IFF_VNET_HDR is a tun/tap flag that allows you to send and receive
large (i.e. GSO) packets and packets with partial checksums. Setting
the flag means that every packet is proceeded by the same header which
virtio uses to communicate GSO/csum metadata.
By enabling this flag on the tap fds we create, we greatly increase
the achievable throughput with virtio_net.
However, we need to be careful to only set the flag when a) QEMU has
support for this ABI and b) the value of the flag is queryable using
the TUNGETIFF ioctl.
It's nearly five months since kvm-74 - the first KVM release with this
feature - was released. Up until now, we've not added libvirt support
because there is no clean way to detect support for this in QEMU at
runtime. A brief attempt to add a "info capabilities" monitor command
to QEMU floundered. Perfect is the enemy of good enough. Probing the
KVM version will suffice for now.
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
* src/virterror.c (virStrerror): New function.
(virReportSystemErrorFull): Don't leak "combined".
In fact, don't even attempt allocation.
Do include the result of formatted print in final diagnostic.
* src/proxy_internal.c (xenProxyCommand): Mark a diagnostic.
* src/xen_unified.c (xenUnifiedOpen, xenUnifiedAddDomainInfo):
Fix unmarked diagnostics by removing the diagnostic altogether:
replace each xenUnifiedError(...,VIR_ERR_NO_MEMORY call with a
call to virReportOOMError.
* .cvsignore: Don't ignore *.orig or *.rej. They're not build products.
Don't ignore the entire m4 directory.
* Makefile.maint (sync-vcs-ignore-files): Correct quoting.
Use sed rather than a for loop.
Search only version-controled files (for reproducibility)
* gnulib/lib/netinet/.cvsignore: Append missing newline-at-EOF,
so that the use of sed doesn't mistakenly concatenate lines.
* gnulib/lib/sys/.cvsignore: Likewise.
* m4/.cvsignore: Ignore acinclude.m4 and aclocal.m4, not *.m4.
* .hgignore: Regenerate.
* all .gitignore files: Regenerate.
from the post-install script:
Installing : libvirt
ln: creating symbolic link `/etc/libvirt/qemu/networks/autostart/default.xml': File exists
See https://bugzilla.redhat.com/462011
* src/network_driver.c (networkEnableIpForwarding): This function
would always return 1 (success). Now it returns 0 upon write or
close failure, being careful to preserve any errno from a write failure.
Add comments.
domain shutdown and read saved vm status on libvirtd startup,
last 2 patches from Guido Günther finishing up the surviving the
libvirt daemon restart
daniel
* qemud/qemud.c (remoteReadConfigFile): Don't return 0 (success)
when the config file is unreadable or nonexistent
Return -1, not 0, upon virConfReadFile failure.
(main): If remote_config_file is not specified via --config(-f),
use the default config file only if it exists. Otherwise,
use /dev/null.
* src/conf.c (virConfReadFile): Don't diagnose virFileReadAll
failure, since it already does that.
* qemud/Makefile.am (check-local): Prefix use of test_libvirtd.aug
with $(srcdir)/. Add quotes around $(AUGPARSE), in case it expands
to something unusual.
* docs/formatdomain.html.in docs/libvirt.rng docs/formatdomain.html:
document the new PCI passthrough format
* tests/qemuxml2argvtest.c tests/qemuxml2xmltest.c
tests/qemuxml2argvdata/qemuxml2argv-hostdev-pci-address.args
tests/qemuxml2argvdata/qemuxml2argv-hostdev-pci-address.xml:
and add a regression test for it.
Daniel
* gnulib/m4/multiarch.m4: New file.
* gnulib/m4/hostent.m4: New file.
* gnulib/m4/servent.m4: New file.
* .x-sc_prohibit_gethostby: Exempt all gnulib/m4/* files, so
hostent.m4's uses of gethostbyname and gethostbyaddr do not
trigger a "make syntax-check" failure.
* bootstrap: Invoke gnulib-tool with its --no-vc-files
option, since we version-control the files it imports.
This tells it not to mark imported files as cvs-ignored.
(modules): Add ioctl.
* gnulib/lib/ioctl.c: New file.
* gnulib/lib/sys_ioctl.in.h: New file.
* gnulib/tests/sys_ioctl.in.h: Removed.
* gnulib/tests/ioctl.c: Removed.
* gnulib/lib/Makefile.am: Update.
* gnulib/m4/gnulib-cache.m4: Likewise.
* gnulib/m4/gnulib-comp.m4: Likewise.
* gnulib/tests/Makefile.am: Likewise.
* gnulib/lib/.cvsignore: Ignore sys_ioctl.h.
* gnulib/lib/.gitignore: Regenerate.
src/uml_driver.c src/util.c src/util.h src/xen_unified.c:
unify hostname lookup using virGetHostname convenience function,
patch by David Lutterkort
daniel
* configure.in: Provide a new --with-bridge option.
* src/Makefile.am (PRIVSYMFILES) [WITH_BRIDGE]: Append libvirt_bridge.syms.
* src/bridge.c: Test WITH_BRIDGE rather than particular drivers.
* src/libvirt_private.syms: Move bridge-related symbols into...
* src/libvirt_bridge.syms: ...this new file.
Author: John Levon
the logging system
* docs/auth.html docs/deployment.html docs/remote.html
docs/sitemap.html docs/sitemap.html.in docs/uri.html
docs/windows.html: integration in the existing set
Daniel
* qemud/qemud.c qemud/test_libvirtd_qemu.aug tests/Makefile.am
tests/daemon-conf: fix make check with the new logging support
the messages now carry a timestamp which need to be removed,
the daemon needs to exit if the log configuration informations
are wrong and we also look at the LIBVIRT_DEBUG environment
variable
Daniel
* src/logging.h src/logging.c: add the infrastructure and internal
APIs for logging, some of those APIs should be made public later
* src/libvirt_sym.version.in: flag the internal APIs as such
daniel
qemudCapsInitNUMA and umlCapsInitNUMA were identical, so this change
factors them into a new function, virCapsInitNUMA, and puts it in
nodeinfo.c.
In addition to factoring out the duplicates, this change also
adjusts that function definition (along with its macros) so
that it works with Fedora 9's numactl version 1, and makes it
so the code will work even if someone builds the kernel with
CONFIG_NR_CPUS > 4096.
Finally, also perform this NUMA initialization for the lxc
and openvz drivers.
* src/nodeinfo.c: Include <stdint.h>, <numa.h> and "memory.h".
(virCapsInitNUMA): Rename from qemudCapsInitNUMA and umlCapsInitNUMA.
(NUMA_MAX_N_CPUS): Define depending on NUMA API version.
(n_bits, MASK_CPU_ISSET): Define, adjust, use uint64 rather than long.
* src/nodeinfo.h: Include "capabilities.h".
(virCapsInitNUMA): Declare it.
* examples/domain-events/events-c/Makefile.am:
* src/Makefile.am: Add $(NUMACTL_CFLAGS) and $(NUMACTL_LIBS) to various
compile/link-related variables.
* src/qemu_conf.c: Include "nodeinfo.h".
(qemudCapsInitNUMA): Remove duplicate code. Adjust caller.
* src/uml_conf.c (umlCapsInitNUMA): Likewise.
Include "nodeinfo.h".
* src/lxc_conf.c: Include "nodeinfo.h".
(lxcCapsInit): Perform NUMA initialization here, too.
* src/openvz_conf.c (openvzCapsInit): And here.
Include "nodeinfo.h".
* src/libvirt_sym.version.in: Add virCapsInitNUMA so that libvirtd
can link to this function.
* src/storage_backend_fs.c: Don't include <endian.h> or <byteswap.h>.
(LV_BIG_ENDIAN, LV_LITTLE_ENDIAN): Define.
Use those instead of __BIG_ENDIAN and __LITTLE_ENDIAN.
* qemud/Makefile.am: Ensure that the generated remote_protocol.c
includes <config.h> first.
* Makefile.maint (sc_require_config_h_first): New rule, so that
"make syntax-check" enforces this.
* .x-sc_require_config_h_first: New file.
* Makefile.am (.x-sc_require_config_h_first): Add it.
* qemud/remote_protocol.x: Don't include <config.h> here.
* qemud/remote_protocol.c: Regenerate.
This change is required on some kernels due to the way a change in
the kernel's CONFIG_NR_CPUS propagates through the numa library.
* src/qemu_conf.c (qemudCapsInitNUMA): Pass numa_all_cpus_ptr->size/8
as the buffer-length-in-bytes in the call to numa_node_to_cpus, since
that's what is required on second and subseqent calls.
* src/uml_conf.c (umlCapsInitNUMA): Likewise.
This is the second part of the change mentioned here:
http://thread.gmane.org/gmane.comp.emulators.libvirt/10331
It removes the essentially redundant "optname" parameter
from each of the vshCommandOptNetworkBy and vshCommandOptDomainBy
functions as well as the correspond macros (without "By" suffix).
Now, instead of always passing the optname, "domain", to
vshCommandOptDomainBy, that function requires that its command
argument refer to an opts array containing a "domain" option.
This normalization makes one more help-related change:
it renames the net-start "name" argument to the more
sensible and consistent "network".
* src/virsh.c (VSH_BYNAME, vshCommandOptDomain)
(cmd_has_option): New function, used in vshCommandOptDomainBy
and vshCommandOptNetworkBy.
(vshCommandOptDomainBy, vshCommandOptNetworkBy): Remove the optname
parameter, it's always "domain" ("network"). Update all callers.
Call cmd_has_option.
(vshCommandOptNetwork, cmdAutostart, cmdConsole, cmdDomstate)
(cmdDomblkstat, cmdDomIfstat, cmdSuspend, cmdUndefine, cmdStart)
(cmdSave, cmdSchedinfo, cmdDump, cmdResume, cmdShutdown)
(cmdReboot, cmdDestroy, cmdDominfo, cmdVcpuinfo, cmdVcpupin)
(cmdSetvcpus, cmdSetmem, cmdSetmaxmem, cmdDumpXML, cmdDomname)
(cmdDomid, cmdDomuuid, cmdMigrate, cmdNetworkAutostart)
(cmdNetworkDestroy, cmdNetworkDumpXML, cmdNetworkName)
(opts_network_start, cmdNetworkStart, cmdNetworkUndefine)
(cmdNetworkUuid, cmdVNCDisplay, cmdTTYConsole, cmdAttachDevice)
(cmdDetachDevice, cmdAttachInterface, cmdDetachInterface)
(cmdAttachDisk, cmdDetachDisk, cmdEdit)
* src/Makefile.am (virsh-pool-edit.c): This code is generated
from cmdEdit, and cmdEdit uses the vshCommandOptDomain macro which
now, with the changes above, has only 3 (was 4) arguments, yet the
macro use is mapped to vshCommandOptPool, which still requires 4
arguments. So this change adjusts the sed code to reinsert the
just-removed argument -- we're not changing pool-related code right
now, because it's not as straight-forward.
* Makefile.maint (sc_prohibit_cvs_keyword): New rule.
Suggested by Daniel Veillard.
The new test exposed two uses of $Date$.
* docs/Goals: Don't use $Date$.
* python/TODO: Likewise.
* qemud/Makefile.am (libvirtd_LDADD): Add gnulib's libgnu.la last.
Otherwise, building with -g (no inlining) would fail due to a use
of c_isspace in libvirt_driver_qemu.a, which used to followed
libgnu.a in the link command.
