Configurable TCP_NODELAY option on network connections
[u/mdw/putty] / winnet.c
1 /*
2 * Windows networking abstraction.
3 *
4 * Due to this clean abstraction it was possible
5 * to easily implement IPv6 support :)
6 *
7 * IPv6 patch 1 (27 October 2000) Jeroen Massar <jeroen@unfix.org>
8 * - Preliminary hacked IPv6 support.
9 * - Connecting to IPv6 address (eg fec0:4242:4242:100:2d0:b7ff:fe8f:5d42) works.
10 * - Connecting to IPv6 hostname (eg heaven.ipv6.unfix.org) works.
11 * - Compiles as either IPv4 or IPv6.
12 *
13 * IPv6 patch 2 (29 October 2000) Jeroen Massar <jeroen@unfix.org>
14 * - When compiled as IPv6 it also allows connecting to IPv4 hosts.
15 * - Added some more documentation.
16 *
17 * IPv6 patch 3 (18 November 2000) Jeroen Massar <jeroen@unfix.org>
18 * - It now supports dynamically loading the IPv6 resolver dll's.
19 * This way we should be able to distribute one (1) binary
20 * which supports both IPv4 and IPv6.
21 * - getaddrinfo() and getnameinfo() are loaded dynamicaly if possible.
22 * - in6addr_any is defined in this file so we don't need to link to wship6.lib
23 * - The patch is now more unified so that we can still
24 * remove all IPv6 support by undef'ing IPV6.
25 * But where it fallsback to IPv4 it uses the IPv4 code which is already in place...
26 * - Canonical name resolving works.
27 *
28 * IPv6 patch 4 (07 January 2001) Jeroen Massar <jeroen@unfix.org>
29 * - patch against CVS of today, will be submitted to the bugs list
30 * as a 'cvs diff -u' on Simon's request...
31 *
32 */
33
34 /*
35 * Define IPV6 to have IPv6 on-the-fly-loading support.
36 * This means that one doesn't have to have an IPv6 stack to use it.
37 * But if an IPv6 stack is found it is used with a fallback to IPv4.
38 */
39 /* #define IPV6 1 */
40
41 #ifdef IPV6
42 #include <winsock2.h>
43 #include <ws2tcpip.h>
44 #include <tpipv6.h>
45 #else
46 #include <winsock.h>
47 #endif
48 #include <windows.h>
49 #include <stdio.h>
50 #include <stdlib.h>
51 #include <assert.h>
52
53 #define DEFINE_PLUG_METHOD_MACROS
54 #include "putty.h"
55 #include "network.h"
56 #include "tree234.h"
57
58 struct Socket_tag {
59 struct socket_function_table *fn;
60 /* the above variable absolutely *must* be the first in this structure */
61 char *error;
62 SOCKET s;
63 Plug plug;
64 void *private_ptr;
65 bufchain output_data;
66 int connected;
67 int writable;
68 int frozen; /* this causes readability notifications to be ignored */
69 int frozen_readable; /* this means we missed at least one readability
70 * notification while we were frozen */
71 int localhost_only; /* for listening sockets */
72 char oobdata[1];
73 int sending_oob;
74 int oobinline;
75 int pending_error; /* in case send() returns error */
76 };
77
78 /*
79 * We used to typedef struct Socket_tag *Socket.
80 *
81 * Since we have made the networking abstraction slightly more
82 * abstract, Socket no longer means a tcp socket (it could mean
83 * an ssl socket). So now we must use Actual_Socket when we know
84 * we are talking about a tcp socket.
85 */
86 typedef struct Socket_tag *Actual_Socket;
87
88 struct SockAddr_tag {
89 char *error;
90 /* address family this belongs to, AF_INET for IPv4, AF_INET6 for IPv6. */
91 int family;
92 unsigned long address; /* Address IPv4 style. */
93 #ifdef IPV6
94 struct addrinfo *ai; /* Address IPv6 style. */
95 #endif
96 };
97
98 static tree234 *sktree;
99
100 static int cmpfortree(void *av, void *bv)
101 {
102 Actual_Socket a = (Actual_Socket) av, b = (Actual_Socket) bv;
103 unsigned long as = (unsigned long) a->s, bs = (unsigned long) b->s;
104 if (as < bs)
105 return -1;
106 if (as > bs)
107 return +1;
108 return 0;
109 }
110
111 static int cmpforsearch(void *av, void *bv)
112 {
113 Actual_Socket b = (Actual_Socket) bv;
114 unsigned long as = (unsigned long) av, bs = (unsigned long) b->s;
115 if (as < bs)
116 return -1;
117 if (as > bs)
118 return +1;
119 return 0;
120 }
121
122 void sk_init(void)
123 {
124 sktree = newtree234(cmpfortree);
125 }
126
127 char *winsock_error_string(int error)
128 {
129 switch (error) {
130 case WSAEACCES:
131 return "Network error: Permission denied";
132 case WSAEADDRINUSE:
133 return "Network error: Address already in use";
134 case WSAEADDRNOTAVAIL:
135 return "Network error: Cannot assign requested address";
136 case WSAEAFNOSUPPORT:
137 return
138 "Network error: Address family not supported by protocol family";
139 case WSAEALREADY:
140 return "Network error: Operation already in progress";
141 case WSAECONNABORTED:
142 return "Network error: Software caused connection abort";
143 case WSAECONNREFUSED:
144 return "Network error: Connection refused";
145 case WSAECONNRESET:
146 return "Network error: Connection reset by peer";
147 case WSAEDESTADDRREQ:
148 return "Network error: Destination address required";
149 case WSAEFAULT:
150 return "Network error: Bad address";
151 case WSAEHOSTDOWN:
152 return "Network error: Host is down";
153 case WSAEHOSTUNREACH:
154 return "Network error: No route to host";
155 case WSAEINPROGRESS:
156 return "Network error: Operation now in progress";
157 case WSAEINTR:
158 return "Network error: Interrupted function call";
159 case WSAEINVAL:
160 return "Network error: Invalid argument";
161 case WSAEISCONN:
162 return "Network error: Socket is already connected";
163 case WSAEMFILE:
164 return "Network error: Too many open files";
165 case WSAEMSGSIZE:
166 return "Network error: Message too long";
167 case WSAENETDOWN:
168 return "Network error: Network is down";
169 case WSAENETRESET:
170 return "Network error: Network dropped connection on reset";
171 case WSAENETUNREACH:
172 return "Network error: Network is unreachable";
173 case WSAENOBUFS:
174 return "Network error: No buffer space available";
175 case WSAENOPROTOOPT:
176 return "Network error: Bad protocol option";
177 case WSAENOTCONN:
178 return "Network error: Socket is not connected";
179 case WSAENOTSOCK:
180 return "Network error: Socket operation on non-socket";
181 case WSAEOPNOTSUPP:
182 return "Network error: Operation not supported";
183 case WSAEPFNOSUPPORT:
184 return "Network error: Protocol family not supported";
185 case WSAEPROCLIM:
186 return "Network error: Too many processes";
187 case WSAEPROTONOSUPPORT:
188 return "Network error: Protocol not supported";
189 case WSAEPROTOTYPE:
190 return "Network error: Protocol wrong type for socket";
191 case WSAESHUTDOWN:
192 return "Network error: Cannot send after socket shutdown";
193 case WSAESOCKTNOSUPPORT:
194 return "Network error: Socket type not supported";
195 case WSAETIMEDOUT:
196 return "Network error: Connection timed out";
197 case WSAEWOULDBLOCK:
198 return "Network error: Resource temporarily unavailable";
199 case WSAEDISCON:
200 return "Network error: Graceful shutdown in progress";
201 default:
202 return "Unknown network error";
203 }
204 }
205
206 SockAddr sk_namelookup(char *host, char **canonicalname)
207 {
208 SockAddr ret = smalloc(sizeof(struct SockAddr_tag));
209 unsigned long a;
210 struct hostent *h = NULL;
211 char realhost[8192];
212
213 /* Clear the structure and default to IPv4. */
214 memset(ret, 0, sizeof(struct SockAddr_tag));
215 ret->family = 0; /* We set this one when we have resolved the host. */
216 *realhost = '\0';
217
218 if ((a = inet_addr(host)) == (unsigned long) INADDR_NONE) {
219 #ifdef IPV6
220
221 /* Try to get the getaddrinfo() function from wship6.dll */
222 /* This way one doesn't need to have IPv6 dll's to use PuTTY and
223 * it will fallback to IPv4. */
224 typedef int (CALLBACK * FGETADDRINFO) (const char *nodename,
225 const char *servname,
226 const struct addrinfo *
227 hints,
228 struct addrinfo ** res);
229 FGETADDRINFO fGetAddrInfo = NULL;
230
231 HINSTANCE dllWSHIP6 = LoadLibrary("wship6.dll");
232 if (dllWSHIP6)
233 fGetAddrInfo = (FGETADDRINFO) GetProcAddress(dllWSHIP6,
234 "getaddrinfo");
235
236 /*
237 * Use fGetAddrInfo when it's available (which usually also
238 * means IPv6 is installed...)
239 */
240 if (fGetAddrInfo) {
241 /*debug(("Resolving \"%s\" with getaddrinfo() (IPv4+IPv6 capable)...\n", host)); */
242 if (fGetAddrInfo(host, NULL, NULL, &ret->ai) == 0)
243 ret->family = ret->ai->ai_family;
244 } else
245 #endif
246 {
247 /*
248 * Otherwise use the IPv4-only gethostbyname...
249 * (NOTE: we don't use gethostbyname as a
250 * fallback!)
251 */
252 if (ret->family == 0) {
253 /*debug(("Resolving \"%s\" with gethostbyname() (IPv4 only)...\n", host)); */
254 if ( (h = gethostbyname(host)) )
255 ret->family = AF_INET;
256 }
257 }
258 /*debug(("Done resolving...(family is %d) AF_INET = %d, AF_INET6 = %d\n", ret->family, AF_INET, AF_INET6)); */
259
260 if (ret->family == 0) {
261 DWORD err = WSAGetLastError();
262 ret->error = (err == WSAENETDOWN ? "Network is down" :
263 err ==
264 WSAHOST_NOT_FOUND ? "Host does not exist" : err
265 == WSATRY_AGAIN ? "Host not found" :
266 #ifdef IPV6
267 fGetAddrInfo ? "getaddrinfo: unknown error" :
268 #endif
269 "gethostbyname: unknown error");
270 #ifdef DEBUG
271 {
272 LPVOID lpMsgBuf;
273 FormatMessage(FORMAT_MESSAGE_ALLOCATE_BUFFER |
274 FORMAT_MESSAGE_FROM_SYSTEM |
275 FORMAT_MESSAGE_IGNORE_INSERTS, NULL, err,
276 MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT),
277 (LPTSTR) & lpMsgBuf, 0, NULL);
278 /*debug(("Error %ld: %s (h=%lx)\n", err, lpMsgBuf, h)); */
279 /* Free the buffer. */
280 LocalFree(lpMsgBuf);
281 }
282 #endif
283 } else {
284 ret->error = NULL;
285
286 #ifdef IPV6
287 /* If we got an address info use that... */
288 if (ret->ai) {
289 typedef int (CALLBACK * FGETNAMEINFO)
290 (const struct sockaddr FAR * sa, socklen_t salen,
291 char FAR * host, size_t hostlen, char FAR * serv,
292 size_t servlen, int flags);
293 FGETNAMEINFO fGetNameInfo = NULL;
294
295 /* Are we in IPv4 fallback mode? */
296 /* We put the IPv4 address into the a variable so we can further-on use the IPv4 code... */
297 if (ret->family == AF_INET)
298 memcpy(&a,
299 (char *) &((SOCKADDR_IN *) ret->ai->
300 ai_addr)->sin_addr, sizeof(a));
301
302 /* Now let's find that canonicalname... */
303 if ((dllWSHIP6)
304 && (fGetNameInfo =
305 (FGETNAMEINFO) GetProcAddress(dllWSHIP6,
306 "getnameinfo"))) {
307 if (fGetNameInfo
308 ((struct sockaddr *) ret->ai->ai_addr,
309 ret->family ==
310 AF_INET ? sizeof(SOCKADDR_IN) :
311 sizeof(SOCKADDR_IN6), realhost,
312 sizeof(realhost), NULL, 0, 0) != 0) {
313 strncpy(realhost, host, sizeof(realhost));
314 }
315 }
316 }
317 /* We used the IPv4-only gethostbyname()... */
318 else
319 #endif
320 {
321 memcpy(&a, h->h_addr, sizeof(a));
322 /* This way we are always sure the h->h_name is valid :) */
323 strncpy(realhost, h->h_name, sizeof(realhost));
324 }
325 }
326 #ifdef IPV6
327 FreeLibrary(dllWSHIP6);
328 #endif
329 } else {
330 /*
331 * This must be a numeric IPv4 address because it caused a
332 * success return from inet_addr.
333 */
334 ret->family = AF_INET;
335 strncpy(realhost, host, sizeof(realhost));
336 }
337 ret->address = ntohl(a);
338 realhost[lenof(realhost)-1] = '\0';
339 *canonicalname = smalloc(1+strlen(realhost));
340 strcpy(*canonicalname, realhost);
341 return ret;
342 }
343
344 void sk_getaddr(SockAddr addr, char *buf, int buflen)
345 {
346 #ifdef IPV6
347 if (addr->family == AF_INET) {
348 #endif
349 struct in_addr a;
350 a.s_addr = htonl(addr->address);
351 strncpy(buf, inet_ntoa(a), buflen);
352 #ifdef IPV6
353 } else {
354 FIXME; /* I don't know how to get a text form of an IPv6 address. */
355 }
356 #endif
357 }
358
359 void sk_addr_free(SockAddr addr)
360 {
361 sfree(addr);
362 }
363
364 static Plug sk_tcp_plug(Socket sock, Plug p)
365 {
366 Actual_Socket s = (Actual_Socket) sock;
367 Plug ret = s->plug;
368 if (p)
369 s->plug = p;
370 return ret;
371 }
372
373 static void sk_tcp_flush(Socket s)
374 {
375 /*
376 * We send data to the socket as soon as we can anyway,
377 * so we don't need to do anything here. :-)
378 */
379 }
380
381 static void sk_tcp_close(Socket s);
382 static int sk_tcp_write(Socket s, char *data, int len);
383 static int sk_tcp_write_oob(Socket s, char *data, int len);
384 static char *sk_tcp_socket_error(Socket s);
385
386 extern char *do_select(SOCKET skt, int startup);
387
388 Socket sk_register(void *sock, Plug plug)
389 {
390 static struct socket_function_table fn_table = {
391 sk_tcp_plug,
392 sk_tcp_close,
393 sk_tcp_write,
394 sk_tcp_write_oob,
395 sk_tcp_flush,
396 sk_tcp_socket_error
397 };
398
399 DWORD err;
400 char *errstr;
401 Actual_Socket ret;
402
403 /*
404 * Create Socket structure.
405 */
406 ret = smalloc(sizeof(struct Socket_tag));
407 ret->fn = &fn_table;
408 ret->error = NULL;
409 ret->plug = plug;
410 bufchain_init(&ret->output_data);
411 ret->writable = 1; /* to start with */
412 ret->sending_oob = 0;
413 ret->frozen = 1;
414 ret->frozen_readable = 0;
415 ret->localhost_only = 0; /* unused, but best init anyway */
416 ret->pending_error = 0;
417
418 ret->s = (SOCKET)sock;
419
420 if (ret->s == INVALID_SOCKET) {
421 err = WSAGetLastError();
422 ret->error = winsock_error_string(err);
423 return (Socket) ret;
424 }
425
426 ret->oobinline = 0;
427
428 /* Set up a select mechanism. This could be an AsyncSelect on a
429 * window, or an EventSelect on an event object. */
430 errstr = do_select(ret->s, 1);
431 if (errstr) {
432 ret->error = errstr;
433 return (Socket) ret;
434 }
435
436 add234(sktree, ret);
437
438 return (Socket) ret;
439 }
440
441 Socket sk_new(SockAddr addr, int port, int privport, int oobinline,
442 int nodelay, Plug plug)
443 {
444 static struct socket_function_table fn_table = {
445 sk_tcp_plug,
446 sk_tcp_close,
447 sk_tcp_write,
448 sk_tcp_write_oob,
449 sk_tcp_flush,
450 sk_tcp_socket_error
451 };
452
453 SOCKET s;
454 #ifdef IPV6
455 SOCKADDR_IN6 a6;
456 #endif
457 SOCKADDR_IN a;
458 DWORD err;
459 char *errstr;
460 Actual_Socket ret;
461 short localport;
462
463 /*
464 * Create Socket structure.
465 */
466 ret = smalloc(sizeof(struct Socket_tag));
467 ret->fn = &fn_table;
468 ret->error = NULL;
469 ret->plug = plug;
470 bufchain_init(&ret->output_data);
471 ret->connected = 0; /* to start with */
472 ret->writable = 0; /* to start with */
473 ret->sending_oob = 0;
474 ret->frozen = 0;
475 ret->frozen_readable = 0;
476 ret->localhost_only = 0; /* unused, but best init anyway */
477 ret->pending_error = 0;
478
479 /*
480 * Open socket.
481 */
482 s = socket(addr->family, SOCK_STREAM, 0);
483 ret->s = s;
484
485 if (s == INVALID_SOCKET) {
486 err = WSAGetLastError();
487 ret->error = winsock_error_string(err);
488 return (Socket) ret;
489 }
490
491 ret->oobinline = oobinline;
492 if (oobinline) {
493 BOOL b = TRUE;
494 setsockopt(s, SOL_SOCKET, SO_OOBINLINE, (void *) &b, sizeof(b));
495 }
496
497 if (nodelay) {
498 BOOL b = TRUE;
499 setsockopt(s, IPPROTO_TCP, TCP_NODELAY, (void *) &b, sizeof(b));
500 }
501
502 /*
503 * Bind to local address.
504 */
505 if (privport)
506 localport = 1023; /* count from 1023 downwards */
507 else
508 localport = 0; /* just use port 0 (ie winsock picks) */
509
510 /* Loop round trying to bind */
511 while (1) {
512 int retcode;
513
514 #ifdef IPV6
515 if (addr->family == AF_INET6) {
516 memset(&a6, 0, sizeof(a6));
517 a6.sin6_family = AF_INET6;
518 /*a6.sin6_addr = in6addr_any; *//* == 0 */
519 a6.sin6_port = htons(localport);
520 } else
521 #endif
522 {
523 a.sin_family = AF_INET;
524 a.sin_addr.s_addr = htonl(INADDR_ANY);
525 a.sin_port = htons(localport);
526 }
527 #ifdef IPV6
528 retcode = bind(s, (addr->family == AF_INET6 ?
529 (struct sockaddr *) &a6 :
530 (struct sockaddr *) &a),
531 (addr->family ==
532 AF_INET6 ? sizeof(a6) : sizeof(a)));
533 #else
534 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
535 #endif
536 if (retcode != SOCKET_ERROR) {
537 err = 0;
538 break; /* done */
539 } else {
540 err = WSAGetLastError();
541 if (err != WSAEADDRINUSE) /* failed, for a bad reason */
542 break;
543 }
544
545 if (localport == 0)
546 break; /* we're only looping once */
547 localport--;
548 if (localport == 0)
549 break; /* we might have got to the end */
550 }
551
552 if (err) {
553 ret->error = winsock_error_string(err);
554 return (Socket) ret;
555 }
556
557 /*
558 * Connect to remote address.
559 */
560 #ifdef IPV6
561 if (addr->family == AF_INET6) {
562 memset(&a, 0, sizeof(a));
563 a6.sin6_family = AF_INET6;
564 a6.sin6_port = htons((short) port);
565 a6.sin6_addr =
566 ((struct sockaddr_in6 *) addr->ai->ai_addr)->sin6_addr;
567 } else
568 #endif
569 {
570 a.sin_family = AF_INET;
571 a.sin_addr.s_addr = htonl(addr->address);
572 a.sin_port = htons((short) port);
573 }
574
575 /* Set up a select mechanism. This could be an AsyncSelect on a
576 * window, or an EventSelect on an event object. */
577 errstr = do_select(s, 1);
578 if (errstr) {
579 ret->error = errstr;
580 return (Socket) ret;
581 }
582
583 if ((
584 #ifdef IPV6
585 connect(s, ((addr->family == AF_INET6) ?
586 (struct sockaddr *) &a6 : (struct sockaddr *) &a),
587 (addr->family == AF_INET6) ? sizeof(a6) : sizeof(a))
588 #else
589 connect(s, (struct sockaddr *) &a, sizeof(a))
590 #endif
591 ) == SOCKET_ERROR) {
592 err = WSAGetLastError();
593 /*
594 * We expect a potential EWOULDBLOCK here, because the
595 * chances are the front end has done a select for
596 * FD_CONNECT, so that connect() will complete
597 * asynchronously.
598 */
599 if ( err != WSAEWOULDBLOCK ) {
600 ret->error = winsock_error_string(err);
601 return (Socket) ret;
602 }
603 } else {
604 /*
605 * If we _don't_ get EWOULDBLOCK, the connect has completed
606 * and we should set the socket as writable.
607 */
608 ret->writable = 1;
609 }
610
611 add234(sktree, ret);
612
613 return (Socket) ret;
614 }
615
616 Socket sk_newlistener(int port, Plug plug, int local_host_only)
617 {
618 static struct socket_function_table fn_table = {
619 sk_tcp_plug,
620 sk_tcp_close,
621 sk_tcp_write,
622 sk_tcp_write_oob,
623 sk_tcp_flush,
624 sk_tcp_socket_error
625 };
626
627 SOCKET s;
628 #ifdef IPV6
629 SOCKADDR_IN6 a6;
630 #endif
631 SOCKADDR_IN a;
632 DWORD err;
633 char *errstr;
634 Actual_Socket ret;
635 int retcode;
636 int on = 1;
637
638 /*
639 * Create Socket structure.
640 */
641 ret = smalloc(sizeof(struct Socket_tag));
642 ret->fn = &fn_table;
643 ret->error = NULL;
644 ret->plug = plug;
645 bufchain_init(&ret->output_data);
646 ret->writable = 0; /* to start with */
647 ret->sending_oob = 0;
648 ret->frozen = 0;
649 ret->frozen_readable = 0;
650 ret->localhost_only = local_host_only;
651 ret->pending_error = 0;
652
653 /*
654 * Open socket.
655 */
656 s = socket(AF_INET, SOCK_STREAM, 0);
657 ret->s = s;
658
659 if (s == INVALID_SOCKET) {
660 err = WSAGetLastError();
661 ret->error = winsock_error_string(err);
662 return (Socket) ret;
663 }
664
665 ret->oobinline = 0;
666
667
668 setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (const char *)&on, sizeof(on));
669
670
671 #ifdef IPV6
672 if (addr->family == AF_INET6) {
673 memset(&a6, 0, sizeof(a6));
674 a6.sin6_family = AF_INET6;
675 if (local_host_only)
676 a6.sin6_addr = in6addr_loopback;
677 else
678 a6.sin6_addr = in6addr_any;
679 a6.sin6_port = htons(port);
680 } else
681 #endif
682 {
683 a.sin_family = AF_INET;
684 if (local_host_only)
685 a.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
686 else
687 a.sin_addr.s_addr = htonl(INADDR_ANY);
688 a.sin_port = htons((short)port);
689 }
690 #ifdef IPV6
691 retcode = bind(s, (addr->family == AF_INET6 ?
692 (struct sockaddr *) &a6 :
693 (struct sockaddr *) &a),
694 (addr->family ==
695 AF_INET6 ? sizeof(a6) : sizeof(a)));
696 #else
697 retcode = bind(s, (struct sockaddr *) &a, sizeof(a));
698 #endif
699 if (retcode != SOCKET_ERROR) {
700 err = 0;
701 } else {
702 err = WSAGetLastError();
703 }
704
705 if (err) {
706 ret->error = winsock_error_string(err);
707 return (Socket) ret;
708 }
709
710
711 if (listen(s, SOMAXCONN) == SOCKET_ERROR) {
712 closesocket(s);
713 ret->error = winsock_error_string(err);
714 return (Socket) ret;
715 }
716
717 /* Set up a select mechanism. This could be an AsyncSelect on a
718 * window, or an EventSelect on an event object. */
719 errstr = do_select(s, 1);
720 if (errstr) {
721 ret->error = errstr;
722 return (Socket) ret;
723 }
724
725 add234(sktree, ret);
726
727 return (Socket) ret;
728 }
729
730 static void sk_tcp_close(Socket sock)
731 {
732 extern char *do_select(SOCKET skt, int startup);
733 Actual_Socket s = (Actual_Socket) sock;
734
735 del234(sktree, s);
736 do_select(s->s, 0);
737 closesocket(s->s);
738 sfree(s);
739 }
740
741 /*
742 * The function which tries to send on a socket once it's deemed
743 * writable.
744 */
745 void try_send(Actual_Socket s)
746 {
747 while (s->sending_oob || bufchain_size(&s->output_data) > 0) {
748 int nsent;
749 DWORD err;
750 void *data;
751 int len, urgentflag;
752
753 if (s->sending_oob) {
754 urgentflag = MSG_OOB;
755 len = s->sending_oob;
756 data = &s->oobdata;
757 } else {
758 urgentflag = 0;
759 bufchain_prefix(&s->output_data, &data, &len);
760 }
761 nsent = send(s->s, data, len, urgentflag);
762 noise_ultralight(nsent);
763 if (nsent <= 0) {
764 err = (nsent < 0 ? WSAGetLastError() : 0);
765 if ((err < WSABASEERR && nsent < 0) || err == WSAEWOULDBLOCK) {
766 /*
767 * Perfectly normal: we've sent all we can for the moment.
768 *
769 * (Some WinSock send() implementations can return
770 * <0 but leave no sensible error indication -
771 * WSAGetLastError() is called but returns zero or
772 * a small number - so we check that case and treat
773 * it just like WSAEWOULDBLOCK.)
774 */
775 s->writable = FALSE;
776 return;
777 } else if (nsent == 0 ||
778 err == WSAECONNABORTED || err == WSAECONNRESET) {
779 /*
780 * If send() returns CONNABORTED or CONNRESET, we
781 * unfortunately can't just call plug_closing(),
782 * because it's quite likely that we're currently
783 * _in_ a call from the code we'd be calling back
784 * to, so we'd have to make half the SSH code
785 * reentrant. Instead we flag a pending error on
786 * the socket, to be dealt with (by calling
787 * plug_closing()) at some suitable future moment.
788 */
789 s->pending_error = err;
790 return;
791 } else {
792 fatalbox(winsock_error_string(err));
793 }
794 } else {
795 if (s->sending_oob) {
796 if (nsent < len) {
797 memmove(s->oobdata, s->oobdata+nsent, len-nsent);
798 s->sending_oob = len - nsent;
799 } else {
800 s->sending_oob = 0;
801 }
802 } else {
803 bufchain_consume(&s->output_data, nsent);
804 }
805 }
806 }
807 }
808
809 static int sk_tcp_write(Socket sock, char *buf, int len)
810 {
811 Actual_Socket s = (Actual_Socket) sock;
812
813 /*
814 * Add the data to the buffer list on the socket.
815 */
816 bufchain_add(&s->output_data, buf, len);
817
818 /*
819 * Now try sending from the start of the buffer list.
820 */
821 if (s->writable)
822 try_send(s);
823
824 return bufchain_size(&s->output_data);
825 }
826
827 static int sk_tcp_write_oob(Socket sock, char *buf, int len)
828 {
829 Actual_Socket s = (Actual_Socket) sock;
830
831 /*
832 * Replace the buffer list on the socket with the data.
833 */
834 bufchain_clear(&s->output_data);
835 assert(len <= sizeof(s->oobdata));
836 memcpy(s->oobdata, buf, len);
837 s->sending_oob = len;
838
839 /*
840 * Now try sending from the start of the buffer list.
841 */
842 if (s->writable)
843 try_send(s);
844
845 return s->sending_oob;
846 }
847
848 int select_result(WPARAM wParam, LPARAM lParam)
849 {
850 int ret, open;
851 DWORD err;
852 char buf[20480]; /* nice big buffer for plenty of speed */
853 Actual_Socket s;
854 u_long atmark;
855
856 /* wParam is the socket itself */
857 s = find234(sktree, (void *) wParam, cmpforsearch);
858 if (!s)
859 return 1; /* boggle */
860
861 if ((err = WSAGETSELECTERROR(lParam)) != 0) {
862 /*
863 * An error has occurred on this socket. Pass it to the
864 * plug.
865 */
866 return plug_closing(s->plug, winsock_error_string(err), err, 0);
867 }
868
869 noise_ultralight(lParam);
870
871 switch (WSAGETSELECTEVENT(lParam)) {
872 case FD_CONNECT:
873 s->connected = s->writable = 1;
874 break;
875 case FD_READ:
876 /* In the case the socket is still frozen, we don't even bother */
877 if (s->frozen) {
878 s->frozen_readable = 1;
879 break;
880 }
881
882 /*
883 * We have received data on the socket. For an oobinline
884 * socket, this might be data _before_ an urgent pointer,
885 * in which case we send it to the back end with type==1
886 * (data prior to urgent).
887 */
888 if (s->oobinline) {
889 atmark = 1;
890 ioctlsocket(s->s, SIOCATMARK, &atmark);
891 /*
892 * Avoid checking the return value from ioctlsocket(),
893 * on the grounds that some WinSock wrappers don't
894 * support it. If it does nothing, we get atmark==1,
895 * which is equivalent to `no OOB pending', so the
896 * effect will be to non-OOB-ify any OOB data.
897 */
898 } else
899 atmark = 1;
900
901 ret = recv(s->s, buf, sizeof(buf), 0);
902 noise_ultralight(ret);
903 if (ret < 0) {
904 err = WSAGetLastError();
905 if (err == WSAEWOULDBLOCK) {
906 break;
907 }
908 }
909 if (ret < 0) {
910 return plug_closing(s->plug, winsock_error_string(err), err,
911 0);
912 } else if (0 == ret) {
913 return plug_closing(s->plug, NULL, 0, 0);
914 } else {
915 return plug_receive(s->plug, atmark ? 0 : 1, buf, ret);
916 }
917 break;
918 case FD_OOB:
919 /*
920 * This will only happen on a non-oobinline socket. It
921 * indicates that we can immediately perform an OOB read
922 * and get back OOB data, which we will send to the back
923 * end with type==2 (urgent data).
924 */
925 ret = recv(s->s, buf, sizeof(buf), MSG_OOB);
926 noise_ultralight(ret);
927 if (ret <= 0) {
928 fatalbox(ret == 0 ? "Internal networking trouble" :
929 winsock_error_string(WSAGetLastError()));
930 } else {
931 return plug_receive(s->plug, 2, buf, ret);
932 }
933 break;
934 case FD_WRITE:
935 {
936 int bufsize_before, bufsize_after;
937 s->writable = 1;
938 bufsize_before = s->sending_oob + bufchain_size(&s->output_data);
939 try_send(s);
940 bufsize_after = s->sending_oob + bufchain_size(&s->output_data);
941 if (bufsize_after < bufsize_before)
942 plug_sent(s->plug, bufsize_after);
943 }
944 break;
945 case FD_CLOSE:
946 /* Signal a close on the socket. First read any outstanding data. */
947 open = 1;
948 do {
949 ret = recv(s->s, buf, sizeof(buf), 0);
950 if (ret < 0) {
951 err = WSAGetLastError();
952 if (err == WSAEWOULDBLOCK)
953 break;
954 return plug_closing(s->plug, winsock_error_string(err),
955 err, 0);
956 } else {
957 if (ret)
958 open &= plug_receive(s->plug, 0, buf, ret);
959 else
960 open &= plug_closing(s->plug, NULL, 0, 0);
961 }
962 } while (ret > 0);
963 return open;
964 case FD_ACCEPT:
965 {
966 struct sockaddr_in isa;
967 int addrlen = sizeof(struct sockaddr_in);
968 SOCKET t; /* socket of connection */
969
970 memset(&isa, 0, sizeof(struct sockaddr_in));
971 err = 0;
972 t = accept(s->s,&isa,&addrlen);
973 if (t == INVALID_SOCKET)
974 {
975 err = WSAGetLastError();
976 if (err == WSATRY_AGAIN)
977 break;
978 }
979
980 if (s->localhost_only &&
981 ntohl(isa.sin_addr.s_addr) != INADDR_LOOPBACK) {
982 closesocket(t); /* dodgy WinSock let nonlocal through */
983 } else if (plug_accepting(s->plug, (void*)t)) {
984 closesocket(t); /* denied or error */
985 }
986 }
987 }
988
989 return 1;
990 }
991
992 /*
993 * Deal with socket errors detected in try_send().
994 */
995 void net_pending_errors(void)
996 {
997 int i;
998 Actual_Socket s;
999
1000 /*
1001 * This might be a fiddly business, because it's just possible
1002 * that handling a pending error on one socket might cause
1003 * others to be closed. (I can't think of any reason this might
1004 * happen in current SSH implementation, but to maintain
1005 * generality of this network layer I'll assume the worst.)
1006 *
1007 * So what we'll do is search the socket list for _one_ socket
1008 * with a pending error, and then handle it, and then search
1009 * the list again _from the beginning_. Repeat until we make a
1010 * pass with no socket errors present. That way we are
1011 * protected against the socket list changing under our feet.
1012 */
1013
1014 do {
1015 for (i = 0; (s = index234(sktree, i)) != NULL; i++) {
1016 if (s->pending_error) {
1017 /*
1018 * An error has occurred on this socket. Pass it to the
1019 * plug.
1020 */
1021 plug_closing(s->plug,
1022 winsock_error_string(s->pending_error),
1023 s->pending_error, 0);
1024 break;
1025 }
1026 }
1027 } while (s);
1028 }
1029
1030 /*
1031 * Each socket abstraction contains a `void *' private field in
1032 * which the client can keep state.
1033 */
1034 void sk_set_private_ptr(Socket sock, void *ptr)
1035 {
1036 Actual_Socket s = (Actual_Socket) sock;
1037 s->private_ptr = ptr;
1038 }
1039
1040 void *sk_get_private_ptr(Socket sock)
1041 {
1042 Actual_Socket s = (Actual_Socket) sock;
1043 return s->private_ptr;
1044 }
1045
1046 /*
1047 * Special error values are returned from sk_namelookup and sk_new
1048 * if there's a problem. These functions extract an error message,
1049 * or return NULL if there's no problem.
1050 */
1051 char *sk_addr_error(SockAddr addr)
1052 {
1053 return addr->error;
1054 }
1055 static char *sk_tcp_socket_error(Socket sock)
1056 {
1057 Actual_Socket s = (Actual_Socket) sock;
1058 return s->error;
1059 }
1060
1061 void sk_set_frozen(Socket sock, int is_frozen)
1062 {
1063 Actual_Socket s = (Actual_Socket) sock;
1064 if (s->frozen == is_frozen)
1065 return;
1066 s->frozen = is_frozen;
1067 if (!is_frozen && s->frozen_readable) {
1068 char c;
1069 recv(s->s, &c, 1, MSG_PEEK);
1070 }
1071 s->frozen_readable = 0;
1072 }
1073
1074 /*
1075 * For Plink: enumerate all sockets currently active.
1076 */
1077 SOCKET first_socket(int *state)
1078 {
1079 Actual_Socket s;
1080 *state = 0;
1081 s = index234(sktree, (*state)++);
1082 return s ? s->s : INVALID_SOCKET;
1083 }
1084
1085 SOCKET next_socket(int *state)
1086 {
1087 Actual_Socket s = index234(sktree, (*state)++);
1088 return s ? s->s : INVALID_SOCKET;
1089 }