First, this permission never protected a node from being changed, as
generic child-replacing functions don't check it.
Second, it's a strange thing: it presents a permission of parent node
to change its child. But generally, children are replaced by different
mechanisms, like jobs or qmp commands, not by nodes.
Graph-mod permission is hard to understand. All other permissions
describe operations which done by parent node on its child: read,
write, resize. Graph modification operations are something completely
different.
The only place where BLK_PERM_GRAPH_MOD is used as "perm" (not shared
perm) is mirror_start_job, for s->target. Still modern code should use
bdrv_freeze_backing_chain() to protect from graph modification, if we
don't do it somewhere it may be considered as a bug. So, it's a bit
risky to drop GRAPH_MOD, and analyzing of possible loss of protection
is hard. But one day we should do it, let's do it now.
One more bit of information is that locking the corresponding byte in
file-posix doesn't make sense at all.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210902093754.2352-1-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In most of the block layer, especially when traversing down from other
BlockDriverStates, we assume that BdrvChild.bs can never be NULL. When
it becomes NULL, it is expected that the corresponding BdrvChild pointer
also becomes NULL and the BdrvChild object is freed.
Therefore, once bdrv_replace_child_noperm() sets the BdrvChild.bs
pointer to NULL, it should also immediately set the corresponding
BdrvChild pointer (like bs->file or bs->backing) to NULL.
In that context, it also makes sense for this function to free the
child. Sometimes we cannot do so, though, because it is called in a
transactional context where the caller might still want to reinstate the
child in the abort branch (and free it only on commit), so this behavior
has to remain optional.
In bdrv_replace_child_tran()'s abort handler, we now rely on the fact
that the BdrvChild passed to bdrv_replace_child_tran() must have had a
non-NULL .bs pointer initially. Make a note of that and assert it.
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20211111120829.81329-10-hreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20211115145409.176785-10-kwolf@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
As of a future commit, bdrv_replace_child_noperm() will clear the
indirect BdrvChild pointer passed to it if the new child BDS is NULL.
bdrv_replace_child_tran() will want to let it do that, but revert this
change in its abort handler. For that, we need to have it receive a
BdrvChild ** pointer, too, and keep it stored in the
BdrvReplaceChildState object that we attach to the transaction.
Note that we do not need to store it in the BdrvReplaceChildState when
new_bs is not NULL, because then there is nothing to revert. This is
important so that bdrv_replace_node_noperm() can pass a pointer to a
loop-local variable to bdrv_replace_child_tran() without worrying that
this pointer will outlive one loop iteration.
(Of course, for that to work, bdrv_replace_node_noperm() and in turn
bdrv_replace_node() and its relatives may not be called with a NULL @to
node. Luckily, they already are not, but now we should assert this.)
bdrv_remove_file_or_backing_child() on the other hand needs to ensure
that the indirect pointer it passes will stay valid for the duration of
the transaction. Ensure this by keeping a strong reference to the BDS
whose &bs->backing or &bs->file it passes to bdrv_replace_child_tran(),
and giving up that reference only in the transaction .clean() handler.
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20211111120829.81329-9-hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20211115145409.176785-9-kwolf@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
bdrv_replace_child_noperm() modifies BdrvChild.bs, and can potentially
set it to NULL. That is dangerous, because BDS parents generally assume
that their children's .bs pointer is never NULL. We therefore want to
let bdrv_replace_child_noperm() set the corresponding BdrvChild pointer
to NULL, too.
This patch lays the foundation for it by passing a BdrvChild ** pointer
to bdrv_replace_child_noperm() so that it can later use it to NULL the
BdrvChild pointer immediately after setting BdrvChild.bs to NULL.
(We will still need to undertake some intermediate steps, though.)
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20211111120829.81329-6-hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20211115145409.176785-6-kwolf@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
bdrv_attach_child_common_abort() restores the parent's AioContext. To
do so, the child (which was supposed to be attached, but is now detached
again by this abort handler) is added to the ignore list for the
AioContext changing functions.
However, since we modify a BDS's children list in the BdrvChildClass's
.attach and .detach handlers, the child is already effectively detached
from the parent by this point. We do not need to put it into the ignore
list.
Use this opportunity to clean up the empty line structure: Keep setting
the ignore list, invoking the AioContext function, and freeing the
ignore list in blocks separated by empty lines.
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20211111120829.81329-5-hreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20211115145409.176785-5-kwolf@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Now that bdrv_remove_empty_child() no longer removes the child from the
parent's children list but only checks that it is not in such a list, it
is only a wrapper around bdrv_child_free() that checks that the child is
empty and unused. That should apply to all children that we free, so
put those checks into bdrv_child_free() and drop
bdrv_remove_empty_child().
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20211111120829.81329-4-hreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20211115145409.176785-4-kwolf@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
The children list is specific to BDS parents. We should not modify it
in the general children modification code, but let BDS parents deal with
it in their .attach() and .detach() methods.
This also has the advantage that a BdrvChild is removed from the
children list before its .bs pointer can become NULL. BDS parents
generally assume that their children's .bs pointer is never NULL, so
this is actually a bug fix.
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20211111120829.81329-3-hreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20211115145409.176785-3-kwolf@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Use bdrv_new_open_driver_opts() instead of complicated bdrv_open().
Among other extra things bdrv_open() also check for white-listed
formats, which we don't want for internal node creation: currently
backup doesn't work when copy-before-write filter is not white-listed.
As well block-stream doesn't work when copy-on-read is not
white-listed.
Fixes: 751cec7a26
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=2004812
Reported-by: Yanan Fu
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210920115538.264372-5-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
- use ERRP_GUARD(): function calls error_prepend(), so it must use
ERRP_GUARD(), otherwise error_prepend() would not be called when
passed errp is error_fatal
- drop error propagation, handle return code instead
- for symmetry, do error_prepend() for the second failure
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210920115538.264372-3-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add version of bdrv_new_open_driver() that supports QDict options.
We'll use it in further commit.
Simply add one more argument to bdrv_new_open_driver() is worse, as
there are too many invocations of bdrv_new_open_driver() to update
then.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210920115538.264372-2-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We must not inactivate child when parent has write permissions on
it.
Calling .bdrv_inactivate() doesn't help: actually only qcow2 has this
handler and it is used to flush caches, not for permission
manipulations.
So, let's simply check cumulative parent permissions before
inactivating the node.
This commit fixes a crash when we do migration during backup: prior to
the commit nothing prevents all nodes inactivation at migration finish
and following backup write to the target crashes on assertion
"assert(!(bs->open_flags & BDRV_O_INACTIVE));" in
bdrv_co_write_req_prepare().
After the commit, we rely on the fact that copy-before-write filter
keeps write permission on target node to be able to write to it. So
inactivation fails and migration fails as expected.
Corresponding test now passes, so, enable it.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20210911120027.8063-3-vsementsov@virtuozzo.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
As we have attempted before
(https://lists.gnu.org/archive/html/qemu-devel/2019-01/msg06451.html,
"file-posix: Cache lseek result for data regions";
https://lists.nongnu.org/archive/html/qemu-block/2021-02/msg00934.html,
"file-posix: Cache next hole"), this patch seeks to reduce the number of
SEEK_DATA/HOLE operations the file-posix driver has to perform. The
main difference is that this time it is implemented as part of the
general block layer code.
The problem we face is that on some filesystems or in some
circumstances, SEEK_DATA/HOLE is unreasonably slow. Given the
implementation is outside of qemu, there is little we can do about its
performance.
We have already introduced the want_zero parameter to
bdrv_co_block_status() to reduce the number of SEEK_DATA/HOLE calls
unless we really want zero information; but sometimes we do want that
information, because for files that consist largely of zero areas,
special-casing those areas can give large performance boosts. So the
real problem is with files that consist largely of data, so that
inquiring the block status does not gain us much performance, but where
such an inquiry itself takes a lot of time.
To address this, we want to cache data regions. Most of the time, when
bad performance is reported, it is in places where the image is iterated
over from start to end (qemu-img convert or the mirror job), so a simple
yet effective solution is to cache only the current data region.
(Note that only caching data regions but not zero regions means that
returning false information from the cache is not catastrophic: Treating
zeroes as data is fine. While we try to invalidate the cache on zero
writes and discards, such incongruences may still occur when there are
other processes writing to the image.)
We only use the cache for nodes without children (i.e. protocol nodes),
because that is where the problem is: Drivers that rely on block-status
implementations outside of qemu (e.g. SEEK_DATA/HOLE).
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/307
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20210812084148.14458-3-hreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
[hreitz: Added `local_file == bs` assertion, as suggested by Vladimir]
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Currently, the block driver whitelists are only applied for the system
emulator. All other binaries still give unrestricted access to all block
drivers. There are use cases where this made sense because the main
concern was avoiding customers running VMs on less optimised block
drivers and getting bad performance. Allowing the same image format e.g.
as a target for 'qemu-img convert' is not a problem then.
However, if the concern is the supportability of the driver in general,
either in full or when used read-write, not applying the list driver
whitelist in tools doesn't help - especially since qemu-nbd and
qemu-storage-daemon now give access to more or less the same operations
in block drivers as running a system emulator.
In order to address this, introduce a new configure option that enforces
the driver whitelist in all binaries.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210709164141.254097-1-kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
As the BlockReopenQueue can contain nodes in multiple AioContexts, only
one of which may be locked when AIO_WAIT_WHILE() can be called, we can't
let the caller lock the right contexts. Instead, individually lock the
AioContext of a single node when iterating the queue.
Reintroduce bdrv_reopen() as a wrapper for reopening a single node that
drains the node and temporarily drops the AioContext lock for
bdrv_reopen_multiple().
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210708114709.206487-4-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Move the code to free a BlockReopenQueue to a separate function.
It will be used in a subsequent patch.
[ kwolf: Also free explicit_options and options, and explicitly
qobject_ref() the value when it continues to be used. This makes
future memory leaks less likely. ]
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210708114709.206487-3-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Back in commit d9f059aa6c (qemu-img: Deprecate use of -b without -F),
we deprecated the ability to create a file with a backing image that
requires qemu to perform format probing. Qemu can still probe older
files for backwards compatibility, but it is time to finish off the
ability to create such images, due to the potential security risk they
present. Update a couple of iotests affected by the change.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20210503213600.569128-3-eblake@redhat.com>
Reviewed-by: Connor Kuehl <ckuehl@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When the x-blockdev-reopen was added it allowed reconfiguring the
graph by replacing backing files, but changing the 'file' option was
forbidden. Because of this restriction some operations are not
possible, notably inserting and removing block filters.
This patch adds support for replacing the 'file' option. This is
similar to replacing the backing file and the user is likewise
responsible for the correctness of the resulting graph, otherwise this
can lead to data corruption.
Signed-off-by: Alberto Garcia <berto@igalia.com>
[vsementsov: bdrv_reopen_parse_file_or_backing() is modified a lot]
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210610120537.196183-9-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It's used only in bdrv_reopen_commit(). "backing" is covered by the
loop through all children except for case when we removed backing child
during reopen.
Make it more obvious and drop extra boolean field: qdict_del will not
fail if there is no such entry.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210610120537.196183-8-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Move supports_backing check of bdrv_reopen_parse_backing to called
(through bdrv_set_backing_noperm()) bdrv_set_file_or_backing_noperm()
function. The check applies to general case, so it's appropriate for
bdrv_set_file_or_backing_noperm().
We have to declare backing support for two test drivers, otherwise new
check fails.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210610120537.196183-7-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The logic around finding overlay here is not obvious. Actually it does
two simple things:
1. If new bs is already in backing chain, split from parent bs by
several implicit filters we are done, do nothing.
2. Otherwise, don't try to replace implicit filter.
Let's rewrite this in more obvious way.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210610120537.196183-6-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
