09.11.2020 15:20, Max Reitz wrote:
On 06.11.20 13:42, Vladimir Sementsov-Ogievskiy wrote:
We should never set permissions other than cumulative permissions of
parents. During bdrv_reopen_multiple() we _check_ for synthetic
permissions but when we do _set_ the graph is already updated.
Add an assertion to bdrv_reopen_multiple(), other cases are more
obvious.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsement...@virtuozzo.com>
---
block.c | 29 +++++++++++++++--------------
1 file changed, 15 insertions(+), 14 deletions(-)
(Perhaps bdrv_commit_perm() might be a better name then, but I’m afraid such a
name change might be quite invasive (because AFAIR *_set_perm is used quite
often).)
Reviewed-by: Max Reitz <mre...@redhat.com>
Thanks for reviewing!
Actually, I plan to split and organize in a similar transactional way
graph-update operations:
- aio context change
- child replacement
- permission update
So, we'll have a chance to discuss final names later. I think about prepare/commit/abort too, as it is more
common than check/set/abort. Also, check now actually do set permissions in BdrvChild, so it isn't "just
check" (and the fact that we should do "abort" after "check" was always a bit odd).
--
Best regards,
Vladimir
