2 * QEMU live block migration
4 * Copyright IBM, Corp. 2009
7 * Liran Schour <lirans@il.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qapi/error.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "qemu/cutils.h"
21 #include "qemu/queue.h"
23 #include "migration/misc.h"
24 #include "migration.h"
25 #include "migration/register.h"
26 #include "qemu-file.h"
27 #include "migration/vmstate.h"
28 #include "sysemu/block-backend.h"
31 #define BLK_MIG_BLOCK_SIZE (1 << 20)
32 #define BDRV_SECTORS_PER_DIRTY_CHUNK (BLK_MIG_BLOCK_SIZE >> BDRV_SECTOR_BITS)
34 #define BLK_MIG_FLAG_DEVICE_BLOCK 0x01
35 #define BLK_MIG_FLAG_EOS 0x02
36 #define BLK_MIG_FLAG_PROGRESS 0x04
37 #define BLK_MIG_FLAG_ZERO_BLOCK 0x08
39 #define MAX_IS_ALLOCATED_SEARCH (65536 * BDRV_SECTOR_SIZE)
41 #define MAX_IO_BUFFERS 512
42 #define MAX_PARALLEL_IO 16
44 /* #define DEBUG_BLK_MIGRATION */
46 #ifdef DEBUG_BLK_MIGRATION
47 #define DPRINTF(fmt, ...) \
48 do { printf("blk_migration: " fmt, ## __VA_ARGS__); } while (0)
50 #define DPRINTF(fmt, ...) \
54 typedef struct BlkMigDevState {
55 /* Written during setup phase. Can be read without a lock. */
59 int64_t total_sectors;
60 QSIMPLEQ_ENTRY(BlkMigDevState) entry;
63 /* Only used by migration thread. Does not need a lock. */
68 /* Data in the aio_bitmap is protected by block migration lock.
69 * Allocation and free happen during setup and cleanup respectively.
71 unsigned long *aio_bitmap;
73 /* Protected by block migration lock. */
74 int64_t completed_sectors;
76 /* During migration this is protected by iothread lock / AioContext.
77 * Allocation and free happen during setup and cleanup respectively.
79 BdrvDirtyBitmap *dirty_bitmap;
82 typedef struct BlkMigBlock {
83 /* Only used by migration thread. */
91 /* Protected by block migration lock. */
93 QSIMPLEQ_ENTRY(BlkMigBlock) entry;
96 typedef struct BlkMigState {
97 QSIMPLEQ_HEAD(, BlkMigDevState) bmds_list;
98 int64_t total_sector_sum;
101 /* Protected by lock. */
102 QSIMPLEQ_HEAD(, BlkMigBlock) blk_list;
106 /* Only used by migration thread. Does not need a lock. */
111 /* Lock must be taken _inside_ the iothread lock and any AioContexts. */
115 static BlkMigState block_mig_state;
117 static void blk_mig_lock(void)
119 qemu_mutex_lock(&block_mig_state.lock);
122 static void blk_mig_unlock(void)
124 qemu_mutex_unlock(&block_mig_state.lock);
127 /* Must run outside of the iothread lock during the bulk phase,
128 * or the VM will stall.
131 static void blk_send(QEMUFile *f, BlkMigBlock * blk)
134 uint64_t flags = BLK_MIG_FLAG_DEVICE_BLOCK;
136 if (block_mig_state.zero_blocks &&
137 buffer_is_zero(blk->buf, BLK_MIG_BLOCK_SIZE)) {
138 flags |= BLK_MIG_FLAG_ZERO_BLOCK;
141 /* sector number and flags */
142 qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS)
146 len = strlen(blk->bmds->blk_name);
147 qemu_put_byte(f, len);
148 qemu_put_buffer(f, (uint8_t *) blk->bmds->blk_name, len);
150 /* if a block is zero we need to flush here since the network
151 * bandwidth is now a lot higher than the storage device bandwidth.
152 * thus if we queue zero blocks we slow down the migration */
153 if (flags & BLK_MIG_FLAG_ZERO_BLOCK) {
158 qemu_put_buffer(f, blk->buf, BLK_MIG_BLOCK_SIZE);
161 int blk_mig_active(void)
163 return !QSIMPLEQ_EMPTY(&block_mig_state.bmds_list);
166 int blk_mig_bulk_active(void)
168 return blk_mig_active() && !block_mig_state.bulk_completed;
171 uint64_t blk_mig_bytes_transferred(void)
173 BlkMigDevState *bmds;
177 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
178 sum += bmds->completed_sectors;
181 return sum << BDRV_SECTOR_BITS;
184 uint64_t blk_mig_bytes_remaining(void)
186 return blk_mig_bytes_total() - blk_mig_bytes_transferred();
189 uint64_t blk_mig_bytes_total(void)
191 BlkMigDevState *bmds;
194 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
195 sum += bmds->total_sectors;
197 return sum << BDRV_SECTOR_BITS;
201 /* Called with migration lock held. */
203 static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector)
205 int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
207 if (sector < blk_nb_sectors(bmds->blk)) {
208 return !!(bmds->aio_bitmap[chunk / (sizeof(unsigned long) * 8)] &
209 (1UL << (chunk % (sizeof(unsigned long) * 8))));
215 /* Called with migration lock held. */
217 static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num,
218 int nb_sectors, int set)
221 unsigned long val, idx, bit;
223 start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
224 end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
226 for (; start <= end; start++) {
227 idx = start / (sizeof(unsigned long) * 8);
228 bit = start % (sizeof(unsigned long) * 8);
229 val = bmds->aio_bitmap[idx];
233 val &= ~(1UL << bit);
235 bmds->aio_bitmap[idx] = val;
239 static void alloc_aio_bitmap(BlkMigDevState *bmds)
241 BlockBackend *bb = bmds->blk;
244 bitmap_size = blk_nb_sectors(bb) + BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
245 bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
247 bmds->aio_bitmap = g_malloc0(bitmap_size);
250 /* Never hold migration lock when yielding to the main loop! */
252 static void blk_mig_read_cb(void *opaque, int ret)
254 BlkMigBlock *blk = opaque;
259 QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry);
260 bmds_set_aio_inflight(blk->bmds, blk->sector, blk->nr_sectors, 0);
262 block_mig_state.submitted--;
263 block_mig_state.read_done++;
264 assert(block_mig_state.submitted >= 0);
268 /* Called with no lock taken. */
270 static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds)
272 int64_t total_sectors = bmds->total_sectors;
273 int64_t cur_sector = bmds->cur_sector;
274 BlockBackend *bb = bmds->blk;
279 if (bmds->shared_base) {
280 qemu_mutex_lock_iothread();
281 aio_context_acquire(blk_get_aio_context(bb));
282 /* Skip unallocated sectors; intentionally treats failure or
283 * partial sector as an allocated sector */
284 while (cur_sector < total_sectors &&
285 !bdrv_is_allocated(blk_bs(bb), cur_sector * BDRV_SECTOR_SIZE,
286 MAX_IS_ALLOCATED_SEARCH, &count)) {
287 if (count < BDRV_SECTOR_SIZE) {
290 cur_sector += count >> BDRV_SECTOR_BITS;
292 aio_context_release(blk_get_aio_context(bb));
293 qemu_mutex_unlock_iothread();
296 if (cur_sector >= total_sectors) {
297 bmds->cur_sector = bmds->completed_sectors = total_sectors;
301 bmds->completed_sectors = cur_sector;
303 cur_sector &= ~((int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK - 1);
305 /* we are going to transfer a full block even if it is not allocated */
306 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
308 if (total_sectors - cur_sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
309 nr_sectors = total_sectors - cur_sector;
312 blk = g_new(BlkMigBlock, 1);
313 blk->buf = g_malloc(BLK_MIG_BLOCK_SIZE);
315 blk->sector = cur_sector;
316 blk->nr_sectors = nr_sectors;
318 qemu_iovec_init_buf(&blk->qiov, blk->buf, nr_sectors * BDRV_SECTOR_SIZE);
321 block_mig_state.submitted++;
324 /* We do not know if bs is under the main thread (and thus does
325 * not acquire the AioContext when doing AIO) or rather under
326 * dataplane. Thus acquire both the iothread mutex and the
329 * This is ugly and will disappear when we make bdrv_* thread-safe,
330 * without the need to acquire the AioContext.
332 qemu_mutex_lock_iothread();
333 aio_context_acquire(blk_get_aio_context(bmds->blk));
334 bdrv_reset_dirty_bitmap(bmds->dirty_bitmap, cur_sector * BDRV_SECTOR_SIZE,
335 nr_sectors * BDRV_SECTOR_SIZE);
336 blk->aiocb = blk_aio_preadv(bb, cur_sector * BDRV_SECTOR_SIZE, &blk->qiov,
337 0, blk_mig_read_cb, blk);
338 aio_context_release(blk_get_aio_context(bmds->blk));
339 qemu_mutex_unlock_iothread();
341 bmds->cur_sector = cur_sector + nr_sectors;
342 return (bmds->cur_sector >= total_sectors);
345 /* Called with iothread lock taken. */
347 static int set_dirty_tracking(void)
349 BlkMigDevState *bmds;
352 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
353 bmds->dirty_bitmap = bdrv_create_dirty_bitmap(blk_bs(bmds->blk),
356 if (!bmds->dirty_bitmap) {
364 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
365 if (bmds->dirty_bitmap) {
366 bdrv_release_dirty_bitmap(bmds->dirty_bitmap);
372 /* Called with iothread lock taken. */
374 static void unset_dirty_tracking(void)
376 BlkMigDevState *bmds;
378 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
379 bdrv_release_dirty_bitmap(bmds->dirty_bitmap);
383 static int init_blk_migration(QEMUFile *f)
385 BlockDriverState *bs;
386 BlkMigDevState *bmds;
391 BlkMigDevState *bmds;
392 BlockDriverState *bs;
394 Error *local_err = NULL;
397 block_mig_state.submitted = 0;
398 block_mig_state.read_done = 0;
399 block_mig_state.transferred = 0;
400 block_mig_state.total_sector_sum = 0;
401 block_mig_state.prev_progress = -1;
402 block_mig_state.bulk_completed = 0;
403 block_mig_state.zero_blocks = migrate_zero_blocks();
405 for (bs = bdrv_first(&it); bs; bs = bdrv_next(&it)) {
408 bmds_bs = g_malloc0(num_bs * sizeof(*bmds_bs));
410 for (i = 0, bs = bdrv_first(&it); bs; bs = bdrv_next(&it), i++) {
411 if (bdrv_is_read_only(bs)) {
415 sectors = bdrv_nb_sectors(bs);
418 bdrv_next_cleanup(&it);
422 bmds = g_new0(BlkMigDevState, 1);
423 bmds->blk = blk_new(qemu_get_aio_context(),
424 BLK_PERM_CONSISTENT_READ, BLK_PERM_ALL);
425 bmds->blk_name = g_strdup(bdrv_get_device_name(bs));
426 bmds->bulk_completed = 0;
427 bmds->total_sectors = sectors;
428 bmds->completed_sectors = 0;
429 bmds->shared_base = migrate_use_block_incremental();
432 bmds_bs[i].bmds = bmds;
435 block_mig_state.total_sector_sum += sectors;
437 if (bmds->shared_base) {
438 trace_migration_block_init_shared(bdrv_get_device_name(bs));
440 trace_migration_block_init_full(bdrv_get_device_name(bs));
443 QSIMPLEQ_INSERT_TAIL(&block_mig_state.bmds_list, bmds, entry);
446 /* Can only insert new BDSes now because doing so while iterating block
447 * devices may end up in a deadlock (iterating the new BDSes, too). */
448 for (i = 0; i < num_bs; i++) {
449 BlkMigDevState *bmds = bmds_bs[i].bmds;
450 BlockDriverState *bs = bmds_bs[i].bs;
453 ret = blk_insert_bs(bmds->blk, bs, &local_err);
455 error_report_err(local_err);
459 alloc_aio_bitmap(bmds);
460 error_setg(&bmds->blocker, "block device is in use by migration");
461 bdrv_op_block_all(bs, bmds->blocker);
471 /* Called with no lock taken. */
473 static int blk_mig_save_bulked_block(QEMUFile *f)
475 int64_t completed_sector_sum = 0;
476 BlkMigDevState *bmds;
480 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
481 if (bmds->bulk_completed == 0) {
482 if (mig_save_device_bulk(f, bmds) == 1) {
483 /* completed bulk section for this device */
484 bmds->bulk_completed = 1;
486 completed_sector_sum += bmds->completed_sectors;
490 completed_sector_sum += bmds->completed_sectors;
494 if (block_mig_state.total_sector_sum != 0) {
495 progress = completed_sector_sum * 100 /
496 block_mig_state.total_sector_sum;
500 if (progress != block_mig_state.prev_progress) {
501 block_mig_state.prev_progress = progress;
502 qemu_put_be64(f, (progress << BDRV_SECTOR_BITS)
503 | BLK_MIG_FLAG_PROGRESS);
504 DPRINTF("Completed %d %%\r", progress);
510 static void blk_mig_reset_dirty_cursor(void)
512 BlkMigDevState *bmds;
514 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
519 /* Called with iothread lock and AioContext taken. */
521 static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds,
525 int64_t total_sectors = bmds->total_sectors;
530 for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) {
532 if (bmds_aio_inflight(bmds, sector)) {
534 blk_drain(bmds->blk);
538 bdrv_dirty_bitmap_lock(bmds->dirty_bitmap);
539 if (bdrv_dirty_bitmap_get_locked(bmds->dirty_bitmap,
540 sector * BDRV_SECTOR_SIZE)) {
541 if (total_sectors - sector < BDRV_SECTORS_PER_DIRTY_CHUNK) {
542 nr_sectors = total_sectors - sector;
544 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
546 bdrv_reset_dirty_bitmap_locked(bmds->dirty_bitmap,
547 sector * BDRV_SECTOR_SIZE,
548 nr_sectors * BDRV_SECTOR_SIZE);
549 bdrv_dirty_bitmap_unlock(bmds->dirty_bitmap);
551 blk = g_new(BlkMigBlock, 1);
552 blk->buf = g_malloc(BLK_MIG_BLOCK_SIZE);
554 blk->sector = sector;
555 blk->nr_sectors = nr_sectors;
558 qemu_iovec_init_buf(&blk->qiov, blk->buf,
559 nr_sectors * BDRV_SECTOR_SIZE);
561 blk->aiocb = blk_aio_preadv(bmds->blk,
562 sector * BDRV_SECTOR_SIZE,
563 &blk->qiov, 0, blk_mig_read_cb,
567 block_mig_state.submitted++;
568 bmds_set_aio_inflight(bmds, sector, nr_sectors, 1);
571 ret = blk_pread(bmds->blk, sector * BDRV_SECTOR_SIZE,
572 nr_sectors * BDRV_SECTOR_SIZE, blk->buf, 0);
582 sector += nr_sectors;
583 bmds->cur_dirty = sector;
587 bdrv_dirty_bitmap_unlock(bmds->dirty_bitmap);
588 sector += BDRV_SECTORS_PER_DIRTY_CHUNK;
589 bmds->cur_dirty = sector;
592 return (bmds->cur_dirty >= bmds->total_sectors);
595 trace_migration_block_save_device_dirty(sector);
601 /* Called with iothread lock taken.
604 * 0: too much data for max_downtime
605 * 1: few enough data for max_downtime
607 static int blk_mig_save_dirty_block(QEMUFile *f, int is_async)
609 BlkMigDevState *bmds;
612 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
613 aio_context_acquire(blk_get_aio_context(bmds->blk));
614 ret = mig_save_device_dirty(f, bmds, is_async);
615 aio_context_release(blk_get_aio_context(bmds->blk));
624 /* Called with no locks taken. */
626 static int flush_blks(QEMUFile *f)
631 trace_migration_block_flush_blks("Enter", block_mig_state.submitted,
632 block_mig_state.read_done,
633 block_mig_state.transferred);
636 while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
637 if (qemu_file_rate_limit(f)) {
645 QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
653 block_mig_state.read_done--;
654 block_mig_state.transferred++;
655 assert(block_mig_state.read_done >= 0);
659 trace_migration_block_flush_blks("Exit", block_mig_state.submitted,
660 block_mig_state.read_done,
661 block_mig_state.transferred);
665 /* Called with iothread lock taken. */
667 static int64_t get_remaining_dirty(void)
669 BlkMigDevState *bmds;
672 QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
673 aio_context_acquire(blk_get_aio_context(bmds->blk));
674 dirty += bdrv_get_dirty_count(bmds->dirty_bitmap);
675 aio_context_release(blk_get_aio_context(bmds->blk));
683 /* Called with iothread lock taken. */
684 static void block_migration_cleanup_bmds(void)
686 BlkMigDevState *bmds;
689 unset_dirty_tracking();
691 while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) {
692 QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry);
693 bdrv_op_unblock_all(blk_bs(bmds->blk), bmds->blocker);
694 error_free(bmds->blocker);
696 /* Save ctx, because bmds->blk can disappear during blk_unref. */
697 ctx = blk_get_aio_context(bmds->blk);
698 aio_context_acquire(ctx);
699 blk_unref(bmds->blk);
700 aio_context_release(ctx);
702 g_free(bmds->blk_name);
703 g_free(bmds->aio_bitmap);
708 /* Called with iothread lock taken. */
709 static void block_migration_cleanup(void *opaque)
715 block_migration_cleanup_bmds();
718 while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
719 QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
726 static int block_save_setup(QEMUFile *f, void *opaque)
730 trace_migration_block_save("setup", block_mig_state.submitted,
731 block_mig_state.transferred);
733 qemu_mutex_lock_iothread();
734 ret = init_blk_migration(f);
736 qemu_mutex_unlock_iothread();
740 /* start track dirty blocks */
741 ret = set_dirty_tracking();
743 qemu_mutex_unlock_iothread();
750 blk_mig_reset_dirty_cursor();
751 qemu_put_be64(f, BLK_MIG_FLAG_EOS);
756 static int block_save_iterate(QEMUFile *f, void *opaque)
759 int64_t last_bytes = qemu_file_total_transferred(f);
762 trace_migration_block_save("iterate", block_mig_state.submitted,
763 block_mig_state.transferred);
770 blk_mig_reset_dirty_cursor();
772 /* control the rate of transfer */
774 while (block_mig_state.read_done * BLK_MIG_BLOCK_SIZE <
775 qemu_file_get_rate_limit(f) &&
776 block_mig_state.submitted < MAX_PARALLEL_IO &&
777 (block_mig_state.submitted + block_mig_state.read_done) <
780 if (block_mig_state.bulk_completed == 0) {
781 /* first finish the bulk phase */
782 if (blk_mig_save_bulked_block(f) == 0) {
783 /* finished saving bulk on all devices */
784 block_mig_state.bulk_completed = 1;
788 /* Always called with iothread lock taken for
789 * simplicity, block_save_complete also calls it.
791 qemu_mutex_lock_iothread();
792 ret = blk_mig_save_dirty_block(f, 1);
793 qemu_mutex_unlock_iothread();
800 /* no more dirty blocks */
811 qemu_put_be64(f, BLK_MIG_FLAG_EOS);
812 delta_bytes = qemu_file_total_transferred(f) - last_bytes;
813 if (delta_bytes > 0) {
815 } else if (delta_bytes < 0) {
822 /* Called with iothread lock taken. */
824 static int block_save_complete(QEMUFile *f, void *opaque)
828 trace_migration_block_save("complete", block_mig_state.submitted,
829 block_mig_state.transferred);
836 blk_mig_reset_dirty_cursor();
838 /* we know for sure that save bulk is completed and
839 all async read completed */
841 assert(block_mig_state.submitted == 0);
845 ret = blk_mig_save_dirty_block(f, 0);
851 /* report completion */
852 qemu_put_be64(f, (100 << BDRV_SECTOR_BITS) | BLK_MIG_FLAG_PROGRESS);
854 trace_migration_block_save_complete();
856 qemu_put_be64(f, BLK_MIG_FLAG_EOS);
858 /* Make sure that our BlockBackends are gone, so that the block driver
859 * nodes can be inactivated. */
860 block_migration_cleanup_bmds();
865 static void block_save_pending(QEMUFile *f, void *opaque, uint64_t max_size,
866 uint64_t *res_precopy_only,
867 uint64_t *res_compatible,
868 uint64_t *res_postcopy_only)
870 /* Estimate pending number of bytes to send */
873 qemu_mutex_lock_iothread();
874 pending = get_remaining_dirty();
875 qemu_mutex_unlock_iothread();
878 pending += block_mig_state.submitted * BLK_MIG_BLOCK_SIZE +
879 block_mig_state.read_done * BLK_MIG_BLOCK_SIZE;
882 /* Report at least one block pending during bulk phase */
883 if (pending <= max_size && !block_mig_state.bulk_completed) {
884 pending = max_size + BLK_MIG_BLOCK_SIZE;
887 trace_migration_block_save_pending(pending);
888 /* We don't do postcopy */
889 *res_precopy_only += pending;
892 static int block_load(QEMUFile *f, void *opaque, int version_id)
894 static int banner_printed;
896 char device_name[256];
898 BlockBackend *blk, *blk_prev = NULL;
899 Error *local_err = NULL;
901 int64_t total_sectors = 0;
905 int cluster_size = BLK_MIG_BLOCK_SIZE;
908 addr = qemu_get_be64(f);
910 flags = addr & (BDRV_SECTOR_SIZE - 1);
911 addr >>= BDRV_SECTOR_BITS;
913 if (flags & BLK_MIG_FLAG_DEVICE_BLOCK) {
914 /* get device name */
915 len = qemu_get_byte(f);
916 qemu_get_buffer(f, (uint8_t *)device_name, len);
917 device_name[len] = '\0';
919 blk = blk_by_name(device_name);
921 fprintf(stderr, "Error unknown block device %s\n",
926 if (blk != blk_prev) {
928 total_sectors = blk_nb_sectors(blk);
929 if (total_sectors <= 0) {
930 error_report("Error getting length of block device %s",
935 blk_activate(blk, &local_err);
937 error_report_err(local_err);
941 ret = bdrv_get_info(blk_bs(blk), &bdi);
942 if (ret == 0 && bdi.cluster_size > 0 &&
943 bdi.cluster_size <= BLK_MIG_BLOCK_SIZE &&
944 BLK_MIG_BLOCK_SIZE % bdi.cluster_size == 0) {
945 cluster_size = bdi.cluster_size;
947 cluster_size = BLK_MIG_BLOCK_SIZE;
951 if (total_sectors - addr < BDRV_SECTORS_PER_DIRTY_CHUNK) {
952 nr_sectors = total_sectors - addr;
954 nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
957 if (flags & BLK_MIG_FLAG_ZERO_BLOCK) {
958 ret = blk_pwrite_zeroes(blk, addr * BDRV_SECTOR_SIZE,
959 nr_sectors * BDRV_SECTOR_SIZE,
966 buf = g_malloc(BLK_MIG_BLOCK_SIZE);
967 qemu_get_buffer(f, buf, BLK_MIG_BLOCK_SIZE);
968 for (i = 0; i < BLK_MIG_BLOCK_SIZE / cluster_size; i++) {
969 cur_addr = addr * BDRV_SECTOR_SIZE + i * cluster_size;
970 cur_buf = buf + i * cluster_size;
972 if ((!block_mig_state.zero_blocks ||
973 cluster_size < BLK_MIG_BLOCK_SIZE) &&
974 buffer_is_zero(cur_buf, cluster_size)) {
975 ret = blk_pwrite_zeroes(blk, cur_addr,
979 ret = blk_pwrite(blk, cur_addr, cluster_size, cur_buf,
992 } else if (flags & BLK_MIG_FLAG_PROGRESS) {
993 if (!banner_printed) {
994 printf("Receiving block device images\n");
997 printf("Completed %d %%%c", (int)addr,
998 (addr == 100) ? '\n' : '\r');
1000 } else if (!(flags & BLK_MIG_FLAG_EOS)) {
1001 fprintf(stderr, "Unknown block migration flags: 0x%x\n", flags);
1004 ret = qemu_file_get_error(f);
1008 } while (!(flags & BLK_MIG_FLAG_EOS));
1013 static bool block_is_active(void *opaque)
1015 return migrate_use_block();
1018 static SaveVMHandlers savevm_block_handlers = {
1019 .save_setup = block_save_setup,
1020 .save_live_iterate = block_save_iterate,
1021 .save_live_complete_precopy = block_save_complete,
1022 .save_live_pending = block_save_pending,
1023 .load_state = block_load,
1024 .save_cleanup = block_migration_cleanup,
1025 .is_active = block_is_active,
1028 void blk_mig_init(void)
1030 QSIMPLEQ_INIT(&block_mig_state.bmds_list);
1031 QSIMPLEQ_INIT(&block_mig_state.blk_list);
1032 qemu_mutex_init(&block_mig_state.lock);
1034 register_savevm_live("block", 0, 1, &savevm_block_handlers,