The refactoring of QEMU command startup was comitted with
a couple of VIR_WARN lines left in from debugging.
* src/qemu/qemu_driver.c: Remove log warning lines
When qemuMonitorSetCapabilities() fails, there is no need to
call qemuMonitorClose(), because the caller will already see
the error code and tear down the entire VM. The extra call to
qemuMonitorClose resulted in a double-free due to it removing
a ref count prematurely.
* src/qemu/qemu_driver.c: Remove premature close of monitor
If the event loop takes a very long time todo something, it is
possible for the 'self pipe' buffer to become full at which
point the entire event loop + remote driver deadlock. Use a
boolean flag to ensure we have strict one-in, one-out behaviour
on writes/reads of the 'self pipe'
Regression in commit caa805ea let a lot of bad messages slip in.
* cfg.mk (msg_gen_function): Fix function name.
* src/qemu/qemu_cgroup.c (qemuRemoveCgroup): Fix fallout from
'make syntax-check'.
* src/qemu/qemu_driver.c (qemudDomainGetInfo)
(qemuDomainWaitForMigrationComplete, qemudStartVMDaemon)
(qemudDomainSaveFlag, qemudDomainAttachDevice)
(qemuDomainUpdateDeviceFlags): Likewise.
* src/qemu/qemu_hotplug.c (qemuDomainAttachHostUsbDevice)
(qemuDomainDetachPciDiskDevice, qemuDomainDetachSCSIDiskDevice):
Likewise.
When attaching device from a xml file and the device is mis-configured,
virsh gives mis-leading message "out of memory". This patch fixes this.
Signed-off-by: Eric Blake <eblake@redhat.com>
If the memory of guest OS is changed constantly, the live migration
can not be ended ever for ever.
We can use the command 'virsh migrate-setmaxdowntime' to control the
live migration. But the value of maxdowntime is diffcult to calculate
because it depends on the transfer speed of network and constantly
changing memroy size. We need a easy way to control the live migration.
This patch adds the support of forcing guest to suspend at timeout.
With this patch, when we migrate the guest OS, we can specify a
timeout. If the live migration timeouts, auto-suspend the guest OS,
where the migration will complete offline.
While migration is in progress and virsh is waiting for its
completion, user may want to terminate the progress by pressing
Ctrl-C. But virsh just exits on user's Ctrl-C leaving migration
in background that user isn't even aware of. It's not reasonable.
This patch changes the behaviour for migration. For other
commands Ctrl-C still terminates virsh itself.
* src/qemu/qemu_command.c (qemuBuildChrChardevStr): Alter the
chardev alias.
(qemuBuildCommandLine): Output an id for the chardev counterpart.
* tests/qemuxml2argvdata/*: Update tests to match.
Reported by Daniel P. Berrange.
Steps to reproduce this bug:
1. service libvirtd start
2. virsh start <domain>
3. kill -STOP $(cat /var/run/libvirt/qemu/<domain>.pid)
4. service libvirtd restart
5. kill -9 $(cat /var/run/libvirt/qemu/<domain>.pid)
Then libvirtd will core dump or be in deadlock state.
Make sure that json is built into libvirt and the version
of qemu is newer than 0.13.0.
The reason of libvirtd cores dump is that:
We add vm->refs when we alloc the memory, and decrease it
in the function qemuHandleMonitorEOF() in other thread.
We add vm->refs in the function qemuConnectMonitor() and
decrease it when the vm is inactive.
The libvirtd will block in the function qemuMonitorSetCapabilities()
because the vm is stopped by signal SIGSTOP. Now the vm->refs is 2.
Then we kill the vm by signal SIGKILL. The function
qemuMonitorSetCapabilities() failed, and then we will decrease vm->refs
in the function qemuMonitorClose().
In another thread, mon->fd is broken and the function
qemuHandleMonitorEOF() is called.
If qemuHandleMonitorEOF() decreases vm->refs before qemuConnectMonitor()
returns, vm->refs will be decrease to 0 and the memory is freed.
We will call qemudShutdownVMDaemon() as qemuConnectMonitor() failed.
The memory has been freed, so qemudShutdownVMDaemon() is too dangerous.
We will reference NULL pointer in the function virDomainConfVMNWFilterTeardown():
=============
void
virDomainConfVMNWFilterTeardown(virDomainObjPtr vm) {
int i;
if (nwfilterDriver != NULL) {
for (i = 0; i < vm->def->nnets; i++)
virDomainConfNWFilterTeardown(vm->def->nets[i]);
}
}
============
vm->def->nnets is not 0 but vm->def->nets is NULL(We don't set vm->def->nnets
to 0 when we free vm).
We should add an extra reference of vm to avoid vm to be deleted if
qemuConnectMonitor() failed.
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
As noticed by Eric, commit 8e28c5d402,
which fixed generation of virtio-serial port numbers, forgot to adjust
test files which resulted in make check failure.
Regression introduced in commit e6b68d7 (Nov 2010).
Prior to that point, handlesAlloc was always a multiple of
EVENT_ALLOC_EXTENT (10), and was an int (so even if the subtraction
had been able to wrap, a negative value would be less than the count
not try to free the handles array). But after that point,
VIR_RESIZE_N made handlesAlloc grow geometrically (with a pattern of
10, 20, 30, 45 for the handles array) but still freed in multiples of
EVENT_ALLOC_EXTENT; and the count changed to size_t. Which means that
after 31 handles have been created, then 30 handles destroyed,
handlesAlloc is 5 while handlesCount is 1, and since (size_t)(1 - 5)
is indeed greater than 1, this then tried to free 10 elements, which
had the awful effect of nuking the handles array while there were
still live handles.
Nuking live handles puts libvirtd in an inconsistent state, and was
easily reproducible by starting and then stopping 60 faqemu guests.
* daemon/event.c (virEventCleanupTimeouts, virEventCleanupHandles):
Avoid integer wrap-around causing us to delete the entire array
while entries are still active.
* tests/eventtest.c (mymain): Expose the bug.
This new parameter allows user specifies where the client
cerficate, client key, CA certificate of x509 is, instead of
hardcoding it. If 'pkipath' is not specified, and the user
is not root, try to find files in $HOME/.pki/libvirt, as long
as one of client cerficate, client key, CA certificate can
not be found, use default global location (LIBVIRT_CACERT,
LIBVIRT_CLIENTCERT, LIBVIRT_CLIENTKEY, see
src/remote/remote_driver.h)
Example of use:
[root@Osier client]# virsh -c qemu+tls://10.66.93.111/system?pkipath=/tmp/pki/client
error: Cannot access CA certificate '/tmp/pki/client/cacert.pem': No such file
or directory
error: failed to connect to the hypervisor
[root@Osier client]# ls -l
total 24
-rwxrwxr-x. 1 root root 6424 Jan 24 21:35 a.out
-rw-r--r--. 1 root root 1245 Jan 23 19:04 clientcert.pem
-rw-r--r--. 1 root root 132 Jan 23 19:04 client.info
-rw-r--r--. 1 root root 1679 Jan 23 19:04 clientkey.pem
[root@Osier client]# cp /tmp/cacert.pem .
[root@Osier client]# virsh -c qemu+tls://10.66.93.111/system?pkipath=/tmp/pki/client
Welcome to virsh, the virtualization interactive terminal.
Type: 'help' for help with commands
'quit' to quit
virsh #
* src/remote/remote_driver.c: adds support for the new pkipath URI parameter
If vol->capacity is odd, the capacity will be rounded down
by devision, this patch is to round it up instead of rounding
down, to be safer in case of one writes to the volume with the
size he used to create.
- src/storage/storage_backend_logical.c: make sure size is not rounded down
If a guest image is saved in compressed format, and the restore fails
in some way after the intermediate process used to uncompress the
image has been started, but before qemu has been started to hook up to
the uncompressor, libvirt will endlessly wait for the uncompressor to
finish, but it never will because it's still waiting to have something
hooked up to drain its output.
The solution is to close the pipes on both sides of the uncompressor,
then send a SIGTERM before calling waitpid on it (only if the restore
has failed, of course).
On x86_64 hosts, /usr/lib64 must be used instead of /usr/lib
Rather than attempt to whitelist architectures, just check
for existance of /usr/lib64
* autogen.sh: Fix to use /usr/lib64 if it exists
Add a hook to the error reporting APIs to allow specific
error messages to be filtered out. Wire up libvirtd to
remove VIR_ERR_NO_DOMAIN & similar error codes from the
logs. They are still logged at DEBUG level.
* daemon/libvirtd.c: Filter VIR_ERR_NO_DOMAIN and friends
* src/libvirt_private.syms, src/util/virterror.c,
src/util/virterror_internal.h: Hook for changing error
reporting level
This reverts the additions in commit
abff683f78
taking us back to state where all errors are fully logged
in both libvirtd and normal clients.
THe intent was to stop VIR_ERR_NO_DOMAIN (No such domain
with UUID XXXX) messages from client apps polluting syslog
The change affected all error codes, but more seriously,
it also impacted errors from internal libvirtd infrastructure
For example guest autostart no longer logged errors. The
libvirtd network code no longer logged some errors. This
makes debugging incredibly hard
* daemon/libvirtd.c: Remove error log priority filter
* src/util/virterror.c, src/util/virterror_internal.h: Remove
callback for overriding log priority
This patch is a partial resolution to the following bug:
https://bugzilla.redhat.com/show_bug.cgi?id=667756
(to complete the fix, an updated selinux-policy package is required,
to add the policy that allows libvirt to set the context of a fifo,
which was previously not allowed).
Explanation : When an incoming migration is over a pipe (for example,
if the image was compressed and is being fed through gzip, or was on a
root-squash nfs server, so needed to be opened by a child process
running as a different uid), qemu cannot read it unless the selinux
context label for the pipe has been set properly.
The solution is to check the fd used as the source of the migration
just before passing it to qemu; if it's a fifo (implying that it's a
pipe), we call the newly added virSecurityManagerSetFDLabel() function
to set the context properly.
A need was found to set the SELinux context label on an open fd (a
pipe, as a matter of fact). This patch adds a function to the security
driver API that will set the label on an open fd to secdef.label. For
all drivers other than the SELinux driver, it's a NOP. For the SElinux
driver, it calls fsetfilecon().
If the return is a failure, it only returns error up to the caller if
1) the desired label is different from the existing label, 2) the
destination fd is of a type that supports setting the selinux context,
and 3) selinux is in enforcing mode. Otherwise it will return
success. This follows the pattern of the existing function
SELinuxSetFilecon().
The problem was introduced by commit 4303c91, which removed the checking
of domain state, this patch is to fix it.
Otherwise, improper error will be thrown, e.g.
error: Failed to save domain rhel6 state
error: cannot resolve symlink /var/lib/libvirt/qemu/save/rhel6.save: No such
file or directory
Running 'make check' can sometimes fail in the gnulib/tests
subdirectory, when doing an incremental build, because
./bootstrap generates a Makefile.am that tries to refer to
../../.. instead of ../.., and gets lost.
This may be an upstream gnulib bug, where a more elegant
solution will present itself in the future:
http://thread.gmane.org/gmane.comp.lib.gnulib.bugs/24898
But in the meantime, I was able to reproduce both the issue,
and this solution to work around it.
* bootstrap.conf (bootstrap_epilogue): Ensure that no stray
../../.. components remain in gnulib/tests/Makefile.in.
Reported by Serge Hallyn.
In QEMU, the card itself is a PCI device, but it requires a codec
(either -device hda-output or -device hda-duplex) to actually output
sound. Specifying <sound model='ich6'/> gives us -device intel-hda
-device hda-duplex I think it's important that a simple <sound model='ich6'/>
sets up a useful codec, to have consistent behavior with all other sound cards.
This is basically Dan's proposal of
<sound model='ich6'>
<codec type='output' slot='0'/>
<codec type='duplex' slot='3'/>
</sound>
without the codec bits implemented.
The important thing is to keep a consistent API here, we don't want some
<sound> devs require tweaking codecs but not others. Steps I see to
accomplishing this:
- every <sound> device has a <codec type='default'/> (unless codecs are
manually specified)
- <codec type='none'/> is required to specify 'no codecs'
- new audio settings like mic=on|off could then be exposed in
<sound> or <codec> in a consistent manner for all sound models
v2:
Use model='ich6'
v3:
Use feature detection, from eblake
Set codec id, bus, and cad values
v4:
intel-hda isn't supported if -device isn't available
v5:
Comment spelling fixes
This bug has been present since before the time that commit
f8a519 (Dec 2008) tried to make the dispatch loop re-entrant.
Dereferencing eventLoop.handles outside the lock risks crashing, since
any other thread could have reallocated the array in the meantime.
It's a narrow race window, however, and one that would have most
likely resulted in passing bogus data to the callback rather than
actually causing a segv, which is probably why it has gone undetected
this long.
* daemon/event.c (virEventDispatchHandles): Cache data while
inside the lock, as the array might be reallocated once outside.
If vnc_auto_unix_socket is enabled, any VNC devices without a hardcoded
listen or socket value will be setup to serve over a unix socket in
/var/lib/libvirt/qemu/$vmname.vnc.
We store the generated socket path in the transient VM definition at
CLI build time.
QEMU supports serving VNC over a unix domain socket rather than traditional
TCP host/port. This is specified with:
<graphics type='vnc' socket='/foo/bar/baz'/>
This provides better security access control than VNC listening on
127.0.0.1, but will cause issues with tools that rely on the lax security
(virt-manager in fedora runs as regular user by default, and wouldn't be
able to access a socket owned by 'qemu' or 'root').
Also not currently supported by any clients, though I have patches for
virt-manager, and virt-viewer should be simple to update.
v2:
schema: Make listen vs. socket a <choice>
This will allow us to record transient runtime state in vm->def, like
default VNC parameters. Accomplish this by adding an extra 'live' parameter
to SetDefTransient, with similar semantics to the 'live' flag for
AssignDef.
https://bugzilla.redhat.com/show_bug.cgi?id=562743
Also, fixes gnulib bug in dealing with strerror_r from glibc 2.13.
* .gnulib: Update to latest, for improved bootstrap.
* bootstrap: Resync from gnulib.
* autogen.sh (bootstrap): Add --bootstrap-sync, to make it easier
to keep bootstrap up-to-date. Pass optional --no-git through.
Reported by Aleksey Avdeev.
When restoring a saved qemu instance via JSON monitor, the vm is
left in a paused state. Turns out the 'cont' cmd was failing with
"MigrationExpected" error class and "An incoming migration is
expected before this command can be executed" error description
due to migration (restore) not yet complete.
Detect if 'cont' cmd fails with "MigrationExpecte" error class and
retry 'cont' cmd.
V2: Fix potential double-free noted by Laine Stump
Report VIR_ERR_CONFIG_UNSUPPORTED instead of VIR_ERR_INTERNAL_ERROR,
as it's valid in our domain schema, just unsupported by hypervisor
here.
* src/qemu/qemu_command.c
The code which set VNC passwords correctly had fallback for
the set_password command, but was lacking it for the
expire_password command. This made it impossible to start
a guest. It also failed to check whether QEMU was still
running after the initial 'set_password' command completed
* src/qemu/qemu_hotplug.c: Fix error handling when
password expiry fails
* src/qemu/qemu_monitor_json.c, src/qemu/qemu_monitor_text.c: Fix
return code for missing expire_password command
Avoid overwriting the real error message with a generic
OOM failure message, when machine type probe fails
* src/qemu/qemu_driver.c: Don't overwrite error
If the XML security model is NULL, it is assumed that the current
model will be used with dynamic labelling. The verify step is
meaningless and potentially crashes if dereferencing NULL
* src/security/security_manager.c: Skip NULL model on verify
The function virUnrefConnect() may call virReleaseConnect() to release
the dest connection, and the function virReleaseConnect() will call
conn->driver->close().
So the function virUnrefConnect() should be surrounded by
qemuDomainObjEnterRemoteWithDriver() and
qemuDomainObjExitRemoteWithDriver() to prevent possible deadlock between
two communicating libvirt daemons.
See commit f0c8e1cb37 for further details.
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
