Creating an image that requires format probing of the backing image is
potentially unsafe (we've had several CVEs over the years based on
probes leaking information to the guest on a subsequent boot, although
these days tools like libvirt are aware of the issue enough to prevent
the worst effects). For example, if our probing algorithm ever
changes, or if other tools like libvirt determine a different probe
result than we do, then subsequent use of that backing file under a
different format will present corrupted data to the guest.
Fortunately, the worst effects occur only when the backing image is
originally raw, and we at least prevent commit into a probed raw
backing file that would change its probed type.
Still, it is worth starting a deprecation clock so that future
qemu-img can refuse to create backing chains that would rely on
probing, to encourage clients to avoid unsafe practices. Most
warnings are intentionally emitted from bdrv_img_create() in the block
layer, but qemu-img convert uses bdrv_create() which cannot emit its
own warning without causing spurious warnings on other code paths. In
the end, all command-line image creation or backing file rewriting now
performs a check.
Furthermore, if we probe a backing file as non-raw, then it is safe to
explicitly record that result (rather than relying on future probes);
only where we probe a raw image do we care about further warnings to
the user when using such an image (for example, commits into a
probed-raw backing file are prevented), to help them improve their
tooling. But whether or not we make the probe results explicit, we
still warn the user to remind them to upgrade their workflow to supply
-F always.
iotest 114 specifically wants to create an unsafe image for later
amendment rather than defaulting to our new default of recording a
probed format, so it needs an update. While touching it, expand it to
cover all of the various warnings enabled by this patch. iotest 301
also shows a change to qcow messages.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200706203954.341758-11-eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For now, this is a mechanical addition; all callers pass false. But
the next patch will use it to improve 'qemu-img rebase -u' when
selecting a backing file with no format.
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Reviewed-by: Ján Tomko <jtomko@redhat.com>
Message-Id: <20200706203954.341758-10-eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Back in commit 6e6e55f5 (Jul 2017, v2.10), we tweaked the code to warn
if the backing file could not be opened but the user gave a size,
unless the user also passes the -u option to bypass the open of the
backing file. As one common reason for failure to open the backing
file is when there is mismatch in the requested backing format in
relation to what the backing file actually contains, we actually want
to open the backing file and ensure that it has the right format in as
many cases as possible. iotest 301 for qcow demonstrates how
detecting explicit format mismatch is useful to prevent the creation
of an image that would probe differently than the user requested. Now
is the time to finally turn the warning an error, as promised.
Note that the original warning was added prior to our documentation of
an official deprecation policy (eb22aeca, also Jul 2017), and because
the warning didn't mention the word "deprecated", we never actually
remembered to document it as such. But the warning has been around
long enough that I don't see prolonging it another two releases.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200706203954.341758-7-eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
During 'qemu-img create ... 2>&1', if --quiet is not in force, we can
end up with buffered I/O in stdout that was produced before failure,
but which appears in output after failure. This is confusing; the fix
is to flush stdout prior to attempting anything that might produce an
error message. Several iotests demonstrate the resulting ordering
change now that the merged outputs now reflect chronology. (An even
better fix would be to avoid printf from within block.c altogether,
but that's much more invasive...)
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200706203954.341758-2-eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When all we do with an Error we receive into a local variable is
propagating to somewhere else, we can just as well receive it there
right away, even when we need to keep error_propagate() for other
error paths.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200707160613.848843-38-armbru@redhat.com>
When all we do with an Error we receive into a local variable is
propagating to somewhere else, we can just as well receive it there
right away. The previous commit did that with a Coccinelle script I
consider fairly trustworthy. This commit uses the same script with
the matching of return taken out, i.e. we convert
if (!foo(..., &err)) {
...
error_propagate(errp, err);
...
}
to
if (!foo(..., errp)) {
...
...
}
This is unsound: @err could still be read between afterwards. I don't
know how to express "no read of @err without an intervening write" in
Coccinelle. Instead, I manually double-checked for uses of @err.
Suboptimal line breaks tweaked manually. qdev_realize() simplified
further to placate scripts/checkpatch.pl.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200707160613.848843-36-armbru@redhat.com>
When all we do with an Error we receive into a local variable is
propagating to somewhere else, we can just as well receive it there
right away. Convert
if (!foo(..., &err)) {
...
error_propagate(errp, err);
...
return ...
}
to
if (!foo(..., errp)) {
...
...
return ...
}
where nothing else needs @err. Coccinelle script:
@rule1 forall@
identifier fun, err, errp, lbl;
expression list args, args2;
binary operator op;
constant c1, c2;
symbol false;
@@
if (
(
- fun(args, &err, args2)
+ fun(args, errp, args2)
|
- !fun(args, &err, args2)
+ !fun(args, errp, args2)
|
- fun(args, &err, args2) op c1
+ fun(args, errp, args2) op c1
)
)
{
... when != err
when != lbl:
when strict
- error_propagate(errp, err);
... when != err
(
return;
|
return c2;
|
return false;
)
}
@rule2 forall@
identifier fun, err, errp, lbl;
expression list args, args2;
expression var;
binary operator op;
constant c1, c2;
symbol false;
@@
- var = fun(args, &err, args2);
+ var = fun(args, errp, args2);
... when != err
if (
(
var
|
!var
|
var op c1
)
)
{
... when != err
when != lbl:
when strict
- error_propagate(errp, err);
... when != err
(
return;
|
return c2;
|
return false;
|
return var;
)
}
@depends on rule1 || rule2@
identifier err;
@@
- Error *err = NULL;
... when != err
Not exactly elegant, I'm afraid.
The "when != lbl:" is necessary to avoid transforming
if (fun(args, &err)) {
goto out
}
...
out:
error_propagate(errp, err);
even though other paths to label out still need the error_propagate().
For an actual example, see sclp_realize().
Without the "when strict", Coccinelle transforms vfio_msix_setup(),
incorrectly. I don't know what exactly "when strict" does, only that
it helps here.
The match of return is narrower than what I want, but I can't figure
out how to express "return where the operand doesn't use @err". For
an example where it's too narrow, see vfio_intx_enable().
Silently fails to convert hw/arm/armsse.c, because Coccinelle gets
confused by ARMSSE being used both as typedef and function-like macro
there. Converted manually.
Line breaks tidied up manually. One nested declaration of @local_err
deleted manually. Preexisting unwanted blank line dropped in
hw/riscv/sifive_e.c.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200707160613.848843-35-armbru@redhat.com>
When creating an image fails because the format doesn't support option
"backing_file" or "backing_fmt", bdrv_img_create() first has
qemu_opt_set() put a generic error into @local_err, then puts the real
error into @errp with error_setg(), and then propagates the former to
the latter, which throws away the generic error. A bit complicated,
but works.
Now that qemu_opt_set() returns a useful value, we can simply ignore
the generic error instead.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200707160613.848843-16-armbru@redhat.com>
The function has only one user: bdrv_co_block_status(). Inline it to
simplify reviewing of the following patches, which will finally drop
unallocated_blocks_are_zero field too.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200528094405.145708-3-vsementsov@virtuozzo.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
'force' option will be used for some unsafe amend operations.
This includes things like erasing last keyslot in luks based formats
(which destroys the data, unless the master key is backed up
by external means), but that _might_ be desired result.
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200608094030.670121-4-mlevitsk@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Commit 93676c88 relaxed our NBD client code to request export names up
to the NBD protocol maximum of 4096 bytes without NUL terminator, even
though the block layer can't store anything longer than 4096 bytes
including NUL terminator for display to the user. Since this means
there are some export names where we have to truncate things, we can
at least try to make the truncation a bit more obvious for the user.
Note that in spite of the truncated display name, we can still
communicate with an NBD server using such a long export name; this was
deemed nicer than refusing to even connect to such a server (since the
server may not be under our control, and since determining our actual
length limits gets tricky when nbd://host:port/export and
nbd+unix:///export?socket=/path are themselves variable-length
expansions beyond the export name but count towards the block layer
name length).
Reported-by: Xueqiang Wei <xuwei@redhat.com>
Fixes: https://bugzilla.redhat.com/1843684
Signed-off-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200610163741.3745251-3-eblake@redhat.com>
Implementations should decide the necessary permissions based on @role.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-35-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-33-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-32-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-31-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Replace child_file by child_of_bds in all remaining places (excluding
tests).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-28-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-25-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Make all parents of backing files pass the appropriate BdrvChildRole.
By doing so, we can switch their BdrvChildClass over to the generic
child_of_bds, which will do the right thing when given a correct
BdrvChildRole.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-24-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-23-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This callback can be used by BDSs that use child_of_bds with the
appropriate BdrvChildRole for their children.
Also, make bdrv_format_default_perms() use it for child_of_bds children
(just a temporary solution until we can drop bdrv_format_default_perms()
altogether).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-20-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We can be less restrictive about pure data children than those with
metadata on them, so let bdrv_default_perms_for_storage() handle
metadata children differently from pure data children.
As explained in the code, the restrictions on metadata children are
strictly stricter than those for pure data children, so in theory we
just have to distinguish between pure-data and all other storage
children (pure metadata or data+metadata). In practice, that is not
obvious, though, so we have two independent code paths for metadata and
for data children, and data+metadata children will go through both
(without the path for data children doing anything meaningful).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-19-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Right now, bdrv_format_default_perms() is used by format parents
(generally). We want to switch to a model where most parents use a
single BdrvChildClass, which then decides the permissions based on the
child role. To do so, we have to split bdrv_format_default_perms() into
separate functions for each such role.
Note that bdrv_default_perms_for_storage() currently handles all DATA |
METADATA children. A follow-up patch is going to split it further into
one function for each case.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-18-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Right now, bdrv_format_default_perms() is used by format parents
(generally). We want to switch to a model where most parents use a
single BdrvChildClass, which then decides the permissions based on the
child role. To do so, we have to split bdrv_format_default_perms() into
separate functions for each such role.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-17-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_format_default_perms() has one code path for backing files, and one
for storage files. We want to pull them out into their own functions,
so make sure they are completely distinct before so the next patches
will be a bit cleaner.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-16-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Any current user of child_file, child_format, and child_backing can and
should use this generic BdrvChildClass instead, as it can handle all of
these cases. However, to be able to do so, the users must pass the
appropriate BdrvChildRole when the child is created/attached. (The
following commits will take care of that.)
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-15-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Make bdrv_child_cb_detach() call bdrv_backing_detach() for children with
a COW role (and drop the reverse call from bdrv_backing_detach()), so it
can be used for any child (with a proper role set).
Because so far no child has a proper role set, we need a temporary new
callback for child_backing.detach that ensures bdrv_backing_detach() is
called for all COW children that do not have their role set yet.
(Also, move bdrv_child_cb_detach() down to group it with
bdrv_inherited_options() and bdrv_child_cb_attach().)
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-14-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Make bdrv_child_cb_attach() call bdrv_backing_attach() for children with
a COW role (and drop the reverse call from bdrv_backing_attach()), so it
can be used for any child (with a proper role set).
Because so far no child has a proper role set, we need a temporary new
callback for child_backing.attach that ensures bdrv_backing_attach() is
called for all COW children that do not have their role set yet.
(Also, move bdrv_child_cb_attach() down to group it with
bdrv_inherited_options().)
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-13-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Let child_file's, child_format's, and child_backing's .inherit_options()
implementations fall back to bdrv_inherited_options() to show that it
would really work for all of these cases, if only the parents passed the
appropriate BdrvChildRole and parent_is_format values.
(Also, make bdrv_open_inherit(), the only place to explicitly call
bdrv_backing_options(), call bdrv_inherited_options() instead.)
This patch should incur only two visible changes, both for child_format
children, both of which are effectively bug fixes:
First, they no longer have discard=unmap set by default. This reason it
was set is because bdrv_inherited_fmt_options() fell through to
bdrv_protocol_options(), and that set it because "format drivers take
care to send flushes and respect unmap policy". None of the drivers
that use child_format for their children (quorum and blkverify) are
format drivers, though, so this reasoning does not apply here.
Second, they no longer have BDRV_O_NO_IO force-cleared. child_format
was used solely for children that do not store any metadata and as such
will not be accessed by their parents as long as those parents do not
receive I/O themselves. Thus, such children should inherit
BDRV_O_NO_IO.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-12-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
After the series this patch belongs to, we want to have a common
BdrvChildClass that encompasses all of child_file, child_format, and
child_backing. Such a single class needs a single .inherit_options()
implementation, and this patch introduces it.
The next patch will show how the existing implementations can fall back
to it just by passing appropriate BdrvChildRole and parent_is_format
values.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200513110544.176672-11-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The other two .inherit_options implementations specify exactly for what
case they are used in their name, so do it for this one as well.
(The actual intention behind this patch is to follow it up with a
generic bdrv_inherited_options() that works for all three cases.)
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-10-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We plan to unify the generic .inherit_options() functions. The
resulting common function will need to decide whether to force-enable
format probing, force-disable it, or leave it as-is. To make this
decision, it will need to know whether the parent node is a format node
or not (because we never want format probing if the parent is a format
node already (except for the backing chain)).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-9-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For now, all callers (effectively) pass 0 and no callee evaluates thie
value. Later patches will change both.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-8-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For now, all callers pass 0 and no callee evaluates this value. Later
patches will change both.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-7-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
For now, it is always set to 0. Later patches in this series will
ensure that all callers pass an appropriate combination of flags.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200513110544.176672-6-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This structure nearly only contains parent callbacks for child state
changes. It cannot really reflect a child's role, because different
roles may overlap (as we will see when real roles are introduced), and
because parents can have custom callbacks even when the child fulfills a
standard role.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-Id: <20200513110544.176672-4-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Right now, all users of bdrv_make_empty() call the BlockDriver method
directly. That is not only bad style, it is also wrong, unless the
caller has a BdrvChild with a WRITE or WRITE_UNCHANGED permission.
(WRITE_UNCHANGED suffices, because callers generally use this function
to clear a node with a backing file after a commit operation.)
Introduce bdrv_make_empty() that verifies that it does.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200429141126.85159-2-mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that there are no clients of bdrv_has_zero_init_truncate, none of
the drivers need to worry about providing it.
What's more, this eliminates a source of some confusion: a literal
reading of the documentation as written in ceaca56f and implemented in
commit 1dcaf527 claims that a driver which returns 0 for
bdrv_has_zero_init_truncate() must not return 1 for
bdrv_has_zero_init(); this condition was violated for parallels, qcow,
and sometimes for vdi, although in practice it did not matter since
those drivers also lacked .bdrv_co_truncate.
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200428202905.770727-10-eblake@redhat.com>
Acked-by: Richard W.M. Jones <rjones@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that node level interface bdrv_truncate() supports passing request
flags to the block driver, expose this on the BlockBackend level, too.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200424125448.63318-4-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The previous few commits have made this more obvious, and removed the
one exception. Time to clarify the documentation, and drop dead error
checking.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20200424084338.26803-13-armbru@redhat.com>
bdrv_root_attach_child promises to drop child_bs reference on failure.
It does it on first handled failure path, but not on the second. Fix
that.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200324155921.23822-1-vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of checking the .bdrv_co_create_opts to see if we need the
fallback, just implement the .bdrv_co_create_opts in the drivers that
need it.
This way we don't break various places that need to know if the
underlying protocol/format really supports image creation, and this way
we still allow some drivers to not support image creation.
Fixes: fd17146cd9
Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1816007
Note that technically this driver reverts the image creation fallback
for the vxhs driver since I don't have a means to test it, and IMHO it
is better to leave it not supported as it was prior to generic image
creation patches.
Also drop iscsi_create_opts which was left accidentally.
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20200326011218.29230-3-mlevitsk@redhat.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
[mreitz: Fixed alignment, and moved bdrv_co_create_opts_simple() and
bdrv_create_opts_simple from block.h into block_int.h]
Signed-off-by: Max Reitz <mreitz@redhat.com>
This will allow the reuse of a single generic .bdrv_co_create
implementation for several drivers.
No functional changes.
Signed-off-by: Maxim Levitsky <mlevitsk@redhat.com>
Message-Id: <20200326011218.29230-2-mlevitsk@redhat.com>
Reviewed-by: Denis V. Lunev <den@openvz.org>
Signed-off-by: Max Reitz <mreitz@redhat.com>
There is a use-after-free possible: bdrv_unref_child() leaves
bs->backing freed but not NULL. bdrv_attach_child may produce nested
polling loop due to drain, than access of freed pointer is possible.
I've produced the following crash on 30 iotest with modified code. It
does not reproduce on master, but still seems possible:
#0 __strcmp_avx2 () at /lib64/libc.so.6
#1 bdrv_backing_overridden (bs=0x55c9d3cc2060) at block.c:6350
#2 bdrv_refresh_filename (bs=0x55c9d3cc2060) at block.c:6404
#3 bdrv_backing_attach (c=0x55c9d48e5520) at block.c:1063
#4 bdrv_replace_child_noperm
(child=child@entry=0x55c9d48e5520,
new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2290
#5 bdrv_replace_child
(child=child@entry=0x55c9d48e5520,
new_bs=new_bs@entry=0x55c9d3cc2060) at block.c:2320
#6 bdrv_root_attach_child
(child_bs=child_bs@entry=0x55c9d3cc2060,
child_name=child_name@entry=0x55c9d241d478 "backing",
child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
ctx=<optimized out>, perm=<optimized out>, shared_perm=21,
opaque=0x55c9d3c5a3d0, errp=0x7ffd117108e0) at block.c:2424
#7 bdrv_attach_child
(parent_bs=parent_bs@entry=0x55c9d3c5a3d0,
child_bs=child_bs@entry=0x55c9d3cc2060,
child_name=child_name@entry=0x55c9d241d478 "backing",
child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
errp=errp@entry=0x7ffd117108e0) at block.c:5876
#8 in bdrv_set_backing_hd
(bs=bs@entry=0x55c9d3c5a3d0,
backing_hd=backing_hd@entry=0x55c9d3cc2060,
errp=errp@entry=0x7ffd117108e0)
at block.c:2576
#9 stream_prepare (job=0x55c9d49d84a0) at block/stream.c:150
#10 job_prepare (job=0x55c9d49d84a0) at job.c:761
#11 job_txn_apply (txn=<optimized out>, fn=<optimized out>) at
job.c:145
#12 job_do_finalize (job=0x55c9d49d84a0) at job.c:778
#13 job_completed_txn_success (job=0x55c9d49d84a0) at job.c:832
#14 job_completed (job=0x55c9d49d84a0) at job.c:845
#15 job_completed (job=0x55c9d49d84a0) at job.c:836
#16 job_exit (opaque=0x55c9d49d84a0) at job.c:864
#17 aio_bh_call (bh=0x55c9d471a160) at util/async.c:117
#18 aio_bh_poll (ctx=ctx@entry=0x55c9d3c46720) at util/async.c:117
#19 aio_poll (ctx=ctx@entry=0x55c9d3c46720,
blocking=blocking@entry=true)
at util/aio-posix.c:728
#20 bdrv_parent_drained_begin_single (poll=true, c=0x55c9d3d558f0)
at block/io.c:121
#21 bdrv_parent_drained_begin_single (c=c@entry=0x55c9d3d558f0,
poll=poll@entry=true)
at block/io.c:114
#22 bdrv_replace_child_noperm
(child=child@entry=0x55c9d3d558f0,
new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2258
#23 bdrv_replace_child
(child=child@entry=0x55c9d3d558f0,
new_bs=new_bs@entry=0x55c9d3d27300) at block.c:2320
#24 bdrv_root_attach_child
(child_bs=child_bs@entry=0x55c9d3d27300,
child_name=child_name@entry=0x55c9d241d478 "backing",
child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
ctx=<optimized out>, perm=<optimized out>, shared_perm=21,
opaque=0x55c9d3cc2060, errp=0x7ffd11710c60) at block.c:2424
#25 bdrv_attach_child
(parent_bs=parent_bs@entry=0x55c9d3cc2060,
child_bs=child_bs@entry=0x55c9d3d27300,
child_name=child_name@entry=0x55c9d241d478 "backing",
child_role=child_role@entry=0x55c9d26ecee0 <child_backing>,
errp=errp@entry=0x7ffd11710c60) at block.c:5876
#26 bdrv_set_backing_hd
(bs=bs@entry=0x55c9d3cc2060,
backing_hd=backing_hd@entry=0x55c9d3d27300,
errp=errp@entry=0x7ffd11710c60)
at block.c:2576
#27 stream_prepare (job=0x55c9d495ead0) at block/stream.c:150
...
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200316060631.30052-2-vsementsov@virtuozzo.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_do_find_format() calls strcmp() using BlockDriver::format_name
as argument, which must not be NULL. Assert this field is not null
when we register a block driver in bdrv_register().
Reported-by: Mansour Ahmadi <ManSoSec@gmail.com>
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200318222235.23856-1-philmd@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Using the new 'bdrv_co_delete_file' interface, a pure co_routine function
'bdrv_co_delete_file' inside block.c can can be used in a way similar of
the existing bdrv_create_file to to clean up a created file.
We're creating a pure co_routine because the only caller of
'bdrv_co_delete_file' will be already in co_routine context, thus there
is no need to add all the machinery to check for qemu_in_coroutine() and
create a separated co_routine to do the job.
Suggested-by: Daniel P. Berrangé <berrange@redhat.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Message-Id: <20200130213907.2830642-3-danielhb413@gmail.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
external_snapshot_prepare() tries to move the overlay to the AioContext
of the backing file (the snapshotted node). However, it's possible that
this doesn't work, but the backing file can instead be moved to the
overlay's AioContext (e.g. opening the backing chain for a mirror
target).
bdrv_append() already indirectly uses bdrv_attach_node(), which takes
care to move nodes to make sure they use the same AioContext and which
tries both directions.
So the problem has a simple fix: Just delete the unnecessary extra
bdrv_try_set_aio_context() call in external_snapshot_prepare() and
instead assert in bdrv_append() that both nodes were indeed moved to the
same AioContext.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200310113831.27293-6-kwolf@redhat.com>
Tested-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20200310113831.27293-2-kwolf@redhat.com>
Reviewed-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch allows bdrv_reopen() (and therefore the x-blockdev-reopen QMP
command) to attach a node as the new backing file even if the node is in
a different AioContext than the parent if one of both nodes can be moved
to the AioContext of the other node.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Tested-by: Peter Krempa <pkrempa@redhat.com>
Message-Id: <20200306141413.30705-3-kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add another step in the reopen process where driver can execute code
after permission changes are comitted.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Message-Id: <adc02cf591c3cb34e98e33518eb1c540a0f27db1.1582893284.git.pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
@options is leaked by the first two return statements in this function.
Note that blk_new_open() takes the reference to @options even on
failure, so all we need to do to fix the leak is to move the QDict
allocation down to where we actually need it.
Reported-by: Coverity (CID 1419884)
Fixes: fd17146cd9
("block: Generic file creation fallback")
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200225155618.133412-1-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
QLIST_REMOVE() assumes the element is in a list. It also leaves the
element's linked list pointers dangling.
Introduce a safe version of QLIST_REMOVE() and convert open-coded
instances of this pattern.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Sergio Lopez <slp@redhat.com>
Message-id: 20200214171712.541358-4-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
If a protocol driver does not support image creation, we can see whether
maybe the file exists already. If so, just truncating it will be
sufficient.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20200122164532.178040-3-mreitz@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
When a management application manages node names there's no reason to
recurse into backing images in the output of query-named-block-nodes.
Add a parameter to the command which will return just the top level
structs.
Signed-off-by: Peter Krempa <pkrempa@redhat.com>
Message-Id: <4470f8c779abc404dcf65e375db195cd91a80651.1579509782.git.pkrempa@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[mreitz: Fixed coding style]
Signed-off-by: Max Reitz <mreitz@redhat.com>
It no longer has any users.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200218103454.296704-11-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Let check_to_replace_node() use the more specialized
bdrv_recurse_can_replace() instead of
bdrv_recurse_is_first_non_filter(), which is too restrictive (or, in the
case of quorum, sometimes not restrictive enough).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200218103454.296704-10-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
After a couple of follow-up patches, this function will replace
bdrv_recurse_is_first_non_filter() in check_to_replace_node().
bdrv_recurse_is_first_non_filter() is both not sufficiently specific for
check_to_replace_node() (it allows cases that should not be allowed,
like replacing child nodes of quorum with dissenting data that have more
parents than just quorum), and it is too restrictive (it is perfectly
fine to replace filters).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200218103454.296704-7-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It is unused now. (And it was ugly because it needed to explore all BDS
chains from the top.)
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20200218103454.296704-4-mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Fixes: 132ada80c4
Signed-off-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Message-Id: <20200218094402.26625-4-philmd@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If we call the qmp 'query-block' while qemu is working on
'block-commit', it will cause memleaks, the memory leak stack is as
follow:
Indirect leak of 12360 byte(s) in 3 object(s) allocated from:
#0 0x7f80f0b6d970 in __interceptor_calloc (/lib64/libasan.so.5+0xef970)
#1 0x7f80ee86049d in g_malloc0 (/lib64/libglib-2.0.so.0+0x5249d)
#2 0x55ea95b5bb67 in qdict_new /mnt/sdb/qemu-4.2.0-rc0/qobject/qdict.c:29
#3 0x55ea956cd043 in bdrv_refresh_filename /mnt/sdb/qemu-4.2.0-rc0/block.c:6427
#4 0x55ea956cc950 in bdrv_refresh_filename /mnt/sdb/qemu-4.2.0-rc0/block.c:6399
#5 0x55ea956cc950 in bdrv_refresh_filename /mnt/sdb/qemu-4.2.0-rc0/block.c:6399
#6 0x55ea956cc950 in bdrv_refresh_filename /mnt/sdb/qemu-4.2.0-rc0/block.c:6399
#7 0x55ea958818ea in bdrv_block_device_info /mnt/sdb/qemu-4.2.0-rc0/block/qapi.c:56
#8 0x55ea958879de in bdrv_query_info /mnt/sdb/qemu-4.2.0-rc0/block/qapi.c:392
#9 0x55ea9588b58f in qmp_query_block /mnt/sdb/qemu-4.2.0-rc0/block/qapi.c:578
#10 0x55ea95567392 in qmp_marshal_query_block qapi/qapi-commands-block-core.c:95
Indirect leak of 4120 byte(s) in 1 object(s) allocated from:
#0 0x7f80f0b6d970 in __interceptor_calloc (/lib64/libasan.so.5+0xef970)
#1 0x7f80ee86049d in g_malloc0 (/lib64/libglib-2.0.so.0+0x5249d)
#2 0x55ea95b5bb67 in qdict_new /mnt/sdb/qemu-4.2.0-rc0/qobject/qdict.c:29
#3 0x55ea956cd043 in bdrv_refresh_filename /mnt/sdb/qemu-4.2.0-rc0/block.c:6427
#4 0x55ea956cc950 in bdrv_refresh_filename /mnt/sdb/qemu-4.2.0-rc0/block.c:6399
#5 0x55ea956cc950 in bdrv_refresh_filename /mnt/sdb/qemu-4.2.0-rc0/block.c:6399
#6 0x55ea9569f301 in bdrv_backing_attach /mnt/sdb/qemu-4.2.0-rc0/block.c:1064
#7 0x55ea956a99dd in bdrv_replace_child_noperm /mnt/sdb/qemu-4.2.0-rc0/block.c:2283
#8 0x55ea956b9b53 in bdrv_replace_node /mnt/sdb/qemu-4.2.0-rc0/block.c:4196
#9 0x55ea956b9e49 in bdrv_append /mnt/sdb/qemu-4.2.0-rc0/block.c:4236
#10 0x55ea958c3472 in commit_start /mnt/sdb/qemu-4.2.0-rc0/block/commit.c:306
#11 0x55ea94b68ab0 in qmp_block_commit /mnt/sdb/qemu-4.2.0-rc0/blockdev.c:3459
#12 0x55ea9556a7a7 in qmp_marshal_block_commit qapi/qapi-commands-block-core.c:407
Fixes: bb808d5f5c
Reported-by: Euler Robot <euler.robot@huawei.com>
Signed-off-by: Pan Nengyuan <pannengyuan@huawei.com>
Message-id: 20200116085600.24056-1-pannengyuan@huawei.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This is a bit more efficient than having to allocate and free memory
for each new permission.
The default size (30) is enough for "consistent read, write, resize".
Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 20200110171518.22168-1-berto@igalia.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
We can save some LoC in xdbg_graph_add_edge() by using
bdrv_qapi_perm_to_blk_perm().
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191108123455.39445-3-mreitz@redhat.com
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
We need some way to correlate QAPI BlockPermission values with
BLK_PERM_* flags. We could:
(1) have the same order in the QAPI definition as the the BLK_PERM_*
flags are in LSb-first order. However, then there is no guarantee
that they actually match (e.g. when someone modifies the QAPI schema
without thinking of the BLK_PERM_* definitions).
We could add static assertions, but these would break what’s good
about this solution, namely its simplicity.
(2) define the BLK_PERM_* flags based on the BlockPermission values.
But this way whenever someone were to modify the QAPI order
(perfectly sensible in theory), the BLK_PERM_* values would change.
Because these values are used for file locking, this might break
file locking between different qemu versions.
Therefore, go the slightly more cumbersome way: Add a function to
translate from the QAPI constants to the BLK_PERM_* flags.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20191108123455.39445-2-mreitz@redhat.com
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_invalidate_cache_all() assumes that all nodes in a given subtree
are either active or inactive when it starts. Therefore, as soon as it
arrives at an already active node, it stops.
However, this assumption is wrong. For example, it's possible to take a
snapshot of an inactive node, which results in an active overlay over an
inactive backing file. The active overlay is probably also the root node
of an inactive BlockBackend (blk->disable_perm == true).
In this case, bdrv_invalidate_cache_all() does not need to do anything
to activate the overlay node, but it still needs to recurse into the
children and the parents to make sure that after returning success,
really everything is activated.
Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
If both the create options (qemu-img create -o ...) and the size
parameter were given, the size parameter was silently ignored. Instead,
make specifying two sizes an error.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
bs->options and bs->explicit_options shouldn't contain any options for
child nodes. bdrv_open_inherited() takes care to remove any options that
match a child name after opening the image and the same is done when
reopening.
However, we miss the case of 'backing': null, which is a child option,
but results in no child being created. This means that a 'backing': null
remains in bs->options and bs->explicit_options.
A typical use for 'backing': null is in live snapshots: blockdev-add for
the qcow2 overlay makes sure not to open the backing file (because it is
already opened and blockdev-snapshot will attach it). After doing a
blockdev-snapshot, bs->options and bs->explicit_options become
inconsistent with the actual state (bs has a backing file now, but the
options still say null). On the next occasion that the image is
reopened, e.g. switching it from read-write to read-only when another
snapshot is taken, the option will take effect again and the node
incorrectly loses its backing file.
Fix bdrv_open_inherited() to remove the 'backing' option from
bs->options and bs->explicit_options even for the case where it
specifies that no backing file is wanted.
Reported-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Tested-by: Peter Krempa <pkrempa@redhat.com>
There are three page size in qemu:
real host page size
host page size
target page size
All of them have dedicate variable to represent. For the last two, we
use the same form in the whole qemu project, while for the first one we
use two forms: qemu_real_host_page_size and getpagesize().
qemu_real_host_page_size is defined to be a replacement of
getpagesize(), so let it serve the role.
[Note] Not fully tested for some arch or device.
Signed-off-by: Wei Yang <richardw.yang@linux.intel.com>
Message-Id: <20191013021145.16011-3-richardw.yang@linux.intel.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
The only reason I can imagine for this strange code at the very-end of
bdrv_reopen_commit is the fact that bs->read_only updated after
calling drv->bdrv_reopen_commit in bdrv_reopen_commit. And in the same
time, prior to previous commit, qcow2_reopen_bitmaps_rw did a wrong
check for being writable, when actually it only need writable file
child not self.
So, as it's fixed, let's move things to correct place.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Acked-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190927122355.7344-10-vsementsov@virtuozzo.com
Signed-off-by: John Snow <jsnow@redhat.com>
It's needed to fix reopening qcow2 with bitmaps to RW. Currently it
can't work, as qcow2 needs write access to file child, to mark bitmaps
in-image with IN_USE flag. But usually children goes after parents in
reopen queue and file child is still RO on qcow2 reopen commit. Reverse
reopen order to fix it.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Acked-by: Max Reitz <mreitz@redhat.com>
Acked-by: John Snow <jsnow@redhat.com>
Message-id: 20190927122355.7344-3-vsementsov@virtuozzo.com
Signed-off-by: John Snow <jsnow@redhat.com>
We'll need reverse-foreach in the following commit, QTAILQ support it,
so move to QTAILQ.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190927122355.7344-2-vsementsov@virtuozzo.com
Signed-off-by: John Snow <jsnow@redhat.com>
bdrv_dirty_bitmap_next is always used in same pattern. So, split it
into _next and _first, instead of combining two functions into one and
add FOR_EACH_DIRTY_BITMAP macro.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20190916141911.5255-5-vsementsov@virtuozzo.com
Signed-off-by: John Snow <jsnow@redhat.com>
block/dirty-bitmap.c seems to be more appropriate for it and
bdrv_remove_persistent_dirty_bitmap already in it.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Message-id: 20190920082543.23444-2-vsementsov@virtuozzo.com
Signed-off-by: John Snow <jsnow@redhat.com>
A block driver can provide a callback to report driver-specific
statistics.
file-posix driver now reports discard statistics
Signed-off-by: Anton Nefedov <anton.nefedov@virtuozzo.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Acked-by: Markus Armbruster <armbru@redhat.com>
Message-id: 20190923121737.83281-10-anton.nefedov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Teach bdrv_debug_breakpoint and bdrv_debug_remove_breakpoint skip
filters with backing. This is needed to implement and use in backup job
it's own backup_top filter driver (like mirror already has one), and
without this improvement, breakpoint removal will fail at least in 55
iotest.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190920142056.12778-9-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
It's wrong to OR shared permissions. It may lead to crash on further
permission updates.
Also, no needs to consider previously calculated permissions, as at
this point we already bind all new parents and bdrv_get_cumulative_perm
result is enough. So fix the bug by just set permissions by
bdrv_get_cumulative_perm result.
Bug was introduced in long ago 234ac1a902, in 2.9.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20190824100740.61635-1-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
No .bdrv_has_zero_init() implementation returns 1 if growing the file
would add non-zero areas (at least with PREALLOC_MODE_OFF), so using it
in lieu of this new function was always safe.
But on the other hand, it is possible that growing an image that is not
zero-initialized would still add a zero-initialized area, like when
using nonpreallocating truncation on a preallocated image. For callers
that care only about truncation, not about creation with potential
preallocation, this new function is useful.
Alternatively, we could have added a PreallocMode parameter to
bdrv_has_zero_init(). But the only user would have been qemu-img
convert, which does not have a plain PreallocMode value right now -- it
would have to parse the creation option to obtain it. Therefore, the
simpler solution is to let bdrv_has_zero_init() inquire the
preallocation status and add the new bdrv_has_zero_init_truncate() that
presupposes PREALLOC_MODE_OFF.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190724171239.8764-4-mreitz@redhat.com
Reviewed-by: Maxim Levitsky <mlevitsk@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
It is used to do transactional movement of the bitmap (which is
possible in conjunction with merge command). Transactional bitmap
movement is needed in scenarios with external snapshot, when we don't
want to leave copy of the bitmap in the base image.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: John Snow <jsnow@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190708220502.12977-3-jsnow@redhat.com
[Edited "since" version to 4.2 --js]
Signed-off-by: John Snow <jsnow@redhat.com>
- file-posix: Fix O_DIRECT alignment detection
- Fixes for concurrent block jobs
- block-backend: Queue requests while drained (fix IDE vs. job crashes)
- qemu-img convert: Deprecate using -n and -o together
- iotests: Migration tests with filter nodes
- iotests: More media change tests
-----BEGIN PGP SIGNATURE-----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=T7L5
-----END PGP SIGNATURE-----
Merge remote-tracking branch 'remotes/kevin/tags/for-upstream' into staging
Block layer patches:
- file-posix: Fix O_DIRECT alignment detection
- Fixes for concurrent block jobs
- block-backend: Queue requests while drained (fix IDE vs. job crashes)
- qemu-img convert: Deprecate using -n and -o together
- iotests: Migration tests with filter nodes
- iotests: More media change tests
# gpg: Signature made Fri 16 Aug 2019 10:29:18 BST
# gpg: using RSA key 7F09B272C88F2FD6
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>" [full]
# Primary key fingerprint: DC3D EB15 9A9A F95D 3D74 56FE 7F09 B272 C88F 2FD6
* remotes/kevin/tags/for-upstream:
file-posix: Handle undetectable alignment
qemu-img convert: Deprecate using -n and -o together
block-backend: Queue requests while drained
mirror: Keep mirror_top_bs drained after dropping permissions
block: Remove blk_pread_unthrottled()
iotests: Add test for concurrent stream/commit
tests: Test mid-drain bdrv_replace_child_noperm()
tests: Test polling in bdrv_drop_intermediate()
block: Reduce (un)drains when replacing a child
block: Keep subtree drained in drop_intermediate
block: Simplify bdrv_filter_default_perms()
iotests: Test migration with all kinds of filter nodes
iotests: Move migration helpers to iotests.py
iotests/118: Add -blockdev based tests
iotests/118: Create test classes dynamically
iotests/118: Test media change for scsi-cd
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
In my "build everything" tree, changing qemu/main-loop.h triggers a
recompile of some 5600 out of 6600 objects (not counting tests and
objects that don't depend on qemu/osdep.h). It includes block/aio.h,
which in turn includes qemu/event_notifier.h, qemu/notify.h,
qemu/processor.h, qemu/qsp.h, qemu/queue.h, qemu/thread-posix.h,
qemu/thread.h, qemu/timer.h, and a few more.
Include qemu/main-loop.h only where it's needed. Touching it now
recompiles only some 1700 objects. For block/aio.h and
qemu/event_notifier.h, these numbers drop from 5600 to 2800. For the
others, they shrink only slightly.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20190812052359.30071-21-armbru@redhat.com>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Tested-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Currently, bdrv_replace_child_noperm() undrains the parent until it is
completely undrained, then re-drains it after attaching the new child
node.
This is a problem with bdrv_drop_intermediate(): We want to keep the
whole subtree drained, including parents, while the operation is
under way. bdrv_replace_child_noperm() breaks this by allowing every
parent to become unquiesced briefly, and then redraining it.
In fact, there is no reason why the parent should become unquiesced and
be allowed to submit requests to the new child node if that new node is
supposed to be kept drained. So if anything, we have to drain the
parent before detaching the old child node. Conversely, we have to
undrain it only after attaching the new child node.
Thus, change the whole drain algorithm here: Calculate the number of
times we have to drain/undrain the parent before replacing the child
node then drain it (if necessary), replace the child node, and then
undrain it.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_drop_intermediate() calls BdrvChildRole.update_filename(). That
may poll, thus changing the graph, which potentially breaks the
QLIST_FOREACH_SAFE() loop.
Just keep the whole subtree drained. This is probably the right thing
to do anyway (dropping nodes while the subtree is not drained seems
wrong).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The same change as commit 2b23f28639 ('block/copy-on-read: Fix
permissions for inactive node') made for the copy-on-read driver can be
made for bdrv_filter_default_perms(): Retaining the old permissions from
the BdrvChild if it is given complicates things unnecessarily when in
the end this only means that the options set in the c == NULL case (i.e.
during child creation) are retained.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
bdrv_set_aio_context_ignore() can only work in the main loop:
bdrv_drained_begin() only works in the main loop and the node's (old)
AioContext; and bdrv_drained_end() really only works in the main loop
and the node's (new) AioContext (contrary to its current comment, which
is just wrong).
Consequentially, bdrv_set_aio_context_ignore() must be called from the
main loop. Luckily, assuming that we can make block graph changes only
from the main loop as well, all its callers do that already.
Note that changing a node's context in a sense is an operation that
changes the block graph, so it actually makes sense to require this
function to be called from the main loop.
Also, fix bdrv_drained_end()'s description. You can only use it from
the main loop or the node's AioContext, and in the latter case, the
whole subtree must be in the same context.
Fixes: e037c09c78
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190722133054.21781-3-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
We should never poll anywhere in bdrv_do_drained_end() (including its
recursive callees like bdrv_drain_invoke()), because it does not cope
well with graph changes. In fact, it has been written based on the
postulation that no graph changes will happen in it.
Instead, the callers that want to poll must poll, i.e. all currently
globally available wrappers: bdrv_drained_end(),
bdrv_subtree_drained_end(), bdrv_unapply_subtree_drain(), and
bdrv_drain_all_end(). Graph changes there do not matter.
They can poll simply by passing a pointer to a drained_end_counter and
wait until it reaches 0.
This patch also adds a non-polling global wrapper for
bdrv_do_drained_end() that takes a drained_end_counter pointer. We need
such a variant because now no function called anywhere from
bdrv_do_drained_end() must poll. This includes
BdrvChildRole.drained_end(), which already must not poll according to
its interface documentation, but bdrv_child_cb_drained_end() just
violates that by invoking bdrv_drained_end() (which does poll).
Therefore, BdrvChildRole.drained_end() must take a *drained_end_counter
parameter, which bdrv_child_cb_drained_end() can pass on to the new
bdrv_drained_end_no_poll() function.
Note that we now have a pattern of all drained_end-related functions
either polling or receiving a *drained_end_counter to let the caller
poll based on that.
A problem with a single poll loop is that when the drained section in
bdrv_set_aio_context_ignore() ends, some nodes in the subgraph may be in
the old contexts, while others are in the new context already. To let
the collective poll in bdrv_drained_end() work correctly, we must not
hold a lock to the old context, so that the old context can make
progress in case it is different from the current context.
(In the process, remove the comment saying that the current context is
always the old context, because it is wrong.)
In all other places, all nodes in a subtree must be in the same context,
so we can just poll that. The exception of course is
bdrv_drain_all_end(), but that always runs in the main context, so we
can just poll NULL (like bdrv_drain_all_begin() does).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit 5cb2737e92 laid out why
bdrv_do_drained_end() must decrement the quiesce_counter after
bdrv_drain_invoke(). It did not give a very good reason why it has to
happen after bdrv_parent_drained_end(), instead only claiming symmetry
to bdrv_do_drained_begin().
It turns out that delaying it for so long is wrong.
Situation: We have an active commit job (i.e. a mirror job) from top to
base for the following graph:
filter
|
[file]
|
v
top --[backing]--> base
Now the VM is closed, which results in the job being cancelled and a
bdrv_drain_all() happening pretty much simultaneously.
Beginning the drain means the job is paused once whenever one of its
nodes is quiesced. This is reversed when the drain ends.
With how the code currently is, after base's drain ends (which means
that it will have unpaused the job once), its quiesce_counter remains at
1 while it goes to undrain its parents (bdrv_parent_drained_end()). For
some reason or another, undraining filter causes the job to be kicked
and enter mirror_exit_common(), where it proceeds to invoke
block_job_remove_all_bdrv().
Now base will be detached from the job. Because its quiesce_counter is
still 1, it will unpause the job once more. So in total, undraining
base will unpause the job twice. Eventually, this will lead to the
job's pause_count going negative -- well, it would, were there not an
assertion against this, which crashes qemu.
The general problem is that if in bdrv_parent_drained_end() we undrain
parent A, and then undrain parent B, which then leads to A detaching the
child, bdrv_replace_child_noperm() will undrain A as if we had not done
so yet; that is, one time too many.
It follows that we cannot decrement the quiesce_counter after invoking
bdrv_parent_drained_end().
Unfortunately, decrementing it before bdrv_parent_drained_end() would be
wrong, too. Imagine the above situation in reverse: Undraining A leads
to B detaching the child. If we had already decremented the
quiesce_counter by that point, bdrv_replace_child_noperm() would undrain
B one time too little; because it expects bdrv_parent_drained_end() to
issue this undrain. But bdrv_parent_drained_end() won't do that,
because B is no longer a parent.
Therefore, we have to do something else. This patch opts for
introducing a second quiesce_counter that counts how many times a
child's parent has been quiesced (though c->role->drained_*). With
that, bdrv_replace_child_noperm() just has to undrain the parent exactly
that many times when removing a child, and it will always be right.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
BDS.inherits_from does not always point to an immediate parent node.
When launching a block job with a filter node, for example, the node
directly below the filter will not point to the filter, but keep its old
pointee (above the filter).
If that pointee goes away while the job is still running, the node's
inherits_from will not be updated and thus point to garbage. To fix
this, bdrv_unref_child() has to check not only the parent node's
immediate children for nodes whose inherits_from needs to be cleared,
but its whole subtree.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190703172813.6868-7-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
The commit and the mirror block job must be able to drop their filter
node at any point. However, this will not be possible if any of the
BdrvChild links to them is frozen. Therefore, we need to prevent them
from ever becoming frozen.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190703172813.6868-2-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
We generally assume that loosening permission restrictions can never
fail. We have seen in the past that this assumption is wrong. This has
led to crashes because we generally pass &error_abort when loosening
permissions.
However, a failure in such a case should actually be handled in quite
the opposite way: It is very much not fatal, so qemu may report it, but
still consider the operation successful. The only realistic problem is
that qemu may then retain permissions and thus locks on images it
actually does not require. But again, that is not fatal.
To implement this behavior, we make all functions that change
permissions and that pass &error_abort to the initiating function
(bdrv_check_perm() or bdrv_child_check_perm()) evaluate the
@loosen_restrictions value introduced in the previous patch. If it is
true and an error did occur, we abort the permission update, discard the
error, and instead report success to the caller.
bdrv_child_try_set_perm() itself does not pass &error_abort, but it is
the only public function to change permissions. As such, callers may
pass &error_abort to it, expecting dropping permission restrictions to
never fail.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch makes three functions report whether the necessary permission
change tightens restrictions or not. These functions are:
- bdrv_check_perm()
- bdrv_check_update_perm()
- bdrv_child_check_perm()
Callers can use this result to decide whether a failure is fatal or not
(see the next patch).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We have to start by applying the permission restrictions to new_bs
before we can loosen them on old_bs. See the comment for the
explanation.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If a block node uses bdrv_child_try_set_perm() to change the permission
it takes on its child, the result may be very short-lived. If anything
makes the block layer recalculate the permissions internally, it will
invoke the node driver's .bdrv_child_perm() implementation. The
permission/shared permissions masks that returns will then override the
values previously passed to bdrv_child_try_set_perm().
If drivers want a child edge to have specific values for the
permissions/shared permissions mask, it must return them in
.bdrv_child_perm(). Consequentially, there is no need for them to pass
the same values to bdrv_child_try_set_perm() then: It is better to have
a function that invokes .bdrv_child_perm() and calls
bdrv_child_try_set_perm() with the result. This patch adds such a
function under the name of bdrv_child_refresh_perms().
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Drop remaining users of bs->job:
1. assertions actually duplicated by assert(!bs->refcnt)
2. trace-point seems not enough reason to change stream_start to return
BlockJob pointer
3. Restricting creation of two jobs based on same bs is bad idea, as
3.1 Some jobs creates filters to be their main node, so, this check
don't actually prevent creating second job on same real node (which
will create another filter node) (but I hope it is restricted by
other mechanisms)
3.2 Even without bs->job we have two systems of permissions:
op-blockers and BLK_PERM
3.3 We may want to run several jobs on one node one day
And finally, drop bs->job pointer itself. Hurrah!
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
All callers of bdrv_set_aio_context() are eliminated now, they have
moved to bdrv_try_set_aio_context() and related safe functions. Remove
bdrv_set_aio_context().
With this, we can now know that the .set_aio_ctx callback must be
present in bdrv_set_aio_context_ignore() because
bdrv_can_set_aio_context() would have returned false previously, so
instead of checking the condition, we can assert it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The mirror and commit block jobs use bdrv_set_aio_context() to move
their filter node into the right AioContext before hooking it up in the
graph. Similarly, bdrv_open_backing_file() explicitly moves the backing
file node into the right AioContext first.
This isn't necessary any more, they get automatically moved into the
right context now when attaching them.
However, in the case of bdrv_open_backing_file() with a node reference,
it's actually not only unnecessary, but even wrong: The unchecked
bdrv_set_aio_context() changes the AioContext of the child node even if
other parents require it to retain the old context. So this is not only
a simplification, but a bug fix, too.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1684342
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
A node should only be in a non-default AioContext if a user is attached
to it that requires this. When the last parent of a node is gone, it can
move back to the main AioContext.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
So far, we only made sure that updating the AioContext of a node
affected the whole subtree. However, if a node is newly attached to a
new parent, we also need to make sure that both the subtree of the node
and the parent are in the same AioContext. This tries to move the new
child node to the parent AioContext and returns an error if this isn't
possible.
BlockBackends now actually apply their AioContext to their root node.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This adds a new parameter to blk_new() which requires its callers to
declare from which AioContext this BlockBackend is going to be used (or
the locks of which AioContext need to be taken anyway).
The given context is only stored and kept up to date when changing
AioContexts. Actually applying the stored AioContext to the root node
is saved for another commit.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of just asserting that no requests are in flight in
bdrv_replace_node(), which is a requirement that most callers ignore, we
can just drain the source node right there. This fixes at least starting
a commit job while I/O is active on the backing chain, but probably
other callers, too.
Having requests in flight on the target node isn't a problem because the
target just gets new parents, but the call path of running requests
isn't modified. So we can just drop this assertion without a replacement.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1711643
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
A consequence of the previous patch is that bdrv_attach_child()
transfers the reference to child_bs from the caller to parent_bs,
which will drop it on bdrv_close() or when someone calls
bdrv_unref_child().
But this only happens when bdrv_attach_child() succeeds. If it fails
then the caller is responsible for dropping the reference to child_bs.
This patch makes bdrv_attach_child() take the reference also when
there is an error, freeing the caller for having to do it.
A similar situation happens with bdrv_root_attach_child(), so the
changes on this patch affect both functions.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 20dfb3d9ccec559cdd1a9690146abad5d204a186.1557754872.git.berto@igalia.com
[mreitz: Removed now superfluous BdrvChild * variable in
bdrv_open_child()]
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_unref_child() does the following things:
- Updates the child->bs->inherits_from pointer.
- Calls bdrv_detach_child() to remove the BdrvChild from bs->children.
- Calls bdrv_unref() to unref the child BlockDriverState.
When bdrv_unref_child() was introduced in commit 33a604075c it was not
used in bdrv_close() because the drivers that had additional children
(like quorum or blkverify) had already called bdrv_unref() on their
children during their own close functions.
This was changed later (in 0bd6e91a7e for quorum, in 3e586be0b2 for
blkverify) so there's no reason not to use bdrv_unref_child() in
bdrv_close() anymore.
After this there's also no need to remove bs->backing and bs->file
separately from the rest of the children, so bdrv_close() can be
simplified.
Now bdrv_close() unrefs all children (before this patch it was only
bs->file and bs->backing). As a result, none of the callers of
brvd_attach_child() should remove their reference to child_bs (because
this function effectively steals that reference). This patch updates a
couple of tests that were doing their own bdrv_unref().
Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 6d1d5feaa53aa1ab127adb73d605dc4503e3abd5.1557754872.git.berto@igalia.com
[mreitz: s/where/were/]
Signed-off-by: Max Reitz <mreitz@redhat.com>
This message does not make any sense when it appears as the response to
making an R/W node read-only. We should detect that case and emit a
different message, then.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
All block nodes and users in any connected component of the block graph
must be in the same AioContext, so changing the AioContext of one node
must not only change all of its children, but all of its parents (and
in turn their children etc.) as well.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of having two recursions, in bdrv_attach_aio_context() and in
bdrv_detach_aio_context(), just having one recursion is enough. Said
functions are only about a single node now.
It is important that the recursion doesn't happen between detaching and
attaching a context to the current node because the nested call will
drain the node, and draining with a NULL context would segfault.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Since commit b97511c7bc, there is no reason for block drivers any more
to call these functions (see the function comment in block_int.h). They
are now just internal helper functions for bdrv_set_aio_context()
and can be made static.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Eventually, we want to make sure that all parents and all children of a
node are in the same AioContext as the node itself. This means that
changing the AioContext may fail because one of the other involved
parties (e.g. a guest device that was configured with an iothread)
cannot allow switching to a different AioContext.
Introduce a set of functions that allow to first check whether all
involved nodes can switch to a new context and only then do the actual
switch. The check recursively covers children and parents.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Close involves flush that can be performed asynchronously and bs
must be protected from being referenced before it is deleted.
Signed-off-by: Anton Kuchin <antonkuchin@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_check_co_entry calls bdrv_co_check, which is a coroutine function.
Thus, it also needs to be marked as a coroutine.
Signed-off-by: Nikita Alekseev <n.alekseev2104@gmail.com>
Message-id: 20190401093051.16488-1-n.alekseev2104@gmail.com
Message-Id: <20190401093051.16488-1-n.alekseev2104@gmail.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
There is no need to check for this because all block drivers that have
children implement bdrv_child_perm and all callers already ensure that
bs->drv is set.
Furthermore, if this check would fail then the callers would end up
with uninitialized values for nperm and nshared.
This patch replaces the check with an assertion.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190404112953.4058-1-berto@igalia.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Even for block nodes with bs->drv == NULL, we can't just ignore a
bdrv_set_aio_context() call. Leaving the node in its old context can
mean that it's still in an iothread context in bdrv_close_all() during
shutdown, resulting in an attempted unlock of the AioContext lock which
we don't hold.
This is an example stack trace of a related crash:
#0 0x00007ffff59da57f in raise () at /lib64/libc.so.6
#1 0x00007ffff59c4895 in abort () at /lib64/libc.so.6
#2 0x0000555555b97b1e in error_exit (err=<optimized out>, msg=msg@entry=0x555555d386d0 <__func__.19059> "qemu_mutex_unlock_impl") at util/qemu-thread-posix.c:36
#3 0x0000555555b97f7f in qemu_mutex_unlock_impl (mutex=mutex@entry=0x5555568002f0, file=file@entry=0x555555d378df "util/async.c", line=line@entry=507) at util/qemu-thread-posix.c:97
#4 0x0000555555b92f55 in aio_context_release (ctx=ctx@entry=0x555556800290) at util/async.c:507
#5 0x0000555555b05cf8 in bdrv_prwv_co (child=child@entry=0x7fffc80012f0, offset=offset@entry=131072, qiov=qiov@entry=0x7fffffffd4f0, is_write=is_write@entry=true, flags=flags@entry=0)
at block/io.c:833
#6 0x0000555555b060a9 in bdrv_pwritev (qiov=0x7fffffffd4f0, offset=131072, child=0x7fffc80012f0) at block/io.c:990
#7 0x0000555555b060a9 in bdrv_pwrite (child=0x7fffc80012f0, offset=131072, buf=<optimized out>, bytes=<optimized out>) at block/io.c:990
#8 0x0000555555ae172b in qcow2_cache_entry_flush (bs=bs@entry=0x555556810680, c=c@entry=0x5555568cc740, i=i@entry=0) at block/qcow2-cache.c:51
#9 0x0000555555ae18dd in qcow2_cache_write (bs=bs@entry=0x555556810680, c=0x5555568cc740) at block/qcow2-cache.c:248
#10 0x0000555555ae15de in qcow2_cache_flush (bs=0x555556810680, c=<optimized out>) at block/qcow2-cache.c:259
#11 0x0000555555ae16b1 in qcow2_cache_flush_dependency (c=0x5555568a1700, c=0x5555568a1700, bs=0x555556810680) at block/qcow2-cache.c:194
#12 0x0000555555ae16b1 in qcow2_cache_entry_flush (bs=bs@entry=0x555556810680, c=c@entry=0x5555568a1700, i=i@entry=0) at block/qcow2-cache.c:194
#13 0x0000555555ae18dd in qcow2_cache_write (bs=bs@entry=0x555556810680, c=0x5555568a1700) at block/qcow2-cache.c:248
#14 0x0000555555ae15de in qcow2_cache_flush (bs=bs@entry=0x555556810680, c=<optimized out>) at block/qcow2-cache.c:259
#15 0x0000555555ad242c in qcow2_inactivate (bs=bs@entry=0x555556810680) at block/qcow2.c:2124
#16 0x0000555555ad2590 in qcow2_close (bs=0x555556810680) at block/qcow2.c:2153
#17 0x0000555555ab0c62 in bdrv_close (bs=0x555556810680) at block.c:3358
#18 0x0000555555ab0c62 in bdrv_delete (bs=0x555556810680) at block.c:3542
#19 0x0000555555ab0c62 in bdrv_unref (bs=0x555556810680) at block.c:4598
#20 0x0000555555af4d72 in blk_remove_bs (blk=blk@entry=0x5555568103d0) at block/block-backend.c:785
#21 0x0000555555af4dbb in blk_remove_all_bs () at block/block-backend.c:483
#22 0x0000555555aae02f in bdrv_close_all () at block.c:3412
#23 0x00005555557f9796 in main (argc=<optimized out>, argv=<optimized out>, envp=<optimized out>) at vl.c:4776
The reproducer I used is a qcow2 image on gluster volume, where the
virtual disk size (4 GB) is larger than the gluster volume size (64M),
so we can easily trigger an ENOSPC. This backend is assigned to a
virtio-blk device using an iothread, and then from the guest a
'dd if=/dev/zero of=/dev/vda bs=1G count=1' causes the VM to stop
because of an I/O error. qemu_gluster_co_flush_to_disk() sets
bs->drv = NULL on error, so when virtio-blk stops the dataplane, the
block nodes stay in the iothread AioContext. A 'quit' monitor command
issued from this paused state crashes the process.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1631227
Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
When bdrv_temp_snapshot_options() is called for snapshot=on, the
'discard' option in the options QDict hasn't been parsed and merged into
the flags yet. So copy the dict entry to make sure that the temporary
overlay enables discard when it was requested for the drive.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
All three functions that handle the BdrvChild.frozen attribute walk
the backing chain from 'bs' to 'base' and stop either when 'base' is
found or at the end of the chain if 'base' is NULL.
However if 'base' is not found then the functions return without
errors as if it was NULL.
This is wrong: if the caller passed an incorrect parameter that means
that there is a bug in the code.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Coverity doesn't like that the return value of bdrv_check_update_perm()
stays unused only in this place (CID 1399710).
Even if checking local_err should be equivalent to checking ret < 0,
let's switch to using the return value to be more consistent (and in
case of a bug somewhere down the call chain, forgetting to assign errp
is more likely than returning 0 for an error case).
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
bdrv_reopen_prepare() receives a BDRVReopenState with (among other
things) a new set of options to be applied to that BlockDriverState.
If an option is missing then it means that we want to reset it to its
default value rather than keep the previous one. This way the state
of the block device after being reopened is comparable to that of a
device added with "blockdev-add" using the same set of options.
Not all options from all drivers can be changed this way, however.
If the user attempts to reset an immutable option to its default value
using this method then we must forbid it.
This new function takes a BlockDriverState and a new set of options
and checks if there's any option that was previously set but is
missing from the new set of options.
If the option is present in both sets we don't need to check that they
have the same value. The loop at the end of bdrv_reopen_prepare()
already takes care of that.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch allows the user to change the backing file of an image that
is being reopened. Here's what it does:
- In bdrv_reopen_prepare(): check that the value of 'backing' points
to an existing node or is null. If it points to an existing node it
also needs to make sure that replacing the backing file will not
create a cycle in the node graph (i.e. you cannot reach the parent
from the new backing file).
- In bdrv_reopen_commit(): perform the actual node replacement by
calling bdrv_set_backing_hd().
There may be temporary implicit nodes between a BDS and its backing
file (e.g. a commit filter node). In these cases bdrv_reopen_prepare()
looks for the real (non-implicit) backing file and requires that the
'backing' option points to it. Replacing or detaching a backing file
is forbidden if there are implicit nodes in the middle.
Although x-blockdev-reopen is meant to be used like blockdev-add,
there's an important thing that must be taken into account: the only
way to set a new backing file is by using a reference to an existing
node (previously added with e.g. blockdev-add). If 'backing' contains
a dictionary with a new set of options ({"driver": "qcow2", "file": {
... }}) then it is interpreted that the _existing_ backing file must
be reopened with those options.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Of all options of type BlockdevRef used to specify children in
BlockdevOptions, 'backing' is the only one that is optional.
For "x-blockdev-reopen" we want that if an option is omitted then it
must be reset to its default value. The default value of 'backing'
means that QEMU opens the backing file specified in the image
metadata, but this is not something that we want to support for the
reopen operation.
Because of this the 'backing' option has to be specified during
reopen, pointing to the existing backing file if we want to keep it,
or pointing to a different one (or NULL) if we want to replace it (to
be implemented in a subsequent patch).
In order to simplify things a bit and not to require that the user
passes the 'backing' option to every single block device even when
it's clearly not necessary, this patch allows omitting this option if
the block device being reopened doesn't have a backing file attached
_and_ no default backing file is specified in the image metadata.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Children in QMP are specified with BlockdevRef / BlockdevRefOrNull,
which can contain a set of child options, a child reference, or
NULL. In optional attributes like "backing" it can also be missing.
Only the first case (set of child options) is being handled properly
by bdrv_reopen_queue(). This patch deals with all the others.
Here's how these cases should be handled when bdrv_reopen_queue() is
deciding what to do with each child of a BlockDriverState:
1) Set of child options: if the child was implicitly created (i.e
inherits_from points to the parent) then the options are removed
from the parent's options QDict and are passed to the child with
a recursive bdrv_reopen_queue() call. This case was already
working fine.
2) Child reference: there's two possibilites here.
2a) Reference to the current child: if the child was implicitly
created then it is put in the reopen queue, keeping its
current set of options (since this was a child reference
there was no way to specify a different set of options).
If the child is not implicit then it keeps its current set
of options but it is not reopened (and therefore does not
inherit any new option from the parent).
2b) Reference to a different BDS: the current child is not put
in the reopen queue at all. Passing a reference to a
different BDS can be used to replace a child, although at
the moment no driver implements this, so it results in an
error. In any case, the current child is not going to be
reopened (and might in fact disappear if it's replaced)
3) NULL: This is similar to (2b). Although no driver allows this
yet it can be used to detach the current child so it should not
be put in the reopen queue.
4) Missing option: at the moment "backing" is the only case where
this can happen. With "blockdev-add", leaving "backing" out
means that the default backing file is opened. We don't want to
open a new image during reopen, so we require that "backing" is
always present. We'll relax this requirement a bit in the next
patch. If keep_old_opts is true and "backing" is missing then
this behaves like 2a (the current child is reopened).
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The bdrv_reopen_queue() function is used to create a queue with
the BDSs that are going to be reopened and their new options. Once
the queue is ready bdrv_reopen_multiple() is called to perform the
operation.
The original options from each one of the BDSs are kept, with the new
options passed to bdrv_reopen_queue() applied on top of them.
For "x-blockdev-reopen" we want a function that behaves much like
"blockdev-add". We want to ignore the previous set of options so that
only the ones actually specified by the user are applied, with the
rest having their default values.
One of the things that we need is a way to tell bdrv_reopen_queue()
whether we want to keep the old set of options or not, and that's what
this patch does. All current callers are setting this new parameter to
true and x-blockdev-reopen will set it to false.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Our permission system is useful to define what operations are allowed
on a certain block node and includes things like BLK_PERM_WRITE or
BLK_PERM_RESIZE among others.
One of the permissions is BLK_PERM_GRAPH_MOD which allows "changing
the node that this BdrvChild points to". The exact meaning of this has
never been very clear, but it can be understood as "change any of the
links connected to the node". This can be used to prevent changing a
backing link, but it's too coarse.
This patch adds a new 'frozen' attribute to BdrvChild, which forbids
detaching the link from the node it points to, and new API to freeze
and unfreeze a backing chain.
After this change a few functions can fail, so they need additional
checks.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Using a different read-only setting for bs->open_flags than for the
flags to the driver's open function is just inconsistent and a bad idea.
After this patch, the temporary snapshot keeps being opened read-only if
read-only=on,snapshot=on is passed.
If we wanted to change this behaviour to make only the orginal image
file read-only, but the temporary overlay read-write (as the comment in
the removed code suggests), that change would have to be made in
bdrv_temp_snapshot_options() (where the comment suggests otherwise).
Addressing this inconsistency before introducing dynamic auto-read-only
is important because otherwise we would immediately try to reopen the
temporary overlay even though the file is already unlinked.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The way that reopen interacts with permission changes has one big
problem: Both operations are recursive, and the permissions are changes
for each node in the reopen queue.
For a simple graph that consists just of parent and child,
.bdrv_check_perm will be called twice for the child, once recursively
when adjusting the permissions of parent, and once again when the child
itself is reopened.
Even worse, the first .bdrv_check_perm call happens before
.bdrv_reopen_prepare was called for the child and the second one is
called afterwards.
Making sure that .bdrv_check_perm (and the other permission callbacks)
are called only once is hard. We can cope with multiple calls right now,
but as soon as file-posix gets a dynamic auto-read-only that may need to
open a new file descriptor, we get the additional requirement that all
of them are after the .bdrv_reopen_prepare call.
So reorder things in bdrv_reopen_multiple() to first call
.bdrv_reopen_prepare for all involved nodes and only then adjust
permissions.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_iterate_format (which is currently only used for printing out the
formats supported by the block layer) doesn't take format whitelisting
into account.
This creates a problem for tests: they enumerate supported formats to
decide which tests to enable, but then discover that QEMU doesn't let
them actually use some of those formats.
To avoid that, exclude formats that are not whitelisted from
enumeration, if whitelisting is in use. Since we have separate
whitelists for r/w and r/o, take this a parameter to
bdrv_iterate_format, and print two lists of supported formats (r/w and
r/o) in main qemu.
Signed-off-by: Roman Kagan <rkagan@virtuozzo.com>
Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When BDSs are created by qemu itself (e.g. as filters in block jobs),
they may not have a "driver" option in their options QDict. When
generating a json:{} filename, however, it must always be present.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-31-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
If a format BDS's file BDS is in turn a format BDS, we cannot simply use
the same filename, because when opening a BDS tree based on a filename
alone, qemu will create only one format node on top of one protocol node
(disregarding a potential backing file).
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-26-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Currently, BlockDriver.bdrv_refresh_filename() is supposed to both
refresh the filename (BDS.exact_filename) and set BDS.full_open_options.
Now that we have generic code in the central bdrv_refresh_filename() for
creating BDS.full_open_options, we can drop the latter part from all
BlockDriver.bdrv_refresh_filename() implementations.
This also means that we can drop all of the existing default code for
this from the global bdrv_refresh_filename() itself.
Furthermore, we now have to call BlockDriver.bdrv_refresh_filename()
after having set BDS.full_open_options, because the block driver's
implementation should now be allowed to depend on BDS.full_open_options
being set correctly.
Finally, with this patch we can drop the @options parameter from
BlockDriver.bdrv_refresh_filename(); also, add a comment on this
function's purpose in block/block_int.h while touching its interface.
This completely obsoletes blklogwrite's implementation of
.bdrv_refresh_filename().
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190201192935.18394-25-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Instead of having every block driver which implements
bdrv_refresh_filename() copy all of the strong runtime options over to
bs->full_open_options, implement this process generically in
bdrv_refresh_filename().
This patch only adds this new generic implementation, it does not remove
the old functionality. This is done in a follow-up patch.
With this patch, some superfluous information (that should never have
been there) may be removed from some JSON filenames, as can be seen in
the change to iotests 110's and 228's reference outputs.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190201192935.18394-24-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_get_full_backing_filename_from_filename() breaks down when it comes
to JSON filenames. Using bdrv_dirname() as the basis is better because
since we have BDS, we can descend through the BDS tree to the protocol
layer, which gives us a greater probability of finding a non-JSON name;
also, bdrv_dirname() is more correct as it allows block drivers to
override the generation of that directory name in a protocol-specific
way.
We still need to keep bdrv_get_full_backing_filename_from_filename(),
though, because it has valid callers which need it during image creation
when no BDS is available yet.
This makes a test case in qemu-iotest 110, which was supposed to fail,
work. That is actually good, but we need to change the reference output
(and the comment in 110) accordingly.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-20-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This function may be implemented by block drivers to derive a directory
name from a BDS. Concatenating this g_free()-able string with a relative
filename must result in a valid (not necessarily existing) filename, so
this is a function that should generally be not implemented by format
drivers, because this is protocol-specific.
If a BDS's driver does not implement this function, bdrv_dirname() will
fall through to the BDS's file if it exists. If it does not, the
exact_filename field will be used to generate a directory name.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-15-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_find_backing_image() should use bdrv_get_full_backing_filename() or
bdrv_make_absolute_filename() instead of trying to do what those
functions do by itself.
path_combine_deprecated() can now be dropped, so let's do that.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-14-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
This is a general function for making a filename that is relative to a
certain BDS absolute.
It calls bdrv_get_full_backing_filename_from_filename() for now, but
that will be changed in a follow-up patch.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-13-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Make bdrv_get_full_backing_filename() return an allocated string instead
of placing the result in a caller-provided buffer.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-12-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Make bdrv_get_full_backing_filename_from_filename() return an allocated
string instead of placing the result in a caller-provided buffer.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190201192935.18394-11-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Besides being safe for arbitrary path lengths, after some follow-up
patches all callers will want a freshly allocated buffer anyway.
In the meantime, path_combine_deprecated() is added which has the same
interface as path_combine() had before this patch. All callers to that
function will be converted in follow-up patches.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-id: 20190201192935.18394-10-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Basically, bdrv_refresh_filename() should respect all children of a
BlockDriverState. However, generally those children are driver-specific,
so this function cannot handle the general case. On the other hand,
there are only few drivers which use other children than @file and
@backing (that being vmdk, quorum, and blkverify).
Most block drivers only use @file and/or @backing (if they use any
children at all). Both can be implemented directly in
bdrv_refresh_filename.
The user overriding the file's filename is already handled, however, the
user overriding the backing file is not. If this is done, opening the
BDS with the plain filename of its file will not be correct, so we may
not set bs->exact_filename in that case.
iotest 051 contains test cases for overriding the backing file, and so
its output changes with this patch applied.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-6-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
If the backing file is overridden, this most probably does change the
guest-visible data of a BDS. Therefore, we will need to consider this
in bdrv_refresh_filename().
To see whether it has been overridden, we might want to compare
bs->backing_file and bs->backing->bs->filename. However,
bs->backing_file is changed by bdrv_set_backing_hd() (which is just used
to change the backing child at runtime, without modifying the image
header), so bs->backing_file most of the time simply contains a copy of
bs->backing->bs->filename anyway, so it is useless for such a
comparison.
This patch adds an auto_backing_file BDS field which contains the
backing file path as indicated by the image header, which is not changed
by bdrv_set_backing_hd().
Because of bdrv_refresh_filename() magic, however, a BDS's filename may
differ from what has been specified during bdrv_open(). Then, the
comparison between bs->auto_backing_file and bs->backing->bs->filename
may fail even though bs->backing was opened from bs->auto_backing_file.
To mitigate this, we can copy the real BDS's filename (after the whole
bdrv_open() and bdrv_refresh_filename() process) into
bs->auto_backing_file, if we know the former has been opened based on
the latter. This is only possible if no options modifying the backing
file's behavior have been specified, though. To simplify things, this
patch only copies the filename from the backing file if no options have
been specified for it at all.
Furthermore, there are cases where an overlay is created by qemu which
already contains a BDS's filename (e.g. in blockdev-snapshot-sync). We
do not need to worry about updating the overlay's bs->auto_backing_file
there, because we actually wrote a post-bdrv_refresh_filename() filename
into the image header.
So all in all, there will be false negatives where (as of a future
patch) bdrv_refresh_filename() will assume that the backing file differs
from what was specified in the image header, even though it really does
not. However, these cases should be limited to where (1) the user
actually did override something in the backing chain (e.g. by specifying
options for the backing file), or (2) the user executed a QMP command to
change some node's backing file (e.g. change-backing-file or
block-commit with @backing-file given) where the given filename does not
happen to coincide with qemu's idea of the backing BDS's filename.
Then again, (1) really is limited to -drive. With -blockdev or
blockdev-add, you have to adhere to the schema, so a user cannot give
partial "unimportant" options (e.g. by just setting backing.node-name
and leaving the rest to the image header). Therefore, trying to fix
this would mean trying to fix something for -drive only.
To improve on (2), we would need a full infrastructure to "canonicalize"
an arbitrary filename (+ options), so it can be compared against
another. That seems a bit over the top, considering that filenames
nowadays are there mostly for the user's entertainment.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-5-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_refresh_filename() should simply skip all implicit nodes. They are
supposed to be invisible to the user, so they should not appear in
filename information.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-4-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_refresh_filename() should invoke itself recursively on all
children, not just on file.
With that change, we can remove the manual invocations in blkverify,
quorum, commit, mirror, and blklogwrites.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20190201192935.18394-3-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Before this patch, bdrv_refresh_filename() is used in a pushing manner:
Whenever the BDS graph is modified, the parents of the modified edges
are supposed to be updated (recursively upwards). However, that is
nonviable, considering that we want child changes not to concern
parents.
Also, in the long run we want a pull model anyway: Here, we would have a
bdrv_filename() function which returns a BDS's filename, freshly
constructed.
This patch is an intermediate step. It adds bdrv_refresh_filename()
calls before every place a BDS.filename value is used. The only
exceptions are protocol drivers that use their own filename, which
clearly would not profit from refreshing that filename before.
Also, bdrv_get_encrypted_filename() is removed along the way (as a user
of BDS.filename), since it is completely unused.
In turn, all of the calls to bdrv_refresh_filename() before this patch
are removed, because we no longer have to call this function on graph
changes.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20190201192935.18394-2-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_check_perm in it's recursion checks each node in context of new
permissions for one parent, because of nature of DFS. It works well,
while children subgraph of top-most updated node is a tree, i.e. it
doesn't have any kind of loops. But if we have a loop (not oriented,
of course), i.e. we have two different ways from top-node to some
child-node, then bdrv_check_perm will do wrong thing:
top
| \
| |
v v
A B
| |
v v
node
It will once check new permissions of node in context of new A
permissions and old B permissions and once visa-versa. It's a wrong way
and may lead to corruption of permission system. We may start with
no-permissions and all-shared for both A->node and B->node relations
and finish up with non shared write permission for both ways.
The following commit will add a test, which shows this bug.
To fix this situation, let's really set BdrvChild permissions during
bdrv_check_perm procedure. And we are happy here, as check-perm is
already written in transaction manner, so we just need to restore
backed-up permissions in _abort.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
As it already said in the comment, we don't want to create loops in
parent->child relations. So, when we try to append @to to @c, we should
check that @c is not in @to children subtree, and we should check it
recursively, not only the first level. The patch provides BFS-based
search, to check the relations.
This is needed for further fleecing-hook filter usage: we need to
append it to source, when the hook is already a parent of target, and
source may be in a backing chain of target (fleecing-scheme). So, on
appending, the hook should not became a child (direct or through
children subtree) of the target.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that bdrv_set_aio_context() works inside drained sections, it can
also use the real drain function instead of open coding something
similar.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When a drained node changes its AioContext, we need to move its
aio_disable_external() to the new context, too.
Without this fix, drain_end will try to reenable the new context, which
has never been disabled, so an assertion failure is triggered.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
The explicit aio_poll() call in bdrv_set_aio_context() was added in
commit c2b6428d38 as a workaround for bdrv_drain() failing to achieve
to actually quiesce everything (specifically the NBD client code to
switch AioContext).
Now that the NBD client has been fixed to complete this operation during
bdrv_drain(), we don't need the workaround any more.
It was wrong anyway: aio_poll() must always be run in the home thread of
the AioContext.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Adds a fast path on aio context setting preventing
unnecessary context setting routine.
Also, it prevents issues with cyclic walk of child
bds-es appeared because of registering aio walking
notifiers:
Call stack:
0 __GI_raise
1 __GI_abort
2 __assert_fail_base
3 __GI___assert_fail
4 bdrv_detach_aio_context (bs=0x55f54d65c000) <<<
5 bdrv_detach_aio_context (bs=0x55f54fc8a800)
6 bdrv_set_aio_context (bs=0x55f54fc8a800, ...)
7 block_job_attached_aio_context
8 bdrv_attach_aio_context (bs=0x55f54d65c000, ...) <<<
9 bdrv_set_aio_context (bs=0x55f54d65c000)
10 blk_set_aio_context
11 virtio_blk_data_plane_stop
12 virtio_bus_stop_ioeventfd
13 virtio_vmstate_change
14 vm_state_notify (running=0, state=RUN_STATE_SHUTDOWN)
15 do_vm_stop (state=RUN_STATE_SHUTDOWN, send_stop=true)
16 vm_stop (state=RUN_STATE_SHUTDOWN)
17 main_loop_should_exit
18 main_loop
19 main
This can happen because of "new" context attachment to VM disk bds.
When attaching a new context the corresponding aio context handler is
called for each of aio_notifiers registered on the VM disk bds context.
Among those handlers, there is the block_job_attached_aio_context handler
which sets a new aio context for the block job bds. When doing so,
the old context is detached from all the block job bds children and one of
them is the VM disk bds, serving as backing store for the blockjob bds,
although the VM disk bds is actually the initializer of that process.
Since the VM disk bds is protected with walking_aio_notifiers flag
from double processing in recursive calls, the assert fires.
Signed-off-by: Denis Plotnikov <dplotnikov@virtuozzo.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Inform a user in case qcow2_get_specific_info fails to obtain
QCOW2 image specific information. This patch is preliminary to
the one "qcow2: Add list of bitmaps to ImageInfoSpecificQCow2".
Signed-off-by: Andrey Shinkevich <andrey.shinkevich@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <1549638368-530182-2-git-send-email-andrey.shinkevich@virtuozzo.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
bdrv_co_invalidate_cache() clears the BDRV_O_INACTIVE flag before
actually activating a node so that the correct permissions etc. are
taken. In case of errors, the flag must be restored so that the next
call to bdrv_co_invalidate_cache() retries activation.
Restoring the flag was missing in the error path for a failed
parent->role->activate() call. The consequence is that this attempt to
activate all images correctly fails because we still set errp, however
on the next attempt BDRV_O_INACTIVE is already clear, so we return
success without actually retrying the failed action.
An example where this is observable in practice is migration to a QEMU
instance that has a raw format block node attached to a guest device
with share-rw=off (the default) while another process holds
BLK_PERM_WRITE for the same image. In this case, all activation steps
before parent->role->activate() succeed because raw can tolerate other
writers to the image. Only the parent callback (in particular
blk_root_activate()) tries to implement the share-rw=on property and
requests exclusive write permissions. This fails when the migration
completes and correctly displays an error. However, a manual 'cont' will
incorrectly resume the VM without calling blk_root_activate() again.
This case is described in more detail in the following bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=1531888
Fix this by correctly restoring the BDRV_O_INACTIVE flag in the error
path.
Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Tested-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
If QEMU was configured with a driver in --block-drv-ro-whitelist, trying
to use that driver read-write resulted in an error message even if
auto-read-only=on was set.
Consider auto-read-only=on for the whitelist checking and use it to
automatically degrade to read-only for block drivers on the read-only
whitelist.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
In the block layer, synchronous APIs are often implemented by creating a
coroutine that calls the asynchronous coroutine-based implementation and
then waiting for completion with BDRV_POLL_WHILE().
For this to work with iothreads (more specifically, when the synchronous
API is called in a thread that is not the home thread of the block
device, so that the coroutine will run in a different thread), we must
make sure to call aio_wait_kick() at the end of the operation. Many
places are missing this, so that BDRV_POLL_WHILE() keeps hanging even if
the condition has long become false.
Note that bdrv_dec_in_flight() involves an aio_wait_kick() call. This
corresponds to the BDRV_POLL_WHILE() in the drain functions, but it is
generally not enough for most other operations because they haven't set
the return value in the coroutine entry stub yet. To avoid race
conditions there, we need to kick after setting the return value.
The race window is small enough that the problem doesn't usually surface
in the common path. However, it does surface and causes easily
reproducible hangs if the operation can return early before even calling
bdrv_inc/dec_in_flight, which many of them do (trivial error or no-op
success paths).
The bug in bdrv_truncate(), bdrv_check() and bdrv_invalidate_cache() is
slightly different: These functions even neglected to schedule the
coroutine in the home thread of the node. This avoids the hang, but is
obviously wrong, too. Fix those to schedule the coroutine in the right
AioContext in addition to adding aio_wait_kick() calls.
Cc: qemu-stable@nongnu.org
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Add a new command, returning block nodes (and their users) graph.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-id: 20181221170909.25584-2-vsementsov@virtuozzo.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
Towards the end of bdrv_reopen_queue_child(), before starting to
process the children, the update_flags_from_options() function is
called in order to have BDRVReopenState.flags in sync with the options
from the QDict.
This is necessary because during the reopen process flags must be
updated for all nodes in the queue so bdrv_is_writable_after_reopen()
and the permission checks work correctly.
Because of that, calling update_flags_from_options() again in
bdrv_reopen_prepare() doesn't really change the flags (they are
already up-to-date). But we need to call it in order to remove those
options from QemuOpts and that way indicate that they have been
processed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This function takes four options (cache.direct, cache.no-flush,
read-only and auto-read-only) from a QemuOpts object and updates the
flags accordingly.
If any of those options is not set (because it was missing from the
original QDict or because it had an invalid value) then the function
aborts with a failed assertion:
$ qemu-io -c 'reopen -o read-only=foo' hd.qcow2
block.c:1126: update_flags_from_options: Assertion `qemu_opt_find(opts, BDRV_OPT_CACHE_DIRECT)' failed.
Aborted
This assertion is unnecessary, and it forces any caller of
bdrv_reopen() to pass all the aforementioned four options. This may
have made sense in order to remove ambiguity when bdrv_reopen() was
taking both flags and options, but that's not the case anymore.
It's also unnecessary if we want to validate the option values,
because bdrv_reopen_prepare() already takes care of that, as we can
see if we remove the assertions:
$ qemu-io -c 'reopen -o read-only=foo' hd.qcow2
Parameter 'read-only' expects 'on' or 'off'
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that all callers are passing the new options using the QDict we no
longer need the 'flags' parameter.
This patch makes the following changes:
1) The update_options_from_flags() call is no longer necessary
so it can be removed.
2) The update_flags_from_options() call is now used in all cases,
and is moved down a few lines so it happens after the options
QDict contains the final set of values.
3) The flags parameter is removed. Now the flags are initialized
using the current value (for the top-level node) or the parent
flags (after inherit_options()). In both cases the initial
values are updated to reflect the new options in the QDict. This
happens in bdrv_reopen_queue_child() (as explained above) and in
bdrv_reopen_prepare().
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that all callers are passing all flag changes as QDict options,
the flags parameter is no longer necessary, so we can get rid of it.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
No one is using this function anymore, so we can safely remove it.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This patch replaces the bdrv_reopen() calls that set and remove the
BDRV_O_RDWR flag with the new bdrv_reopen_set_read_only() function.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Most callers of bdrv_reopen() only use it to switch a BlockDriverState
between read-only and read-write, so this patch adds a new function
that does just that.
We also want to get rid of the flags parameter in the bdrv_reopen()
API, so this function sets the "read-only" option and passes the
original flags (which will then be updated in bdrv_reopen_prepare()).
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_child_cb_inactivate() asserts that parents are already inactive
when children get inactivated. This precondition is necessary because
parents could still issue requests in their inactivation code.
When block nodes are created individually with -blockdev, all of them
are monitor owned and will be returned by bdrv_next() in an undefined
order (in practice, in the order of their creation, which is usually
children before parents), which obviously fails the assertion:
qemu: block.c:899: bdrv_child_cb_inactivate: Assertion `bs->open_flags & BDRV_O_INACTIVE' failed.
This patch fixes the ordering by skipping nodes with still active
parents in bdrv_inactivate_recurse() because we know that they will be
covered by recursion when the last active parent becomes inactive.
With the correct parents-before-children ordering, we also got rid of
the reason why commit aad0b7a0bf introduced two passes, so we can go
back to a single-pass recursion. This is necessary so we can rely on the
BDRV_O_INACTIVE flag to skip nodes with active parents (the flag used
to be set only in pass 2, so we would always skip non-root nodes in
pass 1 because all parents would still be considered active; setting the
flag in pass 1 would mean, that we never skip anything in pass 2 because
all parents are already considered inactive).
Because of the change to single pass, this patch is best reviewed with
whitespace changes ignored.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
The previous patch fixed the inherits_from pointer after block-stream,
and this one does the same for block-commit.
When block-commit finishes and the 'top' node is not the topmost one
from the backing chain then all nodes above 'base' up to and including
'top' are removed from the chain.
The bdrv_drop_intermediate() call converts a chain like this one:
base <- intermediate <- top <- active
into this one:
base <- active
In a simple scenario each backing file from the first chain has the
inherits_from attribute pointing to its parent. This means that
reopening 'active' will recursively reopen all its children, whose
options can be changed in the process.
However after the 'block-commit' call base.inherits_from is NULL and
the chain is broken, so 'base' does not inherit from 'active' and will
not be reopened automatically:
$ qemu-img create -f qcow2 hd0.qcow2 1M
$ qemu-img create -f qcow2 -b hd0.qcow2 hd1.qcow2
$ qemu-img create -f qcow2 -b hd1.qcow2 hd2.qcow2
$ $QEMU -drive if=none,file=hd2.qcow2
{ 'execute': 'block-commit',
'arguments': {
'device': 'none0',
'top': 'hd1.qcow2' } }
{ 'execute': 'human-monitor-command',
'arguments': {
'command-line':
'qemu-io none0 "reopen -o backing.l2-cache-size=2M"' } }
{ "return": "Cannot change the option 'backing.l2-cache-size'\r\n"}
This patch updates base.inherits_from in this scenario, and adds a
test case.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When a BlockDriverState's child is opened (be it a backing file, the
protocol layer, or any other) inherits_from is set to point to the
parent node. Children opened separately and then attached to a parent
don't have this pointer set.
bdrv_reopen_queue_child() uses this to determine whether a node's
children must also be reopened inheriting the options from the parent
or not. If inherits_from points to the parent then the child is
reopened and its options can be changed, like in this example:
$ qemu-img create -f qcow2 hd0.qcow2 1M
$ qemu-img create -f qcow2 hd1.qcow2 1M
$ $QEMU -drive if=none,node-name=hd0,file=hd0.qcow2,\
backing.driver=qcow2,backing.file.filename=hd1.qcow2
(qemu) qemu-io hd0 "reopen -o backing.l2-cache-size=2M"
If the child does not inherit from the parent then it does not get
reopened and its options cannot be changed:
$ $QEMU -drive if=none,node-name=hd1,file=hd1.qcow2
-drive if=none,node-name=hd0,file=hd0.qcow2,backing=hd1
(qemu) qemu-io hd0 "reopen -o backing.l2-cache-size=2M"
Cannot change the option 'backing.l2-cache-size'
If a disk image has a chain of backing files then all of them are also
connected through their inherits_from pointers (i.e. it's possible to
walk the chain in reverse order from base to top).
However this is broken if the intermediate nodes are removed using
e.g. block-stream because the inherits_from pointer from the base node
becomes NULL:
$ qemu-img create -f qcow2 hd0.qcow2 1M
$ qemu-img create -f qcow2 -b hd0.qcow2 hd1.qcow2
$ qemu-img create -f qcow2 -b hd1.qcow2 hd2.qcow2
$ $QEMU -drive if=none,file=hd2.qcow2
(qemu) qemu-io none0 "reopen -o backing.l2-cache-size=2M"
(qemu) block_stream none0 0 hd0.qcow2
(qemu) qemu-io none0 "reopen -o backing.l2-cache-size=2M"
Cannot change the option 'backing.l2-cache-size'
This patch updates the inherits_from pointer if the intermediate nodes
of a backing chain are removed using bdrv_set_backing_hd(), and adds a
test case for this scenario.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit e35bdc123a added the auto-read-only option and the
code to update its corresponding flag in update_flags_from_options(),
but forgot to clear the flag if auto-read-only is false.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reported-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_reopen_multiple() does not invoke bdrv_reopen_abort() for the
element of the reopen queue for which bdrv_reopen_prepare() failed,
because it assumes that the prepare function will have rolled back all
changes already.
However, bdrv_reopen_prepare() does not do this in every case: It may
notice an error after BlockDriver.bdrv_reopen_prepare() succeeded, and
it will not invoke BlockDriver.bdrv_reopen_abort() then; and neither
will bdrv_reopen_multiple(), as explained above.
This is wrong because we must always call .bdrv_reopen_commit() or
.bdrv_reopen_abort() after .bdrv_reopen_prepare() has succeeded.
Otherwise, the block driver has no chance to undo what it has done in
its implementation of .bdrv_reopen_prepare().
To fix this, bdrv_reopen_prepare() has to call .bdrv_reopen_abort() if
it wants to return an error after .bdrv_reopen_prepare() has succeeded.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Some block drivers have traditionally changed their node to read-only
mode without asking the user. This behaviour has been marked deprecated
since 2.11, expecting users to provide an explicit read-only=on option.
Now that we have auto-read-only=on, enable these drivers to make use of
the option.
This is the only use of bdrv_set_read_only(), so we can make it a bit
more specific and turn it into a bdrv_apply_auto_read_only() that is
more convenient for drivers to use.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
If a management application builds the block graph node by node, the
protocol layer doesn't inherit its read-only option from the format
layer any more, so it must be set explicitly.
Backing files should work on read-only storage, but at the same time, a
block job like commit should be able to reopen them read-write if they
are on read-write storage. However, without option inheritance, reopen
only changes the read-only option for the root node (typically the
format layer), but not the protocol layer, so reopening fails (the
format layer wants to get write permissions, but the protocol layer is
still read-only).
A simple workaround for the problem in the management tool would be to
open the protocol layer always read-write and to make only the format
layer read-only for backing files. However, sometimes the file is
actually stored on read-only storage and we don't know whether the image
can be opened read-write (for example, for NBD it depends on the server
we're trying to connect to). This adds an option that makes QEMU try to
open the image read-write, but allows it to degrade to a read-only mode
without returning an error.
The documentation for this option is consciously phrased in a way that
allows QEMU to switch to a better model eventually: Instead of trying
when the image is first opened, making the read-only flag dynamic and
changing it automatically whenever the first BLK_PERM_WRITE user is
attached or the last one is detached would be much more useful
behaviour.
Unfortunately, this more useful behaviour is also a lot harder to
implement, and libvirt needs a solution now before it can switch to
-blockdev, so let's start with this easier approach for now.
Instead of adding a new auto-read-only option, turning the existing
read-only into an enum (with a bool alternate for compatibility) was
considered, but it complicated the implementation to the point that it
didn't seem to be worth it.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
To fully change the read-only state of a node, we must not only change
bs->read_only, but also update bs->open_flags.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
This patch aims to bring the following behavior:
1. We don't load bitmaps, when started in inactive mode. It's the case
of incoming migration. In this case we wait for bitmaps migration
through migration channel (if 'dirty-bitmaps' capability is enabled) or
for invalidation (to load bitmaps from the image).
2. We don't remove persistent bitmaps on inactivation. Instead, we only
remove bitmaps after storing. This is the only way to restore bitmaps,
if we decided to resume source after [failed] migration with
'dirty-bitmaps' capability enabled (which means, that bitmaps were not
stored).
3. We load bitmaps on open and any invalidation, it's ok for all cases:
- normal open
- migration target invalidation with dirty-bitmaps capability
(bitmaps are migrating through migration channel, the are not
stored, so they should have IN_USE flag set and will be skipped
when loading. However, it would fail if bitmaps are read-only[1])
- migration target invalidation without dirty-bitmaps capability
(normal load of the bitmaps, if migrated with shared storage)
- source invalidation with dirty-bitmaps capability
(skip because IN_USE)
- source invalidation without dirty-bitmaps capability
(bitmaps were dropped, reload them)
[1]: to accurately handle this, migration of read-only bitmaps is
explicitly forbidden in this patch.
New mechanism for not storing bitmaps when migrate with dirty-bitmaps
capability is introduced: migration filed in BdrvDirtyBitmap.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Signed-off-by: John Snow <jsnow@redhat.com>
bdrv_img_create() takes an Error ** argument and uses it in the
conventional way, except for one place: when qemu_opts_do_parse()
fails, it first reports its error to stderr or the HMP monitor with
error_report_err(), then error_setg()'s a generic error.
When the caller reports that second error similarly, this produces two
consecutive error messages on stderr or the HMP monitor.
When the caller does something else with it, such as send it via QMP,
the first error still goes to stderr or the HMP monitor. Fortunately,
no such caller exists.
Simply use the first error as is. Update expected output of
qemu-iotest 049 accordingly.
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Max Reitz <mreitz@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20181017082702.5581-37-armbru@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
From include/qapi/error.h:
* Pass an existing error to the caller with the message modified:
* error_propagate(errp, err);
* error_prepend(errp, "Could not frobnicate '%s': ", name);
Fei Li pointed out that doing error_propagate() first doesn't work
well when @errp is &error_fatal or &error_abort: the error_prepend()
is never reached.
Since I doubt fixing the documentation will stop people from getting
it wrong, introduce error_propagate_prepend(), in the hope that it
lures people away from using its constituents in the wrong order.
Update the instructions in error.h accordingly.
Convert existing error_prepend() next to error_propagate to
error_propagate_prepend(). If any of these get reached with
&error_fatal or &error_abort, the error messages improve. I didn't
check whether that's the case anywhere.
Cc: Fei Li <fli@suse.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20181017082702.5581-2-armbru@redhat.com>
'detect-zeroes' is one of the basic BlockdevOptions available for all
drivers, but it's not handled by bdrv_reopen_prepare(), so any attempt
to change it results in an error:
(qemu) qemu-io virtio0 "reopen -o detect-zeroes=on"
Cannot change the option 'detect-zeroes'
Since there's no reason why we shouldn't allow changing it and the
implementation is simple let's just do it.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
'discard' is one of the basic BlockdevOptions available for all
drivers, but it's not handled by bdrv_reopen_prepare() so any attempt
to change it results in an error:
(qemu) qemu-io virtio0 "reopen -o discard=on"
Cannot change the option 'discard'
Since there's no reason why we shouldn't allow changing it and the
implementation is simple let's just do it.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The bdrv_reopen_prepare() function checks all options passed to each
BlockDriverState (in the reopen_state->options QDict) and makes all
necessary preparations to apply the option changes requested by the
user.
Options are removed from the QDict as they are processed, so at the
end of bdrv_reopen_prepare() only the options that can't be changed
are left. Then a loop goes over all remaining options and verifies
that the old and new values are identical, returning an error if
they're not.
The problem is that at the moment there are options that are removed
from the QDict although they can't be changed. The consequence of this
is any modification to any of those options is silently ignored:
(qemu) qemu-io virtio0 "reopen -o discard=on"
This happens when all options from bdrv_runtime_opts are removed
from the QDict but then only a few of them are processed. Since
it's especially important that "node-name" and "driver" are not
changed, the code puts them back into the QDict so they are checked
at the end of the function. Instead of putting only those two options
back into the QDict, this patch puts all unprocessed options using
qemu_opts_to_qdict().
update_flags_from_options() also needs to be modified to prevent
BDRV_OPT_CACHE_NO_FLUSH, BDRV_OPT_CACHE_DIRECT and BDRV_OPT_READ_ONLY
from going back to the QDict.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In the previous patches we removed all child references from
bs->{options,explicit_options} because keeping them is useless and
wrong.
Because of this, any attempt to reopen a BlockDriverState using a
child reference as one of its options would result in a failure,
because bdrv_reopen_prepare() would detect that there's a new option
(the child reference) that wasn't present in bs->options.
But passing child references on reopen can be useful. It's a way to
specify a BDS's child without having to pass recursively all of the
child's options, and if the reference points to a different BDS then
this can allow us to replace the child.
However, replacing the child is something that needs to be implemented
case by case and only when it makes sense. For now, this patch allows
passing a child reference as long as it points to the current child of
the BlockDriverState.
It's also important to remember that, as a consequence of the
previous patches, this child reference will be removed from
bs->{options,explicit_options} after the reopening has been completed.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In the previous patch we removed child references from bs->options, so
there's no need to look for them here anymore.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Block drivers allow opening their children using a reference to an
existing BlockDriverState. These references remain stored in the
'options' and 'explicit_options' QDicts, but we don't need to keep
them once everything is open.
What is more important, these values can become wrong if the children
change:
$ qemu-img create -f qcow2 hd0.qcow2 10M
$ qemu-img create -f qcow2 hd1.qcow2 10M
$ qemu-img create -f qcow2 hd2.qcow2 10M
$ $QEMU -drive if=none,file=hd0.qcow2,node-name=hd0 \
-drive if=none,file=hd1.qcow2,node-name=hd1,backing=hd0 \
-drive file=hd2.qcow2,node-name=hd2,backing=hd1
After this hd2 has hd1 as its backing file. Now let's remove it using
block_stream:
(qemu) block_stream hd2 0 hd0.qcow2
Now hd0 is the backing file of hd2, but hd2's options QDicts still
contain backing=hd1.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When draining a block node, we recurse to its parent and for subtree
drains also to its children. A single AIO_WAIT_WHILE() is then used to
wait for bdrv_drain_poll() to become true, which depends on all of the
nodes we recursed to. However, if the respective child or parent becomes
quiescent and calls bdrv_wakeup(), only the AioWait of the child/parent
is checked, while AIO_WAIT_WHILE() depends on the AioWait of the
original node.
Fix this by using a single AioWait for all callers of AIO_WAIT_WHILE().
This may mean that the draining thread gets a few more unnecessary
wakeups because an unrelated operation got completed, but we already
wake it up when something _could_ have changed rather than only if it
has certainly changed.
Apart from that, drain is a slow path anyway. In theory it would be
possible to use wakeups more selectively and still correctly, but the
gains are likely not worth the additional complexity. In fact, this
patch is a nice simplification for some places in the code.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Max Reitz <mreitz@redhat.com>
When a block device is opened with BDRV_O_SNAPSHOT and the
bdrv_append_temp_snapshot() call fails then the error code path tries
to unref the already destroyed 'options' QDict.
This can be reproduced easily by setting TMPDIR to a location where
the QEMU process can't write:
$ TMPDIR=/nonexistent $QEMU -drive driver=null-co,snapshot=on
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The last case where qobject_from_json() & friends return null without
setting an error is empty or blank input. Callers:
* block.c's parse_json_protocol() reports "Could not parse the JSON
options". It's marked as a work-around, because it also covered
actual bugs, but they got fixed in the previous few commits.
* qobject_input_visitor_new_str() reports "JSON parse error". Also
marked as work-around. The recent fixes have made this unreachable,
because it currently gets called only for input starting with '{'.
* check-qjson.c's empty_input() and blank_input() demonstrate the
behavior.
* The other callers are not affected since they only pass input with
exactly one JSON value or, in the case of negative tests, one error.
Fail with "Expecting a JSON value" instead of returning null, and
simplify callers.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20180823164025.12553-48-armbru@redhat.com>
This function returns a BDS's driver-specific options, excluding also
those from its children. Since we have just removed all children
options from bs->options there's no need to do this last step.
We allow references to children, though ("backing": "node0"), so those
we still have to remove.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If bdrv_reopen() succeeds then bs->explicit_options is updated with
the new values, but bs->options never changes.
Here's an example:
{ "execute": "blockdev-add",
"arguments": {
"driver": "qcow2",
"node-name": "hd0",
"overlap-check": "all",
"file": {
"driver": "file",
"filename": "hd0.qcow2"
}
}
}
After this, both bs->options and bs->explicit_options contain
"overlap-check": "all".
Now let's change that using qemu-io's reopen command:
(qemu) qemu-io hd0 "reopen -o overlap-check=none"
After this, bs->explicit_options contains the new value but
bs->options still keeps the old one.
This patch updates bs->options after a BDS has been successfully
reopened.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If a bdrv_reopen_multiple() call fails, then the explicit_options
QDict has to be deleted for every entry in the reopen queue. This must
happen regardless of whether that entry's bdrv_reopen_prepare() call
succeeded or not.
This patch simplifies the cleanup code a bit.
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
When bdrv_open_inherit() opens a BlockDriverState the options QDict
can contain options for some of its children, passed in the form of
child-name.option=value
So while each child is opened with that subset of options, those same
options remain stored in the parent BDS, leaving (at least) two copies
of each one of them ("child-name.option=value" in the parent and
"option=value" in the child).
Having the children options stored in the parent is unnecessary and it
can easily lead to an inconsistent state:
$ qemu-img create -f qcow2 hd0.qcow2 10M
$ qemu-img create -f qcow2 -b hd0.qcow2 hd1.qcow2
$ qemu-img create -f qcow2 -b hd1.qcow2 hd2.qcow2
$ $QEMU -drive file=hd2.qcow2,node-name=hd2,backing.node-name=hd1
This opens a chain of images hd0 <- hd1 <- hd2. Now let's remove hd1
using block_stream:
(qemu) block_stream hd2 0 hd0.qcow2
After this hd2 contains backing.node-name=hd1, which is no longer
correct because hd1 doesn't exist anymore.
This patch removes all children options from the parent dictionaries
at the end of bdrv_open_inherit() and bdrv_reopen_queue_child().
Signed-off-by: Alberto Garcia <berto@igalia.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Weil <sw@weilnetz.de>
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit dcf94a23b1 ('block: Don't poll in parent drain callbacks')
removed polling in bdrv_child_cb_drained_begin() on the grounds that the
original bdrv_drain() already will poll and BdrvChildRole.drained_begin
calls must not cause graph changes (and therefore must not call
aio_poll() or the recursion through the graph will break.
This reasoning is correct for calls through bdrv_do_drained_begin().
However, BdrvChildRole.drained_begin is also called when a node that is
already in a drained section (i.e. bdrv_do_drained_begin() has already
returned and therefore can't poll any more) is attached to a new parent.
In this case, we must explicitly poll to have all requests completed
before the drained new child can be attached to the parent.
In bdrv_replace_child_noperm(), we know that we're not inside the
recursion of bdrv_do_drained_begin() because graph changes are not
allowed there, and bdrv_replace_child_noperm() is a graph change. The
call of BdrvChildRole.drained_begin() must therefore be followed by a
BDRV_POLL_WHILE() that waits for the completion of requests.
Reported-by: Max Reitz <mreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the user passes a too long node name string, we silently truncate it
to fit into BlockDriverState.node_name, i.e. to 31 characters. Apart
from surprising the user when the node has a different name than
requested, this also bypasses the check for duplicate names, so that the
same name can be assigned to multiple nodes.
Fix this by just making too long node names an error.
Reported-by: Peter Krempa <pkrempa@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This allows using the two constants outside of block.c, which will
happen in a subsequent patch.
Signed-off-by: Ari Sundholm <ari@tuxera.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This moves the bdrv_truncate() implementation from block.c to block/io.c
so it can have access to the tracked requests infrastructure.
This involves making refresh_total_sectors() public (in block_int.h).
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
bdrv_truncate() is an operation that can block (even for a quite long
time, depending on the PreallocMode) in I/O paths that shouldn't block.
Convert it to a coroutine_fn so that we have the infrastructure for
drivers to make their .bdrv_co_truncate implementation asynchronous.
This change could potentially introduce new race conditions because
bdrv_truncate() isn't necessarily executed atomically any more. Whether
this is a problem needs to be evaluated for each block driver that
supports truncate:
* file-posix/win32, gluster, iscsi, nfs, rbd, ssh, sheepdog: The
protocol drivers are trivially safe because they don't actually yield
yet, so there is no change in behaviour.
* copy-on-read, crypto, raw-format: Essentially just filter drivers that
pass the request to a child node, no problem.
* qcow2: The implementation modifies metadata, so it needs to hold
s->lock to be safe with concurrent I/O requests. In order to avoid
double locking, this requires pulling the locking out into
preallocate_co() and using qcow2_write_caches() instead of
bdrv_flush().
* qed: Does a single header update, this is fine without locking.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Currently, bdrv_replace_node() refuses to create loops from one BDS to
itself if the BDS to be replaced is the backing node of the BDS to
replace it: Say there is a node A and a node B. Replacing B by A means
making all references to B point to A. If B is a child of A (i.e. A has
a reference to B), that would mean we would have to make this reference
point to A itself -- so we'd create a loop.
bdrv_replace_node() (through should_update_child()) refuses to do so if
B is the backing node of A. There is no reason why we should create
loops if B is not the backing node of A, though. The BDS graph should
never contain loops, so we should always refuse to create them.
If B is a child of A and B is to be replaced by A, we should simply
leave B in place there because it is the most sensible choice.
A more specific argument would be: Putting filter drivers into the BDS
graph is basically the same as appending an overlay to a backing chain.
But the main child BDS of a filter driver is not "backing" but "file",
so restricting the no-loop rule to backing nodes would fail here.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Reviewed-by: Fam Zheng <famz@redhat.com>
Reviewed-by: Alberto Garcia <berto@igalia.com>
Message-id: 20180613181823.13618-7-mreitz@redhat.com
Signed-off-by: Max Reitz <mreitz@redhat.com>
bdrv_drain_all_*() used bdrv_next() to iterate over all root nodes and
did a subtree drain for each of them. This works fine as long as the
graph is static, but sadly, reality looks different.
If the graph changes so that root nodes are added or removed, we would
have to compensate for this. bdrv_next() returns each root node only
once even if it's the root node for multiple BlockBackends or for a
monitor-owned block driver tree, which would only complicate things.
The much easier and more obviously correct way is to fundamentally
change the way the functions work: Iterate over all BlockDriverStates,
no matter who owns them, and drain them individually. Compensation is
only necessary when a new BDS is created inside a drain_all section.
Removal of a BDS doesn't require any action because it's gone afterwards
anyway.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
In the future, bdrv_drained_all_begin/end() will drain all invidiual
nodes separately rather than whole subtrees. This means that we don't
want to propagate the drain to all parents any more: If the parent is a
BDS, it will already be drained separately. Recursing to all parents is
unnecessary work and would make it an O(n²) operation.
Prepare the drain function for the changed drain_all by adding an
ignore_bds_parents parameter to the internal implementation that
prevents the propagation of the drain to BDS parents. We still (have to)
propagate it to non-BDS parents like BlockBackends or Jobs because those
are not drained separately.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_do_drained_begin() is only safe if we have a single
BDRV_POLL_WHILE() after quiescing all affected nodes. We cannot allow
that parent callbacks introduce a nested polling loop that could cause
graph changes while we're traversing the graph.
Split off bdrv_do_drained_begin_quiesce(), which only quiesces a single
node without waiting for its requests to complete. These requests will
be waited for in the BDRV_POLL_WHILE() call down the call chain.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Anything can happen inside BDRV_POLL_WHILE(), including graph
changes that may interfere with its callers (e.g. child list iteration
in recursive callers of bdrv_do_drained_begin).
Switch to a single BDRV_POLL_WHILE() call for the whole subtree at the
end of bdrv_do_drained_begin() to avoid such effects. The recursion
happens now inside the loop condition. As the graph can only change
between bdrv_drain_poll() calls, but not inside of it, doing the
recursion here is safe.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We already requested that block jobs be paused in .bdrv_drained_begin,
but no guarantee was made that the job was actually inactive at the
point where bdrv_drained_begin() returned.
This introduces a new callback BdrvChildRole.bdrv_drained_poll() and
uses it to make bdrv_drain_poll() consider block jobs using the node to
be drained.
For the test case to work as expected, we have to switch from
block_job_sleep_ns() to qemu_co_sleep_ns() so that the test job is even
considered active and must be waited for when draining the node.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
There are numerous QDict functions that have been introduced for and are
used only by the block layer. Move their declarations into an own
header file to reflect that.
While qdict_extract_subqdict() is in fact used outside of the block
layer (in util/qemu-config.c), it is still a function related very
closely to how the block layer works with nested QDicts, namely by
sometimes flattening them. Therefore, its declaration is put into this
header as well and util/qemu-config.c includes it with a comment stating
exactly which function it needs.
Suggested-by: Markus Armbruster <armbru@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-Id: <20180509165530.29561-7-mreitz@redhat.com>
[Copyright note tweaked, superfluous includes dropped]
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This is a useful function for the whole block layer, so make it public.
At the same time, users outside of block.c probably do not need to make
use of the reopen functionality, so rename the current function to
bdrv_is_writable_after_reopen() create a new bdrv_is_writable() function
that just passes NULL to it for the reopen queue.
Cc: qemu-stable@nongnu.org
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180606193702.7113-2-mreitz@redhat.com
Reviewed-by: John Snow <jsnow@redhat.com>
Reviewed-by: Jeff Cody <jcody@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
Looking at the qcow2 code that is riddled with error_report() calls,
this is really how it should have been from the start.
Along the way, turn the target_version/current_version comparisons at
the beginning of qcow2_downgrade() into assertions (the caller has to
make sure these conditions are met), and rephrase the error message on
using compat=1.1 to get refcount widths other than 16 bits.
Signed-off-by: Max Reitz <mreitz@redhat.com>
Message-id: 20180509210023.20283-3-mreitz@redhat.com
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: John Snow <jsnow@redhat.com>
Signed-off-by: Max Reitz <mreitz@redhat.com>
