#include "qemu-common.h"
#include "block/block_int.h"
#include "block/qcow2.h"
+#include "qemu/range.h"
+#include "qapi/qmp/types.h"
static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size);
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
int64_t offset, int64_t length,
- int addend);
+ int addend, enum qcow2_discard_type type);
/*********************************************************/
} else {
/* Described somewhere else. This can recurse at most twice before we
* arrive at a block that describes itself. */
- ret = update_refcount(bs, new_block, s->cluster_size, 1);
+ ret = update_refcount(bs, new_block, s->cluster_size, 1,
+ QCOW2_DISCARD_NEVER);
if (ret < 0) {
goto fail_block;
}
/* Free old table. Remember, we must not change free_cluster_index */
uint64_t old_free_cluster_index = s->free_cluster_index;
- qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
+ qcow2_free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t),
+ QCOW2_DISCARD_OTHER);
s->free_cluster_index = old_free_cluster_index;
ret = load_refcount_block(bs, new_block, (void**) refcount_block);
return ret;
}
+void qcow2_process_discards(BlockDriverState *bs, int ret)
+{
+ BDRVQcowState *s = bs->opaque;
+ Qcow2DiscardRegion *d, *next;
+
+ QTAILQ_FOREACH_SAFE(d, &s->discards, next, next) {
+ QTAILQ_REMOVE(&s->discards, d, next);
+
+ /* Discard is optional, ignore the return value */
+ if (ret >= 0) {
+ bdrv_discard(bs->file,
+ d->offset >> BDRV_SECTOR_BITS,
+ d->bytes >> BDRV_SECTOR_BITS);
+ }
+
+ g_free(d);
+ }
+}
+
+static void update_refcount_discard(BlockDriverState *bs,
+ uint64_t offset, uint64_t length)
+{
+ BDRVQcowState *s = bs->opaque;
+ Qcow2DiscardRegion *d, *p, *next;
+
+ QTAILQ_FOREACH(d, &s->discards, next) {
+ uint64_t new_start = MIN(offset, d->offset);
+ uint64_t new_end = MAX(offset + length, d->offset + d->bytes);
+
+ if (new_end - new_start <= length + d->bytes) {
+ /* There can't be any overlap, areas ending up here have no
+ * references any more and therefore shouldn't get freed another
+ * time. */
+ assert(d->bytes + length == new_end - new_start);
+ d->offset = new_start;
+ d->bytes = new_end - new_start;
+ goto found;
+ }
+ }
+
+ d = g_malloc(sizeof(*d));
+ *d = (Qcow2DiscardRegion) {
+ .bs = bs,
+ .offset = offset,
+ .bytes = length,
+ };
+ QTAILQ_INSERT_TAIL(&s->discards, d, next);
+
+found:
+ /* Merge discard requests if they are adjacent now */
+ QTAILQ_FOREACH_SAFE(p, &s->discards, next, next) {
+ if (p == d
+ || p->offset > d->offset + d->bytes
+ || d->offset > p->offset + p->bytes)
+ {
+ continue;
+ }
+
+ /* Still no overlap possible */
+ assert(p->offset == d->offset + d->bytes
+ || d->offset == p->offset + p->bytes);
+
+ QTAILQ_REMOVE(&s->discards, p, next);
+ d->offset = MIN(d->offset, p->offset);
+ d->bytes += p->bytes;
+ }
+}
+
/* XXX: cache several refcount block clusters ? */
static int QEMU_WARN_UNUSED_RESULT update_refcount(BlockDriverState *bs,
- int64_t offset, int64_t length, int addend)
+ int64_t offset, int64_t length, int addend, enum qcow2_discard_type type)
{
BDRVQcowState *s = bs->opaque;
int64_t start, last, cluster_offset;
s->free_cluster_index = cluster_index;
}
refcount_block[block_index] = cpu_to_be16(refcount);
+
+ if (refcount == 0 && s->discard_passthrough[type]) {
+ update_refcount_discard(bs, cluster_offset, s->cluster_size);
+ }
}
ret = 0;
fail:
+ if (!s->cache_discards) {
+ qcow2_process_discards(bs, ret);
+ }
+
/* Write last changed block to disk */
if (refcount_block) {
int wret;
*/
if (ret < 0) {
int dummy;
- dummy = update_refcount(bs, offset, cluster_offset - offset, -addend);
+ dummy = update_refcount(bs, offset, cluster_offset - offset, -addend,
+ QCOW2_DISCARD_NEVER);
(void)dummy;
}
*/
static int update_cluster_refcount(BlockDriverState *bs,
int64_t cluster_index,
- int addend)
+ int addend,
+ enum qcow2_discard_type type)
{
BDRVQcowState *s = bs->opaque;
int ret;
- ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend);
+ ret = update_refcount(bs, cluster_index << s->cluster_bits, 1, addend,
+ type);
if (ret < 0) {
return ret;
}
return offset;
}
- ret = update_refcount(bs, offset, size, 1);
+ ret = update_refcount(bs, offset, size, 1, QCOW2_DISCARD_NEVER);
if (ret < 0) {
return ret;
}
old_free_cluster_index = s->free_cluster_index;
s->free_cluster_index = cluster_index + i;
- ret = update_refcount(bs, offset, i << s->cluster_bits, 1);
+ ret = update_refcount(bs, offset, i << s->cluster_bits, 1,
+ QCOW2_DISCARD_NEVER);
if (ret < 0) {
return ret;
}
if (free_in_cluster == 0)
s->free_byte_offset = 0;
if ((offset & (s->cluster_size - 1)) != 0)
- update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
+ update_cluster_refcount(bs, offset >> s->cluster_bits, 1,
+ QCOW2_DISCARD_NEVER);
} else {
offset = qcow2_alloc_clusters(bs, s->cluster_size);
if (offset < 0) {
if ((cluster_offset + s->cluster_size) == offset) {
/* we are lucky: contiguous data */
offset = s->free_byte_offset;
- update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
+ update_cluster_refcount(bs, offset >> s->cluster_bits, 1,
+ QCOW2_DISCARD_NEVER);
s->free_byte_offset += size;
} else {
s->free_byte_offset = offset;
}
void qcow2_free_clusters(BlockDriverState *bs,
- int64_t offset, int64_t size)
+ int64_t offset, int64_t size,
+ enum qcow2_discard_type type)
{
int ret;
BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_FREE);
- ret = update_refcount(bs, offset, size, -1);
+ ret = update_refcount(bs, offset, size, -1, type);
if (ret < 0) {
fprintf(stderr, "qcow2_free_clusters failed: %s\n", strerror(-ret));
/* TODO Remember the clusters to free them later and avoid leaking */
* Free a cluster using its L2 entry (handles clusters of all types, e.g.
* normal cluster, compressed cluster, etc.)
*/
-void qcow2_free_any_clusters(BlockDriverState *bs,
- uint64_t l2_entry, int nb_clusters)
+void qcow2_free_any_clusters(BlockDriverState *bs, uint64_t l2_entry,
+ int nb_clusters, enum qcow2_discard_type type)
{
BDRVQcowState *s = bs->opaque;
s->csize_mask) + 1;
qcow2_free_clusters(bs,
(l2_entry & s->cluster_offset_mask) & ~511,
- nb_csectors * 512);
+ nb_csectors * 512, type);
}
break;
case QCOW2_CLUSTER_NORMAL:
qcow2_free_clusters(bs, l2_entry & L2E_OFFSET_MASK,
- nb_clusters << s->cluster_bits);
+ nb_clusters << s->cluster_bits, type);
break;
case QCOW2_CLUSTER_UNALLOCATED:
case QCOW2_CLUSTER_ZERO:
l1_table = NULL;
l1_size2 = l1_size * sizeof(uint64_t);
+ s->cache_discards = true;
+
/* WARNING: qcow2_snapshot_goto relies on this function not using the
* l1_table_offset when it is the current s->l1_table_offset! Be careful
* when changing this! */
}
for(j = 0; j < s->l2_size; j++) {
+ uint64_t cluster_index;
+
offset = be64_to_cpu(l2_table[j]);
- if (offset != 0) {
- old_offset = offset;
- offset &= ~QCOW_OFLAG_COPIED;
- if (offset & QCOW_OFLAG_COMPRESSED) {
+ old_offset = offset;
+ offset &= ~QCOW_OFLAG_COPIED;
+
+ switch (qcow2_get_cluster_type(offset)) {
+ case QCOW2_CLUSTER_COMPRESSED:
nb_csectors = ((offset >> s->csize_shift) &
s->csize_mask) + 1;
if (addend != 0) {
int ret;
ret = update_refcount(bs,
(offset & s->cluster_offset_mask) & ~511,
- nb_csectors * 512, addend);
+ nb_csectors * 512, addend,
+ QCOW2_DISCARD_SNAPSHOT);
if (ret < 0) {
goto fail;
}
}
/* compressed clusters are never modified */
refcount = 2;
- } else {
- uint64_t cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
+ break;
+
+ case QCOW2_CLUSTER_NORMAL:
+ case QCOW2_CLUSTER_ZERO:
+ cluster_index = (offset & L2E_OFFSET_MASK) >> s->cluster_bits;
+ if (!cluster_index) {
+ /* unallocated */
+ refcount = 0;
+ break;
+ }
if (addend != 0) {
- refcount = update_cluster_refcount(bs, cluster_index, addend);
+ refcount = update_cluster_refcount(bs, cluster_index, addend,
+ QCOW2_DISCARD_SNAPSHOT);
} else {
refcount = get_refcount(bs, cluster_index);
}
ret = refcount;
goto fail;
}
- }
+ break;
- if (refcount == 1) {
- offset |= QCOW_OFLAG_COPIED;
- }
- if (offset != old_offset) {
- if (addend > 0) {
- qcow2_cache_set_dependency(bs, s->l2_table_cache,
- s->refcount_block_cache);
- }
- l2_table[j] = cpu_to_be64(offset);
- qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
+ case QCOW2_CLUSTER_UNALLOCATED:
+ refcount = 0;
+ break;
+
+ default:
+ abort();
+ }
+
+ if (refcount == 1) {
+ offset |= QCOW_OFLAG_COPIED;
+ }
+ if (offset != old_offset) {
+ if (addend > 0) {
+ qcow2_cache_set_dependency(bs, s->l2_table_cache,
+ s->refcount_block_cache);
}
+ l2_table[j] = cpu_to_be64(offset);
+ qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
}
}
if (addend != 0) {
- refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
+ refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend,
+ QCOW2_DISCARD_SNAPSHOT);
} else {
refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
}
qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
}
+ s->cache_discards = false;
+ qcow2_process_discards(bs, ret);
+
/* Update L1 only if it isn't deleted anyway (addend = -1) */
if (ret == 0 && addend >= 0 && l1_modified) {
for (i = 0; i < l1_size; i++) {
if (num_fixed) {
ret = update_refcount(bs, i << s->cluster_bits, 1,
- refcount2 - refcount1);
+ refcount2 - refcount1,
+ QCOW2_DISCARD_ALWAYS);
if (ret >= 0) {
(*num_fixed)++;
continue;
return ret;
}
+#define overlaps_with(ofs, sz) \
+ ranges_overlap(offset, size, ofs, sz)
+
+/*
+ * Checks if the given offset into the image file is actually free to use by
+ * looking for overlaps with important metadata sections (L1/L2 tables etc.),
+ * i.e. a sanity check without relying on the refcount tables.
+ *
+ * The chk parameter specifies exactly what checks to perform (being a bitmask
+ * of QCow2MetadataOverlap values).
+ *
+ * Returns:
+ * - 0 if writing to this offset will not affect the mentioned metadata
+ * - a positive QCow2MetadataOverlap value indicating one overlapping section
+ * - a negative value (-errno) indicating an error while performing a check,
+ * e.g. when bdrv_read failed on QCOW2_OL_INACTIVE_L2
+ */
+int qcow2_check_metadata_overlap(BlockDriverState *bs, int chk, int64_t offset,
+ int64_t size)
+{
+ BDRVQcowState *s = bs->opaque;
+ int i, j;
+
+ if (!size) {
+ return 0;
+ }
+
+ if (chk & QCOW2_OL_MAIN_HEADER) {
+ if (offset < s->cluster_size) {
+ return QCOW2_OL_MAIN_HEADER;
+ }
+ }
+
+ /* align range to test to cluster boundaries */
+ size = align_offset(offset_into_cluster(s, offset) + size, s->cluster_size);
+ offset = start_of_cluster(s, offset);
+
+ if ((chk & QCOW2_OL_ACTIVE_L1) && s->l1_size) {
+ if (overlaps_with(s->l1_table_offset, s->l1_size * sizeof(uint64_t))) {
+ return QCOW2_OL_ACTIVE_L1;
+ }
+ }
+
+ if ((chk & QCOW2_OL_REFCOUNT_TABLE) && s->refcount_table_size) {
+ if (overlaps_with(s->refcount_table_offset,
+ s->refcount_table_size * sizeof(uint64_t))) {
+ return QCOW2_OL_REFCOUNT_TABLE;
+ }
+ }
+
+ if ((chk & QCOW2_OL_SNAPSHOT_TABLE) && s->snapshots_size) {
+ if (overlaps_with(s->snapshots_offset, s->snapshots_size)) {
+ return QCOW2_OL_SNAPSHOT_TABLE;
+ }
+ }
+
+ if ((chk & QCOW2_OL_INACTIVE_L1) && s->snapshots) {
+ for (i = 0; i < s->nb_snapshots; i++) {
+ if (s->snapshots[i].l1_size &&
+ overlaps_with(s->snapshots[i].l1_table_offset,
+ s->snapshots[i].l1_size * sizeof(uint64_t))) {
+ return QCOW2_OL_INACTIVE_L1;
+ }
+ }
+ }
+
+ if ((chk & QCOW2_OL_ACTIVE_L2) && s->l1_table) {
+ for (i = 0; i < s->l1_size; i++) {
+ if ((s->l1_table[i] & L1E_OFFSET_MASK) &&
+ overlaps_with(s->l1_table[i] & L1E_OFFSET_MASK,
+ s->cluster_size)) {
+ return QCOW2_OL_ACTIVE_L2;
+ }
+ }
+ }
+
+ if ((chk & QCOW2_OL_REFCOUNT_BLOCK) && s->refcount_table) {
+ for (i = 0; i < s->refcount_table_size; i++) {
+ if ((s->refcount_table[i] & REFT_OFFSET_MASK) &&
+ overlaps_with(s->refcount_table[i] & REFT_OFFSET_MASK,
+ s->cluster_size)) {
+ return QCOW2_OL_REFCOUNT_BLOCK;
+ }
+ }
+ }
+
+ if ((chk & QCOW2_OL_INACTIVE_L2) && s->snapshots) {
+ for (i = 0; i < s->nb_snapshots; i++) {
+ uint64_t l1_ofs = s->snapshots[i].l1_table_offset;
+ uint32_t l1_sz = s->snapshots[i].l1_size;
+ uint64_t *l1 = g_malloc(l1_sz * sizeof(uint64_t));
+ int ret;
+
+ ret = bdrv_read(bs->file, l1_ofs / BDRV_SECTOR_SIZE, (uint8_t *)l1,
+ l1_sz * sizeof(uint64_t) / BDRV_SECTOR_SIZE);
+
+ if (ret < 0) {
+ g_free(l1);
+ return ret;
+ }
+
+ for (j = 0; j < l1_sz; j++) {
+ if ((l1[j] & L1E_OFFSET_MASK) &&
+ overlaps_with(l1[j] & L1E_OFFSET_MASK, s->cluster_size)) {
+ g_free(l1);
+ return QCOW2_OL_INACTIVE_L2;
+ }
+ }
+
+ g_free(l1);
+ }
+ }
+
+ return 0;
+}
+
+static const char *metadata_ol_names[] = {
+ [QCOW2_OL_MAIN_HEADER_BITNR] = "qcow2_header",
+ [QCOW2_OL_ACTIVE_L1_BITNR] = "active L1 table",
+ [QCOW2_OL_ACTIVE_L2_BITNR] = "active L2 table",
+ [QCOW2_OL_REFCOUNT_TABLE_BITNR] = "refcount table",
+ [QCOW2_OL_REFCOUNT_BLOCK_BITNR] = "refcount block",
+ [QCOW2_OL_SNAPSHOT_TABLE_BITNR] = "snapshot table",
+ [QCOW2_OL_INACTIVE_L1_BITNR] = "inactive L1 table",
+ [QCOW2_OL_INACTIVE_L2_BITNR] = "inactive L2 table",
+};
+
+/*
+ * First performs a check for metadata overlaps (through
+ * qcow2_check_metadata_overlap); if that fails with a negative value (error
+ * while performing a check), that value is returned. If an impending overlap
+ * is detected, the BDS will be made unusable, the qcow2 file marked corrupt
+ * and -EIO returned.
+ *
+ * Returns 0 if there were neither overlaps nor errors while checking for
+ * overlaps; or a negative value (-errno) on error.
+ */
+int qcow2_pre_write_overlap_check(BlockDriverState *bs, int chk, int64_t offset,
+ int64_t size)
+{
+ int ret = qcow2_check_metadata_overlap(bs, chk, offset, size);
+
+ if (ret < 0) {
+ return ret;
+ } else if (ret > 0) {
+ int metadata_ol_bitnr = ffs(ret) - 1;
+ char *message;
+ QObject *data;
+
+ assert(metadata_ol_bitnr < QCOW2_OL_MAX_BITNR);
+
+ fprintf(stderr, "qcow2: Preventing invalid write on metadata (overlaps "
+ "with %s); image marked as corrupt.\n",
+ metadata_ol_names[metadata_ol_bitnr]);
+ message = g_strdup_printf("Prevented %s overwrite",
+ metadata_ol_names[metadata_ol_bitnr]);
+ data = qobject_from_jsonf("{ 'device': %s, 'msg': %s, 'offset': %"
+ PRId64 ", 'size': %" PRId64 " }", bs->device_name, message,
+ offset, size);
+ monitor_protocol_event(QEVENT_BLOCK_IMAGE_CORRUPTED, data);
+ g_free(message);
+ qobject_decref(data);
+
+ qcow2_mark_corrupt(bs);
+ bs->drv = NULL; /* make BDS unusable */
+ return -EIO;
+ }
+
+ return 0;
+}