[Qemu-devel] [PATCH] block/qcow2: fix logic around dirty_bitmaps_loaded

Vladimir Sementsov-Ogievskiy posted 1 patch 5 years, 9 months ago
Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/20180612172614.60107-1-vsementsov@virtuozzo.com
Test checkpatch passed
Test docker-mingw@fedora passed
Test docker-quick@centos7 passed
Test s390x passed
block/qcow2.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
[Qemu-devel] [PATCH] block/qcow2: fix logic around dirty_bitmaps_loaded
Posted by Vladimir Sementsov-Ogievskiy 5 years, 9 months ago
First: this variable was introduced to handle reopens. We need it on
following qcow2_do_open, to don't try loading bitmaps again. So, we are
fixing qcow2_invalidate_cache().

However, if we fix only qcow2_invalidate_cache, iotest 169 fails on
case test__persistent__not_migbitmap__online_shared, because on first
target open, source is not yet closed, and is not yet stored bitmaps,
so, we are load nothing. And on second open, dirty_bitmaps_loaded is
already true, but we have no bitmaps to reopen with
qcow2_reopen_bitmaps_rw_hint(). So, we are fixing qcow2_do_open() too.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
 block/qcow2.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/block/qcow2.c b/block/qcow2.c
index 6fa5e1d71a..b4216a5830 100644
--- a/block/qcow2.c
+++ b/block/qcow2.c
@@ -1499,7 +1499,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
         {
             qcow2_reopen_bitmaps_rw_hint(bs, &header_updated, &local_err);
         }
-    } else {
+    } else if (s->nb_bitmaps > 0) {
         header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
         s->dirty_bitmaps_loaded = true;
     }
@@ -2178,6 +2178,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
     QDict *options;
     Error *local_err = NULL;
     int ret;
+    bool dirty_bitmaps_loaded = s->dirty_bitmaps_loaded;
 
     /*
      * Backing files are read-only which makes all of their metadata immutable,
@@ -2190,6 +2191,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
     qcow2_close(bs);
 
     memset(s, 0, sizeof(BDRVQcow2State));
+    s->dirty_bitmaps_loaded = dirty_bitmaps_loaded;
     options = qdict_clone_shallow(bs->options);
 
     flags &= ~BDRV_O_INACTIVE;
-- 
2.11.1


Re: [Qemu-devel] [Qemu-block] [PATCH] block/qcow2: fix logic around dirty_bitmaps_loaded
Posted by John Snow 5 years, 9 months ago

On 06/12/2018 01:26 PM, Vladimir Sementsov-Ogievskiy wrote:
> First: this variable was introduced to handle reopens. We need it on
> following qcow2_do_open, to don't try loading bitmaps again. So, we are
> fixing qcow2_invalidate_cache().
> 
> However, if we fix only qcow2_invalidate_cache, iotest 169 fails on
> case test__persistent__not_migbitmap__online_shared, because on first
> target open, source is not yet closed, and is not yet stored bitmaps,
> so, we are load nothing. And on second open, dirty_bitmaps_loaded is
> already true, but we have no bitmaps to reopen with
> qcow2_reopen_bitmaps_rw_hint(). So, we are fixing qcow2_do_open() too.
> 
> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> ---
>  block/qcow2.c | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/block/qcow2.c b/block/qcow2.c
> index 6fa5e1d71a..b4216a5830 100644
> --- a/block/qcow2.c
> +++ b/block/qcow2.c
> @@ -1499,7 +1499,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
>          {
>              qcow2_reopen_bitmaps_rw_hint(bs, &header_updated, &local_err);
>          }
> -    } else {
> +    } else if (s->nb_bitmaps > 0) {
>          header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
>          s->dirty_bitmaps_loaded = true;
>      }
> @@ -2178,6 +2178,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>      QDict *options;
>      Error *local_err = NULL;
>      int ret;
> +    bool dirty_bitmaps_loaded = s->dirty_bitmaps_loaded;
>  
>      /*
>       * Backing files are read-only which makes all of their metadata immutable,
> @@ -2190,6 +2191,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>      qcow2_close(bs);
>  
>      memset(s, 0, sizeof(BDRVQcow2State));
> +    s->dirty_bitmaps_loaded = dirty_bitmaps_loaded;
>      options = qdict_clone_shallow(bs->options);
>  
>      flags &= ~BDRV_O_INACTIVE;
> 

Ah, tch... I didn't realize that invalidate completely wipes the qcow2
metadata.

I guess the other three fields, nb_bitmaps, bitmap_directory_size and
bitmap_directory_offset get initialized again on re-open.

(I suppose this means I need to rethink caching bm_list more carefully,
too...)

I think this looks correct mechanically.

Re: [Qemu-devel] [Qemu-block] [PATCH] block/qcow2: fix logic around dirty_bitmaps_loaded
Posted by John Snow 5 years, 9 months ago

On 06/12/2018 06:11 PM, John Snow wrote:
> 
> 
> On 06/12/2018 01:26 PM, Vladimir Sementsov-Ogievskiy wrote:
>> First: this variable was introduced to handle reopens. We need it on
>> following qcow2_do_open, to don't try loading bitmaps again. So, we are
>> fixing qcow2_invalidate_cache().
>>
>> However, if we fix only qcow2_invalidate_cache, iotest 169 fails on
>> case test__persistent__not_migbitmap__online_shared, because on first
>> target open, source is not yet closed, and is not yet stored bitmaps,
>> so, we are load nothing. And on second open, dirty_bitmaps_loaded is
>> already true, but we have no bitmaps to reopen with
>> qcow2_reopen_bitmaps_rw_hint(). So, we are fixing qcow2_do_open() too.
>>
>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
>> ---
>>  block/qcow2.c | 4 +++-
>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>
>> diff --git a/block/qcow2.c b/block/qcow2.c
>> index 6fa5e1d71a..b4216a5830 100644
>> --- a/block/qcow2.c
>> +++ b/block/qcow2.c
>> @@ -1499,7 +1499,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
>>          {
>>              qcow2_reopen_bitmaps_rw_hint(bs, &header_updated, &local_err);
>>          }
>> -    } else {
>> +    } else if (s->nb_bitmaps > 0) {
>>          header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
>>          s->dirty_bitmaps_loaded = true;
>>      }
>> @@ -2178,6 +2178,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>>      QDict *options;
>>      Error *local_err = NULL;
>>      int ret;
>> +    bool dirty_bitmaps_loaded = s->dirty_bitmaps_loaded;
>>  
>>      /*
>>       * Backing files are read-only which makes all of their metadata immutable,
>> @@ -2190,6 +2191,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>>      qcow2_close(bs);
>>  
>>      memset(s, 0, sizeof(BDRVQcow2State));
>> +    s->dirty_bitmaps_loaded = dirty_bitmaps_loaded;
>>      options = qdict_clone_shallow(bs->options);
>>  
>>      flags &= ~BDRV_O_INACTIVE;
>>
> 
> Ah, tch... I didn't realize that invalidate completely wipes the qcow2
> metadata.
> 
> I guess the other three fields, nb_bitmaps, bitmap_directory_size and
> bitmap_directory_offset get initialized again on re-open.
> 
> (I suppose this means I need to rethink caching bm_list more carefully,
> too...)
> 
> I think this looks correct mechanically.
> 

Oh, I meant:

Reviewed-by: John Snow <jsnow@redhat.com>

Re: [Qemu-devel] [Qemu-block] [PATCH] block/qcow2: fix logic around dirty_bitmaps_loaded
Posted by Vladimir Sementsov-Ogievskiy 5 years, 9 months ago
13.06.2018 01:18, John Snow wrote:
>
> On 06/12/2018 06:11 PM, John Snow wrote:
>>
>> On 06/12/2018 01:26 PM, Vladimir Sementsov-Ogievskiy wrote:
>>> First: this variable was introduced to handle reopens. We need it on
>>> following qcow2_do_open, to don't try loading bitmaps again. So, we are
>>> fixing qcow2_invalidate_cache().
>>>
>>> However, if we fix only qcow2_invalidate_cache, iotest 169 fails on
>>> case test__persistent__not_migbitmap__online_shared, because on first
>>> target open, source is not yet closed, and is not yet stored bitmaps,
>>> so, we are load nothing. And on second open, dirty_bitmaps_loaded is
>>> already true, but we have no bitmaps to reopen with
>>> qcow2_reopen_bitmaps_rw_hint(). So, we are fixing qcow2_do_open() too.
>>>
>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
>>> ---
>>>   block/qcow2.c | 4 +++-
>>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/block/qcow2.c b/block/qcow2.c
>>> index 6fa5e1d71a..b4216a5830 100644
>>> --- a/block/qcow2.c
>>> +++ b/block/qcow2.c
>>> @@ -1499,7 +1499,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
>>>           {
>>>               qcow2_reopen_bitmaps_rw_hint(bs, &header_updated, &local_err);
>>>           }
>>> -    } else {
>>> +    } else if (s->nb_bitmaps > 0) {
>>>           header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
>>>           s->dirty_bitmaps_loaded = true;
>>>       }
>>> @@ -2178,6 +2178,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>>>       QDict *options;
>>>       Error *local_err = NULL;
>>>       int ret;
>>> +    bool dirty_bitmaps_loaded = s->dirty_bitmaps_loaded;
>>>   
>>>       /*
>>>        * Backing files are read-only which makes all of their metadata immutable,
>>> @@ -2190,6 +2191,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>>>       qcow2_close(bs);
>>>   
>>>       memset(s, 0, sizeof(BDRVQcow2State));
>>> +    s->dirty_bitmaps_loaded = dirty_bitmaps_loaded;
>>>       options = qdict_clone_shallow(bs->options);
>>>   
>>>       flags &= ~BDRV_O_INACTIVE;
>>>
>> Ah, tch... I didn't realize that invalidate completely wipes the qcow2
>> metadata.
>>
>> I guess the other three fields, nb_bitmaps, bitmap_directory_size and
>> bitmap_directory_offset get initialized again on re-open.
>>
>> (I suppose this means I need to rethink caching bm_list more carefully,
>> too...)
>>
>> I think this looks correct mechanically.
>>
> Oh, I meant:
>
> Reviewed-by: John Snow <jsnow@redhat.com>

It is interesting, this patch also fixes iotest 169. It's not broken 
formally, but if we look at vm output of vm_b after migration, we'll see

qemu-system-x86_64: Could not reopen qcow2 layer: Bitmap already exists: 
bitmap0

which leads to failed invalidation and bs->drv = NULL :)

I'll improve iotest somehow, to show this failure.

-- 
Best regards,
Vladimir


Re: [Qemu-devel] [Qemu-block] [PATCH] block/qcow2: fix logic around dirty_bitmaps_loaded
Posted by Vladimir Sementsov-Ogievskiy 5 years, 9 months ago
20.06.2018 20:17, Vladimir Sementsov-Ogievskiy wrote:
> 13.06.2018 01:18, John Snow wrote:
>>
>> On 06/12/2018 06:11 PM, John Snow wrote:
>>>
>>> On 06/12/2018 01:26 PM, Vladimir Sementsov-Ogievskiy wrote:
>>>> First: this variable was introduced to handle reopens. We need it on
>>>> following qcow2_do_open, to don't try loading bitmaps again. So, we 
>>>> are
>>>> fixing qcow2_invalidate_cache().
>>>>
>>>> However, if we fix only qcow2_invalidate_cache, iotest 169 fails on
>>>> case test__persistent__not_migbitmap__online_shared, because on first
>>>> target open, source is not yet closed, and is not yet stored bitmaps,
>>>> so, we are load nothing. And on second open, dirty_bitmaps_loaded is
>>>> already true, but we have no bitmaps to reopen with
>>>> qcow2_reopen_bitmaps_rw_hint(). So, we are fixing qcow2_do_open() too.
>>>>
>>>> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
>>>> ---
>>>>   block/qcow2.c | 4 +++-
>>>>   1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/block/qcow2.c b/block/qcow2.c
>>>> index 6fa5e1d71a..b4216a5830 100644
>>>> --- a/block/qcow2.c
>>>> +++ b/block/qcow2.c
>>>> @@ -1499,7 +1499,7 @@ static int coroutine_fn 
>>>> qcow2_do_open(BlockDriverState *bs, QDict *options,
>>>>           {
>>>>               qcow2_reopen_bitmaps_rw_hint(bs, &header_updated, 
>>>> &local_err);
>>>>           }
>>>> -    } else {
>>>> +    } else if (s->nb_bitmaps > 0) {
>>>>           header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
>>>>           s->dirty_bitmaps_loaded = true;
>>>>       }
>>>> @@ -2178,6 +2178,7 @@ static void coroutine_fn 
>>>> qcow2_co_invalidate_cache(BlockDriverState *bs,
>>>>       QDict *options;
>>>>       Error *local_err = NULL;
>>>>       int ret;
>>>> +    bool dirty_bitmaps_loaded = s->dirty_bitmaps_loaded;
>>>>         /*
>>>>        * Backing files are read-only which makes all of their 
>>>> metadata immutable,
>>>> @@ -2190,6 +2191,7 @@ static void coroutine_fn 
>>>> qcow2_co_invalidate_cache(BlockDriverState *bs,
>>>>       qcow2_close(bs);
>>>>         memset(s, 0, sizeof(BDRVQcow2State));
>>>> +    s->dirty_bitmaps_loaded = dirty_bitmaps_loaded;
>>>>       options = qdict_clone_shallow(bs->options);
>>>>         flags &= ~BDRV_O_INACTIVE;
>>>>
>>> Ah, tch... I didn't realize that invalidate completely wipes the qcow2
>>> metadata.
>>>
>>> I guess the other three fields, nb_bitmaps, bitmap_directory_size and
>>> bitmap_directory_offset get initialized again on re-open.
>>>
>>> (I suppose this means I need to rethink caching bm_list more carefully,
>>> too...)
>>>
>>> I think this looks correct mechanically.
>>>
>> Oh, I meant:
>>
>> Reviewed-by: John Snow <jsnow@redhat.com>
>
> It is interesting, this patch also fixes iotest 169. It's not broken 
> formally, but if we look at vm output of vm_b after migration, we'll see
>
> qemu-system-x86_64: Could not reopen qcow2 layer: Bitmap already 
> exists: bitmap0
>
> which leads to failed invalidation and bs->drv = NULL :)
>
> I'll improve iotest somehow, to show this failure.
>

the bug can be shown with the following workaround:

diff --git a/tests/qemu-iotests/169 b/tests/qemu-iotests/169
index f243db9955..6dce24d29f 100755
--- a/tests/qemu-iotests/169
+++ b/tests/qemu-iotests/169
@@ -131,9 +131,13 @@ class TestDirtyBitmapMigration(iotests.QMPTestCase):
                  break

          self.check_bitmap(self.vm_b, sha256 if should_migrate else False)
+        self.vm_b.hmp_qemu_io('drive0', 'write 0 512')

          if should_migrate:
              self.vm_b.shutdown()
+            print '===='
+            print self.vm_b.get_log()
+            print '===='
              # recreate vm_b, as we don't want -incoming option (this 
will lead
              # to "cat" process left alive after test finish)
              self.vm_b = iotests.VM(path_suffix='b')
@@ -152,7 +156,8 @@ for cmb in list(itertools.product((True, False), 
repeat=4)):
      name += '_online' if cmb[2] else '_offline'
      name += '_shared' if cmb[3] else '_nonshared'

-    inject_test_case(TestDirtyBitmapMigration, name, 'do_test_migration',
+    if name == '_persistent__not_migbitmap__offline_shared':
+        inject_test_case(TestDirtyBitmapMigration, name, 
'do_test_migration',
                       *list(cmb))




-- 
Best regards,
Vladimir


[Qemu-devel] drop Re: [PATCH] block/qcow2: fix logic around dirty_bitmaps_loaded
Posted by Vladimir Sementsov-Ogievskiy 5 years, 9 months ago
12.06.2018 20:26, Vladimir Sementsov-Ogievskiy wrote:
> First: this variable was introduced to handle reopens. We need it on
> following qcow2_do_open, to don't try loading bitmaps again. So, we are
> fixing qcow2_invalidate_cache().
>
> However, if we fix only qcow2_invalidate_cache, iotest 169 fails on
> case test__persistent__not_migbitmap__online_shared, because on first
> target open, source is not yet closed, and is not yet stored bitmaps,
> so, we are load nothing. And on second open, dirty_bitmaps_loaded is
> already true, but we have no bitmaps to reopen with
> qcow2_reopen_bitmaps_rw_hint(). So, we are fixing qcow2_do_open() too.
>
> Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
> ---
>   block/qcow2.c | 4 +++-
>   1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/block/qcow2.c b/block/qcow2.c
> index 6fa5e1d71a..b4216a5830 100644
> --- a/block/qcow2.c
> +++ b/block/qcow2.c
> @@ -1499,7 +1499,7 @@ static int coroutine_fn qcow2_do_open(BlockDriverState *bs, QDict *options,
>           {
>               qcow2_reopen_bitmaps_rw_hint(bs, &header_updated, &local_err);
>           }
> -    } else {
> +    } else if (s->nb_bitmaps > 0) {
>           header_updated = qcow2_load_dirty_bitmaps(bs, &local_err);
>           s->dirty_bitmaps_loaded = true;
>       }
> @@ -2178,6 +2178,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>       QDict *options;
>       Error *local_err = NULL;
>       int ret;
> +    bool dirty_bitmaps_loaded = s->dirty_bitmaps_loaded;
>   
>       /*
>        * Backing files are read-only which makes all of their metadata immutable,
> @@ -2190,6 +2191,7 @@ static void coroutine_fn qcow2_co_invalidate_cache(BlockDriverState *bs,
>       qcow2_close(bs);
>   
>       memset(s, 0, sizeof(BDRVQcow2State));
> +    s->dirty_bitmaps_loaded = dirty_bitmaps_loaded;
>       options = qdict_clone_shallow(bs->options);
>   
>       flags &= ~BDRV_O_INACTIVE;

Drop this, I'll resend better patch.

-- 
Best regards,
Vladimir