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_BACKEND_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_BACKEND_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_BACKEND_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_BACKEND_NONE
= 0,
117 VHOST_USER_BACKEND_IOTLB_MSG
= 1,
118 VHOST_USER_BACKEND_CONFIG_CHANGE_MSG
= 2,
119 VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG
= 3,
120 VHOST_USER_BACKEND_VRING_CALL
= 4,
121 VHOST_USER_BACKEND_VRING_ERR
= 5,
122 VHOST_USER_BACKEND_MAX
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 #define VHOST_USER_MEM_REG_SIZE (sizeof(VhostUserMemoryRegion))
134 typedef struct VhostUserMemory
{
137 VhostUserMemoryRegion regions
[VHOST_MEMORY_BASELINE_NREGIONS
];
140 typedef struct VhostUserMemRegMsg
{
142 VhostUserMemoryRegion region
;
143 } VhostUserMemRegMsg
;
145 typedef struct VhostUserLog
{
147 uint64_t mmap_offset
;
150 typedef struct VhostUserConfig
{
154 uint8_t region
[VHOST_USER_MAX_CONFIG_SIZE
];
157 static VhostUserConfig c
__attribute__ ((unused
));
158 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
162 typedef struct VhostUserVringArea
{
166 } VhostUserVringArea
;
168 typedef struct VhostUserInflight
{
170 uint64_t mmap_offset
;
175 #if defined(_WIN32) && (defined(__x86_64__) || defined(__i386__))
176 # define VU_PACKED __attribute__((gcc_struct, packed))
178 # define VU_PACKED __attribute__((packed))
181 typedef struct VhostUserMsg
{
184 #define VHOST_USER_VERSION_MASK (0x3)
185 #define VHOST_USER_REPLY_MASK (0x1 << 2)
186 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
188 uint32_t size
; /* the following payload size */
191 #define VHOST_USER_VRING_IDX_MASK (0xff)
192 #define VHOST_USER_VRING_NOFD_MASK (0x1 << 8)
194 struct vhost_vring_state state
;
195 struct vhost_vring_addr addr
;
196 VhostUserMemory memory
;
197 VhostUserMemRegMsg memreg
;
199 VhostUserConfig config
;
200 VhostUserVringArea area
;
201 VhostUserInflight inflight
;
204 int fds
[VHOST_MEMORY_BASELINE_NREGIONS
];
207 } VU_PACKED VhostUserMsg
;
209 typedef struct VuDevRegion
{
210 /* Guest Physical address. */
212 /* Memory region size. */
214 /* QEMU virtual address (userspace). */
216 /* Starting offset in our mmaped space. */
217 uint64_t mmap_offset
;
218 /* Start address of mmaped space. */
222 typedef struct VuDev VuDev
;
224 typedef uint64_t (*vu_get_features_cb
) (VuDev
*dev
);
225 typedef void (*vu_set_features_cb
) (VuDev
*dev
, uint64_t features
);
226 typedef int (*vu_process_msg_cb
) (VuDev
*dev
, VhostUserMsg
*vmsg
,
228 typedef bool (*vu_read_msg_cb
) (VuDev
*dev
, int sock
, VhostUserMsg
*vmsg
);
229 typedef void (*vu_queue_set_started_cb
) (VuDev
*dev
, int qidx
, bool started
);
230 typedef bool (*vu_queue_is_processed_in_order_cb
) (VuDev
*dev
, int qidx
);
231 typedef int (*vu_get_config_cb
) (VuDev
*dev
, uint8_t *config
, uint32_t len
);
232 typedef int (*vu_set_config_cb
) (VuDev
*dev
, const uint8_t *data
,
233 uint32_t offset
, uint32_t size
,
236 typedef struct VuDevIface
{
237 /* called by VHOST_USER_GET_FEATURES to get the features bitmask */
238 vu_get_features_cb get_features
;
239 /* enable vhost implementation features */
240 vu_set_features_cb set_features
;
241 /* get the protocol feature bitmask from the underlying vhost
243 vu_get_features_cb get_protocol_features
;
244 /* enable protocol features in the underlying vhost implementation. */
245 vu_set_features_cb set_protocol_features
;
246 /* process_msg is called for each vhost-user message received */
247 /* skip libvhost-user processing if return value != 0 */
248 vu_process_msg_cb process_msg
;
249 /* tells when queues can be processed */
250 vu_queue_set_started_cb queue_set_started
;
252 * If the queue is processed in order, in which case it will be
253 * resumed to vring.used->idx. This can help to support resuming
254 * on unmanaged exit/crash.
256 vu_queue_is_processed_in_order_cb queue_is_processed_in_order
;
257 /* get the config space of the device */
258 vu_get_config_cb get_config
;
259 /* set the config space of the device */
260 vu_set_config_cb set_config
;
263 typedef void (*vu_queue_handler_cb
) (VuDev
*dev
, int qidx
);
265 typedef struct VuRing
{
267 struct vring_desc
*desc
;
268 struct vring_avail
*avail
;
269 struct vring_used
*used
;
270 uint64_t log_guest_addr
;
274 typedef struct VuDescStateSplit
{
275 /* Indicate whether this descriptor is inflight or not.
276 * Only available for head-descriptor. */
282 /* Maintain a list for the last batch of used descriptors.
283 * Only available when batching is used for submitting */
286 /* Used to preserve the order of fetching available descriptors.
287 * Only available for head-descriptor. */
291 typedef struct VuVirtqInflight
{
292 /* The feature flags of this region. Now it's initialized to 0. */
295 /* The version of this region. It's 1 currently.
296 * Zero value indicates a vm reset happened. */
299 /* The size of VuDescStateSplit array. It's equal to the virtqueue
300 * size. Slave could get it from queue size field of VhostUserInflight. */
303 /* The head of list that track the last batch of used descriptors. */
304 uint16_t last_batch_head
;
306 /* Storing the idx value of used ring */
309 /* Used to track the state of each descriptor in descriptor table */
310 VuDescStateSplit desc
[];
313 typedef struct VuVirtqInflightDesc
{
316 } VuVirtqInflightDesc
;
318 typedef struct VuVirtq
{
321 VuVirtqInflight
*inflight
;
323 VuVirtqInflightDesc
*resubmit_list
;
325 uint16_t resubmit_num
;
329 /* Next head to pop */
330 uint16_t last_avail_idx
;
332 /* Last avail_idx read from VQ. */
333 uint16_t shadow_avail_idx
;
337 /* Last used index value we have signalled on */
338 uint16_t signalled_used
;
340 /* Last used index value we have signalled on */
341 bool signalled_used_valid
;
343 /* Notification enabled? */
348 vu_queue_handler_cb handler
;
356 /* Guest addresses of our ring */
357 struct vhost_vring_addr vra
;
360 enum VuWatchCondtion
{
361 VU_WATCH_IN
= POLLIN
,
362 VU_WATCH_OUT
= POLLOUT
,
363 VU_WATCH_PRI
= POLLPRI
,
364 VU_WATCH_ERR
= POLLERR
,
365 VU_WATCH_HUP
= POLLHUP
,
368 typedef void (*vu_panic_cb
) (VuDev
*dev
, const char *err
);
369 typedef void (*vu_watch_cb
) (VuDev
*dev
, int condition
, void *data
);
370 typedef void (*vu_set_watch_cb
) (VuDev
*dev
, int fd
, int condition
,
371 vu_watch_cb cb
, void *data
);
372 typedef void (*vu_remove_watch_cb
) (VuDev
*dev
, int fd
);
374 typedef struct VuDevInflightInfo
{
383 VuDevRegion regions
[VHOST_USER_MAX_RAM_SLOTS
];
385 VuDevInflightInfo inflight_info
;
387 /* Must be held while using slave_fd */
388 pthread_mutex_t slave_mutex
;
393 uint64_t protocol_features
;
398 * @read_msg: custom method to read vhost-user message
400 * Read data from vhost_user socket fd and fill up
401 * the passed VhostUserMsg *vmsg struct.
403 * If reading fails, it should close the received set of file
404 * descriptors as socket message's auxiliary data.
406 * For the details, please refer to vu_message_read in libvhost-user.c
407 * which will be used by default if not custom method is provided when
410 * Returns: true if vhost-user message successfully received,
411 * otherwise return false.
414 vu_read_msg_cb read_msg
;
417 * @set_watch: add or update the given fd to the watch set,
418 * call cb when condition is met.
420 vu_set_watch_cb set_watch
;
422 /* @remove_watch: remove the given fd from the watch set */
423 vu_remove_watch_cb remove_watch
;
426 * @panic: encountered an unrecoverable error, you may try to re-initialize
429 const VuDevIface
*iface
;
433 bool postcopy_listening
;
436 typedef struct VuVirtqElement
{
438 unsigned int out_num
;
441 struct iovec
*out_sg
;
446 * @dev: a VuDev context
447 * @max_queues: maximum number of virtqueues
448 * @socket: the socket connected to vhost-user master
449 * @panic: a panic callback
450 * @set_watch: a set_watch callback
451 * @remove_watch: a remove_watch callback
452 * @iface: a VuDevIface structure with vhost-user device callbacks
454 * Initializes a VuDev vhost-user context.
456 * Returns: true on success, false on failure.
458 bool vu_init(VuDev
*dev
,
462 vu_read_msg_cb read_msg
,
463 vu_set_watch_cb set_watch
,
464 vu_remove_watch_cb remove_watch
,
465 const VuDevIface
*iface
);
470 * @dev: a VuDev context
472 * Cleans up the VuDev context
474 void vu_deinit(VuDev
*dev
);
478 * vu_request_to_string: return string for vhost message request
479 * @req: VhostUserMsg request
481 * Returns a const string, do not free.
483 const char *vu_request_to_string(unsigned int req
);
487 * @dev: a VuDev context
489 * Process one vhost-user message.
491 * Returns: TRUE on success, FALSE on failure.
493 bool vu_dispatch(VuDev
*dev
);
497 * @dev: a VuDev context
498 * @plen: guest memory size
499 * @guest_addr: guest address
501 * Translate a guest address to a pointer. Returns NULL on failure.
503 void *vu_gpa_to_va(VuDev
*dev
, uint64_t *plen
, uint64_t guest_addr
);
507 * @dev: a VuDev context
510 * Returns the queue number @qidx.
512 VuVirtq
*vu_get_queue(VuDev
*dev
, int qidx
);
515 * vu_set_queue_handler:
516 * @dev: a VuDev context
517 * @vq: a VuVirtq queue
518 * @handler: the queue handler callback
520 * Set the queue handler. This function may be called several times
521 * for the same queue. If called with NULL @handler, the handler is
524 void vu_set_queue_handler(VuDev
*dev
, VuVirtq
*vq
,
525 vu_queue_handler_cb handler
);
528 * vu_set_queue_host_notifier:
529 * @dev: a VuDev context
530 * @vq: a VuVirtq queue
531 * @fd: a file descriptor
532 * @size: host page size
533 * @offset: notifier offset in @fd file
535 * Set queue's host notifier. This function may be called several
536 * times for the same queue. If called with -1 @fd, the notifier
539 bool vu_set_queue_host_notifier(VuDev
*dev
, VuVirtq
*vq
, int fd
,
540 int size
, int offset
);
543 * vu_queue_set_notification:
544 * @dev: a VuDev context
545 * @vq: a VuVirtq queue
548 * Set whether the queue notifies (via event index or interrupt)
550 void vu_queue_set_notification(VuDev
*dev
, VuVirtq
*vq
, int enable
);
554 * @dev: a VuDev context
555 * @vq: a VuVirtq queue
557 * Returns: whether the queue is enabled.
559 bool vu_queue_enabled(VuDev
*dev
, VuVirtq
*vq
);
563 * @dev: a VuDev context
564 * @vq: a VuVirtq queue
566 * Returns: whether the queue is started.
568 bool vu_queue_started(const VuDev
*dev
, const VuVirtq
*vq
);
572 * @dev: a VuDev context
573 * @vq: a VuVirtq queue
575 * Returns: true if the queue is empty or not ready.
577 bool vu_queue_empty(VuDev
*dev
, VuVirtq
*vq
);
581 * @dev: a VuDev context
582 * @vq: a VuVirtq queue
584 * Request to notify the queue via callfd (skipped if unnecessary)
586 void vu_queue_notify(VuDev
*dev
, VuVirtq
*vq
);
589 * vu_queue_notify_sync:
590 * @dev: a VuDev context
591 * @vq: a VuVirtq queue
593 * Request to notify the queue via callfd (skipped if unnecessary)
594 * or sync message if possible.
596 void vu_queue_notify_sync(VuDev
*dev
, VuVirtq
*vq
);
600 * @dev: a VuDev context
601 * @vq: a VuVirtq queue
602 * @sz: the size of struct to return (must be >= VuVirtqElement)
604 * Returns: a VuVirtqElement filled from the queue or NULL. The
605 * returned element must be free()-d by the caller.
607 void *vu_queue_pop(VuDev
*dev
, VuVirtq
*vq
, size_t sz
);
612 * @dev: a VuDev context
613 * @vq: a VuVirtq queue
614 * @elem: The #VuVirtqElement
615 * @len: number of bytes written
617 * Pretend the most recent element wasn't popped from the virtqueue. The next
618 * call to vu_queue_pop() will refetch the element.
620 void vu_queue_unpop(VuDev
*dev
, VuVirtq
*vq
, VuVirtqElement
*elem
,
625 * @dev: a VuDev context
626 * @vq: a VuVirtq queue
627 * @num: number of elements to push back
629 * Pretend that elements weren't popped from the virtqueue. The next
630 * virtqueue_pop() will refetch the oldest element.
632 * Returns: true on success, false if @num is greater than the number of in use
635 bool vu_queue_rewind(VuDev
*dev
, VuVirtq
*vq
, unsigned int num
);
639 * @dev: a VuDev context
640 * @vq: a VuVirtq queue
641 * @elem: a VuVirtqElement
642 * @len: length in bytes to write
643 * @idx: optional offset for the used ring index (0 in general)
645 * Fill the used ring with @elem element.
647 void vu_queue_fill(VuDev
*dev
, VuVirtq
*vq
,
648 const VuVirtqElement
*elem
,
649 unsigned int len
, unsigned int idx
);
653 * @dev: a VuDev context
654 * @vq: a VuVirtq queue
655 * @elem: a VuVirtqElement
656 * @len: length in bytes to write
658 * Helper that combines vu_queue_fill() with a vu_queue_flush().
660 void vu_queue_push(VuDev
*dev
, VuVirtq
*vq
,
661 const VuVirtqElement
*elem
, unsigned int len
);
665 * @dev: a VuDev context
666 * @vq: a VuVirtq queue
667 * @num: number of elements to flush
669 * Mark the last number of elements as done (used.idx is updated by
672 void vu_queue_flush(VuDev
*dev
, VuVirtq
*vq
, unsigned int num
);
675 * vu_queue_get_avail_bytes:
676 * @dev: a VuDev context
677 * @vq: a VuVirtq queue
678 * @in_bytes: in bytes
679 * @out_bytes: out bytes
680 * @max_in_bytes: stop counting after max_in_bytes
681 * @max_out_bytes: stop counting after max_out_bytes
683 * Count the number of available bytes, up to max_in_bytes/max_out_bytes.
685 void vu_queue_get_avail_bytes(VuDev
*vdev
, VuVirtq
*vq
, unsigned int *in_bytes
,
686 unsigned int *out_bytes
,
687 unsigned max_in_bytes
, unsigned max_out_bytes
);
690 * vu_queue_avail_bytes:
691 * @dev: a VuDev context
692 * @vq: a VuVirtq queue
693 * @in_bytes: expected in bytes
694 * @out_bytes: expected out bytes
696 * Returns: true if in_bytes <= in_total && out_bytes <= out_total
698 bool vu_queue_avail_bytes(VuDev
*dev
, VuVirtq
*vq
, unsigned int in_bytes
,
699 unsigned int out_bytes
);
701 #endif /* LIBVHOST_USER_H */