Peter Krempa [Tue, 27 Sep 2016 14:01:55 +0000 (16:01 +0200)]
qemu: monitor: Properly configure backend for UDP chardevs
Since introduction of chardev hotplug the code was wrong for the UDP
case and basically created a TCP socket instead. Use proper objects and
type for UDP.
Peter Krempa [Tue, 27 Sep 2016 13:49:05 +0000 (15:49 +0200)]
tests: qemumonitorjson: Don't do multiple tests in one virTestRun
The chardev attach test would do all the tests in one virTestRun
instance. If one sub-test failed then the test would report failure
improperly and the error would be hard to debug since the error pointer
was overwritten.
John Ferlan [Mon, 19 Sep 2016 12:27:24 +0000 (08:27 -0400)]
conf: Add a formatting macro for all the blkiotune values
Rather than copy-paste - use a macro
Unfortunately due to how the RNG schema was written keeping the 'value'
and 'value'_max next to each other in the XML causes a schema failure,
so the FORMAT has to write out singly rather than optimizing to write
out both values at once
John Ferlan [Mon, 19 Sep 2016 21:46:07 +0000 (17:46 -0400)]
qemu: Adjust how supportMaxOptions is used.
We're about to add more options, let's avoid having multiple if-then-else
which each try to set up the qemuMonitorJSONMakeCommand call with all the
parameters it knows about.
Instead, use the fact that when a NULL is found in the argument list that
processing of the remaining arguments stops and just have call.
John Ferlan [Fri, 23 Sep 2016 12:07:53 +0000 (08:07 -0400)]
qemu: Convert from shorthand to longer throttling names
We're about to add 6 new options and it appears (from testing) one cannot
utilize both the shorthand (alias) and (much) longer names for the arguments.
So modify the command builder to use the longer name and of course alter the
test output .args to have the similarly innocuous long name.
Also utilize a macro to build that name makes it so much more visually
appealing and saves a few characters or potential cut-n-paste issues.
John Ferlan [Mon, 19 Sep 2016 13:52:10 +0000 (09:52 -0400)]
tests: Add blkdeviotune-max xml2xmltest
It was missing... Also since I'm using the soft link from qemuxml2xmloutdata
to the qemuxml2argvdata file, modify the output file to have the necessary
<address> elements plus the mouse and keyboard.
Laine Stump [Wed, 5 Oct 2016 15:26:07 +0000 (11:26 -0400)]
qemu: allow 32 slots on pcie-expander-bus, not just 1
When I added support for the pcie-expander-bus controller in commit bc07251f, I incorrectly thought that it only had a single slot
available. Actually it has 32 slots, just like the root complex aka
pcie-root (the part that I *did* get correct is that unlike pcie-root
a pcie-expander-bus doesn't allow any integrated endpoint devices -
only pcie-root-ports and dmi-to-pci-controllers are allowed).
Jiri Denemark [Wed, 5 Oct 2016 07:19:19 +0000 (09:19 +0200)]
vsh: Fix warnings in command line completer
GCC complained that
vsh.c: In function 'vshReadlineOptionsGenerator':
vsh.c:2622:29: warning: unused variable 'opt' [-Wunused-variable]
const vshCmdOptDef *opt = &cmd->opts[list_index];
^
vsh.c: In function 'vshReadlineParse':
vsh.c:2830:44: warning: 'opt' may be used uninitialized in this function
[-Wmaybe-uninitialized]
completed_list = opt->completer(autoCompleteOpaque,
John Ferlan [Tue, 27 Sep 2016 13:39:21 +0000 (09:39 -0400)]
qemu: Create common code for JSON "query-block" call
Reduce some cut-n-paste code by creating common helper. Make use of the
recently added virJSONValueObjectStealArray to grab the devices list as
part of the common code (we we can Free the reply) and return devices for
each of the callers to continue to parse.
NB: This also adds error checking to qemuMonitorJSONDiskNameLookup
John Ferlan [Mon, 3 Oct 2016 18:45:13 +0000 (14:45 -0400)]
util: Introduce virJSONValueObjectStealArray
Provide the Steal API for any code paths that will desire to grab the
object array and then free it afterwards rather than relying to freeing
the whole chain from the reply.
John Ferlan [Wed, 14 Sep 2016 17:56:31 +0000 (13:56 -0400)]
rbd: Change to using heap allocated state contexts
Rather than use stack allocated state context pointers, let's allocate and
free the state context pointer. In doing so, we'll shrink the code a bit
since many routines perform the same initialization sequence.
Peter Krempa [Tue, 20 Sep 2016 12:08:55 +0000 (14:08 +0200)]
lib: Introduce VIR_DOMAIN_VCPU_HOTPLUGGABLE for virDomainSetVcpusFlags
For compatibility reasons virDomainSetVcpus needs to add vcpus as non
hotpluggable which means that the users will not be able to unplug it
after the VM has started. Add a flag that will allow to tell the API
that the unpluggable vcpus are okay.
Peter Krempa [Mon, 3 Oct 2016 11:11:47 +0000 (13:11 +0200)]
qemu: attach: Close monitor socket on connection failure
If attaching to a qemu process fails after opening the monitor socket
libvirt does not clean up the monitor. As the monitor also holds a
reference to the domain object the qemu attach API basically leaks it.
QEMU also does not interact on a second monitor connection and thus a
further attempt to attach to it would lock up.
Prevent libvirt from leaking the monitor by explicitly closing it.
Peter Krempa [Fri, 30 Sep 2016 10:45:59 +0000 (12:45 +0200)]
qemu: Don't strictly require JSON monitor for vCPU detection
Attaching to a existing qemu process allows to get us into a situation
when qemu is new enough to have JSON monitor and new vCPU hotplug but
the json monitor is not used. The vCPU detection code would require it
though. This broke attaching to qemu processes.
Make the condition less strict and just skip the vCPU hotplug detection
if JSON monitor is not available.
Resolves one of the symptoms in:
https://bugzilla.redhat.com/show_bug.cgi?id=1378401
Don't drop expired lease while reading custom leases file
Libvirt, on its own, shouldn't decide whether an expired lease should
stay in the custom leases database or not. It should rather rely on
the 'DEL' event from dnsmasq.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com> Reviewed-by: Michal Privoznik <mprivozn@redhat.com>
John Ferlan [Tue, 4 Oct 2016 11:07:20 +0000 (07:07 -0400)]
remote: Increase bound limit for virDomainGetBlockIoTune
We are about to add 6 new values to fetch. This will put us over the
current limit of 16 (we're at 13 now).
Once there are more than 16 parameters, this will affect existing clients
that attempt to fetch blockiotune config values for the domain from the
remote host since the server side has no mechanism to determine whether
the capability for the emulator exists and thus would attempt to return
all known values from the persistentDef. If attempting to fetch the
blockiotune values from a running domain, the code will check the emulator
capabilities and set maxparams (in qemuDomainGetBlockIoTune) appropriately.
On the client side of the remote connection, it uses this constant in
xdr_remote_domain_get_block_io_tune_ret and virTypedParamsDeserialize
calls, so if a remote server returns more than 16 parameters, then the
client will fail with "Unable to decode message payload".
John Ferlan [Tue, 4 Oct 2016 11:05:20 +0000 (07:05 -0400)]
remote: Fix erroneous usage of constant
The REMOTE_DOMAIN_MEMORY_PARAMETERS_MAX was erroneously used in the
remoteDomainBlockStatsFlags and remoteDomainGetBlockIoTune calls. Change
the constant to be the right one.
Revert "conf: Skip post parse callbacks when creating copy"
This breaks vCPU hotplug, because when starting a domain, we
create a copy of domain definition (which becomes live XML) and
during the post parse callbacks we might adjust some tunings so
that vCPU hotplug is possible.
This breaks vCPU hotplug, because when starting a domain, we
create a copy of domain definition (which becomes live XML) and
during the post parse callbacks we might adjust some tunings so
that vCPU hotplug is possible.
Peter Krempa [Wed, 21 Sep 2016 05:59:57 +0000 (07:59 +0200)]
qemu: vcpu: Clear vcpu order information rather than making it invalid
Certain operations may make the vcpu order information invalid. Since
the order is primarily used to ensure migration compatibility and has
basically no other user benefits, clear the order prior to certain
operations and document that it may be cleared.
All the operations that would clear the order can still be properly
executed by defining a new domain configuration rather than using the
helper APIs.
Peter Krempa [Fri, 16 Sep 2016 13:51:14 +0000 (15:51 +0200)]
qemu: process: Enforce 'vcpu' order range to <1,maxvcpus>
The current code that validates duplicate vcpu order would not work
properly if the order would exceed def->maxvcpus. Limit the order to the
interval described.
Peter Krempa [Wed, 14 Sep 2016 03:58:33 +0000 (05:58 +0200)]
qemu: process: Fix off-by-one in vcpu order duplicate error message
The bitmap indexes for the order duplicate check are shifted to 0 since
vcpu order 0 is not allowed. The error message doesn't need such
treating though.
Laine Stump [Thu, 29 Sep 2016 17:38:07 +0000 (13:38 -0400)]
docs: correct version requirements for <kvm><hidden='on'/></kvm>
When support was added for the kvm hidden='on' attribute in commit
d07116, the version requirement was listed as "2.1.0 (QEMU
only)". However, this was added when libvirt was at version 1.2.8 - it
is *QEMU* that must be at version 2.1.0 or later.
This went unnoticed for a very long time (over 2 years). Then a week
or two ago a new Windows convert in the #virt channel on OFTC was told
he needed to use this feature (to prevent nvidia drivers in a guest
from refusing to work due to being run in a virtual machine). There
was some problem with it being recognized and "someone" (it may have
been me, or may have been someone else, I don't remember) pointed out
that the documentation at
http://www.libvirt.org/formatdomain.html
says that it requires libvirt 2.1.0. The next several days were filled
with agony as a new convert to Linux first tried to upgrade a Linux
Mint host running their "LTS" version to something newer, then tried
to install a libvirt build built for Ubuntu onto this, and later back
to the old LTS Linux Mint. After this he tried building his own
libvirt from source (with all the expected problems), and finally
switched to Fedora. In the end it was hours and hours of everybody's
lives that they will never get back. To now learn that he didn't need
to do this (his original libvirt version was 1.3.3, so whatever his
problem was, it was elsewhere) makes the pain all that much worse.
To prevent this from happening again, this simple patch changes the
version requirement for the kvm hidden attribute from "2.1.0 (QEMU
only)" to "1.2.8 (QEMU 2.1.0)".
Hold on to your hats, because this is gonna be wild.
In bd3e16a3 I've tried to expose sanlock io_timeout. What I had
not realized (because there is like no documentation for sanlock
at all) was very unusual way their APIs work. Basically, what we
do currently is:
sanlock_add_lockspace_timeout(&ls, io_timeout);
which adds a lockspace to sanlock daemon. One would expect that
io_timeout sets the io_timeout for it. Nah! That's where you are
completely off the tracks. It sets timeout for next lockspace you
will probably add later. Therefore:
sanlock_add_lockspace_timeout(&ls, io_timeout = 10);
/* adds new lockspace with default io_timeout */
sanlock_add_lockspace_timeout(&ls, io_timeout = 20);
/* adds new lockspace with io_timeout = 10 */
sanlock_add_lockspace_timeout(&ls, io_timeout = 40);
/* adds new lockspace with io_timeout = 20 */
And so on. You get the picture.
Fortunately, we don't allow setting io_timeout per domain or per
domain disk. So we just need to set the default used in the very
first step and hope for the best (as all the io_timeout-s used
later will have the same value).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Michal Privoznik [Thu, 15 Sep 2016 08:45:26 +0000 (10:45 +0200)]
m4: Check for sanlock_write_lockspace
Currently, we are checking for sanlock_add_lockspace_timeout
which is good for now. But in a subsequent patch we are going to
use sanlock_write_lockspace (which sets an initial value for io
timeout for sanlock). Now, there is no reason to check for both
functions in sanlock library as the sanlock_write_lockspace was
introduced in 2.7 release and the one we are currently checking
for in the 2.5 release. Therefore it is safe to assume presence
of sanlock_add_lockspace_timeout when sanlock_write_lockspace
is detected.
Moreover, the macro for conditional compilation is renamed to
HAVE_SANLOCK_IO_TIMEOUT (as it now encapsulates two functions).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
There is a logic in place that if there is no real need for
memory-backend-file, qemuBuildMemoryBackendStr() returns 0. However
that wasn't the case with hugepage backing. The reason for that was
that we abused the 'pagesize' variable for storing that information, but
we should rather have a separate one that specifies whether we really
need the new object for hugepage backing. And that variable should be
set only if this particular NUMA cell needs special treatment WRT
hugepages.
During the code motion and creation of vsh.c, the function 'vshDeinit()'
in the (new) vsh.c was altered from whence it came in virsh.c such that
calling 'vshReadlineDeinit(ctl)' was conditional on "ctl->imode".
This causes a problem for the interactive running if the "quit" and "exit"
commands are used because 'cmdQuit' will clear ctl->imode, thus when the
interactive loop in main() of virsh.c exits because ctl->mode is clear and
virshDeinit is called which calls vshDeinit, the history file is now not
written. Conversely, if one had exited the interactive loop via pressing
<ctrl>D the file would be created because loop control is broken on EOF
and ctl->imode is not set to false.
This patch will remove the conditional call to vshReadlineDeinit and
restore the former behaviour.
Jim Fehlig [Wed, 28 Sep 2016 04:43:51 +0000 (22:43 -0600)]
doc: fix note about Xen credit scheduler
Commit 6c504d6a added a note to the virsh man page about the
deprecation of 'cap' and 'weight' settings for the credit
scheduler. To this day, the default scheduler in Xen is credit
and it supports setting 'cap' and 'weight'. Remove the deprecation
notice from the note on the Xen credit scheduler.
Jim Fehlig [Wed, 28 Sep 2016 04:18:50 +0000 (22:18 -0600)]
libxl: fix param assignment in domainGetSchedulerParameters
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
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>
The qemucapsprobe helper calls virQEMUCapsNewForBinaryInternal with
caps == NULL, causing the following crash:
Program received signal SIGSEGV, Segmentation fault.
#0 0x00007ffff788775f in virQEMUCapsInitHostCPUModel
(qemuCaps=qemuCaps@entry=0x649680, host=host@entry=0x10) at
src/qemu/qemu_capabilities.c:2969
#1 0x00007ffff7889dbf in virQEMUCapsNewForBinaryInternal
(caps=caps@entry=0x0, binary=<optimized out>,
libDir=libDir@entry=0x4033f6 "/tmp", cacheDir=cacheDir@entry=0x0,
runUid=runUid@entry=4294967295, runGid=runGid@entry=4294967295,
qmpOnly=true) at src/qemu/qemu_capabilities.c:4039
#2 0x0000000000401702 in main (argc=2, argv=0x7fffffffd968) at
tests/qemucapsprobe.c:73
So far, when it comes to 'virsh update-device --config' of disks
we are limiting ourselves for just the disk source update and
just for CDROMs and floppies. This makes no sense. Especially if
you look around and see that we already allow full update to
graphics and net devices. So let's just take whatever XML user
wants to have there and replace our internal definition with it.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Jim Fehlig [Wed, 21 Sep 2016 21:02:34 +0000 (15:02 -0600)]
libxl: find virDomainObj in libxlDomainShutdownThread
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
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>
qemu: make qemuGetCompressionProgram return int not an enum
enum types are unsigned and the qemuGetCompressionProgram
function can return -1 on error. It is therefore inappropriate
to return an enum type. This fixes a build error where the
internal 'ret' variable was used in a comparison with -1
../../src/qemu/qemu_driver.c: In function 'qemuGetCompressionProgram':
../../src/qemu/qemu_driver.c:3280:5: error: comparison of unsigned expression < 0 is always false [-Werror=type-limits]
../../src/qemu/qemu_driver.c:3289:5: error: comparison of unsigned expression < 0 is always false [-Werror=type-limits]
cc1: all warnings being treated as errors
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Michal Privoznik [Tue, 20 Sep 2016 12:21:26 +0000 (14:21 +0200)]
conf: Skip post parse callbacks when creating copy
When creating a copy of virDomainDef we save ourselves the
trouble of writing deep-copy functions and just format and parse
back domain/device XML. However, the XML we are parsing was
already fully formatted - there is no reason to run post parse
callbacks (which fill in blanks - there are none!).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
This is an internal flag that prevents our two entry points to
XML parsing (virDomainDefParse and virDomainDeviceDefParse) from
running post parse callbacks. This is expected to be used in
cases when we already have full domain/device XML and we are just
parsing it back (i.e. virDomainDefCopy or virDomainDeviceDefCopy)
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Michal Privoznik [Fri, 16 Sep 2016 14:29:41 +0000 (16:29 +0200)]
qemuDomainDefPostParse: Fetch qemuCaps from domain object
We can't rely on def->emulator path. It may be provided by user
as we give them opportunity to provide their own XML for
migration. Therefore the path may point to just whatever binary
(or even to a non-existent file). Moreover, this path is meant
for destination, but the capabilities lookup is done on source.
What we can do is to assume same capabilities for post parse
callbacks as the running domain has. They will be used just to
add some default models/controllers/devices/... anyway.
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>
John Ferlan [Tue, 13 Sep 2016 16:05:39 +0000 (12:05 -0400)]
qemu: Get/return compressedpath program
Based upon a patch from Chen Hanxiao <chenhanxiao@gmail.com>, rather than
need to call virFindFileInPath twice, let's just save the path and pass it
along with the compressed type. (NB: the second call would be in virExec as
called from virCommandRunAsync which is called from qemuMigrationToFile
using the argument 'compressor' which up to this point would be the string
from the cfg file that isn't the fully qualified path).
Since we now have the path, we can remove qemuCompressProgramName which
would return NULL or the string representation of the compress type.
John Ferlan [Tue, 13 Sep 2016 15:51:54 +0000 (11:51 -0400)]
qemu: Remove qemuCompressProgramAvailable
There's only one caller and the code is duplicitous just converting the
recently converted cfg image name back into it's string value in order to
get/find the path to the image. A subsequent patch can return this path.
John Ferlan [Tue, 13 Sep 2016 14:11:00 +0000 (10:11 -0400)]
qemu: Use qemuGetCompressionProgram for error paths
Let's do some more code reuse - there are 3 other callers that care to
check/get the compress program. Each of those though cares whether the
requested cfg image is valid and exists. So, add a parameter to handle
those cases.
NB: We won't need to initialize the returned value in the case where
the cfg image doesn't exist since the called program will handle that.
John Ferlan [Fri, 23 Sep 2016 11:09:03 +0000 (07:09 -0400)]
qemu: Alter qemuGetCompressionProgram warning message
Add a new parameter 'styleFormat' to be used when printing the
warning message so that it's "clearer" what style of compression
call caused the error. Add that style to both messages as a paremter.
Also a VIR_WARN error message doesn't need to be translated
(e.g. inside _()), so remove the need for the translation.
John Ferlan [Tue, 13 Sep 2016 14:31:37 +0000 (10:31 -0400)]
qemu: Remove getCompressionType
There's only one caller now anyway... Besides it's just a shell for
getting the compress type. Subsequent patches will return the path
to the compression program.
John Ferlan [Tue, 13 Sep 2016 14:01:47 +0000 (10:01 -0400)]
qemu: Introduce helper qemuGetCompressionProgram
Split out the guts of getCompressionType to perform the same functionality
in the new helper program with a subsequent patch goal to be reusable for
other callers making similar checks/calls to ensure the compression type
is valid and that the compression program cannot be found.
libxl: increase usbdevice list only when finding such an input device
If passing an empty usbdevice_list to libxl, qemu will always get an
-usb parameter for HVM guests with only non-USB input devices. This
causes qemu to crash when passing pvusb device on HVM guests.
The solution is to allocate the list only when an item to put in it
is found.
Pavel Hrdina [Thu, 22 Sep 2016 14:36:33 +0000 (16:36 +0200)]
qemuhelptest: regenerate data for qemu-1.2.0
This patch also removes device data for qemu-1.2.0 as it was removed for
qemu-kvm-1.2.0 by commit ae3e29e6e. They are not required because we
parse only version from help output and return with error that this qemu
is too new to use help parsing.
Jiri Denemark [Wed, 22 Jun 2016 13:53:48 +0000 (15:53 +0200)]
qemu: Update guest CPU def in live XML
Storing the updated CPU definition in the live domain definition saves
us from having to update it over and over when we need it. Not to
mention that we will soon further update the CPU definition according to
QEMU once it's started.
A highly wanted side effect of this patch, libvirt will pass all CPU
features explicitly specified in domain XML to QEMU, even those that are
already included in the host model.
This patch should fix the following bugs:
https://bugzilla.redhat.com/show_bug.cgi?id=1207095
https://bugzilla.redhat.com/show_bug.cgi?id=1339680
https://bugzilla.redhat.com/show_bug.cgi?id=1371039
https://bugzilla.redhat.com/show_bug.cgi?id=1373849
https://bugzilla.redhat.com/show_bug.cgi?id=1375524
https://bugzilla.redhat.com/show_bug.cgi?id=1377913
Jiri Denemark [Tue, 9 Aug 2016 11:26:53 +0000 (13:26 +0200)]
cpu: Rework cpuCompare* APIs
Both cpuCompare* APIs are renamed to virCPUCompare*. And they should now
work for any guest CPU definition, i.e., even for host-passthrough
(trivial) and host-model CPUs. The implementation in x86 driver is
enhanced to provide a hint about -noTSX Broadwell and Haswell models
when appropriate.
The function is similar to virCPUDataCheckFeature, but it works directly
on CPU definition rather than requiring it to be transformed into CPU
data first.