Re: Perl syswrite and IO/SSL

On May 19, 2009, at 2:31 PM, Dale Moore wrote:
> I am having a problem with LDAP.pm and wondering
> if any other folks are experiencing similar problems.
>
> The problems are the result of code that was modified
> in this patch
>
> http://git.goingon.net/?p=perl-ldap.git;a=commitdiff;h=8c6c9cfcb78adf229437507f2e8ffe86d15e712d
>
> There are some style issues that I don’t like about this patch.
>  - We do a length() call on each time through the loop rather than  
> once
>  - We do a substr call each time through the loop, which isnt  
> necessary
>    if we would instead use the offset parameter to syswrite
>  - No reference in where the number 15000 came from.
>
> But the patch appears to basically sound, even if it does offend my  
> sense
> of asthetics.

Patches welcome.

>  The only problem I am having is that after a successful
> bind and several passes through this code, it starts to misbehave.
> The calls to syswrite return the value 1 instead of the number of  
> bytes
> in the buffer as I might have expected.
>
> The problem is
>  - The call to syswrite() of a buffer of several hundred bytes  
> always returns 1
>    and not the number of the several hundred bytes as expected.
>    This causes the syswrite to be retried with a slightly smaller  
> buffer.
>    The remote server appears to receive the original request, and  
> responds
>    but subsequent writes of the smaller buffer confuse it and it  
> drops the connection.
>  - A system call trace of perl process reveals that it is calling  
> write()
>    with several hundred bytes, and it is succeeding in writing all  
> of the bytes,
>    NOT just 1 or a smaller number of bytes...

Do you have the latest Net::SSLeay and IO::Socket::SSL installed ?

When using SSL, the syswrite is effectively a call to  
Net::SSLeay::write_partial

> I've tried judicious use of 'use bytes;' and a few other common  
> tricks.

use bytes would not make a difference here.

> If you'll look at the previous version, before this patch, you'll  
> see where
> we only looked to see if syswrite returned zero or non-zero.  Not  
> the actual
> number of bytes.

Which was a bug as syswrite does not guarantee to send all the data

Graham.

>  If I tweak my private copy of LDAP.pm to assume that
> syswrite returned what I was expecting, it works as expected, and  
> the client
> and server successfully communicate for as long as expected.
>
> I've seen this problem with Perl 5.8.8 on Solaris 9, 10, and 32Bit  
> i386 Redhat Fc5.
> If you've seen this problem, or similar problems, or have a strong  
> clue
> as to what might be wrong, I'd appreciate your comments.
>
> Comments welcome,
>
>