January 2025 - Devel - libvirt List Archives

[libvirt PATCH] meson: error out when ch driver is requested but unsupported
by Ján Tomko 20 Jan '25

20 Jan '25

Signed-off-by: Ján Tomko <jtomko(a)redhat.com> --- meson.build | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) diff --git a/meson.build b/meson.build index 16ec3ba7e2..326d9dd0f9 100644 --- a/meson.build +++ b/meson.build @@ -1582,8 +1582,12 @@ elif get_option('driver_lxc').enabled() error('linux and remote_driver are required for LXC') endif -if not get_option('driver_ch').disabled() and host_machine.system() == 'linux' and (host_machine.cpu_family() == 'x86_64' or host_machine.cpu_family() == 'aarch64') - use_ch = true +if not get_option('driver_ch').disabled() + if host_machine.system() == 'linux' and (host_machine.cpu_family() == 'x86_64' or host_machine.cpu_family() == 'aarch64') + use_ch = true + elif get_option('driver_ch').enabled() + error('linux on x86_64 or aarch64 is required to build Cloud-Hypervisor driver') + endif if not conf.has('WITH_LIBVIRTD') use_ch = false -- 2.47.0

2 1

[PATCH] spec: Only build ch driver on x86_64 and aarch64
by Cole Robinson 20 Jan '25

20 Jan '25

This matches the cpu_family() check in `meson.build` Signed-off-by: Cole Robinson <crobinso(a)redhat.com> --- note: something probably needs to be fixed on meson.build side too, since an explicit `driver_ch=enabled` doesn't generate a build error. libvirt.spec.in | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/libvirt.spec.in b/libvirt.spec.in index 2bc8dfd323..5a1755d74f 100644 --- a/libvirt.spec.in +++ b/libvirt.spec.in @@ -29,6 +29,7 @@ %define arches_zfs %{arches_x86} %{power64} %{arm} %define arches_numactl %{arches_x86} %{power64} aarch64 s390x %define arches_numad %{arches_x86} %{power64} aarch64 +%define arches_ch x86_64 aarch64 # The hypervisor drivers that run in libvirtd %define with_qemu 0%{!?_without_qemu:1} @@ -124,6 +125,9 @@ %ifnarch %{arches_ceph} %define with_storage_rbd 0 %endif +%ifnarch %{arches_ch} + %define with_ch 0 +%endif # RHEL doesn't ship many hypervisor drivers %if 0%{?rhel} -- 2.48.1

2 1

[PATCH 0/4] Implement virsh hypervisor-cpu-models
by Boris Fiuczynski 17 Jan '25

17 Jan '25

Allows for the query of hypervisor-known CPU models via the simple command: virsh hypervisor-cpu-models. For the QEMU driver, the models are queried via the capabilities file. Each model is printed to the terminal on its own line similar to the cpu-models command, and there is no order to the listing. There is a related libvirt-python merge request at https://gitlab.com/libvirt/libvirt-python/-/merge_requests/159 David Judkovics (4): libvirt: Introduce virConnectGetHypervisorCPUModelNames public API remote: Implement virConnectGetHypervisorCPUModelNames qemu_driver: Implement qemuConnectGetHypervisorCPUModelNames virsh: Introduce new hypervisor-cpu-models command docs/manpages/virsh.rst | 20 ++++++++ include/libvirt/libvirt-host.h | 7 +++ src/driver-hypervisor.h | 10 ++++ src/libvirt-host.c | 65 +++++++++++++++++++++++++ src/libvirt_public.syms | 5 ++ src/qemu/qemu_driver.c | 59 +++++++++++++++++++++++ src/remote/remote_daemon_dispatch.c | 62 ++++++++++++++++++++++++ src/remote/remote_driver.c | 52 ++++++++++++++++++++ src/remote/remote_protocol.x | 26 +++++++++- src/remote_protocol-structs | 16 +++++++ tools/virsh-host.c | 74 +++++++++++++++++++++++++++++ 11 files changed, 395 insertions(+), 1 deletion(-) -- 2.47.0

5 10

[PATCH 0/3] Remove deprecated machines pc-i440fx-2.4 up to pc-i440fx-2.12
by Thomas Huth 17 Jan '25

17 Jan '25

While our new auto-disablement of old machine types will only kick in with the next (v10.1) release, the pc-i440fx-2.* machine types have been explicitly marked as deprecated via our old deprecation policy mechanism before (two releases ago), so it should be fine to remove them now already. Note that we can not do much additional clean ups on top yet since the corresponding q35 machines (which share the same compatibility knobs) are still around and only will be removed for the 10.1 release instead. So the bigger clean-up can only be done for 10.1, but removing the i440fx machine types now will still have at least a small benefit of accelerating our CI a little bit (since we don't have to run tests for these old machine types anymore). Thomas Huth (3): tests/qtest/test-x86-cpuid-compat: Remove tests related to pc-i440fx-2.3 hw/i386/pc_piix: Remove pc-i440fx-2.4 up to pc-i440fx-2.12 tests/qtest/test-x86-cpuid-compat: Replaced the removed pc-i440fx-2.* machines docs/about/deprecated.rst | 7 --- docs/about/removed-features.rst | 4 +- docs/interop/firmware.json | 2 +- hw/i386/pc_piix.c | 95 ----------------------------- tests/qtest/test-x86-cpuid-compat.c | 52 ++++++---------- qemu-options.hx | 10 +-- 6 files changed, 25 insertions(+), 145 deletions(-) -- 2.47.1

5 11

[PATCH 0/1] Support cloning saved VMs
by Felipe Franciosi 17 Jan '25

17 Jan '25

Hello! I have a use case which I'm struggling to support with libvirt: saving a VM to a file, cloning it (which renames the VM), and restoring it. My search revealed a number of tutorials for using virt-clone [1], but that doesn't seem to cover VMs which are _saved_ (only running or paused). [1] https://github.com/virt-manager/virt-manager/blob/main/virtinst/virtclone.py In a nutshell, I want to power on a VM and do some setup, then save its full state to disk (e.g., with virsh save). Finally I want to modify the XML to: - rename the VM - change which bridge its NICs are on (while maintaining mac addresses) - change the disk image to a copy (done while the VM is saved) But the restore operation fails because of a target domain name check implemented in virDomainDefCheckABIStabilityFlags(). I've debated how to best address this and I'm looking for your views. I suspect this call is there for a reason (which may still be relevant), although the name is clearly not part of the ABI; it's the host identifier for that domain and not guest-visible. My first stab at this is therefore just to drop this check (patch attached). I'm open to suggestions, for example by plumbing through a flag which makes the check optional. Please let me know how you prefer that I take this forward. Thanks, F. Felipe Franciosi (1): conf: remove domain name change check on CheckABI src/conf/domain_conf.c | 11 ----------- 1 file changed, 11 deletions(-) -- 2.39.5

2 15

[PATCH] cpu_x86: Probe host CPU for all MSR features
by Jiri Denemark 17 Jan '25

17 Jan '25

The list of CPU features we probe from various MSR grew significantly over time and the CPU map currently mentions 11 distinct MSR indexes. But the code for directly probing host CPU features was still reading only the original 0x10a index. Thus the CPU model in host capabilities was missing a lot of features. Instead of specifying a static list of indexes to read (which we would forget to update in the future), let's just read all indexes found in the CPU map. Signed-off-by: Jiri Denemark <jdenemar(a)redhat.com> --- src/cpu/cpu_x86.c | 62 ++++++++++++++++++++++++++++++++++++++++------- 1 file changed, 53 insertions(+), 9 deletions(-) diff --git a/src/cpu/cpu_x86.c b/src/cpu/cpu_x86.c index 6d72d446c9..1ce97d9dcf 100644 --- a/src/cpu/cpu_x86.c +++ b/src/cpu/cpu_x86.c @@ -2836,15 +2836,61 @@ cpuidSet(uint32_t base, virCPUData *data) } +static size_t +virCPUx86ListMSRs(virCPUx86Map *map, + unsigned long **msrs) +{ + size_t n = 0; + size_t i, j, k; + + /* These three nested loops look horrible, but data->len is always 1 here + * and thanks to grouping features in the CPU map by their MSR index the + * third loop will run more than one iteration only once for each distinct + * index. + */ + for (i = 0; i < map->nfeatures; i++) { + virCPUx86Data *data = &map->features[i]->data; + + for (j = 0; j < data->len; j++) { + virCPUx86DataItem *item = &data->items[j]; + bool found = false; + unsigned long msr; + + if (item->type != VIR_CPU_X86_DATA_MSR) + continue; + + msr = item->data.msr.index; + + for (k = n; k > 0; k--) { + if ((*msrs)[k - 1] == msr) { + found = true; + break; + } + } + + if (!found) + VIR_APPEND_ELEMENT(*msrs, n, msr); + } + } + + return n; +} + + static int virCPUx86GetHost(virCPUDef *cpu, virDomainCapsCPUModels *models) { g_autoptr(virCPUData) cpuData = NULL; unsigned int addrsz; + virCPUx86Map *map; + g_autofree unsigned long *msrs = NULL; + size_t nmsrs; + uint64_t msr; + size_t i; int ret; - if (virCPUx86DriverInitialize() < 0) + if (!(map = virCPUx86GetMap())) return -1; if (!(cpuData = virCPUDataNew(archs[0]))) @@ -2854,18 +2900,16 @@ virCPUx86GetHost(virCPUDef *cpu, cpuidSet(CPUX86_EXTENDED, cpuData) < 0) return -1; - /* Read the IA32_ARCH_CAPABILITIES MSR (0x10a) if supported. - * This is best effort since there might be no way to read the MSR - * when we are not running as root. */ - if (virCPUx86DataCheckFeature(cpuData, "arch-capabilities") == 1) { - uint64_t msr; - unsigned long index = 0x10a; + nmsrs = virCPUx86ListMSRs(map, &msrs); - if (virHostCPUGetMSR(index, &msr) == 0) { + /* This is best effort since there might be no way to read the MSR + * when we are not running as root. */ + for (i = 0; i < nmsrs; i++) { + if (virHostCPUGetMSR(msrs[i], &msr) == 0) { virCPUx86DataItem item = { .type = VIR_CPU_X86_DATA_MSR, .data.msr = { - .index = index, + .index = msrs[i], .eax = msr & 0xffffffff, .edx = msr >> 32, }, -- 2.47.1

2 2

[PATCH v3 0/4] fix AppArmor policy restore for runtime rules
by Georgia Garcia 16 Jan '25

16 Jan '25

Some rules are generated dynamically during boot and added to the AppArmor policy. An example of that is macvtap devices that call the AppArmorSetFDLabel hook to add a rule for the tap device path. Since this information is dynamic, it is not available in the xml config, therefore whenever a "Restore" hook is called, the entire profile is regenerated by virt-aa-helper based only the information from the VM definition, so the dynamic/runtime information is lost. This patchset fixes that by storing these rules in a different file called libvirt-uuid.runtime_files, which is included by libvirt-uuid.files that already exists. It also includes other fixes like memory leaks, adoption of the GLib API in the apparmor files and a fix on the AppArmor policy that incorrectly applies apparmor policy syntax. Georgia Garcia (4): security_apparmor: fix memleaks in AppArmorSetFDLabel security: replace uses of label and VIR_FREE by g_autofree apparmor: fix UUID specification virt-aa-helper: store dynamically generated rules .../usr.lib.libvirt.virt-aa-helper.in | 5 +- src/security/apparmor/usr.sbin.libvirtd.in | 7 +- src/security/apparmor/usr.sbin.virtqemud.in | 6 +- src/security/security_apparmor.c | 84 +++++----- src/security/virt-aa-helper.c | 145 +++++++++--------- 5 files changed, 125 insertions(+), 122 deletions(-) -- 2.43.0

4 15

[PATCH] qemu: Report error from both sides of migration
by Jiri Denemark 16 Jan '25

16 Jan '25

When migration fails in Perform phase, we call Finish on the destination host with cancelled=1 and get the error from there and report it to the user. This works well if the error on the destination caused the migration to fail. But in other cases the main error may reported by the source and the destination would just be complaining about broken migration stream. In other words, we don't really know which error caused the migration to fail and we have no way of detecting that. So instead of choosing one error, this patch will combine the error messages from both sides of migration into a single message and report it to the user. The result would be, for example: operation failed: migration failed. Message from the source host: operation failed: job 'migration out' failed: Certificate does not match the hostname ble.bla. Message from the destination host: operation failed: job 'migration in' failed: load of migration failed: Invalid argument And yes, this is ugly, but I wasn't able to come up with a better way of fixing this issue. https://issues.redhat.com/browse/RHEL-58933 Signed-off-by: Jiri Denemark <jdenemar(a)redhat.com> --- src/libvirt-domain.c | 26 +++++++++++++------------- src/qemu/qemu_migration.c | 26 +++++++++++++------------- 2 files changed, 26 insertions(+), 26 deletions(-) diff --git a/src/libvirt-domain.c b/src/libvirt-domain.c index e8e5379672..efccafc4d2 100644 --- a/src/libvirt-domain.c +++ b/src/libvirt-domain.c @@ -3430,26 +3430,26 @@ virDomainMigrateVersion3Full(virDomainPtr domain, if (ddomain) { VIR_ERROR(_("finish step ignored that migration was cancelled")); } else { - /* If Finish reported a useful error, use it instead of the - * original "migration unexpectedly failed" error. + virErrorPtr err = virGetLastError(); + /* When both Confirm and Finish reported an error in QEMU driver, + * we don't really know which error is the root cause. Let's report + * both errors to the user. * * This is ugly but we can't do better with the APIs we have. We * only replace the error if Finish was called with cancelled == 1 * and reported a real error (old libvirt would report an error - * from RPC instead of MIGRATE_FINISH_OK), which only happens when - * the domain died on destination. To further reduce a possibility - * of false positives we also check that Perform returned - * VIR_ERR_OPERATION_FAILED. + * from RPC instead of MIGRATE_FINISH_OK). */ if (orig_err && orig_err->domain == VIR_FROM_QEMU && - orig_err->code == VIR_ERR_OPERATION_FAILED) { - virErrorPtr err = virGetLastError(); - if (err && - err->domain == VIR_FROM_QEMU && - err->code != VIR_ERR_MIGRATE_FINISH_OK) { - g_clear_pointer(&orig_err, virFreeError); - } + orig_err->code == VIR_ERR_OPERATION_FAILED && + err && + err->domain == VIR_FROM_QEMU && + err->code != VIR_ERR_MIGRATE_FINISH_OK) { + virReportError(VIR_ERR_OPERATION_FAILED, + _("migration failed. Message from the source host: %1$s. Message from the destination host: %2$s"), + orig_err->message, err->message); + g_clear_pointer(&orig_err, virFreeError); } } } diff --git a/src/qemu/qemu_migration.c b/src/qemu/qemu_migration.c index 1582a738a3..8c0e522828 100644 --- a/src/qemu/qemu_migration.c +++ b/src/qemu/qemu_migration.c @@ -5904,26 +5904,26 @@ qemuMigrationSrcPerformPeer2Peer3(virQEMUDriver *driver, if (ddomain) { VIR_ERROR(_("finish step ignored that migration was cancelled")); } else { - /* If Finish reported a useful error, use it instead of the - * original "migration unexpectedly failed" error. + virErrorPtr err = virGetLastError(); + /* When both Confirm and Finish reported an error in QEMU driver, + * we don't really know which error is the root cause. Let's report + * both errors to the user. * * This is ugly but we can't do better with the APIs we have. We * only replace the error if Finish was called with cancelled == 1 * and reported a real error (old libvirt would report an error - * from RPC instead of MIGRATE_FINISH_OK), which only happens when - * the domain died on destination. To further reduce a possibility - * of false positives we also check that Perform returned - * VIR_ERR_OPERATION_FAILED. + * from RPC instead of MIGRATE_FINISH_OK). */ if (orig_err && orig_err->domain == VIR_FROM_QEMU && - orig_err->code == VIR_ERR_OPERATION_FAILED) { - virErrorPtr err = virGetLastError(); - if (err && - err->domain == VIR_FROM_QEMU && - err->code != VIR_ERR_MIGRATE_FINISH_OK) { - g_clear_pointer(&orig_err, virFreeError); - } + orig_err->code == VIR_ERR_OPERATION_FAILED && + err && + err->domain == VIR_FROM_QEMU && + err->code != VIR_ERR_MIGRATE_FINISH_OK) { + virReportError(VIR_ERR_OPERATION_FAILED, + _("migration failed. Message from the source host: %1$s. Message from the destination host: %2$s"), + orig_err->message, err->message); + g_clear_pointer(&orig_err, virFreeError); } } } -- 2.47.1

2 1

[PATCH 0/2] qemu_migration: Do not automatically resume domain after I/O error
by Jiri Denemark 16 Jan '25

16 Jan '25

Jiri Denemark (2): qemu_migration: Refactor qemuMigrationSrcRestoreDomainState qemu_migration: Do not automatically resume domain after I/O error src/qemu/qemu_migration.c | 69 +++++++++++++++++++-------------------- 1 file changed, 33 insertions(+), 36 deletions(-) -- 2.47.1

2 3

[PATCH 0/2] tests: Add qemu-9.2 dev cycle caps on s390x
by Michal Privoznik 16 Jan '25

16 Jan '25

NB, the capabilities were captured in a VM running on an actual s390x machine so there might be some CPU features (incorrectly) missing. But it's the best I can do. Michal Prívozník (2): qemuxmlconftest: Switch s390-default-cpu-...-virtio-2.7 to virtio-2.9 tests: qemucapabilities: Add test data for the qemu-9.2 dev cycle on s390x tests/domaincapsdata/qemu_9.3.0.s390x.xml | 436 + .../caps_9.3.0_s390x.replies | 38101 ++++++++++++++++ .../qemucapabilitiesdata/caps_9.3.0_s390x.xml | 4268 ++ ...-deprecated-features-off.s390x-latest.args | 2 +- ...default-video-type-s390x.s390x-latest.args | 2 +- ...vfio-zpci-ccw-memballoon.s390x-latest.args | 2 +- .../launch-security-s390-pv.s390x-latest.args | 2 +- ...-cpu-kvm-ccw-virtio-2.9.s390x-latest.args} | 4 +- ...t-cpu-kvm-ccw-virtio-2.9.s390x-latest.xml} | 4 +- ...> s390-default-cpu-kvm-ccw-virtio-2.9.xml} | 2 +- ...t-cpu-kvm-ccw-virtio-4.2.s390x-latest.args | 2 +- ...-cpu-tcg-ccw-virtio-2.9.s390x-latest.args} | 2 +- ...t-cpu-tcg-ccw-virtio-2.9.s390x-latest.xml} | 2 +- ...> s390-default-cpu-tcg-ccw-virtio-2.9.xml} | 2 +- .../s390-defaultconsole.s390x-latest.args | 2 +- .../s390-panic.s390x-latest.args | 2 +- tests/qemuxmlconftest.c | 4 +- 17 files changed, 42822 insertions(+), 17 deletions(-) create mode 100644 tests/domaincapsdata/qemu_9.3.0.s390x.xml create mode 100644 tests/qemucapabilitiesdata/caps_9.3.0_s390x.replies create mode 100644 tests/qemucapabilitiesdata/caps_9.3.0_s390x.xml rename tests/qemuxmlconfdata/{s390-default-cpu-kvm-ccw-virtio-2.7.s390x-latest.args => s390-default-cpu-kvm-ccw-virtio-2.9.s390x-latest.args} (70%) rename tests/qemuxmlconfdata/{s390-default-cpu-kvm-ccw-virtio-2.7.s390x-latest.xml => s390-default-cpu-kvm-ccw-virtio-2.9.s390x-latest.xml} (86%) rename tests/qemuxmlconfdata/{s390-default-cpu-kvm-ccw-virtio-2.7.xml => s390-default-cpu-kvm-ccw-virtio-2.9.xml} (85%) rename tests/qemuxmlconfdata/{s390-default-cpu-tcg-ccw-virtio-2.7.s390x-latest.args => s390-default-cpu-tcg-ccw-virtio-2.9.s390x-latest.args} (94%) rename tests/qemuxmlconfdata/{s390-default-cpu-tcg-ccw-virtio-2.7.s390x-latest.xml => s390-default-cpu-tcg-ccw-virtio-2.9.s390x-latest.xml} (92%) rename tests/qemuxmlconfdata/{s390-default-cpu-tcg-ccw-virtio-2.7.xml => s390-default-cpu-tcg-ccw-virtio-2.9.xml} (85%) -- 2.45.2

3 4