Fix more ident lookup bugs

[michaelortmann]

Found by: michaelortmann
Patch by: michaelortmann
Fixes:

One-line summary:
Fix more ident lookup bugs

Additional description (if needed):

If a socket is set O_NONBLOCK and connect() returns EINPROGRESS, it must be checked for ECONNREFUSED.
Here, this was done properly:

eggdrop/src/net.c

Lines 560 to 580 in 2a6a368

	rc = connect(sock, &addr->addr.sa, addr->addrlen);
	/* To minimize a proven race condition, call ident here (especially when
	* rc < 0 and errno == EINPROGRESS)
	*/
	if (dcc[i].status & STAT_SERV) {
	check_tcl_event("ident");
	}
	if (rc < 0) {
	if (errno == EINPROGRESS) {
	/* Async connection... don't return socket descriptor
	* until after we confirm if it was successful or not */
	tv.tv_sec = 1;
	tv.tv_usec = 0;
	FD_ZERO(&sockset);
	FD_SET(sock, &sockset);
	select(sock + 1, &sockset, NULL, NULL, &tv);
	res_len = sizeof(res);
	getsockopt(sock, SOL_SOCKET, SO_ERROR, &res, &res_len);
	if (res == EINPROGRESS) /* Operation now in progress */
	return sock; /* This could probably fail somewhere */
	if (res == ECONNREFUSED) { /* Connection refused */

But here it was not and a write was committed:

eggdrop/src/dcc.c

Lines 1383 to 1404 in 2a6a368

	if (connect(dcc[j].sock, &dcc[j].sockname.addr.sa,
	dcc[j].sockname.addrlen) < 0 && (errno != EINPROGRESS)) {
	killsock(dcc[j].sock);
	lostdcc(j);
	putlog(LOG_MISC, "*", DCC_IDENTFAIL, dcc[i].host, strerror(errno));
	j = 0;
	}
	sock = dcc[j].sock;
	}
	}
	if (j < 0) {
	dcc_telnet_got_ident(i, userhost);
	return;
	}
	dcc[j].sock = sock;
	dcc[j].port = 113;
	dcc[j].addr = dcc[i].addr;
	strcpy(dcc[j].host, dcc[i].host);
	strcpy(dcc[j].nick, "*");
	dcc[j].u.ident_sock = dcc[i].sock;
	dcc[j].timeval = now;
	dprintf(j, "%d, %d\n", dcc[i].port, dcc[idx].port);

Additionally O_NONBLOCK is POSIX 2001 and old quirk code is removed, see: https://www.gaertner.de/~neitzel/susv3/basedefs/fcntl.h.html

Additionally also the proper code mentioned above had an issue, it did select() with an timeout of 1 sec, when it should be nonblocking. to have it fully nonblocking, the main loop select() would have to handle the econnrefused. to keep this PR small, this PR lowers the timeout to 0.5 sec, to mitigate long blocking but still detect econnrefused.

Additionally there was a bug here:

eggdrop/src/dcc.c

Lines 1388 to 1403 in 2a6a368

	j = 0;
	}
	sock = dcc[j].sock;
	}
	}
	if (j < 0) {
	dcc_telnet_got_ident(i, userhost);
	return;
	}
	dcc[j].sock = sock;
	dcc[j].port = 113;
	dcc[j].addr = dcc[i].addr;
	strcpy(dcc[j].host, dcc[i].host);
	strcpy(dcc[j].nick, "*");
	dcc[j].u.ident_sock = dcc[i].sock;
	dcc[j].timeval = now;

Were this line j = 0 ever executed, the gates of tartaros would have opened for dcc[0].

Test cases demonstrating functionality (if applicable):
Without this fix:
When telnetting to the bot and the user has no identd running, the bot would log like:

[21:01:18] Telnet connection: localhost.home.arpa/45052
[21:01:18] net: eof!(write) socket 13 (Broken pipe,32)

With this fix:
Test without identd:

[12:24:02] Telnet connection: localhost.home.arpa/52216
[12:24:02] net: attempted socket connection refused: 127.0.0.1:113
[12:24:02] Ident failed for localhost.home.arpa: Connection refused
[12:24:09] pbkdf2 method SHA256 rounds 16000, user 12.493ms sys 0.416ms
[12:24:09] Logged in: testuser (telnet@localhost.home.arpa/52216)
*** testuser joined the party line.

Test wih identd:

[12:24:25] Telnet connection: localhost.home.arpa/42944
[12:24:28] pbkdf2 method SHA256 rounds 16000, user 16.223ms sys 6.315ms
[12:24:28] Logged in: testuser (michael@localhost.home.arpa/42944)
*** testuser joined the party line.

Additionally i applied this patch to windrop and it also works under cygwin.

[michaelortmann]

Maybe related/fixing #809?

[vanosg]

Please resolve conflicts, thanks

[michaelortmann]

Please resolve conflicts, thanks

done. ready for review again.

[thommey]

Overall very good change, but this does change the behavior of open_telnet_raw which previously had a select timeout of 1s which would detect ECONNREFUSED within that timeframe, now the timeout is 0 (return instantly) which never detects that and always returns a valid socket (unless it's localhost I guess).
That makes every invocation rely on the mainloop properly detecting and handling that, not sure if that's a regression risk or not, any opinion?

[michaelortmann]

eggdrop did select() with a timeout of 1 sec, when it should be nonblocking. to have it fully nonblocking, the main loop select() would have to handle the econnrefused. to keep this PR small, this PR lowers the timeout to 0.5 sec, to mitigate long blocking but still detect econnrefused.