The subtree drain was introduced in commit b1e1af394d as a way to avoid
graph changes between finding the base node and changing the block graph
as necessary on completion of the image streaming job.
The block graph could change between these two points because
bdrv_set_backing_hd() first drains the parent node, which involved
polling and can do anything.
Subtree draining was an imperfect way to make this less likely (because
with it, fewer callbacks are called during this window). Everyone agreed
that it's not really the right solution, and it was only committed as a
stopgap solution.
This replaces the subtree drain with a solution that simply drains the
parent node before we try to find the base node, and then call a version
of bdrv_set_backing_hd() that doesn't drain, but just asserts that the
parent node is already drained.
This way, any graph changes caused by draining happen before we start
looking at the graph and things stay consistent between finding the base
node and changing the graph.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221118174110.55183-10-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We want to use bdrv_child_get_parent_aio_context() from
bdrv_parent_drained_{begin,end}_single(), both of which are "I/O or GS"
functions.
Prior to 3ed4f708fe, all the implementations were I/O code anyway.
3ed4f708fe has put block jobs' AioContext field under the job mutex, so
to make child_job_get_parent_aio_context() work in an I/O context, we
need to take that lock there.
Furthermore, blk_root_get_parent_aio_context() is not marked as
anything, but is safe to run in an I/O context, so mark it that way now.
(blk_get_aio_context() is an I/O code function.)
With that done, all implementations explicitly are I/O code, so we can
mark bdrv_child_get_parent_aio_context() as I/O code, too, so callers
know it is safe to run from both GS and I/O contexts.
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20221107151321.211175-2-hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
No functional change intended.
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221025084952.2139888-11-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
No functional changes intended.
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221025084952.2139888-10-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Together with all _can_set_ and _set_ APIs, as they are not needed
anymore.
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221025084952.2139888-9-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Minor performance improvement, but given that we have hash tables
available, avoid iterating in the visited nodes list every time just
to check if a node has been already visited.
The data structure is not actually a proper hash map, but an hash set,
as we are just adding nodes and not key,value pairs.
Suggested-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20221025084952.2139888-4-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Simplify the way the aiocontext can be changed in a BDS graph.
There are currently two problems in bdrv_try_set_aio_context:
- There is a confusion of AioContext locks taken and released, because
we assume that old aiocontext is always taken and new one is
taken inside.
- It doesn't look very safe to call bdrv_drained_begin while some
nodes have already switched to the new aiocontext and others haven't.
This could be especially dangerous because bdrv_drained_begin polls, so
something else could be executed while graph is in an inconsistent
state.
Additional minor nitpick: can_set and set_ callbacks both traverse the
graph, both using the ignored list of visited nodes in a different way.
Therefore, get rid of all of this and introduce a new callback,
change_aio_context, that uses transactions to efficiently, cleanly
and most importantly safely change the aiocontext of a graph.
This new callback is a "merge" of the two previous ones:
- Just like can_set_aio_context, recursively traverses the graph.
Marks all nodes that are visited using a GList, and checks if
they *could* change the aio_context.
- For each node that passes the above check, drain it and add a new transaction
that implements a callback that effectively changes the aiocontext.
- Once done, the recursive function returns if *all* nodes can change
the AioContext. If so, commit the above transactions.
Regardless of the outcome, call transaction.clean() to undo all drains
done in the recursion.
- The transaction list is scanned only after all nodes are being drained, so
we are sure that they all are in the same context, and then
we switch their AioContext, concluding the drain only after all nodes
switched to the new AioContext. In this way we make sure that
bdrv_drained_begin() is always called under the old AioContext, and
bdrv_drained_end() under the new one.
- Because of the above, we don't need to release and re-acquire the
old AioContext every time, as everything is done once (and not
per-node drain and aiocontext change).
Note that the "change" API is not yet invoked anywhere.
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Message-Id: <20221025084952.2139888-3-eesposit@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Almost all drivers call bdrv_open_child() similarly. Let's create a
helper for this.
The only not updated drivers that call bdrv_open_child() to set
bs->file are raw-format and snapshot-access:
raw-format sometimes want to have filtered child but
don't set drv->is_filter to true.
snapshot-access wants only DATA | PRIMARY
Possibly we should implement drv->is_filter_func() handler, to consider
raw-format as filter when it works as filter.. But it's another story.
Note also, that we decrease assignments to bs->file in code: it helps
us restrict modifying this field in further commit.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20220726201134.924743-3-vsementsov@yandex-team.ru>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Registering an I/O buffer is only a performance optimization hint but it
is still necessary to return errors when it fails.
Later patches will need to detect errors when registering buffers but an
immediate advantage is that error_report() calls are no longer needed in
block driver .bdrv_register_buf() functions.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20221013185908.1297568-8-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The only implementor of bdrv_register_buf() is block/nvme.c, where the
size is not needed when unregistering a buffer. This is because
util/vfio-helpers.c can look up mappings by address.
Future block drivers that implement bdrv_register_buf() may not be able
to do their job given only the buffer address. Add a size argument to
bdrv_unregister_buf().
Also document the assumptions about
bdrv_register_buf()/bdrv_unregister_buf() calls. The same <host, size>
values that were given to bdrv_register_buf() must be given to
bdrv_unregister_buf().
gcc 11.2.1 emits a spurious warning that img_bench()'s buf_size local
variable might be uninitialized, so it's necessary to silence the
compiler.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Message-id: 20221013185908.1297568-5-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This function is safe to call in an I/O context, and qcow2_do_open()
does so (invoked in an I/O context by qcow2_co_invalidate_cache()).
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20220427114057.36651-2-hreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
block.h currently contains a mix of functions:
some of them run under the BQL and modify the block layer graph,
others are instead thread-safe and perform I/O in iothreads.
Some others can only be called by either the main loop or the
iothread running the AioContext (and not other iothreads),
and using them in another thread would cause deadlocks, and therefore
it is not ideal to define them as I/O.
It is not easy to understand which function is part of which
group (I/O vs GS vs "I/O or GS"), and this patch aims to clarify it.
The "GS" functions need the BQL, and often use
aio_context_acquire/release and/or drain to be sure they
can modify the graph safely.
The I/O function are instead thread safe, and can run in
any AioContext.
"I/O or GS" functions run instead in the main loop or in
a single iothread, and use BDRV_POLL_WHILE().
By splitting the header in two files, block-io.h
and block-global-state.h we have a clearer view on what
needs what kind of protection. block-common.h
contains common structures shared by both headers.
block.h is left there for legacy and to avoid changing
all includes in all c files that use the block APIs.
Assertions are added in the next patch.
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Message-Id: <20220303151616.325444-4-eesposit@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
