Skip to contentSkip to conversation

migration/block-dirty-bitmap: fix add_bitmaps_to_list

PATCH In review Vladimir Sementsov-Ogievskiy 2 participants 3 messages Jun 26, 2020

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 migration/block-dirty-bitmap: fix add_bitmaps_to_list

We shouldn't fail, if found unnamed bitmap in a unnamed node or node with auto-generated node name, as bitmap migration ignores such bitmaps at all.

Fixes 82640edb88faa
Fixes 4ff5cc121b089
---
 migration/block-dirty-bitmap.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
migration/block-dirty-bitmap.c +5 -1
index 47bc0f650c..b0dbf9eeed 100644
@@ -274,7 +274,11 @@ static int add_bitmaps_to_list(BlockDriverState *bs, const char *bs_name)
     DirtyBitmapMigBitmapState *dbms;
     Error *local_err = NULL;
 
-    bitmap = bdrv_dirty_bitmap_first(bs);
+    FOR_EACH_DIRTY_BITMAP(bs, bitmap) {
+        if (bdrv_dirty_bitmap_name(bitmap)) {
+            break;
+        }
+    }
     if (!bitmap) {
         return 0;
     }
Signature · 1 line
2.18.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 1 line · Vladimir Sementsov-Ogievskiy

We shouldn't fail, if found unnamed bitmap in a unnamed node or node

We shouldn't fail when finding an unnamed

Quoted 2 lines

with auto-generated node name, as bitmap migration ignores such bitmaps at all.

such bitmaps in the first place.

Quoted 8 lines
Fixes 82640edb88faa
Fixes 4ff5cc121b089
---
  migration/block-dirty-bitmap.c | 6 +++++-
  1 file changed, 5 insertions(+), 1 deletion(-)

Will queue through the bitmaps tree.

Is this easy enough to reproduce that it would be worth having iotest coverage?

Quoted 18 lines
migration/block-dirty-bitmap.c +5 -1
index 47bc0f650c..b0dbf9eeed 100644
@@ -274,7 +274,11 @@ static int add_bitmaps_to_list(BlockDriverState *bs, const char *bs_name)
      DirtyBitmapMigBitmapState *dbms;
      Error *local_err = NULL;
 
-    bitmap = bdrv_dirty_bitmap_first(bs);
+    FOR_EACH_DIRTY_BITMAP(bs, bitmap) {
+        if (bdrv_dirty_bitmap_name(bitmap)) {
+            break;
+        }
+    }
      if (!bitmap) {
          return 0;
      }
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#2 lore raw
Quoted 19 lines · Eric Blake
Quoted 1 line · Vladimir Sementsov-Ogievskiy

We shouldn't fail, if found unnamed bitmap in a unnamed node or node

We shouldn't fail when finding an unnamed

Quoted 2 lines

with auto-generated node name, as bitmap migration ignores such bitmaps at all.

such bitmaps in the first place.

Quoted 8 lines
Fixes 82640edb88faa
Fixes 4ff5cc121b089

---
  migration/block-dirty-bitmap.c | 6 +++++-

  1 file changed, 5 insertions(+), 1 deletion(-)
Quoted 5 lines

Will queue through the bitmaps tree.

Is this easy enough to reproduce that it would be worth having iotest coverage?

I don't think it worth it. This all is so bad, I only hope for the future of explicit mapping.

This patch just fixes wrong code. Actually, today I've committed to downstream also a change, which when see auto-generated node-name a few lines above just returns 0 instead of an error, otherwise, in some specific case, bitmaps which were in backing files break migrations (and when skipped, they migrate through storage, as our specific case is storage migration).. Don't know, how much it applies to upstream.

I've already said about my idea of "lazy" disabled bitmap: we shouldn't load them all on start, but on demand. Actually same thing may be said about bitmaps in read-only nodes. And of course migrating bitmaps from r-o backing chain, when it is shared-migration is nonsense.

Quoted 10 lines
Quoted 10 lines
migration/block-dirty-bitmap.c +1 -1
index 47bc0f650c..b0dbf9eeed 100644
@@ -274,7 +274,11 @@ static int add_bitmaps_to_list(BlockDriverState *bs, const char *bs_name)
       DirtyBitmapMigBitmapState *dbms;
       Error *local_err = NULL;
-    bitmap = bdrv_dirty_bitmap_first(bs);
+    FOR_EACH_DIRTY_BITMAP(bs, bitmap) {
Quoted 9 lines

migration/block-dirty-bitmap.c static int add_bitmaps_to_list(BlockDriverState *bs, const char *bs_name) · line 284

Quoted 8 lines
migration/block-dirty-bitmap.c +4 -0
+        if (bdrv_dirty_bitmap_name(bitmap)) {
+            break;
+        }
+    }
       if (!bitmap) {
           return 0;
       }
 

Here is my today's downstream fix:

migration/block-dirty-bitmap.c +6 -4
index e8ce791278..66f5d6365a 100644
@@ -312,10 +312,12 @@ static int add_bitmaps_to_list(DBMSaveState *s,
     }
 
     if (bs_name[0] == '#') {
-        error_report("Found bitmap '%s' in a node with auto-generated "
-                     "name: %s. It can't be migrated",
-                     bdrv_dirty_bitmap_name(bitmap), bs_name);
-        return -1;
+        /*
+         * Reporting error here breaks shared migration of dispatcher-created
+         * bitmaps. In u13 we just ignored them (ignored all bitmaps in backing
+         * files). So, just ignore the node.
+         */
+        return 0;
     }
 
     for ( ; bitmap; bitmap = bdrv_dirty_bitmap_next(bs, bitmap)) {

I can send a patch, if you think it worth it. For me it seems too hacky and too specific. With this change, bitmaps in backing chain can be migrated though shared-storage when dirty-bitmaps capability is enabled.

(for clarity: u13 didn't have patch
592203e7cfbd1a "migration/dirty-bitmaps: change bitmap enumeration method", and in u14 we've backported it)

Signature · 2 lines
Best regards,
Vladimir