Re: [Qemu-devel] [PATCH 3/3] block/qcow2-bitmap: rewrite bitmap reopening logic

[John Snow]

On 5/23/19 11:47 AM, Vladimir Sementsov-Ogievskiy wrote:
> Current logic
> =============
> 
> Reopen rw -> ro:
> 
> Store bitmaps and release BdrvDirtyBitmap's.
> 
> Reopen ro -> rw:
> 
> Load bitmap list
> Skip bitmaps which for which we don't have BdrvDirtyBitmap [this is
>    the worst thing]

...ah.

> A kind of fail with error message if we see not readonly bitmap
> 
> This leads to at least the following bug:
> 
> Assume we have bitmap0.

Presumably a normal, persistent bitmap.

> Create external snapshot
>   bitmap0 is stored and therefore removed

Written out to the backing file and removed from memory, because we've
reopened rw->ro; this is because of the migration "safety" clause where
we simply drop these bitmaps.

...Ah, and then we don't actually open them readonly again; that entire
stanza in reopen_ro never fires off at all because the bitmaps are
already gone.

> Commit snapshot
>   now we have no bitmaps

When we reopen the base as rw as you note, we skipped bitmaps for which
we had no in-memory bitmap for -- because the readonly logic was really
expecting to have these in-memory.

I should probably say that for the sake of migration correctness, the
way we flush things to disk and remove it from memory on write is really
bothersome to keep correct. The migration logic is so particular that it
keeps causing issues elsewhere, of which this is another symptom.

> Do some writes from guest (*)
>   they are not marked in bitmap

Yikes, right.

> Shutdown
> Start
>   bitmap0 is loaded as valid, but it is actually broken! It misses
>   writes (*)

Yikes; because it was consistent on flush and we skipped it on load,
it's not marked as IN_USE and we are free to load it up again.

> Incremental backup
>   it will be inconsistent
> 

Good writeup, thank you.

> New logic
> =========
> 
> Reopen rw -> ro:
> 
> Store bitmaps and don't remove them from RAM, only mark them readonly
> (the latter is already done, but didn't work properly because of
> removing in storing function)
> 

Yes! I like this change.

> Reopen to rw (don't filter case with reopen rw -> rw, it is supported
> now in qcow2_reopen_bitmaps_rw)
> 

OK, so we allow rw --> rw without trying to be fussy about it. That
seems fine to me.

> Load bitmap list
> 
> Carefully consider all possible combinations of bitmaps in RAM and in
> the image, try to fix corruptions, print corresponding error messages.
> 
> Also, call qcow2_reopen_bitmaps_rw after the whole reopen queue
> commited, as otherwise we can't write to the qcow2 file child on reopen
> ro -> rw.
> 

I take it this is the change that moved the invocation logic into
bdrv_reopen_multiple instead of bdrv_reopen_commit; also notably we no
longer check the transition edge which is much simpler.

oh, I see why you're doing it there now...

> Also, add corresponding test-cases to 255.
> 
> Signed-off-by: Vladimir Sementsov-Ogievskiy <address@hidden>
> ---
>  block/qcow2.h              |   5 +-
>  include/block/block_int.h  |   2 +-
>  block.c                    |  29 ++----
>  block/qcow2-bitmap.c       | 197 ++++++++++++++++++++++++++-----------
>  block/qcow2.c              |   2 +-
>  tests/qemu-iotests/255     |   2 +
>  tests/qemu-iotests/255.out |  35 +++++++
>  7 files changed, 193 insertions(+), 79 deletions(-)
> 
> diff --git a/block/qcow2.h b/block/qcow2.h
> index 8d92ef1fee..5928306e62 100644
> --- a/block/qcow2.h
> +++ b/block/qcow2.h
> @@ -725,9 +725,10 @@ Qcow2BitmapInfoList 
> *qcow2_get_bitmap_info_list(BlockDriverState *bs,
>                                                  Error **errp);
>  int qcow2_reopen_bitmaps_rw_hint(BlockDriverState *bs, bool *header_updated,
>                                   Error **errp);
> -int qcow2_reopen_bitmaps_rw(BlockDriverState *bs, Error **errp);
> +void qcow2_reopen_bitmaps_rw(BlockDriverState *bs);
>  int qcow2_truncate_bitmaps_check(BlockDriverState *bs, Error **errp);
> -void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs, Error 
> **errp);
> +void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs,
> +                                          bool release_stored, Error **errp);
>  int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, Error **errp);
>  bool qcow2_can_store_new_dirty_bitmap(BlockDriverState *bs,
>                                        const char *name,
> diff --git a/include/block/block_int.h b/include/block/block_int.h
> index 1eebc7c8f3..9a694f3da0 100644
> --- a/include/block/block_int.h
> +++ b/include/block/block_int.h
> @@ -536,7 +536,7 @@ struct BlockDriver {
>       * as rw. This handler should realize it. It also should unset readonly
>       * field of BlockDirtyBitmap's in case of success.
>       */
> -    int (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs, Error **errp);
> +    void (*bdrv_reopen_bitmaps_rw)(BlockDriverState *bs);
>      bool (*bdrv_can_store_new_dirty_bitmap)(BlockDriverState *bs,
>                                              const char *name,
>                                              uint32_t granularity,
> diff --git a/block.c b/block.c
> index cb11537029..db1ec0c673 100644
> --- a/block.c
> +++ b/block.c
> @@ -3334,6 +3334,16 @@ int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, 
> Error **errp)
>          bdrv_reopen_commit(&bs_entry->state);
>      }
>  
> +    QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) {
> +        BlockDriverState *bs = bs_entry->state.bs;
> +
> +        if (!bdrv_is_writable(bs) || !bs->drv->bdrv_reopen_bitmaps_rw) {
> +            continue;
> +        }
> +
> +        bs->drv->bdrv_reopen_bitmaps_rw(bs);
> +    }
> +

OK, so this has been moved up into the main body of the loop because we
cannot trust it to run in bdrv_reopen_commit because of the order in
which the nodes are reopened might leave us unable to write to our child
nodes to issue the IN_USE flag.

I have kept out of these discussions in the past; is there a general
solution that allows us to sort the block DAG leaf-up to avoid this
scenario?

Anyway, since the block graph organization isn't my problem I will say
that I think this is fine by me; but I'm not the one to impress here.

>      ret = 0;
>  cleanup_perm:
>      QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) {
> @@ -3770,16 +3780,12 @@ void bdrv_reopen_commit(BDRVReopenState *reopen_state)
>      BlockDriver *drv;
>      BlockDriverState *bs;
>      BdrvChild *child;
> -    bool old_can_write, new_can_write;
>  
>      assert(reopen_state != NULL);
>      bs = reopen_state->bs;
>      drv = bs->drv;
>      assert(drv != NULL);
>  
> -    old_can_write =
> -        !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
> -
>      /* If there are any driver level actions to take */
>      if (drv->bdrv_reopen_commit) {
>          drv->bdrv_reopen_commit(reopen_state);
> @@ -3823,21 +3829,6 @@ void bdrv_reopen_commit(BDRVReopenState *reopen_state)
>      }
>  
>      bdrv_refresh_limits(bs, NULL);
> -
> -    new_can_write =
> -        !bdrv_is_read_only(bs) && !(bdrv_get_flags(bs) & BDRV_O_INACTIVE);
> -    if (!old_can_write && new_can_write && drv->bdrv_reopen_bitmaps_rw) {
> -        Error *local_err = NULL;
> -        if (drv->bdrv_reopen_bitmaps_rw(bs, &local_err) < 0) {
> -            /* This is not fatal, bitmaps just left read-only, so all 
> following
> -             * writes will fail. User can remove read-only bitmaps to unblock
> -             * writes.
> -             */
> -            error_reportf_err(local_err,
> -                              "%s: Failed to make dirty bitmaps writable: ",
> -                              bdrv_get_node_name(bs));
> -        }
> -    }
>  }
>  

Certainly the new code is simpler here, which is good.

>  /*
> diff --git a/block/qcow2-bitmap.c b/block/qcow2-bitmap.c
> index 2b84bfa007..4e565db11f 100644
> --- a/block/qcow2-bitmap.c
> +++ b/block/qcow2-bitmap.c
> @@ -28,6 +28,7 @@
>  #include "qemu/osdep.h"
>  #include "qapi/error.h"
>  #include "qemu/cutils.h"
> +#include "qemu/error-report.h"
>  
>  #include "block/block_int.h"
>  #include "qcow2.h"
> @@ -951,6 +952,12 @@ static void set_readonly_helper(gpointer bitmap, 
> gpointer value)
>      bdrv_dirty_bitmap_set_readonly(bitmap, (bool)value);
>  }
>  
> +/* for g_slist_foreach for GSList of const char* elements */
> +static void error_report_helper(gpointer message, gpointer _unused)
> +{
> +    error_report("%s", (const char *)message);
> +}
> +
>  /* qcow2_load_dirty_bitmaps()
>   * Return value is a hint for caller: true means that the Qcow2 header was
>   * updated. (false doesn't mean that the header should be updated by the
> @@ -1103,76 +1110,150 @@ Qcow2BitmapInfoList 
> *qcow2_get_bitmap_info_list(BlockDriverState *bs,
>      return list;
>  }
>  
> -int qcow2_reopen_bitmaps_rw_hint(BlockDriverState *bs, bool *header_updated,
> -                                 Error **errp)
> +/*
> + * qcow2_reopen_bitmaps_rw
> + *
> + * The function is called in bdrv_reopen_multiple after all calls to
> + * bdrv_reopen_commit, when final bs is writable. It is done so, as we need
> + * write access to child bs, and with current reopening architecture, when
> + * reopen ro -> rw it is possible only after all reopen commits.
> + *
> + * So, we can't fail here. On the other hand, we may face different kinds of
> + * corruptions and/or just can't write IN_USE flags to the image due to EIO.
> + *
> + * Try to handle as many cases as we can.

Hm, I think you're right, but it does make me really uncomfortable that
we lose the ability to report errors back up the stack. I guess we
already always did ignore them, so this is no worse, but I don't like
the idea of adding new error_report_err calls if we can help it.

I guess we can't help it, though.

> + */
> +void qcow2_reopen_bitmaps_rw(BlockDriverState *bs)
>  {
>      BDRVQcow2State *s = bs->opaque;
>      Qcow2BitmapList *bm_list;
>      Qcow2Bitmap *bm;
> -    GSList *ro_dirty_bitmaps = NULL;
> +    GSList *ro_dirty_bitmaps = NULL, *corrupted_bitmaps = NULL;
> +    Error *local_err = NULL;
>      int ret = 0;
> -
> -    if (header_updated != NULL) {
> -        *header_updated = false;
> -    }
> +    bool need_update = false, updated_ok = false;
>  
>      if (s->nb_bitmaps == 0) {
>          /* No bitmaps - nothing to do */
> -        return 0;
> -    }
> -
> -    if (!can_write(bs)) {
> -        error_setg(errp, "Can't write to the image on reopening bitmaps rw");
> -        return -EINVAL;
> +        return;
>      }
>  
>      bm_list = bitmap_list_load(bs, s->bitmap_directory_offset,
> -                               s->bitmap_directory_size, errp);
> +                               s->bitmap_directory_size, &local_err);
>      if (bm_list == NULL) {
> -        return -EINVAL;
> +        error_reportf_err(local_err, "Failed to reopen bitmaps rw: "
> +                          "cannot load bitmap list: ");
> +        return;
>      }
>  
>      QSIMPLEQ_FOREACH(bm, bm_list, entry) {
>          BdrvDirtyBitmap *bitmap = bdrv_find_dirty_bitmap(bs, bm->name);
> -        if (bitmap == NULL) {
> -            continue;
> -        }
> +        const char *corruption = NULL;
>  
> -        if (!bdrv_dirty_bitmap_readonly(bitmap)) {
> -            error_setg(errp, "Bitmap %s was loaded prior to rw-reopen, but 
> was "
> -                       "not marked as readonly. This is a bug, something 
> went "
> -                       "wrong. All of the bitmaps may be corrupted", 
> bm->name);
> -            ret = -EINVAL;
> -            goto out;
> -        }
> +        if (bm->flags & BME_FLAG_IN_USE) {
> +            if (bitmap == NULL) {
> +                /*
> +                 * It's an unexpected inconsistent bitmap,
> +                 * it is safe to ignore it.
> +                 */
> +                continue;
> +            }

This is supposed to be a reopen but we've found a bitmap we didn't load,
and it's IN_USE. Should we make any attempt to load it as an
inconsistent bitmap so the user can know about it?

If we're here, it means we DID reopen the image rw, so we ought to have
exclusive ownership of this file; the IN_USE flag here signals an
inconsistency, no?

>  
> -        bm->flags |= BME_FLAG_IN_USE;
> -        ro_dirty_bitmaps = g_slist_append(ro_dirty_bitmaps, bitmap);
> +            /*
> +             * It's either an inconsistent bitmap, or we are reopening rw -> 
> rw,
> +             * or we just didn't save bitmap for some buggy reason. Still, no
> +             * reason to consider in-RAM bitmap inconsistent, than, mark it 
> rw.
> +             */

I don't understand lines 2-3 of this comment. As far as I can tell:

- We might be seeing a legitimately corrupt bitmap. It's fine to mark it
as rw, because we can't write to it anyway. (It was marked inconsistent
on open.)
- We might be seeing a bitmap that's already properly rw. this call is
effectively a no-op.

Is that right? If that's true, I'd just simply say:

"This is either an inconsistent bitmap or we are reopening rw -> rw. It
is safe to mark it as not read only in either case."

What's the "or we just didn't save bitmap for some buggy reason" you are
alluding to?

> +            bdrv_dirty_bitmap_set_readonly(bitmap, false);
> +        } else {
> +            /*
> +             * In-image bitmap not marked IN_USE
> +             *
> +             * The only valid case is
> +             *     bitmap && bdrv_dirty_bitmap_readonly(bitmap) &&
> +             *        !bdrv_dirty_bitmap_inconsistent(bitmap))
> +             *
> +             * which means reopen ro -> rw of consistent bitmap.
> +             *
> +             * Other cases a different kinds of corruptions:
> +             */
> +            if (!bitmap) {
> +                corruption =
> +                    "Corruption: unexpected valid bitmap '%s' is found in 
> the "
> +                    "image '%s' on reopen rw. Will try to set IN_USE flag.";
> +

In this case, we find a valid bitmap we expected to have a readonly copy
of in memory, but didn't. We attempt to load it.

> +                bitmap = load_bitmap(bs, bm, NULL);
> +                if (!bitmap) {
> +                    bitmap = bdrv_create_dirty_bitmap(
> +                            bs, bdrv_get_default_bitmap_granularity(bs),
> +                            bm->name, NULL);
> +                }
> +
> +                if (bitmap) {
> +                    bdrv_dirty_bitmap_set_persistence(bitmap, true);
> +                    bdrv_dirty_bitmap_set_readonly(bitmap, true);
> +                    bdrv_dirty_bitmap_set_inconsistent(bitmap);

And we mark it as inconsistent because we're not sure how we missed it
earlier. OK.

> +                }
> +            } else if (!bdrv_dirty_bitmap_readonly(bitmap)) {
> +                corruption =
> +                    "Corruption: bitmap '%s' is not marked IN_USE in the "
> +                    "image '%s' and not marked readonly in RAM. Will try to "
> +                    "set IN_USE flag.";
> +

And in this case, we find the bitmap but it's not marked readonly for
some reason.

> +                bdrv_dirty_bitmap_set_readonly(bitmap, true);

Why set it readonly again?

> +                bdrv_dirty_bitmap_set_inconsistent(bitmap);

And just in case, we mark it inconsistent. (It's my impression that any
such write would have failed earlier, but maybe not.)

> +            } else if (bdrv_dirty_bitmap_inconsistent(bitmap)) {
> +                corruption =
> +                    "Corruption: bitmap '%s' is inconsistent but is not 
> marked "
> +                    "IN_USE in the image. Will try to set IN_USE flag.";
> +
> +                bdrv_dirty_bitmap_set_readonly(bitmap, true);

This one is weirder. We have an inconsistent bitmap but it's not IN_USE;
so we set it readonly again? Why?

> +            }
> +
> +            if (bitmap) {
> +                ro_dirty_bitmaps = g_slist_append(ro_dirty_bitmaps, bitmap);

Oh, all those bitmaps we just set readonly we're going to mark as not
readonly again?

> +            }
> +            bm->flags |= BME_FLAG_IN_USE;
> +            need_update = true;
> +            if (corruption) {
> +                error_report(corruption, bm->name, bs->filename);
> +                corrupted_bitmaps = g_slist_append(corrupted_bitmaps, 
> bm->name);
> +            }
> +        }
>      }
>  
> -    if (ro_dirty_bitmaps != NULL) {
> +    if (need_update) {
> +        if (!can_write(bs->file->bs)) {

I genuinely don't know: is it legitimate to check your child's write
permission in this way? will we always have bs->file->bs?

> +            error_report("Failed to reopen bitmaps rw: cannot write to "
> +                         "the protocol file");
> +            goto handle_corrupted;
> +        }
> +
>          /* in_use flags must be updated */
>          ret = update_ext_header_and_dir_in_place(bs, bm_list);
>          if (ret < 0) {
> -            error_setg_errno(errp, -ret, "Can't update bitmap directory");
> -            goto out;
> -        }
> -        if (header_updated != NULL) {
> -            *header_updated = true;
> +            error_report("Cannot update bitmap directory: %s", 
> strerror(-ret));
> +            goto handle_corrupted;
>          }
> +        updated_ok = true;
>          g_slist_foreach(ro_dirty_bitmaps, set_readonly_helper, false);
>      }

And then here at the end the bulk of the work: we re-write the header if
necessary back out to disk to mark everything as IN_USE.

It seems strange that you are trying to maintain the invariant that we
won't set readonly to false if we fail, because this function "cannot fail."

>  
> -out:
> +handle_corrupted:
> +    if (corrupted_bitmaps) {
> +        if (updated_ok) {
> +            error_report("Corrupted bitmaps in '%s' successfully marked "
> +                         "IN_USE", bs->filename);
> +        } else {
> +            error_report("Failed to mark IN_USE the following corrupted "
> +                         "bitmaps in '%s', DO NOT USE THEM:", bs->filename);
> +            g_slist_foreach(corrupted_bitmaps, error_report_helper, NULL);
> +        }
> +    }
> +
>      g_slist_free(ro_dirty_bitmaps);
> +    g_slist_free(corrupted_bitmaps);
>      bitmap_list_free(bm_list);
> -
> -    return ret;
> -}
> -

I just don't know; for how much error checking this function is doing
now, it seems wrong to be behind an interface that cannot fail and will
actually do different things to the bitmaps depending on what it sees
with no return code to help the caller understand the cases.

It even has a bit at the end where it tries to print in uppercase to
manually scare a user into taking action, which says to me that there is
a deeper problem we need to be able to address without intervention from
the user.

That said, the patch does seem correct otherwise; and it does fix a
nasty bug which lets us use bitmaps with snapshots. I want this in for
4.1 if I can help it. I will talk to Kevin and Max and see if there's
some opinion here.

Thanks!

> -int qcow2_reopen_bitmaps_rw(BlockDriverState *bs, Error **errp)
> -{
> -    return qcow2_reopen_bitmaps_rw_hint(bs, NULL, errp);
>  }
>  
>  /* Checks to see if it's safe to resize bitmaps */
> @@ -1446,7 +1527,8 @@ fail:
>      bitmap_list_free(bm_list);
>  }
>  
> -void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs, Error **errp)
> +void qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs,
> +                                          bool release_stored, Error **errp)
>  {
>      BdrvDirtyBitmap *bitmap;
>      BDRVQcow2State *s = bs->opaque;
> @@ -1559,20 +1641,23 @@ void 
> qcow2_store_persistent_dirty_bitmaps(BlockDriverState *bs, Error **errp)
>          g_free(tb);
>      }
>  
> -    QSIMPLEQ_FOREACH(bm, bm_list, entry) {
> -        /* For safety, we remove bitmap after storing.
> -         * We may be here in two cases:
> -         * 1. bdrv_close. It's ok to drop bitmap.
> -         * 2. inactivation. It means migration without 'dirty-bitmaps'
> -         *    capability, so bitmaps are not marked with
> -         *    BdrvDirtyBitmap.migration flags. It's not bad to drop them too,
> -         *    and reload on invalidation.
> -         */
> -        if (bm->dirty_bitmap == NULL) {
> -            continue;
> -        }
> +    if (release_stored) {
> +        QSIMPLEQ_FOREACH(bm, bm_list, entry) {
> +            /*
> +             * For safety, we remove bitmap after storing.
> +             * We may be here in two cases:
> +             * 1. bdrv_close. It's ok to drop bitmap.
> +             * 2. inactivation. It means migration without 'dirty-bitmaps'
> +             *    capability, so bitmaps are not marked with
> +             *    BdrvDirtyBitmap.migration flags. It's not bad to drop them
> +             *    too, and reload on invalidation.
> +             */

While we're here, I might touch up the comment.

For safety, the BdrvDirtyBitmap can be dropped after storing.
We may be here in two cases, both via qcow2_inactivate:
1. bdrv_close: It's correct to remove bitmaps on close.
2. migration: This implies we are migrating without the
   'dirty-bitmaps' capability, because bitmap->migration was unset.
   If needed, these bitmaps will be reloaded on invalidation.

I just wanted to clarify these points:

- The new boolean obviously means we don't /always/ drop them
- qcow2_inactivate is the only caller that instructs us to drop them
- both cases in the comment are through qcow2_inactivate.

you can take any or none of my wording suggestions as you feel is
appropriate.

> +            if (bm->dirty_bitmap == NULL) {
> +                continue;
> +            }
>  
> -        bdrv_release_dirty_bitmap(bs, bm->dirty_bitmap);
> +            bdrv_release_dirty_bitmap(bs, bm->dirty_bitmap);
> +        }
>      }
>  
>  success:
> @@ -1600,7 +1685,7 @@ int qcow2_reopen_bitmaps_ro(BlockDriverState *bs, Error 
> **errp)
>      BdrvDirtyBitmap *bitmap;
>      Error *local_err = NULL;
>  
> -    qcow2_store_persistent_dirty_bitmaps(bs, &local_err);
> +    qcow2_store_persistent_dirty_bitmaps(bs, false, &local_err);
>      if (local_err != NULL) {
>          error_propagate(errp, local_err);
>          return -EINVAL;
> diff --git a/block/qcow2.c b/block/qcow2.c
> index d39882785d..f0a8479874 100644
> --- a/block/qcow2.c
> +++ b/block/qcow2.c
> @@ -2273,7 +2273,7 @@ static int qcow2_inactivate(BlockDriverState *bs)
>      int ret, result = 0;
>      Error *local_err = NULL;
>  
> -    qcow2_store_persistent_dirty_bitmaps(bs, &local_err);
> +    qcow2_store_persistent_dirty_bitmaps(bs, true, &local_err);
>      if (local_err != NULL) {
>          result = -EINVAL;
>          error_reportf_err(local_err, "Lost persistent bitmaps during "
> diff --git a/tests/qemu-iotests/255 b/tests/qemu-iotests/255
> index 36712689d3..e8b0c9d4bb 100755
> --- a/tests/qemu-iotests/255
> +++ b/tests/qemu-iotests/255
> @@ -79,3 +79,5 @@ def test(persistent, restart):
>  
>  
>  test(persistent=False, restart=False)
> +test(persistent=True, restart=False)
> +test(persistent=True, restart=True)
> diff --git a/tests/qemu-iotests/255.out b/tests/qemu-iotests/255.out
> index 2bffb486d2..46e2e3ad4e 100644
> --- a/tests/qemu-iotests/255.out
> +++ b/tests/qemu-iotests/255.out
> @@ -15,3 +15,38 @@ check, that no bitmaps in snapshot: not found
>  {"data": {"device": "drive0", "len": 65536, "offset": 65536, "speed": 0, 
> "type": "commit"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": 
> {"microseconds": "USECS", "seconds": "SECS"}}
>  check merged bitmap: name=bitmap0 dirty-clusters=2
>  check updated bitmap: name=bitmap0 dirty-clusters=3
> +
> +Testcase persistent without restart
> +
> +{"execute": "block-dirty-bitmap-add", "arguments": {"name": "bitmap0", 
> "node": "drive0", "persistent": true}}
> +{"return": {}}
> +initial bitmap: name=bitmap0 dirty-clusters=1
> +{"execute": "blockdev-snapshot-sync", "arguments": {"device": "drive0", 
> "format": "qcow2", "snapshot-file": "TEST_DIR/PID-top"}}
> +{"return": {}}
> +check, that no bitmaps in snapshot: not found
> +{"execute": "block-commit", "arguments": {"device": "drive0", "top": 
> "TEST_DIR/PID-top"}}
> +{"return": {}}
> +{"data": {"device": "drive0", "len": 65536, "offset": 65536, "speed": 0, 
> "type": "commit"}, "event": "BLOCK_JOB_READY", "timestamp": {"microseconds": 
> "USECS", "seconds": "SECS"}}
> +{"execute": "block-job-complete", "arguments": {"device": "drive0"}}
> +{"return": {}}
> +{"data": {"device": "drive0", "len": 65536, "offset": 65536, "speed": 0, 
> "type": "commit"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": 
> {"microseconds": "USECS", "seconds": "SECS"}}
> +check merged bitmap: name=bitmap0 dirty-clusters=2
> +check updated bitmap: name=bitmap0 dirty-clusters=3
> +
> +Testcase persistent with restart
> +
> +{"execute": "block-dirty-bitmap-add", "arguments": {"name": "bitmap0", 
> "node": "drive0", "persistent": true}}
> +{"return": {}}
> +initial bitmap: name=bitmap0 dirty-clusters=1
> +{"execute": "blockdev-snapshot-sync", "arguments": {"device": "drive0", 
> "format": "qcow2", "snapshot-file": "TEST_DIR/PID-top"}}
> +{"return": {}}
> +check, that no bitmaps in snapshot: not found
> +... Restart ...
> +{"execute": "block-commit", "arguments": {"device": "drive0", "top": 
> "TEST_DIR/PID-top"}}
> +{"return": {}}
> +{"data": {"device": "drive0", "len": 65536, "offset": 65536, "speed": 0, 
> "type": "commit"}, "event": "BLOCK_JOB_READY", "timestamp": {"microseconds": 
> "USECS", "seconds": "SECS"}}
> +{"execute": "block-job-complete", "arguments": {"device": "drive0"}}
> +{"return": {}}
> +{"data": {"device": "drive0", "len": 65536, "offset": 65536, "speed": 0, 
> "type": "commit"}, "event": "BLOCK_JOB_COMPLETED", "timestamp": 
> {"microseconds": "USECS", "seconds": "SECS"}}
> +check merged bitmap: name=bitmap0 dirty-clusters=2
> +check updated bitmap: name=bitmap0 dirty-clusters=3
> 

-- 
—js