[SOLVED] What does 'Exec format error' mean?

RandomTroll · 12-19-2014, 09:32 PM

I see that 'Exec format error' is in the kernel, ENOEXEC.

What does it mean? What causes it?

rknichols · 12-19-2014, 09:47 PM

It's from trying to execute a file that has a magic number that the kernel doesn't know how to handle. Typical examples are a 64-bit binary on a 32-bit kernel, a file intended for a different OS (Windows, Mac OS, etc.). Also, interpreter scripts that lack the "#!/bin/whatever" header will cause that response from the kernel, but a shell that sees that error may try a fallback of trying the file as a script in its own language, sometimes with bizarre results if it is not.

RandomTroll · 12-20-2014, 12:21 AM

How do I know a file's magic number?

Are there any other circumstances?

Beginning with kernel 3.10 the fsck.ext4 in my distribution's package returns that error at boot time when /etc/rc.d/rc.S tries to run it, but it works afterwards, even during the maintenance session linux lets me run on that boot; if I create /etc/fastboot (which prevents file-system-checking) I can boot and run fsck.ext4. I build my own e2fsprogs with the same version and it works at boot. How can I figure out the difference?

rknichols · 12-20-2014, 12:21 PM

What does the "file" command report for each executable (e.g., "file /sbin/fsck.ext4")? What distribution are you running? What is the output from "uname -rm"?

RandomTroll · 12-21-2014, 12:01 AM

uname -rm returns

Quote:

3.18.1 i686

'file' on the fsck.ext4 I build returns

Quote:

/sbin/fsck.ext4: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked (uses shared libs), not stripped

fsck.ext4 from the package is script that calls e2fsck, a 'file' of which returns the same thing. I just noticed that the fsck.ext4 I build is identical to e2fsck I build; I suspect that means that the problem is the script.

Code:

#!/bin/sh
exec /sbin/e2fsck -C 0 $*

/bin/sh is a link to /bin/bash

rknichols · 12-21-2014, 05:07 PM

That script looks fine. I presume this is Slackware.

Does the error message mention the name of the file that caused the error? The only thing that comes to mind is that e2fsck might depend on some kernel modules that are not loaded until later in the /etc/rc.d/rc.S script, but I don't think that would cause an ENOEXEC error. Beyond that, I have no idea.

RandomTroll · 12-22-2014, 12:50 AM

Quote:

Originally Posted by rknichols

That script looks fine.

Looked good to me; works after init.

Quote:

Originally Posted by rknichols

I presume this is Slackware.

My identifying information in the left column reveals this.

Quote:

Originally Posted by rknichols

Does the error message mention the name of the file that caused the error?

fsck.ext4

Quote:

Originally Posted by rknichols

The only thing that comes to mind is that e2fsck might depend on some kernel modules that are not loaded until later in the /etc/rc.d/rc.S script, but I don't think that would cause an ENOEXEC error.

Absent libraries have always returned a different error for me. I suppose a library e2fsck called could be unexecutable. The package version has more dependencies.

Quote:

Originally Posted by rknichols

Beyond that, I have no idea.

I'm puzzled too. Especially since no one else seems to have the same problem. And it's been going on for more than a year.

rknichols · 12-22-2014, 10:14 AM

Quote:

Originally Posted by RandomTroll

My identifying information in the left column reveals this.

I've learned not to put total trust in that. People change distros without updating their profile info, or try distros other than the one they are currently running.

Quote:

Absent libraries have always returned a different error for me.

Not libraries, kernel modules. If you look at rc.S you'll see that modprobe is run after the file system checks. Is this problem just with the "generic" kernel, or do you see it with "huge" also? (The generic kernel can't even find the root filesystem on my hardware, so I can't test that without building an initrd.)

RandomTroll · 12-22-2014, 08:00 PM

Quote:

Originally Posted by rknichols

Not libraries, kernel modules.

Whoops! Sorry. Absent modules have never given me ENOEXEC errors either; a module being un-executable could though.

Quote:

Originally Posted by rknichols

Is this problem just with the "generic" kernel, or do you see it with "huge" also?

I build the kernel. I make everything possible a module and attempt not-to-build everything I don't need.

rknichols · 12-22-2014, 08:15 PM

If you have CONF_BINFMT_SCRIPT as a module and don't load it in an initramfs, then the exec() of that fsck.ext4 script will fail with ENOEXEC. Once the modules have been loaded (near the end of /etc/rc.d/rc.S), it will work fine.

RandomTroll · 12-23-2014, 12:14 AM

Quote:

Originally Posted by rknichols

If you have CONF_BINFMT_SCRIPT as a module and don't load it in an initramfs, then the exec() of that fsck.ext4 script will fail with ENOEXEC. Once the modules have been loaded (near the end of /etc/rc.d/rc.S), it will work fine.

Hooray! You found it. CONFIG_BINFMT_SCRIPT (note different spelling) appeared for the first time in kernel 3.10, the point at which this problem started to happen. So the 'cure' is to replace trying to run the script with executing e2fsck directly. This parameter is poorly-documented.