Bug 1183780

Summary:	guestfs needs README.fixed to actually work
Product:	[Community] Virtualization Tools	Reporter:	Daniel Augustin <hossie>
Component:	libguestfs	Assignee:	Richard W.M. Jones <rjones>
Status:	CLOSED CURRENTRELEASE	QA Contact:
Severity:	unspecified	Docs Contact:
Priority:	unspecified
Version:	unspecified	CC:	hossie, mbooth, ptoscano, rbalakri
Target Milestone:	---
Target Release:	---
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:		Doc Type:	Bug Fix
Doc Text:		Story Points:	---
Clone Of:		Environment:
Last Closed:	2015-05-28 07:15:27 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:

Description Daniel Augustin 2015-01-19 19:19:10 UTC

Description of problem:
When having an appliance without the README.fixed file (just root, kernel, initrd) all libguestfs commands fail


Version-Release number of selected component (if applicable):
1.28.5
Appliance: 1.28.1

How reproducible:
always

Steps to Reproduce:
 # ls -l
total 335452
-rw-r--r-- 1 root root    1208832 Oct 18 13:21 initrd
-rw-r--r-- 1 root root    6263432 Oct 18 13:21 kernel
-rw-r--r-- 1 root root        978 Jan 19 19:30 README.fixed
-rw-r--r-- 1 qemu qemu 4294967296 Oct 18 13:21 root
 # echo $LIBGUESTFS_PATH 
/usr/share/guestfs/appliance/
 # pwd
/usr/share/guestfs/appliance

-> everythings fine:
[...]
===== TEST FINISHED OK =====

Delete the README.fixed (like the gentoo ebuild installs it, under /usr/share/doc where it imho belongs):

libguestfs-test-tool:
error: cannot find any suitable libguestfs supermin, fixed or old-style appliance on LIBGUESTFS_PATH (search path: /usr/share/guestfs/appliance/)

For some reason, libguestfs now looks for initramfs.x86_64.img and vmlinuz.x86_64 (I straced it)

When creating two symlinks for those, everything goes completly mad, as it gets totally confused.
Output libguestfs-test-tool then: http://pastebin.com/Gv1gRMrs

OS here is Gentoo with the ebuilds from the tree, but I don't think this is a gentoo specific issue as depending on that readme file in the guestfs-appliance directory seems strange.

Comment 1 Richard W.M. Jones 2015-01-19 19:28:41 UTC

Ah, so this is the Gentoo problem?

Indeed, libguestfs.git/src/appliance.c contains this function:

static int
contains_fixed_appliance (guestfs_h *g, const char *path, void *data)
{
  return dir_contains_files (path,
                             "README.fixed",
                             "kernel", "initrd", "root", NULL);
}

All of those files (including README.fixed) need to exist
on the $LIBGUESTFS_PATH for libguestfs to recognize that the
directory contains a fixed appliance.

> For some reason, libguestfs now looks for initramfs.x86_64.img
> and vmlinuz.x86_64 (I straced it)

That's because if you delete README.fixed, then libguestfs looks
for what is called an "old style" appliance.  This is totally
obsolete, and in any case won't work with the fixed appliance
that we supply on http://libguestfs.org/download/binaries/appliance
So don't use this.

Anyhow, I'm not really sure what the bug is here.  README.fixed
is part of the fixed appliance, so don't delete it.  If you delete
it, you'll have trouble as you've shown :-)

Comment 2 Richard W.M. Jones 2015-01-19 19:29:46 UTC

Convoluted upstream code here ...

https://github.com/libguestfs/libguestfs/blob/master/src/appliance.c

Comment 3 Daniel Augustin 2015-01-19 19:35:23 UTC

For me, depending on a readme file to detect the appliance format is just wrong. For me, the gentoo ebuild does the right thing to put a readme into /usr/share/doc.

It should imho either have a better detection logic (like actually looking at the files, like the header or something) or not caring about README.fixed; at least it should not require it's existence.

Comment 4 Richard W.M. Jones 2015-01-19 19:40:19 UTC

If we drop or change README.fixed then we end up breaking the
existing appliances, which isn't nice.

I see a few solutions here:

Gentoo could carry this simple non-upstream patch:

 static int
 contains_fixed_appliance (guestfs_h *g, const char *path, void *data)
 {
   return dir_contains_files (path,
-                             "README.fixed",
                              "kernel", "initrd", "root", NULL);
 }

Or Gentoo could not move the README.fixed file.

Or we could work on something upstream (not sure exactly what -- I
don't want to introduce yet another "type" of appliance, we have
too many of them already).

Or Gentoo could build a supermin appliance, but I appreciate that
is likely to be difficult since supermin does make a lot of assumptions
about binary packages which certainly don't apply to Gentoo.

Comment 5 Richard W.M. Jones 2015-05-28 07:15:27 UTC

Since the Gentoo bug was resolved (by not deleting README.fixed),
I am closing this bug.