#define BLK_MIG_FLAG_DEVICE_BLOCK 0x01
#define BLK_MIG_FLAG_EOS 0x02
#define BLK_MIG_FLAG_PROGRESS 0x04
+#define BLK_MIG_FLAG_ZERO_BLOCK 0x08
#define MAX_IS_ALLOCATED_SEARCH 65536
int64_t cur_sector;
int64_t cur_dirty;
- /* Protected by iothread lock. */
+ /* Protected by block migration lock. */
unsigned long *aio_bitmap;
int64_t completed_sectors;
} BlkMigDevState;
QEMUIOVector qiov;
BlockDriverAIOCB *aiocb;
- /* Protected by iothread lock. */
+ /* Protected by block migration lock. */
int ret;
QSIMPLEQ_ENTRY(BlkMigBlock) entry;
} BlkMigBlock;
int shared_base;
QSIMPLEQ_HEAD(bmds_list, BlkMigDevState) bmds_list;
int64_t total_sector_sum;
+ bool zero_blocks;
- /* Protected by iothread lock. */
+ /* Protected by lock. */
QSIMPLEQ_HEAD(blk_list, BlkMigBlock) blk_list;
int submitted;
int read_done;
int transferred;
int prev_progress;
int bulk_completed;
+
+ /* Lock must be taken _inside_ the iothread lock. */
+ QemuMutex lock;
} BlkMigState;
static BlkMigState block_mig_state;
+static void blk_mig_lock(void)
+{
+ qemu_mutex_lock(&block_mig_state.lock);
+}
+
+static void blk_mig_unlock(void)
+{
+ qemu_mutex_unlock(&block_mig_state.lock);
+}
+
+/* Must run outside of the iothread lock during the bulk phase,
+ * or the VM will stall.
+ */
+
static void blk_send(QEMUFile *f, BlkMigBlock * blk)
{
int len;
+ uint64_t flags = BLK_MIG_FLAG_DEVICE_BLOCK;
+
+ if (block_mig_state.zero_blocks &&
+ buffer_is_zero(blk->buf, BLOCK_SIZE)) {
+ flags |= BLK_MIG_FLAG_ZERO_BLOCK;
+ }
/* sector number and flags */
qemu_put_be64(f, (blk->sector << BDRV_SECTOR_BITS)
- | BLK_MIG_FLAG_DEVICE_BLOCK);
+ | flags);
/* device name */
len = strlen(blk->bmds->bs->device_name);
qemu_put_byte(f, len);
qemu_put_buffer(f, (uint8_t *)blk->bmds->bs->device_name, len);
+ /* if a block is zero we need to flush here since the network
+ * bandwidth is now a lot higher than the storage device bandwidth.
+ * thus if we queue zero blocks we slow down the migration */
+ if (flags & BLK_MIG_FLAG_ZERO_BLOCK) {
+ qemu_fflush(f);
+ return;
+ }
+
qemu_put_buffer(f, blk->buf, BLOCK_SIZE);
}
BlkMigDevState *bmds;
uint64_t sum = 0;
+ blk_mig_lock();
QSIMPLEQ_FOREACH(bmds, &block_mig_state.bmds_list, entry) {
sum += bmds->completed_sectors;
}
+ blk_mig_unlock();
return sum << BDRV_SECTOR_BITS;
}
return sum << BDRV_SECTOR_BITS;
}
+
+/* Called with migration lock held. */
+
static int bmds_aio_inflight(BlkMigDevState *bmds, int64_t sector)
{
int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
}
}
+/* Called with migration lock held. */
+
static void bmds_set_aio_inflight(BlkMigDevState *bmds, int64_t sector_num,
int nb_sectors, int set)
{
bmds->aio_bitmap = g_malloc0(bitmap_size);
}
+/* Never hold migration lock when yielding to the main loop! */
+
static void blk_mig_read_cb(void *opaque, int ret)
{
BlkMigBlock *blk = opaque;
+ blk_mig_lock();
blk->ret = ret;
QSIMPLEQ_INSERT_TAIL(&block_mig_state.blk_list, blk, entry);
block_mig_state.submitted--;
block_mig_state.read_done++;
assert(block_mig_state.submitted >= 0);
+ blk_mig_unlock();
}
+/* Called with no lock taken. */
+
static int mig_save_device_bulk(QEMUFile *f, BlkMigDevState *bmds)
{
int64_t total_sectors = bmds->total_sectors;
int nr_sectors;
if (bmds->shared_base) {
+ qemu_mutex_lock_iothread();
while (cur_sector < total_sectors &&
!bdrv_is_allocated(bs, cur_sector, MAX_IS_ALLOCATED_SEARCH,
&nr_sectors)) {
cur_sector += nr_sectors;
}
+ qemu_mutex_unlock_iothread();
}
if (cur_sector >= total_sectors) {
blk->iov.iov_len = nr_sectors * BDRV_SECTOR_SIZE;
qemu_iovec_init_external(&blk->qiov, &blk->iov, 1);
+ blk_mig_lock();
block_mig_state.submitted++;
+ blk_mig_unlock();
+ qemu_mutex_lock_iothread();
blk->aiocb = bdrv_aio_readv(bs, cur_sector, &blk->qiov,
nr_sectors, blk_mig_read_cb, blk);
bdrv_reset_dirty(bs, cur_sector, nr_sectors);
- bmds->cur_sector = cur_sector + nr_sectors;
+ qemu_mutex_unlock_iothread();
+ bmds->cur_sector = cur_sector + nr_sectors;
return (bmds->cur_sector >= total_sectors);
}
+/* Called with iothread lock taken. */
+
static void set_dirty_tracking(int enable)
{
BlkMigDevState *bmds;
bmds->completed_sectors = 0;
bmds->shared_base = block_mig_state.shared_base;
alloc_aio_bitmap(bmds);
- drive_get_ref(drive_get_by_blockdev(bs));
bdrv_set_in_use(bs, 1);
+ bdrv_ref(bs);
block_mig_state.total_sector_sum += sectors;
block_mig_state.total_sector_sum = 0;
block_mig_state.prev_progress = -1;
block_mig_state.bulk_completed = 0;
+ block_mig_state.zero_blocks = migrate_zero_blocks();
bdrv_iterate(init_blk_migration_it, NULL);
}
+/* Called with no lock taken. */
+
static int blk_mig_save_bulked_block(QEMUFile *f)
{
int64_t completed_sector_sum = 0;
}
}
+/* Called with iothread lock taken. */
+
static int mig_save_device_dirty(QEMUFile *f, BlkMigDevState *bmds,
int is_async)
{
int ret = -EIO;
for (sector = bmds->cur_dirty; sector < bmds->total_sectors;) {
+ blk_mig_lock();
if (bmds_aio_inflight(bmds, sector)) {
+ blk_mig_unlock();
bdrv_drain_all();
+ } else {
+ blk_mig_unlock();
}
if (bdrv_get_dirty(bmds->bs, sector)) {
blk->aiocb = bdrv_aio_readv(bmds->bs, sector, &blk->qiov,
nr_sectors, blk_mig_read_cb, blk);
+
+ blk_mig_lock();
block_mig_state.submitted++;
bmds_set_aio_inflight(bmds, sector, nr_sectors, 1);
+ blk_mig_unlock();
} else {
ret = bdrv_read(bmds->bs, sector, blk->buf, nr_sectors);
if (ret < 0) {
return ret;
}
-/* return value:
+/* Called with iothread lock taken.
+ *
+ * return value:
* 0: too much data for max_downtime
* 1: few enough data for max_downtime
*/
return ret;
}
+/* Called with no locks taken. */
+
static int flush_blks(QEMUFile *f)
{
BlkMigBlock *blk;
__FUNCTION__, block_mig_state.submitted, block_mig_state.read_done,
block_mig_state.transferred);
+ blk_mig_lock();
while ((blk = QSIMPLEQ_FIRST(&block_mig_state.blk_list)) != NULL) {
if (qemu_file_rate_limit(f)) {
break;
}
QSIMPLEQ_REMOVE_HEAD(&block_mig_state.blk_list, entry);
+ blk_mig_unlock();
blk_send(f, blk);
+ blk_mig_lock();
g_free(blk->buf);
g_free(blk);
block_mig_state.transferred++;
assert(block_mig_state.read_done >= 0);
}
+ blk_mig_unlock();
DPRINTF("%s Exit submitted %d read_done %d transferred %d\n", __FUNCTION__,
block_mig_state.submitted, block_mig_state.read_done,
return ret;
}
+/* Called with iothread lock taken. */
+
static int64_t get_remaining_dirty(void)
{
BlkMigDevState *bmds;
return dirty << BDRV_SECTOR_BITS;
}
+/* Called with iothread lock taken. */
+
static void blk_mig_cleanup(void)
{
BlkMigDevState *bmds;
set_dirty_tracking(0);
+ blk_mig_lock();
while ((bmds = QSIMPLEQ_FIRST(&block_mig_state.bmds_list)) != NULL) {
QSIMPLEQ_REMOVE_HEAD(&block_mig_state.bmds_list, entry);
bdrv_set_in_use(bmds->bs, 0);
- drive_put_ref(drive_get_by_blockdev(bmds->bs));
+ bdrv_unref(bmds->bs);
g_free(bmds->aio_bitmap);
g_free(bmds);
}
g_free(blk->buf);
g_free(blk);
}
+ blk_mig_unlock();
}
static void block_migration_cancel(void *opaque)
DPRINTF("Enter save live setup submitted %d transferred %d\n",
block_mig_state.submitted, block_mig_state.transferred);
+ qemu_mutex_lock_iothread();
init_blk_migration(f);
/* start track dirty blocks */
set_dirty_tracking(1);
+ qemu_mutex_unlock_iothread();
ret = flush_blks(f);
blk_mig_reset_dirty_cursor();
blk_mig_reset_dirty_cursor();
/* control the rate of transfer */
+ blk_mig_lock();
while ((block_mig_state.submitted +
block_mig_state.read_done) * BLOCK_SIZE <
qemu_file_get_rate_limit(f)) {
+ blk_mig_unlock();
if (block_mig_state.bulk_completed == 0) {
/* first finish the bulk phase */
if (blk_mig_save_bulked_block(f) == 0) {
}
ret = 0;
} else {
+ /* Always called with iothread lock taken for
+ * simplicity, block_save_complete also calls it.
+ */
+ qemu_mutex_lock_iothread();
ret = blk_mig_save_dirty_block(f, 1);
+ qemu_mutex_unlock_iothread();
}
if (ret < 0) {
return ret;
}
+ blk_mig_lock();
if (ret != 0) {
/* no more dirty blocks */
break;
}
}
+ blk_mig_unlock();
ret = flush_blks(f);
if (ret) {
return qemu_ftell(f) - last_ftell;
}
+/* Called with iothread lock taken. */
+
static int block_save_complete(QEMUFile *f, void *opaque)
{
int ret;
/* we know for sure that save bulk is completed and
all async read completed */
+ blk_mig_lock();
assert(block_mig_state.submitted == 0);
+ blk_mig_unlock();
do {
ret = blk_mig_save_dirty_block(f, 0);
/* Estimate pending number of bytes to send */
uint64_t pending;
+ qemu_mutex_lock_iothread();
+ blk_mig_lock();
pending = get_remaining_dirty() +
block_mig_state.submitted * BLOCK_SIZE +
block_mig_state.read_done * BLOCK_SIZE;
if (pending == 0 && !block_mig_state.bulk_completed) {
pending = BLOCK_SIZE;
}
+ blk_mig_unlock();
+ qemu_mutex_unlock_iothread();
DPRINTF("Enter save live pending %" PRIu64 "\n", pending);
return pending;
nr_sectors = BDRV_SECTORS_PER_DIRTY_CHUNK;
}
- buf = g_malloc(BLOCK_SIZE);
-
- qemu_get_buffer(f, buf, BLOCK_SIZE);
- ret = bdrv_write(bs, addr, buf, nr_sectors);
+ if (flags & BLK_MIG_FLAG_ZERO_BLOCK) {
+ ret = bdrv_write_zeroes(bs, addr, nr_sectors);
+ } else {
+ buf = g_malloc(BLOCK_SIZE);
+ qemu_get_buffer(f, buf, BLOCK_SIZE);
+ ret = bdrv_write(bs, addr, buf, nr_sectors);
+ g_free(buf);
+ }
- g_free(buf);
if (ret < 0) {
return ret;
}
{
QSIMPLEQ_INIT(&block_mig_state.bmds_list);
QSIMPLEQ_INIT(&block_mig_state.blk_list);
+ qemu_mutex_init(&block_mig_state.lock);
register_savevm_live(NULL, "block", 0, 1, &savevm_block_handlers,
&block_mig_state);
projects / qemu.git / blobdiff