- Devel - libvirt List Archives

[libvirt] [RFC PATCH 0/4] Implementation of a dependency relation between domains
by Valentin BOUSSON 23 Mar '16

23 Mar '16

This patch series adds a feature that enables libvirt domains to have other domains as dependencies. One possible target is the simulation of a complex computing system composed of multiple CPUs, each of which represented by a different libvirt domain. CPUs in this example are virtualized separately, by a dedicated QEMU instance, because all of them are booting independently. Moreover, if those CPU must share some physical property (for example power supply or ACPI), it is meaningful to consider them as a group. Another possible target is Asymmetric Multi Processor (AMP) systems. Where a master processor uses one or more slave processors to offload computation. Master and slave have often different architectures. This patch series makes possible to define the slave domains as dependencies of the master domain. And the whole system to be booted with one single libvirt command. A use-case for this libvirt extension are the recent patch series in QEMU that enable modeling an AMP-like system with multiple QEMU instances. [RFC v2 0/6] SDM Interface [RFC v2 0/6] QEMU shared-memory-backend [RFC v2 1/1] backend: multi-client-socket Features : ---------- - When a domain A depends on a domain B (A=>B), signals are forward-propagated from A to B. So, when a user asks the domain A to be: - instantiated - destroyed - suspended - resumed domain B will automatically receive the same signal. - Single-level dependency: By construction domain dependencies are restricted to one single level, so to avoid wrong recursive (infinite) domain dependencies. This is also because there is no clear benefit in supporting multiple level of dependencies between domains. More specifically, given to domains, A and B, if A depends on B (A=>B), then B can not have any dependency (B => C is forbidden). How to test this patch series : ------------------------------- The source code can be downloaded by: git clone https://git.virtualopensystems.com/dev/libvirt.git \ -b domain-dependency Note: The patch series relies on the commit 95ca4fe, but with the patch proposed by Michal Privoznik, "[PATCH]qemuBuildVideoCommandLine: Don't access def->videos without check" that resolves a critical bug. To add a dependency between two domains (for example between domaninA and domainB), it suffices to add in the XML description of the first one: domainA.xml: <devices> ... <domaindependency> /absolute/path/to/the/domainB.xml </domaindependency> ... </devices> Note: The two domains mus be virtualized thanks to QEMU, and be hosted on the same filesystem. Finally, instanciate domainA with the command: virsh create domainA.xml And observe that both domains are running in the same time with: virsh list You can trigger the implemented signals and check that both domains are working as one: virsh suspend domainA virsh resume domainA virsh destroy domainA Note: as mentioned, all signals are forward-propagated, so if you manipulate the domainB directly, domainA will not be affected. This work has been sponsored by Huawei Technologies Duesseldorf GmbH. Valentin BOUSSON (4): Add XML description of the new dependency device Add the inner structure managing dependency devices Add primitives for manipulation of dependency structures Implement signal propagation to dependancy domains docs/schemas/domaincommon.rng | 5 ++ src/conf/domain_conf.c | 83 +++++++++++++++++++++++++- src/conf/domain_conf.h | 18 ++++++ src/qemu/qemu_driver.c | 135 ++++++++++++++++++++++++++++++++++++++++-- src/qemu/qemu_hotplug.c | 1 + 5 files changed, 237 insertions(+), 5 deletions(-) -- 1.9.1

2 5

[libvirt] [PATCH] qemuxml2argv: Mock virSCSIDeviceGetSgName
by Michal Privoznik 23 Mar '16

23 Mar '16

When constructing SCSI hostdev command line for qemu, the /sys/bus/scsi/devices/... dir is scanned. Unfortunately, even in the tests. This is needed to determine the name of SCSI device to passthrough to qemu, because in the domain XML we were given its address instead. Anyway, we should not be touching live system data in our test suite as it produced unpredictable results. The test is regressing from 1e9a083742efe on. Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com> --- tests/qemuxml2argvmock.c | 14 ++++++++++++++ 1 file changed, 14 insertions(+) diff --git a/tests/qemuxml2argvmock.c b/tests/qemuxml2argvmock.c index 8426108..b7dfebb 100644 --- a/tests/qemuxml2argvmock.c +++ b/tests/qemuxml2argvmock.c @@ -26,6 +26,7 @@ #include "virutil.h" #include "virstring.h" #include "virtpm.h" +#include "virscsi.h" #include <time.h> #include <unistd.h> @@ -84,3 +85,16 @@ virMemoryMaxValue(bool capped ATTRIBUTE_UNUSED) { return LLONG_MAX; } + +char * +virSCSIDeviceGetSgName(const char *sysfs_prefix ATTRIBUTE_UNUSED, + const char *adapter ATTRIBUTE_UNUSED, + unsigned int bus ATTRIBUTE_UNUSED, + unsigned int target ATTRIBUTE_UNUSED, + unsigned long long unit ATTRIBUTE_UNUSED) +{ + char *ret; + + ignore_value(VIR_STRDUP(ret, "sg0")); + return ret; +} -- 2.7.3

2 1

[libvirt] [libvirt-sandbox PATCH] Ignore generated file
by Guido Günther 23 Mar '16

23 Mar '16

--- .gitignore | 1 + 1 file changed, 1 insertion(+) diff --git a/.gitignore b/.gitignore index 390d65c..83831bb 100644 --- a/.gitignore +++ b/.gitignore @@ -31,6 +31,7 @@ po/stamp-it po/libvirt-sandbox.pot po/*.gmo stamp-h1 +build-aux/test-driver INSTALL libvirt-sandbox-1.0.pc libvirt-sandbox.spec -- 2.8.0.rc3

1 0

[libvirt] [PATCH 1/3 trivial] fix typo in virt-aa-helper helptext
by Serge Hallyn 23 Mar '16

23 Mar '16

it's --dryrun not --dry-run Signed-off-by: Serge Hallyn <serge.hallyn(a)ubuntu.com> --- src/security/virt-aa-helper.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/src/security/virt-aa-helper.c b/src/security/virt-aa-helper.c index a2d7226..b466626 100644 --- a/src/security/virt-aa-helper.c +++ b/src/security/virt-aa-helper.c @@ -107,7 +107,7 @@ vah_usage(void) " Options:\n" " -a | --add load profile\n" " -c | --create create profile from template\n" - " -d | --dry-run dry run\n" + " -d | --dryrun dry run\n" " -D | --delete unload and delete profile\n" " -f | --add-file <file> add file to profile\n" " -F | --append-file <file> append file to profile\n" -- 2.7.0

3 5

[libvirt] [libvirt-sandbox PATCH] Allow to switch to a specific user id
by Guido Günther 22 Mar '16

22 Mar '16

Useful if you want to run programs as the same user than outside the sandbox. --- bin/virt-sandbox.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/bin/virt-sandbox.c b/bin/virt-sandbox.c index 4c400d5..9495e85 100644 --- a/bin/virt-sandbox.c +++ b/bin/virt-sandbox.c @@ -24,6 +24,8 @@ #include <libvirt-sandbox/libvirt-sandbox.h> #include <glib/gi18n.h> +#include <sys/types.h> +#include <pwd.h> static gboolean do_close(GVirSandboxConsole *con G_GNUC_UNUSED, gboolean error G_GNUC_UNUSED, @@ -92,6 +94,7 @@ int main(int argc, char **argv) { gchar *kernver = NULL; gchar *kernpath = NULL; gchar *kmodpath = NULL; + gchar *switchto = NULL; gboolean verbose = FALSE; gboolean debug = FALSE; gboolean shell = FALSE; @@ -126,6 +129,8 @@ int main(int argc, char **argv) { N_("security properties"), "PATH", }, { "privileged", 'p', 0, G_OPTION_ARG_NONE, &privileged, N_("run the command privileged"), NULL }, + { "switchto", 'S', 0, G_OPTION_ARG_STRING, &switchto, + N_("swith to the given user"), "USER" }, { "shell", 'l', 0, G_OPTION_ARG_NONE, &shell, N_("start a shell"), NULL, }, { "kernver", 0, 0, G_OPTION_ARG_STRING, &kernver, @@ -139,6 +144,7 @@ int main(int argc, char **argv) { { NULL, 0, 0, G_OPTION_ARG_NONE, NULL, NULL, NULL } }; const char *help_msg = N_("Run 'virt-sandbox --help' to see a full list of available command line options"); + struct passwd *pw; setlocale(LC_ALL, ""); bindtextdomain(PACKAGE, LOCALEDIR); @@ -198,10 +204,25 @@ int main(int argc, char **argv) { if (kmodpath) gvir_sandbox_config_set_kmodpath(cfg, kmodpath); + if (privileged && switchto) { + g_printerr(_("'switchto' and 'privileged' are incompatible options\n")); + goto cleanup; + } + if (privileged) { gvir_sandbox_config_set_userid(cfg, 0); gvir_sandbox_config_set_groupid(cfg, 0); gvir_sandbox_config_set_username(cfg, "root"); + } else if (switchto) { + pw = getpwnam(switchto); + if (!pw) { + g_printerr(_("Failed to resolve user %s\n"), switchto); + goto cleanup; + } + gvir_sandbox_config_set_userid(cfg, pw->pw_uid); + gvir_sandbox_config_set_groupid(cfg, pw->pw_gid); + gvir_sandbox_config_set_username(cfg, pw->pw_name); + gvir_sandbox_config_set_homedir(cfg, pw->pw_dir); } if (envs && @@ -541,6 +562,11 @@ to this path to locate the modules. Retain root privileges inside the sandbox, rather than dropping privileges to match the current user identity. +=item B<-S USER>, B<--switchto=USER> + +Swith to the given user inside the sandbox and setup $HOME +accordingly. + =item B<-l>, B<--shell> Launch an interactive shell on a secondary console device -- 2.8.0.rc3

2 2

[libvirt] [PATCH] nss: use more strict check for automatic enabling
by Roman Bogorodskiy 22 Mar '16

22 Mar '16

Currently nss plugin is automatically enabled if nss.h header exists and --without-nss-plugin is not specified. However, presence of the nss.h header does not guarantee a compatible interface, so check for struct gaih_addrtuple to make sure it's enabled only on platforms where it could be built. --- m4/virt-nss.m4 | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/m4/virt-nss.m4 b/m4/virt-nss.m4 index 3fa4ad3..bc054db 100644 --- a/m4/virt-nss.m4 +++ b/m4/virt-nss.m4 @@ -25,12 +25,14 @@ AC_DEFUN([LIBVIRT_CHECK_NSS],[ fail=0 if test "x$with_nss_plugin" != "xno" ; then - AC_CHECK_HEADERS([nss.h], [ + AC_CHECK_TYPE([struct gaih_addrtuple], [ with_nss_plugin=yes ],[ if test "x$with_nss_plugin" = "xyes" ; then fail = 1 fi + ],[ + #include <nss.h> ]) if test $fail = 1 ; then -- 2.4.6

2 3

[libvirt] Does the proposed solution belong in KVM input devices?
by bancfc＠openmailbox.org 22 Mar '16

22 Mar '16

qemu-discuss doesn't seem the best place to reach the devs. I saw most of the work on KVM input devices was done by Gerd Hoffman and Dan Berrange. Dan do you think the QEMU virtual keyboard is the right place for a solution similar to the one described below? https://paul.reviews/behavioral-profiling-the-password-you-cant-change/ Would this be considered feature creep and best implemented in its own program?

3 3

[libvirt] [libvirt-php 0/3] small improvements
by Vasiliy Tolstov 22 Mar '16

22 Mar '16

Add some more libvirt functions support to binding. Also add some checks to prevent malloc SIZE_MAX memory in case of errors Vasiliy Tolstov (3): add some checks to prevent overflow add block_commit support and needed const add libvirt_domain_block_job_info src/libvirt-php.c | 152 ++++++++++++++++++++++++++++++++++++++++++++++++------ src/libvirt-php.h | 2 + 2 files changed, 138 insertions(+), 16 deletions(-) -- 2.7.0

4 12

[libvirt] [PATCH 00/12] qemu_process cleanups and test improvement
by Pavel Hrdina 22 Mar '16

22 Mar '16

The first patch is unrelated but I've found it while working on this series. Pavel Hrdina (12): qemu_process: check for correct return value while starting domain nvram: generate it's path in qemuDomainDefPostParse tests: cleanup qemuxml2argvtest qemu_process: introduce qemuProcessPrepareDomain qemu_process: introduce qemuProcessPrepareHost qemu: update callers of qemuProcessStartValidate to use virDomainObjPtr qemu_process: move checks to qemuProcessStartValidate qemu_process: move qemuDomainSetPrivatePaths to qemuProcessInit qemu_process: introduce qemuProcessCreateCmd qemu_driver: cleanup qemuConnectDomainXMLToNative qemuxml2argvtest: use driver.config and priv for qemuDomainSetPrivatePaths qemuxml2argvtest: cleanup test src/qemu/qemu_domain.c | 10 + src/qemu/qemu_driver.c | 61 +- src/qemu/qemu_migration.c | 9 +- src/qemu/qemu_process.c | 706 +++++++++++---------- src/qemu/qemu_process.h | 26 +- .../qemuxml2argv-channel-virtio-unix.xml | 2 +- tests/qemuxml2argvtest.c | 126 ++-- tests/testutilsqemu.c | 7 + 8 files changed, 494 insertions(+), 453 deletions(-) -- 2.7.2

3 27

[libvirt] [PATCH 0/2] qemu: support SPICE+unix socket
by Cole Robinson 22 Mar '16

22 Mar '16

qemu has supported SPICE+unix socket listening since 2.3.0. Patch #1 wires it up from domain_conf.c down to qemu, patch #2 adds qemu.conf spice_auto_unix_socket option Full disclosure: this duplicates the logic of VNC socket usage, which has some pre-existing warts: - When the VM is running with graphics socket=FOO, the runtime XML still lists <listen type='address' address='127.0.0.1'/>. - If using vnc/spice auto socket, it looks like ports can still be allocated in the code, though they are never used. - In fact I suspect there's lots of issues with stray XML port/listen being parsed and ending up in the device config, but selectively being ignored later. We should probably explicitly ignore/error about those combos when parsing the XML so we never have to ignore specific values later on. - The auto socket path is set in qemu_command.c but it should probably be set in qemu_process.c with the port stuff. Though there are references to vncListen in qemu_command.c which should probably be moved as well. The issue with moving them to qemu_process.c is now there isn't any pre-existing way to test those bits like via qemuxml2argvtest for example. Though finding a way to test the graphics prep stuff would be valuable anyways - teuf suggests: converting <graphics socket=X/> to <graphics><listen type='socket' path='FOO'/></graphics> like was done with listen= address - tuef: virsh domdisplay doesn't handle sockets To avoid yet another rabbit hole I didn't touch any of the above issues, which basically means spice+unix is no less operationally worse than vnc+unix :) Cole Robinson (2): qemu: Support SPICE listen over unix socket qemu: Add qemu.conf option spice_auto_unix_socket docs/schemas/domaincommon.rng | 5 ++++ src/conf/domain_conf.c | 26 +++++++++++++-------- src/conf/domain_conf.h | 1 + src/qemu/libvirtd_qemu.aug | 1 + src/qemu/qemu.conf | 11 +++++++++ src/qemu/qemu_command.c | 15 +++++++++--- src/qemu/qemu_conf.c | 1 + src/qemu/qemu_conf.h | 1 + src/qemu/qemu_process.c | 3 +++ src/qemu/test_libvirtd_qemu.aug.in | 1 + src/security/virt-aa-helper.c | 5 ++++ .../qemuxml2argv-graphics-spice-unix-auto.args | 21 +++++++++++++++++ .../qemuxml2argv-graphics-spice-unix-auto.xml | 27 ++++++++++++++++++++++ .../qemuxml2argv-graphics-spice-unix.args | 21 +++++++++++++++++ .../qemuxml2argv-graphics-spice-unix.xml | 27 ++++++++++++++++++++++ tests/qemuxml2argvtest.c | 10 ++++++++ 16 files changed, 163 insertions(+), 13 deletions(-) create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-graphics-spice-unix-auto.args create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-graphics-spice-unix-auto.xml create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-graphics-spice-unix.args create mode 100644 tests/qemuxml2argvdata/qemuxml2argv-graphics-spice-unix.xml -- 2.5.0

3 9