(INET_FLAGS_TOS) | |
284 | #define IP_CMSG_RECVOPTS BIT(INET_FLAGS_RECVOPTS) | |
285 | #define IP_CMSG_RETOPTS BIT(INET_FLAGS_RETOPTS) | |
286 | #define IP_CMSG_PASSSEC BIT(INET_FLAGS_PASSSEC) | |
287 | #define IP_CMSG_ORIGDSTADDR BIT(INET_FLAGS_ORIGDSTADDR) | |
288 | #define IP_CMSG_CHECKSUM BIT(INET_FLAGS_CHECKSUM) | |
289 | #define IP_CMSG_RECVFRAGSIZE BIT(INET_FLAGS_RECVFRAGSIZE) | |
290 | ||
291 | #define IP_CMSG_ALL (IP_CMSG_PKTINFO | IP_CMSG_TTL | \ | |
292 | IP_CMSG_TOS | IP_CMSG_RECVOPTS | \ | |
293 | IP_CMSG_RETOPTS | IP_CMSG_PASSSEC | \ | |
294 | IP_CMSG_ORIGDSTADDR | IP_CMSG_CHECKSUM | \ | |
295 | IP_CMSG_RECVFRAGSIZE) | |
296 | ||
297 | static inline unsigned long inet_cmsg_flags(const struct inet_sock *inet) | |
298 | { | |
299 | return READ_ONCE(inet->inet_flags) & IP_CMSG_ALL; | |
300 | } | |
301 | ||
302 | #define inet_test_bit(nr, sk) \ | |
303 | test_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags) | |
304 | #define inet_set_bit(nr, sk) \ | |
305 | set_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags) | |
306 | #define inet_clear_bit(nr, sk) \ | |
307 | clear_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags) | |
308 | #define inet_assign_bit(nr, sk, val) \ | |
309 | assign_bit(INET_FLAGS_##nr, &inet_sk(sk)->inet_flags, val) | |
c44d13d6 | 310 | |
bd5eb35f ED |
311 | /** |
312 | * sk_to_full_sk - Access to a full socket | |
313 | * @sk: pointer to a socket | |
314 | * | |
315 | * SYNACK messages might be attached to request sockets. | |
54abc686 ED |
316 | * Some places want to reach the listener in this case. |
317 | */ | |
bd5eb35f | 318 | static inline struct sock *sk_to_full_sk(struct sock *sk) |
54abc686 | 319 | { |
bd5eb35f | 320 | #ifdef CONFIG_INET |
54abc686 ED |
321 | if (sk && sk->sk_state == TCP_NEW_SYN_RECV) |
322 | sk = inet_reqsk(sk)->rsk_listener; | |
bd5eb35f ED |
323 | #endif |
324 | return sk; | |
325 | } | |
326 | ||
327 | /* sk_to_full_sk() variant with a const argument */ | |
328 | static inline const struct sock *sk_const_to_full_sk(const struct sock *sk) | |
329 | { | |
330 | #ifdef CONFIG_INET | |
331 | if (sk && sk->sk_state == TCP_NEW_SYN_RECV) | |
332 | sk = ((const struct request_sock *)sk)->rsk_listener; | |
333 | #endif | |
54abc686 ED |
334 | return sk; |
335 | } | |
336 | ||
bd5eb35f ED |
337 | static inline struct sock *skb_to_full_sk(const struct sk_buff *skb) |
338 | { | |
339 | return sk_to_full_sk(skb->sk); | |
340 | } | |
341 | ||
abc17a11 | 342 | #define inet_sk(ptr) container_of_const(ptr, struct inet_sock, sk) |
14c85021 ACM |
343 | |
344 | static inline void __inet_sk_copy_descendant(struct sock *sk_to, | |
345 | const struct sock *sk_from, | |
346 | const int ancestor_size) | |
347 | { | |
348 | memcpy(inet_sk(sk_to) + 1, inet_sk(sk_from) + 1, | |
349 | sk_from->sk_prot->obj_size - ancestor_size); | |
350 | } | |
14c85021 | 351 | |
1fd51155 | 352 | int inet_sk_rebuild_header(struct sock *sk); |
986ffdfd YS |
353 | |
354 | /** | |
355 | * inet_sk_state_load - read sk->sk_state for lockless contexts | |
356 | * @sk: socket pointer | |
357 | * | |
358 | * Paired with inet_sk_state_store(). Used in places we don't hold socket lock: | |
359 | * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ... | |
360 | */ | |
361 | static inline int inet_sk_state_load(const struct sock *sk) | |
362 | { | |
363 | /* state change might impact lockless readers. */ | |
364 | return smp_load_acquire(&sk->sk_state); | |
365 | } | |
366 | ||
367 | /** | |
368 | * inet_sk_state_store - update sk->sk_state | |
369 | * @sk: socket pointer | |
370 | * @newstate: new state | |
371 | * | |
372 | * Paired with inet_sk_state_load(). Should be used in contexts where | |
373 | * state change might impact lockless readers. | |
374 | */ | |
563e0bb0 | 375 | void inet_sk_state_store(struct sock *sk, int newstate); |
14c85021 | 376 | |
986ffdfd YS |
377 | void inet_sk_set_state(struct sock *sk, int state); |
378 | ||
65cd8033 HFS |
379 | static inline unsigned int __inet_ehashfn(const __be32 laddr, |
380 | const __u16 lport, | |
381 | const __be32 faddr, | |
382 | const __be16 fport, | |
383 | u32 initval) | |
14c85021 | 384 | { |
7adc3830 DM |
385 | return jhash_3words((__force __u32) laddr, |
386 | (__force __u32) faddr, | |
b3da2cf3 | 387 | ((__u32) lport) << 16 | (__force __u32)fport, |
65cd8033 | 388 | initval); |
14c85021 ACM |
389 | } |
390 | ||
e49bb337 | 391 | struct request_sock *inet_reqsk_alloc(const struct request_sock_ops *ops, |
a1a5344d ED |
392 | struct sock *sk_listener, |
393 | bool attach_listener); | |
ce4a7d0d | 394 | |
88ef4a5a KK |
395 | static inline __u8 inet_sk_flowi_flags(const struct sock *sk) |
396 | { | |
a4daad6b DM |
397 | __u8 flags = 0; |
398 | ||
4bd0623f | 399 | if (inet_test_bit(TRANSPARENT, sk) || inet_test_bit(HDRINCL, sk)) |
a4daad6b | 400 | flags |= FLOWI_FLAG_ANYSRC; |
a4daad6b | 401 | return flags; |
88ef4a5a KK |
402 | } |
403 | ||
224d019c TH |
404 | static inline void inet_inc_convert_csum(struct sock *sk) |
405 | { | |
406 | inet_sk(sk)->convert_csum++; | |
407 | } | |
408 | ||
409 | static inline void inet_dec_convert_csum(struct sock *sk) | |
410 | { | |
411 | if (inet_sk(sk)->convert_csum > 0) | |
412 | inet_sk(sk)->convert_csum--; | |
413 | } | |
414 | ||
415 | static inline bool inet_get_convert_csum(struct sock *sk) | |
416 | { | |
417 | return !!inet_sk(sk)->convert_csum; | |
418 | } | |
419 | ||
83ba4645 VB |
420 | |
421 | static inline bool inet_can_nonlocal_bind(struct net *net, | |
422 | struct inet_sock *inet) | |
423 | { | |
289d3b21 | 424 | return READ_ONCE(net->ipv4.sysctl_ip_nonlocal_bind) || |
3f7e7532 | 425 | test_bit(INET_FLAGS_FREEBIND, &inet->inet_flags) || |
4bd0623f | 426 | test_bit(INET_FLAGS_TRANSPARENT, &inet->inet_flags); |
83ba4645 VB |
427 | } |
428 | ||
8ff978b8 RPB |
429 | static inline bool inet_addr_valid_or_nonlocal(struct net *net, |
430 | struct inet_sock *inet, | |
431 | __be32 addr, | |
432 | int addr_type) | |
433 | { | |
434 | return inet_can_nonlocal_bind(net, inet) || | |
435 | addr == htonl(INADDR_ANY) || | |
436 | addr_type == RTN_LOCAL || | |
437 | addr_type == RTN_MULTICAST || | |
438 | addr_type == RTN_BROADCAST; | |
439 | } | |
440 | ||
14c85021 | 441 | #endif /* _INET_SOCK_H */ |