diff options
Diffstat (limited to 'bugs/efault.html')
| -rw-r--r-- | bugs/efault.html | 44 |
1 files changed, 44 insertions, 0 deletions
diff --git a/bugs/efault.html b/bugs/efault.html new file mode 100644 index 0000000..06be3c1 --- /dev/null +++ b/bugs/efault.html @@ -0,0 +1,44 @@ +<h2>The impossible EFAULT</h2> +<p>I have written a program; suppose it's called <code>worker</code>. +(While the program is written in Haskell, I don't think that's particularly relevant to this post.)</p> +<p>When run, <code>worker</code> starts a bunch of copies of a script. +Under normal circumstances this script sets up a container using Linux cgroups and Linux user namespaces, but none of that is relevant because the strange behaviour in question occurs just fine without all of that -- in fact, we'll let it start the following script, say <code>./sleep.sh</code>:</p> +<pre><code class="language-bash">#!/bin/bash +sleep 10 +</code></pre> +<p>Clearly, there is no weird behaviour here, assuming that the system has <code>bash</code> under <code>/bin</code>, and mine does.</p> +<p>The copies of <code>sleep.sh</code> are started by passing <code>./sleep.sh</code> to <code>posix_spawnp(3)</code>. +(The Haskell <code>process</code> library does this for me.) +The thing is, occasionally (once every 5 to 10 invocations of <code>./worker</code>, approximately), <code>posix_spawnp</code> returns <code>EFAULT</code> ("Bad Address"). +The manpage for <code>posix_spawnp</code> says that:</p> +<blockquote> +<p><strong>ERRORS</strong></p> +<p>The posix_spawn() and posix_spawnp() functions fail only in the case where the underlying fork(2), vfork(2) or clone(2) call fails; in these cases, these functions return an error number, which will be one of the errors described for fork(2), vfork(2) or clone(2).</p> +<p>In addition, these functions fail if:</p> +<p><strong>ENOSYS</strong> Function not supported on this system.</p> +</blockquote> +<p>Okay, so I should look for <code>EFAULT</code> in <code>fork(2)</code>, <code>vfork(2)</code> and <code>clone(2)</code> to figure out what goes wrong, right? +Wrong. +Or, in any case, none of those manpages mention <code>EFAULT</code>. +I've looked through the source code of <code>posix_spawnp</code> in glibc and it at least doesn't throw <code>EFAULT</code> directly; presumably, one of the subroutines it calls does. +glibc is large and I don't think looking through the entire call tree will be very productive, so I tried to diagnose the issue from the outside instead.</p> +<p>And this is where the weirdness starts. +Whenever my program encounters <code>EFAULT</code> from <code>posix_spawnp</code>, it prints <code>Oops EFAULT</code>; hence grepping for <code>EFAULT</code> gives output precisely if the error occurred in this run. +I get the following observations:</p> +<ul> +<li><code>./worker 2>&1 | grep EFAULT</code>: errors occur.</li> +<li><code>./worker 2>&1 | grep EFAULT | cat</code>: errors DO NOT occur.</li> +<li><code>./worker 2>&1 | grep --line-buffered EFAULT | cat</code>: errors occur.</li> +<li><code>./worker 2>&1 | grep --line-buffered EFAULT</code>: errors occur.</li> +</ul> +<p>("errors occur" means that once every few executions I get output indicating that <code>EFAULT</code> occurred; in the negative case I've run it for >20x the number of invocations that are necessary to produce <code>EFAULT</code> in the other cases, without any <code>EFAULT</code>.)</p> +<p>The only situation in which <code>posix_spawnp</code> seems to always succeed, is when <code>stdout</code> of the process that <code>worker</code>'s output is piped to, is block-buffered. +But this makes no sense: there shouldn't even be a reasonable way in which <code>worker</code> <em>can</em> even determine whether this is the case! +Surely it can distinguish between <code>./worker | cat</code> and <code>./worker</code> (using <code>isatty(3)</code> -- this is precisely what <code>grep</code> does when not passed <code>--line-buffered</code>), but in all of the above cases the output is piped to another process anyway.</p> +<p>This is already spooky, but it gets even spookier: if I replace the invocation of <code>./sleep.sh</code> by an invocation of <code>sleep</code> (i.e. removing the indirection of the shell script), errors occur in none of the above setups. +Somehow, starting a script is different from starting a native process (and changing <code>bash</code> to <code>dash</code> in <code>sleep.sh</code> doesn't change anything). +<code>posix_spawnp</code> shouldn't care what it is starting! +That's the job of the loader, as far as I know. +So what gives?</p> +<p>I'll try to reduce my own program to a minimal reproducer, and if I find anything I'll post an update to this post. +In the meantime, spookiness.</p> |