qcow2_get_cluster_offset() checks how many contiguous bytes are
available at a given offset. The returned number of bytes is limited
by the amount that can be addressed without having to load more than
one L2 table.

Since we'll be loading L2 slices instead of full tables this patch
changes the limit accordingly using the size of the L2 slice for the
calculations instead of the full table size.

The l2_table variable is also renamed to l2_slice to reflect this, and
offset_to_l2_index() is replaced with offset_to_l2_slice_index().

Signed-off-by: Alberto Garcia <berto@igalia.com>
---
 block/qcow2-cluster.c | 30 +++++++++++++++---------------
 1 file changed, 15 insertions(+), 15 deletions(-)

diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
index 3d43dc2307..f167bb6336 100644
--- a/block/qcow2-cluster.c
+++ b/block/qcow2-cluster.c
@@ -530,8 +530,8 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
 {
     BDRVQcow2State *s = bs->opaque;
     unsigned int l2_index;
-    uint64_t l1_index, l2_offset, *l2_table;
-    int l1_bits, c;
+    uint64_t l1_index, l2_offset, *l2_slice;
+    int c;
     unsigned int offset_in_cluster;
     uint64_t bytes_available, bytes_needed, nb_clusters;
     QCow2ClusterType type;
@@ -540,12 +540,12 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
     offset_in_cluster = offset_into_cluster(s, offset);
     bytes_needed = (uint64_t) *bytes + offset_in_cluster;
 
-    l1_bits = s->l2_bits + s->cluster_bits;
-
     /* compute how many bytes there are between the start of the cluster
-     * containing offset and the end of the l1 entry */
-    bytes_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1))
-                    + offset_in_cluster;
+     * containing offset and the end of the l2 slice that contains
+     * the entry pointing to it */
+    bytes_available =
+        ((uint64_t) (s->l2_slice_size - offset_to_l2_slice_index(s, offset)))
+        << s->cluster_bits;
 
     if (bytes_needed > bytes_available) {
         bytes_needed = bytes_available;
@@ -574,17 +574,17 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
         return -EIO;
     }
 
-    /* load the l2 table in memory */
+    /* load the l2 slice in memory */
 
-    ret = l2_load(bs, offset, l2_offset, &l2_table);
+    ret = l2_load(bs, offset, l2_offset, &l2_slice);
     if (ret < 0) {
         return ret;
     }
 
     /* find the cluster offset for the given disk offset */
 
-    l2_index = offset_to_l2_index(s, offset);
-    *cluster_offset = be64_to_cpu(l2_table[l2_index]);
+    l2_index = offset_to_l2_slice_index(s, offset);
+    *cluster_offset = be64_to_cpu(l2_slice[l2_index]);
 
     nb_clusters = size_to_clusters(s, bytes_needed);
     /* bytes_needed <= *bytes + offset_in_cluster, both of which are unsigned
@@ -611,14 +611,14 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
     case QCOW2_CLUSTER_UNALLOCATED:
         /* how many empty clusters ? */
         c = count_contiguous_clusters_unallocated(nb_clusters,
-                                                  &l2_table[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(nb_clusters, s->cluster_size,
-                                      &l2_table[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,
@@ -634,7 +634,7 @@ int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
         abort();
     }
 
-    qcow2_cache_put(s->l2_table_cache, (void **) &l2_table);
+    qcow2_cache_put(s->l2_table_cache, (void **) &l2_slice);
 
     bytes_available = (int64_t)c * s->cluster_size;
 
@@ -652,7 +652,7 @@ out:
     return type;
 
 fail:
-    qcow2_cache_put(s->l2_table_cache, (void **)&l2_table);
+    qcow2_cache_put(s->l2_table_cache, (void **)&l2_slice);
     return ret;
 }
 
-- 
2.11.0

On 12/15/2017 06:53 AM, Alberto Garcia wrote:
> qcow2_get_cluster_offset() checks how many contiguous bytes are
> available at a given offset. The returned number of bytes is limited
> by the amount that can be addressed without having to load more than
> one L2 table.
> 
> Since we'll be loading L2 slices instead of full tables this patch
> changes the limit accordingly using the size of the L2 slice for the
> calculations instead of the full table size.
> 
> The l2_table variable is also renamed to l2_slice to reflect this, and
> offset_to_l2_index() is replaced with offset_to_l2_slice_index().
> 
> Signed-off-by: Alberto Garcia <berto@igalia.com>
> ---
>  block/qcow2-cluster.c | 30 +++++++++++++++---------------
>  1 file changed, 15 insertions(+), 15 deletions(-)
> 

Callers like 'qemu-img map' will potentially have to iterate in more
steps over an entire image; but that's not a severe limitation.  The
block_status() functions already document that as long as drivers make
progress, callers must be prepared for back-to-back calls that return
the same status, so you are not breaking semantics with a shorter
clamping limit.

Maybe some of that is worth mentioning in the commit message.  Either way,

Reviewed-by: Eric Blake <eblake@redhat.com>

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

On Tue 16 Jan 2018 11:42:45 PM CET, Eric Blake wrote:
> Callers like 'qemu-img map' will potentially have to iterate in more
> steps over an entire image; but that's not a severe limitation.  The
> block_status() functions already document that as long as drivers make
> progress, callers must be prepared for back-to-back calls that return
> the same status, so you are not breaking semantics with a shorter
> clamping limit.
>
> Maybe some of that is worth mentioning in the commit message.  Either
> way,

Yes indeed, for 4KB slices and 64KB clusters each slice can only map
32MB. Then again I don't think that it makes sense to change the slice
size for qemu-img map, or in general for anything that does pure
sequential I/O.

But it's probably worth mentioning in the commit message.

Berto