Eric Blake [Sat, 5 Oct 2013 19:41:44 +0000 (13:41 -0600)]
maint: avoid 'const fooPtr' in virnet files
'const fooPtr' is the same as 'foo * const' (the pointer won't
change, but it's contents can). But in general, if an interface
is trying to be const-correct, it should be using 'const foo *'
(the pointer is to data that can't be changed).
Eric Blake [Sat, 5 Oct 2013 15:51:55 +0000 (09:51 -0600)]
maint: avoid 'const fooPtr' in several util files
'const fooPtr' is the same as 'foo * const' (the pointer won't
change, but it's contents can). But in general, if an interface
is trying to be const-correct, it should be using 'const foo *'
(the pointer is to data that can't be changed).
Fix up offenders in src/util outside of the virnet namespace.
Also, make a few virSocketAddr functions const-correct, for easier
conversions in future patches.
Eric Blake [Sat, 5 Oct 2013 02:30:35 +0000 (20:30 -0600)]
maint: avoid 'const fooPtr' in hashes
'const fooPtr' is the same as 'foo * const' (the pointer won't
change, but it's contents can). But in general, if an interface
is trying to be const-correct, it should be using 'const foo *'
(the pointer is to data that can't be changed).
Fix up virhash to provide a const-correct interface: all actions
that don't modify the table take a const table. Note that in
one case (virHashSearch), we actually strip const away - we aren't
modifying the contents of the table, so much as associated data
for ensuring that the code uses the table correctly (if this were
C++, it would be a case for the 'mutable' keyword).
Eric Blake [Mon, 7 Oct 2013 18:36:00 +0000 (12:36 -0600)]
maint: avoid 'const fooPtr' in public API
'const fooPtr' is the same as 'foo * const' (the pointer won't
change, but it's contents can). But in general, if an interface
is trying to be const-correct, it should be using 'const foo *'
(the pointer is to data that can't be changed).
Fix up offenders in the public API. Note that this is an API change;
but see commit 6ac6f59, where we first argued that this change is
harmless (but with that commit not actually making the change that it
claimed to be making):
Although this is an API change (not ABI though), real callers won't be
impacted. Why?
1. these callback members are read-only, so it is less likely that
someone is trying to assign into the struct members.
2. The only way to register a virConnectDomainEventGraphicsCallback is
to cast it through a call to virConnectDomainEventRegisterAny. That is,
even if the user's callback function leaves out the const, we never use
the typedef as the direct type of any API parameter. Since they are
already casting their function pointer into a munged type before
registering it, their code will continue to compile.
* include/libvirt/libvirt.h.in
(virConnectDomainEventGraphicsCallback): Use intended type.
Improve log filtering in virLXCProcessReadLogOutputData
Make the virLXCProcessReadLogOutputData method ignore the log
lines about the container startup argv, ignore the generic
error message from libvirt_lxc when lxcContainerMain fails
and skip over blank lines.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The lxcContainerResolveSymlinks method merely logged some errors
as debug messages, rather than reporting them as proper errors.
This meant startup failures were not diagnosed at all.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Revert "VMX: Some serial ports are not actually connected"
This reverts commit dba04e7fa070a79415ea16141ad2e2ebd4f23033.
This change was unfortunately not correct. We should have been
changing the boolean argument supplied.
Peter Krempa [Wed, 9 Oct 2013 16:05:43 +0000 (18:05 +0200)]
qemu: snapshot: Add support for compressing external snapshot memory
The regular save image code has the support to compress images using a
specified algorithm. This was not implemented for external checkpoints
although it shares most of the backend code.
Peter Krempa [Wed, 9 Oct 2013 16:05:43 +0000 (18:05 +0200)]
qemu: managedsave: Add support for compressing managed save images
The regular save image code has the support to compress images using a
specified algorithm. This was not implemented for managed save although
it shares most of the backend code.
Initialize threading & error layer in LXC controller
In Fedora 20, libvirt_lxc crashes immediately at startup with a
trace
#0 0x00007f0cddb653ec in free () from /lib64/libc.so.6
#1 0x00007f0ce0e16f4a in virFree (ptrptr=ptrptr@entry=0x7f0ce1830058) at util/viralloc.c:580
#2 0x00007f0ce0e2764b in virResetError (err=0x7f0ce1830030) at util/virerror.c:354
#3 0x00007f0ce0e27a5a in virResetLastError () at util/virerror.c:387
#4 0x00007f0ce0e28858 in virEventRegisterDefaultImpl () at util/virevent.c:233
#5 0x00007f0ce0db47c6 in main (argc=11, argv=0x7fff4596c328) at lxc/lxc_controller.c:2352
Normally virInitialize calls virErrorInitialize and
virThreadInitialize, but we don't link to libvirt.so
in libvirt_lxc, and nor did we ever call the error
or thread initializers.
I have absolutely no idea how this has ever worked, let alone
what caused it to stop working in Fedora 20.
In addition not all code paths from virLogSetFromEnv will
ensure virLogInitialize is called correctly, which is another
possible crash scenario.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Made the cgroups code fallback to non-systemd based setup
when dbus is not running. It was too big a hammer though,
as it did not check what error code was received when the
dbus connection failed. Thus it silently ignored serious
errors from dbus such as "too many client connections",
which should always be treated as fatal.
We only want to ignore errors if the dbus unix socket does
not exist, or if nothing is listening on it.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The LXC code would read the log file if an LXC guest failed to
startup. There were a number of failure cases where the guest
will not start and libvirtd never gets as far as looking at the
log file.
Fix this by replacing some earlier generic errors with messages
from the log.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The LXC controller main() method initialized 'rc' to 1
rather than '-1'. In the cleanup path it will print any
error to stderr, if-and-only-if rc < 0. Hence the incorrect
initialization caused errors to be lost.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Which is clearly not what it should have done. This caused the code to
skip over things which are not log messages. The solution is to simply
add brackets.
A test case is also added to validate correctness.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Make LXC controller use a private dbus connection & close it
The LXC controller uses dbus to talk to systemd to create
cgroups. This means that each LXC controller instance has
a dbus connection. The DBus daemon is limited to 256
connections by default and we want to be able to run many
1000 of containers.
While the dbus limit could be raised in the config files,
it is simpler to make libvirt LXC controller close its
dbus connection once everything is configured.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The dbus_bus_get() function returns a shared bus connection that
all libraries in a process can use. You are forbidden from calling
close on this connection though, since you can never know if any
other code might be using it.
Add an option to use private dbus bus connections, if the app
wants to be able to close the connection.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Doug Goldstein [Thu, 10 Oct 2013 21:31:47 +0000 (16:31 -0500)]
rpc: Fix getsockopt on Snow Leopard and lower
Since 5a468b38b6 we use SOL_LOCAL for the 2nd argument of getsockopt()
however Lion added the define SOL_LOCAL set to 0, which is the value to
the 2nd argument of getsockopt() for Unix sockets on Mac OS X. So
instead of using the define just pass 0 so we restore compatibility
with Snow Leopard and Leopard.
Reported at https://github.com/mxcl/homebrew/pull/23141
Doug Goldstein [Tue, 1 Oct 2013 02:13:17 +0000 (21:13 -0500)]
VMware: Simplify array walk for driver type
Rather than walking the possible driver backends by handle, use a helper
function. Additionally I've done a bit of refactoring in the code over
the past few commits so add myself to the copyright line.
Michal Privoznik [Fri, 11 Oct 2013 12:15:32 +0000 (14:15 +0200)]
qemu: Include listenAddress in debug prints
After my patches, some functions gained one more argument
(@listenAddress) which wasn't included in debug printing of
arguments they were called with. Functions in question are:
qemuMigrationPrepareDirect and qemuMigrationPerform.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
qemu_migration: Avoid crashing if domain dies too quickly
I've noticed a SIGSEGV-ing libvirtd on the destination when the qemu
died too quickly = in Prepare phase. What is happening here is:
1) [Thread 3493] We are in qemuMigrationPrepareAny() and calling
qemuProcessStart() which subsequently calls qemuProcessWaitForMonitor()
and qemuConnectMonitor(). So far so good. The qemuMonitorOpen()
succeeds, however switching monitor to QMP mode fails as qemu died
meanwhile. That is qemuMonitorSetCapabilities() returns -1.
2) [Thread 3262] The event loop is trying to do the talking to monitor.
However, qemu is dead already, remember?
2013-10-08 15:54:13.436+0000: 3262: error : qemuMonitorIORead:551 : Unable to read from monitor: Connection reset by peer
2013-10-08 15:54:13.516+0000: 3262: debug : virFileClose:90 : Closed fd 25
...
2013-10-08 15:54:13.533+0000: 3493: debug : qemuMonitorSend:968 : Send command resulted in error internal error: early end of file from monitor: possible problem:
3) [Thread 3493] qemuProcessStart() failed. No big deal. Go to the
'endjob' label and subsequently to the 'cleanup'. Since the domain is
not persistent and ret is -1, the qemuDomainRemoveInactive() is called.
This has an (unpleasant) effect of virObjectUnref()-in the @vm object.
Unpleasant because the event loop which is about to trigger EOF callback
still holds a pointer to the @vm (not the reference). See the valgrind
output below.
4) [Thread 3262] So the event loop starts triggering EOF:
And the monitor is cleaned up. This results in calling
qemuProcessHandleMonitorEOF with the @vm pointer passed. The pointer is
kept in qemuMonitor struct.
==3262== Thread 1:
==3262== Invalid read of size 4
==3262== at 0x77ECCAA: pthread_mutex_lock (in /lib64/libpthread-2.15.so)
==3262== by 0x52FAA06: virMutexLock (virthreadpthread.c:85)
==3262== by 0x52E3891: virObjectLock (virobject.c:320)
==3262== by 0x11626743: qemuProcessHandleMonitorEOF (qemu_process.c:296)
==3262== by 0x11642593: qemuMonitorIO (qemu_monitor.c:730)
==3262== by 0x52BD526: virEventPollDispatchHandles (vireventpoll.c:501)
==3262== by 0x52BDD49: virEventPollRunOnce (vireventpoll.c:648)
==3262== by 0x52BBC68: virEventRunDefaultImpl (virevent.c:274)
==3262== by 0x542D3D9: virNetServerRun (virnetserver.c:1112)
==3262== by 0x11F368: main (libvirtd.c:1513)
==3262== Address 0x14549128 is 24 bytes inside a block of size 136 free'd
==3262== at 0x4C2AF5C: free (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==3262== by 0x529B1FF: virFree (viralloc.c:580)
==3262== by 0x52E3703: virObjectUnref (virobject.c:270)
==3262== by 0x531557E: virDomainObjListRemove (domain_conf.c:2355)
==3262== by 0x1160E899: qemuDomainRemoveInactive (qemu_domain.c:2061)
==3262== by 0x1163A0C6: qemuMigrationPrepareAny (qemu_migration.c:2450)
==3262== by 0x1163A923: qemuMigrationPrepareDirect (qemu_migration.c:2626)
==3262== by 0x11682D71: qemuDomainMigratePrepare3Params (qemu_driver.c:10309)
==3262== by 0x53B0976: virDomainMigratePrepare3Params (libvirt.c:7266)
==3262== by 0x1502D3: remoteDispatchDomainMigratePrepare3Params (remote.c:4797)
==3262== by 0x12DECA: remoteDispatchDomainMigratePrepare3ParamsHelper (remote_dispatch.h:5741)
==3262== by 0x54322EB: virNetServerProgramDispatchCall (virnetserverprogram.c:435)
The mon->vm is set in qemuMonitorOpenInternal() which is the correct
place to increase @vm ref counter. The correct place to decrease the ref
counter is then qemuMonitorDispose().
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Bing Bu Cao [Fri, 11 Oct 2013 06:50:33 +0000 (14:50 +0800)]
util: fix two virCompareLimitUlong bugs
The helper function virCompareLimitUlong compares limit values,
where value of 0 is equal to unlimited. If the latter parameter is 0,
it should return -1 instead of 1, hence the user can only set hard_limit when
swap_hard_limit currently is unlimited.
Worse, all callers pass 2 64-bit values, but on 32-bit platforms,
the second argument was silently truncated to 32 bits, which
could lead to incorrect computations.
Signed-off-by: Bing Bu Cao <mars@linux.vnet.ibm.com> Signed-off-by: Eric Blake <eblake@redhat.com>
This function takes exactly one argument: an address to check.
It returns true, if the address is an IPv4 or IPv6 address in numeric
format, false otherwise (e.g. for "examplehost").
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Whenever we check for ABI stability, we have new xml (e.g. provided by
user, or obtained from snapshot, whatever) which we compare to old xml
and see if ABI won't break. However, if the new xml was produced via
virDomainGetXMLDesc(..., VIR_DOMAIN_XML_MIGRATABLE) it lacks some
devices, e.g. 'pci-root' controller. Hence, the ABI stability check
fails even though it is stable. Moreover, we can't simply fix
virDomainDefCheckABIStability because removing the correct devices is
task for the driver. For instance, qemu driver wants to remove the usb
controller too, while LXC driver doesn't. That's why we need special
qemu wrapper over virDomainDefCheckABIStability which removes the
correct devices from domain XML, produces MIGRATABLE xml and calls the
check ABI stability function.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
qemumonitorjsontest: Extend the test for yet another monitor commands
So far, we're unit testing some basic functions and some (so called)
simple functions (e.g. "qmp_capabilities", "system_powerdown"). However,
there are more functions which expect simple "{'return': {}}" reply, but
takes more args to construct the command (for instance "set_link"). This
patch aims on such functions.
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
John Ferlan [Wed, 9 Oct 2013 13:03:25 +0000 (09:03 -0400)]
storage_backend: Fix issue with allocation of 0 length volume
Commit id '532fef36' added a call to fallocate() and some error
handling based on whether or not the function existed. This new
call resulted in libvirt-cim/cimtest failures when attempting to
create a volume with "0" (zero) allocation value. The failure is
logged as:
Oct 9 07:51:33 localhost libvirtd[8030]: cannot allocate 0 bytes in
file '/var/lib/libvirt/images/cimtest-vol.img': Invalid argument
This can also be seen with virsh vol-create-as:
error: Failed to create vol test
error: cannot allocate 0 bytes in file '/home/vm-images/test': Invalid
argument
error: Failed to create vol test
error: cannot allocate 0 bytes in file '/home/vm-images/test': Invalid
argument
It turns out fallocate() will return EINVAL when the incoming 'len'
(or allocation) value is 0 (or less).
Michal Privoznik [Thu, 10 Oct 2013 10:32:49 +0000 (12:32 +0200)]
qemu: Init @pcidevs in qemuPrepareHostdevPCIDevices
At the beginning of the function qemuPrepareHostdevPCICheckSupport() is
called. After that @pcidevs is initialized. However, if the very first
command fails, we go to 'cleanup' label where virObjectUnref(pcidevs) is
called. Obviously, it is called before @pcidevs was able to get
initialized. Compiler warns about it:
CC qemu/libvirt_driver_qemu_impl_la-qemu_hostdev.lo
qemu/qemu_hostdev.c: In function 'qemuPrepareHostdevPCIDevices':
qemu/qemu_hostdev.c:824:19: error: 'pcidevs' may be used uninitialized in this function [-Werror=maybe-uninitialized]
virObjectUnref(pcidevs);
^
cc1: all warnings being treated as errors
Peter Krempa [Fri, 20 Sep 2013 08:39:51 +0000 (10:39 +0200)]
qemu: Prefer VFIO for PCI device passthrough
Prefer using VFIO (if available) to the legacy KVM device passthrough.
With this patch a PCI passthrough device without the driver configured
will be started with VFIO if it's available on the host. If not legacy
KVM passthrough is checked and error is reported if it's not available.
Peter Krempa [Thu, 19 Sep 2013 15:01:17 +0000 (17:01 +0200)]
qemu: hostdev: Add checks if PCI passthrough is available in the host
Add code to check availability of PCI passhthrough using VFIO and the
legacy KVM passthrough and use it when starting VMs and hotplugging
devices to live machine.
Ján Tomko [Wed, 9 Oct 2013 12:17:13 +0000 (14:17 +0200)]
LXC: Fix handling of RAM filesystem size units
Since 76b644c when the support for RAM filesystems was introduced,
libvirt accepted the following XML:
<source usage='1024' unit='KiB'/>
This was parsed correctly and internally stored in bytes, but it
was formatted as (with an extra 's'):
<source usage='1024' units='KiB'/>
When read again, this was treated as if the units were missing,
meaning libvirt was unable to parse its own XML correctly.
The usage attribute was documented as being in KiB, but it was not
scaled if the unit was missing. Transient domains still worked,
because this was balanced by an extra 'k' in the mount options.
This patch:
Changes the parser to use 'units' instead of 'unit', as the latter
was never documented (fixing persistent domains) and some programs
(libvirt-glib, libvirt-sandbox) already parse the 'units' attribute.
Removes the extra 'k' from the tmpfs mount options, which is needed
because now we parse our own XML correctly.
Changes the default input unit to KiB to match documentation, fixing:
https://bugzilla.redhat.com/show_bug.cgi?id=1015689
Commit 2d74822a9eb4856c7f5216bb92bcb76630660f72 renamed
"freebsdNodeGetCPUCount" to "appleFreebsdNodeGetCPUCount", leaving one
call to "freebsdNodeGetCPUCount". Fix this other case.
Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com> Signed-off-by: Eric Blake <eblake@redhat.com>
The current OOM test impl is too inefficient to be used on the
large test suites. It loops running 'main' multiple times, once
for each alloc in the 'main' method. This has complexity
'n * (n + 1) / 2' in terms of total alloc count. It will be
replaced by a more efficient impl whicih runs individual test
cases instead. This will have same complexity but the values
of 'n' will be much smaller, so a net win.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Hongwei Bi [Mon, 7 Oct 2013 14:04:26 +0000 (22:04 +0800)]
fix a ambiguous output of the command:'virsh vol-create-as'
I created a storage volume(eg: test) from a storage pool(eg:vg10) using
the following command:"virsh vol-create-as --pool vg10 --name test --capacity 300M."
When I re-executed the above command, the output was as the following:
"error: Failed to create vol test
error: Storage volume not found: storage vol 'test' already exists"
I think the output "Storage volume not found" is not appropriate. Because in fact storage
vol test has been found at this time. And then I think virErrorNumber should includes
VIR_ERR_STORAGE_EXIST which can also be used elsewhere. So I make this patch. The result
is as following:
"error: Failed to create vol test
error: storage volume 'test' exists already"
Eric Blake [Fri, 4 Oct 2013 17:51:41 +0000 (11:51 -0600)]
build: add configure --without-readline
Make it much easier to test a configuration built without readline
support, by reusing our existing library probe machinery. It gets
a bit tricky with readline, which does not provide a pkg-config
snippet, and which on some platforms requires one of several
terminal libraries as a prerequiste, but the end result should be
the same default behavior but now with the option to disable things.
* m4/virt-readline.m4 (LIBVIRT_CHECK_READLINE): Simplify by using
LIBVIRT_CHECK_LIB.
* tools/virsh.c: Convert USE_READLINE to WITH_READLINE.
because when it is disabled, the user gets no indication if they
touch a file that would normally require a rebuild. Automake
1.11 changed things so that AM_MAINTAINER_MODE([enable]) will set
the mode to enabled by default; but RHEL 5 still uses automake 1.9,
where AM_MAINTAINER_MODE did not recognize an argument, and
therefore disables maintainer mode by default. Having the default
be different according to which version of automake built the
project is annoying, and I _have_ been bitten on RHEL 5 rebuilds
where the default disabled mode led to silently incorrect builds.
The automake manual admits that being able to disable maintainer
mode still makes sense for projects that still store generated
files from the autotools in version control; but we have dropped
that for several years now. As such, it's finally time to just
ditch the whole idea of maintainer mode, and unconditionally
rebuild autotools files if a dependency changes, without offering
a configure option to disable that mode.
Remove use of virConnectPtr from all remaining nwfilter code
The virConnectPtr is passed around loads of nwfilter code in
order to provide it as a parameter to the callback registered
by the virt drivers. None of the virt drivers use this param
though, so it serves no purpose.
Avoiding the need to pass a virConnectPtr means that the
nwfilterStateReload method no longer needs to open a bogus
QEMU driver connection. This addresses a race condition that
can lead to a crash on startup.
The nwfilter driver starts before the QEMU driver and registers
some callbacks with DBus to detect firewalld reload. If the
firewalld reload happens while the QEMU driver is still starting
up though, the nwfilterStateReload method will open a connection
to the partially initialized QEMU driver and cause a crash.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Don't pass virConnectPtr in nwfilter 'struct domUpdateCBStruct'
The nwfilter driver only needs a reference to its private
state object, not a full virConnectPtr. Update the domUpdateCBStruct
struct to have a 'void *opaque' field instead of a virConnectPtr.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Claudio Bley [Mon, 7 Oct 2013 10:13:00 +0000 (12:13 +0200)]
Adjust legacy max payload size to account for header information
Commit 27e81517a87 set the payload size to 256 KB, which is
actually the max packet size, including the size of the header.
Reduce this by VIR_NET_MESSAGE_HEADER_MAX (24) and set
VIR_NET_MESSAGE_LEGACY_PAYLOAD_MAX to 262120, which was the original
value before increasing the limit in commit eb635de1fed.
Ryota Ozaki [Sat, 5 Oct 2013 05:56:37 +0000 (14:56 +0900)]
nodeinfo: make freebsdNodeGetCPUCount work on Mac OS X
This fixes the following error:
error : nodeGetInfo:933 : this function is not supported
by the connection driver: node info not implemented on this platform
The freebsdNodeGetCPUCount was renamed to appleFreebsdNodeGetCPUCount
in order to make more visible the fact, that it works on Mac OS X too.
Mac OS X can use sysctlbyname as same as FreeBSD to get the CPU
frequency. However, the MIB style name is different from FreeBSD's.
And the unit of the return frequency is also different.
Signed-off-by: Ryota Ozaki <ozaki.ryota@gmail.com> Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Ryota Ozaki [Sat, 5 Oct 2013 05:56:36 +0000 (14:56 +0900)]
rpc: fix getsockopt for LOCAL_PEERCRED on Mac OS X
This fixes the following error:
error : virGetUserEnt:703 : Failed to find user record for uid '32654'
'32654' (it's random and varies) comes from getsockopt with
LOCAL_PEERCRED option. getsockopt returns w/o error but seems
to not set any value to the buffer for uid.
For Mac OS X, LOCAL_PEERCRED has to be used with SOL_LOCAL level.
With SOL_LOCAL, getsockopt returns a correct uid.
Note that SOL_LOCAL can be found in
/System/Library/Frameworks/Kernel.framework/Versions/A/Headers/sys/un.h.
Signed-off-by: Ryota Ozaki <ozaki.ryota@gmail.com> Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
Eric Blake [Fri, 4 Oct 2013 23:46:43 +0000 (17:46 -0600)]
build: fix build on 32-bit platforms
qemumonitorjsontest.c: In function 'testQemuMonitorJSONqemuMonitorJSONGetBalloonInfo':
qemumonitorjsontest.c:1134: warning: integer constant is too large for 'long' type
* tests/qemumonitorjsontest.c
(testQemuMonitorJSONqemuMonitorJSONGetBalloonInfo)
(testQemuMonitorJSONqemuMonitorJSONGetBlockStatsInfo): Use correct
type.
Eric Blake [Fri, 4 Oct 2013 23:25:13 +0000 (17:25 -0600)]
build: fix build on RHEL 5
On RHEL 5, compilation fails with:
storage/storage_backend.c: In function 'createRawFile':
storage/storage_backend.c:339: warning: implicit declaration of function 'fallocate'
storage/storage_backend.c:339: warning: nested extern declaration of 'fallocate' [-Wnested-externs]
Huh? It turns out that in kernels that old, fallocate() is not
implemented (config.h is correct), but <linux/fs.h> defines
HAVE_FALLOCATE as an empty witness macro for a completely
different purpose. Since storage_backend.c is including
<linux/fs.h> on RHEL 5, we are hosed by the kernel definition.
Newer kernels no longer pollute the namespace, and it's fairly
easy to convert to an expression that works with both the old
kernel witness and the new-style config.h (undefined or 1).
Eric Blake [Fri, 27 Sep 2013 23:09:20 +0000 (17:09 -0600)]
build: fix build --without-remote
I tried to test ./configure --without-lxc --without-remote.
First, the build failed with some odd errors, such as an
inability to build xen, or link failures for virNetTLSInit.
But when you think about it, once there is no remote code,
all of libvirtd is useless, any stateful driver that depends
on libvirtd is also not worth compiling, and any libraries
used only by RPC code are not needed. So I patched
configure.ac to make for some saner defaults when an
explicit disable is attempted. Similarly, since we have
migrated virnetdevbridge into generic code, the workaround
for Linux kernel stupidity must not depend on stateful
drivers being in use.
Then there's 'make check' that needs segregation.
Wow - quite a bit of cleanup to make --without-remote useful :)
* configure.ac: Let --without-remote toggle defaults on stateful
drivers and other libraries. Pick up Linux kernel workarounds
even when qemu and lxc are not being compiled.
* tests/Makefile.am (test_programs): Factor out programs that
require remote.
* src/libvirt_private.syms (rpc/virnet*.h): Move...
* src/libvirt_remote.syms: ...into new file.
* src/Makefile.am (SYM_FILES): Ship new syms file.
storage: fix file allocation behavior in file cloning
Fixed the safezero call for allocating the rest of the file after cloning
an existing volume; it used to always use a zero offset, causing it to
only allocate the beginning of the file.
Also modified file creation to try to use fallocate(2) to pre-allocate
disk space before copying any data to make sure it fails early on if disk
is full and makes sure we can skip zero blocks when copying file contents.
If fallocate isn't available we will zero out the rest of the file after
cloning and only use sparse cloning if client requested a lower allocation
than the input volume's capacity.
My previous commit 7dc1d4ab was supposed to change safezero to allocate
1 megabyte at maximum, but had the logic reversed and will allocate 1
megabyte at minimum (and a lot more at maximum.)
virfile: safezero: fall back to writing block by block if mmap fails
mmap can fail on 32-bit systems if we're trying to zero out a lot of data.
Fall back to using block-by-block writing in that case. While we could map
smaller blocks it's unlikely that this code is used a lot and its easier to
just fall back to one of the existing methods.
Also modified the block-by-block zeroing to not allocate a megabyte of
zeroes if we're writing less than that.
Again stolen from qemu_driver.c, but dropping all the unneeded bits.
This aims to copy all the current qemu validation checks since that's
the most commonly used real driver, but some of the checks are
completely artificial in the test driver.
This only supports creation of internal snapshots for initial
simplicity.
but the fix only worked for <interface type='hostdev'>, *not* for
<interface type='network'> where the network itself was a pool of
hostdevs.
The symptom in both cases was this error message:
internal error: Unable to determine device index for network device
In both cases the cause was lack of proper handling for netdevs
(<interface>) of type='hostdev' when scanning the netdev list looking
for alias names in qemuAssignDeviceNetAlias() - those that aren't
type='hostdev' have an alias of the form "net%d", while those that are
hostdev use "hostdev%d". This special handling was completely lacking
prior to the fix for Bug 827519 which was:
When searching for the highest alias index, libvirt looks at the alias
for each netdev and if it is type='hostdev' it ignores the entry. If
the type is not hostdev, then it expects the "net%d" form; if it
doesn't find that, it fails and logs the above error message.
That fix works except in the case of <interface type='network'> where
the network uses hostdev (i.e. the network is a pool of VFs to be
assigned to the guests via PCI passthrough). In this case, the check
for type='hostdev' would fail because it was done as:
def->net[i]->type == VIR_DOMAIN_NET_TYPE_HOSTDEV
(which compares what was written in the config) when it actually
should have been:
(which compares the type of netdev that was actually allocated from
the network at runtime).
Of course the latter wouldn't be of any use if the netdevs of
type='network' hadn't already acquired their actual network connection
yet, but manual examination of the code showed that this is never the
case.
While looking through qemu_command.c, two other places were found to
directly compare the net[i]->type field rather than getting actualType:
* qemuAssignDeviceAliases() - in this case, the incorrect comparison
would cause us to create a "net%d" alias for a netdev with
type='network' but actualType='hostdev'. This alias would be
subsequently overwritten by the proper "hostdev%d" form, so
everything would operate properly, but a string would be
leaked. This patch also fixes this problem.
* qemuAssignDevicePCISlots() - would defer assigning a PCI address to
a netdev if it was type='hostdev', but not for type='network +
actualType='hostdev'. In this case, the actual device usually hasn't
been acquired yet anyway, and even in the case that it has, there is
no practical difference between assigning a PCI address while
traversing the netdev list or while traversing the hostdev
list. Because changing it would be an effective NOP (but potentially
cause some unexpected regression), this usage was left unchanged.
Join the worker thread no matter if it is running or zombie already.
With current implementation the thread is joined iff @running is true.
However, when worker executes the last line, @running is set to false.
Hence qemuMonitorTestFree() won't join it (and free resources) even
though we can clearly see worker has run (nobody else sets @running =
false).
Signed-off-by: Michal Privoznik <mprivozn@redhat.com>
qemumonitorjsontest: Test qemuMonitorJSONSystemPowerdown
Right now, we are testing qemuMonitorSystemPowerdown instead of
qemuMonitorJSONSystemPowerdown. It makes no harm, as both functions have
the same header and the former is just a wrapper over the latter. But we
should be consistent as we're testing the JSON functions only in here.
The XML parser reserves 'vnet' as a prefix for automatically
generated NIC device names. Switch the veth device creation
to use this prefix, so it does not have to worry about clashes
with user specified names in the XML.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
The veth device creation code run in two steps, first it looks
for two free veth device names, then it runs ip link to create
the veth pair. There is an obvious race between finding free
names and creating them, when guests are started in parallel.
Rewrite the code to loop and re-try creation if it fails, to
deal with the race condition.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
If veth device allocation has a fatal error, the veths
array may contain NULL device names. Avoid calling the
virNetDevVethDelete function on such names.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
Avoid reporting an error if veth device is already deleted
The kernel automatically destroys veth devices when cleaning
up the container network namespace. During normal shutdown, it
is thus likely that the attempt to run 'ip link del vethN'
will fail. If it fails, check if the device exists, and avoid
reporting an error if it has gone. This switches to use the
virCommand APIs instead of virRun too.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
During container cleanup there is a race where the kernel may
have destroyed the veth device before we try to set it offline.
This causes log error messages. Given that we're about to
delete the device entirely, setting it offline is pointless.
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
projects / libvirt.git / log