4 Copyright (c) 2014 Virtual Open Systems Sarl.
6 This work is licensed under the terms of the GNU GPL, version 2 or later.
7 See the COPYING file in the top-level directory.
10 This protocol is aiming to complement the ioctl interface used to control the
11 vhost implementation in the Linux kernel. It implements the control plane needed
12 to establish virtqueue sharing with a user space process on the same host. It
13 uses communication over a Unix domain socket to share file descriptors in the
14 ancillary data of the message.
16 The protocol defines 2 sides of the communication, master and slave. Master is
17 the application that shares its virtqueues, in our case QEMU. Slave is the
18 consumer of the virtqueues.
20 In the current implementation QEMU is the Master, and the Slave is intended to
21 be a software Ethernet switch running in user space, such as Snabbswitch.
23 Master and slave can be either a client (i.e. connecting) or server (listening)
24 in the socket communication.
29 Note that all numbers are in the machine native byte order. A vhost-user message
30 consists of 3 header fields and a payload:
32 ------------------------------------
33 | request | flags | size | payload |
34 ------------------------------------
36 * Request: 32-bit type of the request
37 * Flags: 32-bit bit field:
38 - Lower 2 bits are the version (currently 0x01)
39 - Bit 2 is the reply flag - needs to be sent on each reply from the slave
40 - Bit 3 is the need_reply flag - see VHOST_USER_PROTOCOL_F_REPLY_ACK for
42 * Size - 32-bit size of the payload
45 Depending on the request type, payload can be:
47 * A single 64-bit integer
52 u64: a 64-bit unsigned integer
54 * A vring state description
62 * A vring address description
63 --------------------------------------------------------------
64 | index | flags | size | descriptor | used | available | log |
65 --------------------------------------------------------------
67 Index: a 32-bit vring index
68 Flags: a 32-bit vring flags
69 Descriptor: a 64-bit user address of the vring descriptor table
70 Used: a 64-bit user address of the vring used ring
71 Available: a 64-bit user address of the vring available ring
72 Log: a 64-bit guest address for logging
74 * Memory regions description
75 ---------------------------------------------------
76 | num regions | padding | region0 | ... | region7 |
77 ---------------------------------------------------
79 Num regions: a 32-bit number of regions
83 -----------------------------------------------------
84 | guest address | size | user address | mmap offset |
85 -----------------------------------------------------
87 Guest address: a 64-bit guest address of the region
89 User address: a 64-bit user address
90 mmap offset: 64-bit offset where region starts in the mapped memory
93 ---------------------------
94 | log size | log offset |
95 ---------------------------
96 log size: size of area used for logging
97 log offset: offset from start of supplied file descriptor
98 where logging starts (i.e. where guest address 0 would be logged)
101 ---------------------------------------------------------
102 | iova | size | user address | permissions flags | type |
103 ---------------------------------------------------------
105 IOVA: a 64-bit I/O virtual address programmed by the guest
107 User address: a 64-bit user address
108 Permissions: a 8-bit value:
112 - 3: Read/Write access
113 Type: a 8-bit IOTLB message type:
116 - 3: IOTLB invalidate
117 - 4: IOTLB access fail
119 In QEMU the vhost-user message is implemented with the following struct:
121 typedef struct VhostUserMsg {
122 VhostUserRequest request;
127 struct vhost_vring_state state;
128 struct vhost_vring_addr addr;
129 VhostUserMemory memory;
131 struct vhost_iotlb_msg iotlb;
133 } QEMU_PACKED VhostUserMsg;
138 The protocol for vhost-user is based on the existing implementation of vhost
139 for the Linux Kernel. Most messages that can be sent via the Unix domain socket
140 implementing vhost-user have an equivalent ioctl to the kernel implementation.
142 The communication consists of master sending message requests and slave sending
143 message replies. Most of the requests don't require replies. Here is a list of
146 * VHOST_USER_GET_FEATURES
147 * VHOST_USER_GET_PROTOCOL_FEATURES
148 * VHOST_USER_GET_VRING_BASE
149 * VHOST_USER_SET_LOG_BASE (if VHOST_USER_PROTOCOL_F_LOG_SHMFD)
151 [ Also see the section on REPLY_ACK protocol extension. ]
153 There are several messages that the master sends with file descriptors passed
154 in the ancillary data:
156 * VHOST_USER_SET_MEM_TABLE
157 * VHOST_USER_SET_LOG_BASE (if VHOST_USER_PROTOCOL_F_LOG_SHMFD)
158 * VHOST_USER_SET_LOG_FD
159 * VHOST_USER_SET_VRING_KICK
160 * VHOST_USER_SET_VRING_CALL
161 * VHOST_USER_SET_VRING_ERR
162 * VHOST_USER_SET_SLAVE_REQ_FD
164 If Master is unable to send the full message or receives a wrong reply it will
165 close the connection. An optional reconnection mechanism can be implemented.
167 Any protocol extensions are gated by protocol feature bits,
168 which allows full backwards compatibility on both master
170 As older slaves don't support negotiating protocol features,
171 a feature bit was dedicated for this purpose:
172 #define VHOST_USER_F_PROTOCOL_FEATURES 30
174 Starting and stopping rings
175 ----------------------
176 Client must only process each ring when it is started.
178 Client must only pass data between the ring and the
179 backend, when the ring is enabled.
181 If ring is started but disabled, client must process the
182 ring without talking to the backend.
184 For example, for a networking device, in the disabled state
185 client must not supply any new RX packets, but must process
186 and discard any TX packets.
188 If VHOST_USER_F_PROTOCOL_FEATURES has not been negotiated, the ring is initialized
191 If VHOST_USER_F_PROTOCOL_FEATURES has been negotiated, the ring is initialized
192 in a disabled state. Client must not pass data to/from the backend until ring is enabled by
193 VHOST_USER_SET_VRING_ENABLE with parameter 1, or after it has been disabled by
194 VHOST_USER_SET_VRING_ENABLE with parameter 0.
196 Each ring is initialized in a stopped state, client must not process it until
197 ring is started, or after it has been stopped.
199 Client must start ring upon receiving a kick (that is, detecting that file
200 descriptor is readable) on the descriptor specified by
201 VHOST_USER_SET_VRING_KICK, and stop ring upon receiving
202 VHOST_USER_GET_VRING_BASE.
204 While processing the rings (whether they are enabled or not), client must
205 support changing some configuration aspects on the fly.
207 Multiple queue support
208 ----------------------
210 Multiple queue is treated as a protocol extension, hence the slave has to
211 implement protocol features first. The multiple queues feature is supported
212 only when the protocol feature VHOST_USER_PROTOCOL_F_MQ (bit 0) is set.
214 The max number of queues the slave supports can be queried with message
215 VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the number of
216 requested queues is bigger than that.
218 As all queues share one connection, the master uses a unique index for each
219 queue in the sent message to identify a specified queue. One queue pair
220 is enabled initially. More queues are enabled dynamically, by sending
221 message VHOST_USER_SET_VRING_ENABLE.
226 During live migration, the master may need to track the modifications
227 the slave makes to the memory mapped regions. The client should mark
228 the dirty pages in a log. Once it complies to this logging, it may
229 declare the VHOST_F_LOG_ALL vhost feature.
231 To start/stop logging of data/used ring writes, server may send messages
232 VHOST_USER_SET_FEATURES with VHOST_F_LOG_ALL and VHOST_USER_SET_VRING_ADDR with
233 VHOST_VRING_F_LOG in ring's flags set to 1/0, respectively.
235 All the modifications to memory pointed by vring "descriptor" should
236 be marked. Modifications to "used" vring should be marked if
237 VHOST_VRING_F_LOG is part of ring's flags.
239 Dirty pages are of size:
240 #define VHOST_LOG_PAGE 0x1000
242 The log memory fd is provided in the ancillary data of
243 VHOST_USER_SET_LOG_BASE message when the slave has
244 VHOST_USER_PROTOCOL_F_LOG_SHMFD protocol feature.
246 The size of the log is supplied as part of VhostUserMsg
247 which should be large enough to cover all known guest
248 addresses. Log starts at the supplied offset in the
249 supplied file descriptor.
250 The log covers from address 0 to the maximum of guest
251 regions. In pseudo-code, to mark page at "addr" as dirty:
253 page = addr / VHOST_LOG_PAGE
254 log[page / 8] |= 1 << page % 8
256 Where addr is the guest physical address.
258 Use atomic operations, as the log may be concurrently manipulated.
260 Note that when logging modifications to the used ring (when VHOST_VRING_F_LOG
261 is set for this ring), log_guest_addr should be used to calculate the log
262 offset: the write to first byte of the used ring is logged at this offset from
263 log start. Also note that this value might be outside the legal guest physical
264 address range (i.e. does not have to be covered by the VhostUserMemory table),
265 but the bit offset of the last byte of the ring must fall within
266 the size supplied by VhostUserLog.
268 VHOST_USER_SET_LOG_FD is an optional message with an eventfd in
269 ancillary data, it may be used to inform the master that the log has
272 Once the source has finished migration, rings will be stopped by
273 the source. No further update must be done before rings are
279 When the VIRTIO_F_IOMMU_PLATFORM feature has been negotiated, the master
280 sends IOTLB entries update & invalidation by sending VHOST_USER_IOTLB_MSG
281 requests to the slave with a struct vhost_iotlb_msg as payload. For update
282 events, the iotlb payload has to be filled with the update message type (2),
283 the I/O virtual address, the size, the user virtual address, and the
284 permissions flags. Addresses and size must be within vhost memory regions set
285 via the VHOST_USER_SET_MEM_TABLE request. For invalidation events, the iotlb
286 payload has to be filled with the invalidation message type (3), the I/O virtual
287 address and the size. On success, the slave is expected to reply with a zero
288 payload, non-zero otherwise.
290 The slave relies on the slave communcation channel (see "Slave communication"
291 section below) to send IOTLB miss and access failure events, by sending
292 VHOST_USER_SLAVE_IOTLB_MSG requests to the master with a struct vhost_iotlb_msg
293 as payload. For miss events, the iotlb payload has to be filled with the miss
294 message type (1), the I/O virtual address and the permissions flags. For access
295 failure event, the iotlb payload has to be filled with the access failure
296 message type (4), the I/O virtual address and the permissions flags.
297 For synchronization purpose, the slave may rely on the reply-ack feature,
298 so the master may send a reply when operation is completed if the reply-ack
299 feature is negotiated and slaves requests a reply. For miss events, completed
300 operation means either master sent an update message containing the IOTLB entry
301 containing requested address and permission, or master sent nothing if the IOTLB
302 miss message is invalid (invalid IOVA or permission).
304 The master isn't expected to take the initiative to send IOTLB update messages,
305 as the slave sends IOTLB miss messages for the guest virtual memory areas it
311 An optional communication channel is provided if the slave declares
312 VHOST_USER_PROTOCOL_F_SLAVE_REQ protocol feature, to allow the slave to make
313 requests to the master.
315 The fd is provided via VHOST_USER_SET_SLAVE_REQ_FD ancillary data.
317 A slave may then send VHOST_USER_SLAVE_* messages to the master
318 using this fd communication channel.
323 #define VHOST_USER_PROTOCOL_F_MQ 0
324 #define VHOST_USER_PROTOCOL_F_LOG_SHMFD 1
325 #define VHOST_USER_PROTOCOL_F_RARP 2
326 #define VHOST_USER_PROTOCOL_F_REPLY_ACK 3
327 #define VHOST_USER_PROTOCOL_F_MTU 4
328 #define VHOST_USER_PROTOCOL_F_SLAVE_REQ 5
329 #define VHOST_USER_PROTOCOL_F_CROSS_ENDIAN 6
334 * VHOST_USER_GET_FEATURES
337 Equivalent ioctl: VHOST_GET_FEATURES
341 Get from the underlying vhost implementation the features bitmask.
342 Feature bit VHOST_USER_F_PROTOCOL_FEATURES signals slave support for
343 VHOST_USER_GET_PROTOCOL_FEATURES and VHOST_USER_SET_PROTOCOL_FEATURES.
345 * VHOST_USER_SET_FEATURES
348 Ioctl: VHOST_SET_FEATURES
351 Enable features in the underlying vhost implementation using a bitmask.
352 Feature bit VHOST_USER_F_PROTOCOL_FEATURES signals slave support for
353 VHOST_USER_GET_PROTOCOL_FEATURES and VHOST_USER_SET_PROTOCOL_FEATURES.
355 * VHOST_USER_GET_PROTOCOL_FEATURES
358 Equivalent ioctl: VHOST_GET_FEATURES
362 Get the protocol feature bitmask from the underlying vhost implementation.
363 Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
364 VHOST_USER_GET_FEATURES.
365 Note: slave that reported VHOST_USER_F_PROTOCOL_FEATURES must support
366 this message even before VHOST_USER_SET_FEATURES was called.
368 * VHOST_USER_SET_PROTOCOL_FEATURES
371 Ioctl: VHOST_SET_FEATURES
374 Enable protocol features in the underlying vhost implementation.
375 Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
376 VHOST_USER_GET_FEATURES.
377 Note: slave that reported VHOST_USER_F_PROTOCOL_FEATURES must support
378 this message even before VHOST_USER_SET_FEATURES was called.
380 * VHOST_USER_SET_OWNER
383 Equivalent ioctl: VHOST_SET_OWNER
386 Issued when a new connection is established. It sets the current Master
387 as an owner of the session. This can be used on the Slave as a
388 "session start" flag.
390 * VHOST_USER_RESET_OWNER
395 This is no longer used. Used to be sent to request disabling
396 all rings, but some clients interpreted it to also discard
397 connection state (this interpretation would lead to bugs).
398 It is recommended that clients either ignore this message,
399 or use it to disable all rings.
401 * VHOST_USER_SET_MEM_TABLE
404 Equivalent ioctl: VHOST_SET_MEM_TABLE
405 Master payload: memory regions description
407 Sets the memory map regions on the slave so it can translate the vring
408 addresses. In the ancillary data there is an array of file descriptors
409 for each memory mapped region. The size and ordering of the fds matches
410 the number and ordering of memory regions.
412 * VHOST_USER_SET_LOG_BASE
415 Equivalent ioctl: VHOST_SET_LOG_BASE
419 Sets logging shared memory space.
420 When slave has VHOST_USER_PROTOCOL_F_LOG_SHMFD protocol
421 feature, the log memory fd is provided in the ancillary data of
422 VHOST_USER_SET_LOG_BASE message, the size and offset of shared
423 memory area provided in the message.
426 * VHOST_USER_SET_LOG_FD
429 Equivalent ioctl: VHOST_SET_LOG_FD
432 Sets the logging file descriptor, which is passed as ancillary data.
434 * VHOST_USER_SET_VRING_NUM
437 Equivalent ioctl: VHOST_SET_VRING_NUM
438 Master payload: vring state description
440 Set the size of the queue.
442 * VHOST_USER_SET_VRING_ADDR
445 Equivalent ioctl: VHOST_SET_VRING_ADDR
446 Master payload: vring address description
449 Sets the addresses of the different aspects of the vring.
451 * VHOST_USER_SET_VRING_BASE
454 Equivalent ioctl: VHOST_SET_VRING_BASE
455 Master payload: vring state description
457 Sets the base offset in the available vring.
459 * VHOST_USER_GET_VRING_BASE
462 Equivalent ioctl: VHOST_USER_GET_VRING_BASE
463 Master payload: vring state description
464 Slave payload: vring state description
466 Get the available vring base offset.
468 * VHOST_USER_SET_VRING_KICK
471 Equivalent ioctl: VHOST_SET_VRING_KICK
474 Set the event file descriptor for adding buffers to the vring. It
475 is passed in the ancillary data.
476 Bits (0-7) of the payload contain the vring index. Bit 8 is the
477 invalid FD flag. This flag is set when there is no file descriptor
478 in the ancillary data. This signals that polling should be used
479 instead of waiting for a kick.
481 * VHOST_USER_SET_VRING_CALL
484 Equivalent ioctl: VHOST_SET_VRING_CALL
487 Set the event file descriptor to signal when buffers are used. It
488 is passed in the ancillary data.
489 Bits (0-7) of the payload contain the vring index. Bit 8 is the
490 invalid FD flag. This flag is set when there is no file descriptor
491 in the ancillary data. This signals that polling will be used
492 instead of waiting for the call.
494 * VHOST_USER_SET_VRING_ERR
497 Equivalent ioctl: VHOST_SET_VRING_ERR
500 Set the event file descriptor to signal when error occurs. It
501 is passed in the ancillary data.
502 Bits (0-7) of the payload contain the vring index. Bit 8 is the
503 invalid FD flag. This flag is set when there is no file descriptor
504 in the ancillary data.
506 * VHOST_USER_GET_QUEUE_NUM
509 Equivalent ioctl: N/A
513 Query how many queues the backend supports. This request should be
514 sent only when VHOST_USER_PROTOCOL_F_MQ is set in queried protocol
515 features by VHOST_USER_GET_PROTOCOL_FEATURES.
517 * VHOST_USER_SET_VRING_ENABLE
520 Equivalent ioctl: N/A
521 Master payload: vring state description
523 Signal slave to enable or disable corresponding vring.
524 This request should be sent only when VHOST_USER_F_PROTOCOL_FEATURES
527 * VHOST_USER_SEND_RARP
530 Equivalent ioctl: N/A
533 Ask vhost user backend to broadcast a fake RARP to notify the migration
534 is terminated for guest that does not support GUEST_ANNOUNCE.
535 Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
536 VHOST_USER_GET_FEATURES and protocol feature bit VHOST_USER_PROTOCOL_F_RARP
537 is present in VHOST_USER_GET_PROTOCOL_FEATURES.
538 The first 6 bytes of the payload contain the mac address of the guest to
539 allow the vhost user backend to construct and broadcast the fake RARP.
541 * VHOST_USER_NET_SET_MTU
544 Equivalent ioctl: N/A
547 Set host MTU value exposed to the guest.
548 This request should be sent only when VIRTIO_NET_F_MTU feature has been
549 successfully negotiated, VHOST_USER_F_PROTOCOL_FEATURES is present in
550 VHOST_USER_GET_FEATURES and protocol feature bit
551 VHOST_USER_PROTOCOL_F_NET_MTU is present in
552 VHOST_USER_GET_PROTOCOL_FEATURES.
553 If VHOST_USER_PROTOCOL_F_REPLY_ACK is negotiated, slave must respond
554 with zero in case the specified MTU is valid, or non-zero otherwise.
556 * VHOST_USER_SET_SLAVE_REQ_FD
559 Equivalent ioctl: N/A
562 Set the socket file descriptor for slave initiated requests. It is passed
563 in the ancillary data.
564 This request should be sent only when VHOST_USER_F_PROTOCOL_FEATURES
565 has been negotiated, and protocol feature bit VHOST_USER_PROTOCOL_F_SLAVE_REQ
566 bit is present in VHOST_USER_GET_PROTOCOL_FEATURES.
567 If VHOST_USER_PROTOCOL_F_REPLY_ACK is negotiated, slave must respond
568 with zero for success, non-zero otherwise.
570 * VHOST_USER_IOTLB_MSG
573 Equivalent ioctl: N/A (equivalent to VHOST_IOTLB_MSG message type)
574 Master payload: struct vhost_iotlb_msg
577 Send IOTLB messages with struct vhost_iotlb_msg as payload.
578 Master sends such requests to update and invalidate entries in the device
579 IOTLB. The slave has to acknowledge the request with sending zero as u64
580 payload for success, non-zero otherwise.
581 This request should be send only when VIRTIO_F_IOMMU_PLATFORM feature
582 has been successfully negotiated.
584 * VHOST_USER_SET_VRING_ENDIAN
587 Equivalent ioctl: VHOST_SET_VRING_ENDIAN
588 Master payload: vring state description
590 Set the endianess of a VQ for legacy devices. Little-endian is indicated
591 with state.num set to 0 and big-endian is indicated with state.num set
592 to 1. Other values are invalid.
593 This request should be sent only when VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
595 Backends that negotiated this feature should handle both endianesses
596 and expect this message once (per VQ) during device configuration
597 (ie. before the master starts the VQ).
602 * VHOST_USER_SLAVE_IOTLB_MSG
605 Equivalent ioctl: N/A (equivalent to VHOST_IOTLB_MSG message type)
606 Slave payload: struct vhost_iotlb_msg
609 Send IOTLB messages with struct vhost_iotlb_msg as payload.
610 Slave sends such requests to notify of an IOTLB miss, or an IOTLB
611 access failure. If VHOST_USER_PROTOCOL_F_REPLY_ACK is negotiated,
612 and slave set the VHOST_USER_NEED_REPLY flag, master must respond with
613 zero when operation is successfully completed, or non-zero otherwise.
614 This request should be send only when VIRTIO_F_IOMMU_PLATFORM feature
615 has been successfully negotiated.
617 VHOST_USER_PROTOCOL_F_REPLY_ACK:
618 -------------------------------
619 The original vhost-user specification only demands replies for certain
620 commands. This differs from the vhost protocol implementation where commands
621 are sent over an ioctl() call and block until the client has completed.
623 With this protocol extension negotiated, the sender (QEMU) can set the
624 "need_reply" [Bit 3] flag to any command. This indicates that
625 the client MUST respond with a Payload VhostUserMsg indicating success or
626 failure. The payload should be set to zero on success or non-zero on failure,
627 unless the message already has an explicit reply body.
629 The response payload gives QEMU a deterministic indication of the result
630 of the command. Today, QEMU is expected to terminate the main vhost-user
631 loop upon receiving such errors. In future, qemu could be taught to be more
632 resilient for selective requests.
634 For the message types that already solicit a reply from the client, the
635 presence of VHOST_USER_PROTOCOL_F_REPLY_ACK or need_reply bit being set brings
636 no behavioural change. (See the 'Communication' section for details.)