When qemu switched to using OptsVisitor for -numa parameter, it did
two things in the same patch. One of them is that the numa parameter
is now visible in "query-command-line-options", the second one is that
it enabled using disjoint cpu ranges for -numa specification. This
will be used in later patch.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
The numa patch series in qemu adds "memory-backend-ram" object type by
which we can tell whether we can use such objects.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
That can be lately achieved with by having .param == NULL in the
virQEMUCapsCommandLineProps struct.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
There were numerous places where numatune configuration (and thus
domain config as well) was changed in different ways. On some
places this even resulted in persistent domain definition not to be
stable (it would change with daemon's restart).
In order to uniformly change how numatune config is dealt with, all
the internals are now accessible directly only in numatune_conf.c and
outside this file accessors must be used.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Since there was already public virDomainNumatune*, I changed the
private virNumaTune to match the same, so all the uses are unified and
public API is kept:
s/vir\(Domain\)\?Numa[tT]une/virDomainNumatune/g
then shrunk long lines, and mainly functions, that were created after
that:
sed -i 's/virDomainNumatuneMemPlacementMode/virDomainNumatunePlacement/g'
And to cope with the enum name, I haad to change the constants as
well:
s/VIR_NUMA_TUNE_MEM_PLACEMENT_MODE/VIR_DOMAIN_NUMATUNE_PLACEMENT/g
Last thing I did was at least a little shortening of already long
name:
s/virDomainNumatuneDef/virDomainNumatune/g
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
There are many places with numatune-related code that should be put
into special numatune_conf and this patch creates a basis for that.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
In XML format, by definition, order of fields should not matter, so
order of parsing the elements doesn't affect the end result. When
specifying guest NUMA cells, we depend only on the order of the 'cell'
elements. With this patch all older domain XMLs are parsed as before,
but with the 'id' attribute they are parsed and formatted according to
that field. This will be useful when we have tuning settings for
particular guest NUMA node.
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
Excerpt from the virCommandAddArgBuffer() description: "Correctly
transfers memory errors or contents from buf to cmd."
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
This patch adds support for the QEMU vhost-user feature to libvirt.
vhost-user enables the communication between a QEMU virtual machine
and other userspace process using the Virtio transport protocol.
It uses a char dev (e.g. Unix socket) for the control plane,
while the data plane based on shared memory.
The XML looks like:
<interface type='vhostuser'>
<mac address='52:54:00:3b:83:1a'/>
<source type='unix' path='/tmp/vhost.sock' mode='server'/>
<model type='virtio'/>
</interface>
Signed-off-by: Michele Paolino <m.paolino@virtualopensystems.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1119173 documents that
commit eaba79d was flawed in the implementation of the
VIR_DOMAIN_BLOCK_JOB_ABORT_ASYNC flag when it comes to completing
a blockcopy. Basically, the qemu pivot action is async (the QMP
command returns immediately, but the user must wait for the
BLOCK_JOB_COMPLETE event to know that all I/O related to the job
has finally been flushed), but the libvirt command was documented
as synchronous by default. As active block commit will also be
using this code, it is worth fixing now.
* src/qemu/qemu_driver.c (qemuDomainBlockJobImpl): Don't skip wait
loop after pivot.
Signed-off-by: Eric Blake <eblake@redhat.com>
Now that we've finally fixed all the violators, it's time to
enforce that any pointer to a const object is never freed (it
is aliasing some other memory, where the non-const original
should be freed instead). Alas, the code still needs a normal
vs. Coverity version, but at least we are still guaranteeing
that the macro call evaluates its argument exactly once.
I verified that we still get the following compiler warnings,
which in turn halts the build thanks to -Werror on gcc (hmm,
gcc 4.8.3's placement of the ^ for ?: type mismatch is a bit
off, but that's not our problem):
int oops1 = 0;
VIR_FREE(oops1);
const char *oops2 = NULL;
VIR_FREE(oops2);
struct blah { int dummy; } oops3;
VIR_FREE(oops3);
util/virauthconfig.c:159:35: error: pointer/integer type mismatch in conditional expression [-Werror]
VIR_FREE(oops1);
^
util/virauthconfig.c:161:5: error: passing argument 1 of 'virFree' discards 'const' qualifier from pointer target type [-Werror]
VIR_FREE(oops2);
^
In file included from util/virauthconfig.c:28:0:
util/viralloc.h:79:6: note: expected 'void *' but argument is of type 'const void *'
void virFree(void *ptrptr) ATTRIBUTE_NONNULL(1);
^
util/virauthconfig.c:163:35: error: type mismatch in conditional expression
VIR_FREE(oops3);
^
* src/util/viralloc.h (VIR_FREE): No longer cast away const.
* src/xenapi/xenapi_utils.c (xenSessionFree): Work around bogus
header.
Signed-off-by: Eric Blake <eblake@redhat.com>
Add 'nocow' to storage volume xml so that user can have an option
to set NOCOW flag to the newly created volume. It's useful on btrfs
file system to enhance performance.
Btrfs has low performance when hosting VM images, even more when the guest
in those VM are also using btrfs as file system. One way to mitigate this
bad performance is to turn off COW attributes on VM files. Generally, there
are two ways to turn off COW on btrfs: a) by mounting fs with nodatacow,
then all newly created files will be NOCOW. b) per file. Add the NOCOW file
attribute. It could only be done to empty or new files.
This patch tries the second way, according to 'nocow' option, it could set
NOCOW flag per file:
for raw file images, handle 'nocow' in libvirt code; for non-raw file images,
pass 'nocow=on' option to qemu-img, and let qemu-img to handle that (requires
qemu-img version >= 2.1).
Signed-off-by: Chunyan Liu <cyliu@suse.com>
In many places we define a variable as a 'const char *' when in fact
we modify it just a few lines below. Or even free it. We should not do
that.
There's one exception though, in xenSessionFree() xenapi_utils.c. We
are freeing the xen_session structure which is defined in
xen/api/xen_common.h public header. The structure contains session_id
which is type of 'const char *' when in fact it should have been just
'char *'. So I'm leaving this unmodified, just noticing the fact in
comment.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
When the backing store of a volume wasn't accessible while updating the
volume definition the call would fail altogether. In cases where we
currently (incorrectly) treat remote backing stores as local one this
might lead to strange errors.
Ignore the opening errors until we figure out how to track proper volume
metadata.
Use the backing store parser to properly create the information about a
volume's backing store. Unfortunately as the storage driver isn't
prepared to allow volumes backed by networked filesystems add a
workaround that will avoid changing the XML output.
Rework the apparmor lxc profile abstraction to mimic ubuntu's container-default.
This profile allows quite a lot, but strives to restrict access to
dangerous resources.
Removing the explicit authorizations to bash, systemd and cron files,
forces them to keep the lxc profile for all applications inside the
container. PUx permissions where leading to running systemd (and others
tasks) unconfined.
Put the generic files, network and capabilities restrictions directly
in the TEMPLATE.lxc: this way, users can restrict them on a per
container basis.
Rename linuxDomainInterfaceStats to virNetInterfaceStats in order
to allow adding platform specific implementations without
making consumer worrying about specific implementation to be used.
Also, rename util/virstatslinux.c to util/virstats.c so placing
other platform specific implementations into this file don't
look unexpected from the file name.
Code logic in libxlDomainAttachDeviceFlags and libxlDomainDetachDeviceFlags
is wrong with return value in error cases.
'ret' was being set to 0 if 'flags & VIR_DOMAIN_DEVICE_MODIFY_CONFIG' was
false. Then if something like virDomainDeviceDefParse() failed in the
VIR_DOMAIN_DEVICE_MODIFY_LIVE logic, the error would be reported but the
function would return success.
Signed-off-by: Chunyan Liu <cyliu@suse.com>
4cc1f1a01f introduced a crash when doing a
block copy as virStorageSourceInitChainElement was called on
"disk->mirror" that is still NULL at that point instead of "mirror"
which temporarily holds the mirror source struct until it's fully
initialized. This resulted into a crash as a NULL was dereferenced.
Reported by: Shanzi Yu <shyu@redhat.com>
Commit id '3ea661de' refactored the code to use the 'disk->src->path'
instead of getting the path from virDomainDiskGetSource(). The one
call to qemuOpenFile() didn't use the disk source path, rather it used
the path as passed from the caller (in this case 'vda') - this caused
a failure with the virt-test/tp-libvirt as follows:
$ virsh domblkinfo virt-tests-vm1 vda
error: cannot stat file '/home/virt-test/shared/data/images/jeos-20-64.qcow2': Bad file descriptor
$
If the openvswitch service is stopped, and is followed by destroying a
VM, the openvswitch bridge translates into a state where it doesn't
recover the port configuration. While it successfully fetches data
from the internal DB, since the corresponding virtual interface does
not exists anymore the whole recovery process fails leaving restarted
VM with inability to connect to the bridge. The following set of
commands will trigger the problem:
virsh start vm
service openvswitch-switch stop
virsh destroy vm
service openvswitch-switch start
virsh start vm
Signed-off-by: Chunhe Li <lichunhe@huawei.com>
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Instead of allocating the virSecurityLabelDef structure ourselves, we
can utilize virSecurityLabelDefNew which even sets the default values
for us.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1113860
We've always done that. Well, until 990e46c45. Point is, if we don't
format model, we may lose a domain on libvirtd restart. If the
seclabel is implicit however, we should skip it's formatting.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Snapshots and block-copy have a flag that forces qemu to re-use existing
file. Our docs weren't exactly clear on what the existing file should
contain for this to actually work.
Re-word the docs a bit to state that the file needs to be pre-created in
the desired format and the backing chain metadata needs to be set prior
to handing it over to qemu.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1084360
Commit dae1568c6c converted the perms
member of the virStorageVolTarget struct into a pointer to make it
optional. But virStorageVolTargetDefFormat did not check perms for
NULL before dereferencing it.
Don't fail when there is nothing to do, as a tweak to the previous
patch regarding output of libvirt-UUID.files for LXC apparmor profiles
Signed-off-by: Eric Blake <eblake@redhat.com>
This was converted to a typedef in 5a2bd4c917 "conf: more enum
cleanups in "src/conf/domain_conf.h"" causing:
libxl/libxl_conf.c: In function 'libxlDiskSetDiscard':
libxl/libxl_conf.c:724:19: error: conversion to incomplete type
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
In lxc, we could not use setmem command
with --config options.
This patch will add support for this.
Signed-off-by: Chen Hanxiao <chenhanxiao@cn.fujitsu.com>
https://bugzilla.redhat.com/show_bug.cgi?id=1066894
With current code it's possible to have for instance:
virsh dumpxml mydomain | grep seclabel
<seclabel type='dynamic' model='selinux' relabel='yes'/>
<seclabel type='dynamic' model='selinux' relabel='yes'/>
<seclabel type='dynamic' model='selinux' relabel='yes'/>
<seclabel type='dynamic' model='selinux' relabel='yes'/>
<seclabel type='dynamic' model='selinux' relabel='yes'/>
what doesn't make any sense. We should reject the XML in the config
parsing phase.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
This negation in names of boolean variables is driving me insane. The
code is much more readable if we drop the 'no-' prefix. Well, at least
for me.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
For non-local storage drivers we can't expect to use the FDStream
backend for up/downloading volumes. Split the code into a separate
backend function so that we can add protocol specific code later.
This saves a few lines of code and catches the error when:
<spice autoport ='yes' defaultMode='any' ..>
<channel name='main' mode='secure'/>
</spice>
is specified with spice_tls = 0 in qemu.conf.
Instead of this error in qemuBuildGraphicsSPICECommandLine:
error: unsupported configuration: spice secure channels set in XML
configuration, but TLS port is not provided
an error is reported in qemuProcessSPICEAllocatePorts:
error: unsupported configuration: Auto allocation of spice TLS port
requested but spice TLS is disabled in qemu.conf
Inspired by:
https://www.redhat.com/archives/libvir-list/2014-June/msg01408.html
When creating cgroups for vcpu and emulator threads whilst starting a
domain, we explicitly skip creating those cgroups in case priv->cgroup
is NULL (cgroups not supported) because SetAffinity() serves the same
purpose. If the host supports only some cgroups (the ones we need are
either unmounted or disabled in qemu.conf), we error out with weird
message even though we could continue starting the domain.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1097028
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
The commit referenced above changed function arguments of
virStorageFileGetMetadataFromBuf() but didn't tweak the
ATTRIBUTE_NONNULL tied to them. This was caught by coverity as it
actually obeys them. We disabled them for GCC and thus it didn't show
up.
Additionally in commit 3ea661deea I passed
NULL to the backingFormat argument which was also marked as nonnull. Use
a dummy int's address when the argument isn't supplied so that the code
doesn't need to change much.
Split out checking of invalid metadata type from the switch statement so
that we can use the typecasted enum value to allow tracking addition of
new items by the compliler.
Also avoids two dead-code break statements.
The default graphics channel mode is 'any', so as to defaultMode attribute.
If defaultMode and channel mode are all the default value 'any',
qemuConnectDomainXMLToNative will set TLSPort.
But in qemuBuildGraphicsSPICECommandLine, if spice_tls is not enabled, libvirtd
will report an error to tell the user that spice TLS is disabled in qemu.conf.
So qemuConnectDomainXMLToNative should check spice_tls is enabled,
then decide to allocate an tlsPort number to this graphics.
If user specified defaultMode is 'secure', qemuConnectDomainXMLToNative
could allocate tlsPort, and then let qemuBuildGraphicsSPICECommandLine reports
the spice_tls disabled error.
The related bug is:
https://bugzilla.redhat.com/show_bug.cgi?id=1113868
Signed-off-by: Jincheng Miao <jmiao@redhat.com>
Signed-off-by: Ján Tomko <jtomko@redhat.com>
Now that cgroups/security driver/locking driver support labelling of
individual images and tolerate network storage we don't have to refrain
from passing all image files to it. This allows removing the checking
code as we already make sure that the snapshot function won't be called
with unsupported options.
Now that security, cgroup and locking APIs support working on individual
images and we track the backing chain security info on a per-image basis
we can finally kill swapping the disk source in virDomainDiskDef and use
the virStorageSource directly.
Until now we were changing information about the disk source via
multiple steps of copying data. Now that we changed to a pointer to
store the disk source we might use it to change the approach to track
the data.
Additionally this will allow proper tracking of the backing chain.
When pivoting to a new disk source after a block commit (and possibly
after a soon-to-be-added active block commit) we changed just a few
fields to the new target. In case we'd copy a network disk to a local
file we'd not change the type properly.
To avoid such problems, switch to tracking of the source via changing of
the complete source struct to the one tracking the mirroring info.
Use the source struct and the corresponding function so that we can
avoid using the path separately. Now that
qemuDomainPrepareDiskChainElementPath isn't use anywhere, we can safely
remove it.
Additionally, the removal fixes a misaligned comment as the removed
function was added under a comment for a different function.
Add security driver functions to label separate storage images using the
virStorageSource definition. This will help to avoid the need to do ugly
changes to the disk struct and use the source directly.
Add functions that will allow to set all the required cgroup stuff on
individual images taking a virStorageSourcePtr. Also convert functions
designed to setup whole backing chain to take advantage of the change.
When dispatching events from the event loop, the array of registered
handles is searched to see what handles happened an event on. However,
the array is searched in weird way: the check for the array boundaries
is at the end, so we may touch the elements after the end of the
array:
==10434== Invalid read of size 4
==10434== at 0x52D06B6: virEventPollDispatchHandles (vireventpoll.c:486)
==10434== by 0x52D10E4: virEventPollRunOnce (vireventpoll.c:660)
==10434== by 0x52CF207: virEventRunDefaultImpl (virevent.c:308)
==10434== by 0x1639D1: virNetServerRun (virnetserver.c:1139)
==10434== by 0x1220DC: main (libvirtd.c:1507)
==10434== Address 0xc11ff04 is 4 bytes after a block of size 960 alloc'd
==10434== at 0x4C2CA5E: realloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==10434== by 0x52AD378: virReallocN (viralloc.c:245)
==10434== by 0x52AD46E: virExpandN (viralloc.c:294)
==10434== by 0x52AD5B1: virResizeN (viralloc.c:352)
==10434== by 0x52CF2EC: virEventPollAddHandle (vireventpoll.c:116)
==10434== by 0x52CEF5B: virEventAddHandle (virevent.c:78)
==10434== by 0x11F69A90: nodeStateInitialize (node_device_udev.c:1797)
==10434== by 0x53C3C89: virStateInitialize (libvirt.c:743)
==10434== by 0x120563: daemonRunStateInit (libvirtd.c:919)
==10434== by 0x5317719: virThreadHelper (virthread.c:197)
==10434== by 0x8376F39: start_thread (in /lib64/libpthread-2.17.so)
==10434== by 0x8A7F9FC: clone (in /lib64/libc-2.17.so)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Commit a48f445100 introduced a helper
function to convert cgroup device mode to string. The function was only
conditionally compiled on platforms that support cgroup. This broke the
build when attempting to export the symbol:
CCLD libvirt.la
Cannot export virCgroupGetDevicePermsString: symbol not defined
Move the function out of the ifdef, as it doesn't really depend on the
cgroup code being present.
In libxlDomainMigrationConfirm(), a transient domain is removed
from the domain list after successful migration. Later in cleanup,
the domain object is unlocked, resulting in a crash
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x7fb4208ed700 (LWP 12044)]
0x00007fb4267251e6 in virClassIsDerivedFrom (klass=0xdeadbeef,
parent=0x7fb42830d0c0) at util/virobject.c:169
169 if (klass->magic == parent->magic)
(gdb) bt
0 0x00007fb4267251e6 in virClassIsDerivedFrom (klass=0xdeadbeef,
parent=0x7fb42830d0c0) at util/virobject.c:169
1 0x00007fb42672591b in virObjectIsClass (anyobj=0x7fb4100082b0,
klass=0x7fb42830d0c0) at util/virobject.c:365
2 0x00007fb42672583c in virObjectUnlock (anyobj=0x7fb4100082b0)
at util/virobject.c:338
3 0x00007fb41a8c7d7a in libxlDomainMigrationConfirm (driver=0x7fb4100404c0,
vm=0x7fb4100082b0, flags=1, cancelled=0) at libxl/libxl_migration.c:583
Fix by setting the virDomainObjPtr to NULL after removing it from
the domain list.
During migration, the libxl driver starts a modify job in the
begin phase, ending the job in the confirm phase. This is
essentially VIR_MIGRATE_CHANGE_PROTECTION semantics, but the
driver does not support that flag. Without CHANGE_PROTECTION
support, the job would never be terminated in error conditions
where migrate confirm phase is not executed. Further attempts
to modify the domain would result in failure to acquire a job
after LIBXL_JOB_WAIT_TIME.
Similar to the qemu driver, end the job in the begin phase.
Protecting the domain object across all phases of migration can
be done in a future patch adding CHANGE_PROTECTION support.
In libxlDomainMigrationPrepare(), a new virDomainObj is created
from the incoming domain def and added to the driver's domain
list, but never removed if there are subsequent failures during
the prepare phase.
targethost# virsh list --all
sourcehost# virsh migrate --live dom xen+ssh://targethost/system
error: operation failed: Fail to create socket for incoming migration.
targethost# virsh list --all
error: Failed to list domains
error: name in virGetDomain must not be NULL
After adding code to remove the domain on prepare failure, noticed
that libvirtd crashed due to double free of the virDomainDef. Similar
to the qemu driver, pass a pointer to virDomainDefPtr so it can be set
to NULL once a virDomainObj is created from it.
Qemu will fallback to aio=threads when the cache mode doesn't use
O_DIRECT, even if aio=native was explictly set.
Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1086704
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Cgroups code uses VIR_CGROUP_DEVICE_* flags to specify the mode but in
the end it needs to be converted to a string. Add a helper to do it and
use it in the cgroup code before introducing it into the rest of the
code.
When discovering a disk backing chain the parent disk's metadata need to
be populated into the guest images so that each piece of the backing
chain contains a copy of those. This will allow us to refactor the
security driver so that it will not need to carry around the original
disk definition.
We are going to modify storage source chains in place. Add a helper that
will copy relevant information such as security labels to the new
element if that doesn't contain it.
In the future we might need to track state of individual images. Move
the readonly and shared flags to the virStorageSource struct so that we
can keep them in a per-image basis.
Some of the further changes will propagate seclabels from a disk source
element into the backing store elements. This would change the XML
output of the backing store as the seclabels would be formatted for each
backing store element. Skip the seclabels formatting until we decide
that it's necessary.
Now that we are able to select images from the backing chain via indexed
access we should also convert possible network sources to
qemu-compatible strings before passing them to qemu.
Now that we are able to select images from the backing chain via indexed
access we should also convert possible network sources to
qemu-compatible strings before passing them to qemu.
Introduce flag for the block rebase API to allow the rebase operation to
leave the chain relatively addressed. Also adds a virsh switch to enable
this behavior.
Introduce flag for the block commit API to allow the commit operation to
leave the chain relatively addressed. Also adds a virsh switch to enable
this behavior.
The qemu block info function relied on working with local storage. Break
this assumption by adding support for remote volumes. Unfortunately we
still need to take a hybrid approach as some of the operations require a
filedescriptor.
Previously you'd get:
$ virsh domblkinfo gl vda
error: cannot stat file '/img10': Bad file descriptor
Now you get some stats:
$ virsh domblkinfo gl vda
Capacity: 10485760
Allocation: 197120
Physical: 197120
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1110198
To allow reusing this function in the qemu driver we need to allow
specifying the storage format. Also separate return of the backing store
path now isn't necessary.
