@@ -32,6 +32,46 @@
#include "qemu/bswap.h"
#include "trace.h"
+int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t exact_size)
+{
+ BDRVQcow2State *s = bs->opaque;
+ int new_l1_size, i, ret;
+
+ if (exact_size >= s->l1_size) {
+ return 0;
+ }
+
+ new_l1_size = exact_size;
+
+#ifdef DEBUG_ALLOC2
+ fprintf(stderr, "shrink l1_table from %d to %d\n", s->l1_size, new_l1_size);
+#endif
+
+ BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_WRITE_TABLE);
+ ret = bdrv_pwrite_zeroes(bs->file, s->l1_table_offset +
+ sizeof(uint64_t) * new_l1_size,
+ (s->l1_size - new_l1_size) * sizeof(uint64_t), 0);
+ if (ret < 0) {
+ return ret;
+ }
+
+ ret = bdrv_flush(bs->file->bs);
+ if (ret < 0) {
+ return ret;
+ }
+
+ BLKDBG_EVENT(bs->file, BLKDBG_L1_SHRINK_FREE_L2_CLUSTERS);
+ for (i = s->l1_size - 1; i > new_l1_size - 1; i--) {
+ if ((s->l1_table[i] & L1E_OFFSET_MASK) == 0) {
+ continue;
+ }
+ qcow2_free_clusters(bs, s->l1_table[i] & L1E_OFFSET_MASK,
+ s->cluster_size, QCOW2_DISCARD_ALWAYS);
+ s->l1_table[i] = 0;
+ }
+ return 0;
+}
+
int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
bool exact_size)
{
@@ -29,6 +29,7 @@
#include "block/qcow2.h"
#include "qemu/range.h"
#include "qemu/bswap.h"
+#include "qemu/cutils.h"
static int64_t alloc_clusters_noref(BlockDriverState *bs, uint64_t size);
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
@@ -2945,3 +2946,105 @@ done:
qemu_vfree(new_refblock);
return ret;
}
+
+int qcow2_shrink_reftable(BlockDriverState *bs)
+{
+ BDRVQcow2State *s = bs->opaque;
+ uint64_t *reftable_tmp =
+ g_try_malloc(sizeof(uint64_t) * s->refcount_table_size);
+ int i, ret;
+
+ if (s->refcount_table_size && reftable_tmp == NULL) {
+ return -ENOMEM;
+ }
+
+ for (i = 0; i < s->refcount_table_size; i++) {
+ int64_t refblock_offs = s->refcount_table[i] & REFT_OFFSET_MASK;
+ void *refblock;
+ bool unused_block;
+
+ if (refblock_offs == 0) {
+ reftable_tmp[i] = 0;
+ continue;
+ }
+ ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
+ &refblock);
+ if (ret < 0) {
+ goto out;
+ }
+
+ /* the refblock has own reference */
+ if (i == offset_to_reftable_index(s, refblock_offs)) {
+ uint64_t block_index = (refblock_offs >> s->cluster_bits) &
+ (s->refcount_block_size - 1);
+ uint64_t refcount = s->get_refcount(refblock, block_index);
+
+ s->set_refcount(refblock, block_index, 0);
+
+ unused_block = buffer_is_zero(refblock, s->cluster_size);
+
+ s->set_refcount(refblock, block_index, refcount);
+ } else {
+ unused_block = buffer_is_zero(refblock, s->cluster_size);
+ }
+ qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
+
+ reftable_tmp[i] = unused_block ? 0 : cpu_to_be64(s->refcount_table[i]);
+ }
+
+ ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset, reftable_tmp,
+ sizeof(uint64_t) * s->refcount_table_size);
+ if (ret < 0) {
+ goto out;
+ }
+
+ for (i = 0; i < s->refcount_table_size; i++) {
+ if (s->refcount_table[i] && !reftable_tmp[i]) {
+ uint64_t discard_offs = s->refcount_table[i] & REFT_OFFSET_MASK;
+ uint64_t refblock_offs = get_refblock_offset(s, discard_offs);
+ uint64_t cluster_index = discard_offs >> s->cluster_bits;
+ uint32_t block_index = cluster_index & (s->refcount_block_size - 1);
+ void *refblock;
+
+ assert(discard_offs != 0);
+
+ ret = qcow2_cache_get(bs, s->refcount_block_cache, refblock_offs,
+ &refblock);
+ if (ret < 0) {
+ goto out;
+ }
+
+ if (s->get_refcount(refblock, block_index) != 1) {
+ qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
+ ret = -EINVAL;
+ goto out;
+ }
+ s->set_refcount(refblock, block_index, 0);
+
+ qcow2_cache_entry_mark_dirty(bs, s->refcount_block_cache, refblock);
+
+ qcow2_cache_put(bs, s->refcount_block_cache, &refblock);
+
+ if (cluster_index < s->free_cluster_index) {
+ s->free_cluster_index = cluster_index;
+ }
+
+ refblock = qcow2_cache_is_table_offset(bs, s->refcount_block_cache,
+ discard_offs);
+ if (refblock) {
+ /* discard refblock from the cache if refblock is cached */
+ qcow2_cache_discard(bs, s->refcount_block_cache, refblock);
+ }
+ update_refcount_discard(bs, discard_offs, s->cluster_size);
+ s->refcount_table[i] = 0;
+ }
+ }
+
+ if (!s->cache_discards) {
+ qcow2_process_discards(bs, ret);
+ }
+
+out:
+ g_free(reftable_tmp);
+ return ret;
+}
@@ -2589,6 +2589,7 @@ static int qcow2_truncate(BlockDriverState *bs, int64_t offset, Error **errp)
{
BDRVQcow2State *s = bs->opaque;
int64_t new_l1_size;
+ uint64_t total_size;
int ret;
if (offset & 511) {
@@ -2602,17 +2603,38 @@ static int qcow2_truncate(BlockDriverState *bs, int64_t offset, Error **errp)
return -ENOTSUP;
}
- /* shrinking is currently not supported */
- if (offset < bs->total_sectors * 512) {
- error_setg(errp, "qcow2 doesn't support shrinking images yet");
- return -ENOTSUP;
- }
-
new_l1_size = size_to_l1(s, offset);
- ret = qcow2_grow_l1_table(bs, new_l1_size, true);
- if (ret < 0) {
- error_setg_errno(errp, -ret, "Failed to grow the L1 table");
- return ret;
+ total_size = bs->total_sectors << BDRV_SECTOR_BITS;
+
+ if (offset < total_size) {
+ ret = qcow2_cluster_discard(bs, ROUND_UP(offset, s->cluster_size),
+ total_size - ROUND_UP(offset,
+ s->cluster_size),
+ QCOW2_DISCARD_ALWAYS, true);
+ if (ret < 0) {
+ error_setg_errno(errp, -ret, "Failed to discard cropped clusters");
+ return ret;
+ }
+
+ ret = qcow2_shrink_l1_table(bs, new_l1_size);
+ if (ret < 0) {
+ error_setg_errno(errp, -ret,
+ "Failed to reduce the number of L2 tables");
+ return ret;
+ }
+
+ ret = qcow2_shrink_reftable(bs);
+ if (ret < 0) {
+ error_setg_errno(errp, -ret,
+ "Failed to discard unused refblocks");
+ return ret;
+ }
+ } else {
+ ret = qcow2_grow_l1_table(bs, new_l1_size, true);
+ if (ret < 0) {
+ error_setg_errno(errp, -ret, "Failed to grow the L1 table");
+ return ret;
+ }
}
/* write updated header.size */
@@ -488,6 +488,18 @@ static inline uint64_t refcount_diff(uint64_t r1, uint64_t r2)
return r1 > r2 ? r1 - r2 : r2 - r1;
}
+static inline
+uint32_t offset_to_reftable_index(BDRVQcow2State *s, uint64_t offset)
+{
+ return offset >> (s->refcount_block_bits + s->cluster_bits);
+}
+
+static inline uint64_t get_refblock_offset(BDRVQcow2State *s, uint64_t offset)
+{
+ uint32_t index = offset_to_reftable_index(s, offset);
+ return s->refcount_table[index] & REFT_OFFSET_MASK;
+}
+
/* qcow2.c functions */
int qcow2_backing_read1(BlockDriverState *bs, QEMUIOVector *qiov,
int64_t sector_num, int nb_sectors);
@@ -538,10 +550,12 @@ int qcow2_pre_write_overlap_check(BlockDriverState *bs, int ign, int64_t offset,
int qcow2_change_refcount_order(BlockDriverState *bs, int refcount_order,
BlockDriverAmendStatusCB *status_cb,
void *cb_opaque, Error **errp);
+int qcow2_shrink_reftable(BlockDriverState *bs);
/* qcow2-cluster.c functions */
int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
bool exact_size);
+int qcow2_shrink_l1_table(BlockDriverState *bs, uint64_t max_size);
int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index);
int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
int qcow2_encrypt_sectors(BDRVQcow2State *s, int64_t sector_num,
@@ -2372,7 +2372,8 @@
'cluster_alloc_bytes', 'cluster_free', 'flush_to_os',
'flush_to_disk', 'pwritev_rmw_head', 'pwritev_rmw_after_head',
'pwritev_rmw_tail', 'pwritev_rmw_after_tail', 'pwritev',
- 'pwritev_zero', 'pwritev_done', 'empty_image_prepare' ] }
+ 'pwritev_zero', 'pwritev_done', 'empty_image_prepare',
+ 'l1_shrink_write_table', 'l1_shrink_free_l2_clusters' ] }
##
# @BlkdebugInjectErrorOptions:
This patch add shrinking of the image file for qcow2. As a result, this allows us to reduce the virtual image size and free up space on the disk without copying the image. Image can be fragmented and shrink is done by punching holes in the image file. Signed-off-by: Pavel Butsykin <pbutsykin@virtuozzo.com> --- block/qcow2-cluster.c | 40 +++++++++++++++++++ block/qcow2-refcount.c | 103 +++++++++++++++++++++++++++++++++++++++++++++++++ block/qcow2.c | 42 +++++++++++++++----- block/qcow2.h | 14 +++++++ qapi/block-core.json | 3 +- 5 files changed, 191 insertions(+), 11 deletions(-)