Part of the resolution to:
https://bugzilla.redhat.com/show_bug.cgi?id=1003983
Although most devices available in qemu area defined as PCI devices,
and strictly speaking should only be attached via a PCI slot, in
practice qemu allows them to be attached to a PCIe slot and sometimes
this makes sense.
For example, The UHCI and EHCI USB controllers are usually attached
directly to the PCIe "root complex" (i.e. PCIe slots) on real
hardware, so that should be possible for a Q35-based qemu virtual
machine as well.
We still want to prefer a standard PCI slot when auto-assigning
addresses, though, and in general to disallow attaching PCI devices
via PCIe slots.
This patch makes that possible by adding a new
QEMU_PCI_CONNECT_TYPE_EITHER_IF_CONFIG flag. Three things are done
with this flag:
1) It is set for the "pcie-root" controller
2) qemuCollectPCIAddress() now has a set of nested switches that set
this "EITHER" flag for devices that we want to allow connecting to
pcie-root when specifically requested in the config.
3) qemuDomainPCIAddressFlagsCompatible() adds this new flag to the
"flagsMatchMask" if the address being checked came from config rather
than being newly auto-allocated by libvirt (this knowledge is
conveniently already available in the "fromConfig" arg).
Now any device having the EITHER flag set can be connected to
pcie-root if explicitly requested, but auto-allocated addresses for
those devices will still be standard PCI slots instead.
This patch only loosens the restrictions on devices that have been
specifically requested, but the setup is such that it should be fairly
easy to add new devices.
Replace them with switch cases. This will make it more efficient when
we add exceptions for more controller types, and other device types.
This is a prerequisite for patches to resolve:
https://bugzilla.redhat.com/show_bug.cgi?id=1003983
Packets sent by guests on virbrN, *or* by dnsmasq on the same, to
- 255.255.255.255/32 (netmask-independent local network broadcast
address), or to
- 224.0.0.0/24 (local subnetwork multicast range)
are never forwarded, hence it is not necessary to masquerade them.
In fact we must not masquerade them: translating their source addresses or
source ports (where applicable) may confuse receivers on virbrN.
One example is the DHCP client in OVMF (= UEFI firmware for virtual
machines):
http://thread.gmane.org/gmane.comp.bios.tianocore.devel/1506/focus=2640
It expects DHCP replies to arrive from remote source port 67. Even though
dnsmasq conforms to that, the destination address (255.255.255.255) and
the source address (eg. 192.168.122.1) in the reply allow the UDP
masquerading rule to match, which rewrites the source port to or above
1024. This prevents the DHCP client in OVMF from accepting the packet.
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=709418
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
The functions
- iptablesAddForwardDontMasquerade(),
- iptablesRemoveForwardDontMasquerade
handle exceptions in the masquerading implemented in the POSTROUTING chain
of the "nat" table. Such exceptions should be added as chronologically
latest, logically top-most rules.
The bridge driver will call these functions beginning with the next patch:
some special destination IP addresses always refer to the local
subnetwork, even though they don't match any practical subnetwork's
netmask. Packets from virbrN targeting such IP addresses are never routed
outwards, but the current rules treat them as non-virbrN-destined packets
and masquerade them. This causes problems for some receivers on virbrN.
Signed-off-by: Laszlo Ersek <lersek@redhat.com>
The previous patches added infrastructure to report better errors from
monitor in some cases. This patch finalizes this "feature" by enabling
this enhanced error reporting on early phases of VM startup. In these
phases the possibility of qemu producing a useful error message is
really high compared to running it during the whole life cycle. After
the start up is complete, the feature is disabled to provide the usual
error messages so that users are not confused by possibly irrelevant
messages that may be in the domain log.
The original motivation to do this enhancement is to capture errors when
using VFIO device passthrough, where qemu reports errors after the
monitor is initialized and the existing error catching code couldn't
catch this producing a unhelpful message:
# virsh start test
error: Failed to start domain test
error: Unable to read from monitor: Connection reset by peer
With this change, the message is changed to:
# virsh start test
error: Failed to start domain test
error: internal error: early end of file from monitor: possible problem:
qemu-system-x86_64: -device vfio-pci,host=00:1a.0,id=hostdev0,bus=pci.0,addr=0x5: vfio: error, group 8 is not viable, please ensure all devices within the iommu_group are bound to their vfio bus driver.
qemu-system-x86_64: -device vfio-pci,host=00:1a.0,id=hostdev0,bus=pci.0,addr=0x5: vfio: failed to get group 8
qemu-system-x86_64: -device vfio-pci,host=00:1a.0,id=hostdev0,bus=pci.0,addr=0x5: Device 'vfio-pci' could not be initialized
Change the monitor error code to add the ability to access the qemu log
file using a file descriptor so that we can dig in it for a more useful
error message. The error is now logged on monitor hangups and overwrites
a possible lesser error. A hangup on the monitor usualy means that qemu
has crashed and there's a significant chance it produced a useful error
message.
The functionality will be latent until the next patch.
Early VM startup errors usually produce a better error message in the
machine log file. Currently we were accessing it only when the process
exited during certain phases of startup. This will help adding a more
comprehensive error extraction for early qemu startup phases.
This patch adds infrastructure to keep a file descriptor for the machine
log file that will be used in case an error happens.
Teach the function to skip character device definitions printed by qemu
at startup in addition to libvirt log messages and make it usable from
outside of qemu_process.c. Also add documentation about the func.
The parsing of '-usb' did not check for failure of the
virDomainControllerInsert method. As a result on OOM, the
parser mistakenly attached USB disks to the IDE controller.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The code formatting NUMA args was ignoring the return value
of virBitmapFormat, so on OOM, it would silently drop the
NUMA cpumask arg.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
When building boot menu args, if OOM occurred the CLI args
would end up containing 'order=(null)' due to a missing
call to 'virBufferError'.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
On win32, using text mode for binary files might result in short
reads since ASCII character 0x1A is interpreted as EOF. Also, it
could lead to problems using the seek functions because of the \r
handling.
Signed-off-by: Claudio Bley <cbley@av-test.de>
N.B. This had no ill effects as long as O_RDONLY is defined to
to be 0, such that the expression (O_RDONLY < 0) yielded 0
again.
Signed-off-by: Claudio Bley <cbley@av-test.de>
The virCommandAddEnvPassCommon method ignored the failure to
pre-allocate the env variable array with VIR_RESIZE_N. While
this is harmless, it confuses the test harness which is trying
to validate OOM handling of every individual allocation call.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
When the various viralloc.c functions were changed to use the
normal error reporting code, the OOM injection code paths
were not updated to report errors.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The qemuParseCommandLine method did not check the return value of
virStringSplit to see if OOM had occurred. This lead to dereference
of a NULL pointer on OOM.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Most callers of qemuParseKeywords were assigning its return
value to a 'size_t' variable. Then then also checked '< 0'
for error condition, but this will never be true with the
unsigned size_t variable. Rather than using 'ssize_t', change
qemuParseKeywords so that the element count is returned via
an output parameter, leaving the return value solely as an
error indicator.
This avoids a crash accessing beyond the end of an error
upon OOM.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
In
commit 41b5505679
Author: Eric Blake <eblake@redhat.com>
Date: Wed Aug 28 15:01:23 2013 -0600
qemu: simplify list cleanup
The qemuStringToArgvEnv method was changed to use virStringFreeList
to free the 'arglist' array. This method assumes the string list
array is NULL terminated, however, qemuStringToArgvEnv was not
ensuring this when populating 'arglist'. This caused an out of
bounds access by virStringFreeList when OOM occured in the initial
loop of qemuStringToArgvEnv
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
When parsing the RBD hosts, it increments the 'nhosts' counter
before increasing the 'hosts' array allocation. If an OOM then
occurs when increasing the array allocation, the cleanup block
will attempt to access beyond the end of the array. Switch
to using VIR_EXPAND_N instead of VIR_REALLOC_N to protect against
this mistake
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If OOM occurs in qemuDomainCCWAddressSetCreate, it jumps to
a cleanup block and frees the partially initialized object.
It then mistakenly returns the address of the just free'd
pointer instead of NULL.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If a OOM error occurs in virGetConnect, this may cause the
virConnectDispose method to de-reference a NULL pointer,
since the close callback will not have been initialized.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The virDomainDefParseXML method did not check the return value
of the virBitmapNew API call for NULL. This lead to a crash on
OOM
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If an OOM error occurs in virSecurityDeviceLabelDefParseXML the
cleanup code may free an uninitialized pointer, causing a crash
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
To allow creation of a virNetSocketPtr instance from a pre-opened
socketpair FD, add a virNetSocketNewConnectSockFD method.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The new function virConnectGetCPUModelNames allows to retrieve the list
of CPU models known by the hypervisor for a specific architecture.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
The fix for CVE-2013-4311 had a pre-requisite enhancement
to the identity code
commit db7a5688c0
Author: Daniel P. Berrange <berrange@redhat.com>
Date: Thu Aug 22 16:00:01 2013 +0100
Also store user & group ID values in virIdentity
This had a typo which caused the group ID to overwrite the
user ID string. This meant any checks using this would have
the wrong ID value. This only affected the ACL code, not the
initial polkit auth. It also leaked memory.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If a dir does not exist, raise an immediate error in logs
rather than letting virFileResolveAllLinks fail, since this
gives better error reporting to the user.
Signed-off-by: Chen Hanxiao <chenhanxiao@cn.fujitsu.com>
When FUSE is enabled, the LXC container is setup with
a custom /proc/meminfo file. This file uses "KB" as a
suffix, rather than "kB" which is the kernel's style.
Fix this inconsistency to avoid confusing apps.
Signed-off-by: Chen Hanxiao <chenhanxiao@cn.fujitsu.com>
The ABI compatibility check for domain features didn't check the
expanded HyperV and APIC EOI values, thus possibly allowing change in
guest ABI.
Add the check and use typecasted switch statement to warn developers
when adding a new HyperV feature.
Since the wait is done during migration (still inside
QEMU_ASYNC_JOB_MIGRATION_OUT), the code should enter the monitor as such
in order to prohibit all other jobs from interfering in the meantime.
This patch fixes bug #1009886 in which qemuDomainGetBlockInfo was
waiting on the monitor condition and after GetSpiceMigrationStatus
mangled its internal data, the daemon crashed.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1009886
This splits up the version parsing code into a callable API like QEMU
help/version string parsing so that we can test it as we need to add
additional patterns for newer versions/products.
Rather than looking up the path to vmrun each time we call it, look it
up once and save it. This sets up the ability for us to detect where the
path is on Mac OS X and not have to look it up each time we execute it.
The VMware driver supports multiple backends for the VMware Player and
VMware Workstation, convert this logic into enum and use VIR_ENUM_IMPL()
to provide conversions to and from strings.
This resolves https://bugzilla.redhat.com/show_bug.cgi?id=1008903
The Q35 machinetype has an implicit SATA controller at 00:1F.2 which
isn't given the "expected" id of ahci0 by qemu when it's created. The
original suggested solution to this problem was to not specify any
controller for the disks that use the default controller and just
specify "unit=n" instead; qemu should then use the first IDE or SATA
controller for the disk.
Unfortunately, this "solution" is ignorant of the fact that in the
case of SATA disks, the "unit" attribute in the disk XML is actually
*not* being used for the unit, but is instead used to specify the
"bus" number; each SATA controller has 6 buses, and each bus only
allows a single unit. This makes it nonsensical to specify unit='n'
where n is anything other than 0. It also means that the only way to
connect more than a single device to the implicit SATA controller is
to explicitly give the bus names, which happen to be "ide.$n", where
$n can be replaced by the disk's "unit" number.
In commit 6d41cb8, the interface for virEventAddHandleFunc was changed.
This patch updates the documentation for virEventAddHandle to reflect
the new significance of the return value. Also, both functions now
mention -1 for failure.
With the existing pkcheck (pid, start time) tuple for identifying
the process, there is a race condition, where a process can make
a libvirt RPC call and in another thread exec a setuid application,
causing it to change to effective UID 0. This in turn causes polkit
to do its permission check based on the wrong UID.
To address this, libvirt must get the UID the caller had at time
of connect() (from SO_PEERCRED) and pass a (pid, start time, uid)
triple to the pkcheck program.
This fix requires that libvirt is re-built against a version of
polkit that has the fix for its CVE-2013-4288, so that libvirt
can see 'pkg-config --variable pkcheck_supports_uid polkit-gobject-1'
Signed-off-by: Colin Walters <walters@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The polkit access driver will want to use the process start
time field. This was already set for network identities, but
not for the system identity.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Future improvements to the polkit code will require access to
the numeric user ID, not merely user name.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
virDomainSetBlockIoTuneEnsureACL was incorrectly called after we already
started a job. As a result of this, the job was not cleaned up when an
access driver had forbidden the action.
Useful to set custom forwarders instead of using the contents of
/etc/resolv.conf. It helps me to setup dnsmasq as local nameserver to
resolve VM domain names from domain 0, when domain option is used.
Signed-off-by: Diego Woitasen <diego.woitasen@vhgroup.net>
Signed-off-by: Eric Blake <eblake@redhat.com>
VMWare Fusion 5 can set the CD-ROM's device name to be 'auto detect' when
using the physical drive via 'cdrom-raw' device type. VMWare will then
connect to first available host CD-ROM to the virtual machine upon start
up according to VMWare documentation. If no device is available, it
appears that the device will remain disconnected.
To better model this a CD-ROM that is marked as "auto detect" when in
the off state would be modeled as the following with this patch:
<disk type='block' device='lun'>
<source startupPolicy='optional'/>
<target dev='hda' bus='ide'/>
<address type='drive' controller='0' bus='0' target='0' unit='0'/>
</disk>
Once the domain transitions to the powered on state, libvirt can
populate the remaining source data with what is connected, if anything.
However future power cycles, the domain may not always start with that
device attached.
Currently the XML parser already allows the following syntax:
<disk type='block' device='cdrom'>
<source startupPolicy='optional'/>
<target dev='hda' bus='ide'/>
<address type='drive' controller='0' bus='0' target='0' unit='0'/>
</disk>
But it if the dev value is NULL then it would not have the leading
"<source ", resulting in invalid XML.
qemu/KVM also supports a tftp URL while specifying the cdrom ISO image.
The xml should be as following:
<disk type='network' device='cdrom'>
<source protocol='tftp' name='/url/path'>
<host name='host.name' port='69'/>
</source>
</disk>
Signed-off-by: Aline Manera <alinefm@br.ibm.com>
The ftps protocol is another protocol supported by qemu/KVM while specifying
the cdrom ISO image.
The xml should be as following:
<disk type='network' device='cdrom'>
<source protocol='ftps' name='/url/path'>
<host name='host.name' port='990'/>
</source>
</disk>
Signed-off-by: Aline Manera <alinefm@br.ibm.com>
The https protocol is also accepted by qemu/KVM when specifying the cdrom ISO
image.
The xml should be as following:
<disk type='network' device='cdrom'>
<source protocol='https' name='/url/path'>
<host name='host.name' port='443'/>
</source>
</disk>
Signed-off-by: Aline Manera <alinefm@br.ibm.com>
GCC 4.8.0+ whines about variable "new" being uninitialized since
commit 73bfac0e71. This is a false positive as the
xmlFreeNode(new) statement can be only reached if new was actually
allocated successfully.
CC conf/libvirt_conf_la-domain_conf.lo
conf/domain_conf.c: In function 'virDomainDefSetMetadata':
conf/domain_conf.c:18650:24: error: 'new' may be used uninitialized in this function [-Werror=maybe-uninitialized]
xmlFreeNode(new);
Reported independently by John Ferlan and Michal Privoznik.
Commit 073e1575 tried to set things up so that 1) generated files
to be shipped in the tarball always live in srcdir, and 2) we have
no files in SOURCES that depend on any other files with a literal
$(srcdir) in the name, because that situation can cause confusing
results for the make expansion of $@ depending on whether the file
is found locally or via VPATH. But all my testing for that patch
was done incrementally, where all the protocol.[ch] files had
already been generated prior to the patch and were up-to-date in
the srcdir, and thus I missed one case where $@ causes grief in a
VPATH build from a fresh checkout:
We have a pattern rule for generating remote_protocol.[ch], and
what's more, the rule for protocol.c depends on protocol.h AND
on the protocol.x file. The pattern for protocol.c is only
satisfied via the VPATH lookup for protocol.x, and if protocol.h
doesn't yet exist, the VPATH rule kicks in and we end up with a
dependency on a file with $(srcdir) in the name. Based on make's
rules for $@, this resulted in make building remote_protocol.h
into srcdir (where we want it), then remote_protocol.c into
builddir (oops, not so good for the tarball), and also causes
the build to fail (the compiler can't find the .h if it lives
in a different directory than the .c):
CC remote/libvirt_driver_remote_la-remote_protocol.lo
remote/remote_protocol.c:7:29: fatal error: remote_protocol.h: No such file or directory
#include "remote_protocol.h"
^
compilation terminated.
As before, the fix is to hard-code the output file to go into
srcdir in spite of $@; but since this is in a pattern rule, we
are forced to use $@ in the recipe, so the patch is a bit
trickier than what was done in commit 073e1575.
* src/Makefile.am (%protocol.c, %protocol.h): Force output to srcdir.
Signed-off-by: Eric Blake <eblake@redhat.com>
Eric Blake suggested that we could do a little better in case copying of
the metadata to be set fails. With this patch, the old metadata is
discarded after the new string is copied successfuly.
If the ABI compatibility check with the "migratable" user XML is
successful, we would leak the originally parsed XML from the user that
would not be used in this case.
Reported by Ján Tomko.
virDomainSetMetadata when operating on the metadata element was
requesting the @key argument to be passed even if @metadata was NULL
used to delete the corresponding metadata element. This is not needed as
the key is only used when adding the element and matching is done via
the XML namespace.
The virDomainGetMetadata function was designed to support also retrieval
of app specific metadata from the <metadata> element. This functionality
was never implemented originally.
The function implemented common behavior that can be reused for other
hypervisor drivers that use the virDomainObj data structures. Factor out
the core into a separate helper func.
The function implemented common behavior that can be reused for other
hypervisor drivers that use the virDomainObj data structures. Factor out
the core into a separate helper func.
In the original implementation of external checkpoints I've mistakenly
used the live definition to be stored in the save image. The normal
approach is to use the "migratable" definition. This was discovered when
commit 07966f6a8b changed the behavior to
use a converted XML from the user to do the compatibility check to fix
problem when using the regular machine saving.
As the previous patch added a compatibility layer, we can now change the
type of the XML in the image.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1008340
External checkpoints have a bug in the implementation where they use the
normal definition instead of the "migratable" one. This causes errors
when the snapshot is being reverted using the workaround method via
qemuDomainRestoreFlags() with a custom XML. This issue was introduced
when commit 07966f6a8b changed the code to
compare "migratable" XMLs from the user as we should have used
migratable in the image too.
This patch adds a compatibility layer, so that fixing the snapshot code
won't make existing snapshots fail to load.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1008340
CD-ROMs and Floppies are allowed to have no source to imply they are
empty or disconnected. Since the LUN type is used for raw CD-ROM access
with QEMU (and VMWare in the future), it also needs to allow an empty
source when the raw CD-ROM device is disconnected from the domain.
qemuMigrationEatCookie has flags to control if these should
be parsed, but it does not fill mig->flags. These cookies might
get leaked if these flags are not set by qemuMigrationBakeCookie.
42 (32 direct, 10 indirect) bytes in 1 blocks are definitely lost in
loss record 361 of 662
==123== by 0x1BA33FCA: qemuMigrationEatCookie (qemu_migration.c:678)
==123== by 0x1BA34A1E: qemuMigrationRun (qemu_migration.c:3108)
==123== by 0x1BA3622B: doNativeMigrate (qemu_migration.c:3343)
==123== by 0x1BA3B408: qemuMigrationPerform (qemu_migration.c:4138)
https://bugzilla.redhat.com/show_bug.cgi?id=1008619
1,003 bytes in 1 blocks are definitely lost in loss record 599 of 635
==404== by 0x50728A7: virBufferAddChar (virbuffer.c:185)
==404== by 0x50BC466: virSystemdEscapeName (virsystemd.c:67)
==404== by 0x50BC6B2: virSystemdMakeSliceName (virsystemd.c:108)
==404== by 0x50BC870: virSystemdCreateMachine (virsystemd.c:169)
==404== by 0x5078267: virCgroupNewMachine (vircgroup.c:1498)
Bother those kernel developers. In the latest rawhide, kernel
and glibc have now been unified so that <netinet/in.h> and
<linux/in6.h> no longer clash; but <linux/if_bridge.h> is still
not self-contained. Because of the latest header change, the
build is failing with:
checking for linux/param.h... no
configure: error: You must install kernel-headers in order to compile libvirt with QEMU or LXC support
with details:
In file included from conftest.c:561:0:
/usr/include/linux/in6.h:71:18: error: field 'flr_dst' has incomplete type
struct in6_addr flr_dst;
We need a workaround to avoid our workaround :)
* configure.ac (NETINET_LINUX_WORKAROUND): New test.
* src/util/virnetdevbridge.c (includes): Use it.
Signed-off-by: Eric Blake <eblake@redhat.com>
Since virnetsocket conditionally uses selinux we need to link against it
otherwise the build fails with:
CCLD libvirtd
/usr/bin/ld: ../src/.libs/libvirt-lxc.so: undefined reference to symbol 'freecon'
/lib/i386-linux-gnu/libselinux.so.1: error adding symbols: DSO missing from command line
collect2: error: ld returned 1 exit status
make[3]: *** [libvirtd] Error 1
An off-list bug report mentioned some confusion where the public
documentation of libvirt.c:virConnectGetHostname did not match
the private documentation of util/virutil.c:virGetHostname.
* src/libvirt.c (virConnectGetHostname): Tweak docs.
Signed-off-by: Eric Blake <eblake@redhat.com>
The VIR_ACCESS_PERM_CONNECT_DETECT_STORAGE_POOLS enum
constant had its string format be 'detect_storage_pool',
note the missing trailing 's'. This prevent the ACL
check from ever succeeding. Fix this and add a simple
test script to validate this problem of matching names.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Right now we mount selinuxfs even user namespace is enabled and
ignore the error. But we shouldn't ignore these errors when user
namespace is not enabled.
This patch skips mounting selinuxfs when user namespace enabled.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
When reverting a live internal snapshot with a live guest the ABI
compatiblity check was comparing a "migratable" definition with a normal
one. This resulted in the check failing with:
revert requires force: Target device address type none does not match source pci
This patch generates a "migratable" definition from the actual one to
check against the definition from the snapshot to avoid this problem.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1006886
This resolves: https://bugzilla.redhat.com/show_bug.cgi?id=983026
The netcf interface driver previously had no state driver associated
with it - as a connection was opened, it would create a new netcf
instance just for that connection, and close it when it was
finished. the problem with this is that each connection to libvirt
used up a netlink socket, and there is a per process maximum of ~1000
netlink sockets.
The solution is to create a state driver to go along with the netcf
driver. The state driver will opens a netcf instance, then all
connections share that same netcf instance, thus only a single
netlink socket will be used no matter how many connections are mde to
libvirtd.
This was rather simple to do - a new virObjectLockable class is
created for the single driverState object, which is created in
netcfStateInitialize and contains the single netcf handle; instead of
creating a new object for each client connection, netcfInterfaceOpen
now just increments the driverState object's reference count and puts
a pointer to it into the connection's privateData. Similarly,
netcfInterfaceClose() just un-refs the driverState object (as does
netcfStateCleanup()), and virNetcfInterfaceDriverStateDispose()
handles closing the netcf instance. Since all the functions already
have locking around them, the static lock functions used by all
functions just needed to be changed to call virObjectLock() and
virObjectUnlock() instead of directly calling the virMutex* functions.
This better fits the modern naming scheme in libvirt, and anticipates
an upcoming change where a single instance of this state will be
maintained by a separate state driver, and every instance of the netcf
driver will share the same state.
If the guest is configured with
<filesystem type='mount'>
<source dir='/'/>
<target dir='/'/>
<readonly/>
</filesystem>
Then any submounts under / should also end up readonly, except
for those setup as basic mounts. eg if the user has /home on a
separate volume, they'd expect /home to be readonly, but we
should not touch the /sys, /proc, etc dirs we setup ourselves.
Users can selectively make sub-mounts read-write again by
simply listing them as new mounts without the <readonly>
flag set
<filesystem type='mount'>
<source dir='/home'/>
<target dir='/home'/>
</filesystem>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Move the array of basic mounts out of the lxcContainerMountBasicFS
function, to a global variable. This is to allow it to be referenced
by other methods wanting to know what the basic mount paths are.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Debian systems may run the 'systemd-logind' daemon, which causes the
/sys/fs/cgroup/systemd mount to be setup, but no other cgroup
controllers are created. While the LXC driver considers cgroups to
be mandatory, the QEMU driver is supposed to accept them as optional.
We detect whether they are present by looking in /proc/mounts for
any mounts of type 'cgroups', but this is not sufficient. We need to
skip any named mounts (as seen by a name=XXX string in the mount
options), so that we only detect actual resource controllers.
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=721979
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The polkit access driver used the wrong permission names for checks
on storage pools, volumes and node devices. This led to them always
being denied access.
The 'dettach' permission was also mis-spelt and should have been
'detach'. While permission names are ABI sensitive, the fact that
the code used the wrong object name for checking node device
permissions, means that no one could have used the mis-spelt
'dettach' permission.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
This fixes the description of virConnectGetType() API function in
API documentation to match the real functionality that it can be
used to get driver name, and provide a hint on how to learn about
full capabilities.
Signed-off-by: Michal Novotny <minovotn@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
This patch introduces virDBusIsServiceEnabled, we can use
this method to get if the service is supported.
In one case, if org.freedesktop.machine1 is unavailable on
host, we should skip creating machine through systemd.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
The devpts, dev and fuse filesystems are mounted temporarily.
there is no need to export them to container if container shares
the root directory with host.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
Some users in Ubuntu/Debian seem to have a setup where all the
cgroup controllers are mounted on /sys/fs/cgroup rather than
any /sys/fs/cgroup/<controller> name. In the loop which detects
which controllers are present for a mount point we were modifying
'mnt_dir' field in the 'struct mntent' var, but not always restoring
the original value. This caused detection to break in the all-in-one
mount setup.
Fix that logic bug and add test case coverage for this mount
setup.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
After freeing the bitmap pointer, it must set the pointer to NULL.
This will avoid any other use of the freed memory of the bitmap pointer.
https://bugzilla.redhat.com/show_bug.cgi?id=1006710
Signed-off-by: Liuji (Jeremy) <jeremy.liu@huawei.com>
Osier Yang pointed out that ever since commit 31cb030, the
signature of qemuDomainObjEndJob was changed to return a bool.
While comparison against 0 or > 0 still gives the right results,
it looks fishy; we also had one place that was comparing < 0
which is effectively dead code.
* src/qemu/qemu_migration.c (qemuMigrationPrepareAny): Fix dead
code bug.
(qemuMigrationBegin): Use more canonical form of bool check.
* src/qemu/qemu_driver.c (qemuAutostartDomain)
(qemuDomainCreateXML, qemuDomainSuspend, qemuDomainResume)
(qemuDomainShutdownFlags, qemuDomainReboot, qemuDomainReset)
(qemuDomainDestroyFlags, qemuDomainSetMemoryFlags)
(qemuDomainSetMemoryStatsPeriod, qemuDomainInjectNMI)
(qemuDomainSendKey, qemuDomainGetInfo, qemuDomainScreenshot)
(qemuDomainSetVcpusFlags, qemuDomainGetVcpusFlags)
(qemuDomainRestoreFlags, qemuDomainGetXMLDesc)
(qemuDomainCreateWithFlags, qemuDomainAttachDeviceFlags)
(qemuDomainUpdateDeviceFlags, qemuDomainDetachDeviceFlags)
(qemuDomainBlockResize, qemuDomainBlockStats)
(qemuDomainBlockStatsFlags, qemuDomainMemoryStats)
(qemuDomainMemoryPeek, qemuDomainGetBlockInfo)
(qemuDomainAbortJob, qemuDomainMigrateSetMaxDowntime)
(qemuDomainMigrateGetCompressionCache)
(qemuDomainMigrateSetCompressionCache)
(qemuDomainMigrateSetMaxSpeed)
(qemuDomainSnapshotCreateActiveInternal)
(qemuDomainRevertToSnapshot, qemuDomainSnapshotDelete)
(qemuDomainQemuMonitorCommand, qemuDomainQemuAttach)
(qemuDomainBlockJobImpl, qemuDomainBlockCopy)
(qemuDomainBlockCommit, qemuDomainOpenGraphics)
(qemuDomainGetBlockIoTune, qemuDomainGetDiskErrors)
(qemuDomainPMSuspendForDuration, qemuDomainPMWakeup)
(qemuDomainQemuAgentCommand, qemuDomainFSTrim): Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
Automake 2.0 will enable subdir-objects by default; in preparation
for that change, automake 1.14 outputs LOADS of warnings:
daemon/Makefile.am:38: warning: source file '../src/remote/remote_protocol.c' is in a subdirectory,
daemon/Makefile.am:38: but option 'subdir-objects' is disabled
automake-1.14: warning: possible forward-incompatibility.
automake-1.14: At least a source file is in a subdirectory, but the 'subdir-objects'
automake-1.14: automake option hasn't been enabled. For now, the corresponding output
automake-1.14: object file(s) will be placed in the top-level directory. However,
automake-1.14: this behaviour will change in future Automake versions: they will
automake-1.14: unconditionally cause object files to be placed in the same subdirectory
automake-1.14: of the corresponding sources.
automake-1.14: You are advised to start using 'subdir-objects' option throughout your
automake-1.14: project, to avoid future incompatibilities.
daemon/Makefile.am:38: warning: source file '../src/remote/lxc_protocol.c' is in a subdirectory,
daemon/Makefile.am:38: but option 'subdir-objects' is disabled
...
As automake 1.9 also supported this option, and the previous patches
fixed up the code base to work with it, it is safe to now turn it on
unconditionally.
* configure.ac (AM_INIT_AUTOMAKE): Enable subdir-objects.
* .gitignore: Ignore .dirstamp directories.
* src/Makefile.am (PDWTAGS, *-protocol-struct): Adjust to
new subdir-object location of .lo files.
Signed-off-by: Eric Blake <eblake@redhat.com>
We have been adding new .x files without keeping the list of
*-structs files up-to-date. This adds the support for the
recent additions.
In the process of testing this, I also noticed that Fedora 19's
use of dwarves-1.10 (providing pdwtags version 1.9) was producing
a single line on stderr but still giving enough useful info on
stdout that we could check structs; the real goal of checking
stderr separately from stdout was to avoid the bug in dwarves-1.9
where stdout was empty (see bug http://bugzilla.redhat.com/772358).
* src/Makefile.am (struct_prefix, PROTOCOL_STRUCTS): Add missing
struct tests.
(PDWTAGS): Work with Fedora 19 pdwtags.
(lxc_monitor_protocol-struct, lock_protocol-struct): New rules.
* src/lxc_monitor_protocol-structs: New file.
* src/lock_protocol-structs): Likewise.
* cfg.mk (generated_files): Enlarge list.
Signed-off-by: Eric Blake <eblake@redhat.com>
Trying to enable automake's subdir-objects option resulted in
the creation of literal directories such as src/$(srcdir)/remote/.
I traced this to the fact that we had used a literal $(srcdir)
in a location that later fed an automake *_SOURCES variable.
This has also been reported as an automake bug:
http://debbugs.gnu.org/cgi/bugreport.cgi?bug=13928
but it's better to fix our code than to wait for an automake fix.
Some things to remember that affect VPATH builds, and where an
in-tree build is blissfully unaware of the issues: if a VPATH
build fails to find a file that was used as a prereq of any
other target, then the rule for that file will expand $@ to
prefer the current build dir (bad because a VPATH build on a
fresh checkout will then stick $@ in the current directory
instead of the desired srcdir); conversely, if a VPATH build
finds the file in srcdir but decides it needs to be rebuilt,
then the rule for that file will expand $@ to include the
directory where it was found out-of-date (bad for an explicit
listing of $(srcdir)/$@ because an incremental VPATH build will
then expand srcdir twice). As we want these files to go into
srcdir unconditionally, we have to massage or avoid $@ for any
recipe that involves one of these files.
Therefore, this patch removes all uses of $(srcdir) from any
generated file name that later feeds a *_SOURCES variable, and
then rewrites all the recipes to generate those files to
hard-code their creation into srcdir without the use of $@.
* src/Makefile.am (REMOTE_DRIVER_GENERATED): Drop $(srcdir); VPATH
builds know how to find the files, and automake subdir-objects
fails with it in place.
(LXC_MONITOR_PROTOCOL_GENERATED, (LXC_MONITOR_GENERATED)
(ACCESS_DRIVER_GENERATED, LOCK_PROTOCOL_GENERATED): Likewise.
(*_client_bodies.h): Hard-code rules to write into srcdir, as
VPATH tries to build $@ locally if missing.
(util/virkeymaps.h): Likewise.
(lxc/lxc_monitor_dispatch.h): Likewise.
(access/viraccessapi*): Likewise.
(locking/lock_daemon_dispatch_stubs.h): Likewise.
* daemon/Makeflie.am (DAEMON_GENERATED, remote_dispatch.h):
Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
fixup DAEMON_GENERATED
Failure to attach to a domain during 'virsh qemu-attach' left
the list of domains in an odd state:
$ virsh qemu-attach 4176
error: An error occurred, but the cause is unknown
$ virsh list --all
Id Name State
----------------------------------------------------
2 foo shut off
$ virsh qemu-attach 4176
error: Requested operation is not valid: domain is already active as 'foo'
$ virsh undefine foo
error: Failed to undefine domain foo
error: Requested operation is not valid: cannot undefine transient domain
$ virsh shutdown foo
error: Failed to shutdown domain foo
error: invalid argument: monitor must not be NULL
It all stems from leaving the list of domains unmodified on
the initial failure; we should follow the lead of createXML
which removes vm on failure (the actual initial failure still
needs to be fixed in a later patch, but at least this patch
gets us to the point where we aren't getting stuck with an
unremovable "shut off" transient domain).
While investigating, I also found a leak in qemuDomainCreateXML;
the two functions should behave similarly. Note that there are
still two unusual paths: if dom is not allocated, the user will
see an OOM error even though the vm remains registered (but oom
errors already indicate tricky cleanup); and if the vm starts
and then quits again all before the job ends, it is possible
to return a non-NULL dom even though the dom will no longer be
useful for anything (but this at least lets the user know their
short-lived vm ran).
* src/qemu/qemu_driver.c (qemuDomainCreateXML): Don't leak vm on
failure to obtain job.
(qemuDomainQemuAttach): Match cleanup of qemuDomainCreateXML.
Signed-off-by: Eric Blake <eblake@redhat.com>
ARM v7 can operate in either little or big endian modes. Add
support for the big-endian version known as armv7b from uname.
Signed-off-by: Yogesh Tillu <tillu.yogesh@gmail.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Currently, only X86 provides users CPU features with CPUID instruction.
If users specify the features for non-x86, it should tell users to
remove them.
This patch is to report one error if features are specified by
users for non-x86 platform.
Signed-off-by: Li Zhang <zhlcindy@linux.vnet.ibm.com>
While debugging a failure of 'virsh qemu-attach', I noticed that
we were leaking the count of active domains on failure. This
means that a libvirtd session that is supposed to quit after
active domains disappear will hang around forever.
* src/qemu/qemu_process.c (qemuProcessAttach): Undo count of
active domains on failure.
Signed-off-by: Eric Blake <eblake@redhat.com>
In Fedora 19, 'qemu-kvm' is a simple wrapper that calls
'qemu-system-x86_64 -machine accel=kvm'. Attempting
to use 'virsh qemu-attach $pid' to a machine started as:
qemu-kvm -cdrom /var/lib/libvirt/images/foo.img \
-monitor unix:/tmp/demo,server,nowait -name foo \
--uuid cece4f9f-dff0-575d-0e8e-01fe380f12ea
was failing with:
error: XML error: No PCI buses available
because we did not see 'kvm' in the executable name read from
/proc/$pid/cmdline, and tried to assign os.machine as
"accel=kvm" instead of "pc"; this in turn led to refusal to
recognize the pci bus.
Noticed while investigating https://bugzilla.redhat.com/995312
although there are still other issues to fix before that bug
will be completely solved.
I've concluded that the existing parser code for native-to-xml
is a horrendous hodge-podge of ad-hoc approaches; I basically
rewrote the -machine section to be a bit saner.
* src/qemu/qemu_command.c (qemuParseCommandLine): Don't assume
-machine argument is always appropriate for os.machine; set
virtType if accel is present.
Signed-off-by: Eric Blake <eblake@redhat.com>
'virsh domxml-from-native' and 'virsh qemu-attach' could misbehave
for an emulator installed in (a somewhat unlikely) location
such as /usr/local/qemu-1.6/qemu-system-x86_64 or (an even less
likely) /opt/notxen/qemu-system-x86_64. Limit the strstr seach
to just the basename of the file where we are assuming details
about the binary based on its name.
While testing, I accidentally triggered a core dump during strcmp
when I forgot to set os.type on one of my code paths; this patch
changes such a coding error to raise a nicer internal error instead.
* src/qemu/qemu_command.c (qemuParseCommandLine): Compute basename
earlier.
* src/conf/domain_conf.c (virDomainDefPostParseInternal): Avoid
NULL deref.
Signed-off-by: Eric Blake <eblake@redhat.com>
The regular expression used to determine guest capabilities
was compiled in libxlCapsInitHost() but used in libxlCapsInitGuests().
Move compilation to libxlCapsInitGuests() where it is used, and free
the compiled regex after use. Ensure not to free the regex if
compilation fails.
On Power platform, Power7+ can support Power7 guest.
It needs to define XML configuration to specify guest's CPU model.
For exmaple:
<cpu match='exact'>
<model>POWER7_v2.1</model>
<vendor>IBM</vendor>
</cpu>
Signed-off-by: Li Zhang <zhlcindy@linux.vnet.ibm.com>
CPU features are not supported on non-x86 and hasFeatures will be NULL.
This patch is to remove CPU features functions calling to avoid errors.
Signed-off-by: Li Zhang <zhlcindy@linux.vnet.ibm.com>
This patch changes virFileLoopDeviceOpen() to use the new loop-control
device to allocate a new loop device. If this behavior is unsupported
we fall back to the previous method of searching /dev for a free device.
With this patch you can start as many image based LXC domains as you
like (well almost).
Fixes bug https://bugzilla.redhat.com/show_bug.cgi?id=995543
Right now, securityfs is disallowed to be mounted in non-initial
user namespace, so we must avoid trying to mount securityfs in
a container which has user namespace enabled.
Signed-off-by: Gao feng <gaofeng@cn.fujitsu.com>
The ESX code has a method esxVI_Alloc which would call
virAllocN directly, instead of using the VIR_ALLOC_N
macro. Remove this method and make the callers just
use VIR_ALLOC as is normal practice.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Delete the USB controller check from the USB Device checklist in
virDomainDeviceIsUSB as USB controller is a PCI device rather than
a USB one.
Signed-off-by: Liu Ji <jeremy.liu@huawei.com>
The s390, ppc and arm CPU drivers never set the 'arch' field
in their impl of cpuArchNodeData. This leads to error messages
being reported from cpuDataFree later, due to trying to use
VIR_ARCH_NONE.
#0 virRaiseErrorFull (filename=filename@entry=0x76f94434 "cpu/cpu.c", funcname=funcname@entry=0x76f942dc <__FUNCTION__.18096> "cpuGetSubDriver", linenr=linenr@entry=58,
domain=domain@entry=31, code=code@entry=1, level=level@entry=VIR_ERR_ERROR, str1=0x76f70e18 "internal error: %s",
str2=str2@entry=0x7155f2ec "undefined hardware architecture", str3=str3@entry=0x0, int1=int1@entry=-1, int2=int2@entry=-1, fmt=0x76f70e18 "internal error: %s")
at util/virerror.c:646
#1 0x76e682ea in virReportErrorHelper (domcode=domcode@entry=31, errorcode=errorcode@entry=1, filename=0x76f94434 "cpu/cpu.c",
funcname=0x76f942dc <__FUNCTION__.18096> "cpuGetSubDriver", linenr=linenr@entry=58, fmt=0x76f7e7e4 "%s") at util/virerror.c:1292
#2 0x76ed82d4 in cpuGetSubDriver (arch=<optimized out>) at cpu/cpu.c:57
#3 cpuGetSubDriver (arch=VIR_ARCH_NONE) at cpu/cpu.c:51
#4 0x76ed8818 in cpuDataFree (data=data@entry=0x70c22d78) at cpu/cpu.c:216
#5 0x716aaec0 in virQEMUCapsInitCPU (arch=VIR_ARCH_ARMV7L, caps=0x70c29a08) at qemu/qemu_capabilities.c:867
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The VIR_FREE() macro will cast away any const-ness. This masked a
number of places where we passed a 'const char *' string to
VIR_FREE. Fortunately in all of these cases, the variable was not
in fact const data, but a heap allocated string. Fix all the
variable declarations to reflect this.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
No need to open code now that we have a nice function.
Interestingly, our virStringFreeList function is typed correctly
(a malloc'd list of malloc'd strings is NOT const, whether at the
point where it is created, or at the point where it is cleand up),
so using it with a 'const char **' argument would require a cast
to keep the compiler. I chose instead to remove const from code
even where we don't modify the argument, just to avoid the need
to cast.
* src/qemu/qemu_command.h (qemuParseCommandLine): Drop declaration.
* src/qemu/qemu_command.c (qemuParseProcFileStrings)
(qemuStringToArgvEnv): Don't force malloc'd result to be const.
(qemuParseCommandLinePid, qemuParseCommandLineString): Simplify
cleanup.
(qemuParseCommandLine, qemuFindEnv): Drop const-correctness to
avoid the need to cast in callers.
Signed-off-by: Eric Blake <eblake@redhat.com>
In commit 991270db99 I've used virDomainNetGetActualHostdev() to get
the actual hostdev from a network when removing the network from the
list to avoid leaving the hostdev in the list. I didn't notice that this
function doesn't check if the actual network is allocated and
dereferences it. This crashes the daemon when cleaning up a domain
object in early startup phases when the actual network definition isn't
allocated. When the actual definition isn't present, the hostdev that
might correspond to it won't be present anyways so it's safe to return
NULL.
Thanks to Cole Robinson for noticing this problem.
No need to check if privileged when reading hostsysinfo, since
that check was already done in libxlDriverShouldLoad(). The
libxl driver fails to load if not privileged.
John Ferlan reported the following Coverity warning:
In libxlDomainCoreDump() Coverity has noted a FORWARD_NULL reference:
2004 if ((flags & VIR_DUMP_CRASH) && !vm->persistent) {
2005 virDomainObjListRemove(driver->domains, vm);
(20) Event assign_zero: Assigning: "vm" = "NULL".
Also see events: [var_deref_model]
2006 vm = NULL;
2007 }
2008
2009 ret = 0;
2010
2011 cleanup_unpause:
(21) Event var_deref_model: Passing null pointer "vm" to function
"virDomainObjIsActive(virDomainObjPtr)", which dereferences it. [details]
Also see events: [assign_zero]
2012 if (virDomainObjIsActive(vm) && paused) {
2013 if (libxl_domain_unpause(priv->ctx, dom->id) != 0) {
2014 virReportError(VIR_ERR_INTERNAL_ERROR,
Removing the vm from domain obj list and setting it to NULL can be
done in the previous 'if (flags & VIR_DUMP_CRASH)' conditional. Fix
the Coverity warning by ensuring vm is not NULL before testing if it
is still active.
daemon/Makefile.am installs a .policy file if WITH_LIBVIRTD and
WITH_POLKIT are both set. src/Makefile.am, on the other hand,
installs a .policy file if WITH_POLKIT1 is set, but without checking
WITH_LIBVIRTD. When running 'make rpm' with client_only manually
set, on a Fedora 19 box, that leads to a failure:
RPM build errors:
Installed (but unpackaged) file(s) found:
/usr/share/polkit-1/actions/org.libvirt.api.policy
Fix it by adding another conditional.
* src/Makefile.am (polkitaction_DATA): Make conditional.
Signed-off-by: Eric Blake <eblake@redhat.com>
When virGetUserEnt() and virGetGroupEnt() fail due to the uid or gid not
existing on the machine they'll print a message like:
$ virsh -c vbox:///session list
error: failed to connect to the hypervisor
error: Failed to find user record for uid '32655': Success
The success at the end is a bit confusing. This changes it to:
$ virsh -c vbox:///session list
error: failed to connect to the hypervisor
error: Failed to find user record for uid '32655'
Automake has builtin support to prevent botched conditional nesting,
but only if you use:
if FOO
else !FOO
endif !FOO
An example error message when using the wrong name:
daemon/Makefile.am:378: error: else reminder (LIBVIRT_INIT_SCRIPT_SYSTEMD_TRUE) incompatible with current conditional: LIBVIRT_INIT_SCRIPT_SYSTEMD_FALSE
daemon/Makefile.am:381: error: endif reminder (LIBVIRT_INIT_SCRIPT_SYSTEMD_TRUE) incompatible with current conditional: LIBVIRT_INIT_SCRIPT_SYSTEMD_FALSE
As our makefiles tend to have quite a bit of nested conditionals,
it's better to take advantage of the benefits of the build system
double-checking that our conditionals are well-nested, but that
requires a syntax check to enforce our usage style.
Alas, unlike C preprocessor and spec files, we can't use indentation
to make it easier to see how deeply nesting goes.
* cfg.mk (sc_makefile_conditionals): New rule.
* daemon/Makefile.am: Enforce the style.
* gnulib/tests/Makefile.am: Likewise.
* python/Makefile.am: Likewise.
* src/Makefile.am: Likewise.
* tests/Makefile.am: Likewise.
* tools/Makefile.am: Likewise.
Signed-off-by: Eric Blake <eblake@redhat.com>
The 'uuid' field in virDomainDefPtr is not a pointer, it is a
fixed length array. Calling VIR_ALLOC on it is thus wrong and
leaks memory.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Commit 50348e6edf reused the code to remove the hostdev portion of a
network definition on multiple places but forgot to take into account
that sometimes the "actual" network is passed and in some cases the
parent of that.
This patch uses the virDomainNetGetActualHostdev() helper to acquire the
correct pointer all the time while removing the hostdev portion from the
list.
https://bugzilla.redhat.com/show_bug.cgi?id=999352
Since commit v1.0.5-56-g449e6b1 (Pull parsing of migration xml up into
QEMU driver APIs) any attempt to rename a domain during migration fails
with the following error message:
internal error Incoming cookie data had unexpected name DOM vs DOM2
This is because migration cookies always use the original domain name
and the mentioned commit failed to propagate the name back to
qemuMigrationPrepareAny.
Now that most fields of libxlDriverPrivate struct are immutable
or self-locking, there is no need to acquire the driver lock in
much of the libxl driver.
The libxlDriverPrivate struct contains an variety of data with
varying access needs. Similar to the QEMU and LXC drivers,
move all the static config data into a dedicated libxlDriverConfig
object. The only locking requirement is to hold the driver lock
while obtaining an instance of libxlDriverConfig. Once a reference
is held on the config object, it can be used completely lockless
since it is immutable.
libxlDomainGetInfo() uses the driver-wide libxl ctx when
it would be more appropriate to use the per-domain ctx
associated with the domain. Switch to using the per-domain
libxl ctx.
libxlMakeDomCreateInfo() uses the driver-wide libxl ctx when
it would be more appropriate to use the per-domain ctx
associated with the domain. Switch to using the per-domain
libxl ctx.
libxl version info is static data as far as the libxl driver
is concerned, so retrieve this info when the driver is initialized
and stash it in the libxlDriverPrivate object. Subsequently use
the stashed info instead of repeatedly calling libxl_get_version_info().
Detect early on in libxl driver initialization if the driver
should be loaded at all, avoiding needless initialization steps
that only have to be undone later. While at it, move the
detection to a helper function to improve readability.
After detecting that the driver should be loaded, subsequent
failures such as initializing the log stream, allocating libxl
ctx, etc. should be treated as failure to initialize the driver.
Create libxl_domain.[ch] and move all functions operating on
libxlDomainObjPrivate to these files. This will be useful for
future patches that e.g. add job support for libxlDomainObjPrivate.
New coverity installation determined that the muliple if condition for
"*Alloc" and "*AppendToList" could fail during AppendToList thus leaking
memory.
Currently, kernel supports up to 8 queues for a multiqueue tap device.
However, if user tries to enter a huge number (e.g. one million) the tap
allocation fails, as expected. But what is not expected is the log full
of warnings:
warning : virFileClose:83 : Tried to close invalid fd 0
The problem is, upon error we iterate over an array of FDs (handlers to
queues) and VIR_FORCE_CLOSE() over each item. However, the array is
pre-filled with zeros. Hence, we repeatedly close stdin. Ouch.
But there's more. The queues allocation is done in virNetDevTapCreate()
which cleans up the FDs in case of error. Then, its caller, the
virNetDevTapCreateInBridgePort() iterates over the FD array and tries to
close them too. And so does qemuNetworkIfaceConnect() and
qemuBuildInterfaceCommandLine().
According to VMWare's documentation 'cdrom-raw' is an acceptable value
for deviceType for a CD-ROM drive. The documentation states that the VMX
configuration for a CD-ROM deviceType is as follows:
ide|scsi(n):(n).deviceType = "cdrom-raw|atapi-cdrom|cdrom-image"
From the documentation it appears the following is true:
- cdrom-image = Provides the ISO to the VM
- atapi-cdrom = Provides a NEC emulated ATAPI CD-ROM on top of the host
CD-ROM
- cdrom-raw = Passthru for a host CD-ROM drive. Allows CD-R burning from
within the guest.
A CD-ROM prior to this patch would always provide an 'atapi-cdrom' is
modeled as:
<disk type='block' device='cdrom'>
<source dev='/dev/scd0'/>
<target dev='hda' bus='ide'/>
<address type='drive' controller='0' bus='0' target='0' unit='0'/>
</disk>
This patch allows the 'device' attribute to be set to 'lun' for a raw
acccess CD-ROM such as:
<disk type='block' device='lun'>
<source dev='/dev/scd0'/>
<target dev='hda' bus='ide'/>
<address type='drive' controller='0' bus='0' target='0' unit='0'/>
</disk>
Sometimes a serial port might not be actually wired to a device when the
user does not have the VM powered on and we should not consider this a
fatal error.
Starting with qemu 1.6, the qemu-system-arm vexpress-a9 model has a
hardcoded virtio-mmio transport which enables attaching all virtio
devices.
On the command line, we have to use virtio-XXX-device rather than
virtio-XXX-pci, thankfully s390 already set the precedent here so
it's fairly straight forward.
At the XML level, this adds a new device address type virtio-mmio.
The controller and addressing don't have any subelements at the
moment because we they aren't needed for this usecase, but could
be added later if needed.
Add a test case for an ARM guest with one of every virtio device
enabled.
Similar to the chardev bit, ARM boards depend on the old style '-net nic'
for actually instantiating net devices. But we can't block out
-netdev altogether since it's needed for upcoming virtio support.
And add tests for working ARM XML with console, disk, and networking.
This corresponds to '-sd' and '-drive if=sd' on the qemu command line.
Needed for many ARM boards which don't provide any other way to
pass in storage.
QEMU ARM boards don't give us any way to explicitly wire in
a -chardev, so use the old style -serial options.
Unfortunately this isn't as simple as just turning off the CHARDEV flag
for qemu-system-arm, as upcoming virtio support _will_ use device/chardev.
On my machine, a guest fails to boot if it has a sound card, but not
graphical device/display is configured, because pulseaudio fails to
initialize since it can't access $HOME.
A workaround is removing the audio device, however on ARM boards there
isn't any option to do that, so -nographic always fails.
Set QEMU_AUDIO_DRV=none if no <graphics> are configured. Unfortunately
this has massive test suite fallout.
Add a qemu.conf parameter nographics_allow_host_audio, that if enabled
will pass through QEMU_AUDIO_DRV from sysconfig (similar to
vnc_allow_host_audio)
Add an attribute named 'removable' to the 'target' element of disks,
which controls the removable flag. For instance, on a Linux guest it
controls the value of /sys/block/$dev/removable. This option is only
valid for USB disks (i.e. bus='usb'), and its default value is 'off',
which is the same behaviour as before.
To achieve this, 'removable=on' (or 'off') is appended to the '-device
usb-storage' parameter sent to qemu when adding a USB disk via
'-disk'. A capability flag QEMU_CAPS_USB_STORAGE_REMOVABLE was added
to keep track if this option is supported by the qemu version used.
Bug: https://bugzilla.redhat.com/show_bug.cgi?id=922495
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Allow use of the usb-storage device only if the new capability flag
QEMU_CAPS_DEVICE_USB_STORAGE is set, which it is for qemu(-kvm)
versions >= 0.12.1.2-rhel62-beta.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
virVMXFormatHardDisk() and virVMXFormatCDROM() duplicated a lot of code
from each other and made a lot of nested if checks to build each part of
the VMX file. This hopefully simplifies the code path while combining
the two functions with no net difference.
Since virtlockd is only built when libvirtd is built, we should
not install its auxiliary files unconditionally. This solves
two failures. 1. 'make distcheck' complains:
rm -f Makefile
ERROR: files left in build directory after distclean:
./src/virtlockd.8
2. './autobuild.sh' complains:
Checking for unpackaged file(s): /usr/lib/rpm/check-files
/home/eblake/rpmbuild/BUILDROOT/mingw-libvirt-1.1.1-1.fc19.eblake1377879911.x86_64
error: Installed (but unpackaged) file(s) found:
/usr/i686-w64-mingw32/sys-root/mingw/etc/libvirt/virtlockd.conf
/usr/i686-w64-mingw32/sys-root/mingw/share/augeas/lenses/tests/test_virtlockd.aug
/usr/i686-w64-mingw32/sys-root/mingw/share/augeas/lenses/virtlockd.aug
/usr/i686-w64-mingw32/sys-root/mingw/share/man/man8/virtlockd.8
/usr/x86_64-w64-mingw32/sys-root/mingw/etc/libvirt/virtlockd.conf
/usr/x86_64-w64-mingw32/sys-root/mingw/share/augeas/lenses/tests/test_virtlockd.aug
/usr/x86_64-w64-mingw32/sys-root/mingw/share/augeas/lenses/virtlockd.aug
/usr/x86_64-w64-mingw32/sys-root/mingw/share/man/man8/virtlockd.8
* src/Makefile.am (CLEANFILES): Add virtlockd.8.
(man8_MANS, conf_DATA, augeas_DATA, augeastest_DATA): Only install
virtlockd files when daemon is built.
Signed-off-by: Eric Blake <eblake@redhat.com>
vhost only works in KVM mode at the moment, and is infact compiled
out if the emulator is built for non-native architecture. While it
may work at some point in the future for plain qemu, for now it's
just noise on the command line (and which contributes to arm cli
breakage).
FreeBSD 10 recently changed their definition of RAND_MAX, to try
and cover the fact that their evenly distributed results of rand()
really are a smaller range than a full power of 2. As a result,
I did some investigation, and learned:
1. POSIX requires random() to be evenly distributed across exactly
31 bits. glibc also guarantees this for rand(), but the two are
unrelated, and POSIX only associates RAND_MAX with rand().
Avoiding RAND_MAX altogether thus avoids a build failure on
FreeBSD 10.
2. Concatenating random bits from a PRNG will NOT provide uniform
coverage over the larger value UNLESS the period of the original
PRNG is at least as large as the number of bits being concatenated.
Simple example: suppose that RAND_MAX were 1 with a period of 2**1
(which means that the PRNG merely alternates between 0 and 1).
Concatenating two successive rand() calls would then invariably
result in 01 or 10, which is a rather non-uniform distribution
(00 and 11 are impossible) and an even worse period (2**0, since
our second attempt will get the same number as our first attempt).
But a RAND_MAX of 1 with a period of 2**2 (alternating between
0, 1, 1, 0) provides sane coverage of all four values, if properly
tempered. (Back-to-back calls would still only see half the values
if we don't do some tempering). We therefore want to guarantee a
period of at least 2**64, preferably larger (as a tempering factor);
POSIX only makes this guarantee for random() with 256 bytes of info.
* src/util/virrandom.c (virRandomBits): Use constants that are
accurate for the PRNG we are using, not an unrelated PRNG.
(randomState): Ensure the period of our PRNG exceeds our usage.
Signed-off-by: Eric Blake <eblake@redhat.com>
Commit 29fe5d7 (released in 1.1.1) introduced a latent problem
for any caller of virSecurityManagerSetProcessLabel and where
the domain already had a uid:gid label to be parsed. Such a
setup would collect the list of supplementary groups during
virSecurityManagerPreFork, but then ignores that information,
and thus fails to call setgroups() to adjust the supplementary
groups of the process.
Upstream does not use virSecurityManagerSetProcessLabel for
qemu (it uses virSecurityManagerSetChildProcessLabel instead),
so this problem remained latent until backporting the initial
commit into v0.10.2-maint (commit c061ff5, released in 0.10.2.7),
where virSecurityManagerSetChildProcessLabel has not been
backported. As a result of using a different code path in the
backport, attempts to start a qemu domain that runs as qemu:qemu
will end up with supplementary groups unchanged from the libvirtd
parent process, rather than the desired supplementary groups of
the qemu user. This can lead to failure to start a domain
(typical Fedora setup assigns user 107 'qemu' to both group 107
'qemu' and group 36 'kvm', so a disk image that is only readable
under kvm group rights is locked out). Worse, it is a security
hole (the qemu process will inherit supplemental group rights
from the parent libvirtd process, which means it has access
rights to files owned by group 0 even when such files should
not normally be visible to user qemu).
LXC does not use the DAC security driver, so it is not vulnerable
at this time. Still, it is better to plug the latent hole on
the master branch first, before cherry-picking it to the only
vulnerable branch v0.10.2-maint.
* src/security/security_dac.c (virSecurityDACGetIds): Always populate
groups and ngroups, rather than only when no label is parsed.
Signed-off-by: Eric Blake <eblake@redhat.com>
The return values for the virConnectListAllSecrets call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virConnectListAllNWFilters call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virConnectListAllNodeDevices call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virConnectListAllInterfaces call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virConnectListAllNetworks call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virStoragePoolListAllVolumes call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virConnectListAllStoragePools call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virConnectListAllDomains call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virDomain{SnapshotListAllChildren,ListAllSnapshots}
calls were not bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The return values for the virDomainGetJobStats call were not
bounds checked. This is a robustness issue for clients if
something where to cause corruption of the RPC stream data.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The parameters for the virDomainMigrate*Params RPC calls were
not bounds checks, meaning a malicious client can cause libvirtd
to consume arbitrary memory
This issue was introduced in the 1.1.0 release of libvirt
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Similarly to qemu_driver.c, we can join often repeating code of looking
up network into one function: networkObjFromNetwork.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
When using a <interface type="network"> that points to a network with
hostdev forwarding mode a hostdev alias is created for the network. This
allias is inserted into the hostdev list, but is backed with a part of
the network object that it is connected to.
When a VM is being stopped qemuProcessStop() calls
networkReleaseActualDevice() which eventually frees the memory for the
hostdev object. Afterwards when the domain definition is being freed by
virDomainDefFree() an invalid pointer is accessed by
virDomainHostdevDefFree() and may cause a crash of the daemon.
This patch removes the entry in the hostdev list before freeing the
depending memory to avoid this issue.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1000973
QEMU commit 3984890 introduced the "pci-hole64-size" property,
to i440FX-pcihost and q35-pcihost with a default setting of 2 GB.
Translate <pcihole64>x<pcihole64/> to:
-global q35-pcihost.pci-hole64-size=x for q35 machines and
-global i440FX-pcihost.pci-hole64-size=x for i440FX-based machines.
Error out on other machine types or if the size was specified
but the pcihost device lacks 'pci-hole64-size' property.
https://bugzilla.redhat.com/show_bug.cgi?id=990418
<controller type='pci' index='0' model='pci-root'>
<pcihole64 unit='KiB'>1048576</pcihole64>
</controller>
It can be used to adjust (or disable) the size of the 64-bit
PCI hole. The size attribute is in kilobytes (different unit
can be specified on input), but it gets rounded up to
the nearest GB by QEMU.
Disabling it will be needed for guests that crash with the
64-bit PCI hole (like Windows XP), see:
https://bugzilla.redhat.com/show_bug.cgi?id=990418
The ftp protocol is already recognized by qemu/KVM so add this support to
libvirt as well.
The xml should be as following:
<disk type='network' device='cdrom'>
<source protocol='ftp' name='/url/path'>
<host name='host.name' port='21'/>
</source>
</disk>
Signed-off-by: Aline Manera <alinefm@br.ibm.com>
QEMU/KVM already allows a HTTP URL for the cdrom ISO image so add this support
to libvirt as well.
The xml should be as following:
<disk type='network' device='cdrom'>
<source protocol='http' name='/url/path'>
<host name='host.name' port='80'/>
</source>
</disk>
Signed-off-by: Aline Manera <alinefm@br.ibm.com>
qemu-img is going to switch the default for QCOW2
to QCOW2v3 (compat=1.1)
Extend the probing for qemu-img command line options to check
if -o compat is supported. If the volume definition specifies
the qcow2 format but no compat level and -o compat is supported,
specify -o compat=0.10 to create a QCOW2v2 image.
https://bugzilla.redhat.com/show_bug.cgi?id=997977
If there's no hard_limit set and domain uses VFIO we still must lock
the guest memory (prerequisite from qemu). Hence, we should compute
the amount to be locked from max_balloon.
When cpu hotplug fails without reporting an error, we would fail the
command but update the count of vCPUs anyways.
Commit 761fc48136 fixed the case when CPU
hot-unplug failed silently, but forgot to fix up the value in this case.
Resolves: https://bugzilla.redhat.com/show_bug.cgi?id=1000357
The virDomainOpenGraphics method accepts a UNIX socket FD from
the client app. It must set the label on this FD otherwise QEMU
will be prevented from receiving it with recvmsg.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
