[openssl-dev] [openssl.org #3739] regression: syswrite payloads >90kb can trigger EFAULT "Bad address" error on 1.0.2

[Erik Forsberg]

I found the same issue, seems unique to 1.0.2 MULTIBLOCK
The attached patch (developed by Matt Caswell) appears to fix it for me.
An alternative, to avoid the issue is compiling with OPENSSL_NO_MULTIBLOCK
or only writing small blocks of memory per SSL_write.

Flow control was not handled properly in the new multiblock code.

>-- Original Message --
>
>I had a hard time nailing down this problem, so apologies in advance if
>this bug is confusing.
>
>1.0.2 seems to have a problem exhibited in various places doing HTTPS
>uploads over 90k ( ballpark ).
>
>1.0.1l did not suffer this error.
>
>The problem is exhibited in both ruby openssl bindings (
>https://github.com/excon/excon/issues/467 ) and perl openssl bindings (
>https://rt.cpan.org/Ticket/Display.html?id=102640 )
>
>As to where this problem is cropping up and where it needs fixing is
>uncertain.
>
>I am personally replicating this issue on Gentoo X86_64 ( with a few gentoo
>specified patches ) , but it is apparent that other people are replicating
>the issue on OSX Homebrew. ( as per the ruby issue ).
>
>
>
>-- 
>Kent
>
>*KENTNL* - https://metacpan.org/author/KENTNL
>
>_______________________________________________
>openssl-dev mailing list
>To unsubscribe: https://mta.openssl.org/mailman/listinfo/openssl-dev

-------------- next part --------------
A non-text attachment was scrubbed...
Name: multiblock.patch
Type: application/octet-stream
Size: 489 bytes
Desc: not available
URL: <http://mta.openssl.org/pipermail/openssl-dev/attachments/20150308/93922a22/attachment.obj>