qemuBuildTPMOpenBackendFDs: Construct 'cancel_path' internally

Since 'cancel_path' is constructed from the 'tpmdev' argument, we can
push it down into the function opening the FDs.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

virTPMCreateCancelPath: Refactor value returning

Automatically free 'path' inside the loop which fills it and return the
values directly.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: Honour 'virtlogd' use when hotplugging chardevs

Setup the chardev similarly to how we do it on startup so that virtlogd
is properly used with chardevs which are hotplugged to a VM.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: process: Add a hotplug version of qemuProcessPrepareHostBackendChardev

When hotplugging a chardev we need the same form of setup for the
character device. Export a version which takes a 'virDomainDeviceDef'.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuMonitorJSONTestAttachChardev: Add tests for FD passing of file backend

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

tests: Move testPrepareHostBackendChardevOne into test utils

Move the function doing the fake setup of chardev backend for FD passing
into the collection of qemu test helpers so that it can be used in
qemumonitorjsontest.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuMonitorJSONAttachCharDevGetProps: Properly handle private data

FD passing and TLS is normally setup via private data for the chardev
source. The monitor implementation didn't support it.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuMonitorJSONTestAttachChardev: Add logfile to some tests

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuMonitorJSONTestAttachChardev: Add test for TLS-secured TCP chardev

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemumonitorjsontest: Refactor chardev hotplug testing

The main objective of this patch is to use a proper instance of
virDomainChrSourceDef allocated with the private data.

To achieve this the test cases are grouped into blocks by how much they
fill in the chardev definition. Some test cases are moved around so
that the resulting sequence doesn't need extra clearing.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemumonitorjsontest: chardev: Remove need to allow unused commands

Don't add the command to the test monitor when we don't expect to invoke
it rather than bypassing the test monitor.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuDomainRemoveChrDevice: Detach fdset after chardev hot-unplug

Our code uses fdsets for the pipe passed from virtlogd to qemu, but the
chardev hot-unplug code neglected to detach the fdset after the chardev
was removed. This kept the FDs open by qemu even after they were not
used any more.

After the refactor to use qemuFDPass for chardevs we now configure the
'opaque' field for fdsets used for chardevs so we can use
qemuHotplugRemoveFDSet to remove the unused fdset.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: Rewrite chardev startup code to use qemuFDPass

Rewrite the parts which already pass FDs via fdset or directly to use
the new infrastructure.

Apart from simpler code this also adds the appropriate names to the fds
in the fdsets which will allow us to properly remove the fdsets won
hot-unplug of chardevs, which we didn't do for now and resulted in
leaking the FDs.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuBuildInterfaceCommandLine: Use new pattern for naming the VDPA fdset

Prefix the file descriptor name with the alias of the network device so
that it's similar to other upcoming use.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

QEMUHotplugRemoveFDSet: Prepare for proper FD unplug handling

For now we have only one code path ('vdpa' interface) which actually
cleans up the fdset after it's done, but there are more device types
using fdsets.

In order to unify the handling of fdsets the removal code will now be
able to remove fdsets based on a prefix of the 'opaque' field, which
we'll always prefix with a device alias or e.g. node name once fdsets
are also used for disk backing.

To keep compatibility with old QEMUs, retain the possibility for the
VDPA interface to use the path.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: hotplug: Extract code for unplugging fdsets

The code unplugging the fdset for a 'vdpa' network device can be later
reused. Extract it into 'qemuHotplugRemoveFDSet'.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuBuildInterfaceCommandLine: Use qemuFDPass for the vdpa fd

Use the new helpers for passing of the file descriptor needed for 'vdpa'
interfaces.

Apart from the simplification in this case it will allow further changes
to unify all fdset handling.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: Introduce helper functions for passing FDs to qemu

The existing helpers we have are very clumsy and there's no integration
with the monitor.

This patch introduces new helpers to bridge the gap and simplify handing
of fdsets and classic FD passing when generating commandline/hotplug
arguments.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: domain: Add helper for generating 'fdset' ids for VM startup

When starting a VM we must assign unique IDs for fdsets we add via
'-add-fd'. For now it was done by using the index of the filedescriptor
passed to the virCommand. That approach is not very flexible, because
you need to have already passed the 'fd' to virCommand before generating
the fdset path, and also won't nicely work with fdsets containing two or
more fds.

This patch introduces a counter into the private data of a qemu domain
so that we can allocate unique ids without relying on virCommand.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: monitor: Make 'id' in 'struct _qemuMonitorFdsetInfo' unsigned

Similarly to the 'qemuMonitorRemoveFdset', it doesn't make sense
to store it as signed when only unsigned values are expected.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuMonitorRemoveFdset: Convert @fdset to unsigned int to avoid error

'qemuMonitorRemoveFdset' validates that the 'fdset' argument isn't less
than 0. We can turn it to unsigned and thus avoid the error message
completely.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuMonitorJSONQueryFdsetsParse: Don't check value passed to g_strdup

'g_strdup()' is NULL-tolerant.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuProcessPrepareHostBackendChardev: Drop unneeded arguments

Caller passes 'driver->securityManager', and 'priv->qemuCaps' as
arguments along with 'vm', but both aforementioned objects are
accessible directly from 'vm'.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuProcessPrepareHostBackendChardevFileHelper: Always use FD passing

Code paths which don't wish to use FD passing are supposed to not call
the function which sets up the chardev for FD passing.

This is ensured by calling it only in the host prepare step.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

scripts/mock-noinline: Use full name of the required annotation in error message

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: domain: Change 'Id' to 'ID' in qemuDomainStorageIdNew/Reset

In a patch adding similarly named APIs I was asked to use 'ID' instead
of 'Id'. Since the code is being put together fix
qemuDomainStorageIdNew/Reset first.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: domain: Move and unexport 'qemuDomainStorageIdNew/Reset'

They're used only inside qemu_domain.c. Move it before their usage,
and unexport them.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

virnetsockettest: Increase coverage

Add test cases for quotes appearing in the netcat parameter,
for the default behavior of proxy=auto where virt-ssh-helper
is used if available, and for proxy=native.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virnetsockettest: Allow changing the proxy parameter

Currently the test cases all follow the proxy=auto behavior, but
we want to add coverage for other proxy modes as well.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virnetsockettest: Tweak input for test 7

The important part of the value we assign to "netcat" is that it
contains whitespace, so drop everything else to highlight this
fact.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virnetsockettest: Improve indentation

Having the actual script indented and the closing quote on a
separate line, like

  sh -c '
    if foo; then
      bar;
    fi
  '

makes things more readable and easier to scan visually.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virnetsockettest: Move opening quote

Make this test case consistent with all the other ones.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virnetsockettest: Drop unnecessary backslash

No need to escape a single quote.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virbuffer: Simplify virBufferEscapeShell()

We can exit early when the input is an empty string, and we can
avoid storing the string length in a variable since we only use
that information once.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virbuftest: Increase coverage

Test the behavior of virBufferEscapeShell for different types of
quotes as well as the empty string.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

util: Fix getting CPU frequency on Apple Silicon

The hw.cpufrequency sysctl, which we use to obtain the CPU
frequency on macOS, is not available when running on Apple
Silicon, and as a consequence we currently report an error
whenever such information is requested.

The virNodeInfo.mhz field, where the CPU frequency gets stored,
is documented as being zero when the information could not be
obtained, and we already do that for Linux on aarch64. Extend
this behavior to macOS on Apple Silicon.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

qemucapabilitiestest: Hack/fix version in qemu-3.1 tests

The test dumps for x86_64 and ppc64 were generated from pre-release
qemu-3.0-rc1/rc2 and thus wouldn't pass our minimum version check.

As these are very old, fix the version info we use for our check to 3.1
without re-generating them and keep the version tag intact.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemuxml2xmltest: Remove unnecessary versioned invocation of 'disk-cache'

The 'disk-cache' output file is identical in the interesting parts
(everything besides CPU config) to the '-latest' version, so the
versioned invocation can be dropped.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemumonitortestutils: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

qemusecuritymock: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

tools: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

vbox: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virtpm: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virlockspace: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

secret: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

secret: Factor out mutex

If the mutex is part of the `driver` object, it cannot guard that
object's creation and destruction perfectly.

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

vmware: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

vz: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

conf: Avoid NULL-dereference in virDomainObjGetMessages

All callers currently guarantee flags passed to virDomainObjGetMessages
are either zero or contain at least one of the supported flags. But it
doesn't mean we should not check for the possibility an unknown flag was
the only one passed to virDomainObjGetMessages.

Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

qemu_migration_cookie: Properly fetch cert DN

If 1024 was not enough to fit the DN, gnutls_x509_crt_get_dn would store
the required size in subjectlen. And since we're not checking the return
value of this function, we would happily overwrite some random memory.

Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

qemu_migration_cookie: Rename ret in qemuDomainExtractTLSSubject

We use 'ret' for storing values to be returned from a function. Return
values from called functions that are not supposed to be returned
further are usually called 'rv' (or 'rc').

Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

tests: Fix fd leaks in virpcivpdtest

Tests testVirPCIVPDReadVPDBytes and testVirPCIVPDParseFullVPDInvalid
failed to properly close open fildescriptors in some cases. Let's fix it
by switching to VIR_AUTOCLOSE in the whole file.

Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

src: Initialize stack allocated virPCIDeviceAddress variables

There are few places where a virPCIDeviceAddress typed variable
is allocated on the stack but it's not initialized. This can lead
to random values of its members which in turn can lead to a
random behaviour.

Generated with help of the following spatch:

  @@
  identifier I;
  @@
  - virPCIDeviceAddress I;
  + virPCIDeviceAddress I = { 0 };

And then fixing bhyveAssignDevicePCISlots() which does declare
the variable and then explicitly zero it by calling memset() only
to set a specific member afterwards.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>

qemuPrepareNVRAM: Drop cleanup label

After previous commits, the cleanup label shrank to plain
'return' statement. There's no point in having such label, so
drop it.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

qemuPrepareNVRAM: Switch to VIR_AUTOCLOSE

Nothing inside the qemuPrepareNVRAM function relies on @srcFD
being closed early and nothing closes it early. It's okay then to
close it automatically when leaving the function.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

qemuPrepareNVRAM: Us virFileRewrite() to write NVRAM

After previous commits there is no need for qemuPrepareNVRAM() to
open code virFileRewrite(). Deduplicate the code by calling the
function.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

virFileRewrite: Move error reporting into callback

When rewriting a file using virFileRewrite() and error occurs
while writing into a temporary file it's actually the callback
that can report the most accurate error. Move error reporting
into very few callback we have currently. Those callbacks are
trivial so the benefit of this change is not obvious, but this
will change shortly when slightly more complicated callback is
introduced.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

virFileRewrite: Allow setting owner

Currently, due to the way virFileRewrite() works, the rewritten
file is owned by user and group that the daemon runs under. So
far, this is not a problem, because the function is used to write
XML files or secrets for persistent objects (domains, networks,
etc.) and we don't need other users to read/write those files.

But shortly, this function is going to be used for creating files
for QEMU domains. There we want the QEMU process (i.e. different
user) to read the file.

Therefore, introduce two new arguments: @uid and @gid that allow
setting desired owner of the file. Pass -1 to preserve current
behaviour (i.e. create the file owned by the user running the
daemon).

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

ch: Report VIR_DRV_FEATURE_NETWORK_UPDATE_HAS_CORRECT_ORDER as supported

Even though the CH driver doesn't implement virNetworkUpdate()
API, when it does it will see the arguments in correct order.
This is similar to other drivers that don't implement the API,
like ESX, libxl, LXC, etc. Enabling this driver feature stops
clients from swapping the arguments (see comment in the API for
more info).

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>

node_device: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

bhyve_driver: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

network: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

ch: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

ch: Prepare virCHDriverGetCapabilities for automatic mutex management

No functional change intended. This change makes the refactoring to
automatic mutex management easier to follow.

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

lxc: Use automatic mutex management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

lxc: Prepare virLXCDriverGetCapabilities for automatic mutex management

No functional change intended. This change makes the recfatoring to
automatic mutex management easier to follow.

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

libxl: Use automatic memory management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virthreadpool: Cleanup

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

virthreadpool: Use automatic memory management

Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

ch_process: Check whether domain is already running before starting it

There are two places where a domain can be started in CH driver:
chDomainCreateXML() and chDomainCreateWithFlags(). Both acquire a
job (good), but neither of them checks whether the domain isn't
already running. This is wrong. Fortunately, both function call
the very same virCHProcessStart() rendering it the best place for
such check.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

ch_driver: Introduce and use virCHDomainRemoveInactive()

There are few places where a call to virDomainObjListRemove() is
guarded with !vm->persistent check. And there are some places
which are missing this check completely (leading us to losing a
domain). To prevent such mistakes introduce
virCHDomainRemoveInactive() which does the check for us. Also
replace all occurrences of virDomainObjListRemove() with the call
to the new function.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

ch_driver: End job properly on failed chDomainCreateXML()

When creating a domain failed, then the virCHDomainObjEndJob()
would be jumped over. Fix this by creating enjob label and fixing
one goto.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

chDomainCreateXML: Drop spurious driver unlock

Inside chDomainCreateXML(), towards the end, the driver is
unlocked even though there is no corresponding driver lock call
before that. Drop it.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

ch_driver: Don't lock driver when looking up domains

There is no need to lock whole driver when accessing
virDomainObjList. Those APIs were specifically tailored to be
thread safe (when we were dropping QEMU driver lock). Don't
resurrect old history.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

ch_driver: Don't lock driver when getting version

In chConnectGetVersion() the CH driver is locked in order to read
driver->version. This is needless, because not only is the
version set with driver unlocked (chStateInitialize() calls
chExtractVersion() which sets the version), but the version is
practically immutable. Once driver initialized itself it's never
changed.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

qemuFirmwareFillDomain: Use FW descriptors to lookup template when resetting NVRAM

If VIR_QEMU_PROCESS_START_RESET_NVRAM flag is passed when
starting a domain, then user requested to overwrite the domain
specific NVRAM with the one from template. But it is very likely
that the path to the template is not stored in the domain
definition, which in turn makes the copy function
(qemuPrepareNVRAM()) fail.

The solution is simple - when preparing domain, specifically when
deciding whether the path to the template should be autofilled,
ignore any existing NVRAM file.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

qemu_process: Accept VIR_QEMU_PROCESS_START_RESET_NVRAM flag when starting a domain

In one of my previous commits I've fixed the value of
VIR_QEMU_PROCESS_START_RESET_NVRAM flag (which was masking
another value). But what I forgot to do is update virCheckFlags()
calls in two places where the flag is passed: qemuProcessLaunch()
and qemuProcessStart().

Fixes: 1b636593c76f443169ef7bdb3644fd670379d04e
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

docs: Clarify our stance on backported packages

The repositories containing them are usually offered with lower
guarantees, so we don't consider them when it comes to figuring
out the minimum targeted version of our dependencies.

Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>

qemuMonitorOpen: Refactor cleanup

Use VIR_AUTOCLOSE to avoid a 'cleanup' label.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

qemuMonitorOpenUnix: Refactor cleanup

Use VIR_AUTOFREE for the temp socket so that the 'error:' label can be
removed.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Tim Wiederhake <twiederh@redhat.com>

virNetDevOpenvswitchUpdateVlan: Use space for indentation

Breaks syntax-check:

TAB_in_indentation
/home/pipo/libvirt/src/util/virnetdevopenvswitch.c:610: if (virtVlan && virtVlan->nTags > 0)
/home/pipo/libvirt/src/util/virnetdevopenvswitch.c:611: virCommandAddArgList(cmd, "--", "--if-exists", "set", "Port", ifname, NULL);
make: Leaving directory '/home/pipo/build/libvirt/gcc/build-aux'

Fixes: 21c55a45efab47f2c8edc25d4fcabc1b99b6628c
Signed-off-by: Peter Krempa <pkrempa@redhat.com>

virNetDevOpenvswitchUpdateVlan: fix vlan tag update error

We try to update vlan tag by running virsh update-device command,
libvirtd will report ovs-vsctl arguments error. Vlan tag update
funtion does't consider the xml with no vlan configured circumstances.

The steps to reproduce the problem:
1 define and start domain with its vlan configured as:
    <interface type='bridge'>
      <mac address='52:54:00:9e:bb:ac'/>
      <source bridge='ovs-br0'/>
      <vlan>
        <tag id='10'/>
      </vlan>
      <virtualport type='openvswitch'>
      </virtualport>
      <target dev='vnet4.0'/>
      <model type='virtio'/>
      <driver name='vhost'/>
    </interface>
2 define and run virsh update-device command with no vlan configured as:
    <interface type='bridge'>
      <mac address='52:54:00:9e:bb:ac'/>
      <source bridge='ovs-br0'/>
      <virtualport type='openvswitch'>
      </virtualport>
      <target dev='vnet4.0'/>
      <model type='virtio'/>
      <driver name='vhost'/>
    </interface>
   #virsh update-device dom-id novlan.xml
3 virsh command returned error, and we got an error in libvirtd.log:
  error : virCommandWait:2584 : internal error: exit status 1: ovs-vsctl: 'set' command requires at least 3 arguments
  . Child process (ovs-vsctl --timeout=5 -- --if-exists clear Port vnet4.0 tag -- --if-exists clear Port vnet4.0 trunk
  -- --if-exists clear Port vnet4.0 vlan_mode -- --if-exists set Port vnet4.0) unexpected
  error : virNetDevOpenvswitchUpdateVlan:540 : internal error: Unable to set vlan configuration on port vnet4.0

Signed-off-by: Tu Qiang <tu.qiang35@zte.com.cn>
Signed-off-by: Yi Wang <wang.yi59@zte.com.cn>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

docs: Fix HMAT example

There are dupliacated and non-continuous CPU IDs used in HMAT
example. Fix that.

Signed-off-by: Jing Qi <jinqi@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>

qemu: Ignore missing vm.unprivileged_userfaultfd sysctl

Older kernels did not support this sysctl, but they did not restrict
userfaultfd in any way so everything worked as if
vm.unprivileged_userfaultfd was set to 1. Thus we can safely ignore
errors when setting the value.

Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu_process: Fix theoretical overflow in uint to bool typecast

The qemuPrepareNVRAM() function accepts three arguments and the
last one being a boolean type. However, when the function is
called from qemuProcessPrepareHost() the argument passed is a
result of logical and of @flags (unsigned int) and
VIR_QEMU_PROCESS_START_RESET_NVRAM value. In theory this is
unsafe to do because if the value of the flag is ever changed
then this expression might overflow. Do what we do elsewhere:
double negation.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>

qemu_process.c: Fix VIR_QEMU_PROCESS_START_RESET_NVRAM value

In one of recent commits qemuProcessStartFlags enum gained new
value: VIR_QEMU_PROCESS_START_RESET_NVRAM but due to a typo it
has the same value as another member of the enum. Fix that.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>

libxl: Release auto-allocated spice ports

While VNC ports auto-allocated by the libxl driver are released in
libxlDomainCleanup, spice ports are overlooked. Rework the existing
logic to release any auto-allocated graphics ports, not just the VNC
port of the first graphics device.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: fix formatting of pflash readonly attribute

When the <loader> had an explicit readonly='no' attribute we
accidentally still marked the plfash as readonly due to the
bad conversion from virTristateBool to bool. This was missed
because the test cases run with no capabilities set and thus
are validated the -drive approach for pflash configuration,
not the -blockdev approach.

This affected the following config:

  <os>
    <loader readonly='no' type='pflash'>/var/lib/libvirt/qemu/nvram/test-bios.fd</loader>
  </os>

for the sake of completeness, we also add a test XML config
with no readonly attribute at all, to demonstrate that the
default for pflash is intended to be r/w.

Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>

tools: add --reset-nvram arg to several virsh commands

This wires up support for resetting NVRAM for all APIs that allow
this feature.

Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>

qemu: wire up support for resetting NVRAM

We can now replace the existing NVRAM file on startup when
the API requests this.

Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>

include: define constants for resetting NVRAM state

When starting a guest with pflash based firmware, we will initialize
NVRAM from a template if it does not already exist. In theory if the
firmware code file is updated, the existing NVRAM variables should
continue to work correctly. It is inevitable that this could break
accidentally one day. Or a bug in the firmware might corrupt the
NVRAM storage. Or user might make bad changes to the settings that
prevent booting. Or the user might have re-configured the XML to
point to a different firmware file incompatible with the current
variables.

In all these cases it would be useful to delete the existing NVRAM
and initialize it from the pristine template.

To support this introduce a VIR_DOMAIN_START_RESET_NVRAM constant
for use with virDomainCreate / virDomainCreateXML, along with
VIR_DOMAIN_SAVE_RESET_NVRAM for use with virDomainRestore and
VIR_DOMAIN_SNAPSHOT_REVERT_RESET_NVRAM for use with
virDomainSnapshotRevert.

Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>

qemu: do crash safe creation of NVRAM file

If we crash part way through writing the NVRAM file we end up with an
unusable NVRAM on file. To avoid this we need to write to a temporary
file and fsync(2) at the end, then rename to the real NVRAM file path.

Reviewed-by: Ján Tomko <jtomko@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>

Translated using Weblate (Swedish)

Currently translated at 32.4% (3394 of 10455 strings)

Translation: libvirt/libvirt
Translate-URL: https://translate.fedoraproject.org/projects/libvirt/libvirt/sv/

Co-authored-by: Göran Uddeborg <goeran@uddeborg.se>
Signed-off-by: Göran Uddeborg <goeran@uddeborg.se>

cpu_conf: unbreak XPath in virCPUDefParseXML()

In one of my previous commits, I've changed an XPath in
virCPUDefParseXML() from "boolean(./counter...)" to
"./counter...)". Notice the dangling closing bracket? Well, I
didn't back then.

Fixes: 0fe2d8dd335054fae38b46bbbac58a4662e1a1d0
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

lib: Use g_clear_pointer() more

This change was generated using the following spatch:

  @ rule1 @
  expression a;
  identifier f;
  @@
    <...
  - f(*a);
    ... when != a;
  - *a = NULL;
  + g_clear_pointer(a, f);
    ...>

  @ rule2 @
  expression a;
  identifier f;
  @@
    <...
  - f(a);
    ... when != a;
  - a = NULL;
  + g_clear_pointer(&a, f);
    ...>

Then, I left some of the changes out, like tools/nss/ (which
doesn't link with glib) and put back a comment in
qemuBlockJobProcessEventCompletedActiveCommit() which coccinelle
decided to remove (I have no idea why).

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

vbox_common: Drop needless set to NULL

There are two places where a variable passed to VBOX_RELEASE()
macro is set to NULL explicitly. There is no need for that
because the macro sets the variable to NULL already.

Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

libxl: Set auto-allocated graphics ports to used on reconnect

The libxl driver reconnects to all running VMs when libvirtd is restarted,
but it failed to mark auto-allocated graphics ports as set in the port
allocator. If many VMs are running that use port auto-allocation and
libvirtd is restarted, the port allocator is likely to hand out a port
already in use when a new VM is created that uses auto-allocation. VM
creation will fail due to the port clash.

When reconnecting to running VMs after a libvirtd restart, let the port
allocator know about previously allocated ports.

Signed-off-by: Jim Fehlig <jfehlig@suse.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: process: Don't use hardcoded QOM path for cpu for probing flags

Modify 'qemuProcessGetVCPUQOMPath' to take the detected QOM path of the
first vCPU which is always present as the QOM path used our code probing
CPU flags via 'qom-get'.

This is needed as upcoming qemu will change it.

Resolves: https://gitlab.com/libvirt/libvirt/-/issues/272
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=2051451
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: process: Move call to qemuProcessRefreshCPU after cpu probe

Similarly to previous commit we need to probe the vcpus first.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>

qemu: process: Move cpu flag querying after code probing cpus

Upcoming changes will require that we have a proper QOM path for cpus
when querying the flags as qemu is going to change it.

By moving the flag probing code later we'll already probe the QOM paths
so no re-query will be needed.

Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>