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
  55     VHOST_USER_PROTOCOL_F_MAX
  56 };
  57
  58 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
  59
  60 typedef enum VhostUserRequest {
  61     VHOST_USER_NONE = 0,
  62     VHOST_USER_GET_FEATURES = 1,
  63     VHOST_USER_SET_FEATURES = 2,
  64     VHOST_USER_SET_OWNER = 3,
  65     VHOST_USER_RESET_OWNER = 4,
  66     VHOST_USER_SET_MEM_TABLE = 5,
  67     VHOST_USER_SET_LOG_BASE = 6,
  68     VHOST_USER_SET_LOG_FD = 7,
  69     VHOST_USER_SET_VRING_NUM = 8,
  70     VHOST_USER_SET_VRING_ADDR = 9,
  71     VHOST_USER_SET_VRING_BASE = 10,
  72     VHOST_USER_GET_VRING_BASE = 11,
  73     VHOST_USER_SET_VRING_KICK = 12,
  74     VHOST_USER_SET_VRING_CALL = 13,
  75     VHOST_USER_SET_VRING_ERR = 14,
  76     VHOST_USER_GET_PROTOCOL_FEATURES = 15,
  77     VHOST_USER_SET_PROTOCOL_FEATURES = 16,
  78     VHOST_USER_GET_QUEUE_NUM = 17,
  79     VHOST_USER_SET_VRING_ENABLE = 18,
  80     VHOST_USER_SEND_RARP = 19,
  81     VHOST_USER_NET_SET_MTU = 20,
  82     VHOST_USER_SET_SLAVE_REQ_FD = 21,
  83     VHOST_USER_IOTLB_MSG = 22,
  84     VHOST_USER_SET_VRING_ENDIAN = 23,
  85     VHOST_USER_GET_CONFIG = 24,
  86     VHOST_USER_SET_CONFIG = 25,
  87     VHOST_USER_CREATE_CRYPTO_SESSION = 26,
  88     VHOST_USER_CLOSE_CRYPTO_SESSION = 27,
  89     VHOST_USER_POSTCOPY_ADVISE  = 28,
  90     VHOST_USER_POSTCOPY_LISTEN  = 29,
  91     VHOST_USER_POSTCOPY_END     = 30,
  92     VHOST_USER_MAX
  93 } VhostUserRequest;
  94
  95 typedef struct VhostUserMemoryRegion {
  96     uint64_t guest_phys_addr;
  97     uint64_t memory_size;
  98     uint64_t userspace_addr;
  99     uint64_t mmap_offset;
 100 } VhostUserMemoryRegion;
 101
 102 typedef struct VhostUserMemory {
 103     uint32_t nregions;
 104     uint32_t padding;
 105     VhostUserMemoryRegion regions[VHOST_MEMORY_MAX_NREGIONS];
 106 } VhostUserMemory;
 107
 108 typedef struct VhostUserLog {
 109     uint64_t mmap_size;
 110     uint64_t mmap_offset;
 111 } VhostUserLog;
 112
 113 typedef struct VhostUserConfig {
 114     uint32_t offset;
 115     uint32_t size;
 116     uint32_t flags;
 117     uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
 118 } VhostUserConfig;
 119
 120 static VhostUserConfig c __attribute__ ((unused));
 121 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
 122                                    + sizeof(c.size) \
 123                                    + sizeof(c.flags))
 124
 125 #if defined(_WIN32)
 126 # define VU_PACKED __attribute__((gcc_struct, packed))
 127 #else
 128 # define VU_PACKED __attribute__((packed))
 129 #endif
 130
 131 typedef struct VhostUserMsg {
 132     VhostUserRequest request;
 133
 134 #define VHOST_USER_VERSION_MASK     (0x3)
 135 #define VHOST_USER_REPLY_MASK       (0x1 << 2)
 136     uint32_t flags;
 137     uint32_t size; /* the following payload size */
 138
 139     union {
 140 #define VHOST_USER_VRING_IDX_MASK   (0xff)
 141 #define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
 142         uint64_t u64;
 143         struct vhost_vring_state state;
 144         struct vhost_vring_addr addr;
 145         VhostUserMemory memory;
 146         VhostUserLog log;
 147         VhostUserConfig config;
 148     } payload;
 149
 150     int fds[VHOST_MEMORY_MAX_NREGIONS];
 151     int fd_num;
 152     uint8_t *data;
 153 } VU_PACKED VhostUserMsg;
 154
 155 typedef struct VuDevRegion {
 156     /* Guest Physical address. */
 157     uint64_t gpa;
 158     /* Memory region size. */
 159     uint64_t size;
 160     /* QEMU virtual address (userspace). */
 161     uint64_t qva;
 162     /* Starting offset in our mmaped space. */
 163     uint64_t mmap_offset;
 164     /* Start address of mmaped space. */
 165     uint64_t mmap_addr;
 166 } VuDevRegion;
 167
 168 typedef struct VuDev VuDev;
 169
 170 typedef uint64_t (*vu_get_features_cb) (VuDev *dev);
 171 typedef void (*vu_set_features_cb) (VuDev *dev, uint64_t features);
 172 typedef int (*vu_process_msg_cb) (VuDev *dev, VhostUserMsg *vmsg,
 173                                   int *do_reply);
 174 typedef void (*vu_queue_set_started_cb) (VuDev *dev, int qidx, bool started);
 175 typedef bool (*vu_queue_is_processed_in_order_cb) (VuDev *dev, int qidx);
 176 typedef int (*vu_get_config_cb) (VuDev *dev, uint8_t *config, uint32_t len);
 177 typedef int (*vu_set_config_cb) (VuDev *dev, const uint8_t *data,
 178                                  uint32_t offset, uint32_t size,
 179                                  uint32_t flags);
 180
 181 typedef struct VuDevIface {
 182     /* called by VHOST_USER_GET_FEATURES to get the features bitmask */
 183     vu_get_features_cb get_features;
 184     /* enable vhost implementation features */
 185     vu_set_features_cb set_features;
 186     /* get the protocol feature bitmask from the underlying vhost
 187      * implementation */
 188     vu_get_features_cb get_protocol_features;
 189     /* enable protocol features in the underlying vhost implementation. */
 190     vu_set_features_cb set_protocol_features;
 191     /* process_msg is called for each vhost-user message received */
 192     /* skip libvhost-user processing if return value != 0 */
 193     vu_process_msg_cb process_msg;
 194     /* tells when queues can be processed */
 195     vu_queue_set_started_cb queue_set_started;
 196     /*
 197      * If the queue is processed in order, in which case it will be
 198      * resumed to vring.used->idx. This can help to support resuming
 199      * on unmanaged exit/crash.
 200      */
 201     vu_queue_is_processed_in_order_cb queue_is_processed_in_order;
 202     /* get the config space of the device */
 203     vu_get_config_cb get_config;
 204     /* set the config space of the device */
 205     vu_set_config_cb set_config;
 206 } VuDevIface;
 207
 208 typedef void (*vu_queue_handler_cb) (VuDev *dev, int qidx);
 209
 210 typedef struct VuRing {
 211     unsigned int num;
 212     struct vring_desc *desc;
 213     struct vring_avail *avail;
 214     struct vring_used *used;
 215     uint64_t log_guest_addr;
 216     uint32_t flags;
 217 } VuRing;
 218
 219 typedef struct VuVirtq {
 220     VuRing vring;
 221
 222     /* Next head to pop */
 223     uint16_t last_avail_idx;
 224
 225     /* Last avail_idx read from VQ. */
 226     uint16_t shadow_avail_idx;
 227
 228     uint16_t used_idx;
 229
 230     /* Last used index value we have signalled on */
 231     uint16_t signalled_used;
 232
 233     /* Last used index value we have signalled on */
 234     bool signalled_used_valid;
 235
 236     /* Notification enabled? */
 237     bool notification;
 238
 239     int inuse;
 240
 241     vu_queue_handler_cb handler;
 242
 243     int call_fd;
 244     int kick_fd;
 245     int err_fd;
 246     unsigned int enable;
 247     bool started;
 248 } VuVirtq;
 249
 250 enum VuWatchCondtion {
 251     VU_WATCH_IN = POLLIN,
 252     VU_WATCH_OUT = POLLOUT,
 253     VU_WATCH_PRI = POLLPRI,
 254     VU_WATCH_ERR = POLLERR,
 255     VU_WATCH_HUP = POLLHUP,
 256 };
 257
 258 typedef void (*vu_panic_cb) (VuDev *dev, const char *err);
 259 typedef void (*vu_watch_cb) (VuDev *dev, int condition, void *data);
 260 typedef void (*vu_set_watch_cb) (VuDev *dev, int fd, int condition,
 261                                  vu_watch_cb cb, void *data);
 262 typedef void (*vu_remove_watch_cb) (VuDev *dev, int fd);
 263
 264 struct VuDev {
 265     int sock;
 266     uint32_t nregions;
 267     VuDevRegion regions[VHOST_MEMORY_MAX_NREGIONS];
 268     VuVirtq vq[VHOST_MAX_NR_VIRTQUEUE];
 269     int log_call_fd;
 270     int slave_fd;
 271     uint64_t log_size;
 272     uint8_t *log_table;
 273     uint64_t features;
 274     uint64_t protocol_features;
 275     bool broken;
 276
 277     /* @set_watch: add or update the given fd to the watch set,
 278      * call cb when condition is met */
 279     vu_set_watch_cb set_watch;
 280
 281     /* @remove_watch: remove the given fd from the watch set */
 282     vu_remove_watch_cb remove_watch;
 283
 284     /* @panic: encountered an unrecoverable error, you may try to
 285      * re-initialize */
 286     vu_panic_cb panic;
 287     const VuDevIface *iface;
 288
 289     /* Postcopy data */
 290     int postcopy_ufd;
 291     bool postcopy_listening;
 292 };
 293
 294 typedef struct VuVirtqElement {
 295     unsigned int index;
 296     unsigned int out_num;
 297     unsigned int in_num;
 298     struct iovec *in_sg;
 299     struct iovec *out_sg;
 300 } VuVirtqElement;
 301
 302 /**
 303  * vu_init:
 304  * @dev: a VuDev context
 305  * @socket: the socket connected to vhost-user master
 306  * @panic: a panic callback
 307  * @set_watch: a set_watch callback
 308  * @remove_watch: a remove_watch callback
 309  * @iface: a VuDevIface structure with vhost-user device callbacks
 310  *
 311  * Intializes a VuDev vhost-user context.
 312  **/
 313 void vu_init(VuDev *dev,
 314              int socket,
 315              vu_panic_cb panic,
 316              vu_set_watch_cb set_watch,
 317              vu_remove_watch_cb remove_watch,
 318              const VuDevIface *iface);
 319
 320
 321 /**
 322  * vu_deinit:
 323  * @dev: a VuDev context
 324  *
 325  * Cleans up the VuDev context
 326  */
 327 void vu_deinit(VuDev *dev);
 328
 329 /**
 330  * vu_dispatch:
 331  * @dev: a VuDev context
 332  *
 333  * Process one vhost-user message.
 334  *
 335  * Returns: TRUE on success, FALSE on failure.
 336  */
 337 bool vu_dispatch(VuDev *dev);
 338
 339 /**
 340  * vu_gpa_to_va:
 341  * @dev: a VuDev context
 342  * @plen: guest memory size
 343  * @guest_addr: guest address
 344  *
 345  * Translate a guest address to a pointer. Returns NULL on failure.
 346  */
 347 void *vu_gpa_to_va(VuDev *dev, uint64_t *plen, uint64_t guest_addr);
 348
 349 /**
 350  * vu_get_queue:
 351  * @dev: a VuDev context
 352  * @qidx: queue index
 353  *
 354  * Returns the queue number @qidx.
 355  */
 356 VuVirtq *vu_get_queue(VuDev *dev, int qidx);
 357
 358 /**
 359  * vu_set_queue_handler:
 360  * @dev: a VuDev context
 361  * @vq: a VuVirtq queue
 362  * @handler: the queue handler callback
 363  *
 364  * Set the queue handler. This function may be called several times
 365  * for the same queue. If called with NULL @handler, the handler is
 366  * removed.
 367  */
 368 void vu_set_queue_handler(VuDev *dev, VuVirtq *vq,
 369                           vu_queue_handler_cb handler);
 370
 371
 372 /**
 373  * vu_queue_set_notification:
 374  * @dev: a VuDev context
 375  * @vq: a VuVirtq queue
 376  * @enable: state
 377  *
 378  * Set whether the queue notifies (via event index or interrupt)
 379  */
 380 void vu_queue_set_notification(VuDev *dev, VuVirtq *vq, int enable);
 381
 382 /**
 383  * vu_queue_enabled:
 384  * @dev: a VuDev context
 385  * @vq: a VuVirtq queue
 386  *
 387  * Returns: whether the queue is enabled.
 388  */
 389 bool vu_queue_enabled(VuDev *dev, VuVirtq *vq);
 390
 391 /**
 392  * vu_queue_started:
 393  * @dev: a VuDev context
 394  * @vq: a VuVirtq queue
 395  *
 396  * Returns: whether the queue is started.
 397  */
 398 bool vu_queue_started(const VuDev *dev, const VuVirtq *vq);
 399
 400 /**
 401  * vu_queue_empty:
 402  * @dev: a VuDev context
 403  * @vq: a VuVirtq queue
 404  *
 405  * Returns: true if the queue is empty or not ready.
 406  */
 407 bool vu_queue_empty(VuDev *dev, VuVirtq *vq);
 408
 409 /**
 410  * vu_queue_notify:
 411  * @dev: a VuDev context
 412  * @vq: a VuVirtq queue
 413  *
 414  * Request to notify the queue via callfd (skipped if unnecessary)
 415  */
 416 void vu_queue_notify(VuDev *dev, VuVirtq *vq);
 417
 418 /**
 419  * vu_queue_pop:
 420  * @dev: a VuDev context
 421  * @vq: a VuVirtq queue
 422  * @sz: the size of struct to return (must be >= VuVirtqElement)
 423  *
 424  * Returns: a VuVirtqElement filled from the queue or NULL. The
 425  * returned element must be free()-d by the caller.
 426  */
 427 void *vu_queue_pop(VuDev *dev, VuVirtq *vq, size_t sz);
 428
 429 /**
 430  * vu_queue_rewind:
 431  * @dev: a VuDev context
 432  * @vq: a VuVirtq queue
 433  * @num: number of elements to push back
 434  *
 435  * Pretend that elements weren't popped from the virtqueue.  The next
 436  * virtqueue_pop() will refetch the oldest element.
 437  *
 438  * Returns: true on success, false if @num is greater than the number of in use
 439  * elements.
 440  */
 441 bool vu_queue_rewind(VuDev *dev, VuVirtq *vq, unsigned int num);
 442
 443 /**
 444  * vu_queue_fill:
 445  * @dev: a VuDev context
 446  * @vq: a VuVirtq queue
 447  * @elem: a VuVirtqElement
 448  * @len: length in bytes to write
 449  * @idx: optional offset for the used ring index (0 in general)
 450  *
 451  * Fill the used ring with @elem element.
 452  */
 453 void vu_queue_fill(VuDev *dev, VuVirtq *vq,
 454                    const VuVirtqElement *elem,
 455                    unsigned int len, unsigned int idx);
 456
 457 /**
 458  * vu_queue_push:
 459  * @dev: a VuDev context
 460  * @vq: a VuVirtq queue
 461  * @elem: a VuVirtqElement
 462  * @len: length in bytes to write
 463  *
 464  * Helper that combines vu_queue_fill() with a vu_queue_flush().
 465  */
 466 void vu_queue_push(VuDev *dev, VuVirtq *vq,
 467                    const VuVirtqElement *elem, unsigned int len);
 468
 469 /**
 470  * vu_queue_flush:
 471  * @dev: a VuDev context
 472  * @vq: a VuVirtq queue
 473  * @num: number of elements to flush
 474  *
 475  * Mark the last number of elements as done (used.idx is updated by
 476  * num elements).
 477 */
 478 void vu_queue_flush(VuDev *dev, VuVirtq *vq, unsigned int num);
 479
 480 /**
 481  * vu_queue_get_avail_bytes:
 482  * @dev: a VuDev context
 483  * @vq: a VuVirtq queue
 484  * @in_bytes: in bytes
 485  * @out_bytes: out bytes
 486  * @max_in_bytes: stop counting after max_in_bytes
 487  * @max_out_bytes: stop counting after max_out_bytes
 488  *
 489  * Count the number of available bytes, up to max_in_bytes/max_out_bytes.
 490  */
 491 void vu_queue_get_avail_bytes(VuDev *vdev, VuVirtq *vq, unsigned int *in_bytes,
 492                               unsigned int *out_bytes,
 493                               unsigned max_in_bytes, unsigned max_out_bytes);
 494
 495 /**
 496  * vu_queue_avail_bytes:
 497  * @dev: a VuDev context
 498  * @vq: a VuVirtq queue
 499  * @in_bytes: expected in bytes
 500  * @out_bytes: expected out bytes
 501  *
 502  * Returns: true if in_bytes <= in_total && out_bytes <= out_total
 503  */
 504 bool vu_queue_avail_bytes(VuDev *dev, VuVirtq *vq, unsigned int in_bytes,
 505                           unsigned int out_bytes);
 506
 507 #endif /* LIBVHOST_USER_H */