Re: bug in passing file descriptors

[Steve Rago <sar@nec-labs.com>]

On 10/08/2013 12:41 PM, Andi Kleen wrote:
>> I just want the semantics to be consistent.  If you want Linux to
>> always require applications that call recvmsg to provide a buffer
>> size of CMSG_SPACE bytes long to retrieve control information, then
>> fail the system call when the buffer is smaller.  But if you do
>> this, you risk breaking applications that work with FreeBSD, Mac OS
>> X, Solaris, and probably a few others.
>
> The primary concern is to be binary compatible with Linux.
>
> But not being compatible between 32bit and 64bit Linux processes on the same
> host would seem like a serious problem to me.
>
>> Regardless, copying 20 bytes and telling me you copied 24 is misleading and wrong.
>
> The question is could it break existing Linux applications to change it?
> And would it help with the 32/64bit compatibility?
>
> If not some other way to fix the compat layer would need to be found.
>
> -Andi
>

I'm not sure if a 64-bit process and a 32-bit process exchange file descriptors on the same system has a problem.  It 
certainly looks like the compat code does the right thing.  I can test this tonight if you want.  The discrepancy arises 
because file descriptors are 4-byte quantities and treated differently (for example, when more than one is specified, 
each one isn't padded to an 8-byte boundary).

The way I discovered the problem is that I had an example program in APUE that was validating that msg_controllen had 
the correct value after calling recvmsg().  It worked on my 32-bit platform, but when I recompiled it and ran it on my 
64-bit platform, the test failed, because msg_controllen was larger than the size that was sent via sendmsg().

Steve