Commit 4454a9efc7 introduced bad
behaviour on the VIR_EVENT_HANDLE_ERROR condition. This condition
is only hit when an invalid FD is used in poll() (typically due
to a double-close bug). The QEMU monitor code was treating this
condition as non-fatal, and thus libvirt would poll() in a fast
loop forever burning 100% CPU. VIR_EVENT_HANDLE_ERROR must be
handled in the same way as VIR_EVENT_HANDLE_HANGUP, killing the
QEMU instance.
* src/qemu/qemu_monitor.c: Treat VIR_EVENT_HANDLE_ERROR as EOF
Currently the QEMU monitor I/O handler code uses errno values
to report errors. This results in a sub-optimal error messages
on certain conditions, in particular when parsing JSON strings
malformed data simply results in 'EINVAL'.
This changes the code to use the standard libvirt error reporting
APIs. The virError is stored against the qemuMonitorPtr struct,
and when a monitor API is run, any existing stored error is copied
into that thread's error local
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.c, src/qemu/qemu_monitor_text.c: Use
virError APIs for all monitor I/O handling code
Currently whenever there is any failure with parsing the monitor,
this is treated in the same was as end-of-file (ie QEMU quit).
The domain is terminated, if not already dead.
With this change, failures in parsing the monitor stream do not
result in the death of QEMU. The guest continues running unchanged,
but all further use of the monitor will be disabled.
The VMM_FAILURE event will be emitted, and the mgmt application
can decide when to kill/restart the guest to re-gain control
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h: Run a
different callback for monitor EOF vs error conditions.
* src/qemu/qemu_process.c: Emit VMM_FAILURE event when monitor
fails
Use the graphics information from the QEMU migration cookie to
issue a 'client_migrate_info' monitor command to QEMU. This causes
the SPICE client to automatically reconnect to the target host
when migration completes
* src/qemu/qemu_migration.c: Set data for SPICE client relocation
before starting migration on src
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.c, src/qemu/qemu_monitor_json.h,
src/qemu/qemu_monitor_text.c, src/qemu/qemu_monitor_text.h: Add
new qemuMonitorGraphicsRelocate() command
These VIR_XXXX0 APIs make us confused, use the non-0-suffix APIs instead.
How do these coversions works? The magic is using the gcc extension of ##.
When __VA_ARGS__ is empty, "##" will swallow the "," in "fmt," to
avoid compile error.
example: origin after CPP
high_level_api("%d", a_int) low_level_api("%d", a_int)
high_level_api("a string") low_level_api("a string")
About 400 conversions.
8 special conversions:
VIR_XXXX0("") -> VIR_XXXX("msg") (avoid empty format) 2 conversions
VIR_XXXX0(string_literal_with_%) -> VIR_XXXX(%->%%) 0 conversions
VIR_XXXX0(non_string_literal) -> VIR_XXXX("%s", non_string_literal)
(for security) 6 conversions
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
We already have virAsprintf, so picking a similar name helps for
seeing a similar purpose. Furthermore, the prefix V before printf
generally implies 'va_list', even though this variant was '...', and
the old name got in the way of adding a new va_list version.
global rename performed with:
$ git grep -l virBufferVSprintf \
| xargs -L1 sed -i 's/virBufferVSprintf/virBufferAsprintf/g'
then revert the changes in ChangeLog-old.
If qemu quited unexpectedly when we call qemuMonitorJSONHMP(),
libvirt will crash.
Steps to reproduce this bug:
1. use gdb to attach libvirtd, and set a breakpoint in the function
qemuMonitorSetCapabilities()
2. start a vm
3. let the libvirtd to run until qemuMonitorJSONSetCapabilities() returns.
4. kill the qemu process
5. continue running libvirtd
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
If the monitor met a error, and we will call qemuProcessHandleMonitorEOF().
But we may try to send monitor command after qemuProcessHandleMonitorEOF()
returned. Then libvirtd will be blocked in qemuMonitorSend().
Steps to reproduce this bug:
1. use gdb to attach libvirtd, and set a breakpoint in the function
qemuConnectMonitor()
2. start a vm
3. let the libvirtd to run until qemuMonitorOpen() returns.
4. kill the qemu process
5. continue running libvirtd
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
Add the compiler attribute to ensure we don't introduce any more
ref bugs like were just patched in commit 9741f34, then explicitly
mark the remaining places in code that are safe.
* src/qemu/qemu_monitor.h (qemuMonitorUnref): Mark
ATTRIBUTE_RETURN_CHECK.
* src/conf/domain_conf.h (virDomainObjUnref): Likewise.
* src/conf/domain_conf.c (virDomainObjParseXML)
(virDomainLoadStatus): Fix offenders.
* src/openvz/openvz_conf.c (openvzLoadDomains): Likewise.
* src/vmware/vmware_conf.c (vmwareLoadDomains): Likewise.
* src/qemu/qemu_domain.c (qemuDomainObjBeginJob)
(qemuDomainObjBeginJobWithDriver)
(qemuDomainObjExitRemoteWithDriver): Likewise.
* src/qemu/qemu_monitor.c (QEMU_MONITOR_CALLBACK): Likewise.
Suggested by Daniel P. Berrange.
A future patch will change reference counting idioms; consolidating
this pattern now makes the next patch smaller (touch only the new
macro rather than every caller).
* src/qemu/qemu_monitor.c (QEMU_MONITOR_CALLBACK): New helper.
(qemuMonitorGetDiskSecret, qemuMonitorEmitShutdown)
(qemuMonitorEmitReset, qemuMonitorEmitPowerdown)
(qemuMonitorEmitStop, qemuMonitorEmitRTCChange)
(qemuMonitorEmitWatchdog, qemuMonitorEmitIOError)
(qemuMonitorEmitGraphics): Use it to reduce duplication.
With only a single caller to these two monitor commands, I
didn't need to wrap a new WithFds version, but just change
the command itself.
* src/qemu/qemu_monitor.h (qemuMonitorAddNetdev)
(qemuMonitorAddHostNetwork): Add parameters.
* src/qemu/qemu_monitor.c (qemuMonitorAddNetdev)
(qemuMonitorAddHostNetwork): Add support for fd passing.
* src/qemu/qemu_hotplug.c (qemuDomainAttachNetDevice): Use it to
simplify code.
This is also a bug fix - on the error path, qemu_hotplug would
leave the configfd file leaked into qemu. At least the next
attempt to hotplug a PCI device would reuse the same fdname,
and when the qemu getfd monitor command gets a new fd by the
same name as an earlier one, it closes the earlier one, so there
is no risk of qemu running out of fds.
* src/qemu/qemu_monitor.h (qemuMonitorAddDeviceWithFd): New
prototype.
* src/qemu/qemu_monitor.c (qemuMonitorAddDevice): Move guts...
(qemuMonitorAddDeviceWithFd): ...to new function, and add support
for fd passing.
* src/qemu/qemu_hotplug.c (qemuDomainAttachHostPciDevice): Use it
to simplify code.
Suggested by Daniel P. Berrange.
qemu_monitor was already returning -1 and setting errno to EINVAL
on any attempt to send an fd without a unix socket, but this was
a silent failure in the case of qemuDomainAttachHostPciDevice.
Meanwhile, qemuDomainAttachNetDevice was doing some sanity checking
for a better error message; it's better to consolidate that to a
central point in the API.
* src/qemu/qemu_hotplug.c (qemuDomainAttachNetDevice): Move sanity
checking...
* src/qemu/qemu_monitor.c (qemuMonitorSendFileHandle): ...into
central location.
Suggested by Chris Wright.
Steps to reproduce this bug:
# virsh qemu-monitor-command domain 'cpu_set 2 online' --hmp
The domain has 2 cpus, and we try to set the third cpu online.
The qemu crashes, and this command will hang.
The reason is that the refs is not 1 when we unwatch the monitor.
We lock the monitor, but we do not unlock it. So virCondWait()
will be blocked.
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
POSIX states about dd:
If the bs=expr operand is specified and no conversions other than
sync, noerror, or notrunc are requested, the data returned from each
input block shall be written as a separate output block; if the read
returns less than a full block and the sync conversion is not
specified, the resulting output block shall be the same size as the
input block. If the bs=expr operand is not specified, or a conversion
other than sync, noerror, or notrunc is requested, the input shall be
processed and collected into full-sized output blocks until the end of
the input is reached.
Since we aren't using conv=sync, there is no zero-padding, but our
use of bs= means that a short read results in a short write. If
instead we use ibs= and obs=, then short reads are collected and dd
only has to do a single write, which can make dd more efficient.
* src/qemu/qemu_monitor.c (qemuMonitorMigrateToFile):
Avoid 'dd bs=', since it can cause short writes.
JSON monitor command implementation can now just directly call text
monitor implementation and it will be automatically encapsulated into
QMP's human-monitor-command.
Done mechanically with:
$ git grep -l '\bDEBUG0\? *(' | xargs -L1 sed -i 's/\bDEBUG0\? *(/VIR_&/'
followed by manual deletion of qemudDebug in daemon/libvirtd.c, along
with a single 'make syntax-check' fallout in the same file, and the
actual deletion in src/util/logging.h.
* src/util/logging.h (DEBUG, DEBUG0): Delete.
* daemon/libvirtd.h (qemudDebug): Likewise.
* global: Change remaining clients over to VIR_DEBUG counterpart.
Suspending a VM which contains shell meta characters doesn't work with
libvirt-0.8.7:
/var/log/libvirt/qemu/andreas_231-ne\ doch\ nicht.log:
sh: -c: line 0: syntax error near unexpected token `doch'
sh: -c: line 0: `cat | { dd bs=4096 seek=1 if=/dev/null && dd bs=1048576; }
Although target="andreas_231-ne doch nicht" contains shell meta
characters (here: blanks), they are not properly escaped by
src/qemu/qemu_monitor_{json,text}.c#qemuMonitor{JSON,Text}MigrateToFile()
First, the filename needs to be properly escaped for the shell, than
this command line has to be properly escaped for qemu again.
For this to work, remove the old qemuMonitorEscapeArg() wrapper, rename
qemuMonitorEscape() to it removing the handling for shell=TRUE, and
implement a new qemuMonitorEscapeShell() returning strings using single
quotes.
Using double quotes or escaping special shell characters with backslashes
would also be possible, but the set of special characters heavily
depends on the concrete shell (dsh, bash, zsh) and its setting (history
expansion, interactive use, ...)
Signed-off-by: Philipp Hahn <hahn@univention.de>
Currently users who want to use virDomainQemuMonitorCommand() API or
it's virsh equivalent has to use the same protocol as libvirt uses for
communication to qemu. Since the protocol is QMP with current qemu and
HMP much more usable for humans, one ends up typing something like the
following:
virsh qemu-monitor-command DOM \
'{"execute":"human-monitor-command","arguments":{"command-line":"info kvm"}}'
which is not a very convenient way of debugging qemu.
This patch introduces --hmp option to qemu-monitor-command, which says
that the provided command is in HMP. If libvirt uses QMP to talk with
qemu, the command will automatically be converted into QMP. So the
example above is simplified to just
virsh qemu-monitor-command --hmp DOM "info kvm"
Also the result is converted from
{"return":"kvm support: enabled\r\n"}
to just plain HMP:
kvm support: enabled
If libvirt talks to qemu in HMP, --hmp flag is obviously a noop.
Currently libvirt doesn't confirm whether the guest has responded to the
disk removal request. In some cases this can leave the guest with
continued access to the device while the mgmt layer believes that it has
been removed. With a recent qemu monitor command[1] we can
deterministically revoke a guests access to the disk (on the QEMU side)
to ensure no futher access is permitted.
This patch adds support for the drive_del() command and introduces it
in the disk removal paths. If the guest is running in a QEMU without this
command we currently explicitly check for unknown command/CommandNotFound
and log the issue.
If QEMU supports the command we issue the drive_del command after we attempt
to remove the device. The guest may respond and remove the block device
before we get to attempt to call drive_del. In that case, we explicitly check
for 'Device not found' from the monitor indicating that the target drive
was auto-deleted upon guest responds to the device removal notification.
1. http://thread.gmane.org/gmane.comp.emulators.qemu/84745
Signed-off-by: Ryan Harper <ryanh@us.ibm.com>
Currently libvirt doesn't confirm whether the guest has responded to the
disk removal request. In some cases this can leave the guest with
continued access to the device while the mgmt layer believes that it has
been removed. With a recent qemu monitor command[1] we can
deterministically revoke a guests access to the disk (on the QEMU side)
to ensure no futher access is permitted.
This patch adds support for the drive_unplug() command and introduces it
in the disk removal paths. There is some discussion to be had about how
to handle the case where the guest is running in a QEMU without this
command (and the fact that we currently don't have a way of detecting
what monitor commands are available).
Changes since v2:
- use VIR_ERROR to report when unplug command not found
Changes since v1:
- return > 0 when command isn't present, < 0 on command failure
- detect when drive_unplug command isn't present and log error
instead of failing entire command
Signed-off-by: Ryan Harper <ryanh@us.ibm.com>
I am replacing the last instances of close() I found with VIR_CLOSE() / VIR_FORCE_CLOSE respectively.
The first part patches virsh, which I missed out on previously.
The 2nd patch I had left out intentionally to look at it more carefully:
The 'closed' variable could be easily removed since it wasn't used anywhere else. The possible race condition that could result from the filedescriptor being closed and not set to -1 (and possibly let us write into 'something' totally different if the fd was allocated by another thread) seems to be prevented by the qemuMonitorLock() already placed around the code that reads from or writes to the fd. So the change of this code as shown in the patch should not have any side-effects.
QEMU allows forcing a CDROM eject even if the guest has locked the device.
Expose this via a new UpdateDevice flag, VIR_DOMAIN_DEVICE_MODIFY_FORCE.
This has been requested for RHEV:
https://bugzilla.redhat.com/show_bug.cgi?id=626305
v2: Change flag name, bool cleanups
Using automated replacement with sed and editing I have now replaced all
occurrences of close() with VIR_(FORCE_)CLOSE() except for one, of
course. Some replacements were straight forward, others I needed to pay
attention. I hope I payed attention in all the right places... Please
have a look. This should have at least solved one more double-close
error.
Implement the qemu driver's virDomainQemuMonitorCommand
and hook it into the API entry point.
Changes since v1:
- Rename the (external) qemuMonitorCommand to qemuDomainMonitorCommand
- Add virCheckFlags to qemuDomainMonitorCommand
Changes since v2:
- Drop ATTRIBUTE_UNUSED from the flags
Changes since v3:
- Add a flag to priv so we only print out monitor command warning once. Note
that this has not been plumbed into qemuDomainObjPrivateXMLFormat or
qemuDomainObjPrivateXMLParse, which means that if you run a monitor command,
restart libvirtd, and then run another monitor command, you may get an
an erroneous VIR_INFO. It's a pretty minor matter, and I didn't think it
warranted the additional code.
- Add BeginJob/EndJob calls around EnterMonitor/ExitMonitor
Signed-off-by: Chris Lalancette <clalance@redhat.com>
Some, but not all, codepaths in the qemuMonitorOpen() method
would trigger the destroy callback. The caller does not expect
this to be invoked if construction fails, only during normal
release of the monitor. This resulted in a possible double-unref
of the virDomainObjPtr, because the caller explicitly unrefs
the virDomainObjPtr if qemuMonitorOpen() fails
* src/qemu/qemu_monitor.c: Don't invoke destroy callback from
qemuMonitorOpen() failure paths
The patches for shared storage migration were not correctly written
for json mode. Thus the 'blk' and 'inc' parameters were never being
set. In addition they didn't set the QEMU_MONITOR_MIGRATE_BACKGROUND
so migration was synchronous. Due to multiple bugs in QEMU's JSON
impl this wasn't noticed because it treated the sync migration requst
as asynchronous anyway. Finally 'background' parameter was converted
to take arbitrary flags but not renamed, and not all uses were changed
to unsigned int.
* src/qemu/qemu_driver.c: Set QEMU_MONITOR_MIGRATE_BACKGROUND in
doNativeMigrate
* src/qemu/qemu_monitor_json.c: Process QEMU_MONITOR_MIGRATE_NON_SHARED_DISK
and QEMU_MONITOR_MIGRATE_NON_SHARED_INC flags
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.h, src/qemu/qemu_monitor_text.c,
src/qemu/qemu_monitor_text.h: change 'int background' to
'unsigned int flags' in migration APIs. Add logging of flags
parameter
The current code pattern requires that callers of qemuMonitorClose
check for the return value == 0, and if so, set priv->mon = NULL
and release the reference held on the associated virDomainObjPtr
The change d84bb6d6a3 violated that
requirement, meaning that priv->mon never gets set to NULL, and
a reference count is leaked on virDomainObjPtr.
This design was a bad one, so remove the need to check the return
valueof qemuMonitorClose(). Instead allow registration of a
callback that's invoked just when the last reference on qemuMonitorPtr
is released.
Finally there was a potential reference leak in qemuConnectMonitor
in the failure path.
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h: Add a destroy
callback invoked from qemuMonitorFree
* src/qemu/qemu_driver.c: Use the destroy callback to release the
reference on virDomainObjPtr when the monitor is freed. Fix other
potential reference count leak in connecting to monitor
The virDomainGetBlockInfo API allows query physical block
extent and allocated block extent. These are normally the
same value unless storing a special format like qcow2
inside a block device. In this scenario we can query QEMU
to get the actual allocated extent.
Since last time:
- Return fatal error in text monitor
- Only invoke monitor command for block devices
- Fix error handling JSON code
* src/qemu/qemu_driver.c: Fill in block aloction extent when VM
is running
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.c, src/qemu/qemu_monitor_json.h,
src/qemu/qemu_monitor_text.c, src/qemu/qemu_monitor_text.h: Add
API to query the highest block extent via info blockstats
History has shown that there are frequent bugs in the QEMU driver
code leading to the monitor being invoked with a NULL pointer.
Although the QEMU driver code should always report an error in
this case before invoking the monitor, as a safety net put in a
generic check in the monitor code entry points.
* src/qemu/qemu_monitor.c: Safety net to check for NULL monitor
object
QEMU is gaining a new monitor command netdev_add for hotplugging
NICs using the netdev backend code. We already support this on
the command this, though it is disabled. This adds support for
hotplug too, also to remain disabled until 0.13 QEMU is released
* src/qemu/qemu_driver.c: Support netdev hotplug for NICs
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.c, src/qemu/qemu_monitor_json.h,
src/qemu/qemu_monitor_text.c, src/qemu/qemu_monitor_text.h: Add
support for netdev_add and netdev_remove commands
When closing a monitor using qemuMonitorClose(), we are aware of
the possibility the monitor is still being used somewhere:
/* NB: ordinarily one might immediately set mon->watch to -1
* and mon->fd to -1, but there may be a callback active
* that is still relying on these fields being valid. So
* we merely close them, but not clear their values and
* use this explicit 'closed' flag to track this state */
but since we call virEventAddHandle() on that monitor without increasing
its ref counter, the monitor is still freed which makes possible users
of it quite unhappy. The unhappiness can lead to a hang if qemuMonitorIO
tries to lock mutex which no longer exists.
Support for live migration between hosts that do not share storage was
added to qemu-kvm release 0.12.1.
It supports two flags:
-b migration without shared storage with full disk copy
-i migration without shared storage with incremental copy (same base image
shared between source and destination).
I tested the live migration without shared storage (both flags) for native
and p2p with and without tunnelling. I also verified that the fix doesn't
affect normal migration with shared storage.
This introduces a new event type
VIR_DOMAIN_EVENT_ID_IO_ERROR_REASON
This event is the same as the previous VIR_DOMAIN_ID_IO_ERROR
event, but also includes a string describing the cause of
the event.
Thus there is a new callback definition for this event type
typedef void (*virConnectDomainEventIOErrorReasonCallback)(virConnectPtr conn,
virDomainPtr dom,
const char *srcPath,
const char *devAlias,
int action,
const char *reason,
void *opaque);
This is currently wired up to the QEMU block IO error events
* daemon/remote.c: Dispatch IO error events to client
* examples/domain-events/events-c/event-test.c: Watch for
IO error events
* include/libvirt/libvirt.h.in: Define new IO error event ID
and callback signature
* src/conf/domain_event.c, src/conf/domain_event.h,
src/libvirt_private.syms: Extend API to handle IO error events
* src/qemu/qemu_driver.c: Connect to the QEMU monitor event
for block IO errors and emit a libvirt IO error event
* src/remote/remote_driver.c: Receive and dispatch IO error
events to application
* src/remote/remote_protocol.x: Wire protocol definition for
IO error events
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.c: Watch for BLOCK_IO_ERROR event
from QEMU monitor
The save process was relying on use of the shell >> append
operator to ensure the save data was placed after the libvirt
header + XML. This doesn't work for block devices though.
Replace this code with use of 'dd' and its 'seek' parameter.
This means that we need to pad the header + XML out to a
multiple of dd block size (in this case we choose 512).
The qemuMonitorMigateToCommand() monitor API is used for both
save/coredump, and migration via UNIX socket. We can't simply
switch this to use 'dd' since this causes problems with the
migration usage. Thus, create a dedicated qemuMonitorMigateToFile
which can accept an filename + offset, and remove the filename
from the current qemuMonitorMigateToCommand() API
* src/qemu/qemu_driver.c: Switch to qemuMonitorMigateToFile
for save and core dump
* src/qemu/qemu_monitor.c, src/qemu/qemu_monitor.h,
src/qemu/qemu_monitor_json.c, src/qemu/qemu_monitor_json.h,
src/qemu/qemu_monitor_text.c, src/qemu/qemu_monitor_text.h: Create
a new qemuMonitorMigateToFile, separate from the existing
qemuMonitorMigateToCommand to allow handling file offsets
