Skip to contentSkip to conversation

block: Update image size in bdrv_invalidate_cache()

PATCH Applied Kevin Wolf 4 participants 6 messages Mar 11, 2014 – Mar 12, 2014 Block layer

lore
Share & tools
lore

Kevin Wolf

author
#0 lore raw

PATCH 1/1 block: Update image size in bdrv_invalidate_cache()

After migration has completed, we call bdrv_invalidate_cache() so that drivers which cache some data drop their stale copy of the data and reread it from the image file to get a new version of data that the source modified while the migration was running.

Reloading metadata from the image file is useless, though, if the size of the image file stays stale (this is a value that is cached for all image formats in block.c). Reads from (meta)data after the old EOF return only zeroes, causing image corruption.

We need to update bs->total_sectors in all layers that could potentially have changed their size (i.e. backing files are not a concern - if they are changed, we're in bigger trouble)

---
 block.c       | 10 +++++++++-
 block/qcow2.c |  2 ++
 block/qed.c   |  3 +++
 3 files changed, 14 insertions(+), 1 deletion(-)
block.c +9 -1
index f1ef4b0..7b306fb 100644
@@ -4776,9 +4776,17 @@ flush_parent:
 
 void bdrv_invalidate_cache(BlockDriverState *bs)
 {
-    if (bs->drv && bs->drv->bdrv_invalidate_cache) {
+    if (!bs->drv)  {
+        return;
+    }
+
+    if (bs->drv->bdrv_invalidate_cache) {
         bs->drv->bdrv_invalidate_cache(bs);
+    } else if (bs->file) {
+        bdrv_invalidate_cache(bs->file);
     }
+
+    refresh_total_sectors(bs, bs->total_sectors);
 }
 
 void bdrv_invalidate_cache_all(void)
block/qcow2.c +2 -0
index cfe80be..b5b1e8c 100644
@@ -1176,6 +1176,8 @@ static void qcow2_invalidate_cache(BlockDriverState *bs)
 
     qcow2_close(bs);
 
+    bdrv_invalidate_cache(bs->file);
+
     options = qdict_new();
     qdict_put(options, QCOW2_OPT_LAZY_REFCOUNTS,
               qbool_from_int(s->use_lazy_refcounts));
block/qed.c +3 -0
index 8802ad3..837accd 100644
@@ -1563,6 +1563,9 @@ static void bdrv_qed_invalidate_cache(BlockDriverState *bs)
     BDRVQEDState *s = bs->opaque;
 
     bdrv_qed_close(bs);
+
+    bdrv_invalidate_cache(bs->file);
+
     memset(s, 0, sizeof(BDRVQEDState));
     bdrv_qed_open(bs, NULL, bs->open_flags, NULL);
 }
Signature · 1 line
1.8.1.4
Reviewed-by Tested-by on v1 1/1
#1 lore raw
Quoted 10 lines · Kevin Wolf

After migration has completed, we call bdrv_invalidate_cache() so that drivers which cache some data drop their stale copy of the data and reread it from the image file to get a new version of data that the source modified while the migration was running.

Reloading metadata from the image file is useless, though, if the size of the image file stays stale (this is a value that is cached for all image formats in block.c). Reads from (meta)data after the old EOF return only zeroes, causing image corruption.

Quoted 5 lines

We need to update bs->total_sectors in all layers that could potentially have changed their size (i.e. backing files are not a concern - if they are changed, we're in bigger trouble)

I had problems with migration with qcow2 files, and this fixes it.

#2 lore raw
In reply to Kevin Wolf
Quoted 14 lines · Kevin Wolf

After migration has completed, we call bdrv_invalidate_cache() so that drivers which cache some data drop their stale copy of the data and reread it from the image file to get a new version of data that the source modified while the migration was running.

Reloading metadata from the image file is useless, though, if the size of the image file stays stale (this is a value that is cached for all image formats in block.c). Reads from (meta)data after the old EOF return only zeroes, causing image corruption.

We need to update bs->total_sectors in all layers that could potentially have changed their size (i.e. backing files are not a concern - if they are changed, we're in bigger trouble)

Quoted 6 lines
---
 block.c       | 10 +++++++++-
 block/qcow2.c |  2 ++
 block/qed.c   |  3 +++
 3 files changed, 14 insertions(+), 1 deletion(-)

Thanks, applied to my block tree:
https://github.com/stefanha/qemu/commits/block

Stefan

Reviewed-by Benoit Canet on v1 1/1
#3 lore raw
In reply to Kevin Wolf
Quoted 75 lines · Kevin Wolf

After migration has completed, we call bdrv_invalidate_cache() so that drivers which cache some data drop their stale copy of the data and reread it from the image file to get a new version of data that the source modified while the migration was running.

Reloading metadata from the image file is useless, though, if the size of the image file stays stale (this is a value that is cached for all image formats in block.c). Reads from (meta)data after the old EOF return only zeroes, causing image corruption.

We need to update bs->total_sectors in all layers that could potentially have changed their size (i.e. backing files are not a concern - if they are changed, we're in bigger trouble)

---
 block.c       | 10 +++++++++-
 block/qcow2.c |  2 ++
 block/qed.c   |  3 +++
 3 files changed, 14 insertions(+), 1 deletion(-)
block.c +9 -1
index f1ef4b0..7b306fb 100644
@@ -4776,9 +4776,17 @@ flush_parent:
 
 void bdrv_invalidate_cache(BlockDriverState *bs)
 {
-    if (bs->drv && bs->drv->bdrv_invalidate_cache) {
+    if (!bs->drv)  {
+        return;
+    }
+
+    if (bs->drv->bdrv_invalidate_cache) {
         bs->drv->bdrv_invalidate_cache(bs);
+    } else if (bs->file) {
+        bdrv_invalidate_cache(bs->file);
     }
+
+    refresh_total_sectors(bs, bs->total_sectors);
 }
 
 void bdrv_invalidate_cache_all(void)
block/qcow2.c +2 -0
index cfe80be..b5b1e8c 100644
@@ -1176,6 +1176,8 @@ static void qcow2_invalidate_cache(BlockDriverState *bs)
 
     qcow2_close(bs);
 
+    bdrv_invalidate_cache(bs->file);
+
     options = qdict_new();
     qdict_put(options, QCOW2_OPT_LAZY_REFCOUNTS,
               qbool_from_int(s->use_lazy_refcounts));
block/qed.c +3 -0
index 8802ad3..837accd 100644
@@ -1563,6 +1563,9 @@ static void bdrv_qed_invalidate_cache(BlockDriverState *bs)
     BDRVQEDState *s = bs->opaque;
 
     bdrv_qed_close(bs);
+
+    bdrv_invalidate_cache(bs->file);
+
     memset(s, 0, sizeof(BDRVQEDState));
     bdrv_qed_open(bs, NULL, bs->open_flags, NULL);
 }
Signature · 1 line
1.8.1.4

I have the impression that you are silently fixing other nits;

However:

author
#4 lore raw

Am 11.03.2014 um 17:41 hat Benoît Canet geschrieben:

Quoted 1 line

I have the impression that you are silently fixing other nits;

Am I?

Not intentionally at least. I think I can justify every single line of the patch with respect to the bug described in the commit message. We need to recursive into the protocol layer, and we need to refresh bs->total_sectors. For drivers implementing the callback, we need to recurse to bs->file between the internal open/close pair.

I can't see what could be left out in this patch.

Kevin

#5 lore raw
Quoted 5 lines · Kevin Wolf

Am 11.03.2014 um 17:41 hat Benoît Canet geschrieben:

Quoted 1 line

I have the impression that you are silently fixing other nits;

Am I?

Quoted 7 lines

Not intentionally at least. I think I can justify every single line of the patch with respect to the bug described in the commit message. We need to recursive into the protocol layer, and we need to refresh bs->total_sectors. For drivers implementing the callback, we need to recurse to bs->file between the internal open/close pair.

I can't see what could be left out in this patch.

I didn't understood the implication on recursion at first sight.

Best regards

Benoît

Quoted 3 lines

Kevin