Now that, after the recent changes, the test passes, its old
name is no longer accurate.
While at it, enable the xml2xml part for it as well.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Due to the way the information is stored by the XML parser, we've
had this quirk where specifying any information about the loader
or NVRAM would implicitly set its format to raw. That is,
<nvram>/path/to/guest_VARS.fd</nvram>
would effectively be interpreted as
<nvram format='raw'>/path/to/guest_VARS.fd</nvram>
forcing the use of raw format firmware even when qcow2 format
would normally be preferred based on the ordering of firmware
descriptors. This behavior can be worked around in a number of
ways, but it's fairly unintuitive.
In order to remove this quirk, move the selection of the default
firmware format from the parser down to the individual drivers.
Most drivers only support raw firmware images, so they can
unconditionally set the format early and be done with it; the
QEMU driver, however, supports multiple formats and so in that
case we want this default to be applied as late as possible,
when we have already ruled out the possibility of using qcow2
formatted firmware images.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Keep things consistent by using the same file extension for the
generated NVRAM path as the NVRAM template.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
If the user included loader.readonly=no in the domain XML, we
should not pick a firmware build that expects to work with
loader.readonly=yes.
https://bugzilla.redhat.com/show_bug.cgi?id=2196178
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Right now, we only generate it after finding a matching entry
either among firmware descriptors or in the legacy firmware
list.
Even if the domain is configured to use a custom firmware build
that we know nothing about, however, we should still automatically
generate the NVRAM path instead of requiring the user to provide
it manually.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
libvirt doesn't really support the microvm machine type, but
it can parse the firmware descriptor just fine.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
These are imported from Fedora 38's edk2 package.
The files that are being replaced date back to RHEL 7 and no
longer represent what libvirt is likely to encounter on an
actual production system.
Notably, the paths have all changed, with both x86_64 and
aarch64 builds now living under /usr/share/edk2 and the AAVMF
name being having been phased out.
Additionally, the 4MB qcow2 format builds have been introduced
on x86_64 and given high priority, effectively making qcow2
the default format across architectures.
The impact of these changes on the test suite is, predictably,
quite severe.
For the cases where paths to firmware files were explicitly
provided as part of the input, they have been adjusted so that
the modern paths are used instead of the legacy ones. Other
than that, input files have been left untouched.
The following expected changes can be seen in output files:
* where qcow2 firmware was used on x86_64, Secure Boot
support is now enabled;
* all ABI_UPDATE test cases for x86_64 now use qcow2
formatted firmware;
* test cases where legacy paths were manually provided
no longer get additional information about the firmware
added to the output XML.
Some of the changes described above highlight why, in order
to guarantee a stable guest ABI over time and regardless of
changes to the host's configuration, it was necessary to move
firmware selection from VM startup time to VM creation time.
In a few cases, updating the firmware descriptors changes the
behavior in a way that's undesired and uncovers latent bugs
in libvirt:
* firmware-manual-efi-secboot-legacy-paths ends up with
Secure Boot disabled, despite the input XML specifically
requesting it to be enabled;
* firmware-manual-efi-rw-modern-paths loses the
loader.readonly=no part of the configuration and starts
using an NVRAM file;
* firmware-manual-efi-nvram-template-nonstandard starts
failing altogether with a fairly obscure error message.
We're going to address all these issues with upcoming changes.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Most of these are just additional coverage, but a few demonstrate
bugs in libvirt:
* firmware-manual-efi-nvram-template-nonstandard sees the NVRAM
template path, which was explicitly provided in the XML,
being overridden by the firmware selection machinery;
* firmware-auto-efi-rw* and firmware-manual-efi-rw-legacy-paths
lose the loader.readonly=no setting and thus behave
differently than requested;
* firmware-manual-efi-loader-path-nonstandard fails because an
NVRAM path doesn't get generated.
We're going to address all these issues with upcoming changes.
Note that the firmware-auto-efi-nvram-template-nonstandard
failure is expected: firmware autoselection has been enabled, but
the NVRAM template points to a custom path that's not mentioned
in any of the firmware descriptors and so it can't succeed.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Just like the more common split builds, these are of type
QEMU_FIRMWARE_DEVICE_FLASH; however, they have no associated
NVRAM template, so we can't access the corresponding structure
member unconditionally or we'll trigger a crash.
https://bugzilla.redhat.com/show_bug.cgi?id=2196178
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The documentation states that, just like the Modern() variant,
this function should return 1 if a match wasn't found. It
currently doesn't do that, and returns 0 instead.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
The new name better describes the test scenario and will fit
better with the additional tests that we're about to introduce.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Since the idea behind introducing the abi-update variant of
a test is showing that libvirt behaves differently based on
whether the configuration is for a newly-defined domain or an
existing one, we don't want the input files to ever go out of
sync.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
That's what we already use in almost all cases.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Using the unversioned machine type means that firmware
descriptors can't be used to discover additional information
about the chosen firmware build.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Firmware selection is not relevant to these tests, so adopt
the most convenient approach.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
We have a number of tests that can benefit from this macro
instead of open-coding it.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Update to v8.1.0-rc4
Notable changes:
- 'dirty-limit' migration feature added
- 'vcpu-dirty-limit', 'x-vcpu-dirty-limit-period' parameters added
- 'dirty-limit-ring-full-time', 'dirty-limit-throttle-time-per-round' statistics added
- migration statistic of number of skipped zero pages is now deprecated
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In mu previous commits I've moved internals of
qemuDomainChrDefDropDefaultPath() into a separate function
(qemuDomainChrMatchDefaultPath()) but forgot to remove @buf and
@regexp variables which are now unused.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
For historical reasons (i.e. unknown reason) we put channel
sockets into a path derived from cfg->libDir which is a path that
survives host reboots (e.g. /var/lib/libvirt/...). This is not
necessary and in fact for session daemon creates a longer prefix:
XDG_CONFIG_HOME -> /home/user/.config
XDG_RUNTIME_DIR -> /run/user/1000
Worse, if host is rebooted suddenly (e.g. due to power loss) then
we leave files behind and nobody will ever remove them.
Therefore, place the channel target dir into state dir.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2173980
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
A <channel/> device is basically an UNIX socket into guest.
Whatever is sent from the host, appears in the guest and vice
versa. But because of that, the length of the path to the socket
is important (underscored by fact that we derive the path from
domain short name). But there are still cases where we might not
fit into UNIX_PATH_MAX limit (usually 108 characters), because
the path is derived also from other variables, e.g.
XDG_CONFIG_HOME for session domains.
There are two components though, that are needless: "/target/"
and "domain-" prefix. Drop them. This is safe to do, because
running domains have their path saved in status XML and even
though paths are dropped on migration, they are not part of guest
ABI and thus we are free to change them.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
Now that we have a local OS target override for lcitool in place, we
can bump the cirrus FreeBSD image version in GitLab CI.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
We've reached a point in lcitool where we can't steer its development
based solely on libvirt's needs IOW there will be times where a local
override of value (e.g. package mapping) will be necessary - an example
of this would be QEMU.
In case of this particular patch we need to add an override for the
cirrus FreeBSD 13 image we request in our CI to fix:
/usr/local/lib/libtasn1.so.6: Undefined symbol "strverscmp@FBSD_1.7"
The reason why we can't/should not make the fix in upstream lcitool
just yet is that we store a libosinfo ID in lcitool's OS target YAML
configs and at the time of writing this patch libosinfo does not have
a corresponding entry/ID for FreeBSD 13.2 so we have to stick with 13.1
in lcitool until they do so.
For the time being, the fix can easily be done on libvirt side as does
this patch.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Let's move our Debian CI workloads to Debian-12 since it's the latest
release and mark Debian-11 jobs as optional.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
It is quite confusing seeing these two in a call like this one:
$ meson build $MESON_OPTS $MESON_ARGS
One has to ask 'how are they different' and 'shouldn't these be
merged'. In fact, these variables hold very different things and we
should make it more obvious. The problem is that renaming MESON_OPTS to
something more meaningful, like 'MESON_CROSS_OPTS' which is what
MESON_OPTS really does would require changes to lcitool and would
impact Dockerfile generation which in turn might have an impact on
other projects which rely on this lcitool functionality which is risky.
Instead, provide a docstring for the former to supplement the latter
and join the two variables in a single one MESON_ARGS which is then
passed to meson's command line so it's a little less confusing.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Although it is currently consistent with the other variables we define
when running ci in a local container environment, it isn't consistent
with the variable naming we use in GitLab recipes. Since the idea is
to unite the two, we're likely going to drop a few other variables from
the local env configuration anyway, hence this renaming.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
There's no harm in always building in system mode, i.e. setting the
right paths.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Even though 'setup' is assumed when no other command is given, we're
being explicit in our GitLab recipes, so do the same for the local
build.sh script too.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
build.sh is not the place where this should be mentioned as the
official entrypoint for this script locally is ci/helper which can
download the right image from our upstream CI registry. Since the idea
is to ultimately drop the usage of a Makefile for the local executions,
this patch doesn't provide an alternative place for the comment in
question as the functionality is going to be altered substantially in
the future.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
This is continuation of what commit b56e2be68e started. If we stick to
only calling meson commands directly, we can achieve much better
consistency in passing arguments to meson especially if we unify the
recipes run in gitlab CI and what we can currently run locally in
containers using docker/podman.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
We still get MRs in Gitlab from individual contributors on a regular
basis which in some ways just makes maintainer's or reviewer's life
just a bit more complicated. This ultimately means our guidelines are
probably not visible enough on the main page
(or some people wouldn't read them anyway). While this patch can't make
the problem go away, it can at least attempt to mitigate it by creating
a quick link to the 'hacking' page, skipping a lot of TL;DR contents
in contributing.rst which we link from the main page.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Kristina Hanicova <khanicov@redhat.com>
We're already past Fedora 35 and so all new fedora's default to
modular daemon setup.
Signed-off-by: Erik Skultety <eskultet@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Users need to enable non-shared-storage migration, otherwise the disks
specified via '--migrate-disks' will be ignored.
Add an error message to inform the users of their wrong config.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Convert the flags declaratively as in the vast majority of cases it's a
simple binary addition if the flag exists.
In one instance there was also an additional check, which was moved up
after the new code, and the error message was fixed.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
If a user passes a list of disks to migrate but don't actually use
'VIR_MIGRATE_NON_SHARED_DISK' or 'VIR_MIGRATE_NON_SHARED_INC' flags the
parameter would be simply ignored without informing the user of the
error.
Add a proper error in such case.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When VIR_MIGRATE_TUNNELLED is used without
VIR_MIGRATE_NON_SHARED_DISK/VIR_MIGRATE_NON_SHARED_INC
an error was reported without actually returning failure.
This was caused by a refactor which dropped many error paths.
Fixes: 6111b23522
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The '/dev' filesystem convenience directory for a LVM volume group is
not created when the volume group is empty.
The logic in 'virStorageBackendLogicalCheckPool' which is used to see
whether a pool is active was first checking presence of the directory,
which failed for an empty VG.
Since the second step is virStorageBackendLogicalMatchPoolSource which
is checking mapping between configured PVs and the VG, we can simply
rely on the function to also check presence of the pool.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2228223
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
In case of invalid placement its value should
be passed as a parameter of virReportError
instead of mode.
Fixes: 93e82727ec ("numatune: Encapsulate numatune configuration in order to unify results")
Signed-off-by: Anastasia Belova <abelova@astralinux.ru>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
When spawning a new container (via clone()) we allocate stack for
lxcContainerChild(). So far, we allocate 4 pages for the stack
and this used to be enough until we started rewriting everything
to glib. With glib we switched to g_strerror() which localizes
errno strings and thus increases stack usage, while the
previously used strerror_r() was more compact.
Fortunately, the solution is easy - just increase how much stack
the child can use (16 pages ought to be enough for anybody).
And while at it, lets use mmap() for allocation which offer some
nice features:
MAP_STACK - align allocation to be suitable for stack (even
though, currently ignored on Linux),
MAP_GROWSDOWN - kernel guards out of bounds access from child
Resolves: https://gitlab.com/libvirt/libvirt/-/issues/511
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
This fixes
commit 38abf9c34d
Author: Daniel P. Berrangé <berrange@redhat.com>
Date: Wed Jun 21 13:22:40 2023 +0100
src: set max open file limit to match systemd >= 240 defaults
The bug referenced in that commit had suggested to set
LimitNOFile=512000:1024
on the basis that matches current systemd default behaviour and is
compatible with old systemd. That was good except
* The setting is LimitNOFILE and these are case sensitive
* The hard and soft limits were inverted - soft must come
first and so it would have been ignored even if the
setting name was correct.
* The default hard limit is 524288 not 512000
Reported-by: Olaf Hering <olaf@aepfle.de>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
When a query for an interface via virInterfaceLookupByMACString finds
multiple interfaces an error is returned. Treat such error with the same
'debug' priority as we treat when the interface was not found to avoid
spamming logs with such configurations.
Closes: https://gitlab.com/libvirt/libvirt/-/issues/514
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
Since the error message originates from a log file it contains a
trailing newline. Strip it as all error handling adds it's own newline.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
The caller passes in a 1k buffer, which when debug logging is in use is
easily filled with debug messages only. Thus after the first pass which
is common if the controller process already terminated the buffer will
not contain the real error, but rather a truncated debug message,
which will result in an error such as:
error: internal error: guest failed to start: 2023-08-01 12:58:31.948+0000: 798195: i
instead of the proper error:
error: internal error: guest failed to start: Failure in libvirt_lxc startup: Failed to create /home/rootfs/.oldroot: Permission denied
To fix the above retry the reading loop if the filtering function made
space in the buffer.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
