Merge tag 'pull-vfio-20250626' of https://github.com/legoater/qemu into staging

vfio queue:

* Added several small fixes and cleanups
* Added support for vfio-user client device

# -----BEGIN PGP SIGNATURE-----
#
# iQIzBAABCAAdFiEEoPZlSPBIlev+awtgUaNDx8/77KEFAmhc+mwACgkQUaNDx8/7
# 7KHQShAAwGTjc6yzBaFr9DTKaL6Vszwby5tYdV3vWTsnFj2zBks+5BswohxbnYdk
# Smy5E/F+yCiHI6X4ohI4FRjJgBJplDxZ6bbEgTwZa6ADs7xWAwFWv349KQY9uLpJ
# XV/ACot2b6FZUskv3w1SxQrpzho4ICm3DuLOdAFAvBPRtxyC2aQMLsXTlT+7+5cC
# X8zJB/9mtjWIomYWKRXnYUP/uM1g7QLtyU7d01szvqCfSVUilVlg6Ys7RxnqLG0k
# A1/kxYOrEPHHxMO+YwFuapIfE8Gqihes2K1GfM871JaBT14dMIAZkajmVasbKD16
# Iljz89nEV3UehDP9HADhx3QuXO7fhJ3cxcHvTH0xhUeoks3EgTlUq0VNRRYzu6rQ
# 3P1E3cVaPTmwfoSrhecNIFcln4v/bENdwzYcjh96r9fcFwE+ro4oUTGNKCPYv2t0
# yOoc6PqgiZN7DM89/N2hcesgOun7oOVpMnKhiqHjVe53HoM8bfLojWECKNq9Cz1u
# m0YEHn2gEuEB5l03IguRnAywZq76Jivd6WFmAeXGrHRZ9sfxQCwvImbqMa7QxYpI
# rt+j7RAyP57WVoBPoW8hlaIQmLuIvIgdWwWkwQd2BTIprLpdHJd4SWkL6eqGozpE
# rsaHw+WQZqFoddrl7EUSVY/Z2CfIRr1g/Zo5z4RU9YLtxVxjSPw=
# =sX2P
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 26 Jun 2025 03:44:44 EDT
# gpg:                using RSA key A0F66548F04895EBFE6B0B6051A343C7CFFBECA1
# gpg: Good signature from "Cédric Le Goater <clg@redhat.com>" [full]
# gpg:                 aka "Cédric Le Goater <clg@kaod.org>" [full]
# Primary key fingerprint: A0F6 6548 F048 95EB FE6B  0B60 51A3 43C7 CFFB ECA1

* tag 'pull-vfio-20250626' of https://github.com/legoater/qemu: (25 commits)
  vfio-user: introduce vfio-user protocol specification
  docs: add vfio-user documentation
  vfio-user: add coalesced posted writes
  vfio-user: support posted writes
  vfio-user: add 'x-msg-timeout' option
  vfio-user: implement VFIO_USER_DMA_READ/WRITE
  vfio-user: implement VFIO_USER_DMA_MAP/UNMAP
  vfio-user: implement VFIO_USER_DEVICE_RESET
  vfio-user: set up container access to the proxy
  vfio-user: forward MSI-X PBA BAR accesses to server
  vfio-user: implement VFIO_USER_DEVICE_GET/SET_IRQ*
  vfio-user: set up PCI in vfio_user_pci_realize()
  vfio-user: implement VFIO_USER_REGION_READ/WRITE
  vfio-user: implement VFIO_USER_DEVICE_GET_REGION_INFO
  vfio-user: implement VFIO_USER_DEVICE_GET_INFO
  vfio-user: implement message send infrastructure
  vfio-user: implement message receive infrastructure
  vfio-user: connect vfio proxy to remote server
  vfio-user: add vfio-user class and container
  vfio/container: fails mdev hotplug if add migration blocker failed
  ...

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>

31 files changed, 4717 insertions, 15 deletions

diff --git a/MAINTAINERS b/MAINTAINERS
index ebb8af7..d1672fd 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -4258,13 +4258,22 @@ F: hw/remote/proxy-memory-listener.c
 F: include/hw/remote/proxy-memory-listener.h
 F: hw/remote/iohub.c
 F: include/hw/remote/iohub.h
-F: subprojects/libvfio-user
 F: hw/remote/vfio-user-obj.c
 F: include/hw/remote/vfio-user-obj.h
 F: hw/remote/iommu.c
 F: include/hw/remote/iommu.h
 F: tests/functional/test_multiprocess.py
 
+VFIO-USER:
+M: John Levon <john.levon@nutanix.com>
+M: Thanos Makatos <thanos.makatos@nutanix.com>
+S: Supported
+F: docs/interop/vfio-user.rst
+F: docs/system/devices/vfio-user.rst
+F: hw/vfio-user/*
+F: include/hw/vfio-user/*
+F: subprojects/libvfio-user
+
 EBPF:
 M: Jason Wang <jasowang@redhat.com>
 R: Andrew Melnychenko <andrew@daynix.com>
diff --git a/docs/interop/index.rst b/docs/interop/index.rst
index 972f3e4..d830c5c 100644
--- a/docs/interop/index.rst
+++ b/docs/interop/index.rst
@@ -25,6 +25,7 @@ are useful for making QEMU interoperate with other software.
    qemu-ga-ref
    qemu-qmp-ref
    qemu-storage-daemon-qmp-ref
+   vfio-user
    vhost-user
    vhost-user-gpu
    vhost-vdpa
diff --git a/docs/interop/vfio-user.rst b/docs/interop/vfio-user.rst
new file mode 100644
index 0000000..0b06f02
--- /dev/null
+++ b/docs/interop/vfio-user.rst
@@ -0,0 +1,1520 @@
+.. include:: <isonum.txt>
+.. SPDX-License-Identifier: GPL-2.0-or-later
+
+================================
+vfio-user Protocol Specification
+================================
+
+.. contents:: Table of Contents
+
+Introduction
+============
+vfio-user is a protocol that allows a device to be emulated in a separate
+process outside of a Virtual Machine Monitor (VMM). vfio-user devices consist
+of a generic VFIO device type, living inside the VMM, which we call the client,
+and the core device implementation, living outside the VMM, which we call the
+server.
+
+The vfio-user specification is partly based on the
+`Linux VFIO ioctl interface <https://www.kernel.org/doc/html/latest/driver-api/vfio.html>`_.
+
+VFIO is a mature and stable API, backed by an extensively used framework. The
+existing VFIO client implementation in QEMU (``qemu/hw/vfio/``) can be largely
+re-used, though there is nothing in this specification that requires that
+particular implementation. None of the VFIO kernel modules are required for
+supporting the protocol, on either the client or server side. Some source
+definitions in VFIO are re-used for vfio-user.
+
+The main idea is to allow a virtual device to function in a separate process in
+the same host over a UNIX domain socket. A UNIX domain socket (``AF_UNIX``) is
+chosen because file descriptors can be trivially sent over it, which in turn
+allows:
+
+* Sharing of client memory for DMA with the server.
+* Sharing of server memory with the client for fast MMIO.
+* Efficient sharing of eventfd's for triggering interrupts.
+
+Other socket types could be used which allow the server to run in a separate
+guest in the same host (``AF_VSOCK``) or remotely (``AF_INET``). Theoretically
+the underlying transport does not necessarily have to be a socket, however we do
+not examine such alternatives. In this protocol version we focus on using a UNIX
+domain socket and introduce basic support for the other two types of sockets
+without considering performance implications.
+
+While passing of file descriptors is desirable for performance reasons, support
+is not necessary for either the client or the server in order to implement the
+protocol. There is always an in-band, message-passing fall back mechanism.
+
+Overview
+========
+
+VFIO is a framework that allows a physical device to be securely passed through
+to a user space process; the device-specific kernel driver does not drive the
+device at all.  Typically, the user space process is a VMM and the device is
+passed through to it in order to achieve high performance. VFIO provides an API
+and the required functionality in the kernel. QEMU has adopted VFIO to allow a
+guest to directly access physical devices, instead of emulating them in
+software.
+
+vfio-user reuses the core VFIO concepts defined in its API, but implements them
+as messages to be sent over a socket. It does not change the kernel-based VFIO
+in any way, in fact none of the VFIO kernel modules need to be loaded to use
+vfio-user. It is also possible for the client to concurrently use the current
+kernel-based VFIO for one device, and vfio-user for another device.
+
+VFIO Device Model
+-----------------
+
+A device under VFIO presents a standard interface to the user process. Many of
+the VFIO operations in the existing interface use the ``ioctl()`` system call, and
+references to the existing interface are called the ``ioctl()`` implementation in
+this document.
+
+The following sections describe the set of messages that implement the vfio-user
+interface over a socket. In many cases, the messages are analogous to data
+structures used in the ``ioctl()`` implementation. Messages derived from the
+``ioctl()`` will have a name derived from the ``ioctl()`` command name.  E.g., the
+``VFIO_DEVICE_GET_INFO`` ``ioctl()`` command becomes a
+``VFIO_USER_DEVICE_GET_INFO`` message.  The purpose of this reuse is to share as
+much code as feasible with the ``ioctl()`` implementation``.
+
+Connection Initiation
+^^^^^^^^^^^^^^^^^^^^^
+
+After the client connects to the server, the initial client message is
+``VFIO_USER_VERSION`` to propose a protocol version and set of capabilities to
+apply to the session. The server replies with a compatible version and set of
+capabilities it supports, or closes the connection if it cannot support the
+advertised version.
+
+Device Information
+^^^^^^^^^^^^^^^^^^
+
+The client uses a ``VFIO_USER_DEVICE_GET_INFO`` message to query the server for
+information about the device. This information includes:
+
+* The device type and whether it supports reset (``VFIO_DEVICE_FLAGS_``),
+* the number of device regions, and
+* the device presents to the client the number of interrupt types the device
+  supports.
+
+Region Information
+^^^^^^^^^^^^^^^^^^
+
+The client uses ``VFIO_USER_DEVICE_GET_REGION_INFO`` messages to query the
+server for information about the device's regions. This information describes:
+
+* Read and write permissions, whether it can be memory mapped, and whether it
+  supports additional capabilities (``VFIO_REGION_INFO_CAP_``).
+* Region index, size, and offset.
+
+When a device region can be mapped by the client, the server provides a file
+descriptor which the client can ``mmap()``. The server is responsible for
+polling for client updates to memory mapped regions.
+
+Region Capabilities
+"""""""""""""""""""
+
+Some regions have additional capabilities that cannot be described adequately
+by the region info data structure. These capabilities are returned in the
+region info reply in a list similar to PCI capabilities in a PCI device's
+configuration space.
+
+Sparse Regions
+""""""""""""""
+A region can be memory-mappable in whole or in part. When only a subset of a
+region can be mapped by the client, a ``VFIO_REGION_INFO_CAP_SPARSE_MMAP``
+capability is included in the region info reply. This capability describes
+which portions can be mapped by the client.
+
+.. Note::
+   For example, in a virtual NVMe controller, sparse regions can be used so
+   that accesses to the NVMe registers (found in the beginning of BAR0) are
+   trapped (an infrequent event), while allowing direct access to the doorbells
+   (an extremely frequent event as every I/O submission requires a write to
+   BAR0), found in the next page after the NVMe registers in BAR0.
+
+Device-Specific Regions
+"""""""""""""""""""""""
+
+A device can define regions additional to the standard ones (e.g. PCI indexes
+0-8). This is achieved by including a ``VFIO_REGION_INFO_CAP_TYPE`` capability
+in the region info reply of a device-specific region. Such regions are reflected
+in ``struct vfio_user_device_info.num_regions``. Thus, for PCI devices this
+value can be equal to, or higher than, ``VFIO_PCI_NUM_REGIONS``.
+
+Region I/O via file descriptors
+-------------------------------
+
+For unmapped regions, region I/O from the client is done via
+``VFIO_USER_REGION_READ/WRITE``.  As an optimization, ioeventfds or ioregionfds
+may be configured for sub-regions of some regions. A client may request
+information on these sub-regions via ``VFIO_USER_DEVICE_GET_REGION_IO_FDS``; by
+configuring the returned file descriptors as ioeventfds or ioregionfds, the
+server can be directly notified of I/O (for example, by KVM) without taking a
+trip through the client.
+
+Interrupts
+^^^^^^^^^^
+
+The client uses ``VFIO_USER_DEVICE_GET_IRQ_INFO`` messages to query the server
+for the device's interrupt types. The interrupt types are specific to the bus
+the device is attached to, and the client is expected to know the capabilities
+of each interrupt type. The server can signal an interrupt by directly injecting
+interrupts into the guest via an event file descriptor. The client configures
+how the server signals an interrupt with ``VFIO_USER_SET_IRQS`` messages.
+
+Device Read and Write
+^^^^^^^^^^^^^^^^^^^^^
+
+When the guest executes load or store operations to an unmapped device region,
+the client forwards these operations to the server with
+``VFIO_USER_REGION_READ`` or ``VFIO_USER_REGION_WRITE`` messages. The server
+will reply with data from the device on read operations or an acknowledgement on
+write operations. See `Read and Write Operations`_.
+
+Client memory access
+--------------------
+
+The client uses ``VFIO_USER_DMA_MAP`` and ``VFIO_USER_DMA_UNMAP`` messages to
+inform the server of the valid DMA ranges that the server can access on behalf
+of a device (typically, VM guest memory). DMA memory may be accessed by the
+server via ``VFIO_USER_DMA_READ`` and ``VFIO_USER_DMA_WRITE`` messages over the
+socket. In this case, the "DMA" part of the naming is a misnomer.
+
+Actual direct memory access of client memory from the server is possible if the
+client provides file descriptors the server can ``mmap()``. Note that ``mmap()``
+privileges cannot be revoked by the client, therefore file descriptors should
+only be exported in environments where the client trusts the server not to
+corrupt guest memory.
+
+See `Read and Write Operations`_.
+
+Client/server interactions
+==========================
+
+Socket
+------
+
+A server can serve:
+
+1) one or more clients, and/or
+2) one or more virtual devices, belonging to one or more clients.
+
+The current protocol specification requires a dedicated socket per
+client/server connection. It is a server-side implementation detail whether a
+single server handles multiple virtual devices from the same or multiple
+clients. The location of the socket is implementation-specific. Multiplexing
+clients, devices, and servers over the same socket is not supported in this
+version of the protocol.
+
+Authentication
+--------------
+
+For ``AF_UNIX``, we rely on OS mandatory access controls on the socket files,
+therefore it is up to the management layer to set up the socket as required.
+Socket types that span guests or hosts will require a proper authentication
+mechanism. Defining that mechanism is deferred to a future version of the
+protocol.
+
+Command Concurrency
+-------------------
+
+A client may pipeline multiple commands without waiting for previous command
+replies.  The server will process commands in the order they are received.  A
+consequence of this is if a client issues a command with the *No_reply* bit,
+then subsequently issues a command without *No_reply*, the older command will
+have been processed before the reply to the younger command is sent by the
+server.  The client must be aware of the device's capability to process
+concurrent commands if pipelining is used.  For example, pipelining allows
+multiple client threads to concurrently access device regions; the client must
+ensure these accesses obey device semantics.
+
+An example is a frame buffer device, where the device may allow concurrent
+access to different areas of video memory, but may have indeterminate behavior
+if concurrent accesses are performed to command or status registers.
+
+Note that unrelated messages sent from the server to the client can appear in
+between a client to server request/reply and vice versa.
+
+Implementers should be prepared for certain commands to exhibit potentially
+unbounded latencies.  For example, ``VFIO_USER_DEVICE_RESET`` may take an
+arbitrarily long time to complete; clients should take care not to block
+unnecessarily.
+
+Socket Disconnection Behavior
+-----------------------------
+The server and the client can disconnect from each other, either intentionally
+or unexpectedly. Both the client and the server need to know how to handle such
+events.
+
+Server Disconnection
+^^^^^^^^^^^^^^^^^^^^
+A server disconnecting from the client may indicate that:
+
+1) A virtual device has been restarted, either intentionally (e.g. because of a
+   device update) or unintentionally (e.g. because of a crash).
+2) A virtual device has been shut down with no intention to be restarted.
+
+It is impossible for the client to know whether or not a failure is
+intermittent or innocuous and should be retried, therefore the client should
+reset the VFIO device when it detects the socket has been disconnected.
+Error recovery will be driven by the guest's device error handling
+behavior.
+
+Client Disconnection
+^^^^^^^^^^^^^^^^^^^^
+The client disconnecting from the server primarily means that the client
+has exited. Currently, this means that the guest is shut down so the device is
+no longer needed therefore the server can automatically exit. However, there
+can be cases where a client disconnection should not result in a server exit:
+
+1) A single server serving multiple clients.
+2) A multi-process QEMU upgrading itself step by step, which is not yet
+   implemented.
+
+Therefore in order for the protocol to be forward compatible, the server should
+respond to a client disconnection as follows:
+
+ - all client memory regions are unmapped and cleaned up (including closing any
+   passed file descriptors)
+ - all IRQ file descriptors passed from the old client are closed
+ - the device state should otherwise be retained
+
+The expectation is that when a client reconnects, it will re-establish IRQ and
+client memory mappings.
+
+If anything happens to the client (such as qemu really did exit), the control
+stack will know about it and can clean up resources accordingly.
+
+Security Considerations
+-----------------------
+
+Speaking generally, vfio-user clients should not trust servers, and vice versa.
+Standard tools and mechanisms should be used on both sides to validate input and
+prevent against denial of service scenarios, buffer overflow, etc.
+
+Request Retry and Response Timeout
+----------------------------------
+A failed command is a command that has been successfully sent and has been
+responded to with an error code. Failure to send the command in the first place
+(e.g. because the socket is disconnected) is a different type of error examined
+earlier in the disconnect section.
+
+.. Note::
+   QEMU's VFIO retries certain operations if they fail. While this makes sense
+   for real HW, we don't know for sure whether it makes sense for virtual
+   devices.
+
+Defining a retry and timeout scheme is deferred to a future version of the
+protocol.
+
+Message sizes
+-------------
+
+Some requests have an ``argsz`` field. In a request, it defines the maximum
+expected reply payload size, which should be at least the size of the fixed
+reply payload headers defined here. The *request* payload size is defined by the
+usual ``msg_size`` field in the header, not the ``argsz`` field.
+
+In a reply, the server sets ``argsz`` field to the size needed for a full
+payload size. This may be less than the requested maximum size. This may be
+larger than the requested maximum size: in that case, the full payload is not
+included in the reply, but the ``argsz`` field in the reply indicates the needed
+size, allowing a client to allocate a larger buffer for holding the reply before
+trying again.
+
+In addition, during negotiation (see  `Version`_), the client and server may
+each specify a ``max_data_xfer_size`` value; this defines the maximum data that
+may be read or written via one of the ``VFIO_USER_DMA/REGION_READ/WRITE``
+messages; see `Read and Write Operations`_.
+
+Protocol Specification
+======================
+
+To distinguish from the base VFIO symbols, all vfio-user symbols are prefixed
+with ``vfio_user`` or ``VFIO_USER``. In this revision, all data is in the
+endianness of the host system, although this may be relaxed in future
+revisions in cases where the client and server run on different hosts
+with different endianness.
+
+Unless otherwise specified, all sizes should be presumed to be in bytes.
+
+.. _Commands:
+
+Commands
+--------
+The following table lists the VFIO message command IDs, and whether the
+message command is sent from the client or the server.
+
+======================================  =========  =================
+Name                                    Command    Request Direction
+======================================  =========  =================
+``VFIO_USER_VERSION``                   1          client -> server
+``VFIO_USER_DMA_MAP``                   2          client -> server
+``VFIO_USER_DMA_UNMAP``                 3          client -> server
+``VFIO_USER_DEVICE_GET_INFO``           4          client -> server
+``VFIO_USER_DEVICE_GET_REGION_INFO``    5          client -> server
+``VFIO_USER_DEVICE_GET_REGION_IO_FDS``  6          client -> server
+``VFIO_USER_DEVICE_GET_IRQ_INFO``       7          client -> server
+``VFIO_USER_DEVICE_SET_IRQS``           8          client -> server
+``VFIO_USER_REGION_READ``               9          client -> server
+``VFIO_USER_REGION_WRITE``              10         client -> server
+``VFIO_USER_DMA_READ``                  11         server -> client
+``VFIO_USER_DMA_WRITE``                 12         server -> client
+``VFIO_USER_DEVICE_RESET``              13         client -> server
+``VFIO_USER_REGION_WRITE_MULTI``        15         client -> server
+======================================  =========  =================
+
+Header
+------
+
+All messages, both command messages and reply messages, are preceded by a
+16-byte header that contains basic information about the message. The header is
+followed by message-specific data described in the sections below.
+
++----------------+--------+-------------+
+| Name           | Offset | Size        |
++================+========+=============+
+| Message ID     | 0      | 2           |
++----------------+--------+-------------+
+| Command        | 2      | 2           |
++----------------+--------+-------------+
+| Message size   | 4      | 4           |
++----------------+--------+-------------+
+| Flags          | 8      | 4           |
++----------------+--------+-------------+
+|                | +-----+------------+ |
+|                | | Bit | Definition | |
+|                | +=====+============+ |
+|                | | 0-3 | Type       | |
+|                | +-----+------------+ |
+|                | | 4   | No_reply   | |
+|                | +-----+------------+ |
+|                | | 5   | Error      | |
+|                | +-----+------------+ |
++----------------+--------+-------------+
+| Error          | 12     | 4           |
++----------------+--------+-------------+
+| <message data> | 16     | variable    |
++----------------+--------+-------------+
+
+* *Message ID* identifies the message, and is echoed in the command's reply
+  message. Message IDs belong entirely to the sender, can be re-used (even
+  concurrently) and the receiver must not make any assumptions about their
+  uniqueness.
+* *Command* specifies the command to be executed, listed in Commands_. It is
+  also set in the reply header.
+* *Message size* contains the size of the entire message, including the header.
+* *Flags* contains attributes of the message:
+
+  * The *Type* bits indicate the message type.
+
+    *  *Command* (value 0x0) indicates a command message.
+    *  *Reply* (value 0x1) indicates a reply message acknowledging a previous
+       command with the same message ID.
+  * *No_reply* in a command message indicates that no reply is needed for this
+    command.  This is commonly used when multiple commands are sent, and only
+    the last needs acknowledgement.
+  * *Error* in a reply message indicates the command being acknowledged had
+    an error. In this case, the *Error* field will be valid.
+
+* *Error* in a reply message is an optional UNIX errno value. It may be zero
+  even if the Error bit is set in Flags. It is reserved in a command message.
+
+Each command message in Commands_ must be replied to with a reply message,
+unless the message sets the *No_Reply* bit.  The reply consists of the header
+with the *Reply* bit set, plus any additional data.
+
+If an error occurs, the reply message must only include the reply header.
+
+As the header is standard in both requests and replies, it is not included in
+the command-specific specifications below; each message definition should be
+appended to the standard header, and the offsets are given from the end of the
+standard header.
+
+``VFIO_USER_VERSION``
+---------------------
+
+.. _Version:
+
+This is the initial message sent by the client after the socket connection is
+established; the same format is used for the server's reply.
+
+Upon establishing a connection, the client must send a ``VFIO_USER_VERSION``
+message proposing a protocol version and a set of capabilities. The server
+compares these with the versions and capabilities it supports and sends a
+``VFIO_USER_VERSION`` reply according to the following rules.
+
+* The major version in the reply must be the same as proposed. If the client
+  does not support the proposed major, it closes the connection.
+* The minor version in the reply must be equal to or less than the minor
+  version proposed.
+* The capability list must be a subset of those proposed. If the server
+  requires a capability the client did not include, it closes the connection.
+
+The protocol major version will only change when incompatible protocol changes
+are made, such as changing the message format. The minor version may change
+when compatible changes are made, such as adding new messages or capabilities,
+Both the client and server must support all minor versions less than the
+maximum minor version it supports. E.g., an implementation that supports
+version 1.3 must also support 1.0 through 1.2.
+
+When making a change to this specification, the protocol version number must
+be included in the form "added in version X.Y"
+
+Request
+^^^^^^^
+
+==============  ======  ====
+Name            Offset  Size
+==============  ======  ====
+version major   0       2
+version minor   2       2
+version data    4       variable (including terminating NUL). Optional.
+==============  ======  ====
+
+The version data is an optional UTF-8 encoded JSON byte array with the following
+format:
+
++--------------+--------+-----------------------------------+
+| Name         | Type   | Description                       |
++==============+========+===================================+
+| capabilities | object | Contains common capabilities that |
+|              |        | the sender supports. Optional.    |
++--------------+--------+-----------------------------------+
+
+Capabilities:
+
++--------------------+---------+------------------------------------------------+
+| Name               | Type    | Description                                    |
++====================+=========+================================================+
+| max_msg_fds        | number  | Maximum number of file descriptors that can be |
+|                    |         | received by the sender in one message.         |
+|                    |         | Optional. If not specified then the receiver   |
+|                    |         | must assume a value of ``1``.                  |
++--------------------+---------+------------------------------------------------+
+| max_data_xfer_size | number  | Maximum ``count`` for data transfer messages;  |
+|                    |         | see `Read and Write Operations`_. Optional,    |
+|                    |         | with a default value of 1048576 bytes.         |
++--------------------+---------+------------------------------------------------+
+| pgsizes            | number  | Page sizes supported in DMA map operations     |
+|                    |         | or'ed together. Optional, with a default value |
+|                    |         | of supporting only 4k pages.                   |
++--------------------+---------+------------------------------------------------+
+| max_dma_maps       | number  | Maximum number DMA map windows that can be     |
+|                    |         | valid simultaneously.  Optional, with a        |
+|                    |         | value of 65535 (64k-1).                        |
++--------------------+---------+------------------------------------------------+
+| migration          | object  | Migration capability parameters. If missing    |
+|                    |         | then migration is not supported by the sender. |
++--------------------+---------+------------------------------------------------+
+| write_multiple     | boolean | ``VFIO_USER_REGION_WRITE_MULTI`` messages      |
+|                    |         | are supported if the value is ``true``.        |
++--------------------+---------+------------------------------------------------+
+
+The migration capability contains the following name/value pairs:
+
++-----------------+--------+--------------------------------------------------+
+| Name            | Type   | Description                                      |
++=================+========+==================================================+
+| pgsize          | number | Page size of dirty pages bitmap. The smallest    |
+|                 |        | between the client and the server is used.       |
++-----------------+--------+--------------------------------------------------+
+| max_bitmap_size | number | Maximum bitmap size in ``VFIO_USER_DIRTY_PAGES`` |
+|                 |        | and ``VFIO_DMA_UNMAP`` messages.  Optional,      |
+|                 |        | with a default value of 256MB.                   |
++-----------------+--------+--------------------------------------------------+
+
+Reply
+^^^^^
+
+The same message format is used in the server's reply with the semantics
+described above.
+
+``VFIO_USER_DMA_MAP``
+---------------------
+
+This command message is sent by the client to the server to inform it of the
+memory regions the server can access. It must be sent before the server can
+perform any DMA to the client. It is normally sent directly after the version
+handshake is completed, but may also occur when memory is added to the client,
+or if the client uses a vIOMMU.
+
+Request
+^^^^^^^
+
+The request payload for this message is a structure of the following format:
+
++-------------+--------+-------------+
+| Name        | Offset | Size        |
++=============+========+=============+
+| argsz       | 0      | 4           |
++-------------+--------+-------------+
+| flags       | 4      | 4           |
++-------------+--------+-------------+
+|             | +-----+------------+ |
+|             | | Bit | Definition | |
+|             | +=====+============+ |
+|             | | 0   | readable   | |
+|             | +-----+------------+ |
+|             | | 1   | writeable  | |
+|             | +-----+------------+ |
++-------------+--------+-------------+
+| offset      | 8      | 8           |
++-------------+--------+-------------+
+| address     | 16     | 8           |
++-------------+--------+-------------+
+| size        | 24     | 8           |
++-------------+--------+-------------+
+
+* *argsz* is the size of the above structure. Note there is no reply payload,
+  so this field differs from other message types.
+* *flags* contains the following region attributes:
+
+  * *readable* indicates that the region can be read from.
+
+  * *writeable* indicates that the region can be written to.
+
+* *offset* is the file offset of the region with respect to the associated file
+  descriptor, or zero if the region is not mappable
+* *address* is the base DMA address of the region.
+* *size* is the size of the region.
+
+This structure is 32 bytes in size, so the message size is 16 + 32 bytes.
+
+If the DMA region being added can be directly mapped by the server, a file
+descriptor must be sent as part of the message meta-data. The region can be
+mapped via the mmap() system call. On ``AF_UNIX`` sockets, the file descriptor
+must be passed as ``SCM_RIGHTS`` type ancillary data.  Otherwise, if the DMA
+region cannot be directly mapped by the server, no file descriptor must be sent
+as part of the message meta-data and the DMA region can be accessed by the
+server using ``VFIO_USER_DMA_READ`` and ``VFIO_USER_DMA_WRITE`` messages,
+explained in `Read and Write Operations`_. A command to map over an existing
+region must be failed by the server with ``EEXIST`` set in error field in the
+reply.
+
+Reply
+^^^^^
+
+There is no payload in the reply message.
+
+``VFIO_USER_DMA_UNMAP``
+-----------------------
+
+This command message is sent by the client to the server to inform it that a
+DMA region, previously made available via a ``VFIO_USER_DMA_MAP`` command
+message, is no longer available for DMA. It typically occurs when memory is
+subtracted from the client or if the client uses a vIOMMU. The DMA region is
+described by the following structure:
+
+Request
+^^^^^^^
+
+The request payload for this message is a structure of the following format:
+
++--------------+--------+------------------------+
+| Name         | Offset | Size                   |
++==============+========+========================+
+| argsz        | 0      | 4                      |
++--------------+--------+------------------------+
+| flags        | 4      | 4                      |
++--------------+--------+------------------------+
+| address      | 8      | 8                      |
++--------------+--------+------------------------+
+| size         | 16     | 8                      |
++--------------+--------+------------------------+
+
+* *argsz* is the maximum size of the reply payload.
+* *flags* is unused in this version.
+* *address* is the base DMA address of the DMA region.
+* *size* is the size of the DMA region.
+
+The address and size of the DMA region being unmapped must match exactly a
+previous mapping.
+
+Reply
+^^^^^
+
+Upon receiving a ``VFIO_USER_DMA_UNMAP`` command, if the file descriptor is
+mapped then the server must release all references to that DMA region before
+replying, which potentially includes in-flight DMA transactions.
+
+The server responds with the original DMA entry in the request.
+
+
+``VFIO_USER_DEVICE_GET_INFO``
+-----------------------------
+
+This command message is sent by the client to the server to query for basic
+information about the device.
+
+Request
+^^^^^^^
+
++-------------+--------+--------------------------+
+| Name        | Offset | Size                     |
++=============+========+==========================+
+| argsz       | 0      | 4                        |
++-------------+--------+--------------------------+
+| flags       | 4      | 4                        |
++-------------+--------+--------------------------+
+|             | +-----+-------------------------+ |
+|             | | Bit | Definition              | |
+|             | +=====+=========================+ |
+|             | | 0   | VFIO_DEVICE_FLAGS_RESET | |
+|             | +-----+-------------------------+ |
+|             | | 1   | VFIO_DEVICE_FLAGS_PCI   | |
+|             | +-----+-------------------------+ |
++-------------+--------+--------------------------+
+| num_regions | 8      | 4                        |
++-------------+--------+--------------------------+
+| num_irqs    | 12     | 4                        |
++-------------+--------+--------------------------+
+
+* *argsz* is the maximum size of the reply payload
+* all other fields must be zero.
+
+Reply
+^^^^^
+
++-------------+--------+--------------------------+
+| Name        | Offset | Size                     |
++=============+========+==========================+
+| argsz       | 0      | 4                        |
++-------------+--------+--------------------------+
+| flags       | 4      | 4                        |
++-------------+--------+--------------------------+
+|             | +-----+-------------------------+ |
+|             | | Bit | Definition              | |
+|             | +=====+=========================+ |
+|             | | 0   | VFIO_DEVICE_FLAGS_RESET | |
+|             | +-----+-------------------------+ |
+|             | | 1   | VFIO_DEVICE_FLAGS_PCI   | |
+|             | +-----+-------------------------+ |
++-------------+--------+--------------------------+
+| num_regions | 8      | 4                        |
++-------------+--------+--------------------------+
+| num_irqs    | 12     | 4                        |
++-------------+--------+--------------------------+
+
+* *argsz* is the size required for the full reply payload (16 bytes today)
+* *flags* contains the following device attributes.
+
+  * ``VFIO_DEVICE_FLAGS_RESET`` indicates that the device supports the
+    ``VFIO_USER_DEVICE_RESET`` message.
+  * ``VFIO_DEVICE_FLAGS_PCI`` indicates that the device is a PCI device.
+
+* *num_regions* is the number of memory regions that the device exposes.
+* *num_irqs* is the number of distinct interrupt types that the device supports.
+
+This version of the protocol only supports PCI devices. Additional devices may
+be supported in future versions.
+
+``VFIO_USER_DEVICE_GET_REGION_INFO``
+------------------------------------
+
+This command message is sent by the client to the server to query for
+information about device regions. The VFIO region info structure is defined in
+``<linux/vfio.h>`` (``struct vfio_region_info``).
+
+Request
+^^^^^^^
+
++------------+--------+------------------------------+
+| Name       | Offset | Size                         |
++============+========+==============================+
+| argsz      | 0      | 4                            |
++------------+--------+------------------------------+
+| flags      | 4      | 4                            |
++------------+--------+------------------------------+
+| index      | 8      | 4                            |
++------------+--------+------------------------------+
+| cap_offset | 12     | 4                            |
++------------+--------+------------------------------+
+| size       | 16     | 8                            |
++------------+--------+------------------------------+
+| offset     | 24     | 8                            |
++------------+--------+------------------------------+
+
+* *argsz* the maximum size of the reply payload
+* *index* is the index of memory region being queried, it is the only field
+  that is required to be set in the command message.
+* all other fields must be zero.
+
+Reply
+^^^^^
+
++------------+--------+------------------------------+
+| Name       | Offset | Size                         |
++============+========+==============================+
+| argsz      | 0      | 4                            |
++------------+--------+------------------------------+
+| flags      | 4      | 4                            |
++------------+--------+------------------------------+
+|            | +-----+-----------------------------+ |
+|            | | Bit | Definition                  | |
+|            | +=====+=============================+ |
+|            | | 0   | VFIO_REGION_INFO_FLAG_READ  | |
+|            | +-----+-----------------------------+ |
+|            | | 1   | VFIO_REGION_INFO_FLAG_WRITE | |
+|            | +-----+-----------------------------+ |
+|            | | 2   | VFIO_REGION_INFO_FLAG_MMAP  | |
+|            | +-----+-----------------------------+ |
+|            | | 3   | VFIO_REGION_INFO_FLAG_CAPS  | |
+|            | +-----+-----------------------------+ |
++------------+--------+------------------------------+
++------------+--------+------------------------------+
+| index      | 8      | 4                            |
++------------+--------+------------------------------+
+| cap_offset | 12     | 4                            |
++------------+--------+------------------------------+
+| size       | 16     | 8                            |
++------------+--------+------------------------------+
+| offset     | 24     | 8                            |
++------------+--------+------------------------------+
+
+* *argsz* is the size required for the full reply payload (region info structure
+  plus the size of any region capabilities)
+* *flags* are attributes of the region:
+
+  * ``VFIO_REGION_INFO_FLAG_READ`` allows client read access to the region.
+  * ``VFIO_REGION_INFO_FLAG_WRITE`` allows client write access to the region.
+  * ``VFIO_REGION_INFO_FLAG_MMAP`` specifies the client can mmap() the region.
+    When this flag is set, the reply will include a file descriptor in its
+    meta-data. On ``AF_UNIX`` sockets, the file descriptors will be passed as
+    ``SCM_RIGHTS`` type ancillary data.
+  * ``VFIO_REGION_INFO_FLAG_CAPS`` indicates additional capabilities found in the
+    reply.
+
+* *index* is the index of memory region being queried, it is the only field
+  that is required to be set in the command message.
+* *cap_offset* describes where additional region capabilities can be found.
+  cap_offset is relative to the beginning of the VFIO region info structure.
+  The data structure it points is a VFIO cap header defined in
+  ``<linux/vfio.h>``.
+* *size* is the size of the region.
+* *offset* is the offset that should be given to the mmap() system call for
+  regions with the MMAP attribute. It is also used as the base offset when
+  mapping a VFIO sparse mmap area, described below.
+
+VFIO region capabilities
+""""""""""""""""""""""""
+
+The VFIO region information can also include a capabilities list. This list is
+similar to a PCI capability list - each entry has a common header that
+identifies a capability and where the next capability in the list can be found.
+The VFIO capability header format is defined in ``<linux/vfio.h>`` (``struct
+vfio_info_cap_header``).
+
+VFIO cap header format
+""""""""""""""""""""""
+
++---------+--------+------+
+| Name    | Offset | Size |
++=========+========+======+
+| id      | 0      | 2    |
++---------+--------+------+
+| version | 2      | 2    |
++---------+--------+------+
+| next    | 4      | 4    |
++---------+--------+------+
+
+* *id* is the capability identity.
+* *version* is a capability-specific version number.
+* *next* specifies the offset of the next capability in the capability list. It
+  is relative to the beginning of the VFIO region info structure.
+
+VFIO sparse mmap cap header
+"""""""""""""""""""""""""""
+
++------------------+----------------------------------+
+| Name             | Value                            |
++==================+==================================+
+| id               | VFIO_REGION_INFO_CAP_SPARSE_MMAP |
++------------------+----------------------------------+
+| version          | 0x1                              |
++------------------+----------------------------------+
+| next             | <next>                           |
++------------------+----------------------------------+
+| sparse mmap info | VFIO region info sparse mmap     |
++------------------+----------------------------------+
+
+This capability is defined when only a subrange of the region supports
+direct access by the client via mmap(). The VFIO sparse mmap area is defined in
+``<linux/vfio.h>`` (``struct vfio_region_sparse_mmap_area`` and ``struct
+vfio_region_info_cap_sparse_mmap``).
+
+VFIO region info cap sparse mmap
+""""""""""""""""""""""""""""""""
+
++----------+--------+------+
+| Name     | Offset | Size |
++==========+========+======+
+| nr_areas | 0      | 4    |
++----------+--------+------+
+| reserved | 4      | 4    |
++----------+--------+------+
+| offset   | 8      | 8    |
++----------+--------+------+
+| size     | 16     | 8    |
++----------+--------+------+
+| ...      |        |      |
++----------+--------+------+
+
+* *nr_areas* is the number of sparse mmap areas in the region.
+* *offset* and size describe a single area that can be mapped by the client.
+  There will be *nr_areas* pairs of offset and size. The offset will be added to
+  the base offset given in the ``VFIO_USER_DEVICE_GET_REGION_INFO`` to form the
+  offset argument of the subsequent mmap() call.
+
+The VFIO sparse mmap area is defined in ``<linux/vfio.h>`` (``struct
+vfio_region_info_cap_sparse_mmap``).
+
+
+``VFIO_USER_DEVICE_GET_REGION_IO_FDS``
+--------------------------------------
+
+Clients can access regions via ``VFIO_USER_REGION_READ/WRITE`` or, if provided, by
+``mmap()`` of a file descriptor provided by the server.
+
+``VFIO_USER_DEVICE_GET_REGION_IO_FDS`` provides an alternative access mechanism via
+file descriptors. This is an optional feature intended for performance
+improvements where an underlying sub-system (such as KVM) supports communication
+across such file descriptors to the vfio-user server, without needing to
+round-trip through the client.
+
+The server returns an array of sub-regions for the requested region. Each
+sub-region describes a span (offset and size) of a region, along with the
+requested file descriptor notification mechanism to use.  Each sub-region in the
+response message may choose to use a different method, as defined below.  The
+two mechanisms supported in this specification are ioeventfds and ioregionfds.
+
+The server in addition returns a file descriptor in the ancillary data; clients
+are expected to configure each sub-region's file descriptor with the requested
+notification method. For example, a client could configure KVM with the
+requested ioeventfd via a ``KVM_IOEVENTFD`` ``ioctl()``.
+
+Request
+^^^^^^^
+
++-------------+--------+------+
+| Name        | Offset | Size |
++=============+========+======+
+| argsz       | 0      | 4    |
++-------------+--------+------+
+| flags       | 4      | 4    |
++-------------+--------+------+
+| index       | 8      | 4    |
++-------------+--------+------+
+| count       | 12     | 4    |
++-------------+--------+------+
+
+* *argsz* the maximum size of the reply payload
+* *index* is the index of memory region being queried
+* all other fields must be zero
+
+The client must set ``flags`` to zero and specify the region being queried in
+the ``index``.
+
+Reply
+^^^^^
+
++-------------+--------+------+
+| Name        | Offset | Size |
++=============+========+======+
+| argsz       | 0      | 4    |
++-------------+--------+------+
+| flags       | 4      | 4    |
++-------------+--------+------+
+| index       | 8      | 4    |
++-------------+--------+------+
+| count       | 12     | 4    |
++-------------+--------+------+
+| sub-regions | 16     | ...  |
++-------------+--------+------+
+
+* *argsz* is the size of the region IO FD info structure plus the
+  total size of the sub-region array. Thus, each array entry "i" is at offset
+  i * ((argsz - 32) / count). Note that currently this is 40 bytes for both IO
+  FD types, but this is not to be relied on. As elsewhere, this indicates the
+  full reply payload size needed.
+* *flags* must be zero
+* *index* is the index of memory region being queried
+* *count* is the number of sub-regions in the array
+* *sub-regions* is the array of Sub-Region IO FD info structures
+
+The reply message will additionally include at least one file descriptor in the
+ancillary data. Note that more than one sub-region may share the same file
+descriptor.
+
+Note that it is the client's responsibility to verify the requested values (for
+example, that the requested offset does not exceed the region's bounds).
+
+Each sub-region given in the response has one of two possible structures,
+depending whether *type* is ``VFIO_USER_IO_FD_TYPE_IOEVENTFD`` or
+``VFIO_USER_IO_FD_TYPE_IOREGIONFD``:
+
+Sub-Region IO FD info format (ioeventfd)
+""""""""""""""""""""""""""""""""""""""""
+
++-----------+--------+------+
+| Name      | Offset | Size |
++===========+========+======+
+| offset    | 0      | 8    |
++-----------+--------+------+
+| size      | 8      | 8    |
++-----------+--------+------+
+| fd_index  | 16     | 4    |
++-----------+--------+------+
+| type      | 20     | 4    |
++-----------+--------+------+
+| flags     | 24     | 4    |
++-----------+--------+------+
+| padding   | 28     | 4    |
++-----------+--------+------+
+| datamatch | 32     | 8    |
++-----------+--------+------+
+
+* *offset* is the offset of the start of the sub-region within the region
+  requested ("physical address offset" for the region)
+* *size* is the length of the sub-region. This may be zero if the access size is
+  not relevant, which may allow for optimizations
+* *fd_index* is the index in the ancillary data of the FD to use for ioeventfd
+  notification; it may be shared.
+* *type* is ``VFIO_USER_IO_FD_TYPE_IOEVENTFD``
+* *flags* is any of:
+
+  * ``KVM_IOEVENTFD_FLAG_DATAMATCH``
+  * ``KVM_IOEVENTFD_FLAG_PIO``
+  * ``KVM_IOEVENTFD_FLAG_VIRTIO_CCW_NOTIFY`` (FIXME: makes sense?)
+
+* *datamatch* is the datamatch value if needed
+
+See https://www.kernel.org/doc/Documentation/virtual/kvm/api.txt, *4.59
+KVM_IOEVENTFD* for further context on the ioeventfd-specific fields.
+
+Sub-Region IO FD info format (ioregionfd)
+"""""""""""""""""""""""""""""""""""""""""
+
++-----------+--------+------+
+| Name      | Offset | Size |
++===========+========+======+
+| offset    | 0      | 8    |
++-----------+--------+------+
+| size      | 8      | 8    |
++-----------+--------+------+
+| fd_index  | 16     | 4    |
++-----------+--------+------+
+| type      | 20     | 4    |
++-----------+--------+------+
+| flags     | 24     | 4    |
++-----------+--------+------+
+| padding   | 28     | 4    |
++-----------+--------+------+
+| user_data | 32     | 8    |
++-----------+--------+------+
+
+* *offset* is the offset of the start of the sub-region within the region
+  requested ("physical address offset" for the region)
+* *size* is the length of the sub-region. This may be zero if the access size is
+  not relevant, which may allow for optimizations; ``KVM_IOREGION_POSTED_WRITES``
+  must be set in *flags* in this case
+* *fd_index* is the index in the ancillary data of the FD to use for ioregionfd
+  messages; it may be shared
+* *type* is ``VFIO_USER_IO_FD_TYPE_IOREGIONFD``
+* *flags* is any of:
+
+  * ``KVM_IOREGION_PIO``
+  * ``KVM_IOREGION_POSTED_WRITES``
+
+* *user_data* is an opaque value passed back to the server via a message on the
+  file descriptor
+
+For further information on the ioregionfd-specific fields, see:
+https://lore.kernel.org/kvm/cover.1613828726.git.eafanasova@gmail.com/
+
+(FIXME: update with final API docs.)
+
+``VFIO_USER_DEVICE_GET_IRQ_INFO``
+---------------------------------
+
+This command message is sent by the client to the server to query for
+information about device interrupt types. The VFIO IRQ info structure is
+defined in ``<linux/vfio.h>`` (``struct vfio_irq_info``).
+
+Request
+^^^^^^^
+
++-------+--------+---------------------------+
+| Name  | Offset | Size                      |
++=======+========+===========================+
+| argsz | 0      | 4                         |
++-------+--------+---------------------------+
+| flags | 4      | 4                         |
++-------+--------+---------------------------+
+|       | +-----+--------------------------+ |
+|       | | Bit | Definition               | |
+|       | +=====+==========================+ |
+|       | | 0   | VFIO_IRQ_INFO_EVENTFD    | |
+|       | +-----+--------------------------+ |
+|       | | 1   | VFIO_IRQ_INFO_MASKABLE   | |
+|       | +-----+--------------------------+ |
+|       | | 2   | VFIO_IRQ_INFO_AUTOMASKED | |
+|       | +-----+--------------------------+ |
+|       | | 3   | VFIO_IRQ_INFO_NORESIZE   | |
+|       | +-----+--------------------------+ |
++-------+--------+---------------------------+
+| index | 8      | 4                         |
++-------+--------+---------------------------+
+| count | 12     | 4                         |
++-------+--------+---------------------------+
+
+* *argsz* is the maximum size of the reply payload (16 bytes today)
+* index is the index of IRQ type being queried (e.g. ``VFIO_PCI_MSIX_IRQ_INDEX``)
+* all other fields must be zero
+
+Reply
+^^^^^
+
++-------+--------+---------------------------+
+| Name  | Offset | Size                      |
++=======+========+===========================+
+| argsz | 0      | 4                         |
++-------+--------+---------------------------+
+| flags | 4      | 4                         |
++-------+--------+---------------------------+
+|       | +-----+--------------------------+ |
+|       | | Bit | Definition               | |
+|       | +=====+==========================+ |
+|       | | 0   | VFIO_IRQ_INFO_EVENTFD    | |
+|       | +-----+--------------------------+ |
+|       | | 1   | VFIO_IRQ_INFO_MASKABLE   | |
+|       | +-----+--------------------------+ |
+|       | | 2   | VFIO_IRQ_INFO_AUTOMASKED | |
+|       | +-----+--------------------------+ |
+|       | | 3   | VFIO_IRQ_INFO_NORESIZE   | |
+|       | +-----+--------------------------+ |
++-------+--------+---------------------------+
+| index | 8      | 4                         |
++-------+--------+---------------------------+
+| count | 12     | 4                         |
++-------+--------+---------------------------+
+
+* *argsz* is the size required for the full reply payload (16 bytes today)
+* *flags* defines IRQ attributes:
+
+  * ``VFIO_IRQ_INFO_EVENTFD`` indicates the IRQ type can support server eventfd
+    signalling.
+  * ``VFIO_IRQ_INFO_MASKABLE`` indicates that the IRQ type supports the ``MASK``
+    and ``UNMASK`` actions in a ``VFIO_USER_DEVICE_SET_IRQS`` message.
+  * ``VFIO_IRQ_INFO_AUTOMASKED`` indicates the IRQ type masks itself after being
+    triggered, and the client must send an ``UNMASK`` action to receive new
+    interrupts.
+  * ``VFIO_IRQ_INFO_NORESIZE`` indicates ``VFIO_USER_SET_IRQS`` operations setup
+    interrupts as a set, and new sub-indexes cannot be enabled without disabling
+    the entire type.
+* index is the index of IRQ type being queried
+* count describes the number of interrupts of the queried type.
+
+``VFIO_USER_DEVICE_SET_IRQS``
+-----------------------------
+
+This command message is sent by the client to the server to set actions for
+device interrupt types. The VFIO IRQ set structure is defined in
+``<linux/vfio.h>`` (``struct vfio_irq_set``).
+
+Request
+^^^^^^^
+
++-------+--------+------------------------------+
+| Name  | Offset | Size                         |
++=======+========+==============================+
+| argsz | 0      | 4                            |
++-------+--------+------------------------------+
+| flags | 4      | 4                            |
++-------+--------+------------------------------+
+|       | +-----+-----------------------------+ |
+|       | | Bit | Definition                  | |
+|       | +=====+=============================+ |
+|       | | 0   | VFIO_IRQ_SET_DATA_NONE      | |
+|       | +-----+-----------------------------+ |
+|       | | 1   | VFIO_IRQ_SET_DATA_BOOL      | |
+|       | +-----+-----------------------------+ |
+|       | | 2   | VFIO_IRQ_SET_DATA_EVENTFD   | |
+|       | +-----+-----------------------------+ |
+|       | | 3   | VFIO_IRQ_SET_ACTION_MASK    | |
+|       | +-----+-----------------------------+ |
+|       | | 4   | VFIO_IRQ_SET_ACTION_UNMASK  | |
+|       | +-----+-----------------------------+ |
+|       | | 5   | VFIO_IRQ_SET_ACTION_TRIGGER | |
+|       | +-----+-----------------------------+ |
++-------+--------+------------------------------+
+| index | 8      | 4                            |
++-------+--------+------------------------------+
+| start | 12     | 4                            |
++-------+--------+------------------------------+
+| count | 16     | 4                            |
++-------+--------+------------------------------+
+| data  | 20     | variable                     |
++-------+--------+------------------------------+
+
+* *argsz* is the size of the VFIO IRQ set request payload, including any *data*
+  field. Note there is no reply payload, so this field differs from other
+  message types.
+* *flags* defines the action performed on the interrupt range. The ``DATA``
+  flags describe the data field sent in the message; the ``ACTION`` flags
+  describe the action to be performed. The flags are mutually exclusive for
+  both sets.
+
+  * ``VFIO_IRQ_SET_DATA_NONE`` indicates there is no data field in the command.
+    The action is performed unconditionally.
+  * ``VFIO_IRQ_SET_DATA_BOOL`` indicates the data field is an array of boolean
+    bytes. The action is performed if the corresponding boolean is true.
+  * ``VFIO_IRQ_SET_DATA_EVENTFD`` indicates an array of event file descriptors
+    was sent in the message meta-data. These descriptors will be signalled when
+    the action defined by the action flags occurs. In ``AF_UNIX`` sockets, the
+    descriptors are sent as ``SCM_RIGHTS`` type ancillary data.
+    If no file descriptors are provided, this de-assigns the specified
+    previously configured interrupts.
+  * ``VFIO_IRQ_SET_ACTION_MASK`` indicates a masking event. It can be used with
+    ``VFIO_IRQ_SET_DATA_BOOL`` or ``VFIO_IRQ_SET_DATA_NONE`` to mask an interrupt,
+    or with ``VFIO_IRQ_SET_DATA_EVENTFD`` to generate an event when the guest masks
+    the interrupt.
+  * ``VFIO_IRQ_SET_ACTION_UNMASK`` indicates an unmasking event. It can be used
+    with ``VFIO_IRQ_SET_DATA_BOOL`` or ``VFIO_IRQ_SET_DATA_NONE`` to unmask an
+    interrupt, or with ``VFIO_IRQ_SET_DATA_EVENTFD`` to generate an event when the
+    guest unmasks the interrupt.
+  * ``VFIO_IRQ_SET_ACTION_TRIGGER`` indicates a triggering event. It can be used
+    with ``VFIO_IRQ_SET_DATA_BOOL`` or ``VFIO_IRQ_SET_DATA_NONE`` to trigger an
+    interrupt, or with ``VFIO_IRQ_SET_DATA_EVENTFD`` to generate an event when the
+    server triggers the interrupt.
+
+* *index* is the index of IRQ type being setup.
+* *start* is the start of the sub-index being set.
+* *count* describes the number of sub-indexes being set. As a special case, a
+  count (and start) of 0, with data flags of ``VFIO_IRQ_SET_DATA_NONE`` disables
+  all interrupts of the index.
+* *data* is an optional field included when the
+  ``VFIO_IRQ_SET_DATA_BOOL`` flag is present. It contains an array of booleans
+  that specify whether the action is to be performed on the corresponding
+  index. It's used when the action is only performed on a subset of the range
+  specified.
+
+Not all interrupt types support every combination of data and action flags.
+The client must know the capabilities of the device and IRQ index before it
+sends a ``VFIO_USER_DEVICE_SET_IRQ`` message.
+
+In typical operation, a specific IRQ may operate as follows:
+
+1. The client sends a ``VFIO_USER_DEVICE_SET_IRQ`` message with
+   ``flags=(VFIO_IRQ_SET_DATA_EVENTFD|VFIO_IRQ_SET_ACTION_TRIGGER)`` along
+   with an eventfd. This associates the IRQ with a particular eventfd on the
+   server side.
+
+#. The client may send a ``VFIO_USER_DEVICE_SET_IRQ`` message with
+   ``flags=(VFIO_IRQ_SET_DATA_EVENTFD|VFIO_IRQ_SET_ACTION_MASK/UNMASK)`` along
+   with another eventfd. This associates the given eventfd with the
+   mask/unmask state on the server side.
+
+#. The server may trigger the IRQ by writing 1 to the eventfd.
+
+#. The server may mask/unmask an IRQ which will write 1 to the corresponding
+   mask/unmask eventfd, if there is one.
+
+5. A client may trigger a device IRQ itself, by sending a
+   ``VFIO_USER_DEVICE_SET_IRQ`` message with
+   ``flags=(VFIO_IRQ_SET_DATA_NONE/BOOL|VFIO_IRQ_SET_ACTION_TRIGGER)``.
+
+6. A client may mask or unmask the IRQ, by sending a
+   ``VFIO_USER_DEVICE_SET_IRQ`` message with
+   ``flags=(VFIO_IRQ_SET_DATA_NONE/BOOL|VFIO_IRQ_SET_ACTION_MASK/UNMASK)``.
+
+Reply
+^^^^^
+
+There is no payload in the reply.
+
+.. _Read and Write Operations:
+
+Note that all of these operations must be supported by the client and/or server,
+even if the corresponding memory or device region has been shared as mappable.
+
+The ``count`` field must not exceed the value of ``max_data_xfer_size`` of the
+peer, for both reads and writes.
+
+``VFIO_USER_REGION_READ``
+-------------------------
+
+If a device region is not mappable, it's not directly accessible by the client
+via ``mmap()`` of the underlying file descriptor. In this case, a client can
+read from a device region with this message.
+
+Request
+^^^^^^^
+
++--------+--------+----------+
+| Name   | Offset | Size     |
++========+========+==========+
+| offset | 0      | 8        |
++--------+--------+----------+
+| region | 8      | 4        |
++--------+--------+----------+
+| count  | 12     | 4        |
++--------+--------+----------+
+
+* *offset* into the region being accessed.
+* *region* is the index of the region being accessed.
+* *count* is the size of the data to be transferred.
+
+Reply
+^^^^^
+
++--------+--------+----------+
+| Name   | Offset | Size     |
++========+========+==========+
+| offset | 0      | 8        |
++--------+--------+----------+
+| region | 8      | 4        |
++--------+--------+----------+
+| count  | 12     | 4        |
++--------+--------+----------+
+| data   | 16     | variable |
++--------+--------+----------+
+
+* *offset* into the region accessed.
+* *region* is the index of the region accessed.
+* *count* is the size of the data transferred.
+* *data* is the data that was read from the device region.
+
+``VFIO_USER_REGION_WRITE``
+--------------------------
+
+If a device region is not mappable, it's not directly accessible by the client
+via mmap() of the underlying fd. In this case, a client can write to a device
+region with this message.
+
+Request
+^^^^^^^
+
++--------+--------+----------+
+| Name   | Offset | Size     |
++========+========+==========+
+| offset | 0      | 8        |
++--------+--------+----------+
+| region | 8      | 4        |
++--------+--------+----------+
+| count  | 12     | 4        |
++--------+--------+----------+
+| data   | 16     | variable |
++--------+--------+----------+
+
+* *offset* into the region being accessed.
+* *region* is the index of the region being accessed.
+* *count* is the size of the data to be transferred.
+* *data* is the data to write
+
+Reply
+^^^^^
+
++--------+--------+----------+
+| Name   | Offset | Size     |
++========+========+==========+
+| offset | 0      | 8        |
++--------+--------+----------+
+| region | 8      | 4        |
++--------+--------+----------+
+| count  | 12     | 4        |
++--------+--------+----------+
+
+* *offset* into the region accessed.
+* *region* is the index of the region accessed.
+* *count* is the size of the data transferred.
+
+``VFIO_USER_DMA_READ``
+-----------------------
+
+If the client has not shared mappable memory, the server can use this message to
+read from guest memory.
+
+Request
+^^^^^^^
+
++---------+--------+----------+
+| Name    | Offset | Size     |
++=========+========+==========+
+| address | 0      | 8        |
++---------+--------+----------+
+| count   | 8      | 8        |
++---------+--------+----------+
+
+* *address* is the client DMA memory address being accessed. This address must have
+  been previously exported to the server with a ``VFIO_USER_DMA_MAP`` message.
+* *count* is the size of the data to be transferred.
+
+Reply
+^^^^^
+
++---------+--------+----------+
+| Name    | Offset | Size     |
++=========+========+==========+
+| address | 0      | 8        |
++---------+--------+----------+
+| count   | 8      | 8        |
++---------+--------+----------+
+| data    | 16     | variable |
++---------+--------+----------+
+
+* *address* is the client DMA memory address being accessed.
+* *count* is the size of the data transferred.
+* *data* is the data read.
+
+``VFIO_USER_DMA_WRITE``
+-----------------------
+
+If the client has not shared mappable memory, the server can use this message to
+write to guest memory.
+
+Request
+^^^^^^^
+
++---------+--------+----------+
+| Name    | Offset | Size     |
++=========+========+==========+
+| address | 0      | 8        |
++---------+--------+----------+
+| count   | 8      | 8        |
++---------+--------+----------+
+| data    | 16     | variable |
++---------+--------+----------+
+
+* *address* is the client DMA memory address being accessed. This address must have
+  been previously exported to the server with a ``VFIO_USER_DMA_MAP`` message.
+* *count* is the size of the data to be transferred.
+* *data* is the data to write
+
+Reply
+^^^^^
+
++---------+--------+----------+
+| Name    | Offset | Size     |
++=========+========+==========+
+| address | 0      | 8        |
++---------+--------+----------+
+| count   | 8      | 4        |
++---------+--------+----------+
+
+* *address* is the client DMA memory address being accessed.
+* *count* is the size of the data transferred.
+
+``VFIO_USER_DEVICE_RESET``
+--------------------------
+
+This command message is sent from the client to the server to reset the device.
+Neither the request or reply have a payload.
+
+``VFIO_USER_REGION_WRITE_MULTI``
+--------------------------------
+
+This message can be used to coalesce multiple device write operations
+into a single messgage.  It is only used as an optimization when the
+outgoing message queue is relatively full.
+
+Request
+^^^^^^^
+
++---------+--------+----------+
+| Name    | Offset | Size     |
++=========+========+==========+
+| wr_cnt  | 0      | 8        |
++---------+--------+----------+
+| wrs     | 8      | variable |
++---------+--------+----------+
+
+* *wr_cnt* is the number of device writes coalesced in the message
+* *wrs* is an array of device writes defined below
+
+Single Device Write Format
+""""""""""""""""""""""""""
+
++--------+--------+----------+
+| Name   | Offset | Size     |
++========+========+==========+
+| offset | 0      | 8        |
++--------+--------+----------+
+| region | 8      | 4        |
++--------+--------+----------+
+| count  | 12     | 4        |
++--------+--------+----------+
+| data   | 16     | 8        |
++--------+--------+----------+
+
+* *offset* into the region being accessed.
+* *region* is the index of the region being accessed.
+* *count* is the size of the data to be transferred.  This format can
+  only describe writes of 8 bytes or less.
+* *data* is the data to write.
+
+Reply
+^^^^^
+
++---------+--------+----------+
+| Name    | Offset | Size     |
++=========+========+==========+
+| wr_cnt  | 0      | 8        |
++---------+--------+----------+
+
+* *wr_cnt* is the number of device writes completed.
+
+
+Appendices
+==========
+
+Unused VFIO ``ioctl()`` commands
+--------------------------------
+
+The following VFIO commands do not have an equivalent vfio-user command:
+
+* ``VFIO_GET_API_VERSION``
+* ``VFIO_CHECK_EXTENSION``
+* ``VFIO_SET_IOMMU``
+* ``VFIO_GROUP_GET_STATUS``
+* ``VFIO_GROUP_SET_CONTAINER``
+* ``VFIO_GROUP_UNSET_CONTAINER``
+* ``VFIO_GROUP_GET_DEVICE_FD``
+* ``VFIO_IOMMU_GET_INFO``
+
+However, once support for live migration for VFIO devices is finalized some
+of the above commands may have to be handled by the client in their
+corresponding vfio-user form. This will be addressed in a future protocol
+version.
+
+VFIO groups and containers
+^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+The current VFIO implementation includes group and container idioms that
+describe how a device relates to the host IOMMU. In the vfio-user
+implementation, the IOMMU is implemented in SW by the client, and is not
+visible to the server. The simplest idea would be that the client put each
+device into its own group and container.
+
+Backend Program Conventions
+---------------------------
+
+vfio-user backend program conventions are based on the vhost-user ones.
+
+* The backend program must not daemonize itself.
+* No assumptions must be made as to what access the backend program has on the
+  system.
+* File descriptors 0, 1 and 2 must exist, must have regular
+  stdin/stdout/stderr semantics, and can be redirected.
+* The backend program must honor the SIGTERM signal.
+* The backend program must accept the following commands line options:
+
+  * ``--socket-path=PATH``: path to UNIX domain socket,
+  * ``--fd=FDNUM``: file descriptor for UNIX domain socket, incompatible with
+    ``--socket-path``
+* The backend program must be accompanied with a JSON file stored under
+  ``/usr/share/vfio-user``.
+
+TODO add schema similar to docs/interop/vhost-user.json.
diff --git a/docs/system/device-emulation.rst b/docs/system/device-emulation.rst
index a1b0d79..9113816 100644
--- a/docs/system/device-emulation.rst
+++ b/docs/system/device-emulation.rst
@@ -85,6 +85,7 @@ Emulated Devices
    devices/can.rst
    devices/ccid.rst
    devices/cxl.rst
+   devices/vfio-user.rst
    devices/ivshmem.rst
    devices/ivshmem-flat.rst
    devices/keyboard.rst
diff --git a/docs/system/devices/vfio-user.rst b/docs/system/devices/vfio-user.rst
new file mode 100644
index 0000000..b6dcaa5
--- /dev/null
+++ b/docs/system/devices/vfio-user.rst
@@ -0,0 +1,26 @@
+.. SPDX-License-Identifier: GPL-2.0-or-later
+
+=========
+vfio-user
+=========
+
+QEMU includes a ``vfio-user`` client. The ``vfio-user`` specification allows for
+implementing (PCI) devices in userspace outside of QEMU; it is similar to
+``vhost-user`` in this respect (see :doc:`vhost-user`), but can emulate arbitrary
+PCI devices, not just ``virtio``. Whereas ``vfio`` is handled by the host
+kernel, ``vfio-user``, while similar in implementation, is handled entirely in
+userspace.
+
+For example, SPDK includes a virtual PCI NVMe controller implementation; by
+setting up a ``vfio-user`` UNIX socket between QEMU and SPDK, a VM can send NVMe
+I/O to the SPDK process.
+
+Presuming a suitable ``vfio-user`` server has opened a socket at
+``/tmp/vfio-user.sock``, a device can be configured with for example:
+
+.. code-block:: console
+
+-device '{"driver": "vfio-user-pci","socket": {"path": "/tmp/vfio-user.sock", "type": "unix"}}'
+
+See `libvfio-user <https://github.com/nutanix/libvfio-user/>`_ for further
+information.
diff --git a/hw/Kconfig b/hw/Kconfig
index 9a86a6a..9e6c789 100644
--- a/hw/Kconfig
+++ b/hw/Kconfig
@@ -42,6 +42,7 @@ source ufs/Kconfig
 source usb/Kconfig
 source virtio/Kconfig
 source vfio/Kconfig
+source vfio-user/Kconfig
 source vmapple/Kconfig
 source xen/Kconfig
 source watchdog/Kconfig
diff --git a/hw/meson.build b/hw/meson.build
index b91f761..791ce21 100644
--- a/hw/meson.build
+++ b/hw/meson.build
@@ -39,6 +39,7 @@ subdir('uefi')
 subdir('ufs')
 subdir('usb')
 subdir('vfio')
+subdir('vfio-user')
 subdir('virtio')
 subdir('vmapple')
 subdir('watchdog')
diff --git a/hw/vfio-user/Kconfig b/hw/vfio-user/Kconfig
new file mode 100644
index 0000000..24bdf7a
--- /dev/null
+++ b/hw/vfio-user/Kconfig
@@ -0,0 +1,7 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+config VFIO_USER
+    bool
+    default y
+    depends on VFIO_PCI
+
diff --git a/hw/vfio-user/container.c b/hw/vfio-user/container.c
new file mode 100644
index 0000000..3133fef
--- /dev/null
+++ b/hw/vfio-user/container.c
@@ -0,0 +1,370 @@
+/*
+ * Container for vfio-user IOMMU type: rather than communicating with the kernel
+ * vfio driver, we communicate over a socket to a server using the vfio-user
+ * protocol.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include <sys/ioctl.h>
+#include <linux/vfio.h>
+#include "qemu/osdep.h"
+
+#include "hw/vfio-user/container.h"
+#include "hw/vfio-user/device.h"
+#include "hw/vfio-user/trace.h"
+#include "hw/vfio/vfio-cpr.h"
+#include "hw/vfio/vfio-device.h"
+#include "hw/vfio/vfio-listener.h"
+#include "qapi/error.h"
+
+/*
+ * When DMA space is the physical address space, the region add/del listeners
+ * will fire during memory update transactions.  These depend on BQL being held,
+ * so do any resulting map/demap ops async while keeping BQL.
+ */
+static void vfio_user_listener_begin(VFIOContainerBase *bcontainer)
+{
+    VFIOUserContainer *container = container_of(bcontainer, VFIOUserContainer,
+                                                 bcontainer);
+
+    container->proxy->async_ops = true;
+}
+
+static void vfio_user_listener_commit(VFIOContainerBase *bcontainer)
+{
+    VFIOUserContainer *container = container_of(bcontainer, VFIOUserContainer,
+                                            bcontainer);
+
+    /* wait here for any async requests sent during the transaction */
+    container->proxy->async_ops = false;
+    vfio_user_wait_reqs(container->proxy);
+}
+
+static int vfio_user_dma_unmap(const VFIOContainerBase *bcontainer,
+                               hwaddr iova, ram_addr_t size,
+                               IOMMUTLBEntry *iotlb, bool unmap_all)
+{
+    VFIOUserContainer *container = container_of(bcontainer, VFIOUserContainer,
+                                            bcontainer);
+    Error *local_err = NULL;
+    int ret = 0;
+
+    VFIOUserDMAUnmap *msgp = g_malloc(sizeof(*msgp));
+
+    vfio_user_request_msg(&msgp->hdr, VFIO_USER_DMA_UNMAP, sizeof(*msgp), 0);
+    msgp->argsz = sizeof(struct vfio_iommu_type1_dma_unmap);
+    msgp->flags = unmap_all ? VFIO_DMA_UNMAP_FLAG_ALL : 0;
+    msgp->iova = iova;
+    msgp->size = size;
+    trace_vfio_user_dma_unmap(msgp->iova, msgp->size, msgp->flags,
+                              container->proxy->async_ops);
+
+    if (container->proxy->async_ops) {
+        if (!vfio_user_send_nowait(container->proxy, &msgp->hdr, NULL,
+                              0, &local_err)) {
+            error_report_err(local_err);
+            ret = -EFAULT;
+        } else {
+            ret = 0;
+        }
+    } else {
+        if (!vfio_user_send_wait(container->proxy, &msgp->hdr, NULL,
+                                 0, &local_err)) {
+                error_report_err(local_err);
+                ret = -EFAULT;
+        }
+
+        if (msgp->hdr.flags & VFIO_USER_ERROR) {
+            ret = -msgp->hdr.error_reply;
+        }
+
+        g_free(msgp);
+    }
+
+    return ret;
+}
+
+static int vfio_user_dma_map(const VFIOContainerBase *bcontainer, hwaddr iova,
+                             ram_addr_t size, void *vaddr, bool readonly,
+                             MemoryRegion *mrp)
+{
+    VFIOUserContainer *container = container_of(bcontainer, VFIOUserContainer,
+                                                bcontainer);
+    int fd = memory_region_get_fd(mrp);
+    Error *local_err = NULL;
+    int ret;
+
+    VFIOUserFDs *fds = NULL;
+    VFIOUserDMAMap *msgp = g_malloc0(sizeof(*msgp));
+
+    vfio_user_request_msg(&msgp->hdr, VFIO_USER_DMA_MAP, sizeof(*msgp), 0);
+    msgp->argsz = sizeof(struct vfio_iommu_type1_dma_map);
+    msgp->flags = VFIO_DMA_MAP_FLAG_READ;
+    msgp->offset = 0;
+    msgp->iova = iova;
+    msgp->size = size;
+
+    /*
+     * vaddr enters as a QEMU process address; make it either a file offset
+     * for mapped areas or leave as 0.
+     */
+    if (fd != -1) {
+        msgp->offset = qemu_ram_block_host_offset(mrp->ram_block, vaddr);
+    }
+
+    if (!readonly) {
+        msgp->flags |= VFIO_DMA_MAP_FLAG_WRITE;
+    }
+
+    trace_vfio_user_dma_map(msgp->iova, msgp->size, msgp->offset, msgp->flags,
+                            container->proxy->async_ops);
+
+    /*
+     * The async_ops case sends without blocking. They're later waited for in
+     * vfio_send_wait_reqs.
+     */
+    if (container->proxy->async_ops) {
+        /* can't use auto variable since we don't block */
+        if (fd != -1) {
+            fds = vfio_user_getfds(1);
+            fds->send_fds = 1;
+            fds->fds[0] = fd;
+        }
+
+        if (!vfio_user_send_nowait(container->proxy, &msgp->hdr, fds,
+                              0, &local_err)) {
+            error_report_err(local_err);
+            ret = -EFAULT;
+        } else {
+            ret = 0;
+        }
+    } else {
+        VFIOUserFDs local_fds = { 1, 0, &fd };
+
+        fds = fd != -1 ? &local_fds : NULL;
+
+        if (!vfio_user_send_wait(container->proxy, &msgp->hdr, fds,
+                                 0, &local_err)) {
+                error_report_err(local_err);
+                ret = -EFAULT;
+        }
+
+        if (msgp->hdr.flags & VFIO_USER_ERROR) {
+            ret = -msgp->hdr.error_reply;
+        }
+
+        g_free(msgp);
+    }
+
+    return ret;
+}
+
+static int
+vfio_user_set_dirty_page_tracking(const VFIOContainerBase *bcontainer,
+                                    bool start, Error **errp)
+{
+    error_setg_errno(errp, ENOTSUP, "Not supported");
+    return -ENOTSUP;
+}
+
+static int vfio_user_query_dirty_bitmap(const VFIOContainerBase *bcontainer,
+                                         VFIOBitmap *vbmap, hwaddr iova,
+                                         hwaddr size, Error **errp)
+{
+    error_setg_errno(errp, ENOTSUP, "Not supported");
+    return -ENOTSUP;
+}
+
+static bool vfio_user_setup(VFIOContainerBase *bcontainer, Error **errp)
+{
+    VFIOUserContainer *container = container_of(bcontainer, VFIOUserContainer,
+                                                bcontainer);
+
+    assert(container->proxy->dma_pgsizes != 0);
+    bcontainer->pgsizes = container->proxy->dma_pgsizes;
+    bcontainer->dma_max_mappings = container->proxy->max_dma;
+
+    /* No live migration support yet. */
+    bcontainer->dirty_pages_supported = false;
+    bcontainer->max_dirty_bitmap_size = container->proxy->max_bitmap;
+    bcontainer->dirty_pgsizes = container->proxy->migr_pgsize;
+
+    return true;
+}
+
+static VFIOUserContainer *vfio_user_create_container(VFIODevice *vbasedev,
+                                                     Error **errp)
+{
+    VFIOUserContainer *container;
+
+    container = VFIO_IOMMU_USER(object_new(TYPE_VFIO_IOMMU_USER));
+    container->proxy = vbasedev->proxy;
+    return container;
+}
+
+/*
+ * Try to mirror vfio_container_connect() as much as possible.
+ */
+static VFIOUserContainer *
+vfio_user_container_connect(AddressSpace *as, VFIODevice *vbasedev,
+                            Error **errp)
+{
+    VFIOContainerBase *bcontainer;
+    VFIOUserContainer *container;
+    VFIOAddressSpace *space;
+    VFIOIOMMUClass *vioc;
+    int ret;
+
+    space = vfio_address_space_get(as);
+
+    container = vfio_user_create_container(vbasedev, errp);
+    if (!container) {
+        goto put_space_exit;
+    }
+
+    bcontainer = &container->bcontainer;
+
+    if (!vfio_cpr_register_container(bcontainer, errp)) {
+        goto free_container_exit;
+    }
+
+    ret = ram_block_uncoordinated_discard_disable(true);
+    if (ret) {
+        error_setg_errno(errp, -ret, "Cannot set discarding of RAM broken");
+        goto unregister_container_exit;
+    }
+
+    vioc = VFIO_IOMMU_GET_CLASS(bcontainer);
+    assert(vioc->setup);
+
+    if (!vioc->setup(bcontainer, errp)) {
+        goto enable_discards_exit;
+    }
+
+    vfio_address_space_insert(space, bcontainer);
+
+    if (!vfio_listener_register(bcontainer, errp)) {
+        goto listener_release_exit;
+    }
+
+    bcontainer->initialized = true;
+
+    return container;
+
+listener_release_exit:
+    vfio_listener_unregister(bcontainer);
+    if (vioc->release) {
+        vioc->release(bcontainer);
+    }
+
+enable_discards_exit:
+    ram_block_uncoordinated_discard_disable(false);
+
+unregister_container_exit:
+    vfio_cpr_unregister_container(bcontainer);
+
+free_container_exit:
+    object_unref(container);
+
+put_space_exit:
+    vfio_address_space_put(space);
+
+    return NULL;
+}
+
+static void vfio_user_container_disconnect(VFIOUserContainer *container)
+{
+    VFIOContainerBase *bcontainer = &container->bcontainer;
+    VFIOIOMMUClass *vioc = VFIO_IOMMU_GET_CLASS(bcontainer);
+    VFIOAddressSpace *space = bcontainer->space;
+
+    ram_block_uncoordinated_discard_disable(false);
+
+    vfio_listener_unregister(bcontainer);
+    if (vioc->release) {
+        vioc->release(bcontainer);
+    }
+
+    vfio_cpr_unregister_container(bcontainer);
+    object_unref(container);
+
+    vfio_address_space_put(space);
+}
+
+static bool vfio_user_device_get(VFIOUserContainer *container,
+                                 VFIODevice *vbasedev, Error **errp)
+{
+    struct vfio_device_info info = { .argsz = sizeof(info) };
+
+
+    if (!vfio_user_get_device_info(vbasedev->proxy, &info, errp)) {
+        return false;
+    }
+
+    vbasedev->fd = -1;
+
+    vfio_device_prepare(vbasedev, &container->bcontainer, &info);
+
+    return true;
+}
+
+/*
+ * vfio_user_device_attach: attach a device to a new container.
+ */
+static bool vfio_user_device_attach(const char *name, VFIODevice *vbasedev,
+                                    AddressSpace *as, Error **errp)
+{
+    VFIOUserContainer *container;
+
+    container = vfio_user_container_connect(as, vbasedev, errp);
+    if (container == NULL) {
+        error_prepend(errp, "failed to connect proxy");
+        return false;
+    }
+
+    return vfio_user_device_get(container, vbasedev, errp);
+}
+
+static void vfio_user_device_detach(VFIODevice *vbasedev)
+{
+    VFIOUserContainer *container = container_of(vbasedev->bcontainer,
+                                                VFIOUserContainer, bcontainer);
+
+    vfio_device_unprepare(vbasedev);
+
+    vfio_user_container_disconnect(container);
+}
+
+static int vfio_user_pci_hot_reset(VFIODevice *vbasedev, bool single)
+{
+    /* ->needs_reset is always false for vfio-user. */
+    return 0;
+}
+
+static void vfio_iommu_user_class_init(ObjectClass *klass, const void *data)
+{
+    VFIOIOMMUClass *vioc = VFIO_IOMMU_CLASS(klass);
+
+    vioc->setup = vfio_user_setup;
+    vioc->listener_begin = vfio_user_listener_begin,
+    vioc->listener_commit = vfio_user_listener_commit,
+    vioc->dma_map = vfio_user_dma_map;
+    vioc->dma_unmap = vfio_user_dma_unmap;
+    vioc->attach_device = vfio_user_device_attach;
+    vioc->detach_device = vfio_user_device_detach;
+    vioc->set_dirty_page_tracking = vfio_user_set_dirty_page_tracking;
+    vioc->query_dirty_bitmap = vfio_user_query_dirty_bitmap;
+    vioc->pci_hot_reset = vfio_user_pci_hot_reset;
+};
+
+static const TypeInfo types[] = {
+    {
+        .name = TYPE_VFIO_IOMMU_USER,
+        .parent = TYPE_VFIO_IOMMU,
+        .instance_size = sizeof(VFIOUserContainer),
+        .class_init = vfio_iommu_user_class_init,
+    },
+};
+
+DEFINE_TYPES(types)
diff --git a/hw/vfio-user/container.h b/hw/vfio-user/container.h
new file mode 100644
index 0000000..2bb1fa1
--- /dev/null
+++ b/hw/vfio-user/container.h
@@ -0,0 +1,23 @@
+/*
+ * vfio-user specific definitions.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#ifndef HW_VFIO_USER_CONTAINER_H
+#define HW_VFIO_USER_CONTAINER_H
+
+#include "qemu/osdep.h"
+
+#include "hw/vfio/vfio-container-base.h"
+#include "hw/vfio-user/proxy.h"
+
+/* MMU container sub-class for vfio-user. */
+typedef struct VFIOUserContainer {
+    VFIOContainerBase bcontainer;
+    VFIOUserProxy *proxy;
+} VFIOUserContainer;
+
+OBJECT_DECLARE_SIMPLE_TYPE(VFIOUserContainer, VFIO_IOMMU_USER);
+
+#endif /* HW_VFIO_USER_CONTAINER_H */
diff --git a/hw/vfio-user/device.c b/hw/vfio-user/device.c
new file mode 100644
index 0000000..0609a7d
--- /dev/null
+++ b/hw/vfio-user/device.c
@@ -0,0 +1,441 @@
+/*
+ * vfio protocol over a UNIX socket device handling.
+ *
+ * Copyright © 2018, 2021 Oracle and/or its affiliates.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include "qemu/osdep.h"
+#include "qapi/error.h"
+#include "qemu/error-report.h"
+#include "qemu/lockable.h"
+#include "qemu/thread.h"
+
+#include "hw/vfio-user/device.h"
+#include "hw/vfio-user/trace.h"
+
+/*
+ * These are to defend against a malign server trying
+ * to force us to run out of memory.
+ */
+#define VFIO_USER_MAX_REGIONS   100
+#define VFIO_USER_MAX_IRQS      50
+
+bool vfio_user_get_device_info(VFIOUserProxy *proxy,
+                               struct vfio_device_info *info, Error **errp)
+{
+    VFIOUserDeviceInfo msg;
+    uint32_t argsz = sizeof(msg) - sizeof(msg.hdr);
+
+    memset(&msg, 0, sizeof(msg));
+    vfio_user_request_msg(&msg.hdr, VFIO_USER_DEVICE_GET_INFO, sizeof(msg), 0);
+    msg.argsz = argsz;
+
+    if (!vfio_user_send_wait(proxy, &msg.hdr, NULL, 0, errp)) {
+        return false;
+    }
+
+    if (msg.hdr.flags & VFIO_USER_ERROR) {
+        error_setg_errno(errp, -msg.hdr.error_reply,
+                         "VFIO_USER_DEVICE_GET_INFO failed");
+        return false;
+    }
+
+    trace_vfio_user_get_info(msg.num_regions, msg.num_irqs);
+
+    memcpy(info, &msg.argsz, argsz);
+
+    /* defend against a malicious server */
+    if (info->num_regions > VFIO_USER_MAX_REGIONS ||
+        info->num_irqs > VFIO_USER_MAX_IRQS) {
+        error_setg_errno(errp, EINVAL, "invalid reply");
+        return false;
+    }
+
+    return true;
+}
+
+void vfio_user_device_reset(VFIOUserProxy *proxy)
+{
+    Error *local_err = NULL;
+    VFIOUserHdr hdr;
+
+    vfio_user_request_msg(&hdr, VFIO_USER_DEVICE_RESET, sizeof(hdr), 0);
+
+    if (!vfio_user_send_wait(proxy, &hdr, NULL, 0, &local_err)) {
+        error_prepend(&local_err, "%s: ", __func__);
+        error_report_err(local_err);
+        return;
+    }
+
+    if (hdr.flags & VFIO_USER_ERROR) {
+        error_printf("reset reply error %d\n", hdr.error_reply);
+    }
+}
+
+static int vfio_user_get_region_info(VFIOUserProxy *proxy,
+                                     struct vfio_region_info *info,
+                                     VFIOUserFDs *fds)
+{
+    g_autofree VFIOUserRegionInfo *msgp = NULL;
+    Error *local_err = NULL;
+    uint32_t size;
+
+    /* data returned can be larger than vfio_region_info */
+    if (info->argsz < sizeof(*info)) {
+        error_printf("vfio_user_get_region_info argsz too small\n");
+        return -E2BIG;
+    }
+    if (fds != NULL && fds->send_fds != 0) {
+        error_printf("vfio_user_get_region_info can't send FDs\n");
+        return -EINVAL;
+    }
+
+    size = info->argsz + sizeof(VFIOUserHdr);
+    msgp = g_malloc0(size);
+
+    vfio_user_request_msg(&msgp->hdr, VFIO_USER_DEVICE_GET_REGION_INFO,
+                          sizeof(*msgp), 0);
+    msgp->argsz = info->argsz;
+    msgp->index = info->index;
+
+    if (!vfio_user_send_wait(proxy, &msgp->hdr, fds, size, &local_err)) {
+        error_prepend(&local_err, "%s: ", __func__);
+        error_report_err(local_err);
+        return -EFAULT;
+    }
+
+    if (msgp->hdr.flags & VFIO_USER_ERROR) {
+        return -msgp->hdr.error_reply;
+    }
+    trace_vfio_user_get_region_info(msgp->index, msgp->flags, msgp->size);
+
+    memcpy(info, &msgp->argsz, info->argsz);
+
+    /*
+     * If at least one region is directly mapped into the VM, then we can no
+     * longer rely on the sequential nature of vfio-user request handling to
+     * ensure that posted writes are completed before a subsequent read. In this
+     * case, disable posted write support. This is a per-device property, not
+     * per-region.
+     */
+    if (info->flags & VFIO_REGION_INFO_FLAG_MMAP) {
+        vfio_user_disable_posted_writes(proxy);
+    }
+
+    return 0;
+}
+
+static int vfio_user_device_io_get_region_info(VFIODevice *vbasedev,
+                                               struct vfio_region_info *info,
+                                               int *fd)
+{
+    VFIOUserFDs fds = { 0, 1, fd};
+    int ret;
+
+    if (info->index > vbasedev->num_regions) {
+        return -EINVAL;
+    }
+
+    ret = vfio_user_get_region_info(vbasedev->proxy, info, &fds);
+    if (ret) {
+        return ret;
+    }
+
+    /* cap_offset in valid area */
+    if ((info->flags & VFIO_REGION_INFO_FLAG_CAPS) &&
+        (info->cap_offset < sizeof(*info) || info->cap_offset > info->argsz)) {
+        return -EINVAL;
+    }
+
+    return 0;
+}
+
+static int vfio_user_device_io_get_irq_info(VFIODevice *vbasedev,
+                                            struct vfio_irq_info *info)
+{
+    VFIOUserProxy *proxy = vbasedev->proxy;
+    Error *local_err = NULL;
+    VFIOUserIRQInfo msg;
+
+    memset(&msg, 0, sizeof(msg));
+    vfio_user_request_msg(&msg.hdr, VFIO_USER_DEVICE_GET_IRQ_INFO,
+                          sizeof(msg), 0);
+    msg.argsz = info->argsz;
+    msg.index = info->index;
+
+    if (!vfio_user_send_wait(proxy, &msg.hdr, NULL, 0, &local_err)) {
+        error_prepend(&local_err, "%s: ", __func__);
+        error_report_err(local_err);
+        return -EFAULT;
+    }
+
+    if (msg.hdr.flags & VFIO_USER_ERROR) {
+        return -msg.hdr.error_reply;
+    }
+    trace_vfio_user_get_irq_info(msg.index, msg.flags, msg.count);
+
+    memcpy(info, &msg.argsz, sizeof(*info));
+    return 0;
+}
+
+static int irq_howmany(int *fdp, uint32_t cur, uint32_t max)
+{
+    int n = 0;
+
+    if (fdp[cur] != -1) {
+        do {
+            n++;
+        } while (n < max && fdp[cur + n] != -1);
+    } else {
+        do {
+            n++;
+        } while (n < max && fdp[cur + n] == -1);
+    }
+
+    return n;
+}
+
+static int vfio_user_device_io_set_irqs(VFIODevice *vbasedev,
+                                        struct vfio_irq_set *irq)
+{
+    VFIOUserProxy *proxy = vbasedev->proxy;
+    g_autofree VFIOUserIRQSet *msgp = NULL;
+    uint32_t size, nfds, send_fds, sent_fds, max;
+    Error *local_err = NULL;
+
+    if (irq->argsz < sizeof(*irq)) {
+        error_printf("vfio_user_set_irqs argsz too small\n");
+        return -EINVAL;
+    }
+
+    /*
+     * Handle simple case
+     */
+    if ((irq->flags & VFIO_IRQ_SET_DATA_EVENTFD) == 0) {
+        size = sizeof(VFIOUserHdr) + irq->argsz;
+        msgp = g_malloc0(size);
+
+        vfio_user_request_msg(&msgp->hdr, VFIO_USER_DEVICE_SET_IRQS, size, 0);
+        msgp->argsz = irq->argsz;
+        msgp->flags = irq->flags;
+        msgp->index = irq->index;
+        msgp->start = irq->start;
+        msgp->count = irq->count;
+        trace_vfio_user_set_irqs(msgp->index, msgp->start, msgp->count,
+                                 msgp->flags);
+
+        if (!vfio_user_send_wait(proxy, &msgp->hdr, NULL, 0, &local_err)) {
+            error_prepend(&local_err, "%s: ", __func__);
+            error_report_err(local_err);
+            return -EFAULT;
+        }
+
+        if (msgp->hdr.flags & VFIO_USER_ERROR) {
+            return -msgp->hdr.error_reply;
+        }
+
+        return 0;
+    }
+
+    /*
+     * Calculate the number of FDs to send
+     * and adjust argsz
+     */
+    nfds = (irq->argsz - sizeof(*irq)) / sizeof(int);
+    irq->argsz = sizeof(*irq);
+    msgp = g_malloc0(sizeof(*msgp));
+    /*
+     * Send in chunks if over max_send_fds
+     */
+    for (sent_fds = 0; nfds > sent_fds; sent_fds += send_fds) {
+        VFIOUserFDs *arg_fds, loop_fds;
+
+        /* must send all valid FDs or all invalid FDs in single msg */
+        max = nfds - sent_fds;
+        if (max > proxy->max_send_fds) {
+            max = proxy->max_send_fds;
+        }
+        send_fds = irq_howmany((int *)irq->data, sent_fds, max);
+
+        vfio_user_request_msg(&msgp->hdr, VFIO_USER_DEVICE_SET_IRQS,
+                              sizeof(*msgp), 0);
+        msgp->argsz = irq->argsz;
+        msgp->flags = irq->flags;
+        msgp->index = irq->index;
+        msgp->start = irq->start + sent_fds;
+        msgp->count = send_fds;
+        trace_vfio_user_set_irqs(msgp->index, msgp->start, msgp->count,
+                                 msgp->flags);
+
+        loop_fds.send_fds = send_fds;
+        loop_fds.recv_fds = 0;
+        loop_fds.fds = (int *)irq->data + sent_fds;
+        arg_fds = loop_fds.fds[0] != -1 ? &loop_fds : NULL;
+
+        if (!vfio_user_send_wait(proxy, &msgp->hdr, arg_fds, 0, &local_err)) {
+            error_prepend(&local_err, "%s: ", __func__);
+            error_report_err(local_err);
+            return -EFAULT;
+        }
+
+        if (msgp->hdr.flags & VFIO_USER_ERROR) {
+            return -msgp->hdr.error_reply;
+        }
+    }
+
+    return 0;
+}
+
+static int vfio_user_device_io_region_read(VFIODevice *vbasedev, uint8_t index,
+                                           off_t off, uint32_t count,
+                                           void *data)
+{
+    g_autofree VFIOUserRegionRW *msgp = NULL;
+    VFIOUserProxy *proxy = vbasedev->proxy;
+    int size = sizeof(*msgp) + count;
+    Error *local_err = NULL;
+
+    if (count > proxy->max_xfer_size) {
+        return -EINVAL;
+    }
+
+    msgp = g_malloc0(size);
+    vfio_user_request_msg(&msgp->hdr, VFIO_USER_REGION_READ, sizeof(*msgp), 0);
+    msgp->offset = off;
+    msgp->region = index;
+    msgp->count = count;
+    trace_vfio_user_region_rw(msgp->region, msgp->offset, msgp->count);
+
+    if (!vfio_user_send_wait(proxy, &msgp->hdr, NULL, size, &local_err)) {
+        error_prepend(&local_err, "%s: ", __func__);
+        error_report_err(local_err);
+        return -EFAULT;
+    }
+
+    if (msgp->hdr.flags & VFIO_USER_ERROR) {
+        return -msgp->hdr.error_reply;
+    } else if (msgp->count > count) {
+        return -E2BIG;
+    } else {
+        memcpy(data, &msgp->data, msgp->count);
+    }
+
+    return msgp->count;
+}
+
+/*
+ * If this is a posted write, and VFIO_PROXY_NO_POST is not set, then we are OK
+ * to send the write to the socket without waiting for the server's reply:
+ * a subsequent read (of any region) will not pass the posted write, as all
+ * messages are handled sequentially.
+ */
+static int vfio_user_device_io_region_write(VFIODevice *vbasedev, uint8_t index,
+                                            off_t off, unsigned count,
+                                            void *data, bool post)
+{
+    VFIOUserRegionRW *msgp = NULL;
+    VFIOUserProxy *proxy = vbasedev->proxy;
+    int size = sizeof(*msgp) + count;
+    Error *local_err = NULL;
+    bool can_multi;
+    int flags = 0;
+    int ret;
+
+    if (count > proxy->max_xfer_size) {
+        return -EINVAL;
+    }
+
+    if (proxy->flags & VFIO_PROXY_NO_POST) {
+        post = false;
+    }
+
+    if (post) {
+        flags |= VFIO_USER_NO_REPLY;
+    }
+
+    /* write eligible to be in a WRITE_MULTI msg ? */
+    can_multi = (proxy->flags & VFIO_PROXY_USE_MULTI) && post &&
+        count <= VFIO_USER_MULTI_DATA;
+
+    /*
+     * This should be a rare case, so first check without the lock,
+     * if we're wrong, vfio_send_queued() will flush any posted writes
+     * we missed here
+     */
+    if (proxy->wr_multi != NULL ||
+        (proxy->num_outgoing > VFIO_USER_OUT_HIGH && can_multi)) {
+
+        /*
+         * re-check with lock
+         *
+         * if already building a WRITE_MULTI msg,
+         *  add this one if possible else flush pending before
+         *  sending the current one
+         *
+         * else if outgoing queue is over the highwater,
+         *  start a new WRITE_MULTI message
+         */
+        WITH_QEMU_LOCK_GUARD(&proxy->lock) {
+            if (proxy->wr_multi != NULL) {
+                if (can_multi) {
+                    vfio_user_add_multi(proxy, index, off, count, data);
+                    return count;
+                }
+                vfio_user_flush_multi(proxy);
+            } else if (proxy->num_outgoing > VFIO_USER_OUT_HIGH && can_multi) {
+                vfio_user_create_multi(proxy);
+                vfio_user_add_multi(proxy, index, off, count, data);
+                return count;
+            }
+        }
+    }
+
+    msgp = g_malloc0(size);
+    vfio_user_request_msg(&msgp->hdr, VFIO_USER_REGION_WRITE, size, flags);
+    msgp->offset = off;
+    msgp->region = index;
+    msgp->count = count;
+    memcpy(&msgp->data, data, count);
+    trace_vfio_user_region_rw(msgp->region, msgp->offset, msgp->count);
+
+    /* async send will free msg after it's sent */
+    if (post) {
+        if (!vfio_user_send_async(proxy, &msgp->hdr, NULL, &local_err)) {
+            error_prepend(&local_err, "%s: ", __func__);
+            error_report_err(local_err);
+            return -EFAULT;
+        }
+
+        return count;
+    }
+
+    if (!vfio_user_send_wait(proxy, &msgp->hdr, NULL, 0, &local_err)) {
+        error_prepend(&local_err, "%s: ", __func__);
+        error_report_err(local_err);
+        g_free(msgp);
+        return -EFAULT;
+    }
+
+    if (msgp->hdr.flags & VFIO_USER_ERROR) {
+        ret = -msgp->hdr.error_reply;
+    } else {
+        ret = count;
+    }
+
+    g_free(msgp);
+    return ret;
+}
+
+/*
+ * Socket-based io_ops
+ */
+VFIODeviceIOOps vfio_user_device_io_ops_sock = {
+    .get_region_info = vfio_user_device_io_get_region_info,
+    .get_irq_info = vfio_user_device_io_get_irq_info,
+    .set_irqs = vfio_user_device_io_set_irqs,
+    .region_read = vfio_user_device_io_region_read,
+    .region_write = vfio_user_device_io_region_write,
+
+};
diff --git a/hw/vfio-user/device.h b/hw/vfio-user/device.h
new file mode 100644
index 0000000..d183a39
--- /dev/null
+++ b/hw/vfio-user/device.h
@@ -0,0 +1,24 @@
+#ifndef VFIO_USER_DEVICE_H
+#define VFIO_USER_DEVICE_H
+
+/*
+ * vfio protocol over a UNIX socket device handling.
+ *
+ * Copyright © 2018, 2021 Oracle and/or its affiliates.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include "qemu/osdep.h"
+#include "linux/vfio.h"
+
+#include "hw/vfio-user/proxy.h"
+
+bool vfio_user_get_device_info(VFIOUserProxy *proxy,
+                               struct vfio_device_info *info, Error **errp);
+
+void vfio_user_device_reset(VFIOUserProxy *proxy);
+
+extern VFIODeviceIOOps vfio_user_device_io_ops_sock;
+
+#endif /* VFIO_USER_DEVICE_H */
diff --git a/hw/vfio-user/meson.build b/hw/vfio-user/meson.build
new file mode 100644
index 0000000..2ed0ae5
--- /dev/null
+++ b/hw/vfio-user/meson.build
@@ -0,0 +1,11 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+vfio_user_ss = ss.source_set()
+vfio_user_ss.add(files(
+  'container.c',
+  'device.c',
+  'pci.c',
+  'proxy.c',
+))
+
+system_ss.add_all(when: 'CONFIG_VFIO_USER', if_true: vfio_user_ss)
diff --git a/hw/vfio-user/pci.c b/hw/vfio-user/pci.c
new file mode 100644
index 0000000..be71c77
--- /dev/null
+++ b/hw/vfio-user/pci.c
@@ -0,0 +1,475 @@
+/*
+ * vfio PCI device over a UNIX socket.
+ *
+ * Copyright © 2018, 2021 Oracle and/or its affiliates.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include <sys/ioctl.h>
+#include "qemu/osdep.h"
+#include "qapi-visit-sockets.h"
+#include "qemu/error-report.h"
+
+#include "hw/qdev-properties.h"
+#include "hw/vfio/pci.h"
+#include "hw/vfio-user/device.h"
+#include "hw/vfio-user/proxy.h"
+
+#define TYPE_VFIO_USER_PCI "vfio-user-pci"
+OBJECT_DECLARE_SIMPLE_TYPE(VFIOUserPCIDevice, VFIO_USER_PCI)
+
+struct VFIOUserPCIDevice {
+    VFIOPCIDevice device;
+    SocketAddress *socket;
+    bool send_queued;   /* all sends are queued */
+    uint32_t wait_time; /* timeout for message replies */
+    bool no_post;       /* all region writes are sync */
+};
+
+/*
+ * The server maintains the device's pending interrupts,
+ * via its MSIX table and PBA, so we treat these accesses
+ * like PCI config space and forward them.
+ */
+static uint64_t vfio_user_pba_read(void *opaque, hwaddr addr,
+                                   unsigned size)
+{
+    VFIOPCIDevice *vdev = opaque;
+    VFIORegion *region = &vdev->bars[vdev->msix->pba_bar].region;
+    uint64_t data;
+
+    /* server copy is what matters */
+    data = vfio_region_read(region, addr + vdev->msix->pba_offset, size);
+    return data;
+}
+
+static void vfio_user_pba_write(void *opaque, hwaddr addr,
+                                  uint64_t data, unsigned size)
+{
+    /* dropped */
+}
+
+static const MemoryRegionOps vfio_user_pba_ops = {
+    .read = vfio_user_pba_read,
+    .write = vfio_user_pba_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+};
+
+static void vfio_user_msix_setup(VFIOPCIDevice *vdev)
+{
+    MemoryRegion *vfio_reg, *msix_reg, *pba_reg;
+
+    pba_reg = g_new0(MemoryRegion, 1);
+    vdev->msix->pba_region = pba_reg;
+
+    vfio_reg = vdev->bars[vdev->msix->pba_bar].mr;
+    msix_reg = &vdev->pdev.msix_pba_mmio;
+    memory_region_init_io(pba_reg, OBJECT(vdev), &vfio_user_pba_ops, vdev,
+                          "VFIO MSIX PBA", int128_get64(msix_reg->size));
+    memory_region_add_subregion_overlap(vfio_reg, vdev->msix->pba_offset,
+                                        pba_reg, 1);
+}
+
+static void vfio_user_msix_teardown(VFIOPCIDevice *vdev)
+{
+    MemoryRegion *mr, *sub;
+
+    mr = vdev->bars[vdev->msix->pba_bar].mr;
+    sub = vdev->msix->pba_region;
+    memory_region_del_subregion(mr, sub);
+
+    g_free(vdev->msix->pba_region);
+    vdev->msix->pba_region = NULL;
+}
+
+static void vfio_user_dma_read(VFIOPCIDevice *vdev, VFIOUserDMARW *msg)
+{
+    PCIDevice *pdev = &vdev->pdev;
+    VFIOUserProxy *proxy = vdev->vbasedev.proxy;
+    VFIOUserDMARW *res;
+    MemTxResult r;
+    size_t size;
+
+    if (msg->hdr.size < sizeof(*msg)) {
+        vfio_user_send_error(proxy, &msg->hdr, EINVAL);
+        return;
+    }
+    if (msg->count > proxy->max_xfer_size) {
+        vfio_user_send_error(proxy, &msg->hdr, E2BIG);
+        return;
+    }
+
+    /* switch to our own message buffer */
+    size = msg->count + sizeof(VFIOUserDMARW);
+    res = g_malloc0(size);
+    memcpy(res, msg, sizeof(*res));
+    g_free(msg);
+
+    r = pci_dma_read(pdev, res->offset, &res->data, res->count);
+
+    switch (r) {
+    case MEMTX_OK:
+        if (res->hdr.flags & VFIO_USER_NO_REPLY) {
+            g_free(res);
+            return;
+        }
+        vfio_user_send_reply(proxy, &res->hdr, size);
+        break;
+    case MEMTX_ERROR:
+        vfio_user_send_error(proxy, &res->hdr, EFAULT);
+        break;
+    case MEMTX_DECODE_ERROR:
+        vfio_user_send_error(proxy, &res->hdr, ENODEV);
+        break;
+    case MEMTX_ACCESS_ERROR:
+        vfio_user_send_error(proxy, &res->hdr, EPERM);
+        break;
+    default:
+        error_printf("vfio_user_dma_read unknown error %d\n", r);
+        vfio_user_send_error(vdev->vbasedev.proxy, &res->hdr, EINVAL);
+    }
+}
+
+static void vfio_user_dma_write(VFIOPCIDevice *vdev, VFIOUserDMARW *msg)
+{
+    PCIDevice *pdev = &vdev->pdev;
+    VFIOUserProxy *proxy = vdev->vbasedev.proxy;
+    MemTxResult r;
+
+    if (msg->hdr.size < sizeof(*msg)) {
+        vfio_user_send_error(proxy, &msg->hdr, EINVAL);
+        return;
+    }
+    /* make sure transfer count isn't larger than the message data */
+    if (msg->count > msg->hdr.size - sizeof(*msg)) {
+        vfio_user_send_error(proxy, &msg->hdr, E2BIG);
+        return;
+    }
+
+    r = pci_dma_write(pdev, msg->offset, &msg->data, msg->count);
+
+    switch (r) {
+    case MEMTX_OK:
+        if ((msg->hdr.flags & VFIO_USER_NO_REPLY) == 0) {
+            vfio_user_send_reply(proxy, &msg->hdr, sizeof(msg->hdr));
+        } else {
+            g_free(msg);
+        }
+        break;
+    case MEMTX_ERROR:
+        vfio_user_send_error(proxy, &msg->hdr, EFAULT);
+        break;
+    case MEMTX_DECODE_ERROR:
+        vfio_user_send_error(proxy, &msg->hdr, ENODEV);
+        break;
+    case MEMTX_ACCESS_ERROR:
+        vfio_user_send_error(proxy, &msg->hdr, EPERM);
+        break;
+    default:
+        error_printf("vfio_user_dma_write unknown error %d\n", r);
+        vfio_user_send_error(vdev->vbasedev.proxy, &msg->hdr, EINVAL);
+    }
+}
+
+/*
+ * Incoming request message callback.
+ *
+ * Runs off main loop, so BQL held.
+ */
+static void vfio_user_pci_process_req(void *opaque, VFIOUserMsg *msg)
+{
+    VFIOPCIDevice *vdev = opaque;
+    VFIOUserHdr *hdr = msg->hdr;
+
+    /* no incoming PCI requests pass FDs */
+    if (msg->fds != NULL) {
+        vfio_user_send_error(vdev->vbasedev.proxy, hdr, EINVAL);
+        vfio_user_putfds(msg);
+        return;
+    }
+
+    switch (hdr->command) {
+    case VFIO_USER_DMA_READ:
+        vfio_user_dma_read(vdev, (VFIOUserDMARW *)hdr);
+        break;
+    case VFIO_USER_DMA_WRITE:
+        vfio_user_dma_write(vdev, (VFIOUserDMARW *)hdr);
+        break;
+    default:
+        error_printf("vfio_user_pci_process_req unknown cmd %d\n",
+                     hdr->command);
+        vfio_user_send_error(vdev->vbasedev.proxy, hdr, ENOSYS);
+    }
+}
+
+/*
+ * Emulated devices don't use host hot reset
+ */
+static void vfio_user_compute_needs_reset(VFIODevice *vbasedev)
+{
+    vbasedev->needs_reset = false;
+}
+
+static Object *vfio_user_pci_get_object(VFIODevice *vbasedev)
+{
+    VFIOUserPCIDevice *vdev = container_of(vbasedev, VFIOUserPCIDevice,
+                                           device.vbasedev);
+
+    return OBJECT(vdev);
+}
+
+static VFIODeviceOps vfio_user_pci_ops = {
+    .vfio_compute_needs_reset = vfio_user_compute_needs_reset,
+    .vfio_eoi = vfio_pci_intx_eoi,
+    .vfio_get_object = vfio_user_pci_get_object,
+    /* No live migration support yet. */
+    .vfio_save_config = NULL,
+    .vfio_load_config = NULL,
+};
+
+static void vfio_user_pci_realize(PCIDevice *pdev, Error **errp)
+{
+    ERRP_GUARD();
+    VFIOUserPCIDevice *udev = VFIO_USER_PCI(pdev);
+    VFIOPCIDevice *vdev = VFIO_PCI_BASE(pdev);
+    VFIODevice *vbasedev = &vdev->vbasedev;
+    const char *sock_name;
+    AddressSpace *as;
+    SocketAddress addr;
+    VFIOUserProxy *proxy;
+
+    if (!udev->socket) {
+        error_setg(errp, "No socket specified");
+        error_append_hint(errp, "e.g. -device '{"
+            "\"driver\":\"vfio-user-pci\", "
+            "\"socket\": {\"path\": \"/tmp/vfio-user.sock\", "
+            "\"type\": \"unix\"}'"
+            "}'\n");
+        return;
+    }
+
+    sock_name = udev->socket->u.q_unix.path;
+
+    vbasedev->name = g_strdup_printf("vfio-user:%s", sock_name);
+
+    memset(&addr, 0, sizeof(addr));
+    addr.type = SOCKET_ADDRESS_TYPE_UNIX;
+    addr.u.q_unix.path = (char *)sock_name;
+    proxy = vfio_user_connect_dev(&addr, errp);
+    if (!proxy) {
+        return;
+    }
+    vbasedev->proxy = proxy;
+    vfio_user_set_handler(vbasedev, vfio_user_pci_process_req, vdev);
+
+    vbasedev->name = g_strdup_printf("vfio-user:%s", sock_name);
+
+    if (udev->send_queued) {
+        proxy->flags |= VFIO_PROXY_FORCE_QUEUED;
+    }
+
+    if (udev->no_post) {
+        proxy->flags |= VFIO_PROXY_NO_POST;
+    }
+
+    /* user specified or 5 sec default */
+    proxy->wait_time = udev->wait_time;
+
+    if (!vfio_user_validate_version(proxy, errp)) {
+        goto error;
+    }
+
+    /*
+     * Use socket-based device I/O instead of vfio kernel driver.
+     */
+    vbasedev->io_ops = &vfio_user_device_io_ops_sock;
+
+    /*
+     * vfio-user devices are effectively mdevs (don't use a host iommu).
+     */
+    vbasedev->mdev = true;
+
+    /*
+     * Enable per-region fds.
+     */
+    vbasedev->use_region_fds = true;
+
+    as = pci_device_iommu_address_space(pdev);
+    if (!vfio_device_attach_by_iommu_type(TYPE_VFIO_IOMMU_USER,
+                                          vbasedev->name, vbasedev,
+                                          as, errp)) {
+        goto error;
+    }
+
+    if (!vfio_pci_populate_device(vdev, errp)) {
+        goto error;
+    }
+
+    if (!vfio_pci_config_setup(vdev, errp)) {
+        goto error;
+    }
+
+    /*
+     * vfio_pci_config_setup will have registered the device's BARs
+     * and setup any MSIX BARs, so errors after it succeeds must
+     * use out_teardown
+     */
+
+    if (!vfio_pci_add_capabilities(vdev, errp)) {
+        goto out_teardown;
+    }
+
+    if (vdev->msix != NULL) {
+        vfio_user_msix_setup(vdev);
+    }
+
+    if (!vfio_pci_interrupt_setup(vdev, errp)) {
+        goto out_teardown;
+    }
+
+    vfio_pci_register_err_notifier(vdev);
+    vfio_pci_register_req_notifier(vdev);
+
+    return;
+
+out_teardown:
+    vfio_pci_teardown_msi(vdev);
+    vfio_pci_bars_exit(vdev);
+error:
+    error_prepend(errp, VFIO_MSG_PREFIX, vdev->vbasedev.name);
+    vfio_pci_put_device(vdev);
+}
+
+static void vfio_user_instance_init(Object *obj)
+{
+    PCIDevice *pci_dev = PCI_DEVICE(obj);
+    VFIOPCIDevice *vdev = VFIO_PCI_BASE(obj);
+    VFIODevice *vbasedev = &vdev->vbasedev;
+
+    device_add_bootindex_property(obj, &vdev->bootindex,
+                                  "bootindex", NULL,
+                                  &pci_dev->qdev);
+    vdev->host.domain = ~0U;
+    vdev->host.bus = ~0U;
+    vdev->host.slot = ~0U;
+    vdev->host.function = ~0U;
+
+    vfio_device_init(vbasedev, VFIO_DEVICE_TYPE_PCI, &vfio_user_pci_ops,
+                     DEVICE(vdev), false);
+
+    vdev->nv_gpudirect_clique = 0xFF;
+
+    /*
+     * QEMU_PCI_CAP_EXPRESS initialization does not depend on QEMU command
+     * line, therefore, no need to wait to realize like other devices.
+     */
+    pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
+}
+
+static void vfio_user_instance_finalize(Object *obj)
+{
+    VFIOPCIDevice *vdev = VFIO_PCI_BASE(obj);
+    VFIODevice *vbasedev = &vdev->vbasedev;
+
+    if (vdev->msix != NULL) {
+        vfio_user_msix_teardown(vdev);
+    }
+
+    vfio_pci_put_device(vdev);
+
+    if (vbasedev->proxy != NULL) {
+        vfio_user_disconnect(vbasedev->proxy);
+    }
+}
+
+static void vfio_user_pci_reset(DeviceState *dev)
+{
+    VFIOPCIDevice *vdev = VFIO_PCI_BASE(dev);
+    VFIODevice *vbasedev = &vdev->vbasedev;
+
+    vfio_pci_pre_reset(vdev);
+
+    if (vbasedev->reset_works) {
+        vfio_user_device_reset(vbasedev->proxy);
+    }
+
+    vfio_pci_post_reset(vdev);
+}
+
+static const Property vfio_user_pci_dev_properties[] = {
+    DEFINE_PROP_UINT32("x-pci-vendor-id", VFIOPCIDevice,
+                       vendor_id, PCI_ANY_ID),
+    DEFINE_PROP_UINT32("x-pci-device-id", VFIOPCIDevice,
+                       device_id, PCI_ANY_ID),
+    DEFINE_PROP_UINT32("x-pci-sub-vendor-id", VFIOPCIDevice,
+                       sub_vendor_id, PCI_ANY_ID),
+    DEFINE_PROP_UINT32("x-pci-sub-device-id", VFIOPCIDevice,
+                       sub_device_id, PCI_ANY_ID),
+    DEFINE_PROP_BOOL("x-send-queued", VFIOUserPCIDevice, send_queued, false),
+    DEFINE_PROP_UINT32("x-msg-timeout", VFIOUserPCIDevice, wait_time, 5000),
+    DEFINE_PROP_BOOL("x-no-posted-writes", VFIOUserPCIDevice, no_post, false),
+};
+
+static void vfio_user_pci_set_socket(Object *obj, Visitor *v, const char *name,
+                                     void *opaque, Error **errp)
+{
+    VFIOUserPCIDevice *udev = VFIO_USER_PCI(obj);
+    bool success;
+
+    if (udev->device.vbasedev.proxy) {
+        error_setg(errp, "Proxy is connected");
+        return;
+    }
+
+    qapi_free_SocketAddress(udev->socket);
+
+    udev->socket = NULL;
+
+    success = visit_type_SocketAddress(v, name, &udev->socket, errp);
+
+    if (!success) {
+        return;
+    }
+
+    if (udev->socket->type != SOCKET_ADDRESS_TYPE_UNIX) {
+        error_setg(errp, "Unsupported socket type %s",
+                   SocketAddressType_str(udev->socket->type));
+        qapi_free_SocketAddress(udev->socket);
+        udev->socket = NULL;
+        return;
+    }
+}
+
+static void vfio_user_pci_dev_class_init(ObjectClass *klass, const void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    PCIDeviceClass *pdc = PCI_DEVICE_CLASS(klass);
+
+    device_class_set_legacy_reset(dc, vfio_user_pci_reset);
+    device_class_set_props(dc, vfio_user_pci_dev_properties);
+
+    object_class_property_add(klass, "socket", "SocketAddress", NULL,
+                              vfio_user_pci_set_socket, NULL, NULL);
+    object_class_property_set_description(klass, "socket",
+                                          "SocketAddress (UNIX sockets only)");
+
+    dc->desc = "VFIO over socket PCI device assignment";
+    pdc->realize = vfio_user_pci_realize;
+}
+
+static const TypeInfo vfio_user_pci_dev_info = {
+    .name = TYPE_VFIO_USER_PCI,
+    .parent = TYPE_VFIO_PCI_BASE,
+    .instance_size = sizeof(VFIOUserPCIDevice),
+    .class_init = vfio_user_pci_dev_class_init,
+    .instance_init = vfio_user_instance_init,
+    .instance_finalize = vfio_user_instance_finalize,
+};
+
+static void register_vfio_user_dev_type(void)
+{
+    type_register_static(&vfio_user_pci_dev_info);
+}
+
+ type_init(register_vfio_user_dev_type)
diff --git a/hw/vfio-user/protocol.h b/hw/vfio-user/protocol.h
new file mode 100644
index 0000000..3249a4a
--- /dev/null
+++ b/hw/vfio-user/protocol.h
@@ -0,0 +1,242 @@
+#ifndef VFIO_USER_PROTOCOL_H
+#define VFIO_USER_PROTOCOL_H
+
+/*
+ * vfio protocol over a UNIX socket.
+ *
+ * Copyright © 2018, 2021 Oracle and/or its affiliates.
+ *
+ * Each message has a standard header that describes the command
+ * being sent, which is almost always a VFIO ioctl().
+ *
+ * The header may be followed by command-specific data, such as the
+ * region and offset info for read and write commands.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+typedef struct {
+    uint16_t id;
+    uint16_t command;
+    uint32_t size;
+    uint32_t flags;
+    uint32_t error_reply;
+} VFIOUserHdr;
+
+/* VFIOUserHdr commands */
+enum vfio_user_command {
+    VFIO_USER_VERSION                   = 1,
+    VFIO_USER_DMA_MAP                   = 2,
+    VFIO_USER_DMA_UNMAP                 = 3,
+    VFIO_USER_DEVICE_GET_INFO           = 4,
+    VFIO_USER_DEVICE_GET_REGION_INFO    = 5,
+    VFIO_USER_DEVICE_GET_REGION_IO_FDS  = 6,
+    VFIO_USER_DEVICE_GET_IRQ_INFO       = 7,
+    VFIO_USER_DEVICE_SET_IRQS           = 8,
+    VFIO_USER_REGION_READ               = 9,
+    VFIO_USER_REGION_WRITE              = 10,
+    VFIO_USER_DMA_READ                  = 11,
+    VFIO_USER_DMA_WRITE                 = 12,
+    VFIO_USER_DEVICE_RESET              = 13,
+    VFIO_USER_DIRTY_PAGES               = 14,
+    VFIO_USER_REGION_WRITE_MULTI        = 15,
+    VFIO_USER_MAX,
+};
+
+/* VFIOUserHdr flags */
+#define VFIO_USER_REQUEST       0x0
+#define VFIO_USER_REPLY         0x1
+#define VFIO_USER_TYPE          0xF
+
+#define VFIO_USER_NO_REPLY      0x10
+#define VFIO_USER_ERROR         0x20
+
+
+/*
+ * VFIO_USER_VERSION
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint16_t major;
+    uint16_t minor;
+    char capabilities[];
+} VFIOUserVersion;
+
+#define VFIO_USER_MAJOR_VER     0
+#define VFIO_USER_MINOR_VER     0
+
+#define VFIO_USER_CAP           "capabilities"
+
+/* "capabilities" members */
+#define VFIO_USER_CAP_MAX_FDS   "max_msg_fds"
+#define VFIO_USER_CAP_MAX_XFER  "max_data_xfer_size"
+#define VFIO_USER_CAP_PGSIZES   "pgsizes"
+#define VFIO_USER_CAP_MAP_MAX   "max_dma_maps"
+#define VFIO_USER_CAP_MIGR      "migration"
+#define VFIO_USER_CAP_MULTI     "write_multiple"
+
+/* "migration" members */
+#define VFIO_USER_CAP_PGSIZE            "pgsize"
+#define VFIO_USER_CAP_MAX_BITMAP        "max_bitmap_size"
+
+/*
+ * Max FDs mainly comes into play when a device supports multiple interrupts
+ * where each ones uses an eventfd to inject it into the guest.
+ * It is clamped by the the number of FDs the qio channel supports in a
+ * single message.
+ */
+#define VFIO_USER_DEF_MAX_FDS   8
+#define VFIO_USER_MAX_MAX_FDS   16
+
+/*
+ * Max transfer limits the amount of data in region and DMA messages.
+ * Region R/W will be very small (limited by how much a single instruction
+ * can process) so just use a reasonable limit here.
+ */
+#define VFIO_USER_DEF_MAX_XFER  (1024 * 1024)
+#define VFIO_USER_MAX_MAX_XFER  (64 * 1024 * 1024)
+
+/*
+ * Default pagesizes supported is 4k.
+ */
+#define VFIO_USER_DEF_PGSIZE    4096
+
+/*
+ * Default max number of DMA mappings is stolen from the
+ * linux kernel "dma_entry_limit"
+ */
+#define VFIO_USER_DEF_MAP_MAX   65535
+
+/*
+ * Default max bitmap size is also take from the linux kernel,
+ * where usage of signed ints limits the VA range to 2^31 bytes.
+ * Dividing that by the number of bits per byte yields 256MB
+ */
+#define VFIO_USER_DEF_MAX_BITMAP (256 * 1024 * 1024)
+
+/*
+ * VFIO_USER_DMA_MAP
+ * imported from struct vfio_iommu_type1_dma_map
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint32_t argsz;
+    uint32_t flags;
+    uint64_t offset;    /* FD offset */
+    uint64_t iova;
+    uint64_t size;
+} VFIOUserDMAMap;
+
+/*
+ * VFIO_USER_DMA_UNMAP
+ * imported from struct vfio_iommu_type1_dma_unmap
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint32_t argsz;
+    uint32_t flags;
+    uint64_t iova;
+    uint64_t size;
+} VFIOUserDMAUnmap;
+
+/*
+ * VFIO_USER_DEVICE_GET_INFO
+ * imported from struct vfio_device_info
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint32_t argsz;
+    uint32_t flags;
+    uint32_t num_regions;
+    uint32_t num_irqs;
+} VFIOUserDeviceInfo;
+
+/*
+ * VFIO_USER_DEVICE_GET_REGION_INFO
+ * imported from struct vfio_region_info
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint32_t argsz;
+    uint32_t flags;
+    uint32_t index;
+    uint32_t cap_offset;
+    uint64_t size;
+    uint64_t offset;
+} VFIOUserRegionInfo;
+
+/*
+ * VFIO_USER_DEVICE_GET_IRQ_INFO
+ * imported from struct vfio_irq_info
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint32_t argsz;
+    uint32_t flags;
+    uint32_t index;
+    uint32_t count;
+} VFIOUserIRQInfo;
+
+/*
+ * VFIO_USER_DEVICE_SET_IRQS
+ * imported from struct vfio_irq_set
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint32_t argsz;
+    uint32_t flags;
+    uint32_t index;
+    uint32_t start;
+    uint32_t count;
+} VFIOUserIRQSet;
+
+/*
+ * VFIO_USER_REGION_READ
+ * VFIO_USER_REGION_WRITE
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint64_t offset;
+    uint32_t region;
+    uint32_t count;
+    char data[];
+} VFIOUserRegionRW;
+
+/*
+ * VFIO_USER_DMA_READ
+ * VFIO_USER_DMA_WRITE
+ */
+typedef struct {
+    VFIOUserHdr hdr;
+    uint64_t offset;
+    uint32_t count;
+    char data[];
+} VFIOUserDMARW;
+
+/* imported from struct vfio_bitmap */
+typedef struct {
+    uint64_t pgsize;
+    uint64_t size;
+    char data[];
+} VFIOUserBitmap;
+
+/*
+ * VFIO_USER_REGION_WRITE_MULTI
+ */
+#define VFIO_USER_MULTI_DATA  8
+#define VFIO_USER_MULTI_MAX   200
+
+typedef struct {
+    uint64_t offset;
+    uint32_t region;
+    uint32_t count;
+    char data[VFIO_USER_MULTI_DATA];
+} VFIOUserWROne;
+
+typedef struct {
+    VFIOUserHdr hdr;
+    uint64_t wr_cnt;
+    VFIOUserWROne wrs[VFIO_USER_MULTI_MAX];
+} VFIOUserWRMulti;
+
+#endif /* VFIO_USER_PROTOCOL_H */
diff --git a/hw/vfio-user/proxy.c b/hw/vfio-user/proxy.c
new file mode 100644
index 0000000..c418954
--- /dev/null
+++ b/hw/vfio-user/proxy.c
@@ -0,0 +1,1356 @@
+/*
+ * vfio protocol over a UNIX socket.
+ *
+ * Copyright © 2018, 2021 Oracle and/or its affiliates.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include "qemu/osdep.h"
+#include <sys/ioctl.h>
+
+#include "hw/vfio/vfio-device.h"
+#include "hw/vfio-user/proxy.h"
+#include "hw/vfio-user/trace.h"
+#include "qapi/error.h"
+#include "qobject/qbool.h"
+#include "qobject/qdict.h"
+#include "qobject/qjson.h"
+#include "qobject/qnum.h"
+#include "qemu/error-report.h"
+#include "qemu/lockable.h"
+#include "qemu/main-loop.h"
+#include "qemu/thread.h"
+#include "system/iothread.h"
+
+static IOThread *vfio_user_iothread;
+
+static void vfio_user_shutdown(VFIOUserProxy *proxy);
+static VFIOUserMsg *vfio_user_getmsg(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                                     VFIOUserFDs *fds);
+static void vfio_user_recycle(VFIOUserProxy *proxy, VFIOUserMsg *msg);
+
+static void vfio_user_recv(void *opaque);
+static void vfio_user_send(void *opaque);
+static void vfio_user_cb(void *opaque);
+
+static void vfio_user_request(void *opaque);
+
+static inline void vfio_user_set_error(VFIOUserHdr *hdr, uint32_t err)
+{
+    hdr->flags |= VFIO_USER_ERROR;
+    hdr->error_reply = err;
+}
+
+/*
+ * Functions called by main, CPU, or iothread threads
+ */
+
+static void vfio_user_shutdown(VFIOUserProxy *proxy)
+{
+    qio_channel_shutdown(proxy->ioc, QIO_CHANNEL_SHUTDOWN_READ, NULL);
+    qio_channel_set_aio_fd_handler(proxy->ioc, proxy->ctx, NULL,
+                                   proxy->ctx, NULL, NULL);
+}
+
+/*
+ * Same return values as qio_channel_writev_full():
+ *
+ * QIO_CHANNEL_ERR_BLOCK: *errp not set
+ * -1: *errp will be populated
+ * otherwise: bytes written
+ */
+static ssize_t vfio_user_send_qio(VFIOUserProxy *proxy, VFIOUserMsg *msg,
+                                  Error **errp)
+{
+    VFIOUserFDs *fds =  msg->fds;
+    struct iovec iov = {
+        .iov_base = msg->hdr,
+        .iov_len = msg->hdr->size,
+    };
+    size_t numfds = 0;
+    int *fdp = NULL;
+    ssize_t ret;
+
+    if (fds != NULL && fds->send_fds != 0) {
+        numfds = fds->send_fds;
+        fdp = fds->fds;
+    }
+
+    ret = qio_channel_writev_full(proxy->ioc, &iov, 1, fdp, numfds, 0, errp);
+
+    if (ret == -1) {
+        vfio_user_set_error(msg->hdr, EIO);
+        vfio_user_shutdown(proxy);
+    }
+    trace_vfio_user_send_write(msg->hdr->id, ret);
+
+    return ret;
+}
+
+static VFIOUserMsg *vfio_user_getmsg(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                                     VFIOUserFDs *fds)
+{
+    VFIOUserMsg *msg;
+
+    msg = QTAILQ_FIRST(&proxy->free);
+    if (msg != NULL) {
+        QTAILQ_REMOVE(&proxy->free, msg, next);
+    } else {
+        msg = g_malloc0(sizeof(*msg));
+        qemu_cond_init(&msg->cv);
+    }
+
+    msg->hdr = hdr;
+    msg->fds = fds;
+    return msg;
+}
+
+/*
+ * Recycle a message list entry to the free list.
+ */
+static void vfio_user_recycle(VFIOUserProxy *proxy, VFIOUserMsg *msg)
+{
+    if (msg->type == VFIO_MSG_NONE) {
+        error_printf("vfio_user_recycle - freeing free msg\n");
+        return;
+    }
+
+    /* free msg buffer if no one is waiting to consume the reply */
+    if (msg->type == VFIO_MSG_NOWAIT || msg->type == VFIO_MSG_ASYNC) {
+        g_free(msg->hdr);
+        if (msg->fds != NULL) {
+            g_free(msg->fds);
+        }
+    }
+
+    msg->type = VFIO_MSG_NONE;
+    msg->hdr = NULL;
+    msg->fds = NULL;
+    msg->complete = false;
+    msg->pending = false;
+    QTAILQ_INSERT_HEAD(&proxy->free, msg, next);
+}
+
+VFIOUserFDs *vfio_user_getfds(int numfds)
+{
+    VFIOUserFDs *fds = g_malloc0(sizeof(*fds) + (numfds * sizeof(int)));
+
+    fds->fds = (int *)((char *)fds + sizeof(*fds));
+
+    return fds;
+}
+
+/*
+ * Functions only called by iothread
+ */
+
+/*
+ * Process a received message.
+ */
+static void vfio_user_process(VFIOUserProxy *proxy, VFIOUserMsg *msg,
+                              bool isreply)
+{
+
+    /*
+     * Replies signal a waiter, if none just check for errors
+     * and free the message buffer.
+     *
+     * Requests get queued for the BH.
+     */
+    if (isreply) {
+        msg->complete = true;
+        if (msg->type == VFIO_MSG_WAIT) {
+            qemu_cond_signal(&msg->cv);
+        } else {
+            if (msg->hdr->flags & VFIO_USER_ERROR) {
+                error_printf("vfio_user_process: error reply on async ");
+                error_printf("request command %x error %s\n",
+                             msg->hdr->command,
+                             strerror(msg->hdr->error_reply));
+            }
+            /* youngest nowait msg has been ack'd */
+            if (proxy->last_nowait == msg) {
+                proxy->last_nowait = NULL;
+            }
+            vfio_user_recycle(proxy, msg);
+        }
+    } else {
+        QTAILQ_INSERT_TAIL(&proxy->incoming, msg, next);
+        qemu_bh_schedule(proxy->req_bh);
+    }
+}
+
+/*
+ * Complete a partial message read
+ */
+static int vfio_user_complete(VFIOUserProxy *proxy, Error **errp)
+{
+    VFIOUserMsg *msg = proxy->part_recv;
+    size_t msgleft = proxy->recv_left;
+    bool isreply;
+    char *data;
+    int ret;
+
+    data = (char *)msg->hdr + (msg->hdr->size - msgleft);
+    while (msgleft > 0) {
+        ret = qio_channel_read(proxy->ioc, data, msgleft, errp);
+
+        /* error or would block */
+        if (ret <= 0) {
+            /* try for rest on next iternation */
+            if (ret == QIO_CHANNEL_ERR_BLOCK) {
+                proxy->recv_left = msgleft;
+            }
+            return ret;
+        }
+        trace_vfio_user_recv_read(msg->hdr->id, ret);
+
+        msgleft -= ret;
+        data += ret;
+    }
+
+    /*
+     * Read complete message, process it.
+     */
+    proxy->part_recv = NULL;
+    proxy->recv_left = 0;
+    isreply = (msg->hdr->flags & VFIO_USER_TYPE) == VFIO_USER_REPLY;
+    vfio_user_process(proxy, msg, isreply);
+
+    /* return positive value */
+    return 1;
+}
+
+/*
+ * Receive and process one incoming message.
+ *
+ * For replies, find matching outgoing request and wake any waiters.
+ * For requests, queue in incoming list and run request BH.
+ */
+static int vfio_user_recv_one(VFIOUserProxy *proxy, Error **errp)
+{
+    VFIOUserMsg *msg = NULL;
+    g_autofree int *fdp = NULL;
+    VFIOUserFDs *reqfds;
+    VFIOUserHdr hdr;
+    struct iovec iov = {
+        .iov_base = &hdr,
+        .iov_len = sizeof(hdr),
+    };
+    bool isreply = false;
+    int i, ret;
+    size_t msgleft, numfds = 0;
+    char *data = NULL;
+    char *buf = NULL;
+
+    /*
+     * Complete any partial reads
+     */
+    if (proxy->part_recv != NULL) {
+        ret = vfio_user_complete(proxy, errp);
+
+        /* still not complete, try later */
+        if (ret == QIO_CHANNEL_ERR_BLOCK) {
+            return ret;
+        }
+
+        if (ret <= 0) {
+            goto fatal;
+        }
+        /* else fall into reading another msg */
+    }
+
+    /*
+     * Read header
+     */
+    ret = qio_channel_readv_full(proxy->ioc, &iov, 1, &fdp, &numfds, 0,
+                                 errp);
+    if (ret == QIO_CHANNEL_ERR_BLOCK) {
+        return ret;
+    }
+
+    /* read error or other side closed connection */
+    if (ret <= 0) {
+        goto fatal;
+    }
+
+    if (ret < sizeof(hdr)) {
+        error_setg(errp, "short read of header");
+        goto fatal;
+    }
+
+    /*
+     * Validate header
+     */
+    if (hdr.size < sizeof(VFIOUserHdr)) {
+        error_setg(errp, "bad header size");
+        goto fatal;
+    }
+    switch (hdr.flags & VFIO_USER_TYPE) {
+    case VFIO_USER_REQUEST:
+        isreply = false;
+        break;
+    case VFIO_USER_REPLY:
+        isreply = true;
+        break;
+    default:
+        error_setg(errp, "unknown message type");
+        goto fatal;
+    }
+    trace_vfio_user_recv_hdr(proxy->sockname, hdr.id, hdr.command, hdr.size,
+                             hdr.flags);
+
+    /*
+     * For replies, find the matching pending request.
+     * For requests, reap incoming FDs.
+     */
+    if (isreply) {
+        QTAILQ_FOREACH(msg, &proxy->pending, next) {
+            if (hdr.id == msg->id) {
+                break;
+            }
+        }
+        if (msg == NULL) {
+            error_setg(errp, "unexpected reply");
+            goto err;
+        }
+        QTAILQ_REMOVE(&proxy->pending, msg, next);
+
+        /*
+         * Process any received FDs
+         */
+        if (numfds != 0) {
+            if (msg->fds == NULL || msg->fds->recv_fds < numfds) {
+                error_setg(errp, "unexpected FDs");
+                goto err;
+            }
+            msg->fds->recv_fds = numfds;
+            memcpy(msg->fds->fds, fdp, numfds * sizeof(int));
+        }
+    } else {
+        if (numfds != 0) {
+            reqfds = vfio_user_getfds(numfds);
+            memcpy(reqfds->fds, fdp, numfds * sizeof(int));
+        } else {
+            reqfds = NULL;
+        }
+    }
+
+    /*
+     * Put the whole message into a single buffer.
+     */
+    if (isreply) {
+        if (hdr.size > msg->rsize) {
+            error_setg(errp, "reply larger than recv buffer");
+            goto err;
+        }
+        *msg->hdr = hdr;
+        data = (char *)msg->hdr + sizeof(hdr);
+    } else {
+        if (hdr.size > proxy->max_xfer_size + sizeof(VFIOUserDMARW)) {
+            error_setg(errp, "vfio_user_recv request larger than max");
+            goto err;
+        }
+        buf = g_malloc0(hdr.size);
+        memcpy(buf, &hdr, sizeof(hdr));
+        data = buf + sizeof(hdr);
+        msg = vfio_user_getmsg(proxy, (VFIOUserHdr *)buf, reqfds);
+        msg->type = VFIO_MSG_REQ;
+    }
+
+    /*
+     * Read rest of message.
+     */
+    msgleft = hdr.size - sizeof(hdr);
+    while (msgleft > 0) {
+        ret = qio_channel_read(proxy->ioc, data, msgleft, errp);
+
+        /* prepare to complete read on next iternation */
+        if (ret == QIO_CHANNEL_ERR_BLOCK) {
+            proxy->part_recv = msg;
+            proxy->recv_left = msgleft;
+            return ret;
+        }
+
+        if (ret <= 0) {
+            goto fatal;
+        }
+        trace_vfio_user_recv_read(hdr.id, ret);
+
+        msgleft -= ret;
+        data += ret;
+    }
+
+    vfio_user_process(proxy, msg, isreply);
+    return 0;
+
+    /*
+     * fatal means the other side closed or we don't trust the stream
+     * err means this message is corrupt
+     */
+fatal:
+    vfio_user_shutdown(proxy);
+    proxy->state = VFIO_PROXY_ERROR;
+
+    /* set error if server side closed */
+    if (ret == 0) {
+        error_setg(errp, "server closed socket");
+    }
+
+err:
+    for (i = 0; i < numfds; i++) {
+        close(fdp[i]);
+    }
+    if (isreply && msg != NULL) {
+        /* force an error to keep sending thread from hanging */
+        vfio_user_set_error(msg->hdr, EINVAL);
+        msg->complete = true;
+        qemu_cond_signal(&msg->cv);
+    }
+    return -1;
+}
+
+static void vfio_user_recv(void *opaque)
+{
+    VFIOUserProxy *proxy = opaque;
+
+    QEMU_LOCK_GUARD(&proxy->lock);
+
+    if (proxy->state == VFIO_PROXY_CONNECTED) {
+        Error *local_err = NULL;
+
+        while (vfio_user_recv_one(proxy, &local_err) == 0) {
+            ;
+        }
+
+        if (local_err != NULL) {
+            error_report_err(local_err);
+        }
+    }
+}
+
+/*
+ * Send a single message, same return semantics as vfio_user_send_qio().
+ *
+ * Sent async messages are freed, others are moved to pending queue.
+ */
+static ssize_t vfio_user_send_one(VFIOUserProxy *proxy, Error **errp)
+{
+    VFIOUserMsg *msg;
+    ssize_t ret;
+
+    msg = QTAILQ_FIRST(&proxy->outgoing);
+    ret = vfio_user_send_qio(proxy, msg, errp);
+    if (ret < 0) {
+        return ret;
+    }
+
+    QTAILQ_REMOVE(&proxy->outgoing, msg, next);
+    proxy->num_outgoing--;
+    if (msg->type == VFIO_MSG_ASYNC) {
+        vfio_user_recycle(proxy, msg);
+    } else {
+        QTAILQ_INSERT_TAIL(&proxy->pending, msg, next);
+        msg->pending = true;
+    }
+
+    return ret;
+}
+
+/*
+ * Send messages from outgoing queue when the socket buffer has space.
+ * If we deplete 'outgoing', remove ourselves from the poll list.
+ */
+static void vfio_user_send(void *opaque)
+{
+    VFIOUserProxy *proxy = opaque;
+
+    QEMU_LOCK_GUARD(&proxy->lock);
+
+    if (proxy->state == VFIO_PROXY_CONNECTED) {
+        while (!QTAILQ_EMPTY(&proxy->outgoing)) {
+            Error *local_err = NULL;
+            int ret;
+
+            ret = vfio_user_send_one(proxy, &local_err);
+
+            if (ret == QIO_CHANNEL_ERR_BLOCK) {
+                return;
+            } else if (ret == -1) {
+                error_report_err(local_err);
+                return;
+            }
+        }
+        qio_channel_set_aio_fd_handler(proxy->ioc, proxy->ctx,
+                                       vfio_user_recv, NULL, NULL, proxy);
+
+        /* queue empty - send any pending multi write msgs */
+        if (proxy->wr_multi != NULL) {
+            vfio_user_flush_multi(proxy);
+        }
+    }
+}
+
+static void vfio_user_cb(void *opaque)
+{
+    VFIOUserProxy *proxy = opaque;
+
+    QEMU_LOCK_GUARD(&proxy->lock);
+
+    proxy->state = VFIO_PROXY_CLOSED;
+    qemu_cond_signal(&proxy->close_cv);
+}
+
+
+/*
+ * Functions called by main or CPU threads
+ */
+
+/*
+ * Process incoming requests.
+ *
+ * The bus-specific callback has the form:
+ *    request(opaque, msg)
+ * where 'opaque' was specified in vfio_user_set_handler
+ * and 'msg' is the inbound message.
+ *
+ * The callback is responsible for disposing of the message buffer,
+ * usually by re-using it when calling vfio_send_reply or vfio_send_error,
+ * both of which free their message buffer when the reply is sent.
+ *
+ * If the callback uses a new buffer, it needs to free the old one.
+ */
+static void vfio_user_request(void *opaque)
+{
+    VFIOUserProxy *proxy = opaque;
+    VFIOUserMsgQ new, free;
+    VFIOUserMsg *msg, *m1;
+
+    /* reap all incoming */
+    QTAILQ_INIT(&new);
+    WITH_QEMU_LOCK_GUARD(&proxy->lock) {
+        QTAILQ_FOREACH_SAFE(msg, &proxy->incoming, next, m1) {
+            QTAILQ_REMOVE(&proxy->incoming, msg, next);
+            QTAILQ_INSERT_TAIL(&new, msg, next);
+        }
+    }
+
+    /* process list */
+    QTAILQ_INIT(&free);
+    QTAILQ_FOREACH_SAFE(msg, &new, next, m1) {
+        QTAILQ_REMOVE(&new, msg, next);
+        trace_vfio_user_recv_request(msg->hdr->command);
+        proxy->request(proxy->req_arg, msg);
+        QTAILQ_INSERT_HEAD(&free, msg, next);
+    }
+
+    /* free list */
+    WITH_QEMU_LOCK_GUARD(&proxy->lock) {
+        QTAILQ_FOREACH_SAFE(msg, &free, next, m1) {
+            vfio_user_recycle(proxy, msg);
+        }
+    }
+}
+
+/*
+ * Messages are queued onto the proxy's outgoing list.
+ *
+ * It handles 3 types of messages:
+ *
+ * async messages - replies and posted writes
+ *
+ * There will be no reply from the server, so message
+ * buffers are freed after they're sent.
+ *
+ * nowait messages - map/unmap during address space transactions
+ *
+ * These are also sent async, but a reply is expected so that
+ * vfio_wait_reqs() can wait for the youngest nowait request.
+ * They transition from the outgoing list to the pending list
+ * when sent, and are freed when the reply is received.
+ *
+ * wait messages - all other requests
+ *
+ * The reply to these messages is waited for by their caller.
+ * They also transition from outgoing to pending when sent, but
+ * the message buffer is returned to the caller with the reply
+ * contents.  The caller is responsible for freeing these messages.
+ *
+ * As an optimization, if the outgoing list and the socket send
+ * buffer are empty, the message is sent inline instead of being
+ * added to the outgoing list.  The rest of the transitions are
+ * unchanged.
+ */
+static bool vfio_user_send_queued(VFIOUserProxy *proxy, VFIOUserMsg *msg,
+                                  Error **errp)
+{
+    int ret;
+
+    /* older coalesced writes go first */
+    if (proxy->wr_multi != NULL &&
+        ((msg->hdr->flags & VFIO_USER_TYPE) == VFIO_USER_REQUEST)) {
+        vfio_user_flush_multi(proxy);
+    }
+
+    /*
+     * Unsent outgoing msgs - add to tail
+     */
+    if (!QTAILQ_EMPTY(&proxy->outgoing)) {
+        QTAILQ_INSERT_TAIL(&proxy->outgoing, msg, next);
+        proxy->num_outgoing++;
+        return true;
+    }
+
+    /*
+     * Try inline - if blocked, queue it and kick send poller
+     */
+    if (proxy->flags & VFIO_PROXY_FORCE_QUEUED) {
+        ret = QIO_CHANNEL_ERR_BLOCK;
+    } else {
+        ret = vfio_user_send_qio(proxy, msg, errp);
+    }
+
+    if (ret == QIO_CHANNEL_ERR_BLOCK) {
+        QTAILQ_INSERT_HEAD(&proxy->outgoing, msg, next);
+        proxy->num_outgoing = 1;
+        qio_channel_set_aio_fd_handler(proxy->ioc, proxy->ctx,
+                                       vfio_user_recv, proxy->ctx,
+                                       vfio_user_send, proxy);
+        return true;
+    }
+    if (ret == -1) {
+        return false;
+    }
+
+    /*
+     * Sent - free async, add others to pending
+     */
+    if (msg->type == VFIO_MSG_ASYNC) {
+        vfio_user_recycle(proxy, msg);
+    } else {
+        QTAILQ_INSERT_TAIL(&proxy->pending, msg, next);
+        msg->pending = true;
+    }
+
+    return true;
+}
+
+/*
+ * nowait send - vfio_wait_reqs() can wait for it later
+ *
+ * Returns false if we did not successfully receive a reply message, in which
+ * case @errp will be populated.
+ *
+ * In either case, ownership of @hdr and @fds is taken, and the caller must
+ * *not* free them itself.
+ */
+bool vfio_user_send_nowait(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                           VFIOUserFDs *fds, int rsize, Error **errp)
+{
+    VFIOUserMsg *msg;
+
+    QEMU_LOCK_GUARD(&proxy->lock);
+
+    msg = vfio_user_getmsg(proxy, hdr, fds);
+    msg->id = hdr->id;
+    msg->rsize = rsize ? rsize : hdr->size;
+    msg->type = VFIO_MSG_NOWAIT;
+
+    if (hdr->flags & VFIO_USER_NO_REPLY) {
+        error_setg_errno(errp, EINVAL, "%s on NO_REPLY message", __func__);
+        vfio_user_recycle(proxy, msg);
+        return false;
+    }
+
+    if (!vfio_user_send_queued(proxy, msg, errp)) {
+        vfio_user_recycle(proxy, msg);
+        return false;
+    }
+
+    proxy->last_nowait = msg;
+
+    return true;
+}
+
+/*
+ * Returns false if we did not successfully receive a reply message, in which
+ * case @errp will be populated.
+ *
+ * In either case, the caller must free @hdr and @fds if needed.
+ */
+bool vfio_user_send_wait(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                         VFIOUserFDs *fds, int rsize, Error **errp)
+{
+    VFIOUserMsg *msg;
+    bool ok = false;
+
+    if (hdr->flags & VFIO_USER_NO_REPLY) {
+        error_setg_errno(errp, EINVAL, "%s on NO_REPLY message", __func__);
+        return false;
+    }
+
+    qemu_mutex_lock(&proxy->lock);
+
+    msg = vfio_user_getmsg(proxy, hdr, fds);
+    msg->id = hdr->id;
+    msg->rsize = rsize ? rsize : hdr->size;
+    msg->type = VFIO_MSG_WAIT;
+
+    ok = vfio_user_send_queued(proxy, msg, errp);
+
+    if (ok) {
+        while (!msg->complete) {
+            if (!qemu_cond_timedwait(&msg->cv, &proxy->lock,
+                                     proxy->wait_time)) {
+                VFIOUserMsgQ *list;
+
+                list = msg->pending ? &proxy->pending : &proxy->outgoing;
+                QTAILQ_REMOVE(list, msg, next);
+                error_setg_errno(errp, ETIMEDOUT,
+                                 "timed out waiting for reply");
+                ok = false;
+                break;
+            }
+        }
+    }
+
+    vfio_user_recycle(proxy, msg);
+
+    qemu_mutex_unlock(&proxy->lock);
+
+    return ok;
+}
+
+/*
+ * async send - msg can be queued, but will be freed when sent
+ *
+ * Returns false on failure, in which case @errp will be populated.
+ *
+ * In either case, ownership of @hdr and @fds is taken, and the caller must
+ * *not* free them itself.
+ */
+bool vfio_user_send_async(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                          VFIOUserFDs *fds, Error **errp)
+{
+    VFIOUserMsg *msg;
+
+    QEMU_LOCK_GUARD(&proxy->lock);
+
+    msg = vfio_user_getmsg(proxy, hdr, fds);
+    msg->id = hdr->id;
+    msg->rsize = 0;
+    msg->type = VFIO_MSG_ASYNC;
+
+    if (!(hdr->flags & (VFIO_USER_NO_REPLY | VFIO_USER_REPLY))) {
+        error_setg_errno(errp, EINVAL, "%s on sync message", __func__);
+        vfio_user_recycle(proxy, msg);
+        return false;
+    }
+
+    if (!vfio_user_send_queued(proxy, msg, errp)) {
+        vfio_user_recycle(proxy, msg);
+        return false;
+    }
+
+    return true;
+}
+
+void vfio_user_wait_reqs(VFIOUserProxy *proxy)
+{
+    VFIOUserMsg *msg;
+
+    /*
+     * Any DMA map/unmap requests sent in the middle
+     * of a memory region transaction were sent nowait.
+     * Wait for them here.
+     */
+    qemu_mutex_lock(&proxy->lock);
+    if (proxy->last_nowait != NULL) {
+        /*
+         * Change type to WAIT to wait for reply
+         */
+        msg = proxy->last_nowait;
+        msg->type = VFIO_MSG_WAIT;
+        proxy->last_nowait = NULL;
+        while (!msg->complete) {
+            if (!qemu_cond_timedwait(&msg->cv, &proxy->lock,
+                                     proxy->wait_time)) {
+                VFIOUserMsgQ *list;
+
+                list = msg->pending ? &proxy->pending : &proxy->outgoing;
+                QTAILQ_REMOVE(list, msg, next);
+                error_printf("vfio_wait_reqs - timed out\n");
+                break;
+            }
+        }
+
+        if (msg->hdr->flags & VFIO_USER_ERROR) {
+            error_printf("vfio_user_wait_reqs - error reply on async ");
+            error_printf("request: command %x error %s\n", msg->hdr->command,
+                         strerror(msg->hdr->error_reply));
+        }
+
+        /*
+         * Change type back to NOWAIT to free
+         */
+        msg->type = VFIO_MSG_NOWAIT;
+        vfio_user_recycle(proxy, msg);
+    }
+
+    qemu_mutex_unlock(&proxy->lock);
+}
+
+/*
+ * Reply to an incoming request.
+ */
+void vfio_user_send_reply(VFIOUserProxy *proxy, VFIOUserHdr *hdr, int size)
+{
+    Error *local_err = NULL;
+
+    if (size < sizeof(VFIOUserHdr)) {
+        error_printf("%s: size too small", __func__);
+        g_free(hdr);
+        return;
+    }
+
+    /*
+     * convert header to associated reply
+     */
+    hdr->flags = VFIO_USER_REPLY;
+    hdr->size = size;
+
+    if (!vfio_user_send_async(proxy, hdr, NULL, &local_err)) {
+        error_report_err(local_err);
+    }
+}
+
+/*
+ * Send an error reply to an incoming request.
+ */
+void vfio_user_send_error(VFIOUserProxy *proxy, VFIOUserHdr *hdr, int error)
+{
+    Error *local_err = NULL;
+
+    /*
+     * convert header to associated reply
+     */
+    hdr->flags = VFIO_USER_REPLY;
+    hdr->flags |= VFIO_USER_ERROR;
+    hdr->error_reply = error;
+    hdr->size = sizeof(*hdr);
+
+    if (!vfio_user_send_async(proxy, hdr, NULL, &local_err)) {
+        error_report_err(local_err);
+    }
+}
+
+/*
+ * Close FDs erroneously received in an incoming request.
+ */
+void vfio_user_putfds(VFIOUserMsg *msg)
+{
+    VFIOUserFDs *fds = msg->fds;
+    int i;
+
+    for (i = 0; i < fds->recv_fds; i++) {
+        close(fds->fds[i]);
+    }
+    g_free(fds);
+    msg->fds = NULL;
+}
+
+void
+vfio_user_disable_posted_writes(VFIOUserProxy *proxy)
+{
+    WITH_QEMU_LOCK_GUARD(&proxy->lock) {
+         proxy->flags |= VFIO_PROXY_NO_POST;
+    }
+}
+
+static QLIST_HEAD(, VFIOUserProxy) vfio_user_sockets =
+    QLIST_HEAD_INITIALIZER(vfio_user_sockets);
+
+VFIOUserProxy *vfio_user_connect_dev(SocketAddress *addr, Error **errp)
+{
+    VFIOUserProxy *proxy;
+    QIOChannelSocket *sioc;
+    QIOChannel *ioc;
+    char *sockname;
+
+    if (addr->type != SOCKET_ADDRESS_TYPE_UNIX) {
+        error_setg(errp, "vfio_user_connect - bad address family");
+        return NULL;
+    }
+    sockname = addr->u.q_unix.path;
+
+    sioc = qio_channel_socket_new();
+    ioc = QIO_CHANNEL(sioc);
+    if (qio_channel_socket_connect_sync(sioc, addr, errp)) {
+        object_unref(OBJECT(ioc));
+        return NULL;
+    }
+    qio_channel_set_blocking(ioc, false, NULL);
+
+    proxy = g_malloc0(sizeof(VFIOUserProxy));
+    proxy->sockname = g_strdup_printf("unix:%s", sockname);
+    proxy->ioc = ioc;
+
+    /* init defaults */
+    proxy->max_xfer_size = VFIO_USER_DEF_MAX_XFER;
+    proxy->max_send_fds = VFIO_USER_DEF_MAX_FDS;
+    proxy->max_dma = VFIO_USER_DEF_MAP_MAX;
+    proxy->dma_pgsizes = VFIO_USER_DEF_PGSIZE;
+    proxy->max_bitmap = VFIO_USER_DEF_MAX_BITMAP;
+    proxy->migr_pgsize = VFIO_USER_DEF_PGSIZE;
+
+    proxy->flags = VFIO_PROXY_CLIENT;
+    proxy->state = VFIO_PROXY_CONNECTED;
+
+    qemu_mutex_init(&proxy->lock);
+    qemu_cond_init(&proxy->close_cv);
+
+    if (vfio_user_iothread == NULL) {
+        vfio_user_iothread = iothread_create("VFIO user", errp);
+    }
+
+    proxy->ctx = iothread_get_aio_context(vfio_user_iothread);
+    proxy->req_bh = qemu_bh_new(vfio_user_request, proxy);
+
+    QTAILQ_INIT(&proxy->outgoing);
+    QTAILQ_INIT(&proxy->incoming);
+    QTAILQ_INIT(&proxy->free);
+    QTAILQ_INIT(&proxy->pending);
+    QLIST_INSERT_HEAD(&vfio_user_sockets, proxy, next);
+
+    return proxy;
+}
+
+void vfio_user_set_handler(VFIODevice *vbasedev,
+                           void (*handler)(void *opaque, VFIOUserMsg *msg),
+                           void *req_arg)
+{
+    VFIOUserProxy *proxy = vbasedev->proxy;
+
+    proxy->request = handler;
+    proxy->req_arg = req_arg;
+    qio_channel_set_aio_fd_handler(proxy->ioc, proxy->ctx,
+                                   vfio_user_recv, NULL, NULL, proxy);
+}
+
+void vfio_user_disconnect(VFIOUserProxy *proxy)
+{
+    VFIOUserMsg *r1, *r2;
+
+    qemu_mutex_lock(&proxy->lock);
+
+    /* our side is quitting */
+    if (proxy->state == VFIO_PROXY_CONNECTED) {
+        vfio_user_shutdown(proxy);
+        if (!QTAILQ_EMPTY(&proxy->pending)) {
+            error_printf("vfio_user_disconnect: outstanding requests\n");
+        }
+    }
+    object_unref(OBJECT(proxy->ioc));
+    proxy->ioc = NULL;
+    qemu_bh_delete(proxy->req_bh);
+    proxy->req_bh = NULL;
+
+    proxy->state = VFIO_PROXY_CLOSING;
+    QTAILQ_FOREACH_SAFE(r1, &proxy->outgoing, next, r2) {
+        qemu_cond_destroy(&r1->cv);
+        QTAILQ_REMOVE(&proxy->outgoing, r1, next);
+        g_free(r1);
+    }
+    QTAILQ_FOREACH_SAFE(r1, &proxy->incoming, next, r2) {
+        qemu_cond_destroy(&r1->cv);
+        QTAILQ_REMOVE(&proxy->incoming, r1, next);
+        g_free(r1);
+    }
+    QTAILQ_FOREACH_SAFE(r1, &proxy->pending, next, r2) {
+        qemu_cond_destroy(&r1->cv);
+        QTAILQ_REMOVE(&proxy->pending, r1, next);
+        g_free(r1);
+    }
+    QTAILQ_FOREACH_SAFE(r1, &proxy->free, next, r2) {
+        qemu_cond_destroy(&r1->cv);
+        QTAILQ_REMOVE(&proxy->free, r1, next);
+        g_free(r1);
+    }
+
+    /*
+     * Make sure the iothread isn't blocking anywhere
+     * with a ref to this proxy by waiting for a BH
+     * handler to run after the proxy fd handlers were
+     * deleted above.
+     */
+    aio_bh_schedule_oneshot(proxy->ctx, vfio_user_cb, proxy);
+    qemu_cond_wait(&proxy->close_cv, &proxy->lock);
+
+    /* we now hold the only ref to proxy */
+    qemu_mutex_unlock(&proxy->lock);
+    qemu_cond_destroy(&proxy->close_cv);
+    qemu_mutex_destroy(&proxy->lock);
+
+    QLIST_REMOVE(proxy, next);
+    if (QLIST_EMPTY(&vfio_user_sockets)) {
+        iothread_destroy(vfio_user_iothread);
+        vfio_user_iothread = NULL;
+    }
+
+    g_free(proxy->sockname);
+    g_free(proxy);
+}
+
+void vfio_user_request_msg(VFIOUserHdr *hdr, uint16_t cmd,
+                           uint32_t size, uint32_t flags)
+{
+    static uint16_t next_id;
+
+    hdr->id = qatomic_fetch_inc(&next_id);
+    hdr->command = cmd;
+    hdr->size = size;
+    hdr->flags = (flags & ~VFIO_USER_TYPE) | VFIO_USER_REQUEST;
+    hdr->error_reply = 0;
+}
+
+struct cap_entry {
+    const char *name;
+    bool (*check)(VFIOUserProxy *proxy, QObject *qobj, Error **errp);
+};
+
+static bool caps_parse(VFIOUserProxy *proxy, QDict *qdict,
+                       struct cap_entry caps[], Error **errp)
+{
+    QObject *qobj;
+    struct cap_entry *p;
+
+    for (p = caps; p->name != NULL; p++) {
+        qobj = qdict_get(qdict, p->name);
+        if (qobj != NULL) {
+            if (!p->check(proxy, qobj, errp)) {
+                return false;
+            }
+            qdict_del(qdict, p->name);
+        }
+    }
+
+    /* warning, for now */
+    if (qdict_size(qdict) != 0) {
+        warn_report("spurious capabilities");
+    }
+    return true;
+}
+
+static bool check_migr_pgsize(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QNum *qn = qobject_to(QNum, qobj);
+    uint64_t pgsize;
+
+    if (qn == NULL || !qnum_get_try_uint(qn, &pgsize)) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_PGSIZE);
+        return false;
+    }
+
+    /* must be larger than default */
+    if (pgsize & (VFIO_USER_DEF_PGSIZE - 1)) {
+        error_setg(errp, "pgsize 0x%"PRIx64" too small", pgsize);
+        return false;
+    }
+
+    proxy->migr_pgsize = pgsize;
+    return true;
+}
+
+static bool check_bitmap(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QNum *qn = qobject_to(QNum, qobj);
+    uint64_t bitmap_size;
+
+    if (qn == NULL || !qnum_get_try_uint(qn, &bitmap_size)) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_MAX_BITMAP);
+        return false;
+    }
+
+    /* can only lower it */
+    if (bitmap_size > VFIO_USER_DEF_MAX_BITMAP) {
+        error_setg(errp, "%s too large", VFIO_USER_CAP_MAX_BITMAP);
+        return false;
+    }
+
+    proxy->max_bitmap = bitmap_size;
+    return true;
+}
+
+static struct cap_entry caps_migr[] = {
+    { VFIO_USER_CAP_PGSIZE, check_migr_pgsize },
+    { VFIO_USER_CAP_MAX_BITMAP, check_bitmap },
+    { NULL }
+};
+
+static bool check_max_fds(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QNum *qn = qobject_to(QNum, qobj);
+    uint64_t max_send_fds;
+
+    if (qn == NULL || !qnum_get_try_uint(qn, &max_send_fds) ||
+        max_send_fds > VFIO_USER_MAX_MAX_FDS) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_MAX_FDS);
+        return false;
+    }
+    proxy->max_send_fds = max_send_fds;
+    return true;
+}
+
+static bool check_max_xfer(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QNum *qn = qobject_to(QNum, qobj);
+    uint64_t max_xfer_size;
+
+    if (qn == NULL || !qnum_get_try_uint(qn, &max_xfer_size) ||
+        max_xfer_size > VFIO_USER_MAX_MAX_XFER) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_MAX_XFER);
+        return false;
+    }
+    proxy->max_xfer_size = max_xfer_size;
+    return true;
+}
+
+static bool check_pgsizes(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QNum *qn = qobject_to(QNum, qobj);
+    uint64_t pgsizes;
+
+    if (qn == NULL || !qnum_get_try_uint(qn, &pgsizes)) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_PGSIZES);
+        return false;
+    }
+
+    /* must be larger than default */
+    if (pgsizes & (VFIO_USER_DEF_PGSIZE - 1)) {
+        error_setg(errp, "pgsize 0x%"PRIx64" too small", pgsizes);
+        return false;
+    }
+
+    proxy->dma_pgsizes = pgsizes;
+    return true;
+}
+
+static bool check_max_dma(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QNum *qn = qobject_to(QNum, qobj);
+    uint64_t max_dma;
+
+    if (qn == NULL || !qnum_get_try_uint(qn, &max_dma)) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_MAP_MAX);
+        return false;
+    }
+
+    /* can only lower it */
+    if (max_dma > VFIO_USER_DEF_MAP_MAX) {
+        error_setg(errp, "%s too large", VFIO_USER_CAP_MAP_MAX);
+        return false;
+    }
+
+    proxy->max_dma = max_dma;
+    return true;
+}
+
+static bool check_migr(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QDict *qdict = qobject_to(QDict, qobj);
+
+    if (qdict == NULL) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_MAX_FDS);
+        return true;
+    }
+    return caps_parse(proxy, qdict, caps_migr, errp);
+}
+
+static bool check_multi(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+    QBool *qb = qobject_to(QBool, qobj);
+
+    if (qb == NULL) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP_MULTI);
+        return false;
+    }
+    if (qbool_get_bool(qb)) {
+        proxy->flags |= VFIO_PROXY_USE_MULTI;
+    }
+    return true;
+}
+
+static struct cap_entry caps_cap[] = {
+    { VFIO_USER_CAP_MAX_FDS, check_max_fds },
+    { VFIO_USER_CAP_MAX_XFER, check_max_xfer },
+    { VFIO_USER_CAP_PGSIZES, check_pgsizes },
+    { VFIO_USER_CAP_MAP_MAX, check_max_dma },
+    { VFIO_USER_CAP_MIGR, check_migr },
+    { VFIO_USER_CAP_MULTI, check_multi },
+    { NULL }
+};
+
+static bool check_cap(VFIOUserProxy *proxy, QObject *qobj, Error **errp)
+{
+   QDict *qdict = qobject_to(QDict, qobj);
+
+    if (qdict == NULL) {
+        error_setg(errp, "malformed %s", VFIO_USER_CAP);
+        return false;
+    }
+    return caps_parse(proxy, qdict, caps_cap, errp);
+}
+
+static struct cap_entry ver_0_0[] = {
+    { VFIO_USER_CAP, check_cap },
+    { NULL }
+};
+
+static bool caps_check(VFIOUserProxy *proxy, int minor, const char *caps,
+                       Error **errp)
+{
+    QObject *qobj;
+    QDict *qdict;
+    bool ret;
+
+    qobj = qobject_from_json(caps, NULL);
+    if (qobj == NULL) {
+        error_setg(errp, "malformed capabilities %s", caps);
+        return false;
+    }
+    qdict = qobject_to(QDict, qobj);
+    if (qdict == NULL) {
+        error_setg(errp, "capabilities %s not an object", caps);
+        qobject_unref(qobj);
+        return false;
+    }
+    ret = caps_parse(proxy, qdict, ver_0_0, errp);
+
+    qobject_unref(qobj);
+    return ret;
+}
+
+static GString *caps_json(void)
+{
+    QDict *dict = qdict_new();
+    QDict *capdict = qdict_new();
+    QDict *migdict = qdict_new();
+    GString *str;
+
+    qdict_put_int(migdict, VFIO_USER_CAP_PGSIZE, VFIO_USER_DEF_PGSIZE);
+    qdict_put_int(migdict, VFIO_USER_CAP_MAX_BITMAP, VFIO_USER_DEF_MAX_BITMAP);
+    qdict_put_obj(capdict, VFIO_USER_CAP_MIGR, QOBJECT(migdict));
+
+    qdict_put_int(capdict, VFIO_USER_CAP_MAX_FDS, VFIO_USER_MAX_MAX_FDS);
+    qdict_put_int(capdict, VFIO_USER_CAP_MAX_XFER, VFIO_USER_DEF_MAX_XFER);
+    qdict_put_int(capdict, VFIO_USER_CAP_PGSIZES, VFIO_USER_DEF_PGSIZE);
+    qdict_put_int(capdict, VFIO_USER_CAP_MAP_MAX, VFIO_USER_DEF_MAP_MAX);
+    qdict_put_bool(capdict, VFIO_USER_CAP_MULTI, true);
+
+    qdict_put_obj(dict, VFIO_USER_CAP, QOBJECT(capdict));
+
+    str = qobject_to_json(QOBJECT(dict));
+    qobject_unref(dict);
+    return str;
+}
+
+bool vfio_user_validate_version(VFIOUserProxy *proxy, Error **errp)
+{
+    g_autofree VFIOUserVersion *msgp = NULL;
+    GString *caps;
+    char *reply;
+    int size, caplen;
+
+    caps = caps_json();
+    caplen = caps->len + 1;
+    size = sizeof(*msgp) + caplen;
+    msgp = g_malloc0(size);
+
+    vfio_user_request_msg(&msgp->hdr, VFIO_USER_VERSION, size, 0);
+    msgp->major = VFIO_USER_MAJOR_VER;
+    msgp->minor = VFIO_USER_MINOR_VER;
+    memcpy(&msgp->capabilities, caps->str, caplen);
+    g_string_free(caps, true);
+    trace_vfio_user_version(msgp->major, msgp->minor, msgp->capabilities);
+
+    if (!vfio_user_send_wait(proxy, &msgp->hdr, NULL, 0, errp)) {
+        return false;
+    }
+
+    if (msgp->hdr.flags & VFIO_USER_ERROR) {
+        error_setg_errno(errp, msgp->hdr.error_reply, "version reply");
+        return false;
+    }
+
+    if (msgp->major != VFIO_USER_MAJOR_VER ||
+        msgp->minor > VFIO_USER_MINOR_VER) {
+        error_setg(errp, "incompatible server version");
+        return false;
+    }
+
+    reply = msgp->capabilities;
+    if (reply[msgp->hdr.size - sizeof(*msgp) - 1] != '\0') {
+        error_setg(errp, "corrupt version reply");
+        return false;
+    }
+
+    if (!caps_check(proxy, msgp->minor, reply, errp)) {
+        return false;
+    }
+
+    trace_vfio_user_version(msgp->major, msgp->minor, msgp->capabilities);
+    return true;
+}
+
+void vfio_user_flush_multi(VFIOUserProxy *proxy)
+{
+    VFIOUserMsg *msg;
+    VFIOUserWRMulti *wm = proxy->wr_multi;
+    Error *local_err = NULL;
+
+    proxy->wr_multi = NULL;
+
+    /* adjust size for actual # of writes */
+    wm->hdr.size -= (VFIO_USER_MULTI_MAX - wm->wr_cnt) * sizeof(VFIOUserWROne);
+
+    msg = vfio_user_getmsg(proxy, &wm->hdr, NULL);
+    msg->id = wm->hdr.id;
+    msg->rsize = 0;
+    msg->type = VFIO_MSG_ASYNC;
+    trace_vfio_user_wrmulti("flush", wm->wr_cnt);
+
+    if (!vfio_user_send_queued(proxy, msg, &local_err)) {
+        error_report_err(local_err);
+        vfio_user_recycle(proxy, msg);
+    }
+}
+
+void vfio_user_create_multi(VFIOUserProxy *proxy)
+{
+    VFIOUserWRMulti *wm;
+
+    wm = g_malloc0(sizeof(*wm));
+    vfio_user_request_msg(&wm->hdr, VFIO_USER_REGION_WRITE_MULTI,
+                          sizeof(*wm), VFIO_USER_NO_REPLY);
+    proxy->wr_multi = wm;
+}
+
+void vfio_user_add_multi(VFIOUserProxy *proxy, uint8_t index,
+                         off_t offset, uint32_t count, void *data)
+{
+    VFIOUserWRMulti *wm = proxy->wr_multi;
+    VFIOUserWROne *w1 = &wm->wrs[wm->wr_cnt];
+
+    w1->offset = offset;
+    w1->region = index;
+    w1->count = count;
+    memcpy(&w1->data, data, count);
+
+    wm->wr_cnt++;
+    trace_vfio_user_wrmulti("add", wm->wr_cnt);
+    if (wm->wr_cnt == VFIO_USER_MULTI_MAX ||
+        proxy->num_outgoing < VFIO_USER_OUT_LOW) {
+        vfio_user_flush_multi(proxy);
+    }
+}
diff --git a/hw/vfio-user/proxy.h b/hw/vfio-user/proxy.h
new file mode 100644
index 0000000..61e64a0
--- /dev/null
+++ b/hw/vfio-user/proxy.h
@@ -0,0 +1,135 @@
+#ifndef VFIO_USER_PROXY_H
+#define VFIO_USER_PROXY_H
+
+/*
+ * vfio protocol over a UNIX socket.
+ *
+ * Copyright © 2018, 2021 Oracle and/or its affiliates.
+ *
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+
+#include "io/channel.h"
+#include "io/channel-socket.h"
+
+#include "qemu/queue.h"
+#include "qemu/sockets.h"
+#include "qemu/thread.h"
+#include "hw/vfio/vfio-device.h"
+#include "hw/vfio-user/protocol.h"
+
+typedef struct {
+    int send_fds;
+    int recv_fds;
+    int *fds;
+} VFIOUserFDs;
+
+enum msg_type {
+    VFIO_MSG_NONE,
+    VFIO_MSG_ASYNC,
+    VFIO_MSG_WAIT,
+    VFIO_MSG_NOWAIT,
+    VFIO_MSG_REQ,
+};
+
+typedef struct VFIOUserMsg {
+    QTAILQ_ENTRY(VFIOUserMsg) next;
+    VFIOUserHdr *hdr;
+    VFIOUserFDs *fds;
+    uint32_t rsize;
+    uint32_t id;
+    QemuCond cv;
+    bool complete;
+    bool pending;
+    enum msg_type type;
+} VFIOUserMsg;
+
+
+enum proxy_state {
+    VFIO_PROXY_CONNECTED = 1,
+    VFIO_PROXY_ERROR = 2,
+    VFIO_PROXY_CLOSING = 3,
+    VFIO_PROXY_CLOSED = 4,
+};
+
+typedef QTAILQ_HEAD(VFIOUserMsgQ, VFIOUserMsg) VFIOUserMsgQ;
+
+typedef struct VFIOUserProxy {
+    QLIST_ENTRY(VFIOUserProxy) next;
+    char *sockname;
+    struct QIOChannel *ioc;
+    void (*request)(void *opaque, VFIOUserMsg *msg);
+    void *req_arg;
+    uint64_t max_xfer_size;
+    uint64_t max_send_fds;
+    uint64_t max_dma;
+    uint64_t dma_pgsizes;
+    uint64_t max_bitmap;
+    uint64_t migr_pgsize;
+    int flags;
+    uint32_t wait_time;
+    QemuCond close_cv;
+    AioContext *ctx;
+    QEMUBH *req_bh;
+    bool async_ops;
+
+    /*
+     * above only changed when BQL is held
+     * below are protected by per-proxy lock
+     */
+    QemuMutex lock;
+    VFIOUserMsgQ free;
+    VFIOUserMsgQ pending;
+    VFIOUserMsgQ incoming;
+    VFIOUserMsgQ outgoing;
+    VFIOUserMsg *last_nowait;
+    VFIOUserMsg *part_recv;
+    size_t recv_left;
+    VFIOUserWRMulti *wr_multi;
+    int num_outgoing;
+    enum proxy_state state;
+} VFIOUserProxy;
+
+/* VFIOProxy flags */
+#define VFIO_PROXY_CLIENT        0x1
+#define VFIO_PROXY_FORCE_QUEUED  0x4
+#define VFIO_PROXY_NO_POST       0x8
+#define VFIO_PROXY_USE_MULTI     0x16
+
+/* coalescing high and low water marks for VFIOProxy num_outgoing */
+#define VFIO_USER_OUT_HIGH       1024
+#define VFIO_USER_OUT_LOW        128
+
+typedef struct VFIODevice VFIODevice;
+
+VFIOUserProxy *vfio_user_connect_dev(SocketAddress *addr, Error **errp);
+void vfio_user_disconnect(VFIOUserProxy *proxy);
+void vfio_user_set_handler(VFIODevice *vbasedev,
+                           void (*handler)(void *opaque, VFIOUserMsg *msg),
+                           void *reqarg);
+bool vfio_user_validate_version(VFIOUserProxy *proxy, Error **errp);
+
+VFIOUserFDs *vfio_user_getfds(int numfds);
+void vfio_user_putfds(VFIOUserMsg *msg);
+
+void vfio_user_disable_posted_writes(VFIOUserProxy *proxy);
+
+void vfio_user_request_msg(VFIOUserHdr *hdr, uint16_t cmd,
+                           uint32_t size, uint32_t flags);
+void vfio_user_wait_reqs(VFIOUserProxy *proxy);
+bool vfio_user_send_wait(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                         VFIOUserFDs *fds, int rsize, Error **errp);
+bool vfio_user_send_nowait(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                           VFIOUserFDs *fds, int rsize, Error **errp);
+bool vfio_user_send_async(VFIOUserProxy *proxy, VFIOUserHdr *hdr,
+                          VFIOUserFDs *fds, Error **errp);
+
+void vfio_user_send_reply(VFIOUserProxy *proxy, VFIOUserHdr *hdr, int size);
+void vfio_user_send_error(VFIOUserProxy *proxy, VFIOUserHdr *hdr, int error);
+
+void vfio_user_flush_multi(VFIOUserProxy *proxy);
+void vfio_user_create_multi(VFIOUserProxy *proxy);
+void vfio_user_add_multi(VFIOUserProxy *proxy, uint8_t index,
+                         off_t offset, uint32_t count, void *data);
+
+#endif /* VFIO_USER_PROXY_H */
diff --git a/hw/vfio-user/trace-events b/hw/vfio-user/trace-events
new file mode 100644
index 0000000..abb67f4
--- /dev/null
+++ b/hw/vfio-user/trace-events
@@ -0,0 +1,20 @@
+# See docs/devel/tracing.rst for syntax documentation.
+#
+# SPDX-License-Identifier: GPL-2.0-or-later
+
+# common.c
+vfio_user_recv_hdr(const char *name, uint16_t id, uint16_t cmd, uint32_t size, uint32_t flags) " (%s) id 0x%x cmd 0x%x size 0x%x flags 0x%x"
+vfio_user_recv_read(uint16_t id, int read) " id 0x%x read 0x%x"
+vfio_user_recv_request(uint16_t cmd) " command 0x%x"
+vfio_user_send_write(uint16_t id, int wrote) " id 0x%x wrote 0x%x"
+vfio_user_version(uint16_t major, uint16_t minor, const char *caps) " major %d minor %d caps: %s"
+vfio_user_get_info(uint32_t nregions, uint32_t nirqs) " #regions %d #irqs %d"
+vfio_user_get_region_info(uint32_t index, uint32_t flags, uint64_t size) " index %d flags 0x%x size 0x%"PRIx64
+vfio_user_region_rw(uint32_t region, uint64_t off, uint32_t count) " region %d offset 0x%"PRIx64" count %d"
+vfio_user_get_irq_info(uint32_t index, uint32_t flags, uint32_t count) " index %d flags 0x%x count %d"
+vfio_user_set_irqs(uint32_t index, uint32_t start, uint32_t count, uint32_t flags) " index %d start %d count %d flags 0x%x"
+vfio_user_wrmulti(const char *s, uint64_t wr_cnt) " %s count 0x%"PRIx64
+
+# container.c
+vfio_user_dma_map(uint64_t iova, uint64_t size, uint64_t off, uint32_t flags, bool async_ops) " iova 0x%"PRIx64" size 0x%"PRIx64" off 0x%"PRIx64" flags 0x%x async_ops %d"
+vfio_user_dma_unmap(uint64_t iova, uint64_t size, uint32_t flags,  bool async_ops) " iova 0x%"PRIx64" size 0x%"PRIx64" flags 0x%x async_ops %d"
diff --git a/hw/vfio-user/trace.h b/hw/vfio-user/trace.h
new file mode 100644
index 0000000..9cf02d9
--- /dev/null
+++ b/hw/vfio-user/trace.h
@@ -0,0 +1,4 @@
+/*
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
+#include "trace/trace-hw_vfio_user.h"
diff --git a/hw/vfio/Kconfig b/hw/vfio/Kconfig
index 7cdba05..91d9023 100644
--- a/hw/vfio/Kconfig
+++ b/hw/vfio/Kconfig
@@ -1,3 +1,5 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
 config VFIO
     bool
     depends on LINUX
diff --git a/hw/vfio/ap.c b/hw/vfio/ap.c
index 874e0d1..1df4438 100644
--- a/hw/vfio/ap.c
+++ b/hw/vfio/ap.c
@@ -52,6 +52,11 @@ static QTAILQ_HEAD(, APConfigChgEvent) cfg_chg_events =
 
 static QemuMutex cfg_chg_events_lock;
 
+static void __attribute__((constructor)) vfio_ap_global_init(void)
+{
+    qemu_mutex_init(&cfg_chg_events_lock);
+}
+
 OBJECT_DECLARE_SIMPLE_TYPE(VFIOAPDevice, VFIO_AP_DEVICE)
 
 static void vfio_ap_compute_needs_reset(VFIODevice *vdev)
@@ -230,13 +235,6 @@ static void vfio_ap_realize(DeviceState *dev, Error **errp)
     VFIOAPDevice *vapdev = VFIO_AP_DEVICE(dev);
     VFIODevice *vbasedev = &vapdev->vdev;
 
-    static bool lock_initialized;
-
-    if (!lock_initialized) {
-        qemu_mutex_init(&cfg_chg_events_lock);
-        lock_initialized = true;
-    }
-
     if (!vfio_device_get_name(vbasedev, errp)) {
         return;
     }
diff --git a/hw/vfio/container.c b/hw/vfio/container.c
index 3e8d645..3e13fea 100644
--- a/hw/vfio/container.c
+++ b/hw/vfio/container.c
@@ -710,7 +710,9 @@ static bool vfio_container_connect(VFIOGroup *group, AddressSpace *as,
     return true;
 
 fail:
-    vfio_listener_unregister(bcontainer);
+    if (new_container) {
+        vfio_listener_unregister(bcontainer);
+    }
 
     if (group_was_added) {
         vfio_container_group_del(container, group);
@@ -990,12 +992,16 @@ static bool vfio_legacy_attach_device(const char *name, VFIODevice *vbasedev,
     if (vbasedev->mdev) {
         error_setg(&vbasedev->cpr.mdev_blocker,
                    "CPR does not support vfio mdev %s", vbasedev->name);
-        migrate_add_blocker_modes(&vbasedev->cpr.mdev_blocker, &error_fatal,
-                                  MIG_MODE_CPR_TRANSFER, -1);
+        if (migrate_add_blocker_modes(&vbasedev->cpr.mdev_blocker, errp,
+                                      MIG_MODE_CPR_TRANSFER, -1) < 0) {
+            goto hiod_unref_exit;
+        }
     }
 
     return true;
 
+hiod_unref_exit:
+    object_unref(vbasedev->hiod);
 device_put_exit:
     vfio_device_put(vbasedev);
 group_put_exit:
diff --git a/hw/vfio/device.c b/hw/vfio/device.c
index d32600e..d91c695 100644
--- a/hw/vfio/device.c
+++ b/hw/vfio/device.c
@@ -243,6 +243,13 @@ retry:
     return 0;
 }
 
+int vfio_device_get_region_fd(VFIODevice *vbasedev, int index)
+{
+        return vbasedev->region_fds ?
+               vbasedev->region_fds[index] :
+               vbasedev->fd;
+}
+
 int vfio_device_get_region_info_type(VFIODevice *vbasedev, uint32_t type,
                                      uint32_t subtype, struct vfio_region_info **info)
 {
diff --git a/hw/vfio/meson.build b/hw/vfio/meson.build
index 73d29f9..63ea393 100644
--- a/hw/vfio/meson.build
+++ b/hw/vfio/meson.build
@@ -1,3 +1,5 @@
+# SPDX-License-Identifier: GPL-2.0-or-later
+
 vfio_ss = ss.source_set()
 vfio_ss.add(files(
   'listener.c',
diff --git a/hw/vfio/pci.h b/hw/vfio/pci.h
index d3dc227..5ba7330 100644
--- a/hw/vfio/pci.h
+++ b/hw/vfio/pci.h
@@ -116,6 +116,7 @@ typedef struct VFIOMSIXInfo {
     uint32_t pba_offset;
     unsigned long *pending;
     bool noresize;
+    MemoryRegion *pba_region;
 } VFIOMSIXInfo;
 
 /*
diff --git a/hw/vfio/region.c b/hw/vfio/region.c
index f5b8e3c..d04c57d 100644
--- a/hw/vfio/region.c
+++ b/hw/vfio/region.c
@@ -273,10 +273,7 @@ int vfio_region_mmap(VFIORegion *region)
             goto no_mmap;
         }
 
-        /* Use the per-region fd if set, or the shared fd. */
-        fd = region->vbasedev->region_fds ?
-             region->vbasedev->region_fds[region->nr] :
-             region->vbasedev->fd,
+        fd = vfio_device_get_region_fd(region->vbasedev, region->nr);
 
         map_align = (void *)ROUND_UP((uintptr_t)map_base, (uintptr_t)align);
         munmap(map_base, map_align - map_base);
diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events
index f06236f..e1728c4 100644
--- a/hw/vfio/trace-events
+++ b/hw/vfio/trace-events
@@ -1,4 +1,6 @@
 # See docs/devel/tracing.rst for syntax documentation.
+#
+# SPDX-License-Identifier: GPL-2.0-or-later
 
 # pci.c
 vfio_intx_interrupt(const char *name, char line) " (%s) Pin %c"
diff --git a/hw/vfio/trace.h b/hw/vfio/trace.h
index 5a343aa..b34b61d 100644
--- a/hw/vfio/trace.h
+++ b/hw/vfio/trace.h
@@ -1 +1,4 @@
+/*
+ * SPDX-License-Identifier: GPL-2.0-or-later
+ */
 #include "trace/trace-hw_vfio.h"
diff --git a/include/hw/vfio/vfio-container-base.h b/include/hw/vfio/vfio-container-base.h
index f023265..3cd86ec 100644
--- a/include/hw/vfio/vfio-container-base.h
+++ b/include/hw/vfio/vfio-container-base.h
@@ -109,6 +109,7 @@ vfio_container_get_page_size_mask(const VFIOContainerBase *bcontainer)
 #define TYPE_VFIO_IOMMU_LEGACY TYPE_VFIO_IOMMU "-legacy"
 #define TYPE_VFIO_IOMMU_SPAPR TYPE_VFIO_IOMMU "-spapr"
 #define TYPE_VFIO_IOMMU_IOMMUFD TYPE_VFIO_IOMMU "-iommufd"
+#define TYPE_VFIO_IOMMU_USER TYPE_VFIO_IOMMU "-user"
 
 OBJECT_DECLARE_TYPE(VFIOContainerBase, VFIOIOMMUClass, VFIO_IOMMU)
 
diff --git a/include/hw/vfio/vfio-device.h b/include/hw/vfio/vfio-device.h
index d45e5a6..c616652 100644
--- a/include/hw/vfio/vfio-device.h
+++ b/include/hw/vfio/vfio-device.h
@@ -47,6 +47,7 @@ typedef struct VFIOMigration VFIOMigration;
 
 typedef struct IOMMUFDBackend IOMMUFDBackend;
 typedef struct VFIOIOASHwpt VFIOIOASHwpt;
+typedef struct VFIOUserProxy VFIOUserProxy;
 
 typedef struct VFIODevice {
     QLIST_ENTRY(VFIODevice) next;
@@ -88,6 +89,7 @@ typedef struct VFIODevice {
     struct vfio_region_info **reginfo;
     int *region_fds;
     VFIODeviceCPR cpr;
+    VFIOUserProxy *proxy;
 } VFIODevice;
 
 struct VFIODeviceOps {
@@ -240,6 +242,7 @@ struct VFIODeviceIOOps {
      * @off: offset within the region
      * @size: size in bytes to write
      * @data: buffer to write from
+     * @post: true if this is a posted write
      *
      * Returns number of bytes write on success or -errno.
      */
@@ -256,6 +259,18 @@ int vfio_device_get_region_info(VFIODevice *vbasedev, int index,
                                 struct vfio_region_info **info);
 int vfio_device_get_region_info_type(VFIODevice *vbasedev, uint32_t type,
                                      uint32_t subtype, struct vfio_region_info **info);
+
+/**
+ * Return the fd for mapping this region. This is either the device's fd (for
+ * e.g. kernel vfio), or a per-region fd (for vfio-user).
+ *
+ * @vbasedev: #VFIODevice to use
+ * @index: region index
+ *
+ * Returns the fd.
+ */
+int vfio_device_get_region_fd(VFIODevice *vbasedev, int index);
+
 bool vfio_device_has_region_cap(VFIODevice *vbasedev, int region, uint16_t cap_type);
 
 int vfio_device_get_irq_info(VFIODevice *vbasedev, int index,
diff --git a/meson.build b/meson.build
index 4676908..dbc97bf 100644
--- a/meson.build
+++ b/meson.build
@@ -3683,6 +3683,7 @@ if have_system
     'hw/ufs',
     'hw/usb',
     'hw/vfio',
+    'hw/vfio-user',
     'hw/virtio',
     'hw/vmapple',
     'hw/watchdog',