The vhost-user-blk export implement AioContext switches in its drain
implementation. This means that on drain_begin, it detaches the server
from its AioContext and on drain_end, attaches it again and schedules
the server->co_trip coroutine in the updated AioContext.
However, nothing guarantees that server->co_trip is even safe to be
scheduled. Not only is it unclear that the coroutine is actually in a
state where it can be reentered externally without causing problems, but
with two consecutive drains, it is possible that the scheduled coroutine
didn't have a chance yet to run and trying to schedule an already
scheduled coroutine a second time crashes with an assertion failure.
Following the model of NBD, this commit makes the vhost-user-blk export
shut down server->co_trip during drain so that resuming the export means
creating and scheduling a new coroutine, which is always safe.
There is one exception: If the drain call didn't poll (for example, this
happens in the context of bdrv_graph_wrlock()), then the coroutine
didn't have a chance to shut down. However, in this case the AioContext
can't have changed; changing the AioContext always involves a polling
drain. So in this case we can simply assert that the AioContext is
unchanged and just leave the coroutine running or wake it up if it has
yielded to wait for the AioContext to be attached again.
Fixes: e1054cd4aa
Fixes: https://issues.redhat.com/browse/RHEL-1708
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-ID: <20231127115755.22846-1-kwolf@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Fix:
util/vhost-user-server.c: In function ‘set_watch’:
util/vhost-user-server.c:274:20: warning: declaration of ‘vu_fd_watch’ shadows a previous local [-Wshadow=compatible-local]
274 | VuFdWatch *vu_fd_watch = g_new0(VuFdWatch, 1);
| ^~~~~~~~~~~
util/vhost-user-server.c:271:16: note: shadowed declaration is here
271 | VuFdWatch *vu_fd_watch = find_vu_fd_watch(server, fd);
| ^~~~~~~~~~~
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-ID: <20230904161235.84651-18-philmd@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
The ongoing QEMU multi-queue block layer effort makes it possible for multiple
threads to process I/O in parallel. The nbd block driver is not compatible with
the multi-queue block layer yet because QIOChannel cannot be used easily from
coroutines running in multiple threads. This series changes the QIOChannel API
to make that possible.
In the current API, calling qio_channel_attach_aio_context() sets the
AioContext where qio_channel_yield() installs an fd handler prior to yielding:
qio_channel_attach_aio_context(ioc, my_ctx);
...
qio_channel_yield(ioc); // my_ctx is used here
...
qio_channel_detach_aio_context(ioc);
This API design has limitations: reading and writing must be done in the same
AioContext and moving between AioContexts involves a cumbersome sequence of API
calls that is not suitable for doing on a per-request basis.
There is no fundamental reason why a QIOChannel needs to run within the
same AioContext every time qio_channel_yield() is called. QIOChannel
only uses the AioContext while inside qio_channel_yield(). The rest of
the time, QIOChannel is independent of any AioContext.
In the new API, qio_channel_yield() queries the AioContext from the current
coroutine using qemu_coroutine_get_aio_context(). There is no need to
explicitly attach/detach AioContexts anymore and
qio_channel_attach_aio_context() and qio_channel_detach_aio_context() are gone.
One coroutine can read from the QIOChannel while another coroutine writes from
a different AioContext.
This API change allows the nbd block driver to use QIOChannel from any thread.
It's important to keep in mind that the block driver already synchronizes
QIOChannel access and ensures that two coroutines never read simultaneously or
write simultaneously.
This patch updates all users of qio_channel_attach_aio_context() to the
new API. Most conversions are simple, but vhost-user-server requires a
new qemu_coroutine_yield() call to quiesce the vu_client_trip()
coroutine when not attached to any AioContext.
While the API is has become simpler, there is one wart: QIOChannel has a
special case for the iohandler AioContext (used for handlers that must not run
in nested event loops). I didn't find an elegant way preserve that behavior, so
I added a new API called qio_channel_set_follow_coroutine_ctx(ioc, true|false)
for opting in to the new AioContext model. By default QIOChannel uses the
iohandler AioHandler. Code that formerly called
qio_channel_attach_aio_context() now calls
qio_channel_set_follow_coroutine_ctx(ioc, true) once after the QIOChannel is
created.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Acked-by: Daniel P. Berrangé <berrange@redhat.com>
Message-ID: <20230830224802.493686-5-stefanha@redhat.com>
[eblake: also fix migration/rdma.c]
Signed-off-by: Eric Blake <eblake@redhat.com>
All callers now pass is_external=false to aio_set_fd_handler() and
aio_set_event_notifier(). The aio_disable_external() API that
temporarily disables fd handlers that were registered is_external=true
is therefore dead code.
Remove aio_disable_external(), aio_enable_external(), and the
is_external arguments to aio_set_fd_handler() and
aio_set_event_notifier().
The entire test-fdmon-epoll test is removed because its sole purpose was
testing aio_disable_external().
Parts of this patch were generated using the following coccinelle
(https://coccinelle.lip6.fr/) semantic patch:
@@
expression ctx, fd, is_external, io_read, io_write, io_poll, io_poll_ready, opaque;
@@
- aio_set_fd_handler(ctx, fd, is_external, io_read, io_write, io_poll, io_poll_ready, opaque)
+ aio_set_fd_handler(ctx, fd, io_read, io_write, io_poll, io_poll_ready, opaque)
@@
expression ctx, notifier, is_external, io_read, io_poll, io_poll_ready;
@@
- aio_set_event_notifier(ctx, notifier, is_external, io_read, io_poll, io_poll_ready)
+ aio_set_event_notifier(ctx, notifier, io_read, io_poll, io_poll_ready)
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20230516190238.8401-21-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
vhost-user activity must be suspended during bdrv_drained_begin/end().
This prevents new requests from interfering with whatever is happening
in the drained section.
Previously this was done using aio_set_fd_handler()'s is_external
argument. In a multi-queue block layer world the aio_disable_external()
API cannot be used since multiple AioContext may be processing I/O, not
just one.
Switch to BlockDevOps->drained_begin/end() callbacks.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20230516190238.8401-8-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Each vhost-user-blk request runs in a coroutine. When the BlockBackend
enters a drained section we need to enter a quiescent state. Currently
any in-flight requests race with bdrv_drained_begin() because it is
unaware of vhost-user-blk requests.
When blk_co_preadv/pwritev()/etc returns it wakes the
bdrv_drained_begin() thread but vhost-user-blk request processing has
not yet finished. The request coroutine continues executing while the
main loop thread thinks it is in a drained section.
One example where this is unsafe is for blk_set_aio_context() where
bdrv_drained_begin() is called before .aio_context_detached() and
.aio_context_attach(). If request coroutines are still running after
bdrv_drained_begin(), then the AioContext could change underneath them
and they race with new requests processed in the new AioContext. This
could lead to virtqueue corruption, for example.
(This example is theoretical, I came across this while reading the
code and have not tried to reproduce it.)
It's easy to make bdrv_drained_begin() wait for in-flight requests: add
a .drained_poll() callback that checks the VuServer's in-flight counter.
VuServer just needs an API that returns true when there are requests in
flight. The in-flight counter needs to be atomic.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20230516190238.8401-7-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
The VuServer object has a refcount field and ref/unref APIs. The name is
confusing because it's actually an in-flight request counter instead of
a refcount.
Normally a refcount destroys the object upon reaching zero. The VuServer
counter is used to wake up the vhost-user coroutine when there are no
more requests.
Avoid confusing by renaming refcount and ref/unref to in_flight and
inc/dec.
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20230516190238.8401-6-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
vhost_user_server_stop() uses AIO_WAIT_WHILE(). AIO_WAIT_WHILE()
requires that AioContext is only acquired once.
Since blk_exp_request_shutdown() already acquires the AioContext it
shouldn't be acquired again in vhost_user_server_stop().
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20230323145853.1345527-1-stefanha@redhat.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Include it in the .c files instead that use the error reporting
functions.
Message-Id: <20230210111931.1115489-1-thuth@redhat.com>
Reviewed-by: Markus Armbruster <armbru@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Signed-off-by: Thomas Huth <thuth@redhat.com>
MSG_PEEK peeks at the channel, The data is treated as unread and
the next read shall still return this data. This support is
currently added only for socket class. Extra parameter 'flags'
is added to io_readv calls to pass extra read flags like MSG_PEEK.
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Daniel P. Berrange <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Suggested-by: Daniel P. Berrange <berrange@redhat.com>
Signed-off-by: manish.mishra <manish.mishra@nutanix.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
The qemu_*block() functions are meant to be be used with sockets (the
win32 implementation expects SOCKET)
Over time, those functions where used with Win32 SOCKET or
file-descriptors interchangeably. But for portability, they must only be
used with socket-like file-descriptors. FDs can use
g_unix_set_fd_nonblocking() instead.
Rename the functions with "socket" in the name to prevent bad usages.
This is effectively reverting commit f9e8cacc55 ("oslib-posix:
rename socket_set_nonblock() to qemu_set_nonblock()").
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
The vhost-user-blk export runs requests asynchronously in their own
coroutine. When the vhost connection goes away and we want to stop the
vhost-user server, we need to wait for these coroutines to stop before
we can unmap the shared memory. Otherwise, they would still access the
unmapped memory and crash.
This introduces a refcount to VuServer which is increased when spawning
a new request coroutine and decreased before the coroutine exits. The
memory is only unmapped when the refcount reaches zero.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220125151435.48792-1-kwolf@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Adaptive polling measures the execution time of the polling check plus
handlers called when a polled event becomes ready. Handlers can take a
significant amount of time, making it look like polling was running for
a long time when in fact the event handler was running for a long time.
For example, on Linux the io_submit(2) syscall invoked when a virtio-blk
device's virtqueue becomes ready can take 10s of microseconds. This
can exceed the default polling interval (32 microseconds) and cause
adaptive polling to stop polling.
By excluding the handler's execution time from the polling check we make
the adaptive polling calculation more accurate. As a result, the event
loop now stays in polling mode where previously it would have fallen
back to file descriptor monitoring.
The following data was collected with virtio-blk num-queues=2
event_idx=off using an IOThread. Before:
168k IOPS, IOThread syscalls:
9837.115 ( 0.020 ms): IO iothread1/620155 io_submit(ctx_id: 140512552468480, nr: 16, iocbpp: 0x7fcb9f937db0) = 16
9837.158 ( 0.002 ms): IO iothread1/620155 write(fd: 103, buf: 0x556a2ef71b88, count: 8) = 8
9837.161 ( 0.001 ms): IO iothread1/620155 write(fd: 104, buf: 0x556a2ef71b88, count: 8) = 8
9837.163 ( 0.001 ms): IO iothread1/620155 ppoll(ufds: 0x7fcb90002800, nfds: 4, tsp: 0x7fcb9f1342d0, sigsetsize: 8) = 3
9837.164 ( 0.001 ms): IO iothread1/620155 read(fd: 107, buf: 0x7fcb9f939cc0, count: 512) = 8
9837.174 ( 0.001 ms): IO iothread1/620155 read(fd: 105, buf: 0x7fcb9f939cc0, count: 512) = 8
9837.176 ( 0.001 ms): IO iothread1/620155 read(fd: 106, buf: 0x7fcb9f939cc0, count: 512) = 8
9837.209 ( 0.035 ms): IO iothread1/620155 io_submit(ctx_id: 140512552468480, nr: 32, iocbpp: 0x7fca7d0cebe0) = 32
174k IOPS (+3.6%), IOThread syscalls:
9809.566 ( 0.036 ms): IO iothread1/623061 io_submit(ctx_id: 140539805028352, nr: 32, iocbpp: 0x7fd0cdd62be0) = 32
9809.625 ( 0.001 ms): IO iothread1/623061 write(fd: 103, buf: 0x5647cfba5f58, count: 8) = 8
9809.627 ( 0.002 ms): IO iothread1/623061 write(fd: 104, buf: 0x5647cfba5f58, count: 8) = 8
9809.663 ( 0.036 ms): IO iothread1/623061 io_submit(ctx_id: 140539805028352, nr: 32, iocbpp: 0x7fd0d0388b50) = 32
Notice that ppoll(2) and eventfd read(2) syscalls are eliminated because
the IOThread stays in polling mode instead of falling back to file
descriptor monitoring.
As usual, polling is not implemented on Windows so this patch ignores
the new io_poll_read() callback in aio-win32.c.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Message-id: 20211207132336.36627-2-stefanha@redhat.com
[Fixed up aio_set_event_notifier() calls in
tests/unit/test-fdmon-epoll.c added after this series was queued.
--Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Headers used by other subsystems are located in include/. Also add the
vhost-user-server and vhost-user-blk-server headers to MAINTAINERS.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20200924151549.913737-13-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Use the new QAPI block exports API instead of defining our own QOM
objects.
This is a large change because the lifecycle of VuBlockDev needs to
follow BlockExportDriver. QOM properties are replaced by QAPI options
objects.
VuBlockDev is renamed VuBlkExport and contains a BlockExport field.
Several fields can be dropped since BlockExport already has equivalents.
The file names and meson build integration will be adjusted in a future
patch. libvhost-user should probably be built as a static library that
is linked into QEMU instead of as a .c file that results in duplicate
compilation.
The new command-line syntax is:
$ qemu-storage-daemon \
--blockdev file,node-name=drive0,filename=test.img \
--export vhost-user-blk,node-name=drive0,id=export0,unix-socket=/tmp/vhost-user-blk.sock
Note that unix-socket is optional because we may wish to accept chardevs
too in the future.
Markus noted that supported address families are not explicit in the
QAPI schema. It is unlikely that support for more address families will
be added since file descriptor passing is required and few address
families support it. If a new address family needs to be added, then the
QAPI 'features' syntax can be used to advertize them.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Acked-by: Markus Armbruster <armbru@redhat.com>
Message-id: 20200924151549.913737-12-stefanha@redhat.com
[Skip test on big-endian host architectures because this device doesn't
support them yet (as already mentioned in a code comment).
--Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The vu_client_trip() coroutine is leaked during AioContext switching. It
is also unsafe to destroy the vu_dev in panic_cb() since its callers
still access it in some cases.
Rework the lifecycle to solve these safety issues.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20200924151549.913737-10-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Unexpected EOF is an error that must be reported.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20200924151549.913737-9-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
fds[] is leaked when qio_channel_readv_full() fails.
Use vmsg->fds[] instead of keeping a local fds[] array. Then we can
reuse goto fail to clean up fds. vmsg->fd_num must be zeroed before the
loop to make this safe.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20200924151549.913737-8-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
The device panic notifier callback is not used. Drop it.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20200924151549.913737-7-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Explicitly deleting watches is not necessary since libvhost-user calls
remove_watch() during vu_deinit(). Add an assertion to check this
though.
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20200924151549.913737-5-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
We already have access to the value with the correct type (ioc and sioc
are the same QIOChannel).
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 20200924151549.913737-4-stefanha@redhat.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Sharing QEMU devices via vhost-user protocol.
Only one vhost-user client can connect to the server one time.
Suggested-by: Kevin Wolf <kwolf@redhat.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Coiby Xu <coiby.xu@gmail.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-id: 20200918080912.321299-4-coiby.xu@gmail.com
[Fixed size_t %lu -> %zu format string compiler error.
--Stefan]
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
