/* Round offset and/or bytes to target cluster if COW is needed, and
* return the offset of the adjusted tail against original. */
static int mirror_cow_align(MirrorBlockJob *s, int64_t *offset,
uint64_t *bytes)
{
bool need_cow;
int ret = 0;
int64_t align_offset = *offset;
int64_t align_bytes = *bytes;
int max_bytes = s->granularity * s->max_iov;
need_cow = !test_bit(*offset / s->granularity, s->cow_bitmap);
need_cow |= !test_bit((*offset + *bytes - 1) / s->granularity,
s->cow_bitmap);
if (need_cow) {
bdrv_round_to_clusters(blk_bs(s->target), *offset, *bytes,
&align_offset, &align_bytes);
}
if (align_bytes > max_bytes) {
align_bytes = max_bytes;
if (need_cow) {
align_bytes = QEMU_ALIGN_DOWN(align_bytes, s->target_cluster_size);
}
}
/* Clipping may result in align_bytes unaligned to chunk boundary, but
* that doesn't matter because it's already the end of source image. */
align_bytes = mirror_clip_bytes(s, align_offset, align_bytes);
ret = align_offset + align_bytes - (*offset + *bytes);
*offset = align_offset;
*bytes = align_bytes;
assert(ret >= 0);
return ret;
}
static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
{
BlockDriverState *source = s->source;
int64_t offset, first_chunk;
uint64_t delay_ns = 0;
/* At least the first dirty chunk is mirrored in one iteration. */
int nb_chunks = 1;
bool write_zeroes_ok = bdrv_can_write_zeroes_with_unmap(blk_bs(s->target));
int max_io_bytes = MAX(s->buf_size / MAX_IN_FLIGHT, MAX_IO_BYTES);
bdrv_dirty_bitmap_lock(s->dirty_bitmap);
offset = bdrv_dirty_iter_next(s->dbi);
if (offset < 0) {
bdrv_set_dirty_iter(s->dbi, 0);
offset = bdrv_dirty_iter_next(s->dbi);
trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
assert(offset >= 0);
}
bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
first_chunk = offset / s->granularity;
while (test_bit(first_chunk, s->in_flight_bitmap)) {
trace_mirror_yield_in_flight(s, offset, s->in_flight);
mirror_wait_for_io(s);
}
block_job_pause_point(&s->common);
/* Find the number of consective dirty chunks following the first dirty
* one, and wait for in flight requests in them. */
bdrv_dirty_bitmap_lock(s->dirty_bitmap);
while (nb_chunks * s->granularity < s->buf_size) {
int64_t next_dirty;
int64_t next_offset = offset + nb_chunks * s->granularity;
int64_t next_chunk = next_offset / s->granularity;
if (next_offset >= s->bdev_length ||
!bdrv_get_dirty_locked(source, s->dirty_bitmap, next_offset)) {
break;
}
if (test_bit(next_chunk, s->in_flight_bitmap)) {
break;
}
next_dirty = bdrv_dirty_iter_next(s->dbi);
if (next_dirty > next_offset || next_dirty < 0) {
/* The bitmap iterator's cache is stale, refresh it */
bdrv_set_dirty_iter(s->dbi, next_offset);
next_dirty = bdrv_dirty_iter_next(s->dbi);
}
assert(next_dirty == next_offset);
nb_chunks++;
}
/* Clear dirty bits before querying the block status, because
* calling bdrv_block_status_above could yield - if some blocks are
* marked dirty in this window, we need to know.
*/
bdrv_reset_dirty_bitmap_locked(s->dirty_bitmap, offset,
nb_chunks * s->granularity);
bdrv_dirty_bitmap_unlock(s->dirty_bitmap);
bitmap_set(s->in_flight_bitmap, offset / s->granularity, nb_chunks);
while (nb_chunks > 0 && offset < s->bdev_length) {
int ret;
int64_t io_bytes;
int64_t io_bytes_acct;
enum MirrorMethod {
MIRROR_METHOD_COPY,
MIRROR_METHOD_ZERO,
MIRROR_METHOD_DISCARD
} mirror_method = MIRROR_METHOD_COPY;
assert(!(offset % s->granularity));
ret = bdrv_block_status_above(source, NULL, offset,
nb_chunks * s->granularity,
&io_bytes, NULL, NULL);
if (ret < 0) {
io_bytes = MIN(nb_chunks * s->granularity, max_io_bytes);
} else if (ret & BDRV_BLOCK_DATA) {
io_bytes = MIN(io_bytes, max_io_bytes);
}
io_bytes -= io_bytes % s->granularity;
if (io_bytes < s->granularity) {
io_bytes = s->granularity;
} else if (ret >= 0 && !(ret & BDRV_BLOCK_DATA)) {
int64_t target_offset;
int64_t target_bytes;
bdrv_round_to_clusters(blk_bs(s->target), offset, io_bytes,
&target_offset, &target_bytes);
if (target_offset == offset &&
target_bytes == io_bytes) {
mirror_method = ret & BDRV_BLOCK_ZERO ?
MIRROR_METHOD_ZERO :
MIRROR_METHOD_DISCARD;
}
}
while (s->in_flight >= MAX_IN_FLIGHT) {
trace_mirror_yield_in_flight(s, offset, s->in_flight);
mirror_wait_for_io(s);
}
if (s->ret < 0) {
return 0;
}
io_bytes = mirror_clip_bytes(s, offset, io_bytes);
switch (mirror_method) {
case MIRROR_METHOD_COPY:
io_bytes = io_bytes_acct = mirror_do_read(s, offset, io_bytes);
break;
case MIRROR_METHOD_ZERO:
case MIRROR_METHOD_DISCARD:
mirror_do_zero_or_discard(s, offset, io_bytes,
mirror_method == MIRROR_METHOD_DISCARD);
if (write_zeroes_ok) {
io_bytes_acct = 0;
} else {
io_bytes_acct = io_bytes;
}
break;
default:
abort();
}
assert(io_bytes);
offset += io_bytes;
nb_chunks -= DIV_ROUND_UP(io_bytes, s->granularity);
delay_ns = block_job_ratelimit_get_delay(&s->common, io_bytes_acct);
}
return delay_ns;
}
static uint64_t coroutine_fn mirror_iteration(MirrorBlockJob *s)
{
BlockDriverState *source = s->common.bs;
int nb_sectors, sectors_per_chunk, nb_chunks;
int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector;
uint64_t delay_ns = 0;
MirrorOp *op;
int pnum;
int64_t ret;
s->sector_num = hbitmap_iter_next(&s->hbi);
if (s->sector_num < 0) {
bdrv_dirty_iter_init(s->dirty_bitmap, &s->hbi);
s->sector_num = hbitmap_iter_next(&s->hbi);
trace_mirror_restart_iter(s, bdrv_get_dirty_count(s->dirty_bitmap));
assert(s->sector_num >= 0);
}
hbitmap_next_sector = s->sector_num;
sector_num = s->sector_num;
sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
end = s->bdev_length / BDRV_SECTOR_SIZE;
/* Extend the QEMUIOVector to include all adjacent blocks that will
* be copied in this operation.
*
* We have to do this if we have no backing file yet in the destination,
* and the cluster size is very large. Then we need to do COW ourselves.
* The first time a cluster is copied, copy it entirely. Note that,
* because both the granularity and the cluster size are powers of two,
* the number of sectors to copy cannot exceed one cluster.
*
* We also want to extend the QEMUIOVector to include more adjacent
* dirty blocks if possible, to limit the number of I/O operations and
* run efficiently even with a small granularity.
*/
nb_chunks = 0;
nb_sectors = 0;
next_sector = sector_num;
next_chunk = sector_num / sectors_per_chunk;
/* Wait for I/O to this cluster (from a previous iteration) to be done. */
while (test_bit(next_chunk, s->in_flight_bitmap)) {
trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
s->waiting_for_io = true;
qemu_coroutine_yield();
s->waiting_for_io = false;
}
do {
int added_sectors, added_chunks;
if (!bdrv_get_dirty(source, s->dirty_bitmap, next_sector) ||
test_bit(next_chunk, s->in_flight_bitmap)) {
assert(nb_sectors > 0);
break;
}
added_sectors = sectors_per_chunk;
if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) {
bdrv_round_to_clusters(s->target,
next_sector, added_sectors,
&next_sector, &added_sectors);
/* On the first iteration, the rounding may make us copy
* sectors before the first dirty one.
*/
if (next_sector < sector_num) {
assert(nb_sectors == 0);
sector_num = next_sector;
next_chunk = next_sector / sectors_per_chunk;
}
}
added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors));
added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk;
/* When doing COW, it may happen that there is not enough space for
* a full cluster. Wait if that is the case.
*/
while (nb_chunks == 0 && s->buf_free_count < added_chunks) {
trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight);
s->waiting_for_io = true;
qemu_coroutine_yield();
s->waiting_for_io = false;
}
if (s->buf_free_count < nb_chunks + added_chunks) {
trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight);
break;
}
if (IOV_MAX < nb_chunks + added_chunks) {
trace_mirror_break_iov_max(s, nb_chunks, added_chunks);
break;
}
/* We have enough free space to copy these sectors. */
bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks);
nb_sectors += added_sectors;
nb_chunks += added_chunks;
next_sector += added_sectors;
next_chunk += added_chunks;
if (!s->synced && s->common.speed) {
delay_ns = ratelimit_calculate_delay(&s->limit, added_sectors);
}
} while (delay_ns == 0 && next_sector < end);
/* Allocate a MirrorOp that is used as an AIO callback. */
op = g_new(MirrorOp, 1);
op->s = s;
op->sector_num = sector_num;
op->nb_sectors = nb_sectors;
/* Now make a QEMUIOVector taking enough granularity-sized chunks
* from s->buf_free.
*/
qemu_iovec_init(&op->qiov, nb_chunks);
next_sector = sector_num;
while (nb_chunks-- > 0) {
MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
size_t remaining = (nb_sectors * BDRV_SECTOR_SIZE) - op->qiov.size;
QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
s->buf_free_count--;
qemu_iovec_add(&op->qiov, buf, MIN(s->granularity, remaining));
/* Advance the HBitmapIter in parallel, so that we do not examine
* the same sector twice.
*/
if (next_sector > hbitmap_next_sector
&& bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) {
hbitmap_next_sector = hbitmap_iter_next(&s->hbi);
}
next_sector += sectors_per_chunk;
}
bdrv_reset_dirty_bitmap(s->dirty_bitmap, sector_num, nb_sectors);
/* Copy the dirty cluster. */
s->in_flight++;
s->sectors_in_flight += nb_sectors;
trace_mirror_one_iteration(s, sector_num, nb_sectors);
ret = bdrv_get_block_status_above(source, NULL, sector_num,
nb_sectors, &pnum);
if (ret < 0 || pnum < nb_sectors ||
(ret & BDRV_BLOCK_DATA && !(ret & BDRV_BLOCK_ZERO))) {
bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors,
mirror_read_complete, op);
} else if (ret & BDRV_BLOCK_ZERO) {
bdrv_aio_write_zeroes(s->target, sector_num, op->nb_sectors,
s->unmap ? BDRV_REQ_MAY_UNMAP : 0,
mirror_write_complete, op);
} else {
assert(!(ret & BDRV_BLOCK_DATA));
bdrv_aio_discard(s->target, sector_num, op->nb_sectors,
mirror_write_complete, op);
}
return delay_ns;
}
static void coroutine_fn mirror_iteration(MirrorBlockJob *s)
{
BlockDriverState *source = s->common.bs;
int nb_sectors, sectors_per_chunk, nb_chunks;
int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector;
MirrorOp *op;
s->sector_num = hbitmap_iter_next(&s->hbi);
if (s->sector_num < 0) {
bdrv_dirty_iter_init(source, &s->hbi);
s->sector_num = hbitmap_iter_next(&s->hbi);
trace_mirror_restart_iter(s, bdrv_get_dirty_count(source));
assert(s->sector_num >= 0);
}
hbitmap_next_sector = s->sector_num;
sector_num = s->sector_num;
sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
end = s->common.len >> BDRV_SECTOR_BITS;
/* Extend the QEMUIOVector to include all adjacent blocks that will
* be copied in this operation.
*
* We have to do this if we have no backing file yet in the destination,
* and the cluster size is very large. Then we need to do COW ourselves.
* The first time a cluster is copied, copy it entirely. Note that,
* because both the granularity and the cluster size are powers of two,
* the number of sectors to copy cannot exceed one cluster.
*
* We also want to extend the QEMUIOVector to include more adjacent
* dirty blocks if possible, to limit the number of I/O operations and
* run efficiently even with a small granularity.
*/
nb_chunks = 0;
nb_sectors = 0;
next_sector = sector_num;
next_chunk = sector_num / sectors_per_chunk;
/* Wait for I/O to this cluster (from a previous iteration) to be done. */
while (test_bit(next_chunk, s->in_flight_bitmap)) {
trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
qemu_coroutine_yield();
}
do {
int added_sectors, added_chunks;
if (!bdrv_get_dirty(source, next_sector) ||
test_bit(next_chunk, s->in_flight_bitmap)) {
assert(nb_sectors > 0);
break;
}
added_sectors = sectors_per_chunk;
if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) {
bdrv_round_to_clusters(s->target,
next_sector, added_sectors,
&next_sector, &added_sectors);
/* On the first iteration, the rounding may make us copy
* sectors before the first dirty one.
*/
if (next_sector < sector_num) {
assert(nb_sectors == 0);
sector_num = next_sector;
next_chunk = next_sector / sectors_per_chunk;
}
}
added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors));
added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk;
/* When doing COW, it may happen that there is not enough space for
* a full cluster. Wait if that is the case.
*/
while (nb_chunks == 0 && s->buf_free_count < added_chunks) {
trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight);
qemu_coroutine_yield();
}
if (s->buf_free_count < nb_chunks + added_chunks) {
trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight);
break;
}
/* We have enough free space to copy these sectors. */
bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks);
nb_sectors += added_sectors;
nb_chunks += added_chunks;
next_sector += added_sectors;
next_chunk += added_chunks;
} while (next_sector < end);
/* Allocate a MirrorOp that is used as an AIO callback. */
op = g_slice_new(MirrorOp);
op->s = s;
op->sector_num = sector_num;
op->nb_sectors = nb_sectors;
/* Now make a QEMUIOVector taking enough granularity-sized chunks
* from s->buf_free.
*/
qemu_iovec_init(&op->qiov, nb_chunks);
next_sector = sector_num;
while (nb_chunks-- > 0) {
MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
s->buf_free_count--;
qemu_iovec_add(&op->qiov, buf, s->granularity);
/* Advance the HBitmapIter in parallel, so that we do not examine
* the same sector twice.
*/
if (next_sector > hbitmap_next_sector && bdrv_get_dirty(source, next_sector)) {
hbitmap_next_sector = hbitmap_iter_next(&s->hbi);
}
next_sector += sectors_per_chunk;
}
bdrv_reset_dirty(source, sector_num, nb_sectors);
/* Copy the dirty cluster. */
s->in_flight++;
trace_mirror_one_iteration(s, sector_num, nb_sectors);
bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors,
mirror_read_complete, op);
}