Re: [Qemu-devel] [PATCH 2/5] block/dirty-bitmap: Refactor bdrv_can_store_new_bitmap

[John Snow]

On 6/7/19 2:17 PM, Vladimir Sementsov-Ogievskiy wrote:
> 07.06.2019 21:10, John Snow wrote:
>>
>>
>> On 6/7/19 10:29 AM, Vladimir Sementsov-Ogievskiy wrote:
>>> 06.06.2019 21:41, John Snow wrote:
>>>> Instead of bdrv_can_store_new_bitmap, rework this as
>>>> bdrv_add_persistent_dirty_bitmap. This makes a more obvious symmetry
>>>> with bdrv_remove_persistent_dirty_bitmap. Most importantly, we are free
>>>> to modify the driver state because we know we ARE adding a bitmap
>>>> instead of simply being asked if we CAN store one.
>>>>
>>>> As part of this symmetry, move this function next to
>>>> bdrv_remove_persistent_bitmap in block/dirty-bitmap.c.
>>>>
>>>> To cement this semantic distinction, use a bitmap object instead of the
>>>> (name, granularity) pair as this allows us to set persistence as a
>>>> condition of the "add" succeeding.
>>>>
>>>> Notably, the qcow2 implementation still does not actually store the new
>>>> bitmap at add time, but merely ensures that it will be able to at store
>>>> time.
>>>>
>>>> Signed-off-by: John Snow <address@hidden>
>>>> ---
>>>>    block/qcow2.h                |  5 ++---
>>>>    include/block/block.h        |  2 --
>>>>    include/block/block_int.h    |  5 ++---
>>>>    include/block/dirty-bitmap.h |  3 +++
>>>>    block.c                      | 22 ---------------------
>>>>    block/dirty-bitmap.c         | 38 ++++++++++++++++++++++++++++++++++++
>>>>    block/qcow2-bitmap.c         | 24 ++++++++++++++---------
>>>>    block/qcow2.c                |  2 +-
>>>>    blockdev.c                   | 19 +++++++-----------
>>>>    9 files changed, 68 insertions(+), 52 deletions(-)
>>>>
>>>> diff --git a/block/qcow2.h b/block/qcow2.h
>>>> index fc1b0d3c1e..95d723d3c0 100644
>>>> --- a/block/qcow2.h
>>>> +++ b/block/qcow2.h
>>>> @@ -742,9 +742,8 @@ int qcow2_reopen_bitmaps_rw(BlockDriverState *bs, 
>>>> Error **errp);
>>>>    int qcow2_truncate_bitmaps_check(BlockDriverState *bs, Error **errp);
>>>>    void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs, Error 
>>>> **errp);
>>>>    int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, Error **errp);
>>>> -bool qcow2_can_store_new_dirty_bitmap(BlockDriverState *bs,
>>>> -                                      const char *name,
>>>> -                                      uint32_t granularity,
>>>> +int qcow2_add_persistent_dirty_bitmap(BlockDriverState *bs,
>>>> +                                      BdrvDirtyBitmap *bitmap,
>>>>                                          Error **errp);
>>>>    void qcow2_remove_persistent_dirty_bitmap(BlockDriverState *bs,
>>>>                                              const char *name,
>>>> diff --git a/include/block/block.h b/include/block/block.h
>>>> index f9415ed740..6d520f3b59 100644
>>>> --- a/include/block/block.h
>>>> +++ b/include/block/block.h
>>>> @@ -683,8 +683,6 @@ void bdrv_add_child(BlockDriverState *parent, 
>>>> BlockDriverState *child,
>>>>                        Error **errp);
>>>>    void bdrv_del_child(BlockDriverState *parent, BdrvChild *child, Error 
>>>> **errp);
>>>>    
>>>> -bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char 
>>>> *name,
>>>> -                                     uint32_t granularity, Error **errp);
>>>>    /**
>>>>     *
>>>>     * bdrv_register_buf/bdrv_unregister_buf:
>>>> diff --git a/include/block/block_int.h b/include/block/block_int.h
>>>> index 06df2bda1b..93bbb66cd0 100644
>>>> --- a/include/block/block_int.h
>>>> +++ b/include/block/block_int.h
>>>> @@ -537,9 +537,8 @@ struct BlockDriver {
>>>>         * field of BlockDirtyBitmap's in case of success.
>>>>         */
>>>>        int (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs, Error **errp);
>>>> -    bool (*bdrv_can_store_new_dirty_bitmap)(BlockDriverState *bs,
>>>> -                                            const char *name,
>>>> -                                            uint32_t granularity,
>>>> +    int (*bdrv_add_persistent_dirty_bitmap)(BlockDriverState *bs,
>>>> +                                            BdrvDirtyBitmap *bitmap,
>>>>                                                Error **errp);
>>>>        void (*bdrv_remove_persistent_dirty_bitmap)(BlockDriverState *bs,
>>>>                                                    const char *name,
>>>> diff --git a/include/block/dirty-bitmap.h b/include/block/dirty-bitmap.h
>>>> index 8044ace63e..c37edbe05f 100644
>>>> --- a/include/block/dirty-bitmap.h
>>>> +++ b/include/block/dirty-bitmap.h
>>>> @@ -38,6 +38,9 @@ int bdrv_dirty_bitmap_check(const BdrvDirtyBitmap 
>>>> *bitmap, uint32_t flags,
>>>>                                Error **errp);
>>>>    void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap 
>>>> *bitmap);
>>>>    void bdrv_release_named_dirty_bitmaps(BlockDriverState *bs);
>>>> +int bdrv_add_persistent_dirty_bitmap(BlockDriverState *bs,
>>>> +                                      BdrvDirtyBitmap *bitmap,
>>>> +                                      Error **errp);
>>>>    void bdrv_remove_persistent_dirty_bitmap(BlockDriverState *bs,
>>>>                                             const char *name,
>>>>                                             Error **errp);
>>>> diff --git a/block.c b/block.c
>>>> index e3e77feee0..6aec36b7c9 100644
>>>> --- a/block.c
>>>> +++ b/block.c
>>>> @@ -6379,25 +6379,3 @@ void bdrv_del_child(BlockDriverState *parent_bs, 
>>>> BdrvChild *child, Error **errp)
>>>>    
>>>>        parent_bs->drv->bdrv_del_child(parent_bs, child, errp);
>>>>    }
>>>> -
>>>> -bool bdrv_can_store_new_dirty_bitmap(BlockDriverState *bs, const char 
>>>> *name,
>>>> -                                     uint32_t granularity, Error **errp)
>>>> -{
>>>> -    BlockDriver *drv = bs->drv;
>>>> -
>>>> -    if (!drv) {
>>>> -        error_setg_errno(errp, ENOMEDIUM,
>>>> -                         "Can't store persistent bitmaps to %s",
>>>> -                         bdrv_get_device_or_node_name(bs));
>>>> -        return false;
>>>> -    }
>>>> -
>>>> -    if (!drv->bdrv_can_store_new_dirty_bitmap) {
>>>> -        error_setg_errno(errp, ENOTSUP,
>>>> -                         "Can't store persistent bitmaps to %s",
>>>> -                         bdrv_get_device_or_node_name(bs));
>>>> -        return false;
>>>> -    }
>>>> -
>>>> -    return drv->bdrv_can_store_new_dirty_bitmap(bs, name, granularity, 
>>>> errp);
>>>> -}
>>>> diff --git a/block/dirty-bitmap.c b/block/dirty-bitmap.c
>>>> index 49646a30e6..615f8480b2 100644
>>>> --- a/block/dirty-bitmap.c
>>>> +++ b/block/dirty-bitmap.c
>>>> @@ -466,6 +466,44 @@ void 
>>>> bdrv_remove_persistent_dirty_bitmap(BlockDriverState *bs,
>>>>        }
>>>>    }
>>>>    
>>>> +int bdrv_add_persistent_dirty_bitmap(BlockDriverState *bs,
>>>> +                                     BdrvDirtyBitmap *bitmap,
>>>> +                                     Error **errp)
>>>
>>> strange thing for me: "bitmap" in function name is _not_ the same
>>> thing as @bitmap argument.. as if it the same,
>>> function adds "persistent bitmap", but @bitmap is not persistent yet,
>>> so may be function, don't add persistent bitmap, but marks bitmap 
>>> persistent?
>>>
>>>
>>> Ok, I can think of it like "add persistent dirty bitmap to driver. if 
>>> succeed, set
>>> bitmap.persistent flag"
>>>
>>
>> Yeah, I meant "Add bitmap to file", where the persistence is implied
>> because it's part of the file. The bitmap is indeed not YET persistent,
>> but becomes so as a condition of this call succeeding.
>>
>> Do you have a suggestion for a better name? I liked the symmetry with
>> 'remove' so that there was an obvious "add bitmap" and "remove bitmap".
>>
>> Of course, when you remove, it is indeed persistent, so
>> "remove_persistent_dirty_bitmap" makes sense there.
>>
>>>
>>>> +{
>>>> +    BlockDriver *drv = bs->drv;
>>>> +    int ret;
>>>> +
>>>> +    if (bdrv_dirty_bitmap_get_persistence(bitmap)) {
>>>> +        error_setg(errp, "Bitmap '%s' is already persistent, "
>>>> +                   "and cannot be re-added to node '%s'",
>>>> +                   bdrv_dirty_bitmap_name(bitmap),
>>>> +                   bdrv_get_device_or_node_name(bs));
>>>> +        return -EINVAL;
>>>> +    }
>>>> +
>>>> +    if (!drv) {
>>>> +        error_setg_errno(errp, ENOMEDIUM,
>>>> +                         "Can't store persistent bitmaps to %s",
>>>> +                         bdrv_get_device_or_node_name(bs));
>>>> +        return -ENOMEDIUM;
>>>> +    }
>>>> +
>>>> +    if (!drv->bdrv_add_persistent_dirty_bitmap) {
>>>> +        error_setg_errno(errp, ENOTSUP,
>>>> +                         "Can't store persistent bitmaps to %s",
>>>> +                         bdrv_get_device_or_node_name(bs));
>>>> +        return -ENOTSUP;
>>>> +    }
>>>> +
>>>> +    ret = drv->bdrv_add_persistent_dirty_bitmap(bs, bitmap, errp);
>>>> +    if (ret == 0) {
>>>> +        bdrv_dirty_bitmap_set_persistence(bitmap, true);
>>>> +    }
>>>> +
>>>> +    return ret;
>>>> +}
>>>> +
>>>> +
>>>>    void bdrv_disable_dirty_bitmap(BdrvDirtyBitmap *bitmap)
>>>>    {
>>>>        bdrv_dirty_bitmap_lock(bitmap);
>>>> diff --git a/block/qcow2-bitmap.c b/block/qcow2-bitmap.c
>>>> index 83aee1a42a..c3a72ca782 100644
>>>> --- a/block/qcow2-bitmap.c
>>>> +++ b/block/qcow2-bitmap.c
>>>> @@ -1607,14 +1607,16 @@ int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, 
>>>> Error **errp)
>>>>        return 0;
>>>>    }
>>>>    
>>>> -bool qcow2_can_store_new_dirty_bitmap(BlockDriverState *bs,
>>>> -                                      const char *name,
>>>> -                                      uint32_t granularity,
>>>> +int qcow2_add_persistent_dirty_bitmap(BlockDriverState *bs,
>>>> +                                      BdrvDirtyBitmap *bitmap,
>>>>                                          Error **errp)
>>>>    {
>>>>        BDRVQcow2State *s = bs->opaque;
>>>>        bool found;
>>>>        Qcow2BitmapList *bm_list;
>>>> +    const char *name = bdrv_dirty_bitmap_name(bitmap);
>>>> +    uint32_t granularity = bdrv_dirty_bitmap_granularity(bitmap);
>>>> +    int ret = 0;
>>>
>>> dead assignment
>>>
>>
>> Will take care of.
>>
>>>>    
>>>>        if (s->qcow_version < 3) {
>>>>            /* Without autoclear_features, we would always have to assume
>>>> @@ -1623,20 +1625,23 @@ bool 
>>>> qcow2_can_store_new_dirty_bitmap(BlockDriverState *bs,
>>>>             * have to drop all dirty bitmaps (defeating their purpose).
>>>>             */
>>>>            error_setg(errp, "Cannot store dirty bitmaps in qcow2 v2 
>>>> files");
>>>> +        ret = -ENOTSUP;
>>>>            goto fail;
>>>>        }
>>>>    
>>>> -    if (check_constraints_on_bitmap(bs, name, granularity, errp) != 0) {
>>>> +    ret = check_constraints_on_bitmap(bs, name, granularity, errp);
>>>> +    if (ret) {
>>>
>>> hmm, in other places you prefere
>>> if ((ret = ...)) {
>>> syntax
>>>
>>
>> I have to get rid of those anyway, they violate our coding style. I'll
>> be replacing them all with this full style.
>>
>>>>            goto fail;
>>>>        }
>>>>    
>>>>        if (s->nb_bitmaps == 0) {
>>>> -        return true;
>>>> +        return 0;
>>>>        }
>>>>    
>>>>        if (s->nb_bitmaps >= QCOW2_MAX_BITMAPS) {
>>>>            error_setg(errp,
>>>>                       "Maximum number of persistent bitmaps is already 
>>>> reached");
>>>> +        ret = -EOVERFLOW;
>>>>            goto fail;
>>>>        }
>>>>    
>>>> @@ -1644,24 +1649,25 @@ bool 
>>>> qcow2_can_store_new_dirty_bitmap(BlockDriverState *bs,
>>>>            QCOW2_MAX_BITMAP_DIRECTORY_SIZE)
>>>>        {
>>>>            error_setg(errp, "Not enough space in the bitmap directory");
>>>> +        ret = -EOVERFLOW;
>>>>            goto fail;
>>>>        }
>>>>    
>>>> -    if (bitmap_list_load(bs, &bm_list, errp)) {
>>>> +    if ((ret = bitmap_list_load(bs, &bm_list, errp))) {
>>>
>>> interesting, now we load all bitmaps, so, may be, we don't need to load 
>>> list every time,
>>> but may use bs.dirty_bitmaps to check such things.. But it seems unsafe
>>>
>>
>> Yeah, I would like to cache this list eventually, but I ran into a lot
>> of hassle with it last time I tried and put it off for now.
>>
>> I think we should definitely be able to.
>>
>> And in fact, if we did, we could even add these bitmaps to the bm_list
>> as soon as _add_ is called and drop the need for the queued counters.
> 
> Personally I like the old _can_add_ :)
> 
> But you want to make this function really do something with driver, so 
> "_can_" is not good
> ofcourse and _add_ is better. And any kind of _mark_persistent_ or 
> _make_persistent_ will
> be in a bad relation with simple setter _set_persistence() which we already 
> have and which
> doesn't imply any complicated logic..
> 

I think it's a simpler design to have "add" and "remove" callbacks and
be more direct about it. Of course, the qcow2 implementation is free to
avoid a write to disk on add if it wishes, but I don't like the idea of
having to call "can_store" and then later a "do_store" or any such thing.

You could say that can_store is the check and then mark persistent is
the actual action, but then why keep them separate?

I like the link between calling the driver and the later store to be
obvious. I feel like marking the bitmap persistent without the knowledge
or consent of the driver is bound to cause trouble sooner or later, so
I'd rather make the persistent call a condition of the store check
succeeding.

> On the other hand I still not sure that we need track "queued" bitmaps in 
> qcow2 driver, as we
> can calculate their number and needed size of directory in any moment, not 
> extending the qcow2
> state..
> 

Do we want to add an O(N) check because we don't want to spend 12 bytes?

--js