Differences from qemu:
* "vmx-ept-uc" (bit 8) and "vmx-ept-wb" (bit 14) are not added to
qemu's list of named features yet, but used in several qemu cpu
models never the less. Add to libvirt regardless.
* "vmx-invvpid-single-context" (bit 41) is erroneously called
"vmx-invept-single-context" in qemu. This is the name of the
feature associated with bit 25 in both libvirt and qemu.
* "vmx-invvpid-single-context-noglobals" (bit 43) is erroneously
called "vmx-invept-single-context-noglobals". Use the correct name.
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Some guest OSes require cpu features from the vmx-* family,
e.g. vmx-xsaves. Up to now, libvirt ignored these features as they
were not required yet. qemu does not automatically enable e.g.
"vmx-xsaves" when requesting "xsaves":
qmp="qemu-kvm -machine accel=kvm -nodefaults -nographic -qmp stdio"
$(qmp) <<-EOF | jq | grep "xsaves"
{ "execute": "qmp_capabilities" }
{
"execute": "query-cpu-model-expansion",
"arguments": {
"type": "full",
"model": {
"name": "Skylake-Client-v1",
"props": { "xsaves": true } `# set to "true" or "false"`
}
}
}
{ "execute": "quit" }
EOF
with xsaves "false":
"xsaves": false,
"vmx-xsaves": false,
with xsaves "true":
"xsaves": true,
"vmx-xsaves": false,
Stop ignoring vmx-* features and begin adding them to libvirt's
database.
Signed-off-by: Tim Wiederhake <twiederh@redhat.com>
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
While glibc provides qsort(), which usually is just a mergesort,
until sorting arrays so huge that temporary array used by
mergesort would not fit into physical memory (which in our case
is never), we are not guaranteed it'll use mergesort. The
advantage of mergesort is clear - it's stable. IOW, if we have an
array of values parsed from XML, qsort() it and produce some
output based on those values, we can then compare the output with
some expected output, line by line.
But with newer glibc this is all history. After [1], qsort() is
no longer mergesort but introsort instead, which is not stable.
This is suboptimal, because in some cases we want to preserve
order of equal items. For instance, in ebiptablesApplyNewRules(),
nwfilter rules are sorted by their priority. But if two rules
have the same priority, we want to keep them in the order they
appear in the XML. Since it's hard/needless work to identify
places where stable or unstable sorting is needed, let's just
play it safe and use stable sorting everywhere.
Fortunately, glib provides g_qsort_with_data() which indeed
implement mergesort and it's a drop in replacement for qsort(),
almost. It accepts fifth argument (pointer to opaque data), that
is passed to comparator function, which then accepts three
arguments.
We have to keep one occurance of qsort() though - in NSS module
which deliberately does not link with glib.
1: https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=03bf8357e8291857a435afcc3048e0b697b6cc04
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Martin Kletzander <mkletzan@redhat.com>
Originally the disk hotplug code didn't know how to attach a CD-ROM
drive, thus didn't have the necessary logic to handle empty cdroms.
Other disks can't be empty which is enforced by the parser validation
logic.
When support for hotplugging cdroms was added the code was not adjusted
to deal with empty drives thus attempted to setup the blockdev backend
for it.
Fixes: 3078799fef
Resolves: https://issues.redhat.com/browse/RHEL-16870
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Commit allowing hotplug of CDROMs moved the logic forbidding the hotplug
to the appropriate blocks based on the disk frontend but forgot to
actually bail out on such error.
Fixes: 3078799fef
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When I've originally refactored the function in commit 0d981bcefc
the logic was still correct, but then later in commit 52f8655439
I've moved most of the image setup logic into the function neglecting to
add the 'goto cleanup;' needed to skip over the setup of the disk
images.
Fixes: 52f8655439
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reword the error message to clearly state that the machine type doesn't
support the address type. It doesn't matter which device it's for.
Additionally the alias may be still NULL at the point when the error is
being reported misleading users that they have something wrong with a
specific device.
Resolves: https://issues.redhat.com/browse/RHEL-16878
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Boris Fiuczynski <fiuczy@linux.ibm.com>
A domain name is expected to be non-empty, and we validate this when
parsing XML, or accepting a new name during renames. We fail to
enforce this property, however, when performing a migration. This
was discovered when a user complained about inaccessible VMs after
migrating with the Rust APIs which mistakenly hardcoded 'dname' to
the empty string.
Fixes: https://gitlab.com/libvirt/libvirt-rust/-/issues/11
Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
As a guard against programming errors, one part of the condition
only dereferences srcpool if it exists, other one does not.
Move the check up one level so that it actually has a chance to do
something useful.
Fixes: 19b1c0d319
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Now that we have virXMLParseWithIndent() and
virXMLParseStringCtxtWithIndent(), we can use them directly and
drop calls to xmlKeepBlanksDefault().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When parsing an XML it may be important to keep indentation to
produce a better looking result when formatting the XML back.
Just look at all those xmlKeepBlanksDefault() calls just before
virXMLParse() is called.
Anyway, as of libxml2 commit v2.12.0~108 xmlKeepBlanksDefault()
is deprecated. Therefore, introduce virXMLParse...WithIndent()
variants which would do exactly xmlKeepBlanksDefault() did but
with non-deprecated APIs.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The virXMLParseHelper() can work in two modes: either it parses a
file or a string. Either way, the same set of flags is specified
in call of corresponding function. Save flags in a local variable
instead.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
After libxml2's commit of v2.12.0~101 we no longer get
xmlIndentTreeOutput declaration by us including just
libxml/xpathInternals.h and libxml2's header files leakage.
Resolves: https://bugs.gentoo.org/917516
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
As mentioned in previous commit, VirtualBox has its own snapshot
XML which we parse, change and then format back. During this, we
ought to keep the indentation to produce better looking result
(especially when we want to compare the output in tests later on,
like we do in vboxsnapshotxmltest).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
When working with VirtualBox's snapshots, the snapshot XML is
firstly parsed, stored in memory (with some parts being stored as
verbatim XML snippets, strings), requested changes are made and
then this modified XML is formatted via
virVBoxSnapshotConfSaveVboxFile() which calls
xmlParseInNodeContext() to format those previously stored XML
snippets.
The first parse of whole VirtualBox snapshot file is done using
virXMLParse() (in virVBoxSnapshotConfLoadVboxFile()) and thus
with XML_PARSE_NONET specified.
But those ad-hoc parsings when formatting the XML back pass zero
flags mask: xmlParseInNodeContext(..., options = 0, ...);
This is potentially dangerous.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
gpg-agent can be used instead of ssh-agent to authenticate
against an SSH server, but in order to do so the GPG_TTY and
TERM environment variables need to be passed through.
For obvious reasons, we avoid doing that when no_tty=1 is found
in the connection URI.
https://bugs.debian.org/843863https://gitlab.com/libvirt/libvirt/-/merge_requests/290
Thanks: Guilhem Moulin <guilhem@guilhem.org>
Thanks: Kunwu Chan <chentao@kylinos.cn>
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
When reverting to inactive snapshot updating the domain definition needs
to happen after the new overlays are created otherwise qemu-img will
correctly fail with error:
Trying to create an image with the same filename as the backing file
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
When we revert to non-leaf snapshot and create new branch or branches
the overlay in snapshot metadata is no longer usable as a disk source
for deletion of that snapshot. We need to use other places to figure out
the correct storage source.
Fixes: https://gitlab.com/libvirt/libvirt/-/issues/534
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
CCW addresses need to be also checked for ABI stability.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
We recently unified all services and sockets, except a couple
were missed. Finish the job.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Problem with HW_PHYSMEM sysctl on 64-bit macOS is that it
returns a 32-bit signed value. Thus it overflows. Switching to
HW_MEMSIZE is recommended as it's of an uint_64 type [1].
1: https://github.com/apple-oss-distributions/xnu/blob/xnu-10002.1.13/bsd/sys/sysctl.h
Reported-by: Jaroslav Suchanek <jsuchane@redhat.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Xcode 15, which provides the compiler toolchain for building libvirt
on macOS has switched to a new linker that warns about duplicated
"-lblah" options on the ld commandline. In practice this is impossible
to prevent in a large project, and also harmless.
Fortunately the new ld command also has an option,
-no_warn_duplicate_libraries, that supresses this harmless/pointless
warning, meson has a simple way to check if that option is supported,
and libvirt's meson.build files already have examples of adding an
option to the ld commandline if it's available.
Signed-off-by: Laine Stump <laine@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Andrea Bolognani <abologna@redhat.com>
Currently some, but not all, methods have a call to the
xdr_free function, for the 'ret' variable. This is done
on methods where there are complex structs containing
allocated memory. In other cases the structs contain
allocated memory, but the pointer is stolen, so xdr_free
is not called. In other cases no allocated memory is
present, so xdr_free.
This is hard to reason about, because the definition of
the struct is not visible in the client stubs.
Switch to use g_auto() for the 'ret' variable, which
means 'xdr_free' is always going to be called. Some
places now need to use g_steal_pointer as a result.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Currently some, but not all, methods have a call to the
xdr_free function, for the 'ret' variable. This is done
on methods where there are complex structs containing
allocated memory. In other cases the structs contain
allocated memory, but the pointer is stolen, so xdr_free
is not called. In other cases no allocated memory is
present, so xdr_free.
This is hard to reason about, because the definition of
the struct is not visible in the client stubs.
Switch to use g_auto() for the 'ret' variable, which
means 'xdr_free' is always going to be called. Some
places now need to use g_steal_pointer as a result.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Currently some, but not all, methods have a call to the
xdr_free function, for the 'ret' variable. This is done
on methods where there are complex structs containing
allocated memory. In other cases the structs contain
allocated memory, but the pointer is stolen, so xdr_free
is not called. In other cases no allocated memory is
present, so xdr_free.
This is hard to reason about, because the definition of
the struct is not visible in the client stubs.
Switch to use g_auto() for the 'ret' variable, which
means 'xdr_free' is always going to be called. Some
places now need to use g_steal_pointer as a result.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
This replaces use of 'rpcgen' with our new python impl of
the RPC code generator. Since the new impl generates code
that matches our style/coding rules, and does not contain
long standing bugs, we no longer need to post-process the
output.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
The current RPC code is post-processed to introduce an
intermediate variable, rather than casting directly
to char ** at time of use. This is said to be a workaround
for type-puning warnings that the compiler emitted.
Neither GCC or CLang emit any warnings for the code in
question today, across any of the architectures we
test in CI. Thus it is presumed that somewhere in the
15 years since the workaround was done, the compilers
have got smarter and do the right thing.
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Commit changing the code to allow passing NULL as @data into
qemuSaveImageDecompressionStart() was not correct as it left the
original call into the function as well.
Introduced-by: 2f3e582a1a
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=2247754
Signed-off-by: Pavel Hrdina <phrdina@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
This is the default for the version of rpcgen shipped with
Linux distributions, but the one in macOS and possibly others
default to K&R C, which modern compilers don't appreciate.
Luckily, all versions of rpcgen shipped with our target
platforms seem to support the -C option.
Signed-off-by: Andrea Bolognani <abologna@redhat.com>
Reviewed-by: Laine Stump <laine@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
QEMU_BLOCK_STORAGE_SOURCE_BACKEND_PROPS_SKIP_UNMAP is no longer
referenced inside the code.
QEMU_BLOCK_STORAGE_SOURCE_BACKEND_PROPS_AUTO_READONLY is passed from
various code paths to the qemuBlockStorageSourceGetBackendProps helper,
but it's no longer used.
Both thus can be removed.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The code was refactored to format the 'read-only' and 'auto-read-only'
flags via the common helper, so the logic determining their values can
be removed.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Use the qemuBlockStorageSourceAddBlockdevCommonProps helper when
formatting protocol layer both when it's used as backing for a format
node and when it's used as the effective node.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The resulting properties are identical, as the hostdev backend code
doesn't set any of the extra properties.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Introduce a mode where the protocol layer -blockdev will be formatted
so that it can be used as the effective node (used to access data from
the device). For this new mode we'll use
qemuBlockStorageSourceAddBlockdevCommonProps.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Use the new helper in qemuBlockStorageSourceGetBlockdevStorageSliceProps
to format the common bits.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The new helper replaces qemuBlockStorageSourceGetBlockdevFormatCommonProps
and the two inline instances generating the common properties for a
blockdev layer.
The new helper is to be used for both the format layer and the storage
backing layer, thus a new parameter 'effective' switches between the
modes.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Use the same ordering of the relevant fields as we do for the format
layer -blockdev so that later they can be refactored without test
fallout.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Change the return value type to 'virDomainDiskGetDetectZeroes'.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
At this point only a single code path (for formatting -drive for legacy
SD cards) uses the 'legacy' output and that code path doesn't populate
the node name. Thus we can unify the code block and simplify the JSON
formatters.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
The 'systemd-analyze security' command looks at the unit file
configuration and reports on any settings which increase the
attack surface for the daemon. Since most systemd units are
fairly minimalist, this is generally informing us about settings
that we never put any thought into using before.
In its current configuration it reports
# systemd-analyze security virtlogd.service
...snip...
→ Overall exposure level for virtlogd.service: 9.6 UNSAFE 😨
which is pretty terrible as a score.
If we apply all of the recommendations that appear possible
without (knowingly) breaking functionality it reports:
# systemd-analyze security virtlogd.service
...snip...
→ Overall exposure level for virtlogd.service: 2.2 OK 🙂
which is a pretty decent improvement.
Some of the settings we would like to enable require a systemd
version that is newer than that available in our oldest distro
target - RHEL-8 at v239.
NB, RestrictSUIDSGID is technically newer than 239, but RHEL-8
backported it, and other distros we target have it by default.
Remaining recommendations are
✗ CapabilityBoundingSet=~CAP_(DAC_*|FOWNER|IPC_OWNER)
We block FOWNER/IPC_OWNER, but can't block the two DAC
capabilities. Historically apps/users might point QEMU
to log files in $HOME, pre-created with their own user
ID.
✗ IPAddressDeny=
Not required since RestrictAddressFamilies blocks IP
usage. Ignoring this avoids the overhead of creating
a traffic filter than will never be used.
✗ NoNewPrivileges=
Highly desirable, but cannot enable it yet, because it
will block the ability to transition to the virtlogd_t
SELinux domain during execve. The SELinux policy needs
fixing to permit this transition under NNP first.
✗ PrivateTmp=
There is a decent chance people have VMs configured
with a serial port logfile pointing at /tmp. We would
cause a regression to use private /tmp for logging
✗ PrivateUsers=
This would put virtlogd inside a user namespace where
its root is in fact unprivileged. Same problem as the
User= setting below
✗ ProcSubset=
Libraries we link to might read certain non-PID related
files from /proc
✗ ProtectClock=
Requires v245
✗ ProtectHome=
Same problem as PrivateTmp=. There's a decent chance
that someone has a VM configured to write a logfile
to /home
✗ ProtectHostname=
Requires v241
✗ ProtectKernelLogs
Requires v244
✗ ProtectProc
Requires v247
✗ ProtectSystem=
We only set it to 'full', as 'strict' is not viable for
our required usage
✗ RootDirectory=/RootImage=
We are not capable of running inside a custom chroot
given needs to write log files to arbitrary places
✗ RestrictAddressFamilies=~AF_UNIX
We need AF_UNIX to communicate with other libvirt daemons
✗ SystemCallFilter=~@resources
We link to libvirt.so which links to libnuma.so which has
a constructor that calls set_mempolicy. This is highly
undesirable todo during a constructor.
✗ User=/DynamicUser=
This is highly desirable, but we currently read/write
logs as root, and directories we're told to write into
could be anywhere. So using a non-root user would have
a major risk of regressions for applications and also
have upgrade implications
Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Setup the VDPA bits of the appropriate part of the image chain for block
copy.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
