At a couple of hundred bytes per notifier allocating one for every
potential queue is very wasteful as most devices only have a few
queues. Instead of having this handled statically dynamically assign
them and track in a GPtrArray.
[AJB: it's hard to trigger the vhost notifiers code, I assume as it
requires a KVM guest with appropriate backend]
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20220321153037.3622127-14-alex.bennee@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Previously we would silently suppress VHOST_USER_PROTOCOL_F_CONFIG
during the protocol negotiation if the QEMU stub hadn't implemented
the vhost_dev_config_notifier. However this isn't the only way we can
handle config messages, the existing vdc->get/set_config can do this
as well.
Lightly re-factor the code to check for both potential methods and
instead of silently squashing the feature error out. It is unlikely
that a vhost-user backend expecting to handle CONFIG messages will
behave correctly if they never get sent.
Fixes: 1c3e5a2617 ("vhost-user: back SET/GET_CONFIG requests with a protocol feature")
Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20220321153037.3622127-13-alex.bennee@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
These are useful when trying to debug the initial vhost-user
negotiation, especially when it hard to get logging from the low level
library on the other side.
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
Message-Id: <20220321153037.3622127-4-alex.bennee@linaro.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
The correct name of the macro is TARGET_PPC64.
Fixes: 27598393a2 ("Lift max memory slots limit imposed by vhost-user")
Reported-by: Fabiano Rosas <farosas@linux.ibm.com>
Signed-off-by: Murilo Opsfelder Araujo <muriloo@linux.ibm.com>
Cc: Raphael Norwitz <raphael.norwitz@nutanix.com>
Cc: Peter Turschmid <peter.turschm@nutanix.com>
Reviewed-by: Daniel Henrique Barboza <danielhb413@gmail.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <20220503180108.34506-1-muriloo@linux.ibm.com>
Signed-off-by: Daniel Henrique Barboza <danielhb413@gmail.com>
The spec clarifies now that QEMU should not send a file descriptor in a
request to remove a memory region. Change it accordingly.
For libvhost-user, this is a bug fix that makes it compatible with
rust-vmm's implementation that doesn't send a file descriptor. Keep
accepting, but ignoring a file descriptor for compatibility with older
QEMU versions.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220407133657.155281-4-kwolf@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@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>
Replace the global variables with inlined helper functions. getpagesize() is very
likely annotated with a "const" function attribute (at least with glibc), and thus
optimization should apply even better.
This avoids the need for a constructor initialization too.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-Id: <20220323155743.1585078-12-marcandre.lureau@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
When vhost-user device cleanup, remove notifier MR and munmaps notifier
address in the event-handling thread, VM CPU thread writing the notifier
in concurrent fails with an error of accessing invalid address. It
happens because MR is still being referenced and accessed in another
thread while the underlying notifier mmap address is being freed and
becomes invalid.
This patch calls RCU and munmap notifiers in the callback after the
memory flatview update finish.
Fixes: 44866521bd ("vhost-user: support registering external host notifiers")
Cc: qemu-stable@nongnu.org
Signed-off-by: Xueming Li <xuemingl@nvidia.com>
Message-Id: <20220207071929.527149-3-xuemingl@nvidia.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Notifier set when vhost-user backend asks qemu to mmap an FD and
offset. When vhost-user backend restart or getting killed, VQ notifier
FD and mmap addresses become invalid. After backend restart, MR contains
the invalid address will be restored and fail on notifier access.
On the other hand, qemu should munmap the notifier, release underlying
hardware resources to enable backend restart and allocate hardware
notifier resources correctly.
Qemu shouldn't reference and use resources of disconnected backend.
This patch removes VQ notifier restore, uses the default vhost-user
notifier to avoid invalid address access.
After backend restart, the backend should ask qemu to install a hardware
notifier if needed.
Fixes: 44866521bd ("vhost-user: support registering external host notifiers")
Cc: qemu-stable@nongnu.org
Signed-off-by: Xueming Li <xuemingl@nvidia.com>
Message-Id: <20220207071929.527149-2-xuemingl@nvidia.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
VhostOps methods in user_ops are not very consistent in their error
returns: some return negated errno while others just -1.
Make sure all of them consistently return negated errno. This also
helps error propagation from the functions being called inside.
Besides, this synchronizes the error return convention with the other
two vhost backends, kernel and vdpa, and will therefore allow for
consistent error propagation in the generic vhost code (in a followup
patch).
Signed-off-by: Roman Kagan <rvkagan@yandex-team.ru>
Message-Id: <20211111153354.18807-9-rvkagan@yandex-team.ru>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
In case of device resume after suspend, VQ notifier MR still valid.
Duplicated registrations explode memory block list and slow down device
resume.
Fixes: 44866521bd ("vhost-user: support registering external host notifiers")
Cc: tiwei.bie@intel.com
Cc: qemu-stable@nongnu.org
Cc: Yuwei Zhang <zhangyuwei.9149@bytedance.com>
Signed-off-by: Xueming Li <xuemingl@nvidia.com>
Message-Id: <20211008080215.590292-1-xuemingl@nvidia.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Qemu will crash on vhost backend unexpected exit and re-connect │
in some case due to access released memory.
Signed-off-by: Yuwei Zhang <zhangyuwei.9149@bytedance.com>
Message-Id: <20210830123433.45727-1-zhangyuwei.9149@bytedance.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This would previously give error messages like
> Received unexpected msg type.Expected 0 received 1
Signed-off-by: Alyssa Ross <hi@alyssa.is>
Message-Id: <20210806143926.315725-1-hi@alyssa.is>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
On vhost-user-blk migration, qemu normally sends a number of commands
to enable logging if VHOST_USER_PROTOCOL_F_LOG_SHMFD is negotiated.
Qemu sends VHOST_USER_SET_FEATURES to enable buffers logging and
VHOST_USER_SET_VRING_ADDR per each started ring to enable "used ring"
data logging.
The issue is that qemu doesn't wait for reply from the vhost daemon
for these commands which may result in races between qemu expectation
of logging starting and actual login starting in vhost daemon.
The race can appear as follows: on migration setup, qemu enables dirty page
logging by sending VHOST_USER_SET_FEATURES. The command doesn't arrive to a
vhost-user-blk daemon immediately and the daemon needs some time to turn the
logging on internally. If qemu doesn't wait for reply, after sending the
command, qemu may start migrateing memory pages to a destination. At this time,
the logging may not be actually turned on in the daemon but some guest pages,
which the daemon is about to write to, may have already been transferred
without logging to the destination. Since the logging wasn't turned on,
those pages won't be transferred again as dirty. So we may end up with
corrupted data on the destination.
The same scenario is applicable for "used ring" data logging, which is
turned on with VHOST_USER_SET_VRING_ADDR command.
To resolve this issue, this patch makes qemu wait for the command result
explicitly if VHOST_USER_PROTOCOL_F_REPLY_ACK is negotiated and logging enabled.
Signed-off-by: Denis Plotnikov <den-plotnikov@yandex-team.ru>
Message-Id: <20210809104824.78830-1-den-plotnikov@yandex-team.ru>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
If call virtio_queue_set_host_notifier_mr fails, should free
host-notifier memory-region.
Fixes: 44866521bd ("vhost-user: support registering external host notifiers")
Signed-off-by: Yajun Wu <yajunw@nvidia.com>
Message-Id: <1629077555-19907-1-git-send-email-yajunw@nvidia.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vhost_user_backend_init() can fail without setting an error. Unclean.
Its caller vhost_dev_init() compensates by substituting a generic
error then. Goes back to commit 28770ff935 "vhost: Distinguish errors
in vhost_backend_init()".
Clean up by moving the generic error from vhost_dev_init() to all the
failure paths that neglect to set an error.
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20210720125408.387910-14-armbru@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
vhost_user_get_config() can fail without setting an error. Unclean.
Its caller vhost_dev_get_config() compensates by substituting a
generic error then. Goes back to commit 50de51387f "vhost:
Distinguish errors in vhost_dev_get_config()".
Clean up by moving the generic error from vhost_dev_get_config() to
all the failure paths that neglect to set an error.
Cc: Kevin Wolf <kwolf@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Markus Armbruster <armbru@redhat.com>
Message-Id: <20210720125408.387910-13-armbru@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
[Sign of error_setg_errno()'s second argument fixed in both calls]
Since commit 9894dc0cdc "char: convert
from GIOChannel to QIOChannel", the first argument to the watch callback
can actually be a QIOChannel, which is not a GIOChannel (but a QEMU
Object).
Even though we never used that pointer, change the callback type to warn
the users. Possibly a better fix later, we may want to store the
callback and call it from intermediary functions.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
dev->max_queues was never initialised for backends that don't support
VHOST_USER_PROTOCOL_F_MQ, so it would use 0 as the maximum number of
queues to check against and consequently fail for any such backend.
Set it to 1 if the backend doesn't have multiqueue support.
Fixes: c90bd505a3
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210705171429.29286-1-kwolf@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of just returning 0/-1 and letting the caller make up a
meaningless error message, add an Error parameter to allow reporting the
real error and switch to 0/-errno so that different kind of errors can
be distinguished in the caller.
config_len in vhost_user_get_config() is defined by the device, so if
it's larger than VHOST_USER_MAX_CONFIG_SIZE, this is a programming
error. Turn the corresponding check into an assertion.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210609154658.350308-6-kwolf@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of just returning 0/-1 and letting the caller make up a
meaningless error message, switch to 0/-errno so that different kinds of
errors can be distinguished in the caller.
This involves changing a few more callbacks in VhostOps to return
0/-errno: .vhost_set_owner(), .vhost_get_features() and
.vhost_virtqueue_set_busyloop_timeout(). The implementations of these
functions are trivial as they generally just send a message to the
backend.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210609154658.350308-4-kwolf@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Instead of just returning 0/-1 and letting the caller make up a
meaningless error message, add an Error parameter to allow reporting the
real error and switch to 0/-errno so that different kind of errors can
be distinguished in the caller.
Specifically, in vhost-user, EPROTO is used for all errors that relate
to the connection itself, whereas other error codes are used for errors
relating to the content of the connection. This will allow us later to
automatically reconnect when the connection goes away, without ending up
in an endless loop if it's a permanent error in the configuration.
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20210609154658.350308-3-kwolf@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Creating a device with a number of queues that isn't supported by the
backend is pointless, the device won't work properly and the error
messages are rather confusing.
Just fail to create the device if num-queues is higher than what the
backend supports.
Since the relationship between num-queues and the number of virtqueues
depends on the specific device, this is an additional value that needs
to be initialised by the device. For convenience, allow leaving it 0 if
the check should be skipped. This makes sense for vhost-user-net where
separate vhost devices are used for the queues and custom initialisation
code is needed to perform the check.
Fixes: https://bugzilla.redhat.com/show_bug.cgi?id=1935031
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <20210429171316.162022-7-kwolf@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
Now that everything is in place, have the nested event loop to monitor
the slave channel. The source in the main event loop is destroyed and
recreated to ensure any pending even for the slave channel that was
previously detected is purged. This guarantees that the main loop
wont invoke slave_read() based on an event that was already handled
by the nested loop.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20210312092212.782255-7-groug@kaod.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
A deadlock condition potentially exists if a vhost-user process needs
to request something to QEMU on the slave channel while processing a
vhost-user message.
This doesn't seem to affect any vhost-user implementation so far, but
this is currently biting the upcoming enablement of DAX with virtio-fs.
The issue is being observed when the guest does an emergency reboot while
a mapping still exits in the DAX window, which is very easy to get with
a busy enough workload (e.g. as simulated by blogbench [1]) :
- QEMU sends VHOST_USER_GET_VRING_BASE to virtiofsd.
- In order to complete the request, virtiofsd then asks QEMU to remove
the mapping on the slave channel.
All these dialogs are synchronous, hence the deadlock.
As pointed out by Stefan Hajnoczi:
When QEMU's vhost-user master implementation sends a vhost-user protocol
message, vhost_user_read() does a "blocking" read during which slave_fd
is not monitored by QEMU.
The natural solution for this issue is an event loop. The main event
loop cannot be nested though since we have no guarantees that its
fd handlers are prepared for re-entrancy.
Introduce a new event loop that only monitors the chardev I/O for now
in vhost_user_read() and push the actual reading to a one-shot handler.
A subsequent patch will teach the loop to monitor and process messages
from the slave channel as well.
[1] https://github.com/jedisct1/Blogbench
Suggested-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20210312092212.782255-6-groug@kaod.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
The slave channel is implemented with socketpair() : QEMU creates
the pair, passes one of the socket to virtiofsd and monitors the
other one with the main event loop using qemu_set_fd_handler().
In order to fix a potential deadlock between QEMU and a vhost-user
external process (e.g. virtiofsd with DAX), we want to be able to
monitor and service the slave channel while handling vhost-user
requests.
Prepare ground for this by converting the slave channel to be a
QIOChannelSocket. This will make monitoring of the slave channel
as simple as calling qio_channel_add_watch_source(). Since the
connection is already established between the two sockets, only
incoming I/O (G_IO_IN) and disconnect (G_IO_HUP) need to be
serviced.
This also allows to get rid of the ancillary data parsing since
QIOChannelSocket can do this for us. Note that the MSG_CTRUNC
check is dropped on the way because QIOChannelSocket ignores this
case. This isn't a problem since slave_read() provisions space for
8 file descriptors, but affected vhost-user slave protocol messages
generally only convey one. If for some reason a buggy implementation
passes more file descriptors, no need to break the connection, just
like we don't break it if some other type of ancillary data is
received : this isn't explicitely violating the protocol per-se so
it seems better to ignore it.
The current code errors out on short reads and writes. Use the
qio_channel_*_all() variants to address this on the way.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20210312092212.782255-5-groug@kaod.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20210312092212.782255-4-groug@kaod.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Some message types, e.g. VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
can convey file descriptors. These must be closed before returning
from slave_read() to avoid being leaked. This can currently be done
in two different places:
[1] just after the request has been processed
[2] on the error path, under the goto label err:
These path are supposed to be mutually exclusive but they are not
actually. If the VHOST_USER_NEED_REPLY_MASK flag was passed and the
sending of the reply fails, both [1] and [2] are performed with the
same descriptor values. This can potentially cause subtle bugs if one
of the descriptor was recycled by some other thread in the meantime.
This code duplication complicates rollback for no real good benefit.
Do the closing in a unique place, under a new fdcleanup: goto label
at the end of the function.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20210312092212.782255-3-groug@kaod.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
slave_read() checks EAGAIN when reading or writing to the socket
fails. This gives the impression that the slave channel is in
non-blocking mode, which is certainly not the case with the current
code base. And the rest of the code isn't actually ready to cope
with non-blocking I/O.
Just drop the checks everywhere in this function for the sake of
clarity.
Signed-off-by: Greg Kurz <groug@kaod.org>
Message-Id: <20210312092212.782255-2-groug@kaod.org>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
QEMU currently truncates the mmap_offset field when sending
VHOST_USER_ADD_MEM_REG and VHOST_USER_REM_MEM_REG messages. The struct
layout looks like this:
typedef struct VhostUserMemoryRegion {
uint64_t guest_phys_addr;
uint64_t memory_size;
uint64_t userspace_addr;
uint64_t mmap_offset;
} VhostUserMemoryRegion;
typedef struct VhostUserMemRegMsg {
uint32_t padding;
/* WARNING: there is a 32-bit hole here! */
VhostUserMemoryRegion region;
} VhostUserMemRegMsg;
The payload size is calculated as follows when sending the message in
hw/virtio/vhost-user.c:
msg->hdr.size = sizeof(msg->payload.mem_reg.padding) +
sizeof(VhostUserMemoryRegion);
This calculation produces an incorrect result of only 36 bytes.
sizeof(VhostUserMemRegMsg) is actually 40 bytes.
The consequence of this is that the final field, mmap_offset, is
truncated. This breaks x86_64 TCG guests on s390 hosts. Other guest/host
combinations may get lucky if either of the following holds:
1. The guest memory layout does not need mmap_offset != 0.
2. The host is little-endian and mmap_offset <= 0xffffffff so the
truncation has no effect.
Fix this by extending the existing 32-bit padding field to 64-bit. Now
the padding reflects the actual compiler padding. This can be verified
using pahole(1).
Also document the layout properly in the vhost-user specification. The
vhost-user spec did not document the exact layout. It would be
impossible to implement the spec without looking at the QEMU source
code.
Existing vhost-user frontends and device backends continue to work after
this fix has been applied. The only change in the wire protocol is that
QEMU now sets hdr.size to 40 instead of 36. If a vhost-user
implementation has a hardcoded size check for 36 bytes, then it will
fail with new QEMUs. Both QEMU and DPDK/SPDK don't check the exact
payload size, so they continue to work.
Fixes: f1aeb14b08 ("Transmit vhost-user memory regions individually")
Cc: Raphael Norwitz <raphael.norwitz@nutanix.com>
Cc: Cornelia Huck <cohuck@redhat.com>
Cc: Michael S. Tsirkin <mst@redhat.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20201109174355.1069147-1-stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Fixes: f1aeb14b08 ("Transmit vhost-user memory regions individually")
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
The VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS vhost-user protocol
feature introduced a shadow-table, used by the backend to dynamically
determine how a vdev's memory regions have changed since the last
vhost_user_set_mem_table() call. On hot-remove, a memmove() operation
is used to overwrite the removed shadow region descriptor(s). The size
parameter of this memmove was off by 1 such that if a VM with a backend
supporting the VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS filled it's
shadow-table (by performing the maximum number of supported hot-add
operatons) and attempted to remove the last region, Qemu would read an
out of bounds value and potentially crash.
This change fixes the memmove() bounds such that this erroneous read can
never happen.
Signed-off-by: Peter Turschmid <peter.turschm@nutanix.com>
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1594799958-31356-1-git-send-email-raphael.norwitz@nutanix.com>
Fixes: f1aeb14b08 ("Transmit vhost-user memory regions individually")
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: qemu-stable@nongnu.org
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Prior to this change, the vhost_user_fill_msg_region function filled out
all elements of the VhostUserMemoryRegion struct except the mmap_offset.
This function is often called on uninitialized structs, which are then
copied into VHOST_USER_SET_MEM_TABLE and VHOST_USER_ADD/REM_MEM_REG
messages. In some cases, where the mmap_offset was not needed, it was
left uninitialized, causing QEMU to send the backend uninitialized data,
which Coverity flagged as a series of issues.
This change augments the vhost_user_fill_msg_region API, adding a
mmap_offset paramenter, forcing the caller to initialize mmap_offset.
Fixes: ece99091c2
Fixes: f1aeb14b08
Reported-by: Coverity (CIDs 1429802, 1429803 and 1429804)
Suggested-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1592650156-25845-1-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Historically, sending all memory regions to vhost-user backends in a
single message imposed a limitation on the number of times memory
could be hot-added to a VM with a vhost-user device. Now that backends
which support the VHOST_USER_PROTOCOL_F_CONFIGURE_SLOTS send memory
regions individually, we no longer need to impose this limitation on
devices which support this feature.
With this change, VMs with a vhost-user device which supports the
VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS can support a configurable
number of memory slots, up to the maximum allowed by the target
platform.
Existing backends which do not support
VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS are unaffected.
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Peter Turschmid <peter.turschm@nutanix.com>
Suggested-by: Mike Cui <cui@nutanix.com>
Message-Id: <1588533678-23450-6-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
With this change, when the VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
protocol feature has been negotiated, Qemu no longer sends the backend
all the memory regions in a single message. Rather, when the memory
tables are set or updated, a series of VHOST_USER_ADD_MEM_REG and
VHOST_USER_REM_MEM_REG messages are sent to transmit the regions to map
and/or unmap instead of sending send all the regions in one fixed size
VHOST_USER_SET_MEM_TABLE message.
The vhost_user struct maintains a shadow state of the VM’s memory
regions. When the memory tables are modified, the
vhost_user_set_mem_table() function compares the new device memory state
to the shadow state and only sends regions which need to be unmapped or
mapped in. The regions which must be unmapped are sent first, followed
by the new regions to be mapped in. After all the messages have been
sent, the shadow state is set to the current virtual device state.
Existing backends which do not support
VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS are unaffected.
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Swapnil Ingle <swapnil.ingle@nutanix.com>
Signed-off-by: Peter Turschmid <peter.turschm@nutanix.com>
Suggested-by: Mike Cui <cui@nutanix.com>
Message-Id: <1588533678-23450-5-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Marc-André Lureau <marcandre.lureau@redhat.com>
This change introduces a new feature to the vhost-user protocol allowing
a backend device to specify the maximum number of ram slots it supports.
At this point, the value returned by the backend will be capped at the
maximum number of ram slots which can be supported by vhost-user, which
is currently set to 8 because of underlying protocol limitations.
The returned value will be stored inside the VhostUserState struct so
that on device reconnect we can verify that the ram slot limitation
has not decreased since the last time the device connected.
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Peter Turschmid <peter.turschm@nutanix.com>
Message-Id: <1588533678-23450-4-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
When setting the memory tables, qemu uses a memory region's userspace
address to look up the region's MemoryRegion struct. Among other things,
the MemoryRegion contains the region's offset and associated file
descriptor, all of which need to be sent to the backend.
With VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS, this logic will be
needed in multiple places, so before feature support is added it
should be moved to a helper function.
This helper is also used to simplify the vhost_user_can_merge()
function.
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1588533678-23450-3-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
When setting vhost-user memory tables, memory region descriptors must be
copied from the vhost_dev struct to the vhost-user message. To avoid
duplicating code in setting the memory tables, we should use a helper to
populate this field. This change adds this helper.
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1588533678-23450-2-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
vhost_user_set_mem_table() and vhost_user_set_mem_table_postcopy() have
gotten convoluted, and have some identical code.
This change moves the logic populating the VhostUserMemory struct and
fds array from vhost_user_set_mem_table() and
vhost_user_set_mem_table_postcopy() to a new function,
vhost_user_fill_set_mem_table_msg().
No functionality is impacted.
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Peter Turschmid <peter.turschm@nutanix.com>
Message-Id: <1585132506-13316-1-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The current vhost_user_set_mem_table_postcopy() implementation
populates each region of the VHOST_USER_SET_MEM_TABLE message without
first checking if there are more than VHOST_MEMORY_MAX_NREGIONS already
populated. This can cause memory corruption if too many regions are
added to the message during the postcopy step.
This change moves an existing assert up such that attempting to
construct a VHOST_USER_SET_MEM_TABLE message with too many memory
regions will gracefully bring down qemu instead of corrupting memory.
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Signed-off-by: Peter Turschmid <peter.turschm@nutanix.com>
Message-Id: <1579143426-18305-2-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When multiqueue is enabled, a vhost_dev is created for each queue
pair. However, only one slave channel is needed.
Fixes: 4bbeeba023 (vhost-user: add slave-req-fd support)
Cc: marcandre.lureau@redhat.com
Signed-off-by: Adrian Moreno <amorenoz@redhat.com>
Message-Id: <20200121214553.28459-1-amorenoz@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When we receive an unexpected message type on the slave fd, print
the type.
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Add a VHOST_USER_RESET_DEVICE message which will reset the vhost user
backend. Disabling all rings, and resetting all internal state, ready
for the backend to be reinitialized.
A backend has to report it supports this features with the
VHOST_USER_PROTOCOL_F_RESET_DEVICE protocol feature bit. If it does
so, the new message is used instead of sending a RESET_OWNER which has
had inconsistent implementations.
Signed-off-by: David Vrabel <david.vrabel@nutanix.com>
Signed-off-by: Raphael Norwitz <raphael.norwitz@nutanix.com>
Message-Id: <1572385083-5254-2-git-send-email-raphael.norwitz@nutanix.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
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>
Add a new vhost-user message to give a unix socket to a vhost-user
backend for GPU display updates.
Back when I started that work, I added a new GPU channel because the
vhost-user protocol wasn't bidirectional. Since then, there is a
vhost-user-slave channel for the slave to send requests to the master.
We could extend it with GPU messages. However, the GPU protocol is
quite orthogonal to vhost-user, thus I chose to have a new dedicated
channel.
See vhost-user-gpu.rst for the protocol details.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Message-id: 20190524130946.31736-2-marcandre.lureau@redhat.com
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
This patch introduces two new messages VHOST_USER_GET_INFLIGHT_FD
and VHOST_USER_SET_INFLIGHT_FD to support transferring a shared
buffer between qemu and backend.
Firstly, qemu uses VHOST_USER_GET_INFLIGHT_FD to get the
shared buffer from backend. Then qemu should send it back
through VHOST_USER_SET_INFLIGHT_FD each time we start vhost-user.
This shared buffer is used to track inflight I/O by backend.
Qemu should retrieve a new one when vm reset.
Signed-off-by: Xie Yongji <xieyongji@baidu.com>
Signed-off-by: Chai Wen <chaiwen@baidu.com>
Signed-off-by: Zhang Yu <zhangyu31@baidu.com>
Message-Id: <20190228085355.9614-2-xieyongji@baidu.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Split vhost_user_read(), so only header can be read with
vhost_user_read_header().
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Message-Id: <20190308140454.32437-8-marcandre.lureau@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Message-Id: <20190308140454.32437-6-marcandre.lureau@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Take a VhostUserState* that can be pre-allocated, and initialize it
with the associated chardev.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Reviewed-by: Tiwei Bie <tiwei.bie@intel.com>
Message-Id: <20190308140454.32437-4-marcandre.lureau@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vhost-user does not depend on Linux; it can run on any POSIX system. Restrict
vhost-kernel to Linux in hw/virtio/vhost-backend.c, everything else can be
compiled on all POSIX systems.
Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com>
Reviewed-by: Thomas Huth <thuth@redhat.com>
Message-Id: <1543851204-41186-4-git-send-email-pbonzini@redhat.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Message-Id: <1550165756-21617-4-git-send-email-pbonzini@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Alex Bennée
Murilo Opsfelder Araujo
Kevin Wolf
Marc-André Lureau
Xueming Li
Roman Kagan
Yuwei Zhang
Alyssa Ross
Denis Plotnikov
Yajun Wu
Markus Armbruster
Greg Kurz
Stefan Hajnoczi
Chen Qun
Raphael Norwitz
Adrian Moreno
Dr. David Alan Gilbert
Xie Yongji
Paolo Bonzini