1 /*****************************************************************************\
4 * Author : Chris Koeritz *
6 *******************************************************************************
7 * Copyright (c) 1991-$now By Author. This program is free software; you can *
8 * redistribute it and/or modify it under the terms of the GNU General Public *
9 * License as published by the Free Software Foundation; either version 2 of *
10 * the License or (at your option) any later version. This is online at: *
11 * http://www.fsf.org/copyleft/gpl.html *
12 * Please send any updates to: fred@gruntose.com *
13 \*****************************************************************************/
15 #include "raw_socket.h"
16 #include "tcpip_stack.h"
18 #include <basis/functions.h>
19 #include <loggers/critical_events.h>
20 #include <loggers/program_wide_logger.h>
21 #include <timely/time_stamp.h>
28 #include <arpa/inet.h>
30 #include <netinet/tcp.h>
31 #include <sys/ioctl.h>
32 #include <sys/socket.h>
34 #define OPTYPE (void *)
37 #define OPTYPE (char *)
40 using namespace basis;
41 using namespace loggers;
42 using namespace timely;
46 //#define DEBUG_RAW_SOCKET
47 // uncomment for noisy diagnostics.
50 #define LOG(to_print) CLASS_EMERGENCY_LOG(program_wide_logger::get(), to_print)
52 const int MULTIPLE_DISCONNECT_CHECKS = 28;
53 // we will make certain that select really says there's data and ioctl
54 // really says there's no data waiting before we believe there's been
55 // a disconnect. we'll check that state the number of times specified.
57 class fd_set_wrapper : public fd_set {};
61 const basis::un_int NON_BLOCKING = FIONBIO;
62 const basis::un_int IOCTL_READ = FIONREAD;
66 // defined by winsock header but not present in the winsock dll.
67 int PASCAL FAR __WSAFDIsSet(SOCKET fd, fd_set FAR *the_set)
69 int i = the_set->fd_count;
71 if (the_set->fd_array[i] == fd)
80 raw_socket::raw_socket()
81 : _stack(new tcpip_stack())
84 raw_socket::~raw_socket()
89 int raw_socket::close(basis::un_int &socket)
93 to_return = closesocket(socket);
96 to_return = ::close(socket);
102 //move this into parser bits as a OR combiner or something.
103 void combine(astring &existing, const astring &addition)
106 if (existing.t()) existing += " | ";
107 existing += addition;
111 astring raw_socket::interest_name(int interest)
114 if (interest & SI_CONNECTED) combine(to_return, "CONNECTED");
115 if (interest & SI_DISCONNECTED) combine(to_return, "DISCONNECTED");
116 if (interest & SI_WRITABLE) combine(to_return, "WRITABLE");
117 if (interest & SI_READABLE) combine(to_return, "READABLE");
118 if (interest & SI_ERRONEOUS) combine(to_return, "ERRONEOUS");
119 if (!interest) combine(to_return, "NORMAL");
123 int raw_socket::ioctl(basis::un_int socket, int request, void *argp) const
126 return ::ioctl(socket, request, argp);
129 return ioctlsocket(socket, request, (un_int *)argp);
133 bool raw_socket::set_non_blocking(basis::un_int socket, bool non_blocking)
135 FUNCDEF("set_non_blocking");
137 int curr_flags = fcntl(socket, F_GETFL, 0);
138 if (fcntl(socket, F_SETFL, curr_flags | O_NONBLOCK) < 0) return false;
140 int arg = int(non_blocking);
141 if (negative(ioctl(socket, NON_BLOCKING, &arg))) {
142 LOG(a_sprintf("Could not set non-blocking (FIONBIO) option on raw_socket %u.", socket));
149 bool raw_socket::set_nagle_algorithm(basis::un_int socket, bool use_nagle)
151 FUNCDEF("set_nagle_algorithm");
152 int arg = int(!use_nagle); // opposite of the flag, since we set no-delay.
153 if (negative(setsockopt(socket, IPPROTO_TCP, TCP_NODELAY, OPTYPE &arg,
155 LOG(a_sprintf("Could not change nagle coalescing mode on %u.", socket));
161 bool raw_socket::set_broadcast(basis::un_int socket, bool broadcasting)
163 FUNCDEF("set_broadcast");
164 int arg = int(broadcasting);
165 if (negative(setsockopt(socket, SOL_SOCKET, SO_BROADCAST, OPTYPE &arg,
167 LOG(a_sprintf("Could not change broadcast mode on %u.", socket));
173 bool raw_socket::set_reuse_address(basis::un_int socket, bool reuse)
175 FUNCDEF("set_reuse_address");
176 int arg = int(reuse);
177 if (negative(setsockopt(socket, SOL_SOCKET, SO_REUSEADDR, OPTYPE &arg,
179 LOG(a_sprintf("Could not set reuse address mode on %u.", socket));
185 bool raw_socket::set_keep_alive(basis::un_int socket, bool keep_alive)
187 FUNCDEF("set_keep_alive");
188 int arg = int(keep_alive);
189 if (negative(setsockopt(socket, SOL_SOCKET, SO_KEEPALIVE, OPTYPE &arg,
191 LOG(a_sprintf("Could not set keep alive mode on %u.", socket));
197 int raw_socket::select(basis::un_int socket, int mode, int timeout) const
199 FUNCDEF("select [single]");
200 if (!socket) return SI_ERRONEOUS;
201 fd_set_wrapper read_list, write_list, exceps;
202 int ret = inner_select(socket, mode, timeout, read_list, write_list, exceps);
203 if (!ret) return 0; // nothing is happening.
204 if (ret == SI_ERRONEOUS) return SI_ERRONEOUS; // something bad happened.
205 // otherwise we should be at base-line status.
206 return analyze_select_result(socket, mode, read_list, write_list, exceps);
209 int raw_socket::inner_select(basis::un_int socket, int mode, int timeout,
210 fd_set_wrapper &read_list, fd_set_wrapper &write_list,
211 fd_set_wrapper &exceptions) const
213 FUNCDEF("inner_select");
214 // setup the file descriptor sets for the select. we check readability,
215 // writability and exception status.
216 FD_ZERO(&read_list); FD_SET(socket, &read_list);
217 FD_ZERO(&write_list); FD_SET(socket, &write_list);
218 FD_ZERO(&exceptions); FD_SET(socket, &exceptions);
220 timeval base_time_out;
221 time_stamp::fill_timeval_ms(base_time_out, timeout);
222 // timeval has tv_sec=seconds, tv_usec=microseconds.
223 #if !defined(__GNU_WINDOWS__)
224 timeval *time_out = &base_time_out;
225 #elif defined(__GNU_WINDOWS__)
226 __ms_timeval win_time_out;
227 win_time_out.tv_sec = base_time_out.tv_sec;
228 win_time_out.tv_usec = base_time_out.tv_usec;
229 __ms_timeval *time_out = &win_time_out;
232 // select will tell us about the socket.
233 int ret = ::select(socket + 1,
234 (mode & SELECTING_JUST_WRITE)? NIL : &read_list,
235 (mode & SELECTING_JUST_READ)? NIL : &write_list,
236 &exceptions, time_out);
237 int error = critical_events::system_error();
238 if (!ret) return 0; // nothing to report.
240 if (ret == SOCKET_ERROR) {
242 // all real errors fall out to the error handling stuff.
243 case SOCK_EFAULT: // intentional fall-through.
244 case SOCK_ENETDOWN: // intentional fall-through.
245 case SOCK_EINVAL: // intentional fall-through.
246 case SOCK_EINTR: // intentional fall-through.
248 case SOCK_NOTINITIALISED: // intentional fall-through.
253 // hopefully all these others are bogus errors...
254 case SOCK_EINPROGRESS: // intentional fall-through.
255 case 0: // intentional fall-through.
257 #ifdef DEBUG_RAW_SOCKET
258 LOG("got to weird case, in progress or zero.");
260 return 0; // not really an error.
262 #ifdef DEBUG_RAW_SOCKET
263 LOG(a_sprintf("socket %u had error %d in select: %s.",
264 socket, error, _stack->tcpip_error_name(error).s()));
269 // if we got to here, then there are some things to report...
273 int raw_socket::test_readability(basis::un_int socket) const
275 FUNCDEF("test_readability");
277 if (negative(ioctl(socket, IOCTL_READ, &len))) {
278 LOG(astring(astring::SPRINTF, "socket %u had ioctl error: %s.",
279 socket, _stack->tcpip_error_name(critical_events::system_error()).s()));
282 if (positive(len)) return SI_READABLE;
283 else return SI_DISCONNECTED;
287 int raw_socket::analyze_select_result(basis::un_int socket, int mode,
288 fd_set_wrapper &read_list, fd_set_wrapper &write_list,
289 fd_set_wrapper &exceptions) const
291 #ifdef DEBUG_RAW_SOCKET
292 FUNCDEF("analyze_select_result");
296 // in case of an exception, we return an error.
297 if (FD_ISSET(socket, &exceptions)) {
298 #ifdef DEBUG_RAW_SOCKET
299 LOG(astring(astring::SPRINTF, "exception seen for socket %u!", socket));
303 // check to see if there are bytes to read.
304 if ( ! (mode & SELECTING_JUST_WRITE) && FD_ISSET(socket, &read_list)) {
305 // make sure we have data. if no data is available, it means a
306 // disconnect occurred.
308 int readable = test_readability(socket);
309 if (readable == SI_ERRONEOUS)
310 to_return |= SI_ERRONEOUS;
311 else if (readable == SI_READABLE)
312 to_return |= SI_READABLE;
313 else if (readable == SI_DISCONNECTED) {
314 // we need to check multiple times to be sure the OS really means this.
315 // either windoze seems to report an erroneous disconnect every few
316 // days or there's a bad synchronization issue as yet uncovered.
317 bool really_disconnected = true;
318 for (int i = 0; i < MULTIPLE_DISCONNECT_CHECKS; i++) {
319 fd_set_wrapper read_list, write_list, exceps;
320 int temp_ret = inner_select(socket, SELECTING_JUST_READ, 0, read_list,
322 // check the return value first...
324 // nothing happening (a zero return) means the socket's no longer
325 // claiming to have a readable state; our disconnect condition is
326 // thus violated and we can leave.
327 really_disconnected = false;
330 if (temp_ret == SI_ERRONEOUS) {
331 // this, on the other hand, sounds really bad. the socket doesn't
332 // seem to exist any more or something else horrid happened.
333 really_disconnected = true;
336 // if the select worked, we can check the fd_set now for readability.
337 if (!FD_ISSET(socket, &read_list)) {
338 // we are not in a disconnected state without being told we're
339 // readable. supposedly.
340 really_disconnected = false;
343 // now we really test the socket for readability by making sure there
344 // really is data pending on the socket. if it's readable but there's
345 // no data, then either a disconnection has occurred or is in progress.
346 readable = test_readability(socket);
347 if (readable != SI_DISCONNECTED) {
348 // we are not disconnected if there's really data waiting.
349 really_disconnected = false;
353 if (really_disconnected) {
354 #ifdef DEBUG_RAW_SOCKET
355 LOG(a_sprintf("connection closed on socket %u.", socket));
357 to_return |= SI_DISCONNECTED;
362 // check writability state.
363 if (! (mode & SELECTING_JUST_READ) && FD_ISSET(socket, &write_list)) {
364 to_return |= SI_WRITABLE;
370 int raw_socket::select(int_array &read_sox, int_array &write_sox,
373 #ifdef DEBUG_RAW_SOCKET
374 FUNCDEF("select [multiple]");
376 if (!read_sox.length() && !write_sox.length())
377 return 0; // nothing happened to nothing.
379 int to_return = 0; // will get bits slammed into it to report results.
381 // setup the file descriptor sets for the select. we check readability,
382 // writability and exception status.
383 fd_set_wrapper read_list; FD_ZERO(&read_list);
384 fd_set_wrapper write_list; FD_ZERO(&write_list);
385 fd_set_wrapper exceptions; FD_ZERO(&exceptions);
386 // set up the lists with the sets we were handed.
387 basis::un_int highest = 0;
389 for (i = 0; i < read_sox.length(); i++) {
390 basis::un_int sock = (basis::un_int)read_sox[i];
391 if (sock > highest) highest = sock;
392 FD_SET(sock, &read_list);
394 for (i = 0; i < write_sox.length(); i++) {
395 basis::un_int sock = (basis::un_int)write_sox[i];
396 if (sock > highest) highest = sock;
397 FD_SET(sock, &write_list);
400 timeval base_time_out;
401 time_stamp::fill_timeval_ms(base_time_out, timeout);
402 // timeval has tv_sec=seconds, tv_usec=microseconds.
403 #if !defined(__GNU_WINDOWS__)
404 timeval *time_out = &base_time_out;
405 #elif defined(__GNU_WINDOWS__)
406 __ms_timeval win_time_out;
407 win_time_out.tv_sec = base_time_out.tv_sec;
408 win_time_out.tv_usec = base_time_out.tv_usec;
409 __ms_timeval *time_out = &win_time_out;
412 // select will tell us about the socket.
413 int ret = ::select(highest + 1,
414 (read_sox.length())? &read_list : NIL,
415 (write_sox.length())? &write_list : NIL,
416 &exceptions, time_out);
417 int error = critical_events::system_error();
419 if (ret == SOCKET_ERROR) {
421 // all real errors fall out to the error handling stuff.
422 case SOCK_EFAULT: // intentional fall-through.
423 case SOCK_ENETDOWN: // intentional fall-through.
424 case SOCK_EINVAL: // intentional fall-through.
425 case SOCK_EINTR: // intentional fall-through.
427 case SOCK_NOTINITIALISED: // intentional fall-through.
432 // hopefully all these others are bogus errors...
433 case SOCK_EINPROGRESS: // intentional fall-through.
434 case 0: // intentional fall-through.
436 #ifdef DEBUG_RAW_SOCKET
437 LOG("got to weird case, in progress or zero.");
440 //hmmm: fix retd sox? what's this outcome mean for the list?
442 return 0; // not really an error.
444 #ifdef DEBUG_RAW_SOCKET
445 LOG(a_sprintf("sockets had error %d in select: %s.",
446 error, _stack->tcpip_error_name(error).s()));
449 //hmmm: fix retd sox? what's this outcome mean for the list?
453 // we know of nothing exciting for any of these.
456 return 0; // nothing to report.
459 // if we got to here, then there are some things to report...
460 // iterate through the lists and check results.
461 for (int k = 0; k < read_sox.length(); k++) {
462 basis::un_int socket = read_sox[k];
463 int interim = analyze_select_result(socket, SELECTING_JUST_READ, read_list,
464 write_list, exceptions);
466 // nothing happened on that guy.
468 k--; // skip back to before the whack.
471 to_return |= interim; // join the results with overall result.
473 for (int p = 0; p < write_sox.length(); p++) {
474 basis::un_int socket = write_sox[p];
475 int interim = analyze_select_result(socket, SELECTING_JUST_WRITE, read_list,
476 write_list, exceptions);
478 // nothing happened on that guy.
480 p--; // skip back to before the whack.
483 to_return |= interim; // join the results with overall result.