Skip to contentSkip to conversation

Add subcluster allocation to qcow2

RFC PATCH Superseded Alberto Garcia 3 participants 65 messages Oct 26, 2019 – Nov 15, 2019 Block layer

v2 of 11 Superseded Latest is v11 →

lore
Share & tools
lore
#61 lore raw

RFC PATCH v2 25/26 qcow2: Allow preallocation and backing files if extended_l2 is set

In reply to Alberto Garcia

Traditional qcow2 images don't allow preallocation if a backing file is set. This is because once a cluster is allocated there is no way to tell that its data should be read from the backing file.

Extended L2 entries have individual allocation bits for each subcluster, and therefore it is perfectly possible to have an allocated cluster with all its subclusters unallocated.

---
 block/qcow2.c              | 7 ++++---
 tests/qemu-iotests/206.out | 2 +-
 2 files changed, 5 insertions(+), 4 deletions(-)
block/qcow2.c +4 -3
index b1fa7ab5da..8cf51c5d64 100644
@@ -3307,10 +3307,11 @@ qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
     }
     if (qcow2_opts->has_backing_file &&
-        qcow2_opts->preallocation != PREALLOC_MODE_OFF)
+        qcow2_opts->preallocation != PREALLOC_MODE_OFF &&
+        !qcow2_opts->extended_l2)
     {
-        error_setg(errp, "Backing file and preallocation cannot be used at "
-                   "the same time");
+        error_setg(errp, "Backing file and preallocation can only be used at "
+                   "the same time if extended_l2 is on");
         ret = -EINVAL;
         goto out;
     }
tests/qemu-iotests/206.out +1 -1
index d2efc0394a..cfddfbfaa4 100644
@@ -198,7 +198,7 @@ Job failed: Different refcount widths than 16 bits require compatibility level 1
 === Invalid backing file options ===
 {"execute": "blockdev-create", "arguments": {"job-id": "job0", "options": {"backing-file": "/dev/null", "driver": "qcow2", "file": "node0", "preallocation": "full", "size": 67108864}}}
 {"return": {}}
-Job failed: Backing file and preallocation cannot be used at the same time
+Job failed: Backing file and preallocation can only be used at the same time if extended_l2 is on
 {"execute": "job-dismiss", "arguments": {"id": "job0"}}
 {"return": {}}
 
Signature · 1 line
2.20.1
#42 lore raw

RFC PATCH v2 16/26 qcow2: Add subcluster support to discard_in_l2_slice()

In reply to Alberto Garcia

Setting the QCOW_OFLAG_ZERO bit of the L2 entry is forbidden if an image has subclusters. Instead, the individual 'all zeroes' bits must be used.

---
 block/qcow2-cluster.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
block/qcow2-cluster.c +5 -1
index 3e4ba8d448..aa3eb727a5 100644
@@ -1772,7 +1772,11 @@ static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset,
 
         /* First remove L2 entries */
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
-        if (!full_discard && s->qcow_version >= 3) {
+        if (has_subclusters(s)) {
+            set_l2_entry(s, l2_slice, l2_index + i, 0);
+            set_l2_bitmap(s, l2_slice, l2_index + i,
+                          full_discard ? 0 : QCOW_L2_BITMAP_ALL_ZEROES);
+        } else if (!full_discard && s->qcow_version >= 3) {
             set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
         } else {
             set_l2_entry(s, l2_slice, l2_index + i, 0);
Signature · 1 line
2.20.1
#31 lore raw

RFC PATCH v2 13/26 qcow2: Add subcluster support to calculate_l2_meta()

In reply to Alberto Garcia

If an image has subclusters then there are more copy-on-write scenarios that we need to consider. Let's say we have a write request from the middle of subcluster #3 until the end of the cluster:

- If the cluster is new, then subclusters #0 to #3 from the old cluster must be copied into the new one.

- If the cluster is new but the old cluster was unallocated, then only subcluster #3 needs copy-on-write. #0 to #2 are marked as unallocated in the bitmap of the new L2 entry.

- If we are overwriting an old cluster and subcluster #3 is unallocated or has the all-zeroes bit set then we need copy-on-write on subcluster #3.

- If we are overwriting an old cluster and subcluster #3 was allocated then there is no need to copy-on-write.

---
 block/qcow2-cluster.c | 136 +++++++++++++++++++++++++++++++++---------
 1 file changed, 108 insertions(+), 28 deletions(-)
block/qcow2-cluster.c +108 -28
index 1f509bda15..990bc070af 100644
@@ -1034,14 +1034,16 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
  * If @keep_old is true it means that the clusters were already
  * allocated and will be overwritten. If false then the clusters are
  * new and we have to decrease the reference count of the old ones.
+ *
+ * Returns 1 on success, -errno on failure.
  */
-static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
-                              uint64_t guest_offset, uint64_t bytes,
-                              uint64_t *l2_slice, QCowL2Meta **m, bool keep_old)
+static int calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
+                             uint64_t guest_offset, uint64_t bytes,
+                             uint64_t *l2_slice, QCowL2Meta **m, bool keep_old)
 {
     BDRVQcow2State *s = bs->opaque;
-    int l2_index = offset_to_l2_slice_index(s, guest_offset);
-    uint64_t l2_entry;
+    int sc_index, l2_index = offset_to_l2_slice_index(s, guest_offset);
+    uint64_t l2_entry, l2_bitmap;
     unsigned cow_start_from, cow_end_to;
     unsigned cow_start_to = offset_into_cluster(s, guest_offset);
     unsigned cow_end_from = cow_start_to + bytes;
@@ -1049,38 +1051,108 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
     QCowL2Meta *old_m = *m;
     QCow2ClusterType type;
 
-    /* Return if there's no COW (all clusters are normal and we keep them) */
+    /* Return if there's no COW (all subclusters are normal and we are
+     * keeping the clusters) */
     if (keep_old) {
+        unsigned first_sc = cow_start_to / s->subcluster_size;
+        unsigned last_sc = (cow_end_from - 1) / s->subcluster_size;
         int i;
-        for (i = 0; i < nb_clusters; i++) {
-            l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
-            if (qcow2_get_cluster_type(bs, l2_entry) != QCOW2_CLUSTER_NORMAL) {
+        for (i = first_sc; i <= last_sc; i++) {
+            unsigned c = i / s->subclusters_per_cluster;
+            unsigned sc = i % s->subclusters_per_cluster;
+            l2_entry = get_l2_entry(s, l2_slice, l2_index + c);
+            l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + c);
+            type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc);
+            if (type == QCOW2_CLUSTER_INVALID) {
+                l2_index += c; /* Point to the invalid entry */
+                goto fail;
+            }
+            if (type != QCOW2_CLUSTER_NORMAL) {
                 break;
             }
         }
-        if (i == nb_clusters) {
-            return;
+        if (i == last_sc + 1) {
+            return 1;
         }
     }
 
     /* Get the L2 entry from the first cluster */
     l2_entry = get_l2_entry(s, l2_slice, l2_index);
-    type = qcow2_get_cluster_type(bs, l2_entry);
+    l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
+    sc_index = offset_to_sc_index(s, guest_offset);
+    type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index);
 
-    if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
-        cow_start_from = cow_start_to;
+    if (type == QCOW2_CLUSTER_INVALID) {
+        goto fail;
+    }
+
+    if (!keep_old) {
+        switch (type) {
+        case QCOW2_CLUSTER_NORMAL:
+        case QCOW2_CLUSTER_COMPRESSED:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+        case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
+            cow_start_from = 0;
+            break;
+        case QCOW2_CLUSTER_ZERO_PLAIN:
+        case QCOW2_CLUSTER_UNALLOCATED:
+            cow_start_from = sc_index << s->subcluster_bits;
+            break;
+        default:
+            g_assert_not_reached();
+        }
     } else {
-        cow_start_from = 0;
+        switch (type) {
+        case QCOW2_CLUSTER_NORMAL:
+            cow_start_from = cow_start_to;
+            break;
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+        case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
+            cow_start_from = sc_index << s->subcluster_bits;
+            break;
+        default:
+            g_assert_not_reached();
+        }
     }
 
     /* Get the L2 entry from the last cluster */
-    l2_entry = get_l2_entry(s, l2_slice, l2_index + nb_clusters - 1);
-    type = qcow2_get_cluster_type(bs, l2_entry);
+    l2_index += nb_clusters - 1;
+    l2_entry = get_l2_entry(s, l2_slice, l2_index);
+    l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
+    sc_index = offset_to_sc_index(s, guest_offset + bytes - 1);
+    type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index);
 
-    if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
-        cow_end_to = cow_end_from;
+    if (type == QCOW2_CLUSTER_INVALID) {
+        goto fail;
+    }
+
+    if (!keep_old) {
+        switch (type) {
+        case QCOW2_CLUSTER_NORMAL:
+        case QCOW2_CLUSTER_COMPRESSED:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+        case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
+            cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
+            break;
+        case QCOW2_CLUSTER_ZERO_PLAIN:
+        case QCOW2_CLUSTER_UNALLOCATED:
+            cow_end_to = ROUND_UP(cow_end_from, s->subcluster_size);
+            break;
+        default:
+            g_assert_not_reached();
+        }
     } else {
-        cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
+        switch (type) {
+        case QCOW2_CLUSTER_NORMAL:
+            cow_end_to = cow_end_from;
+            break;
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+        case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
+            cow_end_to = ROUND_UP(cow_end_from, s->subcluster_size);
+            break;
+        default:
+            g_assert_not_reached();
+        }
     }
 
     *m = g_malloc0(sizeof(**m));
@@ -1105,6 +1177,18 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
 
     qemu_co_queue_init(&(*m)->dependent_requests);
     QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
+
+fail:
+    if (type == QCOW2_CLUSTER_INVALID) {
+        uint64_t l1_index = offset_to_l1_index(s, guest_offset);
+        uint64_t l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
+        qcow2_signal_corruption(bs, true, -1, -1, "Invalid cluster entry found "
+                                " (L2 offset: %#" PRIx64 ", L2 index: %#x)",
+                                l2_offset, l2_index);
+        return -EIO;
+    }
+
+    return 1;
 }
 
 /* Returns true if the cluster is unallocated or has refcount > 1 */
@@ -1313,10 +1397,8 @@ static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
                  - offset_into_cluster(s, guest_offset));
         assert(*bytes != 0);
 
-        calculate_l2_meta(bs, cluster_offset & L2E_OFFSET_MASK, guest_offset,
-                          *bytes, l2_slice, m, true);
-
-        ret = 1;
+        ret = calculate_l2_meta(bs, cluster_offset & L2E_OFFSET_MASK,
+                                guest_offset, *bytes, l2_slice, m, true);
     } else {
         ret = 0;
     }
@@ -1491,10 +1573,8 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
     assert(*bytes != 0);
 
-    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes, l2_slice,
-                      m, false);
-
-    ret = 1;
+    ret = calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
+                            l2_slice, m, false);
 
 out:
     qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
Signature · 1 line
2.20.1
#54 lore raw

RFC PATCH v2 21/26 qcow2: Clear the L2 bitmap when allocating a compressed cluster

In reply to Alberto Garcia

Compressed clusters always have the bitmap part of the extended L2 entry set to 0.

---
 block/qcow2-cluster.c | 1 +
 1 file changed, 1 insertion(+)
block/qcow2-cluster.c +1 -0
index acb7226e03..3ba8a98073 100644
@@ -783,6 +783,7 @@ int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
     qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
     set_l2_entry(s, l2_slice, l2_index, cluster_offset);
+    set_l2_bitmap(s, l2_slice, l2_index, 0);
     qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
 
     *host_offset = cluster_offset & s->cluster_offset_mask;
Signature · 1 line
2.20.1
#34 lore raw

RFC PATCH v2 14/26 qcow2: Add subcluster support to qcow2_get_cluster_offset()

In reply to Alberto Garcia

The logic of this function remains pretty much the same, except that it uses count_contiguous_subclusters(), which combines the logic of count_contiguous_clusters() / count_contiguous_clusters_unallocated() and checks individual subclusters.

---
 block/qcow2-cluster.c | 111 ++++++++++++++++++++----------------------
 1 file changed, 52 insertions(+), 59 deletions(-)
block/qcow2-cluster.c +52 -59
index 990bc070af..e67559152f 100644
@@ -372,66 +372,51 @@ fail:
 }
 
 /*
- * Checks how many clusters in a given L2 slice are contiguous in the image
- * file. As soon as one of the flags in the bitmask stop_flags changes compared
- * to the first cluster, the search is stopped and the cluster is not counted
- * as contiguous. (This allows it, for example, to stop at the first compressed
- * cluster which may require a different handling)
+ * Return the number of contiguous subclusters of the exact same type
+ * in a given L2 slice, starting from cluster @l2_index, subcluster
+ * @sc_index. At most @nb_clusters are checked. Allocated clusters are
+ * also required to be contiguous in the image file.
  */
-static int count_contiguous_clusters(BlockDriverState *bs, int nb_clusters,
-        int cluster_size, uint64_t *l2_slice, int l2_index, uint64_t stop_flags)
+static int count_contiguous_subclusters(BlockDriverState *bs, int nb_clusters,
+                                        unsigned sc_index, uint64_t *l2_slice,
+                                        int l2_index)
 {
     BDRVQcow2State *s = bs->opaque;
-    int i;
-    QCow2ClusterType first_cluster_type;
-    uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
-    uint64_t first_entry = get_l2_entry(s, l2_slice, l2_index);
-    uint64_t offset = first_entry & mask;
-
-    first_cluster_type = qcow2_get_cluster_type(bs, first_entry);
-    if (first_cluster_type == QCOW2_CLUSTER_UNALLOCATED) {
-        return 0;
+    int i, j, count = 0;
+    uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index);
+    uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
+    uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK;
+    bool check_offset = true;
+    QCow2ClusterType type =
+        qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index);
+
+    assert(type != QCOW2_CLUSTER_INVALID); /* The caller should check this */
+
+    if (type == QCOW2_CLUSTER_COMPRESSED) {
+        return 1; /* Compressed clusters are always counted one by one */
     }
 
-    /* must be allocated */
-    assert(first_cluster_type == QCOW2_CLUSTER_NORMAL ||
-           first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC);
-
-    for (i = 0; i < nb_clusters; i++) {
-        uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index + i) & mask;
-        if (offset + (uint64_t) i * cluster_size != l2_entry) {
-            break;
-        }
+    if (type == QCOW2_CLUSTER_UNALLOCATED || type == QCOW2_CLUSTER_ZERO_PLAIN) {
+        check_offset = false;
     }
 
-        return i;
-}
-
-/*
- * Checks how many consecutive unallocated clusters in a given L2
- * slice have the same cluster type.
- */
-static int count_contiguous_clusters_unallocated(BlockDriverState *bs,
-                                                 int nb_clusters,
-                                                 uint64_t *l2_slice,
-                                                 int l2_index,
-                                                 QCow2ClusterType wanted_type)
-{
-    BDRVQcow2State *s = bs->opaque;
-    int i;
-
-    assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN ||
-           wanted_type == QCOW2_CLUSTER_UNALLOCATED);
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t entry = get_l2_entry(s, l2_slice, l2_index + i);
-        QCow2ClusterType type = qcow2_get_cluster_type(bs, entry);
-
-        if (type != wanted_type) {
-            break;
+        l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
+        l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
+        if (check_offset && expected_offset != (l2_entry & L2E_OFFSET_MASK)) {
+            goto out;
+        }
+        for (j = (i == 0) ? sc_index : 0; j < s->subclusters_per_cluster; j++) {
+            if (qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, j) != type) {
+                goto out;
+            }
+            count++;
         }
+        expected_offset += s->cluster_size;
     }
 
-    return i;
+out:
+    return count;
 }
 
 static int coroutine_fn do_perform_cow_read(BlockDriverState *bs,
@@ -514,8 +499,8 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
                              unsigned int *bytes, uint64_t *cluster_offset)
 {
     BDRVQcow2State *s = bs->opaque;
-    unsigned int l2_index;
-    uint64_t l1_index, l2_offset, *l2_slice;
+    unsigned int l2_index, sc_index;
+    uint64_t l1_index, l2_offset, *l2_slice, l2_bitmap;
     int c;
     unsigned int offset_in_cluster;
     uint64_t bytes_available, bytes_needed, nb_clusters;
@@ -569,7 +554,9 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
     /* find the cluster offset for the given disk offset */
 
     l2_index = offset_to_l2_slice_index(s, offset);
+    sc_index = offset_to_sc_index(s, offset);
     *cluster_offset = get_l2_entry(s, l2_slice, l2_index);
+    l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
 
     nb_clusters = size_to_clusters(s, bytes_needed);
     /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned
@@ -577,7 +564,7 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
      * true */
     assert(nb_clusters <= INT_MAX);
 
-    type = qcow2_get_cluster_type(bs, *cluster_offset);
+    type = qcow2_get_subcluster_type(bs, *cluster_offset, l2_bitmap, sc_index);
     if (s->qcow_version < 3 && (type == QCOW2_CLUSTER_ZERO_PLAIN ||
                                 type == QCOW2_CLUSTER_ZERO_ALLOC)) {
         qcow2_signal_corruption(bs, true, -1, -1, "Zero cluster entry found"
@@ -587,6 +574,13 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
         goto fail;
     }
     switch (type) {
+    case QCOW2_CLUSTER_INVALID:
+        qcow2_signal_corruption(bs, true, -1, -1, "Invalid cluster entry found "
+                                " (L2 offset: %#" PRIx64 ", L2 index: %#x)",
+                                l2_offset, l2_index);
+        ret = -EIO;
+        goto fail;
+        break;
     case QCOW2_CLUSTER_COMPRESSED:
         if (has_data_file(bs)) {
             qcow2_signal_corruption(bs, true, -1, -1, "Compressed cluster "
@@ -602,16 +596,15 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
         break;
     case QCOW2_CLUSTER_ZERO_PLAIN:
     case QCOW2_CLUSTER_UNALLOCATED:
-        /* how many empty clusters ? */
-        c = count_contiguous_clusters_unallocated(bs, nb_clusters,
-                                                  l2_slice, l2_index, type);
+        c = count_contiguous_subclusters(bs, nb_clusters, sc_index,
+                                         l2_slice, l2_index);
         *cluster_offset = 0;
         break;
     case QCOW2_CLUSTER_ZERO_ALLOC:
     case QCOW2_CLUSTER_NORMAL:
-        /* how many allocated clusters ? */
-        c = count_contiguous_clusters(bs, nb_clusters, s->cluster_size,
-                                      l2_slice, l2_index, QCOW_OFLAG_ZERO);
+    case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
+        c = count_contiguous_subclusters(bs, nb_clusters, sc_index,
+                                         l2_slice, l2_index);
         *cluster_offset &= L2E_OFFSET_MASK;
         if (offset_into_cluster(s, *cluster_offset)) {
             qcow2_signal_corruption(bs, true, -1, -1,
@@ -640,7 +633,7 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
 
     qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
 
-    bytes_available = (int64_t)c * s->cluster_size;
+    bytes_available = ((int64_t)c + sc_index) * s->subcluster_size;
 
 out:
     if (bytes_available > bytes_needed) {
Signature · 1 line
2.20.1
#51 lore raw

RFC PATCH v2 20/26 qcow2: Update L2 bitmap in qcow2_alloc_cluster_link_l2()

In reply to Alberto Garcia

The L2 bitmap needs to be updated after each write to indicate what new subclusters are now allocated.

This needs to happen even if the cluster was already allocated and the L2 entry was otherwise valid.

---
 block/qcow2-cluster.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
block/qcow2-cluster.c +16 -0
index fb6cf8df17..acb7226e03 100644
@@ -980,6 +980,22 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
 
         set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_COPIED |
                      (cluster_offset + (i << s->cluster_bits)));
+
+        /* Update bitmap with the subclusters that were just written */
+        if (has_subclusters(s)) {
+            uint64_t written_from = m->cow_start.offset;
+            uint64_t written_to = m->cow_end.offset + m->cow_end.nb_bytes;
+            uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
+            int sc;
+            for (sc = 0; sc < s->subclusters_per_cluster; sc++) {
+                uint64_t sc_off = i * s->cluster_size + sc * s->subcluster_size;
+                if (sc_off >= written_from && sc_off < written_to) {
+                    l2_bitmap |= QCOW_OFLAG_SUB_ALLOC(sc);
+                    l2_bitmap &= ~QCOW_OFLAG_SUB_ZERO(sc);
+                }
+            }
+            set_l2_bitmap(s, l2_slice, l2_index + i, l2_bitmap);
+        }
      }
 
 
Signature · 1 line
2.20.1
#38 lore raw

RFC PATCH v2 15/26 qcow2: Add subcluster support to zero_in_l2_slice()

In reply to Alberto Garcia

Setting the QCOW_OFLAG_ZERO bit of the L2 entry is forbidden if an image has subclusters. Instead, the individual 'all zeroes' bits must be used.

---
 block/qcow2-cluster.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
block/qcow2-cluster.c +10 -4
index e67559152f..3e4ba8d448 100644
@@ -1852,7 +1852,7 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
     assert(nb_clusters <= INT_MAX);
 
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t old_offset;
+        uint64_t old_offset, l2_entry = 0;
         QCow2ClusterType cluster_type;
 
         old_offset = get_l2_entry(s, l2_slice, l2_index + i);
@@ -1869,12 +1869,18 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
 
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
         if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) {
-            set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
             qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
         } else {
-            uint64_t entry = get_l2_entry(s, l2_slice, l2_index + i);
-            set_l2_entry(s, l2_slice, l2_index + i, entry | QCOW_OFLAG_ZERO);
+            l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
         }
+
+        if (has_subclusters(s)) {
+            set_l2_bitmap(s, l2_slice, l2_index + i, QCOW_L2_BITMAP_ALL_ZEROES);
+        } else {
+            l2_entry |= QCOW_OFLAG_ZERO;
+        }
+
+        set_l2_entry(s, l2_slice, l2_index + i, l2_entry);
     }
 
     qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
Signature · 1 line
2.20.1
#50 lore raw

RFC PATCH v2 19/26 qcow2: Fix offset calculation in handle_dependencies()

In reply to Alberto Garcia

l2meta_cow_start() and l2meta_cow_end() are not necessarily cluster-aligned if the image has subclusters, so update the calculation of old_start and old_end to guarantee that no two requests try to write on the same cluster.

---
 block/qcow2-cluster.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
block/qcow2-cluster.c +2 -2
index 62f2a9fcc0..fb6cf8df17 100644
@@ -1262,8 +1262,8 @@ static int handle_dependencies(BlockDriverState *bs, uint64_t guest_offset,
 
         uint64_t start = guest_offset;
         uint64_t end = start + bytes;
-        uint64_t old_start = l2meta_cow_start(old_alloc);
-        uint64_t old_end = l2meta_cow_end(old_alloc);
+        uint64_t old_start = start_of_cluster(s, l2meta_cow_start(old_alloc));
+        uint64_t old_end = ROUND_UP(l2meta_cow_end(old_alloc), s->cluster_size);
 
         if (end <= old_start || start >= old_end) {
             /* No intersection */
Signature · 1 line
2.20.1
#8 lore raw

RFC PATCH v2 3/26 qcow2: Process QCOW2_CLUSTER_ZERO_ALLOC clusters in handle_copied()

In reply to Alberto Garcia

When writing to a qcow2 file there are two functions that take a virtual offset and return a host offset, possibly allocating new clusters if necessary:

- handle_copied() looks for normal data clusters that are already allocated and have a reference count of 1. In those clusters we can simply write the data and there is no need to perform any copy-on-write.

- handle_alloc() looks for clusters that do need copy-on-write, either because they haven't been allocated yet, because their reference count is != 1 or because they are ZERO_ALLOC clusters.

The ZERO_ALLOC case is a bit special because those are clusters that are already allocated and they could perfectly be dealt with in handle_copied() (as long as copy-on-write is performed when required).

In fact, there is extra code specifically for them in handle_alloc() that tries to reuse the existing allocation if possible and frees them otherwise.

This patch changes the handling of ZERO_ALLOC clusters so the semantics of these two functions are now like this:

- handle_copied() looks for clusters that are already allocated and which we can overwrite (NORMAL and ZERO_ALLOC clusters with a reference count of 1).

- handle_alloc() looks for clusters for which we need a new allocation (all other cases).

One importante difference after this change is that clusters found in handle_copied() may now require copy-on-write, but this will be anyway necessary once we add support for subclusters.

---
 block/qcow2-cluster.c | 177 +++++++++++++++++++++++-------------------
 1 file changed, 96 insertions(+), 81 deletions(-)
block/qcow2-cluster.c +96 -81
index aa1010a515..ee6b46f917 100644
@@ -1021,7 +1021,8 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
 
 /*
  * For a given write request, create a new QCowL2Meta structure and
- * add it to @m.
+ * add it to @m. If the write request does not need copy-on-write or
+ * changes to the L2 metadata then this function does nothing.
  *
  * @host_offset points to the beginning of the first cluster.
  *
@@ -1034,15 +1035,51 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
  */
 static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
                               uint64_t guest_offset, uint64_t bytes,
-                              QCowL2Meta **m, bool keep_old)
+                              uint64_t *l2_slice, QCowL2Meta **m, bool keep_old)
 {
     BDRVQcow2State *s = bs->opaque;
-    unsigned cow_start_from = 0;
+    int l2_index = offset_to_l2_slice_index(s, guest_offset);
+    uint64_t l2_entry;
+    unsigned cow_start_from, cow_end_to;
     unsigned cow_start_to = offset_into_cluster(s, guest_offset);
     unsigned cow_end_from = cow_start_to + bytes;
-    unsigned cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
     unsigned nb_clusters = size_to_clusters(s, cow_end_from);
     QCowL2Meta *old_m = *m;
+    QCow2ClusterType type;
+
+    /* Return if there's no COW (all clusters are normal and we keep them) */
+    if (keep_old) {
+        int i;
+        for (i = 0; i < nb_clusters; i++) {
+            l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
+            if (qcow2_get_cluster_type(bs, l2_entry) != QCOW2_CLUSTER_NORMAL) {
+                break;
+            }
+        }
+        if (i == nb_clusters) {
+            return;
+        }
+    }
+
+    /* Get the L2 entry from the first cluster */
+    l2_entry = be64_to_cpu(l2_slice[l2_index]);
+    type = qcow2_get_cluster_type(bs, l2_entry);
+
+    if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
+        cow_start_from = cow_start_to;
+    } else {
+        cow_start_from = 0;
+    }
+
+    /* Get the L2 entry from the last cluster */
+    l2_entry = be64_to_cpu(l2_slice[l2_index + nb_clusters - 1]);
+    type = qcow2_get_cluster_type(bs, l2_entry);
+
+    if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
+        cow_end_to = cow_end_from;
+    } else {
+        cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
+    }
 
     *m = g_malloc0(sizeof(**m));
     **m = (QCowL2Meta) {
@@ -1068,18 +1105,18 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
     QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
 }
 
-/* Returns true if writing to a cluster requires COW */
+/* Returns true if the cluster is unallocated or has refcount > 1 */
 static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry)
 {
     switch (qcow2_get_cluster_type(bs, l2_entry)) {
     case QCOW2_CLUSTER_NORMAL:
+    case QCOW2_CLUSTER_ZERO_ALLOC:
         if (l2_entry & QCOW_OFLAG_COPIED) {
             return false;
         }
     case QCOW2_CLUSTER_UNALLOCATED:
     case QCOW2_CLUSTER_COMPRESSED:
     case QCOW2_CLUSTER_ZERO_PLAIN:
-    case QCOW2_CLUSTER_ZERO_ALLOC:
         return true;
     default:
         abort();
@@ -1087,20 +1124,34 @@ static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry)
 }
 
 /*
- * Returns the number of contiguous clusters that can be used for an allocating
- * write, but require COW to be performed (this includes yet unallocated space,
- * which must copy from the backing file)
+ * Returns the number of contiguous clusters that can be written to
+ * using one single write request, starting from @l2_index.
+ * At most @nb_clusters are checked.
+ *
+ * If @want_cow is true this counts clusters that are either
+ * unallocated, or allocated but with refcount > 1.
+ *
+ * If @want_cow is false this counts clusters that are already
+ * allocated and can be written to using their current locations
+ * (including QCOW2_CLUSTER_ZERO_ALLOC).
  */
 static int count_cow_clusters(BlockDriverState *bs, int nb_clusters,
-    uint64_t *l2_slice, int l2_index)
+                              uint64_t *l2_slice, int l2_index, bool want_cow)
 {
+    BDRVQcow2State *s = bs->opaque;
+    uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index]);
+    uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK;
     int i;
 
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
-        if (!cluster_needs_cow(bs, l2_entry)) {
+        l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
+        if (cluster_needs_cow(bs, l2_entry) != want_cow) {
             break;
         }
+        if (!want_cow && expected_offset != (l2_entry & L2E_OFFSET_MASK)) {
+            break;
+        }
+        expected_offset += s->cluster_size;
     }
 
     assert(i <= nb_clusters);
@@ -1228,18 +1279,17 @@ static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
 
     cluster_offset = be64_to_cpu(l2_slice[l2_index]);
 
-    /* Check how many clusters are already allocated and don't need COW */
-    if (qcow2_get_cluster_type(bs, cluster_offset) == QCOW2_CLUSTER_NORMAL
-        && (cluster_offset & QCOW_OFLAG_COPIED))
-    {
+    if (!cluster_needs_cow(bs, cluster_offset)) {
         /* If a specific host_offset is required, check it */
         bool offset_matches =
             (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
 
         if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) {
-            qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
+            qcow2_signal_corruption(bs, true, -1, -1, "%s cluster offset "
                                     "%#llx unaligned (guest offset: %#" PRIx64
-                                    ")", cluster_offset & L2E_OFFSET_MASK,
+                                    ")", cluster_offset & QCOW_OFLAG_ZERO ?
+                                    "Preallocated zero" : "Data",
+                                    cluster_offset & L2E_OFFSET_MASK,
                                     guest_offset);
             ret = -EIO;
             goto out;
@@ -1252,15 +1302,17 @@ static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
         }
 
         /* We keep all QCOW_OFLAG_COPIED clusters */
-        keep_clusters =
-            count_contiguous_clusters(bs, nb_clusters, s->cluster_size,
-                                      &l2_slice[l2_index],
-                                      QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
+        keep_clusters = count_cow_clusters(bs, nb_clusters, l2_slice,
+                                           l2_index, false);
         assert(keep_clusters <= nb_clusters);
 
         *bytes = MIN(*bytes,
                  keep_clusters * s->cluster_size
                  - offset_into_cluster(s, guest_offset));
+        assert(*bytes != 0);
+
+        calculate_l2_meta(bs, cluster_offset & L2E_OFFSET_MASK, guest_offset,
+                          *bytes, l2_slice, m, true);
 
         ret = 1;
     } else {
@@ -1361,12 +1413,10 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     BDRVQcow2State *s = bs->opaque;
     int l2_index;
     uint64_t *l2_slice;
-    uint64_t entry;
     uint64_t nb_clusters;
     int ret;
-    bool keep_old_clusters = false;
 
-    uint64_t alloc_cluster_offset = INV_OFFSET;
+    uint64_t alloc_cluster_offset;
 
     trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
                              *bytes);
@@ -1392,67 +1442,31 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
         return ret;
     }
 
-    entry = be64_to_cpu(l2_slice[l2_index]);
-    nb_clusters = count_cow_clusters(bs, nb_clusters, l2_slice, l2_index);
+    nb_clusters = count_cow_clusters(bs, nb_clusters, l2_slice, l2_index, true);
 
     /* This function is only called when there were no non-COW clusters, so if
      * we can't find any unallocated or COW clusters either, something is
      * wrong with our code. */
     assert(nb_clusters > 0);
 
-    if (qcow2_get_cluster_type(bs, entry) == QCOW2_CLUSTER_ZERO_ALLOC &&
-        (entry & QCOW_OFLAG_COPIED) &&
-        (*host_offset == INV_OFFSET ||
-         start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK)))
-    {
-        int preallocated_nb_clusters;
-
-        if (offset_into_cluster(s, entry & L2E_OFFSET_MASK)) {
-            qcow2_signal_corruption(bs, true, -1, -1, "Preallocated zero "
-                                    "cluster offset %#llx unaligned (guest "
-                                    "offset: %#" PRIx64 ")",
-                                    entry & L2E_OFFSET_MASK, guest_offset);
-            ret = -EIO;
-            goto fail;
-        }
-
-        /* Try to reuse preallocated zero clusters; contiguous normal clusters
-         * would be fine, too, but count_cow_clusters() above has limited
-         * nb_clusters already to a range of COW clusters */
-        preallocated_nb_clusters =
-            count_contiguous_clusters(bs, nb_clusters, s->cluster_size,
-                                      &l2_slice[l2_index], QCOW_OFLAG_COPIED);
-        assert(preallocated_nb_clusters > 0);
-
-        nb_clusters = preallocated_nb_clusters;
-        alloc_cluster_offset = entry & L2E_OFFSET_MASK;
-
-        /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2()
-         * should not free them. */
-        keep_old_clusters = true;
+    /* Allocate, if necessary at a given offset in the image file */
+    alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET :
+        start_of_cluster(s, *host_offset);
+    ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
+                                  &nb_clusters);
+    if (ret < 0) {
+        goto out;
     }
 
-    qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
-
-    if (alloc_cluster_offset == INV_OFFSET) {
-        /* Allocate, if necessary at a given offset in the image file */
-        alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET :
-                               start_of_cluster(s, *host_offset);
-        ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
-                                      &nb_clusters);
-        if (ret < 0) {
-            goto fail;
-        }
-
-        /* Can't extend contiguous allocation */
-        if (nb_clusters == 0) {
-            *bytes = 0;
-            return 0;
-        }
-
-        assert(alloc_cluster_offset != INV_OFFSET);
+    /* Can't extend contiguous allocation */
+    if (nb_clusters == 0) {
+        *bytes = 0;
+        ret = 0;
+        goto out;
     }
 
+    assert(alloc_cluster_offset != INV_OFFSET);
+
     /*
      * Save info needed for meta data update.
      *
@@ -1475,13 +1489,14 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
     assert(*bytes != 0);
 
-    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
-                      m, keep_old_clusters);
+    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes, l2_slice,
+                      m, false);
 
-    return 1;
+    ret = 1;
 
-fail:
-    if (*m && (*m)->nb_clusters > 0) {
+out:
+    qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
+    if (ret < 0 && *m && (*m)->nb_clusters > 0) {
         QLIST_REMOVE(*m, next_in_flight);
     }
     return ret;
Signature · 1 line
2.20.1
#6 lore raw

RFC PATCH v2 2/26 qcow2: Split cluster_needs_cow() out of count_cow_clusters()

In reply to Alberto Garcia

We are going to need it in other places.

---
 block/qcow2-cluster.c | 34 +++++++++++++++++++---------------
 1 file changed, 19 insertions(+), 15 deletions(-)
block/qcow2-cluster.c +19 -15
index 6c1dcdc781..aa1010a515 100644
@@ -1068,6 +1068,24 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
     QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
 }
 
+/* Returns true if writing to a cluster requires COW */
+static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry)
+{
+    switch (qcow2_get_cluster_type(bs, l2_entry)) {
+    case QCOW2_CLUSTER_NORMAL:
+        if (l2_entry & QCOW_OFLAG_COPIED) {
+            return false;
+        }
+    case QCOW2_CLUSTER_UNALLOCATED:
+    case QCOW2_CLUSTER_COMPRESSED:
+    case QCOW2_CLUSTER_ZERO_PLAIN:
+    case QCOW2_CLUSTER_ZERO_ALLOC:
+        return true;
+    default:
+        abort();
+    }
+}
+
 /*
  * Returns the number of contiguous clusters that can be used for an allocating
  * write, but require COW to be performed (this includes yet unallocated space,
@@ -1080,25 +1098,11 @@ static int count_cow_clusters(BlockDriverState *bs, int nb_clusters,
 
     for (i = 0; i < nb_clusters; i++) {
         uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
-        QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry);
-
-        switch(cluster_type) {
-        case QCOW2_CLUSTER_NORMAL:
-            if (l2_entry & QCOW_OFLAG_COPIED) {
-                goto out;
-            }
+        if (!cluster_needs_cow(bs, l2_entry)) {
             break;
-        case QCOW2_CLUSTER_UNALLOCATED:
-        case QCOW2_CLUSTER_COMPRESSED:
-        case QCOW2_CLUSTER_ZERO_PLAIN:
-        case QCOW2_CLUSTER_ZERO_ALLOC:
-            break;
-        default:
-            abort();
         }
     }
 
-out:
     assert(i <= nb_clusters);
     return i;
 }
Signature · 1 line
2.20.1
#63 lore raw

RFC PATCH v2 26/26 iotests: Add tests for qcow2 images with extended L2 entries

In reply to Alberto Garcia
---
 tests/qemu-iotests/271     | 235 +++++++++++++++++++++++++++++++++++++
 tests/qemu-iotests/271.out | 183 +++++++++++++++++++++++++++++
 tests/qemu-iotests/group   |   1 +
 3 files changed, 419 insertions(+)
 create mode 100755 tests/qemu-iotests/271
 create mode 100644 tests/qemu-iotests/271.out
tests/qemu-iotests/271 new file +235 -0
new file mode 100755
index 0000000000..c49433cdc9
@@ -0,0 +1,235 @@
+#!/bin/bash
+#
+# Test qcow2 images with extended L2 entries
+#
+# Copyright (C) 2019 Igalia, S.L.
+# Author: Alberto Garcia <[email protected]>
+#
+# This program is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 2 of the License, or
+# (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program.  If not, see <http://www.gnu.org/licenses/>.
+#
+
+# creator
[email protected]
+
+seq="$(basename $0)"
+echo "QA output created by $seq"
+
+here="$PWD"
+status=1	# failure is the default!
+
+_cleanup()
+{
+	_cleanup_test_img
+        rm -f "$TEST_IMG.raw"
+        rm -f "$TEST_IMG.backing"
+}
+trap "_cleanup; exit \$status" 0 1 2 3 15
+
+# get standard environment, filters and checks
+. ./common.rc
+. ./common.filter
+
+_supported_fmt qcow2
+_supported_proto file nfs
+_supported_os Linux
+
+IMGOPTS="extended_l2=on"
+l2_offset=262144 # 0x40000
+
+_verify_img()
+{
+    $QEMU_IMG compare "$TEST_IMG" "$TEST_IMG.raw" | grep -v 'Images are identical'
+    $QEMU_IMG check "$TEST_IMG" | _filter_qemu_img_check | \
+        grep -v 'No errors were found on the image'
+}
+
+_read_l2_entry()
+{
+    entry_no=$1
+    nentries=$2
+    offset=$(($l2_offset + $entry_no * 16))
+    length=$((nentries * 16))
+    $QEMU_IO -f raw -c "read -v $offset $length" "$TEST_IMG" | _filter_qemu_io | head -n -2
+}
+
+_test_write()
+{
+    cmd="$1"
+    l2_entry_idx="$2"
+    l2_entry_num="$3"
+    raw_cmd=`echo $1 | sed s/-c//` # Raw images don't support -c
+    echo "$cmd"
+    $QEMU_IO -c "$cmd" "$TEST_IMG" | _filter_qemu_io
+    $QEMU_IO -c "$raw_cmd" -f raw "$TEST_IMG.raw" | _filter_qemu_io
+    _verify_img
+    if [ -n "$l2_entry_idx" ]; then
+        _read_l2_entry "$l2_entry_idx" "$l2_entry_num"
+    fi
+}
+
+_reset_img()
+{
+    $QEMU_IMG create -f raw "$TEST_IMG.raw" 1M | _filter_img_create
+    if [ "$use_backing_file" = "yes" ]; then
+        $QEMU_IMG create -f raw "$TEST_IMG.backing" 1M | _filter_img_create
+        $QEMU_IO -c 'write -q -P 0xFF 0 1M' -f raw "$TEST_IMG.backing" | _filter_qemu_io
+        $QEMU_IO -c 'write -q -P 0xFF 0 1M' -f raw "$TEST_IMG.raw" | _filter_qemu_io
+        _make_test_img -b "$TEST_IMG.backing" 1M
+    else
+        _make_test_img 1M
+    fi
+}
+
+# Test that writing to an image with subclusters produces the expected
+# results, in images with and without backing files
+for use_backing_file in yes no; do
+    echo
+    echo "### Standard write tests (backing file: $use_backing_file) ###"
+    echo
+    _reset_img
+    ### Write subcluster #0 (beginning of subcluster) ###
+    _test_write 'write -q -P 1 0 1k' 0 1
+
+    ### Write subcluster #1 (middle of subcluster) ###
+    _test_write 'write -q -P 2 3k 512' 0 1
+
+    ### Write subcluster #2 (end of subcluster) ###
+    _test_write 'write -q -P 3 5k 1k' 0 1
+
+    ### Write subcluster #3 (full subcluster) ###
+    _test_write 'write -q -P 4 6k 2k' 0 1
+
+    ### Write subclusters #4-6 (full subclusters) ###
+    _test_write 'write -q -P 5 8k 6k' 0 1
+
+    ### Write subclusters #7-9 (partial subclusters) ###
+    _test_write 'write -q -P 6 15k 4k' 0 1
+
+    ### Write subcluster #16 (partial subcluster) ###
+    _test_write 'write -q -P 7 32k 1k' 0 2
+
+    ### Write subcluster #31-#34 (cluster overlap) ###
+    _test_write 'write -q -P 8 63k 4k' 0 2
+
+    ### Zero subcluster #1 (TODO: use the "all zeros" bit)
+    _test_write 'write -q -z 2k 2k' 0 1
+
+    ### Zero cluster #0
+    _test_write 'write -q -z 0 64k' 0 1
+
+    ### Fill cluster #0 with data
+    _test_write 'write -q -P 9 0 64k' 0 1
+
+    ### Zero and unmap half of cluster #0 (this won't unmap it)
+    _test_write 'write -q -z -u 0 32k' 0 1
+
+    ### Zero and unmap cluster #0
+    _test_write 'write -q -z -u 0 64k' 0 1
+
+    ### Write subcluster #2 (middle of subcluster)
+    _test_write 'write -q -P 10 3k 512' 0 1
+
+    ### Fill cluster #0 with data
+    _test_write 'write -q -P 11 0 64k' 0 1
+
+    ### Discard cluster #0
+    _test_write 'discard -q 0 64k' 0 1
+
+    ### Write compressed data to cluster #0
+    _test_write 'write -q -c -P 12 0 64k' 0 1
+
+    ### Write subcluster #2 (middle of subcluster)
+    _test_write 'write -q -P 13 3k 512' 0 1
+done
+
+# Test that corrupted L2 entries are detected in both read and write
+# operations
+for corruption_test_cmd in read write; do
+    echo
+    echo "### Corrupted L2 entries - $corruption_test_cmd test (allocated) ###"
+    echo
+    echo "# 'cluster is zero' bit set on the standard cluster descriptor"
+    echo
+    _make_test_img 1M
+    $QEMU_IO -c 'write -q 0 2k' "$TEST_IMG"
+    poke_file "$TEST_IMG" $(($l2_offset+7)) "\x01"
+    $QEMU_IO -c "$corruption_test_cmd 0 1k" "$TEST_IMG"
+
+    echo
+    echo "# Both 'subcluster is zero' and 'subcluster is allocated' bits set"
+    echo
+    _make_test_img 1M
+    $QEMU_IO -c 'write -q 0 2k' "$TEST_IMG"
+    poke_file "$TEST_IMG" $(($l2_offset+8)) "\x80"
+    $QEMU_IO -c "$corruption_test_cmd 0 1k" "$TEST_IMG"
+
+    echo
+    echo "### Corrupted L2 entries - $corruption_test_cmd test (unallocated) ###"
+    echo
+    echo "# 'cluster is zero' bit set on the standard cluster descriptor"
+    echo
+    _make_test_img 1M
+    # Write to cluster #4 in order to initialize the L2 table
+    $QEMU_IO -c 'write -q 256k 1k' "$TEST_IMG"
+    poke_file "$TEST_IMG" $(($l2_offset+7)) "\x01"
+    $QEMU_IO -c "$corruption_test_cmd 0 1k" "$TEST_IMG"
+
+    echo
+    echo "# 'subcluster is allocated' bit set"
+    echo
+    _make_test_img 1M
+    # Write to cluster #4 in order to initialize the L2 table
+    $QEMU_IO -c 'write -q 256k 1k' "$TEST_IMG"
+    poke_file "$TEST_IMG" $(($l2_offset+12)) "\x80"
+    _read_l2_entry 0 1
+    $QEMU_IO -c "$corruption_test_cmd 0 1k" "$TEST_IMG"
+
+    echo
+    echo "# Both 'subcluster is zero' and 'subcluster is allocated' bits set"
+    echo
+    _make_test_img 1M
+    # Write to cluster #4 in order to initialize the L2 table
+    $QEMU_IO -c 'write -q 256k 1k' "$TEST_IMG"
+    poke_file "$TEST_IMG" $(($l2_offset+8)) "\x80\x00\x00\x00\x80"
+    $QEMU_IO -c "$corruption_test_cmd 0 1k" "$TEST_IMG"
+
+    echo
+    echo "### Compressed cluster with subcluster bitmap != 0 - $corruption_test_cmd test ###"
+    echo
+    _make_test_img 1M
+    $QEMU_IO -c 'write -q -c 0 64k' "$TEST_IMG"
+    poke_file "$TEST_IMG" $(($l2_offset+8)) "\x01"
+    $QEMU_IO -c "$corruption_test_cmd 0 1k" "$TEST_IMG"
+done
+
+echo
+echo "### Image creation options ###"
+echo
+echo "# cluster_size < 16k"
+IMGOPTS="extended_l2=on,cluster_size=8k" _make_test_img 1M
+
+echo "# backing file and preallocation=metadata"
+IMGOPTS="extended_l2=on,preallocation=metadata" _make_test_img -b "$TEST_IMG.backing" 1M
+
+echo "# backing file and preallocation=falloc"
+IMGOPTS="extended_l2=on,preallocation=falloc" _make_test_img -b "$TEST_IMG.backing" 1M
+
+echo "# backing file and preallocation=full"
+IMGOPTS="extended_l2=on,preallocation=full" _make_test_img -b "$TEST_IMG.backing" 1M
+
+# success, all done
+echo "*** done"
+rm -f $seq.full
+status=0
+
tests/qemu-iotests/271.out new file +183 -0

183 lines not on this page · raw message

tests/qemu-iotests/group +1 -0
index af322af756..945e6d68cc 100644
@@ -282,3 +282,4 @@
 267 rw auto quick snapshot
 268 rw auto quick
 270 rw backing quick
+271 rw auto
Signature · 1 line
2.20.1
#58 lore raw

RFC PATCH v2 24/26 qcow2: Add the 'extended_l2' option and the QCOW2_INCOMPAT_EXTL2 bit

In reply to Alberto Garcia

Now that the implementation of subclusters is complete we can finally add the necessary options to create and read images with this feature, which we call "extended L2 entries".

---
 block/qcow2.c                    |  46 ++++++++++++++
 block/qcow2.h                    |   8 ++-
 include/block/block_int.h        |   1 +
 qapi/block-core.json             |   6 ++
 tests/qemu-iotests/031.out       |   8 +--
 tests/qemu-iotests/036.out       |   4 +-
 tests/qemu-iotests/049.out       | 102 +++++++++++++++----------------
 tests/qemu-iotests/060.out       |   1 +
 tests/qemu-iotests/061.out       |  20 +++---
 tests/qemu-iotests/065           |  18 ++++--
 tests/qemu-iotests/082.out       |  48 ++++++++++++---
 tests/qemu-iotests/085.out       |  38 ++++++------
 tests/qemu-iotests/144.out       |   4 +-
 tests/qemu-iotests/182.out       |   2 +-
 tests/qemu-iotests/185.out       |   8 +--
 tests/qemu-iotests/198.out       |   2 +
 tests/qemu-iotests/206.out       |   4 ++
 tests/qemu-iotests/242.out       |   5 ++
 tests/qemu-iotests/255.out       |   8 +--
 tests/qemu-iotests/common.filter |   1 +
 20 files changed, 224 insertions(+), 110 deletions(-)
block/qcow2.c +46 -0
index 537569ce88..b1fa7ab5da 100644
@@ -1347,6 +1347,12 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
     s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
     s->subcluster_bits = ctz32(s->subcluster_size);
 
+    if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
+        error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
+        ret = -EINVAL;
+        goto fail;
+    }
+
     /* Check support for various header values */
     if (header.refcount_order > 6) {
         error_setg(errp, "Reference count entry width too large; may not "
@@ -2806,6 +2812,11 @@ int qcow2_update_header(BlockDriverState *bs)
                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
                 .name = "lazy refcounts",
             },
+            {
+                .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
+                .bit  = QCOW2_INCOMPAT_EXTL2_BITNR,
+                .name = "extended L2 entries",
+            },
         };
 
         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
@@ -3271,6 +3282,27 @@ qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
         goto out;
     }
 
+    if (!qcow2_opts->has_extended_l2) {
+        qcow2_opts->extended_l2 = false;
+    }
+    if (qcow2_opts->extended_l2) {
+        unsigned min_cluster_size =
+            (1 << MIN_CLUSTER_BITS) * QCOW_MAX_SUBCLUSTERS_PER_CLUSTER;
+        if (version < 3) {
+            error_setg(errp, "Extended L2 entries are only supported with "
+                       "compatibility level 1.1 and above (use version=v3 or "
+                       "greater)");
+            ret = -EINVAL;
+            goto out;
+        }
+        if (cluster_size < min_cluster_size) {
+            error_setg(errp, "Extended L2 entries are only supported with "
+                       "cluster sizes of at least %u bytes", min_cluster_size);
+            ret = -EINVAL;
+            goto out;
+        }
+    }
+
     if (!qcow2_opts->has_preallocation) {
         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
     }
@@ -3392,6 +3424,11 @@ qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
             cpu_to_be64(QCOW2_AUTOCLEAR_DATA_FILE_RAW);
     }
 
+    if (qcow2_opts->extended_l2) {
+        header->incompatible_features |=
+            cpu_to_be64(QCOW2_INCOMPAT_EXTL2);
+    }
+
     ret = blk_pwrite(blk, 0, header, cluster_size, 0);
     g_free(header);
     if (ret < 0) {
@@ -3569,6 +3606,7 @@ static int coroutine_fn qcow2_co_create_opts(const char *filename, QemuOpts *opt
         { BLOCK_OPT_BACKING_FMT,        "backing-fmt" },
         { BLOCK_OPT_CLUSTER_SIZE,       "cluster-size" },
         { BLOCK_OPT_LAZY_REFCOUNTS,     "lazy-refcounts" },
+        { BLOCK_OPT_EXTL2,              "extended-l2" },
         { BLOCK_OPT_REFCOUNT_BITS,      "refcount-bits" },
         { BLOCK_OPT_ENCRYPT,            BLOCK_OPT_ENCRYPT_FORMAT },
         { BLOCK_OPT_COMPAT_LEVEL,       "version" },
@@ -4772,6 +4810,8 @@ static ImageInfoSpecific *qcow2_get_specific_info(BlockDriverState *bs,
             .corrupt            = s->incompatible_features &
                                   QCOW2_INCOMPAT_CORRUPT,
             .has_corrupt        = true,
+            .has_extended_l2    = true,
+            .extended_l2        = has_subclusters(s),
             .refcount_bits      = s->refcount_bits,
             .has_bitmaps        = !!bitmaps,
             .bitmaps            = bitmaps,
@@ -5365,6 +5405,12 @@ static QemuOptsList qcow2_create_opts = {
             .help = "Postpone refcount updates",
             .def_value_str = "off"
         },
+        {
+            .name = BLOCK_OPT_EXTL2,
+            .type = QEMU_OPT_BOOL,
+            .help = "Extended L2 tables",
+            .def_value_str = "off"
+        },
         {
             .name = BLOCK_OPT_REFCOUNT_BITS,
             .type = QEMU_OPT_NUMBER,
block/qcow2.h +5 -3
index 23a2532ff2..3832e8ccc9 100644
@@ -220,13 +220,16 @@ enum {
     QCOW2_INCOMPAT_DIRTY_BITNR      = 0,
     QCOW2_INCOMPAT_CORRUPT_BITNR    = 1,
     QCOW2_INCOMPAT_DATA_FILE_BITNR  = 2,
+    QCOW2_INCOMPAT_EXTL2_BITNR      = 3,
     QCOW2_INCOMPAT_DIRTY            = 1 << QCOW2_INCOMPAT_DIRTY_BITNR,
     QCOW2_INCOMPAT_CORRUPT          = 1 << QCOW2_INCOMPAT_CORRUPT_BITNR,
     QCOW2_INCOMPAT_DATA_FILE        = 1 << QCOW2_INCOMPAT_DATA_FILE_BITNR,
+    QCOW2_INCOMPAT_EXTL2            = 1 << QCOW2_INCOMPAT_EXTL2_BITNR,
 
     QCOW2_INCOMPAT_MASK             = QCOW2_INCOMPAT_DIRTY
                                     | QCOW2_INCOMPAT_CORRUPT
-                                    | QCOW2_INCOMPAT_DATA_FILE,
+                                    | QCOW2_INCOMPAT_DATA_FILE
+                                    | QCOW2_INCOMPAT_EXTL2,
 };
 
 /* Compatible feature bits */
@@ -506,8 +509,7 @@ typedef enum QCow2MetadataOverlap {
 
 static inline bool has_subclusters(BDRVQcow2State *s)
 {
-    /* FIXME: Return false until this feature is complete */
-    return false;
+    return s->incompatible_features & QCOW2_INCOMPAT_EXTL2;
 }
 
 static inline size_t l2_entry_size(BDRVQcow2State *s)
include/block/block_int.h +1 -0
index ca4ccac4c1..bf11764155 100644
@@ -57,6 +57,7 @@
 #define BLOCK_OPT_REFCOUNT_BITS     "refcount_bits"
 #define BLOCK_OPT_DATA_FILE         "data_file"
 #define BLOCK_OPT_DATA_FILE_RAW     "data_file_raw"
+#define BLOCK_OPT_EXTL2             "extended_l2"
 
 #define BLOCK_PROBE_BUF_SIZE        512
 
qapi/block-core.json +6 -0
index aa97ee2641..6e4c81ae1c 100644
@@ -66,6 +66,8 @@
 #                 standalone (read-only) raw image without looking at qcow2
 #                 metadata (since: 4.0)
 #
+# @extended-l2: true if the image has extended L2 entries (since 4.2)
+#
 # @lazy-refcounts: on or off; only valid for compat >= 1.1
 #
 # @corrupt: true if the image has been marked corrupt; only valid for
@@ -85,6 +87,7 @@
       'compat': 'str',
       '*data-file': 'str',
       '*data-file-raw': 'bool',
+      '*extended-l2': 'bool',
       '*lazy-refcounts': 'bool',
       '*corrupt': 'bool',
       'refcount-bits': 'int',
@@ -4372,6 +4375,8 @@
 # @data-file-raw    True if the external data file must stay valid as a
 #                   standalone (read-only) raw image without looking at qcow2
 #                   metadata (default: false; since: 4.0)
+# @extended-l2      True if the image has extended L2 entries
+#                   (default: false; since 4.2)
 # @size             Size of the virtual disk in bytes
 # @version          Compatibility level (default: v3)
 # @backing-file     File name of the backing file if a backing file
@@ -4390,6 +4395,7 @@
   'data': { 'file':             'BlockdevRef',
             '*data-file':       'BlockdevRef',
             '*data-file-raw':   'bool',
+            '*extended-l2':     'bool',
             'size':             'size',
             '*version':         'BlockdevQcow2Version',
             '*backing-file':    'str',
tests/qemu-iotests/031.out +4 -4
index 68a74d03b9..614950be56 100644
@@ -117,7 +117,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 Header extension:
@@ -150,7 +150,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 Header extension:
@@ -164,7 +164,7 @@ No errors were found on the image.
 
 magic                     0x514649fb
 version                   3
-backing_file_offset       0x178
+backing_file_offset       0x1a8
 backing_file_size         0x17
 cluster_bits              16
 size                      67108864
@@ -188,7 +188,7 @@ data                      'host_device'
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 Header extension:
tests/qemu-iotests/036.out +2 -2
index e489b44386..c7e6512b43 100644
@@ -58,7 +58,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 
@@ -86,7 +86,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 *** done
tests/qemu-iotests/049.out +51 -51

225 lines not on this page · raw message

tests/qemu-iotests/060.out +1 -0
index 0f6b0658a1..1f6ae50027 100644
@@ -20,6 +20,7 @@ Format specific information:
     lazy refcounts: false
     refcount bits: 16
     corrupt: true
+    extended l2: false
 qemu-io: can't open device TEST_DIR/t.IMGFMT: IMGFMT: Image is corrupt; cannot be opened read/write
 no file open, try 'help open'
 read 512/512 bytes at offset 0
tests/qemu-iotests/061.out +13 -7
index d6a7c2af95..96a6933554 100644
@@ -26,7 +26,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 magic                     0x514649fb
@@ -84,7 +84,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 magic                     0x514649fb
@@ -140,7 +140,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 ERROR cluster 5 refcount=0 reference=1
@@ -195,7 +195,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 magic                     0x514649fb
@@ -264,7 +264,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 read 65536/65536 bytes at offset 44040192
@@ -298,7 +298,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 ERROR cluster 5 refcount=0 reference=1
@@ -327,7 +327,7 @@ header_length             104
 
 Header extension:
 magic                     0x6803f857
-length                    192
+length                    240
 data                      <binary>
 
 read 131072/131072 bytes at offset 0
@@ -496,6 +496,7 @@ Format specific information:
     data file: TEST_DIR/t.IMGFMT.data
     data file raw: false
     corrupt: false
+    extended l2: false
 No errors were found on the image.
 
 === Try changing the external data file ===
@@ -516,6 +517,7 @@ Format specific information:
     data file: foo
     data file raw: false
     corrupt: false
+    extended l2: false
 
 qemu-img: Could not open 'TEST_DIR/t.IMGFMT': 'data-file' is required for this image
 image: TEST_DIR/t.IMGFMT
@@ -528,6 +530,7 @@ Format specific information:
     refcount bits: 16
     data file raw: false
     corrupt: false
+    extended l2: false
 
 === Clearing and setting data-file-raw ===
 
@@ -543,6 +546,7 @@ Format specific information:
     data file: TEST_DIR/t.IMGFMT.data
     data file raw: true
     corrupt: false
+    extended l2: false
 No errors were found on the image.
 image: TEST_DIR/t.IMGFMT
 file format: IMGFMT
@@ -555,6 +559,7 @@ Format specific information:
     data file: TEST_DIR/t.IMGFMT.data
     data file raw: false
     corrupt: false
+    extended l2: false
 No errors were found on the image.
 qemu-img: data-file-raw cannot be set on existing images
 image: TEST_DIR/t.IMGFMT
@@ -568,5 +573,6 @@ Format specific information:
     data file: TEST_DIR/t.IMGFMT.data
     data file raw: false
     corrupt: false
+    extended l2: false
 No errors were found on the image.
 *** done
tests/qemu-iotests/065 +12 -6
index 5b21eb96bd..7d3a137434 100755
@@ -95,17 +95,21 @@ class TestQCow3NotLazy(TestQemuImgInfo):
     '''Testing a qcow2 version 3 image with lazy refcounts disabled'''
     img_options = 'compat=1.1,lazy_refcounts=off'
     json_compare = { 'compat': '1.1', 'lazy-refcounts': False,
-                     'refcount-bits': 16, 'corrupt': False }
+                     'refcount-bits': 16, 'corrupt': False,
+                     'extended-l2': False }
     human_compare = [ 'compat: 1.1', 'lazy refcounts: false',
-                      'refcount bits: 16', 'corrupt: false' ]
+                      'refcount bits: 16', 'corrupt: false',
+                      'extended l2: false' ]
 
 class TestQCow3Lazy(TestQemuImgInfo):
     '''Testing a qcow2 version 3 image with lazy refcounts enabled'''
     img_options = 'compat=1.1,lazy_refcounts=on'
     json_compare = { 'compat': '1.1', 'lazy-refcounts': True,
-                     'refcount-bits': 16, 'corrupt': False }
+                     'refcount-bits': 16, 'corrupt': False,
+                     'extended-l2': False }
     human_compare = [ 'compat: 1.1', 'lazy refcounts: true',
-                      'refcount bits: 16', 'corrupt: false' ]
+                      'refcount bits: 16', 'corrupt: false',
+                      'extended l2: false' ]
 
 class TestQCow3NotLazyQMP(TestQMP):
     '''Testing a qcow2 version 3 image with lazy refcounts disabled, opening
@@ -113,7 +117,8 @@ class TestQCow3NotLazyQMP(TestQMP):
     img_options = 'compat=1.1,lazy_refcounts=off'
     qemu_options = 'lazy-refcounts=on'
     compare = { 'compat': '1.1', 'lazy-refcounts': False,
-                'refcount-bits': 16, 'corrupt': False }
+                'refcount-bits': 16, 'corrupt': False,
+                'extended-l2': False }
 
 
 class TestQCow3LazyQMP(TestQMP):
@@ -122,7 +127,8 @@ class TestQCow3LazyQMP(TestQMP):
     img_options = 'compat=1.1,lazy_refcounts=on'
     qemu_options = 'lazy-refcounts=off'
     compare = { 'compat': '1.1', 'lazy-refcounts': True,
-                'refcount-bits': 16, 'corrupt': False }
+                'refcount-bits': 16, 'corrupt': False,
+                'extended-l2': False }
 
 TestImageInfoSpecific = None
 TestQemuImgInfo = None
tests/qemu-iotests/082.out +41 -7

282 lines not on this page · raw message

tests/qemu-iotests/085.out +19 -19
index 2a5f256cd3..62e7fd82a7 100644
@@ -11,7 +11,7 @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=134217728
 
 === Create a single snapshot on virtio0 ===
 
-Formatting 'TEST_DIR/1-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/t.qcow2.1 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/1-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/t.qcow2.1 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
 
 === Invalid command - missing device and nodename ===
@@ -25,32 +25,32 @@ Formatting 'TEST_DIR/1-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file
 
 === Create several transactional group snapshots ===
 
-Formatting 'TEST_DIR/2-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/1-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/2-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/t.qcow2.2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/2-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/1-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/2-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/t.qcow2.2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/3-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/2-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/3-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/2-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/3-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/2-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/3-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/2-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/4-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/3-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/4-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/3-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/4-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/3-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/4-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/3-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/5-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/4-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/5-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/4-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/5-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/4-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/5-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/4-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/6-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/5-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/6-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/5-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/6-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/5-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/6-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/5-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/7-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/6-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/7-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/6-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/7-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/6-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/7-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/6-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/8-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/7-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/8-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/7-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/8-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/7-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/8-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/7-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/9-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/8-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/9-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/8-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/9-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/8-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/9-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/8-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
-Formatting 'TEST_DIR/10-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/9-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
-Formatting 'TEST_DIR/10-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/9-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/10-snapshot-v0.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/9-snapshot-v0.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
+Formatting 'TEST_DIR/10-snapshot-v1.qcow2', fmt=qcow2 size=134217728 backing_file=TEST_DIR/9-snapshot-v1.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
 
 === Create a couple of snapshots using blockdev-snapshot ===
tests/qemu-iotests/144.out +2 -2
index a9a8216bea..7bf05253cc 100644
@@ -7,7 +7,7 @@ Formatting 'TEST_DIR/t.IMGFMT', fmt=IMGFMT size=536870912
 === Performing Live Snapshot 1 ===
 
 {"return": {}}
-Formatting 'TEST_DIR/tmp.qcow2', fmt=qcow2 size=536870912 backing_file=TEST_DIR/t.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/tmp.qcow2', fmt=qcow2 size=536870912 backing_file=TEST_DIR/t.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
 
 === Performing block-commit on active layer ===
@@ -26,6 +26,6 @@ Formatting 'TEST_DIR/tmp.qcow2', fmt=qcow2 size=536870912 backing_file=TEST_DIR/
 
 === Performing Live Snapshot 2 ===
 
-Formatting 'TEST_DIR/tmp2.qcow2', fmt=qcow2 size=536870912 backing_file=TEST_DIR/t.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/tmp2.qcow2', fmt=qcow2 size=536870912 backing_file=TEST_DIR/t.qcow2 backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
 *** done
tests/qemu-iotests/182.out +1 -1
index ffef23e32b..78e2d51761 100644
@@ -10,7 +10,7 @@ Is another process using the image [TEST_DIR/t.qcow2]?
 
 {"return": {}}
 {"return": {}}
-Formatting 'TEST_DIR/t.qcow2.overlay', fmt=qcow2 size=197120 backing_file=TEST_DIR/t.qcow2 backing_fmt=file cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/t.qcow2.overlay', fmt=qcow2 size=197120 backing_file=TEST_DIR/t.qcow2 backing_fmt=file cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
 {"return": {}}
 {"return": {}}
tests/qemu-iotests/185.out +4 -4
index ddfbf3c765..eaab057fed 100644
@@ -7,12 +7,12 @@ Formatting 'TEST_DIR/t.IMGFMT.base', fmt=IMGFMT size=67108864
 
 === Creating backing chain ===
 
-Formatting 'TEST_DIR/t.qcow2.mid', fmt=qcow2 size=67108864 backing_file=TEST_DIR/t.qcow2.base backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/t.qcow2.mid', fmt=qcow2 size=67108864 backing_file=TEST_DIR/t.qcow2.base backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
 wrote 4194304/4194304 bytes at offset 0
 4 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
 {"return": ""}
-Formatting 'TEST_DIR/t.qcow2', fmt=qcow2 size=67108864 backing_file=TEST_DIR/t.qcow2.mid backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/t.qcow2', fmt=qcow2 size=67108864 backing_file=TEST_DIR/t.qcow2.mid backing_fmt=qcow2 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"return": {}}
 
 === Start commit job and exit qemu ===
@@ -37,7 +37,7 @@ Formatting 'TEST_DIR/t.qcow2', fmt=qcow2 size=67108864 backing_file=TEST_DIR/t.q
 === Start mirror job and exit qemu ===
 
 {"return": {}}
-Formatting 'TEST_DIR/t.qcow2.copy', fmt=qcow2 size=67108864 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/t.qcow2.copy', fmt=qcow2 size=67108864 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"timestamp": {"seconds":  TIMESTAMP, "microseconds":  TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "created", "id": "disk"}}
 {"timestamp": {"seconds":  TIMESTAMP, "microseconds":  TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "running", "id": "disk"}}
 {"return": {}}
@@ -48,7 +48,7 @@ Formatting 'TEST_DIR/t.qcow2.copy', fmt=qcow2 size=67108864 cluster_size=65536 l
 === Start backup job and exit qemu ===
 
 {"return": {}}
-Formatting 'TEST_DIR/t.qcow2.copy', fmt=qcow2 size=67108864 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/t.qcow2.copy', fmt=qcow2 size=67108864 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 {"timestamp": {"seconds":  TIMESTAMP, "microseconds":  TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "created", "id": "disk"}}
 {"timestamp": {"seconds":  TIMESTAMP, "microseconds":  TIMESTAMP}, "event": "JOB_STATUS_CHANGE", "data": {"status": "running", "id": "disk"}}
 {"return": {}}
tests/qemu-iotests/198.out +2 -0
index e86b175e39..e46dccdb08 100644
@@ -72,6 +72,7 @@ Format specific information:
                 key offset: 1810432
         payload offset: 2068480
         master key iters: 1024
+    extended l2: false
 
 == checking image layer ==
 image: json:{"encrypt.key-secret": "sec1", "driver": "IMGFMT", "file": {"driver": "file", "filename": "TEST_DIR/t.IMGFMT"}}
@@ -115,4 +116,5 @@ Format specific information:
                 key offset: 1810432
         payload offset: 2068480
         master key iters: 1024
+    extended l2: false
 *** done
tests/qemu-iotests/206.out +4 -0
index 61e7241e0b..d2efc0394a 100644
@@ -21,6 +21,7 @@ Format specific information:
     lazy refcounts: false
     refcount bits: 16
     corrupt: false
+    extended l2: false
 
 === Successful image creation (inline blockdev-add, explicit defaults) ===
 
@@ -43,6 +44,7 @@ Format specific information:
     lazy refcounts: false
     refcount bits: 16
     corrupt: false
+    extended l2: false
 
 === Successful image creation (v3 non-default options) ===
 
@@ -65,6 +67,7 @@ Format specific information:
     lazy refcounts: true
     refcount bits: 1
     corrupt: false
+    extended l2: false
 
 === Successful image creation (v2 non-default options) ===
 
@@ -141,6 +144,7 @@ Format specific information:
         payload offset: 528384
         master key iters: XXX
     corrupt: false
+    extended l2: false
 
 === Invalid BlockdevRef ===
 
tests/qemu-iotests/242.out +5 -0
index 7ac8404d11..0d32dd9148 100644
@@ -15,6 +15,7 @@ Format specific information:
     lazy refcounts: false
     refcount bits: 16
     corrupt: false
+    extended l2: false
 
 No bitmap in JSON format output
 
@@ -40,6 +41,7 @@ Format specific information:
             granularity: 32768
     refcount bits: 16
     corrupt: false
+    extended l2: false
 
 The same bitmaps in JSON format:
 [
@@ -77,6 +79,7 @@ Format specific information:
             granularity: 65536
     refcount bits: 16
     corrupt: false
+    extended l2: false
 
 The same bitmaps in JSON format:
 [
@@ -119,6 +122,7 @@ Format specific information:
             granularity: 65536
     refcount bits: 16
     corrupt: false
+    extended l2: false
 
 The same bitmaps in JSON format:
 [
@@ -162,5 +166,6 @@ Format specific information:
             granularity: 16384
     refcount bits: 16
     corrupt: false
+    extended l2: false
 
 Test complete
tests/qemu-iotests/255.out +4 -4
index 348909fdef..4e1b917a0f 100644
@@ -3,9 +3,9 @@ Finishing a commit job with background reads
 
 === Create backing chain and start VM ===
 
-Formatting 'TEST_DIR/PID-t.qcow2.mid', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/PID-t.qcow2.mid', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 
-Formatting 'TEST_DIR/PID-t.qcow2', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/PID-t.qcow2', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 
 === Start background read requests ===
 
@@ -23,9 +23,9 @@ Closing the VM while a job is being cancelled
 
 === Create images and start VM ===
 
-Formatting 'TEST_DIR/PID-src.qcow2', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/PID-src.qcow2', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 
-Formatting 'TEST_DIR/PID-dst.qcow2', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off refcount_bits=16
+Formatting 'TEST_DIR/PID-dst.qcow2', fmt=qcow2 size=134217728 cluster_size=65536 lazy_refcounts=off extended_l2=off refcount_bits=16
 
 wrote 1048576/1048576 bytes at offset 0
 1 MiB, X ops; XX:XX:XX.X (XXX YYY/sec and XXX ops/sec)
tests/qemu-iotests/common.filter +1 -0
index 9f418b4881..45c0fe746b 100644
@@ -137,6 +137,7 @@ _filter_img_create()
         -e "s# adapter_type=[^ ]*##g" \
         -e "s# hwversion=[^ ]*##g" \
         -e "s# lazy_refcounts=\\(on\\|off\\)##g" \
+        -e "s# extended_l2=\\(on\\|off\\)##g" \
         -e "s# block_size=[0-9]\\+##g" \
         -e "s# block_state_zero=\\(on\\|off\\)##g" \
         -e "s# log_size=[0-9]\\+##g" \
Signature · 1 line
2.20.1
#17 lore raw

RFC PATCH v2 8/26 qcow2: Add offset_to_sc_index()

In reply to Alberto Garcia

For a given offset, return the subcluster number within its cluster (i.e. with 32 subclusters per cluster it returns a number between 0 and 31).

---
 block/qcow2.h | 5 +++++
 1 file changed, 5 insertions(+)
block/qcow2.h +5 -0
index 278ca41314..e25758079c 100644
@@ -559,6 +559,11 @@ static inline int offset_to_l2_slice_index(BDRVQcow2State *s, int64_t offset)
     return (offset >> s->cluster_bits) & (s->l2_slice_size - 1);
 }
 
+static inline int offset_to_sc_index(BDRVQcow2State *s, int64_t offset)
+{
+    return (offset >> s->subcluster_bits) & (s->subclusters_per_cluster - 1);
+}
+
 static inline int64_t qcow2_vm_state_offset(BDRVQcow2State *s)
 {
     return (int64_t)s->l1_vm_state_index << (s->cluster_bits + s->l2_bits);
Signature · 1 line
2.20.1
#55 lore raw

RFC PATCH v2 22/26 qcow2: Add subcluster support to handle_alloc_space()

In reply to Alberto Garcia

The bdrv_co_pwrite_zeroes() call here fills complete clusters with zeroes, but it can happen that some subclusters are not part of the write request or the copy-on-write. This patch makes sure that only the affected subclusters are overwritten.

A potential improvement would be to also fill with zeroes the other subclusters if we can guarantee that we are not overwriting existing data. However this would waste more disk space, so we should first evaluate if it's really worth doing.

---
 block/qcow2.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
block/qcow2.c +5 -4
index 0261e87709..01322ca449 100644
@@ -2304,6 +2304,9 @@ static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
 
     for (m = l2meta; m != NULL; m = m->next) {
         int ret;
+        uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
+        uint64_t nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
+            m->cow_start.offset;
 
         if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
             continue;
@@ -2318,16 +2321,14 @@ static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
          * efficiently zero out the whole clusters
          */
 
-        ret = qcow2_pre_write_overlap_check(bs, 0, m->alloc_offset,
-                                            m->nb_clusters * s->cluster_size,
+        ret = qcow2_pre_write_overlap_check(bs, 0, start_offset, nb_bytes,
                                             true);
         if (ret < 0) {
             return ret;
         }
 
         BLKDBG_EVENT(bs->file, BLKDBG_CLUSTER_ALLOC_SPACE);
-        ret = bdrv_co_pwrite_zeroes(s->data_file, m->alloc_offset,
-                                    m->nb_clusters * s->cluster_size,
+        ret = bdrv_co_pwrite_zeroes(s->data_file, start_offset, nb_bytes,
                                     BDRV_REQ_NO_FALLBACK);
         if (ret < 0) {
             if (ret != -ENOTSUP && ret != -EAGAIN) {
Signature · 1 line
2.20.1
#15 lore raw

RFC PATCH v2 6/26 qcow2: Add dummy has_subclusters() function

In reply to Alberto Garcia

This function will be used by the qcow2 code to check if an image has subclusters or not.

At the moment this simply returns false. Once all patches needed for subcluster support are ready then QEMU will be able to create and read images with subclusters and this function will return the actual value.

---
 block/qcow2.h | 6 ++++++
 1 file changed, 6 insertions(+)
block/qcow2.h +6 -0
index 940cd4c236..b3826b37c1 100644
@@ -488,6 +488,12 @@ typedef enum QCow2MetadataOverlap {
 
 #define INV_OFFSET (-1ULL)
 
+static inline bool has_subclusters(BDRVQcow2State *s)
+{
+    /* FIXME: Return false until this feature is complete */
+    return false;
+}
+
 static inline uint64_t get_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
                                     int idx)
 {
Signature · 1 line
2.20.1
#12 lore raw

RFC PATCH v2 5/26 qcow2: Document the Extended L2 Entries feature

In reply to Alberto Garcia

Subcluster allocation in qcow2 is implemented by extending the existing L2 table entries and adding additional information to indicate the allocation status of each subcluster.

This patch documents the changes to the qcow2 format and how they affect the calculation of the L2 cache size.

---
 docs/interop/qcow2.txt | 68 ++++++++++++++++++++++++++++++++++++++++--
 docs/qcow2-cache.txt   | 19 +++++++++++-
 2 files changed, 83 insertions(+), 4 deletions(-)
docs/interop/qcow2.txt +65 -3
index af5711e533..d34261f955 100644
@@ -39,6 +39,9 @@ The first cluster of a qcow2 image contains the file header:
                     as the maximum cluster size and won't be able to open images
                     with larger cluster sizes.
 
+                    Note: if the image has Extended L2 Entries then cluster_bits
+                    must be at least 14 (i.e. 16384 byte clusters).
+
          24 - 31:   size
                     Virtual disk size in bytes.
 
@@ -109,7 +112,12 @@ in the description of a field.
                                 An External Data File Name header extension may
                                 be present if this bit is set.
 
-                    Bits 3-63:  Reserved (set to 0)
+                    Bit 3:      Extended L2 Entries.  If this bit is set then
+                                L2 table entries use an extended format that
+                                allows subcluster-based allocation. See the
+                                Extended L2 Entries section for more details.
+
+                    Bits 4-63:  Reserved (set to 0)
 
          80 -  87:  compatible_features
                     Bitmask of compatible features. An implementation can
@@ -437,7 +445,7 @@ cannot be relaxed without an incompatible layout change).
 Given an offset into the virtual disk, the offset into the image file can be
 obtained as follows:
 
-    l2_entries = (cluster_size / sizeof(uint64_t))
+    l2_entries = (cluster_size / sizeof(uint64_t))        [*]
 
     l2_index = (offset / cluster_size) % l2_entries
     l1_index = (offset / cluster_size) / l2_entries
@@ -447,6 +455,8 @@ obtained as follows:
 
     return cluster_offset + (offset % cluster_size)
 
+    [*] this changes if Extended L2 Entries are enabled, see next section
+
 L1 table entry:
 
     Bit  0 -  8:    Reserved (set to 0)
@@ -487,7 +497,8 @@ Standard Cluster Descriptor:
                     nor is data read from the backing file if the cluster is
                     unallocated.
 
-                    With version 2, this is always 0.
+                    With version 2 or with extended L2 entries (see the next
+                    section), this is always 0.
 
          1 -  8:    Reserved (set to 0)
 
@@ -524,6 +535,57 @@ file (except if bit 0 in the Standard Cluster Descriptor is set). If there is
 no backing file or the backing file is smaller than the image, they shall read
 zeros for all parts that are not covered by the backing file.
 
+== Extended L2 Entries ==
+
+An image uses Extended L2 Entries if bit 3 is set on the incompatible_features
+field of the header.
+
+In these images standard data clusters are divided into 32 subclusters of the
+same size. They are contiguous and start from the beginning of the cluster.
+Subclusters can be allocated independently and the L2 entry contains information
+indicating the status of each one of them. Compressed data clusters don't have
+subclusters so they are treated like in images without this feature.
+
+The size of an extended L2 entry is 128 bits so the number of entries per table
+is calculated using this formula:
+
+    l2_entries = (cluster_size / (2 * sizeof(uint64_t)))
+
+The first 64 bits have the same format as the standard L2 table entry described
+in the previous section, with the exception of bit 0 of the standard cluster
+descriptor.
+
+The last 64 bits contain a subcluster allocation bitmap with this format:
+
+Subcluster Allocation Bitmap (for standard clusters):
+
+    Bit  0 -  31:   Allocation status (one bit per subcluster)
+
+                    1: the subcluster is allocated. In this case the
+                       host cluster offset field must contain a valid
+                       offset.
+                    0: the subcluster is not allocated. In this case
+                       read requests shall go to the backing file or
+                       return zeros if there is no backing file data.
+
+                    Bits are assigned starting from the most significant one.
+                    (i.e. bit x is used for subcluster 31 - x)
+
+        32 -  63    Subcluster reads as zeros (one bit per subcluster)
+
+                    1: the subcluster reads as zeros. In this case the
+                       allocation status bit must be unset. The host
+                       cluster offset field may or may not be set.
+                    0: no effect.
+
+                    Bits are assigned starting from the most significant one.
+                    (i.e. bit x is used for subcluster 63 - x)
+
+Subcluster Allocation Bitmap (for compressed clusters):
+
+    Bit  0 -  63:   Reserved (set to 0)
+                    Compressed clusters don't have subclusters,
+                    so this field is not used.
 
 == Snapshots ==
 
docs/qcow2-cache.txt +18 -1
index d57f409861..04eb4ce2f1 100644
@@ -1,6 +1,6 @@
 qcow2 L2/refcount cache configuration
 =====================================
-Copyright (C) 2015, 2018 Igalia, S.L.
+Copyright (C) 2015, 2018-2019 Igalia, S.L.
 Author: Alberto Garcia <[email protected]>
 
 This work is licensed under the terms of the GNU GPL, version 2 or
@@ -222,3 +222,20 @@ support this functionality, and is 0 (disabled) on other platforms.
 This functionality currently relies on the MADV_DONTNEED argument for
 madvise() to actually free the memory. This is a Linux-specific feature,
 so cache-clean-interval is not supported on other systems.
+
+
+Extended L2 Entries
+-------------------
+All numbers shown in this document are valid for qcow2 images with normal
+64-bit L2 entries.
+
+Images with extended L2 entries need twice as much L2 metadata, so the L2
+cache size must be twice as large for the same disk space.
+
+   disk_size = l2_cache_size * cluster_size / 16
+
+i.e.
+
+   l2_cache_size = disk_size * 16 / cluster_size
+
+Refcount blocks are not affected by this.
Signature · 1 line
2.20.1
#47 lore raw

RFC PATCH v2 18/26 qcow2: Add subcluster support to expand_zero_clusters_in_l1()

In reply to Alberto Garcia

Two changes are needed in order to add subcluster support to this
function: deallocated clusters must have their bitmaps cleared, and expanded clusters must have all the "subcluster allocated" bits set.

---
 block/qcow2-cluster.c | 2 ++
 1 file changed, 2 insertions(+)
block/qcow2-cluster.c +2 -0
index aa3eb727a5..62f2a9fcc0 100644
@@ -2036,6 +2036,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
                         /* not backed; therefore we can simply deallocate the
                          * cluster */
                         set_l2_entry(s, l2_slice, j, 0);
+                        set_l2_bitmap(s, l2_slice, j, 0);
                         l2_dirty = true;
                         continue;
                     }
@@ -2102,6 +2103,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
                 } else {
                     set_l2_entry(s, l2_slice, j, offset);
                 }
+                set_l2_bitmap(s, l2_slice, j, QCOW_L2_BITMAP_ALL_ALLOC);
                 l2_dirty = true;
             }
 
Signature · 1 line
2.20.1
#23 lore raw

RFC PATCH v2 11/26 qcow2: Add qcow2_get_subcluster_type()

In reply to Alberto Garcia

This function returns the type of an individual subcluster. If an image does not have subclusters then this returns the exact same value as qcow2_get_cluster_type().

The information in standard and extended L2 entries is encoded in a slightly different way, but all existing QCow2ClusterType values are also valid for subclusters and have the same meanings (although they typically only apply to the requested subcluster).

There are two important exceptions to this:

a) QCOW2_CLUSTER_COMPRESSED means that the whole cluster is compressed. We do not support compression at the subcluster level.

b) QCOW2_CLUSTER_UNALLOCATED means that the cluster is unallocated, that is, the offset field of the L2 entry does not point to a host cluster. All subclusters are obviously unallocated too but any of them could be of type QCOW2_CLUSTER_ZERO_PLAIN.

In addition to that, extended L2 entries allow one new scenario where the cluster is normally allocated but an individual subcluster is not. This is very different from (b) and because of that this patch adds a new value called QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER.

As a last thing, this patch adds QCOW2_CLUSTER_INVALID to detect the cases where an L2 entry has a value that violates the spec. The caller is responsible for handling these situations.

To prevent compatibility problems with images that have invalid values but are currently being read by QEMU without causing side effects, QCOW2_CLUSTER_INVALID is only returned for images with extended L2 entries.

---
 block/qcow2.h | 62 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 62 insertions(+)
block/qcow2.h +62 -0
index 741c41c80f..23a2532ff2 100644
@@ -77,6 +77,15 @@
 
 #define QCOW_MAX_SUBCLUSTERS_PER_CLUSTER 32
 
+/* The subcluster X [0..31] reads as zeroes */
+#define QCOW_OFLAG_SUB_ZERO(X)    ((1ULL << 63) >> (X))
+/* The subcluster X [0..31] is allocated */
+#define QCOW_OFLAG_SUB_ALLOC(X)   ((1ULL << 31) >> (X))
+/* L2 entry bitmap with all "read as zeroes" bits set */
+#define QCOW_L2_BITMAP_ALL_ZEROES 0xFFFFFFFF00000000ULL
+/* L2 entry bitmap with all allocation bits set */
+#define QCOW_L2_BITMAP_ALL_ALLOC  0x00000000FFFFFFFFULL
+
 #define MIN_CLUSTER_BITS 9
 #define MAX_CLUSTER_BITS 21
 
@@ -438,10 +447,12 @@ typedef struct QCowL2Meta
 
 typedef enum QCow2ClusterType {
     QCOW2_CLUSTER_UNALLOCATED,
+    QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER,
     QCOW2_CLUSTER_ZERO_PLAIN,
     QCOW2_CLUSTER_ZERO_ALLOC,
     QCOW2_CLUSTER_NORMAL,
     QCOW2_CLUSTER_COMPRESSED,
+    QCOW2_CLUSTER_INVALID,
 } QCow2ClusterType;
 
 typedef enum QCow2MetadataOverlap {
@@ -621,6 +632,57 @@ static inline QCow2ClusterType qcow2_get_cluster_type(BlockDriverState *bs,
     }
 }
 
+/* In an image without subsclusters this returns the same value as
+ * qcow2_get_cluster_type() */
+static inline int qcow2_get_subcluster_type(BlockDriverState *bs,
+                                            uint64_t l2_entry,
+                                            uint64_t l2_bitmap,
+                                            unsigned sc_index)
+{
+    BDRVQcow2State *s = bs->opaque;
+    QCow2ClusterType type = qcow2_get_cluster_type(bs, l2_entry);
+    assert(sc_index < s->subclusters_per_cluster);
+
+    if (has_subclusters(s)) {
+        bool sc_zero  = l2_bitmap & QCOW_OFLAG_SUB_ZERO(sc_index);
+        bool sc_alloc = l2_bitmap & QCOW_OFLAG_SUB_ALLOC(sc_index);
+        switch (type) {
+        case QCOW2_CLUSTER_COMPRESSED:
+            if (l2_bitmap != 0) {
+                return QCOW2_CLUSTER_INVALID;
+            }
+            break;
+        case QCOW2_CLUSTER_ZERO_PLAIN:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+            return QCOW2_CLUSTER_INVALID;
+        case QCOW2_CLUSTER_NORMAL:
+            if (!sc_zero && !sc_alloc) {
+                return QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER;
+            } else if (!sc_zero && sc_alloc) {
+                return QCOW2_CLUSTER_NORMAL;
+            } else if (sc_zero && !sc_alloc) {
+                return QCOW2_CLUSTER_ZERO_ALLOC;
+            } else { /* sc_zero && sc_alloc */
+                return QCOW2_CLUSTER_INVALID;
+            }
+        case QCOW2_CLUSTER_UNALLOCATED:
+            if (!sc_zero && !sc_alloc) {
+                return QCOW2_CLUSTER_UNALLOCATED;
+            } else if (!sc_zero && sc_alloc) {
+                return QCOW2_CLUSTER_INVALID;
+            } else if (sc_zero && !sc_alloc) {
+                return QCOW2_CLUSTER_ZERO_PLAIN;
+            } else { /* sc_zero && sc_alloc */
+                return QCOW2_CLUSTER_INVALID;
+            }
+        default:
+            g_assert_not_reached();
+        }
+    }
+
+    return type;
+}
+
 /* Check whether refcounts are eager or lazy */
 static inline bool qcow2_need_accurate_refcounts(BDRVQcow2State *s)
 {
Signature · 1 line
2.20.1
#18 lore raw

RFC PATCH v2 9/26 qcow2: Add l2_entry_size()

In reply to Alberto Garcia

qcow2 images with subclusters have 128-bit L2 entries. The first 64 bits contain the same information as traditional images and the last 64 bits form a bitmap with the status of each individual subcluster.

Because of that we cannot assume that L2 entries are sizeof(uint64_t) anymore. This function returns the proper value for the image.

---
 block/qcow2-cluster.c  | 12 ++++++------
 block/qcow2-refcount.c | 14 ++++++++------
 block/qcow2.c          |  6 +++---
 block/qcow2.h          |  5 +++++
 4 files changed, 22 insertions(+), 15 deletions(-)
block/qcow2-cluster.c +6 -6
index 581fa90ab1..1f509bda15 100644
@@ -209,7 +209,7 @@ static int l2_load(BlockDriverState *bs, uint64_t offset,
                    uint64_t l2_offset, uint64_t **l2_slice)
 {
     BDRVQcow2State *s = bs->opaque;
-    int start_of_slice = sizeof(uint64_t) *
+    int start_of_slice = l2_entry_size(s) *
         (offset_to_l2_index(s, offset) - offset_to_l2_slice_index(s, offset));
 
     return qcow2_cache_get(bs, s->l2_table_cache, l2_offset + start_of_slice,
@@ -277,7 +277,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index)
 
     /* allocate a new l2 entry */
 
-    l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
+    l2_offset = qcow2_alloc_clusters(bs, s->l2_size * l2_entry_size(s));
     if (l2_offset < 0) {
         ret = l2_offset;
         goto fail;
@@ -301,7 +301,7 @@ static int l2_allocate(BlockDriverState *bs, int l1_index)
 
     /* allocate a new entry in the l2 cache */
 
-    slice_size2 = s->l2_slice_size * sizeof(uint64_t);
+    slice_size2 = s->l2_slice_size * l2_entry_size(s);
     n_slices = s->cluster_size / slice_size2;
 
     trace_qcow2_l2_allocate_get_empty(bs, l1_index);
@@ -365,7 +365,7 @@ fail:
     }
     s->l1_table[l1_index] = old_l2_offset;
     if (l2_offset > 0) {
-        qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
+        qcow2_free_clusters(bs, l2_offset, s->l2_size * l2_entry_size(s),
                             QCOW2_DISCARD_ALWAYS);
     }
     return ret;
@@ -708,7 +708,7 @@ static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
 
         /* Then decrease the refcount of the old table */
         if (l2_offset) {
-            qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t),
+            qcow2_free_clusters(bs, l2_offset, s->l2_size * l2_entry_size(s),
                                 QCOW2_DISCARD_OTHER);
         }
 
@@ -1883,7 +1883,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
     int ret;
     int i, j;
 
-    slice_size2 = s->l2_slice_size * sizeof(uint64_t);
+    slice_size2 = s->l2_slice_size * l2_entry_size(s);
     n_slices = s->cluster_size / slice_size2;
 
     if (!is_active_l1) {
block/qcow2-refcount.c +8 -6
index 84fe02d388..621318447d 100644
@@ -1253,7 +1253,7 @@ int qcow2_update_snapshot_refcount(BlockDriverState *bs,
     l2_slice = NULL;
     l1_table = NULL;
     l1_size2 = l1_size * sizeof(uint64_t);
-    slice_size2 = s->l2_slice_size * sizeof(uint64_t);
+    slice_size2 = s->l2_slice_size * l2_entry_size(s);
     n_slices = s->cluster_size / slice_size2;
 
     s->cache_discards = true;
@@ -1604,7 +1604,7 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
     int i, l2_size, nb_csectors, ret;
 
     /* Read L2 table from disk */
-    l2_size = s->l2_size * sizeof(uint64_t);
+    l2_size = s->l2_size * l2_entry_size(s);
     l2_table = g_malloc(l2_size);
 
     ret = bdrv_pread(bs->file, l2_offset, l2_table, l2_size);
@@ -1679,15 +1679,16 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                             fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR",
                             offset);
                     if (fix & BDRV_FIX_ERRORS) {
+                        int idx = i * (l2_entry_size(s) / sizeof(uint64_t));
                         uint64_t l2e_offset =
-                            l2_offset + (uint64_t)i * sizeof(uint64_t);
+                            l2_offset + (uint64_t)i * l2_entry_size(s);
                         int ign = active ? QCOW2_OL_ACTIVE_L2 :
                                            QCOW2_OL_INACTIVE_L2;
 
                         l2_entry = QCOW_OFLAG_ZERO;
                         set_l2_entry(s, l2_table, i, l2_entry);
                         ret = qcow2_pre_write_overlap_check(bs, ign,
-                                l2e_offset, sizeof(uint64_t), false);
+                                l2e_offset, l2_entry_size(s), false);
                         if (ret < 0) {
                             fprintf(stderr, "ERROR: Overlap check failed\n");
                             res->check_errors++;
@@ -1697,7 +1698,8 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                         }
 
                         ret = bdrv_pwrite_sync(bs->file, l2e_offset,
-                                               &l2_table[i], sizeof(uint64_t));
+                                               &l2_table[idx],
+                                               l2_entry_size(s));
                         if (ret < 0) {
                             fprintf(stderr, "ERROR: Failed to overwrite L2 "
                                     "table entry: %s\n", strerror(-ret));
@@ -1904,7 +1906,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
         }
 
         ret = bdrv_pread(bs->file, l2_offset, l2_table,
-                         s->l2_size * sizeof(uint64_t));
+                         s->l2_size * l2_entry_size(s));
         if (ret < 0) {
             fprintf(stderr, "ERROR: Could not read L2 table: %s\n",
                     strerror(-ret));
block/qcow2.c +3 -3
index ed8b81c7b7..ab40ae36ea 100644
@@ -832,7 +832,7 @@ static void read_cache_sizes(BlockDriverState *bs, QemuOpts *opts,
     uint64_t max_l2_entries = DIV_ROUND_UP(virtual_disk_size, s->cluster_size);
     /* An L2 table is always one cluster in size so the max cache size
      * should be a multiple of the cluster size. */
-    uint64_t max_l2_cache = ROUND_UP(max_l2_entries * sizeof(uint64_t),
+    uint64_t max_l2_cache = ROUND_UP(max_l2_entries * l2_entry_size(s),
                                      s->cluster_size);
 
     combined_cache_size_set = qemu_opt_get(opts, QCOW2_OPT_CACHE_SIZE);
@@ -993,7 +993,7 @@ static int qcow2_update_options_prepare(BlockDriverState *bs,
         }
     }
 
-    r->l2_slice_size = l2_cache_entry_size / sizeof(uint64_t);
+    r->l2_slice_size = l2_cache_entry_size / l2_entry_size(s);
     r->l2_table_cache = qcow2_cache_create(bs, l2_cache_size,
                                            l2_cache_entry_size);
     r->refcount_block_cache = qcow2_cache_create(bs, refcount_cache_size,
@@ -1387,7 +1387,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
         bs->encrypted = true;
     }
 
-    s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
+    s->l2_bits = s->cluster_bits - ctz32(l2_entry_size(s));
     s->l2_size = 1 << s->l2_bits;
     /* 2^(s->refcount_order - 3) is the refcount width in bytes */
     s->refcount_block_bits = s->cluster_bits - (s->refcount_order - 3);
block/qcow2.h +5 -0
index e25758079c..29a253bfb9 100644
@@ -499,6 +499,11 @@ static inline bool has_subclusters(BDRVQcow2State *s)
     return false;
 }
 
+static inline size_t l2_entry_size(BDRVQcow2State *s)
+{
+    return has_subclusters(s) ? sizeof(uint64_t) * 2 : sizeof(uint64_t);
+}
+
 static inline uint64_t get_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
                                     int idx)
 {
Signature · 1 line
2.20.1
#46 lore raw

RFC PATCH v2 17/26 qcow2: Add subcluster support to check_refcounts_l2()

In reply to Alberto Garcia

Setting the QCOW_OFLAG_ZERO bit of the L2 entry is forbidden if an image has subclusters. Instead, the individual 'all zeroes' bits must be used.

---
 block/qcow2-refcount.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)
block/qcow2-refcount.c +7 -2
index 621318447d..bc73125f70 100644
@@ -1685,8 +1685,13 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                         int ign = active ? QCOW2_OL_ACTIVE_L2 :
                                            QCOW2_OL_INACTIVE_L2;
 
-                        l2_entry = QCOW_OFLAG_ZERO;
-                        set_l2_entry(s, l2_table, i, l2_entry);
+                        if (has_subclusters(s)) {
+                            set_l2_entry(s, l2_table, i, 0);
+                            set_l2_bitmap(s, l2_table, i,
+                                          QCOW_L2_BITMAP_ALL_ZEROES);
+                        } else {
+                            set_l2_entry(s, l2_table, i, QCOW_OFLAG_ZERO);
+                        }
                         ret = qcow2_pre_write_overlap_check(bs, ign,
                                 l2e_offset, l2_entry_size(s), false);
                         if (ret < 0) {
Signature · 1 line
2.20.1
#11 lore raw

RFC PATCH v2 4/26 qcow2: Add get_l2_entry() and set_l2_entry()

In reply to Alberto Garcia

The size of an L2 entry is 64 bits, but if we want to have subclusters we need extended L2 entries. This means that we have to access L2 tables and slices differently depending on whether an image has extended L2 entries or not.

This patch replaces all l2_slice[] accesses with calls to get_l2_entry() and set_l2_entry().

---
 block/qcow2-cluster.c  | 65 ++++++++++++++++++++++--------------------
 block/qcow2-refcount.c | 17 +++++------
 block/qcow2.h          | 12 ++++++++
 3 files changed, 55 insertions(+), 39 deletions(-)
block/qcow2-cluster.c +34 -31
index ee6b46f917..581fa90ab1 100644
@@ -379,12 +379,13 @@ fail:
  * cluster which may require a different handling)
  */
 static int count_contiguous_clusters(BlockDriverState *bs, int nb_clusters,
-        int cluster_size, uint64_t *l2_slice, uint64_t stop_flags)
+        int cluster_size, uint64_t *l2_slice, int l2_index, uint64_t stop_flags)
 {
+    BDRVQcow2State *s = bs->opaque;
     int i;
     QCow2ClusterType first_cluster_type;
     uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
-    uint64_t first_entry = be64_to_cpu(l2_slice[0]);
+    uint64_t first_entry = get_l2_entry(s, l2_slice, l2_index);
     uint64_t offset = first_entry & mask;
 
     first_cluster_type = qcow2_get_cluster_type(bs, first_entry);
@@ -397,7 +398,7 @@ static int count_contiguous_clusters(BlockDriverState *bs, int nb_clusters,
            first_cluster_type == QCOW2_CLUSTER_ZERO_ALLOC);
 
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t l2_entry = be64_to_cpu(l2_slice[i]) & mask;
+        uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index + i) & mask;
         if (offset + (uint64_t) i * cluster_size != l2_entry) {
             break;
         }
@@ -413,14 +414,16 @@ static int count_contiguous_clusters(BlockDriverState *bs, int nb_clusters,
 static int count_contiguous_clusters_unallocated(BlockDriverState *bs,
                                                  int nb_clusters,
                                                  uint64_t *l2_slice,
+                                                 int l2_index,
                                                  QCow2ClusterType wanted_type)
 {
+    BDRVQcow2State *s = bs->opaque;
     int i;
 
     assert(wanted_type == QCOW2_CLUSTER_ZERO_PLAIN ||
            wanted_type == QCOW2_CLUSTER_UNALLOCATED);
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t entry = be64_to_cpu(l2_slice[i]);
+        uint64_t entry = get_l2_entry(s, l2_slice, l2_index + i);
         QCow2ClusterType type = qcow2_get_cluster_type(bs, entry);
 
         if (type != wanted_type) {
@@ -566,7 +569,7 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
     /* find the cluster offset for the given disk offset */
 
     l2_index = offset_to_l2_slice_index(s, offset);
-    *cluster_offset = be64_to_cpu(l2_slice[l2_index]);
+    *cluster_offset = get_l2_entry(s, l2_slice, l2_index);
 
     nb_clusters = size_to_clusters(s, bytes_needed);
     /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned
@@ -601,14 +604,14 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
     case QCOW2_CLUSTER_UNALLOCATED:
         /* how many empty clusters ? */
         c = count_contiguous_clusters_unallocated(bs, nb_clusters,
-                                                  &l2_slice[l2_index], type);
+                                                  l2_slice, l2_index, type);
         *cluster_offset = 0;
         break;
     case QCOW2_CLUSTER_ZERO_ALLOC:
     case QCOW2_CLUSTER_NORMAL:
         /* how many allocated clusters ? */
         c = count_contiguous_clusters(bs, nb_clusters, s->cluster_size,
-                                      &l2_slice[l2_index], QCOW_OFLAG_ZERO);
+                                      l2_slice, l2_index, QCOW_OFLAG_ZERO);
         *cluster_offset &= L2E_OFFSET_MASK;
         if (offset_into_cluster(s, *cluster_offset)) {
             qcow2_signal_corruption(bs, true, -1, -1,
@@ -761,7 +764,7 @@ int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
 
     /* Compression can't overwrite anything. Fail if the cluster was already
      * allocated. */
-    cluster_offset = be64_to_cpu(l2_slice[l2_index]);
+    cluster_offset = get_l2_entry(s, l2_slice, l2_index);
     if (cluster_offset & L2E_OFFSET_MASK) {
         qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
         return -EIO;
@@ -786,7 +789,7 @@ int qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
 
     BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
     qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
-    l2_slice[l2_index] = cpu_to_be64(cluster_offset);
+    set_l2_entry(s, l2_slice, l2_index, cluster_offset);
     qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
 
     *host_offset = cluster_offset & s->cluster_offset_mask;
@@ -978,12 +981,12 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
          * cluster the second one has to do RMW (which is done above by
          * perform_cow()), update l2 table with its cluster pointer and free
          * old cluster. This is what this loop does */
-        if (l2_slice[l2_index + i] != 0) {
-            old_cluster[j++] = l2_slice[l2_index + i];
+        if (get_l2_entry(s, l2_slice, l2_index + i) != 0) {
+            old_cluster[j++] = get_l2_entry(s, l2_slice, l2_index + i);
         }
 
-        l2_slice[l2_index + i] = cpu_to_be64((cluster_offset +
-                    (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
+        set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_COPIED |
+                     (cluster_offset + (i << s->cluster_bits)));
      }
 
 
@@ -997,8 +1000,7 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
      */
     if (!m->keep_old_clusters && j != 0) {
         for (i = 0; i < j; i++) {
-            qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1,
-                                    QCOW2_DISCARD_NEVER);
+            qcow2_free_any_clusters(bs, old_cluster[i], 1, QCOW2_DISCARD_NEVER);
         }
     }
 
@@ -1051,7 +1053,7 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
     if (keep_old) {
         int i;
         for (i = 0; i < nb_clusters; i++) {
-            l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
+            l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
             if (qcow2_get_cluster_type(bs, l2_entry) != QCOW2_CLUSTER_NORMAL) {
                 break;
             }
@@ -1062,7 +1064,7 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
     }
 
     /* Get the L2 entry from the first cluster */
-    l2_entry = be64_to_cpu(l2_slice[l2_index]);
+    l2_entry = get_l2_entry(s, l2_slice, l2_index);
     type = qcow2_get_cluster_type(bs, l2_entry);
 
     if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
@@ -1072,7 +1074,7 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
     }
 
     /* Get the L2 entry from the last cluster */
-    l2_entry = be64_to_cpu(l2_slice[l2_index + nb_clusters - 1]);
+    l2_entry = get_l2_entry(s, l2_slice, l2_index + nb_clusters - 1);
     type = qcow2_get_cluster_type(bs, l2_entry);
 
     if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
@@ -1139,12 +1141,12 @@ static int count_cow_clusters(BlockDriverState *bs, int nb_clusters,
                               uint64_t *l2_slice, int l2_index, bool want_cow)
 {
     BDRVQcow2State *s = bs->opaque;
-    uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index]);
+    uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index);
     uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK;
     int i;
 
     for (i = 0; i < nb_clusters; i++) {
-        l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
+        l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
         if (cluster_needs_cow(bs, l2_entry) != want_cow) {
             break;
         }
@@ -1277,7 +1279,7 @@ static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
         return ret;
     }
 
-    cluster_offset = be64_to_cpu(l2_slice[l2_index]);
+    cluster_offset = get_l2_entry(s, l2_slice, l2_index);
 
     if (!cluster_needs_cow(bs, cluster_offset)) {
         /* If a specific host_offset is required, check it */
@@ -1658,7 +1660,7 @@ static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset,
     for (i = 0; i < nb_clusters; i++) {
         uint64_t old_l2_entry;
 
-        old_l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
+        old_l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
 
         /*
          * If full_discard is false, make sure that a discarded area reads back
@@ -1698,9 +1700,9 @@ static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset,
         /* First remove L2 entries */
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
         if (!full_discard && s->qcow_version >= 3) {
-            l2_slice[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
+            set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
         } else {
-            l2_slice[l2_index + i] = cpu_to_be64(0);
+            set_l2_entry(s, l2_slice, l2_index + i, 0);
         }
 
         /* Then decrease the refcount */
@@ -1780,7 +1782,7 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
         uint64_t old_offset;
         QCow2ClusterType cluster_type;
 
-        old_offset = be64_to_cpu(l2_slice[l2_index + i]);
+        old_offset = get_l2_entry(s, l2_slice, l2_index + i);
 
         /*
          * Minimize L2 changes if the cluster already reads back as
@@ -1794,10 +1796,11 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
 
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
         if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) {
-            l2_slice[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
+            set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
             qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);
         } else {
-            l2_slice[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
+            uint64_t entry = get_l2_entry(s, l2_slice, l2_index + i);
+            set_l2_entry(s, l2_slice, l2_index + i, entry | QCOW_OFLAG_ZERO);
         }
     }
 
@@ -1935,7 +1938,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
             }
 
             for (j = 0; j < s->l2_slice_size; j++) {
-                uint64_t l2_entry = be64_to_cpu(l2_slice[j]);
+                uint64_t l2_entry = get_l2_entry(s, l2_slice, j);
                 int64_t offset = l2_entry & L2E_OFFSET_MASK;
                 QCow2ClusterType cluster_type =
                     qcow2_get_cluster_type(bs, l2_entry);
@@ -1949,7 +1952,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
                     if (!bs->backing) {
                         /* not backed; therefore we can simply deallocate the
                          * cluster */
-                        l2_slice[j] = 0;
+                        set_l2_entry(s, l2_slice, j, 0);
                         l2_dirty = true;
                         continue;
                     }
@@ -2012,9 +2015,9 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
                 }
 
                 if (l2_refcount == 1) {
-                    l2_slice[j] = cpu_to_be64(offset | QCOW_OFLAG_COPIED);
+                    set_l2_entry(s, l2_slice, j, offset | QCOW_OFLAG_COPIED);
                 } else {
-                    l2_slice[j] = cpu_to_be64(offset);
+                    set_l2_entry(s, l2_slice, j, offset);
                 }
                 l2_dirty = true;
             }
block/qcow2-refcount.c +9 -8
index 0d64bf5a5e..84fe02d388 100644
@@ -1309,7 +1309,7 @@ int qcow2_update_snapshot_refcount(BlockDriverState *bs,
                     uint64_t cluster_index;
                     uint64_t offset;
 
-                    entry = be64_to_cpu(l2_slice[j]);
+                    entry = get_l2_entry(s, l2_slice, j);
                     old_entry = entry;
                     entry &= ~QCOW_OFLAG_COPIED;
                     offset = entry & L2E_OFFSET_MASK;
@@ -1383,7 +1383,7 @@ int qcow2_update_snapshot_refcount(BlockDriverState *bs,
                             qcow2_cache_set_dependency(bs, s->l2_table_cache,
                                                        s->refcount_block_cache);
                         }
-                        l2_slice[j] = cpu_to_be64(entry);
+                        set_l2_entry(s, l2_slice, j, entry);
                         qcow2_cache_entry_mark_dirty(s->l2_table_cache,
                                                      l2_slice);
                     }
@@ -1616,7 +1616,7 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
 
     /* Do the actual checks */
     for(i = 0; i < s->l2_size; i++) {
-        l2_entry = be64_to_cpu(l2_table[i]);
+        l2_entry = get_l2_entry(s, l2_table, i);
 
         switch (qcow2_get_cluster_type(bs, l2_entry)) {
         case QCOW2_CLUSTER_COMPRESSED:
@@ -1685,7 +1685,7 @@ static int check_refcounts_l2(BlockDriverState *bs, BdrvCheckResult *res,
                                            QCOW2_OL_INACTIVE_L2;
 
                         l2_entry = QCOW_OFLAG_ZERO;
-                        l2_table[i] = cpu_to_be64(l2_entry);
+                        set_l2_entry(s, l2_table, i, l2_entry);
                         ret = qcow2_pre_write_overlap_check(bs, ign,
                                 l2e_offset, sizeof(uint64_t), false);
                         if (ret < 0) {
@@ -1913,7 +1913,7 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
         }
 
         for (j = 0; j < s->l2_size; j++) {
-            uint64_t l2_entry = be64_to_cpu(l2_table[j]);
+            uint64_t l2_entry = get_l2_entry(s, l2_table, j);
             uint64_t data_offset = l2_entry & L2E_OFFSET_MASK;
             QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry);
 
@@ -1936,9 +1936,10 @@ static int check_oflag_copied(BlockDriverState *bs, BdrvCheckResult *res,
                             "l2_entry=%" PRIx64 " refcount=%" PRIu64 "\n",
                             repair ? "Repairing" : "ERROR", l2_entry, refcount);
                     if (repair) {
-                        l2_table[j] = cpu_to_be64(refcount == 1
-                                    ? l2_entry |  QCOW_OFLAG_COPIED
-                                    : l2_entry & ~QCOW_OFLAG_COPIED);
+                        set_l2_entry(s, l2_table, j,
+                                     refcount == 1 ?
+                                     l2_entry |  QCOW_OFLAG_COPIED :
+                                     l2_entry & ~QCOW_OFLAG_COPIED);
                         l2_dirty++;
                     }
                 }
block/qcow2.h +12 -0
index 5cccd87162..940cd4c236 100644
@@ -488,6 +488,18 @@ typedef enum QCow2MetadataOverlap {
 
 #define INV_OFFSET (-1ULL)
 
+static inline uint64_t get_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
+                                    int idx)
+{
+    return be64_to_cpu(l2_slice[idx]);
+}
+
+static inline void set_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
+                                int idx, uint64_t entry)
+{
+    l2_slice[idx] = cpu_to_be64(entry);
+}
+
 static inline bool has_data_file(BlockDriverState *bs)
 {
     BDRVQcow2State *s = bs->opaque;
Signature · 1 line
2.20.1
#20 lore raw

RFC PATCH v2 10/26 qcow2: Update get/set_l2_entry() and add get/set_l2_bitmap()

In reply to Alberto Garcia

Extended L2 entries are 128-bit wide: 64 bits for the entry itself and 64 bits for the subcluster allocation bitmap.

In order to support them correctly get/set_l2_entry() need to be updated so they take the entry width into account in order to calculate the correct offset.

This patch also adds the get/set_l2_bitmap() functions that are used to access the bitmaps. For convenience, these functions are no-ops when used in traditional qcow2 images.

---
 block/qcow2.h | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)
block/qcow2.h +22 -0
index 29a253bfb9..741c41c80f 100644
@@ -507,15 +507,37 @@ static inline size_t l2_entry_size(BDRVQcow2State *s)
 static inline uint64_t get_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
                                     int idx)
 {
+    idx *= l2_entry_size(s) / sizeof(uint64_t);
     return be64_to_cpu(l2_slice[idx]);
 }
 
+static inline uint64_t get_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice,
+                                     int idx)
+{
+    if (has_subclusters(s)) {
+        idx *= l2_entry_size(s) / sizeof(uint64_t);
+        return be64_to_cpu(l2_slice[idx + 1]);
+    } else {
+        return 0;
+    }
+}
+
 static inline void set_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
                                 int idx, uint64_t entry)
 {
+    idx *= l2_entry_size(s) / sizeof(uint64_t);
     l2_slice[idx] = cpu_to_be64(entry);
 }
 
+static inline void set_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice,
+                                 int idx, uint64_t bitmap)
+{
+    if (has_subclusters(s)) {
+        idx *= l2_entry_size(s) / sizeof(uint64_t);
+        l2_slice[idx + 1] = cpu_to_be64(bitmap);
+    }
+}
+
 static inline bool has_data_file(BlockDriverState *bs)
 {
     BDRVQcow2State *s = bs->opaque;
Signature · 1 line
2.20.1
#1 lore raw

RFC PATCH v2 1/26 qcow2: Add calculate_l2_meta()

In reply to Alberto Garcia

handle_alloc() creates a QCowL2Meta structure in order to update the image metadata and perform the necessary copy-on-write operations.

This patch moves that code to a separate function so it can be used from other places.

---
 block/qcow2-cluster.c | 76 +++++++++++++++++++++++++++++--------------
 1 file changed, 52 insertions(+), 24 deletions(-)
block/qcow2-cluster.c +52 -24
index 8982b7b762..6c1dcdc781 100644
@@ -1019,6 +1019,55 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
                         QCOW2_DISCARD_NEVER);
 }
 
+/*
+ * For a given write request, create a new QCowL2Meta structure and
+ * add it to @m.
+ *
+ * @host_offset points to the beginning of the first cluster.
+ *
+ * @guest_offset and @bytes indicate the offset and length of the
+ * request.
+ *
+ * If @keep_old is true it means that the clusters were already
+ * allocated and will be overwritten. If false then the clusters are
+ * new and we have to decrease the reference count of the old ones.
+ */
+static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
+                              uint64_t guest_offset, uint64_t bytes,
+                              QCowL2Meta **m, bool keep_old)
+{
+    BDRVQcow2State *s = bs->opaque;
+    unsigned cow_start_from = 0;
+    unsigned cow_start_to = offset_into_cluster(s, guest_offset);
+    unsigned cow_end_from = cow_start_to + bytes;
+    unsigned cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
+    unsigned nb_clusters = size_to_clusters(s, cow_end_from);
+    QCowL2Meta *old_m = *m;
+
+    *m = g_malloc0(sizeof(**m));
+    **m = (QCowL2Meta) {
+        .next           = old_m,
+
+        .alloc_offset   = host_offset,
+        .offset         = start_of_cluster(s, guest_offset),
+        .nb_clusters    = nb_clusters,
+
+        .keep_old_clusters = keep_old,
+
+        .cow_start = {
+            .offset     = cow_start_from,
+            .nb_bytes   = cow_start_to - cow_start_from,
+        },
+        .cow_end = {
+            .offset     = cow_end_from,
+            .nb_bytes   = cow_end_to - cow_end_from,
+        },
+    };
+
+    qemu_co_queue_init(&(*m)->dependent_requests);
+    QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
+}
+
 /*
  * Returns the number of contiguous clusters that can be used for an allocating
  * write, but require COW to be performed (this includes yet unallocated space,
@@ -1417,35 +1466,14 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset);
     int avail_bytes = nb_clusters << s->cluster_bits;
     int nb_bytes = MIN(requested_bytes, avail_bytes);
-    QCowL2Meta *old_m = *m;
-
-    *m = g_malloc0(sizeof(**m));
-
-    **m = (QCowL2Meta) {
-        .next           = old_m,
-
-        .alloc_offset   = alloc_cluster_offset,
-        .offset         = start_of_cluster(s, guest_offset),
-        .nb_clusters    = nb_clusters,
-
-        .keep_old_clusters  = keep_old_clusters,
-
-        .cow_start = {
-            .offset     = 0,
-            .nb_bytes   = offset_into_cluster(s, guest_offset),
-        },
-        .cow_end = {
-            .offset     = nb_bytes,
-            .nb_bytes   = avail_bytes - nb_bytes,
-        },
-    };
-    qemu_co_queue_init(&(*m)->dependent_requests);
-    QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
 
     *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
     *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
     assert(*bytes != 0);
 
+    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
+                      m, keep_old_clusters);
+
     return 1;
 
 fail:
Signature · 1 line
2.20.1
#26 lore raw

RFC PATCH v2 12/26 qcow2: Handle QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER

In reply to Alberto Garcia

In the previous patch we added a new QCow2ClusterType named QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER. There is a couple of places where this new value needs to be handled, and that is what this patch does.

---
 block/qcow2.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
block/qcow2.c +9 -4
index ab40ae36ea..0261e87709 100644
@@ -1938,8 +1938,8 @@ static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
 
     *pnum = bytes;
 
-    if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC) &&
-        !s->crypto) {
+    if ((ret == QCOW2_CLUSTER_NORMAL || ret == QCOW2_CLUSTER_ZERO_ALLOC ||
+         ret == QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER) && !s->crypto) {
         index_in_cluster = offset & (s->cluster_size - 1);
         *map = cluster_offset | index_in_cluster;
         *file = s->data_file->bs;
@@ -1947,7 +1947,8 @@ static int coroutine_fn qcow2_co_block_status(BlockDriverState *bs,
     }
     if (ret == QCOW2_CLUSTER_ZERO_PLAIN || ret == QCOW2_CLUSTER_ZERO_ALLOC) {
         status |= BDRV_BLOCK_ZERO;
-    } else if (ret != QCOW2_CLUSTER_UNALLOCATED) {
+    } else if (ret != QCOW2_CLUSTER_UNALLOCATED &&
+               ret != QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER) {
         status |= BDRV_BLOCK_DATA;
     }
     if (s->metadata_preallocation && (status & BDRV_BLOCK_DATA) &&
@@ -2117,6 +2118,7 @@ static coroutine_fn int qcow2_co_preadv_task(BlockDriverState *bs,
         g_assert_not_reached();
 
     case QCOW2_CLUSTER_UNALLOCATED:
+    case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
         assert(bs->backing); /* otherwise handled in qcow2_co_preadv_part */
 
         BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING_AIO);
@@ -2187,7 +2189,8 @@ static coroutine_fn int qcow2_co_preadv_part(BlockDriverState *bs,
 
         if (ret == QCOW2_CLUSTER_ZERO_PLAIN ||
             ret == QCOW2_CLUSTER_ZERO_ALLOC ||
-            (ret == QCOW2_CLUSTER_UNALLOCATED && !bs->backing))
+            (ret == QCOW2_CLUSTER_UNALLOCATED && !bs->backing) ||
+            (ret == QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER && !bs->backing))
         {
             qemu_iovec_memset(qiov, qiov_offset, 0, cur_bytes);
         } else {
@@ -3701,6 +3704,7 @@ static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
         nr = s->cluster_size;
         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
+            ret != QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER &&
             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
             ret != QCOW2_CLUSTER_ZERO_ALLOC) {
             qemu_co_mutex_unlock(&s->lock);
@@ -3771,6 +3775,7 @@ qcow2_co_copy_range_from(BlockDriverState *bs,
 
         switch (ret) {
         case QCOW2_CLUSTER_UNALLOCATED:
+        case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
             if (bs->backing && bs->backing->bs) {
                 int64_t backing_length = bdrv_getlength(bs->backing->bs);
                 if (src_offset >= backing_length) {
Signature · 1 line
2.20.1
#16 lore raw

RFC PATCH v2 7/26 qcow2: Add subcluster-related fields to BDRVQcow2State

In reply to Alberto Garcia

This patch adds the following new fields to BDRVQcow2State:

- subclusters_per_cluster: Number of subclusters in a cluster
- subcluster_size: The size of each subcluster, in bytes
- subcluster_bits: No. of bits so 1 << subcluster_bits = subcluster_size

Images without subclusters are treated as if they had exactly one, with subcluster_size = cluster_size.

---
 block/qcow2.c | 5 +++++
 block/qcow2.h | 5 +++++
 2 files changed, 10 insertions(+)
block/qcow2.c +5 -0
index 0bc69e6996..ed8b81c7b7 100644
@@ -1342,6 +1342,11 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
         }
     }
 
+    s->subclusters_per_cluster =
+        has_subclusters(s) ? QCOW_MAX_SUBCLUSTERS_PER_CLUSTER : 1;
+    s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
+    s->subcluster_bits = ctz32(s->subcluster_size);
+
     /* Check support for various header values */
     if (header.refcount_order > 6) {
         error_setg(errp, "Reference count entry width too large; may not "
block/qcow2.h +5 -0
index b3826b37c1..278ca41314 100644
@@ -75,6 +75,8 @@
 /* The cluster reads as all zeros */
 #define QCOW_OFLAG_ZERO (1ULL << 0)
 
+#define QCOW_MAX_SUBCLUSTERS_PER_CLUSTER 32
+
 #define MIN_CLUSTER_BITS 9
 #define MAX_CLUSTER_BITS 21
 
@@ -277,6 +279,9 @@ typedef struct BDRVQcow2State {
     int cluster_bits;
     int cluster_size;
     int l2_slice_size;
+    int subcluster_bits;
+    int subcluster_size;
+    int subclusters_per_cluster;
     int l2_bits;
     int l2_size;
     int l1_size;
Signature · 1 line
2.20.1
#57 lore raw

RFC PATCH v2 23/26 qcow2: Restrict qcow2_co_pwrite_zeroes() to full clusters only

In reply to Alberto Garcia

Ideally it should be possible to zero individual subclusters using this function, but this is currently not implemented.

---
 block/qcow2.c | 6 ++++++
 1 file changed, 6 insertions(+)
block/qcow2.c +6 -0
index 01322ca449..537569ce88 100644
@@ -3704,6 +3704,12 @@ static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
         bytes = s->cluster_size;
         nr = s->cluster_size;
         ret = qcow2_get_cluster_offset(bs, offset, &nr, &off);
+        /* TODO: allow zeroing separate subclusters, we only allow
+         * zeroing full clusters at the moment. */
+        if (nr != bytes) {
+            qemu_co_mutex_unlock(&s->lock);
+            return -ENOTSUP;
+        }
         if (ret != QCOW2_CLUSTER_UNALLOCATED &&
             ret != QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER &&
             ret != QCOW2_CLUSTER_ZERO_PLAIN &&
Signature · 1 line
2.20.1
#0 lore raw

v2 cover Add subcluster allocation to qcow2

Hi,

here's the new version of the patches to add subcluster allocation support to qcow2.

Please refer to the cover letter of the first version for a full description of the patches:

https://lists.gnu.org/archive/html/qemu-block/2019-10/msg00983.html

This version includes a few tests, but I'm planning to add more for the next revision.

Berto

v2:
- Patch 12: Update after the changes in 88f468e546.
- Patch 21 (new): Clear the L2 bitmap when allocating a compressed cluster. Compressed clusters should have the bitmap all set to 0.
- Patch 24: Document the new fields in the QAPI documentation [Eric].
- Patch 25: Allow qcow2 preallocation with backing files.
- Patch 26: Add some tests for qcow2 images with extended L2 entries.

v1: https://lists.gnu.org/archive/html/qemu-block/2019-10/msg00983.html

Output of git backport-diff against v1:

Key:
[----] : patches are identical
[####] : number of functional differences between upstream/downstream patch [down] : patch is downstream-only
The flags [FC] indicate (F)unctional and (C)ontextual differences, respectively

001/26:[----] [-C] 'qcow2: Add calculate_l2_meta()'
002/26:[----] [--] 'qcow2: Split cluster_needs_cow() out of count_cow_clusters()' 003/26:[----] [--] 'qcow2: Process QCOW2_CLUSTER_ZERO_ALLOC clusters in handle_copied()' 004/26:[----] [--] 'qcow2: Add get_l2_entry() and set_l2_entry()' 005/26:[----] [--] 'qcow2: Document the Extended L2 Entries feature' 006/26:[----] [--] 'qcow2: Add dummy has_subclusters() function' 007/26:[----] [--] 'qcow2: Add subcluster-related fields to BDRVQcow2State' 008/26:[----] [--] 'qcow2: Add offset_to_sc_index()'
009/26:[----] [--] 'qcow2: Add l2_entry_size()'
010/26:[----] [--] 'qcow2: Update get/set_l2_entry() and add get/set_l2_bitmap()' 011/26:[----] [--] 'qcow2: Add qcow2_get_subcluster_type()'
012/26:[0005] [FC] 'qcow2: Handle QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER' 013/26:[----] [--] 'qcow2: Add subcluster support to calculate_l2_meta()' 014/26:[----] [--] 'qcow2: Add subcluster support to qcow2_get_cluster_offset()' 015/26:[----] [--] 'qcow2: Add subcluster support to zero_in_l2_slice()' 016/26:[----] [--] 'qcow2: Add subcluster support to discard_in_l2_slice()' 017/26:[----] [--] 'qcow2: Add subcluster support to check_refcounts_l2()' 018/26:[----] [--] 'qcow2: Add subcluster support to expand_zero_clusters_in_l1()' 019/26:[----] [--] 'qcow2: Fix offset calculation in handle_dependencies()' 020/26:[----] [--] 'qcow2: Update L2 bitmap in qcow2_alloc_cluster_link_l2()' 021/26:[down] 'qcow2: Clear the L2 bitmap when allocating a compressed cluster' 022/26:[----] [--] 'qcow2: Add subcluster support to handle_alloc_space()' 023/26:[----] [--] 'qcow2: Restrict qcow2_co_pwrite_zeroes() to full clusters only' 024/26:[0007] [FC] 'qcow2: Add the 'extended_l2' option and the QCOW2_INCOMPAT_EXTL2 bit' 025/26:[down] 'qcow2: Allow preallocation and backing files if extended_l2 is set' 026/26:[down] 'iotests: Add tests for qcow2 images with extended L2 entries'

Alberto Garcia (26):
qcow2: Add calculate_l2_meta()
qcow2: Split cluster_needs_cow() out of count_cow_clusters()
qcow2: Process QCOW2_CLUSTER_ZERO_ALLOC clusters in handle_copied()
qcow2: Add get_l2_entry() and set_l2_entry()
qcow2: Document the Extended L2 Entries feature
qcow2: Add dummy has_subclusters() function
qcow2: Add subcluster-related fields to BDRVQcow2State
qcow2: Add offset_to_sc_index()
qcow2: Add l2_entry_size()
qcow2: Update get/set_l2_entry() and add get/set_l2_bitmap()
qcow2: Add qcow2_get_subcluster_type()
qcow2: Handle QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER
qcow2: Add subcluster support to calculate_l2_meta()
qcow2: Add subcluster support to qcow2_get_cluster_offset()
qcow2: Add subcluster support to zero_in_l2_slice()
qcow2: Add subcluster support to discard_in_l2_slice()
qcow2: Add subcluster support to check_refcounts_l2()
qcow2: Add subcluster support to expand_zero_clusters_in_l1()
qcow2: Fix offset calculation in handle_dependencies()
qcow2: Update L2 bitmap in qcow2_alloc_cluster_link_l2()
qcow2: Clear the L2 bitmap when allocating a compressed cluster
qcow2: Add subcluster support to handle_alloc_space()
qcow2: Restrict qcow2_co_pwrite_zeroes() to full clusters only
qcow2: Add the 'extended_l2' option and the QCOW2_INCOMPAT_EXTL2 bit
qcow2: Allow preallocation and backing files if extended_l2 is set
iotests: Add tests for qcow2 images with extended L2 entries

 block/qcow2-cluster.c            | 548 ++++++++++++++++++++-----------
 block/qcow2-refcount.c           |  38 ++-
 block/qcow2.c                    |  92 +++++-
 block/qcow2.h                    | 121 ++++++-
 docs/interop/qcow2.txt           |  68 +++-
 docs/qcow2-cache.txt             |  19 +-
 include/block/block_int.h        |   1 +
 qapi/block-core.json             |   6 +
 tests/qemu-iotests/031.out       |   8 +-
 tests/qemu-iotests/036.out       |   4 +-
 tests/qemu-iotests/049.out       | 102 +++---
 tests/qemu-iotests/060.out       |   1 +
 tests/qemu-iotests/061.out       |  20 +-
 tests/qemu-iotests/065           |  18 +-
 tests/qemu-iotests/082.out       |  48 ++-
 tests/qemu-iotests/085.out       |  38 +--
 tests/qemu-iotests/144.out       |   4 +-
 tests/qemu-iotests/182.out       |   2 +-
 tests/qemu-iotests/185.out       |   8 +-
 tests/qemu-iotests/198.out       |   2 +
 tests/qemu-iotests/206.out       |   6 +-
 tests/qemu-iotests/242.out       |   5 +
 tests/qemu-iotests/255.out       |   8 +-
 tests/qemu-iotests/271           | 235 +++++++++++++
 tests/qemu-iotests/271.out       | 183 +++++++++++
 tests/qemu-iotests/common.filter |   1 +
 tests/qemu-iotests/group         |   1 +
 27 files changed, 1247 insertions(+), 340 deletions(-)
 create mode 100755 tests/qemu-iotests/271
 create mode 100644 tests/qemu-iotests/271.out
Signature · 1 line
2.20.1
#2 lore raw

Re: [RFC PATCH v2 01/26] qcow2: Add calculate_l2_meta()

In reply to Alberto Garcia
Quoted 31 lines · Alberto Garcia

handle_alloc() creates a QCowL2Meta structure in order to update the image metadata and perform the necessary copy-on-write operations.

This patch moves that code to a separate function so it can be used from other places.

---
  block/qcow2-cluster.c | 76 +++++++++++++++++++++++++++++--------------
  1 file changed, 52 insertions(+), 24 deletions(-)
block/qcow2-cluster.c +12 -0
index 8982b7b762..6c1dcdc781 100644
@@ -1019,6 +1019,55 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
                          QCOW2_DISCARD_NEVER);
  }
 
+/*
+ * For a given write request, create a new QCowL2Meta structure and
+ * add it to @m.
+ *
+ * @host_offset points to the beginning of the first cluster.
+ *
+ * @guest_offset and @bytes indicate the offset and length of the
+ * request.
+ *
+ * If @keep_old is true it means that the clusters were already
+ * allocated and will be overwritten. If false then the clusters are
+ * new and we have to decrease the reference count of the old ones.

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1043

Quoted 10 lines
block/qcow2-cluster.c +10 -0
+ */
+static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
+                              uint64_t guest_offset, uint64_t bytes,
+                              QCowL2Meta **m, bool keep_old)
+{
+    BDRVQcow2State *s = bs->opaque;
+    unsigned cow_start_from = 0;
+    unsigned cow_start_to = offset_into_cluster(s, guest_offset);
+    unsigned cow_end_from = cow_start_to + bytes;
+    unsigned cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);

New code is easier to read!

Interesting, seems QEMU_ALIGN_UP is more popular.. But comment above ROUND_UP makes sense in using it here..

Quoted 43 lines
block/qcow2-cluster.c +27 -9
+    unsigned nb_clusters = size_to_clusters(s, cow_end_from);
+    QCowL2Meta *old_m = *m;
+
+    *m = g_malloc0(sizeof(**m));
+    **m = (QCowL2Meta) {
+        .next           = old_m,
+
+        .alloc_offset   = host_offset,
+        .offset         = start_of_cluster(s, guest_offset),
+        .nb_clusters    = nb_clusters,
+
+        .keep_old_clusters = keep_old,
+
+        .cow_start = {
+            .offset     = cow_start_from,
+            .nb_bytes   = cow_start_to - cow_start_from,
+        },
+        .cow_end = {
+            .offset     = cow_end_from,
+            .nb_bytes   = cow_end_to - cow_end_from,
+        },
+    };
+
+    qemu_co_queue_init(&(*m)->dependent_requests);
+    QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
+}
+
  /*
   * Returns the number of contiguous clusters that can be used for an allocating
   * write, but require COW to be performed (this includes yet unallocated space,
@@ -1417,35 +1466,14 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
      uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset);
      int avail_bytes = nb_clusters << s->cluster_bits;
      int nb_bytes = MIN(requested_bytes, avail_bytes);
-    QCowL2Meta *old_m = *m;
-
-    *m = g_malloc0(sizeof(**m));
-
-    **m = (QCowL2Meta) {
-        .next           = old_m,
-
-        .alloc_offset   = alloc_cluster_offset,
-        .offset         = start_of_cluster(s, guest_offset),

block/qcow2-cluster.c static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, · line 1438

Quoted 10 lines
block/qcow2-cluster.c +0 -10
-        .nb_clusters    = nb_clusters,
-
-        .keep_old_clusters  = keep_old_clusters,
-
-        .cow_start = {
-            .offset     = 0,
-            .nb_bytes   = offset_into_cluster(s, guest_offset),
-        },
-        .cow_end = {
-            .offset     = nb_bytes,

hmm this logic is changed.. after the patch, it would be not nb_bytes, but

offset_into_cluster(s, guest_offset) + MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset))

Quoted 17 lines
block/qcow2-cluster.c +3 -5
-            .nb_bytes   = avail_bytes - nb_bytes,
-        },
-    };
-    qemu_co_queue_init(&(*m)->dependent_requests);
-    QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
 
      *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
      *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
      assert(*bytes != 0);
 
+    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
+                      m, keep_old_clusters);
+
      return 1;
 
  fail:
Signature · 2 lines
Best regards,
Vladimir
Reviewed-by on v2 2/26
#7 lore raw

Re: [RFC PATCH v2 02/26] qcow2: Split cluster_needs_cow() out of count_cow_clusters()

In reply to Alberto Garcia
Quoted 3 lines · Alberto Garcia

We are going to need it in other places.

Quoted 19 lines
---
  block/qcow2-cluster.c | 34 +++++++++++++++++++---------------
  1 file changed, 19 insertions(+), 15 deletions(-)
block/qcow2-cluster.c +7 -0
index 6c1dcdc781..aa1010a515 100644
@@ -1068,6 +1068,24 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
      QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
  }
 
+/* Returns true if writing to a cluster requires COW */
+static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry)
+{
+    switch (qcow2_get_cluster_type(bs, l2_entry)) {
+    case QCOW2_CLUSTER_NORMAL:
+        if (l2_entry & QCOW_OFLAG_COPIED) {
+            return false;

block/qcow2-cluster.c static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset, · line 1087

Quoted 10 lines
block/qcow2-cluster.c +10 -0
+        }
+    case QCOW2_CLUSTER_UNALLOCATED:
+    case QCOW2_CLUSTER_COMPRESSED:
+    case QCOW2_CLUSTER_ZERO_PLAIN:
+    case QCOW2_CLUSTER_ZERO_ALLOC:
+        return true;
+    default:
+        abort();
+    }
+}

Not sure how much is it better than just
return !(qcow2_get_cluster_type(bs, l2_entry) == QCOW2_CLUSTER_NORMAL && (l2_entry & QCOW_OFLAG_COPIED));

Quoted 32 lines
block/qcow2-cluster.c +2 -15
+
  /*
   * Returns the number of contiguous clusters that can be used for an allocating
   * write, but require COW to be performed (this includes yet unallocated space,
@@ -1080,25 +1098,11 @@ static int count_cow_clusters(BlockDriverState *bs, int nb_clusters,
 
      for (i = 0; i < nb_clusters; i++) {
          uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
-        QCow2ClusterType cluster_type = qcow2_get_cluster_type(bs, l2_entry);
-
-        switch(cluster_type) {
-        case QCOW2_CLUSTER_NORMAL:
-            if (l2_entry & QCOW_OFLAG_COPIED) {
-                goto out;
-            }
+        if (!cluster_needs_cow(bs, l2_entry)) {
              break;
-        case QCOW2_CLUSTER_UNALLOCATED:
-        case QCOW2_CLUSTER_COMPRESSED:
-        case QCOW2_CLUSTER_ZERO_PLAIN:
-        case QCOW2_CLUSTER_ZERO_ALLOC:
-            break;
-        default:
-            abort();
          }
      }
 
-out:
      assert(i <= nb_clusters);
      return i;
  }
Signature · 2 lines
Best regards,
Vladimir
#4 lore raw

Re: [RFC PATCH v2 01/26] qcow2: Add calculate_l2_meta()

In reply to Alberto Garcia
Quoted 16 lines · Alberto Garcia

handle_alloc() creates a QCowL2Meta structure in order to update the image metadata and perform the necessary copy-on-write operations.

This patch moves that code to a separate function so it can be used from other places.

---
 block/qcow2-cluster.c | 76 +++++++++++++++++++++++++++++--------------
 1 file changed, 52 insertions(+), 24 deletions(-)
block/qcow2-cluster.c +0 -0
index 8982b7b762..6c1dcdc781 100644

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1026

Quoted 8 lines
block/qcow2-cluster.c +5 -0
@@ -1019,6 +1019,55 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
                         QCOW2_DISCARD_NEVER);
 }
 
+/*
+ * For a given write request, create a new QCowL2Meta structure and
+ * add it to @m.
+ *
+ * @host_offset points to the beginning of the first cluster.

(I intended not to comment on such things on an RFC, but here I am...)

I’d call it host_cluster_offset to make clear that it points to a cluster and isn’t the host offset for @guest_offset.

And now that I’ve gone this far already, I might as well say that I’d like if it the comment noted that this function not only creates the L2Meta structure but also adds it to the cluster_allocs list.

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1036

Quoted 9 lines
block/qcow2-cluster.c +9 -0
+ * @guest_offset and @bytes indicate the offset and length of the
+ * request.
+ *
+ * If @keep_old is true it means that the clusters were already
+ * allocated and will be overwritten. If false then the clusters are
+ * new and we have to decrease the reference count of the old ones.
+ */
+static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
+                              uint64_t guest_offset, uint64_t bytes,

And now I’m so deep into non-RFC-level review territory, I might as well say I’d prefer @bytes to be an unsigned (or maybe even a plain int), because anything more wouldn’t work. (Not least because cow_end_to is an unsigned).

Sorry...

Max

Quoted 77 lines
block/qcow2-cluster.c +37 -24
+                              QCowL2Meta **m, bool keep_old)
+{
+    BDRVQcow2State *s = bs->opaque;
+    unsigned cow_start_from = 0;
+    unsigned cow_start_to = offset_into_cluster(s, guest_offset);
+    unsigned cow_end_from = cow_start_to + bytes;
+    unsigned cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
+    unsigned nb_clusters = size_to_clusters(s, cow_end_from);
+    QCowL2Meta *old_m = *m;
+
+    *m = g_malloc0(sizeof(**m));
+    **m = (QCowL2Meta) {
+        .next           = old_m,
+
+        .alloc_offset   = host_offset,
+        .offset         = start_of_cluster(s, guest_offset),
+        .nb_clusters    = nb_clusters,
+
+        .keep_old_clusters = keep_old,
+
+        .cow_start = {
+            .offset     = cow_start_from,
+            .nb_bytes   = cow_start_to - cow_start_from,
+        },
+        .cow_end = {
+            .offset     = cow_end_from,
+            .nb_bytes   = cow_end_to - cow_end_from,
+        },
+    };
+
+    qemu_co_queue_init(&(*m)->dependent_requests);
+    QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
+}
+
 /*
  * Returns the number of contiguous clusters that can be used for an allocating
  * write, but require COW to be performed (this includes yet unallocated space,
@@ -1417,35 +1466,14 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     uint64_t requested_bytes = *bytes + offset_into_cluster(s, guest_offset);
     int avail_bytes = nb_clusters << s->cluster_bits;
     int nb_bytes = MIN(requested_bytes, avail_bytes);
-    QCowL2Meta *old_m = *m;
-
-    *m = g_malloc0(sizeof(**m));
-
-    **m = (QCowL2Meta) {
-        .next           = old_m,
-
-        .alloc_offset   = alloc_cluster_offset,
-        .offset         = start_of_cluster(s, guest_offset),
-        .nb_clusters    = nb_clusters,
-
-        .keep_old_clusters  = keep_old_clusters,
-
-        .cow_start = {
-            .offset     = 0,
-            .nb_bytes   = offset_into_cluster(s, guest_offset),
-        },
-        .cow_end = {
-            .offset     = nb_bytes,
-            .nb_bytes   = avail_bytes - nb_bytes,
-        },
-    };
-    qemu_co_queue_init(&(*m)->dependent_requests);
-    QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
 
     *host_offset = alloc_cluster_offset + offset_into_cluster(s, guest_offset);
     *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
     assert(*bytes != 0);
 
+    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
+                      m, keep_old_clusters);
+
     return 1;
 
 fail:
#9 lore raw

Re: [RFC PATCH v2 03/26] qcow2: Process QCOW2_CLUSTER_ZERO_ALLOC clusters in handle_copied()

In reply to Alberto Garcia
Quoted 68 lines · Alberto Garcia

When writing to a qcow2 file there are two functions that take a virtual offset and return a host offset, possibly allocating new clusters if necessary:

- handle_copied() looks for normal data clusters that are already allocated and have a reference count of 1. In those clusters we can simply write the data and there is no need to perform any copy-on-write.

- handle_alloc() looks for clusters that do need copy-on-write, either because they haven't been allocated yet, because their reference count is != 1 or because they are ZERO_ALLOC clusters.

The ZERO_ALLOC case is a bit special because those are clusters that are already allocated and they could perfectly be dealt with in handle_copied() (as long as copy-on-write is performed when required).

In fact, there is extra code specifically for them in handle_alloc() that tries to reuse the existing allocation if possible and frees them otherwise.

This patch changes the handling of ZERO_ALLOC clusters so the semantics of these two functions are now like this:

- handle_copied() looks for clusters that are already allocated and which we can overwrite (NORMAL and ZERO_ALLOC clusters with a reference count of 1).

- handle_alloc() looks for clusters for which we need a new allocation (all other cases).

One importante difference after this change is that clusters found in handle_copied() may now require copy-on-write, but this will be anyway necessary once we add support for subclusters.

---
 block/qcow2-cluster.c | 177 +++++++++++++++++++++++-------------------
 1 file changed, 96 insertions(+), 81 deletions(-)
block/qcow2-cluster.c +6 -3
index aa1010a515..ee6b46f917 100644
@@ -1021,7 +1021,8 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
 
 /*
  * For a given write request, create a new QCowL2Meta structure and
- * add it to @m.
+ * add it to @m. If the write request does not need copy-on-write or
+ * changes to the L2 metadata then this function does nothing.
  *
  * @host_offset points to the beginning of the first cluster.
  *
@@ -1034,15 +1035,51 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
  */
 static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
                               uint64_t guest_offset, uint64_t bytes,
-                              QCowL2Meta **m, bool keep_old)
+                              uint64_t *l2_slice, QCowL2Meta **m, bool keep_old)
 {
     BDRVQcow2State *s = bs->opaque;
-    unsigned cow_start_from = 0;
+    int l2_index = offset_to_l2_slice_index(s, guest_offset);
+    uint64_t l2_entry;
+    unsigned cow_start_from, cow_end_to;
     unsigned cow_start_to = offset_into_cluster(s, guest_offset);
     unsigned cow_end_from = cow_start_to + bytes;

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1054

Quoted 10 lines
block/qcow2-cluster.c +7 -1
-    unsigned cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
     unsigned nb_clusters = size_to_clusters(s, cow_end_from);
     QCowL2Meta *old_m = *m;
+    QCow2ClusterType type;
+
+    /* Return if there's no COW (all clusters are normal and we keep them) */
+    if (keep_old) {
+        int i;
+        for (i = 0; i < nb_clusters; i++) {
+            l2_entry = be64_to_cpu(l2_slice[l2_index + i]);

I’d assert somewhere that l2_index + nb_clusters - 1 won’t overflow.

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1055

Quoted 1 line
block/qcow2-cluster.c +1 -0
+            if (qcow2_get_cluster_type(bs, l2_entry) != QCOW2_CLUSTER_NORMAL) {

Wouldn’t cluster_needs_cow() be better?

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1062

Quoted 7 lines
block/qcow2-cluster.c +7 -0
+                break;
+            }
+        }
+        if (i == nb_clusters) {
+            return;
+        }
+    }

So I understand we always need to create an L2Meta structure in all other cases because we at least need to turn those clusters into normal clusters? (Even if they’re already allocated, as in the case of allocated zero clusters.)

Quoted 26 lines
block/qcow2-cluster.c +20 -0
+
+    /* Get the L2 entry from the first cluster */
+    l2_entry = be64_to_cpu(l2_slice[l2_index]);
+    type = qcow2_get_cluster_type(bs, l2_entry);
+
+    if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
+        cow_start_from = cow_start_to;
+    } else {
+        cow_start_from = 0;
+    }
+
+    /* Get the L2 entry from the last cluster */
+    l2_entry = be64_to_cpu(l2_slice[l2_index + nb_clusters - 1]);
+    type = qcow2_get_cluster_type(bs, l2_entry);
+
+    if (type == QCOW2_CLUSTER_NORMAL && keep_old) {
+        cow_end_to = cow_end_from;
+    } else {
+        cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
+    }
 
     *m = g_malloc0(sizeof(**m));
     **m = (QCowL2Meta) {
@@ -1068,18 +1105,18 @@ static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
     QLIST_INSERT_HEAD(&s->cluster_allocs, *m, next_in_flight);
 }

block/qcow2-cluster.c static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset, · line 1115

Quoted 10 lines
block/qcow2-cluster.c +2 -1
 
-/* Returns true if writing to a cluster requires COW */
+/* Returns true if the cluster is unallocated or has refcount > 1 */
 static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry)
 {
     switch (qcow2_get_cluster_type(bs, l2_entry)) {
     case QCOW2_CLUSTER_NORMAL:
+    case QCOW2_CLUSTER_ZERO_ALLOC:
         if (l2_entry & QCOW_OFLAG_COPIED) {
             return false;

Don’t zero-allocated clusters need COW always? (Because the at the given host offset isn’t guaranteed to be zero.)

Quoted 11 lines
block/qcow2-cluster.c +0 -1
         }
     case QCOW2_CLUSTER_UNALLOCATED:
     case QCOW2_CLUSTER_COMPRESSED:
     case QCOW2_CLUSTER_ZERO_PLAIN:
-    case QCOW2_CLUSTER_ZERO_ALLOC:
         return true;
     default:
         abort();
@@ -1087,20 +1124,34 @@ static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry)
 }
 

block/qcow2-cluster.c static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry) · line 1132

Quoted 10 lines
block/qcow2-cluster.c +6 -3
 /*
- * Returns the number of contiguous clusters that can be used for an allocating
- * write, but require COW to be performed (this includes yet unallocated space,
- * which must copy from the backing file)
+ * Returns the number of contiguous clusters that can be written to
+ * using one single write request, starting from @l2_index.
+ * At most @nb_clusters are checked.
+ *
+ * If @want_cow is true this counts clusters that are either
+ * unallocated, or allocated but with refcount > 1.

+(So they need to be newly allocated and COWed)?

(Or is the past participle of COW COWn? Or maybe CedOW?)

block/qcow2-cluster.c static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry) · line 1135

Quoted 3 lines
block/qcow2-cluster.c +3 -0
+ *
+ * If @want_cow is false this counts clusters that are already
+ * allocated and can be written to using their current locations

s/using their current locations/in-place/?

block/qcow2-cluster.c static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry) · line 1138

Quoted 3 lines
block/qcow2-cluster.c +1 -0
+ * (including QCOW2_CLUSTER_ZERO_ALLOC).
  */
 static int count_cow_clusters(BlockDriverState *bs, int nb_clusters,

Hm, well, it’s not really cow anymore. Would
count_single_write_clusters() work?

Quoted 55 lines
block/qcow2-cluster.c +16 -13
-    uint64_t *l2_slice, int l2_index)
+                              uint64_t *l2_slice, int l2_index, bool want_cow)
 {
+    BDRVQcow2State *s = bs->opaque;
+    uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index]);
+    uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK;
     int i;
 
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
-        if (!cluster_needs_cow(bs, l2_entry)) {
+        l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
+        if (cluster_needs_cow(bs, l2_entry) != want_cow) {
             break;
         }
+        if (!want_cow && expected_offset != (l2_entry & L2E_OFFSET_MASK)) {
+            break;
+        }
+        expected_offset += s->cluster_size;
     }
 
     assert(i <= nb_clusters);
@@ -1228,18 +1279,17 @@ static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
 
     cluster_offset = be64_to_cpu(l2_slice[l2_index]);
 
-    /* Check how many clusters are already allocated and don't need COW */
-    if (qcow2_get_cluster_type(bs, cluster_offset) == QCOW2_CLUSTER_NORMAL
-        && (cluster_offset & QCOW_OFLAG_COPIED))
-    {
+    if (!cluster_needs_cow(bs, cluster_offset)) {
         /* If a specific host_offset is required, check it */
         bool offset_matches =
             (cluster_offset & L2E_OFFSET_MASK) == *host_offset;
 
         if (offset_into_cluster(s, cluster_offset & L2E_OFFSET_MASK)) {
-            qcow2_signal_corruption(bs, true, -1, -1, "Data cluster offset "
+            qcow2_signal_corruption(bs, true, -1, -1, "%s cluster offset "
                                     "%#llx unaligned (guest offset: %#" PRIx64
-                                    ")", cluster_offset & L2E_OFFSET_MASK,
+                                    ")", cluster_offset & QCOW_OFLAG_ZERO ?
+                                    "Preallocated zero" : "Data",
+                                    cluster_offset & L2E_OFFSET_MASK,
                                     guest_offset);
             ret = -EIO;
             goto out;
@@ -1252,15 +1302,17 @@ static int handle_copied(BlockDriverState *bs, uint64_t guest_offset,
         }
 
         /* We keep all QCOW_OFLAG_COPIED clusters */
-        keep_clusters =
-            count_contiguous_clusters(bs, nb_clusters, s->cluster_size,
-                                      &l2_slice[l2_index],
-                                      QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
+        keep_clusters = count_cow_clusters(bs, nb_clusters, l2_slice,

block/qcow2-cluster.c static int handle_copied(BlockDriverState *bs, uint64_t guest_offset, · line 1315

Quoted 10 lines
block/qcow2-cluster.c +5 -0
+                                           l2_index, false);
         assert(keep_clusters <= nb_clusters);
 
         *bytes = MIN(*bytes,
                  keep_clusters * s->cluster_size
                  - offset_into_cluster(s, guest_offset));
+        assert(*bytes != 0);
+
+        calculate_l2_meta(bs, cluster_offset & L2E_OFFSET_MASK, guest_offset,
+                          *bytes, l2_slice, m, true);

We wouldn’t need calculate_l2_meta() here if the clusters really all didn’t need COW.

We do need it because cluster_needs_cow() returns false for zero clusters, which isn’t what the function name says.

Quoted 51 lines
block/qcow2-cluster.c +2 -26
         ret = 1;
     } else {
@@ -1361,12 +1413,10 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     BDRVQcow2State *s = bs->opaque;
     int l2_index;
     uint64_t *l2_slice;
-    uint64_t entry;
     uint64_t nb_clusters;
     int ret;
-    bool keep_old_clusters = false;
 
-    uint64_t alloc_cluster_offset = INV_OFFSET;
+    uint64_t alloc_cluster_offset;
 
     trace_qcow2_handle_alloc(qemu_coroutine_self(), guest_offset, *host_offset,
                              *bytes);
@@ -1392,67 +1442,31 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
         return ret;
     }
 
-    entry = be64_to_cpu(l2_slice[l2_index]);
-    nb_clusters = count_cow_clusters(bs, nb_clusters, l2_slice, l2_index);
+    nb_clusters = count_cow_clusters(bs, nb_clusters, l2_slice, l2_index, true);
 
     /* This function is only called when there were no non-COW clusters, so if
      * we can't find any unallocated or COW clusters either, something is
      * wrong with our code. */
     assert(nb_clusters > 0);
 
-    if (qcow2_get_cluster_type(bs, entry) == QCOW2_CLUSTER_ZERO_ALLOC &&
-        (entry & QCOW_OFLAG_COPIED) &&
-        (*host_offset == INV_OFFSET ||
-         start_of_cluster(s, *host_offset) == (entry & L2E_OFFSET_MASK)))
-    {
-        int preallocated_nb_clusters;
-
-        if (offset_into_cluster(s, entry & L2E_OFFSET_MASK)) {
-            qcow2_signal_corruption(bs, true, -1, -1, "Preallocated zero "
-                                    "cluster offset %#llx unaligned (guest "
-                                    "offset: %#" PRIx64 ")",
-                                    entry & L2E_OFFSET_MASK, guest_offset);
-            ret = -EIO;
-            goto fail;
-        }
-
-        /* Try to reuse preallocated zero clusters; contiguous normal clusters
-         * would be fine, too, but count_cow_clusters() above has limited
-         * nb_clusters already to a range of COW clusters */
-        preallocated_nb_clusters =
-            count_contiguous_clusters(bs, nb_clusters, s->cluster_size,

block/qcow2-cluster.c static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, · line 1452

Quoted 10 lines
block/qcow2-cluster.c +1 -9
-                                      &l2_slice[l2_index], QCOW_OFLAG_COPIED);
-        assert(preallocated_nb_clusters > 0);
-
-        nb_clusters = preallocated_nb_clusters;
-        alloc_cluster_offset = entry & L2E_OFFSET_MASK;
-
-        /* We want to reuse these clusters, so qcow2_alloc_cluster_link_l2()
-         * should not free them. */
-        keep_old_clusters = true;
+    /* Allocate, if necessary at a given offset in the image file */

Well, it’s always necessary now.

Quoted 45 lines
block/qcow2-cluster.c +13 -21
+    alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET :
+        start_of_cluster(s, *host_offset);
+    ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
+                                  &nb_clusters);
+    if (ret < 0) {
+        goto out;
     }
 
-    qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
-
-    if (alloc_cluster_offset == INV_OFFSET) {
-        /* Allocate, if necessary at a given offset in the image file */
-        alloc_cluster_offset = *host_offset == INV_OFFSET ? INV_OFFSET :
-                               start_of_cluster(s, *host_offset);
-        ret = do_alloc_cluster_offset(bs, guest_offset, &alloc_cluster_offset,
-                                      &nb_clusters);
-        if (ret < 0) {
-            goto fail;
-        }
-
-        /* Can't extend contiguous allocation */
-        if (nb_clusters == 0) {
-            *bytes = 0;
-            return 0;
-        }
-
-        assert(alloc_cluster_offset != INV_OFFSET);
+    /* Can't extend contiguous allocation */
+    if (nb_clusters == 0) {
+        *bytes = 0;
+        ret = 0;
+        goto out;
     }
 
+    assert(alloc_cluster_offset != INV_OFFSET);
+
     /*
      * Save info needed for meta data update.
      *
@@ -1475,13 +1489,14 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
     assert(*bytes != 0);
 
-    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
-                      m, keep_old_clusters);

block/qcow2-cluster.c static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, · line 1498

Quoted 10 lines
block/qcow2-cluster.c +5 -3
+    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes, l2_slice,
+                      m, false);
 
-    return 1;
+    ret = 1;
 
-fail:
-    if (*m && (*m)->nb_clusters > 0) {
+out:
+    qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);

Is this a bug fix?

Max

Quoted 5 lines
block/qcow2-cluster.c +1 -0
+    if (ret < 0 && *m && (*m)->nb_clusters > 0) {
         QLIST_REMOVE(*m, next_in_flight);
     }
     return ret;
#3 lore raw

Re: [RFC PATCH v2 01/26] qcow2: Add calculate_l2_meta()

In reply to Vladimir Sementsov-Ogievskiy
Quoted 6 lines · Vladimir Sementsov-Ogievskiy

block/qcow2-cluster.c static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, · line 1438

Quoted 2 lines
block/qcow2-cluster.c +0 -2
-        .cow_end = {
-            .offset     = nb_bytes,

hmm this logic is changed.. after the patch, it would be not nb_bytes, but

offset_into_cluster(s, guest_offset) + MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset))

It hasn't changed. The value of .cow_end.offset before the patch is nb_bytes (these two lines are equivalent):

  nb_bytes = MIN(requested_bytes, avail_bytes);
  nb_bytes = MIN(*bytes + offset_into_cluster, avail_bytes);

After the patch we update *bytes to pass it to calculate_l2_meta(). Here is the value (again, these are all equivalent):

*bytes = MIN(*bytes, nb_bytes - offset_into_cluster) *bytes = MIN(*bytes, MIN(*bytes + offset_into_cluster, avail_bytes) -
offset_into_cluster)
*bytes = MIN(*bytes, MIN(*bytes, avail_bytes - offset_into_cluster)) *bytes = MIN(*bytes, avail_bytes - offset_into_cluster)

And here's the value of .cow_end.offset set in calculate_l2_meta():

cow_end_from = *bytes + offset_into_cluster cow_end_from = MIN(*bytes, avail_bytes - offset_into_cluster) +
offset_into_cluster
cow_end_from = MIN(*bytes + offset_into_cluster, avail_bytes)

This last definition of cow_end_from is the same as nb_bytes as used before the patch.

Berto

#5 lore raw

Re: [RFC PATCH v2 01/26] qcow2: Add calculate_l2_meta()

In reply to Max Reitz
Quoted 1 line · Max Reitz

(I intended not to comment on such things on an RFC, but here I am...)

No problem with that :-)

Quoted 2 lines

I’d call it host_cluster_offset to make clear that it points to a cluster and isn’t the host offset for @guest_offset.

Sure, why not. We can also accept the exact offset within the cluster and then use start_of_cluster(), but I prefer this one.

Quoted 3 lines

And now that I’ve gone this far already, I might as well say that I’d like if it the comment noted that this function not only creates the L2Meta structure but also adds it to the cluster_allocs list.

I'll update the comment.

Quoted 10 lines

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1036

Quoted 9 lines

+ * @guest_offset and @bytes indicate the offset and length of the
+ * request.
+ *
+ * If @keep_old is true it means that the clusters were already
+ * allocated and will be overwritten. If false then the clusters are
+ * new and we have to decrease the reference count of the old ones.
+ */

block/qcow2-cluster.c +2 -0
+static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
+                              uint64_t guest_offset, uint64_t bytes,
Quoted 4 lines

And now I’m so deep into non-RFC-level review territory, I might as well say I’d prefer @bytes to be an unsigned (or maybe even a plain int), because anything more wouldn’t work. (Not least because cow_end_to is an unsigned).

True, I'll correct that too.

Thanks!

Berto

#13 lore raw

Re: [RFC PATCH v2 05/26] qcow2: Document the Extended L2 Entries feature

In reply to Alberto Garcia
Quoted 4 lines · Alberto Garcia

Subcluster allocation in qcow2 is implemented by extending the existing L2 table entries and adding additional information to indicate the allocation status of each subcluster.

Quoted 8 lines

This patch documents the changes to the qcow2 format and how they affect the calculation of the L2 cache size.

---
 docs/interop/qcow2.txt | 68 ++++++++++++++++++++++++++++++++++++++++--
 docs/qcow2-cache.txt   | 19 +++++++++++-
 2 files changed, 83 insertions(+), 4 deletions(-)

Sounds good, just a bit of bit nit-picking from my side.

Quoted 4 lines
docs/interop/qcow2.txt +0 -0
index af5711e533..d34261f955 100644

[...]

Quoted 29 lines
docs/interop/qcow2.txt +25 -0
@@ -524,6 +535,57 @@ file (except if bit 0 in the Standard Cluster Descriptor is set). If there is
 no backing file or the backing file is smaller than the image, they shall read
 zeros for all parts that are not covered by the backing file.
 
+== Extended L2 Entries ==
+
+An image uses Extended L2 Entries if bit 3 is set on the incompatible_features
+field of the header.
+
+In these images standard data clusters are divided into 32 subclusters of the
+same size. They are contiguous and start from the beginning of the cluster.
+Subclusters can be allocated independently and the L2 entry contains information
+indicating the status of each one of them. Compressed data clusters don't have
+subclusters so they are treated like in images without this feature.
+
+The size of an extended L2 entry is 128 bits so the number of entries per table
+is calculated using this formula:
+
+    l2_entries = (cluster_size / (2 * sizeof(uint64_t)))
+
+The first 64 bits have the same format as the standard L2 table entry described
+in the previous section, with the exception of bit 0 of the standard cluster
+descriptor.
+
+The last 64 bits contain a subcluster allocation bitmap with this format:
+
+Subcluster Allocation Bitmap (for standard clusters):
+
+    Bit  0 -  31:   Allocation status (one bit per subcluster)

docs/interop/qcow2.txt file (except if bit 0 in the Standard Cluster Descriptor is set). If there is · line 572

Quoted 10 lines
docs/interop/qcow2.txt +10 -0
+
+                    1: the subcluster is allocated. In this case the
+                       host cluster offset field must contain a valid
+                       offset.
+                    0: the subcluster is not allocated. In this case
+                       read requests shall go to the backing file or
+                       return zeros if there is no backing file data.
+
+                    Bits are assigned starting from the most significant one.
+                    (i.e. bit x is used for subcluster 31 - x)

I seem to remember that someone proposed this bit ordering to you, but I wonder why. So far everything in qcow2 starts from the least significant bit, for example refcounts (“If refcount_bits implies a sub-byte width, note that bit 0 means the least significant bit in this context”), feature bits, and sub-byte structure descriptions in general (which you reference directly with “bit x”).

Soo... What’s the reason for doing it the other way around here?

docs/interop/qcow2.txt file (except if bit 0 in the Standard Cluster Descriptor is set). If there is · line 584

Quoted 11 lines
docs/interop/qcow2.txt +11 -0
+        32 -  63    Subcluster reads as zeros (one bit per subcluster)
+
+                    1: the subcluster reads as zeros. In this case the
+                       allocation status bit must be unset. The host
+                       cluster offset field may or may not be set.
+                    0: no effect.
+
+                    Bits are assigned starting from the most significant one.
+                    (i.e. bit x is used for subcluster 63 - x)
+
+Subcluster Allocation Bitmap (for compressed clusters):

Going from nit-picking to bike shedding: I’d drop the parentheses.

Max

#19 lore raw

Re: [RFC PATCH v2 09/26] qcow2: Add l2_entry_size()

In reply to Alberto Garcia
Quoted 7 lines · Alberto Garcia

qcow2 images with subclusters have 128-bit L2 entries. The first 64 bits contain the same information as traditional images and the last 64 bits form a bitmap with the status of each individual subcluster.

Because of that we cannot assume that L2 entries are sizeof(uint64_t) anymore. This function returns the proper value for the image.

Quoted 7 lines
---
 block/qcow2-cluster.c  | 12 ++++++------
 block/qcow2-refcount.c | 14 ++++++++------
 block/qcow2.c          |  6 +++---
 block/qcow2.h          |  5 +++++
 4 files changed, 22 insertions(+), 15 deletions(-)

I suppose qcow2_calc_prealloc_size(), qcow2_co_truncate() (nb_new_l2_tables), and qcow2_measure() (l2_tables) also need some adjustment.

Max

#21 lore raw

Re: [RFC PATCH v2 10/26] qcow2: Update get/set_l2_entry() and add get/set_l2_bitmap()

In reply to Alberto Garcia
Quoted 10 lines · Alberto Garcia

Extended L2 entries are 128-bit wide: 64 bits for the entry itself and 64 bits for the subcluster allocation bitmap.

In order to support them correctly get/set_l2_entry() need to be updated so they take the entry width into account in order to calculate the correct offset.

This patch also adds the get/set_l2_bitmap() functions that are used to access the bitmaps. For convenience, these functions are no-ops when used in traditional qcow2 images.

Granted, I haven’t seen the following patches yet, but if these functions are indeed called for images that don’t have subclusters, shouldn’t they return 0x0*0f*f then? (i.e. everything allocated)

If they aren’t, they should probably just abort(). Well, set_l2_bitmap() should probably always abort() if there aren’t any subclusters.

Max

Quoted 48 lines
---
 block/qcow2.h | 22 ++++++++++++++++++++++
 1 file changed, 22 insertions(+)
block/qcow2.h +22 -0
index 29a253bfb9..741c41c80f 100644
@@ -507,15 +507,37 @@ static inline size_t l2_entry_size(BDRVQcow2State *s)
 static inline uint64_t get_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
                                     int idx)
 {
+    idx *= l2_entry_size(s) / sizeof(uint64_t);
     return be64_to_cpu(l2_slice[idx]);
 }
 
+static inline uint64_t get_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice,
+                                     int idx)
+{
+    if (has_subclusters(s)) {
+        idx *= l2_entry_size(s) / sizeof(uint64_t);
+        return be64_to_cpu(l2_slice[idx + 1]);
+    } else {
+        return 0;
+    }
+}
+
 static inline void set_l2_entry(BDRVQcow2State *s, uint64_t *l2_slice,
                                 int idx, uint64_t entry)
 {
+    idx *= l2_entry_size(s) / sizeof(uint64_t);
     l2_slice[idx] = cpu_to_be64(entry);
 }
 
+static inline void set_l2_bitmap(BDRVQcow2State *s, uint64_t *l2_slice,
+                                 int idx, uint64_t bitmap)
+{
+    if (has_subclusters(s)) {
+        idx *= l2_entry_size(s) / sizeof(uint64_t);
+        l2_slice[idx + 1] = cpu_to_be64(bitmap);
+    }
+}
+
 static inline bool has_data_file(BlockDriverState *bs)
 {
     BDRVQcow2State *s = bs->opaque;
#14 lore raw

Re: [RFC PATCH v2 05/26] qcow2: Document the Extended L2 Entries feature

In reply to Max Reitz
Quoted 21 lines · Max Reitz

docs/interop/qcow2.txt file (except if bit 0 in the Standard Cluster Descriptor is set). If there is · line 572

Quoted 13 lines
docs/interop/qcow2.txt +13 -0
+Subcluster Allocation Bitmap (for standard clusters):
+
+    Bit  0 -  31:   Allocation status (one bit per subcluster)
+
+                    1: the subcluster is allocated. In this case the
+                       host cluster offset field must contain a valid
+                       offset.
+                    0: the subcluster is not allocated. In this case
+                       read requests shall go to the backing file or
+                       return zeros if there is no backing file data.
+
+                    Bits are assigned starting from the most significant one.
+                    (i.e. bit x is used for subcluster 31 - x)

I seem to remember that someone proposed this bit ordering to you, but I wonder why. So far everything in qcow2 starts from the least significant bit, for example refcounts (“If refcount_bits implies a sub-byte width, note that bit 0 means the least significant bit in this context”), feature bits, and sub-byte structure descriptions in general (which you reference directly with “bit x”).

Quoted 1 line

Soo... What’s the reason for doing it the other way around here?

The reason is that I thought that it would be better for debugging purposes. If I do an hexdump of the L2 table to see what's going on then starting from the most significant bit gives me a better visual image of what subclusters are allocated.

In other words, if the first two subclusters are allocated I think this representation

               11000000 00000000 00000000 00000000  (c0 00 00 00)

is more natural than this one

               00000000 00000000 00000000 00000011  (00 00 00 03)

But I don't have a very strong opinion so I'm open to changing it.

Berto

#24 lore raw

Re: [RFC PATCH v2 11/26] qcow2: Add qcow2_get_subcluster_type()

In reply to Alberto Garcia
Quoted 38 lines · Alberto Garcia

This function returns the type of an individual subcluster. If an image does not have subclusters then this returns the exact same value as qcow2_get_cluster_type().

The information in standard and extended L2 entries is encoded in a slightly different way, but all existing QCow2ClusterType values are also valid for subclusters and have the same meanings (although they typically only apply to the requested subcluster).

There are two important exceptions to this:

a) QCOW2_CLUSTER_COMPRESSED means that the whole cluster is compressed. We do not support compression at the subcluster level.

b) QCOW2_CLUSTER_UNALLOCATED means that the cluster is unallocated, that is, the offset field of the L2 entry does not point to a host cluster. All subclusters are obviously unallocated too but any of them could be of type QCOW2_CLUSTER_ZERO_PLAIN.

In addition to that, extended L2 entries allow one new scenario where the cluster is normally allocated but an individual subcluster is not. This is very different from (b) and because of that this patch adds a new value called QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER.

As a last thing, this patch adds QCOW2_CLUSTER_INVALID to detect the cases where an L2 entry has a value that violates the spec. The caller is responsible for handling these situations.

To prevent compatibility problems with images that have invalid values but are currently being read by QEMU without causing side effects, QCOW2_CLUSTER_INVALID is only returned for images with extended L2 entries.

---
 block/qcow2.h | 62 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 62 insertions(+)

[...]

block/qcow2.h static inline QCow2ClusterType qcow2_get_cluster_type(BlockDriverState *bs, · line 657

Quoted 3 lines
block/qcow2.h +3 -0
+        case QCOW2_CLUSTER_ZERO_PLAIN:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+            return QCOW2_CLUSTER_INVALID;

The spec doesn’t say anything about this, though.

(I would have assumed it’s valid and the subcluster zero status is simply ignored. It makes sense to forbit this case, but the spec should make a note of it.)

Nax

#27 lore raw

Re: [RFC PATCH v2 12/26] qcow2: Handle QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER

In reply to Alberto Garcia
Quoted 9 lines · Alberto Garcia

In the previous patch we added a new QCow2ClusterType named QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER. There is a couple of places where this new value needs to be handled, and that is what this patch does.

---
 block/qcow2.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

This patch deals with everything in qcow2.c. There are more places that reference QCOW2_CLUSTER_* constants elsewhere, and I suppose most of them are handled by the following patches.

But I wonder what the criterion is on where it needs to be handled and where it’s OK not to. Right now it looks to me like it’s a bit arbitrary maybe? But I suppose I’ll just have to wait until after the next patches.

Max

#28 lore raw

Re: [RFC PATCH v2 12/26] qcow2: Handle QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER

Quoted 10 lines · Max Reitz
Quoted 9 lines · Alberto Garcia

In the previous patch we added a new QCow2ClusterType named QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER. There is a couple of places where this new value needs to be handled, and that is what this patch does.

---
 block/qcow2.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
Quoted 8 lines

This patch deals with everything in qcow2.c. There are more places that reference QCOW2_CLUSTER_* constants elsewhere, and I suppose most of them are handled by the following patches.

But I wonder what the criterion is on where it needs to be handled and where it’s OK not to. Right now it looks to me like it’s a bit arbitrary maybe? But I suppose I’ll just have to wait until after the next patches.

This is the part of the series that I'm the least happy about, because the existing qcow2_get_cluster_type() can never return this new value, I only updated the cases where this can actually happen.

I'm still considering a different approach for this.

Berto

#29 lore raw

Re: [RFC PATCH v2 12/26] qcow2: Handle QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER

Quoted 20 lines · Alberto Garcia
Quoted 10 lines · Max Reitz
Quoted 9 lines · Alberto Garcia

In the previous patch we added a new QCow2ClusterType named QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER. There is a couple of places where this new value needs to be handled, and that is what this patch does.

---
 block/qcow2.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
Quoted 8 lines

This patch deals with everything in qcow2.c. There are more places that reference QCOW2_CLUSTER_* constants elsewhere, and I suppose most of them are handled by the following patches.

But I wonder what the criterion is on where it needs to be handled and where it’s OK not to. Right now it looks to me like it’s a bit arbitrary maybe? But I suppose I’ll just have to wait until after the next patches.

Quoted 5 lines

This is the part of the series that I'm the least happy about, because the existing qcow2_get_cluster_type() can never return this new value, I only updated the cases where this can actually happen.

I'm still considering a different approach for this.

I still don’t know what you’re doing in the later patches, but to me it looks a bit like you don’t dare breaking up the existing structure that just deals with clusters.

If that is so, I think it will help to make a clear cut between what concerns subclusters and what concerns clusters at a whole. QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER shouldn’t be in QCow2ClusterType; there should be a separate QCow2SubclusterType.

OTOH, that would require more modifications, but (naïvely) I believe that would make for the cleaner interface in the end.

Max

#32 lore raw

Re: [RFC PATCH v2 13/26] qcow2: Add subcluster support to calculate_l2_meta()

In reply to Alberto Garcia
Quoted 6 lines · Alberto Garcia

If an image has subclusters then there are more copy-on-write scenarios that we need to consider. Let's say we have a write request from the middle of subcluster #3 until the end of the cluster:

- If the cluster is new, then subclusters #0 to #3 from the old cluster must be copied into the new one.

You mean for snapshots?

(That isn’t quite clear, and I only guess this based on the next bullet point which differentiates based on “the old cluster was unallocated”. That’s weird, too, because what does that mean, old cluster and new cluster? I suppose it’s abstract and it just means “There was no old cluster and now we’ve allocated something”. I can only understand the concept of old and new clusters for COW inside of an image, i.e. for snapshots and compressed clusters (theoretically).)

Quoted 21 lines

- If the cluster is new but the old cluster was unallocated, then only subcluster #3 needs copy-on-write. #0 to #2 are marked as unallocated in the bitmap of the new L2 entry.

- If we are overwriting an old cluster and subcluster #3 is unallocated or has the all-zeroes bit set then we need copy-on-write on subcluster #3.

- If we are overwriting an old cluster and subcluster #3 was allocated then there is no need to copy-on-write.

---
 block/qcow2-cluster.c | 136 +++++++++++++++++++++++++++++++++---------
 1 file changed, 108 insertions(+), 28 deletions(-)
block/qcow2-cluster.c +0 -0
index 1f509bda15..990bc070af 100644

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1038

Quoted 5 lines
block/qcow2-cluster.c +2 -0
@@ -1034,14 +1034,16 @@ void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m)
  * If @keep_old is true it means that the clusters were already
  * allocated and will be overwritten. If false then the clusters are
  * new and we have to decrease the reference count of the old ones.
+ *
+ * Returns 1 on success, -errno on failure.

I think there should be a note here on why this doesn’t follow the general 0/-errno schema (i.e., “, because that is what callers generally expect”).

Quoted 1 line
block/qcow2-cluster.c +0 -0
  */

[...]

block/qcow2-cluster.c static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset, · line 1095

Quoted 7 lines
block/qcow2-cluster.c +7 -0
+    if (!keep_old) {
+        switch (type) {
+        case QCOW2_CLUSTER_NORMAL:
+        case QCOW2_CLUSTER_COMPRESSED:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+        case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
+            cow_start_from = 0;

Somehow (I don’t know why) I find this a bit tough to understand.

Wouldn’t it work to let cow_start start from the first subcluster for ZERO_ALLOC and UNALLOCATED_SUBCLUSTER? We don’t need to COW those, it should be sufficient to just make the subclusters before that zero or unallocated, respectively.

(Same for cow_end)

Max

Quoted 8 lines
block/qcow2-cluster.c +8 -0
+            break;
+        case QCOW2_CLUSTER_ZERO_PLAIN:
+        case QCOW2_CLUSTER_UNALLOCATED:
+            cow_start_from = sc_index << s->subcluster_bits;
+            break;
+        default:
+            g_assert_not_reached();
+        }
#35 lore raw

Re: [RFC PATCH v2 14/26] qcow2: Add subcluster support to qcow2_get_cluster_offset()

In reply to Alberto Garcia
Quoted 16 lines · Alberto Garcia

The logic of this function remains pretty much the same, except that it uses count_contiguous_subclusters(), which combines the logic of count_contiguous_clusters() / count_contiguous_clusters_unallocated() and checks individual subclusters.

---
 block/qcow2-cluster.c | 111 ++++++++++++++++++++----------------------
 1 file changed, 52 insertions(+), 59 deletions(-)
block/qcow2-cluster.c +0 -0
index 990bc070af..e67559152f 100644
@@ -372,66 +372,51 @@ fail:
 }

block/qcow2-cluster.c fail: · line 377

Quoted 10 lines
block/qcow2-cluster.c +3 -5
 
 /*
- * Checks how many clusters in a given L2 slice are contiguous in the image
- * file. As soon as one of the flags in the bitmask stop_flags changes compared
- * to the first cluster, the search is stopped and the cluster is not counted
- * as contiguous. (This allows it, for example, to stop at the first compressed
- * cluster which may require a different handling)
+ * Return the number of contiguous subclusters of the exact same type
+ * in a given L2 slice, starting from cluster @l2_index, subcluster
+ * @sc_index. At most @nb_clusters are checked. Allocated clusters are

I’d add a note that reassures that @nb_clusters really is a number of clusters, not subclusters. (Because if the variable were named “x” (i.e., it’d be “At most @x are checked”), you’d assume it refers to a number of subclusters.)

OTOH, what I don’t like so far about this series is that the “cluster logic” is still everywhere when I think it should just be about subclusters now. (Except in few places where it must be about clusters as in something that can have a distinct host offset and/or has an own L2 entry.) So maybe the parameter should really be @nb_subclusters.

But I’m not sure. For how this function is written right now, it makes sense for it to be @nb_clusters, but I think it could be changed so it would work with @nb_subclusters, too. (Just a single loop over the subclusters and then loading l2_entry+l2_bitmap and checking the offset at every cluster boundary.)

Quoted 20 lines
block/qcow2-cluster.c +6 -11
+ * also required to be contiguous in the image file.
  */
-static int count_contiguous_clusters(BlockDriverState *bs, int nb_clusters,
-        int cluster_size, uint64_t *l2_slice, int l2_index, uint64_t stop_flags)
+static int count_contiguous_subclusters(BlockDriverState *bs, int nb_clusters,
+                                        unsigned sc_index, uint64_t *l2_slice,
+                                        int l2_index)
 {
     BDRVQcow2State *s = bs->opaque;
-    int i;
-    QCow2ClusterType first_cluster_type;
-    uint64_t mask = stop_flags | L2E_OFFSET_MASK | QCOW_OFLAG_COMPRESSED;
-    uint64_t first_entry = get_l2_entry(s, l2_slice, l2_index);
-    uint64_t offset = first_entry & mask;
-
-    first_cluster_type = qcow2_get_cluster_type(bs, first_entry);
-    if (first_cluster_type == QCOW2_CLUSTER_UNALLOCATED) {
-        return 0;
+    int i, j, count = 0;
+    uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index);

block/qcow2-cluster.c fail: · line 396

Quoted 10 lines
block/qcow2-cluster.c +10 -0
+    uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
+    uint64_t expected_offset = l2_entry & L2E_OFFSET_MASK;
+    bool check_offset = true;
+    QCow2ClusterType type =
+        qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc_index);
+
+    assert(type != QCOW2_CLUSTER_INVALID); /* The caller should check this */
+
+    if (type == QCOW2_CLUSTER_COMPRESSED) {
+        return 1; /* Compressed clusters are always counted one by one */

Hm, yes, but cluster by cluster, not subcluster by subcluster, so this should be s->subclusters_per_cluster, and perhaps sc_index should be asserted to be 0. (Or it should be s->subclusters_per_cluster - sc_index.)

[...]

block/qcow2-cluster.c int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,

Quoted 1 line
block/qcow2-cluster.c +0 -0
@@ -514,8 +499,8 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,

I suppose this is get_subcluster_offset now.

[...]

block/qcow2-cluster.c int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, · line 583

Quoted 11 lines
block/qcow2-cluster.c +7 -0
@@ -587,6 +574,13 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
         goto fail;
     }
     switch (type) {
+    case QCOW2_CLUSTER_INVALID:
+        qcow2_signal_corruption(bs, true, -1, -1, "Invalid cluster entry found "
+                                " (L2 offset: %#" PRIx64 ", L2 index: %#x)",
+                                l2_offset, l2_index);
+        ret = -EIO;
+        goto fail;
+        break;

No need to break here.

block/qcow2-cluster.c int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset, · line 596

Quoted 5 lines
block/qcow2-cluster.c +0 -0
     case QCOW2_CLUSTER_COMPRESSED:
         if (has_data_file(bs)) {
             qcow2_signal_corruption(bs, true, -1, -1, "Compressed cluster "
@@ -602,16 +596,15 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
         break;

Here (QCOW2_CLUSTER_COMPRESSED), c is 1, even though it should count subclusters.

Max

Quoted 9 lines
block/qcow2-cluster.c +2 -3
     case QCOW2_CLUSTER_ZERO_PLAIN:
     case QCOW2_CLUSTER_UNALLOCATED:
-        /* how many empty clusters ? */
-        c = count_contiguous_clusters_unallocated(bs, nb_clusters,
-                                                  l2_slice, l2_index, type);
+        c = count_contiguous_subclusters(bs, nb_clusters, sc_index,
+                                         l2_slice, l2_index);
         *cluster_offset = 0;
         break;
#25 lore raw

Re: [RFC PATCH v2 11/26] qcow2: Add qcow2_get_subcluster_type()

In reply to Max Reitz
Quoted 40 lines · Max Reitz
Quoted 34 lines · Alberto Garcia

This function returns the type of an individual subcluster. If an image does not have subclusters then this returns the exact same value as qcow2_get_cluster_type().

The information in standard and extended L2 entries is encoded in a slightly different way, but all existing QCow2ClusterType values are also valid for subclusters and have the same meanings (although they typically only apply to the requested subcluster).

There are two important exceptions to this:

a) QCOW2_CLUSTER_COMPRESSED means that the whole cluster is compressed. We do not support compression at the subcluster level.

b) QCOW2_CLUSTER_UNALLOCATED means that the cluster is unallocated, that is, the offset field of the L2 entry does not point to a host cluster. All subclusters are obviously unallocated too but any of them could be of type QCOW2_CLUSTER_ZERO_PLAIN.

In addition to that, extended L2 entries allow one new scenario where the cluster is normally allocated but an individual subcluster is not. This is very different from (b) and because of that this patch adds a new value called QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER.

As a last thing, this patch adds QCOW2_CLUSTER_INVALID to detect the cases where an L2 entry has a value that violates the spec. The caller is responsible for handling these situations.

To prevent compatibility problems with images that have invalid values but are currently being read by QEMU without causing side effects, QCOW2_CLUSTER_INVALID is only returned for images with extended L2 entries.

Quoted 4 lines
---
 block/qcow2.h | 62 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 62 insertions(+)
Quoted 7 lines

[...]

block/qcow2.h static inline QCow2ClusterType qcow2_get_cluster_type(BlockDriverState *bs, · line 657

Quoted 3 lines
block/qcow2.h +3 -0
+        case QCOW2_CLUSTER_ZERO_PLAIN:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+            return QCOW2_CLUSTER_INVALID;

The spec doesn’t say anything about this, though.

It does, I just forgot. O:-)

Max

#39 lore raw

Re: [RFC PATCH v2 15/26] qcow2: Add subcluster support to zero_in_l2_slice()

In reply to Alberto Garcia
Quoted 23 lines · Alberto Garcia

Setting the QCOW_OFLAG_ZERO bit of the L2 entry is forbidden if an image has subclusters. Instead, the individual 'all zeroes' bits must be used.

---
 block/qcow2-cluster.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
block/qcow2-cluster.c +1 -1
index e67559152f..3e4ba8d448 100644
@@ -1852,7 +1852,7 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
     assert(nb_clusters <= INT_MAX);
 
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t old_offset;
+        uint64_t old_offset, l2_entry = 0;
         QCow2ClusterType cluster_type;
 
         old_offset = get_l2_entry(s, l2_slice, l2_index + i);

block/qcow2-cluster.c static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset, · line 1872

Quoted 5 lines
block/qcow2-cluster.c +0 -1
@@ -1869,12 +1869,18 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
 
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
         if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) {
-            set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
             qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);

It feels wrong to me to free the cluster before updating the L2 entry.

Max

Quoted 17 lines
block/qcow2-cluster.c +9 -2
         } else {
-            uint64_t entry = get_l2_entry(s, l2_slice, l2_index + i);
-            set_l2_entry(s, l2_slice, l2_index + i, entry | QCOW_OFLAG_ZERO);
+            l2_entry = get_l2_entry(s, l2_slice, l2_index + i);
         }
+
+        if (has_subclusters(s)) {
+            set_l2_bitmap(s, l2_slice, l2_index + i, QCOW_L2_BITMAP_ALL_ZEROES);
+        } else {
+            l2_entry |= QCOW_OFLAG_ZERO;
+        }
+
+        set_l2_entry(s, l2_slice, l2_index + i, l2_entry);
     }
 
     qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
#43 lore raw

Re: [RFC PATCH v2 16/26] qcow2: Add subcluster support to discard_in_l2_slice()

In reply to Alberto Garcia
Quoted 14 lines · Alberto Garcia

Setting the QCOW_OFLAG_ZERO bit of the L2 entry is forbidden if an image has subclusters. Instead, the individual 'all zeroes' bits must be used.

---
 block/qcow2-cluster.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
block/qcow2-cluster.c +0 -0
index 3e4ba8d448..aa3eb727a5 100644

block/qcow2-cluster.c static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset, · line 1778

Quoted 8 lines
block/qcow2-cluster.c +4 -1
@@ -1772,7 +1772,11 @@ static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset,
 
         /* First remove L2 entries */
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
-        if (!full_discard && s->qcow_version >= 3) {
+        if (has_subclusters(s)) {
+            set_l2_entry(s, l2_slice, l2_index + i, 0);
+            set_l2_bitmap(s, l2_slice, l2_index + i,
+                          full_discard ? 0 : QCOW_L2_BITMAP_ALL_ZEROES);

But only for !full_discard, right?

Max

Quoted 5 lines
block/qcow2-cluster.c +1 -0
+        } else if (!full_discard && s->qcow_version >= 3) {
             set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
         } else {
             set_l2_entry(s, l2_slice, l2_index + i, 0);
#40 lore raw

Re: [RFC PATCH v2 15/26] qcow2: Add subcluster support to zero_in_l2_slice()

In reply to Max Reitz
Quoted 25 lines · Max Reitz
Quoted 24 lines · Alberto Garcia

Setting the QCOW_OFLAG_ZERO bit of the L2 entry is forbidden if an image has subclusters. Instead, the individual 'all zeroes' bits must be used.

---
 block/qcow2-cluster.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
block/qcow2-cluster.c +1 -1
index e67559152f..3e4ba8d448 100644
@@ -1852,7 +1852,7 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
     assert(nb_clusters <= INT_MAX);
 
     for (i = 0; i < nb_clusters; i++) {
-        uint64_t old_offset;
+        uint64_t old_offset, l2_entry = 0;
         QCow2ClusterType cluster_type;
 
         old_offset = get_l2_entry(s, l2_slice, l2_index + i);
@@ -1869,12 +1869,18 @@ static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset,
 
Quoted 6 lines

block/qcow2-cluster.c static int zero_in_l2_slice(BlockDriverState *bs, uint64_t offset, · line 1872

Quoted 4 lines
block/qcow2-cluster.c +0 -1
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
         if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) {
-            set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
             qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);

It feels wrong to me to free the cluster before updating the L2 entry.

(Although it’s pre-existing, as set_l2_entry() is just an in-cache operation anyway :-/)

Max

#48 lore raw

Re: [RFC PATCH v2 18/26] qcow2: Add subcluster support to expand_zero_clusters_in_l1()

In reply to Alberto Garcia
Quoted 22 lines · Alberto Garcia

Two changes are needed in order to add subcluster support to this
function: deallocated clusters must have their bitmaps cleared, and expanded clusters must have all the "subcluster allocated" bits set.

---
 block/qcow2-cluster.c | 2 ++
 1 file changed, 2 insertions(+)
block/qcow2-cluster.c +1 -0
index aa3eb727a5..62f2a9fcc0 100644
@@ -2036,6 +2036,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
                         /* not backed; therefore we can simply deallocate the
                          * cluster */
                         set_l2_entry(s, l2_slice, j, 0);
+                        set_l2_bitmap(s, l2_slice, j, 0);
                         l2_dirty = true;
                         continue;
                     }

block/qcow2-cluster.c static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, · line 2108

Quoted 6 lines
block/qcow2-cluster.c +1 -0
@@ -2102,6 +2103,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
                 } else {
                     set_l2_entry(s, l2_slice, j, offset);
                 }
+                set_l2_bitmap(s, l2_slice, j, QCOW_L2_BITMAP_ALL_ALLOC);
                 l2_dirty = true;
             }

Technically this isn’t needed because this function is only called when downgrading v3 to v2 images (which can’t have subclusters), but of course it won’t hurt.

Max

#52 lore raw

Re: [RFC PATCH v2 20/26] qcow2: Update L2 bitmap in qcow2_alloc_cluster_link_l2()

In reply to Alberto Garcia
Quoted 16 lines · Alberto Garcia

The L2 bitmap needs to be updated after each write to indicate what new subclusters are now allocated.

This needs to happen even if the cluster was already allocated and the L2 entry was otherwise valid.

---
 block/qcow2-cluster.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
block/qcow2-cluster.c +0 -0
index fb6cf8df17..acb7226e03 100644

block/qcow2-cluster.c int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) · line 987

Quoted 8 lines
block/qcow2-cluster.c +5 -0
@@ -980,6 +980,22 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
 
         set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_COPIED |
                      (cluster_offset + (i << s->cluster_bits)));
+
+        /* Update bitmap with the subclusters that were just written */
+        if (has_subclusters(s)) {
+            uint64_t written_from = m->cow_start.offset;
+            uint64_t written_to = m->cow_end.offset + m->cow_end.nb_bytes;

It’s a bit strange to make these uint64_t when all the fields queried are of type unsigned, but more on that below.

block/qcow2-cluster.c int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m) · line 991

Quoted 4 lines
block/qcow2-cluster.c +4 -0
+            uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
+            int sc;
+            for (sc = 0; sc < s->subclusters_per_cluster; sc++) {
+                uint64_t sc_off = i * s->cluster_size + sc * s->subcluster_size;

It’s weird to give this a uint64_t type when all the variables in the term are of type int.

I’m not sure whether it can overflow. handle_alloc() limits everything to INT_MAX, but I’m not sure about handle_copied().

Speaking of handle_copied(); both elements of Qcow2COWRegion are of type unsigned. handle_copied() doesn’t look like it takes any precautions to limit the range to even UINT_MAX (and it should probably limit it to INT_MAX).

Max

Quoted 11 lines
block/qcow2-cluster.c +7 -0
+                if (sc_off >= written_from && sc_off < written_to) {
+                    l2_bitmap |= QCOW_OFLAG_SUB_ALLOC(sc);
+                    l2_bitmap &= ~QCOW_OFLAG_SUB_ZERO(sc);
+                }
+            }
+            set_l2_bitmap(s, l2_slice, l2_index + i, l2_bitmap);
+        }
      }
 
 
#56 lore raw

Re: [RFC PATCH v2 22/26] qcow2: Add subcluster support to handle_alloc_space()

In reply to Alberto Garcia
Quoted 20 lines · Alberto Garcia

The bdrv_co_pwrite_zeroes() call here fills complete clusters with zeroes, but it can happen that some subclusters are not part of the write request or the copy-on-write. This patch makes sure that only the affected subclusters are overwritten.

A potential improvement would be to also fill with zeroes the other subclusters if we can guarantee that we are not overwriting existing data. However this would waste more disk space, so we should first evaluate if it's really worth doing.

---
 block/qcow2.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
block/qcow2.c +0 -0
index 0261e87709..01322ca449 100644

block/qcow2.c static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta) · line 2309

Quoted 6 lines
block/qcow2.c +3 -0
@@ -2304,6 +2304,9 @@ static int handle_alloc_space(BlockDriverState *bs, QCowL2Meta *l2meta)
 
     for (m = l2meta; m != NULL; m = m->next) {
         int ret;
+        uint64_t start_offset = m->alloc_offset + m->cow_start.offset;
+        uint64_t nb_bytes = m->cow_end.offset + m->cow_end.nb_bytes -
+            m->cow_start.offset;

It might be more honest to make nb_bytes an unsigned. (There shouldn’t be any overflows here, because the total size is limited to INT64_MAX by handle_alloc().)

Max

Quoted 2 lines
block/qcow2.c +0 -0
         if (!m->cow_start.nb_bytes && !m->cow_end.nb_bytes) {
             continue;
#59 lore raw

Re: [RFC PATCH v2 24/26] qcow2: Add the 'extended_l2' option and the QCOW2_INCOMPAT_EXTL2 bit

In reply to Alberto Garcia
Quoted 35 lines · Alberto Garcia

Now that the implementation of subclusters is complete we can finally add the necessary options to create and read images with this feature, which we call "extended L2 entries".

---
 block/qcow2.c                    |  46 ++++++++++++++
 block/qcow2.h                    |   8 ++-
 include/block/block_int.h        |   1 +
 qapi/block-core.json             |   6 ++
 tests/qemu-iotests/031.out       |   8 +--
 tests/qemu-iotests/036.out       |   4 +-
 tests/qemu-iotests/049.out       | 102 +++++++++++++++----------------
 tests/qemu-iotests/060.out       |   1 +
 tests/qemu-iotests/061.out       |  20 +++---
 tests/qemu-iotests/065           |  18 ++++--
 tests/qemu-iotests/082.out       |  48 ++++++++++++---
 tests/qemu-iotests/085.out       |  38 ++++++------
 tests/qemu-iotests/144.out       |   4 +-
 tests/qemu-iotests/182.out       |   2 +-
 tests/qemu-iotests/185.out       |   8 +--
 tests/qemu-iotests/198.out       |   2 +
 tests/qemu-iotests/206.out       |   4 ++
 tests/qemu-iotests/242.out       |   5 ++
 tests/qemu-iotests/255.out       |   8 +--
 tests/qemu-iotests/common.filter |   1 +
 20 files changed, 224 insertions(+), 110 deletions(-)
block/qcow2.c +0 -0
index 537569ce88..b1fa7ab5da 100644
@@ -1347,6 +1347,12 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
     s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
     s->subcluster_bits = ctz32(s->subcluster_size);

block/qcow2.c static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, · line 1358

Quoted 10 lines
block/qcow2.c +6 -0
 
+    if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
+        error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
+        ret = -EINVAL;
+        goto fail;
+    }
+
     /* Check support for various header values */
     if (header.refcount_order > 6) {
         error_setg(errp, "Reference count entry width too large; may not "

It feels like this hunk should be part of the patch that added the s->subcluster_size assignment.

Quoted 15 lines
block/qcow2.c +5 -0
@@ -2806,6 +2812,11 @@ int qcow2_update_header(BlockDriverState *bs)
                 .bit  = QCOW2_COMPAT_LAZY_REFCOUNTS_BITNR,
                 .name = "lazy refcounts",
             },
+            {
+                .type = QCOW2_FEAT_TYPE_INCOMPATIBLE,
+                .bit  = QCOW2_INCOMPAT_EXTL2_BITNR,
+                .name = "extended L2 entries",
+            },
         };
 
         ret = header_ext_add(buf, QCOW2_EXT_MAGIC_FEATURE_TABLE,
@@ -3271,6 +3282,27 @@ qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp)
         goto out;
     }

block/qcow2.c qcow2_co_create(BlockdevCreateOptions *create_options, Error **errp) · line 3293

Quoted 10 lines
block/qcow2.c +9 -0
 
+    if (!qcow2_opts->has_extended_l2) {
+        qcow2_opts->extended_l2 = false;
+    }
+    if (qcow2_opts->extended_l2) {
+        unsigned min_cluster_size =
+            (1 << MIN_CLUSTER_BITS) * QCOW_MAX_SUBCLUSTERS_PER_CLUSTER;
+        if (version < 3) {
+            error_setg(errp, "Extended L2 entries are only supported with "
+                       "compatibility level 1.1 and above (use version=v3 or "

Pre-existing, but old-style creation doesn’t allow version=v3. I suppose someone(tm) should fix that...

Quoted 15 lines
block/qcow2.c +12 -0
+                       "greater)");
+            ret = -EINVAL;
+            goto out;
+        }
+        if (cluster_size < min_cluster_size) {
+            error_setg(errp, "Extended L2 entries are only supported with "
+                       "cluster sizes of at least %u bytes", min_cluster_size);
+            ret = -EINVAL;
+            goto out;
+        }
+    }
+
     if (!qcow2_opts->has_preallocation) {
         qcow2_opts->preallocation = PREALLOC_MODE_OFF;
     }

[...]

qapi/block-core.json line 70

Quoted 10 lines
qapi/block-core.json +2 -0
index aa97ee2641..6e4c81ae1c 100644
@@ -66,6 +66,8 @@
 #                 standalone (read-only) raw image without looking at qcow2
 #                 metadata (since: 4.0)
 #
+# @extended-l2: true if the image has extended L2 entries (since 4.2)
+#

I think there should be the same note here as for lazy-refcounts. So this field is not present for compat=0.10, and present otherwise. (Without that note, it looks as if this field maybe is only present if it’s true.)

(Personally, I’d also prefer it to be an “iff”, but who cares.)

Quoted 12 lines
qapi/block-core.json +1 -0
 # @lazy-refcounts: on or off; only valid for compat >= 1.1
 #
 # @corrupt: true if the image has been marked corrupt; only valid for
@@ -85,6 +87,7 @@
       'compat': 'str',
       '*data-file': 'str',
       '*data-file-raw': 'bool',
+      '*extended-l2': 'bool',
       '*lazy-refcounts': 'bool',
       '*corrupt': 'bool',
       'refcount-bits': 'int',

qapi/block-core.json line 4378

Quoted 4 lines
qapi/block-core.json +1 -0
@@ -4372,6 +4375,8 @@
 # @data-file-raw    True if the external data file must stay valid as a
 #                   standalone (read-only) raw image without looking at qcow2
 #                   metadata (default: false; since: 4.0)
+# @extended-l2      True if the image has extended L2 entries

s/if the image has/to make the image have/ (or something like it)

Quoted 4 lines
qapi/block-core.json +1 -0
+#                   (default: false; since 4.2)
 # @size             Size of the virtual disk in bytes
 # @version          Compatibility level (default: v3)
 # @backing-file     File name of the backing file if a backing file

[...]

tests/qemu-iotests/common.filter _filter_img_create() · line 143

Quoted 12 lines
tests/qemu-iotests/common.filter +1 -0
index 9f418b4881..45c0fe746b 100644
@@ -137,6 +137,7 @@ _filter_img_create()
         -e "s# adapter_type=[^ ]*##g" \
         -e "s# hwversion=[^ ]*##g" \
         -e "s# lazy_refcounts=\\(on\\|off\\)##g" \
+        -e "s# extended_l2=\\(on\\|off\\)##g" \
         -e "s# block_size=[0-9]\\+##g" \
         -e "s# block_state_zero=\\(on\\|off\\)##g" \
         -e "s# log_size=[0-9]\\+##g" \

Filtering it seems like a good idea so one could run the iotests with -o extended_l2=on. There are still a ton of tests that don’t really seem filtered, though. Maybe my “Allow ./check -o data_file” series improves that? :?

Max

#62 lore raw

Re: [RFC PATCH v2 25/26] qcow2: Allow preallocation and backing files if extended_l2 is set

In reply to Alberto Garcia
Quoted 8 lines · Alberto Garcia

Traditional qcow2 images don't allow preallocation if a backing file is set. This is because once a cluster is allocated there is no way to tell that its data should be read from the backing file.

Extended L2 entries have individual allocation bits for each subcluster, and therefore it is perfectly possible to have an allocated cluster with all its subclusters unallocated.

Quoted 5 lines
---
 block/qcow2.c              | 7 ++++---
 tests/qemu-iotests/206.out | 2 +-
 2 files changed, 5 insertions(+), 4 deletions(-)

But it doesn’t work, because qcow2_alloc_cluster_offset() always allocates the whole cluster, so the backing file content isn’t visible:

$ ./qemu-img create -f qcow2 base.qcow2 64M
Formatting 'base.qcow2', fmt=qcow2 size=67108864 cluster_size=65536
lazy_refcounts=off extended_l2=off refcount_bits=16
$ ./qemu-io -c 'write -P 42 0 64M' base.qcow2
wrote 67108864/67108864 bytes at offset 0
64 MiB, 1 ops; 00.21 sec (307.344 MiB/sec and 4.8022 ops/sec)
$ ./qemu-img create -f qcow2 -o preallocation=metadata,extended_l2=on \
    top.qcow2
Formatting 'top.qcow2', fmt=qcow2 size=67108864 backing_file=base.qcow2
cluster_size=65536 preallocation=metadata lazy_refcounts=off
extended_l2=on refcount_bits=16
$ ./qemu-io -c 'read -P 42 0 64M' top.qcow2
Pattern verification failed at offset 0, 67108864 bytes
read 67108864/67108864 bytes at offset 0
64 MiB, 1 ops; 00.03 sec (2.498 GiB/sec and 39.9725 ops/sec)

Max

#64 lore raw
In reply to Alberto Garcia
Quoted 5 lines · Alberto Garcia

Hi,

here's the new version of the patches to add subcluster allocation support to qcow2.

Quoted 7 lines

Please refer to the cover letter of the first version for a full description of the patches:

   https://lists.gnu.org/archive/html/qemu-block/2019-10/msg00983.html

This version includes a few tests, but I'm planning to add more for the next revision.

I think what would help most with testing is if it were possible to simply run the iotests with -o extended_l2=on.

In general, the RFC looks OK to me. The one thing I dislike is that I feel that it is a bit, well, uncourageous. Now, after looking at the series, I don’t know whether you really changed everything that needs to be changed so it can deal with subclusters.

To me it feels like this is because you tried to keep everything as it is and only do minimal changes. That is usually a good thing, but here I don’t know, because this way we can’t simply grep for places that need fixing (because they use /\<cluster/ instead of /subcluster/).

To me it feels like with subclusters, the whole design should be different. Note that the following is just a very naïve idea, but
anyway: I feel like what we need to separate isn’t L2 entries vs. clusters vs. subclusters (so a separation based on, well, syntax?) but a separation based on offset vs. allocation status (so a separation based on semantics).

So I imagine there would be one function that sets a whole cluster’s (i.e., a group of subclusters) allocation offset; and another function that sets individual subclusters’ allocation status.

Reversely, there should be a function to query a cluster’s/subcluster’s allocation offset, and another to query a subcluster’s type.

To me it looks like the places where QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER is handled separately from QCOW2_CLUSTER_UNALLOCATED are places where we’re really not interested in the subcluster’s type at all, but just whether there’s an allocation or not. This is the case in qcow2_get_cluster_offset(), calculate_l2_meta(), and qcow2_co_block_status().

(These places should then use the function to query the allocation offset and evaluate the result instead of querying the subcluster type.)

Does that sound in any way acceptable to you? You have more insight into this now and so maybe you know already that it can’t work. (Or maybe it’s just too invasive.)

Right now, all I can do is grep for QCOW2_CLUSTER_ and set_l2_entry(). And all of those places look OK to me. But I just can’t shake off the uneasy feeling of not being able to really know whether this series really got to all the places that need adjustment.

Max

#30 lore raw

Re: [RFC PATCH v2 12/26] qcow2: Handle QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER

In reply to Max Reitz

On Mon 04 Nov 2019 02:10:37 PM CET, Max Reitz wrote:

[QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER]

Quoted 3 lines

I still don’t know what you’re doing in the later patches, but to me it looks a bit like you don’t dare breaking up the existing structure that just deals with clusters.

Yeah, I decided to extend the existing type to make the changes less invasive, but I'm also leaning towards creating a different type now. I'll give it a try.

Berto

#33 lore raw

Re: [RFC PATCH v2 13/26] qcow2: Add subcluster support to calculate_l2_meta()

In reply to Max Reitz
Quoted 7 lines · Max Reitz
Quoted 6 lines

If an image has subclusters then there are more copy-on-write scenarios that we need to consider. Let's say we have a write request from the middle of subcluster #3 until the end of the cluster:

- If the cluster is new, then subclusters #0 to #3 from the old cluster must be copied into the new one.

Quoted 6 lines

You mean for snapshots?

(That isn’t quite clear, and I only guess this based on the next bullet point which differentiates based on “the old cluster was unallocated”. That’s weird, too, because what does that mean, old cluster and new cluster?

Yes, perhaps the terminology is a bit unclear.

When I say "new cluster" is this context I mean that a write request requires that a new cluster is allocated in the qcow2 file.

Then the "old cluster" would be what was there before the write (i.e. a cluster with refcount > 1 or an unallocated cluster). Where we are doing the copy-on-write from.

Quoted 9 lines
Quoted 5 lines

* If @keep_old is true it means that the clusters were already
* allocated and will be overwritten. If false then the clusters are
* new and we have to decrease the reference count of the old ones.
+ *
+ * Returns 1 on success, -errno on failure.

I think there should be a note here on why this doesn’t follow the general 0/-errno schema (i.e., “, because that is what callers generally expect”).

Good idea.

Quoted 8 lines

block/qcow2-cluster.c static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset, · line 1095

Quoted 7 lines
block/qcow2-cluster.c +7 -0
+    if (!keep_old) {
+        switch (type) {
+        case QCOW2_CLUSTER_NORMAL:
+        case QCOW2_CLUSTER_COMPRESSED:
+        case QCOW2_CLUSTER_ZERO_ALLOC:
+        case QCOW2_CLUSTER_UNALLOCATED_SUBCLUSTER:
+            cow_start_from = 0;
Quoted 6 lines

Somehow (I don’t know why) I find this a bit tough to understand.

Wouldn’t it work to let cow_start start from the first subcluster for ZERO_ALLOC and UNALLOCATED_SUBCLUSTER? We don’t need to COW those, it should be sufficient to just make the subclusters before that zero or unallocated, respectively.

Here's one good example why I should probably add a QCow2SubclusterType different from the existing QCow2ClusterType.

In this context, 'type' is the type of the subcluster, and because of that _ZERO_ALLOC means that the subcluster reads as zeros but the cluster itself is allocated. Other subcluster may contain data and that's why we have to copy all of them.

Berto

#36 lore raw

Re: [RFC PATCH v2 14/26] qcow2: Add subcluster support to qcow2_get_cluster_offset()

In reply to Max Reitz
Quoted 10 lines · Max Reitz

OTOH, what I don’t like so far about this series is that the “cluster logic” is still everywhere when I think it should just be about subclusters now. (Except in few places where it must be about clusters as in something that can have a distinct host offset and/or has an own L2 entry.) So maybe the parameter should really be @nb_subclusters.

But I’m not sure. For how this function is written right now, it makes sense for it to be @nb_clusters, but I think it could be changed so it would work with @nb_subclusters, too.

I'm still reviewing your (much appreciated) feedback, but one thing I can tell you is that my initial versions were doing everything with subclusters because of the reasons you mention (i.e. there was @nb_subclusters and all that).

Later when I started to tidy things up I realized that most of those places only needed the number of clusters after all, and in some cases the necessary changes were really minimal (like in handle_copied()).

Quoted 4 lines

block/qcow2-cluster.c fail: · line 383

Quoted 4 lines
block/qcow2-cluster.c +3 -0
+static int count_contiguous_subclusters(BlockDriverState *bs, int nb_clusters,
+                                        unsigned sc_index, uint64_t *l2_slice,
+                                        int l2_index)
 {

/* ... */

Quoted 7 lines

block/qcow2-cluster.c fail: · line 396

Quoted 2 lines
block/qcow2-cluster.c +2 -0
+    if (type == QCOW2_CLUSTER_COMPRESSED) {
+        return 1; /* Compressed clusters are always counted one by one */

Hm, yes, but cluster by cluster, not subcluster by subcluster, so this should be s->subclusters_per_cluster, and perhaps sc_index should be asserted to be 0. (Or it should be s->subclusters_per_cluster - sc_index.)

Right, that's a bug, it forces the caller to decompress the cluster 32 times in order to read it completely! Thanks!

(in reality this is not used because this function doesn't get called for compressed clusters but the same problem happens in the calling function, as you correctly point out. Maybe I should assert here instead)

Quoted 3 lines

Review int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,

Quoted 1 line
(quoted diff) +0 -0
@@ -514,8 +499,8 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,

I suppose this is get_subcluster_offset now.

Hmmm no, this returns the actual host cluster offset, then the caller uses offset_into_cluster() to get the final value (which doesn't need to be subcluster-aligned anyway).

Berto

#37 lore raw

Re: [RFC PATCH v2 14/26] qcow2: Add subcluster support to qcow2_get_cluster_offset()

Quoted 1 line · Alberto Garcia

On Mon 04 Nov 2019 03:58:57 PM CET, Max Reitz wrote:

[...]

Quoted 7 lines
Quoted 3 lines

Review int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,

Quoted 1 line
(quoted diff) +0 -0
@@ -514,8 +499,8 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,

I suppose this is get_subcluster_offset now.

Hmmm no, this returns the actual host cluster offset, then the caller uses offset_into_cluster() to get the final value (which doesn't need to be subcluster-aligned anyway).

It still returns the subcluster type, though. Which makes the whole thing kind of a weird chimera... Maybe it would work better if the subcluster type would be stored in a pointer parameter, so it’d be clear?

But does it have actually have any advantages to on one hand return a cluster-aligned host offset and on the other return a bytes count that starts at the mid-cluster offset? That’s weird already. Maybe the offset that’s returned should just not be cluster-aligned and the whole function be called qcow2_get_host_offset().

I suppose one reason (maybe the only one?) to return aligned offsets is for compressed clusters. It would be another kind of magic to return aligned offsets only for them. But maybe it’s worth it? (Judging from a quick glance, it doesn’t look too difficult to me to modify the callers to accept a mid-cluster host offset.)

(Another thing I just noticed is that the comment above qcow2_get_cluster_offset() needs some fixing, because it still refers to whole clusters.)

Max

#10 lore raw

Re: [RFC PATCH v2 03/26] qcow2: Process QCOW2_CLUSTER_ZERO_ALLOC clusters in handle_copied()

In reply to Max Reitz
Quoted 23 lines · Max Reitz
Quoted 12 lines

static void calculate_l2_meta(BlockDriverState *bs, uint64_t host_offset,
uint64_t guest_offset, uint64_t bytes,

block/qcow2-cluster.c +4 -2
-                              QCowL2Meta **m, bool keep_old)
+                              uint64_t *l2_slice, QCowL2Meta **m, bool keep_old)
 {
     BDRVQcow2State *s = bs->opaque;
-    unsigned cow_start_from = 0;
+    int l2_index = offset_to_l2_slice_index(s, guest_offset);
+    uint64_t l2_entry;
+    unsigned cow_start_from, cow_end_to;
     unsigned cow_start_to = offset_into_cluster(s, guest_offset);
     unsigned cow_end_from = cow_start_to + bytes;

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1054

Quoted 10 lines
block/qcow2-cluster.c +7 -1
-    unsigned cow_end_to = ROUND_UP(cow_end_from, s->cluster_size);
     unsigned nb_clusters = size_to_clusters(s, cow_end_from);
     QCowL2Meta *old_m = *m;
+    QCow2ClusterType type;
+
+    /* Return if there's no COW (all clusters are normal and we keep them) */
+    if (keep_old) {
+        int i;
+        for (i = 0; i < nb_clusters; i++) {
+            l2_entry = be64_to_cpu(l2_slice[l2_index + i]);
Quoted 5 lines

I’d assert somewhere that l2_index + nb_clusters - 1 won’t overflow.

Quoted 1 line

+ if (qcow2_get_cluster_type(bs, l2_entry) != QCOW2_CLUSTER_NORMAL) {

Wouldn’t cluster_needs_cow() be better?

The semantics of cluster_needs_cow() change in this patch (which also updates the documentation). But I should maybe change the name instead.

Quoted 8 lines

block/qcow2-cluster.c void qcow2_alloc_cluster_abort(BlockDriverState *bs, QCowL2Meta *m) · line 1062

Quoted 7 lines
block/qcow2-cluster.c +7 -0
+                break;
+            }
+        }
+        if (i == nb_clusters) {
+            return;
+        }
+    }
Quoted 4 lines

So I understand we always need to create an L2Meta structure in all other cases because we at least need to turn those clusters into normal clusters? (Even if they’re already allocated, as in the case of allocated zero clusters.)

That's correct.

Quoted 4 lines
Quoted 4 lines
block/qcow2-cluster.c +1 -1
-/* Returns true if writing to a cluster requires COW */
+/* Returns true if the cluster is unallocated or has refcount > 1 */
 static bool cluster_needs_cow(BlockDriverState *bs, uint64_t l2_entry)
 {
Quoted 8 lines
Quoted 5 lines
     switch (qcow2_get_cluster_type(bs, l2_entry)) {
     case QCOW2_CLUSTER_NORMAL:
+    case QCOW2_CLUSTER_ZERO_ALLOC:
         if (l2_entry & QCOW_OFLAG_COPIED) {
             return false;

Don’t zero-allocated clusters need COW always? (Because the at the given host offset isn’t guaranteed to be zero.)

Yeah, hence the semantics change I described earlier. I should probably call it cluster_needs_new_allocation() or something like that, which is what this means now ("true if unallocated or refcount > 1").

Quoted 11 lines
Quoted 9 lines

- * Returns the number of contiguous clusters that can be used for an allocating
- * write, but require COW to be performed (this includes yet unallocated space,
- * which must copy from the backing file)
+ * Returns the number of contiguous clusters that can be written to
+ * using one single write request, starting from @l2_index.
+ * At most @nb_clusters are checked.
+ *
+ * If @want_cow is true this counts clusters that are either
+ * unallocated, or allocated but with refcount > 1.

+(So they need to be newly allocated and COWed)?

Yes. Which in this context is the same as "newly allocated" I guess, because every newly allocated cluster requires COW.

Quoted 1 line

(Or is the past participle of COW COWn? Or maybe CedOW?)

:-))

Quoted 4 lines
Quoted 2 lines

+ * If @want_cow is false this counts clusters that are already
+ * allocated and can be written to using their current locations

s/using their current locations/in-place/?

Ok.

Quoted 12 lines
Quoted 12 lines
block/qcow2-cluster.c +3 -3
@@ -1475,13 +1489,14 @@ static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset,
     *bytes = MIN(*bytes, nb_bytes - offset_into_cluster(s, guest_offset));
     assert(*bytes != 0);
 
-    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes,
-                      m, keep_old_clusters);
+    calculate_l2_meta(bs, alloc_cluster_offset, guest_offset, *bytes, l2_slice,
+                      m, false);
 
-    return 1;
+    ret = 1;
 
Quoted 6 lines

block/qcow2-cluster.c static int handle_alloc(BlockDriverState *bs, uint64_t guest_offset, · line 1498

Quoted 4 lines
block/qcow2-cluster.c +2 -2
-fail:
-    if (*m && (*m)->nb_clusters > 0) {
+out:
+    qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);

Is this a bug fix?

No, we call qcow2_cache_put() later because calculate_l2_meta() needs l2_slice.

Berto

#22 lore raw

Re: [RFC PATCH v2 10/26] qcow2: Update get/set_l2_entry() and add get/set_l2_bitmap()

In reply to Max Reitz
Quoted 11 lines · Max Reitz
Quoted 3 lines

This patch also adds the get/set_l2_bitmap() functions that are used to access the bitmaps. For convenience, these functions are no-ops when used in traditional qcow2 images.

Granted, I haven’t seen the following patches yet, but if these functions are indeed called for images that don’t have subclusters, shouldn’t they return 0x0*0f*f then? (i.e. everything allocated)

If they aren’t, they should probably just abort(). Well, set_l2_bitmap() should probably always abort() if there aren’t any subclusters.

Yeah, set_l2_bitmap() should abort (I had this changed already).

About get_l2_bitmap() ... I decided not to abort for convenience, for cases like this one:

    uint64_t l2_entry = get_l2_entry(s, l2_slice, l2_index);
    uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index);
    type = qcow2_get_subcluster_type(bs, l2_entry, l2_bitmap, sc);

Here the value of l2_bitmap is going to be ignored anyway so it doesn't matter what we return, but perhaps for consistency we should return QCOW_OFLAG_SUB_ALLOC(0), which means that the first (and only, in this case) subcluster is allocated.

Berto

#41 lore raw

Re: [RFC PATCH v2 15/26] qcow2: Add subcluster support to zero_in_l2_slice()

In reply to Max Reitz
Quoted 10 lines · Max Reitz
Quoted 7 lines
Quoted 4 lines
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
         if (cluster_type == QCOW2_CLUSTER_COMPRESSED || unmap) {
-            set_l2_entry(s, l2_slice, l2_index + i, QCOW_OFLAG_ZERO);
             qcow2_free_any_clusters(bs, old_offset, 1, QCOW2_DISCARD_REQUEST);

It feels wrong to me to free the cluster before updating the L2 entry.

(Although it’s pre-existing, as set_l2_entry() is just an in-cache operation anyway :-/)

Yes, I think that if you want to do it afterwards you need to add another loop after the qcow2_cache_put() call.

Berto

#44 lore raw

Re: [RFC PATCH v2 16/26] qcow2: Add subcluster support to discard_in_l2_slice()

In reply to Max Reitz
Quoted 9 lines · Max Reitz

block/qcow2-cluster.c static int discard_in_l2_slice(BlockDriverState *bs, uint64_t offset, · line 1778

Quoted 7 lines
         /* First remove L2 entries */
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
block/qcow2-cluster.c +4 -1
-        if (!full_discard && s->qcow_version >= 3) {
+        if (has_subclusters(s)) {
+            set_l2_entry(s, l2_slice, l2_index + i, 0);
+            set_l2_bitmap(s, l2_slice, l2_index + i,
+                          full_discard ? 0 : QCOW_L2_BITMAP_ALL_ZEROES);

But only for !full_discard, right?

Hence the conditional operator in my patch, maybe you didn't see it?

Berto

#49 lore raw

Re: [RFC PATCH v2 18/26] qcow2: Add subcluster support to expand_zero_clusters_in_l1()

In reply to Max Reitz
Quoted 11 lines · Max Reitz

block/qcow2-cluster.c static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table, · line 2108

Quoted 7 lines
block/qcow2-cluster.c +1 -0
@@ -2102,6 +2103,7 @@ static int expand_zero_clusters_in_l1(BlockDriverState *bs, uint64_t *l1_table,
                 } else {
                     set_l2_entry(s, l2_slice, j, offset);
                 }
+                set_l2_bitmap(s, l2_slice, j, QCOW_L2_BITMAP_ALL_ALLOC);
                 l2_dirty = true;
             }

Technically this isn’t needed because this function is only called when downgrading v3 to v2 images (which can’t have subclusters), but of course it won’t hurt.

Right, but we need to change the function anyway to either do this or assert that there are no subclusters. I prefer to do this because it's trivial but I won't oppose if someone prefers the alternative.

Berto

#45 lore raw

Re: [RFC PATCH v2 16/26] qcow2: Add subcluster support to discard_in_l2_slice()

In reply to Alberto Garcia
Quoted 9 lines · Alberto Garcia
Quoted 8 lines · Max Reitz
Quoted 7 lines
         /* First remove L2 entries */
         qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_slice);
block/qcow2-cluster.c +4 -1
-        if (!full_discard && s->qcow_version >= 3) {
+        if (has_subclusters(s)) {
+            set_l2_entry(s, l2_slice, l2_index + i, 0);
+            set_l2_bitmap(s, l2_slice, l2_index + i,
+                          full_discard ? 0 : QCOW_L2_BITMAP_ALL_ZEROES);
Quoted 3 lines
Quoted 1 line

But only for !full_discard, right?

Hence the conditional operator in my patch, maybe you didn't see it?

Oops, yep, I was just hung up on the if conditional.

Max

#53 lore raw

Re: [RFC PATCH v2 20/26] qcow2: Update L2 bitmap in qcow2_alloc_cluster_link_l2()

In reply to Max Reitz
Quoted 4 lines · Max Reitz

Speaking of handle_copied(); both elements of Qcow2COWRegion are of type unsigned. handle_copied() doesn’t look like it takes any precautions to limit the range to even UINT_MAX (and it should probably limit it to INT_MAX).

Or rather, both handle_copied() and handle_alloc() should probably limit it to BDRV_REQUEST_MAX_BYTES.

Berto

#60 lore raw

Re: [RFC PATCH v2 24/26] qcow2: Add the 'extended_l2' option and the QCOW2_INCOMPAT_EXTL2 bit

In reply to Max Reitz
Quoted 4 lines · Max Reitz
Quoted 4 lines
block/qcow2.c +0 -0
@@ -1347,6 +1347,12 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
     s->subcluster_size = s->cluster_size / s->subclusters_per_cluster;
     s->subcluster_bits = ctz32(s->subcluster_size);
 
Quoted 12 lines

block/qcow2.c static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options, · line 1358

Quoted 9 lines
block/qcow2.c +6 -0
+    if (s->subcluster_size < (1 << MIN_CLUSTER_BITS)) {
+        error_setg(errp, "Unsupported subcluster size: %d", s->subcluster_size);
+        ret = -EINVAL;
+        goto fail;
+    }
+
     /* Check support for various header values */
     if (header.refcount_order > 6) {
         error_setg(errp, "Reference count entry width too large; may not "

It feels like this hunk should be part of the patch that added the s->subcluster_size assignment.

I don't have a strong opinion, but that patch simply defines the basic attributes that are used elsewhere in the code, and this is the patch where the values are checked (for creation and opening).

Berto