Open GoogleCodeExporter opened 9 years ago
Can you provide proper backtrace for this crash. Your explanation of this bug
make no sense as the content of the file is not parsed by get_pac(). It simply
stat() the file and read it all at once. I would rather suspect the URI parser
to be broken with windows style file://L:/... (if that is the standard on
windows, but in any case it should not crash).
Original comment by nicolas.dufresne@gmail.com
on 5 Jul 2012 at 6:42
Ok, first thing is that you file:// URI is incorrect.
See this for simplified explanations:
http://blogs.msdn.com/b/ie/archive/2006/12/06/file-uris-in-windows.aspx
But that also mean we need a special code path for URI parsing on windows. As /
need to be changed into \, and file://server/Doc/file need to be interpreted as
UNC \\server\Doc\file while file:///C:/Doc/file has to have a resulting patch
C:\Doc\file. Currently parser has a bug and ind a hostname "/C:". Simply fixing
that bug result in /C:/Doc/File, so the initial / will need to be removed,
along with / \ conversion.
Original comment by nicolas.dufresne@gmail.com
on 5 Jul 2012 at 7:07
r846 fixes the issue where the ":" is confused with port division. I added 2
unit tests, that fails on windows at the moment. If you have a dev environment
on windows, can you change url.cpp to make those tests pass (there is a test
target) ?
Original comment by nicolas.dufresne@gmail.com
on 5 Jul 2012 at 7:17
[deleted comment]
[deleted comment]
[deleted comment]
Yes I agree that stat read the file at once, but the problem on windows
platform is due to ::open function call which opens .pac file in read only
mode. On windows there is carriage return characters which get escaped when you
open file in read only mode. This causes exception in format_pac_response
function with error "string subscript out of range" on windows platform. As
file is not opened in binary mode, the escape sequence characters CRLF causes
this exception may be because of the wrong length of file fetched by stat while
reading it in read only mode.
If I open .pac file in binary mode, above exception gets disappear.
Original comment by udayja...@gmail.com
on 25 Jul 2012 at 4:01
I have also got runtime crash, in get_pac function in url.cpp in following
cases:
When web services of server(where .pac files are kept) is not running (e.g.
Apache server is stopped running on server where .pac files are kept, there
crash while closing socket on windows platform.
I have got following stack frames while debugging libproxy:
libproxy.dll!libproxy::url::get_pac() Line 423 C++
libproxy.dll!libproxy::proxy_factory::_get_proxies(libproxy::url * realurl,
std::vector<std::basic_string<char,std::char_traits<char>,std::allocator<char>
>,std::allocator<std::basic_string<char,std::char_traits<char>,std::allocator<ch
ar> > > > & response) Line 352 + 0xb bytes C++
libproxy.dll!libproxy::proxy_factory::get_proxies(std::basic_string<char,std::ch
ar_traits<char>,std::allocator<char> > url_) Line 196 C++
libproxy.dll!px_proxy_factory_get_proxies(pxProxyFactory_ * self, const char *
url) Line 421 + 0x20 bytes C++
I have also attached snapshot of runtime exception caused.
Original comment by udayja...@gmail.com
on 25 Jul 2012 at 4:17
Attachments:
I have also observed runtime crash in following scenario:
Test step 1:
- Set up DNS server
- Add wpad host entry in the DNS
- Configure DNS setting on the test machine
- Open “Internet option” – “Connections” – “Local Area Network (LAN) Settings”
- Enable “Automatically detect settings”
- Launch the application which uses libproxy
- There is runtime crash
Test step 2:
- Set up DNS server
- Add wpad host entry in the DNS
- Configure DNS setting on the test machine
- Open “Internet option” – “Connections” – “Local Area Network (LAN) Settings”
- Enable “Automatically detect settings”
- Add the wpad host entry in hosts file
- Launch the application which uses libproxy
- There is runtime crash
In both of these cases, stack unwinding information is as follows:
0038e8a4 6f4caf8a 00000000 00000000 00000000 msvcr100!invoke_watson+0x12
0038e8f4 72f1b994 000002bc 717dd718 007e7570
msvcr100!invalid_parameter_noinfo+0xc
0038ea00 72f1fc8a 717dd748 021d1f68 021d10a0
libproxy!libproxy::url::get_pac+0x364
0038ea50 72f18ddc 021d1f9c 717dd1c8 021d1f68
libproxy!libmodman::extension<libproxy::pacrunner_extension,1>::singleton+0x35da
0038ecd0 72f19553 007e73a8 0038ed78 717dd058
libproxy!libmodman::module_manager::module_manager+0x12ec
0038ed40 72f196ec 0038ed78 007e7300 00100022
libproxy!libmodman::module_manager::module_manager+0x1a63
0038edac 003f0325 021d1f68 023dfa88 717d7cb3
libproxy!px_proxy_factory_get_proxies+0x7c
And call stack is:
- libproxy.dll!libproxy::url::get_pac() Line 423 C++
- libproxy.dll!libproxy::proxy_factory::_get_proxies(libproxy::url * realurl,
std::vector<std::basic_string<char,std::char_traits<char>,std::allocator<char>
>,std::allocator<std::basic_string<char,std::char_traits<char>,std::allocator<ch
ar> > > > & response) Line 352 + 0xb bytes C++
-
libproxy.dll!libproxy::proxy_factory::get_proxies(std::basic_string<char,std::ch
ar_traits<char>,std::allocator<char> > url_) Line 196 C++
- libproxy.dll!px_proxy_factory_get_proxies(pxProxyFactory_ * self, const char
* url) Line 421 + 0x20 bytes C++
This runtime crash is at same location i.e. while closing socket. I dont know
the reason of crash behind it.
Original comment by udayja...@gmail.com
on 25 Jul 2012 at 4:33
The are no crash but debug assertion. Unless run with the development version
of msvcrt, they should not be visible to user. Having these may result in leaks
(in this case client socket leaks).
My first impression is that this code is not using the windows closesocket()
function. On windows, unlike all other OS, a socket is not a file descriptor,
and must be closed using closesocket(). Using normal close() causes debug
assertion, and leaks the socket.
It would be nice if you split this bug, as you started reporting many things on
the same bug. Also, have you had time to run the unit test suggested earlier,
and write a fix to url.cpp to make them work ?
Original comment by nicolas.dufresne@gmail.com
on 15 Oct 2012 at 8:27
Also, what I don't get here is why didn't you proposed a patch after testing
the binary mode fix some of the issue?
Original comment by nicolas.dufresne@gmail.com
on 12 Aug 2015 at 2:19
Sorry, but I do not have code base with me which contains fix.
Original comment by udayja...@gmail.com
on 12 Aug 2015 at 2:37
Original issue reported on code.google.com by
udayja...@gmail.com
on 26 Jun 2012 at 10:21Attachments: