diff mbox series

[v4,23/30] qcow2: Update L2 bitmap in qcow2_alloc_cluster_link_l2()

Message ID 6ffd3ee7ca1e53272705ad34d2ea7ac10b50ade0.1584468723.git.berto@igalia.com
State New
Headers show
Series Add subcluster allocation to qcow2 | expand

Commit Message

Alberto Garcia March 17, 2020, 6:16 p.m. UTC
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.

Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
---
 block/qcow2-cluster.c | 17 +++++++++++++++++
 1 file changed, 17 insertions(+)

Comments

Eric Blake April 24, 2020, 7:39 p.m. UTC | #1
On 3/17/20 1:16 PM, Alberto Garcia wrote:
> 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.
> 
> Signed-off-by: Alberto Garcia <berto@igalia.com>
> Reviewed-by: Max Reitz <mreitz@redhat.com>
> ---
>   block/qcow2-cluster.c | 17 +++++++++++++++++
>   1 file changed, 17 insertions(+)
> 
> diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
> index ceacd91ea3..dfd8b66958 100644
> --- a/block/qcow2-cluster.c
> +++ b/block/qcow2-cluster.c
> @@ -1006,6 +1006,23 @@ int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
>           assert((offset & L2E_OFFSET_MASK) == offset);
>   
>           set_l2_entry(s, l2_slice, l2_index + i, offset | QCOW_OFLAG_COPIED);
> +
> +        /* Update bitmap with the subclusters that were just written */
> +        if (has_subclusters(s)) {
> +            unsigned written_from = m->cow_start.offset;
> +            unsigned written_to = m->cow_end.offset + m->cow_end.nb_bytes ?:
> +                m->nb_clusters << s->cluster_bits;
> +            uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
> +            int sc;
> +            for (sc = 0; sc < s->subclusters_per_cluster; sc++) {
> +                int 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);
> +                }
> +            }

Are there more efficient ways to set this series of bits than iterating 
one bit at a time, while still remaining legible?  For example, what if 
we had something like:

l2_bitmap = get_l2_bitmap(...);
int sc_from = OFFSET_TO_SC(written_from);
int sc_to = OFFSET_TO_SC(written_to - 1);
l2_bitmap |= QCOW_OFLAG_SUB_ALLOC_RANGE(sc_from, sc_to);
l2_bitmap &= ~QCOW_OFLAG_SUB_ZERO_RANGE(sc_from, sc_to);

which would require macros:

#define OFFSET_TO_SC(offset) (offset >> (s->cluster_bits - 6))
#define QCOW_OFLAG_SUB_ALLOC_RANGE(from, to) \
   deposit64(0, (from), (len) - (from), -1)
#define QCOW_OFLAG_SUB_ZERO_RANGE(from, to) \
   deposit64(0, (from) + 32, (len) - (from) + 32, -1)


> +            set_l2_bitmap(s, l2_slice, l2_index + i, l2_bitmap);

I'm hoping this function doesn't cause redundant I/O if the L2 entry 
didn't actually change.  But that's not the concern for this patch.
Alberto Garcia April 27, 2020, 1:17 p.m. UTC | #2
On Fri 24 Apr 2020 09:39:25 PM CEST, Eric Blake wrote:
>> +        /* Update bitmap with the subclusters that were just written */
>> +        if (has_subclusters(s)) {
>> +            unsigned written_from = m->cow_start.offset;
>> +            unsigned written_to = m->cow_end.offset + m->cow_end.nb_bytes ?:
>> +                m->nb_clusters << s->cluster_bits;
>> +            uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
>> +            int sc;
>> +            for (sc = 0; sc < s->subclusters_per_cluster; sc++) {
>> +                int 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);
>> +                }
>> +            }
>
> Are there more efficient ways to set this series of bits than iterating 
> one bit at a time, while still remaining legible?  For example, what if 
> we had something like:
>
> l2_bitmap = get_l2_bitmap(...);
> int sc_from = OFFSET_TO_SC(written_from);
> int sc_to = OFFSET_TO_SC(written_to - 1);
> l2_bitmap |= QCOW_OFLAG_SUB_ALLOC_RANGE(sc_from, sc_to);
> l2_bitmap &= ~QCOW_OFLAG_SUB_ZERO_RANGE(sc_from, sc_to);

That's a very good suggestion, thanks!

Berto
diff mbox series

Patch

diff --git a/block/qcow2-cluster.c b/block/qcow2-cluster.c
index ceacd91ea3..dfd8b66958 100644
--- a/block/qcow2-cluster.c
+++ b/block/qcow2-cluster.c
@@ -1006,6 +1006,23 @@  int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
         assert((offset & L2E_OFFSET_MASK) == offset);
 
         set_l2_entry(s, l2_slice, l2_index + i, offset | QCOW_OFLAG_COPIED);
+
+        /* Update bitmap with the subclusters that were just written */
+        if (has_subclusters(s)) {
+            unsigned written_from = m->cow_start.offset;
+            unsigned written_to = m->cow_end.offset + m->cow_end.nb_bytes ?:
+                m->nb_clusters << s->cluster_bits;
+            uint64_t l2_bitmap = get_l2_bitmap(s, l2_slice, l2_index + i);
+            int sc;
+            for (sc = 0; sc < s->subclusters_per_cluster; sc++) {
+                int 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);
+        }
      }