Fix error handling during bitmap postcopy
PATCH v2 22/22 qemu-iotests/199: add source-killed case to bitmaps postcopy
In reply to Vladimir Sementsov-OgievskiyPrevious patches fixes behavior of bitmaps migration, so that errors are handled by just removing unfinished bitmaps, and not fail or try to recover postcopy migration. Add corresponding test.
--- tests/qemu-iotests/199 | 15 +++++++++++++++ tests/qemu-iotests/199.out | 4 ++-- 2 files changed, 17 insertions(+), 2 deletions(-)
tests/qemu-iotests/199 +15 -0
@@ -235,6 +235,21 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.launch() check_bitmaps(self.vm_a, 0) + def test_early_kill_source(self): + self.start_postcopy() + + self.vm_a_events = self.vm_a.get_qmp_events() + self.vm_a.kill() + + self.vm_a.launch() + + match = {'data': {'status': 'completed'}} + e_complete = self.vm_b.event_wait('MIGRATION', match=match) + self.vm_b_events.append(e_complete) + + check_bitmaps(self.vm_a, 0) + check_bitmaps(self.vm_b, 0) + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2'])
tests/qemu-iotests/199.out +2 -2
@@ -1,5 +1,5 @@ -.. +... ---------------------------------------------------------------------- -Ran 2 tests +Ran 3 tests OK
Signature · 1 line
2.21.0
PATCH v2 20/22 qemu-iotests/199: check persistent bitmaps
In reply to Vladimir Sementsov-OgievskiyCheck that persistent bitmaps are not stored on source and that bitmaps are persistent on destination.
--- tests/qemu-iotests/199 | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-)
tests/qemu-iotests/199 +15 -1
@@ -117,7 +117,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): for i in range(nb_bitmaps): result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap{}'.format(i), - granularity=granularity) + granularity=granularity, + persistent=True) self.assert_qmp(result, 'return', {}) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256',
@@ -193,6 +194,19 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): print('downtime:', downtime) print('postcopy_time:', postcopy_time) + # check that there are no bitmaps stored on source + self.vm_a_events += self.vm_a.get_qmp_events() + self.vm_a.shutdown() + self.vm_a.launch() + check_bitmaps(self.vm_a, 0) + + # check that bitmaps are migrated and persistence works + check_bitmaps(self.vm_b, nb_bitmaps) + self.vm_b.shutdown() + # recreate vm_b, so there is no incoming option, which prevents + # loading bitmaps from disk + self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b) + self.vm_b.launch() check_bitmaps(self.vm_b, nb_bitmaps) # Check content of migrated bitmaps. Still, don't waste time checking
Signature · 1 line
2.21.0
PATCH v2 21/22 qemu-iotests/199: add early shutdown case to bitmaps postcopy
In reply to Vladimir Sementsov-OgievskiyPrevious patches fixed two crashes which may occur on shutdown prior to bitmaps postcopy finished. Check that it works now.
--- tests/qemu-iotests/199 | 18 ++++++++++++++++++ tests/qemu-iotests/199.out | 4 ++-- 2 files changed, 20 insertions(+), 2 deletions(-)
tests/qemu-iotests/199 +18 -0
@@ -217,6 +217,24 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): sha = self.discards1_sha256 if i % 2 else self.all_discards_sha256 self.assert_qmp(result, 'return/sha256', sha) + def test_early_shutdown_destination(self): + self.start_postcopy() + + self.vm_b_events += self.vm_b.get_qmp_events() + self.vm_b.shutdown() + # recreate vm_b, so there is no incoming option, which prevents + # loading bitmaps from disk + self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b) + self.vm_b.launch() + check_bitmaps(self.vm_b, 0) + + result = self.vm_a.qmp('query-status') + assert not result['return']['running'] + self.vm_a_events += self.vm_a.get_qmp_events() + self.vm_a.shutdown() + self.vm_a.launch() + check_bitmaps(self.vm_a, 0) + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2'])
tests/qemu-iotests/199.out +2 -2
@@ -1,5 +1,5 @@ -. +.. ---------------------------------------------------------------------- -Ran 1 tests +Ran 2 tests OK
Signature · 1 line
2.21.0
PATCH v2 5/22 migration/block-dirty-bitmap: refactor state global variables
In reply to Vladimir Sementsov-OgievskiyMove all state variables into one global struct. Reduce global variable usage, utilizing opaque pointer where possible.
--- migration/block-dirty-bitmap.c | 171 ++++++++++++++++++--------------- 1 file changed, 95 insertions(+), 76 deletions(-)
migration/block-dirty-bitmap.c +95 -76
@@ -128,6 +128,12 @@ typedef struct DBMSaveState {BdrvDirtyBitmap *prev_bitmap; } DBMSaveState; +typedef struct LoadBitmapState { +BlockDriverState *bs; +BdrvDirtyBitmap *bitmap; +bool migrated; +}LoadBitmapState ; +/* State of the dirty bitmap migration (DBM) during load process */ typedef struct DBMLoadState { uint32_t flags;
@@ -135,18 +141,17 @@ typedef struct DBMLoadState {char bitmap_name[256 ];BlockDriverState *bs;BdrvDirtyBitmap *bitmap; + +GSList *enabled_bitmaps; +QemuMutex finish_lock; } DBMLoadState; -static DBMSaveState dirty_bitmap_mig_state; +typedef struct DBMState { +DBMSaveState save; +DBMLoadState load; +}DBMState ; -/* State of one bitmap during load process */ -typedef struct LoadBitmapState { -BlockDriverState *bs; -BdrvDirtyBitmap *bitmap; -bool migrated; -}LoadBitmapState ; -static GSList *enabled_bitmaps; -QemuMutex finish_lock; +static DBMState dbm_state;static uint32_t qemu_get_bitmap_flags (QEMUFile *f) {
@@ -169,21 +174,21 @@ static void qemu_put_bitmap_flags(QEMUFile *f, uint32_t flags)qemu_put_byte (f, flags); } -static void send_bitmap_header (QEMUFile *f,SaveBitmapState *dbms, -uint32_t additional_flags) +static void send_bitmap_header (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms,uint32_t additional_flags) {BlockDriverState *bs = dbms->bs;BdrvDirtyBitmap *bitmap = dbms->bitmap;uint32_t flags = additional_flags;trace_send_bitmap_header_enter (); -if (bs != dirty_bitmap_mig_state.prev_bs) { - dirty_bitmap_mig_state.prev_bs = bs; +if (bs != s->prev_bs) { + s->prev_bs = bs; flags |=DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME ; } -if (bitmap != dirty_bitmap_mig_state.prev_bitmap) { - dirty_bitmap_mig_state.prev_bitmap = bitmap; +if (bitmap != s->prev_bitmap) { + s->prev_bitmap = bitmap; flags |=DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME ; }
@@ -198,19 +203,22 @@ static void send_bitmap_header(QEMUFile *f, SaveBitmapState *dbms, } } -static void send_bitmap_start (QEMUFile *f,SaveBitmapState *dbms) +static void send_bitmap_start (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms) { -send_bitmap_header (f, dbms,DIRTY_BITMAP_MIG_FLAG_START ); +send_bitmap_header (f, s, dbms,DIRTY_BITMAP_MIG_FLAG_START );qemu_put_be32 (f,bdrv_dirty_bitmap_granularity (dbms->bitmap));qemu_put_byte (f, dbms->flags); } -static void send_bitmap_complete (QEMUFile *f,SaveBitmapState *dbms) +static void send_bitmap_complete (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms) { -send_bitmap_header (f, dbms,DIRTY_BITMAP_MIG_FLAG_COMPLETE ); +send_bitmap_header (f, s, dbms,DIRTY_BITMAP_MIG_FLAG_COMPLETE ); } -static void send_bitmap_bits (QEMUFile *f,SaveBitmapState *dbms, +static void send_bitmap_bits (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms,uint64_t start_sector,uint32_t nr_sectors) {/* align for buffer_is_zero() */
@@ -235,7 +243,7 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms,trace_send_bitmap_bits (flags, start_sector, nr_sectors, buf_size); -send_bitmap_header (f, dbms, flags); +send_bitmap_header (f, s, dbms, flags);qemu_put_be64 (f, start_sector);qemu_put_be32 (f, nr_sectors);
@@ -254,12 +262,12 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms, }/* Called with iothread lock taken. */ -static void dirty_bitmap_do_save_cleanup (void ) +static void dirty_bitmap_do_save_cleanup (DBMSaveState *s) {SaveBitmapState *dbms; -while ((dbms =QSIMPLEQ_FIRST (&dirty_bitmap_mig_state.dbms_list)) !=NULL ) { -QSIMPLEQ_REMOVE_HEAD (&dirty_bitmap_mig_state.dbms_list, entry); +while ((dbms =QSIMPLEQ_FIRST (&s->dbms_list)) !=NULL ) { +QSIMPLEQ_REMOVE_HEAD (&s->dbms_list, entry);bdrv_dirty_bitmap_set_busy (dbms->bitmap, false);bdrv_unref (dbms->bs);g_free (dbms);
@@ -267,17 +275,17 @@ static void dirty_bitmap_do_save_cleanup(void) }/* Called with iothread lock taken. */ -static int init_dirty_bitmap_migration (void ) +static int init_dirty_bitmap_migration (DBMSaveState *s) {BlockDriverState *bs;BdrvDirtyBitmap *bitmap;SaveBitmapState *dbms;Error *local_err =NULL ; - dirty_bitmap_mig_state.bulk_completed = false; - dirty_bitmap_mig_state.prev_bs =NULL ; - dirty_bitmap_mig_state.prev_bitmap =NULL ; - dirty_bitmap_mig_state.no_bitmaps = false; + s->bulk_completed = false; + s->prev_bs =NULL ; + s->prev_bitmap =NULL ; + s->no_bitmaps = false;for (bs =bdrv_next_all_states (NULL ); bs; bs =bdrv_next_all_states (bs)) {const char *name =bdrv_get_device_or_node_name (bs);
@@ -316,35 +324,36 @@ static int init_dirty_bitmap_migration(void) dbms->flags |=DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT ; } -QSIMPLEQ_INSERT_TAIL (&dirty_bitmap_mig_state.dbms_list, +QSIMPLEQ_INSERT_TAIL (&s->dbms_list, dbms, entry); } }/* unset migration flags here, to not roll back it */ - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) {bdrv_dirty_bitmap_skip_store (dbms->bitmap, true); } -if (QSIMPLEQ_EMPTY (&dirty_bitmap_mig_state.dbms_list)) { - dirty_bitmap_mig_state.no_bitmaps = true; +if (QSIMPLEQ_EMPTY (&s->dbms_list)) { + s->no_bitmaps = true; }return 0 ; fail: -dirty_bitmap_do_save_cleanup (); +dirty_bitmap_do_save_cleanup (s);return -1 ; }/* Called with no lock taken. */ -static void bulk_phase_send_chunk (QEMUFile *f,SaveBitmapState *dbms) +static void bulk_phase_send_chunk (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms) {uint32_t nr_sectors =MIN (dbms->total_sectors - dbms->cur_sector, dbms->sectors_per_chunk); -send_bitmap_bits (f, dbms, dbms->cur_sector, nr_sectors); +send_bitmap_bits (f, s, dbms, dbms->cur_sector, nr_sectors); dbms->cur_sector += nr_sectors;if (dbms->cur_sector >= dbms->total_sectors) {
@@ -353,61 +362,66 @@ static void bulk_phase_send_chunk(QEMUFile *f, SaveBitmapState *dbms) }/* Called with no lock taken. */ -static void bulk_phase (QEMUFile *f,bool limit) +static void bulk_phase (QEMUFile *f,DBMSaveState *s,bool limit) {SaveBitmapState *dbms; - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) {while (!dbms->bulk_completed) { -bulk_phase_send_chunk (f, dbms); +bulk_phase_send_chunk (f, s, dbms);if (limit &&qemu_file_rate_limit (f)) {return ; } } } - dirty_bitmap_mig_state.bulk_completed = true; + s->bulk_completed = true; }/* for SaveVMHandlers */ static void dirty_bitmap_save_cleanup (void *opaque) { -dirty_bitmap_do_save_cleanup (); +DBMSaveState *s = &((DBMState *)opaque)->save; + +dirty_bitmap_do_save_cleanup (s); }static int dirty_bitmap_save_iterate (QEMUFile *f,void *opaque) { +DBMSaveState *s = &((DBMState *)opaque)->save; +trace_dirty_bitmap_save_iterate (migration_in_postcopy ()); -if (migration_in_postcopy () && !dirty_bitmap_mig_state.bulk_completed) { -bulk_phase (f, true); +if (migration_in_postcopy () && !s->bulk_completed) { +bulk_phase (f, s, true); }qemu_put_bitmap_flags (f,DIRTY_BITMAP_MIG_FLAG_EOS ); -return dirty_bitmap_mig_state.bulk_completed; +return s->bulk_completed; }/* Called with iothread lock taken. */ static int dirty_bitmap_save_complete (QEMUFile *f,void *opaque) { +DBMSaveState *s = &((DBMState *)opaque)->save;SaveBitmapState *dbms;trace_dirty_bitmap_save_complete_enter (); -if (!dirty_bitmap_mig_state.bulk_completed) { -bulk_phase (f, false); +if (!s->bulk_completed) { +bulk_phase (f, s, false); } - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { -send_bitmap_complete (f, dbms); + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { +send_bitmap_complete (f, s, dbms); }qemu_put_bitmap_flags (f,DIRTY_BITMAP_MIG_FLAG_EOS );trace_dirty_bitmap_save_complete_finish (); -dirty_bitmap_do_save_cleanup (); +dirty_bitmap_save_cleanup (opaque);return 0 ; }
@@ -417,12 +431,13 @@ static void dirty_bitmap_save_pending(QEMUFile *f, void *opaque,uint64_t *res_compatible, uint64_t *res_postcopy_only) { + DBMSaveState *s = &((DBMState *)opaque)->save;SaveBitmapState *dbms;uint64_t pending =0 ;qemu_mutex_lock_iothread (); - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) {uint64_t gran =bdrv_dirty_bitmap_granularity (dbms->bitmap);uint64_t sectors = dbms->bulk_completed ?0 : dbms->total_sectors - dbms->cur_sector;
@@ -481,7 +496,7 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s) b->bs = s->bs; b->bitmap = s->bitmap; b->migrated = false; - enabled_bitmaps =g_slist_prepend (enabled_bitmaps, b); + s->enabled_bitmaps =g_slist_prepend (s->enabled_bitmaps, b); }return 0 ;
@@ -489,11 +504,12 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s)void dirty_bitmap_mig_before_vm_start (void ) { +DBMLoadState *s = &dbm_state.load;GSList *item; -qemu_mutex_lock (&finish_lock); +qemu_mutex_lock (&s->finish_lock); -for (item = enabled_bitmaps; item; item =g_slist_next (item)) { +for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;if (b->migrated) {
@@ -505,10 +521,10 @@ void dirty_bitmap_mig_before_vm_start(void)g_free (b); } -g_slist_free (enabled_bitmaps); - enabled_bitmaps =NULL ; +g_slist_free (s->enabled_bitmaps); + s->enabled_bitmaps =NULL ; -qemu_mutex_unlock (&finish_lock); +qemu_mutex_unlock (&s->finish_lock); }static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s)
@@ -517,9 +533,9 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)trace_dirty_bitmap_load_complete ();bdrv_dirty_bitmap_deserialize_finish (s->bitmap); -qemu_mutex_lock (&finish_lock); +qemu_mutex_lock (&s->finish_lock); -for (item = enabled_bitmaps; item; item =g_slist_next (item)) { +for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;if (b->bitmap == s->bitmap) {
@@ -530,7 +546,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)if (bdrv_dirty_bitmap_has_successor (s->bitmap)) {bdrv_dirty_bitmap_lock (s->bitmap); -if (enabled_bitmaps ==NULL ) { +if (s->enabled_bitmaps ==NULL ) {/* in postcopy */ bdrv_reclaim_dirty_bitmap_locked (s->bitmap, &error_abort);bdrv_enable_dirty_bitmap_locked (s->bitmap);
@@ -549,7 +565,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)bdrv_dirty_bitmap_unlock (s->bitmap); } -qemu_mutex_unlock (&finish_lock); +qemu_mutex_unlock (&s->finish_lock); }static int dirty_bitmap_load_bits (QEMUFile *f,DBMLoadState *s)
@@ -646,7 +662,7 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s)static int dirty_bitmap_load (QEMUFile *f,void *opaque,int version_id) { -static DBMLoadState s; +DBMLoadState *s = &((DBMState *)opaque)->load;int ret =0 ;trace_dirty_bitmap_load_enter ();
@@ -656,17 +672,17 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) } do { - ret = dirty_bitmap_load_header(f, &s); + ret = dirty_bitmap_load_header(f, s); if (ret < 0) { return ret; } - if (s.flags & DIRTY_BITMAP_MIG_FLAG_START) { - ret = dirty_bitmap_load_start(f, &s); - } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) { - dirty_bitmap_load_complete(f, &s); - } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_BITS) { - ret = dirty_bitmap_load_bits(f, &s); + if (s->flags & DIRTY_BITMAP_MIG_FLAG_START) { + ret = dirty_bitmap_load_start(f, s); + } else if (s->flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) { + dirty_bitmap_load_complete(f, s); + } else if (s->flags & DIRTY_BITMAP_MIG_FLAG_BITS) { + ret = dirty_bitmap_load_bits(f, s); } if (!ret) {
@@ -676,7 +692,7 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)if (ret) {return ret; } - }while (!(s.flags &DIRTY_BITMAP_MIG_FLAG_EOS )); + }while (!(s->flags &DIRTY_BITMAP_MIG_FLAG_EOS ));trace_dirty_bitmap_load_success ();return 0 ;
@@ -684,13 +700,14 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)static int dirty_bitmap_save_setup (QEMUFile *f,void *opaque) { +DBMSaveState *s = &((DBMState *)opaque)->save;SaveBitmapState *dbms =NULL ; -if (init_dirty_bitmap_migration () <0 ) { +if (init_dirty_bitmap_migration (s) <0 ) {return -1 ; } - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { -send_bitmap_start (f, dbms); + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { +send_bitmap_start (f, s, dbms); }qemu_put_bitmap_flags (f,DIRTY_BITMAP_MIG_FLAG_EOS );
@@ -699,7 +716,9 @@ static int dirty_bitmap_save_setup(QEMUFile *f, void *opaque)static bool dirty_bitmap_is_active (void *opaque) { -return migrate_dirty_bitmaps () && !dirty_bitmap_mig_state.no_bitmaps; +DBMSaveState *s = &((DBMState *)opaque)->save; + +return migrate_dirty_bitmaps () && !s->no_bitmaps; }static bool dirty_bitmap_is_active_iterate (void *opaque)
@@ -727,10 +746,10 @@ static SaveVMHandlers savevm_dirty_bitmap_handlers = {void dirty_bitmap_mig_init (void ) { -QSIMPLEQ_INIT (&dirty_bitmap_mig_state.dbms_list); -qemu_mutex_init (&finish_lock); +QSIMPLEQ_INIT (&dbm_state.save.dbms_list); +qemu_mutex_init (&dbm_state.load.finish_lock);register_savevm_live ("dirty-bitmap" ,0 ,1 , &savevm_dirty_bitmap_handlers, - &dirty_bitmap_mig_state); + &dbm_state); }
Signature · 1 line
2.21.0
PATCH v2 12/22 qemu-iotests/199: fix style
In reply to Vladimir Sementsov-OgievskiyMostly, satisfy pep8 complains.
--- tests/qemu-iotests/199 | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)
tests/qemu-iotests/199 +7 -6
@@ -28,8 +28,8 @@ disk_b = os.path.join(iotests.test_dir, 'disk_b') size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') -class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): +class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def tearDown(self): self.vm_a.shutdown() self.vm_b.shutdown()
@@ -54,7 +54,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap', granularity=granularity) - self.assert_qmp(result, 'return', {}); + self.assert_qmp(result, 'return', {}) s = 0 while s < write_size:
@@ -71,7 +71,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap') - self.assert_qmp(result, 'return', {}); + self.assert_qmp(result, 'return', {}) s = 0 while s < write_size: self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk))
@@ -104,15 +104,16 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) s += 0x10000 - result = self.vm_b.qmp('query-block'); + result = self.vm_b.qmp('query-block') while len(result['return'][0]['dirty-bitmaps']) > 1: time.sleep(2) - result = self.vm_b.qmp('query-block'); + result = self.vm_b.qmp('query-block') result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap') - self.assert_qmp(result, 'return/sha256', sha256); + self.assert_qmp(result, 'return/sha256', sha256) + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2'], supported_cache_modes=['none'],
Signature · 1 line
2.21.0
PATCH v2 18/22 python/qemu/machine: add kill() method
In reply to Vladimir Sementsov-OgievskiyAdd method to hard-kill vm, without any quit commands.
--- python/qemu/machine.py | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
python/qemu/machine.py +9 -3
@@ -341,7 +341,7 @@ class QEMUMachine(object): self._load_io_log() self._post_shutdown() - def shutdown(self, has_quit=False): + def shutdown(self, has_quit=False, hard=False): """ Terminate the VM and clean up """
@@ -353,7 +353,9 @@ class QEMUMachine(object): self._console_socket = None if self.is_running(): - if self._qmp: + if hard: + self._popen.kill() + elif self._qmp: try: if not has_quit: self._qmp.cmd('quit')
@@ -366,7 +368,8 @@ class QEMUMachine(object): self._post_shutdown() exitcode = self.exitcode() - if exitcode is not None and exitcode < 0: + if exitcode is not None and exitcode < 0 and \ + not (exitcode == -9 and hard): msg = 'qemu received signal %i: %s' if self._qemu_full_args: command = ' '.join(self._qemu_full_args)
@@ -376,6 +379,9 @@ class QEMUMachine(object): self._launched = False + def kill(self): + self.shutdown(hard=True) + def set_qmp_monitor(self, enabled=True): """ Set the QMP monitor.
Signature · 1 line
2.21.0
PATCH v2 6/22 migration/block-dirty-bitmap: rename finish_lock to just lock
In reply to Vladimir Sementsov-Ogievskiyfinish_lock is bad name, as lock used not only on process end.
--- migration/block-dirty-bitmap.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
migration/block-dirty-bitmap.c +6 -6
@@ -143,7 +143,7 @@ typedef struct DBMLoadState {BdrvDirtyBitmap *bitmap;GSList *enabled_bitmaps; -QemuMutex finish_lock; +QemuMutex lock;/* protect enabled_bitmaps */ } DBMLoadState; typedef struct DBMState {
@@ -507,7 +507,7 @@ void dirty_bitmap_mig_before_vm_start(void)DBMLoadState *s = &dbm_state.load;GSList *item; -qemu_mutex_lock (&s->finish_lock); +qemu_mutex_lock (&s->lock);for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;
@@ -524,7 +524,7 @@ void dirty_bitmap_mig_before_vm_start(void)g_slist_free (s->enabled_bitmaps); s->enabled_bitmaps =NULL ; -qemu_mutex_unlock (&s->finish_lock); +qemu_mutex_unlock (&s->lock); }static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s)
@@ -533,7 +533,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)trace_dirty_bitmap_load_complete ();bdrv_dirty_bitmap_deserialize_finish (s->bitmap); -qemu_mutex_lock (&s->finish_lock); +qemu_mutex_lock (&s->lock);for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;
@@ -565,7 +565,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)bdrv_dirty_bitmap_unlock (s->bitmap); } -qemu_mutex_unlock (&s->finish_lock); +qemu_mutex_unlock (&s->lock); }static int dirty_bitmap_load_bits (QEMUFile *f,DBMLoadState *s)
@@ -747,7 +747,7 @@ static SaveVMHandlers savevm_dirty_bitmap_handlers = {void dirty_bitmap_mig_init (void ) {QSIMPLEQ_INIT (&dbm_state.save.dbms_list); -qemu_mutex_init (&dbm_state.load.finish_lock); +qemu_mutex_init (&dbm_state.load.lock); register_savevm_live("dirty-bitmap", 0, 1, &savevm_dirty_bitmap_handlers,
Signature · 1 line
2.21.0
PATCH v2 11/22 migration/savevm: don't worry if bitmap migration postcopy failed
In reply to Vladimir Sementsov-OgievskiyFirst, if only bitmaps postcopy enabled (not ram postcopy) postcopy_pause_incoming crashes on assertion assert(mis->to_src_file).
And anyway, bitmaps postcopy is not prepared to be somehow recovered. The original idea instead is that if bitmaps postcopy failed, we just loss some bitmaps, which is not critical. So, on failure we just need to remove unfinished bitmaps and guest should continue execution on destination.
--- migration/savevm.c | 37 ++++++++++++++++++++++++++++++++----- 1 file changed, 32 insertions(+), 5 deletions(-)
migration/savevm.c +32 -5
@@ -1812,6 +1812,9 @@ static void *postcopy_ram_listen_thread(void *opaque)MigrationIncomingState *mis =migration_incoming_get_current ();QEMUFile *f = mis->from_src_file;int load_res; +MigrationState *migr =migrate_get_current (); + +object_ref (OBJECT (migr));migrate_set_state (&mis->state,MIGRATION_STATUS_ACTIVE ,MIGRATION_STATUS_POSTCOPY_ACTIVE );
@@ -1838,11 +1841,24 @@ static void *postcopy_ram_listen_thread(void *opaque)trace_postcopy_ram_listen_thread_exit ();if (load_res <0 ) { -error_report ("%s: loadvm failed: %d" , __func__, load_res);qemu_file_set_error (f, load_res); -migrate_set_state (&mis->state,MIGRATION_STATUS_POSTCOPY_ACTIVE , -MIGRATION_STATUS_FAILED ); - }else { +dirty_bitmap_mig_cancel_incoming (); +if (postcopy_state_get () ==POSTCOPY_INCOMING_RUNNING && + !migrate_postcopy_ram () &&migrate_dirty_bitmaps ()) + { +error_report ("%s: loadvm failed during postcopy: %d. All state is " +"migrated except for dirty bitmaps. Some dirty " +"bitmaps may be lost, and present migrated dirty " +"bitmaps are correctly migrated and valid." , + __func__, load_res); + load_res =0 ;/* prevent further exit() */ + }else { +error_report ("%s: loadvm failed: %d" , __func__, load_res); +migrate_set_state (&mis->state,MIGRATION_STATUS_POSTCOPY_ACTIVE , +MIGRATION_STATUS_FAILED ); + } + } +if (load_res >=0 ) {/* * This looks good, but it's possible that the device loading in the * main thread hasn't finished yet, and so we might not be in 'RUN'
@@ -1878,6 +1894,8 @@ static void *postcopy_ram_listen_thread(void *opaque) mis->have_listen_thread = false;postcopy_state_set (POSTCOPY_INCOMING_END ); +object_unref (OBJECT (migr)); +return NULL ; }
@@ -2429,6 +2447,8 @@ static bool postcopy_pause_incoming(MigrationIncomingState *mis) {trace_postcopy_pause_incoming (); +assert (migrate_postcopy_ram ()); +/* Clear the triggered bit to allow one recovery */ mis->postcopy_recover_triggered = false;
@@ -2513,15 +2533,22 @@ out:if (ret <0 ) {qemu_file_set_error (f, ret); +/* Cancel bitmaps incoming regardless of recovery */ +dirty_bitmap_mig_cancel_incoming (); +/* * If we are during an active postcopy, then we pause instead * of bail out to at least keep the VM's dirty data. Note * that POSTCOPY_INCOMING_LISTENING stage is still not enough, * during which we're still receiving device states and we * still haven't yet started the VM on destination. +* +* Only RAM postcopy supports recovery. Still, if RAM postcopy is +* enabled, canceled bitmaps postcopy will not affect RAM postcopy +* recovering. */ if (postcopy_state_get () ==POSTCOPY_INCOMING_RUNNING && -postcopy_pause_incoming (mis)) { +migrate_postcopy_ram () &&postcopy_pause_incoming (mis)) {/* Reset f to point to the newly created channel */ f = mis->from_src_file; goto retry;
Signature · 1 line
2.21.0
PATCH v2 14/22 qemu-iotests/199: better catch postcopy time
In reply to Vladimir Sementsov-OgievskiyThe test aims to test _postcopy_ migration, and wants to do some write operations during postcopy time.
Test considers migrate status=complete event on source as start of postcopy. This is completely wrong, completion is completion of the whole migration process. Let's instead consider destination start as start of postcopy, and use RESUME event for it.
Next, as migration finish, let's use migration status=complete event on target, as such method is closer to what libvirt or another user will do, than tracking number of dirty-bitmaps.
Finally, add a possibility to dump events for debug. And if set debug to True, we see, that actual postcopy period is very small relatively to the whole test duration time (~0.2 seconds to >40 seconds for me). This means, that test is very inefficient in what it supposed to do. Let's improve it in following commits.
--- tests/qemu-iotests/199 | 72 +++++++++++++++++++++++++++++++++--------- 1 file changed, 57 insertions(+), 15 deletions(-)
tests/qemu-iotests/199 +57 -15
@@ -20,17 +20,43 @@ import os import iotests -import time from iotests import qemu_img +debug = False + disk_a = os.path.join(iotests.test_dir, 'disk_a') disk_b = os.path.join(iotests.test_dir, 'disk_b') size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +def event_seconds(event): + return event['timestamp']['seconds'] + \ + event['timestamp']['microseconds'] / 1000000.0 + + +def event_dist(e1, e2): + return event_seconds(e2) - event_seconds(e1) + + class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def tearDown(self): + if debug: + self.vm_a_events += self.vm_a.get_qmp_events() + self.vm_b_events += self.vm_b.get_qmp_events() + for e in self.vm_a_events: + e['vm'] = 'SRC' + for e in self.vm_b_events: + e['vm'] = 'DST' + events = (self.vm_a_events + self.vm_b_events) + events = [(e['timestamp']['seconds'], + e['timestamp']['microseconds'], + e['vm'], + e['event'], + e.get('data', '')) for e in events] + for e in sorted(events): + print('{}.{:06} {} {} {}'.format(*e)) + self.vm_a.shutdown() self.vm_b.shutdown() os.remove(disk_a)
@@ -47,6 +73,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.launch() self.vm_b.launch() + # collect received events for debug + self.vm_a_events = [] + self.vm_b_events = [] + def test_postcopy(self): write_size = 0x40000000 granularity = 512
@@ -77,15 +107,13 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) s += 0x10000 - bitmaps_cap = {'capability': 'dirty-bitmaps', 'state': True} - events_cap = {'capability': 'events', 'state': True} + caps = [{'capability': 'dirty-bitmaps', 'state': True}, + {'capability': 'events', 'state': True}] - result = self.vm_a.qmp('migrate-set-capabilities', - capabilities=[bitmaps_cap, events_cap]) + result = self.vm_a.qmp('migrate-set-capabilities', capabilities=caps) self.assert_qmp(result, 'return', {}) - result = self.vm_b.qmp('migrate-set-capabilities', - capabilities=[bitmaps_cap]) + result = self.vm_b.qmp('migrate-set-capabilities', capabilities=caps) self.assert_qmp(result, 'return', {}) result = self.vm_a.qmp('migrate', uri='exec:cat>' + fifo)
@@ -94,24 +122,38 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('migrate-start-postcopy') self.assert_qmp(result, 'return', {}) - while True: - event = self.vm_a.event_wait('MIGRATION') - if event['data']['status'] == 'completed': - break + e_resume = self.vm_b.event_wait('RESUME') + self.vm_b_events.append(e_resume) s = 0x8000 while s < write_size: self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) s += 0x10000 + match = {'data': {'status': 'completed'}} + e_complete = self.vm_b.event_wait('MIGRATION', match=match) + self.vm_b_events.append(e_complete) + + # take queued event, should already been happened + e_stop = self.vm_a.event_wait('STOP') + self.vm_a_events.append(e_stop) + + downtime = event_dist(e_stop, e_resume) + postcopy_time = event_dist(e_resume, e_complete) + + # TODO: assert downtime * 10 < postcopy_time + if debug: + print('downtime:', downtime) + print('postcopy_time:', postcopy_time) + + # Assert that bitmap migration is finished (check that successor bitmap + # is removed) result = self.vm_b.qmp('query-block') - while len(result['return'][0]['dirty-bitmaps']) > 1: - time.sleep(2) - result = self.vm_b.qmp('query-block') + assert len(result['return'][0]['dirty-bitmaps']) == 1 + # Check content of migrated (and updated by new writes) bitmap result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap') - self.assert_qmp(result, 'return/sha256', sha256)
Signature · 1 line
2.21.0
PATCH v2 17/22 qemu-iotests/199: increase postcopy period
In reply to Vladimir Sementsov-OgievskiyTest wants force bitmap postcopy. Still, resulting postcopy period is very small. Let's increase it by adding more bitmaps to migrate. Also, test disabled bitmaps migration.
--- tests/qemu-iotests/199 | 58 ++++++++++++++++++++++++++++-------------- 1 file changed, 39 insertions(+), 19 deletions(-)
tests/qemu-iotests/199 +39 -19
@@ -103,29 +103,45 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def test_postcopy(self): granularity = 512 + nb_bitmaps = 15 - result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', - name='bitmap', granularity=granularity) - self.assert_qmp(result, 'return', {}) + for i in range(nb_bitmaps): + result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', + name='bitmap{}'.format(i), + granularity=granularity) + self.assert_qmp(result, 'return', {}) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', - node='drive0', name='bitmap') + node='drive0', name='bitmap0') empty_sha256 = result['return']['sha256'] - apply_discards(self.vm_a, discards1 + discards2) + apply_discards(self.vm_a, discards1) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', - node='drive0', name='bitmap') - sha256 = result['return']['sha256'] + node='drive0', name='bitmap0') + discards1_sha256 = result['return']['sha256'] # Check, that updating the bitmap by discards works - assert sha256 != empty_sha256 + assert discards1_sha256 != empty_sha256 - result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', - name='bitmap') - self.assert_qmp(result, 'return', {}) + # We want to calculate resulting sha256. Do it in bitmap0, so, disable + # other bitmaps + for i in range(1, nb_bitmaps): + result = self.vm_a.qmp('block-dirty-bitmap-disable', node='drive0', + name='bitmap{}'.format(i)) + self.assert_qmp(result, 'return', {}) - apply_discards(self.vm_a, discards1) + apply_discards(self.vm_a, discards2) + + result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', + node='drive0', name='bitmap0') + all_discards_sha256 = result['return']['sha256'] + + # Now, enable some bitmaps, to be updated during migration + for i in range(2, nb_bitmaps, 2): + result = self.vm_a.qmp('block-dirty-bitmap-enable', node='drive0', + name='bitmap{}'.format(i)) + self.assert_qmp(result, 'return', {}) caps = [{'capability': 'dirty-bitmaps', 'state': True}, {'capability': 'events', 'state': True}]
@@ -145,6 +161,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): e_resume = self.vm_b.event_wait('RESUME') self.vm_b_events.append(e_resume) + # enabled bitmaps should be updated apply_discards(self.vm_b, discards2) match = {'data': {'status': 'completed'}}
@@ -158,7 +175,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): downtime = event_dist(e_stop, e_resume) postcopy_time = event_dist(e_resume, e_complete) - # TODO: assert downtime * 10 < postcopy_time + assert downtime * 10 < postcopy_time if debug: print('downtime:', downtime) print('postcopy_time:', postcopy_time)
@@ -166,12 +183,15 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): # Assert that bitmap migration is finished (check that successor bitmap # is removed) result = self.vm_b.qmp('query-block') - assert len(result['return'][0]['dirty-bitmaps']) == 1 - - # Check content of migrated (and updated by new writes) bitmap - result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', - node='drive0', name='bitmap') - self.assert_qmp(result, 'return/sha256', sha256) + assert len(result['return'][0]['dirty-bitmaps']) == nb_bitmaps + + # Check content of migrated bitmaps. Still, don't waste time checking + # every bitmap + for i in range(0, nb_bitmaps, 5): + result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', + node='drive0', name='bitmap{}'.format(i)) + sha256 = discards1_sha256 if i % 2 else all_discards_sha256 + self.assert_qmp(result, 'return/sha256', sha256) if __name__ == '__main__':
Signature · 1 line
2.21.0
PATCH v2 10/22 migration/block-dirty-bitmap: cancel migration on shutdown
In reply to Vladimir Sementsov-OgievskiyIf target is turned of prior to postcopy finished, target crashes because busy bitmaps are found at shutdown.
Canceling incoming migration helps, as it removes all unfinished (and therefore busy) bitmaps.
Similarly on source we crash in bdrv_close_all which asserts that all bdrv states are removed, because bdrv states involved into dirty bitmap migration are referenced by it. So, we need to cancel outgoing migration as well.
--- migration/migration.h | 2 ++ migration/block-dirty-bitmap.c | 16 ++++++++++++++++ migration/migration.c | 13 +++++++++++++ 3 files changed, 31 insertions(+)
migration/migration.h +2 -0
@@ -332,6 +332,8 @@ void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,void migrate_send_rp_resume_ack (MigrationIncomingState *mis,uint32_t value);void dirty_bitmap_mig_before_vm_start (void ); +void dirty_bitmap_mig_cancel_outgoing (void ); +void dirty_bitmap_mig_cancel_incoming (void );void migrate_add_address (SocketAddress *address);int foreach_not_ignored_block (RAMBlockIterFunc func,void *opaque);
migration/block-dirty-bitmap.c +16 -0
@@ -585,6 +585,22 @@ static void cancel_incoming_locked(DBMLoadState *s) s->bitmaps =NULL ; } +void dirty_bitmap_mig_cancel_outgoing (void ) +{ +dirty_bitmap_do_save_cleanup (&dbm_state.save); +} + +void dirty_bitmap_mig_cancel_incoming (void ) +{ +DBMLoadState *s = &dbm_state.load; + +qemu_mutex_lock (&s->lock); + +cancel_incoming_locked (s); + +qemu_mutex_unlock (&s->lock); +} +static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s) {GSList *item;
migration/migration.c +13 -0
@@ -181,6 +181,19 @@ void migration_shutdown(void)*/ migrate_fd_cancel (current_migration);object_unref (OBJECT (current_migration)); + +/* +* Cancel outgoing migration of dirty bitmaps. It should +* at least unref used block nodes. +*/ +dirty_bitmap_mig_cancel_outgoing (); + +/* +* Cancel incoming migration of dirty bitmaps. Dirty bitmaps +* are non-critical data, and their loss never considered as +* something serious. +*/ +dirty_bitmap_mig_cancel_incoming (); }/* For outgoing */
Signature · 1 line
2.21.0
PATCH v2 7/22 migration/block-dirty-bitmap: simplify dirty_bitmap_load_complete
In reply to Vladimir Sementsov-Ogievskiybdrv_enable_dirty_bitmap_locked() call does nothing, as if we are in postcopy, bitmap successor must be enabled, and reclaim operation will enable the bitmap.
So, actually we need just call _reclaim_ in both if branches, and making differences only to add an assertion seems not really good. The logic becomes simple: on load complete we do reclaim and that's all.
--- migration/block-dirty-bitmap.c | 25 ++++--------------------- 1 file changed, 4 insertions(+), 21 deletions(-)
migration/block-dirty-bitmap.c +4 -21
@@ -535,6 +535,10 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)qemu_mutex_lock (&s->lock); +if (bdrv_dirty_bitmap_has_successor (s->bitmap)) { +bdrv_reclaim_dirty_bitmap (s->bitmap, &error_abort); + } +for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;
@@ -544,27 +548,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) } } -if (bdrv_dirty_bitmap_has_successor (s->bitmap)) { -bdrv_dirty_bitmap_lock (s->bitmap); -if (s->enabled_bitmaps ==NULL ) { -/* in postcopy */ -bdrv_reclaim_dirty_bitmap_locked (s->bitmap, &error_abort); -bdrv_enable_dirty_bitmap_locked (s->bitmap); - }else { -/* target not started, successor must be empty */ -int64_t count =bdrv_get_dirty_count (s->bitmap); -BdrvDirtyBitmap *ret =bdrv_reclaim_dirty_bitmap_locked (s->bitmap, -NULL ); -/* bdrv_reclaim_dirty_bitmap can fail only on no successor (it -* must be) or on merge fail, but merge can't fail when second -* bitmap is empty -*/ -assert (ret == s->bitmap && - count ==bdrv_get_dirty_count (s->bitmap)); - } -bdrv_dirty_bitmap_unlock (s->bitmap); - } -qemu_mutex_unlock (&s->lock); }
Signature · 1 line
2.21.0
PATCH v2 16/22 qemu-iotests/199: change discard patterns
In reply to Vladimir Sementsov-Ogievskiyiotest 40 works too long because of many discard opertion. On the same time, postcopy period is very short, in spite of all these efforts.
So, let's use less discards (and with more interesting patterns) to reduce test timing. In the next commit we'll increase postcopy period.
--- tests/qemu-iotests/199 | 44 +++++++++++++++++++++++++----------------- 1 file changed, 26 insertions(+), 18 deletions(-)
tests/qemu-iotests/199 +26 -18
@@ -30,6 +30,28 @@ size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +GiB = 1024 * 1024 * 1024 + +discards1 = ( + (0, GiB), + (2 * GiB + 512 * 5, 512), + (3 * GiB + 512 * 5, 512), + (100 * GiB, GiB) +) + +discards2 = ( + (3 * GiB + 512 * 8, 512), + (4 * GiB + 512 * 8, 512), + (50 * GiB, GiB), + (100 * GiB + GiB // 2, GiB) +) + + +def apply_discards(vm, discards): + for d in discards: + vm.hmp_qemu_io('drive0', 'discard {} {}'.format(*d)) + + def event_seconds(event): return event['timestamp']['seconds'] + \ event['timestamp']['microseconds'] / 1000000.0
@@ -80,9 +102,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b_events = [] def test_postcopy(self): - discard_size = 0x40000000 granularity = 512 - chunk = 4096 result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap', granularity=granularity)
@@ -92,14 +112,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): node='drive0', name='bitmap') empty_sha256 = result['return']['sha256'] - s = 0 - while s < discard_size: - self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 - s = 0x8000 - while s < discard_size: - self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + apply_discards(self.vm_a, discards1 + discards2) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap')
@@ -111,10 +124,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap') self.assert_qmp(result, 'return', {}) - s = 0 - while s < discard_size: - self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + + apply_discards(self.vm_a, discards1) caps = [{'capability': 'dirty-bitmaps', 'state': True}, {'capability': 'events', 'state': True}]
@@ -134,10 +145,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): e_resume = self.vm_b.event_wait('RESUME') self.vm_b_events.append(e_resume) - s = 0x8000 - while s < discard_size: - self.vm_b.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + apply_discards(self.vm_b, discards2) match = {'data': {'status': 'completed'}} e_complete = self.vm_b.event_wait('MIGRATION', match=match)
Signature · 1 line
2.21.0
PATCH v2 9/22 migration/block-dirty-bitmap: relax error handling in incoming part
In reply to Vladimir Sementsov-OgievskiyBitmaps data is not critical, and we should not fail the migration (or use postcopy recovering) because of dirty-bitmaps migration failure. Instead we should just lose unfinished bitmaps.
Still we have to report io stream violation errors, as they affect the whole migration stream.
--- migration/block-dirty-bitmap.c | 148 +++++++++++++++++++++++++-------- 1 file changed, 113 insertions(+), 35 deletions(-)
migration/block-dirty-bitmap.c +113 -35
@@ -145,6 +145,15 @@ typedef struct DBMLoadState {bool before_vm_start_handled;/* set in dirty_bitmap_mig_before_vm_start */ +/* +* cancelled +* Incoming migration is cancelled for some reason. That means that we +* still should read our chunks from migration stream, to not affect other +* migration objects (like RAM), but just ignore them and do not touch any +* bitmaps or nodes. +*/ +bool cancelled; +GSList *bitmaps;QemuMutex lock;/* protect bitmaps */ } DBMLoadState;
@@ -545,13 +554,47 @@ void dirty_bitmap_mig_before_vm_start(void)qemu_mutex_unlock (&s->lock); } +static void cancel_incoming_locked (DBMLoadState *s) +{ +GSList *item; + +if (s->cancelled) { +return ; + } + + s->cancelled = true; + s->bs =NULL ; + s->bitmap =NULL ; + +/* Drop all unfinished bitmaps */ +for (item = s->bitmaps; item; item =g_slist_next (item)) { +LoadBitmapState *b = item->data; + +/* +* Bitmap must be unfinished, as finished bitmaps should already be +* removed from the list. +*/ +assert (!s->before_vm_start_handled || !b->migrated); +if (bdrv_dirty_bitmap_has_successor (b->bitmap)) { +bdrv_reclaim_dirty_bitmap (b->bitmap, &error_abort); + } +bdrv_release_dirty_bitmap (b->bitmap); + } + +g_slist_free_full (s->bitmaps, g_free); + s->bitmaps =NULL ; +} +static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s) {GSList *item;trace_dirty_bitmap_load_complete (); -bdrv_dirty_bitmap_deserialize_finish (s->bitmap); -qemu_mutex_lock (&s->lock); +if (s->cancelled) { +return ; + } + +bdrv_dirty_bitmap_deserialize_finish (s->bitmap);if (bdrv_dirty_bitmap_has_successor (s->bitmap)) {bdrv_reclaim_dirty_bitmap (s->bitmap, &error_abort);
@@ -569,8 +612,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)break ; } } - -qemu_mutex_unlock (&s->lock); }static int dirty_bitmap_load_bits (QEMUFile *f,DBMLoadState *s)
@@ -582,15 +623,32 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)if (s->flags &DIRTY_BITMAP_MIG_FLAG_ZEROES ) {trace_dirty_bitmap_load_bits_zeroes (); -bdrv_dirty_bitmap_deserialize_zeroes (s->bitmap, first_byte, nr_bytes, - false); +if (!s->cancelled) { +bdrv_dirty_bitmap_deserialize_zeroes (s->bitmap, first_byte, + nr_bytes, false); + } }else {size_t ret;uint8_t *buf;uint64_t buf_size =qemu_get_be64 (f); -uint64_t needed_size = -bdrv_dirty_bitmap_serialization_size (s->bitmap, - first_byte, nr_bytes); +uint64_t needed_size; + + buf =g_malloc (buf_size); + ret =qemu_get_buffer (f, buf, buf_size); +if (ret != buf_size) { +error_report ("Failed to read bitmap bits" ); +g_free (buf); +return -EIO ; + } + +if (s->cancelled) { +g_free (buf); +return 0 ; + } + + needed_size =bdrv_dirty_bitmap_serialization_size (s->bitmap, + first_byte, + nr_bytes);if (needed_size > buf_size || buf_size > QEMU_ALIGN_UP(needed_size, 4 * sizeof(long))
@@ -599,15 +657,8 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)error_report ("Migrated bitmap granularity doesn't " "match the destination bitmap '%s' granularity" ,bdrv_dirty_bitmap_name (s->bitmap)); -return -EINVAL ; - } - - buf =g_malloc (buf_size); - ret =qemu_get_buffer (f, buf, buf_size); -if (ret != buf_size) { -error_report ("Failed to read bitmap bits" ); -g_free (buf); -return -EIO ; +cancel_incoming_locked (s); +return 0 ; } bdrv_dirty_bitmap_deserialize_part(s->bitmap, buf, first_byte, nr_bytes,
@@ -632,14 +683,16 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s)error_report ("Unable to read node name string" );return -EINVAL ; } - s->bs =bdrv_lookup_bs (s->node_name, s->node_name, &local_err); -if (!s->bs) { -error_report_err (local_err); -return -EINVAL ; +if (!s->cancelled) { + s->bs =bdrv_lookup_bs (s->node_name, s->node_name, &local_err); +if (!s->bs) { +error_report_err (local_err); +cancel_incoming_locked (s); + } } - }else if (!s->bs && !nothing) { + }else if (!s->bs && !nothing && !s->cancelled) {error_report ("Error: block device name is not set" ); -return -EINVAL ; +cancel_incoming_locked (s); }if (s->flags &DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME ) {
@@ -647,24 +700,38 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s)error_report ("Unable to read bitmap name string" );return -EINVAL ; } - s->bitmap =bdrv_find_dirty_bitmap (s->bs, s->bitmap_name); - -/* bitmap may be NULL here, it wouldn't be an error if it is the -* first occurrence of the bitmap */ -if (!s->bitmap && !(s->flags &DIRTY_BITMAP_MIG_FLAG_START )) { -error_report ("Error: unknown dirty bitmap " -"'%s' for block device '%s'" , - s->bitmap_name, s->node_name); -return -EINVAL ; +if (!s->cancelled) { + s->bitmap =bdrv_find_dirty_bitmap (s->bs, s->bitmap_name); + +/* +* bitmap may be NULL here, it wouldn't be an error if it is the +* first occurrence of the bitmap +*/ +if (!s->bitmap && !(s->flags &DIRTY_BITMAP_MIG_FLAG_START )) { +error_report ("Error: unknown dirty bitmap " +"'%s' for block device '%s'" , + s->bitmap_name, s->node_name); +cancel_incoming_locked (s); + } } - }else if (!s->bitmap && !nothing) { + }else if (!s->bitmap && !nothing && !s->cancelled) {error_report ("Error: block device name is not set" ); -return -EINVAL ; +cancel_incoming_locked (s); }return 0 ; } +/* +* dirty_bitmap_load +* +* Load sequence of dirty bitmap chunks. Return error only on fatal io stream +* violations. On other errors just cancel bitmaps incoming migration and return +* 0. +* +* Note, than when incoming bitmap migration is canceled, we still must read all +* our chunks (and just ignore them), to not affect other migration objects. +*/ static int dirty_bitmap_load (QEMUFile *f,void *opaque,int version_id) {DBMLoadState *s = &((DBMState *)opaque)->load;
@@ -673,12 +740,19 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)trace_dirty_bitmap_load_enter ();if (version_id !=1 ) { +qemu_mutex_lock (&s->lock); +cancel_incoming_locked (s); +qemu_mutex_unlock (&s->lock);return -EINVAL ; } do { + qemu_mutex_lock(&s->lock); + ret = dirty_bitmap_load_header(f, s); if (ret < 0) { + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return ret; }
@@ -695,8 +769,12 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) }if (ret) { +cancel_incoming_locked (s); +qemu_mutex_unlock (&s->lock);return ret; } + +qemu_mutex_unlock (&s->lock); }while (!(s->flags &DIRTY_BITMAP_MIG_FLAG_EOS ));trace_dirty_bitmap_load_success ();
Signature · 1 line
2.21.0
PATCH v2 1/22 migration/block-dirty-bitmap: fix dirty_bitmap_mig_before_vm_start
In reply to Vladimir Sementsov-OgievskiyNo reason to use _locked version of bdrv_enable_dirty_bitmap, as we don't lock this mutex before. Moreover, the adjacent bdrv_dirty_bitmap_enable_successor do lock the mutex.
--- migration/block-dirty-bitmap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
migration/block-dirty-bitmap.c +1 -1
@@ -498,7 +498,7 @@ void dirty_bitmap_mig_before_vm_start(void)DirtyBitmapLoadBitmapState *b = item->data;if (b->migrated) { -bdrv_enable_dirty_bitmap_locked (b->bitmap); +bdrv_enable_dirty_bitmap (b->bitmap); }else {bdrv_dirty_bitmap_enable_successor (b->bitmap); }
Signature · 1 line
2.21.0
PATCH v2 8/22 migration/block-dirty-bitmap: keep bitmap state for all bitmaps
In reply to Vladimir Sementsov-OgievskiyKeep bitmap state for disabled bitmaps too. Keep the state until the end of the process. It's needed for the following commit to implement bitmap postcopy canceling.
To clean-up the new list the following logic is used:
We need two events to consider bitmap migration finished:
1. chunk with DIRTY_BITMAP_MIG_FLAG_COMPLETE flag should be received
2. dirty_bitmap_mig_before_vm_start should be called
These two events may come in any order, so we understand which one is last, and on the last of them we remove bitmap migration state from the list.
--- migration/block-dirty-bitmap.c | 64 +++++++++++++++++++++++----------- 1 file changed, 43 insertions(+), 21 deletions(-)
migration/block-dirty-bitmap.c +43 -21
@@ -132,6 +132,7 @@ typedef struct LoadBitmapState {BlockDriverState *bs;BdrvDirtyBitmap *bitmap;bool migrated; +bool enabled; } LoadBitmapState;/* State of the dirty bitmap migration (DBM) during load process */
@@ -142,8 +143,10 @@ typedef struct DBMLoadState {BlockDriverState *bs;BdrvDirtyBitmap *bitmap; -GSList *enabled_bitmaps; -QemuMutex lock;/* protect enabled_bitmaps */ +bool before_vm_start_handled;/* set in dirty_bitmap_mig_before_vm_start */ + +GSList *bitmaps; +QemuMutex lock;/* protect bitmaps */ } DBMLoadState; typedef struct DBMState {
@@ -458,6 +461,7 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s)Error *local_err =NULL ;uint32_t granularity =qemu_get_be32 (f);uint8_t flags =qemu_get_byte (f); +LoadBitmapState *b;if (s->bitmap) { error_report("Bitmap with the same name ('%s') already exists on "
@@ -484,45 +488,59 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s)bdrv_disable_dirty_bitmap (s->bitmap);if (flags &DIRTY_BITMAP_MIG_START_FLAG_ENABLED ) { -LoadBitmapState *b; -bdrv_dirty_bitmap_create_successor (s->bitmap, &local_err);if (local_err) {error_report_err (local_err);return -EINVAL ; } - - b =g_new (LoadBitmapState,1 ); - b->bs = s->bs; - b->bitmap = s->bitmap; - b->migrated = false; - s->enabled_bitmaps =g_slist_prepend (s->enabled_bitmaps, b); } + b =g_new (LoadBitmapState,1 ); + b->bs = s->bs; + b->bitmap = s->bitmap; + b->migrated = false; + b->enabled = flags &DIRTY_BITMAP_MIG_START_FLAG_ENABLED , + + s->bitmaps =g_slist_prepend (s->bitmaps, b); +return 0 ; } -void dirty_bitmap_mig_before_vm_start (void ) +/* +* before_vm_start_handle_item +* +* g_slist_foreach helper +* +* item is LoadBitmapState* +* opaque is DBMLoadState* +*/ +static void before_vm_start_handle_item (void *item,void *opaque) { -DBMLoadState *s = &dbm_state.load; -GSList *item; - -qemu_mutex_lock (&s->lock); - -for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) { -LoadBitmapState *b = item->data; +DBMLoadState *s = opaque; +LoadBitmapState *b = item; +if (b->enabled) {if (b->migrated) {bdrv_enable_dirty_bitmap (b->bitmap); }else {bdrv_dirty_bitmap_enable_successor (b->bitmap); } + } +if (b->migrated) { + s->bitmaps =g_slist_remove (s->bitmaps, b);g_free (b); } +} -g_slist_free (s->enabled_bitmaps); - s->enabled_bitmaps =NULL ; +void dirty_bitmap_mig_before_vm_start (void ) +{ +DBMLoadState *s = &dbm_state.load; +qemu_mutex_lock (&s->lock); + +assert (!s->before_vm_start_handled); +g_slist_foreach (s->bitmaps, before_vm_start_handle_item, s); + s->before_vm_start_handled = true;qemu_mutex_unlock (&s->lock); }
@@ -539,11 +557,15 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)bdrv_reclaim_dirty_bitmap (s->bitmap, &error_abort); } -for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) { +for (item = s->bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;if (b->bitmap == s->bitmap) { b->migrated = true; +if (s->before_vm_start_handled) { + s->bitmaps =g_slist_remove (s->bitmaps, b); +g_free (b); + }break ; } }
Signature · 1 line
2.21.0
PATCH v2 13/22 qemu-iotests/199: drop extra constraints
In reply to Vladimir Sementsov-OgievskiyWe don't need any specific format constraints here. Still keep qcow2 for two reasons:
1. No extra calls of format-unrelated test
2. Add some check around persistent bitmap in future (require qcow2)
--- tests/qemu-iotests/199 | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
tests/qemu-iotests/199 +1 -2
@@ -116,5 +116,4 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): if __name__ == '__main__': - iotests.main(supported_fmts=['qcow2'], supported_cache_modes=['none'], - supported_protocols=['file']) + iotests.main(supported_fmts=['qcow2'])
Signature · 1 line
2.21.0
v2 cover Fix error handling during bitmap postcopy
Original idea of bitmaps postcopy migration is that bitmaps are non critical data, and their loss is not serious problem. So, using postcopy method on any failure we should just drop unfinished bitmaps and continue guest execution.
However, it doesn't work so. It crashes, fails, it goes to postcopy-recovery feature. It does anything except for behavior we want. These series fixes at least some problems with error handling during bitmaps migration postcopy.
v1 was "[PATCH 0/7] Fix crashes on early shutdown during bitmaps postcopy"
v2:
Most of patches are new or changed a lot.
Only patches 06,07 mostly unchanged, just rebased on refactorings.
Vladimir Sementsov-Ogievskiy (22):
migration/block-dirty-bitmap: fix dirty_bitmap_mig_before_vm_start migration/block-dirty-bitmap: rename state structure types migration/block-dirty-bitmap: rename dirty_bitmap_mig_cleanup migration/block-dirty-bitmap: move mutex init to dirty_bitmap_mig_init migration/block-dirty-bitmap: refactor state global variables migration/block-dirty-bitmap: rename finish_lock to just lock migration/block-dirty-bitmap: simplify dirty_bitmap_load_complete migration/block-dirty-bitmap: keep bitmap state for all bitmaps migration/block-dirty-bitmap: relax error handling in incoming part migration/block-dirty-bitmap: cancel migration on shutdown migration/savevm: don't worry if bitmap migration postcopy failed qemu-iotests/199: fix style
qemu-iotests/199: drop extra constraints
qemu-iotests/199: better catch postcopy time qemu-iotests/199: improve performance: set bitmap by discard qemu-iotests/199: change discard patterns qemu-iotests/199: increase postcopy period python/qemu/machine: add kill() method
qemu-iotests/199: prepare for new test-cases addition qemu-iotests/199: check persistent bitmaps qemu-iotests/199: add early shutdown case to bitmaps postcopy qemu-iotests/199: add source-killed case to bitmaps postcopy
migration/migration.h | 3 +- migration/block-dirty-bitmap.c | 444 +++++++++++++++++++++------------ migration/migration.c | 15 +- migration/savevm.c | 37 ++- python/qemu/machine.py | 12 +- tests/qemu-iotests/199 | 244 ++++++++++++++---- tests/qemu-iotests/199.out | 4 +- 7 files changed, 529 insertions(+), 230 deletions(-)
Signature · 1 line
2.21.0
PATCH v2 15/22 qemu-iotests/199: improve performance: set bitmap by discard
Discard dirties dirty-bitmap as well as write, but works faster. Let's use it instead.
--- tests/qemu-iotests/199 | 31 ++++++++++++++++++++----------- 1 file changed, 20 insertions(+), 11 deletions(-)
tests/qemu-iotests/199 +20 -11
@@ -67,8 +67,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): os.mkfifo(fifo) qemu_img('create', '-f', iotests.imgfmt, disk_a, size) qemu_img('create', '-f', iotests.imgfmt, disk_b, size) - self.vm_a = iotests.VM(path_suffix='a').add_drive(disk_a) - self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b) + self.vm_a = iotests.VM(path_suffix='a').add_drive(disk_a, + 'discard=unmap') + self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b, + 'discard=unmap') self.vm_b.add_incoming("exec: cat '" + fifo + "'") self.vm_a.launch() self.vm_b.launch()
@@ -78,7 +80,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b_events = [] def test_postcopy(self): - write_size = 0x40000000 + discard_size = 0x40000000 granularity = 512 chunk = 4096
@@ -86,25 +88,32 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): name='bitmap', granularity=granularity) self.assert_qmp(result, 'return', {}) + result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', + node='drive0', name='bitmap') + empty_sha256 = result['return']['sha256'] + s = 0 - while s < write_size: - self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 s = 0x8000 - while s < write_size: - self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap') sha256 = result['return']['sha256'] + # Check, that updating the bitmap by discards works + assert sha256 != empty_sha256 + result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap') self.assert_qmp(result, 'return', {}) s = 0 - while s < write_size: - self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 caps = [{'capability': 'dirty-bitmaps', 'state': True},
@@ -126,8 +135,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b_events.append(e_resume) s = 0x8000 - while s < write_size: - self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_b.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 match = {'data': {'status': 'completed'}}
Signature · 1 line
2.21.0
PATCH v2 19/22 qemu-iotests/199: prepare for new test-cases addition
In reply to Vladimir Sementsov-OgievskiyMove future common part to start_postcopy() method. Move checking number of bitmaps to check_bitmap().
--- tests/qemu-iotests/199 | 36 +++++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 13 deletions(-)
tests/qemu-iotests/199 +23 -13
@@ -29,6 +29,8 @@ disk_b = os.path.join(iotests.test_dir, 'disk_b') size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +granularity = 512 +nb_bitmaps = 15 GiB = 1024 * 1024 * 1024
@@ -61,6 +63,15 @@ def event_dist(e1, e2): return event_seconds(e2) - event_seconds(e1) +def check_bitmaps(vm, count): + result = vm.qmp('query-block') + + if count == 0: + assert 'dirty-bitmaps' not in result['return'][0] + else: + assert len(result['return'][0]['dirty-bitmaps']) == count + + class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def tearDown(self): if debug:
@@ -101,10 +112,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a_events = [] self.vm_b_events = [] - def test_postcopy(self): - granularity = 512 - nb_bitmaps = 15 - + def start_postcopy(self): + """ Run migration until RESUME event on target. Return this event. """ for i in range(nb_bitmaps): result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap{}'.format(i),
@@ -119,10 +128,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap0') - discards1_sha256 = result['return']['sha256'] + self.discards1_sha256 = result['return']['sha256'] # Check, that updating the bitmap by discards works - assert discards1_sha256 != empty_sha256 + assert self.discards1_sha256 != empty_sha256 # We want to calculate resulting sha256. Do it in bitmap0, so, disable # other bitmaps
@@ -135,7 +144,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap0') - all_discards_sha256 = result['return']['sha256'] + self.all_discards_sha256 = result['return']['sha256'] # Now, enable some bitmaps, to be updated during migration for i in range(2, nb_bitmaps, 2):
@@ -160,6 +169,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): e_resume = self.vm_b.event_wait('RESUME') self.vm_b_events.append(e_resume) + return e_resume + + def test_postcopy_success(self): + e_resume = self.start_postcopy() # enabled bitmaps should be updated apply_discards(self.vm_b, discards2)
@@ -180,18 +193,15 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): print('downtime:', downtime) print('postcopy_time:', postcopy_time) - # Assert that bitmap migration is finished (check that successor bitmap - # is removed) - result = self.vm_b.qmp('query-block') - assert len(result['return'][0]['dirty-bitmaps']) == nb_bitmaps + check_bitmaps(self.vm_b, nb_bitmaps) # Check content of migrated bitmaps. Still, don't waste time checking # every bitmap for i in range(0, nb_bitmaps, 5): result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap{}'.format(i)) - sha256 = discards1_sha256 if i % 2 else all_discards_sha256 - self.assert_qmp(result, 'return/sha256', sha256) + sha = self.discards1_sha256 if i % 2 else self.all_discards_sha256 + self.assert_qmp(result, 'return/sha256', sha) if __name__ == '__main__':
Signature · 1 line
2.21.0
PATCH v2 3/22 migration/block-dirty-bitmap: rename dirty_bitmap_mig_cleanup
In reply to Vladimir Sementsov-OgievskiyRename dirty_bitmap_mig_cleanup to dirty_bitmap_do_save_cleanup, to stress that it is on save part.
--- migration/block-dirty-bitmap.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
migration/block-dirty-bitmap.c +4 -4
@@ -259,7 +259,7 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms, }/* Called with iothread lock taken. */ -static void dirty_bitmap_mig_cleanup (void ) +static void dirty_bitmap_do_save_cleanup (void ) {SaveBitmapState *dbms;
@@ -338,7 +338,7 @@ static int init_dirty_bitmap_migration(void)return 0 ; fail: -dirty_bitmap_mig_cleanup (); +dirty_bitmap_do_save_cleanup ();return -1 ; }
@@ -377,7 +377,7 @@ static void bulk_phase(QEMUFile *f, bool limit)/* for SaveVMHandlers */ static void dirty_bitmap_save_cleanup (void *opaque) { -dirty_bitmap_mig_cleanup (); +dirty_bitmap_do_save_cleanup (); }static int dirty_bitmap_save_iterate (QEMUFile *f,void *opaque)
@@ -412,7 +412,7 @@ static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque)trace_dirty_bitmap_save_complete_finish (); -dirty_bitmap_mig_cleanup (); +dirty_bitmap_do_save_cleanup ();return 0 ; }
Signature · 1 line
2.21.0
PATCH v2 2/22 migration/block-dirty-bitmap: rename state structure types
In reply to Vladimir Sementsov-OgievskiyRename types to be symmetrical for load/save part and shorter.
--- migration/block-dirty-bitmap.c | 68 ++++++++++++++++++---------------- 1 file changed, 36 insertions(+), 32 deletions(-)
migration/block-dirty-bitmap.c +36 -32
@@ -100,23 +100,25 @@/* 0x04 was "AUTOLOAD" flags on elder versions, no it is ignored */ #define DIRTY_BITMAP_MIG_START_FLAG_RESERVED_MASK 0xf8 -typedef struct DirtyBitmapMigBitmapState { +/* State of one bitmap during save process */ +typedef struct SaveBitmapState {/* Written during setup phase. */ BlockDriverState *bs;const char *node_name;BdrvDirtyBitmap *bitmap;uint64_t total_sectors;uint64_t sectors_per_chunk; -QSIMPLEQ_ENTRY (DirtyBitmapMigBitmapState) entry; +QSIMPLEQ_ENTRY (SaveBitmapState) entry;uint8_t flags;/* For bulk phase. */ bool bulk_completed;uint64_t cur_sector; -}DirtyBitmapMigBitmapState ; +}SaveBitmapState ; -typedef struct DirtyBitmapMigState { - QSIMPLEQ_HEAD(, DirtyBitmapMigBitmapState) dbms_list; +/* State of the dirty bitmap migration (DBM) during save process */ +typedef struct DBMSaveState { + QSIMPLEQ_HEAD(, SaveBitmapState) dbms_list; bool bulk_completed; bool no_bitmaps;
@@ -124,23 +126,25 @@ typedef struct DirtyBitmapMigState {/* for send_bitmap_bits() */ BlockDriverState *prev_bs;BdrvDirtyBitmap *prev_bitmap; -} DirtyBitmapMigState; +} DBMSaveState; -typedef struct DirtyBitmapLoadState { +/* State of the dirty bitmap migration (DBM) during load process */ +typedef struct DBMLoadState {uint32_t flags;char node_name[256 ];char bitmap_name[256 ];BlockDriverState *bs;BdrvDirtyBitmap *bitmap; -}DirtyBitmapLoadState ; +}DBMLoadState ; -static DirtyBitmapMigState dirty_bitmap_mig_state; +static DBMSaveState dirty_bitmap_mig_state; -typedef struct DirtyBitmapLoadBitmapState { +/* State of one bitmap during load process */ +typedef struct LoadBitmapState {BlockDriverState *bs;BdrvDirtyBitmap *bitmap;bool migrated; -}DirtyBitmapLoadBitmapState ; +}LoadBitmapState ;static GSList *enabled_bitmaps;QemuMutex finish_lock;
@@ -170,7 +174,7 @@ static void qemu_put_bitmap_flags(QEMUFile *f, uint32_t flags)qemu_put_byte (f, flags); } -static void send_bitmap_header (QEMUFile *f,DirtyBitmapMigBitmapState *dbms, +static void send_bitmap_header (QEMUFile *f,SaveBitmapState *dbms,uint32_t additional_flags) {BlockDriverState *bs = dbms->bs;
@@ -199,19 +203,19 @@ static void send_bitmap_header(QEMUFile *f, DirtyBitmapMigBitmapState *dbms, } } -static void send_bitmap_start (QEMUFile *f,DirtyBitmapMigBitmapState *dbms) +static void send_bitmap_start (QEMUFile *f,SaveBitmapState *dbms) {send_bitmap_header (f, dbms,DIRTY_BITMAP_MIG_FLAG_START );qemu_put_be32 (f,bdrv_dirty_bitmap_granularity (dbms->bitmap));qemu_put_byte (f, dbms->flags); } -static void send_bitmap_complete (QEMUFile *f,DirtyBitmapMigBitmapState *dbms) +static void send_bitmap_complete (QEMUFile *f,SaveBitmapState *dbms) {send_bitmap_header (f, dbms,DIRTY_BITMAP_MIG_FLAG_COMPLETE ); } -static void send_bitmap_bits (QEMUFile *f,DirtyBitmapMigBitmapState *dbms, +static void send_bitmap_bits (QEMUFile *f,SaveBitmapState *dbms,uint64_t start_sector,uint32_t nr_sectors) {/* align for buffer_is_zero() */
@@ -257,7 +261,7 @@ static void send_bitmap_bits(QEMUFile *f, DirtyBitmapMigBitmapState *dbms,/* Called with iothread lock taken. */ static void dirty_bitmap_mig_cleanup (void ) { -DirtyBitmapMigBitmapState *dbms; +SaveBitmapState *dbms;while ((dbms =QSIMPLEQ_FIRST (&dirty_bitmap_mig_state.dbms_list)) !=NULL ) {QSIMPLEQ_REMOVE_HEAD (&dirty_bitmap_mig_state.dbms_list, entry);
@@ -272,7 +276,7 @@ static int init_dirty_bitmap_migration(void) {BlockDriverState *bs;BdrvDirtyBitmap *bitmap; -DirtyBitmapMigBitmapState *dbms; +SaveBitmapState *dbms;Error *local_err =NULL ; dirty_bitmap_mig_state.bulk_completed = false;
@@ -303,7 +307,7 @@ static int init_dirty_bitmap_migration(void)bdrv_ref (bs);bdrv_dirty_bitmap_set_busy (bitmap, true); - dbms =g_new0 (DirtyBitmapMigBitmapState,1 ); + dbms =g_new0 (SaveBitmapState,1 ); dbms->bs = bs; dbms->node_name = name; dbms->bitmap = bitmap;
@@ -340,7 +344,7 @@ fail: }/* Called with no lock taken. */ -static void bulk_phase_send_chunk (QEMUFile *f,DirtyBitmapMigBitmapState *dbms) +static void bulk_phase_send_chunk (QEMUFile *f,SaveBitmapState *dbms) {uint32_t nr_sectors =MIN (dbms->total_sectors - dbms->cur_sector, dbms->sectors_per_chunk);
@@ -356,7 +360,7 @@ static void bulk_phase_send_chunk(QEMUFile *f, DirtyBitmapMigBitmapState *dbms)/* Called with no lock taken. */ static void bulk_phase (QEMUFile *f,bool limit) { -DirtyBitmapMigBitmapState *dbms; +SaveBitmapState *dbms; QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) {while (!dbms->bulk_completed) {
@@ -393,7 +397,7 @@ static int dirty_bitmap_save_iterate(QEMUFile *f, void *opaque)static int dirty_bitmap_save_complete (QEMUFile *f,void *opaque) { -DirtyBitmapMigBitmapState *dbms; +SaveBitmapState *dbms;trace_dirty_bitmap_save_complete_enter ();if (!dirty_bitmap_mig_state.bulk_completed) {
@@ -418,7 +422,7 @@ static void dirty_bitmap_save_pending(QEMUFile *f, void *opaque,uint64_t *res_compatible, uint64_t *res_postcopy_only) { - DirtyBitmapMigBitmapState *dbms; + SaveBitmapState *dbms;uint64_t pending =0 ;qemu_mutex_lock_iothread ();
@@ -439,7 +443,7 @@ static void dirty_bitmap_save_pending(QEMUFile *f, void *opaque, }/* First occurrence of this bitmap. It should be created if doesn't exist */ -static int dirty_bitmap_load_start (QEMUFile *f,DirtyBitmapLoadState *s) +static int dirty_bitmap_load_start (QEMUFile *f,DBMLoadState *s) {Error *local_err =NULL ;uint32_t granularity =qemu_get_be32 (f);
@@ -470,7 +474,7 @@ static int dirty_bitmap_load_start(QEMUFile *f, DirtyBitmapLoadState *s)bdrv_disable_dirty_bitmap (s->bitmap);if (flags &DIRTY_BITMAP_MIG_START_FLAG_ENABLED ) { -DirtyBitmapLoadBitmapState *b; +LoadBitmapState *b;bdrv_dirty_bitmap_create_successor (s->bitmap, &local_err);if (local_err) {
@@ -478,7 +482,7 @@ static int dirty_bitmap_load_start(QEMUFile *f, DirtyBitmapLoadState *s)return -EINVAL ; } - b =g_new (DirtyBitmapLoadBitmapState,1 ); + b =g_new (LoadBitmapState,1 ); b->bs = s->bs; b->bitmap = s->bitmap; b->migrated = false;
@@ -495,7 +499,7 @@ void dirty_bitmap_mig_before_vm_start(void)qemu_mutex_lock (&finish_lock);for (item = enabled_bitmaps; item; item =g_slist_next (item)) { -DirtyBitmapLoadBitmapState *b = item->data; +LoadBitmapState *b = item->data;if (b->migrated) {bdrv_enable_dirty_bitmap (b->bitmap);
@@ -512,7 +516,7 @@ void dirty_bitmap_mig_before_vm_start(void)qemu_mutex_unlock (&finish_lock); } -static void dirty_bitmap_load_complete (QEMUFile *f,DirtyBitmapLoadState *s) +static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s) {GSList *item;trace_dirty_bitmap_load_complete ();
@@ -521,7 +525,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DirtyBitmapLoadState *s)qemu_mutex_lock (&finish_lock);for (item = enabled_bitmaps; item; item =g_slist_next (item)) { -DirtyBitmapLoadBitmapState *b = item->data; +LoadBitmapState *b = item->data;if (b->bitmap == s->bitmap) { b->migrated = true;
@@ -553,7 +557,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DirtyBitmapLoadState *s)qemu_mutex_unlock (&finish_lock); } -static int dirty_bitmap_load_bits (QEMUFile *f,DirtyBitmapLoadState *s) +static int dirty_bitmap_load_bits (QEMUFile *f,DBMLoadState *s) {uint64_t first_byte =qemu_get_be64 (f) <<BDRV_SECTOR_BITS ;uint64_t nr_bytes = (uint64_t )qemu_get_be32 (f) <<BDRV_SECTOR_BITS ;
@@ -598,7 +602,7 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DirtyBitmapLoadState *s)return 0 ; } -static int dirty_bitmap_load_header (QEMUFile *f,DirtyBitmapLoadState *s) +static int dirty_bitmap_load_header (QEMUFile *f,DBMLoadState *s) {Error *local_err =NULL ;bool nothing;
@@ -647,7 +651,7 @@ static int dirty_bitmap_load_header(QEMUFile *f, DirtyBitmapLoadState *s)static int dirty_bitmap_load (QEMUFile *f,void *opaque,int version_id) { -static DirtyBitmapLoadState s; +static DBMLoadState s;int ret =0 ;trace_dirty_bitmap_load_enter ();
@@ -685,7 +689,7 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)static int dirty_bitmap_save_setup (QEMUFile *f,void *opaque) { -DirtyBitmapMigBitmapState *dbms =NULL ; +SaveBitmapState *dbms =NULL ;if (init_dirty_bitmap_migration () <0 ) {return -1 ; }
Signature · 1 line
2.21.0
PATCH v2 4/22 migration/block-dirty-bitmap: move mutex init to dirty_bitmap_mig_init
In reply to Vladimir Sementsov-OgievskiyNo reasons to keep two public init functions.
--- migration/migration.h | 1 - migration/block-dirty-bitmap.c | 6 +----- migration/migration.c | 2 -- 3 files changed, 1 insertion(+), 8 deletions(-)
migration/migration.h +0 -1
@@ -332,7 +332,6 @@ void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,void migrate_send_rp_resume_ack (MigrationIncomingState *mis,uint32_t value);void dirty_bitmap_mig_before_vm_start (void ); -void init_dirty_bitmap_incoming_migration (void );void migrate_add_address (SocketAddress *address);int foreach_not_ignored_block (RAMBlockIterFunc func,void *opaque);
migration/block-dirty-bitmap.c +1 -5
@@ -148,11 +148,6 @@ typedef struct LoadBitmapState {static GSList *enabled_bitmaps;QemuMutex finish_lock; -void init_dirty_bitmap_incoming_migration (void ) -{ -qemu_mutex_init (&finish_lock); -} -static uint32_t qemu_get_bitmap_flags (QEMUFile *f) {uint8_t flags =qemu_get_byte (f);
@@ -733,6 +728,7 @@ static SaveVMHandlers savevm_dirty_bitmap_handlers = {void dirty_bitmap_mig_init (void ) {QSIMPLEQ_INIT (&dirty_bitmap_mig_state.dbms_list); +qemu_mutex_init (&finish_lock); register_savevm_live("dirty-bitmap", 0, 1, &savevm_dirty_bitmap_handlers,
migration/migration.c +0 -2
@@ -158,8 +158,6 @@ void migration_object_init(void)qemu_sem_init (¤t_incoming->postcopy_pause_sem_dst,0 );qemu_sem_init (¤t_incoming->postcopy_pause_sem_fault,0 ); -init_dirty_bitmap_incoming_migration (); -if (!migration_object_check (current_migration, &err)) {error_report_err (err);exit (1 );
Signature · 1 line
2.21.0
Re: [PATCH v2 11/22] migration/savevm: don't worry if bitmap migration postcopy failed
In reply to Vladimir Sementsov-OgievskiyQuoted 10 lines · * Vladimir Sementsov-Ogievskiy ([email protected]) wrote:
First, if only bitmaps postcopy enabled (not ram postcopy)
postcopy_pause_incomingcrashes on assertionassert(mis->to_src_file).And anyway, bitmaps postcopy is not prepared to be somehow recovered. The original idea instead is that if bitmaps postcopy failed, we just loss some bitmaps, which is not critical. So, on failure we just need to remove unfinished bitmaps and guest should continue execution on destination.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>
Quoted 92 lines
--- migration/savevm.c | 37 ++++++++++++++++++++++++++++++++----- 1 file changed, 32 insertions(+), 5 deletions(-)migration/savevm.c +32
-5@@ -1812,6 +1812,9 @@ static void *postcopy_ram_listen_thread(void *opaque)MigrationIncomingState *mis =migration_incoming_get_current ();QEMUFile *f = mis->from_src_file;int load_res; +MigrationState *migr =migrate_get_current (); + +object_ref (OBJECT (migr));migrate_set_state (&mis->state,MIGRATION_STATUS_ACTIVE ,MIGRATION_STATUS_POSTCOPY_ACTIVE );@@ -1838,11 +1841,24 @@ static void *postcopy_ram_listen_thread(void *opaque)trace_postcopy_ram_listen_thread_exit ();if (load_res <0 ) { -error_report ("%s: loadvm failed: %d" , __func__, load_res);qemu_file_set_error (f, load_res); -migrate_set_state (&mis->state,MIGRATION_STATUS_POSTCOPY_ACTIVE , -MIGRATION_STATUS_FAILED ); - }else { +dirty_bitmap_mig_cancel_incoming (); +if (postcopy_state_get () ==POSTCOPY_INCOMING_RUNNING && + !migrate_postcopy_ram () &&migrate_dirty_bitmaps ()) + { +error_report ("%s: loadvm failed during postcopy: %d. All state is " +"migrated except for dirty bitmaps. Some dirty " +"bitmaps may be lost, and present migrated dirty " +"bitmaps are correctly migrated and valid." , + __func__, load_res); + load_res =0 ;/* prevent further exit() */ + }else { +error_report ("%s: loadvm failed: %d" , __func__, load_res); +migrate_set_state (&mis->state,MIGRATION_STATUS_POSTCOPY_ACTIVE , +MIGRATION_STATUS_FAILED ); + } + } +if (load_res >=0 ) {/* * This looks good, but it's possible that the device loading in the * main thread hasn't finished yet, and so we might not be in 'RUN' @@ -1878,6 +1894,8 @@ static void *postcopy_ram_listen_thread(void *opaque) mis->have_listen_thread = false;postcopy_state_set (POSTCOPY_INCOMING_END ); +object_unref (OBJECT (migr)); +return NULL ; }@@ -2429,6 +2447,8 @@ static bool postcopy_pause_incoming(MigrationIncomingState *mis) {trace_postcopy_pause_incoming (); +assert (migrate_postcopy_ram ()); +/* Clear the triggered bit to allow one recovery */ mis->postcopy_recover_triggered = false;@@ -2513,15 +2533,22 @@ out:if (ret <0 ) {qemu_file_set_error (f, ret); +/* Cancel bitmaps incoming regardless of recovery */ +dirty_bitmap_mig_cancel_incoming (); +/* * If we are during an active postcopy, then we pause instead * of bail out to at least keep the VM's dirty data. Note * that POSTCOPY_INCOMING_LISTENING stage is still not enough, * during which we're still receiving device states and we * still haven't yet started the VM on destination. +* +* Only RAM postcopy supports recovery. Still, if RAM postcopy is +* enabled, canceled bitmaps postcopy will not affect RAM postcopy +* recovering. */ if (postcopy_state_get () ==POSTCOPY_INCOMING_RUNNING && -postcopy_pause_incoming (mis)) { +migrate_postcopy_ram () &&postcopy_pause_incoming (mis)) {/* Reset f to point to the newly created channel */ f = mis->from_src_file; goto retry;Signature · 1 line
2.21.0
--
Dr. David Alan Gilbert / [email protected] / Manchester, UK
Patchew URL: https://patchew.org/QEMU/[email protected]/
Patchew URL: https://patchew.org/QEMU/[email protected]/
Hi,
This series seems to have some coding style problems. See output below for more information:
Subject: [PATCH v2 00/22] Fix error handling during bitmap postcopy
Message-id: [email protected]
Type: series
=== TEST SCRIPT BEGIN ===
#!/bin/bash
git rev-parse base > /dev/null || exit 0
git config --local diff.renamelimit 0
git config --local diff.renames True
git config --local diff.algorithm histogram./scripts/checkpatch.pl --mailback base..
=== TEST SCRIPT END ===
Updating 3c8cf5a9c21ff8782164d1def7f44bd888713384
fatal: git fetch_pack: expected ACK/NAK, got 'ERR upload-pack: not our ref 247b588c357694c896d056836da2341d75451c4f'
fatal: The remote end hung up unexpectedly
error: Could not fetch 3c8cf5a9c21ff8782164d1def7f44bd888713384
Traceback (most recent call last):
File "patchew-tester/src/patchew-cli", line 521, in test_one
git_clone_repo(clone, r["repo"], r["head"], logf, True)
File "patchew-tester/src/patchew-cli", line 48, in git_clone_repo
stdout=logf, stderr=logf)
File "/opt/rh/rh-python36/root/usr/lib64/python3.6/subprocess.py", line 291, in check_call
raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command '['git', 'remote', 'add', '-f', '--mirror=fetch', '3c8cf5a9c21ff8782164d1def7f44bd888713384', 'https://github.com/patchew-project/qemu']' returned non-zero exit status 1.
The full log is available at
http://patchew.org/logs/[email protected]/testing.checkpatch/?type=message.
---
Email generated automatically by Patchew [https://patchew.org/].
Please send your feedback to [email protected]
Re: [PATCH v2 01/22] migration/block-dirty-bitmap: fix dirty_bitmap_mig_before_vm_start
In reply to Vladimir Sementsov-OgievskiyQuoted 17 lines · Vladimir Sementsov-Ogievskiy
No reason to use
_lockedversion ofbdrv_enable_dirty_bitmap, as we don't lock this mutex before. Moreover, the adjacentbdrv_dirty_bitmap_enable_successordo lock the mutex.Fixes 58f72b965e9e1qCc [email protected] # v3.0Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)migration/block-dirty-bitmap.c +0
-0
migration/block-dirty-bitmap.c void dirty_bitmap_mig_before_vm_start(void) · line 504
Quoted 8 lines
migration/block-dirty-bitmap.c +1
-1@@ -498,7 +498,7 @@ void dirty_bitmap_mig_before_vm_start(void)DirtyBitmapLoadBitmapState *b = item->data;if (b->migrated) { -bdrv_enable_dirty_bitmap_locked (b->bitmap); +bdrv_enable_dirty_bitmap (b->bitmap); }else {bdrv_dirty_bitmap_enable_successor (b->bitmap); }
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 03/22] migration/block-dirty-bitmap: rename dirty_bitmap_mig_cleanup
In reply to Vladimir Sementsov-OgievskiyQuoted 41 lines · Vladimir Sementsov-Ogievskiy
Rename
dirty_bitmap_mig_cleanuptodirty_bitmap_do_save_cleanup, to stress that it is on save part.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)migration/block-dirty-bitmap.c +3
-3@@ -259,7 +259,7 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms, }/* Called with iothread lock taken. */ -static void dirty_bitmap_mig_cleanup (void ) +static void dirty_bitmap_do_save_cleanup (void ) {SaveBitmapState *dbms;@@ -338,7 +338,7 @@ static int init_dirty_bitmap_migration(void)return 0 ; fail: -dirty_bitmap_mig_cleanup (); +dirty_bitmap_do_save_cleanup ();return -1 ; }@@ -377,7 +377,7 @@ static void bulk_phase(QEMUFile *f, bool limit)/* for SaveVMHandlers */ static void dirty_bitmap_save_cleanup (void *opaque) { -dirty_bitmap_mig_cleanup (); +dirty_bitmap_do_save_cleanup (); }static int dirty_bitmap_save_iterate (QEMUFile *f,void *opaque)
migration/block-dirty-bitmap.c static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque) · line 418
Quoted 8 lines
migration/block-dirty-bitmap.c +1
-1@@ -412,7 +412,7 @@ static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque)trace_dirty_bitmap_save_complete_finish (); -dirty_bitmap_mig_cleanup (); +dirty_bitmap_do_save_cleanup ();return 0 ; }
At the next opportunity, I would suggest the name like "dirty_bitmap_do_clean_after_saving()"
and similar for dirty_bitmap_save_cleanup() "dirty_bitmap_clean_after_saving()".
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 04/22] migration/block-dirty-bitmap: move mutex init to dirty_bitmap_mig_init
In reply to Vladimir Sementsov-OgievskiyQuoted 51 lines · Vladimir Sementsov-Ogievskiy
No reasons to keep two public init functions.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/migration.h | 1 - migration/block-dirty-bitmap.c | 6 +----- migration/migration.c | 2 -- 3 files changed, 1 insertion(+), 8 deletions(-)migration/migration.h +0
-1@@ -332,7 +332,6 @@ void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,void migrate_send_rp_resume_ack (MigrationIncomingState *mis,uint32_t value);void dirty_bitmap_mig_before_vm_start (void ); -void init_dirty_bitmap_incoming_migration (void );void migrate_add_address (SocketAddress *address);int foreach_not_ignored_block (RAMBlockIterFunc func,void *opaque);migration/block-dirty-bitmap.c +1
-5@@ -148,11 +148,6 @@ typedef struct LoadBitmapState {static GSList *enabled_bitmaps;QemuMutex finish_lock; -void init_dirty_bitmap_incoming_migration (void ) -{ -qemu_mutex_init (&finish_lock); -} -static uint32_t qemu_get_bitmap_flags (QEMUFile *f) {uint8_t flags =qemu_get_byte (f);@@ -733,6 +728,7 @@ static SaveVMHandlers savevm_dirty_bitmap_handlers = {void dirty_bitmap_mig_init (void ) {QSIMPLEQ_INIT (&dirty_bitmap_mig_state.dbms_list); +qemu_mutex_init (&finish_lock); register_savevm_live("dirty-bitmap", 0, 1, &savevm_dirty_bitmap_handlers,migration/migration.c +0
-0
migration/migration.c void migration_object_init(void) · line 163
Quoted 8 lines
migration/migration.c +0
-2@@ -158,8 +158,6 @@ void migration_object_init(void)qemu_sem_init (¤t_incoming->postcopy_pause_sem_dst,0 );qemu_sem_init (¤t_incoming->postcopy_pause_sem_fault,0 ); -init_dirty_bitmap_incoming_migration (); -if (!migration_object_check (current_migration, &err)) {error_report_err (err);exit (1 );
I rely on that the mutex initialization is being made in proper time.
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 05/22] migration/block-dirty-bitmap: refactor state global variables
In reply to Vladimir Sementsov-OgievskiyQuoted 222 lines · Vladimir Sementsov-Ogievskiy
Move all state variables into one global struct. Reduce global variable usage, utilizing opaque pointer where possible.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 171 ++++++++++++++++++--------------- 1 file changed, 95 insertions(+), 76 deletions(-)migration/block-dirty-bitmap.c +49
-40@@ -128,6 +128,12 @@ typedef struct DBMSaveState {BdrvDirtyBitmap *prev_bitmap; } DBMSaveState; +typedef struct LoadBitmapState { +BlockDriverState *bs; +BdrvDirtyBitmap *bitmap; +bool migrated; +}LoadBitmapState ; +/* State of the dirty bitmap migration (DBM) during load process */ typedef struct DBMLoadState { uint32_t flags;@@ -135,18 +141,17 @@ typedef struct DBMLoadState {char bitmap_name[256 ];BlockDriverState *bs;BdrvDirtyBitmap *bitmap; + +GSList *enabled_bitmaps; +QemuMutex finish_lock; } DBMLoadState; -static DBMSaveState dirty_bitmap_mig_state; +typedef struct DBMState { +DBMSaveState save; +DBMLoadState load; +}DBMState ; -/* State of one bitmap during load process */ -typedef struct LoadBitmapState { -BlockDriverState *bs; -BdrvDirtyBitmap *bitmap; -bool migrated; -}LoadBitmapState ; -static GSList *enabled_bitmaps; -QemuMutex finish_lock; +static DBMState dbm_state;static uint32_t qemu_get_bitmap_flags (QEMUFile *f) {@@ -169,21 +174,21 @@ static void qemu_put_bitmap_flags(QEMUFile *f, uint32_t flags)qemu_put_byte (f, flags); } -static void send_bitmap_header (QEMUFile *f,SaveBitmapState *dbms, -uint32_t additional_flags) +static void send_bitmap_header (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms,uint32_t additional_flags) {BlockDriverState *bs = dbms->bs;BdrvDirtyBitmap *bitmap = dbms->bitmap;uint32_t flags = additional_flags;trace_send_bitmap_header_enter (); -if (bs != dirty_bitmap_mig_state.prev_bs) { - dirty_bitmap_mig_state.prev_bs = bs; +if (bs != s->prev_bs) { + s->prev_bs = bs; flags |=DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME ; } -if (bitmap != dirty_bitmap_mig_state.prev_bitmap) { - dirty_bitmap_mig_state.prev_bitmap = bitmap; +if (bitmap != s->prev_bitmap) { + s->prev_bitmap = bitmap; flags |=DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME ; }@@ -198,19 +203,22 @@ static void send_bitmap_header(QEMUFile *f, SaveBitmapState *dbms, } } -static void send_bitmap_start (QEMUFile *f,SaveBitmapState *dbms) +static void send_bitmap_start (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms) { -send_bitmap_header (f, dbms,DIRTY_BITMAP_MIG_FLAG_START ); +send_bitmap_header (f, s, dbms,DIRTY_BITMAP_MIG_FLAG_START );qemu_put_be32 (f,bdrv_dirty_bitmap_granularity (dbms->bitmap));qemu_put_byte (f, dbms->flags); } -static void send_bitmap_complete (QEMUFile *f,SaveBitmapState *dbms) +static void send_bitmap_complete (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms) { -send_bitmap_header (f, dbms,DIRTY_BITMAP_MIG_FLAG_COMPLETE ); +send_bitmap_header (f, s, dbms,DIRTY_BITMAP_MIG_FLAG_COMPLETE ); } -static void send_bitmap_bits (QEMUFile *f,SaveBitmapState *dbms, +static void send_bitmap_bits (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms,uint64_t start_sector,uint32_t nr_sectors) {/* align for buffer_is_zero() */ @@ -235,7 +243,7 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms,trace_send_bitmap_bits (flags, start_sector, nr_sectors, buf_size); -send_bitmap_header (f, dbms, flags); +send_bitmap_header (f, s, dbms, flags);qemu_put_be64 (f, start_sector);qemu_put_be32 (f, nr_sectors);@@ -254,12 +262,12 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms, }/* Called with iothread lock taken. */ -static void dirty_bitmap_do_save_cleanup (void ) +static void dirty_bitmap_do_save_cleanup (DBMSaveState *s) {SaveBitmapState *dbms; -while ((dbms =QSIMPLEQ_FIRST (&dirty_bitmap_mig_state.dbms_list)) !=NULL ) { -QSIMPLEQ_REMOVE_HEAD (&dirty_bitmap_mig_state.dbms_list, entry); +while ((dbms =QSIMPLEQ_FIRST (&s->dbms_list)) !=NULL ) { +QSIMPLEQ_REMOVE_HEAD (&s->dbms_list, entry);bdrv_dirty_bitmap_set_busy (dbms->bitmap, false);bdrv_unref (dbms->bs);g_free (dbms);@@ -267,17 +275,17 @@ static void dirty_bitmap_do_save_cleanup(void) }/* Called with iothread lock taken. */ -static int init_dirty_bitmap_migration (void ) +static int init_dirty_bitmap_migration (DBMSaveState *s) {BlockDriverState *bs;BdrvDirtyBitmap *bitmap;SaveBitmapState *dbms;Error *local_err =NULL ; - dirty_bitmap_mig_state.bulk_completed = false; - dirty_bitmap_mig_state.prev_bs =NULL ; - dirty_bitmap_mig_state.prev_bitmap =NULL ; - dirty_bitmap_mig_state.no_bitmaps = false; + s->bulk_completed = false; + s->prev_bs =NULL ; + s->prev_bitmap =NULL ; + s->no_bitmaps = false;for (bs =bdrv_next_all_states (NULL ); bs; bs =bdrv_next_all_states (bs)) {const char *name =bdrv_get_device_or_node_name (bs);@@ -316,35 +324,36 @@ static int init_dirty_bitmap_migration(void) dbms->flags |=DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT ; } -QSIMPLEQ_INSERT_TAIL (&dirty_bitmap_mig_state.dbms_list, +QSIMPLEQ_INSERT_TAIL (&s->dbms_list, dbms, entry); } }/* unset migration flags here, to not roll back it */ - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) {bdrv_dirty_bitmap_skip_store (dbms->bitmap, true); } -if (QSIMPLEQ_EMPTY (&dirty_bitmap_mig_state.dbms_list)) { - dirty_bitmap_mig_state.no_bitmaps = true; +if (QSIMPLEQ_EMPTY (&s->dbms_list)) { + s->no_bitmaps = true; }return 0 ; fail: -dirty_bitmap_do_save_cleanup (); +dirty_bitmap_do_save_cleanup (s);return -1 ; }/* Called with no lock taken. */ -static void bulk_phase_send_chunk (QEMUFile *f,SaveBitmapState *dbms) +static void bulk_phase_send_chunk (QEMUFile *f,DBMSaveState *s, +SaveBitmapState *dbms) {uint32_t nr_sectors =MIN (dbms->total_sectors - dbms->cur_sector, dbms->sectors_per_chunk); -send_bitmap_bits (f, dbms, dbms->cur_sector, nr_sectors); +send_bitmap_bits (f, s, dbms, dbms->cur_sector, nr_sectors); dbms->cur_sector += nr_sectors;if (dbms->cur_sector >= dbms->total_sectors) {@@ -353,61 +362,66 @@ static void bulk_phase_send_chunk(QEMUFile *f, SaveBitmapState *dbms) }/* Called with no lock taken. */ -static void bulk_phase (QEMUFile *f,bool limit) +static void bulk_phase (QEMUFile *f,DBMSaveState *s,bool limit) {SaveBitmapState *dbms; - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) {while (!dbms->bulk_completed) { -bulk_phase_send_chunk (f, dbms); +bulk_phase_send_chunk (f, s, dbms);if (limit &&qemu_file_rate_limit (f)) {return ; } } } - dirty_bitmap_mig_state.bulk_completed = true; + s->bulk_completed = true;
migration/block-dirty-bitmap.c static void bulk_phase_send_chunk(QEMUFile *f, SaveBitmapState *dbms) · line 387
Quoted 10 lines
migration/block-dirty-bitmap.c +3
-1}/* for SaveVMHandlers */ static void dirty_bitmap_save_cleanup (void *opaque) { -dirty_bitmap_do_save_cleanup (); +DBMSaveState *s = &((DBMState *)opaque)->save; + +dirty_bitmap_do_save_cleanup (s); }
Why do one need the extra nested "do" function?
Quoted 208 lines
migration/block-dirty-bitmap.c +40
-32static int dirty_bitmap_save_iterate (QEMUFile *f,void *opaque) { +DBMSaveState *s = &((DBMState *)opaque)->save; +trace_dirty_bitmap_save_iterate (migration_in_postcopy ()); -if (migration_in_postcopy () && !dirty_bitmap_mig_state.bulk_completed) { -bulk_phase (f, true); +if (migration_in_postcopy () && !s->bulk_completed) { +bulk_phase (f, s, true); }qemu_put_bitmap_flags (f,DIRTY_BITMAP_MIG_FLAG_EOS ); -return dirty_bitmap_mig_state.bulk_completed; +return s->bulk_completed; }/* Called with iothread lock taken. */ static int dirty_bitmap_save_complete (QEMUFile *f,void *opaque) { +DBMSaveState *s = &((DBMState *)opaque)->save;SaveBitmapState *dbms;trace_dirty_bitmap_save_complete_enter (); -if (!dirty_bitmap_mig_state.bulk_completed) { -bulk_phase (f, false); +if (!s->bulk_completed) { +bulk_phase (f, s, false); } - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { -send_bitmap_complete (f, dbms); + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { +send_bitmap_complete (f, s, dbms); }qemu_put_bitmap_flags (f,DIRTY_BITMAP_MIG_FLAG_EOS );trace_dirty_bitmap_save_complete_finish (); -dirty_bitmap_do_save_cleanup (); +dirty_bitmap_save_cleanup (opaque);return 0 ; }@@ -417,12 +431,13 @@ static void dirty_bitmap_save_pending(QEMUFile *f, void *opaque,uint64_t *res_compatible, uint64_t *res_postcopy_only) { + DBMSaveState *s = &((DBMState *)opaque)->save;SaveBitmapState *dbms;uint64_t pending =0 ;qemu_mutex_lock_iothread (); - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) {uint64_t gran =bdrv_dirty_bitmap_granularity (dbms->bitmap);uint64_t sectors = dbms->bulk_completed ?0 : dbms->total_sectors - dbms->cur_sector;@@ -481,7 +496,7 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s) b->bs = s->bs; b->bitmap = s->bitmap; b->migrated = false; - enabled_bitmaps =g_slist_prepend (enabled_bitmaps, b); + s->enabled_bitmaps =g_slist_prepend (s->enabled_bitmaps, b); }return 0 ;@@ -489,11 +504,12 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s)void dirty_bitmap_mig_before_vm_start (void ) { +DBMLoadState *s = &dbm_state.load;GSList *item; -qemu_mutex_lock (&finish_lock); +qemu_mutex_lock (&s->finish_lock); -for (item = enabled_bitmaps; item; item =g_slist_next (item)) { +for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;if (b->migrated) {@@ -505,10 +521,10 @@ void dirty_bitmap_mig_before_vm_start(void)g_free (b); } -g_slist_free (enabled_bitmaps); - enabled_bitmaps =NULL ; +g_slist_free (s->enabled_bitmaps); + s->enabled_bitmaps =NULL ; -qemu_mutex_unlock (&finish_lock); +qemu_mutex_unlock (&s->finish_lock); }static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s)@@ -517,9 +533,9 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)trace_dirty_bitmap_load_complete ();bdrv_dirty_bitmap_deserialize_finish (s->bitmap); -qemu_mutex_lock (&finish_lock); +qemu_mutex_lock (&s->finish_lock); -for (item = enabled_bitmaps; item; item =g_slist_next (item)) { +for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;if (b->bitmap == s->bitmap) {@@ -530,7 +546,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)if (bdrv_dirty_bitmap_has_successor (s->bitmap)) {bdrv_dirty_bitmap_lock (s->bitmap); -if (enabled_bitmaps ==NULL ) { +if (s->enabled_bitmaps ==NULL ) {/* in postcopy */ bdrv_reclaim_dirty_bitmap_locked (s->bitmap, &error_abort);bdrv_enable_dirty_bitmap_locked (s->bitmap);@@ -549,7 +565,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)bdrv_dirty_bitmap_unlock (s->bitmap); } -qemu_mutex_unlock (&finish_lock); +qemu_mutex_unlock (&s->finish_lock); }static int dirty_bitmap_load_bits (QEMUFile *f,DBMLoadState *s)@@ -646,7 +662,7 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s)static int dirty_bitmap_load (QEMUFile *f,void *opaque,int version_id) { -static DBMLoadState s; +DBMLoadState *s = &((DBMState *)opaque)->load;int ret =0 ;trace_dirty_bitmap_load_enter ();@@ -656,17 +672,17 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) } do { - ret = dirty_bitmap_load_header(f, &s); + ret = dirty_bitmap_load_header(f, s); if (ret < 0) { return ret; } - if (s.flags & DIRTY_BITMAP_MIG_FLAG_START) { - ret = dirty_bitmap_load_start(f, &s); - } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) { - dirty_bitmap_load_complete(f, &s); - } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_BITS) { - ret = dirty_bitmap_load_bits(f, &s); + if (s->flags & DIRTY_BITMAP_MIG_FLAG_START) { + ret = dirty_bitmap_load_start(f, s); + } else if (s->flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) { + dirty_bitmap_load_complete(f, s); + } else if (s->flags & DIRTY_BITMAP_MIG_FLAG_BITS) { + ret = dirty_bitmap_load_bits(f, s); } if (!ret) {@@ -676,7 +692,7 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)if (ret) {return ret; } - }while (!(s.flags &DIRTY_BITMAP_MIG_FLAG_EOS )); + }while (!(s->flags &DIRTY_BITMAP_MIG_FLAG_EOS ));trace_dirty_bitmap_load_success ();return 0 ;@@ -684,13 +700,14 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)static int dirty_bitmap_save_setup (QEMUFile *f,void *opaque) { +DBMSaveState *s = &((DBMState *)opaque)->save;SaveBitmapState *dbms =NULL ; -if (init_dirty_bitmap_migration () <0 ) { +if (init_dirty_bitmap_migration (s) <0 ) {return -1 ; } - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { -send_bitmap_start (f, dbms); + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { +send_bitmap_start (f, s, dbms); }qemu_put_bitmap_flags (f,DIRTY_BITMAP_MIG_FLAG_EOS );@@ -699,7 +716,9 @@ static int dirty_bitmap_save_setup(QEMUFile *f, void *opaque)static bool dirty_bitmap_is_active (void *opaque) { -return migrate_dirty_bitmaps () && !dirty_bitmap_mig_state.no_bitmaps; +DBMSaveState *s = &((DBMState *)opaque)->save; + +return migrate_dirty_bitmaps () && !s->no_bitmaps; }static bool dirty_bitmap_is_active_iterate (void *opaque)@@ -727,10 +746,10 @@ static SaveVMHandlers savevm_dirty_bitmap_handlers = {void dirty_bitmap_mig_init (void ) {
migration/block-dirty-bitmap.c static SaveVMHandlers savevm_dirty_bitmap_handlers = { · line 755
Quoted 10 lines
migration/block-dirty-bitmap.c +3
-3-QSIMPLEQ_INIT (&dirty_bitmap_mig_state.dbms_list); -qemu_mutex_init (&finish_lock); +QSIMPLEQ_INIT (&dbm_state.save.dbms_list); +qemu_mutex_init (&dbm_state.load.finish_lock);register_savevm_live ("dirty-bitmap" ,0 ,1 , &savevm_dirty_bitmap_handlers, - &dirty_bitmap_mig_state); + &dbm_state); }
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 06/22] migration/block-dirty-bitmap: rename finish_lock to just lock
In reply to Vladimir Sementsov-OgievskiyQuoted 58 lines · Vladimir Sementsov-Ogievskiy
finish_lockis bad name, as lock used not only on process end.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)migration/block-dirty-bitmap.c +5
-5@@ -143,7 +143,7 @@ typedef struct DBMLoadState {BdrvDirtyBitmap *bitmap;GSList *enabled_bitmaps; -QemuMutex finish_lock; +QemuMutex lock;/* protect enabled_bitmaps */ } DBMLoadState; typedef struct DBMState {@@ -507,7 +507,7 @@ void dirty_bitmap_mig_before_vm_start(void)DBMLoadState *s = &dbm_state.load;GSList *item; -qemu_mutex_lock (&s->finish_lock); +qemu_mutex_lock (&s->lock);for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;@@ -524,7 +524,7 @@ void dirty_bitmap_mig_before_vm_start(void)g_slist_free (s->enabled_bitmaps); s->enabled_bitmaps =NULL ; -qemu_mutex_unlock (&s->finish_lock); +qemu_mutex_unlock (&s->lock); }static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s)@@ -533,7 +533,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)trace_dirty_bitmap_load_complete ();bdrv_dirty_bitmap_deserialize_finish (s->bitmap); -qemu_mutex_lock (&s->finish_lock); +qemu_mutex_lock (&s->lock);for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;@@ -565,7 +565,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)bdrv_dirty_bitmap_unlock (s->bitmap); } -qemu_mutex_unlock (&s->finish_lock); +qemu_mutex_unlock (&s->lock); }static int dirty_bitmap_load_bits (QEMUFile *f,DBMLoadState *s)
migration/block-dirty-bitmap.c static SaveVMHandlers savevm_dirty_bitmap_handlers = { · line 753
Quoted 8 lines
migration/block-dirty-bitmap.c +1
-1@@ -747,7 +747,7 @@ static SaveVMHandlers savevm_dirty_bitmap_handlers = {void dirty_bitmap_mig_init (void ) {QSIMPLEQ_INIT (&dbm_state.save.dbms_list); -qemu_mutex_init (&dbm_state.load.finish_lock); +qemu_mutex_init (&dbm_state.load.lock); register_savevm_live("dirty-bitmap", 0, 1, &savevm_dirty_bitmap_handlers,
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 07/22] migration/block-dirty-bitmap: simplify dirty_bitmap_load_complete
In reply to Vladimir Sementsov-OgievskiyQuoted 18 lines · Vladimir Sementsov-Ogievskiy
bdrv_enable_dirty_bitmap_locked()call does nothing, as if we are in postcopy, bitmap successor must be enabled, and reclaim operation will enable the bitmap.So, actually we need just call _reclaim_ in both if branches, and making differences only to add an assertion seems not really good. The logic becomes simple: on load complete we do reclaim and that's all.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 25 ++++--------------------- 1 file changed, 4 insertions(+), 21 deletions(-)migration/block-dirty-bitmap.c +0
-0
migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 538
Quoted 4 lines
migration/block-dirty-bitmap.c +1
-0@@ -535,6 +535,10 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)qemu_mutex_lock (&s->lock); +if (bdrv_dirty_bitmap_has_successor (s->bitmap)) {
What about making it sure?
assert(!s->bitmap->successor->disabled);
Quoted 25 lines
migration/block-dirty-bitmap.c +3
-14+bdrv_reclaim_dirty_bitmap (s->bitmap, &error_abort); + } +for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;@@ -544,27 +548,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) } } -if (bdrv_dirty_bitmap_has_successor (s->bitmap)) { -bdrv_dirty_bitmap_lock (s->bitmap); -if (s->enabled_bitmaps ==NULL ) { -/* in postcopy */ -bdrv_reclaim_dirty_bitmap_locked (s->bitmap, &error_abort); -bdrv_enable_dirty_bitmap_locked (s->bitmap); - }else { -/* target not started, successor must be empty */ -int64_t count =bdrv_get_dirty_count (s->bitmap); -BdrvDirtyBitmap *ret =bdrv_reclaim_dirty_bitmap_locked (s->bitmap, -NULL ); -/* bdrv_reclaim_dirty_bitmap can fail only on no successor (it -* must be) or on merge fail, but merge can't fail when second -* bitmap is empty
migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 553
Quoted 10 lines
migration/block-dirty-bitmap.c +0
-7-*/ -assert (ret == s->bitmap && - count ==bdrv_get_dirty_count (s->bitmap)); - } -bdrv_dirty_bitmap_unlock (s->bitmap); - } -qemu_mutex_unlock (&s->lock); }
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 08/22] migration/block-dirty-bitmap: keep bitmap state for all bitmaps
In reply to Vladimir Sementsov-OgievskiyQuoted 136 lines · Vladimir Sementsov-Ogievskiy
Keep bitmap state for disabled bitmaps too. Keep the state until the end of the process. It's needed for the following commit to implement bitmap postcopy canceling.
To clean-up the new list the following logic is used:
We need two events to consider bitmap migration finished:
1. chunk withDIRTY_BITMAP_MIG_FLAG_COMPLETEflag should be received
2.dirty_bitmap_mig_before_vm_startshould be called
These two events may come in any order, so we understand which one is last, and on the last of them we remove bitmap migration state from the list.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 64 +++++++++++++++++++++++----------- 1 file changed, 43 insertions(+), 21 deletions(-)migration/block-dirty-bitmap.c +39
-21@@ -132,6 +132,7 @@ typedef struct LoadBitmapState {BlockDriverState *bs;BdrvDirtyBitmap *bitmap;bool migrated; +bool enabled; } LoadBitmapState;/* State of the dirty bitmap migration (DBM) during load process */ @@ -142,8 +143,10 @@ typedef struct DBMLoadState {BlockDriverState *bs;BdrvDirtyBitmap *bitmap; -GSList *enabled_bitmaps; -QemuMutex lock;/* protect enabled_bitmaps */ +bool before_vm_start_handled;/* set in dirty_bitmap_mig_before_vm_start */ + +GSList *bitmaps; +QemuMutex lock;/* protect bitmaps */ } DBMLoadState; typedef struct DBMState {@@ -458,6 +461,7 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s)Error *local_err =NULL ;uint32_t granularity =qemu_get_be32 (f);uint8_t flags =qemu_get_byte (f); +LoadBitmapState *b;if (s->bitmap) { error_report("Bitmap with the same name ('%s') already exists on "@@ -484,45 +488,59 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s)bdrv_disable_dirty_bitmap (s->bitmap);if (flags &DIRTY_BITMAP_MIG_START_FLAG_ENABLED ) { -LoadBitmapState *b; -bdrv_dirty_bitmap_create_successor (s->bitmap, &local_err);if (local_err) {error_report_err (local_err);return -EINVAL ; } - - b =g_new (LoadBitmapState,1 ); - b->bs = s->bs; - b->bitmap = s->bitmap; - b->migrated = false; - s->enabled_bitmaps =g_slist_prepend (s->enabled_bitmaps, b); } + b =g_new (LoadBitmapState,1 ); + b->bs = s->bs; + b->bitmap = s->bitmap; + b->migrated = false; + b->enabled = flags &DIRTY_BITMAP_MIG_START_FLAG_ENABLED , + + s->bitmaps =g_slist_prepend (s->bitmaps, b); +return 0 ; } -void dirty_bitmap_mig_before_vm_start (void ) +/* +* before_vm_start_handle_item +* +* g_slist_foreach helper +* +* item is LoadBitmapState* +* opaque is DBMLoadState* +*/ +static void before_vm_start_handle_item (void *item,void *opaque) { -DBMLoadState *s = &dbm_state.load; -GSList *item; - -qemu_mutex_lock (&s->lock); - -for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) { -LoadBitmapState *b = item->data; +DBMLoadState *s = opaque; +LoadBitmapState *b = item; +if (b->enabled) {if (b->migrated) {bdrv_enable_dirty_bitmap (b->bitmap); }else {bdrv_dirty_bitmap_enable_successor (b->bitmap); } + } +if (b->migrated) { + s->bitmaps =g_slist_remove (s->bitmaps, b);g_free (b); } +} -g_slist_free (s->enabled_bitmaps); - s->enabled_bitmaps =NULL ; +void dirty_bitmap_mig_before_vm_start (void ) +{ +DBMLoadState *s = &dbm_state.load; +qemu_mutex_lock (&s->lock); + +assert (!s->before_vm_start_handled); +g_slist_foreach (s->bitmaps, before_vm_start_handle_item, s); + s->before_vm_start_handled = true;qemu_mutex_unlock (&s->lock); }@@ -539,11 +557,15 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)bdrv_reclaim_dirty_bitmap (s->bitmap, &error_abort); } -for (item = s->enabled_bitmaps; item; item =g_slist_next (item)) { +for (item = s->bitmaps; item; item =g_slist_next (item)) {LoadBitmapState *b = item->data;
migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 571
Quoted 10 lines
migration/block-dirty-bitmap.c +4
-0if (b->bitmap == s->bitmap) { b->migrated = true; +if (s->before_vm_start_handled) { + s->bitmaps =g_slist_remove (s->bitmaps, b); +g_free (b); + }break ; } }
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 09/22] migration/block-dirty-bitmap: relax error handling in incoming part
In reply to Vladimir Sementsov-OgievskiyQuoted 64 lines · Vladimir Sementsov-Ogievskiy
Bitmaps data is not critical, and we should not fail the migration (or use postcopy recovering) because of dirty-bitmaps migration failure. Instead we should just lose unfinished bitmaps.
Still we have to report io stream violation errors, as they affect the whole migration stream.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 148 +++++++++++++++++++++++++-------- 1 file changed, 113 insertions(+), 35 deletions(-)migration/block-dirty-bitmap.c +37
-0@@ -145,6 +145,15 @@ typedef struct DBMLoadState {bool before_vm_start_handled;/* set in dirty_bitmap_mig_before_vm_start */ +/* +* cancelled +* Incoming migration is cancelled for some reason. That means that we +* still should read our chunks from migration stream, to not affect other +* migration objects (like RAM), but just ignore them and do not touch any +* bitmaps or nodes. +*/ +bool cancelled; +GSList *bitmaps;QemuMutex lock;/* protect bitmaps */ } DBMLoadState;@@ -545,13 +554,47 @@ void dirty_bitmap_mig_before_vm_start(void)qemu_mutex_unlock (&s->lock); } +static void cancel_incoming_locked (DBMLoadState *s) +{ +GSList *item; + +if (s->cancelled) { +return ; + } + + s->cancelled = true; + s->bs =NULL ; + s->bitmap =NULL ; + +/* Drop all unfinished bitmaps */ +for (item = s->bitmaps; item; item =g_slist_next (item)) { +LoadBitmapState *b = item->data; + +/* +* Bitmap must be unfinished, as finished bitmaps should already be +* removed from the list. +*/ +assert (!s->before_vm_start_handled || !b->migrated); +if (bdrv_dirty_bitmap_has_successor (b->bitmap)) { +bdrv_reclaim_dirty_bitmap (b->bitmap, &error_abort); + } +bdrv_release_dirty_bitmap (b->bitmap); + } + +g_slist_free_full (s->bitmaps, g_free);
migration/block-dirty-bitmap.c void dirty_bitmap_mig_before_vm_start(void) · line 554
Quoted 10 lines
migration/block-dirty-bitmap.c +3
-2+ s->bitmaps =NULL ; +} +static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s) {GSList *item;trace_dirty_bitmap_load_complete (); -bdrv_dirty_bitmap_deserialize_finish (s->bitmap); -qemu_mutex_lock (&s->lock);
Why is it safe to remove the critical section?
Quoted 59 lines
migration/block-dirty-bitmap.c +27
-7+if (s->cancelled) { +return ; + } + +bdrv_dirty_bitmap_deserialize_finish (s->bitmap);if (bdrv_dirty_bitmap_has_successor (s->bitmap)) {bdrv_reclaim_dirty_bitmap (s->bitmap, &error_abort);@@ -569,8 +612,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)break ; } } - -qemu_mutex_unlock (&s->lock); }static int dirty_bitmap_load_bits (QEMUFile *f,DBMLoadState *s)@@ -582,15 +623,32 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)if (s->flags &DIRTY_BITMAP_MIG_FLAG_ZEROES ) {trace_dirty_bitmap_load_bits_zeroes (); -bdrv_dirty_bitmap_deserialize_zeroes (s->bitmap, first_byte, nr_bytes, - false); +if (!s->cancelled) { +bdrv_dirty_bitmap_deserialize_zeroes (s->bitmap, first_byte, + nr_bytes, false); + } }else {size_t ret;uint8_t *buf;uint64_t buf_size =qemu_get_be64 (f); -uint64_t needed_size = -bdrv_dirty_bitmap_serialization_size (s->bitmap, - first_byte, nr_bytes); +uint64_t needed_size; + + buf =g_malloc (buf_size); + ret =qemu_get_buffer (f, buf, buf_size); +if (ret != buf_size) { +error_report ("Failed to read bitmap bits" ); +g_free (buf); +return -EIO ; + } + +if (s->cancelled) { +g_free (buf); +return 0 ; + } + + needed_size =bdrv_dirty_bitmap_serialization_size (s->bitmap, + first_byte, + nr_bytes);if (needed_size > buf_size || buf_size > QEMU_ALIGN_UP(needed_size, 4 * sizeof(long))@@ -599,15 +657,8 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)error_report ("Migrated bitmap granularity doesn't " "match the destination bitmap '%s' granularity" ,bdrv_dirty_bitmap_name (s->bitmap));
migration/block-dirty-bitmap.c static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s) · line 660
Quoted 10 lines
migration/block-dirty-bitmap.c +1
-9-return -EINVAL ; - } - - buf =g_malloc (buf_size); - ret =qemu_get_buffer (f, buf, buf_size); -if (ret != buf_size) { -error_report ("Failed to read bitmap bits" ); -g_free (buf); -return -EIO ; +cancel_incoming_locked (s);
/* Continue the VM migration as bitmaps data are not
critical */
Quoted 6 lines
migration/block-dirty-bitmap.c +1
-0+return 0 ; } bdrv_dirty_bitmap_deserialize_part(s->bitmap, buf, first_byte, nr_bytes,@@ -632,14 +683,16 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s)error_report ("Unable to read node name string" );
migration/block-dirty-bitmap.c static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) · line 689
Quoted 10 lines
migration/block-dirty-bitmap.c +4
-4return -EINVAL ; } - s->bs =bdrv_lookup_bs (s->node_name, s->node_name, &local_err); -if (!s->bs) { -error_report_err (local_err); -return -EINVAL ; +if (!s->cancelled) { + s->bs =bdrv_lookup_bs (s->node_name, s->node_name, &local_err); +if (!s->bs) { +error_report_err (local_err);
The error message may be supplemented with a report about the canceled bitmap migration. Also down there at cancel_incoming_locked(s).
Quoted 46 lines
migration/block-dirty-bitmap.c +19
-13+cancel_incoming_locked (s); + } } - }else if (!s->bs && !nothing) { + }else if (!s->bs && !nothing && !s->cancelled) {error_report ("Error: block device name is not set" ); -return -EINVAL ; +cancel_incoming_locked (s); }if (s->flags &DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME ) {@@ -647,24 +700,38 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s)error_report ("Unable to read bitmap name string" );return -EINVAL ; } - s->bitmap =bdrv_find_dirty_bitmap (s->bs, s->bitmap_name); - -/* bitmap may be NULL here, it wouldn't be an error if it is the -* first occurrence of the bitmap */ -if (!s->bitmap && !(s->flags &DIRTY_BITMAP_MIG_FLAG_START )) { -error_report ("Error: unknown dirty bitmap " -"'%s' for block device '%s'" , - s->bitmap_name, s->node_name); -return -EINVAL ; +if (!s->cancelled) { + s->bitmap =bdrv_find_dirty_bitmap (s->bs, s->bitmap_name); + +/* +* bitmap may be NULL here, it wouldn't be an error if it is the +* first occurrence of the bitmap +*/ +if (!s->bitmap && !(s->flags &DIRTY_BITMAP_MIG_FLAG_START )) { +error_report ("Error: unknown dirty bitmap " +"'%s' for block device '%s'" , + s->bitmap_name, s->node_name); +cancel_incoming_locked (s); + } } - }else if (!s->bitmap && !nothing) { + }else if (!s->bitmap && !nothing && !s->cancelled) {error_report ("Error: block device name is not set" ); -return -EINVAL ; +cancel_incoming_locked (s); }return 0 ;
migration/block-dirty-bitmap.c static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) · line 732
Quoted 10 lines
migration/block-dirty-bitmap.c +8
-0} +/* +* dirty_bitmap_load +* +* Load sequence of dirty bitmap chunks. Return error only on fatal io stream +* violations. On other errors just cancel bitmaps incoming migration and return +* 0. +* +* Note, than when incoming bitmap migration is canceled, we still must read all
"than (that)" may be omitted
Quoted 29 lines
migration/block-dirty-bitmap.c +9
-0+* our chunks (and just ignore them), to not affect other migration objects. +*/ static int dirty_bitmap_load (QEMUFile *f,void *opaque,int version_id) {DBMLoadState *s = &((DBMState *)opaque)->load;@@ -673,12 +740,19 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id)trace_dirty_bitmap_load_enter ();if (version_id !=1 ) { +qemu_mutex_lock (&s->lock); +cancel_incoming_locked (s); +qemu_mutex_unlock (&s->lock);return -EINVAL ; } do { + qemu_mutex_lock(&s->lock); + ret = dirty_bitmap_load_header(f, s); if (ret < 0) { + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return ret; }@@ -695,8 +769,12 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) }
migration/block-dirty-bitmap.c static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) · line 780
Quoted 10 lines
migration/block-dirty-bitmap.c +4
-0if (ret) { +cancel_incoming_locked (s); +qemu_mutex_unlock (&s->lock);return ret; } + +qemu_mutex_unlock (&s->lock); }while (!(s->flags &DIRTY_BITMAP_MIG_FLAG_EOS ));trace_dirty_bitmap_load_success ();
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 10/22] migration/block-dirty-bitmap: cancel migration on shutdown
In reply to Vladimir Sementsov-OgievskiyQuoted 35 lines · Vladimir Sementsov-Ogievskiy
If target is turned of prior to postcopy finished, target crashes because busy bitmaps are found at shutdown.
Canceling incoming migration helps, as it removes all unfinished (and therefore busy) bitmaps.Similarly on source we crash in
bdrv_close_allwhich asserts that all bdrv states are removed, because bdrv states involved into dirty bitmap migration are referenced by it. So, we need to cancel outgoing migration as well.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/migration.h | 2 ++ migration/block-dirty-bitmap.c | 16 ++++++++++++++++ migration/migration.c | 13 +++++++++++++ 3 files changed, 31 insertions(+)migration/migration.h +2
-0@@ -332,6 +332,8 @@ void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,void migrate_send_rp_resume_ack (MigrationIncomingState *mis,uint32_t value);void dirty_bitmap_mig_before_vm_start (void ); +void dirty_bitmap_mig_cancel_outgoing (void ); +void dirty_bitmap_mig_cancel_incoming (void );void migrate_add_address (SocketAddress *address);int foreach_not_ignored_block (RAMBlockIterFunc func,void *opaque);migration/block-dirty-bitmap.c +0
-0
migration/block-dirty-bitmap.c static void cancel_incoming_locked(DBMLoadState *s) · line 590
Quoted 6 lines
migration/block-dirty-bitmap.c +3
-0@@ -585,6 +585,22 @@ static void cancel_incoming_locked(DBMLoadState *s) s->bitmaps =NULL ; } +void dirty_bitmap_mig_cancel_outgoing (void ) +{ +dirty_bitmap_do_save_cleanup (&dbm_state.save);
The comment above the dirty_bitmap_do_save_cleanup() says:
"Called with iothread lock taken"
Quoted 31 lines
migration/block-dirty-bitmap.c +13
-0+} + +void dirty_bitmap_mig_cancel_incoming (void ) +{ +DBMLoadState *s = &dbm_state.load; + +qemu_mutex_lock (&s->lock); + +cancel_incoming_locked (s); + +qemu_mutex_unlock (&s->lock); +} +static void dirty_bitmap_load_complete (QEMUFile *f,DBMLoadState *s) {GSList *item;migration/migration.c +6
-0@@ -181,6 +181,19 @@ void migration_shutdown(void)*/ migrate_fd_cancel (current_migration);object_unref (OBJECT (current_migration)); + +/* +* Cancel outgoing migration of dirty bitmaps. It should +* at least unref used block nodes. +*/ +dirty_bitmap_mig_cancel_outgoing ();
migration/migration.c void migration_shutdown(void) · line 199
Quoted 10 lines
migration/migration.c +7
-0+ +/* +* Cancel incoming migration of dirty bitmaps. Dirty bitmaps +* are non-critical data, and their loss never considered as +* something serious. +*/ +dirty_bitmap_mig_cancel_incoming (); }/* For outgoing */
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 11/22] migration/savevm: don't worry if bitmap migration postcopy failed
In reply to Vladimir Sementsov-OgievskiyQuoted 33 lines · Vladimir Sementsov-Ogievskiy
First, if only bitmaps postcopy enabled (not ram postcopy)
postcopy_pause_incomingcrashes on assertionassert(mis->to_src_file).And anyway, bitmaps postcopy is not prepared to be somehow recovered. The original idea instead is that if bitmaps postcopy failed, we just loss some bitmaps, which is not critical. So, on failure we just need to remove unfinished bitmaps and guest should continue execution on destination.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/savevm.c | 37 ++++++++++++++++++++++++++++++++----- 1 file changed, 32 insertions(+), 5 deletions(-)migration/savevm.c +3
-1@@ -1812,6 +1812,9 @@ static void *postcopy_ram_listen_thread(void *opaque)MigrationIncomingState *mis =migration_incoming_get_current ();QEMUFile *f = mis->from_src_file;int load_res; +MigrationState *migr =migrate_get_current (); + +object_ref (OBJECT (migr));migrate_set_state (&mis->state,MIGRATION_STATUS_ACTIVE ,MIGRATION_STATUS_POSTCOPY_ACTIVE );@@ -1838,11 +1841,24 @@ static void *postcopy_ram_listen_thread(void *opaque)trace_postcopy_ram_listen_thread_exit ();if (load_res <0 ) { -error_report ("%s: loadvm failed: %d" , __func__, load_res);
migration/savevm.c static void *postcopy_ram_listen_thread(void *opaque) · line 1850
Quoted 10 lines
migration/savevm.c +6
-3qemu_file_set_error (f, load_res); -migrate_set_state (&mis->state,MIGRATION_STATUS_POSTCOPY_ACTIVE , -MIGRATION_STATUS_FAILED ); - }else { +dirty_bitmap_mig_cancel_incoming (); +if (postcopy_state_get () ==POSTCOPY_INCOMING_RUNNING && + !migrate_postcopy_ram () &&migrate_dirty_bitmaps ()) + { +error_report ("%s: loadvm failed during postcopy: %d. All state is " + "migrated except for dirty bitmaps. Some dirty "
"All states migrated except dirty bitmaps"
Quoted 47 lines
migration/savevm.c +19
-0+"bitmaps may be lost, and present migrated dirty " +"bitmaps are correctly migrated and valid." , + __func__, load_res); + load_res =0 ;/* prevent further exit() */ + }else { +error_report ("%s: loadvm failed: %d" , __func__, load_res); +migrate_set_state (&mis->state,MIGRATION_STATUS_POSTCOPY_ACTIVE , +MIGRATION_STATUS_FAILED ); + } + } +if (load_res >=0 ) {/* * This looks good, but it's possible that the device loading in the * main thread hasn't finished yet, and so we might not be in 'RUN' @@ -1878,6 +1894,8 @@ static void *postcopy_ram_listen_thread(void *opaque) mis->have_listen_thread = false;postcopy_state_set (POSTCOPY_INCOMING_END ); +object_unref (OBJECT (migr)); +return NULL ; }@@ -2429,6 +2447,8 @@ static bool postcopy_pause_incoming(MigrationIncomingState *mis) {trace_postcopy_pause_incoming (); +assert (migrate_postcopy_ram ()); +/* Clear the triggered bit to allow one recovery */ mis->postcopy_recover_triggered = false;@@ -2513,15 +2533,22 @@ out:if (ret <0 ) {qemu_file_set_error (f, ret); +/* Cancel bitmaps incoming regardless of recovery */ +dirty_bitmap_mig_cancel_incoming (); +/* * If we are during an active postcopy, then we pause instead * of bail out to at least keep the VM's dirty data. Note * that POSTCOPY_INCOMING_LISTENING stage is still not enough, * during which we're still receiving device states and we * still haven't yet started the VM on destination. +*
migration/savevm.c out: · line 2554
Quoted 10 lines
migration/savevm.c +4
-1+* Only RAM postcopy supports recovery. Still, if RAM postcopy is +* enabled, canceled bitmaps postcopy will not affect RAM postcopy +* recovering. */ if (postcopy_state_get () ==POSTCOPY_INCOMING_RUNNING && -postcopy_pause_incoming (mis)) { +migrate_postcopy_ram () &&postcopy_pause_incoming (mis)) {/* Reset f to point to the newly created channel */ f = mis->from_src_file; goto retry;
Signature · 2 lines
With the best regards, Andrey Shinkevich
qemu-iotests:$ ./check -qcow2
PASSED
(except always failed 261 and 272)
Andrey
Quoted 64 lines · Vladimir Sementsov-Ogievskiy
Original idea of bitmaps postcopy migration is that bitmaps are non critical data, and their loss is not serious problem. So, using postcopy method on any failure we should just drop unfinished bitmaps and continue guest execution.
However, it doesn't work so. It crashes, fails, it goes to postcopy-recovery feature. It does anything except for behavior we want. These series fixes at least some problems with error handling during bitmaps migration postcopy.
v1 was "[PATCH 0/7] Fix crashes on early shutdown during bitmaps postcopy"
v2:
Most of patches are new or changed a lot.
Only patches 06,07 mostly unchanged, just rebased on refactorings.Vladimir Sementsov-Ogievskiy (22):
migration/block-dirty-bitmap: fixdirty_bitmap_mig_before_vm_startmigration/block-dirty-bitmap: rename state structure types migration/block-dirty-bitmap: renamedirty_bitmap_mig_cleanupmigration/block-dirty-bitmap: move mutex init todirty_bitmap_mig_initmigration/block-dirty-bitmap: refactor state global variables migration/block-dirty-bitmap: renamefinish_lockto just lock migration/block-dirty-bitmap: simplifydirty_bitmap_load_completemigration/block-dirty-bitmap: keep bitmap state for all bitmaps migration/block-dirty-bitmap: relax error handling in incoming part migration/block-dirty-bitmap: cancel migration on shutdown migration/savevm: don't worry if bitmap migration postcopy failed qemu-iotests/199: fix style
qemu-iotests/199: drop extra constraints
qemu-iotests/199: better catch postcopy time qemu-iotests/199: improve performance: set bitmap by discard qemu-iotests/199: change discard patterns qemu-iotests/199: increase postcopy period python/qemu/machine: addkill()method
qemu-iotests/199: prepare for new test-cases addition qemu-iotests/199: check persistent bitmaps qemu-iotests/199: add early shutdown case to bitmaps postcopy qemu-iotests/199: add source-killed case to bitmaps postcopyCc John Snow <[email protected]>Cc Vladimir Sementsov-Ogievskiy <[email protected]>Cc Stefan Hajnoczi <[email protected]>Cc Fam Zheng <[email protected]>Cc Juan Quintela <[email protected]>Cc "Dr. David Alan Gilbert" <[email protected]>Cc Eduardo Habkost <[email protected]>Cc Cleber Rosa <[email protected]>Cc Kevin Wolf <[email protected]>Cc Max Reitz <[email protected]>Cc [email protected] # for patch 01migration/migration.h | 3 +- migration/block-dirty-bitmap.c | 444 +++++++++++++++++++++------------ migration/migration.c | 15 +- migration/savevm.c | 37 ++- python/qemu/machine.py | 12 +- tests/qemu-iotests/199 | 244 ++++++++++++++---- tests/qemu-iotests/199.out | 4 +- 7 files changed, 529 insertions(+), 230 deletions(-)
Signature · 2 lines
With the best regards, Andrey Shinkevich
Quoted 3 lines · Andrey Shinkevich
qemu-iotests:$ ./check -qcow2
PASSED
(except always failed 261 and 272)
Have you reported those failures on the threads that introduced those tests?
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 12/22] qemu-iotests/199: fix style
In reply to Vladimir Sementsov-OgievskiyQuoted 50 lines · Vladimir Sementsov-Ogievskiy
Mostly, satisfy pep8 complains.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)tests/qemu-iotests/199 +5
-5@@ -28,8 +28,8 @@ disk_b = os.path.join(iotests.test_dir, 'disk_b') size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') -class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): +class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def tearDown(self): self.vm_a.shutdown() self.vm_b.shutdown()@@ -54,7 +54,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap', granularity=granularity) - self.assert_qmp(result, 'return', {}); + self.assert_qmp(result, 'return', {}) s = 0 while s < write_size:@@ -71,7 +71,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap') - self.assert_qmp(result, 'return', {}); + self.assert_qmp(result, 'return', {}) s = 0 while s < write_size: self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk))@@ -104,15 +104,16 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) s += 0x10000 - result = self.vm_b.qmp('query-block'); + result = self.vm_b.qmp('query-block') while len(result['return'][0]['dirty-bitmaps']) > 1: time.sleep(2) - result = self.vm_b.qmp('query-block'); + result = self.vm_b.qmp('query-block')
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 119
Quoted 10 lines
tests/qemu-iotests/199 +2
-1result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap') - self.assert_qmp(result, 'return/sha256', sha256); + self.assert_qmp(result, 'return/sha256', sha256) + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2'], supported_cache_modes=['none'],
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 13/22] qemu-iotests/199: drop extra constraints
In reply to Vladimir Sementsov-OgievskiyQuoted 16 lines · Vladimir Sementsov-Ogievskiy
We don't need any specific format constraints here. Still keep qcow2 for two reasons:
1. No extra calls of format-unrelated test
2. Add some check around persistent bitmap in future (require qcow2)Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)tests/qemu-iotests/199 +0
-0
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 119
Quoted 6 lines
tests/qemu-iotests/199 +1
-2@@ -116,5 +116,4 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): if __name__ == '__main__': - iotests.main(supported_fmts=['qcow2'], supported_cache_modes=['none'], - supported_protocols=['file']) + iotests.main(supported_fmts=['qcow2'])
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 14/22] qemu-iotests/199: better catch postcopy time
In reply to Vladimir Sementsov-OgievskiyQuoted 43 lines · Vladimir Sementsov-Ogievskiy
The test aims to test _postcopy_ migration, and wants to do some write operations during postcopy time.
Test considers migrate status=complete event on source as start of postcopy. This is completely wrong, completion is completion of the whole migration process. Let's instead consider destination start as start of postcopy, and use RESUME event for it.
Next, as migration finish, let's use migration status=complete event on target, as such method is closer to what libvirt or another user will do, than tracking number of dirty-bitmaps.
Finally, add a possibility to dump events for debug. And if set debug to True, we see, that actual postcopy period is very small relatively to the whole test duration time (~0.2 seconds to >40 seconds for me). This means, that test is very inefficient in what it supposed to do. Let's improve it in following commits.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 72 +++++++++++++++++++++++++++++++++--------- 1 file changed, 57 insertions(+), 15 deletions(-)tests/qemu-iotests/199 +3
-1@@ -20,17 +20,43 @@ import os import iotests -import time from iotests import qemu_img +debug = False + disk_a = os.path.join(iotests.test_dir, 'disk_a') disk_b = os.path.join(iotests.test_dir, 'disk_b') size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +def event_seconds(event):
tests/qemu-iotests/199 line 43
Quoted 10 lines
tests/qemu-iotests/199 +8
-0+ return event['timestamp']['seconds'] + \ + event['timestamp']['microseconds'] / 1000000.0 + + +def event_dist(e1, e2): + return event_seconds(e2) - event_seconds(e1) + + class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def tearDown(self):
It's common to put the definition of setUp() ahead
Quoted 31 lines
tests/qemu-iotests/199 +20
-0+ if debug: + self.vm_a_events += self.vm_a.get_qmp_events() + self.vm_b_events += self.vm_b.get_qmp_events() + for e in self.vm_a_events: + e['vm'] = 'SRC' + for e in self.vm_b_events: + e['vm'] = 'DST' + events = (self.vm_a_events + self.vm_b_events) + events = [(e['timestamp']['seconds'], + e['timestamp']['microseconds'], + e['vm'], + e['event'], + e.get('data', '')) for e in events] + for e in sorted(events): + print('{}.{:06} {} {} {}'.format(*e)) + self.vm_a.shutdown() self.vm_b.shutdown() os.remove(disk_a)@@ -47,6 +73,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.launch() self.vm_b.launch() + # collect received events for debug + self.vm_a_events = [] + self.vm_b_events = [] + def test_postcopy(self): write_size = 0x40000000 granularity = 512
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 110
Quoted 6 lines
tests/qemu-iotests/199 +1
-2@@ -77,15 +107,13 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) s += 0x10000 - bitmaps_cap = {'capability': 'dirty-bitmaps', 'state': True} - events_cap = {'capability': 'events', 'state': True} + caps = [{'capability': 'dirty-bitmaps', 'state': True},
The name "capabilities" would be an appropriate identifier.
Quoted 14 lines
tests/qemu-iotests/199 +3
-4+ {'capability': 'events', 'state': True}] - result = self.vm_a.qmp('migrate-set-capabilities', - capabilities=[bitmaps_cap, events_cap]) + result = self.vm_a.qmp('migrate-set-capabilities', capabilities=caps) self.assert_qmp(result, 'return', {}) - result = self.vm_b.qmp('migrate-set-capabilities', - capabilities=[bitmaps_cap]) + result = self.vm_b.qmp('migrate-set-capabilities', capabilities=caps) self.assert_qmp(result, 'return', {}) result = self.vm_a.qmp('migrate', uri='exec:cat>' + fifo)
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 125
Quoted 8 lines
tests/qemu-iotests/199 +1
-4@@ -94,24 +122,38 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('migrate-start-postcopy') self.assert_qmp(result, 'return', {}) - while True: - event = self.vm_a.event_wait('MIGRATION') - if event['data']['status'] == 'completed': - break + e_resume = self.vm_b.event_wait('RESUME')
"event_resume" gives a faster understanding
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 134
Quoted 9 lines
tests/qemu-iotests/199 +3
-0+ self.vm_b_events.append(e_resume) s = 0x8000 while s < write_size: self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) s += 0x10000 + match = {'data': {'status': 'completed'}} + e_complete = self.vm_b.event_wait('MIGRATION', match=match)
"event_complete" also
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 138
Quoted 4 lines
tests/qemu-iotests/199 +4
-0+ self.vm_b_events.append(e_complete) + + # take queued event, should already been happened + e_stop = self.vm_a.event_wait('STOP')
"event_stop"
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 144
Quoted 6 lines
tests/qemu-iotests/199 +6
-0+ self.vm_a_events.append(e_stop) + + downtime = event_dist(e_stop, e_resume) + postcopy_time = event_dist(e_resume, e_complete) + + # TODO: assert downtime * 10 < postcopy_time
I got the results below in debug mode:
downtime: 6.194924831390381postcopy_time: 0.1592559814453125
1582102669.764919 SRC MIGRATION {'status': 'setup'} 1582102669.766179 SRC MIGRATION_PASS {'pass': 1} 1582102669.766234 SRC MIGRATION {'status': 'active'} 1582102669.768058 DST MIGRATION {'status': 'active'} 1582102669.801422 SRC MIGRATION {'status': 'postcopy-active'} 1582102669.801510 SRC STOP
1582102675.990041 DST MIGRATION {'status': 'postcopy-active'} 1582102675.996435 DST RESUME
1582102676.111313 SRC MIGRATION {'status': 'completed'} 1582102676.155691 DST MIGRATION {'status': 'completed'}
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 145
Quoted 1 line
tests/qemu-iotests/199 +1
-0+ if debug:
with no usage in the following patches, you can put the whole block of relative code above under the "if debug: section
Quoted 9 lines
tests/qemu-iotests/199 +5
-2+ print('downtime:', downtime) + print('postcopy_time:', postcopy_time) + + # Assert that bitmap migration is finished (check that successor bitmap + # is removed) result = self.vm_b.qmp('query-block') - while len(result['return'][0]['dirty-bitmaps']) > 1: - time.sleep(2)
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 159
Quoted 10 lines
tests/qemu-iotests/199 +2
-2- result = self.vm_b.qmp('query-block') + assert len(result['return'][0]['dirty-bitmaps']) == 1 + # Check content of migrated (and updated by new writes) bitmap result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap') - self.assert_qmp(result, 'return/sha256', sha256)
Signature · 2 lines
With the best regards, Andrey Shinkevich
Quoted 7 lines · Eric Blake
Quoted 3 lines · Andrey Shinkevich
qemu-iotests:$ ./check -qcow2
PASSED
(except always failed 261 and 272)Have you reported those failures on the threads that introduced those tests?
Not yet unfortunately. I have not investigated the case. "$ ./check -qcow2 261" dumps
+od: unrecognized option '--endian=big'
+Try 'od --help' for more information.
+od: invalid -N argument '--endian=big'
+qemu-img: Could not open 'TEST_DIR/t.IMGFMT': IMGFMT header exceeds cluster size
and "$ ./check -qcow2 272" dumps
+od: unrecognized option '--endian=big'
+Try 'od --help' for more information.
+od: invalid -N argument '--endian=big'
+qemu-io: can't open device .../qemu/tests/qemu-iotests/scratch/t.qcow2: Image is not in qcow2 format
Signature · 2 lines
With the best regards, Andrey Shinkevich
Quoted 18 lines · Andrey Shinkevich
Quoted 7 lines · Eric Blake
Quoted 3 lines · Andrey Shinkevich
qemu-iotests:$ ./check -qcow2
PASSED
(except always failed 261 and 272)Have you reported those failures on the threads that introduced those tests?
Not yet unfortunately. I have not investigated the case. "$ ./check -qcow2 261" dumps
(quoted diff) +4
-0+od: unrecognized option '--endian=big' +Try 'od --help' for more information. +od: invalid -N argument '--endian=big' +qemu-img: Could not open 'TEST_DIR/t.IMGFMT': IMGFMT header exceeds
Quoted 1 line
cluster size
Which version of od are you using? I do recall wondering whether reliance on the GNU coreutils extension --endian=big was going to cause problems later - well, here we are, it's later :)
https://lists.gnu.org/archive/html/qemu-devel/2019-07/msg06781.html
Review
Quoted 6 lines
and "$ ./check -qcow2 272" dumps
(quoted diff) +3
-0+od: unrecognized option '--endian=big' +Try 'od --help' for more information. +od: invalid -N argument '--endian=big'
Yay, same problem for both tests. Fix common.rc once, and both tests should start working for you.
Quoted 3 lines
+qemu-io: can't open device .../qemu/tests/qemu-iotests/scratch/t.qcow2:
Image is not in qcow2 format
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Quoted 15 lines · Eric Blake
Quoted 14 lines · Andrey Shinkevich
Quoted 7 lines · Eric Blake
Quoted 3 lines · Andrey Shinkevich
qemu-iotests:$ ./check -qcow2
PASSED
(except always failed 261 and 272)Have you reported those failures on the threads that introduced those tests?
Not yet unfortunately. I have not investigated the case. "$ ./check -qcow2 261" dumps
Quoted 7 lines
Quoted 5 lines
(quoted diff) +4
-0+od: unrecognized option '--endian=big' +Try 'od --help' for more information. +od: invalid -N argument '--endian=big' +qemu-img: Could not open 'TEST_DIR/t.IMGFMT': IMGFMT header exceedscluster size
Which version of od are you using? I do recall wondering whether
$ od --version od (GNU coreutils) 8.22 Copyright (C) 2013 Free Software Foundation, Inc. License GPLv3+: GNU GPL version 3 or later...
Quoted 5 lines
reliance on the GNU coreutils extension --endian=big was going to cause problems later - well, here we are, it's later :)
https://lists.gnu.org/archive/html/qemu-devel/2019-07/msg06781.html
Quoted 1 line
Quoted 7 lines
Review
Quoted 5 lines
and "$ ./check -qcow2 272" dumps
(quoted diff) +3
-0+od: unrecognized option '--endian=big' +Try 'od --help' for more information. +od: invalid -N argument '--endian=big'Yay, same problem for both tests. Fix common.rc once, and both tests should start working for you.
Thank you Eric! I want to sort it out later...
Quoted 5 lines
Quoted 4 lines
+qemu-io: can't open device .../qemu/tests/qemu-iotests/scratch/t.qcow2: Image is not in qcow2 format
Signature · 2 lines
With the best regards, Andrey Shinkevich
Quoted 9 lines · Eric Blake
Quoted 5 lines
(quoted diff) +4
-0+od: unrecognized option '--endian=big' +Try 'od --help' for more information. +od: invalid -N argument '--endian=big' +qemu-img: Could not open 'TEST_DIR/t.IMGFMT': IMGFMT header exceedscluster size
Which version of od are you using? I do recall wondering whether reliance on the GNU coreutils extension --endian=big was going to cause problems later - well, here we are, it's later :)
https://lists.gnu.org/archive/html/qemu-devel/2019-07/msg06781.html
coreutils documents that od --endian was added in 8.23, released in 2014-07-18. Per https://wiki.qemu.org/Supported_Build_Platforms, we still have support for RHEL 7 through 2022, and it was first released 2014-06-09 (all other supported distros have newer releases, but I didn't check what coreutils version they included, or even if the BSD builds which don't use coreutils would also be impacted by this problem). Still, I'd like to know your specific setup, and why the CI tools have not flagged it.
But even one counterexample within the bounds of our supported distro page is a good argument that use of od --endian is not yet portable. Or, if your setup is not on the supported page, it becomes a question of whether it should be added or whether you should upgrade to something that is supported.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 15/22] qemu-iotests/199: improve performance: set bitmap by discard
In reply to Vladimir Sementsov-OgievskiyQuoted 75 lines · Vladimir Sementsov-Ogievskiy
Discard dirties dirty-bitmap as well as write, but works faster. Let's use it instead.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 31 ++++++++++++++++++++----------- 1 file changed, 20 insertions(+), 11 deletions(-)tests/qemu-iotests/199 +18
-9@@ -67,8 +67,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): os.mkfifo(fifo) qemu_img('create', '-f', iotests.imgfmt, disk_a, size) qemu_img('create', '-f', iotests.imgfmt, disk_b, size) - self.vm_a = iotests.VM(path_suffix='a').add_drive(disk_a) - self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b) + self.vm_a = iotests.VM(path_suffix='a').add_drive(disk_a, + 'discard=unmap') + self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b, + 'discard=unmap') self.vm_b.add_incoming("exec: cat '" + fifo + "'") self.vm_a.launch() self.vm_b.launch()@@ -78,7 +80,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b_events = [] def test_postcopy(self): - write_size = 0x40000000 + discard_size = 0x40000000 granularity = 512 chunk = 4096@@ -86,25 +88,32 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): name='bitmap', granularity=granularity) self.assert_qmp(result, 'return', {}) + result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', + node='drive0', name='bitmap') + empty_sha256 = result['return']['sha256'] + s = 0 - while s < write_size: - self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 s = 0x8000 - while s < write_size: - self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap') sha256 = result['return']['sha256'] + # Check, that updating the bitmap by discards works + assert sha256 != empty_sha256 + result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap') self.assert_qmp(result, 'return', {}) s = 0 - while s < write_size: - self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 caps = [{'capability': 'dirty-bitmaps', 'state': True},
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 142
Quoted 10 lines
tests/qemu-iotests/199 +2
-2@@ -126,8 +135,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b_events.append(e_resume) s = 0x8000 - while s < write_size: - self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk)) + while s < discard_size: + self.vm_b.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) s += 0x10000 match = {'data': {'status': 'completed'}}
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 03/22] migration/block-dirty-bitmap: rename dirty_bitmap_mig_cleanup
In reply to Andrey ShinkevichQuoted 39 lines · Andrey Shinkevich
Quoted 38 lines · Vladimir Sementsov-Ogievskiy
Rename
dirty_bitmap_mig_cleanuptodirty_bitmap_do_save_cleanup, to stress that it is on save part.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 8 ++++----1 file changed, 4 insertions(+), 4 deletions(-)migration/block-dirty-bitmap.c +1
-1@@ -259,7 +259,7 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms, } /* Called with iothread lock taken. */ -static void dirty_bitmap_mig_cleanup (void ) +static void dirty_bitmap_do_save_cleanup (void ) { SaveBitmapState *dbms;@@ -338,7 +338,7 @@ static int init_dirty_bitmap_migration(void) return 0;fail:
-dirty_bitmap_mig_cleanup();
+dirty_bitmap_do_save_cleanup();
return -1;
}migration/block-dirty-bitmap.c +2
-2@@ -377,7 +377,7 @@ static void bulk_phase(QEMUFile *f, bool limit) /* for SaveVMHandlers */ static void dirty_bitmap_save_cleanup(void *opaque) { - dirty_bitmap_mig_cleanup(); + dirty_bitmap_do_save_cleanup(); } static int dirty_bitmap_save_iterate(QEMUFile *f, void *opaque)@@ -412,7 +412,7 @@ static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque) trace_dirty_bitmap_save_complete_finish(); - dirty_bitmap_mig_cleanup(); + dirty_bitmap_do_save_cleanup();
Quoted 8 lines
migration/block-dirty-bitmap.c static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque) · line 418
Quoted 3 lines
migration/block-dirty-bitmap.c +0
-0return 0; }At the next opportunity, I would suggest the name like "
dirty_bitmap_do_clean_after_saving()"
and similar fordirty_bitmap_save_cleanup()"dirty_bitmap_clean_after_saving()".
I'd keep my naming, it corresponds to .save_cleanup handler name.
Quoted 2 lines
Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 16/22] qemu-iotests/199: change discard patterns
In reply to Vladimir Sementsov-OgievskiyQuoted 1 line · Vladimir Sementsov-Ogievskiy
iotest 40 works too long because of many discard opertion. On the same
operations
At the same time
Quoted 27 lines
time, postcopy period is very short, in spite of all these efforts.
So, let's use less discards (and with more interesting patterns) to reduce test timing. In the next commit we'll increase postcopy period.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 44 +++++++++++++++++++++++++----------------- 1 file changed, 26 insertions(+), 18 deletions(-)tests/qemu-iotests/199 +9
-0@@ -30,6 +30,28 @@ size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +GiB = 1024 * 1024 * 1024 + +discards1 = ( + (0, GiB), + (2 * GiB + 512 * 5, 512), + (3 * GiB + 512 * 5, 512), + (100 * GiB, GiB) +) +
tests/qemu-iotests/199 size = '256G' · line 51
Quoted 10 lines
tests/qemu-iotests/199 +10
-0+discards2 = ( + (3 * GiB + 512 * 8, 512), + (4 * GiB + 512 * 8, 512), + (50 * GiB, GiB), + (100 * GiB + GiB // 2, GiB) +) + + +def apply_discards(vm, discards): + for d in discards:
If we run qemu-io only once, it will update the bitmap state and will speed the test performance up. Is that wrong idea?
Quoted 48 lines
tests/qemu-iotests/199 +6
-14+ vm.hmp_qemu_io('drive0', 'discard {} {}'.format(*d)) + + def event_seconds(event): return event['timestamp']['seconds'] + \ event['timestamp']['microseconds'] / 1000000.0@@ -80,9 +102,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_b_events = [] def test_postcopy(self): - discard_size = 0x40000000 granularity = 512 - chunk = 4096 result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap', granularity=granularity)@@ -92,14 +112,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): node='drive0', name='bitmap') empty_sha256 = result['return']['sha256'] - s = 0 - while s < discard_size: - self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 - s = 0x8000 - while s < discard_size: - self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + apply_discards(self.vm_a, discards1 + discards2) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap')@@ -111,10 +124,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap') self.assert_qmp(result, 'return', {}) - s = 0 - while s < discard_size: - self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + + apply_discards(self.vm_a, discards1) caps = [{'capability': 'dirty-bitmaps', 'state': True}, {'capability': 'events', 'state': True}]@@ -134,10 +145,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): e_resume = self.vm_b.event_wait('RESUME')
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 151
Quoted 10 lines
tests/qemu-iotests/199 +1
-4self.vm_b_events.append(e_resume) - s = 0x8000 - while s < discard_size: - self.vm_b.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + apply_discards(self.vm_b, discards2) match = {'data': {'status': 'completed'}} e_complete = self.vm_b.event_wait('MIGRATION', match=match)
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 16/22] qemu-iotests/199: change discard patterns
Quoted 35 lines · Andrey Shinkevich
Quoted 1 line · Vladimir Sementsov-Ogievskiy
iotest 40 works too long because of many discard opertion. On the same
operations
At the same timeQuoted 29 lines
time, postcopy period is very short, in spite of all these efforts.
So, let's use less discards (and with more interesting patterns) to reduce test timing. In the next commit we'll increase postcopy period.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
tests/qemu-iotests/199 | 44 +++++++++++++++++++++++++-----------------1 file changed, 26 insertions(+), 18 deletions(-)tests/qemu-iotests/199 +13
-0@@ -30,6 +30,28 @@ size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +GiB = 1024 * 1024 * 1024 + +discards1 = ( + (0, GiB), + (2 * GiB + 512 * 5, 512), + (3 * GiB + 512 * 5, 512), + (100 * GiB, GiB) +) + +discards2 = ( + (3 * GiB + 512 * 8, 512), + (4 * GiB + 512 * 8, 512), + (50 * GiB, GiB),
Quoted 8 lines
tests/qemu-iotests/199 size = '256G' · line 51
Quoted 6 lines
tests/qemu-iotests/199 +6
-0+ (100 * GiB + GiB // 2, GiB) +) + + +def apply_discards(vm, discards): + for d in discards:If we run qemu-io only once, it will update the bitmap state and will speed the test performance up. Is that wrong idea?
Yes, that is. I have seen with the later review.
Andrey
Quoted 62 lines
Quoted 51 lines
tests/qemu-iotests/199 +3
-0+ vm.hmp_qemu_io('drive0', 'discard {} {}'.format(*d)) + +def
event_seconds(event):
return event['timestamp']['seconds'] + \
event['timestamp']['microseconds'] / 1000000.0tests/qemu-iotests/199 +0
-0@@ -80,9 +102,7 @@ classTestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):
self.vm_b_events = []
deftest_postcopy(self):
-discard_size=0x40000000
granularity = 512
- chunk = 4096
result =self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap', granularity=granularity)tests/qemu-iotests/199 +0
-0@@ -92,14 +112,7 @@ classTestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):
node='drive0', name='bitmap')empty_sha256= result['return']['sha256']
- s = 0
- while s <discard_size:
-self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk))
- s +=0x10000
- s =0x8000
- while s <discard_size:
-self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk))
- s +=0x10000
+apply_discards(self.vm_a, discards1 + discards2)result =self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap')tests/qemu-iotests/199 +0
-0@@ -111,10 +124,8 @@ classTestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):
result =self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap')self.assert_qmp(result, 'return', {})
- s = 0
- while s <discard_size:
-self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk))
- s +=0x10000
+
+apply_discards(self.vm_a, discards1)
caps = [{'capability': 'dirty-bitmaps', 'state': True}, {'capability': 'events', 'state': True}]tests/qemu-iotests/199 +0
-0@@ -134,10 +145,7 @@ classTestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):
e_resume=self.vm_b.event_wait('RESUME')self.vm_b_events.append(e_resume)Quoted 9 lines
- s =
0x8000
- while s <discard_size:
-self.vm_b.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk))
- s +=0x10000
+apply_discards(self.vm_b, discards2)
match = {'data': {'status': 'completed'}}e_complete=self.vm_b.event_wait('MIGRATION', match=match)Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 17/22] qemu-iotests/199: increase postcopy period
In reply to Vladimir Sementsov-OgievskiyQuoted 101 lines · Vladimir Sementsov-Ogievskiy
Test wants force bitmap postcopy. Still, resulting postcopy period is very small. Let's increase it by adding more bitmaps to migrate. Also, test disabled bitmaps migration.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 58 ++++++++++++++++++++++++++++-------------- 1 file changed, 39 insertions(+), 19 deletions(-)tests/qemu-iotests/199 +32
-19@@ -103,29 +103,45 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def test_postcopy(self): granularity = 512 + nb_bitmaps = 15 - result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', - name='bitmap', granularity=granularity) - self.assert_qmp(result, 'return', {}) + for i in range(nb_bitmaps): + result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', + name='bitmap{}'.format(i), + granularity=granularity) + self.assert_qmp(result, 'return', {}) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', - node='drive0', name='bitmap') + node='drive0', name='bitmap0') empty_sha256 = result['return']['sha256'] - apply_discards(self.vm_a, discards1 + discards2) + apply_discards(self.vm_a, discards1) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', - node='drive0', name='bitmap') - sha256 = result['return']['sha256'] + node='drive0', name='bitmap0') + discards1_sha256 = result['return']['sha256'] # Check, that updating the bitmap by discards works - assert sha256 != empty_sha256 + assert discards1_sha256 != empty_sha256 - result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', - name='bitmap') - self.assert_qmp(result, 'return', {}) + # We want to calculate resulting sha256. Do it in bitmap0, so, disable + # other bitmaps + for i in range(1, nb_bitmaps): + result = self.vm_a.qmp('block-dirty-bitmap-disable', node='drive0', + name='bitmap{}'.format(i)) + self.assert_qmp(result, 'return', {}) - apply_discards(self.vm_a, discards1) + apply_discards(self.vm_a, discards2) + + result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', + node='drive0', name='bitmap0') + all_discards_sha256 = result['return']['sha256'] + + # Now, enable some bitmaps, to be updated during migration + for i in range(2, nb_bitmaps, 2): + result = self.vm_a.qmp('block-dirty-bitmap-enable', node='drive0', + name='bitmap{}'.format(i)) + self.assert_qmp(result, 'return', {}) caps = [{'capability': 'dirty-bitmaps', 'state': True}, {'capability': 'events', 'state': True}]@@ -145,6 +161,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): e_resume = self.vm_b.event_wait('RESUME') self.vm_b_events.append(e_resume) + # enabled bitmaps should be updated apply_discards(self.vm_b, discards2) match = {'data': {'status': 'completed'}}@@ -158,7 +175,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): downtime = event_dist(e_stop, e_resume) postcopy_time = event_dist(e_resume, e_complete) - # TODO: assert downtime * 10 < postcopy_time + assert downtime * 10 < postcopy_time if debug: print('downtime:', downtime) print('postcopy_time:', postcopy_time)@@ -166,12 +183,15 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): # Assert that bitmap migration is finished (check that successor bitmap # is removed) result = self.vm_b.qmp('query-block') - assert len(result['return'][0]['dirty-bitmaps']) == 1 - - # Check content of migrated (and updated by new writes) bitmap - result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', - node='drive0', name='bitmap') - self.assert_qmp(result, 'return/sha256', sha256) + assert len(result['return'][0]['dirty-bitmaps']) == nb_bitmaps +
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 197
Quoted 10 lines
tests/qemu-iotests/199 +7
-0+ # Check content of migrated bitmaps. Still, don't waste time checking + # every bitmap + for i in range(0, nb_bitmaps, 5): + result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', + node='drive0', name='bitmap{}'.format(i)) + sha256 = discards1_sha256 if i % 2 else all_discards_sha256 + self.assert_qmp(result, 'return/sha256', sha256) if __name__ == '__main__':
The updated test passed.
Signature · 2 lines
With the best regards, Andrey Shinkevich
Quoted 7 lines · Andrey Shinkevich
Quoted 5 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+od: unrecognized option '--endian=big' +Try 'od --help' for more information. +od: invalid -N argument '--endian=big'Yay, same problem for both tests. Fix common.rc once, and both tests should start working for you.
Thank you Eric! I want to sort it out later...
Patch proposed:
https://lists.gnu.org/archive/html/qemu-devel/2020-02/msg05188.html
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 05/22] migration/block-dirty-bitmap: refactor state global variables
In reply to Andrey ShinkevichQuoted 197 lines · Andrey Shinkevich
Quoted 194 lines · Vladimir Sementsov-Ogievskiy
Move all state variables into one global struct. Reduce global variable usage, utilizing opaque pointer where possible.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 171 ++++++++++++++++++---------------1 file changed, 95 insertions(+), 76 deletions(-)migration/block-dirty-bitmap.c +41
-33@@ -128,6 +128,12 @@ typedef struct DBMSaveState { BdrvDirtyBitmap *prev_bitmap; } DBMSaveState; +typedef struct LoadBitmapState { + BlockDriverState *bs; + BdrvDirtyBitmap *bitmap; + bool migrated; +}LoadBitmapState ; + /* State of the dirty bitmap migration (DBM) during load process */ typedef struct DBMLoadState { uint32_t flags;@@ -135,18 +141,17 @@ typedef struct DBMLoadState { char bitmap_name[256]; BlockDriverState *bs; BdrvDirtyBitmap *bitmap; + + GSList *enabled_bitmaps; + QemuMutex finish_lock; } DBMLoadState; -static DBMSaveState dirty_bitmap_mig_state; +typedef struct DBMState { + DBMSaveState save; + DBMLoadState load; +}DBMState ; -/* State of one bitmap during load process */ -typedef struct LoadBitmapState { - BlockDriverState *bs; - BdrvDirtyBitmap *bitmap; - bool migrated; -}LoadBitmapState ; -static GSList *enabled_bitmaps; -QemuMutex finish_lock; +static DBMState dbm_state; static uint32_t qemu_get_bitmap_flags(QEMUFile *f) {@@ -169,21 +174,21 @@ static void qemu_put_bitmap_flags(QEMUFile *f, uint32_t flags) qemu_put_byte(f, flags); } -static void send_bitmap_header (QEMUFile *f,SaveBitmapState *dbms, - uint32_t additional_flags) +static void send_bitmap_header (QEMUFile *f,DBMSaveState *s, + SaveBitmapState *dbms, uint32_t additional_flags) { BlockDriverState *bs = dbms->bs; BdrvDirtyBitmap *bitmap = dbms->bitmap; uint32_t flags = additional_flags; trace_send_bitmap_header_enter(); - if (bs != dirty_bitmap_mig_state.prev_bs) { - dirty_bitmap_mig_state.prev_bs = bs; + if (bs != s->prev_bs) { + s->prev_bs = bs; flags |= DIRTY_BITMAP_MIG_FLAG_DEVICE_NAME; } - if (bitmap != dirty_bitmap_mig_state.prev_bitmap) { - dirty_bitmap_mig_state.prev_bitmap = bitmap; + if (bitmap != s->prev_bitmap) { + s->prev_bitmap = bitmap; flags |= DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME; }@@ -198,19 +203,22 @@ static void send_bitmap_header(QEMUFile *f, SaveBitmapState *dbms, } } -static void send_bitmap_start (QEMUFile *f,SaveBitmapState *dbms) +static void send_bitmap_start (QEMUFile *f,DBMSaveState *s, + SaveBitmapState *dbms) { - send_bitmap_header(f, dbms, DIRTY_BITMAP_MIG_FLAG_START); + send_bitmap_header(f, s, dbms, DIRTY_BITMAP_MIG_FLAG_START); qemu_put_be32(f, bdrv_dirty_bitmap_granularity(dbms->bitmap)); qemu_put_byte(f, dbms->flags); } -static void send_bitmap_complete (QEMUFile *f,SaveBitmapState *dbms) +static void send_bitmap_complete (QEMUFile *f,DBMSaveState *s, + SaveBitmapState *dbms) { - send_bitmap_header(f, dbms, DIRTY_BITMAP_MIG_FLAG_COMPLETE); + send_bitmap_header(f, s, dbms, DIRTY_BITMAP_MIG_FLAG_COMPLETE); } -static void send_bitmap_bits (QEMUFile *f,SaveBitmapState *dbms, +static void send_bitmap_bits (QEMUFile *f,DBMSaveState *s, + SaveBitmapState *dbms, uint64_t start_sector, uint32_t nr_sectors) { /* align for buffer_is_zero() */@@ -235,7 +243,7 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms, trace_send_bitmap_bits(flags, start_sector, nr_sectors, buf_size); - send_bitmap_header(f, dbms, flags); + send_bitmap_header(f, s, dbms, flags); qemu_put_be64(f, start_sector); qemu_put_be32(f, nr_sectors);@@ -254,12 +262,12 @@ static void send_bitmap_bits(QEMUFile *f, SaveBitmapState *dbms, } /* Called with iothread lock taken. */ -static void dirty_bitmap_do_save_cleanup (void ) +static void dirty_bitmap_do_save_cleanup (DBMSaveState *s) { SaveBitmapState *dbms; - while ((dbms = QSIMPLEQ_FIRST(&dirty_bitmap_mig_state.dbms_list)) != NULL) { - QSIMPLEQ_REMOVE_HEAD(&dirty_bitmap_mig_state.dbms_list, entry); + while ((dbms = QSIMPLEQ_FIRST(&s->dbms_list)) != NULL) { + QSIMPLEQ_REMOVE_HEAD(&s->dbms_list, entry); bdrv_dirty_bitmap_set_busy(dbms->bitmap, false); bdrv_unref(dbms->bs); g_free(dbms);@@ -267,17 +275,17 @@ static void dirty_bitmap_do_save_cleanup(void) } /* Called with iothread lock taken. */ -static int init_dirty_bitmap_migration (void ) +static int init_dirty_bitmap_migration (DBMSaveState *s) { BlockDriverState *bs; BdrvDirtyBitmap *bitmap; SaveBitmapState *dbms; Error *local_err = NULL; - dirty_bitmap_mig_state.bulk_completed = false; - dirty_bitmap_mig_state.prev_bs = NULL; - dirty_bitmap_mig_state.prev_bitmap = NULL; - dirty_bitmap_mig_state.no_bitmaps = false; + s->bulk_completed = false; + s->prev_bs = NULL; + s->prev_bitmap = NULL; + s->no_bitmaps = false; for (bs = bdrv_next_all_states(NULL); bs; bs = bdrv_next_all_states(bs)) { const char *name = bdrv_get_device_or_node_name(bs);@@ -316,35 +324,36 @@ static int init_dirty_bitmap_migration(void) dbms->flags |= DIRTY_BITMAP_MIG_START_FLAG_PERSISTENT; } - QSIMPLEQ_INSERT_TAIL(&dirty_bitmap_mig_state.dbms_list, + QSIMPLEQ_INSERT_TAIL(&s->dbms_list, dbms, entry); } } /* unset migration flags here, to not roll back it */ - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { bdrv_dirty_bitmap_skip_store(dbms->bitmap, true); } - if (QSIMPLEQ_EMPTY(&dirty_bitmap_mig_state.dbms_list)) { - dirty_bitmap_mig_state.no_bitmaps = true; + if (QSIMPLEQ_EMPTY(&s->dbms_list)) { + s->no_bitmaps = true; } return 0;fail:
-dirty_bitmap_do_save_cleanup();
+dirty_bitmap_do_save_cleanup(s);
return -1;
}
/* Called with no lock taken. */migration/block-dirty-bitmap.c +2
-1-static void bulk_phase_send_chunk (QEMUFile *f,SaveBitmapState *dbms) +static void bulk_phase_send_chunk (QEMUFile *f,DBMSaveState *s, + SaveBitmapState *dbms){
uint32_tnr_sectors=MIN(dbms->total_sectors - dbms->cur_sector, dbms->sectors_per_chunk);
-send_bitmap_bits(f, dbms, dbms->cur_sector, nr_sectors); +send_bitmap_bits(f, s, dbms, dbms->cur_sector, nr_sectors);dbms->cur_sector+=nr_sectors;
if (dbms->cur_sector>=dbms->total_sectors) {migration/block-dirty-bitmap.c +4
-4@@ -353,61 +362,66 @@ static void bulk_phase_send_chunk(QEMUFile *f, SaveBitmapState *dbms) } /* Called with no lock taken. */ -static void bulk_phase (QEMUFile *f,bool limit) +static void bulk_phase (QEMUFile *f,DBMSaveState *s,bool limit) { SaveBitmapState *dbms; - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { while (!dbms->bulk_completed) { - bulk_phase_send_chunk(f, dbms); + bulk_phase_send_chunk(f, s, dbms); if (limit && qemu_file_rate_limit(f)) { return; } } } - dirty_bitmap_mig_state.bulk_completed = true; + s->bulk_completed = true; } /* for SaveVMHandlers */ static void dirty_bitmap_save_cleanup(void *opaque)
Quoted 8 lines
migration/block-dirty-bitmap.c static void bulk_phase_send_chunk(QEMUFile *f, SaveBitmapState *dbms) · line 387
Quoted 6 lines
migration/block-dirty-bitmap.c +3
-1{ - dirty_bitmap_do_save_cleanup(); + DBMSaveState *s = &((DBMState *)opaque)->save; + + dirty_bitmap_do_save_cleanup(s); }Why do one need the extra nested "do" function?
Because "_do_" has sub-structure argument, and I don't have pointer to the whole structure in another "_do_" caller..
Quoted 180 lines
Quoted 168 lines
migration/block-dirty-bitmap.c +13
-9static int dirty_bitmap_save_iterate(QEMUFile *f, void *opaque) { + DBMSaveState *s = &((DBMState *)opaque)->save; + trace_dirty_bitmap_save_iterate(migration_in_postcopy()); - if (migration_in_postcopy() && !dirty_bitmap_mig_state.bulk_completed) { - bulk_phase(f, true); + if (migration_in_postcopy() && !s->bulk_completed) { + bulk_phase(f, s, true); } qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS); - return dirty_bitmap_mig_state.bulk_completed; + return s->bulk_completed; } /* Called with iothread lock taken. */ static int dirty_bitmap_save_complete(QEMUFile *f, void *opaque) { + DBMSaveState *s = &((DBMState *)opaque)->save; SaveBitmapState *dbms; trace_dirty_bitmap_save_complete_enter(); - if (!dirty_bitmap_mig_state.bulk_completed) { - bulk_phase(f, false); + if (!s->bulk_completed) { + bulk_phase(f, s, false); } - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { - send_bitmap_complete(f, dbms); + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { + send_bitmap_complete(f, s, dbms); } qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS); trace_dirty_bitmap_save_complete_finish(); - dirty_bitmap_do_save_cleanup(); + dirty_bitmap_save_cleanup(opaque); return 0; }@@ -417,12 +431,13 @@ static void dirty_bitmap_save_pending(QEMUFile *f, void *opaque, uint64_t *res_compatible, uint64_t *res_postcopy_only) { + DBMSaveState *s = &((DBMState *)opaque)->save; SaveBitmapState *dbms; uint64_t pending = 0; qemu_mutex_lock_iothread(); - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { uint64_t gran = bdrv_dirty_bitmap_granularity(dbms->bitmap);
uint64_tsectors =dbms->bulk_completed? 0 :
dbms->total_sectors-dbms->cur_sector;migration/block-dirty-bitmap.c +27
-23@@ -481,7 +496,7 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s) b->bs = s->bs; b->bitmap = s->bitmap; b->migrated = false; - enabled_bitmaps = g_slist_prepend(enabled_bitmaps, b); + s->enabled_bitmaps = g_slist_prepend(s->enabled_bitmaps, b); } return 0;@@ -489,11 +504,12 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s) void dirty_bitmap_mig_before_vm_start(void) { + DBMLoadState *s = &dbm_state.load; GSList *item; - qemu_mutex_lock(&finish_lock); + qemu_mutex_lock(&s->finish_lock); - for (item = enabled_bitmaps; item; item = g_slist_next(item)) { + for (item = s->enabled_bitmaps; item; item = g_slist_next(item)) { LoadBitmapState *b = item->data; if (b->migrated) {@@ -505,10 +521,10 @@ void dirty_bitmap_mig_before_vm_start(void) g_free(b); } - g_slist_free(enabled_bitmaps); - enabled_bitmaps = NULL; + g_slist_free(s->enabled_bitmaps); + s->enabled_bitmaps = NULL; - qemu_mutex_unlock(&finish_lock); + qemu_mutex_unlock(&s->finish_lock); } static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s)@@ -517,9 +533,9 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) trace_dirty_bitmap_load_complete(); bdrv_dirty_bitmap_deserialize_finish(s->bitmap); - qemu_mutex_lock(&finish_lock); + qemu_mutex_lock(&s->finish_lock); - for (item = enabled_bitmaps; item; item = g_slist_next(item)) { + for (item = s->enabled_bitmaps; item; item = g_slist_next(item)) { LoadBitmapState *b = item->data; if (b->bitmap == s->bitmap) {@@ -530,7 +546,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) if (bdrv_dirty_bitmap_has_successor(s->bitmap)) { bdrv_dirty_bitmap_lock(s->bitmap); - if (enabled_bitmaps == NULL) { + if (s->enabled_bitmaps == NULL) { /* in postcopy */ bdrv_reclaim_dirty_bitmap_locked(s->bitmap, &error_abort); bdrv_enable_dirty_bitmap_locked(s->bitmap);@@ -549,7 +565,7 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) bdrv_dirty_bitmap_unlock(s->bitmap); } - qemu_mutex_unlock(&finish_lock); + qemu_mutex_unlock(&s->finish_lock); } static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)@@ -646,7 +662,7 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) { - static DBMLoadState s; + DBMLoadState *s = &((DBMState *)opaque)->load; int ret = 0; trace_dirty_bitmap_load_enter();@@ -656,17 +672,17 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) } do { - ret = dirty_bitmap_load_header(f, &s); + ret = dirty_bitmap_load_header(f, s); if (ret < 0) { return ret; } - if (s.flags & DIRTY_BITMAP_MIG_FLAG_START) { - ret = dirty_bitmap_load_start(f, &s); - } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) { - dirty_bitmap_load_complete(f, &s); - } else if (s.flags & DIRTY_BITMAP_MIG_FLAG_BITS) { - ret = dirty_bitmap_load_bits(f, &s); + if (s->flags & DIRTY_BITMAP_MIG_FLAG_START) { + ret = dirty_bitmap_load_start(f, s); + } else if (s->flags & DIRTY_BITMAP_MIG_FLAG_COMPLETE) { + dirty_bitmap_load_complete(f, s); + } else if (s->flags & DIRTY_BITMAP_MIG_FLAG_BITS) { + ret = dirty_bitmap_load_bits(f, s); } if (!ret) {@@ -676,7 +692,7 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) if (ret) { return ret; } - } while (!(s.flags & DIRTY_BITMAP_MIG_FLAG_EOS)); + } while (!(s->flags & DIRTY_BITMAP_MIG_FLAG_EOS)); trace_dirty_bitmap_load_success(); return 0;@@ -684,13 +700,14 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) static int dirty_bitmap_save_setup(QEMUFile *f, void *opaque) { + DBMSaveState *s = &((DBMState *)opaque)->save; SaveBitmapState *dbms = NULL; - if (init_dirty_bitmap_migration() < 0) { + if (init_dirty_bitmap_migration(s) < 0) { return -1; } - QSIMPLEQ_FOREACH(dbms, &dirty_bitmap_mig_state.dbms_list, entry) { - send_bitmap_start(f, dbms); + QSIMPLEQ_FOREACH(dbms, &s->dbms_list, entry) { + send_bitmap_start(f, s, dbms); } qemu_put_bitmap_flags(f, DIRTY_BITMAP_MIG_FLAG_EOS);@@ -699,7 +716,9 @@ static int dirty_bitmap_save_setup(QEMUFile *f, void *opaque) static bool dirty_bitmap_is_active(void *opaque) { - return migrate_dirty_bitmaps() && !dirty_bitmap_mig_state.no_bitmaps; + DBMSaveState *s = &((DBMState *)opaque)->save; + + return migrate_dirty_bitmaps() && !s->no_bitmaps; } static bool dirty_bitmap_is_active_iterate(void *opaque)@@ -727,10 +746,10 @@ static SaveVMHandlers savevm_dirty_bitmap_handlers = { void dirty_bitmap_mig_init(void) {migration/block-dirty-bitmap.c static SaveVMHandlers savevm_dirty_bitmap_handlers = { · line 754
Quoted 10 lines
migration/block-dirty-bitmap.c +3
-3- QSIMPLEQ_INIT(&dirty_bitmap_mig_state.dbms_list); - qemu_mutex_init(&finish_lock); + QSIMPLEQ_INIT(&dbm_state.save.dbms_list); + qemu_mutex_init(&dbm_state.load.finish_lock); register_savevm_live("dirty-bitmap", 0, 1, &savevm_dirty_bitmap_handlers, - &dirty_bitmap_mig_state); + &dbm_state); }Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 07/22] migration/block-dirty-bitmap: simplify dirty_bitmap_load_complete
In reply to Andrey ShinkevichQuoted 14 lines · Andrey Shinkevich
Quoted 13 lines · Vladimir Sementsov-Ogievskiy
bdrv_enable_dirty_bitmap_locked()call does nothing, as if we are in postcopy, bitmap successor must be enabled, and reclaim operation will enable the bitmap.So, actually we need just call _reclaim_ in both if branches, and making differences only to add an assertion seems not really good. The logic becomes simple: on load complete we do reclaim and that's all.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 25 ++++---------------------1 file changed, 4 insertions(+), 21 deletions(-)
Quoted 9 lines
migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 536
Quoted 7 lines
migration/block-dirty-bitmap.c +1
-0@@ -535,6 +535,10 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) qemu_mutex_lock(&s->lock); + if (bdrv_dirty_bitmap_has_successor(s->bitmap)) {What about making it sure?
assert(!s->bitmap->successor->disabled);
I'm afraid we can't as BdrvDirtyBitmap is not public structure
Quoted 2 lines
migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 539
Quoted 2 lines
migration/block-dirty-bitmap.c +1
-0+ bdrv_reclaim_dirty_bitmap(s->bitmap, &error_abort);
But we can assert that resulting bitmap is enabled.
Quoted 33 lines
Quoted 21 lines
migration/block-dirty-bitmap.c +2
-14+ } + for (item = s->enabled_bitmaps; item; item = g_slist_next(item)) { LoadBitmapState *b = item->data;@@ -544,27 +548,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) } } - if (bdrv_dirty_bitmap_has_successor(s->bitmap)) { - bdrv_dirty_bitmap_lock(s->bitmap); - if (s->enabled_bitmaps == NULL) { - /* in postcopy */ - bdrv_reclaim_dirty_bitmap_locked(s->bitmap, &error_abort); - bdrv_enable_dirty_bitmap_locked(s->bitmap); - } else { - /* target not started, successor must be empty */ - int64_t count = bdrv_get_dirty_count(s->bitmap); - BdrvDirtyBitmap *ret = bdrv_reclaim_dirty_bitmap_locked(s->bitmap, - NULL); - /* bdrv_reclaim_dirty_bitmap can fail only on no successor (it -* must be) or on merge fail, but merge can't fail when second -* bitmap is empty migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 552
Quoted 10 lines
migration/block-dirty-bitmap.c +0
-7-*/ - assert(ret == s->bitmap && - count == bdrv_get_dirty_count(s->bitmap)); - } - bdrv_dirty_bitmap_unlock(s->bitmap); - } - qemu_mutex_unlock(&s->lock); }Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 09/22] migration/block-dirty-bitmap: relax error handling in incoming part
In reply to Andrey ShinkevichQuoted 67 lines · Andrey Shinkevich
Quoted 64 lines · Vladimir Sementsov-Ogievskiy
Bitmaps data is not critical, and we should not fail the migration (or use postcopy recovering) because of dirty-bitmaps migration failure. Instead we should just lose unfinished bitmaps.
Still we have to report io stream violation errors, as they affect the whole migration stream.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 148 +++++++++++++++++++++++++--------1 file changed, 113 insertions(+), 35 deletions(-)migration/block-dirty-bitmap.c +0
-0@@ -145,6 +145,15 @@ typedef struct DBMLoadState {bool
before_vm_start_handled; /* set indirty_bitmap_mig_before_vm_start*/ + /*
+ * cancelled
+ * Incoming migration is cancelled for some reason. That means that we + * still should read our chunks from migration stream, to not affect other + * migration objects (like RAM), but just ignore them and do not touch any + * bitmaps or nodes.
+ */
+ bool cancelled;
+
GSList *bitmaps;
QemuMutex lock; /* protect bitmaps */
} DBMLoadState;migration/block-dirty-bitmap.c +31
-0@@ -545,13 +554,47 @@ void dirty_bitmap_mig_before_vm_start(void) qemu_mutex_unlock(&s->lock); } +static void cancel_incoming_locked (DBMLoadState *s) +{ + GSList *item; + + if (s->cancelled) { + return; + } + + s->cancelled = true; + s->bs = NULL; + s->bitmap = NULL; + + /* Drop all unfinished bitmaps */ + for (item = s->bitmaps; item; item = g_slist_next(item)) { + LoadBitmapState *b = item->data; + + /* + * Bitmap must be unfinished, as finished bitmaps should already be + * removed from the list. + */ + assert(!s->before_vm_start_handled || !b->migrated); + if (bdrv_dirty_bitmap_has_successor(b->bitmap)) { + bdrv_reclaim_dirty_bitmap(b->bitmap, &error_abort); + } + bdrv_release_dirty_bitmap(b->bitmap); + } + + g_slist_free_full(s->bitmaps, g_free); + s->bitmaps = NULL; +} +
Quoted 8 lines
migration/block-dirty-bitmap.c void dirty_bitmap_mig_before_vm_start(void) · line 554
Quoted 6 lines
migration/block-dirty-bitmap.c +0
-2static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) { GSList *item; trace_dirty_bitmap_load_complete(); - bdrv_dirty_bitmap_deserialize_finish(s->bitmap); - qemu_mutex_lock(&s->lock);Why is it safe to remove the critical section?
It's not removed, it becomes wider in this patch.
Quoted 60 lines
Quoted 60 lines
migration/block-dirty-bitmap.c +27
-7+ if (s->cancelled) { + return; + } + + bdrv_dirty_bitmap_deserialize_finish(s->bitmap); if (bdrv_dirty_bitmap_has_successor(s->bitmap)) { bdrv_reclaim_dirty_bitmap(s->bitmap, &error_abort);@@ -569,8 +612,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) break; } } - - qemu_mutex_unlock(&s->lock); } static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)@@ -582,15 +623,32 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s) if (s->flags & DIRTY_BITMAP_MIG_FLAG_ZEROES) { trace_dirty_bitmap_load_bits_zeroes(); - bdrv_dirty_bitmap_deserialize_zeroes(s->bitmap, first_byte, nr_bytes, - false); + if (!s->cancelled) { + bdrv_dirty_bitmap_deserialize_zeroes(s->bitmap, first_byte, + nr_bytes, false); + } } else { size_t ret; uint8_t *buf; uint64_t buf_size = qemu_get_be64(f); - uint64_t needed_size = - bdrv_dirty_bitmap_serialization_size(s->bitmap, - first_byte, nr_bytes); + uint64_t needed_size; + + buf = g_malloc(buf_size); + ret = qemu_get_buffer(f, buf, buf_size); + if (ret != buf_size) { + error_report("Failed to read bitmap bits"); + g_free(buf); + return -EIO; + } + + if (s->cancelled) { + g_free(buf); + return 0; + } + + needed_size = bdrv_dirty_bitmap_serialization_size(s->bitmap, + first_byte, + nr_bytes);if (
needed_size>buf_size||
buf_size>QEMU_ALIGN_UP(needed_size, 4 * sizeof(long))migration/block-dirty-bitmap.c +0
-0@@ -599,15 +657,8 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)error_report("Migrated bitmap granularity doesn't " "match the destination bitmap '%s' granularity",
bdrv_dirty_bitmap_name(s->bitmap)); - return-EINVAL;
- }
-
- buf =g_malloc(buf_size);
Quoted 8 lines
Quoted 6 lines
- ret =
qemu_get_buffer(f, buf, buf_size); - if (ret !=buf_size) {
-error_report("Failed to read bitmap bits"); -g_free(buf);
- return-EIO;
+cancel_incoming_locked(s);/* Continue the VM migration as bitmaps data are not critical */
Hmm yes it what this patch does.. But I don't think we should add comment to each call of cancel_..()
Quoted 110 lines
Quoted 6 lines
+ return 0;
}
bdrv_dirty_bitmap_deserialize_part(s->bitmap, buf,first_byte,nr_bytes,migration/block-dirty-bitmap.c +0
-0@@ -632,14 +683,16 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) error_report("Unable to read node name string");migration/block-dirty-bitmap.c static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) · line 689
Quoted 10 lines
migration/block-dirty-bitmap.c +4
-4return -EINVAL; } - s->bs = bdrv_lookup_bs(s->node_name, s->node_name, &local_err); - if (!s->bs) { - error_report_err(local_err); - return -EINVAL; + if (!s->cancelled) { + s->bs = bdrv_lookup_bs(s->node_name, s->node_name, &local_err); + if (!s->bs) { + error_report_err(local_err);The error message may be supplemented with a report about the canceled bitmap migration. Also down there at
cancel_incoming_locked(s).Quoted 43 lines
migration/block-dirty-bitmap.c +19
-13+ cancel_incoming_locked(s); + } } - } else if (!s->bs && !nothing) { + } else if (!s->bs && !nothing && !s->cancelled) { error_report("Error: block device name is not set"); - return -EINVAL; + cancel_incoming_locked(s); } if (s->flags & DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME) {@@ -647,24 +700,38 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) error_report("Unable to read bitmap name string"); return -EINVAL; } - s->bitmap = bdrv_find_dirty_bitmap(s->bs, s->bitmap_name); - - /* bitmap may be NULL here, it wouldn't be an error if it is the - * first occurrence of the bitmap */ - if (!s->bitmap && !(s->flags & DIRTY_BITMAP_MIG_FLAG_START)) { - error_report("Error: unknown dirty bitmap " - "'%s' for block device '%s'", - s->bitmap_name, s->node_name); - return -EINVAL; + if (!s->cancelled) { + s->bitmap = bdrv_find_dirty_bitmap(s->bs, s->bitmap_name); + + /* + * bitmap may be NULL here, it wouldn't be an error if it is the + * first occurrence of the bitmap + */ + if (!s->bitmap && !(s->flags & DIRTY_BITMAP_MIG_FLAG_START)) { + error_report("Error: unknown dirty bitmap " + "'%s' for block device '%s'", + s->bitmap_name, s->node_name); + cancel_incoming_locked(s); + } } - } else if (!s->bitmap && !nothing) { + } else if (!s->bitmap && !nothing && !s->cancelled) { error_report("Error: block device name is not set"); - return -EINVAL; + cancel_incoming_locked(s); }migration/block-dirty-bitmap.c static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) · line 730
Quoted 10 lines
migration/block-dirty-bitmap.c +8
-0return 0; } +/* +* dirty_bitmap_load +* +* Load sequence of dirty bitmap chunks. Return error only on fatal io stream +* violations. On other errors just cancel bitmaps incoming migration and return +* 0. +* +* Note, than when incoming bitmap migration is canceled, we still must read all "than (that)" may be omitted
Quoted 24 lines
migration/block-dirty-bitmap.c +9
-0+* our chunks (and just ignore them), to not affect other migration objects. +*/ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) { DBMLoadState *s = &((DBMState *)opaque)->load;@@ -673,12 +740,19 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) trace_dirty_bitmap_load_enter(); if (version_id != 1) { + qemu_mutex_lock(&s->lock); + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return -EINVAL; } do { + qemu_mutex_lock(&s->lock); + ret = dirty_bitmap_load_header(f, s); if (ret < 0) { + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return ret; }@@ -695,8 +769,12 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) }migration/block-dirty-bitmap.c static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) · line 779
Quoted 10 lines
migration/block-dirty-bitmap.c +4
-0if (ret) { + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return ret; } + + qemu_mutex_unlock(&s->lock); } while (!(s->flags & DIRTY_BITMAP_MIG_FLAG_EOS)); trace_dirty_bitmap_load_success();Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 14/22] qemu-iotests/199: better catch postcopy time
In reply to Andrey ShinkevichQuoted 124 lines · Andrey Shinkevich
Quoted 47 lines · Vladimir Sementsov-Ogievskiy
The test aims to test _postcopy_ migration, and wants to do some write operations during postcopy time.
Test considers migrate status=complete event on source as start of postcopy. This is completely wrong, completion is completion of the whole migration process. Let's instead consider destination start as start of postcopy, and use RESUME event for it.
Next, as migration finish, let's use migration status=complete event on target, as such method is closer to what libvirt or another user will do, than tracking number of dirty-bitmaps.
Finally, add a possibility to dump events for debug. And if set debug to True, we see, that actual postcopy period is very small relatively to the whole test duration time (~0.2 seconds to >40 seconds for me). This means, that test is very inefficient in what it supposed to do. Let's improve it in following commits.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
tests/qemu-iotests/199 | 72 +++++++++++++++++++++++++++++++++---------1 file changed, 57 insertions(+), 15 deletions(-)tests/qemu-iotests/199 +0
-0@@ -20,17 +20,43 @@import os
import iotests
-import time
from iotests importqemu_imgtests/qemu-iotests/199 +2
-0+debug = False +
disk_a=os.path.join(iotests.test_dir, 'disk_a')disk_b=os.path.join(iotests.test_dir, 'disk_b')size = '256G'
fifo =os.path.join(iotests.test_dir, 'mig_fifo')tests/qemu-iotests/199 +9
-0+def event_seconds(event): + return event['timestamp']['seconds'] + \ + event['timestamp']['microseconds'] / 1000000.0 + + +def event_dist(e1, e2): + return event_seconds(e2) - event_seconds(e1) + +Quoted 2 lines
class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):
def tearDown(self):It's common to put the definition of
setUp()aheadQuoted 31 lines
+ if debug:
+ self.vm_a_events +=self.vm_a.get_qmp_events()+ self.vm_b_events +=self.vm_b.get_qmp_events()+ for e in self.vm_a_events:
+ e['vm'] = 'SRC'
+ for e in self.vm_b_events:
+ e['vm'] = 'DST'
+ events = (self.vm_a_events + self.vm_b_events) + events = [(e['timestamp']['seconds'], + e['timestamp']['microseconds'], + e['vm'],
+ e['event'],
+e.get('data', '')) for e in events] + for e in sorted(events):
+print('{}.{:06} {} {} {}'.format(*e))+
self.vm_a.shutdown()self.vm_b.shutdown()
os.remove(disk_a)tests/qemu-iotests/199 +4
-0@@ -47,6 +73,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.launch() self.vm_b.launch() + # collect received events for debug + self.vm_a_events = [] + self.vm_b_events = [] +def
test_postcopy(self):
write_size=0x40000000
granularity = 512tests/qemu-iotests/199 +0
-0@@ -77,15 +107,13 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk))Quoted 4 lines
s +=
0x10000
-bitmaps_cap= {'capability': 'dirty-bitmaps', 'state': True} -events_cap= {'capability': 'events', 'state': True} + caps = [{'capability': 'dirty-bitmaps', 'state': True},The name "capabilities" would be an appropriate identifier.
Quoted 11 lines
+ {'capability': 'events', 'state': True}]
- result = self.vm_a.qmp('migrate-set-capabilities',
- capabilities=[bitmaps_cap,events_cap]) + result =self.vm_a.qmp('migrate-set-capabilities', capabilities=caps)self.assert_qmp(result, 'return', {})
- result = self.vm_b.qmp('migrate-set-capabilities',
- capabilities=[bitmaps_cap])
+ result =self.vm_b.qmp('migrate-set-capabilities', capabilities=caps)self.assert_qmp(result, 'return', {})
result =self.vm_a.qmp('migrate', uri='exec:cat>' + fifo)tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 124
Quoted 7 lines
tests/qemu-iotests/199 +1
-4@@ -94,24 +122,38 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('migrate-start-postcopy') self.assert_qmp(result, 'return', {}) - while True: - event = self.vm_a.event_wait('MIGRATION') - if event['data']['status'] == 'completed': - break + e_resume = self.vm_b.event_wait('RESUME')"
event_resume" gives a faster understandingQuoted 7 lines
tests/qemu-iotests/199 +1
-0+ self.vm_b_events.append(e_resume)s =
0x8000
while s <write_size:
self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk))s +=0x10000
+ match = {'data': {'status': 'completed'}}
+e_complete=self.vm_b.event_wait('MIGRATION', match=match)"
event_complete" alsoQuoted 4 lines
+
self.vm_b_events.append(e_complete)
+
+ # take queued event, should already been happened +e_stop=self.vm_a.event_wait('STOP')"
event_stop"
Quoted 8 lines
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 144
Quoted 6 lines
+
self.vm_a_events.append(e_stop)tests/qemu-iotests/199 +5
-0+ + downtime = event_dist(e_stop, e_resume) + postcopy_time = event_dist(e_resume, e_complete) + + # TODO: assert downtime * 10 < postcopy_timeI got the results below in debug mode:
That's why it's a TODO
Quoted 14 lines
downtime: 6.194924831390381
postcopy_time: 0.1592559814453125
1582102669.764919 SRC MIGRATION {'status': 'setup'} 1582102669.766179 SRCMIGRATION_PASS{'pass': 1} 1582102669.766234 SRC MIGRATION {'status': 'active'} 1582102669.768058 DST MIGRATION {'status': 'active'} 1582102669.801422 SRC MIGRATION {'status': 'postcopy-active'} 1582102669.801510 SRC STOP
1582102675.990041 DST MIGRATION {'status': 'postcopy-active'} 1582102675.996435 DST RESUME
1582102676.111313 SRC MIGRATION {'status': 'completed'} 1582102676.155691 DST MIGRATION {'status': 'completed'}
Quoted 2 lines
Quoted 1 line
+ if debug:
with no usage in the following patches, you can put the whole block of relative code above under the "if debug: section
TODO will be uncommented soon
Quoted 19 lines
Quoted 17 lines
+
print('downtime:', downtime)
+print('postcopy_time:', postcopy_time)
+
+ # Assert that bitmap migration is finished (check that successor bitmap + # is removed)
result =self.vm_b.qmp('query-block')
- whilelen(result['return'][0]['dirty-bitmaps'])> 1:
-time.sleep(2)
- result =self.vm_b.qmp('query-block')
+ assertlen(result['return'][0]['dirty-bitmaps'])== 1
+ # Check content of migrated (and updated by new writes) bitmap result =self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap')
-
self.assert_qmp(result, 'return/sha256', sha256)Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 16/22] qemu-iotests/199: change discard patterns
In reply to Andrey ShinkevichQuoted 35 lines · Andrey Shinkevich
Quoted 1 line · Vladimir Sementsov-Ogievskiy
iotest 40 works too long because of many discard opertion. On the same
operations
At the same timeQuoted 29 lines
time, postcopy period is very short, in spite of all these efforts.
So, let's use less discards (and with more interesting patterns) to reduce test timing. In the next commit we'll increase postcopy period.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
tests/qemu-iotests/199 | 44 +++++++++++++++++++++++++-----------------1 file changed, 26 insertions(+), 18 deletions(-)tests/qemu-iotests/199 +13
-0@@ -30,6 +30,28 @@ size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +GiB = 1024 * 1024 * 1024 + +discards1 = ( + (0, GiB), + (2 * GiB + 512 * 5, 512), + (3 * GiB + 512 * 5, 512), + (100 * GiB, GiB) +) + +discards2 = ( + (3 * GiB + 512 * 8, 512), + (4 * GiB + 512 * 8, 512), + (50 * GiB, GiB),
Quoted 8 lines
tests/qemu-iotests/199 size = '256G' · line 51
Quoted 6 lines
tests/qemu-iotests/199 +6
-0+ (100 * GiB + GiB // 2, GiB) +) + + +def apply_discards(vm, discards): + for d in discards:If we run qemu-io only once, it will update the bitmap state and will speed the test performance up. Is that wrong idea?
But it will be less interesting test. Now we have several regions, which are in different relations in discards1 and discards2.
And four elements are handled fast enough.
Quoted 54 lines
Quoted 42 lines
tests/qemu-iotests/199 +3
-0+ vm.hmp_qemu_io('drive0', 'discard {} {}'.format(*d)) + +def
event_seconds(event):
return event['timestamp']['seconds'] + \
event['timestamp']['microseconds'] / 1000000.0tests/qemu-iotests/199 +0
-0@@ -80,9 +102,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):self.vm_b_events = []
deftest_postcopy(self):
-discard_size=0x40000000
granularity = 512
- chunk = 4096
result =self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap', granularity=granularity)tests/qemu-iotests/199 +0
-0@@ -92,14 +112,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): node='drive0', name='bitmap')
empty_sha256= result['return']['sha256']
- s = 0
- while s <discard_size:
-self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk))- s +=0x10000
- s =0x8000
- while s <discard_size:
-self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk))- s +=0x10000
+apply_discards(self.vm_a, discards1 + discards2)result =self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap')tests/qemu-iotests/199 +2
-4@@ -111,10 +124,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('block-dirty-bitmap-clear', node='drive0', name='bitmap') self.assert_qmp(result, 'return', {}) - s = 0 - while s < discard_size: - self.vm_a.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + + apply_discards(self.vm_a, discards1) caps = [{'capability': 'dirty-bitmaps', 'state': True},{'capability': 'events', 'state': True}]
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 150
Quoted 10 lines
tests/qemu-iotests/199 +1
-4@@ -134,10 +145,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): e_resume = self.vm_b.event_wait('RESUME') self.vm_b_events.append(e_resume) - s = 0x8000 - while s < discard_size: - self.vm_b.hmp_qemu_io('drive0', 'discard %d %d' % (s, chunk)) - s += 0x10000 + apply_discards(self.vm_b, discards2) match = {'data': {'status': 'completed'}} e_complete = self.vm_b.event_wait('MIGRATION', match=match)Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 19/22] qemu-iotests/199: prepare for new test-cases addition
In reply to Vladimir Sementsov-OgievskiyQuoted 96 lines · Vladimir Sementsov-Ogievskiy
Move future common part to
start_postcopy()method. Move checking number of bitmaps tocheck_bitmap().Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 36 +++++++++++++++++++++++------------- 1 file changed, 23 insertions(+), 13 deletions(-)tests/qemu-iotests/199 +21
-11@@ -29,6 +29,8 @@ disk_b = os.path.join(iotests.test_dir, 'disk_b') size = '256G' fifo = os.path.join(iotests.test_dir, 'mig_fifo') +granularity = 512 +nb_bitmaps = 15 GiB = 1024 * 1024 * 1024@@ -61,6 +63,15 @@ def event_dist(e1, e2): return event_seconds(e2) - event_seconds(e1) +def check_bitmaps(vm, count): + result = vm.qmp('query-block') + + if count == 0: + assert 'dirty-bitmaps' not in result['return'][0] + else: + assert len(result['return'][0]['dirty-bitmaps']) == count + + class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): def tearDown(self): if debug:@@ -101,10 +112,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a_events = [] self.vm_b_events = [] - def test_postcopy(self): - granularity = 512 - nb_bitmaps = 15 - + def start_postcopy(self): + """ Run migration until RESUME event on target. Return this event. """ for i in range(nb_bitmaps): result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap{}'.format(i),@@ -119,10 +128,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap0') - discards1_sha256 = result['return']['sha256'] + self.discards1_sha256 = result['return']['sha256'] # Check, that updating the bitmap by discards works - assert discards1_sha256 != empty_sha256 + assert self.discards1_sha256 != empty_sha256 # We want to calculate resulting sha256. Do it in bitmap0, so, disable # other bitmaps@@ -135,7 +144,7 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap0') - all_discards_sha256 = result['return']['sha256'] + self.all_discards_sha256 = result['return']['sha256'] # Now, enable some bitmaps, to be updated during migration for i in range(2, nb_bitmaps, 2):@@ -160,6 +169,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): e_resume = self.vm_b.event_wait('RESUME') self.vm_b_events.append(e_resume) + return e_resume + + def test_postcopy_success(self): + e_resume = self.start_postcopy() # enabled bitmaps should be updated apply_discards(self.vm_b, discards2)@@ -180,18 +193,15 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): print('downtime:', downtime) print('postcopy_time:', postcopy_time) - # Assert that bitmap migration is finished (check that successor bitmap - # is removed) - result = self.vm_b.qmp('query-block') - assert len(result['return'][0]['dirty-bitmaps']) == nb_bitmaps + check_bitmaps(self.vm_b, nb_bitmaps) # Check content of migrated bitmaps. Still, don't waste time checking # every bitmap
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 207
Quoted 10 lines
tests/qemu-iotests/199 +2
-2for i in range(0, nb_bitmaps, 5): result = self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap{}'.format(i)) - sha256 = discards1_sha256 if i % 2 else all_discards_sha256 - self.assert_qmp(result, 'return/sha256', sha256) + sha = self.discards1_sha256 if i % 2 else self.all_discards_sha256 + self.assert_qmp(result, 'return/sha256', sha) if __name__ == '__main__':
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 20/22] qemu-iotests/199: check persistent bitmaps
In reply to Vladimir Sementsov-OgievskiyQuoted 33 lines · Vladimir Sementsov-Ogievskiy
Check that persistent bitmaps are not stored on source and that bitmaps are persistent on destination.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-)tests/qemu-iotests/199 +8
-1@@ -117,7 +117,8 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): for i in range(nb_bitmaps): result = self.vm_a.qmp('block-dirty-bitmap-add', node='drive0', name='bitmap{}'.format(i), - granularity=granularity) + granularity=granularity, + persistent=True) self.assert_qmp(result, 'return', {}) result = self.vm_a.qmp('x-debug-block-dirty-bitmap-sha256',@@ -193,6 +194,19 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): print('downtime:', downtime) print('postcopy_time:', postcopy_time) + # check that there are no bitmaps stored on source + self.vm_a_events += self.vm_a.get_qmp_events() + self.vm_a.shutdown() + self.vm_a.launch() + check_bitmaps(self.vm_a, 0) +
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 212
Quoted 10 lines
tests/qemu-iotests/199 +7
-0+ # check that bitmaps are migrated and persistence works + check_bitmaps(self.vm_b, nb_bitmaps) + self.vm_b.shutdown() + # recreate vm_b, so there is no incoming option, which prevents + # loading bitmaps from disk + self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b) + self.vm_b.launch() check_bitmaps(self.vm_b, nb_bitmaps) # Check content of migrated bitmaps. Still, don't waste time checking
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 07/22] migration/block-dirty-bitmap: simplify dirty_bitmap_load_complete
In reply to Vladimir Sementsov-OgievskiyQuoted 25 lines · Vladimir Sementsov-Ogievskiy
Quoted 14 lines · Andrey Shinkevich
Quoted 13 lines · Vladimir Sementsov-Ogievskiy
bdrv_enable_dirty_bitmap_locked()call does nothing, as if we are in postcopy, bitmap successor must be enabled, and reclaim operation will enable the bitmap.So, actually we need just call _reclaim_ in both if branches, and making differences only to add an assertion seems not really good. The logic becomes simple: on load complete we do reclaim and that's all.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 25 ++++---------------------1 file changed, 4 insertions(+), 21 deletions(-)Quoted 9 lines
migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 536
Quoted 7 lines
migration/block-dirty-bitmap.c +1
-0@@ -535,6 +535,10 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) qemu_mutex_lock(&s->lock); + if (bdrv_dirty_bitmap_has_successor(s->bitmap)) {What about making it sure?
assert(!s->bitmap->successor->disabled);
Quoted 6 lines
I'm afraid we can't as BdrvDirtyBitmap is not public structure
Quoted 2 lines
migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 539
Quoted 2 lines
migration/block-dirty-bitmap.c +1
-0+ bdrv_reclaim_dirty_bitmap(s->bitmap, &error_abort);But we can assert that resulting bitmap is enabled.
Or not, as bitmap may be not yet enabled, if guest is not yet started.
Quoted 36 lines
Quoted 34 lines
Quoted 22 lines
migration/block-dirty-bitmap.c +2
-14+ } + for (item = s->enabled_bitmaps; item; item = g_slist_next(item)) { LoadBitmapState *b = item->data;@@ -544,27 +548,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) } } - if (bdrv_dirty_bitmap_has_successor(s->bitmap)) { - bdrv_dirty_bitmap_lock(s->bitmap); - if (s->enabled_bitmaps == NULL) { - /* in postcopy */ - bdrv_reclaim_dirty_bitmap_locked(s->bitmap, &error_abort); - bdrv_enable_dirty_bitmap_locked(s->bitmap); - } else { - /* target not started, successor must be empty */ - int64_t count = bdrv_get_dirty_count(s->bitmap); - BdrvDirtyBitmap *ret = bdrv_reclaim_dirty_bitmap_locked(s->bitmap, - NULL); - /* bdrv_reclaim_dirty_bitmap can fail only on no successor (it -* must be) or on merge fail, but merge can't fail when second -* bitmap is empty migration/block-dirty-bitmap.c static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) · line 552
Quoted 10 lines
migration/block-dirty-bitmap.c +0
-7-*/ - assert(ret == s->bitmap && - count == bdrv_get_dirty_count(s->bitmap)); - } - bdrv_dirty_bitmap_unlock(s->bitmap); - } - qemu_mutex_unlock(&s->lock); }Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 21/22] qemu-iotests/199: add early shutdown case to bitmaps postcopy
In reply to Vladimir Sementsov-OgievskiyQuoted 44 lines · Vladimir Sementsov-Ogievskiy
Previous patches fixed two crashes which may occur on shutdown prior to bitmaps postcopy finished. Check that it works now.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 18 ++++++++++++++++++ tests/qemu-iotests/199.out | 4 ++-- 2 files changed, 20 insertions(+), 2 deletions(-)tests/qemu-iotests/199 +18
-0@@ -217,6 +217,24 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): sha = self.discards1_sha256 if i % 2 else self.all_discards_sha256 self.assert_qmp(result, 'return/sha256', sha) + def test_early_shutdown_destination(self): + self.start_postcopy() + + self.vm_b_events += self.vm_b.get_qmp_events() + self.vm_b.shutdown() + # recreate vm_b, so there is no incoming option, which prevents + # loading bitmaps from disk + self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b) + self.vm_b.launch() + check_bitmaps(self.vm_b, 0) + + result = self.vm_a.qmp('query-status') + assert not result['return']['running'] + self.vm_a_events += self.vm_a.get_qmp_events() + self.vm_a.shutdown() + self.vm_a.launch() + check_bitmaps(self.vm_a, 0) + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2'])tests/qemu-iotests/199.out +0
-0
tests/qemu-iotests/199.out line 5
Quoted 7 lines
tests/qemu-iotests/199.out +2
-2@@ -1,5 +1,5 @@ -. +.. ---------------------------------------------------------------------- -Ran 1 tests +Ran 2 tests OK
The updated test passed.
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 18/22] python/qemu/machine: add kill() method
In reply to Vladimir Sementsov-OgievskiyQuoted 43 lines · Vladimir Sementsov-Ogievskiy
Add method to hard-kill vm, without any quit commands.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- python/qemu/machine.py | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)python/qemu/machine.py +6
-3@@ -341,7 +341,7 @@ class QEMUMachine(object): self._load_io_log() self._post_shutdown() - def shutdown(self, has_quit=False): + def shutdown(self, has_quit=False, hard=False): """ Terminate the VM and clean up """@@ -353,7 +353,9 @@ class QEMUMachine(object): self._console_socket = None if self.is_running(): - if self._qmp: + if hard: + self._popen.kill() + elif self._qmp: try: if not has_quit: self._qmp.cmd('quit')@@ -366,7 +368,8 @@ class QEMUMachine(object): self._post_shutdown() exitcode = self.exitcode() - if exitcode is not None and exitcode < 0: + if exitcode is not None and exitcode < 0 and \ + not (exitcode == -9 and hard): msg = 'qemu received signal %i: %s' if self._qemu_full_args: command = ' '.join(self._qemu_full_args)
python/qemu/machine.py class QEMUMachine(object): · line 387
Quoted 9 lines
python/qemu/machine.py +3
-0@@ -376,6 +379,9 @@ class QEMUMachine(object): self._launched = False + def kill(self): + self.shutdown(hard=True) + def set_qmp_monitor(self, enabled=True): """ Set the QMP monitor.
It would be reasonable to number this patch the last but one, that's right before a usage in the patch that follows.
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 21/22] qemu-iotests/199: add early shutdown case to bitmaps postcopy
In reply to Vladimir Sementsov-OgievskiyQuoted 18 lines · Vladimir Sementsov-Ogievskiy
Previous patches fixed two crashes which may occur on shutdown prior to bitmaps postcopy finished. Check that it works now.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 18 ++++++++++++++++++ tests/qemu-iotests/199.out | 4 ++-- 2 files changed, 20 insertions(+), 2 deletions(-)tests/qemu-iotests/199 +1
-0@@ -217,6 +217,24 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): sha = self.discards1_sha256 if i % 2 else self.all_discards_sha256 self.assert_qmp(result, 'return/sha256', sha) + def test_early_shutdown_destination(self):
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 230
Quoted 10 lines
tests/qemu-iotests/199 +10
-0+ self.start_postcopy() + + self.vm_b_events += self.vm_b.get_qmp_events() + self.vm_b.shutdown() + # recreate vm_b, so there is no incoming option, which prevents + # loading bitmaps from disk + self.vm_b = iotests.VM(path_suffix='b').add_drive(disk_b) + self.vm_b.launch() + check_bitmaps(self.vm_b, 0) +
Comments would help to understand the idea behind
Quoted 23 lines
tests/qemu-iotests/199 +7
-0+ result = self.vm_a.qmp('query-status') + assert not result['return']['running'] + self.vm_a_events += self.vm_a.get_qmp_events() + self.vm_a.shutdown() + self.vm_a.launch() + check_bitmaps(self.vm_a, 0) + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2'])tests/qemu-iotests/199.out +2
-2@@ -1,5 +1,5 @@ -. +.. ---------------------------------------------------------------------- -Ran 1 tests +Ran 2 tests OK
Signature · 2 lines
With the best regards, Andrey Shinkevich
Re: [PATCH v2 22/22] qemu-iotests/199: add source-killed case to bitmaps postcopy
In reply to Vladimir Sementsov-OgievskiyQuoted 18 lines · Vladimir Sementsov-Ogievskiy
Previous patches fixes behavior of bitmaps migration, so that errors are handled by just removing unfinished bitmaps, and not fail or try to recover postcopy migration. Add corresponding test.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 15 +++++++++++++++ tests/qemu-iotests/199.out | 4 ++-- 2 files changed, 17 insertions(+), 2 deletions(-)tests/qemu-iotests/199 +0
-0@@ -235,6 +235,21 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.launch() check_bitmaps(self.vm_a, 0)
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 247
Quoted 10 lines
tests/qemu-iotests/199 +10
-0+ def test_early_kill_source(self): + self.start_postcopy() + + self.vm_a_events = self.vm_a.get_qmp_events() + self.vm_a.kill() + + self.vm_a.launch() + + match = {'data': {'status': 'completed'}} + e_complete = self.vm_b.event_wait('MIGRATION', match=match)
A failed migration gets the status 'completed'. That misleads a user but is not in the scope of this series, I guess.
Quoted 14 lines
tests/qemu-iotests/199 +5
-0+ self.vm_b_events.append(e_complete) + + check_bitmaps(self.vm_a, 0) + check_bitmaps(self.vm_b, 0) + if __name__ == '__main__': iotests.main(supported_fmts=['qcow2'])tests/qemu-iotests/199.out +0
-0
tests/qemu-iotests/199.out line 5
Quoted 7 lines
tests/qemu-iotests/199.out +2
-2@@ -1,5 +1,5 @@ -.. +... ---------------------------------------------------------------------- -Ran 2 tests +Ran 3 tests OK
The updated test passed.
Signature · 2 lines
With the best regards, Andrey Shinkevich
Quoted 8 lines · Eric Blake
Quoted 6 lines · Andrey Shinkevich
Quoted 5 lines
Review
Quoted 3 lines
(quoted diff) +3
-0+od: unrecognized option '--endian=big' +Try 'od --help' for more information. +od: invalid -N argument '--endian=big'Yay, same problem for both tests. Fix common.rc once, and both tests should start working for you.
Quoted 6 lines
Quoted 1 line
Thank you Eric! I want to sort it out later...
Patch proposed:
https://lists.gnu.org/archive/html/qemu-devel/2020-02/msg05188.html
Thank you Eric, I appreciate.
Signature · 2 lines
With the best regards, Andrey Shinkevich
Ping!
It's a fix, but not a degradation and I'm afraid too big for 5.0.
Still, I think I should ping it anyway. John, I'm afraid, that this all is for your branch :)
Quoted 64 lines · Vladimir Sementsov-Ogievskiy
Original idea of bitmaps postcopy migration is that bitmaps are non critical data, and their loss is not serious problem. So, using postcopy method on any failure we should just drop unfinished bitmaps and continue guest execution.
However, it doesn't work so. It crashes, fails, it goes to postcopy-recovery feature. It does anything except for behavior we want. These series fixes at least some problems with error handling during bitmaps migration postcopy.
v1 was "[PATCH 0/7] Fix crashes on early shutdown during bitmaps postcopy"
v2:
Most of patches are new or changed a lot.
Only patches 06,07 mostly unchanged, just rebased on refactorings.Vladimir Sementsov-Ogievskiy (22):
migration/block-dirty-bitmap: fixdirty_bitmap_mig_before_vm_startmigration/block-dirty-bitmap: rename state structure types migration/block-dirty-bitmap: renamedirty_bitmap_mig_cleanupmigration/block-dirty-bitmap: move mutex init todirty_bitmap_mig_initmigration/block-dirty-bitmap: refactor state global variables migration/block-dirty-bitmap: renamefinish_lockto just lock migration/block-dirty-bitmap: simplifydirty_bitmap_load_completemigration/block-dirty-bitmap: keep bitmap state for all bitmaps migration/block-dirty-bitmap: relax error handling in incoming part migration/block-dirty-bitmap: cancel migration on shutdown migration/savevm: don't worry if bitmap migration postcopy failed qemu-iotests/199: fix style
qemu-iotests/199: drop extra constraints
qemu-iotests/199: better catch postcopy time qemu-iotests/199: improve performance: set bitmap by discard qemu-iotests/199: change discard patterns qemu-iotests/199: increase postcopy period python/qemu/machine: addkill()method
qemu-iotests/199: prepare for new test-cases addition qemu-iotests/199: check persistent bitmaps qemu-iotests/199: add early shutdown case to bitmaps postcopy qemu-iotests/199: add source-killed case to bitmaps postcopyCc John Snow <[email protected]>Cc Vladimir Sementsov-Ogievskiy <[email protected]>Cc Stefan Hajnoczi <[email protected]>Cc Fam Zheng <[email protected]>Cc Juan Quintela <[email protected]>Cc "Dr. David Alan Gilbert" <[email protected]>Cc Eduardo Habkost <[email protected]>Cc Cleber Rosa <[email protected]>Cc Kevin Wolf <[email protected]>Cc Max Reitz <[email protected]>Cc [email protected] # for patch 01migration/migration.h | 3 +- migration/block-dirty-bitmap.c | 444 +++++++++++++++++++++------------ migration/migration.c | 15 +- migration/savevm.c | 37 ++- python/qemu/machine.py | 12 +- tests/qemu-iotests/199 | 244 ++++++++++++++---- tests/qemu-iotests/199.out | 4 +- 7 files changed, 529 insertions(+), 230 deletions(-)
Signature · 2 lines
Best regards, Vladimir
Philippe Mathieu-Daudé
AppliedRe: [PATCH v2 18/22] python/qemu/machine: add kill() method
In reply to Vladimir Sementsov-OgievskiyQuoted 6 lines · Vladimir Sementsov-Ogievskiy
Add method to hard-kill vm, without any quit commands.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- python/qemu/machine.py | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
Thanks, applied to my python-next tree:
https://gitlab.com/philmd/qemu/commits/python-next
Quoted 5 lines · Vladimir Sementsov-Ogievskiy
Ping!
It's a fix, but not a degradation and I'm afraid too big for 5.0.
Still, I think I should ping it anyway. John, I'm afraid, that this all is for your branch :)
Just noticing this thread, now that we've shuffled bitmaps maintainers. Is there anything here that we still need to include in 5.1?
Quoted 69 lines
Quoted 64 lines · Vladimir Sementsov-Ogievskiy
Original idea of bitmaps postcopy migration is that bitmaps are non critical data, and their loss is not serious problem. So, using postcopy method on any failure we should just drop unfinished bitmaps and continue guest execution.
However, it doesn't work so. It crashes, fails, it goes to postcopy-recovery feature. It does anything except for behavior we want. These series fixes at least some problems with error handling during bitmaps migration postcopy.
v1 was "[PATCH 0/7] Fix crashes on early shutdown during bitmaps postcopy"
v2:
Most of patches are new or changed a lot.
Only patches 06,07 mostly unchanged, just rebased on refactorings.Vladimir Sementsov-Ogievskiy (22):
migration/block-dirty-bitmap: fixdirty_bitmap_mig_before_vm_startmigration/block-dirty-bitmap: rename state structure types migration/block-dirty-bitmap: renamedirty_bitmap_mig_cleanupmigration/block-dirty-bitmap: move mutex init todirty_bitmap_mig_initmigration/block-dirty-bitmap: refactor state global variables migration/block-dirty-bitmap: renamefinish_lockto just lock migration/block-dirty-bitmap: simplifydirty_bitmap_load_completemigration/block-dirty-bitmap: keep bitmap state for all bitmaps migration/block-dirty-bitmap: relax error handling in incoming part migration/block-dirty-bitmap: cancel migration on shutdown migration/savevm: don't worry if bitmap migration postcopy failed qemu-iotests/199: fix style
qemu-iotests/199: drop extra constraints
qemu-iotests/199: better catch postcopy time qemu-iotests/199: improve performance: set bitmap by discard qemu-iotests/199: change discard patterns qemu-iotests/199: increase postcopy period python/qemu/machine: addkill()method
qemu-iotests/199: prepare for new test-cases addition qemu-iotests/199: check persistent bitmaps qemu-iotests/199: add early shutdown case to bitmaps postcopy qemu-iotests/199: add source-killed case to bitmaps postcopyCc John Snow <[email protected]>Cc Vladimir Sementsov-Ogievskiy <[email protected]>Cc Stefan Hajnoczi <[email protected]>Cc Fam Zheng <[email protected]>Cc Juan Quintela <[email protected]>Cc "Dr. David Alan Gilbert" <[email protected]>Cc Eduardo Habkost <[email protected]>Cc Cleber Rosa <[email protected]>Cc Kevin Wolf <[email protected]>Cc Max Reitz <[email protected]>Cc [email protected] # for patch 01migration/migration.h | 3 +-
migration/block-dirty-bitmap.c| 444 +++++++++++++++++++++------------ migration/migration.c | 15 +-
migration/savevm.c | 37 ++- python/qemu/machine.py | 12 +- tests/qemu-iotests/199 | 244 ++++++++++++++---- tests/qemu-iotests/199.out | 4 +-7 files changed, 529 insertions(+), 230 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Quoted 8 lines · Eric Blake
Quoted 5 lines · Vladimir Sementsov-Ogievskiy
Ping!
It's a fix, but not a degradation and I'm afraid too big for 5.0.
Still, I think I should ping it anyway. John, I'm afraid, that this all is for your branch :)
Just noticing this thread, now that we've shuffled bitmaps maintainers. Is there anything here that we still need to include in 5.1?
Yes, we need the whole series.
Quoted 71 lines
Quoted 70 lines
Quoted 64 lines · Vladimir Sementsov-Ogievskiy
Original idea of bitmaps postcopy migration is that bitmaps are non critical data, and their loss is not serious problem. So, using postcopy method on any failure we should just drop unfinished bitmaps and continue guest execution.
However, it doesn't work so. It crashes, fails, it goes to postcopy-recovery feature. It does anything except for behavior we want. These series fixes at least some problems with error handling during bitmaps migration postcopy.
v1 was "[PATCH 0/7] Fix crashes on early shutdown during bitmaps postcopy"
v2:
Most of patches are new or changed a lot.
Only patches 06,07 mostly unchanged, just rebased on refactorings.Vladimir Sementsov-Ogievskiy (22):
migration/block-dirty-bitmap: fixdirty_bitmap_mig_before_vm_startmigration/block-dirty-bitmap: rename state structure types migration/block-dirty-bitmap: renamedirty_bitmap_mig_cleanupmigration/block-dirty-bitmap: move mutex init todirty_bitmap_mig_initmigration/block-dirty-bitmap: refactor state global variables migration/block-dirty-bitmap: renamefinish_lockto just lock migration/block-dirty-bitmap: simplifydirty_bitmap_load_completemigration/block-dirty-bitmap: keep bitmap state for all bitmaps migration/block-dirty-bitmap: relax error handling in incoming part migration/block-dirty-bitmap: cancel migration on shutdown migration/savevm: don't worry if bitmap migration postcopy failed qemu-iotests/199: fix style
qemu-iotests/199: drop extra constraints
qemu-iotests/199: better catch postcopy time qemu-iotests/199: improve performance: set bitmap by discard qemu-iotests/199: change discard patterns qemu-iotests/199: increase postcopy period python/qemu/machine: addkill()method
qemu-iotests/199: prepare for new test-cases addition qemu-iotests/199: check persistent bitmaps qemu-iotests/199: add early shutdown case to bitmaps postcopy qemu-iotests/199: add source-killed case to bitmaps postcopyCc John Snow <[email protected]>Cc Vladimir Sementsov-Ogievskiy <[email protected]>Cc Stefan Hajnoczi <[email protected]>Cc Fam Zheng <[email protected]>Cc Juan Quintela <[email protected]>Cc "Dr. David Alan Gilbert" <[email protected]>Cc Eduardo Habkost <[email protected]>Cc Cleber Rosa <[email protected]>Cc Kevin Wolf <[email protected]>Cc Max Reitz <[email protected]>Cc [email protected] # for patch 01migration/migration.h | 3 +-
migration/block-dirty-bitmap.c| 444 +++++++++++++++++++++------------ migration/migration.c | 15 +-
migration/savevm.c | 37 ++- python/qemu/machine.py | 12 +- tests/qemu-iotests/199 | 244 ++++++++++++++---- tests/qemu-iotests/199.out | 4 +-7 files changed, 529 insertions(+), 230 deletions(-)
Signature · 2 lines
Best regards, Vladimir
Quoted 11 lines · Vladimir Sementsov-Ogievskiy
Quoted 8 lines · Eric Blake
Quoted 5 lines · Vladimir Sementsov-Ogievskiy
Ping!
It's a fix, but not a degradation and I'm afraid too big for 5.0.
Still, I think I should ping it anyway. John, I'm afraid, that this all is for your branch :)
Just noticing this thread, now that we've shuffled bitmaps maintainers. Is there anything here that we still need to include in 5.1?
Yes, we need the whole series.
I'm starting to go through it now, to see what is still worth getting in to 5.1-rc2, but no promises as it is a long series and I don't want to introduce last-minute regressions (the fact that this missed 5.0 says that 5.1 will be no worse than 5.0 if we don't get this in until 5.2).
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 02/22] migration/block-dirty-bitmap: rename state structure types
In reply to Vladimir Sementsov-OgievskiyQuoted 6 lines · Vladimir Sementsov-Ogievskiy
Rename types to be symmetrical for load/save part and shorter.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 68 ++++++++++++++++++---------------- 1 file changed, 36 insertions(+), 32 deletions(-)
No longer applies to master, but the mechanical aspect of the change makes sense. If you rebase the series to make review easier,
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Eric Blake
Reviewed-byRe: [PATCH v2 03/22] migration/block-dirty-bitmap: rename dirty_bitmap_mig_cleanup
In reply to Vladimir Sementsov-OgievskiyQuoted 11 lines · Vladimir Sementsov-Ogievskiy
Quoted 10 lines · Andrey Shinkevich
Quoted 7 lines · Vladimir Sementsov-Ogievskiy
Rename
dirty_bitmap_mig_cleanuptodirty_bitmap_do_save_cleanup, to stress that it is on save part.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 8 ++++----1 file changed, 4 insertions(+), 4 deletions(-)
Quoted 6 lines
Quoted 4 lines
At the next opportunity, I would suggest the name like "
dirty_bitmap_do_clean_after_saving()"
and similar fordirty_bitmap_save_cleanup()"dirty_bitmap_clean_after_saving()".I'd keep my naming, it corresponds to .save_cleanup handler name.
I'm fine with that explanation, so no need to rename again.
Quoted 5 lines
Quoted 3 lines
Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 10/22] migration/block-dirty-bitmap: cancel migration on shutdown
In reply to Vladimir Sementsov-OgievskiyQuoted 1 line · Vladimir Sementsov-Ogievskiy
If target is turned of prior to postcopy finished, target crashes
s/of/off/
Quoted 12 lines
because busy bitmaps are found at shutdown.
Canceling incoming migration helps, as it removes all unfinished (and therefore busy) bitmaps.Similarly on source we crash in
bdrv_close_allwhich asserts that all bdrv states are removed, because bdrv states involved into dirty bitmap migration are referenced by it. So, we need to cancel outgoing migration as well.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 08/22] migration/block-dirty-bitmap: keep bitmap state for all bitmaps
In reply to Vladimir Sementsov-OgievskiyQuoted 33 lines · Vladimir Sementsov-Ogievskiy
Keep bitmap state for disabled bitmaps too. Keep the state until the end of the process. It's needed for the following commit to implement bitmap postcopy canceling.
To clean-up the new list the following logic is used:
We need two events to consider bitmap migration finished:
1. chunk withDIRTY_BITMAP_MIG_FLAG_COMPLETEflag should be received
2.dirty_bitmap_mig_before_vm_startshould be called
These two events may come in any order, so we understand which one is last, and on the last of them we remove bitmap migration state from the list.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- migration/block-dirty-bitmap.c | 64 +++++++++++++++++++++++----------- 1 file changed, 43 insertions(+), 21 deletions(-)migration/block-dirty-bitmap.c +0
-7@@ -484,45 +488,59 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s)bdrv_disable_dirty_bitmap (s->bitmap);if (flags &DIRTY_BITMAP_MIG_START_FLAG_ENABLED ) { -LoadBitmapState *b; -bdrv_dirty_bitmap_create_successor (s->bitmap, &local_err);if (local_err) {error_report_err (local_err);return -EINVAL ; } - - b =g_new (LoadBitmapState,1 ); - b->bs = s->bs; - b->bitmap = s->bitmap; - b->migrated = false;
migration/block-dirty-bitmap.c static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s) · line 504
Quoted 10 lines
migration/block-dirty-bitmap.c +7
-1- s->enabled_bitmaps =g_slist_prepend (s->enabled_bitmaps, b); } + b =g_new (LoadBitmapState,1 ); + b->bs = s->bs; + b->bitmap = s->bitmap; + b->migrated = false; + b->enabled = flags &DIRTY_BITMAP_MIG_START_FLAG_ENABLED , + + s->bitmaps =g_slist_prepend (s->bitmaps, b);
Did you really mean to use a comma operator there, or should that be ';'?
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 12/22] qemu-iotests/199: fix style
In reply to Vladimir Sementsov-OgievskiyQuoted 6 lines · Vladimir Sementsov-Ogievskiy
Mostly, satisfy pep8 complains.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-)
With none of your series applied, I get:
$ ./check -qcow2 199 ... 199 not run [16:52:34] [16:52:34] not suitable for this cache mode: writeback Not run: 199 Passed all 0 iotests 199 fail [16:53:37] [16:53:37] output mismatch (see 199.out.bad) --- /home/eblake/qemu/tests/qemu-iotests/199.out 2020-07-23 16:48:56.275529368 -0500 +++ /home/eblake/qemu/build/tests/qemu-iotests/199.out.bad 2020-07-23 16:53:37.728416207 -0500 @@ -1,5 +1,13 @@ -. +E +====================================================================== +ERROR: test_postcopy (__main__.TestDirtyBitmapPostcopyMigration) +---------------------------------------------------------------------- +Traceback (most recent call last): + File "199", line 41, in setUp + os.mkfifo(fifo) +FileExistsError: [Errno 17] File exists + ---------------------------------------------------------------------- Ran 1 tests
-OK
+FAILED (errors=1)
Failures: 199
Failed 1 of 1 iotests
Ah, 'scratch/mig_fifo' was left over from some other aborted run of the test. I removed that file (which implies it might be nice if the test handled that automatically, instead of making me do it), and tried again; now I got the desired:
199 pass [17:00:34] [17:01:48] 74s
Passed all 1 iotests
After trying to rebase your series, I once again got failures, but that could mean I botched the rebase (since quite a few of the code patches earlier in the series were non-trivially changed). If you send a v3 (which would be really nice!), I'd hoist this and 13/22 first in the series, to get to a point where testing 199 works, to then make it easier to demonstrate what the rest of the 199 enhancements do in relation to the non-iotest patches. But I like that you separated the 199 improvements from the code - testing-wise, it's easy to apply the iotests patches first, make sure it fails, then apply the code patches, and make sure it passes, to prove that the enhanced test now covers what the code fixes did.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 17/22] qemu-iotests/199: increase postcopy period
In reply to Vladimir Sementsov-OgievskiyQuoted 1 line · Vladimir Sementsov-Ogievskiy
Test wants force bitmap postcopy. Still, resulting postcopy period is
The test wants to force a bitmap postcopy. Still, the resulting postcopy period is very small.
Quoted 7 lines
very small. Let's increase it by adding more bitmaps to migrate. Also, test disabled bitmaps migration.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 58 ++++++++++++++++++++++++++++-------------- 1 file changed, 39 insertions(+), 19 deletions(-)
Patches 12-17:
As they all work without any other patches in this series, and DO make a dramatic difference (cutting the test from over 70 seconds to just 7, on my machine), I'm inclined to stage them now, even while waiting for you to rebase the rest of the series. And 18 is already in the tree.
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 16/22] qemu-iotests/199: change discard patterns
In reply to Vladimir Sementsov-OgievskiyQuoted 1 line · Vladimir Sementsov-Ogievskiy
iotest 40 works too long because of many discard opertion. On the same
I'm assuming you meant s/40/199/ here, as well as the typo fixes pointed out by Andrey.
Quoted 10 lines
time, postcopy period is very short, in spite of all these efforts.
So, let's use less discards (and with more interesting patterns) to reduce test timing. In the next commit we'll increase postcopy period.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>--- tests/qemu-iotests/199 | 44 +++++++++++++++++++++++++----------------- 1 file changed, 26 insertions(+), 18 deletions(-)
Signature · 3 lines
Eric Blake, Principal Software Engineer Red Hat, Inc. +1-919-301-3226 Virtualization: qemu.org | libvirt.org
Re: [PATCH v2 08/22] migration/block-dirty-bitmap: keep bitmap state for all bitmaps
In reply to Eric BlakeQuoted 36 lines · Eric Blake
Quoted 35 lines · Vladimir Sementsov-Ogievskiy
Keep bitmap state for disabled bitmaps too. Keep the state until the end of the process. It's needed for the following commit to implement bitmap postcopy canceling.
To clean-up the new list the following logic is used:
We need two events to consider bitmap migration finished:
1. chunk withDIRTY_BITMAP_MIG_FLAG_COMPLETEflag should be received
2.dirty_bitmap_mig_before_vm_startshould be called
These two events may come in any order, so we understand which one is last, and on the last of them we remove bitmap migration state from the list.Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 64 +++++++++++++++++++++++-----------1 file changed, 43 insertions(+), 21 deletions(-)migration/block-dirty-bitmap.c +1
-8@@ -484,45 +488,59 @@ static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s) bdrv_disable_dirty_bitmap(s->bitmap); if (flags & DIRTY_BITMAP_MIG_START_FLAG_ENABLED) { - LoadBitmapState *b; - bdrv_dirty_bitmap_create_successor(s->bitmap, &local_err); if (local_err) { error_report_err(local_err); return -EINVAL; } - - b = g_new(LoadBitmapState, 1); - b->bs = s->bs; - b->bitmap = s->bitmap; - b->migrated = false; - s->enabled_bitmaps = g_slist_prepend(s->enabled_bitmaps, b); } + b = g_new(LoadBitmapState, 1);
Quoted 9 lines
migration/block-dirty-bitmap.c static int dirty_bitmap_load_start(QEMUFile *f, DBMLoadState *s) · line 504
Quoted 6 lines
migration/block-dirty-bitmap.c +6
-0+ b->bs = s->bs; + b->bitmap = s->bitmap; + b->migrated = false; + b->enabled = flags & DIRTY_BITMAP_MIG_START_FLAG_ENABLED, + + s->bitmaps = g_slist_prepend(s->bitmaps, b);Did you really mean to use a comma operator there, or should that be ';'?
Of course, it should be ';':)
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 12/22] qemu-iotests/199: fix style
In reply to Eric BlakeQuoted 37 lines · Eric Blake
Quoted 6 lines · Vladimir Sementsov-Ogievskiy
Mostly, satisfy pep8 complains.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
tests/qemu-iotests/199 | 13 +++++++------1 file changed, 7 insertions(+), 6 deletions(-)With none of your series applied, I get:
$ ./check -qcow2 199 ... 199 not run [16:52:34] [16:52:34] not suitable for this cache mode: writeback Not run: 199 Passed all 0 iotests 199 fail [16:53:37] [16:53:37] output mismatch (see 199.out.bad)home/eblake/qemu/build/tests/qemu-iotests/199.out.bad 2020-07-23 16:53:37.728416207 -0500 renamed from home/eblake/qemu/tests/qemu-iotests/199.out 2020-07-23 16:48:56.275529368 -0500 +10
-2@@ -1,5 +1,13 @@ -. +E +====================================================================== +ERROR: test_postcopy (__main__.TestDirtyBitmapPostcopyMigration) +---------------------------------------------------------------------- +Traceback (most recent call last): + File "199", line 41, in setUp + os.mkfifo(fifo) +FileExistsError: [Errno 17] File exists + ---------------------------------------------------------------------- Ran 1 tests -OK +FAILED (errors=1)Failures: 199
Failed 1 of 1 iotests
Quoted 8 lines
Ah, 'scratch/mig_fifo' was left over from some other aborted run of the test. I removed that file (which implies it might be nice if the test handled that automatically, instead of making me do it), and tried again; now I got the desired:
199 pass [17:00:34] [17:01:48] 74s
Passed all 1 iotestsAfter trying to rebase your series, I once again got failures, but that could mean I botched the rebase (since quite a few of the code patches earlier in the series were non-trivially changed).> If you send a v3 (which would be really nice!), I'd hoist this and 13/22 first in the series, to get to a point where testing 199 works, to then make it easier to demonstrate what the rest of the 199 enhancements do in relation to the non-iotest patches. But I like that you separated the 199 improvements from the code - testing-wise, it's easy to apply the iotests patches first, make sure it fails, then apply the code patches, and make sure it passes, to prove that the enhanced test now covers what the code fixes did.
A bit off-topic:
Yes, that's our usual scheme: test goes after bug-fix, so careful reviewers always have to apply patches in different order to check is there a real bug-fix.. And the only benefit of such scheme - not to break git bisect. Still, I think we can do better:
- apply test first, just put it into "bug" group
- check script should ignore "bug" group (unless it explicitly specified, or direct test run)
- bug-fix patch removes test from "bug" group
So bisect is not broken and we achieve native ordering: problem exists (test fails) -> fixing -> no problem (test pass).
I think, I'll add "bug" group as a follow up to my "[PATCH v4 0/9] Rework iotests/check", which I really hope will land one day.
PS: I've successfully rebased the series, and test pass. I'll now fix all review notes and resend soon.
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 14/22] qemu-iotests/199: better catch postcopy time
In reply to Andrey ShinkevichQuoted 43 lines · Andrey Shinkevich
Quoted 42 lines · Vladimir Sementsov-Ogievskiy
The test aims to test _postcopy_ migration, and wants to do some write operations during postcopy time.
Test considers migrate status=complete event on source as start of postcopy. This is completely wrong, completion is completion of the whole migration process. Let's instead consider destination start as start of postcopy, and use RESUME event for it.
Next, as migration finish, let's use migration status=complete event on target, as such method is closer to what libvirt or another user will do, than tracking number of dirty-bitmaps.
Finally, add a possibility to dump events for debug. And if set debug to True, we see, that actual postcopy period is very small relatively to the whole test duration time (~0.2 seconds to >40 seconds for me). This means, that test is very inefficient in what it supposed to do. Let's improve it in following commits.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
tests/qemu-iotests/199 | 72 +++++++++++++++++++++++++++++++++---------1 file changed, 57 insertions(+), 15 deletions(-)tests/qemu-iotests/199 +0
-0@@ -20,17 +20,43 @@import os
import iotests
-import time
from iotests importqemu_imgtests/qemu-iotests/199 +2
-0+debug = False +
disk_a=os.path.join(iotests.test_dir, 'disk_a')disk_b=os.path.join(iotests.test_dir, 'disk_b')size = '256G'
fifo =os.path.join(iotests.test_dir, 'mig_fifo')tests/qemu-iotests/199 +4
-0+def event_seconds(event): + return event['timestamp']['seconds'] + \ + event['timestamp']['microseconds'] / 1000000.0 +
Quoted 8 lines
Quoted 7 lines
tests/qemu-iotests/199 +5
-0+ +def event_dist(e1, e2): + return event_seconds(e2) - event_seconds(e1) + +class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):
def tearDown(self):It's common to put the definition of
setUp()ahead
Preexisting, I don't want to update it in this patch
Quoted 29 lines
Quoted 29 lines
+ if debug:
+ self.vm_a_events +=self.vm_a.get_qmp_events()+ self.vm_b_events +=self.vm_b.get_qmp_events()+ for e in self.vm_a_events:
+ e['vm'] = 'SRC'
+ for e in self.vm_b_events:
+ e['vm'] = 'DST'
+ events = (self.vm_a_events + self.vm_b_events) + events = [(e['timestamp']['seconds'], + e['timestamp']['microseconds'], + e['vm'],
+ e['event'],
+e.get('data', '')) for e in events] + for e in sorted(events):
+print('{}.{:06} {} {} {}'.format(*e))+
self.vm_a.shutdown()self.vm_b.shutdown()
os.remove(disk_a)tests/qemu-iotests/199 +4
-0@@ -47,6 +73,10 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.launch() self.vm_b.launch() + # collect received events for debug + self.vm_a_events = [] + self.vm_b_events = [] +def
test_postcopy(self):
write_size=0x40000000
Quoted 8 lines
Quoted 7 lines
granularity = 512
tests/qemu-iotests/199 +0
-0@@ -77,15 +107,13 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk))s +=
0x10000
-bitmaps_cap= {'capability': 'dirty-bitmaps', 'state': True} -events_cap= {'capability': 'events', 'state': True} + caps = [{'capability': 'dirty-bitmaps', 'state': True},The name "capabilities" would be an appropriate identifier.
This will result in following lines growing and not fit into one line. I'll leave "caps". Also, they are called "caps" in iotest 169 and in migration.c. And here in the context always used together with full word ('capability': or capabilities=).
Quoted 12 lines
Quoted 12 lines
+ {'capability': 'events', 'state': True}]
- result = self.vm_a.qmp('migrate-set-capabilities',
- capabilities=[bitmaps_cap,events_cap]) + result =self.vm_a.qmp('migrate-set-capabilities', capabilities=caps)self.assert_qmp(result, 'return', {})
- result = self.vm_b.qmp('migrate-set-capabilities',
- capabilities=[bitmaps_cap])
+ result =self.vm_b.qmp('migrate-set-capabilities', capabilities=caps)self.assert_qmp(result, 'return', {})
result =self.vm_a.qmp('migrate', uri='exec:cat>' + fifo)tests/qemu-iotests/199 +0
-0@@ -94,24 +122,38 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase):
Quoted 8 lines
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 125
Quoted 7 lines
tests/qemu-iotests/199 +1
-4result = self.vm_a.qmp('migrate-start-postcopy') self.assert_qmp(result, 'return', {}) - while True: - event = self.vm_a.event_wait('MIGRATION') - if event['data']['status'] == 'completed': - break + e_resume = self.vm_b.event_wait('RESUME')"
event_resume" gives a faster understanding
OK, no problem
Quoted 9 lines
Quoted 8 lines
tests/qemu-iotests/199 +1
-0+ self.vm_b_events.append(e_resume)s =
0x8000
while s <write_size:
self.vm_b.hmp_qemu_io('drive0', 'write %d %d' % (s, chunk))s +=0x10000
+ match = {'data': {'status': 'completed'}}
+e_complete=self.vm_b.event_wait('MIGRATION', match=match)"
event_complete" also
OK
Quoted 6 lines
Quoted 5 lines
+
self.vm_b_events.append(e_complete)
+
+ # take queued event, should already been happened +e_stop=self.vm_a.event_wait('STOP')"
event_stop"
OK
Quoted 36 lines
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 144
Quoted 7 lines
+
self.vm_a_events.append(e_stop)tests/qemu-iotests/199 +5
-0+ + downtime = event_dist(e_stop, e_resume) + postcopy_time = event_dist(e_resume, e_complete) + + # TODO: assert downtime * 10 < postcopy_timeI got the results below in debug mode:
downtime: 6.194924831390381
postcopy_time: 0.1592559814453125
1582102669.764919 SRC MIGRATION {'status': 'setup'} 1582102669.766179 SRCMIGRATION_PASS{'pass': 1} 1582102669.766234 SRC MIGRATION {'status': 'active'} 1582102669.768058 DST MIGRATION {'status': 'active'} 1582102669.801422 SRC MIGRATION {'status': 'postcopy-active'} 1582102669.801510 SRC STOP
1582102675.990041 DST MIGRATION {'status': 'postcopy-active'} 1582102675.996435 DST RESUME
1582102676.111313 SRC MIGRATION {'status': 'completed'} 1582102676.155691 DST MIGRATION {'status': 'completed'}Quoted 1 line
+ if debug:
with no usage in the following patches, you can put the whole block of relative code above under the "if debug: section
Quoted 10 lines
+
print('downtime:', downtime)
+print('postcopy_time:', postcopy_time)
+
+ # Assert that bitmap migration is finished (check that successor bitmap + # is removed)
result =self.vm_b.qmp('query-block')
- whilelen(result['return'][0]['dirty-bitmaps'])> 1:
-time.sleep(2)
- result =self.vm_b.qmp('query-block')
+ assertlen(result['return'][0]['dirty-bitmaps'])== 1
Quoted 8 lines
Quoted 6 lines
+ # Check content of migrated (and updated by new writes) bitmap result =
self.vm_b.qmp('x-debug-block-dirty-bitmap-sha256', node='drive0', name='bitmap')-
self.assert_qmp(result, 'return/sha256', sha256)Reviewed-by Andrey Shinkevich <[email protected]>
Thanks!
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 09/22] migration/block-dirty-bitmap: relax error handling in incoming part
In reply to Vladimir Sementsov-OgievskiyQuoted 148 lines · Vladimir Sementsov-Ogievskiy
Quoted 67 lines · Andrey Shinkevich
Quoted 64 lines · Vladimir Sementsov-Ogievskiy
Bitmaps data is not critical, and we should not fail the migration (or use postcopy recovering) because of dirty-bitmaps migration failure. Instead we should just lose unfinished bitmaps.
Still we have to report io stream violation errors, as they affect the whole migration stream.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
migration/block-dirty-bitmap.c| 148 +++++++++++++++++++++++++--------1 file changed, 113 insertions(+), 35 deletions(-)migration/block-dirty-bitmap.c +0
-0@@ -145,6 +145,15 @@ typedef struct DBMLoadState {bool
before_vm_start_handled; /* set indirty_bitmap_mig_before_vm_start*/ + /*
+ * cancelled
+ * Incoming migration is cancelled for some reason. That means that we + * still should read our chunks from migration stream, to not affect other + * migration objects (like RAM), but just ignore them and do not touch any + * bitmaps or nodes.
+ */
+ bool cancelled;
+
GSList *bitmaps;
QemuMutex lock; /* protect bitmaps */
} DBMLoadState;migration/block-dirty-bitmap.c +31
-0@@ -545,13 +554,47 @@ void dirty_bitmap_mig_before_vm_start(void) qemu_mutex_unlock(&s->lock); } +static void cancel_incoming_locked (DBMLoadState *s) +{ + GSList *item; + + if (s->cancelled) { + return; + } + + s->cancelled = true; + s->bs = NULL; + s->bitmap = NULL; + + /* Drop all unfinished bitmaps */ + for (item = s->bitmaps; item; item = g_slist_next(item)) { + LoadBitmapState *b = item->data; + + /* + * Bitmap must be unfinished, as finished bitmaps should already be + * removed from the list. + */ + assert(!s->before_vm_start_handled || !b->migrated); + if (bdrv_dirty_bitmap_has_successor(b->bitmap)) { + bdrv_reclaim_dirty_bitmap(b->bitmap, &error_abort); + } + bdrv_release_dirty_bitmap(b->bitmap); + } + + g_slist_free_full(s->bitmaps, g_free); + s->bitmaps = NULL; +} +Quoted 8 lines
migration/block-dirty-bitmap.c void dirty_bitmap_mig_before_vm_start(void) · line 554
Quoted 6 lines
migration/block-dirty-bitmap.c +0
-2static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) { GSList *item; trace_dirty_bitmap_load_complete(); - bdrv_dirty_bitmap_deserialize_finish(s->bitmap); - qemu_mutex_lock(&s->lock);Why is it safe to remove the critical section?
It's not removed, it becomes wider in this patch.
Quoted 60 lines
Quoted 60 lines
migration/block-dirty-bitmap.c +27
-7+ if (s->cancelled) { + return; + } + + bdrv_dirty_bitmap_deserialize_finish(s->bitmap); if (bdrv_dirty_bitmap_has_successor(s->bitmap)) { bdrv_reclaim_dirty_bitmap(s->bitmap, &error_abort);@@ -569,8 +612,6 @@ static void dirty_bitmap_load_complete(QEMUFile *f, DBMLoadState *s) break; } } - - qemu_mutex_unlock(&s->lock); } static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)@@ -582,15 +623,32 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s) if (s->flags & DIRTY_BITMAP_MIG_FLAG_ZEROES) { trace_dirty_bitmap_load_bits_zeroes(); - bdrv_dirty_bitmap_deserialize_zeroes(s->bitmap, first_byte, nr_bytes, - false); + if (!s->cancelled) { + bdrv_dirty_bitmap_deserialize_zeroes(s->bitmap, first_byte, + nr_bytes, false); + } } else { size_t ret; uint8_t *buf; uint64_t buf_size = qemu_get_be64(f); - uint64_t needed_size = - bdrv_dirty_bitmap_serialization_size(s->bitmap, - first_byte, nr_bytes); + uint64_t needed_size; + + buf = g_malloc(buf_size); + ret = qemu_get_buffer(f, buf, buf_size); + if (ret != buf_size) { + error_report("Failed to read bitmap bits"); + g_free(buf); + return -EIO; + } + + if (s->cancelled) { + g_free(buf); + return 0; + } + + needed_size = bdrv_dirty_bitmap_serialization_size(s->bitmap, + first_byte, + nr_bytes);if (
needed_size>buf_size||
buf_size>QEMU_ALIGN_UP(needed_size, 4 * sizeof(long))migration/block-dirty-bitmap.c +0
-0@@ -599,15 +657,8 @@ static int dirty_bitmap_load_bits(QEMUFile *f, DBMLoadState *s)error_report("Migrated bitmap granularity doesn't " "match the destination bitmap '%s' granularity",
bdrv_dirty_bitmap_name(s->bitmap)); - return-EINVAL;
- }
-
- buf =g_malloc(buf_size);Quoted 8 lines
Quoted 6 lines
- ret =
qemu_get_buffer(f, buf, buf_size); - if (ret !=buf_size) {
-error_report("Failed to read bitmap bits"); -g_free(buf);
- return-EIO;
+cancel_incoming_locked(s);/* Continue the VM migration as bitmaps data are not critical */
Quoted 20 lines
Hmm yes it what this patch does.. But I don't think we should add comment to each call of cancel_..()
Quoted 18 lines
Quoted 6 lines
+ return 0;
}
bdrv_dirty_bitmap_deserialize_part(s->bitmap, buf,first_byte,nr_bytes,migration/block-dirty-bitmap.c +0
-0@@ -632,14 +683,16 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) error_report("Unable to read node name string");migration/block-dirty-bitmap.c static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) · line 689
Quoted 10 lines
migration/block-dirty-bitmap.c +4
-4return -EINVAL; } - s->bs = bdrv_lookup_bs(s->node_name, s->node_name, &local_err); - if (!s->bs) { - error_report_err(local_err); - return -EINVAL; + if (!s->cancelled) { + s->bs = bdrv_lookup_bs(s->node_name, s->node_name, &local_err); + if (!s->bs) { + error_report_err(local_err);The error message may be supplemented with a report about the canceled bitmap migration. Also down there at
cancel_incoming_locked(s).
If we want to log something about cancelling, I think it should be done in cancel_incoming_locked(). I'll keep this patch as is.
Quoted 94 lines
Quoted 92 lines
Quoted 44 lines
migration/block-dirty-bitmap.c +19
-13+ cancel_incoming_locked(s); + } } - } else if (!s->bs && !nothing) { + } else if (!s->bs && !nothing && !s->cancelled) { error_report("Error: block device name is not set"); - return -EINVAL; + cancel_incoming_locked(s); } if (s->flags & DIRTY_BITMAP_MIG_FLAG_BITMAP_NAME) {@@ -647,24 +700,38 @@ static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) error_report("Unable to read bitmap name string"); return -EINVAL; } - s->bitmap = bdrv_find_dirty_bitmap(s->bs, s->bitmap_name); - - /* bitmap may be NULL here, it wouldn't be an error if it is the - * first occurrence of the bitmap */ - if (!s->bitmap && !(s->flags & DIRTY_BITMAP_MIG_FLAG_START)) { - error_report("Error: unknown dirty bitmap " - "'%s' for block device '%s'", - s->bitmap_name, s->node_name); - return -EINVAL; + if (!s->cancelled) { + s->bitmap = bdrv_find_dirty_bitmap(s->bs, s->bitmap_name); + + /* + * bitmap may be NULL here, it wouldn't be an error if it is the + * first occurrence of the bitmap + */ + if (!s->bitmap && !(s->flags & DIRTY_BITMAP_MIG_FLAG_START)) { + error_report("Error: unknown dirty bitmap " + "'%s' for block device '%s'", + s->bitmap_name, s->node_name); + cancel_incoming_locked(s); + } } - } else if (!s->bitmap && !nothing) { + } else if (!s->bitmap && !nothing && !s->cancelled) { error_report("Error: block device name is not set"); - return -EINVAL; + cancel_incoming_locked(s); }migration/block-dirty-bitmap.c static int dirty_bitmap_load_header(QEMUFile *f, DBMLoadState *s) · line 730
Quoted 10 lines
migration/block-dirty-bitmap.c +8
-0return 0; } +/* +* dirty_bitmap_load +* +* Load sequence of dirty bitmap chunks. Return error only on fatal io stream +* violations. On other errors just cancel bitmaps incoming migration and return +* 0. +* +* Note, than when incoming bitmap migration is canceled, we still must read all "than (that)" may be omitted
Quoted 24 lines
migration/block-dirty-bitmap.c +9
-0+* our chunks (and just ignore them), to not affect other migration objects. +*/ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) { DBMLoadState *s = &((DBMState *)opaque)->load;@@ -673,12 +740,19 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) trace_dirty_bitmap_load_enter(); if (version_id != 1) { + qemu_mutex_lock(&s->lock); + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return -EINVAL; } do { + qemu_mutex_lock(&s->lock); + ret = dirty_bitmap_load_header(f, s); if (ret < 0) { + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return ret; }@@ -695,8 +769,12 @@ static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) }migration/block-dirty-bitmap.c static int dirty_bitmap_load(QEMUFile *f, void *opaque, int version_id) · line 779
Quoted 10 lines
migration/block-dirty-bitmap.c +4
-0if (ret) { + cancel_incoming_locked(s); + qemu_mutex_unlock(&s->lock); return ret; } + + qemu_mutex_unlock(&s->lock); } while (!(s->flags & DIRTY_BITMAP_MIG_FLAG_EOS)); trace_dirty_bitmap_load_success();Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir
Re: [PATCH v2 22/22] qemu-iotests/199: add source-killed case to bitmaps postcopy
In reply to Andrey ShinkevichQuoted 22 lines · Andrey Shinkevich
Quoted 21 lines · Vladimir Sementsov-Ogievskiy
Previous patches fixes behavior of bitmaps migration, so that errors are handled by just removing unfinished bitmaps, and not fail or try to recover postcopy migration. Add corresponding test.
Signed-off-by Vladimir Sementsov-Ogievskiy <[email protected]>---
tests/qemu-iotests/199 | 15 +++++++++++++++
tests/qemu-iotests/199.out | 4 ++--2 files changed, 17 insertions(+), 2 deletions(-)tests/qemu-iotests/199 +4
-0@@ -235,6 +235,21 @@ class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): self.vm_a.launch() check_bitmaps(self.vm_a, 0) + def test_early_kill_source(self): + self.start_postcopy() + + self.vm_a_events = self.vm_a.get_qmp_events()
Quoted 8 lines
tests/qemu-iotests/199 class TestDirtyBitmapPostcopyMigration(iotests.QMPTestCase): · line 247
Quoted 6 lines
tests/qemu-iotests/199 +6
-0+ self.vm_a.kill() + + self.vm_a.launch() + + match = {'data': {'status': 'completed'}} + e_complete = self.vm_b.event_wait('MIGRATION', match=match)A failed migration gets the status 'completed'. That misleads a user but is not in the scope of this series, I guess.
It's not failed. Only bitmaps are not migrated, which is not a problem.. Probably we should invent some additional status or QAPI event for this, but yes, not in this series.
Quoted 24 lines
Quoted 18 lines
tests/qemu-iotests/199 +5
-0+ self.vm_b_events.append(e_complete) + + check_bitmaps(self.vm_a, 0) + check_bitmaps(self.vm_b, 0) +if
__name__== '__main__':
iotests.main(supported_fmts=['qcow2'])tests/qemu-iotests/199.out +1
-1@@ -1,5 +1,5 @@ -.. +...----------------------------------------------------------------------
tests/qemu-iotests/199.out +1
-1-Ran 2 tests +Ran 3 testsQuoted 2 lines
OK
The updated test passed.
Reviewed-by Andrey Shinkevich <[email protected]>
Signature · 2 lines
Best regards, Vladimir