jewel: librbd: restore journal access when force disabling mirroring

[trociny]

If mirroring is force disabled on a demoted image, the journal was
being left in an inconsistent ownership state.

Fixes: http://tracker.ceph.com/issues/17767
Signed-off-by: Mykola Golub mgolub@mirantis.com

[trociny]

@dillaman I tested this using this test:

trociny@38b2d40

Note, the test uses workarounds for the following issues found:

1. Removing a demoted image will fiail with the following error until you disable mirroring first (both master and jewel):

2016-11-11 16:18:20.454915 7fba651b8700 10 librbd::exclusive_lock::AcquireRequest: handle_open_journal: r=0
2016-11-11 16:18:20.454916 7fba651b8700 10 librbd::exclusive_lock::AcquireRequest: send_allocate_journal_tag
2016-11-11 16:18:20.454918 7fba651b8700 -1 librbd::journal::StandardPolicy: local image not promoted
2016-11-11 16:18:20.454921 7fba651b8700 10 librbd::exclusive_lock::AcquireRequest: handle_allocate_journal_tag: r=-1
2016-11-11 16:18:20.454923 7fba651b8700 -1 librbd::exclusive_lock::AcquireRequest: failed to allocate journal tag: (1) Operation not permitted

2. If one do mirror_image_demote and mirror_image_disable without reopening the image between these operation this assertion in DisableRequest::send_promote_image is triggered (only master):

  // Not primary -- shouldn't have the journal open
  assert(m_image_ctx->journal == nullptr);

I am planning to report these issues and work on the fix.

[dillaman]

@trociny For (1), that is the correct behavior I would think -- we don't want to allow users to delete non-primary images easily. For (2), that does seem like an issue that the demote step didn't release the exclusive lock (which would close the journal).

[ghost]

@trociny setting DNM while it is being discussed. It would be great if you could shortly explain in the commit message why this back has to be a unique snowflake (as Jason puts it ;-).

[trociny]

@dachary The commit message has been updated.

As for DNM status, actually the discussion we had with Jason was about some other (not related) issues I had observed testing this, and it turned out that one of them was an expected behavior and another one was triggered only in master.

Still, I'd like to see Jason's approve before the merge.

And it needs teuthology run. @dachary Are you going to run teuthology for backport PRs soon? If you are, could you please test this too? If you are not, I can run the tests myself.

[ghost]

@trociny a run finished yesterday and since this is the only PR that was not included in it, it's likely the next run will be in one or two weeks. Feel free to run the relevant jobs earlier if you'd like.

[ghost]

jenkins test this please (general jenkins failure)

[dillaman]

lgtm