Although v4.6.0 was just recently released we do not want to jump to
it due to:
https://bugzilla.redhat.com/show_bug.cgi?id=1614569
Unfortunately the mess around the above defect basically renders that
release useless. We will need to keep an eye for either a v4.7.0 or
grab updates off of the 'v4.5-maint' over the coming weeks.
As usual I have tested using qemu and lxc usecases (define, boot and
'virsh console' onto a guest). The ptests results match those of
v4.3.0:
======================================
Testsuite summary for libvirt 4.5.0
======================================
# TOTAL: 119
# PASS: 114
# SKIP: 0
# XFAIL: 0
# FAIL: 5
# XPASS: 0
# ERROR: 0
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Oddly something seems to have changed since I did the uprev of libvirt
to v4.3.0. Previously using the 'run-ptests' script would only run the
tests and not attempt to build them first. In preparation for the
uprev to v4.5.0 I attempted to run the tests and found that they would
fail to run and instead the tests were attempting to be built. I
suspect this change is caused by an uprev of autotools and the
handling of the 'TESTS' built-in.
To prevent this we are explicitly using the '--with-test-suite'
configure option and this return the previously observed behavior,
where the tests are simply run and not built.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
If our project is in a directory with a pathname of longer than
about 125 characters, configuration will fail with an "Argument
list too long" error. This patch uses the technique that was applied
to the coreutils recipe.
Signed-off-by: Joe Slater <joe.slater@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
xen and xen-inotify have been removed from upstream. We change
PACKAGECONFIG accordingly.
pkg_postinst has been deprecated, use pkg_postinst_ontarget instead.
WARNING: do_rootfs: Intentionally failing postinstall scriptlets of
['libvirt'] to defer them to first boot is deprecated. Please place
them into pkg_postinst_ontarget_${PN} ().
Signed-off-by: He Zhe <zhe.he@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
There is host contamination in that we force XMLLINT to be
'/usr/bin/xmllint' via ac_cv_path_XMLLINT. This results in the error:
/bin/bash: /usr/bin/xmllint: No such file or directory
We only use XMLLINT during the build, it is present in two Makefiles
included with ptest package but they are unused. So we drop the
setting of ac_cv_path_XMLLINT to ensure we are using xmllint in the
recipe sysroot instead of the host's. Should the ptest package ever
evolve to make use of the Makefiles we need to update not only the
XMLLINT path but that for XSLTPROC and others too. (I would suggest
that the Makefiles may have been used on the target as part of ptest
at one point, but are no longer and should be removed, though I
haven't investigated further).
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
We have been using the 1.3.x release series for a long time now which
has been great for stability but is slowly becoming harder and harder
to track and port bug and CVE fixes. This is a big jump to the latest
upstream release which gives us access to a myriad of fixes as well as
puts us in a better position to contribute to the upstream project
when issues are found.
Several patches have been dropped as they are either no longer valid
against this release or have equivalent updates already applied to the
upstream project. Some patches were consolidated which should ease
future uprevs of this recipe. The majority of the updates were related
to ptest patches, which is not a huge surprise given this code has no
upstream equivalent.
The overall runtime behavior remains much the same from v1.3.5 with
the only notable configuration change being for 'seccomp_sandbox'
which has been disabled here but should possibly be revisited in the
near future.
As usual the normal runtime usecases for qemu/kvm and lxc have been
run successfully along with ptest results which are by and large OK:
====================================
Testsuite summary for libvirt 4.3.0
====================================
# TOTAL: 119
# PASS: 115
# SKIP: 0
# XFAIL: 0
# FAIL: 4
# XPASS: 0
# ERROR: 0
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
With Sun RPC Interfaces removed from libc we need to use an external
provider for rpc. Polling other distros there seems to be consensus to
use libtirpc so we follow this trend.
Unfortunately this should only require the DEPENDS addition to the
recipe but the upstream project has a few flaws in their Makefiles and
configure which we need to work around so we add 2 new patches to
address these to allow for the use of a sysroot and to use the
$XDR_CFLAGS in a few more places.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
A while ago changes were merged to meta-openembedded to make
/etc/dnsmasq.d (and specifically the files it contains) referenced
when the main instance of dnsmasq is run (see dnsmasq.service and
commit ba665493a0dd [dnsmasq: allow for dnsmasq instances to reuse
default dnsmasq.conf]).
We, however, continued to modify the global configuration
(/etc/dnsmasq.conf) to keep the main instance of dnsmasq from
attaching to virbr0 and lxcbr0, by using 'bind-dynamic'. This approach
is problematic, since it is common that other instances of dnsmasq
will make use of the global configuration file and may have
incompatible options. We see this for example when attempting to start
lxc-net which will attempt to use 'bind-interface' which is
incompatible with 'bind-dynamic' that we were adding to the global
configuration.
Here we remove our change to the global configuration (leaving it
mostly empty as it should be) and instead have lxc and libvirt
packages instruct the global instance not to bind to virbr0 and lxcbr0
by adding configuration files to /etc/dnsmasq.d (setting
except-interface).
The added benefit to this approach is that if lxc or libvirt are not
part of an image the global configuration will not be modified in such
a way as to expect that they are present.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
The build is failing due to a failing QA check:
ERROR: libvirt-1.3.5-r0 do_package_qa: QA Issue: /usr/lib/libvirt/ptest/tests/tests/check-file-access.pl
contained in package libvirt-ptest requires /usr/bin/perl,
but no providers found in RDEPENDS_libvirt-ptest? [file-rdeps]
Signed-off-by: Chunrong Guo <chunrong.guo@nxp.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Commit 8709d4504c35817ef1d7181b47ba96c3ca0c949e [lvm2: separate
libdevmapper to avoid circular dependency] in meta-oe meant that our
build would fail with
checking for libdevmapper.h... no
configure: error: You must install device-mapper-devel/libdevmapper >= 1.0.0 to compile libvirt
Add the required DEPENDS for libdevmapper to get the build working as
it did prior to this meta-oe change.
NOTE: our previous dependency on lvm2 was not only to get libdevmapper
functionality, so lvm2 remains a valid DEPENDS, so is left in place.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
The DIRFILES variable was a temporary workaround to resolve a do_rootfs
RPM package conflict between libvirt and polkit.
This happened because of the different permissions and ownership that the
packages placed on ${datadir}/usr/share/polkit-1/rules.d.
The behaviour is now conditional based on how PACKAGECONFIG is set. The
polkit rules will be removed from the package if polkit is not enabled.
If polkit is enabled the permissions and ownership are set to match those
set by the polkit recipe.
This uses the useradd enhancements for RSS. It requires that shadow-native
be included as a DEPENDS to provide the useradd command in the native
sysroot.
Signed-off-by: Rob Woolley <rob.woolley@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
The switch to RPM4 and dnf results in this error during rootfs construction:
Error: Transaction check error:
file /usr/share/polkit-1/rules.d conflicts between attempted installs of libvirt-1.3.5-r0.13.core2_64 and polkit-0.113-r0.9.core2_64
We can temporarily work around it by instructing RPM to not package/process
the dirfiles.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
The patch fix the following QA warning:
--
WARNING: QA Issue: libvirt rdepends on libacl, but it isn't a build dependency? [build-deps]
Signed-off-by: Zhenbo Gao <zhenbo.gao@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Systemd untis should be placed as per ${systemd_system_unitdir}.
Patched the source code to support configure variable to set the systemd units
directory.
Signed-off-by: Amarnath Valluri <amarnath.valluri@intel.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
getVar() now defaults to expanding by default.
Signed-off-by: Paul Barker <paul@paulbarker.me.uk>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
libvirt by default has '--with-wireshark-dissector' set to
'check'. Therefor if wireshark is built before vs after libvirt this
will result in a potentially unexpected difference in how libvirt is
configured. Additionally the build of libvirt with wireshark
capabilities enabled is currently broken and thus if attempted will
result in the error:
fatal error: ws_symbol_export.h: No such file or directory
To prevent the random configuration of libvirt with respect to
wireshark we add a new packageconfig for wireshark.
NOTE that the wireshark recipe is currently incomplete so this new
packageconfig will result in a build failure if used, until wireshark
is updated to build libwsutil.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
When do split_and_strip_files, it may calls debugedit to deal elf files.
For ptest sub-package, it fails to run debugedit with .o files. So remove
the .o files which are redundant to run ptest cases.
Signed-off-by: Kai Kang <kai.kang@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Add missing test helpers
We don't support windows encoding so remove that case
Signed-off-by: He Zhe <zhe.he@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Apparently if a license is listed for a package it must be included in
the overall LICENSE variable otherwise you will get WARNINGS such as:
WARNING: LICENSE_libvirt-ptest includes licenses (LGPLv2.1 GPLv2+)
that are not listed in LICENSE
We ensure that the license string is consistent between the LICENSE
variable and the subpackage license string to remove this
warning. Since '2+' is a superset of '2', ie. "version 2 or later",
and actually is more reflective of the text found on the file headers
and such we use GPLv2+ and LGPLv2.1+.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
When build for architecture that don't use gcc atomic ops but pthread,
it fails to build for arm:
| ../tools/nss/.libs/libnss_libvirt_impl.a(libvirt_nss_la-virobject.o): In function `virClassNew':
| /buildarea2/kkang/builds/qemuarm-Aug03/bitbake_build/tmp/work/armv5e-wrs-linux-gnueabi/libvirt/1.3.5-r0/build/src/../../libvirt-1.3.5/src/util/virobject.c:153: undefined reference to `virAtomicLock'
| ../tools/nss/.libs/libnss_libvirt_impl.a(libvirt_nss_la-virobject.o): In function `virObjectNew':
| /buildarea2/kkang/builds/qemuarm-Aug03/bitbake_build/tmp/work/armv5e-wrs-linux-gnueabi/libvirt/1.3.5-r0/build/src/../../libvirt-1.3.5/src/util/virobject.c:205: undefined reference to `virAtomicLock'
| ../tools/nss/.libs/libnss_libvirt_impl.a(libvirt_nss_la-virobject.o): In function `virObjectUnref':
| /buildarea2/kkang/builds/qemuarm-Aug03/bitbake_build/tmp/work/armv5e-wrs-linux-gnueabi/libvirt/1.3.5-r0/build/src/../../libvirt-1.3.5/src/util/virobject.c:277: undefined reference to `virAtomicLock'
| ../tools/nss/.libs/libnss_libvirt_impl.a(libvirt_nss_la-virobject.o): In function `virObjectRef':
| /buildarea2/kkang/builds/qemuarm-Aug03/bitbake_build/tmp/work/armv5e-wrs-linux-gnueabi/libvirt/1.3.5-r0/build/src/../../libvirt-1.3.5/src/util/virobject.c:298: undefined reference to `virAtomicLock'
| collect2: error: ld returned 1 exit status
It is similar with:
http://libvirt.org/git/?p=libvirt.git;a=commit;h=12dc729
Signed-off-by: Kai Kang <kai.kang@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Upgrade libvirt from 1.3.2 to 1.3.5.
* update checksums and remove PR
* remove trailing whitespace
* remove backport patch 0001-fix-crash-in-getAllDomainStats.patch
* update context of patches
* remove libvirtd.socket from SYSTEMD_SERVICE which doesn't exist anymore
* install test data which ptest requires
Signed-off-by: Kai Kang <kai.kang@windriver.com>
Signed-off-by: Bruce Ashfield <bruce@zedd.org>
Currently libvirt python builds in the multilib case, but fails to
install. Switch from PN to BPN, and fix compile to be consistent.
Signed-off-by: Bruce Ashfield <bruce@zedd.org>
openvzutilstest.conf file is needed by openvzutilstest test
Signed-off-by: Catalin Enache <catalin.enache@windriver.com>
Signed-off-by: Bruce Ashfield <bruce@zedd.org>
Upstream libvirt Commits 1d39dbaf and 827ed9b4 broke the libvirt-python API
by removing virDomainRef() and virDomainFree(). virDomainStatsRecordListFree() will
free that domain pointer and later when virDomain (python object) call
its destructor and tries to free that same pointer again.
Signed-off-by: Roy Li <rongqing.li@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
The 'base_contains' is now deprecated and only kept as a compatibility method. It will be removed in future releases.
Signed-off-by: Derek Straka <derek@asterius.io>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
The .la files generated for the ptest libraries have multiple references
to the working dir and continue to trigger QA errors.
Rather than playing whack-a-mole with the references, we can simply remove
the .la files .. since they are not required on the target.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
oe-core 3739aaf644 [insane: Fix populate_sysroot sanity test path] means
that .la files are subject to a QA sanity check that fails if the build
paths (workdir) are found.
The ptest and main libvirt files have silently had this issue with hosts
paths.
One fix is to remove the .la files, but this could impact on target test
and builds, so our first attempt to fix them is to follow the lead of a
similar lttng_tools fix, and modify .la files to have paths and values that
are valid on target.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
We build libvirt without support for sd_notify (dbus) therefor we have
to run libvirtd as a 'forking daemon' (per commit
055744cdbc we pass --daemon)
We need to update the systemd service file to reflect this such that
the libvirtd.service can be properly stopped. If we ever enable
sd_notify support in libvirt we can revert back to using the 'notify'
systemd type. Without this change libvirtd continue to run even if the
stop command is issued, subsequent attempts to start the service will
fail.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
The release notes is http://libvirt.org/news.html
Signed-off-by: Roy Li <rongqing.li@windriver.com>
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
When systemd starts libvirtd we want to make sure we pass the correct
arguments to run the daemon. If we fail to do so the libvirtd service
will restart about once a minute and not allow for VMs to properly be
instantiated. libvirtd now runs with the same args as when we use
sysvinit.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
While testing libvirt with the new systemd init I found that the
libvirt-guests service was failing to run since the libvirt-guests.sh
script was not where it was expected. If we are building for systemd
init system use the default installation location which is used
upstream for the libvirt-guests.sh script.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Commit d114a62dd8 [libvirt: kill
update-rc.d when not using sysvinit] resulted in libvirt services not
starting automatically at boot when using systemd. The upstream
libvirt package already has the necessary systemd service files so it
is just a matter of activating them via the '--with-init-script'
configure option and making the recipe use these.
We always set '--with-init-script' to 'systemd' regardless of
the systemd vs sysvinit DISTRO_FLAG since our sysvinit scripts are
external and thus not affected by this setting when building for
sysvinit.
Both sysvinit and systemd packages will be libvirt-libvirtd, the
bbclasses ensure that only the files for one init system will exist in
the package, so this 'overloading' is acceptable.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
We have had a persistent issue of contention with multiple instances
of dnsmasq. When libvirtd is started it will attempt and fail to
create virbr0 that includes a dnsmasq binding (dnsmasq complains that
the port is already in use). The root cause being that the base
dnsmasq configuration uses the 'default' binding, meaning it will be
greedy and bind to all available interfaces, along with any new
interfaces that appear. This prevents any new instances of dnsmasq to
fail. By changing the base dnsmasq configuration to use 'bind-dynamic'
we will be able to create new interfaces and if we want to attach a
new dnsmasq instance for the new interface we can, without contention.
Overall this change should not change the default behavior (other then
fixing the contention) as the 'bind-dynamic' mode will allow the base
config to attach to new interfaces which are not be bound to new
dnsmasq instances.
Removing dnsmasq_exclude_virbr.conf which was a previous (partial)
attempt that was made to resolve this issue.
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Fixes the following QA warning:
WARNING: QA Issue: libvirt-ptest rdepends on libtasn1, but it isn't a build dependency? [build-deps]
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Configure searches for and sets XSLTPROC and will not get a proper
value unless we have made sure libxslt-native is built prior to
libvirt. So we need to add it as a DEPENDS. Failing to do this we
could end up with libvirt configuring before libxslt-native has built
and installed xlstproc in the -native sysroot, resulting potential
host contamination or in a potential build error:
cat: acl.html.tmp: No such file or directory
Makefile:2215: recipe for target 'acl.html' failed
(xsltproc is used in the generation of acl.html.tmp)
Signed-off-by: Mark Asselstine <mark.asselstine@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
* Fixes a dependency on a nonexistent initscripts-functions package.
Signed-off-by: Noor Ahsan <noor_ahsan@mentor.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Remove libvirt-Fix-cannot-symlink-cpu_map.xml-error.patch, since
rules for install *.xml are changed, the race issue does not exist.
Signed-off-by: Roy Li <rongqing.li@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Fix the following warning by passing correct option to configure script:
WARNING: QA Issue: libvirt: configure was passed unrecognised options:
--with-capnp [unknown-configure-option]
Signed-off-by: Dmitry Eremin-Solenikov <dmitry_eremin@mentor.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
libvirt wants to use pcap-config to locate the exisence and location
of libpcap. oe-core stubs this script and replaces it with pkg-config,
which can lead to the host pcap-config triggering and either breaking
the build or introducing host contamination.
To fix this issue, we patch configure to use 'pkg-config libcap' to
locate the correct libraries.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
In some circumstances where the build tree differs from the source,
libvirt's compile will try to create the symlink for cpu_map.xml before
creating the directory $(abs_builddir)/cpu.
Add a test to create this directory if it hasn't already been created.
Signed-off-by: Amy Fong <amy.fong@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Some package formats explicitly track which files are configuration files
so that they are not overwritten on updates. Explicitly list the
configuration files.
Signed-off by: Rob Woolley <rob.woolley@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Other than updating the checksums, no other significant differences are
present in the 1.2.8 -> 1.2.11 update of libvirt.
Sanity testing passes, so we do the update.
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
Add PACKAGECONFIG for audit, libcap-ng, otherwise there would be
warnings like below:
WARN: libvirt: libvirt-virsh rdepends on audit, but it isn't a build dependency?
WARN: libvirt: libvirt-virsh rdepends on libcap-ng, but it isn't a build dependency?
WARN: libvirt: libvirt rdepends on audit, but it isn't a build dependency?
WARN: libvirt: libvirt rdepends on libcap-ng, but it isn't a build dependency?
WARN: libvirt: libvirt-libvirtd rdepends on audit, but it isn't a build dependency?
WARN: libvirt: libvirt-libvirtd rdepends on libcap-ng, but it isn't a build dependency?
Both audit and libcap-ng should be enabled/disabled along with selinux
since all of them are from meta-selinux.
Signed-off-by: Wenzong Fan <wenzong.fan@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@windriver.com>
