![[first]](../icons/first.png){kind=icon}
![[previous]](../icons/n_left.png){kind=icon}
![[next]](../icons/n_right.png){kind=icon}
![[last]](../icons/last.png){kind=icon}
![[top]](../icons/n_top.png){kind=icon}
![[bottom]](../icons/bottom.png){kind=icon}
![[index]](../icons/index.png){kind=icon}
![[help]](../icons/help.png){kind=icon}
*/
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the AF_INET socket handler.
7 *
8 * Version: @(#)sock.h 1.0.4 05/13/93
9 *
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Corey Minyard <wf-rch!minyard@relay.EU.net>
13 * Florian La Roche <flla@stud.uni-sb.de>
14 *
15 * Fixes:
16 * Alan Cox : Volatiles in skbuff pointers. See
17 * skbuff comments. May be overdone,
18 * better to prove they can be removed
19 * than the reverse.
20 * Alan Cox : Added a zapped field for tcp to note
21 * a socket is reset and must stay shut up
22 * Alan Cox : New fields for options
23 * Pauline Middelink : identd support
24 * Alan Cox : Eliminate low level recv/recvfrom
25 *
26 * This program is free software; you can redistribute it and/or
27 * modify it under the terms of the GNU General Public License
28 * as published by the Free Software Foundation; either version
29 * 2 of the License, or (at your option) any later version.
30 */
31 #ifndef _SOCK_H
32 #define _SOCK_H
33
34 #include <linux/timer.h>
35 #include <linux/ip.h> /* struct options */
36 #include <linux/in.h> /* struct sockaddr_in */
37 #include <linux/tcp.h> /* struct tcphdr */
38 #include <linux/config.h>
39
40 #include <linux/netdevice.h>
41 #include <linux/skbuff.h> /* struct sk_buff */
42 #include <net/protocol.h> /* struct inet_protocol */
43 #ifdef CONFIG_AX25
44 #include <net/ax25.h>
45 #ifdef CONFIG_NETROM
46 #include <net/netrom.h>
47 #endif
48 #endif
49 #ifdef CONFIG_IPX
50 #include <net/ipx.h>
51 #endif
52 #ifdef CONFIG_ATALK
53 #include <linux/atalk.h>
54 #endif
55
56 #include <linux/igmp.h>
57
58 #include <asm/atomic.h>
59
60 /* Think big (also on some systems a byte is faster) */
61 #define SOCK_ARRAY_SIZE 256
62
63
64 /*
65 * The AF_UNIX specific socket options
66 */
67
68 struct unix_opt
69 {
70 int family;
71 char * name;
72 int locks;
73 struct inode * inode;
74 struct semaphore readsem;
75 struct sock * other;
76 };
77
78 /*
79 * IP packet socket options
80 */
81
82 struct inet_packet_opt
83 {
84 struct notifier_block notifier; /* Used when bound */
85 struct device *bound_dev;
86 unsigned long dev_stamp;
87 struct packet_type *prot_hook;
88 char device_name[15];
89 };
90
91 /*
92 * Once the IPX ncpd patches are in these are going into protinfo
93 */
94
95 #ifdef CONFIG_IPX
96 struct ipx_opt
97 {
98 ipx_address dest_addr;
99 ipx_interface *intrfc;
100 unsigned short port;
101 #ifdef CONFIG_IPX_INTERN
102 unsigned char node[IPX_NODE_LEN];
103 #endif
104 unsigned short type;
105 /*
106 * To handle asynchronous messages from the NetWare server, we have to
107 * know the connection this socket belongs to.
108 */
109 struct ncp_server *ncp_server;
110
111 };
112 #endif
113
114 #ifdef CONFIG_NUTCP
115 struct tcp_opt
116 {
117 /*
118 * RFC793 variables by their proper names. This means you can
119 * read the code and the spec side by side (and laugh ...)
120 * See RFC793 and RFC1122. The RFC writes these in capitals.
121 */
122 __u32 rcv_nxt; /* What we want to receive next */
123 __u32 rcv_up; /* The urgent point (may not be valid) */
124 __u32 rcv_wnd; /* Current receiver window */
125 __u32 snd_nxt; /* Next sequence we send */
126 __u32 snd_una; /* First byte we want an ack for */
127 __u32 snd_up; /* Outgoing urgent pointer */
128 __u32 snd_wl1; /* Sequence for window update */
129 __u32 snd_wl2; /* Ack sequence for update */
130 /*
131 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
132 */
133 __u32 snd_cwnd; /* Sending congestion window */
134 __u32 snd_ssthresh; /* Slow start size threshold */
135 /*
136 * Timers used by the TCP protocol layer
137 */
138 struct timer_list delack_timer; /* Ack delay */
139 struct timer_list idle_timer; /* Idle watch */
140 struct timer_list completion_timer; /* Up/Down timer */
141 struct timer_list probe_timer; /* Probes */
142 struct timer_list retransmit_timer; /* Resend (no ack) */
143 };
144 #endif
145
146 /*
147 * This structure really needs to be cleaned up.
148 * Most of it is for TCP, and not used by any of
149 * the other protocols.
150 */
151 struct sock
152 {
153 struct options *opt;
154 atomic_t wmem_alloc;
155 atomic_t rmem_alloc;
156 unsigned long allocation; /* Allocation mode */
157 __u32 write_seq;
158 __u32 sent_seq;
159 __u32 acked_seq;
160 __u32 copied_seq;
161 __u32 rcv_ack_seq;
162 __u32 window_seq;
163 __u32 fin_seq;
164 __u32 urg_seq;
165 __u32 urg_data;
166 int users; /* user count */
167 /*
168 * Not all are volatile, but some are, so we
169 * might as well say they all are.
170 */
171 volatile char dead,
172 urginline,
173 intr,
174 blog,
175 done,
176 reuse,
177 keepopen,
178 linger,
179 delay_acks,
180 destroy,
181 ack_timed,
182 no_check,
183 zapped, /* In ax25 & ipx means not linked */
184 broadcast,
185 nonagle,
186 bsdism;
187 unsigned long lingertime;
188 int proc;
189 struct sock *next;
190 struct sock *prev; /* Doubly linked chain.. */
191 struct sock *pair;
192 struct sk_buff * volatile send_head;
193 struct sk_buff * volatile send_tail;
194 struct sk_buff_head back_log;
195 struct sk_buff *partial;
196 struct timer_list partial_timer;
197 long retransmits;
198 struct sk_buff_head write_queue,
199 receive_queue;
200 struct proto *prot;
201 struct wait_queue **sleep;
202 __u32 daddr;
203 __u32 saddr; /* Sending source */
204 __u32 rcv_saddr; /* Bound address */
205 unsigned short max_unacked;
206 unsigned short window;
207 __u32 lastwin_seq; /* sequence number when we last updated the window we offer */
208 volatile unsigned long ato; /* ack timeout */
209 volatile unsigned long lrcvtime; /* jiffies at last rcv */
210 unsigned short bytes_rcv;
211 /*
212 * mss is min(mtu, max_window)
213 */
214 unsigned short mtu; /* mss negotiated in the syn's */
215 volatile unsigned short mss; /* current eff. mss - can change */
216 volatile unsigned short user_mss; /* mss requested by user in ioctl */
217 volatile unsigned short max_window;
218 unsigned long window_clamp;
219 unsigned short num;
220 volatile unsigned short cong_window;
221 volatile unsigned short cong_count;
222 volatile unsigned short ssthresh;
223 volatile unsigned short packets_out;
224 volatile unsigned short shutdown;
225 volatile unsigned long rtt;
226 volatile unsigned long mdev;
227 volatile unsigned long rto;
228
229 /*
230 * currently backoff isn't used, but I'm maintaining it in case
231 * we want to go back to a backoff formula that needs it
232 */
233
234 volatile unsigned short backoff;
235 volatile int err, err_soft; /* Soft holds errors that don't
236 cause failure but are the cause
237 of a persistent failure not just
238 'timed out' */
239 unsigned char protocol;
240 volatile unsigned char state;
241 unsigned char ack_backlog;
242 unsigned char max_ack_backlog;
243 unsigned char priority;
244 unsigned char debug;
245 unsigned short rcvbuf;
246 unsigned short sndbuf;
247 unsigned short type;
248 unsigned char localroute; /* Route locally only */
249 #ifdef CONFIG_AX25
250 ax25_cb *ax25;
251 #ifdef CONFIG_NETROM
252 nr_cb *nr;
253 #endif
254 #endif
255
256 /*
257 * This is where all the private (optional) areas that don't
258 * overlap will eventually live.
259 */
260
261 union
262 {
263 struct unix_opt af_unix;
264 #ifdef CONFIG_ATALK
265 struct atalk_sock af_at;
266 #endif
267 #ifdef CONFIG_IPX
268 struct ipx_opt af_ipx;
269 #endif
270 #ifdef CONFIG_INET
271 struct inet_packet_opt af_packet;
272 #ifdef CONFIG_NUTCP
273 struct tcp_opt af_tcp;
274 #endif
275 #endif
276 } protinfo;
277
278 /*
279 * IP 'private area' or will be eventually
280 */
281 int ip_ttl; /* TTL setting */
282 int ip_tos; /* TOS */
283 struct tcphdr dummy_th;
284 struct timer_list keepalive_timer; /* TCP keepalive hack */
285 struct timer_list retransmit_timer; /* TCP retransmit timer */
286 struct timer_list ack_timer; /* TCP delayed ack timer */
287 int ip_xmit_timeout; /* Why the timeout is running */
288 struct rtable *ip_route_cache; /* Cached output route */
289 unsigned char ip_hdrincl; /* Include headers ? */
290 #ifdef CONFIG_IP_MULTICAST
291 int ip_mc_ttl; /* Multicasting TTL */
292 int ip_mc_loop; /* Loopback */
293 char ip_mc_name[MAX_ADDR_LEN];/* Multicast device name */
294 struct ip_mc_socklist *ip_mc_list; /* Group array */
295 #endif
296
297 /*
298 * This part is used for the timeout functions (timer.c).
299 */
300
301 int timeout; /* What are we waiting for? */
302 struct timer_list timer; /* This is the TIME_WAIT/receive timer
303 * when we are doing IP
304 */
305 struct timeval stamp;
306
307 /*
308 * Identd
309 */
310
311 struct socket *socket;
312
313 /*
314 * Callbacks
315 */
316
317 void (*state_change)(struct sock *sk);
318 void (*data_ready)(struct sock *sk,int bytes);
319 void (*write_space)(struct sock *sk);
320 void (*error_report)(struct sock *sk);
321
322 };
323
324 /*
325 * IP protocol blocks we attach to sockets.
326 */
327
328 struct proto
329 {
330 void (*close)(struct sock *sk, unsigned long timeout);
331 int (*build_header)(struct sk_buff *skb,
332 __u32 saddr,
333 __u32 daddr,
334 struct device **dev, int type,
335 struct options *opt, int len,
336 int tos, int ttl, struct rtable ** rp);
337 int (*connect)(struct sock *sk,
338 struct sockaddr_in *usin, int addr_len);
339 struct sock * (*accept) (struct sock *sk, int flags);
340 void (*queue_xmit)(struct sock *sk,
341 struct device *dev, struct sk_buff *skb,
342 int free);
343 void (*retransmit)(struct sock *sk, int all);
344 void (*write_wakeup)(struct sock *sk);
345 void (*read_wakeup)(struct sock *sk);
346 int (*rcv)(struct sk_buff *buff, struct device *dev,
347 struct options *opt, __u32 daddr,
348 unsigned short len, __u32 saddr,
349 int redo, struct inet_protocol *protocol);
350 int (*select)(struct sock *sk, int which,
351 select_table *wait);
352 int (*ioctl)(struct sock *sk, int cmd,
353 unsigned long arg);
354 int (*init)(struct sock *sk);
355 void (*shutdown)(struct sock *sk, int how);
356 int (*setsockopt)(struct sock *sk, int level, int optname,
357 char *optval, int optlen);
358 int (*getsockopt)(struct sock *sk, int level, int optname,
359 char *optval, int *option);
360 int (*sendmsg)(struct sock *sk, struct msghdr *msg, int len,
361 int noblock, int flags);
362 int (*recvmsg)(struct sock *sk, struct msghdr *msg, int len,
363 int noblock, int flags, int *addr_len);
364 int (*bind)(struct sock *sk, struct sockaddr *uaddr, int addr_len);
365 unsigned short max_header;
366 unsigned long retransmits;
367 char name[32];
368 int inuse, highestinuse;
369 struct sock * sock_array[SOCK_ARRAY_SIZE];
370 };
371
372 #define TIME_WRITE 1
373 #define TIME_CLOSE 2
374 #define TIME_KEEPOPEN 3
375 #define TIME_DESTROY 4
376 #define TIME_DONE 5 /* Used to absorb those last few packets */
377 #define TIME_PROBE0 6
378
379 /*
380 * About 10 seconds
381 */
382
383 #define SOCK_DESTROY_TIME (10*HZ)
384
385 /*
386 * Sockets 0-1023 can't be bound too unless you are superuser
387 */
388
389 #define PROT_SOCK 1024
390
391 #define SHUTDOWN_MASK 3
392 #define RCV_SHUTDOWN 1
393 #define SEND_SHUTDOWN 2
394
395 /*
396 * Used by processes to "lock" a socket state, so that
397 * interrupts and bottom half handlers won't change it
398 * from under us. It essentially blocks any incoming
399 * packets, so that we won't get any new data or any
400 * packets that change the state of the socket.
401 *
402 * Note the 'barrier()' calls: gcc may not move a lock
403 * "downwards" or a unlock "upwards" when optimizing.
404 */
405 extern void __release_sock(struct sock *sk);
406
407 static inline void lock_sock(struct sock *sk)
/* ![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[top]](../icons/top.png)
*/
408 {
409 #if 1
410 /* debugging code: the test isn't even 100% correct, but it can catch bugs */
411 /* Note that a double lock is ok in theory - it's just _usually_ a bug */
412 if (sk->users) {
413 __label__ here;
414 printk("double lock on socket at %p\n", &&here);
415 here:
416 }
417 #endif
418 sk->users++;
419 barrier();
420 }
421
422 static inline void release_sock(struct sock *sk)
/* ![[previous]](../icons/left.png)
*/
423 {
424 barrier();
425 #if 1
426 /* debugging code: remove me when ok */
427 if (sk->users == 0) {
428 __label__ here;
429 sk->users = 1;
430 printk("trying to unlock unlocked socket at %p\n", &&here);
431 here:
432 }
433 #endif
434 if (!--sk->users)
435 __release_sock(sk);
436 }
437
438
439 extern void destroy_sock(struct sock *sk);
440 extern unsigned short get_new_socknum(struct proto *,
441 unsigned short);
442 extern void put_sock(unsigned short, struct sock *);
443 extern struct sock *get_sock(struct proto *, unsigned short,
444 unsigned long, unsigned short,
445 unsigned long);
446 extern struct sock *get_sock_mcast(struct sock *, unsigned short,
447 unsigned long, unsigned short,
448 unsigned long);
449 extern struct sock *get_sock_raw(struct sock *, unsigned short,
450 unsigned long, unsigned long);
451
452 extern struct sk_buff *sock_wmalloc(struct sock *sk,
453 unsigned long size, int force,
454 int priority);
455 extern struct sk_buff *sock_rmalloc(struct sock *sk,
456 unsigned long size, int force,
457 int priority);
458 extern void sock_wfree(struct sock *sk,
459 struct sk_buff *skb);
460 extern void sock_rfree(struct sock *sk,
461 struct sk_buff *skb);
462 extern unsigned long sock_rspace(struct sock *sk);
463 extern unsigned long sock_wspace(struct sock *sk);
464
465 extern int sock_setsockopt(struct sock *sk, int level,
466 int op, char *optval,
467 int optlen);
468
469 extern int sock_getsockopt(struct sock *sk, int level,
470 int op, char *optval,
471 int *optlen);
472 extern struct sk_buff *sock_alloc_send_skb(struct sock *skb,
473 unsigned long size,
474 unsigned long fallback,
475 int noblock,
476 int *errcode);
477
478 /*
479 * Queue a received datagram if it will fit. Stream and sequenced
480 * protocols can't normally use this as they need to fit buffers in
481 * and play with them.
482 *
483 * Inlined as its very short and called for pretty much every
484 * packet ever received.
485 */
486
487 extern __inline__ int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
/* */
488 {
489 if (sk->rmem_alloc + skb->truesize >= sk->rcvbuf)
490 return -ENOMEM;
491 atomic_add(skb->truesize, &sk->rmem_alloc);
492 skb->sk=sk;
493 skb_queue_tail(&sk->receive_queue,skb);
494 if(!sk->dead)
495 sk->data_ready(sk,skb->len);
496 return 0;
497 }
498
499 /*
500 * Recover an error report and clear atomically
501 */
502
503 extern __inline__ int sock_error(struct sock *sk)
/* ![[last]](../icons/n_last.png)
*/
504 {
505 int err=xchg(&sk->err,0);
506 return -err;
507 }
508
509 /*
510 * Declarations from timer.c
511 */
512
513 extern struct sock *timer_base;
514
515 extern void delete_timer (struct sock *);
516 extern void reset_timer (struct sock *, int, unsigned long);
517 extern void net_timer (unsigned long);
518
519
520 /*
521 * Enable debug/info messages
522 */
523
524 #define NETDEBUG(x) x
525
526 #endif /* _SOCK_H */
![[bottom]](../icons/n_bottom.png){kind=icon}
*/