Integrate with QEMU monitor to provide encryption passphrase when
starting a guest using encrypted qcow volumes
* src/qemu_driver.c (findDomainDiskEncryption,
findVolumeQcowPassphrase,
qemudMonitorSendVolumePassphrase, qemudMonitorSendCont): Send a volume
passphrase if qemu asks for it.
If the <encryption format='qcow'> element does not specify a secret
during volume creation, generate a suitable secret and add it to the
<encryption> tag. The caller can view the updated <encryption> tag
using virStorageVolGetXMLDesc().
Similarly, when <encryption format='default'/> is specified while
creating a qcow or qcow2-formatted volume, change the format to "qcow"
and generate a secret as described above.
* src/storage_encryption_conf.h (VIR_STORAGE_QCOW_PASSPHRASE_SIZE,
virStorageGenerateQcowPasphrase),
src/storage_encryption_conf.c (virStorageGenerateQcowPasphrase),
src/libvirt_private.syms: Add virStorageGenerateQcowPasphrase().
* src/storage_backend.c (virStoragegenerateQcowEncryption,
virStorageBackendCreateQemuImg): Generate a passphrase and
<encryption> when creating a qcow-formatted encrypted volume and the
user did not supply the information.
This implementation stores the secrets in an unencrypted text file,
for simplicity in implementation and debugging.
(Symmetric encryption, e.g. using gpgme, will not be difficult to add.
Because the TLS private key used by libvirtd is stored unencrypted,
encrypting the secrets file does not currently provide much additional
security.)
* include/libvirt/virterror.h, src/virterror.c (VIR_ERR_NO_SECRET): New
error number.
* po/POTFILES.in, src/Makefile.am: Add secret_driver.
* bootstrap: Use gnulib's base64 module.
* src/secret_driver.c, src.secret_driver.h, src/libvirt_private.syms:
Add local secret driver.
* qemud/qemud.c (qemudInitialize): Use the local secret driver.
Add a <secret> XML handling API, separate from the local driver, to
avoid manually generating XML in other parts of libvirt.
* src/secret_conf.c, src/secret_conf.h: New files.
* po/POTFILES.in, src/Makefile.am: Add secret_conf.
Add a VIR_SECRET_GET_VALUE_INTERNAL_CALL flag value, replacing the
originally separate libvirt_internal_call parameter. The flag is used
to differentiate external virSecretGetValue() calls from internal calls
by libvirt drivers that need to use the secret even if it is private.
* src/libvirt_internal.h Remove VIR_DOMAIN_XML_FLAGS_MASK
* src/driver.h Add VIR_SECRET_GET_VALUE_FLAGS_MASK constant and
VIR_SECRET_GET_VALUE_INTERNAL_CALL. Re-add the
VIR_DOMAIN_XML_FLAGS_MASK constant
* src/libvirt.c (virSecretGetValue): Don't allow the user to specify
internal flags.
Remove the bogus dependancy between node_device.c & storage_backend.c
by moving the virWaitForDevices into util.h where it can be shared
safely
* src/storage_backend_disk.c, src/storage_backend_logical.c,
src/storage_backend_mpath.c, src/storage_backend_scsi.c: Replace
virStorageBackendWaitForDevices with virFileWaitForDevices
* src/storage_backend.c, src/storage_backend.h: Remove
virStorageBackendWaitForDevices, virWaitForDevices
* src/util.h, src/util.c: Add virFileWaitForDevices
* configure.in: Move xmlrpc check further down after pkgconfig
is detected
* src/Makefile.am: Add missing XMLRPC_CFLAGS/LIBS to opennebula
* src/libvirt_private.syms: Add many missing exports
The interface allows qemudMonitorSendCont() to report errors that are
not overridden by its callers.
Also fix a potential infinite loop in qemuDomainCoreDump() if sending
cont repeatedly fails.
* src/qemu_driver.c (qemudMonitorSendCont): New function.
(qemudAutostartConfigs): Reset error before each call to
qemudStartVMDaemon().
(qemudInitCpus, qemudDomainResume, qemudDomainCoreDump,
qemudDomainRestore, qemudDomainMigratePerform,
qemudDomainMigrateFinish2): Use qemudMonitorSendCont().
* src/security_selinux.c: Use virReportSystemError whereever an
errno is involved
* src/qemu_driver.c: Don't overwrite error message from the
security driver
* src/security.h: Driver API for relabelling host devices
* src/security_selinux.c: Implement relabelling of PCI and USB
devices
* src/qemu_driver.c: Relabel USB/PCI devices before hotplug
Fix all thread locking bugs reported by object-locking test
case.
NB, some of the driver locking is getting too coarse. Driver
mutexes really need to be turned into RW locks instead to
significantly increase concurrency.
* src/lxc_driver.c: Fix useof driver when unlocked in the methods
lxcDomainGetInfo, lxcSetSchedulerParameters, and
lxcGetSchedulerParameters
* src/opennebula/one_driver.c: Fix missing unlock in oneDomainUndefine.
Fix use of driver when unlocked in oneDomainGetInfo,
oneGetOSType, oneDomainShutdown
* src/qemu_driver.c: Fix use of driver when unlocked in
qemudDomainSavem, qemuGetSchedulerType, qemuSetSchedulerParameters
and qemuGetSchedulerParameters
* src/storage_driver.c: Re-work storagePoolCreate to avoid bogus
lock checking warning. Re-work storageVolumeCreateXMLFrom to
remove a potential NULL de-reference & avoid bogus lock check
warnings
* src/test.c: Remove testDomainAssignDef since it break lock chekc
warnings.
* tests/object-locking.ml: Add oneDriverLock, oneDriverUnlock
and one_driver_t methods/types to allow lock checking on the
OpenNebula drivers
The algorithm is quite simple:
If the emulator matches a guest's domain:
if domain has machine type info:
check the domain's machine type info
else
check the guest's default machine type info
else if the emulator matches the guest's default emulator:
check the guest's default machine type info
The previous implementation was incorrectly falling back to the default
machine type info if the domain's machine type info didn't have an
alias.
* src/qemu_driver.c: simplify and fix qemudCanonicalizeMachine()
Currently we only probe the main qemu binary for machine types, but we
should also probe the kvm binary.
* src/qemu_conf.c: probe kvm binary machines in qemudCapsInitGuest()
We need to look at all the domain infos in guest capabilities, not
just the defaults.
In order to allow that, split out a qemudGetOldMachinesFromInfo()
from qemudGetOldMachines(). We'll make more use of it in the next
patch.
* src/qemu_conf.c: split out qemudGetOldMachinesFromInfo() from
qemudGetOldMachines()
This doesn't have any affect on the current tests because we don't have
any machine aliases in the current test data.
* src/qemu_conf.h, src/qemu_driver.c: expose qemudCanonicalizeMachine()
for the tests
* tests/qemuxml2argvtest.c: canonicalize the machine type
After the mentioned patch was applied, I noticed that
shutting down a kvm guest from inside (i.e. poweroff) caused
the guest to shutdown, but not removed from the list of
active guests. DanB pointed out that the problem is that
the virEventAddHandle() call in the qemu driver was asking
to watch for 0 events, not HANGUP | ERROR as it should. Add
these events so that shutdown works again.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
lzop was removed due to some confusion over whether it provided functional
advantages distinct from xz. This has been addressed in the mailing list post
archived at http://permalink.gmane.org/gmane.comp.emulators.libvirt/16487, and
support for lzop is re-added here.
* libvirt.spec.in: add dependancy on lzop
* src/qemu.conf: update documentation of save_image_format
* src/qemu_driver.c: re-add lzop compression option
* src/qemu_driver.c: drop lzma and lzop images compression options
as they are deprecated by xz
* libvirt.spec.in: add requires for xz/bzip2/gzip as they are needed
to implement the compression options
* src/qemu_driver.c (QEMUD_SAVE_FORMAT_LAST): Define.
(qemudSaveCompressionTypeFromString): Declare.
(qemudSaveCompressionTypeToString): Declare.
(qemudDomainSave): Use those functions rather than open-coding them.
Use "cat >> '%s' ..." in place of equivalent
"dd of='%s' oflag=append conv=notrunc ...".
* src/qemu_driver.c (enum qemud_save_formats) [QEMUD_SAVE_FORMAT_XZ]:
New member.
[QEMUD_SAVE_FORMAT_LZMA]: Mark as deprecated.
Use an explicit value for each member.
(qemudDomainSave, qemudDomainRestore): Handle the new member.
* src/qemu.conf: Mention xz, too.
* configure.in src/Makefile.am src/storage_backend.[ch]
src/storage_conf.[ch] src/storage_backend_mpath.[ch] po/POTFILES.in:
add a new module for storage multipath, it requires device-mapper
* src/xm_internal.c (xenXMDomainConfigParse): After t=strchr...
don't test *t; it's known. This was *not* detected by clang,
but I spotted it since once instance was in the vicinity of the
dead increment of "data".
* src/vbox/vbox_tmpl.c: minor bug in selecting the graphics type. if the
graphics type was desktop it was assumed that display is set for it,
and thus crashed on strdup. Also adds a number of missing OOM checks.
* src/lxc_container.c (lxcContainerMountBasicFS): Don't leak upon failure.
Add "cleanup:" label and change each post-allocation failure to
use "goto cleanup" rather than returning immediately.
* src/qemu_conf.c (qemuBuildHostNetStr): Do not remove the type_sep=','
dead store, since not having it would be a problem if we ever add a
new attribute=VAL option.
* src/esx/esx_vi.[ch]: convert esxVI_RemoteRequest_Execute() to a
simpler esxVI_Context_Execute() version, remove esxVI_RemoteRequest
and convert esxVI_RemoteResponse to esxVI_Response
* src/esx/esx_vi_methods.c: update and simplify callers to use
esxVI_Context_Execute() instead of esxVI_RemoteRequest_Execute()
* src/esx/esx_vi.[ch]: use virXPathNode*() in
esxVI_RemoteRequest_Execute() and remove
esxVI_RemoteResponse_DeserializeXPathObject*()
* src/esx/esx_vi_methods.c: update callers to use the new syntax of
esxVI_RemoteRequest_Execute()
* src/esx/esx_driver.c: add configStatus to the requested properties
to check it in esxVI_GetVirtualMachineIdentity()
* src/esx/esx_vi.[ch]: add esxVI_GetManagedEntityStatus()
and use it in esxVI_GetVirtualMachineIdentity()
* src/esx/esx_vi_types.[ch]: add VI type esxVI_ManagedEntityStatus
* src/esx/esx_driver.c: handle spaces in VMX file path and use a
virBuffer to encode spaces correctly in the resulting URL
* src/esx/esx_vi.c: include the URL in the error message in case
of a download error
* configure.in src/Makefile.am: change detection and flags
* src/phyp/phyp_driver.c src/phyp/phyp_driver.h: connection now
need to be done as part of the driver code, cleaned up by DV
* src/qemu_driver.c: Free the vm->monitor_chr field at VM shutdown.
Unlink the UNIX domain socket at VM shutdown to avoid littering
FS with old sockets
Separate the guest created QEMU monitor socket location
from the libvirtd create XML / PID data files, to improve
security separation when running QEMU non-root
* libvirt.spec.in: Leave /var/run/libvirt/qemu as root:root
* src/qemu_conf.h: Add libDir and cacheDir directory paths
* src/qemu_driver.c: Move QEMU monitor socket from
stateDir to libDir to avoid making security critical directory
accessible to QEMU guests.
* src/util.c: Delay running hook till after damonizing to
ensure pidfile is still written before changing UID/GID
* src/openvz_conf.c (openvzReadNetworkConf): Initialize "net".
Otherwise, upon openvzRead... failure, we would "goto error;"
where an uninitialized "net" could be dereferenced.
* src/test.c (testOpenVolumesForPool): Upon early virAsprintf or
virXPathNodeSet failure, "goto error" would take us to
virStorageVolDefFree(def), but with "def" not defined.
Initialize it to NULL.
* src/storage_backend.c: Include "internal.h".
(virStorageBackendCopyToFD): Mark inputvol parameter as "nonnull".
Remove test for non-NULL inputvol. Both callers ensure it's non-NULL.
* src/libvirt.c (virStoragePoolSetAutostart): Return -1 if the pool
argument is invalid, rather than "goto error" where we could dereference
that possibly-NULL "pool".
(virConnectFindStoragePoolSources): Likewise.
(virConnectNumOfDomains): Likewise.
Daniel P. Berrange spotted that the two latter functions
needed the same treatment.
* docs/schemas/domain.rng: Add <serial> element to disks
* src/domain_conf.h, src/domain_conf.c: XML parsing and
formatting for disk serial numbers
* src/qemu_conf.c: Set serial number when launching guests
* tests/qemuxml2argvdata/qemuxml2argv-disk-drive-shared.args,
tests/qemuxml2argvdata/qemuxml2argv-disk-drive-shared.xml: Add
serial number to XML test
Add option to domain XML for
<memoryBacking>
<hugepages/>
</memoryBacking>
* configure.in: Add check for mntent.h
* qemud/libvirtd_qemu.aug, qemud/test_libvirtd_qemu.aug, src/qemu.conf
Add 'hugetlbfs_mount' config parameter
* src/qemu_conf.c, src/qemu_conf.h: Check for -mem-path flag in QEMU,
and pass it when hugepages are requested.
Load hugetlbfs_mount config parameter, search for mount if not given.
* src/qemu_driver.c: Free hugetlbfs_mount/path parameter in driver shutdown.
Create directory for QEMU hugepage usage, chowning if required.
* docs/formatdomain.html.in: Document memoryBacking/hugepages elements
* docs/schemas/domain.rng: Add memoryBacking/hugepages elements to schema
* src/util.c, src/util.h, src/libvirt_private.syms: Add virFileFindMountPoint
helper API
* tests/qemuhelptest.c: Add -mem-path constants
* tests/qemuxml2argvtest.c, tests/qemuxml2xmltest.c: Add tests for hugepage
handling
* tests/qemuxml2argvdata/qemuxml2argv-hugepages.xml,
tests/qemuxml2argvdata/qemuxml2argv-hugepages.args: Data files for
hugepage tests
* tests/testutils.c: Run test function twice, once to prime it for
static allocations, once to count the non-static allocations.
* tests/testutilsqemu.c: Initialize variable correctl
* src/capabilities.c: Don't free machines variable upon failure
since caller must do that
* src/xm_internal.c: Add missing check for OOM in building VIF
config param
* docs/schemas/domain.rng: augment the video model with an optional
acceleration element with optional accel2d and accel3d flags
* src/domain_conf.c src/domain_conf.h: exten the virDomainVideoDef
structure with an optional accel field, virDomainVideoAccelDefParseXML
and virDomainVideoAccelDefFormat functions to parse and serialize
the structure.
Paolo Bonzini points out that in my refactoring of the code for
virDomainMigrate(), I added a check for the return value from
virDomainMigratePerform(). The problem is that we don't want to
exit if we fail, we actually want to go on and do
virDomainMigrateFinish2() with a non-0 return code to clean things
up. Remove the check.
While reproducing this issue, I also noticed that we wouldn't
always properly propagate an error message. In particular, I
found that if you blocked off the migration ports (with iptables)
and then tried the migration, it would actually fail but we would
get no failure output from Qemu. Therefore, we would think we
succeeded, and leave a huge mess behind us. Execute the monitor
command "info migrate", and look for a failure string in there
as well.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
* src/esx/esx_util.c: esxUtil_ParseQuery() warns if a known query
parameter should be ignored due to the corresponding char/int pointer
being NULL, instead of silently ignoring it. Fix the control flow.
* src/esx/esx_vmx.c: add an extra type of addressType beside 'static'
and 'generated', 'vpx' indicates that the MAC address was generated
by a vCenter.
Calling qsort() on the disks array causes disk to be
unneccessarily re-ordered, potentially breaking the
ability to boot if the boot disk gets moved later in
the list. The new algorithm will insert a new disk as
far to the end of the list as possible, while being
ordered correctly wrt other disks on the same bus.
* src/domain_conf.c, src/domain_conf.h: Remove disk sorting
routines. Add API to insert a disk into existing list at
the optimal position, without resorting disks
* src/libvirt_private.syms: Export virDomainDiskInsert
* src/xend_internal.c, src/xm_internal.c: Remove calls to
qsort, use virDomainDiskInsert instead.
* src/qemu_driver.c: Remove calls to qsort, use virDoaminDiskInsert
instead. Fix reordering bugs when hotunplugging disks and
networks. Fix memory leak in disk/net unplug
* proxy/Makefile.am: Build storage_encryption_conf.c since its a
dependancy of domain_conf.c
* src/storage_encryption_conf.c: Disable XML parsing APis when
build under proxy
* src/test.c: Add a dummy no-op secrets driver for test suite
The if ((nlptr...)) implicitly assumes commptr != NULL (and that "buf"
starts with "cmd"). Make the assumption explicit, it will be broken in
a future patch.
* src/qemu_driver.c: Don't assume buffered monitor output echoes the
command.
The XML allows <encryption format='unencrypted'/>, this implementation
canonicalizes the internal representation so that "disk->encryption" is
non-NULL iff encryption information is available.
A domain with partial encryption information can be defined,
completeness of the information is not verified. The domain won't
start until the remaining information is added, of course.
* docs/formatdomain.html, docs/formatdomain.html.in: Document
new encryption options for disks
* docs/schemas/domain.rng: Pull in storage encryption schema
rules
* src/domain_conf.h, src/domain_conf.c: Wire up storage encryption
XML parsing/formatting APIs
Supports only virStorageVolCreateXML, not virStorageVolCreateXMLFrom.
Curiously, qemu-img does not need the passphrase for anything to create
an encrypted volume. This implementation thus does not need to touch
any secrets to work with cooperating clients. More generic passphrase
handling is added in the next patch.
* src/storage_backend.c: Request encryption when creating qcow/qcow2
files
* src/storage_backend_disk.c, src/storage_backend_fs.c,
src/storage_backend_logical.c: Refuse to create volumes with
encryption params set.
(The implementation is not very generic, but that can be very
easily rectified if/when new encryption formats appear.)
* src/storage_backend_fs.c: Probe for qcow/qcow2 encryption
algorithm field
The XML allows <encryption format='unencrypted'/>, this implementation
canonicalizes the internal representation so that "vol->encryption" is
non-NULL iff the volume is encrypted.
Note that partial encryption information (e.g. specifying an encryption
format, but not the key/passphrase) is valid, libvirt will automatically
choose value for the missing information during volume creation. The
user can read the volume XML, and use the unmodified <encryption> tag in
future operations (without having to be able to understand) its contents.
* docs/formatstorage.html, docs/formatstorage.html.in: Document
storage volume encryption options
* src/storage_conf.c, src/storage_conf.h: Hook up storage
encryption XML handling
* tests/storagevolschemadata/vol-qcow2.xml: Test case for encryption
schema changes
Define an <encryption> tag specifying volume encryption format and
format-depenedent parameters (e.g. passphrase, cipher name, key
length, key).
Currently the only defined parameter is a reference to a "secret"
(passphrase/key) managed using the virSecret* API.
Only the qcow/qcow2 encryption format, and a "default" format used to
let libvirt choose the format during volume creation, is currently
supported.
This patch does not add any users; the <encryption> tag is added in
the following patches to both volumes (to support encrypted volume
creation) and domains.
* docs/*.html: Re-generate
* docs/formatstorageencryption.html.in, docs/sitemap.html.in:
Add page describing storage encryption data format
* docs/schemas/Makefile.am, docs/schemas/storageencryption.rng:
Add RNG schema for storage encryption format
* po/POTFILES.in: Add src/storage_encryption_conf.c
* src/libvirt_private.syms: Export virStorageEncryption* functions
* src/storage_encryption_conf.h, src/storage_encryption_conf.c: Internal
helper APIs for dealing with storage encryption format
* libvirt.spec.in, mingw32-libvirt.spec.in: Add storageencryption.rng
RNG schema
* include/libvirt/virterror.h, src/virterror.c: Add VIR_WAR_NO_SECRET
* src/libvirt_private.syms, src/datatypes.h, src/datatypes.c: Type
virSecret struct definition and helper APIs
* src/driver.h: Sub-driver API definitions for secrets
* src/libvirt.c: Define new sub-driver for secrets
This patch adds a "secret" as a separately managed object, using a
special-purpose API to transfer the secret values between nodes and
libvirt users.
* docs/schemas/secret.rng, docs/schemas/Makefilem.am: Add new
schema for virSecret objects
* docs/*html: Re-generated
* docs/formatsecret.html.in, docs/sitemap.html.in: Add page
describing the virSecret XML schema
* include/libvirt/libvirt.h.in: Define the new virSecret public
API
* src/libvirt_public.syms: Export symbols for new public APIs
* mingw32-libvirt.spec.in, libvirt.spec.in: Add secret.rng to
files list
Per prior discussion -- this was, indeed, trivial.
I'm a little disappointed to be breaking the ordering characteristics of
the enum (as it had been ordered by increasing time requirements and
decreasing output size), but breaking any save files with the old
constants in the headers would of course be worse.
>From 2a9cdcfc88de091a8d34aa3fc3b1208d7681790e Mon Sep 17 00:00:00 2001
From: Charles Duffy <Charles_Duffy@dell.com>
Date: Fri, 28 Aug 2009 11:49:54 -0500
Subject: [PATCH] support lzop save compression for qemu
One of the larger disincentives towards use of compression for migrated-out save
files is performance impact. This patch adds support for lzop; CPU time for
compression is about 5x faster than gzip (the next most performant algorithm)
and decompression is about 3x faster.
Signed-off-by: Charles Duffy <Charles_Duffy@dell.com>
Signed-off-by: Chris Lalancette <clalance@redhat.com>
All of the other drivers that support the getMaxVcpus callback
also accept a NULL value for type. Make openvz also accept a
NULL value.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
qemudExtractMonitorPath() was doing a VIR_ALLOC_N followed by a
strncpy. However, this isn't necessary; we can do the same thing
using strndup, which is much safer.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
The documentation for virNodeGetCellsFreeMemory claims the values
returned are in kilobytes, but that's actually wrong; the value
returned is actually in bytes. Fix up the documentation to be
correct.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
S-expression containing empty lists, e.g. (cpus (() () () ())),
was not being handled properly in sexpr2string() serialization.
Emit an empty list when encountering NIL sexpr kind.
* src/remote_internal.c: Split remoteAuthPolkit into separate
impls for v0 and v1 to avoid compile warnings due to unused
variables/params
* qemud/remote.c: Remove accidental tabs
* configure.in: Check for pkcheck which indicates new policykit
* qemud/Makefile.am: Install different versions of policy
* qemud/libvirtd.policy: Rename to libvirtd.policy-0
* qemud/libvirtd.policy-1: new style policy
* qemud/qemud.c, qemud/qemud.h, qemud/remote.c: Support new
policykit API via external pkcheck helper
* src/remote_internal.c: Don't prompt for polkit auth with new
policykit API
* libvirt.spec.in: deal with new policy install locations & deps
Matthias correctly points out that escape_specialcharaters() takes a
length, and since we are now malloc()'ing string in phypOpen instead of
making it a static array, we can't use sizeof(string) anymore. Calculate
the proper strlen and then use that both to allocate the string and also
pass it to escape_specialcharacters().
Signed-off-by: Chris Lalancette <clalance@redhat.com>
I came across this line in the phypOpen function:
char string[strlen(conn->uri->path)];
Here the path part of the given URI is used without checking it for
NULL, this can cause a segfault as strlen expects a string != NULL.
Beside that uuid_db and connection_data leak in case of an error.
In this line
conn->uri->path = string;
the original path of the URI leaks. The patch adds a VIR_FREE call
before setting the new path.
The attached patch is compile-tested but I don't have a Power
Hypervisor installation at hand to test it for real.
Matthias
Signed-off-by: Chris Lalancette <clalance@redhat.com>
Fix up a small memory leak pointed out by DanB; I was forgetting
to release memory allocated to driver->saveImageFormat.
Also add the "save_image_format" and "security" entries to
the augeas lens.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
https://bugzilla.redhat.com/517371
Matt Booth points out that if you use a non-existent bridge name when
start a guest you get a weird error message:
Failed to add tap interface 'vnet%d' to bridge 'virbr0'
and dev='vnet%d' appears in the dumpxml output.
Fix that by not including 'vnet%d' in the error message and freeing the
'vnet%d' string if adding the tap device to the bridge fails.
* src/qemu_conf.c, src/uml_conf.c: fix qemudNetworkIfaceConnect()
and umlConnectTapDevice() to not expose 'vnet%d' to the user
As we start/shutdown guests, or hotplug/hot-unplug devices, we can add
or delete devices as appropriate from a list of active devices.
Then, in pciReset(), we can use this to determine whether its safe to
reset a device as a side effect of resetting another device.
* src/qemu_conf.h: add activePciHostdevs to qemud_driver
* src/qemu_driver.c: maintain the activePciHostdevs list, and pass it
to pciResetDevice()
* src/pci.[ch]: pass the activeDevs list to pciResetDevice() and use
it to determine whether a Secondary Bus Reset is safe
The qemuPrepareHostDevices() and qemuDomainReAttachHostDevices()
functions are clutter with a bunch of calls to pciGetDevice() and
pciFreeDevice() obscuring the basic logic.
Add a pciDeviceList type and add a qemuGetPciHostDeviceList() function
to build a list from a domain definition. Use this in prepare/re-attach
fto simplify things and eliminate the multiple pciGetDevice calls.
This is especially useful because in the next patch we need to iterate
the hostdevs list a third time and we also need a list type for keeping
track of active devices.
* src/pci.[ch]: add pciDeviceList type and also a per-device 'managed'
property
* src/libvirt_private.syms: export the new functions
* src/qemu_driver.c: add qemuGetPciHostDeviceList() and re-write
qemuPrepareHostDevices() and qemuDomainReAttachHostDevices() to use it
Newer versions of QEMU accept 'pci_add auto', but older versions require
'pci_add pci_addr=auto'
* src/qemu_driver.c: use pci_addr= in qemudDomainAttachHostPciDevice()
for older versions of QEMU
When we hot-unplug a PCI host device from a guest, we should reset it.
Both managed and unmanaged devices should be reset, but only managed
devices should be re-attached.
* src/qemu_driver.c: reset devices in qemudDomainDetachHostPciDevice()
Right now we're only resetting managed devices before hotplug, but we
should reset them irrespective of whether they are managed.
* src/qemu_driver.c: reset all PCI hostdevs before hotplug
It turns out that the previous attempt at this doesn't work well
in the case of hotplug. We need qemuCheckPciHostDevice() to
disallow the reset affecting devices already attach to the guest,
but we still need to avoid double locking the virDomainObjPtr.
This is all getting messy, I've a better idea.
This reverts commit 6318808270 and
c106c8a18c.
* src/qemu_driver.c, src/pci.[ch], src/xen_unified.c,
src/libvirt_private.syms: revert a bunch of stuff.
The current code makes a poor effort at updating the device arrays after
hot-unplug. Fix that and combine the two code paths into one.
* src/qemu_driver.c: fix list updating in qemudDomainDetachNetDevice(),
qemudDomainDetachPciDiskDevice() and qemudDomainDetachHostPciDevice()
Maybe it's just me, but I try to select an item from the tree using
double-click and get annoyed when "+-" gets included in the selection.
* src/virsh.c: add a space between "+-" and the node device name
in 'virsh nodedev-list --tree'
Currently the reference counting for connections is busted. I
first noticed it while trying to use virConnectRef; it would
eventually cause a crash in the remote_internal driver, although
that was really just a victim. Really, we should only call the
close callbacks on the methods when the references drop to 0. To
accomplish this, move all of the close callbacks into
virUnrefConnect (since there are lots of internal users of that
function), and arrange for virConnectClose to call that.
V2: Make sure to drop the connection lock before we call the close
callbacks, otherwise we could deadlock the daemon
V3: Fix up a crash when we got an error from one of the drivers
Signed-off-by: Chris Lalancette <clalance@redhat.com>
* src/domain_conf.c: Make virDomainObjListFree a no-op if list
is NULL
* src/domain_event.c: make virDomainEventCallbackListFree a no-op
if event list is NULL
* src/lxc_driver.c: Log a message if LXC driver does not startup
due to lacking kernel support
Implement a compressed save image format for qemu. While ideally
we would have the choice between compressed/non-compressed
available to the libvirt API, unfortunately there is no "flags"
parameter to the virDomainSave() API. Therefore, implement this
as a qemu.conf option. gzip, bzip2, and lzma are implemented, and
it should be very easy to implement additional compression
methods.
One open question is if/how we should detect the compression
binaries. One way to do it is to do compile-time setting of the
paths (via configure.in), but that doesn't seem like a great thing
to do. My preferred solution is not to detect at all;
when we go to run the commands that need them, if they
aren't available, or aren't available in one of the standard paths,
then we'll fail. That's also the solution implemented in this patch.
In the future, we'll have a more robust (managed) save/restore API,
at which time we can expose this functionality properly in the API.
V2: get rid of redundant dd command and just use >> to append data.
V3: Add back the missing pieces for the enum and bumping the save version.
V4: Make the compressed field in the save_header an int.
Implement LZMA compression.
Signed-off-by: Chris Lalancette <clalance@redhat.com>
If a PCI device reset causes other devices to be reset, allow it so long
as those other devices are note assigned to another active domain.
Note, we need to take the driver lock qemudNodeDeviceReset() because the
check function will iterate over the domain list.
* src/qemu_conf.c: add qemuCheckPciHostDevice() to iterate over active
domains checking whether the affected device is assigned
* src/pci.[ch]: add pciDeviceEquals() helper
When using a Secondary Bus Reset, all devices on the bus are reset.
Extend the pciResetDevice() API so that a 'check' callback can be
supplied which will verify that it is safe to reset the other devices
on the bus.
The virDomainObjPtr parameter is needed so that when the check function
iterates over the domain list, it can avoid double locking.
* src/pci.[ch]: add a 'check' callback to pciResetDevice(), re-work
pciIterDevices() to pass the check function to the iter functions,
use the check function in the bus iterator, return the first unsafe
device from pciBusCheckOtherDevices() and include its details in
the bus reset error message.
* src/qemu_driver.c, src/xen_uninified.c: just pass NULL as the
check function for now
Currently, if we are unable to reset a PCI device we return a fairly
generic 'No PCI reset capability available' error message.
Fix that by returning an error from the individual reset messages and
using that error to construct the higher level error mesage.
* src/pci.c: set errors in pciTryPowerManagementReset() and
pciTrySecondaryBusReset() on failure; use those error messages
in pciResetDevice(), or explain that no reset support is available
When the guest shuts down, we should attempt to restore all PCI host
devices to a sane state.
In the case of managed hostdevs, we should reset and re-attach the
devices. In the case of unmanaged hostdevs, we should just reset them.
Note, KVM will already reset assigned devices when the guest shuts
down using whatever means it can, so we are only doing it to cover the
cases the kernel can't handle.
* src/qemu_driver.c: add qemuDomainReAttachHostDevices() and call
it from qemudShutdownVMDaemon()
It turns out that a PCI Power Management reset only affects individual
functions, and not the whole device.
The PCI Power Management spec talks about resetting the 'device' rather
than the 'function', but Intel's Dexuan Cui informs me that it is
actually a per-function reset.
Also, Yu Zhao has added pci_pm_reset() to the kernel, and it doesn't
reject multi-function devices, so it must be true! :-)
(A side issue is that we could defer the PM reset to the kernel if we
could detect that the kernel has PM reset support, but barring version
number checks we don't have a way to detect that support)
* src/pci.c: remove the pciDeviceContainsOtherFunctions() check from
pciTryPowerManagementReset() and prefer PM reset over bus reset
where both are available
Cc: Cui, Dexuan <dexuan.cui@intel.com>
Cc: Yu Zhao <yu.zhao@intel.com>
PCI device assignment is only supported in KVM's fork of qemu, so we
should really detect its availability and give a nice error if its
not supported.
* src/qemu_conf.[ch]: introduce QEMUD_CMD_FLAG_PCIDEVICE indicating
that the -pcidevice command line option is available
* tests/*: update the tests
Attaching a host PCI device to a qemu guest is done with a
straightforward 'pci_add auto host host=XX:XX.X' command.
Like with NIC and disk hotplug, we need to retain the guest PCI address
assigned by qemu so that we can use it for hot-unplug.
Identifying a device for detach is done using the host PCI address.
Managed mode is handled by detaching/resetting the device before
attaching it to the guest and re-attaching it after detaching it from
the guest.
* src/qemu_driver.c: add qemudDomainAttachHostPciDevice() and
qemudDomainDetachHostPciDevice()
* src/domain_conf.h: add somewhere to store the guest PCI address
* src/domain_conf.c: handle formatting and parsing the guest PCI
address
