<html>
<head>
<meta http-equiv="content-type" content="text/html; charset=utf-8">
</head>
<body bgcolor="#FFFFFF" text="#000000">
<p>Hi, <br>
</p>
<p>In OpenMPI 2.0, Michael Banck discovered that some of the
packages that failed were doing so as openmpi no longer works
under fakeroot.</p>
<p>Upstream (Gilles Gouaillaerdet) point out:</p>
<blockquote>
<p>from <code>validate_cred</code> in <code>opal/mca/pmix/pmix112/pmix/src/sec/pmix_native.c</code></p>
<p><code>getsockopt (peer->sd, SOL_SOCKET, SO_PEERCRED,
&ucred, &crlen);</code></p>
<p>my understanding is that <code>fakeroot</code> tricks the
processes into believing they are ran as <code>root</code>,<br>
as such, <code>orted</code> expects a connection from <code>root</code>
(e.g. <code>ucred.uid</code> is <code>0</code>)<br>
/* in singleton mode, an <code>orted</code> is fork&exec'ed
under the hood */</p>
but <code>fakeroot</code> cannot trick <code>getsockopt</code>,
which returns the user (non zero)<br>
</blockquote>
So its a legitimate tightening of credential-checking in pmix.
However it causes problems for us running tests under binary-arch.<br>
<br>
What do DDs think should be done about this - move all tests outside
binary-arch?<br>
<br>
regards<br>
Alastair<br>
<br>
<pre class="moz-signature" cols="72">--
Alastair McKinstry, <a class="moz-txt-link-rfc2396E" href="mailto:alastair@sceal.ie"><alastair@sceal.ie></a>, <a class="moz-txt-link-rfc2396E" href="mailto:mckinstry@debian.org"><mckinstry@debian.org></a>, <a class="moz-txt-link-freetext" href="https://diaspora.sceal.ie/u/amckinstry">https://diaspora.sceal.ie/u/amckinstry</a>
Misentropy: doubting that the Universe is becoming more disordered. </pre>
</body>
</html>