January 2022 - Devel - libvirt List Archives

[PATCH] rpc: Emit a warning when applying RPC throttling
by Ivan Teterevkov 13 Jan '22

13 Jan '22

Whenever the number of outstanding requests being served reaches max_client_requests, the RPC server applies RX throttling and temporarily stops receiving incoming data. Consequently, it stops receiving the keep-alive responses from the client and eventually may cause a connection timeout on the server-side. This patch emits a warning message in such cases so that the users may root-cause the connection timeout and reconfigure max_client_requests depending on their application. Signed-off-by: Ivan Teterevkov <ivan.teterevkov(a)nutanix.com> --- src/rpc/virnetserverclient.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/src/rpc/virnetserverclient.c b/src/rpc/virnetserverclient.c index 7050430e17..9d86e1ba3e 100644 --- a/src/rpc/virnetserverclient.c +++ b/src/rpc/virnetserverclient.c @@ -1293,6 +1293,10 @@ static virNetMessage *virNetServerClientDispatchRead(virNetServerClient *client) client->rx->buffer = g_new0(char, client->rx->bufferLength); client->nrequests++; } + } else { + VIR_WARN("Applying RPC throttling, consider increasing " + "max_client_requests (client->nrequests_max=%zu)", + client->nrequests_max); } virNetServerClientUpdateEvent(client); -- 2.35.0-rc0

1 0

[PATCH] Fix missing-field-initializers error
by huangy81＠chinatelecom.cn 12 Jan '22

12 Jan '22

From: Hyman Huang(黄勇) <huangy81(a)chinatelecom.cn> When compile libvirt via ninja tool with default configuration, it report the error message as the following: "missing initializer for field 'path' of 'virTPMBinaryInfo'". So initialize the 'path' field in 'virTPMBinaryInfo' with 'NULL'. Signed-off-by: Hyman Huang(黄勇) <huangy81(a)chinatelecom.cn> --- src/util/virtpm.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/src/util/virtpm.c b/src/util/virtpm.c index cf4c7c4..a02d933 100644 --- a/src/util/virtpm.c +++ b/src/util/virtpm.c @@ -122,13 +122,16 @@ typedef struct _virTPMBinaryInfo { static virTPMBinaryInfo swtpmBinaries[VIR_TPM_BINARY_LAST] = { [VIR_TPM_BINARY_SWTPM] = { + .path = NULL, .parm = "socket", .capsParse = virTPMSwtpmFeatureTypeFromString, }, [VIR_TPM_BINARY_SWTPM_SETUP] = { + .path = NULL, .capsParse = virTPMSwtpmSetupFeatureTypeFromString, }, [VIR_TPM_BINARY_SWTPM_IOCTL] = { + .path = NULL, }, }; -- 1.8.3.1

3 4

[PATCH for 8.0 0/3] Fix creation of snapshots without libvirt metadata
by Peter Krempa 12 Jan '22

12 Jan '22

After a recent unreleased refactor we don't remove the temporary snapshot object from the list when finishing a snapshot without metadata thus we should fix it before releasing a buggy release. This patchset fixes it by not adding it into the list in the first place. Peter Krempa (3): virdomainmomentobjlist.h: Convert to modern header style conf: moment: Export helpers to create the virDomainMoment wrapper qemuSnapshotCreate: Don't insert snapshot into list with VIR_DOMAIN_SNAPSHOT_CREATE_NO_METADATA src/conf/virdomainmomentobjlist.c | 4 +- src/conf/virdomainmomentobjlist.h | 121 +++++++++++++++++++----------- src/libvirt_private.syms | 2 + src/qemu/qemu_snapshot.c | 25 +++--- 4 files changed, 97 insertions(+), 55 deletions(-) -- 2.31.1

2 6

[PATCH v4] report error when virProcessGetStatInfo() is unable to parse data
by Ani Sinha 12 Jan '22

12 Jan '22

Currently virProcessGetStatInfo() always returns success and only logs error when it is unable to parse the data. Make this function actually report the error and return a negative value in this error scenario. Fix the callers so that they do not override the error generated. Also fix non-linux implementation of this function so as to report error. Signed-off-by: Ani Sinha <ani(a)anisinha.ca> --- src/ch/ch_driver.c | 2 -- src/qemu/qemu_driver.c | 7 +------ src/util/virprocess.c | 17 +++++++++++++---- 3 files changed, 14 insertions(+), 12 deletions(-) changelog: v4: on freebsd error on the logs is a problem appaently. try to fix that. v3: fix non linux implementation v2: fix callers diff --git a/src/ch/ch_driver.c b/src/ch/ch_driver.c index 53e0872207..3cbc668489 100644 --- a/src/ch/ch_driver.c +++ b/src/ch/ch_driver.c @@ -1073,8 +1073,6 @@ chDomainHelperGetVcpus(virDomainObj *vm, if (virProcessGetStatInfo(&vcpuinfo->cpuTime, &vcpuinfo->cpu, NULL, vm->pid, vcpupid) < 0) { - virReportSystemError(errno, "%s", - _("cannot get vCPU placement & pCPU time")); return -1; } } diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index d3d76c003f..65ac5ef367 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -1359,8 +1359,6 @@ qemuDomainHelperGetVcpus(virDomainObj *vm, if (virProcessGetStatInfo(&vcpuinfo->cpuTime, &vcpuinfo->cpu, NULL, vm->pid, vcpupid) < 0) { - virReportSystemError(errno, "%s", - _("cannot get vCPU placement & pCPU time")); return -1; } } @@ -2521,8 +2519,6 @@ qemuDomainGetInfo(virDomainPtr dom, if (virDomainObjIsActive(vm)) { if (virProcessGetStatInfo(&(info->cpuTime), NULL, NULL, vm->pid, 0) < 0) { - virReportError(VIR_ERR_OPERATION_FAILED, "%s", - _("cannot read cputime for domain")); goto cleanup; } } @@ -10530,8 +10526,7 @@ qemuDomainMemoryStatsInternal(virQEMUDriver *driver, } if (virProcessGetStatInfo(NULL, NULL, &rss, vm->pid, 0) < 0) { - virReportError(VIR_ERR_OPERATION_FAILED, "%s", - _("cannot get RSS for domain")); + virResetLastError(); } else { stats[ret].tag = VIR_DOMAIN_MEMORY_STAT_RSS; stats[ret].val = rss; diff --git a/src/util/virprocess.c b/src/util/virprocess.c index b559a4257e..cbb31441cc 100644 --- a/src/util/virprocess.c +++ b/src/util/virprocess.c @@ -1779,12 +1779,20 @@ virProcessGetStatInfo(unsigned long long *cpuTime, long rss = 0; int cpu = 0; - if (!proc_stat || - virStrToLong_ullp(proc_stat[VIR_PROCESS_STAT_UTIME], NULL, 10, &usertime) < 0 || + if (!proc_stat) { + VIR_WARN("Unable to get proc stats from the kernel"); + return 0; + } + + if (virStrToLong_ullp(proc_stat[VIR_PROCESS_STAT_UTIME], NULL, 10, &usertime) < 0 || virStrToLong_ullp(proc_stat[VIR_PROCESS_STAT_STIME], NULL, 10, &systime) < 0 || virStrToLong_l(proc_stat[VIR_PROCESS_STAT_RSS], NULL, 10, &rss) < 0 || virStrToLong_i(proc_stat[VIR_PROCESS_STAT_PROCESSOR], NULL, 10, &cpu) < 0) { - VIR_WARN("cannot parse process status data"); + virReportError(VIR_ERR_INTERNAL_ERROR, + _("cannot parse process status data for pid '%d/%d'"), + (int) pid, (int) tid); + + return -1; } /* We got jiffies @@ -1881,7 +1889,8 @@ virProcessGetStatInfo(unsigned long long *cpuTime G_GNUC_UNUSED, pid_t pid G_GNUC_UNUSED, pid_t tid G_GNUC_UNUSED) { - errno = ENOSYS; + virReportSystemError(ENOSYS, "%s", + _("Process statistics data is not supported on this platform")); return -1; } -- 2.25.1

2 4

[PATCH v3] report error when virProcessGetStatInfo() is unable to parse data
by Ani Sinha 12 Jan '22

12 Jan '22

Currently virProcessGetStatInfo() always returns success and only logs error when it is unable to parse the data. Make this function actually report the error and return a negative value in this error scenario. Fix the callers so that they do not override the error generated. Also fix non-linux implementation of this function so as to report error. Signed-off-by: Ani Sinha <ani(a)anisinha.ca> --- src/ch/ch_driver.c | 2 -- src/qemu/qemu_driver.c | 7 +------ src/util/virprocess.c | 9 +++++++-- 3 files changed, 8 insertions(+), 10 deletions(-) changelog: v3: fix non-linux implementation v2: fix callers diff --git a/src/ch/ch_driver.c b/src/ch/ch_driver.c index 53e0872207..3cbc668489 100644 --- a/src/ch/ch_driver.c +++ b/src/ch/ch_driver.c @@ -1073,8 +1073,6 @@ chDomainHelperGetVcpus(virDomainObj *vm, if (virProcessGetStatInfo(&vcpuinfo->cpuTime, &vcpuinfo->cpu, NULL, vm->pid, vcpupid) < 0) { - virReportSystemError(errno, "%s", - _("cannot get vCPU placement & pCPU time")); return -1; } } diff --git a/src/qemu/qemu_driver.c b/src/qemu/qemu_driver.c index d3d76c003f..9a17c93b08 100644 --- a/src/qemu/qemu_driver.c +++ b/src/qemu/qemu_driver.c @@ -1359,8 +1359,6 @@ qemuDomainHelperGetVcpus(virDomainObj *vm, if (virProcessGetStatInfo(&vcpuinfo->cpuTime, &vcpuinfo->cpu, NULL, vm->pid, vcpupid) < 0) { - virReportSystemError(errno, "%s", - _("cannot get vCPU placement & pCPU time")); return -1; } } @@ -2521,8 +2519,6 @@ qemuDomainGetInfo(virDomainPtr dom, if (virDomainObjIsActive(vm)) { if (virProcessGetStatInfo(&(info->cpuTime), NULL, NULL, vm->pid, 0) < 0) { - virReportError(VIR_ERR_OPERATION_FAILED, "%s", - _("cannot read cputime for domain")); goto cleanup; } } @@ -10530,8 +10526,7 @@ qemuDomainMemoryStatsInternal(virQEMUDriver *driver, } if (virProcessGetStatInfo(NULL, NULL, &rss, vm->pid, 0) < 0) { - virReportError(VIR_ERR_OPERATION_FAILED, "%s", - _("cannot get RSS for domain")); + return -1; } else { stats[ret].tag = VIR_DOMAIN_MEMORY_STAT_RSS; stats[ret].val = rss; diff --git a/src/util/virprocess.c b/src/util/virprocess.c index b559a4257e..709ec616de 100644 --- a/src/util/virprocess.c +++ b/src/util/virprocess.c @@ -1784,7 +1784,11 @@ virProcessGetStatInfo(unsigned long long *cpuTime, virStrToLong_ullp(proc_stat[VIR_PROCESS_STAT_STIME], NULL, 10, &systime) < 0 || virStrToLong_l(proc_stat[VIR_PROCESS_STAT_RSS], NULL, 10, &rss) < 0 || virStrToLong_i(proc_stat[VIR_PROCESS_STAT_PROCESSOR], NULL, 10, &cpu) < 0) { - VIR_WARN("cannot parse process status data"); + virReportError(VIR_ERR_INTERNAL_ERROR, + _("cannot parse process status data for pid '%d/%d'"), + (int) pid, (int) tid); + + return -1; } /* We got jiffies @@ -1881,7 +1885,8 @@ virProcessGetStatInfo(unsigned long long *cpuTime G_GNUC_UNUSED, pid_t pid G_GNUC_UNUSED, pid_t tid G_GNUC_UNUSED) { - errno = ENOSYS; + virReportSystemError(ENOSYS, "%s", + _("Process statistics data is not supported on this platform")); return -1; } -- 2.25.1

2 3

[libvirt PATCH 0/3] Misc fixes / improvements
by Daniel P. Berrangé 12 Jan '22

12 Jan '22

Just a few things I dealt with as side effect of other work I'm doing wrt AMD SEV. Daniel P. Berrangé (3): docs: split example for <os> schema docs: use virYesNo definition in more schemas qemu: split handling of distinct firmware enum conversions docs/formatdomain.rst | 32 +++++++++++++++++++++++++++++--- docs/schemas/capability.rng | 5 +---- docs/schemas/domainbackup.rng | 5 +---- docs/schemas/domaincaps.rng | 5 +---- docs/schemas/domaincommon.rng | 17 ++++------------- src/conf/domain_conf.h | 4 ++-- src/qemu/qemu_firmware.c | 21 +++++++++++++++++++-- 7 files changed, 57 insertions(+), 32 deletions(-) -- 2.33.1

2 4

[PATCH 0/1] qemu_tpm: Get swtpm pid without binary validation
by Vasiliy Ulyanov 12 Jan '22

12 Jan '22

Hi all and Happy New Year! My name is Vasiliy, I am an engineer at SUSE. I was playing around with TPM in libvirt and trying to enable it in KubeVirt. With the emulator I was always getting "swtpm failed to start" internal error. After debugging the issue I found that the problem was not actually with starting the emulator but rather with retrieving the PID. The code in libvirt currently verifies that /proc/[pid]/exe points to the correct swtpm binary. In my case an attempt to dereference the symlink from procfs always resulted in EACCES. Eventually I found this issue [1]. It appears that libvirt needs CAP_SYS_PTRACE otherwise it will not be able to access the exe link (even if run as root). This can also be observed with the following reproducer: $ docker run -it --rm --security-opt apparmor:unconfined --security-opt seccomp:unconfined busybox / # adduser -D test / # su - test ~ $ sleep infinity & ~ $ exit / # stat /proc/$(pidof sleep)/exe File: stat: /proc/10/exe: cannot read link: Permission denied Size: 0 Blocks: 0 IO Block: 1024 symbolic link Device: 6eh/110d Inode: 187271 Links: 1 Access: (0777/lrwxrwxrwx) Uid: ( 1000/ test) Gid: ( 1000/ test) Access: 2022-01-03 06:52:39.480790247 +0000 Modify: 2022-01-03 06:52:39.480790247 +0000 Change: 2022-01-03 06:52:39.480790247 +0000 $ docker run -it --rm --security-opt apparmor:unconfined --security-opt seccomp:unconfined --cap-add sys_ptrace busybox / # adduser -D test / # su - test ~ $ sleep infinity & ~ $ exit / # stat /proc/$(pidof sleep)/exe File: '/proc/10/exe' -> '/bin/sleep' Size: 0 Blocks: 0 IO Block: 1024 symbolic link Device: 6eh/110d Inode: 195011 Links: 1 Access: (0777/lrwxrwxrwx) Uid: ( 1000/ test) Gid: ( 1000/ test) Access: 2022-01-03 07:13:28.003224653 +0000 Modify: 2022-01-03 07:13:28.003224653 +0000 Change: 2022-01-03 07:13:28.003224653 +0000 I tried to adapt the function that retrieves swtpm PID so it also covers the usecase when libvirt is run in a container without ptrace capability. The patch solved the issue for me and I verified that the error is no more reproducible. So I wanted to propose that solution to handle the issue. Or maybe someone can suggest a better alternative which would be more suitable? Would appreciate any feedback. Thanks. [1] https://github.com/moby/moby/issues/40713 Vasiliy Ulyanov (1): qemu_tpm: Get swtpm pid without binary validation src/qemu/qemu_tpm.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-) -- 2.34.1

3 4

[PATCH v4 00/12] Add riscv kvm accel support
by Yifei Jiang 12 Jan '22

12 Jan '22

This series adds both riscv32 and riscv64 kvm support, and implements migration based on riscv. Because of RISC-V KVM has been merged into the Linux master, so this series are changed from RFC to patch. Several steps to use this: 1. Build emulation $ ./configure --target-list=riscv64-softmmu $ make -j$(nproc) 2. Build kernel 3. Build QEMU VM Cross built in riscv toolchain. $ PKG_CONFIG_LIBDIR=<toolchain pkgconfig path> $ export PKG_CONFIG_SYSROOT_DIR=<toolchain sysroot path> $ ./configure --target-list=riscv64-softmmu --enable-kvm \ --cross-prefix=riscv64-linux-gnu- --disable-libiscsi --disable-glusterfs \ --disable-libusb --disable-usb-redir --audio-drv-list= --disable-opengl \ --disable-libxml2 $ make -j$(nproc) 4. Start emulation $ ./qemu-system-riscv64 -M virt -m 4096M -cpu rv64,x-h=true -nographic \ -name guest=riscv-hyp,debug-threads=on \ -smp 4 \ -bios ./fw_jump.bin \ -kernel ./Image \ -drive file=./hyp.img,format=raw,id=hd0 \ -device virtio-blk-device,drive=hd0 \ -append "root=/dev/vda rw console=ttyS0 earlycon=sbi" 5. Start kvm-acceled QEMU VM in emulation $ ./qemu-system-riscv64 -M virt,accel=kvm -m 1024M -cpu host -nographic \ -name guest=riscv-guset \ -smp 2 \ -bios none \ -kernel ./Image \ -drive file=./guest.img,format=raw,id=hd0 \ -device virtio-blk-device,drive=hd0 \ -append "root=/dev/vda rw console=ttyS0 earlycon=sbi" Changes since patch v3 - Re-write the for-loop in sifive_plic_create(). - Drop unnecessary change in hw/riscv/virt.c. - Use serial to handle console sbi call. Changes since patch v2 - Create a macro for get and put timer csr. - Remove M-mode PLIC contexts when kvm is enabled. - Add get timer frequency. - Move cpu_host_load to vmstate_kvmtimer. Changes since patch v1 - Rebase on recent commit a216e7cf119c91ffdf5931834a1a030ebea40d70 - Sync-up headers with Linux-5.16-rc4. - Fixbug in kvm_arch_init_vcpu. - Create a macro for get and put regs csr. - Start kernel directly when kvm_enabled. - Use riscv_cpu_set_irq to inject KVM interrupts. - Use the Semihosting Console API for RISC-V kvm handle sbi. - Update vmstate_riscv_cpu version id. Placing kvm_timer into a subsection. Changes since RFC v6 - Rebase on recent commit 8627edfb3f1fca24a96a0954148885c3241c10f8 - Sync-up headers with Linux-5.16-rc1 Changes since RFC v5 - Rebase on QEMU v6.1.0-rc1 and kvm-riscv linux v19. - Move kvm interrupt setting to riscv_cpu_update_mip(). - Replace __u64 with uint64_t. Changes since RFC v4 - Rebase on QEMU v6.0.0-rc2 and kvm-riscv linux v17. - Remove time scaling support as software solution is incomplete. Because it will cause unacceptable performance degradation. and We will post a better solution. - Revise according to Alistair's review comments. - Remove compile time XLEN checks in kvm_riscv_reg_id - Surround TYPE_RISCV_CPU_HOST definition by CONFIG_KVM and share it between RV32 and RV64. - Add kvm-stub.c for reduce unnecessary compilation checks. - Add riscv_setup_direct_kernel() to direct boot kernel for KVM. Changes since RFC v3 - Rebase on QEMU v5.2.0-rc2 and kvm-riscv linux v15. - Add time scaling support(New patches 13, 14 and 15). - Fix the bug that guest vm can't reboot. Changes since RFC v2 - Fix checkpatch error at target/riscv/sbi_ecall_interface.h. - Add riscv migration support. Changes since RFC v1 - Add separate SBI ecall interface header. - Add riscv32 kvm accel support. Yifei Jiang (12): update-linux-headers: Add asm-riscv/kvm.h target/riscv: Add target/riscv/kvm.c to place the public kvm interface target/riscv: Implement function kvm_arch_init_vcpu target/riscv: Implement kvm_arch_get_registers target/riscv: Implement kvm_arch_put_registers target/riscv: Support start kernel directly by KVM target/riscv: Support setting external interrupt by KVM target/riscv: Handle KVM_EXIT_RISCV_SBI exit target/riscv: Add host cpu type target/riscv: Add kvm_riscv_get/put_regs_timer target/riscv: Implement virtual time adjusting with vm state changing target/riscv: Support virtual time context synchronization hw/intc/sifive_plic.c | 21 +- hw/riscv/boot.c | 16 +- hw/riscv/virt.c | 83 +++-- include/hw/riscv/boot.h | 1 + linux-headers/asm-riscv/kvm.h | 128 +++++++ meson.build | 2 + target/riscv/cpu.c | 29 +- target/riscv/cpu.h | 11 + target/riscv/kvm-stub.c | 30 ++ target/riscv/kvm.c | 532 +++++++++++++++++++++++++++++ target/riscv/kvm_riscv.h | 25 ++ target/riscv/machine.c | 30 ++ target/riscv/meson.build | 1 + target/riscv/sbi_ecall_interface.h | 72 ++++ 14 files changed, 948 insertions(+), 33 deletions(-) create mode 100644 linux-headers/asm-riscv/kvm.h create mode 100644 target/riscv/kvm-stub.c create mode 100644 target/riscv/kvm.c create mode 100644 target/riscv/kvm_riscv.h create mode 100644 target/riscv/sbi_ecall_interface.h -- 2.19.1

3 20

[libvirt PATCH v4 0/7] cgroup and thread management in ch driver.
by Praveen K Paladugu 11 Jan '22

11 Jan '22

This patchset adds support for cgroup management of ch threads. This version correctly manages cgroups for vcpu and emulator threads created by ch. cgroup management for iothreads is not yet supported. Along with cgroup management, this patchset also enables support for pinning vcpu and emulator threads to selected host cpus. v4: * addressed all open comments in v3 * dropped all the merged commits v3: * addrressed all the formatting comments in v2 patch set * dropped indentation patches are they do not adhere to libvirt coding style * fixed build issue in qemu driver that was introduced in v2 Praveen K Paladugu (3): qemu,hypervisor: refactor some cgroup mgmt methods ch_process: Setup emulator and iothread settings ch_driver: emulator threadinfo & pinning callbacks Vineeth Pillai (4): ch: methods for cgroup mgmt in ch driver ch_driver,ch_domain: vcpupin callback in ch driver ch_driver: enable typed param string for numatune ch_driver: add numatune callbacks for CH driver src/ch/ch_conf.c | 2 + src/ch/ch_conf.h | 4 +- src/ch/ch_domain.c | 64 ++++ src/ch/ch_domain.h | 18 +- src/ch/ch_driver.c | 590 +++++++++++++++++++++++++++++++++ src/ch/ch_monitor.c | 156 +++++++++ src/ch/ch_monitor.h | 56 +++- src/ch/ch_process.c | 385 ++++++++++++++++++++- src/ch/ch_process.h | 3 + src/hypervisor/domain_cgroup.c | 457 ++++++++++++++++++++++++- src/hypervisor/domain_cgroup.h | 72 ++++ src/libvirt_private.syms | 14 +- src/qemu/qemu_cgroup.c | 413 +---------------------- src/qemu/qemu_cgroup.h | 11 - src/qemu/qemu_driver.c | 14 +- src/qemu/qemu_hotplug.c | 7 +- src/qemu/qemu_process.c | 24 +- 17 files changed, 1835 insertions(+), 455 deletions(-) -- 2.27.0

1 7

[PATCH 0/3] virdnsmasq: Lookup DNSMASQ in PATH
by Michal Privoznik 11 Jan '22

11 Jan '22

This is an alternative implementation to Olaf's original patch posted here: https://listman.redhat.com/archives/libvir-list/2021-August/msg00273.html I find my patches to be a bit cleaner, but either way we should merge his or mine. Michal Prívozník (3): virdnsmasq: Drop @binaryPath argument from dnsmasqCapsNewEmpty() virdnsmasq: Lookup DNSMASQ in PATH virdnsmasq: Require non NULL @caps in dnsmasqCapsGetBinaryPath() src/util/virdnsmasq.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) -- 2.34.1

4 13