Hi, replying to IRC:
> <pitti> chris_se: I think the setup should go into adt-virt-qemu, not
> setup-testbed
> <pitti> setup-testbed is used for lxc or openstack instances too
> <pitti> and it's not a strict requirement to run it
> <pitti> chris_se: but I'm fine with mopping this up myself
> <pitti> chris_se: i. e. install the rule and call udevadm trigger to run it
> <pitti> so that it works without rebooting
So yes, I think you're right that it's better for adt-virt-qemu to
configure this, but I think we can even do it better:
- DON'T add the drive to the qemu command
- create the rule
- udevadm control -R (because otherwise we might fall in the 3s
window where udev doesn't reload the config)
- update-initramfs -k all -u (to support reboots)
- use qemu's monitor to add the drive
This way, the wrong symlinks will never be created (because the
drive won't exist at initial boot without the udev rule) and we can
guarantee that this doesn't cause any weird problems.
I've created a patch that does just that and tested it: with
setup-testbed from current git master it properly installs the udev
rule, *then* adds the drive, and we have:
- /dev/baseimage exists
- /dev/disk/by-{part,}uuid/* don't point to the baseimage disk.
Patch is attached.
Would still like to hear a comment about the CPU issue.
Regards,
Christian
From e2aaf0c2d0386d62bc9929f0b84e00cdc706c8f3 Mon Sep 17 00:00:00 2001
From: Christian Seiler <christ...@iwakd.de>
Date: Thu, 3 Mar 2016 22:10:27 +0100
Subject: [PATCH] adt-virt-qemu: Implement support for nested base images.
This adds initial support for nested base images to be passed into the
test environment, so that nested VMs may be used in tests.
A read-only copy of the first image without the overlay is passed to
the VM with a hardware serial number BASEIMAGE. adt-virt-qemu installs
udev rules that create a symbolic link /dev/baseimage for that drive
the first time the testbed is booted. Also, the symlink priority for
that drive is lowered, because the same file system UUIDs will be
present on both the first drive and the readonly baseimage drive.
Closes: #800845
---
debian/changelog | 4 ++++
virt-subproc/adt-virt-qemu | 25 +++++++++++++++++++++++++
virt-subproc/adt-virt-qemu.1 | 5 +++++
3 files changed, 34 insertions(+)
diff --git a/debian/changelog b/debian/changelog
index 36ee620..85a5571 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -1,5 +1,6 @@
autopkgtest (3.19.4) UNRELEASED; urgency=medium
+ [ Martin Pitt ]
* setup-commands/setup-testbed: Ensure that removing cruft does not remove
cloud-init. (LP: #1539126)
* setup-commands/setup-testbed: Purge lxd and lxc.
@@ -14,6 +15,9 @@ autopkgtest (3.19.4) UNRELEASED; urgency=medium
lxc-start-ephemeral got deprecated by that. This now supports reboots in
ephemeral mode.
+ [ Christian Seiler ]
+ * adt-virt-qemu: Implement support for nested base images. (Closes: #800845)
+
-- Martin Pitt <mp...@debian.org> Tue, 23 Feb 2016 18:21:51 +0100
autopkgtest (3.19.3) unstable; urgency=medium
diff --git a/virt-subproc/adt-virt-qemu b/virt-subproc/adt-virt-qemu
index 965e3e8..d676708 100755
--- a/virt-subproc/adt-virt-qemu
+++ b/virt-subproc/adt-virt-qemu
@@ -175,6 +175,30 @@ def login_tty_and_setup_shell():
term.send(b'\nexit\n')
VirtSubproc.expect(term, b'\nlogout', 10)
+def setup_baseimage():
+ '''setup /dev/baseimage in VM'''
+
+ term = VirtSubproc.get_unix_socket(os.path.join(workdir, 'ttyS1'))
+
+ # Setup udev rules for /dev/baseimage; set link_priority to -1024 so
+ # that the duplicate UUIDs of the partitions will have no effect.
+ term.send(b'''mkdir -p -m 0755 /etc/udev/rules.d ; printf '# Created by adt-virt-qemu\\n%s\\n%s\\n' 'KERNEL=="vd*[!0-9]", ENV{ID_SERIAL}=="BASEIMAGE", OPTIONS+="link_priority=-1024", SYMLINK+="baseimage", MODE="0664"' 'KERNEL=="vd*[0-9]", ENV{ID_SERIAL}=="BASEIMAGE", OPTIONS+="link_priority=-1024"' > /etc/udev/rules.d/61-baseimage.rules\n''')
+ VirtSubproc.expect(term, b'#', 10)
+ # Reload udev to make sure the rules take effect (udev only auto-
+ # rereads rules every 3 seconds)
+ term.send(b'udevadm control -R\n')
+ # Update the initramfs to include the new udev rules (to support
+ # reboots properly)
+ term.send(b'update-initramfs -k all -u\n')
+ VirtSubproc.expect(term, b'#', 60)
+
+ monitor = VirtSubproc.get_unix_socket(os.path.join(workdir, 'monitor'))
+
+ # Add the base image as an additional drive
+ monitor.send(('drive_add 0 file=%s,if=none,readonly=on,serial=BASEIMAGE,id=drive-baseimage\n' % args.image[0]).encode())
+ VirtSubproc.expect(monitor, b'(qemu)', 10)
+ monitor.send(b'device_add virtio-blk-pci,drive=drive-baseimage,id=virtio-baseimage\n')
+ VirtSubproc.expect(monitor, b'(qemu)', 10)
def setup_shared(shared_dir):
'''Set up shared dir'''
@@ -501,6 +525,7 @@ def hook_open():
# files; let QEMU run with the deleted inode
os.unlink(overlay)
setup_shell()
+ setup_baseimage()
setup_shared(shareddir)
setup_config(shareddir)
make_auxverb(shareddir)
diff --git a/virt-subproc/adt-virt-qemu.1 b/virt-subproc/adt-virt-qemu.1
index 81bb11f..2de5e3e 100644
--- a/virt-subproc/adt-virt-qemu.1
+++ b/virt-subproc/adt-virt-qemu.1
@@ -26,6 +26,11 @@ does
the given images, but will instead create a temporary overlay for the
primary image, and add all other images as read-only.
+The first image without the overlay is always added as an additional
+read-only hard drive, which will be available for tests as
+.IR /dev/baseimage .
+This allows for tests that require nested VMs to reuse the same image.
+
.SH REQUIREMENTS
.B adt-virt-qemu
assumes that you have already prepared a suitable Debian based QEMU image (see
--
2.1.4
signature.asc
Description: OpenPGP digital signature
