[ 
https://issues.apache.org/jira/browse/THRIFT-3941?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15545886#comment-15545886
 ] 

Ted Wang commented on THRIFT-3941:
----------------------------------

In fact, Thrift specifically targets Windows XP for broadest compatibility, so 
by default, all clients get the Win XP implementation with this undefined 
behavior.

This is in the README:
## Windows version compatibility

The Thrift library targets Windows XP for broadest compatbility. A notable      
                   
difference is in the Windows-specific implementation of the socket poll
function. To target Vista, Win7 or other versions, comment out the line

    #define TARGET_WIN_XP.

And here is the source:
https://github.com/apache/thrift/blob/master/lib/cpp/src/thrift/windows/config.h#L44


> WinXP version of thrift_poll() relies on undefined behavior by passing a 
> destructed variable to select()
> --------------------------------------------------------------------------------------------------------
>
>                 Key: THRIFT-3941
>                 URL: https://issues.apache.org/jira/browse/THRIFT-3941
>             Project: Thrift
>          Issue Type: Bug
>          Components: C++ - Library
>    Affects Versions: 0.9.3
>            Reporter: Ted Wang
>            Assignee: Ted Wang
>
> thrift_poll() for WINVER <= 0x0502 in thrift/windows/WinFnctl.cpp shadows the 
> 'time_out' variable, and it ends up passing the destructed copy to select():
>   timeval time_out;
>   timeval* time_out_ptr = NULL;
>   if (timeout >= 0) {
>     timeval time_out = {timeout / 1000, (timeout % 1000) * 1000};
>     time_out_ptr = &time_out;
>   } else { // to avoid compiler warnings
>     (void)time_out;
>     (void)timeout;
>   }
>   int sktready = select(1, read_fds_ptr, write_fds_ptr, NULL, time_out_ptr);
> Stepping through this code in the debugger, it looks like MSVC reserves a 
> large enough stack frame to avoid overwriting the variable when calling 
> select(), which may be why this hasn't been caught yet.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to