VMSTATE_WITH_TMP is for handling structures where some calculation
or rearrangement of the data needs to be performed before the data
hits the wire.
For example, where the value on the wire is an offset from a
non-migrated base, but the data in the structure is the actual pointer.
To use it, a temporary type is created and a vmsd used on that type.
The first element of the type must be 'parent' a pointer back to the
type of the main structure. VMSTATE_WITH_TMP takes care of allocating
and freeing the temporary before running the child vmsd.
The post_load/pre_save on the child vmsd can copy things from the parent
to the temporary using the parent pointer and do any other calculations
needed; it can then use normal VMSD entries to do the actual data
storage without having to fiddle around with qemu_get_*/qemu_put_*
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20170203160651.19917-3-dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
If the net connection between primary host and secondary host breaks
while COLO/COLO incoming threads are doing read() or write().
It will block until connection is timeout, and the failover process
will be blocked because of it.
So it is necessary to shutdown all the socket fds used by COLO
to avoid this situation. Besides, we should close the corresponding
file descriptors after failvoer BH shutdown them,
Or there will be an error.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <1484657864-21708-3-git-send-email-zhang.zhanghailiang@huawei.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
If we set checkpoint-delay through command 'migrate-set-parameters',
It will not take effect until we finish last sleep chekpoint-delay,
That's will be offensive espeically when we want to change its value
from an extreme big one to a proper value.
Fix it by using timer to realize checkpoint-delay.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Message-Id: <1484657864-21708-2-git-send-email-zhang.zhanghailiang@huawei.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
The member VMStateField.start is used for two things, partial data
migration for VBUFFER data (basically provide migration for a
sub-buffer) and for locating next in QTAILQ.
The implementation of the VBUFFER feature is broken when VMSTATE_ALLOC
is used. This however goes unnoticed because actually partial migration
for VBUFFER is not used at all.
Let's consolidate the usage of VMStateField.start by removing support
for partial migration for VBUFFER.
Signed-off-by: Halil Pasic <pasic@linux.vnet.ibm.com>
Message-Id: <20170203175217.45562-1-pasic@linux.vnet.ibm.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
After the start of postcopy migration there are some non-dirty pages which have
already been migrated. These pages are no longer needed on the source vm so that
we can free them and it doen't hurt to complete the migration.
Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
Message-Id: <20170203152321.19739-4-pbutsykin@virtuozzo.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
This feature frees the migrated memory on the source during postcopy-ram
migration. In the second step of postcopy-ram migration when the source vm
is put on pause we can free unnecessary memory. It will allow, in particular,
to start relaxing the memory stress on the source host in a load-balancing
scenario.
Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com>
Message-Id: <20170203152321.19739-3-pbutsykin@virtuozzo.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Manually merged in Pavel's 'migration: madvise error_report fixup!'
An early postcopy failure can be recovered from as long as we know
we haven't sent the command to run the destination.
We have to undo the bdrv_inactivate_all by calling
bdrv_invalidate_cache_all
Note that I'm not using ms->block_inactive because once we've
sent the postcopy package we dont want anything else to try
and recover the block storage on the source; the destination
might have started writing to it.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20170202155909.31784-3-dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
On a destination host with no userfault support an incoming
postcopy would cause the state to enter ADVISE before
it realised there was no support, and because it was in ADVISE
state it would perform a cleanup at the end. Since there
was no support the cleanup function should be unreachable,
but ends up being called and asserting.
Reset the state when we realise we have no support, thus the
cleanup doesn't happen.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20170202155909.31784-2-dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
The qdev id of a device can be huge if it's on the end of a chain
of bridges; in reality such chains shouldn't occur but they can
be made to by chaining PCIe bridges together.
The migration format has a number of 256 character long format
limits; check we don't hit them (we already use pstrcat/cpy but
that just protects us from buffer overruns, we fairly quickly
hit an assert).
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Message-Id: <20170202125956.21942-3-dgilbert@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
There are a number of unused trace events that
scripts/cleanup-trace-events.pl finds. The "hw/vfio/pci-quirks.c"
filename was typoed and "qapi/qapi-visit-core.c" was missing the qapi/
directory prefix.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-id: 20170126171613.1399-3-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch introduces save_vmstate function to allow saving and loading
vmstates from the replay module.
Signed-off-by: Pavel Dovgalyuk <pavel.dovgaluk@ispras.ru>
Message-Id: <20170124071741.4572.13714.stgit@PASHA-ISP>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Add some tracing to vmstate_subsection_save and vmstate_save_state
to help in debugging when you're not sure if a conditional piece
of data is being saved.
In vmstate_subsection_save I renamed the inner vmsd to avoid the aliasing
and be able to print both names.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Message-Id: <20161212125838.14425-1-dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
commit fe904ea824 fixed a case
which migration aborted QEMU because it didn't regain the control
of images while some errors happened.
Actually, there are another two cases can trigger the same error reports:
" bdrv_co_do_pwritev: Assertion `!(bs->open_flags & 0x0800)' failed",
Case 1, codes path:
migration_thread()
migration_completion()
bdrv_inactivate_all() ----------------> inactivate images
qemu_savevm_state_complete_precopy()
socket_writev_buffer() --------> error because destination fails
qemu_fflush() ----------------> set error on migration stream
-> qmp_migrate_cancel() ----------------> user cancelled migration concurrently
-> migrate_set_state() ------------------> set migrate CANCELLIN
migration_completion() -----------------> go on to fail_invalidate
if (s->state == MIGRATION_STATUS_ACTIVE) -> Jump this branch
Case 2, codes path:
migration_thread()
migration_completion()
bdrv_inactivate_all() ----------------> inactivate images
migreation_completion() finished
-> qmp_migrate_cancel() ---------------> user cancelled migration concurrently
qemu_mutex_lock_iothread();
qemu_bh_schedule (s->cleanup_bh);
As we can see from above, qmp_migrate_cancel can slip in whenever
migration_thread does not hold the global lock. If this happens after
bdrv_inactive_all() been called, the above error reports will appear.
To prevent this, we can call bdrv_invalidate_cache_all() in qmp_migrate_cancel()
directly if we find images become inactive.
Besides, bdrv_invalidate_cache_all() in migration_completion() doesn't have the
protection of big lock, fix it by add the missing qemu_mutex_lock_iothread();
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Message-Id: <1485244792-11248-1-git-send-email-zhang.zhanghailiang@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Currently we cannot directly transfer a QTAILQ instance because of the
limitation in the migration code. Here we introduce an approach to
transfer such structures. We created VMStateInfo vmstate_info_qtailq
for QTAILQ. Similar VMStateInfo can be created for other data structures
such as list.
When a QTAILQ is migrated from source to target, it is appended to the
corresponding QTAILQ structure, which is assumed to have been properly
initialized.
This approach will be used to transfer pending_events and ccs_list in spapr
state.
We also create some macros in qemu/queue.h to access a QTAILQ using pointer
arithmetic. This ensures that we do not depend on the implementation
details about QTAILQ in the migration code.
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Jianjun Duan <duanj@linux.vnet.ibm.com>
Message-Id: <1484852453-12728-3-git-send-email-duanj@linux.vnet.ibm.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Currently the QIOTaskFunc signature takes an Object * for
the source, and an Error * for any error. We also need to
be able to provide a result pointer. Rather than continue
to add parameters to QIOTaskFunc, remove the existing
ones and simply pass the QIOTask object instead. This
has methods to access all the other data items required
in the callback impl.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrange <berrange@redhat.com>
During migration, save state entries are saved/loaded without a specific
order - we just traverse the savevm_state.handlers list and do it one by
one. This might not be enough.
There are requirements that we need to load specific device's vmstate
first before others. For example, VT-d IOMMU contains DMA address
remapping information, which is required by all the PCI devices to do
address translations. We need to make sure IOMMU's device state is
loaded before the rest of the PCI devices, so that DMA address
translation can work properly.
This patch provide a VMStateDescription.priority value to allow specify
the priority of the saved states. The loadvm operation will be done with
those devices with higher vmsd priority.
Before this patch, we are possibly achieving the ordering requirement by
an assumption that the ordering will be the same with the ordering that
objects are created. A better way is to mark it out explicitly in the
VMStateDescription table, like what this patch does.
Current ordering logic is still naive and slow, but after all that's not
a critical path so IMO it's a workable solution for now.
Signed-off-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
qemu_savevm_state_iterate() expects the iterators to return 1
when they are done, and 0 if there is still something left to do.
However, ram_save_iterate() does not obey this rule and returns
the number of saved pages instead. This causes a fatal hang with
ppc64 guests when you run QEMU like this (also works with TCG):
qemu-img create -f qcow2 /tmp/test.qcow2 1M
qemu-system-ppc64 -nographic -nodefaults -m 256 \
-hda /tmp/test.qcow2 -serial mon:stdio
... then switch to the monitor by pressing CTRL-a c and try to
save a snapshot with "savevm test1" for example.
After the first iteration, ram_save_iterate() always returns 0 here,
so that qemu_savevm_state_iterate() hangs in an endless loop and you
can only "kill -9" the QEMU process.
Fix it by using proper return values in ram_save_iterate().
Signed-off-by: Thomas Huth <thuth@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
We forgot to assign true to params->has_x_checkpoint_delay parameter
in qmp_query_migrate_parameters.
Without this, qmp command 'query-migrate-parameters' doesn't show the
default value for x-checkpoint-delay option.
This also fixes the fact that HMP was relying on unspecified behavior by
reading x_checkpoint_delay without checking has_x_checkpoint_delay.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Migration bits from the COLO project
# gpg: Signature made Sun 30 Oct 2016 10:39:55 GMT
# gpg: using RSA key 0xEB0B4DFC657EF670
# gpg: Good signature from "Amit Shah <amit@amitshah.net>"
# gpg: aka "Amit Shah <amit@kernel.org>"
# gpg: aka "Amit Shah <amitshah@gmx.net>"
# Primary key fingerprint: 48CA 3722 5FE7 F4A8 B337 2735 1E9A 3B5F 8540 83B6
# Subkey fingerprint: CC63 D332 AB8F 4617 4529 6534 EB0B 4DFC 657E F670
* remotes/amit-migration/tags/migration-for-2.8:
MAINTAINERS: Add maintainer for COLO framework related files
configure: Support enable/disable COLO feature
docs: Add documentation for COLO feature
COLO: Implement failover work for secondary VM
COLO: Implement the process of failover for primary VM
COLO: Introduce state to record failover process
COLO: Add 'x-colo-lost-heartbeat' command to trigger failover
COLO: Synchronize PVM's state to SVM periodically
COLO: Add checkpoint-delay parameter for migrate-set-parameters
COLO: Load VMState into QIOChannelBuffer before restore it
COLO: Send PVM state to secondary side when do checkpoint
COLO: Add a new RunState RUN_STATE_COLO
COLO: Introduce checkpointing protocol
COLO: Establish a new communicating path for COLO
migration: Switch to COLO process after finishing loadvm
migration: Enter into COLO mode after migration if COLO is enabled
COLO: migrate COLO related info to secondary node
migration: Introduce capability 'x-colo' to migration
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
configure --enable-colo/--disable-colo to switch COLO
support on/off.
COLO feature doesn't depend on any other external libraries,
So here it is reasonable to enable COLO by default, to
avoid re-compile QEMU if users want to use this capability.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
For primary side, if COLO gets failover request from users.
To be exact, gets 'x_colo_lost_heartbeat' command.
COLO thread will exit the loop while the failover BH does the
cleanup work and resumes VM.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
When handling failover, COLO processes differently according to
the different stage of failover process, here we introduce a global
atomic variable to record the status of failover.
We add four failover status to indicate the different stage of failover process.
You should use the helpers to get and set the value.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
We leave users to choose whatever heartbeat solution they want,
if the heartbeat is lost, or other errors they detect, they can use
experimental command 'x_colo_lost_heartbeat' to tell COLO to do failover,
COLO will do operations accordingly.
For example, if the command is sent to the Primary side,
the Primary side will exit COLO mode, does cleanup work,
and then, PVM will take over the service work. If sent to the Secondary side,
the Secondary side will run failover work, then takes over PVM's service work.
Cc: Luiz Capitulino <lcapitulino@redhat.com>
Cc: Eric Blake <eblake@redhat.com>
Cc: Markus Armbruster <armbru@redhat.com>
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
We should not destroy the state of SVM (Secondary VM) until we receive
the complete data of PVM's state, in case the primary fails in the process
of sending the state, so we cache the VM's state in secondary side before
load it into SVM.
Besides, we should call qemu_system_reset() before load VM state,
which can ensure the data is intact.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
VM checkpointing is to synchronize the state of PVM to SVM, just
like migration does, we re-use save helpers to achieve migrating
PVM's state to Secondary side.
COLO need to cache the data of VM's state in the secondary side before
synchronize it to SVM. COLO need the size of the data to determine
how much data should be read in the secondary side.
So here, we can get the size of the data by saving it into I/O channel
before send it to the secondary side.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
We need communications protocol of user-defined to control
the checkpointing process.
The new checkpointing request is started by Primary VM,
and the interactive process like below:
Checkpoint synchronizing points:
Primary Secondary
initial work
'checkpoint-ready' <-------------------- @
'checkpoint-request' @ -------------------->
Suspend (Only in hybrid mode)
'checkpoint-reply' <-------------------- @
Suspend&Save state
'vmstate-send' @ -------------------->
Send state Receive state
'vmstate-received' <-------------------- @
Release packets Load state
'vmstate-load' <-------------------- @
Resume Resume (Only in hybrid mode)
Start Comparing (Only in hybrid mode)
NOTE:
1) '@' who sends the message
2) Every sync-point is synchronized by two sides with only
one handshake(single direction) for low-latency.
If more strict synchronization is required, a opposite direction
sync-point should be added.
3) Since sync-points are single direction, the remote side may
go forward a lot when this side just receives the sync-point.
4) For now, we only support 'periodic' checkpoint, for which
the Secondary VM is not running, later we will support 'hybrid' mode.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Cc: Eric Blake <eblake@redhat.com>
Cc: Markus Armbruster <armbru@redhat.com>
Cc: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
Switch from normal migration loadvm process into COLO checkpoint process if
COLO mode is enabled.
We add three new members to struct MigrationIncomingState,
'have_colo_incoming_thread' and 'colo_incoming_thread' record the COLO
related thread for secondary VM, 'migration_incoming_co' records the
original migration incoming coroutine.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
Add a new migration state: MIGRATION_STATUS_COLO. Migration source side
enters this state after the first live migration successfully finished
if COLO is enabled by command 'migrate_set_capability x-colo on'.
We reuse migration thread, so the process of checkpointing will be handled
in migration thread.
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
We can determine whether or not VM in destination should go into COLO mode
by referring to the info that was migrated.
We skip this section if COLO is not enabled (i.e.
migrate_set_capability colo off), so that, It doesn't break compatibility
with migration no matter whether users configure the --enable-colo/disable-colo
on the source/destination side or not;
Signed-off-by: zhanghailiang <zhang.zhanghailiang@huawei.com>
Signed-off-by: Li Zhijian <lizhijian@cn.fujitsu.com>
Signed-off-by: Gonglei <arei.gonglei@huawei.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
Signed-off-by: Amit Shah <amit@amitshah.net>
All the callers of migrate_fd_error() pass a non-NULL
error parameter, and if any did pass NULL then we would
segfault in error_copy(), so remove the unnecessary
NULL check earlier in the function.
(Spotted by Coverity.)
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>
Merge qio 2016/10/27 v1
# gpg: Signature made Thu 27 Oct 2016 13:54:03 BST
# gpg: using RSA key 0xBE86EBB415104FDF
# gpg: Good signature from "Daniel P. Berrange <dan@berrange.com>"
# gpg: aka "Daniel P. Berrange <berrange@redhat.com>"
# Primary key fingerprint: DAF3 A6FD B26B 6291 2D0E 8E3F BE86 EBB4 1510 4FDF
* remotes/berrange/tags/pull-qio-2016-10-27-1:
main: set names for main loop sources created
vnc: set name for all I/O channels created
migration: set name for all I/O channels created
char: set name for all I/O channels created
nbd: set name for all I/O channels created
io: add ability to set a name for IO channels
io: Add a QIOChannelSocket cleanup test
io: set LISTEN flag explicitly for listen sockets
io: Introduce a qio_channel_set_feature() helper
io: Use qio_channel_has_feature() where applicable
io: Fix double shift usages on QIOChannel features
Conflicts:
qemu-char.c
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
