The qcow2_check_refcounts function has been extended to be able to fix
OFLAG_COPIED errors and multiple references on refcount blocks.
If no corruptions remain after an image repair (and no errors have been
encountered), clear the corrupt flag in qcow2_check.
Signed-off-by: Max Reitz <mre...@redhat.com>
---
block/qcow2-cluster.c | 4 +-
block/qcow2-refcount.c | 249 ++++++++++++++++++++++++++++++++++++++++++-------
block/qcow2.c | 6 +-
block/qcow2.h | 1 +
include/block/block.h | 1 +
5 files changed, 222 insertions(+), 39 deletions(-)
diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
index 7c248aa..2d5aa92 100644
--- a/block/qcow2-cluster.c
+++ b/block/qcow2-cluster.c
@@ -145,7 +145,7 @@ static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
* and we really don't want bdrv_pread to perform a read-modify-write)
*/
#define L1_ENTRIES_PER_SECTOR (512 / 8)
-static int write_l1_entry(BlockDriverState *bs, int l1_index)
+int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index)
{
BDRVQcowState *s = bs->opaque;
uint64_t buf[L1_ENTRIES_PER_SECTOR];
@@ -254,7 +254,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index,
uint64_t **table)
/* update the L1 entry */
trace_qcow2_l2_allocate_write_l1(bs, l1_index);
s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
- ret = write_l1_entry(bs, l1_index);
+ ret = qcow2_write_l1_entry(bs, l1_index);
if (ret < 0) {
goto fail;
}
diff --git a/block/qcow2-refcount.c b/block/qcow2-refcount.c
index d06a9df..333d10b 100644
--- a/block/qcow2-refcount.c
+++ b/block/qcow2-refcount.c
@@ -1034,7 +1034,7 @@ static int check_refcounts_l2(BlockDriverState *bs,
BdrvCheckResult *res,
BDRVQcowState *s = bs->opaque;
uint64_t *l2_table, l2_entry;
uint64_t next_contiguous_offset = 0;
- int i, l2_size, nb_csectors, refcount;
+ int i, l2_size, nb_csectors;
/* Read L2 table from disk */
l2_size = s->l2_size * sizeof(uint64_t);
@@ -1086,23 +1086,8 @@ static int check_refcounts_l2(BlockDriverState *bs,
BdrvCheckResult *res,
case QCOW2_CLUSTER_NORMAL:
{
- /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
uint64_t offset = l2_entry & L2E_OFFSET_MASK;
- if (flags & CHECK_OFLAG_COPIED) {
- refcount = get_refcount(bs, offset >> s->cluster_bits);
- if (refcount < 0) {
- fprintf(stderr, "Can't get refcount for offset %"
- PRIx64 ": %s\n", l2_entry, strerror(-refcount));
- goto fail;
- }
- if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
- fprintf(stderr, "ERROR OFLAG_COPIED: offset=%"
- PRIx64 " refcount=%d\n", l2_entry, refcount);
- res->corruptions++;
- }
- }
-
if (flags & CHECK_FRAG_INFO) {
res->bfi.allocated_clusters++;
if (next_contiguous_offset &&
@@ -1159,7 +1144,7 @@ static int check_refcounts_l1(BlockDriverState *bs,
{
BDRVQcowState *s = bs->opaque;
uint64_t *l1_table, l2_offset, l1_size2;
- int i, refcount, ret;
+ int i, ret;
l1_size2 = l1_size * sizeof(uint64_t);
@@ -1183,22 +1168,6 @@ static int check_refcounts_l1(BlockDriverState *bs,
for(i = 0; i < l1_size; i++) {
l2_offset = l1_table[i];
if (l2_offset) {
- /* QCOW_OFLAG_COPIED must be set iff refcount == 1 */
- if (flags & CHECK_OFLAG_COPIED) {
- refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED)
- >> s->cluster_bits);
- if (refcount < 0) {
- fprintf(stderr, "Can't get refcount for l2_offset %"
- PRIx64 ": %s\n", l2_offset, strerror(-refcount));
- goto fail;
- }
- if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0))
{
- fprintf(stderr, "ERROR OFLAG_COPIED: l2_offset=%" PRIx64
- " refcount=%d\n", l2_offset, refcount);
- res->corruptions++;
- }
- }
-
/* Mark L2 table as used */
l2_offset &= L1E_OFFSET_MASK;
inc_refcounts(bs, res, refcount_table, refcount_table_size,
@@ -1230,6 +1199,40 @@ fail:
}
/*
+ * Writes one sector of the refcount table to the disk
+ */
+#define RT_ENTRIES_PER_SECTOR (512 / sizeof(uint64_t))
+static int write_reftable_entry(BlockDriverState *bs, int rt_index)
+{
+ BDRVQcowState *s = bs->opaque;
+ uint64_t buf[RT_ENTRIES_PER_SECTOR];
+ int rt_start_index;
+ int i, ret;
+
+ rt_start_index = rt_index & ~(RT_ENTRIES_PER_SECTOR - 1);
+ for (i = 0; i < RT_ENTRIES_PER_SECTOR; i++) {
+ buf[i] = cpu_to_be64(s->refcount_table[rt_start_index + i]);
+ }
+
+ ret = qcow2_pre_write_overlap_check(bs,
+ QCOW2_OL_DEFAULT & ~QCOW2_OL_REFCOUNT_TABLE,
+ s->refcount_table_offset + rt_start_index * sizeof(uint64_t),
+ sizeof(buf));
+ if (ret < 0) {
+ return ret;
+ }
+
+ BLKDBG_EVENT(bs->file, BLKDBG_REFTABLE_UPDATE);
+ ret = bdrv_pwrite_sync(bs->file, s->refcount_table_offset +
+ rt_start_index * sizeof(uint64_t), buf, sizeof(buf));
+ if (ret < 0) {
+ return ret;
+ }
+
+ return 0;
+}
+
+/*
* Checks an image for refcount consistency.
*
* Returns 0 if no errors are found, the number of errors in case the image is
@@ -1240,7 +1243,8 @@ int qcow2_check_refcounts(BlockDriverState *bs,
BdrvCheckResult *res,
{
BDRVQcowState *s = bs->opaque;
int64_t size, i, highest_cluster;
- int nb_clusters, refcount1, refcount2;
+ uint64_t *l2_table = NULL;
+ int nb_clusters, refcount1, refcount2, j;
QCowSnapshot *sn;
uint16_t *refcount_table;
int ret;
@@ -1305,10 +1309,85 @@ int qcow2_check_refcounts(BlockDriverState *bs,
BdrvCheckResult *res,
inc_refcounts(bs, res, refcount_table, nb_clusters,
offset, s->cluster_size);
if (refcount_table[cluster] != 1) {
- fprintf(stderr, "ERROR refcount block %" PRId64
+ fprintf(stderr, "%s refcount block %" PRId64
" refcount=%d\n",
+ fix & BDRV_FIX_ERRORS ? "Repairing" :
+ "ERROR",
i, refcount_table[cluster]);
- res->corruptions++;
+ if (fix & BDRV_FIX_ERRORS) {
+ int64_t new_offset;
+ void *refcount_block;
+
+ /* allocate new refcount block */
+ new_offset = qcow2_alloc_clusters(bs, s->cluster_size);
+ if (new_offset < 0) {
+ fprintf(stderr, "Could not allocate new cluster\n");
+ res->corruptions++;
+ continue;
+ }
+ /* fetch current content */
+ ret = qcow2_cache_get(bs, s->refcount_block_cache, offset,
+ &refcount_block);
+ if (ret < 0) {
+ fprintf(stderr, "Could not fetch refcount block\n");
+ qcow2_free_clusters(bs, new_offset, s->cluster_size,
+ QCOW2_DISCARD_ALWAYS);
+ res->corruptions++;
+ continue;
+ }
+ /* new block has not yet been entered into refcount table,
+ * therefore it is no refcount block yet (regarding this
+ * check) */
+ ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_DEFAULT,
+ new_offset, s->cluster_sectors * BDRV_SECTOR_SIZE);
+ if (ret < 0) {
+ fprintf(stderr, "Could not write refcount block (would
"
+ "overlap with existing metadata)\n");
+ /* the image will be marked corrupt here, so don't even
+ * attempt on freeing the cluster */
+ res->corruptions++;
+ goto fail;
+ }
+ /* write to new block */
+ ret = bdrv_write(bs->file, new_offset >> BDRV_SECTOR_BITS,
+ refcount_block, s->cluster_sectors);
+ if (ret < 0) {
+ fprintf(stderr, "Could not write refcount block\n");
+ qcow2_free_clusters(bs, new_offset, s->cluster_size,
+ QCOW2_DISCARD_ALWAYS);
+ res->corruptions++;
+ continue;
+ }
+ /* update refcount table */
+ assert(!(new_offset & (s->cluster_size - 1)));
+ s->refcount_table[i] = new_offset;
+ ret = write_reftable_entry(bs, i);
+ if (ret < 0) {
+ fprintf(stderr, "Could not update refcount table\n");
+ s->refcount_table[i] = offset;
+ qcow2_free_clusters(bs, new_offset, s->cluster_size,
+ QCOW2_DISCARD_ALWAYS);
+ res->corruptions++;
+ continue;
+ }
+ qcow2_cache_put(bs, s->refcount_block_cache,
+ &refcount_block);
+ /* update refcounts */
+ if ((new_offset >> s->cluster_bits) >= nb_clusters) {
+ /* increase refcount_table size if necessary */
+ int old_nb_clusters = nb_clusters;
+ nb_clusters = (new_offset >> s->cluster_bits) + 1;
+ refcount_table = g_realloc(refcount_table,
+ nb_clusters * sizeof(uint16_t));
+ memset(&refcount_table[old_nb_clusters], 0,
(nb_clusters
+ - old_nb_clusters) * sizeof(uint16_t));
+ }
+ refcount_table[cluster]--;
+ inc_refcounts(bs, res, refcount_table, nb_clusters,
+ new_offset, s->cluster_size);
+ } else {
+ res->corruptions++;
+ }
}
}
}
@@ -1364,10 +1443,108 @@ int qcow2_check_refcounts(BlockDriverState *bs,
BdrvCheckResult *res,
}
}
+ l2_table = g_malloc(s->l2_size * sizeof(uint64_t));
+
+ /* check OFLAG_COPIED */
+ for (i = 0; i < s->l1_size; i++) {
+ uint64_t l1_entry = s->l1_table[i];
+ uint64_t l2_offset = l1_entry & L1E_OFFSET_MASK;
+ bool l2_dirty = false;
+ int refcount;
+
+ if (!l2_offset) {
+ continue;
+ }
+
+ refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
+ if (refcount < 0) {
+ /* don't print message nor increment check_errors, since the above
+ * loop will have done this already */
+ continue;
+ }
+ if ((refcount == 1) != ((l1_entry & QCOW_OFLAG_COPIED) != 0)) {
+ fprintf(stderr, "%s OFLAG_COPIED L2 cluster: l1_entry=%" PRIx64
+ " refcount=%d\n",
+ fix & BDRV_FIX_ERRORS ? "Repairing" :
+ "ERROR",
+ l1_entry, refcount);
+ if (fix & BDRV_FIX_ERRORS) {
+ s->l1_table[i] = refcount == 1
+ ? l1_entry | QCOW_OFLAG_COPIED
+ : l1_entry & ~QCOW_OFLAG_COPIED;
+ ret = qcow2_write_l1_entry(bs, i);
+ if (ret < 0) {
+ res->check_errors++;
+ goto fail;
+ }
+ } else {
+ res->corruptions++;
+ }
+ }
+
+ ret = bdrv_pread(bs->file, l2_offset, l2_table,
+ s->l2_size * sizeof(uint64_t));
+ if (ret != s->l2_size * sizeof(uint64_t)) {
+ fprintf(stderr, "ERROR: Could not read L2 table\n");
+ res->check_errors++;
+ if (ret >= 0) {
+ ret = -EIO;
+ }
+ goto fail;
+ }
+
+ for (j = 0; j < s->l2_size; j++) {
+ uint64_t l2_entry = be64_to_cpu(l2_table[j]);
+ uint64_t data_offset;
+
+ if (qcow2_get_cluster_type(l2_entry) != QCOW2_CLUSTER_NORMAL) {
+ continue;
+ }
+
+ data_offset = l2_entry & L2E_OFFSET_MASK;
+
+ refcount = get_refcount(bs, data_offset >> s->cluster_bits);
+ if (refcount < 0) {
+ /* don't print message nor increment check_errors */
+ continue;
+ }
+ if ((refcount == 1) != ((l2_entry & QCOW_OFLAG_COPIED) != 0)) {
+ fprintf(stderr, "%s OFLAG_COPIED data cluster: l2_entry=%"
+ PRIx64 " refcount=%d\n",
+ fix & BDRV_FIX_ERRORS ? "Repairing" :
+ "ERROR",
+ l2_entry, refcount);
+ if (fix & BDRV_FIX_ERRORS) {
+ l2_table[j] = cpu_to_be64(refcount == 1
+ ? l2_entry | QCOW_OFLAG_COPIED
+ : l2_entry & ~QCOW_OFLAG_COPIED);
+ l2_dirty = true;
+ } else {
+ res->corruptions++;
+ }
+ }
+ }
+
+ if (l2_dirty) {
+ ret = bdrv_pwrite(bs->file, l2_offset, l2_table,
+ s->l2_size * sizeof(uint64_t));
+ if (ret != s->l2_size * sizeof(uint64_t)) {
+ fprintf(stderr, "ERROR: Could not write L2 table\n");
+ res->check_errors++;
+ if (ret >= 0) {
+ ret = -EIO;
+ }
+ goto fail;
+ }
+ }
+ }
+
+
res->image_end_offset = (highest_cluster + 1) * s->cluster_size;
ret = 0;
fail:
+ g_free(l2_table);
g_free(refcount_table);
return ret;
diff --git a/block/qcow2.c b/block/qcow2.c
index e860834..7ed4c5d 100644
--- a/block/qcow2.c
+++ b/block/qcow2.c
@@ -312,7 +312,11 @@ static int qcow2_check(BlockDriverState *bs,
BdrvCheckResult *result,
}
if (fix && result->check_errors == 0 && result->corruptions == 0) {
- return qcow2_mark_clean(bs);
+ ret = qcow2_mark_clean(bs);
+ if (ret < 0) {
+ return ret;
+ }
+ return qcow2_mark_consistent(bs);
}
return ret;
}
diff --git a/block/qcow2.h b/block/qcow2.h
index fa2425b..1dd9f2d 100644
--- a/block/qcow2.h
+++ b/block/qcow2.h
@@ -421,6 +421,7 @@ int qcow2_pre_write_overlap_check(BlockDriverState *bs, int
chk, int64_t offset,
/* qcow2-cluster.c functions */
int qcow2_grow_l1_table(BlockDriverState *bs, uint64_t min_size,
bool exact_size);
+int qcow2_write_l1_entry(BlockDriverState *bs, int l1_index);
void qcow2_l2_cache_reset(BlockDriverState *bs);
int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset);
void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
diff --git a/include/block/block.h b/include/block/block.h
index 742fce5..e6b391c 100644
--- a/include/block/block.h
+++ b/include/block/block.h
@@ -413,6 +413,7 @@ typedef enum {
BLKDBG_REFTABLE_LOAD,
BLKDBG_REFTABLE_GROW,
+ BLKDBG_REFTABLE_UPDATE,
BLKDBG_REFBLOCK_LOAD,
BLKDBG_REFBLOCK_UPDATE,
--
1.8.3.1
