docs/specs/vhost-user.txt

   1 Vhost-user Protocol
   2 ===================
   3
   4 Copyright (c) 2014 Virtual Open Systems Sarl.
   5
   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.
   8 ===================
   9
  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.
  15
  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.
  19
  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.
  22
  23 Master and slave can be either a client (i.e. connecting) or server (listening)
  24 in the socket communication.
  25
  26 Message Specification
  27 ---------------------
  28
  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:
  31
  32 ------------------------------------
  33 | request | flags | size | payload |
  34 ------------------------------------
  35
  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
  41      details.
  42  * Size - 32-bit size of the payload
  43
  44
  45 Depending on the request type, payload can be:
  46
  47  * A single 64-bit integer
  48    -------
  49    | u64 |
  50    -------
  51
  52    u64: a 64-bit unsigned integer
  53
  54  * A vring state description
  55    ---------------
  56   | index | num |
  57   ---------------
  58
  59    Index: a 32-bit index
  60    Num: a 32-bit number
  61
  62  * A vring address description
  63    --------------------------------------------------------------
  64    | index | flags | size | descriptor | used | available | log |
  65    --------------------------------------------------------------
  66
  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
  73
  74  * Memory regions description
  75    ---------------------------------------------------
  76    | num regions | padding | region0 | ... | region7 |
  77    ---------------------------------------------------
  78
  79    Num regions: a 32-bit number of regions
  80    Padding: 32-bit
  81
  82    A region is:
  83    -----------------------------------------------------
  84    | guest address | size | user address | mmap offset |
  85    -----------------------------------------------------
  86
  87    Guest address: a 64-bit guest address of the region
  88    Size: a 64-bit size
  89    User address: a 64-bit user address
  90    mmap offset: 64-bit offset where region starts in the mapped memory
  91
  92 * Log description
  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)
  99
 100 In QEMU the vhost-user message is implemented with the following struct:
 101
 102 typedef struct VhostUserMsg {
 103     VhostUserRequest request;
 104     uint32_t flags;
 105     uint32_t size;
 106     union {
 107         uint64_t u64;
 108         struct vhost_vring_state state;
 109         struct vhost_vring_addr addr;
 110         VhostUserMemory memory;
 111         VhostUserLog log;
 112     };
 113 } QEMU_PACKED VhostUserMsg;
 114
 115 Communication
 116 -------------
 117
 118 The protocol for vhost-user is based on the existing implementation of vhost
 119 for the Linux Kernel. Most messages that can be sent via the Unix domain socket
 120 implementing vhost-user have an equivalent ioctl to the kernel implementation.
 121
 122 The communication consists of master sending message requests and slave sending
 123 message replies. Most of the requests don't require replies. Here is a list of
 124 the ones that do:
 125
 126  * VHOST_GET_FEATURES
 127  * VHOST_GET_PROTOCOL_FEATURES
 128  * VHOST_GET_VRING_BASE
 129  * VHOST_SET_LOG_BASE (if VHOST_USER_PROTOCOL_F_LOG_SHMFD)
 130
 131 [ Also see the section on REPLY_ACK protocol extension. ]
 132
 133 There are several messages that the master sends with file descriptors passed
 134 in the ancillary data:
 135
 136  * VHOST_SET_MEM_TABLE
 137  * VHOST_SET_LOG_BASE (if VHOST_USER_PROTOCOL_F_LOG_SHMFD)
 138  * VHOST_SET_LOG_FD
 139  * VHOST_SET_VRING_KICK
 140  * VHOST_SET_VRING_CALL
 141  * VHOST_SET_VRING_ERR
 142
 143 If Master is unable to send the full message or receives a wrong reply it will
 144 close the connection. An optional reconnection mechanism can be implemented.
 145
 146 Any protocol extensions are gated by protocol feature bits,
 147 which allows full backwards compatibility on both master
 148 and slave.
 149 As older slaves don't support negotiating protocol features,
 150 a feature bit was dedicated for this purpose:
 151 #define VHOST_USER_F_PROTOCOL_FEATURES 30
 152
 153 Starting and stopping rings
 154 ----------------------
 155 Client must only process each ring when it is started.
 156
 157 Client must only pass data between the ring and the
 158 backend, when the ring is enabled.
 159
 160 If ring is started but disabled, client must process the
 161 ring without talking to the backend.
 162
 163 For example, for a networking device, in the disabled state
 164 client must not supply any new RX packets, but must process
 165 and discard any TX packets.
 166
 167 If VHOST_USER_F_PROTOCOL_FEATURES has not been negotiated, the ring is initialized
 168 in an enabled state.
 169
 170 If VHOST_USER_F_PROTOCOL_FEATURES has been negotiated, the ring is initialized
 171 in a disabled state. Client must not pass data to/from the backend until ring is enabled by
 172 VHOST_USER_SET_VRING_ENABLE with parameter 1, or after it has been disabled by
 173 VHOST_USER_SET_VRING_ENABLE with parameter 0.
 174
 175 Each ring is initialized in a stopped state, client must not process it until
 176 ring is started, or after it has been stopped.
 177
 178 Client must start ring upon receiving a kick (that is, detecting that file
 179 descriptor is readable) on the descriptor specified by
 180 VHOST_USER_SET_VRING_KICK, and stop ring upon receiving
 181 VHOST_USER_GET_VRING_BASE.
 182
 183 While processing the rings (whether they are enabled or not), client must
 184 support changing some configuration aspects on the fly.
 185
 186 Multiple queue support
 187 ----------------------
 188
 189 Multiple queue is treated as a protocol extension, hence the slave has to
 190 implement protocol features first. The multiple queues feature is supported
 191 only when the protocol feature VHOST_USER_PROTOCOL_F_MQ (bit 0) is set.
 192
 193 The max number of queues the slave supports can be queried with message
 194 VHOST_USER_GET_PROTOCOL_FEATURES. Master should stop when the number of
 195 requested queues is bigger than that.
 196
 197 As all queues share one connection, the master uses a unique index for each
 198 queue in the sent message to identify a specified queue. One queue pair
 199 is enabled initially. More queues are enabled dynamically, by sending
 200 message VHOST_USER_SET_VRING_ENABLE.
 201
 202 Migration
 203 ---------
 204
 205 During live migration, the master may need to track the modifications
 206 the slave makes to the memory mapped regions. The client should mark
 207 the dirty pages in a log. Once it complies to this logging, it may
 208 declare the VHOST_F_LOG_ALL vhost feature.
 209
 210 To start/stop logging of data/used ring writes, server may send messages
 211 VHOST_USER_SET_FEATURES with VHOST_F_LOG_ALL and VHOST_USER_SET_VRING_ADDR with
 212 VHOST_VRING_F_LOG in ring's flags set to 1/0, respectively.
 213
 214 All the modifications to memory pointed by vring "descriptor" should
 215 be marked. Modifications to "used" vring should be marked if
 216 VHOST_VRING_F_LOG is part of ring's flags.
 217
 218 Dirty pages are of size:
 219 #define VHOST_LOG_PAGE 0x1000
 220
 221 The log memory fd is provided in the ancillary data of
 222 VHOST_USER_SET_LOG_BASE message when the slave has
 223 VHOST_USER_PROTOCOL_F_LOG_SHMFD protocol feature.
 224
 225 The size of the log is supplied as part of VhostUserMsg
 226 which should be large enough to cover all known guest
 227 addresses. Log starts at the supplied offset in the
 228 supplied file descriptor.
 229 The log covers from address 0 to the maximum of guest
 230 regions. In pseudo-code, to mark page at "addr" as dirty:
 231
 232 page = addr / VHOST_LOG_PAGE
 233 log[page / 8] |= 1 << page % 8
 234
 235 Where addr is the guest physical address.
 236
 237 Use atomic operations, as the log may be concurrently manipulated.
 238
 239 Note that when logging modifications to the used ring (when VHOST_VRING_F_LOG
 240 is set for this ring), log_guest_addr should be used to calculate the log
 241 offset: the write to first byte of the used ring is logged at this offset from
 242 log start. Also note that this value might be outside the legal guest physical
 243 address range (i.e. does not have to be covered by the VhostUserMemory table),
 244 but the bit offset of the last byte of the ring must fall within
 245 the size supplied by VhostUserLog.
 246
 247 VHOST_USER_SET_LOG_FD is an optional message with an eventfd in
 248 ancillary data, it may be used to inform the master that the log has
 249 been modified.
 250
 251 Once the source has finished migration, rings will be stopped by
 252 the source. No further update must be done before rings are
 253 restarted.
 254
 255 Protocol features
 256 -----------------
 257
 258 #define VHOST_USER_PROTOCOL_F_MQ             0
 259 #define VHOST_USER_PROTOCOL_F_LOG_SHMFD      1
 260 #define VHOST_USER_PROTOCOL_F_RARP           2
 261 #define VHOST_USER_PROTOCOL_F_REPLY_ACK      3
 262
 263 Message types
 264 -------------
 265
 266  * VHOST_USER_GET_FEATURES
 267
 268       Id: 1
 269       Equivalent ioctl: VHOST_GET_FEATURES
 270       Master payload: N/A
 271       Slave payload: u64
 272
 273       Get from the underlying vhost implementation the features bitmask.
 274       Feature bit VHOST_USER_F_PROTOCOL_FEATURES signals slave support for
 275       VHOST_USER_GET_PROTOCOL_FEATURES and VHOST_USER_SET_PROTOCOL_FEATURES.
 276
 277  * VHOST_USER_SET_FEATURES
 278
 279       Id: 2
 280       Ioctl: VHOST_SET_FEATURES
 281       Master payload: u64
 282
 283       Enable features in the underlying vhost implementation using a bitmask.
 284       Feature bit VHOST_USER_F_PROTOCOL_FEATURES signals slave support for
 285       VHOST_USER_GET_PROTOCOL_FEATURES and VHOST_USER_SET_PROTOCOL_FEATURES.
 286
 287  * VHOST_USER_GET_PROTOCOL_FEATURES
 288
 289       Id: 15
 290       Equivalent ioctl: VHOST_GET_FEATURES
 291       Master payload: N/A
 292       Slave payload: u64
 293
 294       Get the protocol feature bitmask from the underlying vhost implementation.
 295       Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
 296       VHOST_USER_GET_FEATURES.
 297       Note: slave that reported VHOST_USER_F_PROTOCOL_FEATURES must support
 298       this message even before VHOST_USER_SET_FEATURES was called.
 299
 300  * VHOST_USER_SET_PROTOCOL_FEATURES
 301
 302       Id: 16
 303       Ioctl: VHOST_SET_FEATURES
 304       Master payload: u64
 305
 306       Enable protocol features in the underlying vhost implementation.
 307       Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
 308       VHOST_USER_GET_FEATURES.
 309       Note: slave that reported VHOST_USER_F_PROTOCOL_FEATURES must support
 310       this message even before VHOST_USER_SET_FEATURES was called.
 311
 312  * VHOST_USER_SET_OWNER
 313
 314       Id: 3
 315       Equivalent ioctl: VHOST_SET_OWNER
 316       Master payload: N/A
 317
 318       Issued when a new connection is established. It sets the current Master
 319       as an owner of the session. This can be used on the Slave as a
 320       "session start" flag.
 321
 322  * VHOST_USER_RESET_OWNER
 323
 324       Id: 4
 325       Master payload: N/A
 326
 327       This is no longer used. Used to be sent to request disabling
 328       all rings, but some clients interpreted it to also discard
 329       connection state (this interpretation would lead to bugs).
 330       It is recommended that clients either ignore this message,
 331       or use it to disable all rings.
 332
 333  * VHOST_USER_SET_MEM_TABLE
 334
 335       Id: 5
 336       Equivalent ioctl: VHOST_SET_MEM_TABLE
 337       Master payload: memory regions description
 338
 339       Sets the memory map regions on the slave so it can translate the vring
 340       addresses. In the ancillary data there is an array of file descriptors
 341       for each memory mapped region. The size and ordering of the fds matches
 342       the number and ordering of memory regions.
 343
 344  * VHOST_USER_SET_LOG_BASE
 345
 346       Id: 6
 347       Equivalent ioctl: VHOST_SET_LOG_BASE
 348       Master payload: u64
 349       Slave payload: N/A
 350
 351       Sets logging shared memory space.
 352       When slave has VHOST_USER_PROTOCOL_F_LOG_SHMFD protocol
 353       feature, the log memory fd is provided in the ancillary data of
 354       VHOST_USER_SET_LOG_BASE message, the size and offset of shared
 355       memory area provided in the message.
 356
 357
 358  * VHOST_USER_SET_LOG_FD
 359
 360       Id: 7
 361       Equivalent ioctl: VHOST_SET_LOG_FD
 362       Master payload: N/A
 363
 364       Sets the logging file descriptor, which is passed as ancillary data.
 365
 366  * VHOST_USER_SET_VRING_NUM
 367
 368       Id: 8
 369       Equivalent ioctl: VHOST_SET_VRING_NUM
 370       Master payload: vring state description
 371
 372       Set the size of the queue.
 373
 374  * VHOST_USER_SET_VRING_ADDR
 375
 376       Id: 9
 377       Equivalent ioctl: VHOST_SET_VRING_ADDR
 378       Master payload: vring address description
 379       Slave payload: N/A
 380
 381       Sets the addresses of the different aspects of the vring.
 382
 383  * VHOST_USER_SET_VRING_BASE
 384
 385       Id: 10
 386       Equivalent ioctl: VHOST_SET_VRING_BASE
 387       Master payload: vring state description
 388
 389       Sets the base offset in the available vring.
 390
 391  * VHOST_USER_GET_VRING_BASE
 392
 393       Id: 11
 394       Equivalent ioctl: VHOST_USER_GET_VRING_BASE
 395       Master payload: vring state description
 396       Slave payload: vring state description
 397
 398       Get the available vring base offset.
 399
 400  * VHOST_USER_SET_VRING_KICK
 401
 402       Id: 12
 403       Equivalent ioctl: VHOST_SET_VRING_KICK
 404       Master payload: u64
 405
 406       Set the event file descriptor for adding buffers to the vring. It
 407       is passed in the ancillary data.
 408       Bits (0-7) of the payload contain the vring index. Bit 8 is the
 409       invalid FD flag. This flag is set when there is no file descriptor
 410       in the ancillary data. This signals that polling should be used
 411       instead of waiting for a kick.
 412
 413  * VHOST_USER_SET_VRING_CALL
 414
 415       Id: 13
 416       Equivalent ioctl: VHOST_SET_VRING_CALL
 417       Master payload: u64
 418
 419       Set the event file descriptor to signal when buffers are used. It
 420       is passed in the ancillary data.
 421       Bits (0-7) of the payload contain the vring index. Bit 8 is the
 422       invalid FD flag. This flag is set when there is no file descriptor
 423       in the ancillary data. This signals that polling will be used
 424       instead of waiting for the call.
 425
 426  * VHOST_USER_SET_VRING_ERR
 427
 428       Id: 14
 429       Equivalent ioctl: VHOST_SET_VRING_ERR
 430       Master payload: u64
 431
 432       Set the event file descriptor to signal when error occurs. It
 433       is passed in the ancillary data.
 434       Bits (0-7) of the payload contain the vring index. Bit 8 is the
 435       invalid FD flag. This flag is set when there is no file descriptor
 436       in the ancillary data.
 437
 438  * VHOST_USER_GET_QUEUE_NUM
 439
 440       Id: 17
 441       Equivalent ioctl: N/A
 442       Master payload: N/A
 443       Slave payload: u64
 444
 445       Query how many queues the backend supports. This request should be
 446       sent only when VHOST_USER_PROTOCOL_F_MQ is set in queried protocol
 447       features by VHOST_USER_GET_PROTOCOL_FEATURES.
 448
 449  * VHOST_USER_SET_VRING_ENABLE
 450
 451       Id: 18
 452       Equivalent ioctl: N/A
 453       Master payload: vring state description
 454
 455       Signal slave to enable or disable corresponding vring.
 456       This request should be sent only when VHOST_USER_F_PROTOCOL_FEATURES
 457       has been negotiated.
 458
 459  * VHOST_USER_SEND_RARP
 460
 461       Id: 19
 462       Equivalent ioctl: N/A
 463       Master payload: u64
 464
 465       Ask vhost user backend to broadcast a fake RARP to notify the migration
 466       is terminated for guest that does not support GUEST_ANNOUNCE.
 467       Only legal if feature bit VHOST_USER_F_PROTOCOL_FEATURES is present in
 468       VHOST_USER_GET_FEATURES and protocol feature bit VHOST_USER_PROTOCOL_F_RARP
 469       is present in VHOST_USER_GET_PROTOCOL_FEATURES.
 470       The first 6 bytes of the payload contain the mac address of the guest to
 471       allow the vhost user backend to construct and broadcast the fake RARP.
 472
 473 VHOST_USER_PROTOCOL_F_REPLY_ACK:
 474 -------------------------------
 475 The original vhost-user specification only demands replies for certain
 476 commands. This differs from the vhost protocol implementation where commands
 477 are sent over an ioctl() call and block until the client has completed.
 478
 479 With this protocol extension negotiated, the sender (QEMU) can set the
 480 "need_reply" [Bit 3] flag to any command. This indicates that
 481 the client MUST respond with a Payload VhostUserMsg indicating success or
 482 failure. The payload should be set to zero on success or non-zero on failure,
 483 unless the message already has an explicit reply body.
 484
 485 The response payload gives QEMU a deterministic indication of the result
 486 of the command. Today, QEMU is expected to terminate the main vhost-user
 487 loop upon receiving such errors. In future, qemu could be taught to be more
 488 resilient for selective requests.
 489
 490 For the message types that already solicit a reply from the client, the
 491 presence of VHOST_USER_PROTOCOL_F_REPLY_ACK or need_reply bit being set brings
 492 no behavioural change. (See the 'Communication' section for details.)