Skip to contentSkip to conversation

qcow2: Make qcow2_alloc_bytes() more explicit

PATCH Applied Max Reitz 2 participants 2 messages Sep 11, 2015 Block layer

lore
Share & tools
lore

Max Reitz

author
#0 lore raw

PATCH 1/1 qcow2: Make qcow2_alloc_bytes() more explicit

In case of -EAGAIN returned by update_refcount(), we should discard the cluster offset we were trying to allocate and request a new one, because in theory that old offset might now be taken by a refcount block.

In practice, this was not the case due to update_refcount() generally returning strictly monotonic increasing cluster offsets. However, this behavior is not set in stone, and it is also not obvious when looking at qcow2_alloc_bytes() alone, so we should not rely on it.

---
 block/qcow2-refcount.c | 6 ++++++
 1 file changed, 6 insertions(+)
block/qcow2-refcount.c +6 -0
index e8430ec..c30bb14 100644
@@ -949,11 +949,17 @@ int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
 
             if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
                 offset = new_cluster;
+                free_in_cluster = s->cluster_size;
+            } else {
+                free_in_cluster += s->cluster_size;
             }
         }
 
         assert(offset);
         ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
+        if (ret < 0) {
+            offset = 0;
+        }
     } while (ret == -EAGAIN);
     if (ret < 0) {
         return ret;
Signature · 1 line
2.5.1
Applied
#1 lore raw

Am 11.09.2015 um 18:47 hat Max Reitz geschrieben:

Quoted 20 lines

In case of -EAGAIN returned by update_refcount(), we should discard the cluster offset we were trying to allocate and request a new one, because in theory that old offset might now be taken by a refcount block.

In practice, this was not the case due to update_refcount() generally returning strictly monotonic increasing cluster offsets. However, this behavior is not set in stone, and it is also not obvious when looking at qcow2_alloc_bytes() alone, so we should not rely on it.

---
 block/qcow2-refcount.c | 6 ++++++
 1 file changed, 6 insertions(+)
block/qcow2-refcount.c +0 -0
index e8430ec..c30bb14 100644

block/qcow2-refcount.c int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) · line 956

Quoted 8 lines
block/qcow2-refcount.c +3 -0
@@ -949,11 +949,17 @@ int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size)
 
             if (!offset || ROUND_UP(offset, s->cluster_size) != new_cluster) {
                 offset = new_cluster;
+                free_in_cluster = s->cluster_size;
+            } else {
+                free_in_cluster += s->cluster_size;
             }
         }

This doesn't actually do anything except confuse the reader, but as the value of free_in_cluster doesn't matter in the second iteration because offset == 0, this is correct.

block/qcow2-refcount.c int64_t qcow2_alloc_bytes(BlockDriverState *bs, int size) · line 965

Quoted 8 lines
block/qcow2-refcount.c +3 -0
         assert(offset);
         ret = update_refcount(bs, offset, size, 1, false, QCOW2_DISCARD_NEVER);
+        if (ret < 0) {
+            offset = 0;
+        }
     } while (ret == -EAGAIN);
     if (ret < 0) {
         return ret;

Thanks, applied to the block branch.

Kevin