Fix crash when deleting monitor while a command is in progress
If QEMU shuts down while we're in the middle of processing a
monitor command, the monitor will be freed, and upon cleaning
up we attempt to do qemuMonitorUnlock(priv->mon) when priv->mon
is NULL.
To address this we introduce proper reference counting into
the qemuMonitorPtr object, and hold an extra reference whenever
executing a command.
* src/qemu/qemu_driver.c: Hold a reference on the monitor while
executing commands, and only NULL-ify the priv->mon field when
the last reference is released
* src/qemu/qemu_monitor.h, src/qemu/qemu_monitor.c: Add reference
counting to handle safe deletion of monitor objects
Daniel Veillard [Tue, 8 Dec 2009 10:08:17 +0000 (11:08 +0100)]
Fix a compilation failure if yajl not avail
configure: yajl: no
CC libvirt_util_la-json.lo
util/json.c:32:27: error: yajl/yajl_gen.h: No such file or directory
util/json.c:33:29: error: yajl/yajl_parse.h: No such file or directory
* src/util/json.c: remove the includes if yajl not configured in
This introduces simple API for handling JSON data. There is
an internal data structure 'virJSONValuePtr' which stores a
arbitrary nested JSON value (number, string, array, object,
nul, etc). There are APIs for constructing/querying objects
and APIs for parsing/formatting string formatted JSON data.
This uses the YAJL library for parsing/formatting from
http://lloyd.github.com/yajl/
* src/util/json.h, src/util/json.c: Data structures and APIs
for representing JSON data, and parsing/formatting it
* configure.in: Add check for yajl library
* libvirt.spec.in: Add build requires for yajl
* src/Makefile.am: Add json.c/h
* src/libvirt_private.syms: Export JSON symbols to drivers
Add --system flag to autogen.sh to make it easy to build with right prefix
Add a --system flag to autogen.sh which gets turned into the args
--prefix=/usr --sysconfdir=/etc --localstatedir=/var to make it
easy to build with settings that match an RPM build
Jiri Denemark [Fri, 4 Dec 2009 15:55:59 +0000 (16:55 +0100)]
Support for interface model='netfront'
Xen HVM guests with PV drivers end up with two network interfaces for
each configured interface. One of them being emulated by qemu and the
other one paravirtual. As this might not be desirable, the attached
patch provides a way for users to specify that only paravirtual network
interface should be presented to the guest.
The configuration was inspired by qemu/kvm driver, for which users can
specify model='virtio' to use paravirtual network interface.
The patch adds support for model='netfront' which results in
type=netfront instead of type=ioemu (or nothing for newer xen versions)
in guests native configuration. Xen's qemu ignores interfaces with
type != ioemu and only paravirtual network device will be seen in the
guest.
Four possible configuration scenarios follow:
- no model specified in domain's XML
- libvirt will behave like before this change; it will set
type=ioemu for HVM guests on xen host which is not newer than
XEND_CONFIG_MAX_VERS_NET_TYPE_IOEMU
- covered by existing tests
- PV guest, any model
- no functional change, model is passed as is (and ignored by the
hypervisor)
- covered by existing tests (e.g., *-net-e1000.*)
- HVM guest, model=netfront
- type is set to "netfront", model is not specified
- covered by new *-net-netfront.* tests
- HVM guest, model != netfront
- type is set to "ioemu", model is passed as is
- covered by new *-net-ioemu.* tests
The fourth scenario feels like a regression for xen newer than
XEND_CONFIG_MAX_VERS_NET_TYPE_IOEMU as users who had a model specified
in their guest's configuration won't see a paravirtual interface in
their guests any more. On the other hand, the reason for specifying a
model is most likely the fact that they want to use such model which
implies emulated interface. Users of older xen won't be affected at all
as their xen provides paravirtual interface regardless of the type used.
- src/xen/xend_internal.c: add netfront support for the xend backend
- src/xen/xm_internal.c: add netfront support for the XM serialization too
Pritesh Kothari [Fri, 4 Dec 2009 13:49:45 +0000 (14:49 +0100)]
vbox: Add support for version 3.1
Also fixed serial port configuration which was broken due to recent
change in virDomainChrDef where targetType was newly added.
* src/Makefile.am: add new files
* src/vbox/vbox_driver.c: add case for version 3.1
* src/vbox/vbox_tmpl.c: refactor common patterns into macros, support for
version 3.1, serial port configuration fix
* src/vbox/vbox_CAPI_v3_1.h, src/vbox/vbox_V3_1.c: generated code
Matthias Bolte [Thu, 3 Dec 2009 16:17:40 +0000 (17:17 +0100)]
Add virIndexToDiskName and fix mapping gap
esxVMX_IndexToDiskName handles indices up to 701. This limit comes
from a mapping gap in virDiskNameToIndex:
sdzy -> 700
sdzz -> 701
sdaaa -> 728
sdaab -> 729
This line in virDiskNameToIndex causes this gap:
idx = (idx + i) * 26;
Fixing it by altering this line to:
idx = (idx + (i < 1 ? 0 : 1)) * 26;
Also add a new version of virIndexToDiskName that handles the inverse
mapping for arbitrary indices.
* src/esx/esx_vmx.[ch]: remove esxVMX_IndexToDiskName
* src/util/util.[ch]: add virIndexToDiskName and fix mapping gap
* tests/esxutilstest.c: update test to verify that the gap is fixed
Matthias Bolte [Wed, 2 Dec 2009 23:03:06 +0000 (00:03 +0100)]
Fix some locking issues
* src/conf/domain_conf.c: don't call virDomainObjUnlock twice
* src/qemu/qemu_driver.c: relock driver lock if an error occurs in
qemuDomainObjBeginJobWithDriver, enter/exit monitor with driver
in qemudDomainSave
Fix event test timer checks on kernels with HZ=100
On kernels with HZ=100, the resolution of sleeps in poll() is
quite bad. Doing a precise check on the expiry time vs the
current time will thus often thing the timer has not expired
even though we're within 10ms of the expected expiry time. This
then causes another pointless sleep in poll() for <10ms. Timers
do not need to have such precise expiration, so we treat a timer
as expired if it is within 20ms of the expected expiry time. This
also fixes the eventtest.c test suite on kernels with HZ=100
* daemon/event.c: Add 20ms fuzz when checking for timer expiry
Wolfgang Mauerer [Tue, 10 Nov 2009 14:53:20 +0000 (15:53 +0100)]
Clarify documentation for private symbols
The instruction "See Makefile.am" in libvirt.private_syms
always makes me think that this file is autogenerated
and should not be touched manually. This patch spares
every reader of libvirt.private_syms the hassle of
reading Makefile.am before augmenting libvirt.private_syms.
Signed-off-by: Wolfgang Mauerer <wolfgang.mauerer@siemens.com> Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
* libvirt-override.c: Add many missing calls to allow threading
when entering C code, otherwise python blocks & then deadlocks
when we have an async event to dispatch back into python code.
Fix return value check for virDomainPinVcpu binding.
Matthias Bolte [Wed, 25 Nov 2009 23:09:40 +0000 (00:09 +0100)]
xen: Fix unconditional freeing in xenDaemonListDefinedDomains()
Commit 790f0b3057787bb64da8c46c111ff8d3eff7b2af causes the contents of
the names array to be freed even on success, resulting in no listing of
defined but inactive Xen domains.
Only read-only disks are supported with virtual FAT mode
* src/conf/domain_conf.c, src/conf/domain_conf.h: Add type="dir"
* docs/schemas/domain.rng: Document new disk type
* src/xen/xend_internal.c, src/xen/xm_internal.c: Raise error for
unsupported disk types
* tests/qemuxml2argvdata/qemuxml2argv-disk-cdrom-empty.args: Fix
empty disk file handling
* tests/qemuxml2argvdata/qemuxml2argv-disk-drive-fat.args,
tests/qemuxml2argvdata/qemuxml2argv-disk-drive-fat.xml,
tests/qemuxml2argvdata/qemuxml2argv-floppy-drive-fat.args,
tests/qemuxml2argvdata/qemuxml2argv-floppy-drive-fat.xml
tests/qemuxml2argvtest.c: Test QEMU vitual FAT driver
* src/qemu/qemu_conf.c: Support generating fat:/some/dir type
disk args
* src/security/security_selinux.c: Temporarily skip labelling
of directory based disks
Pull schedular affinity code out into a separate module
* src/Makefile.am: Add processinfo.h/processinfo.c
* src/util/processinfo.c, src/util/processinfo.h: Module providing
APIs for getting/setting process CPU affinity
* src/qemu/qemu_driver.c: Switch over to new APIs for schedular
affinity
* src/libvirt_private.syms: Export virProcessInfoSetAffinity
and virProcessInfoGetAffinity to internal drivers
* python/generator.py python/libvirt-override-api.xml
python/libvirt-override.c: implement the bindings for
virConnectListInterfaces() and virConnectListDefinedInterfaces()
Daniel Veillard [Wed, 18 Nov 2009 10:53:41 +0000 (11:53 +0100)]
514532 Fix man page, most operation are synchronous
* tools/virsh.pod: the man page was stating that most operations
are asynchronous while in fact most of them are synchronous except
domain shutdown, setvcpus and setmem.
Daniel Veillard [Wed, 18 Nov 2009 10:25:20 +0000 (11:25 +0100)]
512069 fix domain XML schemas for backward compatibility
For backward compatibility we used to add the tty path as
a tty attribute on console of type pty, duplicating the value
now found in source/@path, but the Relax-NG grammar wasn't
extended for this
<console type='pty' tty='/dev/pts/8'>
<source path='/dev/pts/8'/>
<target port='0'/>
</console>
* docs/schemas/domain.rng: allow an optional tty attribute
containing a devicePath
Steve Yarmie [Fri, 20 Nov 2009 14:41:05 +0000 (15:41 +0100)]
qemu-kvm needs -enable-kvm flag for VT optimization
Recent qemu releases require command option '-enable-qemu' in order
for the kvm functionality be activated. Libvirt needs to pass this flag
to qemu when starting a domain. Note that without the option,
even if both the kernel and qemu support KVM, KVM will not be activated
and VMs will be very slow.
* src/qemu/qemu_conf.h src/qemu/qemu_conf.c: parse the extra command
line option from help and add it when running kvm
* tests/qemuhelptest.c: this modified the flags output for qemu-0.10.5
and qemu-kvm-0.11.0-rc2 regression tests
Dave Allan [Thu, 19 Nov 2009 15:02:18 +0000 (16:02 +0100)]
Removing devicePath member from dev struct
I realized that I inadvertently added a member to the def struct to
contain each device's sysfs path when there was an existing member in the
dev struct for "OS specific path to device metadat, eg sysfs" Since the
udev backend needs to record the sysfs path while it's in the process of
creating the device, before the dev struct gets allocated, I chose to
remove the member from the dev struct.
* src/conf/node_device_conf.c src/conf/node_device_conf.h
src/node_device/node_device_driver.c src/node_device/node_device_hal.c
src/node_device/node_device_udev.c: remove devicePath from the
structure and use def->sysfs_path instead
Ensure driver lock is released when entering QEMU monitor
The qemudStartVMDaemon() and several functions it calls use
the QEMU monitor. The QEMU driver is locked while this function
is executing, so it is rquired to release the driver lock and
reacquire it either side of issuing a monitor command. It
failed todo so, leading to deadlock
* qemu/qemu_driver.c: Release driver when in qemudStartVMDaemon
and things it calls
Matthias Bolte [Sat, 14 Nov 2009 18:51:05 +0000 (19:51 +0100)]
esx: Fix MAC address formatting
VMware uses two MAC address prefixes: 00:0c:29 and 00:50:56. The 00:0c:29
prefix is used for ESX server generated addresses. The 00:50:56 prefix is
split into two parts. MAC addresses above 00:50:56:3f:ff:ff are generated
by a vCenter. The rest of the 00:50:56 prefix can be assigned manually.
Any MAC address within the 00:0c:29 and 00:50:56 prefix can be specified
in a domain XML config and the driver will handle the details internally.
* src/esx/esx_vmx.c: fix MAC address formatting
* tests/xml2vmxdata/*: update test files accordingly
Matthias Bolte [Fri, 13 Nov 2009 21:55:25 +0000 (22:55 +0100)]
esx: Handle 'vmxnet3' in esxVMX_FormatEthernet()
In commit 3c80fac2588cbc9e5ee7e7069e1ca4468f2359d3 'vmxnet3' handling
was added to esxVMX_ParseEthernet(), but not to the inverse function
esxVMX_FormatEthernet().
Matthias Bolte [Fri, 13 Nov 2009 21:43:10 +0000 (22:43 +0100)]
esx: Add documentation to the website
* docs/drivers.html.in: list the ESX driver
* docs/drvesx.html.in: the new ESX driver documentation
* docs/hvsupport.html.in: add the ESX driver to the matrix
* docs/index.html.in, docs/sitemap.html.in: list the ESX driver
* src/esx/esx_driver.c: fix and cleanup some comments
Jim Fehlig [Sun, 15 Nov 2009 08:40:36 +0000 (09:40 +0100)]
Fix compilation of libvirt against xen-unstable
* src/xen/xen_hypervisor.c: xen-unstable changeset 19788 removed
MAX_VIRT_CPUS from public headers, breaking compilation of libvirt
on -unstable. Its semanitc was retained with XEN_LEGACY_MAX_VCPUS.
Ensure MAX_VIRT_CPUS is defined accordingly.
Device kit support was removed, but the configure.ac checks were
left in place. A number of the XXX_REQUIRED=X.Y.Z variables were
not declared in the correct location (ie top of the file)
Fix incorrect reference counting logic in qemu monitor open
The QEMU monitor open method would not take a reference on
the virDomainObjPtr until it had successfully opened the
monitor. The cleanup code upon failure to open though would
call qemuMonitorClose() which would in turn decrement the
reference count. This caused the virDoaminObjPtr to be mistakenly
freed and then the whole driver crashes
* src/qemu/qemu_monitor.c: Fix reference counting in
qemuMonitorOpen
Don't return fatal error in HAL driver init if HAL isn't running
The HAL driver returns a fatal error code in the case where HAL
is not running. This causes the entire libvirtd daemon to quit
which isn't desirable. Instead it should simply disable the HAL
driver
* src/node_device/node_device_hal.c: Quietly disable HAL if it is
not running
Jamie Strandboge [Fri, 13 Nov 2009 14:25:30 +0000 (15:25 +0100)]
AppArmor handling of accesses to readonly files
Fixes https://launchpad.net/bugs/453335
* src/security/virt-aa-helper.c: suppress confusing and misleading
apparmor denied message when kvm/qemu tries to open a libvirt specified
readonly file (such as a cdrom) with write permissions. libvirt uses
the readonly attribute for the security driver only, and has no way
of telling kvm/qemu that the device should be opened readonly
Jamie Strandboge [Fri, 13 Nov 2009 14:22:20 +0000 (15:22 +0100)]
AppArmor require absolute paths
Fixes https://launchpad.net/bugs/460271
* src/security/virt-aa-helper.c: require absolute path for dynamic added
files. This is required by AppArmor and conveniently prevents adding
tcp consoles to the profile
Jamie Strandboge [Fri, 13 Nov 2009 14:19:05 +0000 (15:19 +0100)]
AppArmor updates of examples
* examples/apparmor/libvirt-qemu: adds pulseaudio, alsa and preliminary
save/restore to the example apparmor abstraction
* examples/apparmor/usr.sbin.libvirtd: allows libvirtd access to inet
dgram, inet6 dgram, inet6 stream and /usr/lib/libvirt/*
Check that domain is running when starting console
The 'virsh console' command did not check if the domain was
already running before attempting to fetch the XML and extract
the console PTY path. This caused a slightly unhelpful / misleading
error message for the user. The explicit check ensures the user
gets an explicit 'domain is not running' message.
* tools/virsh.c: Validate that state != VIR_DOMAIN_SHUTOFF in
virsh console command
In the scenario where the cgroups were mounted but the
particular group did not exist, and the caller had not
requested auto-creation, the code would fail to return
an error condition. This caused the lxc_controller to
think the cgroup existed, and it then later failed when
attempting to use it
* src/util/cgroup.c: Raise an error if the cgroup path does not
exist
There is a race condition in HAL driver startup where the callback
can get triggered before we have finished startup. This then causes
a deadlock in the driver.
* src/node_device/node_device_hal.c: RElease driver lock before
registering DBus callbacks
If the virDomainDefPtr object has an 'id' of -1, then forcably
set the VIR_DOMAIN_XML_INACTIVE flag to ensure generated XML
does not include any cruft from the previously running guest
such as console PTY path, or VNC port.
* src/conf/domain_conf.c: Set VIR_DOMAIN_XML_INACTIVE if
def->id is -1. Replace checks for def->id == -1 with
check against flags & VIR_DOMAIN_XML_INACTIVE.
Remove capng_lock() call when spawning LXC container init process
The capng_lock() call sets the SECURE_NO_SETUID_FIXUP and SECURE_NOROOT
bits on the process. This prevents the kernel granting capabilities to
processes with an effective UID of 0, or with setuid programs. This is
not actually what we want in the container init process. It should be
allowed to run setuid processes & keep capabilities when root. All that
is required is masking a handful of dangerous capabilities from the
bounding set.
The libvirtd initscript could get confused between the system and
session instances of the daemon. To avoid this it is neccessary
to check the pidfile explicitly.
* daemon/libvirtd.init.in: Always check the pidfile of the system
daemon to avoid confusion with the session daemons
Jamie Strandboge [Fri, 13 Nov 2009 10:04:23 +0000 (11:04 +0100)]
Fix virt-aa-helper when host and os.type arch differ
* src/security/virt-aa-helper.c: get_definition() now calls the new
caps_mockup() function which will parse the XML for os.type,
os.type.arch and then sets the wordsize. These attributes are needed
only to get a valid virCapsPtr for virDomainDefParseString(). The -H
and -b options are now removed from virt-aa-helper (they weren't used
yet anyway).
* tests/virt-aa-helper-test: extend and fixes tests, chmod'ed 755
David Allan [Thu, 12 Nov 2009 22:22:00 +0000 (23:22 +0100)]
Add translation of PCI vendor and product IDs
uses libpciaccess to provide human readable names for PCI vendor and
device IDs
* configure.in: add a requirement for libpciaccess >= 0.10.0
* src/Makefile.am: add the associated compilation flags and link
* src/node_device/node_device_udev.c: lookup the libpciaccess for
vendor name and product name based on their ids
David Allan [Thu, 12 Nov 2009 22:09:24 +0000 (23:09 +0100)]
Remove DevKit node device backend
* configure.in src/Makefile.am: remove the configuration check and
build instructions
* src/node_device/node_device_devkit.c: removed the module
* src/node_device/node_device_driver.c src/node_device/node_device_driver.h:
removed references to the old backend
David Allan [Thu, 12 Nov 2009 22:03:47 +0000 (23:03 +0100)]
Add scsi_target device type
* src/conf/node_device_conf.h src/conf/node_device_conf.c: add specific
support for SCSI target in node device capabilities
* src/node_device/node_device_udev.c: add some extra detection code
when handling udev output
David Allan [Thu, 12 Nov 2009 21:48:24 +0000 (22:48 +0100)]
Implement a node device backend using libudev
* configure.in: add new --with-udev, disabled by default, and requiring
libudev > 145
* src/node_device/node_device_udev.c src/node_device/node_device_udev.h:
the new node device backend
* src/node_device/node_device_linux_sysfs.c: moved node_device_hal_linux.c
to a better file name
* src/conf/node_device_conf.c src/conf/node_device_conf.h: add a couple
of fields in node device definitions, and an API to look them up,
remove a couple of unused fields from previous patch.
* src/node_device/node_device_driver.c src/node_device/node_device_driver.h:
plug the new driver
* po/POTFILES.in src/Makefile.am src/libvirt_private.syms: add the new
files and symbols
* src/util/util.h src/util/util.c: add a new convenience macro
virBuildPath and virBuildPathInternal() function
Cole Robinson [Thu, 12 Nov 2009 15:53:26 +0000 (10:53 -0500)]
Add virConnectGetLibvirtVersion API
There is currently no way to determine the libvirt version of a remote
libvirtd we are connected to. This is a useful piece of data to enable
feature detection.
* src/xen/xen_driver.c: Add support for VIR_MIGRATE_PERSIST_DEST flag
* src/xen/xend_internal.c: Add support for VIR_MIGRATE_UNDEFINE_SOURCE flag
* include/libvirt/virterror.h, src/util/virterror.c: Add new errorcode
VIR_ERR_MIGRATE_PERSIST_FAILED
pritesh [Thu, 12 Nov 2009 14:45:18 +0000 (15:45 +0100)]
Support for SATA Disks in virDomainDiskBus
* src/conf/domain_conf.h src/conf/domain_conf.c: add the new entry in
the enum and lists of virDomainDiskBus
* src/qemu/qemu_conf.c: same for virDomainDiskQEMUBus
The xenstore database sometimes has stale domain IDs which are not
present in the hypervisor anymore. Filter these out to avoid causing
confusion
* src/xen/xs_internal.c: Filter domain IDs against HV's list
* src/xen/xen_hypervisor.h, src/xen/xen_hypervisor.c: Add new
xenHypervisorHasDomain() method for checking ID validity
Jonas Eriksson [Fri, 9 Oct 2009 09:23:23 +0000 (10:23 +0100)]
Fix logic in xenUnifiedNumOfDomains to match xenUnifiedListDomains
The xenUnifiedNumOfDomains and xenUnifiedListDomains methods work
together as a pair, so it is critical they both apply the same
logic. With the current mis-matched logic it is possible to sometimes
get into a state when you miss certain active guests.
* src/xen/xen_driver.c: Change xenUnifiedNumOfDomains ordering to
match xenUnifiedListDomains.
Sometimes getaddrinfo returns IPv4 addresses before IPv6 addresses.
IPv6 sockets default to attempting to bind to IPv4 addresses too.
So if the IPv4 address is activated first, then binding to IPv6
will unneccessarily fail.
* daemon/libvirtd.c: Bind to IPv6 and IPv4 addresses separately
Fix save and restore with non-privileged guests and SELinux
When running qemu:///system instance, libvirtd runs as root,
but QEMU may optionally be configured to run non-root. When
then saving a guest to a state file, the file is initially
created as root, and thus QEMU cannot write to it. It is also
missing labelling required to allow access via SELinux.
* src/qemu/qemu_driver.c: Set ownership on save image before
running migrate command in virDomainSave impl. Call out to
security driver to set save image labelling
* src/security/security_driver.h: Add driver APIs for setting
and restoring saved state file labelling
* src/security/security_selinux.c: Implement saved state file
labelling for SELinux