Re: [Qemu-devel] bug in reopen arch

[Kevin Wolf]

Am 21.06.2018 um 17:55 hat Vladimir Sementsov-Ogievskiy geschrieben:
> 21.06.2018 17:25, Kevin Wolf wrote:
> > Am 15.06.2018 um 20:42 hat Vladimir Sementsov-Ogievskiy geschrieben:
> > > 14.06.2018 13:46, Kevin Wolf wrote:
> > > > Am 12.06.2018 um 20:57 hat Vladimir Sementsov-Ogievskiy geschrieben:
> > > > > Hi all!
> > > > > 
> > > > > I've faced the following problem:
> > > > > 
> > > > >       1. create image with dirty bitmap, a.qcow2 (start qemu and run 
> > > > > qmp
> > > > >       command block-dirty-bitmap-add)
> > > > > 
> > > > >       2. run the following commands:
> > > > > 
> > > > >           qemu-img create -f qcow2 -b a.qcow2 b.qcow2 10M
> > > > >           qemu-io -c 'write 0 512' b.qcow2
> > > > >           qemu-img commit b.qcow2
> > > > > 
> > > > >       3. last command fails with the following output:
> > > > > 
> > > > > Formatting 'b.qcow2', fmt=qcow2 size=68719476736 backing_file=a.qcow2
> > > > > cluster_size=65536 lazy_refcounts=off refcount_bits=16
> > > > > wrote 512/512 bytes at offset 0
> > > > > 512 bytes, 1 ops; 0.0953 sec (5.243 KiB/sec and 10.4867 ops/sec)
> > > > > qemu-img: #block397: Failed to make dirty bitmaps writable: Can't 
> > > > > update
> > > > > bitmap directory: Operation not permitted
> > > > > qemu-img: Block job failed: Operation not permitted
> > > > > 
> > > > > And problem is that children are reopened _after_ parent. But qcow2 
> > > > > reopen
> > > > > needs write access to its file, to write IN_USE flag to dirty-bitmaps
> > > > > extension.
> > > > I was aware of a different instance of this problem: Assume a qcow2
> > > > image with an unknown autoclear flag (so it will be cleared on r/w
> > > > open), which is first opened r/o and then reopened r/w. This will fail
> > > > because .bdrv_reopen_prepare doesn't have the permissions yet.
> > > Hm.. If I understand correctly qcow2_reopen_prepare doesn't deal with
> > > autoclear flags, as it doesn't call qcow2_do_open.
> > Hm, right, not sure what I really meant back then when I added it to my
> > to-do list... Maybe I confused reopen and invalidate_cache.
> > 
> > > > Simply changing the order won't fix this because in the r/w -> r/o, the
> > > > driver will legitimately flush its caches in .bdrv_reopen_prepare, and
> > > > for this it still needs to be able to write.
> > > > 
> > > > We may need to have a way for nodes to access both the old and the new
> > > > state of their children. I'm not completely sure how to achieve this
> > > > best, though.
> > > > 
> > > > When I thought only of permissions, the obvious and simple thing to do
> > > > was to just get combined permissions for the old and new state, i.e.
> > > > 'old_perm | new_perm' and 'old_shared & new_shared'. But I don't think
> > > > this is actually enough when the child node switches between a r/w and
> > > > a r/o file descriptor because even though QEMU's permission system would
> > > > allow the write, you still can't successfully write to a r/o file
> > > > descriptor.
> > > > 
> > > > Kevin
> > > Maybe we want two .bdrv_reopen_prepare: 
> > > .bdrv_reopen_prepare_before_children
> > > and .bdrv_reopen_prepare_after_children. But to write something in
> > > reopen_prepare, we need to move bdrv_set_perm from reopen_commit to
> > > .. Is it possible?
> > Getting the permission problems out of the way can be solved by changing
> > permissions twice, like I said above: First to the combined permissions
> > of old and new, and finally to only the new permissions.
> > 
> > The problem I see with .bdrv_reopen_prepare_after_children is that I
> > don't see how it actually buys you anything: Even if the children
> > already prepared the reopen, any access of the child node still refers
> > to the old file descriptor because the new one only becomes valid with
> > .bdrv_reopen_commit.
> > 
> > > Now, I've found the following workaround, what do you think about 
> > > something
> > > like this as a temporary fix:
> > I honestly don't understand why this workaround makes any difference.
> 
> with this patch, commit for children will be called earlier than for parent,
> so, when reopening bitmaps rw (which is done in commit) bs->file will be
> already completely reopened rw, and all works.

.bdrv_reopen_commit() can't do any I/O because it must not fail.
Therefore the order in which nodes are committed should not matter.

Any I/O that needs to be done has to be in .bdrv_reopen_prepare() (and
possibly be kept in a temporary buffer) and .bdrv_reopen_commit() can
only apply what is already in memory.

I don't see the code for reopening bitmaps in master. Is this a pending
patch?

Kevin