![[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 * Implementation of the Transmission Control Protocol(TCP).
7 *
8 * Version: @(#)tcp.c 1.0.16 05/25/93
9 *
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Charles Hedrick, <hedrick@klinzhai.rutgers.edu>
16 * Linus Torvalds, <torvalds@cs.helsinki.fi>
17 * Alan Cox, <gw4pts@gw4pts.ampr.org>
18 * Matthew Dillon, <dillon@apollo.west.oic.com>
19 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
20 * Jorge Cwik, <jorge@laser.satlink.net>
21 *
22 * Fixes:
23 * Alan Cox : Numerous verify_area() calls
24 * Alan Cox : Set the ACK bit on a reset
25 * Alan Cox : Stopped it crashing if it closed while
26 * sk->inuse=1 and was trying to connect
27 * (tcp_err()).
28 * Alan Cox : All icmp error handling was broken
29 * pointers passed where wrong and the
30 * socket was looked up backwards. Nobody
31 * tested any icmp error code obviously.
32 * Alan Cox : tcp_err() now handled properly. It
33 * wakes people on errors. select
34 * behaves and the icmp error race
35 * has gone by moving it into sock.c
36 * Alan Cox : tcp_send_reset() fixed to work for
37 * everything not just packets for
38 * unknown sockets.
39 * Alan Cox : tcp option processing.
40 * Alan Cox : Reset tweaked (still not 100%) [Had
41 * syn rule wrong]
42 * Herp Rosmanith : More reset fixes
43 * Alan Cox : No longer acks invalid rst frames.
44 * Acking any kind of RST is right out.
45 * Alan Cox : Sets an ignore me flag on an rst
46 * receive otherwise odd bits of prattle
47 * escape still
48 * Alan Cox : Fixed another acking RST frame bug.
49 * Should stop LAN workplace lockups.
50 * Alan Cox : Some tidyups using the new skb list
51 * facilities
52 * Alan Cox : sk->keepopen now seems to work
53 * Alan Cox : Pulls options out correctly on accepts
54 * Alan Cox : Fixed assorted sk->rqueue->next errors
55 * Alan Cox : PSH doesn't end a TCP read. Switched a
56 * bit to skb ops.
57 * Alan Cox : Tidied tcp_data to avoid a potential
58 * nasty.
59 * Alan Cox : Added some better commenting, as the
60 * tcp is hard to follow
61 * Alan Cox : Removed incorrect check for 20 * psh
62 * Michael O'Reilly : ack < copied bug fix.
63 * Johannes Stille : Misc tcp fixes (not all in yet).
64 * Alan Cox : FIN with no memory -> CRASH
65 * Alan Cox : Added socket option proto entries.
66 * Also added awareness of them to accept.
67 * Alan Cox : Added TCP options (SOL_TCP)
68 * Alan Cox : Switched wakeup calls to callbacks,
69 * so the kernel can layer network
70 * sockets.
71 * Alan Cox : Use ip_tos/ip_ttl settings.
72 * Alan Cox : Handle FIN (more) properly (we hope).
73 * Alan Cox : RST frames sent on unsynchronised
74 * state ack error.
75 * Alan Cox : Put in missing check for SYN bit.
76 * Alan Cox : Added tcp_select_window() aka NET2E
77 * window non shrink trick.
78 * Alan Cox : Added a couple of small NET2E timer
79 * fixes
80 * Charles Hedrick : TCP fixes
81 * Toomas Tamm : TCP window fixes
82 * Alan Cox : Small URG fix to rlogin ^C ack fight
83 * Charles Hedrick : Rewrote most of it to actually work
84 * Linus : Rewrote tcp_read() and URG handling
85 * completely
86 * Gerhard Koerting: Fixed some missing timer handling
87 * Matthew Dillon : Reworked TCP machine states as per RFC
88 * Gerhard Koerting: PC/TCP workarounds
89 * Adam Caldwell : Assorted timer/timing errors
90 * Matthew Dillon : Fixed another RST bug
91 * Alan Cox : Move to kernel side addressing changes.
92 * Alan Cox : Beginning work on TCP fastpathing
93 * (not yet usable)
94 * Arnt Gulbrandsen: Turbocharged tcp_check() routine.
95 * Alan Cox : TCP fast path debugging
96 * Alan Cox : Window clamping
97 * Michael Riepe : Bug in tcp_check()
98 * Matt Dillon : More TCP improvements and RST bug fixes
99 * Matt Dillon : Yet more small nasties remove from the
100 * TCP code (Be very nice to this man if
101 * tcp finally works 100%) 8)
102 * Alan Cox : BSD accept semantics.
103 * Alan Cox : Reset on closedown bug.
104 * Peter De Schrijver : ENOTCONN check missing in tcp_sendto().
105 * Michael Pall : Handle select() after URG properly in
106 * all cases.
107 * Michael Pall : Undo the last fix in tcp_read_urg()
108 * (multi URG PUSH broke rlogin).
109 * Michael Pall : Fix the multi URG PUSH problem in
110 * tcp_readable(), select() after URG
111 * works now.
112 * Michael Pall : recv(...,MSG_OOB) never blocks in the
113 * BSD api.
114 * Alan Cox : Changed the semantics of sk->socket to
115 * fix a race and a signal problem with
116 * accept() and async I/O.
117 * Alan Cox : Relaxed the rules on tcp_sendto().
118 * Yury Shevchuk : Really fixed accept() blocking problem.
119 * Craig I. Hagan : Allow for BSD compatible TIME_WAIT for
120 * clients/servers which listen in on
121 * fixed ports.
122 * Alan Cox : Cleaned the above up and shrank it to
123 * a sensible code size.
124 * Alan Cox : Self connect lockup fix.
125 * Alan Cox : No connect to multicast.
126 * Ross Biro : Close unaccepted children on master
127 * socket close.
128 * Alan Cox : Reset tracing code.
129 * Alan Cox : Spurious resets on shutdown.
130 * Alan Cox : Giant 15 minute/60 second timer error
131 * Alan Cox : Small whoops in selecting before an
132 * accept.
133 * Alan Cox : Kept the state trace facility since
134 * it's handy for debugging.
135 * Alan Cox : More reset handler fixes.
136 * Alan Cox : Started rewriting the code based on
137 * the RFC's for other useful protocol
138 * references see: Comer, KA9Q NOS, and
139 * for a reference on the difference
140 * between specifications and how BSD
141 * works see the 4.4lite source.
142 * A.N.Kuznetsov : Don't time wait on completion of tidy
143 * close.
144 * Linus Torvalds : Fin/Shutdown & copied_seq changes.
145 * Linus Torvalds : Fixed BSD port reuse to work first syn
146 * Alan Cox : Reimplemented timers as per the RFC
147 * and using multiple timers for sanity.
148 * Alan Cox : Small bug fixes, and a lot of new
149 * comments.
150 * Alan Cox : Fixed dual reader crash by locking
151 * the buffers (much like datagram.c)
152 * Alan Cox : Fixed stuck sockets in probe. A probe
153 * now gets fed up of retrying without
154 * (even a no space) answer.
155 * Alan Cox : Extracted closing code better
156 * Alan Cox : Fixed the closing state machine to
157 * resemble the RFC.
158 * Alan Cox : More 'per spec' fixes.
159 * Jorge Cwik : Even faster checksumming.
160 * Alan Cox : tcp_data() doesn't ack illegal PSH
161 * only frames. At least one pc tcp stack
162 * generates them.
163 * Alan Cox : Cache last socket.
164 * Alan Cox : Per route irtt.
165 * Matt Day : Select() match BSD precisely on error
166 * Alan Cox : New buffers
167 * Marc Tamsky : Various sk->prot->retransmits and
168 * sk->retransmits misupdating fixed.
169 * Fixed tcp_write_timeout: stuck close,
170 * and TCP syn retries gets used now.
171 * Mark Yarvis : In tcp_read_wakeup(), don't send an
172 * ack if stat is TCP_CLOSED.
173 * Alan Cox : Look up device on a retransmit - routes may
174 * change. Doesn't yet cope with MSS shrink right
175 * but its a start!
176 * Marc Tamsky : Closing in closing fixes.
177 * Mike Shaver : RFC1122 verifications.
178 * Alan Cox : rcv_saddr errors.
179 * Alan Cox : Block double connect().
180 * Alan Cox : Small hooks for enSKIP.
181 * Alexey Kuznetsov: Path MTU discovery.
182 * Alan Cox : Support soft errors.
183 * Alan Cox : Fix MTU discovery pathalogical case
184 * when the remote claims no mtu!
185 * Marc Tamsky : TCP_CLOSE fix.
186 * Colin (G3TNE) : Send a reset on syn ack replies in
187 * window but wrong (fixes NT lpd problems)
188 * Pedro Roque : Better TCP window handling, delayed ack.
189 * Joerg Reuter : No modification of locked buffers in
190 * tcp_do_retransmit()
191 *
192 * To Fix:
193 * Fast path the code. Two things here - fix the window calculation
194 * so it doesn't iterate over the queue, also spot packets with no funny
195 * options arriving in order and process directly.
196 *
197 * Rewrite output state machine to use a single queue.
198 * Speed up input assembly algorithm.
199 * RFC1323 - PAWS and window scaling. PAWS is required for IPv6 so we
200 * could do with it working on IPv4
201 * User settable/learned rtt/max window/mtu
202 *
203 * Change the fundamental structure to a single send queue maintained
204 * by TCP (removing the bogus ip stuff [thus fixing mtu drops on
205 * active routes too]). Cut the queue off in tcp_retransmit/
206 * tcp_transmit.
207 * Change the receive queue to assemble as it goes. This lets us
208 * dispose of most of tcp_sequence, half of tcp_ack and chunks of
209 * tcp_data/tcp_read as well as the window shrink crud.
210 * Separate out duplicated code - tcp_alloc_skb, tcp_build_ack
211 * tcp_queue_skb seem obvious routines to extract.
212 *
213 * This program is free software; you can redistribute it and/or
214 * modify it under the terms of the GNU General Public License
215 * as published by the Free Software Foundation; either version
216 * 2 of the License, or(at your option) any later version.
217 *
218 * Description of States:
219 *
220 * TCP_SYN_SENT sent a connection request, waiting for ack
221 *
222 * TCP_SYN_RECV received a connection request, sent ack,
223 * waiting for final ack in three-way handshake.
224 *
225 * TCP_ESTABLISHED connection established
226 *
227 * TCP_FIN_WAIT1 our side has shutdown, waiting to complete
228 * transmission of remaining buffered data
229 *
230 * TCP_FIN_WAIT2 all buffered data sent, waiting for remote
231 * to shutdown
232 *
233 * TCP_CLOSING both sides have shutdown but we still have
234 * data we have to finish sending
235 *
236 * TCP_TIME_WAIT timeout to catch resent junk before entering
237 * closed, can only be entered from FIN_WAIT2
238 * or CLOSING. Required because the other end
239 * may not have gotten our last ACK causing it
240 * to retransmit the data packet (which we ignore)
241 *
242 * TCP_CLOSE_WAIT remote side has shutdown and is waiting for
243 * us to finish writing our data and to shutdown
244 * (we have to close() to move on to LAST_ACK)
245 *
246 * TCP_LAST_ACK out side has shutdown after remote has
247 * shutdown. There may still be data in our
248 * buffer that we have to finish sending
249 *
250 * TCP_CLOSE socket is finished
251 */
252
253 /*
254 * RFC1122 status:
255 * NOTE: I'm not going to be doing comments in the code for this one except
256 * for violations and the like. tcp.c is just too big... If I say something
257 * "does?" or "doesn't?", it means I'm not sure, and will have to hash it out
258 * with Alan. -- MS 950903
259 *
260 * Use of PSH (4.2.2.2)
261 * MAY aggregate data sent without the PSH flag. (does)
262 * MAY queue data received without the PSH flag. (does)
263 * SHOULD collapse successive PSH flags when it packetizes data. (doesn't)
264 * MAY implement PSH on send calls. (doesn't, thus:)
265 * MUST NOT buffer data indefinitely (doesn't [1 second])
266 * MUST set PSH on last segment (does)
267 * MAY pass received PSH to application layer (doesn't)
268 * SHOULD send maximum-sized segment whenever possible. (almost always does)
269 *
270 * Window Size (4.2.2.3, 4.2.2.16)
271 * MUST treat window size as an unsigned number (does)
272 * SHOULD treat window size as a 32-bit number (does not)
273 * MUST NOT shrink window once it is offered (does not normally)
274 *
275 * Urgent Pointer (4.2.2.4)
276 * **MUST point urgent pointer to last byte of urgent data (not right
277 * after). (doesn't, to be like BSD)
278 * MUST inform application layer asynchronously of incoming urgent
279 * data. (does)
280 * MUST provide application with means of determining the amount of
281 * urgent data pending. (does)
282 * **MUST support urgent data sequence of arbitrary length. (doesn't, but
283 * it's sort of tricky to fix, as urg_ptr is a 16-bit quantity)
284 * [Follows BSD 1 byte of urgent data]
285 *
286 * TCP Options (4.2.2.5)
287 * MUST be able to receive TCP options in any segment. (does)
288 * MUST ignore unsupported options (does)
289 *
290 * Maximum Segment Size Option (4.2.2.6)
291 * MUST implement both sending and receiving MSS. (does)
292 * SHOULD send an MSS with every SYN where receive MSS != 536 (MAY send
293 * it always). (does, even when MSS == 536, which is legal)
294 * MUST assume MSS == 536 if no MSS received at connection setup (does)
295 * MUST calculate "effective send MSS" correctly:
296 * min(physical_MTU, remote_MSS+20) - sizeof(tcphdr) - sizeof(ipopts)
297 * (does - but allows operator override)
298 *
299 * TCP Checksum (4.2.2.7)
300 * MUST generate and check TCP checksum. (does)
301 *
302 * Initial Sequence Number Selection (4.2.2.8)
303 * MUST use the RFC 793 clock selection mechanism. (doesn't, but it's
304 * OK: RFC 793 specifies a 250KHz clock, while we use 1MHz, which is
305 * necessary for 10Mbps networks - and harder than BSD to spoof!)
306 *
307 * Simultaneous Open Attempts (4.2.2.10)
308 * MUST support simultaneous open attempts (does)
309 *
310 * Recovery from Old Duplicate SYN (4.2.2.11)
311 * MUST keep track of active vs. passive open (does)
312 *
313 * RST segment (4.2.2.12)
314 * SHOULD allow an RST segment to contain data (does, but doesn't do
315 * anything with it, which is standard)
316 *
317 * Closing a Connection (4.2.2.13)
318 * MUST inform application of whether connectin was closed by RST or
319 * normal close. (does)
320 * MAY allow "half-duplex" close (treat connection as closed for the
321 * local app, even before handshake is done). (does)
322 * MUST linger in TIME_WAIT for 2 * MSL (does)
323 *
324 * Retransmission Timeout (4.2.2.15)
325 * MUST implement Jacobson's slow start and congestion avoidance
326 * stuff. (does)
327 *
328 * Probing Zero Windows (4.2.2.17)
329 * MUST support probing of zero windows. (does)
330 * MAY keep offered window closed indefinitely. (does)
331 * MUST allow remote window to stay closed indefinitely. (does)
332 *
333 * Passive Open Calls (4.2.2.18)
334 * MUST NOT let new passive open affect other connections. (doesn't)
335 * MUST support passive opens (LISTENs) concurrently. (does)
336 *
337 * Time to Live (4.2.2.19)
338 * MUST make TCP TTL configurable. (does - IP_TTL option)
339 *
340 * Event Processing (4.2.2.20)
341 * SHOULD queue out-of-order segments. (does)
342 * MUST aggregate ACK segments whenever possible. (does but badly)
343 *
344 * Retransmission Timeout Calculation (4.2.3.1)
345 * MUST implement Karn's algorithm and Jacobson's algorithm for RTO
346 * calculation. (does, or at least explains them in the comments 8*b)
347 * SHOULD initialize RTO to 0 and RTT to 3. (does)
348 *
349 * When to Send an ACK Segment (4.2.3.2)
350 * SHOULD implement delayed ACK. (does)
351 * MUST keep ACK delay < 0.5 sec. (does)
352 *
353 * When to Send a Window Update (4.2.3.3)
354 * MUST implement receiver-side SWS. (does)
355 *
356 * When to Send Data (4.2.3.4)
357 * MUST implement sender-side SWS. (does)
358 * SHOULD implement Nagle algorithm. (does)
359 *
360 * TCP Connection Failures (4.2.3.5)
361 * MUST handle excessive retransmissions "properly" (see the RFC). (does)
362 * SHOULD inform application layer of soft errors. (does)
363 *
364 * TCP Keep-Alives (4.2.3.6)
365 * MAY provide keep-alives. (does)
366 * MUST make keep-alives configurable on a per-connection basis. (does)
367 * MUST default to no keep-alives. (does)
368 * **MUST make keep-alive interval configurable. (doesn't)
369 * **MUST make default keep-alive interval > 2 hours. (doesn't)
370 * MUST NOT interpret failure to ACK keep-alive packet as dead
371 * connection. (doesn't)
372 * SHOULD send keep-alive with no data. (does)
373 *
374 * TCP Multihoming (4.2.3.7)
375 * MUST get source address from IP layer before sending first
376 * SYN. (does)
377 * MUST use same local address for all segments of a connection. (does)
378 *
379 * IP Options (4.2.3.8)
380 * MUST ignore unsupported IP options. (does)
381 * MAY support Time Stamp and Record Route. (does)
382 * MUST allow application to specify a source route. (does)
383 * MUST allow receieved Source Route option to set route for all future
384 * segments on this connection. (does not (security issues))
385 *
386 * ICMP messages (4.2.3.9)
387 * MUST act on ICMP errors. (does)
388 * MUST slow transmission upon receipt of a Source Quench. (does)
389 * MUST NOT abort connection upon receipt of soft Destination
390 * Unreachables (0, 1, 5), Time Exceededs and Parameter
391 * Problems. (doesn't)
392 * SHOULD report soft Destination Unreachables etc. to the
393 * application. (does)
394 * SHOULD abort connection upon receipt of hard Destination Unreachable
395 * messages (2, 3, 4). (does)
396 *
397 * Remote Address Validation (4.2.3.10)
398 * MUST reject as an error OPEN for invalid remote IP address. (does)
399 * MUST ignore SYN with invalid source address. (does)
400 * MUST silently discard incoming SYN for broadcast/multicast
401 * address. (does)
402 *
403 * Asynchronous Reports (4.2.4.1)
404 * MUST provide mechanism for reporting soft errors to application
405 * layer. (does)
406 *
407 * Type of Service (4.2.4.2)
408 * MUST allow application layer to set Type of Service. (does IP_TOS)
409 *
410 * (Whew. -- MS 950903)
411 **/
412
413 #include <linux/config.h>
414 #include <linux/types.h>
415 #include <linux/fcntl.h>
416
417 #include <net/icmp.h>
418 #include <net/tcp.h>
419
420 #include <asm/segment.h>
421
422 unsigned long seq_offset;
423 struct tcp_mib tcp_statistics;
424
425 static void tcp_close(struct sock *sk, int timeout);
426
427 /*
428 * The less said about this the better, but it works and will do for 1.2 (and 1.4 ;))
429 */
430
431 struct wait_queue *master_select_wakeup;
432
433 /*
434 * Find someone to 'accept'. Must be called with
435 * sk->inuse=1 or cli()
436 */
437
438 static struct sk_buff *tcp_find_established(struct sock *s)
/* ![[next]](../icons/right.png)
![[first]](../icons/n_first.png)
![[top]](../icons/top.png)
*/
439 {
440 struct sk_buff *p=skb_peek(&s->receive_queue);
441 if(p==NULL)
442 return NULL;
443 do
444 {
445 if(p->sk->state == TCP_ESTABLISHED || p->sk->state >= TCP_FIN_WAIT1)
446 return p;
447 p=p->next;
448 }
449 while(p!=(struct sk_buff *)&s->receive_queue);
450 return NULL;
451 }
452
453 /*
454 * Remove a completed connection and return it. This is used by
455 * tcp_accept() to get connections from the queue.
456 */
457
458 static struct sk_buff *tcp_dequeue_established(struct sock *s)
/* ![[previous]](../icons/left.png)
*/
459 {
460 struct sk_buff *skb;
461 unsigned long flags;
462 save_flags(flags);
463 cli();
464 skb=tcp_find_established(s);
465 if(skb!=NULL)
466 skb_unlink(skb); /* Take it off the queue */
467 restore_flags(flags);
468 return skb;
469 }
470
471 /*
472 * This routine closes sockets which have been at least partially
473 * opened, but not yet accepted. Currently it is only called by
474 * tcp_close, and timeout mirrors the value there.
475 */
476
477 static void tcp_close_pending (struct sock *sk)
/* */
478 {
479 struct sk_buff *skb;
480
481 while ((skb = skb_dequeue(&sk->receive_queue)) != NULL)
482 {
483 skb->sk->dead=1;
484 tcp_close(skb->sk, 0);
485 kfree_skb(skb, FREE_READ);
486 }
487 return;
488 }
489
490 /*
491 * Enter the time wait state.
492 */
493
494 void tcp_time_wait(struct sock *sk)
/* */
495 {
496 tcp_set_state(sk,TCP_TIME_WAIT);
497 sk->shutdown = SHUTDOWN_MASK;
498 if (!sk->dead)
499 sk->state_change(sk);
500 tcp_reset_msl_timer(sk, TIME_CLOSE, TCP_TIMEWAIT_LEN);
501 }
502
503
504 /*
505 * This routine is called by the ICMP module when it gets some
506 * sort of error condition. If err < 0 then the socket should
507 * be closed and the error returned to the user. If err > 0
508 * it's just the icmp type << 8 | icmp code. After adjustment
509 * header points to the first 8 bytes of the tcp header. We need
510 * to find the appropriate port.
511 */
512
513 void tcp_err(int type, int code, unsigned char *header, __u32 daddr,
/* */
514 __u32 saddr, struct inet_protocol *protocol)
515 {
516 struct tcphdr *th = (struct tcphdr *)header;
517 struct sock *sk;
518
519 /*
520 * This one is _WRONG_. FIXME urgently.
521 */
522 #ifndef CONFIG_NO_PATH_MTU_DISCOVERY
523 struct iphdr *iph=(struct iphdr *)(header-sizeof(struct iphdr));
524 #endif
525 th =(struct tcphdr *)header;
526 sk = get_sock(&tcp_prot, th->source, daddr, th->dest, saddr);
527
528 if (sk == NULL)
529 return;
530
531 if (type == ICMP_SOURCE_QUENCH)
532 {
533 /*
534 * FIXME:
535 * For now we will just trigger a linear backoff.
536 * The slow start code should cause a real backoff here.
537 */
538 if (sk->cong_window > 4)
539 sk->cong_window--;
540 return;
541 }
542
543 if (type == ICMP_PARAMETERPROB)
544 {
545 sk->err=EPROTO;
546 sk->error_report(sk);
547 }
548
549 #ifndef CONFIG_NO_PATH_MTU_DISCOVERY
550 if (type == ICMP_DEST_UNREACH && code == ICMP_FRAG_NEEDED)
551 {
552 struct rtable * rt;
553 /*
554 * Ugly trick to pass MTU to protocol layer.
555 * Really we should add argument "info" to error handler.
556 */
557 unsigned short new_mtu = ntohs(iph->id);
558
559 if ((rt = sk->ip_route_cache) != NULL)
560 if (rt->rt_mtu > new_mtu)
561 rt->rt_mtu = new_mtu;
562
563 if (sk->mtu > new_mtu - sizeof(struct iphdr) - sizeof(struct tcphdr)
564 && new_mtu > sizeof(struct iphdr)+sizeof(struct tcphdr))
565 sk->mtu = new_mtu - sizeof(struct iphdr) - sizeof(struct tcphdr);
566
567 return;
568 }
569 #endif
570
571 /*
572 * If we've already connected we will keep trying
573 * until we time out, or the user gives up.
574 */
575
576 if (code < 13)
577 {
578 if(icmp_err_convert[code].fatal || sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
579 {
580 sk->err = icmp_err_convert[code].errno;
581 if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
582 {
583 tcp_statistics.TcpAttemptFails++;
584 tcp_set_state(sk,TCP_CLOSE);
585 sk->error_report(sk); /* Wake people up to see the error (see connect in sock.c) */
586 }
587 }
588 else /* Only an error on timeout */
589 sk->err_soft = icmp_err_convert[code].errno;
590 }
591 }
592
593
594 /*
595 * Walk down the receive queue counting readable data until we hit the end or we find a gap
596 * in the received data queue (ie a frame missing that needs sending to us). Not
597 * sorting using two queues as data arrives makes life so much harder.
598 */
599
600 static int tcp_readable(struct sock *sk)
/* */
601 {
602 unsigned long counted;
603 unsigned long amount;
604 struct sk_buff *skb;
605 int sum;
606 unsigned long flags;
607
608 if(sk && sk->debug)
609 printk("tcp_readable: %p - ",sk);
610
611 save_flags(flags);
612 cli();
613 if (sk == NULL || (skb = skb_peek(&sk->receive_queue)) == NULL)
614 {
615 restore_flags(flags);
616 if(sk && sk->debug)
617 printk("empty\n");
618 return(0);
619 }
620
621 counted = sk->copied_seq; /* Where we are at the moment */
622 amount = 0;
623
624 /*
625 * Do until a push or until we are out of data.
626 */
627
628 do
629 {
630 if (before(counted, skb->seq)) /* Found a hole so stops here */
631 break;
632 sum = skb->len - (counted - skb->seq); /* Length - header but start from where we are up to (avoid overlaps) */
633 if (skb->h.th->syn)
634 sum++;
635 if (sum > 0)
636 { /* Add it up, move on */
637 amount += sum;
638 if (skb->h.th->syn)
639 amount--;
640 counted += sum;
641 }
642 /*
643 * Don't count urg data ... but do it in the right place!
644 * Consider: "old_data (ptr is here) URG PUSH data"
645 * The old code would stop at the first push because
646 * it counted the urg (amount==1) and then does amount--
647 * *after* the loop. This means tcp_readable() always
648 * returned zero if any URG PUSH was in the queue, even
649 * though there was normal data available. If we subtract
650 * the urg data right here, we even get it to work for more
651 * than one URG PUSH skb without normal data.
652 * This means that select() finally works now with urg data
653 * in the queue. Note that rlogin was never affected
654 * because it doesn't use select(); it uses two processes
655 * and a blocking read(). And the queue scan in tcp_read()
656 * was correct. Mike <pall@rz.uni-karlsruhe.de>
657 */
658 if (skb->h.th->urg)
659 amount--; /* don't count urg data */
660 if (amount && skb->h.th->psh) break;
661 skb = skb->next;
662 }
663 while(skb != (struct sk_buff *)&sk->receive_queue);
664
665 restore_flags(flags);
666 if(sk->debug)
667 printk("got %lu bytes.\n",amount);
668 return(amount);
669 }
670
671 /*
672 * LISTEN is a special case for select..
673 */
674 static int tcp_listen_select(struct sock *sk, int sel_type, select_table *wait)
/* */
675 {
676 if (sel_type == SEL_IN) {
677 int retval;
678
679 sk->inuse = 1;
680 retval = (tcp_find_established(sk) != NULL);
681 release_sock(sk);
682 if (!retval)
683 select_wait(&master_select_wakeup,wait);
684 return retval;
685 }
686 return 0;
687 }
688
689
690 /*
691 * Wait for a TCP event.
692 *
693 * Note that we don't need to set "sk->inuse", as the upper select layers
694 * take care of normal races (between the test and the event) and we don't
695 * go look at any of the socket buffers directly.
696 */
697 static int tcp_select(struct sock *sk, int sel_type, select_table *wait)
/* */
698 {
699 if (sk->state == TCP_LISTEN)
700 return tcp_listen_select(sk, sel_type, wait);
701
702 switch(sel_type) {
703 case SEL_IN:
704 if (sk->err)
705 return 1;
706 if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
707 break;
708
709 if (sk->shutdown & RCV_SHUTDOWN)
710 return 1;
711
712 if (sk->acked_seq == sk->copied_seq)
713 break;
714
715 if (sk->urg_seq != sk->copied_seq ||
716 sk->acked_seq != sk->copied_seq+1 ||
717 sk->urginline || !sk->urg_data)
718 return 1;
719 break;
720
721 case SEL_OUT:
722 if (sk->err)
723 return 1;
724 if (sk->shutdown & SEND_SHUTDOWN)
725 return 0;
726 if (sk->state == TCP_SYN_SENT || sk->state == TCP_SYN_RECV)
727 break;
728 /*
729 * This is now right thanks to a small fix
730 * by Matt Dillon.
731 */
732
733 if (sock_wspace(sk) < sk->mtu+128+sk->prot->max_header)
734 break;
735 return 1;
736
737 case SEL_EX:
738 if (sk->urg_data)
739 return 1;
740 break;
741 }
742 select_wait(sk->sleep, wait);
743 return 0;
744 }
745
746 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg)
/* */
747 {
748 int err;
749 switch(cmd)
750 {
751
752 case TIOCINQ:
753 #ifdef FIXME /* FIXME: */
754 case FIONREAD:
755 #endif
756 {
757 unsigned long amount;
758
759 if (sk->state == TCP_LISTEN)
760 return(-EINVAL);
761
762 sk->inuse = 1;
763 amount = tcp_readable(sk);
764 release_sock(sk);
765 err=verify_area(VERIFY_WRITE,(void *)arg, sizeof(int));
766 if(err)
767 return err;
768 put_user(amount, (int *)arg);
769 return(0);
770 }
771 case SIOCATMARK:
772 {
773 int answ = sk->urg_data && sk->urg_seq == sk->copied_seq;
774
775 err = verify_area(VERIFY_WRITE,(void *) arg, sizeof(int));
776 if (err)
777 return err;
778 put_user(answ,(int *) arg);
779 return(0);
780 }
781 case TIOCOUTQ:
782 {
783 unsigned long amount;
784
785 if (sk->state == TCP_LISTEN) return(-EINVAL);
786 amount = sock_wspace(sk);
787 err=verify_area(VERIFY_WRITE,(void *)arg, sizeof(int));
788 if(err)
789 return err;
790 put_user(amount, (int *)arg);
791 return(0);
792 }
793 default:
794 return(-EINVAL);
795 }
796 }
797
798
799 /*
800 * This routine computes a TCP checksum.
801 *
802 * Modified January 1995 from a go-faster DOS routine by
803 * Jorge Cwik <jorge@laser.satlink.net>
804 */
805
806 void tcp_send_check(struct tcphdr *th, unsigned long saddr,
/* */
807 unsigned long daddr, int len, struct sock *sk)
808 {
809 th->check = 0;
810 th->check = tcp_check(th, len, saddr, daddr,
811 csum_partial((char *)th,len,0));
812 return;
813 }
814
815
816 /*
817 * This routine builds a generic TCP header.
818 */
819
820 extern __inline int tcp_build_header(struct tcphdr *th, struct sock *sk, int push)
/* */
821 {
822
823 memcpy(th,(void *) &(sk->dummy_th), sizeof(*th));
824 th->seq = htonl(sk->write_seq);
825 th->psh =(push == 0) ? 1 : 0;
826 th->doff = sizeof(*th)/4;
827 th->ack = 1;
828 th->fin = 0;
829 sk->ack_backlog = 0;
830 sk->bytes_rcv = 0;
831 sk->ack_timed = 0;
832 th->ack_seq = htonl(sk->acked_seq);
833 sk->window = tcp_select_window(sk);
834 th->window = htons(sk->window);
835
836 return(sizeof(*th));
837 }
838
839 /*
840 * This routine copies from a user buffer into a socket,
841 * and starts the transmit system.
842 */
843
844 static int tcp_sendmsg(struct sock *sk, struct msghdr *msg,
/* */
845 int len, int nonblock, int flags)
846 {
847 int copied = 0;
848 int copy;
849 int tmp;
850 int seglen;
851 int iovct=0;
852 struct sk_buff *skb;
853 struct sk_buff *send_tmp;
854 struct proto *prot;
855 struct device *dev = NULL;
856 unsigned char *from;
857
858 /*
859 * Do sanity checking for sendmsg/sendto/send
860 */
861
862 if (flags & ~(MSG_OOB|MSG_DONTROUTE))
863 return -EINVAL;
864 if (msg->msg_name)
865 {
866 struct sockaddr_in *addr=(struct sockaddr_in *)msg->msg_name;
867 if(sk->state == TCP_CLOSE)
868 return -ENOTCONN;
869 if (msg->msg_namelen < sizeof(*addr))
870 return -EINVAL;
871 if (addr->sin_family && addr->sin_family != AF_INET)
872 return -EINVAL;
873 if (addr->sin_port != sk->dummy_th.dest)
874 return -EISCONN;
875 if (addr->sin_addr.s_addr != sk->daddr)
876 return -EISCONN;
877 }
878
879 /*
880 * Ok commence sending
881 */
882
883 while(iovct<msg->msg_iovlen)
884 {
885 seglen=msg->msg_iov[iovct].iov_len;
886 from=msg->msg_iov[iovct++].iov_base;
887 sk->inuse=1;
888 prot = sk->prot;
889 while(seglen > 0)
890 {
891 if (sk->err)
892 { /* Stop on an error */
893 release_sock(sk);
894 if (copied)
895 return(copied);
896 return sock_error(sk);
897 }
898
899 /*
900 * First thing we do is make sure that we are established.
901 */
902
903 if (sk->shutdown & SEND_SHUTDOWN)
904 {
905 release_sock(sk);
906 sk->err = EPIPE;
907 if (copied)
908 return(copied);
909 sk->err = 0;
910 return(-EPIPE);
911 }
912
913 /*
914 * Wait for a connection to finish.
915 */
916
917 while(sk->state != TCP_ESTABLISHED && sk->state != TCP_CLOSE_WAIT)
918 {
919 if (sk->err)
920 {
921 release_sock(sk);
922 if (copied)
923 return(copied);
924 return sock_error(sk);
925 }
926
927 if (sk->state != TCP_SYN_SENT && sk->state != TCP_SYN_RECV)
928 {
929 release_sock(sk);
930 if (copied)
931 return(copied);
932
933 if (sk->err)
934 return sock_error(sk);
935
936 if (sk->keepopen)
937 {
938 send_sig(SIGPIPE, current, 0);
939 }
940 return(-EPIPE);
941 }
942
943 if (nonblock || copied)
944 {
945 release_sock(sk);
946 if (copied)
947 return(copied);
948 return(-EAGAIN);
949 }
950
951 release_sock(sk);
952 cli();
953
954 if (sk->state != TCP_ESTABLISHED &&
955 sk->state != TCP_CLOSE_WAIT && sk->err == 0)
956 {
957 interruptible_sleep_on(sk->sleep);
958 if (current->signal & ~current->blocked)
959 {
960 sti();
961 if (copied)
962 return(copied);
963 return(-ERESTARTSYS);
964 }
965 }
966 sk->inuse = 1;
967 sti();
968 }
969
970 /*
971 * The following code can result in copy <= if sk->mss is ever
972 * decreased. It shouldn't be. sk->mss is min(sk->mtu, sk->max_window).
973 * sk->mtu is constant once SYN processing is finished. I.e. we
974 * had better not get here until we've seen his SYN and at least one
975 * valid ack. (The SYN sets sk->mtu and the ack sets sk->max_window.)
976 * But ESTABLISHED should guarantee that. sk->max_window is by definition
977 * non-decreasing. Note that any ioctl to set user_mss must be done
978 * before the exchange of SYN's. If the initial ack from the other
979 * end has a window of 0, max_window and thus mss will both be 0.
980 */
981
982 /*
983 * Now we need to check if we have a half built packet.
984 */
985 #ifndef CONFIG_NO_PATH_MTU_DISCOVERY
986 /*
987 * FIXME: I'm almost sure that this fragment is BUG,
988 * but it works... I do not know why 8) --ANK
989 *
990 * Really, we should rebuild all the queues...
991 * It's difficult. Temprorary hack is to send all
992 * queued segments with allowed fragmentation.
993 */
994 {
995 int new_mss = min(sk->mtu, sk->max_window);
996 if (new_mss < sk->mss)
997 {
998 tcp_send_partial(sk);
999 sk->mss = new_mss;
1000 }
1001 }
1002 #endif
1003
1004 if ((skb = tcp_dequeue_partial(sk)) != NULL)
1005 {
1006 int hdrlen;
1007
1008 /* IP header + TCP header */
1009 hdrlen = ((unsigned long)skb->h.th - (unsigned long)skb->data)
1010 + sizeof(struct tcphdr);
1011
1012 /* Add more stuff to the end of skb->len */
1013 if (!(flags & MSG_OOB))
1014 {
1015 copy = min(sk->mss - (skb->len - hdrlen), seglen);
1016 if (copy <= 0)
1017 {
1018 printk("TCP: **bug**: \"copy\" <= 0\n");
1019 return -EFAULT;
1020 }
1021 memcpy_fromfs(skb_put(skb,copy), from, copy);
1022 from += copy;
1023 copied += copy;
1024 len -= copy;
1025 sk->write_seq += copy;
1026 seglen -= copy;
1027 }
1028 if ((skb->len - hdrlen) >= sk->mss ||
1029 (flags & MSG_OOB) || !sk->packets_out)
1030 tcp_send_skb(sk, skb);
1031 else
1032 tcp_enqueue_partial(skb, sk);
1033 continue;
1034 }
1035
1036 /*
1037 * We also need to worry about the window.
1038 * If window < 1/2 the maximum window we've seen from this
1039 * host, don't use it. This is sender side
1040 * silly window prevention, as specified in RFC1122.
1041 * (Note that this is different than earlier versions of
1042 * SWS prevention, e.g. RFC813.). What we actually do is
1043 * use the whole MSS. Since the results in the right
1044 * edge of the packet being outside the window, it will
1045 * be queued for later rather than sent.
1046 */
1047
1048 copy = sk->window_seq - sk->write_seq;
1049 if (copy <= 0 || copy < (sk->max_window >> 1) || copy > sk->mss)
1050 copy = sk->mss;
1051 if (copy > seglen)
1052 copy = seglen;
1053
1054 /*
1055 * We should really check the window here also.
1056 */
1057
1058 send_tmp = NULL;
1059 if (copy < sk->mss && !(flags & MSG_OOB) && sk->packets_out)
1060 {
1061 /*
1062 * We will release the socket in case we sleep here.
1063 */
1064 release_sock(sk);
1065 /*
1066 * NB: following must be mtu, because mss can be increased.
1067 * mss is always <= mtu
1068 */
1069 skb = sock_wmalloc(sk, sk->mtu + 128 + prot->max_header + 15, 0, GFP_KERNEL);
1070 sk->inuse = 1;
1071 send_tmp = skb;
1072 }
1073 else
1074 {
1075 /*
1076 * We will release the socket in case we sleep here.
1077 */
1078 release_sock(sk);
1079 skb = sock_wmalloc(sk, copy + prot->max_header + 15 , 0, GFP_KERNEL);
1080 sk->inuse = 1;
1081 }
1082
1083 /*
1084 * If we didn't get any memory, we need to sleep.
1085 */
1086
1087 if (skb == NULL)
1088 {
1089 sk->socket->flags |= SO_NOSPACE;
1090 if (nonblock)
1091 {
1092 release_sock(sk);
1093 if (copied)
1094 return(copied);
1095 return(-EAGAIN);
1096 }
1097
1098 release_sock(sk);
1099 cli();
1100 if (sk->wmem_alloc*2 > sk->sndbuf &&
1101 (sk->state == TCP_ESTABLISHED||sk->state == TCP_CLOSE_WAIT)
1102 && sk->err == 0)
1103 {
1104 sk->socket->flags &= ~SO_NOSPACE;
1105 interruptible_sleep_on(sk->sleep);
1106 if (current->signal & ~current->blocked)
1107 {
1108 sti();
1109 if (copied)
1110 return(copied);
1111 return(-ERESTARTSYS);
1112 }
1113 }
1114 sk->inuse = 1;
1115 sti();
1116 continue;
1117 }
1118
1119 skb->sk = sk;
1120 skb->free = 0;
1121 skb->localroute = sk->localroute|(flags&MSG_DONTROUTE);
1122
1123 /*
1124 * FIXME: we need to optimize this.
1125 * Perhaps some hints here would be good.
1126 */
1127
1128 tmp = prot->build_header(skb, sk->saddr, sk->daddr, &dev,
1129 IPPROTO_TCP, sk->opt, skb->truesize,sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
1130 if (tmp < 0 )
1131 {
1132 sock_wfree(sk, skb);
1133 release_sock(sk);
1134 if (copied)
1135 return(copied);
1136 return(tmp);
1137 }
1138 #ifndef CONFIG_NO_PATH_MTU_DISCOVERY
1139 skb->ip_hdr->frag_off |= htons(IP_DF);
1140 #endif
1141 skb->dev = dev;
1142 skb->h.th =(struct tcphdr *)skb_put(skb,sizeof(struct tcphdr));
1143 tmp = tcp_build_header(skb->h.th, sk, seglen-copy);
1144 if (tmp < 0)
1145 {
1146 sock_wfree(sk, skb);
1147 release_sock(sk);
1148 if (copied)
1149 return(copied);
1150 return(tmp);
1151 }
1152
1153 if (flags & MSG_OOB)
1154 {
1155 skb->h.th->urg = 1;
1156 skb->h.th->urg_ptr = ntohs(copy);
1157 }
1158
1159 memcpy_fromfs(skb_put(skb,copy), from, copy);
1160
1161 from += copy;
1162 copied += copy;
1163 len -= copy;
1164 seglen -= copy;
1165 skb->free = 0;
1166 sk->write_seq += copy;
1167
1168 if (send_tmp != NULL)
1169 {
1170 tcp_enqueue_partial(send_tmp, sk);
1171 continue;
1172 }
1173 tcp_send_skb(sk, skb);
1174 }
1175 }
1176 sk->err = 0;
1177
1178 /*
1179 * Nagle's rule. Turn Nagle off with TCP_NODELAY for highly
1180 * interactive fast network servers. It's meant to be on and
1181 * it really improves the throughput though not the echo time
1182 * on my slow slip link - Alan
1183 */
1184
1185 /*
1186 * Avoid possible race on send_tmp - c/o Johannes Stille
1187 */
1188
1189 if(sk->partial && ((!sk->packets_out)
1190 /* If not nagling we can send on the before case too.. */
1191 || (sk->nonagle && before(sk->write_seq , sk->window_seq))
1192 ))
1193 tcp_send_partial(sk);
1194
1195 release_sock(sk);
1196 return(copied);
1197 }
1198
1199 /*
1200 * Send an ack if one is backlogged at this point. Ought to merge
1201 * this with tcp_send_ack().
1202 * This is called for delayed acks also.
1203 */
1204
1205 void tcp_read_wakeup(struct sock *sk)
/* */
1206 {
1207 int tmp;
1208 struct device *dev = NULL;
1209 struct tcphdr *t1;
1210 struct sk_buff *buff;
1211
1212 if (!sk->ack_backlog)
1213 return;
1214
1215 /*
1216 * If we're closed, don't send an ack, or we'll get a RST
1217 * from the closed destination.
1218 */
1219 if ((sk->state == TCP_CLOSE) || (sk->state == TCP_TIME_WAIT))
1220 return;
1221
1222 /*
1223 * FIXME: we need to put code here to prevent this routine from
1224 * being called. Being called once in a while is ok, so only check
1225 * if this is the second time in a row.
1226 */
1227
1228 /*
1229 * We need to grab some memory, and put together an ack,
1230 * and then put it into the queue to be sent.
1231 */
1232
1233 buff = sock_wmalloc(sk,MAX_ACK_SIZE,1, GFP_ATOMIC);
1234 if (buff == NULL)
1235 {
1236 /* Try again real soon. */
1237 tcp_reset_xmit_timer(sk, TIME_WRITE, HZ);
1238 return;
1239 }
1240
1241 buff->sk = sk;
1242 buff->localroute = sk->localroute;
1243
1244 /*
1245 * Put in the IP header and routing stuff.
1246 */
1247
1248 tmp = sk->prot->build_header(buff, sk->saddr, sk->daddr, &dev,
1249 IPPROTO_TCP, sk->opt, MAX_ACK_SIZE,sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
1250 if (tmp < 0)
1251 {
1252 buff->free = 1;
1253 sock_wfree(sk, buff);
1254 return;
1255 }
1256
1257 t1 =(struct tcphdr *)skb_put(buff,sizeof(struct tcphdr));
1258
1259 memcpy(t1,(void *) &sk->dummy_th, sizeof(*t1));
1260 t1->seq = htonl(sk->sent_seq);
1261 t1->ack = 1;
1262 t1->res1 = 0;
1263 t1->res2 = 0;
1264 t1->rst = 0;
1265 t1->urg = 0;
1266 t1->syn = 0;
1267 t1->psh = 0;
1268
1269
1270 sk->ack_backlog = 0;
1271 sk->bytes_rcv = 0;
1272
1273 sk->window = tcp_select_window(sk);
1274 t1->window = htons(sk->window);
1275 t1->ack_seq = htonl(sk->acked_seq);
1276 t1->doff = sizeof(*t1)/4;
1277 tcp_send_check(t1, sk->saddr, sk->daddr, sizeof(*t1), sk);
1278 sk->prot->queue_xmit(sk, dev, buff, 1);
1279 tcp_statistics.TcpOutSegs++;
1280 }
1281
1282
1283 /*
1284 * FIXME:
1285 * This routine frees used buffers.
1286 * It should consider sending an ACK to let the
1287 * other end know we now have a bigger window.
1288 */
1289
1290 static void cleanup_rbuf(struct sock *sk)
/* */
1291 {
1292 unsigned long flags;
1293 struct sk_buff *skb;
1294 unsigned long rspace;
1295
1296 save_flags(flags);
1297 cli();
1298
1299 /*
1300 * See if we have anything to free up?
1301 */
1302
1303 skb = skb_peek(&sk->receive_queue);
1304 if (!skb || !skb->used || skb->users) {
1305 restore_flags(flags);
1306 return;
1307 }
1308
1309 /*
1310 * We have to loop through all the buffer headers,
1311 * and try to free up all the space we can.
1312 */
1313
1314 do {
1315 skb_unlink(skb);
1316 skb->sk = sk;
1317 kfree_skb(skb, FREE_READ);
1318 skb = skb_peek(&sk->receive_queue);
1319 } while (skb && skb->used && !skb->users);
1320 restore_flags(flags);
1321
1322 /*
1323 * FIXME:
1324 * At this point we should send an ack if the difference
1325 * in the window, and the amount of space is bigger than
1326 * TCP_WINDOW_DIFF.
1327 */
1328
1329 rspace=sock_rspace(sk);
1330 if(sk->debug)
1331 printk("sk->rspace = %lu\n", rspace);
1332 /*
1333 * This area has caused the most trouble. The current strategy
1334 * is to simply do nothing if the other end has room to send at
1335 * least 3 full packets, because the ack from those will auto-
1336 * matically update the window. If the other end doesn't think
1337 * we have much space left, but we have room for at least 1 more
1338 * complete packet than it thinks we do, we will send an ack
1339 * immediately. Otherwise we will wait up to .5 seconds in case
1340 * the user reads some more.
1341 */
1342 sk->ack_backlog++;
1343
1344 /*
1345 * It's unclear whether to use sk->mtu or sk->mss here. They differ only
1346 * if the other end is offering a window smaller than the agreed on MSS
1347 * (called sk->mtu here). In theory there's no connection between send
1348 * and receive, and so no reason to think that they're going to send
1349 * small packets. For the moment I'm using the hack of reducing the mss
1350 * only on the send side, so I'm putting mtu here.
1351 */
1352
1353 if (rspace > (sk->window - sk->bytes_rcv + sk->mtu))
1354 {
1355 /* Send an ack right now. */
1356 tcp_read_wakeup(sk);
1357 }
1358 else
1359 {
1360 /* Force it to send an ack soon. */
1361 int was_active = del_timer(&sk->retransmit_timer);
1362 if (!was_active || jiffies+TCP_ACK_TIME < sk->timer.expires)
1363 {
1364 tcp_reset_xmit_timer(sk, TIME_WRITE, TCP_ACK_TIME);
1365 }
1366 else
1367 add_timer(&sk->retransmit_timer);
1368 }
1369 }
1370
1371
1372 /*
1373 * Handle reading urgent data. BSD has very simple semantics for
1374 * this, no blocking and very strange errors 8)
1375 */
1376
1377 static int tcp_recv_urg(struct sock * sk, int nonblock,
/* */
1378 struct msghdr *msg, int len, int flags, int *addr_len)
1379 {
1380 /*
1381 * No URG data to read
1382 */
1383 if (sk->urginline || !sk->urg_data || sk->urg_data == URG_READ)
1384 return -EINVAL; /* Yes this is right ! */
1385
1386 if (sk->err)
1387 return sock_error(sk);
1388
1389 if (sk->state == TCP_CLOSE || sk->done)
1390 {
1391 if (!sk->done)
1392 {
1393 sk->done = 1;
1394 return 0;
1395 }
1396 return -ENOTCONN;
1397 }
1398
1399 if (sk->shutdown & RCV_SHUTDOWN)
1400 {
1401 sk->done = 1;
1402 return 0;
1403 }
1404 sk->inuse = 1;
1405 if (sk->urg_data & URG_VALID)
1406 {
1407 char c = sk->urg_data;
1408 if (!(flags & MSG_PEEK))
1409 sk->urg_data = URG_READ;
1410 memcpy_toiovec(msg->msg_iov, &c, 1);
1411 if(msg->msg_name)
1412 {
1413 struct sockaddr_in *sin=(struct sockaddr_in *)msg->msg_name;
1414 sin->sin_family=AF_INET;
1415 sin->sin_addr.s_addr=sk->daddr;
1416 sin->sin_port=sk->dummy_th.dest;
1417 }
1418 if(addr_len)
1419 *addr_len=sizeof(struct sockaddr_in);
1420 release_sock(sk);
1421 return 1;
1422 }
1423 release_sock(sk);
1424
1425 /*
1426 * Fixed the recv(..., MSG_OOB) behaviour. BSD docs and
1427 * the available implementations agree in this case:
1428 * this call should never block, independent of the
1429 * blocking state of the socket.
1430 * Mike <pall@rz.uni-karlsruhe.de>
1431 */
1432 return -EAGAIN;
1433 }
1434
1435
1436 /*
1437 * This routine copies from a sock struct into the user buffer.
1438 */
1439
1440 static int tcp_recvmsg(struct sock *sk, struct msghdr *msg,
/* */
1441 int len, int nonblock, int flags, int *addr_len)
1442 {
1443 struct wait_queue wait = { current, NULL };
1444 int copied = 0;
1445 u32 peek_seq;
1446 volatile u32 *seq; /* So gcc doesn't overoptimise */
1447 unsigned long used;
1448
1449 /*
1450 * This error should be checked.
1451 */
1452
1453 if (sk->state == TCP_LISTEN)
1454 return -ENOTCONN;
1455
1456 /*
1457 * Urgent data needs to be handled specially.
1458 */
1459
1460 if (flags & MSG_OOB)
1461 return tcp_recv_urg(sk, nonblock, msg, len, flags, addr_len);
1462
1463 /*
1464 * Copying sequence to update. This is volatile to handle
1465 * the multi-reader case neatly (memcpy_to/fromfs might be
1466 * inline and thus not flush cached variables otherwise).
1467 */
1468
1469 peek_seq = sk->copied_seq;
1470 seq = &sk->copied_seq;
1471 if (flags & MSG_PEEK)
1472 seq = &peek_seq;
1473
1474 add_wait_queue(sk->sleep, &wait);
1475 sk->inuse = 1;
1476 while (len > 0)
1477 {
1478 struct sk_buff * skb;
1479 u32 offset;
1480
1481 /*
1482 * Are we at urgent data? Stop if we have read anything.
1483 */
1484
1485 if (copied && sk->urg_data && sk->urg_seq == *seq)
1486 break;
1487
1488 /*
1489 * Next get a buffer.
1490 */
1491
1492 current->state = TASK_INTERRUPTIBLE;
1493
1494 skb = skb_peek(&sk->receive_queue);
1495 do
1496 {
1497 if (!skb)
1498 break;
1499 if (before(*seq, skb->seq))
1500 break;
1501 offset = *seq - skb->seq;
1502 if (skb->h.th->syn)
1503 offset--;
1504 if (offset < skb->len)
1505 goto found_ok_skb;
1506 if (skb->h.th->fin)
1507 goto found_fin_ok;
1508 if (!(flags & MSG_PEEK))
1509 skb->used = 1;
1510 skb = skb->next;
1511 }
1512 while (skb != (struct sk_buff *)&sk->receive_queue);
1513
1514 if (copied)
1515 break;
1516
1517 if (sk->err)
1518 {
1519 copied = sock_error(sk);
1520 break;
1521 }
1522
1523 if (sk->state == TCP_CLOSE)
1524 {
1525 if (!sk->done)
1526 {
1527 sk->done = 1;
1528 break;
1529 }
1530 copied = -ENOTCONN;
1531 break;
1532 }
1533
1534 if (sk->shutdown & RCV_SHUTDOWN)
1535 {
1536 sk->done = 1;
1537 break;
1538 }
1539
1540 if (nonblock)
1541 {
1542 copied = -EAGAIN;
1543 break;
1544 }
1545
1546 cleanup_rbuf(sk);
1547 release_sock(sk);
1548 sk->socket->flags |= SO_WAITDATA;
1549 schedule();
1550 sk->socket->flags &= ~SO_WAITDATA;
1551 sk->inuse = 1;
1552
1553 if (current->signal & ~current->blocked)
1554 {
1555 copied = -ERESTARTSYS;
1556 break;
1557 }
1558 continue;
1559
1560 found_ok_skb:
1561 /*
1562 * Lock the buffer. We can be fairly relaxed as
1563 * an interrupt will never steal a buffer we are
1564 * using unless I've missed something serious in
1565 * tcp_data.
1566 */
1567
1568 skb->users++;
1569
1570 /*
1571 * Ok so how much can we use ?
1572 */
1573
1574 used = skb->len - offset;
1575 if (len < used)
1576 used = len;
1577 /*
1578 * Do we have urgent data here?
1579 */
1580
1581 if (sk->urg_data)
1582 {
1583 u32 urg_offset = sk->urg_seq - *seq;
1584 if (urg_offset < used)
1585 {
1586 if (!urg_offset)
1587 {
1588 if (!sk->urginline)
1589 {
1590 ++*seq;
1591 offset++;
1592 used--;
1593 }
1594 }
1595 else
1596 used = urg_offset;
1597 }
1598 }
1599
1600 /*
1601 * Copy it - We _MUST_ update *seq first so that we
1602 * don't ever double read when we have dual readers
1603 */
1604
1605 *seq += used;
1606
1607 /*
1608 * This memcpy_tofs can sleep. If it sleeps and we
1609 * do a second read it relies on the skb->users to avoid
1610 * a crash when cleanup_rbuf() gets called.
1611 */
1612
1613 memcpy_toiovec(msg->msg_iov,((unsigned char *)skb->h.th) +
1614 skb->h.th->doff*4 + offset, used);
1615 copied += used;
1616 len -= used;
1617
1618 /*
1619 * We now will not sleep again until we are finished
1620 * with skb. Sorry if you are doing the SMP port
1621 * but you'll just have to fix it neatly ;)
1622 */
1623
1624 skb->users --;
1625
1626 if (after(sk->copied_seq,sk->urg_seq))
1627 sk->urg_data = 0;
1628 if (used + offset < skb->len)
1629 continue;
1630
1631 /*
1632 * Process the FIN.
1633 */
1634
1635 if (skb->h.th->fin)
1636 goto found_fin_ok;
1637 if (flags & MSG_PEEK)
1638 continue;
1639 skb->used = 1;
1640 continue;
1641
1642 found_fin_ok:
1643 ++*seq;
1644 if (flags & MSG_PEEK)
1645 break;
1646
1647 /*
1648 * All is done
1649 */
1650
1651 skb->used = 1;
1652 sk->shutdown |= RCV_SHUTDOWN;
1653 break;
1654
1655 }
1656
1657 if(copied>0 && msg->msg_name)
1658 {
1659 struct sockaddr_in *sin=(struct sockaddr_in *)msg->msg_name;
1660 sin->sin_family=AF_INET;
1661 sin->sin_addr.s_addr=sk->daddr;
1662 sin->sin_port=sk->dummy_th.dest;
1663 }
1664 if(addr_len)
1665 *addr_len=sizeof(struct sockaddr_in);
1666
1667 remove_wait_queue(sk->sleep, &wait);
1668 current->state = TASK_RUNNING;
1669
1670 /* Clean up data we have read: This will do ACK frames */
1671 cleanup_rbuf(sk);
1672 release_sock(sk);
1673 return copied;
1674 }
1675
1676
1677
1678 /*
1679 * State processing on a close. This implements the state shift for
1680 * sending our FIN frame. Note that we only send a FIN for some
1681 * states. A shutdown() may have already sent the FIN, or we may be
1682 * closed.
1683 */
1684
1685 static int tcp_close_state(struct sock *sk, int dead)
/* */
1686 {
1687 int ns=TCP_CLOSE;
1688 int send_fin=0;
1689 switch(sk->state)
1690 {
1691 case TCP_SYN_SENT: /* No SYN back, no FIN needed */
1692 break;
1693 case TCP_SYN_RECV:
1694 case TCP_ESTABLISHED: /* Closedown begin */
1695 ns=TCP_FIN_WAIT1;
1696 send_fin=1;
1697 break;
1698 case TCP_FIN_WAIT1: /* Already closing, or FIN sent: no change */
1699 case TCP_FIN_WAIT2:
1700 case TCP_CLOSING:
1701 ns=sk->state;
1702 break;
1703 case TCP_CLOSE:
1704 case TCP_LISTEN:
1705 break;
1706 case TCP_CLOSE_WAIT: /* They have FIN'd us. We send our FIN and
1707 wait only for the ACK */
1708 ns=TCP_LAST_ACK;
1709 send_fin=1;
1710 }
1711
1712 tcp_set_state(sk,ns);
1713
1714 /*
1715 * This is a (useful) BSD violating of the RFC. There is a
1716 * problem with TCP as specified in that the other end could
1717 * keep a socket open forever with no application left this end.
1718 * We use a 3 minute timeout (about the same as BSD) then kill
1719 * our end. If they send after that then tough - BUT: long enough
1720 * that we won't make the old 4*rto = almost no time - whoops
1721 * reset mistake.
1722 */
1723 if(dead && ns==TCP_FIN_WAIT2)
1724 {
1725 int timer_active=del_timer(&sk->timer);
1726 if(timer_active)
1727 add_timer(&sk->timer);
1728 else
1729 tcp_reset_msl_timer(sk, TIME_CLOSE, TCP_FIN_TIMEOUT);
1730 }
1731
1732 return send_fin;
1733 }
1734
1735 /*
1736 * Shutdown the sending side of a connection. Much like close except
1737 * that we don't receive shut down or set sk->dead=1.
1738 */
1739
1740 void tcp_shutdown(struct sock *sk, int how)
/* */
1741 {
1742 /*
1743 * We need to grab some memory, and put together a FIN,
1744 * and then put it into the queue to be sent.
1745 * Tim MacKenzie(tym@dibbler.cs.monash.edu.au) 4 Dec '92.
1746 */
1747
1748 if (!(how & SEND_SHUTDOWN))
1749 return;
1750
1751 /*
1752 * If we've already sent a FIN, or it's a closed state
1753 */
1754
1755 if (sk->state == TCP_FIN_WAIT1 ||
1756 sk->state == TCP_FIN_WAIT2 ||
1757 sk->state == TCP_CLOSING ||
1758 sk->state == TCP_LAST_ACK ||
1759 sk->state == TCP_TIME_WAIT ||
1760 sk->state == TCP_CLOSE ||
1761 sk->state == TCP_LISTEN
1762 )
1763 {
1764 return;
1765 }
1766 sk->inuse = 1;
1767
1768 /*
1769 * flag that the sender has shutdown
1770 */
1771
1772 sk->shutdown |= SEND_SHUTDOWN;
1773
1774 /*
1775 * Clear out any half completed packets.
1776 */
1777
1778 if (sk->partial)
1779 tcp_send_partial(sk);
1780
1781 /*
1782 * FIN if needed
1783 */
1784
1785 if(tcp_close_state(sk,0))
1786 tcp_send_fin(sk);
1787
1788 release_sock(sk);
1789 }
1790
1791 static void tcp_close(struct sock *sk, int timeout)
/* */
1792 {
1793 /*
1794 * We need to grab some memory, and put together a FIN,
1795 * and then put it into the queue to be sent.
1796 */
1797
1798 sk->inuse = 1;
1799
1800 tcp_cache_zap();
1801 if(sk->state == TCP_LISTEN)
1802 {
1803 /* Special case */
1804 tcp_set_state(sk, TCP_CLOSE);
1805 tcp_close_pending(sk);
1806 release_sock(sk);
1807 return;
1808 }
1809
1810 sk->keepopen = 1;
1811 sk->shutdown = SHUTDOWN_MASK;
1812
1813 if (!sk->dead)
1814 sk->state_change(sk);
1815
1816 if (timeout == 0)
1817 {
1818 struct sk_buff *skb;
1819
1820 /*
1821 * We need to flush the recv. buffs. We do this only on the
1822 * descriptor close, not protocol-sourced closes, because the
1823 * reader process may not have drained the data yet!
1824 */
1825
1826 while((skb=skb_dequeue(&sk->receive_queue))!=NULL)
1827 kfree_skb(skb, FREE_READ);
1828 /*
1829 * Get rid off any half-completed packets.
1830 */
1831
1832 if (sk->partial)
1833 tcp_send_partial(sk);
1834 }
1835
1836
1837 /*
1838 * Timeout is not the same thing - however the code likes
1839 * to send both the same way (sigh).
1840 */
1841
1842 if(timeout)
1843 {
1844 tcp_set_state(sk, TCP_CLOSE); /* Dead */
1845 }
1846 else
1847 {
1848 if(tcp_close_state(sk,1)==1)
1849 {
1850 tcp_send_fin(sk);
1851 }
1852 }
1853 release_sock(sk);
1854 }
1855
1856
1857 /*
1858 * This will accept the next outstanding connection.
1859 */
1860
1861 static struct sock *tcp_accept(struct sock *sk, int flags)
/* */
1862 {
1863 struct sock *newsk;
1864 struct sk_buff *skb;
1865
1866 /*
1867 * We need to make sure that this socket is listening,
1868 * and that it has something pending.
1869 */
1870
1871 if (sk->state != TCP_LISTEN)
1872 {
1873 sk->err = EINVAL;
1874 return(NULL);
1875 }
1876
1877 /* Avoid the race. */
1878 cli();
1879 sk->inuse = 1;
1880
1881 while((skb = tcp_dequeue_established(sk)) == NULL)
1882 {
1883 if (flags & O_NONBLOCK)
1884 {
1885 sti();
1886 release_sock(sk);
1887 sk->err = EAGAIN;
1888 return(NULL);
1889 }
1890
1891 release_sock(sk);
1892 interruptible_sleep_on(sk->sleep);
1893 if (current->signal & ~current->blocked)
1894 {
1895 sti();
1896 sk->err = ERESTARTSYS;
1897 return(NULL);
1898 }
1899 sk->inuse = 1;
1900 }
1901 sti();
1902
1903 /*
1904 * Now all we need to do is return skb->sk.
1905 */
1906
1907 newsk = skb->sk;
1908
1909 kfree_skb(skb, FREE_READ);
1910 sk->ack_backlog--;
1911 release_sock(sk);
1912 return(newsk);
1913 }
1914
1915 /*
1916 * This will initiate an outgoing connection.
1917 */
1918
1919 static int tcp_connect(struct sock *sk, struct sockaddr_in *usin, int addr_len)
/* */
1920 {
1921 struct sk_buff *buff;
1922 struct device *dev=NULL;
1923 unsigned char *ptr;
1924 int tmp;
1925 int atype;
1926 struct tcphdr *t1;
1927 struct rtable *rt;
1928
1929 if (sk->state != TCP_CLOSE)
1930 return(-EISCONN);
1931
1932 /*
1933 * Don't allow a double connect.
1934 */
1935
1936 if(sk->daddr)
1937 return -EINVAL;
1938
1939 if (addr_len < 8)
1940 return(-EINVAL);
1941
1942 if (usin->sin_family && usin->sin_family != AF_INET)
1943 return(-EAFNOSUPPORT);
1944
1945 /*
1946 * connect() to INADDR_ANY means loopback (BSD'ism).
1947 */
1948
1949 if(usin->sin_addr.s_addr==INADDR_ANY)
1950 usin->sin_addr.s_addr=ip_my_addr();
1951
1952 /*
1953 * Don't want a TCP connection going to a broadcast address
1954 */
1955
1956 if ((atype=ip_chk_addr(usin->sin_addr.s_addr)) == IS_BROADCAST || atype==IS_MULTICAST)
1957 return -ENETUNREACH;
1958
1959 sk->inuse = 1;
1960 sk->daddr = usin->sin_addr.s_addr;
1961 sk->write_seq = tcp_init_seq();
1962 sk->window_seq = sk->write_seq;
1963 sk->rcv_ack_seq = sk->write_seq -1;
1964 sk->err = 0;
1965 sk->dummy_th.dest = usin->sin_port;
1966 release_sock(sk);
1967
1968 buff = sock_wmalloc(sk,MAX_SYN_SIZE,0, GFP_KERNEL);
1969 if (buff == NULL)
1970 {
1971 return(-ENOMEM);
1972 }
1973 sk->inuse = 1;
1974 buff->sk = sk;
1975 buff->free = 0;
1976 buff->localroute = sk->localroute;
1977
1978
1979 /*
1980 * Put in the IP header and routing stuff.
1981 */
1982
1983 tmp = sk->prot->build_header(buff, sk->saddr, sk->daddr, &dev,
1984 IPPROTO_TCP, NULL, MAX_SYN_SIZE,sk->ip_tos,sk->ip_ttl,&sk->ip_route_cache);
1985 if (tmp < 0)
1986 {
1987 sock_wfree(sk, buff);
1988 release_sock(sk);
1989 return(-ENETUNREACH);
1990 }
1991 if ((rt = sk->ip_route_cache) != NULL && !sk->saddr)
1992 sk->saddr = rt->rt_src;
1993 sk->rcv_saddr = sk->saddr;
1994
1995 t1 = (struct tcphdr *) skb_put(buff,sizeof(struct tcphdr));
1996
1997 memcpy(t1,(void *)&(sk->dummy_th), sizeof(*t1));
1998 buff->seq = sk->write_seq++;
1999 t1->seq = htonl(buff->seq);
2000 sk->sent_seq = sk->write_seq;
2001 buff->end_seq = sk->write_seq;
2002 t1->ack = 0;
2003 t1->window = 2;
2004 t1->res1=0;
2005 t1->res2=0;
2006 t1->rst = 0;
2007 t1->urg = 0;
2008 t1->psh = 0;
2009 t1->syn = 1;
2010 t1->urg_ptr = 0;
2011 t1->doff = 6;
2012 /* use 512 or whatever user asked for */
2013
2014 if(rt!=NULL && (rt->rt_flags&RTF_WINDOW))
2015 sk->window_clamp=rt->rt_window;
2016 else
2017 sk->window_clamp=0;
2018
2019 if (sk->user_mss)
2020 sk->mtu = sk->user_mss;
2021 else if (rt)
2022 sk->mtu = rt->rt_mtu - sizeof(struct iphdr) - sizeof(struct tcphdr);
2023 else
2024 sk->mtu = 576 - sizeof(struct iphdr) - sizeof(struct tcphdr);
2025
2026 /*
2027 * but not bigger than device MTU
2028 */
2029
2030 if(sk->mtu <32)
2031 sk->mtu = 32; /* Sanity limit */
2032
2033 sk->mtu = min(sk->mtu, dev->mtu - sizeof(struct iphdr) - sizeof(struct tcphdr));
2034
2035 #ifdef CONFIG_SKIP
2036
2037 /*
2038 * SKIP devices set their MTU to 65535. This is so they can take packets
2039 * unfragmented to security process then fragment. They could lie to the
2040 * TCP layer about a suitable MTU, but its easier to let skip sort it out
2041 * simply because the final package we want unfragmented is going to be
2042 *
2043 * [IPHDR][IPSP][Security data][Modified TCP data][Security data]
2044 */
2045
2046 if(skip_pick_mtu!=NULL) /* If SKIP is loaded.. */
2047 sk->mtu=skip_pick_mtu(sk->mtu,dev);
2048 #endif
2049
2050 /*
2051 * Put in the TCP options to say MTU.
2052 */
2053
2054 ptr = skb_put(buff,4);
2055 ptr[0] = 2;
2056 ptr[1] = 4;
2057 ptr[2] = (sk->mtu) >> 8;
2058 ptr[3] = (sk->mtu) & 0xff;
2059 tcp_send_check(t1, sk->saddr, sk->daddr,
2060 sizeof(struct tcphdr) + 4, sk);
2061
2062 /*
2063 * This must go first otherwise a really quick response will get reset.
2064 */
2065
2066 tcp_cache_zap();
2067 tcp_set_state(sk,TCP_SYN_SENT);
2068 if(rt&&rt->rt_flags&RTF_IRTT)
2069 sk->rto = rt->rt_irtt;
2070 else
2071 sk->rto = TCP_TIMEOUT_INIT;
2072 sk->retransmit_timer.function=&tcp_retransmit_timer;
2073 sk->retransmit_timer.data = (unsigned long)sk;
2074 tcp_reset_xmit_timer(sk, TIME_WRITE, sk->rto); /* Timer for repeating the SYN until an answer */
2075 sk->retransmits = 0; /* Now works the right way instead of a hacked
2076 initial setting */
2077
2078 sk->prot->queue_xmit(sk, dev, buff, 0);
2079 tcp_reset_xmit_timer(sk, TIME_WRITE, sk->rto);
2080 tcp_statistics.TcpActiveOpens++;
2081 tcp_statistics.TcpOutSegs++;
2082
2083 release_sock(sk);
2084 return(0);
2085 }
2086
2087 /*
2088 * Socket option code for TCP.
2089 */
2090
2091 int tcp_setsockopt(struct sock *sk, int level, int optname, char *optval, int optlen)
/* */
2092 {
2093 int val,err;
2094
2095 if(level!=SOL_TCP)
2096 return ip_setsockopt(sk,level,optname,optval,optlen);
2097
2098 if (optval == NULL)
2099 return(-EINVAL);
2100
2101 err=verify_area(VERIFY_READ, optval, sizeof(int));
2102 if(err)
2103 return err;
2104
2105 val = get_user((int *)optval);
2106
2107 switch(optname)
2108 {
2109 case TCP_MAXSEG:
2110 /*
2111 * values greater than interface MTU won't take effect. however at
2112 * the point when this call is done we typically don't yet know
2113 * which interface is going to be used
2114 */
2115 if(val<1||val>MAX_WINDOW)
2116 return -EINVAL;
2117 sk->user_mss=val;
2118 return 0;
2119 case TCP_NODELAY:
2120 sk->nonagle=(val==0)?0:1;
2121 return 0;
2122 default:
2123 return(-ENOPROTOOPT);
2124 }
2125 }
2126
2127 int tcp_getsockopt(struct sock *sk, int level, int optname, char *optval, int *optlen)
/* ![[last]](../icons/n_last.png)
*/
2128 {
2129 int val,err;
2130
2131 if(level!=SOL_TCP)
2132 return ip_getsockopt(sk,level,optname,optval,optlen);
2133
2134 switch(optname)
2135 {
2136 case TCP_MAXSEG:
2137 val=sk->user_mss;
2138 break;
2139 case TCP_NODELAY:
2140 val=sk->nonagle;
2141 break;
2142 default:
2143 return(-ENOPROTOOPT);
2144 }
2145 err=verify_area(VERIFY_WRITE, optlen, sizeof(int));
2146 if(err)
2147 return err;
2148 put_user(sizeof(int),(int *) optlen);
2149
2150 err=verify_area(VERIFY_WRITE, optval, sizeof(int));
2151 if(err)
2152 return err;
2153 put_user(val,(int *)optval);
2154
2155 return(0);
2156 }
2157
2158
2159 struct proto tcp_prot = {
2160 tcp_close,
2161 ip_build_header,
2162 tcp_connect,
2163 tcp_accept,
2164 ip_queue_xmit,
2165 tcp_retransmit,
2166 tcp_write_wakeup,
2167 tcp_read_wakeup,
2168 tcp_rcv,
2169 tcp_select,
2170 tcp_ioctl,
2171 NULL,
2172 tcp_shutdown,
2173 tcp_setsockopt,
2174 tcp_getsockopt,
2175 tcp_sendmsg,
2176 tcp_recvmsg,
2177 NULL, /* No special bind() */
2178 128,
2179 0,
2180 "TCP",
2181 0, 0,
2182 {NULL,}
2183 };
![[bottom]](../icons/n_bottom.png){kind=icon}
*/