4 * Copyright (c) 2016 Red Hat, Inc.
7 * Victor Kaplansky <victork@redhat.com>
8 * Marc-André Lureau <mlureau@redhat.com>
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.
14 #ifndef LIBVHOST_USER_H
15 #define LIBVHOST_USER_H
21 #include <linux/vhost.h>
23 #include "standard-headers/linux/virtio_ring.h"
25 /* Based on qemu/hw/virtio/vhost-user.c */
26 #define VHOST_USER_F_PROTOCOL_FEATURES 30
27 #define VHOST_LOG_PAGE 4096
29 #define VIRTQUEUE_MAX_SIZE 1024
31 #define VHOST_MEMORY_BASELINE_NREGIONS 8
34 * Set a reasonable maximum number of ram slots, which will be supported by
37 #define VHOST_USER_MAX_RAM_SLOTS 32
39 #define VHOST_USER_HDR_SIZE offsetof(VhostUserMsg, payload.u64)
41 typedef enum VhostSetConfigType
{
42 VHOST_SET_CONFIG_TYPE_MASTER
= 0,
43 VHOST_SET_CONFIG_TYPE_MIGRATION
= 1,
47 * Maximum size of virtio device config space
49 #define VHOST_USER_MAX_CONFIG_SIZE 256
51 enum VhostUserProtocolFeature
{
52 VHOST_USER_PROTOCOL_F_MQ
= 0,
53 VHOST_USER_PROTOCOL_F_LOG_SHMFD
= 1,
54 VHOST_USER_PROTOCOL_F_RARP
= 2,
55 VHOST_USER_PROTOCOL_F_REPLY_ACK
= 3,
56 VHOST_USER_PROTOCOL_F_NET_MTU
= 4,
57 VHOST_USER_PROTOCOL_F_SLAVE_REQ
= 5,
58 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
= 6,
59 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
= 7,
60 VHOST_USER_PROTOCOL_F_PAGEFAULT
= 8,
61 VHOST_USER_PROTOCOL_F_CONFIG
= 9,
62 VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD
= 10,
63 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
= 11,
64 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
= 12,
65 VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS
= 14,
66 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
= 15,
68 VHOST_USER_PROTOCOL_F_MAX
71 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
73 typedef enum VhostUserRequest
{
75 VHOST_USER_GET_FEATURES
= 1,
76 VHOST_USER_SET_FEATURES
= 2,
77 VHOST_USER_SET_OWNER
= 3,
78 VHOST_USER_RESET_OWNER
= 4,
79 VHOST_USER_SET_MEM_TABLE
= 5,
80 VHOST_USER_SET_LOG_BASE
= 6,
81 VHOST_USER_SET_LOG_FD
= 7,
82 VHOST_USER_SET_VRING_NUM
= 8,
83 VHOST_USER_SET_VRING_ADDR
= 9,
84 VHOST_USER_SET_VRING_BASE
= 10,
85 VHOST_USER_GET_VRING_BASE
= 11,
86 VHOST_USER_SET_VRING_KICK
= 12,
87 VHOST_USER_SET_VRING_CALL
= 13,
88 VHOST_USER_SET_VRING_ERR
= 14,
89 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
90 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
91 VHOST_USER_GET_QUEUE_NUM
= 17,
92 VHOST_USER_SET_VRING_ENABLE
= 18,
93 VHOST_USER_SEND_RARP
= 19,
94 VHOST_USER_NET_SET_MTU
= 20,
95 VHOST_USER_SET_SLAVE_REQ_FD
= 21,
96 VHOST_USER_IOTLB_MSG
= 22,
97 VHOST_USER_SET_VRING_ENDIAN
= 23,
98 VHOST_USER_GET_CONFIG
= 24,
99 VHOST_USER_SET_CONFIG
= 25,
100 VHOST_USER_CREATE_CRYPTO_SESSION
= 26,
101 VHOST_USER_CLOSE_CRYPTO_SESSION
= 27,
102 VHOST_USER_POSTCOPY_ADVISE
= 28,
103 VHOST_USER_POSTCOPY_LISTEN
= 29,
104 VHOST_USER_POSTCOPY_END
= 30,
105 VHOST_USER_GET_INFLIGHT_FD
= 31,
106 VHOST_USER_SET_INFLIGHT_FD
= 32,
107 VHOST_USER_GPU_SET_SOCKET
= 33,
108 VHOST_USER_VRING_KICK
= 35,
109 VHOST_USER_GET_MAX_MEM_SLOTS
= 36,
110 VHOST_USER_ADD_MEM_REG
= 37,
111 VHOST_USER_REM_MEM_REG
= 38,
115 typedef enum VhostUserSlaveRequest
{
116 VHOST_USER_SLAVE_NONE
= 0,
117 VHOST_USER_SLAVE_IOTLB_MSG
= 1,
118 VHOST_USER_SLAVE_CONFIG_CHANGE_MSG
= 2,
119 VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG
= 3,
120 VHOST_USER_SLAVE_VRING_CALL
= 4,
121 VHOST_USER_SLAVE_VRING_ERR
= 5,
123 } VhostUserSlaveRequest
;
125 typedef struct VhostUserMemoryRegion
{
126 uint64_t guest_phys_addr
;
127 uint64_t memory_size
;
128 uint64_t userspace_addr
;
129 uint64_t mmap_offset
;
130 } VhostUserMemoryRegion
;
132 typedef struct VhostUserMemory
{
135 VhostUserMemoryRegion regions
[VHOST_MEMORY_BASELINE_NREGIONS
];
138 typedef struct VhostUserMemRegMsg
{
140 VhostUserMemoryRegion region
;
141 } VhostUserMemRegMsg
;
143 typedef struct VhostUserLog
{
145 uint64_t mmap_offset
;
148 typedef struct VhostUserConfig
{
152 uint8_t region
[VHOST_USER_MAX_CONFIG_SIZE
];
155 static VhostUserConfig c
__attribute__ ((unused
));
156 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
160 typedef struct VhostUserVringArea
{
164 } VhostUserVringArea
;
166 typedef struct VhostUserInflight
{
168 uint64_t mmap_offset
;
173 #if defined(_WIN32) && (defined(__x86_64__) || defined(__i386__))
174 # define VU_PACKED __attribute__((gcc_struct, packed))
176 # define VU_PACKED __attribute__((packed))
179 typedef struct VhostUserMsg
{
182 #define VHOST_USER_VERSION_MASK (0x3)
183 #define VHOST_USER_REPLY_MASK (0x1 << 2)
184 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
186 uint32_t size
; /* the following payload size */
189 #define VHOST_USER_VRING_IDX_MASK (0xff)
190 #define VHOST_USER_VRING_NOFD_MASK (0x1 << 8)
192 struct vhost_vring_state state
;
193 struct vhost_vring_addr addr
;
194 VhostUserMemory memory
;
195 VhostUserMemRegMsg memreg
;
197 VhostUserConfig config
;
198 VhostUserVringArea area
;
199 VhostUserInflight inflight
;
202 int fds
[VHOST_MEMORY_BASELINE_NREGIONS
];
205 } VU_PACKED VhostUserMsg
;
207 typedef struct VuDevRegion
{
208 /* Guest Physical address. */
210 /* Memory region size. */
212 /* QEMU virtual address (userspace). */
214 /* Starting offset in our mmaped space. */
215 uint64_t mmap_offset
;
216 /* Start address of mmaped space. */
220 typedef struct VuDev VuDev
;
222 typedef uint64_t (*vu_get_features_cb
) (VuDev
*dev
);
223 typedef void (*vu_set_features_cb
) (VuDev
*dev
, uint64_t features
);
224 typedef int (*vu_process_msg_cb
) (VuDev
*dev
, VhostUserMsg
*vmsg
,
226 typedef bool (*vu_read_msg_cb
) (VuDev
*dev
, int sock
, VhostUserMsg
*vmsg
);
227 typedef void (*vu_queue_set_started_cb
) (VuDev
*dev
, int qidx
, bool started
);
228 typedef bool (*vu_queue_is_processed_in_order_cb
) (VuDev
*dev
, int qidx
);
229 typedef int (*vu_get_config_cb
) (VuDev
*dev
, uint8_t *config
, uint32_t len
);
230 typedef int (*vu_set_config_cb
) (VuDev
*dev
, const uint8_t *data
,
231 uint32_t offset
, uint32_t size
,
234 typedef struct VuDevIface
{
235 /* called by VHOST_USER_GET_FEATURES to get the features bitmask */
236 vu_get_features_cb get_features
;
237 /* enable vhost implementation features */
238 vu_set_features_cb set_features
;
239 /* get the protocol feature bitmask from the underlying vhost
241 vu_get_features_cb get_protocol_features
;
242 /* enable protocol features in the underlying vhost implementation. */
243 vu_set_features_cb set_protocol_features
;
244 /* process_msg is called for each vhost-user message received */
245 /* skip libvhost-user processing if return value != 0 */
246 vu_process_msg_cb process_msg
;
247 /* tells when queues can be processed */
248 vu_queue_set_started_cb queue_set_started
;
250 * If the queue is processed in order, in which case it will be
251 * resumed to vring.used->idx. This can help to support resuming
252 * on unmanaged exit/crash.
254 vu_queue_is_processed_in_order_cb queue_is_processed_in_order
;
255 /* get the config space of the device */
256 vu_get_config_cb get_config
;
257 /* set the config space of the device */
258 vu_set_config_cb set_config
;
261 typedef void (*vu_queue_handler_cb
) (VuDev
*dev
, int qidx
);
263 typedef struct VuRing
{
265 struct vring_desc
*desc
;
266 struct vring_avail
*avail
;
267 struct vring_used
*used
;
268 uint64_t log_guest_addr
;
272 typedef struct VuDescStateSplit
{
273 /* Indicate whether this descriptor is inflight or not.
274 * Only available for head-descriptor. */
280 /* Maintain a list for the last batch of used descriptors.
281 * Only available when batching is used for submitting */
284 /* Used to preserve the order of fetching available descriptors.
285 * Only available for head-descriptor. */
289 typedef struct VuVirtqInflight
{
290 /* The feature flags of this region. Now it's initialized to 0. */
293 /* The version of this region. It's 1 currently.
294 * Zero value indicates a vm reset happened. */
297 /* The size of VuDescStateSplit array. It's equal to the virtqueue
298 * size. Slave could get it from queue size field of VhostUserInflight. */
301 /* The head of list that track the last batch of used descriptors. */
302 uint16_t last_batch_head
;
304 /* Storing the idx value of used ring */
307 /* Used to track the state of each descriptor in descriptor table */
308 VuDescStateSplit desc
[];
311 typedef struct VuVirtqInflightDesc
{
314 } VuVirtqInflightDesc
;
316 typedef struct VuVirtq
{
319 VuVirtqInflight
*inflight
;
321 VuVirtqInflightDesc
*resubmit_list
;
323 uint16_t resubmit_num
;
327 /* Next head to pop */
328 uint16_t last_avail_idx
;
330 /* Last avail_idx read from VQ. */
331 uint16_t shadow_avail_idx
;
335 /* Last used index value we have signalled on */
336 uint16_t signalled_used
;
338 /* Last used index value we have signalled on */
339 bool signalled_used_valid
;
341 /* Notification enabled? */
346 vu_queue_handler_cb handler
;
354 /* Guest addresses of our ring */
355 struct vhost_vring_addr vra
;
358 enum VuWatchCondtion
{
359 VU_WATCH_IN
= POLLIN
,
360 VU_WATCH_OUT
= POLLOUT
,
361 VU_WATCH_PRI
= POLLPRI
,
362 VU_WATCH_ERR
= POLLERR
,
363 VU_WATCH_HUP
= POLLHUP
,
366 typedef void (*vu_panic_cb
) (VuDev
*dev
, const char *err
);
367 typedef void (*vu_watch_cb
) (VuDev
*dev
, int condition
, void *data
);
368 typedef void (*vu_set_watch_cb
) (VuDev
*dev
, int fd
, int condition
,
369 vu_watch_cb cb
, void *data
);
370 typedef void (*vu_remove_watch_cb
) (VuDev
*dev
, int fd
);
372 typedef struct VuDevInflightInfo
{
381 VuDevRegion regions
[VHOST_USER_MAX_RAM_SLOTS
];
383 VuDevInflightInfo inflight_info
;
385 /* Must be held while using slave_fd */
386 pthread_mutex_t slave_mutex
;
391 uint64_t protocol_features
;
396 * @read_msg: custom method to read vhost-user message
398 * Read data from vhost_user socket fd and fill up
399 * the passed VhostUserMsg *vmsg struct.
401 * If reading fails, it should close the received set of file
402 * descriptors as socket message's auxiliary data.
404 * For the details, please refer to vu_message_read in libvhost-user.c
405 * which will be used by default if not custom method is provided when
408 * Returns: true if vhost-user message successfully received,
409 * otherwise return false.
412 vu_read_msg_cb read_msg
;
415 * @set_watch: add or update the given fd to the watch set,
416 * call cb when condition is met.
418 vu_set_watch_cb set_watch
;
420 /* @remove_watch: remove the given fd from the watch set */
421 vu_remove_watch_cb remove_watch
;
424 * @panic: encountered an unrecoverable error, you may try to re-initialize
427 const VuDevIface
*iface
;
431 bool postcopy_listening
;
434 typedef struct VuVirtqElement
{
436 unsigned int out_num
;
439 struct iovec
*out_sg
;
444 * @dev: a VuDev context
445 * @max_queues: maximum number of virtqueues
446 * @socket: the socket connected to vhost-user master
447 * @panic: a panic callback
448 * @set_watch: a set_watch callback
449 * @remove_watch: a remove_watch callback
450 * @iface: a VuDevIface structure with vhost-user device callbacks
452 * Initializes a VuDev vhost-user context.
454 * Returns: true on success, false on failure.
456 bool vu_init(VuDev
*dev
,
460 vu_read_msg_cb read_msg
,
461 vu_set_watch_cb set_watch
,
462 vu_remove_watch_cb remove_watch
,
463 const VuDevIface
*iface
);
468 * @dev: a VuDev context
470 * Cleans up the VuDev context
472 void vu_deinit(VuDev
*dev
);
476 * @dev: a VuDev context
478 * Process one vhost-user message.
480 * Returns: TRUE on success, FALSE on failure.
482 bool vu_dispatch(VuDev
*dev
);
486 * @dev: a VuDev context
487 * @plen: guest memory size
488 * @guest_addr: guest address
490 * Translate a guest address to a pointer. Returns NULL on failure.
492 void *vu_gpa_to_va(VuDev
*dev
, uint64_t *plen
, uint64_t guest_addr
);
496 * @dev: a VuDev context
499 * Returns the queue number @qidx.
501 VuVirtq
*vu_get_queue(VuDev
*dev
, int qidx
);
504 * vu_set_queue_handler:
505 * @dev: a VuDev context
506 * @vq: a VuVirtq queue
507 * @handler: the queue handler callback
509 * Set the queue handler. This function may be called several times
510 * for the same queue. If called with NULL @handler, the handler is
513 void vu_set_queue_handler(VuDev
*dev
, VuVirtq
*vq
,
514 vu_queue_handler_cb handler
);
517 * vu_set_queue_host_notifier:
518 * @dev: a VuDev context
519 * @vq: a VuVirtq queue
520 * @fd: a file descriptor
521 * @size: host page size
522 * @offset: notifier offset in @fd file
524 * Set queue's host notifier. This function may be called several
525 * times for the same queue. If called with -1 @fd, the notifier
528 bool vu_set_queue_host_notifier(VuDev
*dev
, VuVirtq
*vq
, int fd
,
529 int size
, int offset
);
532 * vu_queue_set_notification:
533 * @dev: a VuDev context
534 * @vq: a VuVirtq queue
537 * Set whether the queue notifies (via event index or interrupt)
539 void vu_queue_set_notification(VuDev
*dev
, VuVirtq
*vq
, int enable
);
543 * @dev: a VuDev context
544 * @vq: a VuVirtq queue
546 * Returns: whether the queue is enabled.
548 bool vu_queue_enabled(VuDev
*dev
, VuVirtq
*vq
);
552 * @dev: a VuDev context
553 * @vq: a VuVirtq queue
555 * Returns: whether the queue is started.
557 bool vu_queue_started(const VuDev
*dev
, const VuVirtq
*vq
);
561 * @dev: a VuDev context
562 * @vq: a VuVirtq queue
564 * Returns: true if the queue is empty or not ready.
566 bool vu_queue_empty(VuDev
*dev
, VuVirtq
*vq
);
570 * @dev: a VuDev context
571 * @vq: a VuVirtq queue
573 * Request to notify the queue via callfd (skipped if unnecessary)
575 void vu_queue_notify(VuDev
*dev
, VuVirtq
*vq
);
578 * vu_queue_notify_sync:
579 * @dev: a VuDev context
580 * @vq: a VuVirtq queue
582 * Request to notify the queue via callfd (skipped if unnecessary)
583 * or sync message if possible.
585 void vu_queue_notify_sync(VuDev
*dev
, VuVirtq
*vq
);
589 * @dev: a VuDev context
590 * @vq: a VuVirtq queue
591 * @sz: the size of struct to return (must be >= VuVirtqElement)
593 * Returns: a VuVirtqElement filled from the queue or NULL. The
594 * returned element must be free()-d by the caller.
596 void *vu_queue_pop(VuDev
*dev
, VuVirtq
*vq
, size_t sz
);
601 * @dev: a VuDev context
602 * @vq: a VuVirtq queue
603 * @elem: The #VuVirtqElement
604 * @len: number of bytes written
606 * Pretend the most recent element wasn't popped from the virtqueue. The next
607 * call to vu_queue_pop() will refetch the element.
609 void vu_queue_unpop(VuDev
*dev
, VuVirtq
*vq
, VuVirtqElement
*elem
,
614 * @dev: a VuDev context
615 * @vq: a VuVirtq queue
616 * @num: number of elements to push back
618 * Pretend that elements weren't popped from the virtqueue. The next
619 * virtqueue_pop() will refetch the oldest element.
621 * Returns: true on success, false if @num is greater than the number of in use
624 bool vu_queue_rewind(VuDev
*dev
, VuVirtq
*vq
, unsigned int num
);
628 * @dev: a VuDev context
629 * @vq: a VuVirtq queue
630 * @elem: a VuVirtqElement
631 * @len: length in bytes to write
632 * @idx: optional offset for the used ring index (0 in general)
634 * Fill the used ring with @elem element.
636 void vu_queue_fill(VuDev
*dev
, VuVirtq
*vq
,
637 const VuVirtqElement
*elem
,
638 unsigned int len
, unsigned int idx
);
642 * @dev: a VuDev context
643 * @vq: a VuVirtq queue
644 * @elem: a VuVirtqElement
645 * @len: length in bytes to write
647 * Helper that combines vu_queue_fill() with a vu_queue_flush().
649 void vu_queue_push(VuDev
*dev
, VuVirtq
*vq
,
650 const VuVirtqElement
*elem
, unsigned int len
);
654 * @dev: a VuDev context
655 * @vq: a VuVirtq queue
656 * @num: number of elements to flush
658 * Mark the last number of elements as done (used.idx is updated by
661 void vu_queue_flush(VuDev
*dev
, VuVirtq
*vq
, unsigned int num
);
664 * vu_queue_get_avail_bytes:
665 * @dev: a VuDev context
666 * @vq: a VuVirtq queue
667 * @in_bytes: in bytes
668 * @out_bytes: out bytes
669 * @max_in_bytes: stop counting after max_in_bytes
670 * @max_out_bytes: stop counting after max_out_bytes
672 * Count the number of available bytes, up to max_in_bytes/max_out_bytes.
674 void vu_queue_get_avail_bytes(VuDev
*vdev
, VuVirtq
*vq
, unsigned int *in_bytes
,
675 unsigned int *out_bytes
,
676 unsigned max_in_bytes
, unsigned max_out_bytes
);
679 * vu_queue_avail_bytes:
680 * @dev: a VuDev context
681 * @vq: a VuVirtq queue
682 * @in_bytes: expected in bytes
683 * @out_bytes: expected out bytes
685 * Returns: true if in_bytes <= in_total && out_bytes <= out_total
687 bool vu_queue_avail_bytes(VuDev
*dev
, VuVirtq
*vq
, unsigned int in_bytes
,
688 unsigned int out_bytes
);
690 #endif /* LIBVHOST_USER_H */