If a remote call fails during event registration (more than likely from
a network failure or remote libvirtd restart timed just right), then when
calling the virObjectEventStateDeregisterID we don't want to call the
registered @freecb function because that breaks our contract that we
would only call it after succesfully returning. If the @freecb routine
were called, it could result in a double free from properly coded
applications that free their opaque data on failure to register, as seen
in the following details:
Program terminated with signal 6, Aborted.
#0 0x00007fc45cba15d7 in raise
#1 0x00007fc45cba2cc8 in abort
#2 0x00007fc45cbe12f7 in __libc_message
#3 0x00007fc45cbe86d3 in _int_free
#4 0x00007fc45d8d292c in PyDict_Fini
#5 0x00007fc45d94f46a in Py_Finalize
#6 0x00007fc45d960735 in Py_Main
#7 0x00007fc45cb8daf5 in __libc_start_main
#8 0x0000000000400721 in _start
The double dereference of 'pyobj_cbData' is triggered in the following way:
(1) libvirt_virConnectDomainEventRegisterAny is invoked.
(2) the event is successfully added to the event callback list
(virDomainEventStateRegisterClient in
remoteConnectDomainEventRegisterAny returns 1 which means ok).
(3) when function remoteConnectDomainEventRegisterAny is hit,
network connection disconnected coincidently (or libvirtd is
restarted) in the context of function 'call' then the connection
is lost and the function 'call' failed, the branch
virObjectEventStateDeregisterID is therefore taken.
(4) 'pyobj_conn' is dereferenced the 1st time in
libvirt_virConnectDomainEventFreeFunc.
(5) 'pyobj_cbData' (refered to pyobj_conn) is dereferenced the
2nd time in libvirt_virConnectDomainEventRegisterAny.
(6) the double free error is triggered.
Resolve this by adding a @doFreeCb boolean in order to avoid calling the
freeCb in virObjectEventStateDeregisterID for any remote call failure in
a remoteConnect*EventRegister* API. For remoteConnect*EventDeregister* calls,
the passed value would be true indicating they should run the freecb if it
exists; whereas, it's false for the remote call failure path.
Patch based on the investigation and initial patch posted by
fangying <fangying1@huawei.com>.
Use ATTRIBUTE_FALLTHROUGH, introduced by commit
5d84f5961b, instead of comments to
indicate that the fall through is an intentional behavior.
Signed-off-by: Marc Hartmayer <mhartmay@linux.vnet.ibm.com>
Reviewed-by: Boris Fiuczynski <fiuczy@linux.vnet.ibm.com>
Reviewed-by: Bjoern Walk <bwalk@linux.vnet.ibm.com>
virDomainXMLOption gains driver specific callbacks for parsing and
formatting save cookies.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
Reviewed-by: Pavel Hrdina <phrdina@redhat.com>
While checking for ABI stability, drivers might pose additional
checks that are not valid for general case. For instance, qemu
driver might check some memory backing attributes because of how
qemu works. But those attributes may work well in other drivers.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Attempting to start a domain with USB hostdevs but no USB controllers
fails with the rather cryptic error
libxl: error: libxl_qmp.c:287:qmp_handle_error_response: received an
error message from QMP server: Bus 'xenusb-0.0' not found
This can be fixed by creating default USB controllers. When no USB
controllers are defined, create the number of 8 port controllers
necessary to accommodate the number of defined USB devices.
Note that USB controllers are already created as needed in the
domainAttachDevice code path. E.g. a USB controller will be created,
if necessary, when attaching a USB device with
'virsh attach-device dom usbdev.xml'.
If a VNC listen address is not specified in domXML, libxl
will default to 127.0.0.1, but this is never reflected in the domXML.
In the case of spice, a missing listen address resulted in listening
on all interfaces, i.e. '0.0.0.0'. If not specified, set the listen
address in virDomainGraphicsDef struct to the libxl default when
creating the frame buffer device. Additionally, set default spice
listen address to 127.0.0.1.
When running on a NUMA machine, populate the sibling node
and distance information using data supplied by Xen.
With locality distances information, under Xen, new host
capabilities would like:
<topology>
<cells num='4'>
<cell id='0'>
<memory unit='KiB'>263902380</memory>
<distances>
<sibling id='0' value='10'/>
<sibling id='1' value='21'/>
</distances>
...
</cell>
...
</cells>
...
</topology>
Signed-off-by: Wim ten Have <wim.ten.have@oracle.com>
Reviewed-by: Joao Martins <joao.m.martins@oracle.com>
Reviewed-by: Jim Fehlig <jfehlig@suse.com>
Xen feature nestedhvm is the option on Xen 4.4+ which enables
nested virtualization when mode host-passthrough is applied.
nested HVM is enabled by adding below on the target domain;
<cpu mode='host-passthrough'/>
Virtualization on target domain can be disabled by specifying
such under feature policy rule on target name;
[On Intel (VT-x) architecture]
<feature policy='disable' name='vmx'/>
or:
[On AMD (AMD-V) architecture]
<feature policy='disable' name='svm'/>
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Wim ten Have <wim.ten.have@oracle.com>
So far our code is full of the following pattern:
dom = virGetDomain(conn, name, uuid)
if (dom)
dom->id = 42;
There is no reasong why it couldn't be just:
dom = virGetDomain(conn, name, uuid, id);
After all, client domain representation consists of tuple (name,
uuid, id).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
There is no reason for it not to be in the utils, all global symbols
under that file already have prefix vir* and there is no reason for it
to be part of DRIVER_SOURCES because that is just a leftover from
older days (pre-driver modules era, I believe).
Signed-off-by: Martin Kletzander <mkletzan@redhat.com>
The CPU driver provides APIs to create and free virCPUDataPtr. Thus all
APIs exported from the driver should work with that rather than
requiring the caller to pass a pointer to an internal part of the
structure.
In other words
virCPUx86DataAddCPUID(cpudata, &cpuid)
is much better than the original
virCPUx86DataAddCPUID(&cpudata->data.x86, &cpuid)
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
The new API is called virCPUDataFree. Individual CPU drivers are no
longer required to implement their own freeing function unless they need
to free architecture specific data from virCPUData.
Signed-off-by: Jiri Denemark <jdenemar@redhat.com>
The libxl code was checking that a 'char *' was != '\0', instead
of checking the first element in the string
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
As discussed here [0][1] Coverity reported two issues:
- On libxlDomainMigrationPrepareTunnel3 @@mig will be leaked on failures
after sucessfull call libxlDomainMigrationPrepareAny hence we free it.
Setting mig = NULL after @mig is assigned plus adding libxlMigrationCookieFree
on error paths addresses the issue. In case virThreadCreate fails,
unref of args frees the cookie on dispose function (libxlMigrationDstArgsDispose)
- On libxlMigrationStartTunnel @tc would be leaked.
Fixed by correctly saving the newly allocated @tc onto @tnl such that
libxlMigrationStopTunnel would free it up.
[0] https://www.redhat.com/archives/libvir-list/2017-February/msg00791.html
[1] https://www.redhat.com/archives/libvir-list/2017-February/msg00833.html
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Commit 4ab0c959 fixed a memory leak in libxlDriverGetDom0MaxmemConf
but introduced a potential double free of mem_tokens
*** Error in `/usr/sbin/libvirtd': double free or corruption (out):
0x00007fffc808cfd0 ***
Avoid double free by setting mem_tokens to NULL after calling
virStringListFree.
Tunnelled migration doesn't require any extra network connections
beside the libvirt daemon. It's capable of strong encryption and the
default option of openstack-nova.
This patch adds the tunnelled migration(Tunnel3params) support to
libxl. On the source side, the data flow is:
* libxlDoMigrateSend() -> pipe libxlTunnel3MigrationFunc() polls pipe
* out and then write to dest stream.
While on the destination side:
* Stream -> pipe -> 'recvfd of libxlDomainStartRestore'
The usage is the same as p2p migration, execpt adding one extra
'--tunnelled' to the libvirt p2p migration command.
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
The newly introduced function libxlDomainMigrationPrepareAny
will be shared between P2P and tunnelled variations.
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
If either the "if (STRPREFIX(mem_tokens[j], "max:"))" is never entered
or the "if (virStrToLong_ull(mem_tokens[j] + 4, &p, 10, maxmem) < 0)" break
is hit, control goes back to the outer loop processing 'cmd_tokens' and
it's possible that the 'mem_tokens' would be overwritten.
Found by Coverity
When the libxl driver is initialized, it creates a virDomainDef
object for dom0 and adds it to the list of domains. Total memory
for dom0 was being set from the max_memkb field of libxl_dominfo
struct retrieved from libxl, but this field can be set to
LIBXL_MEMKB_DEFAULT (~0ULL) if dom0 maximum memory has not been
explicitly set by the user.
This patch adds some simple parsing of the Xen commandline,
looking for a dom0_mem parameter that also specifies a 'max' value.
If not specified, dom0 maximum memory is effectively all physical
host memory.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
The libxl driver reports different values of maximum memory depending
on state of a domain. If inactive, maximum memory value is reported
correctly. When active, maximum memory is derived from max_pages value
returned by the XEN_SYSCTL_getdomaininfolist sysctl operation. But
max_pages can be changed by toolstacks and does not necessarily
represent the maximum memory a domain can use during its active
lifetime.
A better location for determining a domain's maximum memory is the
/local/domain/<id>/memory/static-max node in xenstore. This value
is set from the libxl_domain_build_info.max_memkb field when creating
the domain. Currently it cannot be changed nor can its value be
exceeded by a balloon operation. From libvirt's perspective, always
reporting maximum memory with virDomainDefGetMemoryTotal() will produce
the same results as reading the static-max node in xenstore.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
When a user does not explicitly set a <driver> in the disk config,
libvirt defers selection of a default to libxl. This approach works
fine when starting a domain with such configuration or attaching a
disk to a running domain. But when detaching such a disk, libxl
will fail with "unrecognized disk backend type: 0". libxl makes no
attempt to recalculate a default backend (driver) on detach and
simply fails when uninitialized.
This patch updates the libvirt disk config with the backend selected
by libxl when starting a domain or attaching a disk to a running
domain. Another benefit of this approach is that the live XML is
also updated with the backend driver selected by libxl.
When starting a domian, a libxl_domain_config object is created from
virDomainDef. Any virDomainDiskDef devices with a format of
VIR_STORAGE_FILE_NONE are mapped to LIBXL_DISK_FORMAT_RAW in the
corresponding libxl_disk_device, but the virDomainDiskDef format is
never updated to reflect the change.
A better place to set a default format for disk devices is the
device post-parse callback, ensuring the virDomainDiskDef object
reflects the default format.
The typical pattern when calling libxl functions that populate a
structure is
libxl_foo foo;
libxl_foo_init(&foo);
libxl_get_foo(ctx, &foo);
...
libxl_foo_dispose(&foo);
Fix several instances of libxl_physinfo missing the init and
dispose calls.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
libxlGetAutoballoonConf is supposed to honor user-specified
autoballoon setting in libxl.conf. As written, the user-specified
setting could be overwritten by the subsequent logic to check
dom0_mem parameter. If user-specified setting is present and
correct, accept it. Only fallback to checking Xen dom0_mem
command line parameter if user-specfied setting is not present.
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
xen.git commit 57f8b13c changed several of the libxl memory
get/set functions to take 64 bit parameters. The libvirt
libxl driver still uses uint32_t variables for these various
parameters, which is particularly problematic for the
libxl_set_memory_target() function.
When dom0 autoballooning is enabled, libvirt (like xl) determines
the memory needed to start a domain and the memory available. If
memory available is less than memory needed, dom0 is ballooned
down by passing a negative value to libxl_set_memory_target()
'target_memkb' parameter. Prior to xen.git commit 57f8b13c,
'target_memkb' was an int32_t. Subtracting a larger uint32 from
a smaller uint32 and assigning it to int32 resulted in a negative
number. After commit 57f8b13c, the same subtraction is widened
to a int64, resulting in a large positive number. The simple
fix taken by this patch is to assign the difference of the
uint32 values to a temporary int32 variable, which is then
passed to 'target_memkb' parameter of libxl_set_memory_target().
Note that it is undesirable to change libvirt to use 64 bit
variables since it requires setting LIBXL_API_VERSION to 0x040800.
Currently libvirt supports LIBXL_API_VERSION >= 0x040400,
essentially Xen >= 4.4.
The current logic around configuring timers in libxl based on
virDomainDef object is a bit brain dead. Unsupported timers are
silently ignored and tsc is only recognized if it is the first
timer specified.
Change the logic to reject unsupported timers and honor the tsc
timer regardless of its order when multiple timers are specified.
For HVM domains, pae is only set in libxl_domain_build_info when
explicitly specified in the hypervisor <features> config. This is
fine for i686 machines, but is incorrect behavior for x86_64 machines
where pae must always be enabled. See the following discussion for
additional details
https://www.redhat.com/archives/libvir-list/2017-January/msg00254.html
libxl doesn't provide a way to write one log for each domain. Thus
we need to demux the messages. If our logger doesn't know to which
domain to attribute a message, then it will write it to the default
log file.
Starting with Xen 4.9 (commit f9858025 and following), libxl will
write the domain ID in an easy to grab manner. The logger introduced
by this commit will use it to demux the libxl log messages.
Thanks to the default log file, this logger will also work with older
versions of Xen.
libvirt libxl picks its own default with respect to the default NIC
to use. libxlMakeNic is the one responsible for this and on boot it
picks LIBXL_NIC_TYPE_VIF_IOEMU for HVM domains such that it accomodates
both PV and emulated one. The good behaving guest at boot will then
select the pv and unplug the emulated device.
Now, on HVM when attaching an interface it will pick the same default
that is LIBXL_NIC_TYPE_VIF_IOEMU which as a result will fail the attach
(see xen commit 32e9d0f ("libxl: nic type defaults to vif in hotplug for
hvm guest"). Xen doesn't yet support the hotplug of emulated devices,
but we don't want to rule out that case either, which might get support
in the future. Hence we simply reverse the defaults when we are
attaching the interface which allows libvirt to prefer the PV nic first
without adding "model='netfront'" following the same pattern as above
commit. Also to avoid ruling out the emulated one we set to
LIBXL_NIC_TYPE_IOEMU when setting a model type that is not 'netfront'.
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
In preparation to the code move to virnetdevtap.c, this change:
* renames virNetInterfaceStats to virNetDevTapInterfaceStats
* changes 'path' to 'ifname', to use the same vocable as other
method in virnetdevtap.c.
* Add the attributes checker
New line character in name of domain is now forbidden because it
mess virsh output and can be confusing for users.
Validation of name is done in drivers, after parsing XML to avoid
problems with dissappeared domains which was already created with
new-line char in name.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Although the migration port is immediately released in the
finish phase of migration, it was never set in the domain
private object when allocated in the prepare phase. So
libxlDomainMigrationFinish() always released a 0-initialized
migrationPort, leaking any allocated port. After enough
migrations to exhaust the migration port pool, migration would
fail with
error: internal error: Unable to find an unused port in range
'migration' (49152-49216)
Fix it by setting libxlDomainObjPrivate->migrationPort to the
port allocated in the prepare phase. While at it, also fix
leaking an allocated port if the prepare phase fails.
Change the virDomainChrDef to use a pointer to 'source' and allocate
that pointer during virDomainChrDefNew.
This has tremendous "fallout" in the rest of the code which mainly
has to change source.$field to source->$field.
Signed-off-by: John Ferlan <jferlan@redhat.com>
Modeled after the qemuDomainHostdevPrivatePtr (commit id '27726d8c'),
create a privateData pointer in the _virDomainChardevDef to allow storage
of private data for a hypervisor in order to at least temporarily store
secret data for usage during qemuBuildCommandLine.
NB: Since the qemu_parse_command (qemuParseCommandLine) code is not
expecting to restore the secret data, there's no need to add code
code to handle this new structure there.
Signed-off-by: John Ferlan <jferlan@redhat.com>
This function for some weird reason returns integer instead of
virDomainNetType type. It is important to return the correct type
so that we know what values we can expect.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Due to a copy and paste error, the scheduler 'cap' parameter
was over-writing the 'weight' parameter when preparing the
return parameters in libxlDomainGetSchedulerParametersFlags.
As a result, the scheduler weight was never shown when getting
schedinfo and setting the weight failed as well
virsh schedinfo testvm
Scheduler : credit
cap : 0
virsh schedinfo testvm --cap 50 --weight 500
Scheduler : credit
error: invalid scheduler option: weight
The obvious fix is to assign the 'caps' parameter to the correct
item in the parameter list.
Reported-by: Volo M. <vm@vovs.net>
And allow libxl to handle channel element which creates a Xen
console visible to the guest as a low-bandwitdh communication
channel. If type is PTY we also fetch the tty after boot using
libxl_channel_getinfo to fetch the tty path. On socket case,
we autogenerate a path if not specified in the XML. Path autogenerated
is slightly different from qemu driver: qemu stores also on
"channels/target" but it creates then a directory per domain with
each channel target name. libxl doesn't appear to have a clear
definition of private files associated with each domain, so for
simplicity we do it slightly different. On qemu each autogenerated
channel goes like:
channels/target/<domain-name>/<target name>
Whereas for libxl:
channels/target/<domain-name>-<target name>
Should note that if path is not specified it won't persist,
existing only on live XML, unless user had initially specified it.
Since support for libxl channels only came on Xen >= 4.5 we therefore
need to conditionally compile it with LIBXL_HAVE_DEVICE_CHANNEL.
After this patch and having a qemu guest agent:
$ cat domain.xml | grep -a1 channel | head -n 5 | tail -n 4
<channel type='unix'>
<source mode='bind' path='/tmp/channel'/>
<target type='xen' name='org.qemu.guest_agent.0'/>
</channel>
$ virsh create domain.xml
$ echo '{"execute":"guest-network-get-interfaces"}' | socat
stdio,ignoreeof unix-connect:/tmp/channel
{"execute":"guest-network-get-interfaces"}
{"return": [{"name": "lo", "ip-addresses": [{"ip-address-type": "ipv4",
"ip-address": "127.0.0.1", "prefix": 8}, {"ip-address-type": "ipv6",
"ip-address": "::1", "prefix": 128}], "hardware-address":
"00:00:00:00:00:00"}, {"name": "eth0", "ip-addresses":
[{"ip-address-type": "ipv4", "ip-address": "10.100.0.6", "prefix": 24},
{"ip-address-type": "ipv6", "ip-address": "fe80::216:3eff:fe40:88eb",
"prefix": 64}], "hardware-address": "00:16:3e:40:88:eb"}, {"name":
"sit0"}]}
Signed-off-by: Joao Martins <joao.m.martins@oracle.com>
Signed-off-by: Jim Fehlig <jfehlig@suse.com>
libxl events are delivered to libvirt via the libxlDomainEventHandler
callback registered with libxl. Documenation in
$xensrc/tools/libxl/libxl_event.h states that the callback "may occur
on any thread in which the application calls libxl". This can result
in deadlock since many of the libvirt callees of libxl hold a lock on
the virDomainObj they are working on. When the callback is invoked, it
attempts to find a virDomainObj corresponding to the domain ID provided
by libxl. Searching the domain obj list results in locking each obj
before checking if it is active, and its ID equals the requested ID.
Deadlock is possible when attempting to lock an obj that is already
locked further up the call stack. Indeed, Max Ustermann reported an
instance of this deadlock
https://www.redhat.com/archives/libvir-list/2015-November/msg00130.html
Guido Rossmueller also recently stumbled across it
https://www.redhat.com/archives/libvir-list/2016-September/msg00287.html
Fix the deadlock by moving the lookup of virDomainObj to the
libxlDomainShutdownThread. After this patch, libxl events are
enqueued on the libvirt side and processed by dedicated thread,
avoiding the described deadlock.
Reported-by: Max Ustermann <ustermann78@web.de>
Reported-by: Guido Rossmueller <Guido.Rossmueller@gdata.de>
Just like virDomainDefPostParseCallback has gained new
parseOpaque argument, we need to follow the logic with
virDomainDeviceDefPostParse.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
We want to pass the proper opaque pointer instead of NULL to
virDomainDefParse and subsequently virDomainDefParseNode too.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Some callers might want to pass yet another pointer to opaque
data to post parse callbacks. The driver generic one is not
enough because two threads executing post parse callback might
want to see different data (e.g. domain object pointer that
domain def belongs to).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
