Convert uses of bdrv_pwrite_sync() into bdrv_co_pwrite_sync() when the
callers are already coroutine_fn.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220609152744.3891847-10-afaria@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Also convert bdrv_pwrite_sync() to being implemented using
generated_co_wrapper.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220609152744.3891847-9-afaria@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
bdrv_{pread,pwrite}() now return -EIO instead of -EINVAL when 'bytes' is
negative, making them consistent with bdrv_{preadv,pwritev}() and
bdrv_co_{pread,pwrite,preadv,pwritev}().
bdrv_pwrite_zeroes() now also calls trace_bdrv_co_pwrite_zeroes() and
clears the BDRV_REQ_MAY_UNMAP flag when appropriate, which it didn't
previously.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Message-Id: <20220609152744.3891847-8-afaria@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
For consistency with other I/O functions, and in preparation to
implement bdrv_{pread,pwrite}() using generated_co_wrapper.
unsigned int fits in int64_t, so all callers remain correct.
bdrv_check_request32() is called further down the stack and causes -EIO
to be returned if 'bytes' is negative or greater than
BDRV_REQUEST_MAX_BYTES, which in turns never exceeds SIZE_MAX.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Message-Id: <20220609152744.3891847-7-afaria@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
They currently return the value of their headerlen/buflen parameter on
success. Returning 0 instead makes it clear that short reads/writes are
not possible.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220609152744.3891847-5-afaria@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
They currently return the value of their 'bytes' parameter on success.
Make them return 0 instead, for consistency with other I/O functions and
in preparation to implement them using generated_co_wrapper. This also
makes it clear that short reads/writes are not possible.
The few callers that rely on the previous behavior are adjusted
accordingly by hand.
Signed-off-by: Alberto Faria <afaria@redhat.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220609152744.3891847-4-afaria@redhat.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Jens Axboe has confirmed that short reads are rare but can happen:
https://lore.kernel.org/io-uring/YsU%2FCGkl9ZXUI+Tj@stefanha-x1.localdomain/T/#m729963dc577d709b709c191922e98ec79d7eef54
The luring_resubmit_short_read() comment claimed they were only due to a
specific io_uring bug that was fixed in Linux commit 9d93a3f5a0c
("io_uring: punt short reads to async context"), which is wrong.
Dominique Martinet found that a btrfs bug also causes short reads. There
may be more kernel code paths that result in short reads.
Let's consider short reads fair game.
Cc: Dominique Martinet <dominique.martinet@atmark-techno.com>
Based-on: <20220630010137.2518851-1-dominique.martinet@atmark-techno.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Message-id: 20220706080341.1206476-1-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
sqeq.off here is the offset to read within the disk image, so obviously
not 'nread' (the amount we just read), but as the author meant to write
its current value incremented by the amount we just read.
Normally recent versions of linux will not issue short reads,
but it can happen so we should fix this.
This lead to weird image corruptions when short read happened
Fixes: 6663a0a337 ("block/io_uring: implements interfaces for io_uring")
Link: https://lkml.kernel.org/r/YrrFGO4A1jS0GI0G@atmark-techno.com
Signed-off-by: Dominique Martinet <dominique.martinet@atmark-techno.com>
Message-Id: <20220630010137.2518851-1-dominique.martinet@atmark-techno.com>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This patch makes in_flight field 'unsigned' for BDRVNBDState and
MirrorBlockJob. This matches the definition of this field on BDS
and is generically correct - we should never get negative value here.
Signed-off-by: Denis V. Lunev <den@openvz.org>
CC: John Snow <jsnow@redhat.com>
CC: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Hanna Reitz <hreitz@redhat.com>
CC: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
At the moment there are 2 sources of lengthy operations if configured:
* open connection, which could retry inside and
* reconnect of already opened connection
These operations could be quite lengthy and cumbersome to catch thus
it would be quite natural to add trace points for them.
This patch is based on the original downstream work made by Vladimir.
Signed-off-by: Denis V. Lunev <den@openvz.org>
CC: Eric Blake <eblake@redhat.com>
CC: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
CC: Kevin Wolf <kwolf@redhat.com>
CC: Hanna Reitz <hreitz@redhat.com>
CC: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
In some scenarios, when copy-before-write operations lasts too long
time, it's better to cancel it.
Most useful would be to use the new option together with
on-cbw-error=break-snapshot: this way if cbw operation takes too long
time we'll just cancel backup process but do not disturb the guest too
much.
Note the tricky point of realization: we keep additional point in
bs->in_flight during block_copy operation even if it's timed-out.
Background "cancelled" block_copy operations will finish at some point
and will want to access state. We should care to not free the state in
.bdrv_close() earlier.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
[vsementsov: use bdrv_inc_in_flight()/bdrv_dec_in_flight() instead of
direct manipulation on bs->in_flight]
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Add possibility to limit block_copy() call in time. To be used in the
next commit.
As timed-out block_copy() call will continue in background anyway (we
can't immediately cancel IO operation), it's important also give user a
possibility to pass a callback, to do some additional actions on
block-copy call finish.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Currently, behavior on copy-before-write operation failure is simple:
report error to the guest.
Let's implement alternative behavior: break the whole copy-before-write
process (and corresponding backup job or NBD client) but keep guest
working. It's needed if we consider guest stability as more important.
The realisation is simple: on copy-before-write failure we set
s->snapshot_ret and continue guest operations. s->snapshot_ret being
set will lead to all further snapshot API requests. Note that all
in-flight snapshot-API requests may still success: we do wait for them
on BREAK_SNAPSHOT-failure path in cbw_do_copy_before_write().
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
We are going to add one more option of enum type. Let's refactor option
parsing so that we can simply work with BlockdevOptionsCbw object.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Hanna Reitz <hreitz@redhat.com>
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Currently we use 'id' option as the name of VDUSE device.
It's a bit confusing since we use one value for two different
purposes: the ID to identfy the export within QEMU (must be
distinct from any other exports in the same QEMU process, but
can overlap with names used by other processes), and the VDUSE
name to uniquely identify it on the host (must be distinct from
other VDUSE devices on the same host, but can overlap with other
export types like NBD in the same process). To make it clear,
this patch adds a separate 'name' option to specify the VDUSE
name for the vduse-blk export instead.
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220614051532.92-7-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Add a 'serial' option to allow user to specify this value
explicitly. And the default value is changed to an empty
string as what we did in "hw/block/virtio-blk.c".
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220614051532.92-6-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
CID 1488362 points out that the second 'rc >= 0' check is now dead
code.
Reported-by: Peter Maydell <peter.maydell@linaro.org>
Fixes: 172f5f1a40(nbd: remove peppering of nbd_client_connected)
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20220516210519.76135-1-eblake@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
On 64-bit platforms, assigning SIZE_MAX to the int64_t max_pdiscard
results in a negative value, and the following assertion would trigger
down the line (it's not the same max_pdiscard, but computed from the
other one):
qemu-system-x86_64: ../block/io.c:3166: bdrv_co_pdiscard: Assertion
`max_pdiscard >= bs->bl.request_alignment' failed.
On 32-bit platforms, it's fine to keep using SIZE_MAX.
The assertion in qemu_gluster_co_pdiscard() is checking that the value
of 'bytes' can safely be passed to glfs_discard_async(), which takes a
size_t for the argument in question, so it is kept as is. And since
max_pdiscard is still <= SIZE_MAX, relying on max_pdiscard is still
fine.
Fixes: 0c8022876f ("block: use int64_t instead of int in driver discard handlers")
Cc: qemu-stable@nongnu.org
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
Message-Id: <20220520075922.43972-1-f.ebner@proxmox.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
If the namespace does not exist, rbd_create() fails with -ENOENT and
QEMU reports a generic "error rbd create: No such file or directory":
$ qemu-img create rbd:rbd/namespace/image 1M
Formatting 'rbd:rbd/namespace/image', fmt=raw size=1048576
qemu-img: rbd:rbd/namespace/image: error rbd create: No such file or directory
Unfortunately rados_ioctx_set_namespace() does not fail if the namespace
does not exist, so let's use rbd_namespace_exists() in qemu_rbd_connect()
to check if the namespace exists, reporting a more understandable error:
$ qemu-img create rbd:rbd/namespace/image 1M
Formatting 'rbd:rbd/namespace/image', fmt=raw size=1048576
qemu-img: rbd:rbd/namespace/image: namespace 'namespace' does not exist
Reported-by: Tingting Mao <timao@redhat.com>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220517071012.6120-1-sgarzare@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
To support reconnecting after restart or crash, VDUSE backend
might need to resubmit inflight I/Os. This stores the metadata
such as the index of inflight I/O's descriptors to a shm file so
that VDUSE backend can restore them during reconnecting.
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220523084611.91-9-xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
To support block resize, this uses vduse_dev_update_config()
to update the capacity field in configuration space and inject
config interrupt on the block resize callback.
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220523084611.91-8-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This implements a VDUSE block backends based on
the libvduse library. We can use it to export the BDSs
for both VM and container (host) usage.
The new command-line syntax is:
$ qemu-storage-daemon \
--blockdev file,node-name=drive0,filename=test.img \
--export vduse-blk,node-name=drive0,id=vduse-export0,writable=on
After the qemu-storage-daemon started, we need to use
the "vdpa" command to attach the device to vDPA bus:
$ vdpa dev add name vduse-export0 mgmtdev vduse
Also the device must be removed via the "vdpa" command
before we stop the qemu-storage-daemon.
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220523084611.91-7-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Abstract the common logic of virtio-blk I/O process to a function
named virtio_blk_process_req(). It's needed for the following commit.
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220523084611.91-4-xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now the req->size is set to the correct value only
when handling VIRTIO_BLK_T_GET_ID request. This patch
fixes it.
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220523084611.91-3-xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
This supports passing NULL ops to blk_set_dev_ops()
so that we can remove stale ops in some cases.
Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220523084611.91-2-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We have too much logic to simply check that bitmaps are of the same
size. Let's just define that hbitmap_merge() and
bdrv_dirty_bitmap_merge_internal() require their argument bitmaps be of
same size, this simplifies things.
Let's look through the callers:
For backup_init_bcs_bitmap() we already assert that merge can't fail.
In bdrv_reclaim_dirty_bitmap_locked() we gracefully handle the error
that can't happen: successor always has same size as its parent, drop
this logic.
In bdrv_merge_dirty_bitmap() we already has assertion and separate
check. Make the check explicit and improve error message.
Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Reviewed-by: Nikita Lapshin <nikita.lapshin@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220517111206.23585-4-v.sementsov-og@mail.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
We don't need extra bitmap. All we need is to backup the original
bitmap when we do first merge. So, drop extra temporary bitmap and work
directly with target and backup.
Still to keep old semantics, that on failure target is unchanged and
user don't need to restore, we need a local_backup variable and do
restore ourselves on failure path.
Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Message-Id: <20220517111206.23585-3-v.sementsov-og@mail.ru>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
At the end we ignore failure of bdrv_merge_dirty_bitmap() and report
success. And still set errp. That's wrong.
Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Reviewed-by: Nikita Lapshin <nikita.lapshin@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220517111206.23585-2-v.sementsov-og@mail.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Commit 1b7fd72955 ("block: rename buffer_alignment to
guest_block_size") noted:
At this point, the field is set by the device emulation, but completely
ignored by the block layer.
The last time the value of buffer_alignment/guest_block_size was
actually used was before commit 339064d506 ("block: Don't use guest
sector size for qemu_blockalign()").
This value has not been used since 2013. Get rid of it.
Cc: Xie Yongji <xieyongji@bytedance.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220518130945.2657905-1-stefanha@redhat.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Faria <afaria@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
bdrv_co_drain() has not been used since commit 9a0cec664e ("mirror:
use bdrv_drained_begin/bdrv_drained_end") in 2016. Remove it so there
are fewer drain scenarios to worry about.
Use bdrv_drained_begin()/bdrv_drained_end() instead. They are "mixed"
functions that can be called from coroutine context. Unlike
bdrv_co_drain(), these functions provide control of the length of the
drained section, which is usually the right thing.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220521122714.3837731-1-stefanha@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Alberto Faria <afaria@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It may not be obvious why laio_io_unplug() checks max batch. I discussed
this with Stefano and have added a comment summarizing the reason.
Cc: Stefano Garzarella <sgarzare@redhat.com>
Cc: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Message-id: 20220609164712.1539045-3-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Every laio_io_plug() call has a matching laio_io_unplug() call. There is
a plugged counter that tracks the number of levels of plugging and
allows for nesting.
The plugged counter must reflect the balance between laio_io_plug() and
laio_io_unplug() calls accurately. Otherwise I/O stalls occur since
io_submit(2) calls are skipped while plugged.
Reported-by: Nikolay Tenev <nt@storpool.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Message-id: 20220609164712.1539045-2-stefanha@redhat.com
Cc: Stefano Garzarella <sgarzare@redhat.com>
Fixes: 68d7946648 ("linux-aio: add `dev_max_batch` parameter to laio_io_unplug()")
[Stefano Garzarella suggested adding a Fixes tag.
--Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Linux recently added a new io_uring(7) optimization API that QEMU
doesn't take advantage of yet. The liburing library that QEMU uses
has added a corresponding new API calling io_uring_register_ring_fd().
When this API is called after creating the ring, the io_uring_submit()
library function passes a flag to the io_uring_enter(2) syscall
allowing it to skip the ring file descriptor fdget()/fdput()
operations. This saves some CPU cycles.
Signed-off-by: Sam Li <faithilikerun@gmail.com>
Message-id: 20220531105011.111082-1-faithilikerun@gmail.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
qemu_co_queue_restart_all is basically the same as qemu_co_enter_all
but without a QemuLockable argument. That's perfectly fine, but only as
long as the function is marked coroutine_fn. If used outside coroutine
context, qemu_co_queue_wait will attempt to take the lock and that
is just broken: if you are calling qemu_co_queue_restart_all outside
coroutine context, the lock is going to be a QemuMutex which cannot be
taken twice by the same thread.
The patch adds the marker to qemu_co_queue_restart_all and to its sole
non-coroutine_fn caller; it then reimplements the function in terms of
qemu_co_enter_all_impl, to remove duplicated code and to clarify that the
latter also works in coroutine context.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Message-Id: <20220427130830.150180-4-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Leading underscores are ill-advised because such identifiers are
reserved. Trailing underscores are merely ugly. Strip both.
Our header guards commonly end in _H. Normalize the exceptions.
Macros should be ALL_CAPS. Normalize the exception.
Done with scripts/clean-header-guards.pl.
include/hw/xen/interface/ and tools/virtiofsd/ left alone, because
these were imported from Xen and libfuse respectively.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20220506134911.2856099-3-armbru@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Header guard symbols should match their file name to make guard
collisions less likely.
Cleaned up with scripts/clean-header-guards.pl, followed by some
renaming of new guard symbols picked by the script to better ones.
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20220506134911.2856099-2-armbru@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
[Change to generated file ebpf/rss.bpf.skeleton.h backed out]
VMDK disk data is stored in extents, which may or may not be separate
from bs->file. VmdkExtent.file points to where they are stored. Each
that is stored in bs->file will simply reuse the exact pointer value of
bs->file.
(That is why vmdk_free_extents() will unref VmdkExtent.file (e->file)
only if e->file != bs->file.)
Reopen operations can change bs->file (they will replace the whole
BdrvChild object, not just the BDS stored in that BdrvChild), and then
we will need to change all .file pointers of all such VmdkExtents to
point to the new BdrvChild.
In vmdk_reopen_prepare(), we have to check which VmdkExtents are
affected, and in vmdk_reopen_commit(), we can modify them. We have to
split this because:
- The new BdrvChild is created only after prepare, so we can change
VmdkExtent.file only in commit
- In commit, there no longer is any (valid) reference to the old
BdrvChild object, so there would be nothing to compare VmdkExtent.file
against to see whether it was equal to bs->file before reopening
(There is BDRVReopenState.old_file_bs, but the old bs->file
BdrvChild's .bs pointer will be NULL-ed when the new BdrvChild is
created, and so we cannot compare VmdkExtent.file->bs against
BDRVReopenState.old_file_bs)
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20220314162719.65384-2-hreitz@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
qcow2_co_invalidate_cache() closes and opens the qcow2 file, by calling
qcow2_close() and qcow2_do_open(). These two functions must thus be
usable from both a global-state and an I/O context.
As they are, they are not safe to call in an I/O context, because they
use bdrv_unref_child() and bdrv_open_child() to close/open the data_file
child, respectively, both of which are global-state functions. When
used from qcow2_co_invalidate_cache(), we do not need to close/open the
data_file child, though (we do not do this for bs->file or bs->backing
either), and so we should skip it in the qcow2_co_invalidate_cache()
path.
To do so, add a parameter to qcow2_do_open() and qcow2_close() to make
them skip handling s->data_file, and have qcow2_co_invalidate_cache()
exempt it from the memset() on the BDRVQcow2State.
(Note that the QED driver similarly closes/opens the QED image by
invoking bdrv_qed_close()+bdrv_qed_do_open(), but both functions seem
safe to use in an I/O context.)
Fixes: https://gitlab.com/qemu-project/qemu/-/issues/945
Signed-off-by: Hanna Reitz <hreitz@redhat.com>
Message-Id: <20220427114057.36651-3-hreitz@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
It is only used by block/file-posix.c, move it there.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
requests[].receiving is set by nbd_receive_replies() under the receive_mutex;
Read it under the same mutex as well. Waking up receivers on errors happens
after each reply finishes processing, in nbd_co_receive_one_chunk().
If there is no currently-active reply, there are two cases:
* either there is no active request at all, in which case no
element of request[] can have .receiving = true
* or nbd_receive_replies() must be running and owns receive_mutex;
in that case it will get back to nbd_co_receive_one_chunk() because
the socket has been shutdown, and all waiting coroutines will wake up
in turn.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-9-pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
Remove the confusing, and most likely wrong, atomics. The only function
that used to be somewhat in a hot path was nbd_client_connected(),
but it is not anymore after the previous patches.
The same logic is used both to check if a request had to be reissued
and also in nbd_reconnecting_attempt(). The former cases are outside
requests_lock, while nbd_reconnecting_attempt() does have the lock,
therefore the two have been separated in the previous commit.
nbd_client_will_reconnect() can simply take s->requests_lock, while
nbd_reconnecting_attempt() can inline the access now that no
complicated atomics are involved.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-8-pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
Prepare for the next patch, so that the diff is less confusing.
nbd_client_connecting is moved closer to the definition point.
nbd_client_connecting_wait() is kept only for the reconnection
logic; when it is used to check if a request has to be reissued,
use the renamed function nbd_client_will_reconnect(). In the
next patch, the two cases will have different locking requirements.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-7-pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
The condition for waiting on the s->free_sema queue depends on
both s->in_flight and s->state. The latter is currently using
atomics, but this is quite dubious and probably wrong.
Because s->state is written in the main thread too, for example by
the yank callback, it cannot be protected by a CoMutex. Introduce a
separate lock that can be used by nbd_co_send_request(); later on this
lock will also be used for s->state. There will not be any contention
on the lock unless there is a yank or reconnect, so this is not
performance sensitive.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-6-pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
Elevate s->in_flight early so that other incoming requests will wait
on the CoQueue in nbd_co_send_request; restart them after getting back
from nbd_reconnect_attempt. This could be after the reconnect timer or
nbd_cancel_in_flight have cancelled the attempt, so there is no
need anymore to cancel the requests there.
nbd_co_send_request now handles both stopping and restarting pending
requests after a successful connection, and there is no need to
hold send_mutex in nbd_co_do_establish_connection. The current setup
is confusing because nbd_co_do_establish_connection is called both with
send_mutex taken and without it. Before the patch it uses free_sema which
(at least in theory...) is protected by send_mutex, after the patch it
does not anymore.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-5-pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
[eblake: wrap long line]
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
It is unnecessary to check nbd_client_connected() because every time
s->state is moved out of NBD_CLIENT_CONNECTED the socket is shut down
and all coroutines are resumed.
The only case where it was actually needed is when the NBD
server disconnects and there is no reconnect-delay. In that
case, nbd_receive_replies() does not set s->reply.handle and
nbd_co_do_receive_one_chunk() cannot continue. For that one case,
check the return value of nbd_receive_replies().
As to the others:
* nbd_receive_replies() can put the current coroutine to sleep if another
reply is ongoing; then it will be woken by nbd_channel_error(), called
by the ongoing reply. Or it can try itself to read a reply header and
fail, thus calling nbd_channel_error() itself.
* nbd_co_send_request() will write the body of the request and fail
* nbd_reply_chunk_iter_receive() will call nbd_co_receive_one_chunk()
and then nbd_co_do_receive_one_chunk(), which will handle the failure as
above; or it will just detect a previous call to nbd_iter_channel_error()
via iter->ret < 0.
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-4-pbonzini@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
Several coroutine functions in block/nbd.c are not marked as such. This
patch adds a few more markers; it is not exhaustive, but it focuses
especially on:
- places that wake other coroutines, because aio_co_wake() has very
different semantics inside a coroutine (queuing after yield vs. entering
immediately);
- functions with _co_ in their names, to avoid confusion
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-3-pbonzini@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
The .reply_possible field of s->requests is never set to false. This is
not a problem as it is only a safeguard to detect protocol errors,
but it's sloppy. In fact, the field is actually not necessary at all,
because .coroutine is set to NULL in NBD_FOREACH_REPLY_CHUNK after
receiving the last chunk. Thus, replace .reply_possible with .coroutine
and move the check before deciding the fate of this request.
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <20220414175756.671165-2-pbonzini@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@openvz.org>
Reviewed-by: Lukas Straub <lukasstraub2@web.de>
Signed-off-by: Eric Blake <eblake@redhat.com>
