libvirt correctly ignores these two elements and they will not appear
in the parsed XML.
However, if we use virDomainSetMetadata API and with "" as value for
title or description we will end up with the parsed XML that contains
these empty elements.
Let's fix the behavior of this API to behave the same as
virDomainDefineXML.
Thomas Huth [Fri, 1 Mar 2019 11:10:26 +0000 (12:10 +0100)]
tools/virt-host-validate: Fix IOMMU check on s390x
When running virt-host-validate on an s390x host, the tool currently warns
that it is "Unknown if this platform has IOMMU support". We can use the
common check for entries in /sys/kernel/iommu_groups here, too, but it only
makes sense to check it if there are also PCI devices available. It's also
common on s390x that there are no PCI devices assigned to the LPAR, and in
that case there is no need for the PCI-related IOMMU, so without PCI devices
we should simply skip this test.
Signed-off-by: Thomas Huth <thuth@redhat.com> Reviewed-by: Bjoern Walk <bwalk@linux.ibm.com> Reviewed-by: Boris Fiuczynski <fiuczy@linux.ibm.com> Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Eric Blake [Tue, 5 Mar 2019 03:26:38 +0000 (21:26 -0600)]
snapshot: Add virDomainSnapshotObjListParse
Add a new function to make it possible to parse a list of snapshots
at once. This is a counterpart to an earlier patch making it
possible to produce all snapshots in a single XML string, and
intentionally parses the same top-level element <snapshots> with
an optional attribute current='name'.
Note that since we know we started with no relations at all, and
since checking parent relationships per-snapshot is not viable as
we don't control which order the snapshots appear in, that we are
fine with doing a final pass to update all parent/child
relationships among the definitions.
Signed-off-by: Eric Blake <eblake@redhat.com> Reviewed-by: John Ferlan <jferlan@redhat.com>
Eric Blake [Thu, 7 Mar 2019 22:12:24 +0000 (16:12 -0600)]
snapshot: Split out virDomainSnapshotRedefineValidate helper
Pull out the portion of virDomainSnapshotRefinePrep() that deals
with definition sanity into a separate helper routine that can
be reused with bulk redefine, leaving behind only the code
specific to loop checking and in-place updates that are only
needed in single-definition handling.
Signed-off-by: Eric Blake <eblake@redhat.com> Reviewed-by: John Ferlan <jferlan@redhat.com>
Eric Blake [Tue, 5 Mar 2019 02:38:27 +0000 (20:38 -0600)]
snapshot: Avoid latent use-after-free when cleaning snapshots
Right now, the only callers of qemuDomainSnapshotDiscardAllMetadata()
are right before freeing the virDomainSnapshotObjList, so it did not
matter if the list's metaroot (which points to all the defined root
snapshots) is left inconsistent. But an upcoming patch will want to
clear all snapshots if a bulk redefine fails partway through, in
which case things must be reset. Make this work by teaching the
existing virDomainSnapshotUpdateRelations() to be safe regardless of
the incoming state of the metaroot (since we don't want to leak that
internal detail into qemu code), then fixing the qemu code to use
it after deleting all snapshots. Additionally, the qemu code must
reset vm->current_snapshot if the current snapshot was removed,
regardless of whether the overall removal succeeded or failed later.
Signed-off-by: Eric Blake <eblake@redhat.com> Reviewed-by: John Ferlan <jferlan@redhat.com>
Eric Blake [Fri, 15 Feb 2019 20:43:43 +0000 (14:43 -0600)]
snapshot: Give virDomainSnapshotDefFormat its own flags
virDomainSnapshotDefFormat currently takes two sets of knobs:
an 'unsigned int flags' argument that can currently just be
VIR_DOMAIN_DEF_FORMAT_SECURE, and an 'int internal' argument used as
a bool to determine whether to output an additional element. It
then reuses the 'flags' knob to call into virDomainDefFormatInternal(),
which takes a different set of flags. In fact, prior to commit 0ecd6851
(1.2.12), the 'flags' argument actually took the public
VIR_DOMAIN_XML_SECURE, which was even more confusing. Let's borrow
from the style of that earlier commit, by introducing a function
for translating from the public flags (VIR_DOMAIN_SNAPSHOT_XML_SECURE
was just recently introduced) into a new enum specific to snapshot
formatting, and adjust all callers to use snapshot-specific enum
values when formatting, and where the formatter now uses a new
variable 'domainflags' to make it obvious when we are translating
from snapshot flags back to domain flags. We don't even have to
use the conversion function for drivers that don't accept the
public VIR_DOMAIN_SNAPSHOT_XML_SECURE flag.
Signed-off-by: Eric Blake <eblake@redhat.com> Reviewed-by: John Ferlan <jferlan@redhat.com>
Eric Blake [Thu, 7 Mar 2019 21:12:40 +0000 (15:12 -0600)]
qemu: Use virDomainSnapshotState for switch statements
Clean up the previous patch which abused switch on virDomainState
while working with a variable containing virDomainSnapshotState, by
converting the two affected switch statements to now use the right
enum.
Signed-off-by: Eric Blake <eblake@redhat.com> Reviewed-by: John Ferlan <jferlan@redhat.com>
Eric Blake [Tue, 26 Feb 2019 20:14:36 +0000 (14:14 -0600)]
snapshot: Rework virDomainSnapshotState enum
The existing virDomainSnapshotState is a superset of virDomainState,
adding one more state (disk-snapshot) on top of valid domain states.
But as written, the enum cannot be used for gcc validation that all
enum values are covered in a strongly-typed switch condition, because
the enum does not explicitly include the values it is adding to.
Copy the style used in qemu_blockjob.h of creating new enum names
for every inherited value, and update most clients to use the new
enum names anywhere snapshot state is referenced. The exception is
two switch statements in qemu code, which instead gain a fixme
comment about odd type usage (which will be cleaned up in the next
patch). The rest of the patch is fairly mechanical (I actually did
it by temporarily s/state/xstate/ in snapshot_conf.h to let the
compiler find which spots in the code used the field, did the
obvious search and replace in those functions, then undid the rename).
Signed-off-by: Eric Blake <eblake@redhat.com> Reviewed-by: John Ferlan <jferlan@redhat.com>
Eric Blake [Tue, 26 Feb 2019 19:44:46 +0000 (13:44 -0600)]
qemu: Refactor snapshot check for _LIVE vs. _REDEFINE
The current qemu code rejects the combination of the two flags
VIR_DOMAIN_SNAPSHOT_CREATE_LIVE in tandem with
VIR_DOMAIN_SNAPSHOT_CREATE_REDEFINE, but rather late in the cycle
(after the snapshot was already parsed), and with a rather confusing
message (complaining that live snapshots require external storage,
even if the redefined snapshot already declares external storage).
Hoist the rejection message to occur earlier (before parsing any
XML, which also aids upcoming patches that will implement bulk
redefine), and with a more typical error message about mutually
exclusive flags.
Signed-off-by: Eric Blake <eblake@redhat.com> Reviewed-by: John Ferlan <jferlan@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Maxim Kozin [Wed, 6 Mar 2019 18:39:11 +0000 (21:39 +0300)]
lxc: Try harder to stop/reboot containers
If shutting down a container via setting the runlevel fails, the
control jumps right onto endjob label and doesn't even try
sending the signal. If flags allow it, we should try both
methods.
Signed-off-by: Maxim Kozin <kolomaxes@gmail.com> Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
We can use STRNEQ() instead of STRNEQLEN() since we're only
interested in the trailing part of the string and we've
already verified that the length of file, name and suffix
are those we expect.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> ACKed-by: Peter Krempa <pkrempa@redhat.com>
While this function is not, strictly speaking, a predicate,
it still mostly behaves like one as evidenced by the vast
majority of its callers, so using bool rather than int as
the return type makes sense.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> ACKed-by: Peter Krempa <pkrempa@redhat.com>
Since qemu 2.13 reports the target architecture in a property called
'target' additionally to the property 'arch', that has been used in
qemu 2.12 in the response data of 'query-cpus-fast'.
Libvirts monitor code prefers the 'target' property over 'arch'.
At least for s390(x), target is reported as 's390x' while arch is 's390'.
In a later step a comparison is performed against 's390' which fails for
qemu 2.13 and later.
In consequence the architecture specific data for s390 won't be extracted
from the returned data, leading to incorrect values being reported by
virsh domstats --vcpu.
Changing to check explicitly for 's390' and 's390x'.
Signed-off-by: Viktor Mihajlovski <mihajlov@linux.ibm.com> Reviewed-by: Bjoern Walk <bwalk@linux.ibm.com> Reviewed-by: Boris Fiuczynski <fiuczy@linux.ibm.com>
cpu: Don't access invalid memory in virCPUx86Translate
Problem is that if there are no signatures for a CPU, then we
still allocate cpu->signatures (even though with size 0). Later,
we access cpu->signatures[0] if cpu->signatures is not NULL.
Invalid read of size 4
at 0x5F439D7: virCPUx86Translate (cpu_x86.c:2930)
by 0x5F3C239: virCPUTranslate (cpu.c:927)
by 0x57CE7A1: qemuProcessUpdateGuestCPU (qemu_process.c:5870)
...
Address 0xf752d40 is 0 bytes after a block of size 0 alloc'd
at 0x4C30EC6: calloc (vg_replace_malloc.c:711)
by 0x5DBDE4E: virAllocN (viralloc.c:190)
by 0x5F3E4FA: x86ModelCopySignatures (cpu_x86.c:990)
by 0x5F3E60F: x86ModelCopy (cpu_x86.c:1008)
by 0x5F3E7CB: x86ModelFromCPU (cpu_x86.c:1068)
by 0x5F4397E: virCPUx86Translate (cpu_x86.c:2922)
by 0x5F3C239: virCPUTranslate (cpu.c:927)
by 0x57CE7A1: qemuProcessUpdateGuestCPU (qemu_process.c:5870)
...
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Jiri Denemark <jdenemar@redhat.com>
There is a lot of documentation in the comments about how PPC64 handles
passthrough VFIO devices to calculate the @memLockLimit. And more will
be added with the PPC64 NVLink2 support code.
Let's remove the PPC64 code from qemuDomainGetMemLockLimitBytes()
body and put it into a helper function. This will simplify the
flow of qemuDomainGetMemLockLimitBytes() that handles all the other
platforms and improves readability of the PPC64 specifics.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com> Reviewed-by: Erik Skultety <eskultet@redhat.com>
qemu: domain: Simplify non-VFIO memLockLimit calculation for PPC64
@passthroughLimit is being calculated even if @usesVFIO is false. After
that, an if-else conditional is used to check if we're going to sum it
up with @baseLimit.
This patch initializes @passthroughLimit to zero and always returns
@memKB = @baseLimit + @passthroughLimit. The conditional is then used to
calculate @passthroughLimit if @usesVFIO == true. This results in some
cycles being spared for the @usesVFIO == false scenario, but the real
motivation is to make the code simpler to add an alternative formula to
calculate @passthroughLimit for NVLink2.
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com> Reviewed-by: Erik Skultety <eskultet@redhat.com>
Michal Privoznik [Tue, 26 Feb 2019 09:12:05 +0000 (10:12 +0100)]
virmock: Initialize both symbols in VIR_MOCK_REAL_INIT_ALT
It may happen that both symbols are present. Especially when
chaining mocks. For instance if a test is using virpcimock and
then both stat and __xstat would be present in the address space
as virpcimock implements both. Then, if the test would try to use
say virfilewrapper (which again uses VIR_MOCK_REAL_INIT_ALT() to
init real_stat and real___xstat) it would find stat() from
virpcimock and stop there. The virfilewrapper.c:real___xstat
wouldn't be initialized and thus it may result in a segfault.
The reason for segfault is that sys/stat.h may redefine stat() to
call __xstat().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Pavel Hrdina [Wed, 6 Mar 2019 17:59:39 +0000 (18:59 +0100)]
caps: drop requiredSourceElements from storage pool capabilities
Capabilities should not duplicate data that are obvious from our
documentation and will not change with different QEMU binaries
or the way how we compile libvirt.
When dealing with internal paths we don't need to worry about
whether or not suffixes are lowercase since we have full control
over them, which means we can avoid performing case-insensitive
string comparisons.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> ACKed-by: Peter Krempa <pkrempa@redhat.com>
util: Rename virFileMatchesNameSuffix() to virStringMatchesNameSuffix()
Despite its name, this is really just a general-purpose string
manipulation function, so it should be moved to the virstring
module and renamed accordingly.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> ACKed-by: Peter Krempa <pkrempa@redhat.com>
util: Rename virFileStripSuffix() to virStringStripSuffix()
Despite its name, this is really just a general-purpose string
manipulation function, so it should be moved to the virstring
module and renamed accordingly.
A few trivial whitespace changes are squashed in.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> ACKed-by: Peter Krempa <pkrempa@redhat.com>
util: Rename virFileHasSuffix() to virStringHasCaseSuffix()
Despite its name, this is really just a general-purpose string
manipulation function, so it should be moved to the virstring
module and renamed accordingly.
In addition to the obvious s/File/String/, also tweak the name
to make it clear that the presence of the suffix is verified
using case-insensitive comparison.
A few trivial whitespace changes are squashed in.
Signed-off-by: Andrea Bolognani <abologna@redhat.com> ACKed-by: Peter Krempa <pkrempa@redhat.com>
Jim Fehlig [Fri, 1 Mar 2019 22:05:36 +0000 (15:05 -0700)]
apparmor: Add ptrace and signal rules for named profile
Commit a3ab6d42 changed the libvirtd profile to a named profile
but neglected to accommodate the change in the qemu profile
ptrace and signal rules. As a result, libvirtd is unable to
signal confined qemu processes and hence unable to shutdown
or destroy VMs.
Add ptrace and signal rules that reference the libvirtd profile
by name in addition to full binary path.
Signed-off-by: Jim Fehlig <jfehlig@suse.com> Acked-by: Jamie Strandboge <jamie@canonical.com>
John Ferlan [Thu, 7 Feb 2019 17:07:16 +0000 (12:07 -0500)]
tests: Introduce storage pool capabilites test
Add a new test for the storage pool capabilities. There will be
one test mocked with every backend available (full) and one where
only the file system pool is available.
Signed-off-by: John Ferlan <jferlan@redhat.com> ACKed-by: Michal Privoznik <mprivozn@redhat.com>
John Ferlan [Thu, 7 Feb 2019 17:29:43 +0000 (12:29 -0500)]
conf: Add storage pool capability formatting
Add support to format the storage pool capabilities using
the virStoragePoolTypeInfoPtr to determine what capabilities
exist for the various pools and the driver capabilities to
determine whether the pool is compiled in and supported.
Signed-off-by: John Ferlan <jferlan@redhat.com> ACKed-by: Michal Privoznik <mprivozn@redhat.com>
During storage driver backend initialization, let's save
which backends are available in the storage pool capabilities.
In order to format those, we need add a connectGetCapabilities
processor to the storageHypervisorDriver. This allows a storage
connection, such as "storage:///system" to find the API and
format the results, such as:
John Ferlan [Wed, 6 Feb 2019 22:14:35 +0000 (17:14 -0500)]
conf: Remove volOptions for VIR_STORAGE_POOL_ISCSI[_DIRECT]
The iscsi and iscsi-direct pools are documented as not using
the volume type, so let's just remove it. Besides it would
have produced bad output since formatting uses the Disk types.
Signed-off-by: John Ferlan <jferlan@redhat.com> ACKed-by: Michal Privoznik <mprivozn@redhat.com>
John Ferlan [Wed, 6 Feb 2019 21:57:28 +0000 (16:57 -0500)]
conf: Remove volOptions for VIR_STORAGE_POOL_SHEEPDOG
The sheepdog pool is documented as not using the volume type,
so let's just remove it. Besides it would have produced bad
results since the defaultType is FILE but the formatting used
the Disk types.
Signed-off-by: John Ferlan <jferlan@redhat.com> ACKed-by: Michal Privoznik <mprivozn@redhat.com>
Peter Krempa [Tue, 5 Mar 2019 12:01:26 +0000 (13:01 +0100)]
conf: Split out domain features formatting from virDomainDefFormatInternal
Pure code motion of code for formatting domain features to a function
called virDomainDefFormatFeatures. Best viewed with the '--patience'
option for git show.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Peter Krempa [Fri, 22 Feb 2019 16:36:23 +0000 (17:36 +0100)]
util: xml: Enforce return value check from virXMLFormatElement
The function does not transfer errors from 'attrBuf' and 'childBuf'
arguments into 'buf', but rather reports them right away, thus we need
to make sure that it's always checked.
Signed-off-by: Peter Krempa <pkrempa@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Jiri Denemark [Tue, 5 Mar 2019 21:16:03 +0000 (22:16 +0100)]
qemu: Don't set migration caps when changing postcopy bandwidth
The qemuMigrationParamsApply internal API was designed to apply all
migration parameters and capabilities before we start to migrate a
domain. While migration parameters are only passed to QEMU when we
explicitly want to set a specific value, capabilities are always either
enabled or disabled.
Thus when this API is called outside migration job, e.g., via a call to
qemuDomainMigrateSetMaxSpeed with VIR_DOMAIN_MIGRATE_MAX_SPEED_POSTCOPY
flag, we would call migrate-set-capabilities and disable all
capabilities. However, changing capabilities while migration is already
running does not make sense and our code should never be trying to do
so. In fact QEMU even reports an error if migrate-set-capabilities is
called during migration and qemuDomainMigrateSetMaxSpeed would fail
with:
internal error: unable to execute QEMU command
migrate-set-capabilities: There's a migration process in progress
With this patch qemuMigrationParamsApply never tries to call
migrate-set-capabilities outside of migration job. When the capabilities
bitmap is all zeros (which is its initial value after
qemuMigrationParamsNew), we just skip the command. But when any
capability bit is set to 1 by a non-migration job, we report an error to
highlight a bug in our code.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com> Acked-by: Peter Krempa <pkrempa@redhat.com>
security: aa-helper: gl devices in sysfs at arbitrary depth
Further testing with more devices showed that we sometimes have a
different depth of pci device paths when accessing sysfs for device
attributes.
But since the access is limited to a set of filenames and read only it
is safe to use a wildcard for that.
Related apparmor denies - while we formerly had only considered:
apparmor="DENIED" operation="open"
name="/sys/devices/pci0000:00/0000:00:02.1/uevent"
requested_mask="r"
We now also know of cases like:
apparmor="DENIED" operation="open"
name="/sys/devices/pci0000:00/0000:00:03.1/0000:1c:00.0/uevent"
requested_mask="r"
Fixes: https://bugs.launchpad.net/ubuntu/+source/libvirt/+bug/1817943 Acked-by: Jamie Strandboge <jamie@canonical.com> Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
Now that drivers may call virConnectOpen() on secondary drivers, it
doesn't make much sense to have autostart separated from driver
initialization callback. In fact, it creates a problem because one
driver during its initialization might try to fetch an object from
another driver but since the object is yet to be autostarted the fetch
fails. This has been observed in reality: qemu driver performs
qemuProcessReconnect() during qemu's stateInitialize phase which may
call virDomainDiskTranslateSourcePool() which connects to the storage
driver to look up the volume. But the storage driver did not autostart
its pools yet therefore volume lookup fails and the domain is killed.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
The stateAutoStart callback will be removed in the next commit.
Therefore move autostarting of domains, networks and storage
pools back into stateInitialize callbacks.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
daemon: Register secret driver before storage driver
The order in which drivers are registered is important because
their stateInitialize and stateAutoStart callback are called in
that order. Well, stateAutoStart is going away and therefore if
there is some dependency between two drivers (e.g. when
initializing storage driver expects secret driver to be available
already), the registration of such drivers must happen in correct
order.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Jiri Denemark [Mon, 25 Feb 2019 09:05:34 +0000 (10:05 +0100)]
cputest: Test CPU signatures
The signature computation code is not too complicated and it will likely
never change so testing it is not very important. We do it mostly for a
nice side effect of easily accessible signature numbers for all CPU
data files.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
Jiri Denemark [Tue, 26 Feb 2019 08:22:36 +0000 (09:22 +0100)]
cpu_map: Add hex representation of signatures
The family/model numbers are nice for humans or for comparing with
/proc/cpuinfo, but sometimes there's a need to see the CPUID
representation of the signature. Let's add it into a comment for each
signature in out cpu_map XMLs as the conversion is not exactly
straightforward.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com> Reviewed-by: Ján Tomko <jtomko@redhat.com>
projects / libvirt.git / log