This fixes the "busy loop waiting for an SSL handshake to finish" which the last
"Update to latest Csocket" was already supposed to fix. However, that fix had a
bug if poll() is used instead of select().
poll()'s timeout argument is in milliseconds while select also allows
microseconds. Since Csocket originally used select(), it expects the
microseconds-approach. This means it has to divide by 1000 to get the timeout
argument for poll().
However, the iQuickReset which was used to "fix" (rather: hide) the busy loop
was less than 1ms so this still resulted in a timeout of 0 (= busy loop) because
integer division truncates the result.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Fixes:
- A possible crash bug for empty DNS replies with c-ares. E.g. a AAAA lookup for
google.com doesn't give any reply but is still successful. This might be a
c-ares bug (there is ARES_ENODATA) or c-ares just changed its behavior?
(No bug report, just noticed accidentally)
- Connecting to ipv4-only hosts with a v6 bindhost caused weird errors:
https://github.com/znc/znc/issues/47
- There was a pull request for some DSA server certificate thingy:
https://github.com/znc/znc/pull/46
- Busy loop waiting for an SSL handshake to finish:
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=631590
- Some other stuff? No idea what some of the changes in here are actually doing.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This fixes a problem where Csocket detected socket timeouts (as configured with
SetTimeout()) too late if there was few activity. See csocket's git commit
message for more info, I'm lazy. :-P
Signed-off-by: Uli Schlachter <psychon@znc.in>
We are using a vanilla version of Csocket again. Previously, our copy of
Csocket.h had a new '#include "zncconfig.h"'.
The other changes are mostly cleanup for various compiler warnings. The switch
from inet_ntoa() to inet_ntop() is for thread-safety reasons.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This commit adds a zncconfig.h to ZNC that is automatically generated by
configure. This is done because the -DPACKAGE_STRING=\"znc\ 0.097\" that
configure adds to CXXFLAGS breaks znc-buildmod.
This means that we have to include zncconfig.h as the very first header in every
C++ file that is compiled. This commit kinda cheats and instead adds this
include as the very first thing to all header files we have. This should
hopefully mean that modules don't have to include this.
Because Csocket includes defines.h too late, this commit causes znc to divert
from upstream Csocket once again. :(
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@2250 726aef4b-f618-498e-8847-2d620e286838
select() says a socket is readable when the other end hangs up, with poll() you
only get POLLHUP. These other flags should all be handled like POLLIN.
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@2165 726aef4b-f618-498e-8847-2d620e286838
This fixes a busy-loop bug with openssl and adds support for using poll()
instead of select().
Currently, poll() isn't used unless you use
./configure CXXFLAGS="-DCSOCK_USE_POLL".
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@2096 726aef4b-f618-498e-8847-2d620e286838
This fixes a bug where the wrong errno value was used for SockError(). This also
fixes a busy loop caused by broken c-ares timeout handling on our side.
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@2052 726aef4b-f618-498e-8847-2d620e286838
This adds Csock::ConvertAddress() which produces printable addresses. We use
this function in CZNCSock to strip away 4in6 prefixes (e.g. ::ffff:127.0.0.1).
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@1930 726aef4b-f618-498e-8847-2d620e286838
Every socket which is based on CZNCSock will now use c-ares for its name
resolving. This is possible thanks to CSocket's Csock::GetAddrInfo() which
lets one override the DNS lookup.
This can be disabled with --disable-c-ares.
If IPv6 is enabled and Csocket didn't specify which kind of lookup (ipv4/ipv6)
it wants, we first do an ipv4 lookup. If that lookup doesn't yield any useful
result, we try again with an ipv6 lookup. If one wants to force ipv6 usage on a
domain which also resolves to an ipv4 address, he has to set an ipv6 vhost.
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@1551 726aef4b-f618-498e-8847-2d620e286838
This adds a new file defines.h and moves all the Csocket specific stuff from
main.h into this new file.
Then this makes Csocket use ZNC's DEBUG() macro for its debug output which
means this can now also be enabled by running znc --debug.
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@1445 726aef4b-f618-498e-8847-2d620e286838
This is some stuff I wrote:
- sockets are marked as close-on-exec so that child processes (e.g. from the
shell module) don't get a copy of all of our sockets.
- This adds separate functions for marking sockets as blocking / non-blocking.
- When a socket is marked for closing, we stop reading lines from it.
I don't think any of these two bugs were actually ever triggered by anyone,
but meh, so what?
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@941 726aef4b-f618-498e-8847-2d620e286838
This time it fixes ZNC on BSD* (netbsd and openbsd were reported, only netbsd
tested, but well :) ).
We used an illegal select timeout (tv_usec bigger than 999999).
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@863 726aef4b-f618-498e-8847-2d620e286838
DynamicSelectLoop had a bug causing it to always use the lower bound making it
effectively useless. Now it's useful again :)
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@859 726aef4b-f618-498e-8847-2d620e286838
This fixes an unimportant leak (if ReadBuf or ReadLine returned by throwing an
exception a 4k buffer was leaked).
We only triggered this if we were shutting down and thus this leak is
unimportant to us.
Also remove the svn:keywords property from Csocket.h and .cpp so the $Revision$
strings from upstream aren't changed.
git-svn-id: https://znc.svn.sourceforge.net/svnroot/znc/trunk@854 726aef4b-f618-498e-8847-2d620e286838
Uli Schlachter