When doing a live migration of a Xen guest with libxl, the images for
block devices are locked by the original QEMU process, and this prevent
the QEMU at the destination to take the lock and the migration fail.
>From QEMU point of view, once the RAM of a domain is migrated, there is
two QMP commands, "stop" then "xen-save-devices-state", at which point a
new QEMU is spawned at the destination.
Release locks in "xen-save-devices-state" so the destination can takes
them, if it's a live migration.
This patch add the "live" parameter to "xen-save-devices-state" which
default to true so older version of libxenlight can work with newer
version of QEMU.
Signed-off-by: Anthony PERARD <anthony.perard@citrix.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Route the errors from vsmtate_save_state back up through
vmstate_save and out to the normal device state path.
That's the normal error path done.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20170925112917.21340-6-dgilbert@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Modify the pre_save method on VMStateDescription to return an int
rather than void so that it potentially can fail.
Changed zillions of devices to make them return 0; the only
case I've made it return non-0 is hw/intc/s390_flic_kvm.c that already
had an error_report/return case.
Note: If you add an error exit in your pre_save you must emit
an error_report to say why.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20170925112917.21340-2-dgilbert@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
That tiny refactoring is necessary to be able to set
UFFD_FEATURE_THREAD_ID while requesting features, and then
to create downtime context in case when kernel supports it.
Signed-off-by: Alexey Perevalov <a.perevalov@samsung.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Split common postcopy staff from ram postcopy staff.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Now postcopy-able states are recognized by not NULL
save_live_complete_postcopy handler. But when we have several different
postcopy-able states, it is not convenient. Ram postcopy may be
disabled, while some other postcopy enabled, in this case Ram state
should behave as it is not postcopy-able.
This patch add separate has_postcopy handler to specify behaviour of
savevm state.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
If the bdrv_inactivate_all fails near the end of the migration,
the migration will fail and often the only diagnostics in the log
are an I/O error which you can't distinguish from an error on
the socket connection.
Add an error so we know when it's actually a block problem.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20170822170212.27347-1-dgilbert@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We need to do things at load time and at cleanup time.
Signed-off-by: Juan Quintela <quintela@redhat.com>
--
Move the printing of the error message so we can print the device
giving the error.
Add call to postcopy stuff
Message-Id: <20170628095228.4661-4-quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We need a cleanup for loads, so we rename here to be consistent.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
--
Rename htab_cleanup to htap_save_cleanup as dave suggestion
Message-Id: <20170628095228.4661-3-quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
We are going to use it now for more than save live regions.
Once there rename qemu_savevm_state_begin() to qemu_savevm_state_setup().
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20170628095228.4661-2-quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
These two parameters:
- MachineState::enforce_config_section
- MigrationState::send_configuration
are playing similar role here. This patch merges the first one into
second, then we'll have a single place to reference whether we need to
send the configuration section.
I didn't remove the MachineState.enforce_config_section field since when
applying that machine property (in machine_set_property()) we haven't
yet initialized global properties and migration object. Then, it's
still not easy to pass that boolean to MigrationState at such an early
time.
A natural benefit for current patch is that now we kept the meaning of
"enforce-config-section" since it'll still have the highest
priority (that's what "enforce" mean I guess).
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <1498536619-14548-10-git-send-email-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Move it into MigrationState, revert its meaning and renaming it to
send_section_footer, with a property bound to it. Same trick is played
like previous patches.
Removing savevm_skip_section_footers().
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <1498536619-14548-9-git-send-email-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
It was in SaveState but now moved to MigrationState altogether, reverted
its meaning, then renamed to "send_configuration". Again, using
HW_COMPAT_2_3 for old PC/SPAPR machines, and accel_register_prop() for
xen_init().
Removing savevm_skip_configuration().
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <1498536619-14548-8-git-send-email-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
One less global variable, and it does only matter with migration.
We keep the old "--only-migratable" option, but also now we support:
-global migration.only-migratable=true
Currently still keep the old interface.
Hmm, now vl.c has no way to access migrate_get_current(). Export a
function for it to setup only_migratable.
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Peter Xu <peterx@redhat.com>
Message-Id: <1498536619-14548-7-git-send-email-peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
migration_incoming_state_destroy() uses qemu_fclose() on the vmstate
file. Make sure to call it inside an AioContext acquire/release region.
This fixes an 'qemu: qemu_mutex_unlock: Operation not permitted' abort
in loadvm.
This patch closes the vmstate file before ending the drained region.
Previously we closed the vmstate file after ending the drained region.
The order does not matter.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
blk/bdrv_drain_all() only takes effect for a single instant and then
resumes block jobs, guest devices, and other external clients like the
NBD server. This can be handy when performing a synchronous drain
before terminating the program, for example.
Monitor commands usually need to quiesce I/O across an entire code
region so blk/bdrv_drain_all() is not suitable. They must use
bdrv_drain_all_begin/end() to mark the region. This prevents new I/O
requests from slipping in or worse - block jobs completing and modifying
the graph.
I audited other blk/bdrv_drain_all() callers but did not find anything
that needs a similar fix. This patch fixes the savevm/loadvm commands.
Although I haven't encountered a read world issue this makes the code
safer.
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
AioContext was designed to allow nested acquire/release calls. It uses
a recursive mutex so callers don't need to worry about nesting...or so
we thought.
BDRV_POLL_WHILE() is used to wait for block I/O requests. It releases
the AioContext temporarily around aio_poll(). This gives IOThreads a
chance to acquire the AioContext to process I/O completions.
It turns out that recursive locking and BDRV_POLL_WHILE() don't mix.
BDRV_POLL_WHILE() only releases the AioContext once, so the IOThread
will not be able to acquire the AioContext if it was acquired
multiple times.
Instead of trying to release AioContext n times in BDRV_POLL_WHILE(),
this patch simply avoids nested locking in save_vmstate(). It's the
simplest fix and we should step back to consider the big picture with
all the recent changes to block layer threading.
This patch is the final fix to solve 'savevm' hanging with -object
iothread.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Previously, dst side will immediately try to lock the write byte upon
receiving QEMU_VM_EOF, but at src side, bdrv_inactivate_all() is only
done after sending it. If the src host is under load, dst may fail to
acquire the lock due to racing with the src unlocking it.
Fix this by hoisting the bdrv_inactivate_all() operation before
QEMU_VM_EOF.
N.B. A further improvement could possibly be done to cleanly handover
locks between src and dst, so that there is no window where a third QEMU
could steal the locks and prevent src and dst from running.
N.B. This commit includes a minor improvement to the error handling
by using qemu_file_set_error().
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Fam Zheng <famz@redhat.com>
Message-id: 20170616160658.32290-1-famz@redhat.com
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
[PMM: noted qemu_file_set_error() use in commit as suggested by Daniel]
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Nothing uses it outside of migration.h
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
It don't belong anywhere else, just the global state where everybody
can stick other things.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
They are indpendent, and nowadays almost every device register things
with qdev->vmsd.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Reviewed-by: Peter Xu <peterx@redhat.com>
In load_snapshot, mis->from_src_file is freed twice, the first free is by
qemu_fclose, the second is by migration_incoming_state_destroy and
it causes Illegal instruction exception. The fix is just to remove the
first free.
This problem is found by qemu-iotests case 068 since commit
"660819b migration: shut src return path unconditionally". The error is:
068 1s ... - output mismatch (see 068.out.bad)
--- tests/qemu-iotests/068.out 2017-05-06 01:00:26.417270437 +0200
+++ 068.out.bad 2017-06-03 13:59:55.360274640 +0200
@@ -6,6 +6,8 @@
QEMU X.Y.Z monitor - type 'help' for more information
(qemu) savevm 0
(qemu) quit
+./common.config: line 107: 242472 Illegal instruction (core dumped) ( if [ -n "${QEMU_NEED_PID}" ]; then
+ echo $BASHPID > "${QEMU_TEST_DIR}/qemu-${_QEMU_HANDLE}.pid";
+fi; exec "$QEMU_PROG" $QEMU_OPTIONS "$@" )
QEMU X.Y.Z monitor - type 'help' for more information
-(qemu) quit
-*** done
+(qemu) *** done
Signed-off-by: QingFeng Hao <haoqf@linux.vnet.ibm.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
As a rule, CPU internal state should never be updated when
!cpu->kvm_vcpu_dirty (or the HAX equivalent). If that is done, then
subsequent calls to cpu_synchronize_state() - usually safe and idempotent -
will clobber state.
However, we routinely do this during a loadvm or incoming migration.
Usually this is called shortly after a reset, which will clear all the cpu
dirty flags with cpu_synchronize_all_post_reset(). Nothing is expected
to set the dirty flags again before the cpu state is loaded from the
incoming stream.
This means that it isn't safe to call cpu_synchronize_state() from a
post_load handler, which is non-obvious and potentially inconvenient.
We could cpu_synchronize_all_state() before the loadvm, but that would be
overkill since a) we expect the state to already be synchronized from the
reset and b) we expect to completely rewrite the state with a call to
cpu_synchronize_all_post_init() at the end of qemu_loadvm_state().
To clear this up, this patch introduces cpu_synchronize_pre_loadvm() and
associated helpers, which simply marks the cpu state as dirty without
actually changing anything. i.e. it says we want to discard any existing
KVM (or HAX) state and replace it with what we're going to load.
Cc: Juan Quintela <quintela@redhat.com>
Cc: Dave Gilbert <dgilbert@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Juan Quintela <quintela@redhat.com>
We can replace the four remaining calls of register_savevm() by
calls to register_savevm_live(). So we can remove the function and
as we don't allocate anymore the ops pointer with g_new0()
we don't have to free it then.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
All functions are internal except for ram_mig_init(). Create
migration/misc.h for this kind of functions.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Start removing migration code from sysemu/sysemu.h.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Split the file into public and internal interfaces. I have to rename
the external one because we can't have two include files with the same
name in the same directory. Build system gets confused. The only
exported functions are the ones that handle basic types.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Everything else assumes that we always load a device from its own
savevm handler.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
So we remove all traces of them.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
There is no reason for having the loadvm_handlers at all. There is
only one use, and we can use the savevm handlers.
We will remove the loadvm handlers on a following patch.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
--
- Added load_version_id: version_id read from the stream (laurent)
- Added load_section_id: section_id read from the stream (dave)
This removes last trace of migration functions from sysemu/sysemu.h.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
We want to track why a guest was shutdown; in particular, being able
to tell the difference between a guest request (such as ACPI request)
and host request (such as SIGINT) will prove useful to libvirt.
Since all requests eventually end up changing shutdown_requested in
vl.c, the logical change is to make that value track the reason,
rather than its current 0/1 contents.
Since command-line options control whether a reset request is turned
into a shutdown request instead, the same treatment is given to
reset_requested.
This patch adds an internal enum ShutdownCause that describes reasons
that a shutdown can be requested, and changes qemu_system_reset() to
pass the reason through, although for now nothing is actually changed
with regards to what gets reported. The enum could be exported via
QAPI at a later date, if deemed necessary, but for now, there has not
been a request to expose that much detail to end clients.
For the most part, we turn 0 into SHUTDOWN_CAUSE_NONE, and 1 into
SHUTDOWN_CAUSE_HOST_ERROR; the only specific case where we have enough
information right now to use a different value is when we are reacting
to a host signal. It will take a further patch to edit all call-sites
that can trigger a reset or shutdown request to properly pass in any
other reasons; this patch includes TODOs to point such places out.
qemu_system_reset() trades its 'bool report' parameter for a
'ShutdownCause reason', with all non-zero values having the same
effect; this lets us get rid of the weird #defines for VMRESET_*
as synonyms for bools.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20170515214114.15442-3-eblake@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
It only needed TARGET_PAGE_SIZE/BITS/BITS_MIN values, so just export
them from exec.h
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
That is the only function that we need from exec.c, and having to
include the whole sysemu.h for this.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
---
/me leans to be less sloppy with copyright notices
thanks Dave
Not used anymore after moving block migration to use capabilities.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
We have change in the previous patch to use migration capabilities for
it. Notice that we continue using the old command line flags from
migrate command from the time being. Remove the set_params method as
now it is empty.
For savevm, one can't do a:
savevm -b/-i foo
but now one can do:
migrate_set_capability block on
savevm foo
And we can't use block migration. We could disable block capability
unconditionally, but it would not be much better.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
---
- Maintain shared/enabled dependency (Xu suggestion)
- Now we maintain the dependency on the setter functions
- improve error messages
The function is only used once, and nothing else in migration knows
about objects. Create the function vmstate_device_is_migratable() in
savem.c that really do the bit that is related with migration.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
This way we use the "normal" way of printing errors for hmp commands.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Instead of manually calling blk_resume_after_migration() in migration
code after doing bdrv_invalidate_cache_all(), integrate the BlockBackend
activation with cache invalidation into a single function. This is
achieved with a new callback in BdrvChildRole that is called by
bdrv_invalidate_cache_all().
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Migration code activates all block driver nodes on the destination when
the migration completes. It does so by calling
bdrv_invalidate_cache_all() and blk_resume_after_migration(). There is
one code path for precopy and one for postcopy migration, resulting in
four function calls, which used to have three different failure modes.
This patch unifies the behaviour so that failure to activate all block
nodes is non-fatal, but the error message is logged and the VM isn't
automatically started. 'cont' will retry activating the block nodes.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
It is internal to migration, not intended for other users.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
It only uses block/* functions, nothing from migration.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
It really uses block/* stuff, not migration one.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
It is a monitor command, and has nothing migration specific in it.
Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Laurent Vivier <lvivier@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
