In some cases when the host has, for whatever reason, multiple copies of
the same repo listed the build may fail with an error about running out
of space.
So this commit removes duplicate entries after the host's repos have been
loaded. It also adjusts some of the test repos to use different
temporary repo names for the tests.
(cherry picked from commit 98482e444d)
If systemd's tmpfiles.d timer is executed while lorax is running it will
remove any files and directories older than 30 days. This is what has
been causing the occasional error where /proc/ would seem to vanish
during the install.
Upstream has proposed this solution, https://github.com/systemd/systemd/pull/11482
but until that is released we need a work-around to protect the lorax
files.
This commit does several things:
* Move the default tmpdir from /var/tmp/ to /var/tmp/lorax/
* Add a lorax.conf tmpfiles.d file that prevents systemd-tmpfiles from
removing anything under /var/tmp/lorax/
* Add an exit handler to lorax so that temporary directories are removed on
exit or on a python traceback.
* Use flock to lock access to the tempdir while lorax is running.
* Remove any unlocked tempdirs named /var/tmp/lorax/lorax.* at startup
Note that the exit handler will not remove the tempdir if lorax is
killed with a signal -- those are being caught by dnf and prevent the
exit handler from running.
systemd-tmpfiles cannot clean up the tempdirs at boot time because they
contain files labeled as shadow_t, so we have to remove those when lorax
runs. It uses the flock to prevent removing any directories created by
parallel instances of lorax and only removes ones that are unlocked.
Worst case they will be around until the first run of lorax after a
reboot.
If you want to keep the working directory around for debugging purposes
use --workdir /var/tmp/lorax/my-workdir and it won't be removed by
lorax.
(cherry picked from commit e4fe1aab32)
After a novirt disk image install, we run `setfiles` in the
install root to ensure some SELinux contexts are correct. /dev
is currently excluded from this run. However, as reported and
discussed in https://bugzilla.redhat.com/show_bug.cgi?id=1663040
it seems that with a recent systemd change, startup of many
services will fail if /dev itself is incorrectly labelled, and
in current Rawhide live images, it *is* incorrectly labelled.
Including `/dev` in this setfiles command appears to resolve the
problem in my testing.
Resolves: rhbz#1663040
Signed-off-by: Adam Williamson <awilliam@redhat.com>
(cherry picked from commit 2d3f266373)
Support both
[customizations]
hostname = "whatever"
and
[[customizations]]
hostname = "whatever"
in the blueprint data. The [[ syntax matches the other customization
directives (user, group, sshkey), and as such it's easy to accidentally
use it for the hostname without even realizing it's specifying something
different.
Add some tests for converting customizations to kickstarts.
(cherry picked from commit 35ab6a1336)
Drop running pkill. This causes problems if more than one is running on
a system (eg. in parallel using mock). It can kill off other processes
unrelated to this instance of anaconda.
This reverts commit e3e892ea43.
Run df on the filesystem image after it has been created.
Output will be in program.log, eg:
Running... df /var/tmp/lorax.imgutils.wm04pg_v
Filesystem 1K-blocks Used Available Use% Mounted on
/dev/loop0 1998672 1619508 362780 82% /var/tmp/lorax.imgutils.wm04pg_v
Return code: 0
It ends up that this isn't as easy as you'd think. Anaconda sets up some
signal handlers to handle cleanly exiting, but they are not being run
when sent a TERM after package installation has started. I think DNF
resets them causing it to get ignored.
When the cancel is sent it can take several minutes for it to have an
effect. In my testing it usually takes around 2 minutes for anaconda to
notice and exit.
This sends a TERM to the process and then waits for it to exit. When it
returns it then removed any device-mapper devices that were setup for
image installations, removes any hanging loop devices.
It then kills off any process with pyanaconda. in the cmdline, and
anaconda-bus.conf (because anaconda starts a bunch of helpers and if it
doesn't shut down cleanly they remain running).
Resolves: rhbz#1656691
In addition to monitoring the logs for errors, call a function (or
functions) that tell it to cancel the anaconda process and cleanup.
Also check for a cancel after creating the squashfs image for live-iso
since that's a long running process.
This required adding a new argument to a number of existing functions,
passing it down to QEMUInstall and novirt_install where the function is
called.
Resolves: rhbz#1656691
When there is no run or new symlink do one last check to make sure no
STATUS file was written. If it is missing, go ahead and remove the
results directory.
Related: rhbz#1656691
If another CANCEL request has already been made just exit from
uuid_cancel. If the build is FINISHED before it times out just exit,
don't remove the finished results.
Related: rhbz#1656691
Images don't work without these fixes:
* Enable Network Manager.
* Disable cloud-init.
* Add Hyper-V modules into initramfs.
Fixes specific for RHEL:
* Create ifcfg-eth0 required by waagent.
* Install python3 and net-tools required by waagent.
Recommended changes:
* Use recommended kernel boot args.
* Disable kdump.
When the repository has multiple arches, eg. i686 and x86_64, it should
add a new entry to the project's builds list, not create a new project
in the list.
This handles that by adding a modified insort_left function and
examining the packages returned from dnf to make sure they aren't
already listed in the results. It also handles adding them in sorted
order so that no further sorting needs to be done on the results.
Resolves: rhbz#1656642
(cherry picked from commit d18934775c)
If the system ran out of space, or was rebooted unexpectedly, the state
of the queue symlinks, or the results STATUS files may be inconsistent.
This checks them and:
* Removes broken symlinks from queue/new and queue/run
* Removes symlinks from run and sets the build to FAILED
* Sets builds w/o a STATUS to FAILED
* Sets builds with STATUS of RUNNING to FAILED
* Creates missing queue/new symlinks to results with STATUS of WAITING
So, any builds that were running during the reboot will be FAILED, and
any that were waiting to be started will be started upon rebooting.
Resolves: rhbz#1647985
(cherry picked from commit 4dd9004d13)
This is required to ensure that SELinux is configured properly while
building. It fixes the problem with building tar, and should be
installed in the other image types for consistency.
Resolves: rhbz#1645189
(cherry picked from commit 99d867db65)
SELinux applies the correct labels, setfiles is no longer needed.
This allows lorax to run with SELinux in Enforcing mode.
(cherry picked from commit 4a4a415f88)
Anaconda, Lorax, lorax-composer, and livemedia-creator can all now run
with SELinux in Enforcing mode. It does not need to be disabled and if
there are denials they should be reported as a bug.
Log the current state of SELinux when starting, update the
documentation.
(cherry picked from commit 080705e8e6)
Running lorax-composer --no-system-repos will prevent it from copying
the dnf repositories from /etc/yum.repos.d/ into the lorax-composer repo
directory. It will *only* use repositories setup using the sources api
or written to /var/lib/lorax/composer/repos.d/
If lorax-composer has previously been run without this switch the system
repos will need to be removed from the composer/repos.d/ directory. It
would also be a good idea to remove the cached metadata in
/var/tmp/composer/
Resolves: rhbz#1650363
(cherry picked from commit 43ff505804)
When using package name globs and multiple repos dnf doesn't choose the
highest NEVRA as you would expect, work around this by applying max() to
the duplicate package names in installpkg.
Resolves: rhbz#1548586
Depending on how lorax-composer is run setting up an empty blueprints
directory can fail. So this moves checking/creation until after the
other directories are created and uses make_owned_dir to make sure
ownership is correct.
It needs to be root in order to set the ownership and permissions on the
directories that are under /var/lib/lorax/composer/
Refactor the directory creation into a utility function, and use a umask
of 0o006 to ensure that the parent directories created do not have o+rw
set on them (makedirs behavior is different between Python 3.6 and 3.7
so umask of 0 doesn't work consistently).
Since these images can be used to create multiple machines, they should
not have a unique machine-id attached to them. Replace /etc/machine-id
with an empty file so that it will be regenerated at boot time.
If a package is in multiple repos dnf may return more than 1 of them
when using best...glob so we pick the highest NEVRA one and install
that.
Related: rhbz#1636239
At the end of disk image installs, use fstrim on the generated filesystem to
discard any blocks that were allocated during the install and are now unused.
This will allow tools such as qemu-img to create images that do not include
deleted data.
For raw disk images that do not go through qemu-img, use fallocate --dig-holes
to create sparse holes in place of the unused blocks.
