contrib/libvhost-user/libvhost-user.h

   1 /*
   2  * Vhost User library
   3  *
   4  * Copyright (c) 2016 Red Hat, Inc.
   5  *
   6  * Authors:
   7  *  Victor Kaplansky <victork@redhat.com>
   8  *  Marc-André Lureau <mlureau@redhat.com>
   9  *
  10  * This work is licensed under the terms of the GNU GPL, version 2 or
  11  * later.  See the COPYING file in the top-level directory.
  12  */
  13
  14 #ifndef LIBVHOST_USER_H
  15 #define LIBVHOST_USER_H
  16
  17 #include <stdint.h>
  18 #include <stdbool.h>
  19 #include <stddef.h>
  20 #include <sys/poll.h>
  21 #include <linux/vhost.h>
  22 #include "standard-headers/linux/virtio_ring.h"
  23
  24 /* Based on qemu/hw/virtio/vhost-user.c */
  25 #define VHOST_USER_F_PROTOCOL_FEATURES 30
  26 #define VHOST_LOG_PAGE 4096
  27
  28 #define VHOST_MAX_NR_VIRTQUEUE 8
  29 #define VIRTQUEUE_MAX_SIZE 1024
  30
  31 #define VHOST_MEMORY_MAX_NREGIONS 8
  32
  33 typedef enum VhostSetConfigType {
  34     VHOST_SET_CONFIG_TYPE_MASTER = 0,
  35     VHOST_SET_CONFIG_TYPE_MIGRATION = 1,
  36 } VhostSetConfigType;
  37
  38 /*
  39  * Maximum size of virtio device config space
  40  */
  41 #define VHOST_USER_MAX_CONFIG_SIZE 256
  42
  43 enum VhostUserProtocolFeature {
  44     VHOST_USER_PROTOCOL_F_MQ = 0,
  45     VHOST_USER_PROTOCOL_F_LOG_SHMFD = 1,
  46     VHOST_USER_PROTOCOL_F_RARP = 2,
  47     VHOST_USER_PROTOCOL_F_REPLY_ACK = 3,
  48     VHOST_USER_PROTOCOL_F_NET_MTU = 4,
  49     VHOST_USER_PROTOCOL_F_SLAVE_REQ = 5,
  50     VHOST_USER_PROTOCOL_F_CROSS_ENDIAN = 6,
  51     VHOST_USER_PROTOCOL_F_CRYPTO_SESSION = 7,
  52     VHOST_USER_PROTOCOL_F_PAGEFAULT = 8,
  53     VHOST_USER_PROTOCOL_F_CONFIG = 9,
  54     VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD = 10,
  55     VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
  56     VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12,
  57
  58     VHOST_USER_PROTOCOL_F_MAX
  59 };
  60
  61 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
  62
  63 typedef enum VhostUserRequest {
  64     VHOST_USER_NONE = 0,
  65     VHOST_USER_GET_FEATURES = 1,
  66     VHOST_USER_SET_FEATURES = 2,
  67     VHOST_USER_SET_OWNER = 3,
  68     VHOST_USER_RESET_OWNER = 4,
  69     VHOST_USER_SET_MEM_TABLE = 5,
  70     VHOST_USER_SET_LOG_BASE = 6,
  71     VHOST_USER_SET_LOG_FD = 7,
  72     VHOST_USER_SET_VRING_NUM = 8,
  73     VHOST_USER_SET_VRING_ADDR = 9,
  74     VHOST_USER_SET_VRING_BASE = 10,
  75     VHOST_USER_GET_VRING_BASE = 11,
  76     VHOST_USER_SET_VRING_KICK = 12,
  77     VHOST_USER_SET_VRING_CALL = 13,
  78     VHOST_USER_SET_VRING_ERR = 14,
  79     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
  80     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
  81     VHOST_USER_GET_QUEUE_NUM = 17,
  82     VHOST_USER_SET_VRING_ENABLE = 18,
  83     VHOST_USER_SEND_RARP = 19,
  84     VHOST_USER_NET_SET_MTU = 20,
  85     VHOST_USER_SET_SLAVE_REQ_FD = 21,
  86     VHOST_USER_IOTLB_MSG = 22,
  87     VHOST_USER_SET_VRING_ENDIAN = 23,
  88     VHOST_USER_GET_CONFIG = 24,
  89     VHOST_USER_SET_CONFIG = 25,
  90     VHOST_USER_CREATE_CRYPTO_SESSION = 26,
  91     VHOST_USER_CLOSE_CRYPTO_SESSION = 27,
  92     VHOST_USER_POSTCOPY_ADVISE  = 28,
  93     VHOST_USER_POSTCOPY_LISTEN  = 29,
  94     VHOST_USER_POSTCOPY_END     = 30,
  95     VHOST_USER_GET_INFLIGHT_FD = 31,
  96     VHOST_USER_SET_INFLIGHT_FD = 32,
  97     VHOST_USER_GPU_SET_SOCKET = 33,
  98     VHOST_USER_MAX
  99 } VhostUserRequest;
 100
 101 typedef enum VhostUserSlaveRequest {
 102     VHOST_USER_SLAVE_NONE = 0,
 103     VHOST_USER_SLAVE_IOTLB_MSG = 1,
 104     VHOST_USER_SLAVE_CONFIG_CHANGE_MSG = 2,
 105     VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG = 3,
 106     VHOST_USER_SLAVE_MAX
 107 }  VhostUserSlaveRequest;
 108
 109 typedef struct VhostUserMemoryRegion {
 110     uint64_t guest_phys_addr;
 111     uint64_t memory_size;
 112     uint64_t userspace_addr;
 113     uint64_t mmap_offset;
 114 } VhostUserMemoryRegion;
 115
 116 typedef struct VhostUserMemory {
 117     uint32_t nregions;
 118     uint32_t padding;
 119     VhostUserMemoryRegion regions[VHOST_MEMORY_MAX_NREGIONS];
 120 } VhostUserMemory;
 121
 122 typedef struct VhostUserLog {
 123     uint64_t mmap_size;
 124     uint64_t mmap_offset;
 125 } VhostUserLog;
 126
 127 typedef struct VhostUserConfig {
 128     uint32_t offset;
 129     uint32_t size;
 130     uint32_t flags;
 131     uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
 132 } VhostUserConfig;
 133
 134 static VhostUserConfig c __attribute__ ((unused));
 135 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
 136                                    + sizeof(c.size) \
 137                                    + sizeof(c.flags))
 138
 139 typedef struct VhostUserVringArea {
 140     uint64_t u64;
 141     uint64_t size;
 142     uint64_t offset;
 143 } VhostUserVringArea;
 144
 145 typedef struct VhostUserInflight {
 146     uint64_t mmap_size;
 147     uint64_t mmap_offset;
 148     uint16_t num_queues;
 149     uint16_t queue_size;
 150 } VhostUserInflight;
 151
 152 #if defined(_WIN32) && (defined(__x86_64__) || defined(__i386__))
 153 # define VU_PACKED __attribute__((gcc_struct, packed))
 154 #else
 155 # define VU_PACKED __attribute__((packed))
 156 #endif
 157
 158 typedef struct VhostUserMsg {
 159     int request;
 160
 161 #define VHOST_USER_VERSION_MASK     (0x3)
 162 #define VHOST_USER_REPLY_MASK       (0x1 << 2)
 163 #define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
 164     uint32_t flags;
 165     uint32_t size; /* the following payload size */
 166
 167     union {
 168 #define VHOST_USER_VRING_IDX_MASK   (0xff)
 169 #define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
 170         uint64_t u64;
 171         struct vhost_vring_state state;
 172         struct vhost_vring_addr addr;
 173         VhostUserMemory memory;
 174         VhostUserLog log;
 175         VhostUserConfig config;
 176         VhostUserVringArea area;
 177         VhostUserInflight inflight;
 178     } payload;
 179
 180     int fds[VHOST_MEMORY_MAX_NREGIONS];
 181     int fd_num;
 182     uint8_t *data;
 183 } VU_PACKED VhostUserMsg;
 184
 185 typedef struct VuDevRegion {
 186     /* Guest Physical address. */
 187     uint64_t gpa;
 188     /* Memory region size. */
 189     uint64_t size;
 190     /* QEMU virtual address (userspace). */
 191     uint64_t qva;
 192     /* Starting offset in our mmaped space. */
 193     uint64_t mmap_offset;
 194     /* Start address of mmaped space. */
 195     uint64_t mmap_addr;
 196 } VuDevRegion;
 197
 198 typedef struct VuDev VuDev;
 199
 200 typedef uint64_t (*vu_get_features_cb) (VuDev *dev);
 201 typedef void (*vu_set_features_cb) (VuDev *dev, uint64_t features);
 202 typedef int (*vu_process_msg_cb) (VuDev *dev, VhostUserMsg *vmsg,
 203                                   int *do_reply);
 204 typedef void (*vu_queue_set_started_cb) (VuDev *dev, int qidx, bool started);
 205 typedef bool (*vu_queue_is_processed_in_order_cb) (VuDev *dev, int qidx);
 206 typedef int (*vu_get_config_cb) (VuDev *dev, uint8_t *config, uint32_t len);
 207 typedef int (*vu_set_config_cb) (VuDev *dev, const uint8_t *data,
 208                                  uint32_t offset, uint32_t size,
 209                                  uint32_t flags);
 210
 211 typedef struct VuDevIface {
 212     /* called by VHOST_USER_GET_FEATURES to get the features bitmask */
 213     vu_get_features_cb get_features;
 214     /* enable vhost implementation features */
 215     vu_set_features_cb set_features;
 216     /* get the protocol feature bitmask from the underlying vhost
 217      * implementation */
 218     vu_get_features_cb get_protocol_features;
 219     /* enable protocol features in the underlying vhost implementation. */
 220     vu_set_features_cb set_protocol_features;
 221     /* process_msg is called for each vhost-user message received */
 222     /* skip libvhost-user processing if return value != 0 */
 223     vu_process_msg_cb process_msg;
 224     /* tells when queues can be processed */
 225     vu_queue_set_started_cb queue_set_started;
 226     /*
 227      * If the queue is processed in order, in which case it will be
 228      * resumed to vring.used->idx. This can help to support resuming
 229      * on unmanaged exit/crash.
 230      */
 231     vu_queue_is_processed_in_order_cb queue_is_processed_in_order;
 232     /* get the config space of the device */
 233     vu_get_config_cb get_config;
 234     /* set the config space of the device */
 235     vu_set_config_cb set_config;
 236 } VuDevIface;
 237
 238 typedef void (*vu_queue_handler_cb) (VuDev *dev, int qidx);
 239
 240 typedef struct VuRing {
 241     unsigned int num;
 242     struct vring_desc *desc;
 243     struct vring_avail *avail;
 244     struct vring_used *used;
 245     uint64_t log_guest_addr;
 246     uint32_t flags;
 247 } VuRing;
 248
 249 typedef struct VuDescStateSplit {
 250     /* Indicate whether this descriptor is inflight or not.
 251      * Only available for head-descriptor. */
 252     uint8_t inflight;
 253
 254     /* Padding */
 255     uint8_t padding[5];
 256
 257     /* Maintain a list for the last batch of used descriptors.
 258      * Only available when batching is used for submitting */
 259     uint16_t next;
 260
 261     /* Used to preserve the order of fetching available descriptors.
 262      * Only available for head-descriptor. */
 263     uint64_t counter;
 264 } VuDescStateSplit;
 265
 266 typedef struct VuVirtqInflight {
 267     /* The feature flags of this region. Now it's initialized to 0. */
 268     uint64_t features;
 269
 270     /* The version of this region. It's 1 currently.
 271      * Zero value indicates a vm reset happened. */
 272     uint16_t version;
 273
 274     /* The size of VuDescStateSplit array. It's equal to the virtqueue
 275      * size. Slave could get it from queue size field of VhostUserInflight. */
 276     uint16_t desc_num;
 277
 278     /* The head of list that track the last batch of used descriptors. */
 279     uint16_t last_batch_head;
 280
 281     /* Storing the idx value of used ring */
 282     uint16_t used_idx;
 283
 284     /* Used to track the state of each descriptor in descriptor table */
 285     VuDescStateSplit desc[0];
 286 } VuVirtqInflight;
 287
 288 typedef struct VuVirtqInflightDesc {
 289     uint16_t index;
 290     uint64_t counter;
 291 } VuVirtqInflightDesc;
 292
 293 typedef struct VuVirtq {
 294     VuRing vring;
 295
 296     VuVirtqInflight *inflight;
 297
 298     VuVirtqInflightDesc *resubmit_list;
 299
 300     uint16_t resubmit_num;
 301
 302     uint64_t counter;
 303
 304     /* Next head to pop */
 305     uint16_t last_avail_idx;
 306
 307     /* Last avail_idx read from VQ. */
 308     uint16_t shadow_avail_idx;
 309
 310     uint16_t used_idx;
 311
 312     /* Last used index value we have signalled on */
 313     uint16_t signalled_used;
 314
 315     /* Last used index value we have signalled on */
 316     bool signalled_used_valid;
 317
 318     /* Notification enabled? */
 319     bool notification;
 320
 321     int inuse;
 322
 323     vu_queue_handler_cb handler;
 324
 325     int call_fd;
 326     int kick_fd;
 327     int err_fd;
 328     unsigned int enable;
 329     bool started;
 330 } VuVirtq;
 331
 332 enum VuWatchCondtion {
 333     VU_WATCH_IN = POLLIN,
 334     VU_WATCH_OUT = POLLOUT,
 335     VU_WATCH_PRI = POLLPRI,
 336     VU_WATCH_ERR = POLLERR,
 337     VU_WATCH_HUP = POLLHUP,
 338 };
 339
 340 typedef void (*vu_panic_cb) (VuDev *dev, const char *err);
 341 typedef void (*vu_watch_cb) (VuDev *dev, int condition, void *data);
 342 typedef void (*vu_set_watch_cb) (VuDev *dev, int fd, int condition,
 343                                  vu_watch_cb cb, void *data);
 344 typedef void (*vu_remove_watch_cb) (VuDev *dev, int fd);
 345
 346 typedef struct VuDevInflightInfo {
 347     int fd;
 348     void *addr;
 349     uint64_t size;
 350 } VuDevInflightInfo;
 351
 352 struct VuDev {
 353     int sock;
 354     uint32_t nregions;
 355     VuDevRegion regions[VHOST_MEMORY_MAX_NREGIONS];
 356     VuVirtq vq[VHOST_MAX_NR_VIRTQUEUE];
 357     VuDevInflightInfo inflight_info;
 358     int log_call_fd;
 359     int slave_fd;
 360     uint64_t log_size;
 361     uint8_t *log_table;
 362     uint64_t features;
 363     uint64_t protocol_features;
 364     bool broken;
 365
 366     /* @set_watch: add or update the given fd to the watch set,
 367      * call cb when condition is met */
 368     vu_set_watch_cb set_watch;
 369
 370     /* @remove_watch: remove the given fd from the watch set */
 371     vu_remove_watch_cb remove_watch;
 372
 373     /* @panic: encountered an unrecoverable error, you may try to
 374      * re-initialize */
 375     vu_panic_cb panic;
 376     const VuDevIface *iface;
 377
 378     /* Postcopy data */
 379     int postcopy_ufd;
 380     bool postcopy_listening;
 381 };
 382
 383 typedef struct VuVirtqElement {
 384     unsigned int index;
 385     unsigned int out_num;
 386     unsigned int in_num;
 387     struct iovec *in_sg;
 388     struct iovec *out_sg;
 389 } VuVirtqElement;
 390
 391 /**
 392  * vu_init:
 393  * @dev: a VuDev context
 394  * @socket: the socket connected to vhost-user master
 395  * @panic: a panic callback
 396  * @set_watch: a set_watch callback
 397  * @remove_watch: a remove_watch callback
 398  * @iface: a VuDevIface structure with vhost-user device callbacks
 399  *
 400  * Intializes a VuDev vhost-user context.
 401  **/
 402 void vu_init(VuDev *dev,
 403              int socket,
 404              vu_panic_cb panic,
 405              vu_set_watch_cb set_watch,
 406              vu_remove_watch_cb remove_watch,
 407              const VuDevIface *iface);
 408
 409
 410 /**
 411  * vu_deinit:
 412  * @dev: a VuDev context
 413  *
 414  * Cleans up the VuDev context
 415  */
 416 void vu_deinit(VuDev *dev);
 417
 418 /**
 419  * vu_dispatch:
 420  * @dev: a VuDev context
 421  *
 422  * Process one vhost-user message.
 423  *
 424  * Returns: TRUE on success, FALSE on failure.
 425  */
 426 bool vu_dispatch(VuDev *dev);
 427
 428 /**
 429  * vu_gpa_to_va:
 430  * @dev: a VuDev context
 431  * @plen: guest memory size
 432  * @guest_addr: guest address
 433  *
 434  * Translate a guest address to a pointer. Returns NULL on failure.
 435  */
 436 void *vu_gpa_to_va(VuDev *dev, uint64_t *plen, uint64_t guest_addr);
 437
 438 /**
 439  * vu_get_queue:
 440  * @dev: a VuDev context
 441  * @qidx: queue index
 442  *
 443  * Returns the queue number @qidx.
 444  */
 445 VuVirtq *vu_get_queue(VuDev *dev, int qidx);
 446
 447 /**
 448  * vu_set_queue_handler:
 449  * @dev: a VuDev context
 450  * @vq: a VuVirtq queue
 451  * @handler: the queue handler callback
 452  *
 453  * Set the queue handler. This function may be called several times
 454  * for the same queue. If called with NULL @handler, the handler is
 455  * removed.
 456  */
 457 void vu_set_queue_handler(VuDev *dev, VuVirtq *vq,
 458                           vu_queue_handler_cb handler);
 459
 460 /**
 461  * vu_set_queue_host_notifier:
 462  * @dev: a VuDev context
 463  * @vq: a VuVirtq queue
 464  * @fd: a file descriptor
 465  * @size: host page size
 466  * @offset: notifier offset in @fd file
 467  *
 468  * Set queue's host notifier. This function may be called several
 469  * times for the same queue. If called with -1 @fd, the notifier
 470  * is removed.
 471  */
 472 bool vu_set_queue_host_notifier(VuDev *dev, VuVirtq *vq, int fd,
 473                                 int size, int offset);
 474
 475 /**
 476  * vu_queue_set_notification:
 477  * @dev: a VuDev context
 478  * @vq: a VuVirtq queue
 479  * @enable: state
 480  *
 481  * Set whether the queue notifies (via event index or interrupt)
 482  */
 483 void vu_queue_set_notification(VuDev *dev, VuVirtq *vq, int enable);
 484
 485 /**
 486  * vu_queue_enabled:
 487  * @dev: a VuDev context
 488  * @vq: a VuVirtq queue
 489  *
 490  * Returns: whether the queue is enabled.
 491  */
 492 bool vu_queue_enabled(VuDev *dev, VuVirtq *vq);
 493
 494 /**
 495  * vu_queue_started:
 496  * @dev: a VuDev context
 497  * @vq: a VuVirtq queue
 498  *
 499  * Returns: whether the queue is started.
 500  */
 501 bool vu_queue_started(const VuDev *dev, const VuVirtq *vq);
 502
 503 /**
 504  * vu_queue_empty:
 505  * @dev: a VuDev context
 506  * @vq: a VuVirtq queue
 507  *
 508  * Returns: true if the queue is empty or not ready.
 509  */
 510 bool vu_queue_empty(VuDev *dev, VuVirtq *vq);
 511
 512 /**
 513  * vu_queue_notify:
 514  * @dev: a VuDev context
 515  * @vq: a VuVirtq queue
 516  *
 517  * Request to notify the queue via callfd (skipped if unnecessary)
 518  */
 519 void vu_queue_notify(VuDev *dev, VuVirtq *vq);
 520
 521 /**
 522  * vu_queue_pop:
 523  * @dev: a VuDev context
 524  * @vq: a VuVirtq queue
 525  * @sz: the size of struct to return (must be >= VuVirtqElement)
 526  *
 527  * Returns: a VuVirtqElement filled from the queue or NULL. The
 528  * returned element must be free()-d by the caller.
 529  */
 530 void *vu_queue_pop(VuDev *dev, VuVirtq *vq, size_t sz);
 531
 532
 533 /**
 534  * vu_queue_unpop:
 535  * @dev: a VuDev context
 536  * @vq: a VuVirtq queue
 537  * @elem: The #VuVirtqElement
 538  * @len: number of bytes written
 539  *
 540  * Pretend the most recent element wasn't popped from the virtqueue.  The next
 541  * call to vu_queue_pop() will refetch the element.
 542  */
 543 void vu_queue_unpop(VuDev *dev, VuVirtq *vq, VuVirtqElement *elem,
 544                     size_t len);
 545
 546 /**
 547  * vu_queue_rewind:
 548  * @dev: a VuDev context
 549  * @vq: a VuVirtq queue
 550  * @num: number of elements to push back
 551  *
 552  * Pretend that elements weren't popped from the virtqueue.  The next
 553  * virtqueue_pop() will refetch the oldest element.
 554  *
 555  * Returns: true on success, false if @num is greater than the number of in use
 556  * elements.
 557  */
 558 bool vu_queue_rewind(VuDev *dev, VuVirtq *vq, unsigned int num);
 559
 560 /**
 561  * vu_queue_fill:
 562  * @dev: a VuDev context
 563  * @vq: a VuVirtq queue
 564  * @elem: a VuVirtqElement
 565  * @len: length in bytes to write
 566  * @idx: optional offset for the used ring index (0 in general)
 567  *
 568  * Fill the used ring with @elem element.
 569  */
 570 void vu_queue_fill(VuDev *dev, VuVirtq *vq,
 571                    const VuVirtqElement *elem,
 572                    unsigned int len, unsigned int idx);
 573
 574 /**
 575  * vu_queue_push:
 576  * @dev: a VuDev context
 577  * @vq: a VuVirtq queue
 578  * @elem: a VuVirtqElement
 579  * @len: length in bytes to write
 580  *
 581  * Helper that combines vu_queue_fill() with a vu_queue_flush().
 582  */
 583 void vu_queue_push(VuDev *dev, VuVirtq *vq,
 584                    const VuVirtqElement *elem, unsigned int len);
 585
 586 /**
 587  * vu_queue_flush:
 588  * @dev: a VuDev context
 589  * @vq: a VuVirtq queue
 590  * @num: number of elements to flush
 591  *
 592  * Mark the last number of elements as done (used.idx is updated by
 593  * num elements).
 594 */
 595 void vu_queue_flush(VuDev *dev, VuVirtq *vq, unsigned int num);
 596
 597 /**
 598  * vu_queue_get_avail_bytes:
 599  * @dev: a VuDev context
 600  * @vq: a VuVirtq queue
 601  * @in_bytes: in bytes
 602  * @out_bytes: out bytes
 603  * @max_in_bytes: stop counting after max_in_bytes
 604  * @max_out_bytes: stop counting after max_out_bytes
 605  *
 606  * Count the number of available bytes, up to max_in_bytes/max_out_bytes.
 607  */
 608 void vu_queue_get_avail_bytes(VuDev *vdev, VuVirtq *vq, unsigned int *in_bytes,
 609                               unsigned int *out_bytes,
 610                               unsigned max_in_bytes, unsigned max_out_bytes);
 611
 612 /**
 613  * vu_queue_avail_bytes:
 614  * @dev: a VuDev context
 615  * @vq: a VuVirtq queue
 616  * @in_bytes: expected in bytes
 617  * @out_bytes: expected out bytes
 618  *
 619  * Returns: true if in_bytes <= in_total && out_bytes <= out_total
 620  */
 621 bool vu_queue_avail_bytes(VuDev *dev, VuVirtq *vq, unsigned int in_bytes,
 622                           unsigned int out_bytes);
 623
 624 #endif /* LIBVHOST_USER_H */