4 * Copyright (c) 2013 Virtual Open Systems Sarl.
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.
11 #include "qemu/osdep.h"
12 #include "qapi/error.h"
13 #include "hw/virtio/virtio-dmabuf.h"
14 #include "hw/virtio/vhost.h"
15 #include "hw/virtio/virtio-crypto.h"
16 #include "hw/virtio/vhost-user.h"
17 #include "hw/virtio/vhost-backend.h"
18 #include "hw/virtio/virtio.h"
19 #include "hw/virtio/virtio-net.h"
20 #include "chardev/char-fe.h"
21 #include "io/channel-socket.h"
22 #include "sysemu/kvm.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qemu/uuid.h"
26 #include "qemu/sockets.h"
27 #include "sysemu/runstate.h"
28 #include "sysemu/cryptodev.h"
29 #include "migration/postcopy-ram.h"
31 #include "exec/ramblock.h"
33 #include <sys/ioctl.h>
34 #include <sys/socket.h>
37 #include "standard-headers/linux/vhost_types.h"
40 #include <linux/userfaultfd.h>
43 #define VHOST_MEMORY_BASELINE_NREGIONS 8
44 #define VHOST_USER_F_PROTOCOL_FEATURES 30
45 #define VHOST_USER_BACKEND_MAX_FDS 8
47 #if defined(TARGET_PPC) || defined(TARGET_PPC64)
48 #include "hw/ppc/spapr.h"
49 #define VHOST_USER_MAX_RAM_SLOTS SPAPR_MAX_RAM_SLOTS
52 #define VHOST_USER_MAX_RAM_SLOTS 512
56 * Maximum size of virtio device config space
58 #define VHOST_USER_MAX_CONFIG_SIZE 256
60 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
62 typedef enum VhostUserRequest
{
64 VHOST_USER_GET_FEATURES
= 1,
65 VHOST_USER_SET_FEATURES
= 2,
66 VHOST_USER_SET_OWNER
= 3,
67 VHOST_USER_RESET_OWNER
= 4,
68 VHOST_USER_SET_MEM_TABLE
= 5,
69 VHOST_USER_SET_LOG_BASE
= 6,
70 VHOST_USER_SET_LOG_FD
= 7,
71 VHOST_USER_SET_VRING_NUM
= 8,
72 VHOST_USER_SET_VRING_ADDR
= 9,
73 VHOST_USER_SET_VRING_BASE
= 10,
74 VHOST_USER_GET_VRING_BASE
= 11,
75 VHOST_USER_SET_VRING_KICK
= 12,
76 VHOST_USER_SET_VRING_CALL
= 13,
77 VHOST_USER_SET_VRING_ERR
= 14,
78 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
79 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
80 VHOST_USER_GET_QUEUE_NUM
= 17,
81 VHOST_USER_SET_VRING_ENABLE
= 18,
82 VHOST_USER_SEND_RARP
= 19,
83 VHOST_USER_NET_SET_MTU
= 20,
84 VHOST_USER_SET_BACKEND_REQ_FD
= 21,
85 VHOST_USER_IOTLB_MSG
= 22,
86 VHOST_USER_SET_VRING_ENDIAN
= 23,
87 VHOST_USER_GET_CONFIG
= 24,
88 VHOST_USER_SET_CONFIG
= 25,
89 VHOST_USER_CREATE_CRYPTO_SESSION
= 26,
90 VHOST_USER_CLOSE_CRYPTO_SESSION
= 27,
91 VHOST_USER_POSTCOPY_ADVISE
= 28,
92 VHOST_USER_POSTCOPY_LISTEN
= 29,
93 VHOST_USER_POSTCOPY_END
= 30,
94 VHOST_USER_GET_INFLIGHT_FD
= 31,
95 VHOST_USER_SET_INFLIGHT_FD
= 32,
96 VHOST_USER_GPU_SET_SOCKET
= 33,
97 VHOST_USER_RESET_DEVICE
= 34,
98 /* Message number 35 reserved for VHOST_USER_VRING_KICK. */
99 VHOST_USER_GET_MAX_MEM_SLOTS
= 36,
100 VHOST_USER_ADD_MEM_REG
= 37,
101 VHOST_USER_REM_MEM_REG
= 38,
102 VHOST_USER_SET_STATUS
= 39,
103 VHOST_USER_GET_STATUS
= 40,
104 VHOST_USER_GET_SHARED_OBJECT
= 41,
105 VHOST_USER_SET_DEVICE_STATE_FD
= 42,
106 VHOST_USER_CHECK_DEVICE_STATE
= 43,
110 typedef enum VhostUserBackendRequest
{
111 VHOST_USER_BACKEND_NONE
= 0,
112 VHOST_USER_BACKEND_IOTLB_MSG
= 1,
113 VHOST_USER_BACKEND_CONFIG_CHANGE_MSG
= 2,
114 VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG
= 3,
115 VHOST_USER_BACKEND_SHARED_OBJECT_ADD
= 6,
116 VHOST_USER_BACKEND_SHARED_OBJECT_REMOVE
= 7,
117 VHOST_USER_BACKEND_SHARED_OBJECT_LOOKUP
= 8,
118 VHOST_USER_BACKEND_MAX
119 } VhostUserBackendRequest
;
121 typedef struct VhostUserMemoryRegion
{
122 uint64_t guest_phys_addr
;
123 uint64_t memory_size
;
124 uint64_t userspace_addr
;
125 uint64_t mmap_offset
;
126 } VhostUserMemoryRegion
;
128 typedef struct VhostUserMemory
{
131 VhostUserMemoryRegion regions
[VHOST_MEMORY_BASELINE_NREGIONS
];
134 typedef struct VhostUserMemRegMsg
{
136 VhostUserMemoryRegion region
;
137 } VhostUserMemRegMsg
;
139 typedef struct VhostUserLog
{
141 uint64_t mmap_offset
;
144 typedef struct VhostUserConfig
{
148 uint8_t region
[VHOST_USER_MAX_CONFIG_SIZE
];
151 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN 512
152 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN 64
153 #define VHOST_CRYPTO_ASYM_MAX_KEY_LEN 1024
155 typedef struct VhostUserCryptoSession
{
159 CryptoDevBackendSymSessionInfo session_setup_data
;
160 uint8_t key
[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN
];
161 uint8_t auth_key
[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN
];
164 CryptoDevBackendAsymSessionInfo session_setup_data
;
165 uint8_t key
[VHOST_CRYPTO_ASYM_MAX_KEY_LEN
];
169 /* session id for success, -1 on errors */
171 } VhostUserCryptoSession
;
173 static VhostUserConfig c
__attribute__ ((unused
));
174 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
178 typedef struct VhostUserVringArea
{
182 } VhostUserVringArea
;
184 typedef struct VhostUserInflight
{
186 uint64_t mmap_offset
;
191 typedef struct VhostUserShared
{
192 unsigned char uuid
[16];
196 VhostUserRequest request
;
198 #define VHOST_USER_VERSION_MASK (0x3)
199 #define VHOST_USER_REPLY_MASK (0x1 << 2)
200 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
202 uint32_t size
; /* the following payload size */
203 } QEMU_PACKED VhostUserHeader
;
205 /* Request payload of VHOST_USER_SET_DEVICE_STATE_FD */
206 typedef struct VhostUserTransferDeviceState
{
209 } VhostUserTransferDeviceState
;
212 #define VHOST_USER_VRING_IDX_MASK (0xff)
213 #define VHOST_USER_VRING_NOFD_MASK (0x1 << 8)
215 struct vhost_vring_state state
;
216 struct vhost_vring_addr addr
;
217 VhostUserMemory memory
;
218 VhostUserMemRegMsg mem_reg
;
220 struct vhost_iotlb_msg iotlb
;
221 VhostUserConfig config
;
222 VhostUserCryptoSession session
;
223 VhostUserVringArea area
;
224 VhostUserInflight inflight
;
225 VhostUserShared object
;
226 VhostUserTransferDeviceState transfer_state
;
229 typedef struct VhostUserMsg
{
231 VhostUserPayload payload
;
232 } QEMU_PACKED VhostUserMsg
;
234 static VhostUserMsg m
__attribute__ ((unused
));
235 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
237 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
239 /* The version of the protocol we support */
240 #define VHOST_USER_VERSION (0x1)
243 struct vhost_dev
*dev
;
244 /* Shared between vhost devs of the same virtio device */
245 VhostUserState
*user
;
246 QIOChannel
*backend_ioc
;
247 GSource
*backend_src
;
248 NotifierWithReturn postcopy_notifier
;
249 struct PostCopyFD postcopy_fd
;
250 uint64_t postcopy_client_bases
[VHOST_USER_MAX_RAM_SLOTS
];
251 /* Length of the region_rb and region_rb_offset arrays */
252 size_t region_rb_len
;
253 /* RAMBlock associated with a given region */
254 RAMBlock
**region_rb
;
256 * The offset from the start of the RAMBlock to the start of the
259 ram_addr_t
*region_rb_offset
;
261 /* True once we've entered postcopy_listen */
262 bool postcopy_listen
;
264 /* Our current regions */
265 int num_shadow_regions
;
266 struct vhost_memory_region shadow_regions
[VHOST_USER_MAX_RAM_SLOTS
];
269 struct scrub_regions
{
270 struct vhost_memory_region
*region
;
275 static int vhost_user_read_header(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
277 struct vhost_user
*u
= dev
->opaque
;
278 CharBackend
*chr
= u
->user
->chr
;
279 uint8_t *p
= (uint8_t *) msg
;
280 int r
, size
= VHOST_USER_HDR_SIZE
;
282 r
= qemu_chr_fe_read_all(chr
, p
, size
);
284 int saved_errno
= errno
;
285 error_report("Failed to read msg header. Read %d instead of %d."
286 " Original request %d.", r
, size
, msg
->hdr
.request
);
287 return r
< 0 ? -saved_errno
: -EIO
;
290 /* validate received flags */
291 if (msg
->hdr
.flags
!= (VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
)) {
292 error_report("Failed to read msg header."
293 " Flags 0x%x instead of 0x%x.", msg
->hdr
.flags
,
294 VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
);
298 trace_vhost_user_read(msg
->hdr
.request
, msg
->hdr
.flags
);
303 static int vhost_user_read(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
305 struct vhost_user
*u
= dev
->opaque
;
306 CharBackend
*chr
= u
->user
->chr
;
307 uint8_t *p
= (uint8_t *) msg
;
310 r
= vhost_user_read_header(dev
, msg
);
315 /* validate message size is sane */
316 if (msg
->hdr
.size
> VHOST_USER_PAYLOAD_SIZE
) {
317 error_report("Failed to read msg header."
318 " Size %d exceeds the maximum %zu.", msg
->hdr
.size
,
319 VHOST_USER_PAYLOAD_SIZE
);
324 p
+= VHOST_USER_HDR_SIZE
;
325 size
= msg
->hdr
.size
;
326 r
= qemu_chr_fe_read_all(chr
, p
, size
);
328 int saved_errno
= errno
;
329 error_report("Failed to read msg payload."
330 " Read %d instead of %d.", r
, msg
->hdr
.size
);
331 return r
< 0 ? -saved_errno
: -EIO
;
338 static int process_message_reply(struct vhost_dev
*dev
,
339 const VhostUserMsg
*msg
)
342 VhostUserMsg msg_reply
;
344 if ((msg
->hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) == 0) {
348 ret
= vhost_user_read(dev
, &msg_reply
);
353 if (msg_reply
.hdr
.request
!= msg
->hdr
.request
) {
354 error_report("Received unexpected msg type. "
355 "Expected %d received %d",
356 msg
->hdr
.request
, msg_reply
.hdr
.request
);
360 return msg_reply
.payload
.u64
? -EIO
: 0;
363 static bool vhost_user_per_device_request(VhostUserRequest request
)
366 case VHOST_USER_SET_OWNER
:
367 case VHOST_USER_RESET_OWNER
:
368 case VHOST_USER_SET_MEM_TABLE
:
369 case VHOST_USER_GET_QUEUE_NUM
:
370 case VHOST_USER_NET_SET_MTU
:
371 case VHOST_USER_RESET_DEVICE
:
372 case VHOST_USER_ADD_MEM_REG
:
373 case VHOST_USER_REM_MEM_REG
:
380 /* most non-init callers ignore the error */
381 static int vhost_user_write(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
382 int *fds
, int fd_num
)
384 struct vhost_user
*u
= dev
->opaque
;
385 CharBackend
*chr
= u
->user
->chr
;
386 int ret
, size
= VHOST_USER_HDR_SIZE
+ msg
->hdr
.size
;
389 * Some devices, like virtio-scsi, are implemented as a single vhost_dev,
390 * while others, like virtio-net, contain multiple vhost_devs. For
391 * operations such as configuring device memory mappings or issuing device
392 * resets, which affect the whole device instead of individual VQs,
393 * vhost-user messages should only be sent once.
395 * Devices with multiple vhost_devs are given an associated dev->vq_index
396 * so per_device requests are only sent if vq_index is 0.
398 if (vhost_user_per_device_request(msg
->hdr
.request
)
399 && dev
->vq_index
!= 0) {
400 msg
->hdr
.flags
&= ~VHOST_USER_NEED_REPLY_MASK
;
404 if (qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
) < 0) {
405 error_report("Failed to set msg fds.");
409 ret
= qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
);
411 int saved_errno
= errno
;
412 error_report("Failed to write msg."
413 " Wrote %d instead of %d.", ret
, size
);
414 return ret
< 0 ? -saved_errno
: -EIO
;
417 trace_vhost_user_write(msg
->hdr
.request
, msg
->hdr
.flags
);
422 int vhost_user_gpu_set_socket(struct vhost_dev
*dev
, int fd
)
425 .hdr
.request
= VHOST_USER_GPU_SET_SOCKET
,
426 .hdr
.flags
= VHOST_USER_VERSION
,
429 return vhost_user_write(dev
, &msg
, &fd
, 1);
432 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
433 struct vhost_log
*log
)
435 int fds
[VHOST_USER_MAX_RAM_SLOTS
];
437 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
438 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
441 .hdr
.request
= VHOST_USER_SET_LOG_BASE
,
442 .hdr
.flags
= VHOST_USER_VERSION
,
443 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
444 .payload
.log
.mmap_offset
= 0,
445 .hdr
.size
= sizeof(msg
.payload
.log
),
448 /* Send only once with first queue pair */
449 if (dev
->vq_index
!= 0) {
453 if (shmfd
&& log
->fd
!= -1) {
454 fds
[fd_num
++] = log
->fd
;
457 ret
= vhost_user_write(dev
, &msg
, fds
, fd_num
);
464 ret
= vhost_user_read(dev
, &msg
);
469 if (msg
.hdr
.request
!= VHOST_USER_SET_LOG_BASE
) {
470 error_report("Received unexpected msg type. "
471 "Expected %d received %d",
472 VHOST_USER_SET_LOG_BASE
, msg
.hdr
.request
);
480 static MemoryRegion
*vhost_user_get_mr_data(uint64_t addr
, ram_addr_t
*offset
,
485 assert((uintptr_t)addr
== addr
);
486 mr
= memory_region_from_host((void *)(uintptr_t)addr
, offset
);
487 *fd
= memory_region_get_fd(mr
);
488 *offset
+= mr
->ram_block
->fd_offset
;
493 static void vhost_user_fill_msg_region(VhostUserMemoryRegion
*dst
,
494 struct vhost_memory_region
*src
,
495 uint64_t mmap_offset
)
497 assert(src
!= NULL
&& dst
!= NULL
);
498 dst
->userspace_addr
= src
->userspace_addr
;
499 dst
->memory_size
= src
->memory_size
;
500 dst
->guest_phys_addr
= src
->guest_phys_addr
;
501 dst
->mmap_offset
= mmap_offset
;
504 static int vhost_user_fill_set_mem_table_msg(struct vhost_user
*u
,
505 struct vhost_dev
*dev
,
507 int *fds
, size_t *fd_num
,
508 bool track_ramblocks
)
513 struct vhost_memory_region
*reg
;
514 VhostUserMemoryRegion region_buffer
;
516 msg
->hdr
.request
= VHOST_USER_SET_MEM_TABLE
;
518 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
519 reg
= dev
->mem
->regions
+ i
;
521 mr
= vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
523 if (track_ramblocks
) {
524 assert(*fd_num
< VHOST_MEMORY_BASELINE_NREGIONS
);
525 trace_vhost_user_set_mem_table_withfd(*fd_num
, mr
->name
,
527 reg
->guest_phys_addr
,
530 u
->region_rb_offset
[i
] = offset
;
531 u
->region_rb
[i
] = mr
->ram_block
;
532 } else if (*fd_num
== VHOST_MEMORY_BASELINE_NREGIONS
) {
533 error_report("Failed preparing vhost-user memory table msg");
536 vhost_user_fill_msg_region(®ion_buffer
, reg
, offset
);
537 msg
->payload
.memory
.regions
[*fd_num
] = region_buffer
;
538 fds
[(*fd_num
)++] = fd
;
539 } else if (track_ramblocks
) {
540 u
->region_rb_offset
[i
] = 0;
541 u
->region_rb
[i
] = NULL
;
545 msg
->payload
.memory
.nregions
= *fd_num
;
548 error_report("Failed initializing vhost-user memory map, "
549 "consider using -object memory-backend-file share=on");
553 msg
->hdr
.size
= sizeof(msg
->payload
.memory
.nregions
);
554 msg
->hdr
.size
+= sizeof(msg
->payload
.memory
.padding
);
555 msg
->hdr
.size
+= *fd_num
* sizeof(VhostUserMemoryRegion
);
560 static inline bool reg_equal(struct vhost_memory_region
*shadow_reg
,
561 struct vhost_memory_region
*vdev_reg
)
563 return shadow_reg
->guest_phys_addr
== vdev_reg
->guest_phys_addr
&&
564 shadow_reg
->userspace_addr
== vdev_reg
->userspace_addr
&&
565 shadow_reg
->memory_size
== vdev_reg
->memory_size
;
568 static void scrub_shadow_regions(struct vhost_dev
*dev
,
569 struct scrub_regions
*add_reg
,
571 struct scrub_regions
*rem_reg
,
572 int *nr_rem_reg
, uint64_t *shadow_pcb
,
573 bool track_ramblocks
)
575 struct vhost_user
*u
= dev
->opaque
;
576 bool found
[VHOST_USER_MAX_RAM_SLOTS
] = {};
577 struct vhost_memory_region
*reg
, *shadow_reg
;
578 int i
, j
, fd
, add_idx
= 0, rm_idx
= 0, fd_num
= 0;
584 * Find memory regions present in our shadow state which are not in
585 * the device's current memory state.
587 * Mark regions in both the shadow and device state as "found".
589 for (i
= 0; i
< u
->num_shadow_regions
; i
++) {
590 shadow_reg
= &u
->shadow_regions
[i
];
593 for (j
= 0; j
< dev
->mem
->nregions
; j
++) {
594 reg
= &dev
->mem
->regions
[j
];
596 mr
= vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
598 if (reg_equal(shadow_reg
, reg
)) {
601 if (track_ramblocks
) {
603 * Reset postcopy client bases, region_rb, and
604 * region_rb_offset in case regions are removed.
607 u
->region_rb_offset
[j
] = offset
;
608 u
->region_rb
[j
] = mr
->ram_block
;
609 shadow_pcb
[j
] = u
->postcopy_client_bases
[i
];
611 u
->region_rb_offset
[j
] = 0;
612 u
->region_rb
[j
] = NULL
;
620 * If the region was not found in the current device memory state
621 * create an entry for it in the removed list.
624 rem_reg
[rm_idx
].region
= shadow_reg
;
625 rem_reg
[rm_idx
++].reg_idx
= i
;
630 * For regions not marked "found", create entries in the added list.
632 * Note their indexes in the device memory state and the indexes of their
635 for (i
= 0; i
< dev
->mem
->nregions
; i
++) {
636 reg
= &dev
->mem
->regions
[i
];
637 vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
643 * If the region was in both the shadow and device state we don't
644 * need to send a VHOST_USER_ADD_MEM_REG message for it.
650 add_reg
[add_idx
].region
= reg
;
651 add_reg
[add_idx
].reg_idx
= i
;
652 add_reg
[add_idx
++].fd_idx
= fd_num
;
654 *nr_rem_reg
= rm_idx
;
655 *nr_add_reg
= add_idx
;
660 static int send_remove_regions(struct vhost_dev
*dev
,
661 struct scrub_regions
*remove_reg
,
662 int nr_rem_reg
, VhostUserMsg
*msg
,
663 bool reply_supported
)
665 struct vhost_user
*u
= dev
->opaque
;
666 struct vhost_memory_region
*shadow_reg
;
667 int i
, fd
, shadow_reg_idx
, ret
;
669 VhostUserMemoryRegion region_buffer
;
672 * The regions in remove_reg appear in the same order they do in the
673 * shadow table. Therefore we can minimize memory copies by iterating
674 * through remove_reg backwards.
676 for (i
= nr_rem_reg
- 1; i
>= 0; i
--) {
677 shadow_reg
= remove_reg
[i
].region
;
678 shadow_reg_idx
= remove_reg
[i
].reg_idx
;
680 vhost_user_get_mr_data(shadow_reg
->userspace_addr
, &offset
, &fd
);
683 msg
->hdr
.request
= VHOST_USER_REM_MEM_REG
;
684 vhost_user_fill_msg_region(®ion_buffer
, shadow_reg
, 0);
685 msg
->payload
.mem_reg
.region
= region_buffer
;
687 ret
= vhost_user_write(dev
, msg
, NULL
, 0);
692 if (reply_supported
) {
693 ret
= process_message_reply(dev
, msg
);
701 * At this point we know the backend has unmapped the region. It is now
702 * safe to remove it from the shadow table.
704 memmove(&u
->shadow_regions
[shadow_reg_idx
],
705 &u
->shadow_regions
[shadow_reg_idx
+ 1],
706 sizeof(struct vhost_memory_region
) *
707 (u
->num_shadow_regions
- shadow_reg_idx
- 1));
708 u
->num_shadow_regions
--;
714 static int send_add_regions(struct vhost_dev
*dev
,
715 struct scrub_regions
*add_reg
, int nr_add_reg
,
716 VhostUserMsg
*msg
, uint64_t *shadow_pcb
,
717 bool reply_supported
, bool track_ramblocks
)
719 struct vhost_user
*u
= dev
->opaque
;
720 int i
, fd
, ret
, reg_idx
, reg_fd_idx
;
721 struct vhost_memory_region
*reg
;
724 VhostUserMsg msg_reply
;
725 VhostUserMemoryRegion region_buffer
;
727 for (i
= 0; i
< nr_add_reg
; i
++) {
728 reg
= add_reg
[i
].region
;
729 reg_idx
= add_reg
[i
].reg_idx
;
730 reg_fd_idx
= add_reg
[i
].fd_idx
;
732 mr
= vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
735 if (track_ramblocks
) {
736 trace_vhost_user_set_mem_table_withfd(reg_fd_idx
, mr
->name
,
738 reg
->guest_phys_addr
,
741 u
->region_rb_offset
[reg_idx
] = offset
;
742 u
->region_rb
[reg_idx
] = mr
->ram_block
;
744 msg
->hdr
.request
= VHOST_USER_ADD_MEM_REG
;
745 vhost_user_fill_msg_region(®ion_buffer
, reg
, offset
);
746 msg
->payload
.mem_reg
.region
= region_buffer
;
748 ret
= vhost_user_write(dev
, msg
, &fd
, 1);
753 if (track_ramblocks
) {
756 ret
= vhost_user_read(dev
, &msg_reply
);
761 reply_gpa
= msg_reply
.payload
.mem_reg
.region
.guest_phys_addr
;
763 if (msg_reply
.hdr
.request
!= VHOST_USER_ADD_MEM_REG
) {
764 error_report("%s: Received unexpected msg type."
765 "Expected %d received %d", __func__
,
766 VHOST_USER_ADD_MEM_REG
,
767 msg_reply
.hdr
.request
);
772 * We're using the same structure, just reusing one of the
773 * fields, so it should be the same size.
775 if (msg_reply
.hdr
.size
!= msg
->hdr
.size
) {
776 error_report("%s: Unexpected size for postcopy reply "
777 "%d vs %d", __func__
, msg_reply
.hdr
.size
,
782 /* Get the postcopy client base from the backend's reply. */
783 if (reply_gpa
== dev
->mem
->regions
[reg_idx
].guest_phys_addr
) {
784 shadow_pcb
[reg_idx
] =
785 msg_reply
.payload
.mem_reg
.region
.userspace_addr
;
786 trace_vhost_user_set_mem_table_postcopy(
787 msg_reply
.payload
.mem_reg
.region
.userspace_addr
,
788 msg
->payload
.mem_reg
.region
.userspace_addr
,
789 reg_fd_idx
, reg_idx
);
791 error_report("%s: invalid postcopy reply for region. "
792 "Got guest physical address %" PRIX64
", expected "
793 "%" PRIX64
, __func__
, reply_gpa
,
794 dev
->mem
->regions
[reg_idx
].guest_phys_addr
);
797 } else if (reply_supported
) {
798 ret
= process_message_reply(dev
, msg
);
803 } else if (track_ramblocks
) {
804 u
->region_rb_offset
[reg_idx
] = 0;
805 u
->region_rb
[reg_idx
] = NULL
;
809 * At this point, we know the backend has mapped in the new
810 * region, if the region has a valid file descriptor.
812 * The region should now be added to the shadow table.
814 u
->shadow_regions
[u
->num_shadow_regions
].guest_phys_addr
=
815 reg
->guest_phys_addr
;
816 u
->shadow_regions
[u
->num_shadow_regions
].userspace_addr
=
818 u
->shadow_regions
[u
->num_shadow_regions
].memory_size
=
820 u
->num_shadow_regions
++;
826 static int vhost_user_add_remove_regions(struct vhost_dev
*dev
,
828 bool reply_supported
,
829 bool track_ramblocks
)
831 struct vhost_user
*u
= dev
->opaque
;
832 struct scrub_regions add_reg
[VHOST_USER_MAX_RAM_SLOTS
];
833 struct scrub_regions rem_reg
[VHOST_USER_MAX_RAM_SLOTS
];
834 uint64_t shadow_pcb
[VHOST_USER_MAX_RAM_SLOTS
] = {};
835 int nr_add_reg
, nr_rem_reg
;
838 msg
->hdr
.size
= sizeof(msg
->payload
.mem_reg
);
840 /* Find the regions which need to be removed or added. */
841 scrub_shadow_regions(dev
, add_reg
, &nr_add_reg
, rem_reg
, &nr_rem_reg
,
842 shadow_pcb
, track_ramblocks
);
845 ret
= send_remove_regions(dev
, rem_reg
, nr_rem_reg
, msg
,
853 ret
= send_add_regions(dev
, add_reg
, nr_add_reg
, msg
, shadow_pcb
,
854 reply_supported
, track_ramblocks
);
860 if (track_ramblocks
) {
861 memcpy(u
->postcopy_client_bases
, shadow_pcb
,
862 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS
);
864 * Now we've registered this with the postcopy code, we ack to the
865 * client, because now we're in the position to be able to deal with
866 * any faults it generates.
868 /* TODO: Use this for failure cases as well with a bad value. */
869 msg
->hdr
.size
= sizeof(msg
->payload
.u64
);
870 msg
->payload
.u64
= 0; /* OK */
872 ret
= vhost_user_write(dev
, msg
, NULL
, 0);
881 if (track_ramblocks
) {
882 memcpy(u
->postcopy_client_bases
, shadow_pcb
,
883 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS
);
889 static int vhost_user_set_mem_table_postcopy(struct vhost_dev
*dev
,
890 struct vhost_memory
*mem
,
891 bool reply_supported
,
892 bool config_mem_slots
)
894 struct vhost_user
*u
= dev
->opaque
;
895 int fds
[VHOST_MEMORY_BASELINE_NREGIONS
];
897 VhostUserMsg msg_reply
;
902 .hdr
.flags
= VHOST_USER_VERSION
,
905 if (u
->region_rb_len
< dev
->mem
->nregions
) {
906 u
->region_rb
= g_renew(RAMBlock
*, u
->region_rb
, dev
->mem
->nregions
);
907 u
->region_rb_offset
= g_renew(ram_addr_t
, u
->region_rb_offset
,
909 memset(&(u
->region_rb
[u
->region_rb_len
]), '\0',
910 sizeof(RAMBlock
*) * (dev
->mem
->nregions
- u
->region_rb_len
));
911 memset(&(u
->region_rb_offset
[u
->region_rb_len
]), '\0',
912 sizeof(ram_addr_t
) * (dev
->mem
->nregions
- u
->region_rb_len
));
913 u
->region_rb_len
= dev
->mem
->nregions
;
916 if (config_mem_slots
) {
917 ret
= vhost_user_add_remove_regions(dev
, &msg
, reply_supported
, true);
922 ret
= vhost_user_fill_set_mem_table_msg(u
, dev
, &msg
, fds
, &fd_num
,
928 ret
= vhost_user_write(dev
, &msg
, fds
, fd_num
);
933 ret
= vhost_user_read(dev
, &msg_reply
);
938 if (msg_reply
.hdr
.request
!= VHOST_USER_SET_MEM_TABLE
) {
939 error_report("%s: Received unexpected msg type."
940 "Expected %d received %d", __func__
,
941 VHOST_USER_SET_MEM_TABLE
, msg_reply
.hdr
.request
);
946 * We're using the same structure, just reusing one of the
947 * fields, so it should be the same size.
949 if (msg_reply
.hdr
.size
!= msg
.hdr
.size
) {
950 error_report("%s: Unexpected size for postcopy reply "
951 "%d vs %d", __func__
, msg_reply
.hdr
.size
,
956 memset(u
->postcopy_client_bases
, 0,
957 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS
);
960 * They're in the same order as the regions that were sent
961 * but some of the regions were skipped (above) if they
964 for (msg_i
= 0, region_i
= 0;
965 region_i
< dev
->mem
->nregions
;
967 if (msg_i
< fd_num
&&
968 msg_reply
.payload
.memory
.regions
[msg_i
].guest_phys_addr
==
969 dev
->mem
->regions
[region_i
].guest_phys_addr
) {
970 u
->postcopy_client_bases
[region_i
] =
971 msg_reply
.payload
.memory
.regions
[msg_i
].userspace_addr
;
972 trace_vhost_user_set_mem_table_postcopy(
973 msg_reply
.payload
.memory
.regions
[msg_i
].userspace_addr
,
974 msg
.payload
.memory
.regions
[msg_i
].userspace_addr
,
979 if (msg_i
!= fd_num
) {
980 error_report("%s: postcopy reply not fully consumed "
982 __func__
, msg_i
, fd_num
);
987 * Now we've registered this with the postcopy code, we ack to the
988 * client, because now we're in the position to be able to deal
989 * with any faults it generates.
991 /* TODO: Use this for failure cases as well with a bad value. */
992 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
993 msg
.payload
.u64
= 0; /* OK */
994 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1003 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
1004 struct vhost_memory
*mem
)
1006 struct vhost_user
*u
= dev
->opaque
;
1007 int fds
[VHOST_MEMORY_BASELINE_NREGIONS
];
1009 bool do_postcopy
= u
->postcopy_listen
&& u
->postcopy_fd
.handler
;
1010 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1011 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1012 bool config_mem_slots
=
1013 virtio_has_feature(dev
->protocol_features
,
1014 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
);
1019 * Postcopy has enough differences that it's best done in it's own
1022 return vhost_user_set_mem_table_postcopy(dev
, mem
, reply_supported
,
1026 VhostUserMsg msg
= {
1027 .hdr
.flags
= VHOST_USER_VERSION
,
1030 if (reply_supported
) {
1031 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1034 if (config_mem_slots
) {
1035 ret
= vhost_user_add_remove_regions(dev
, &msg
, reply_supported
, false);
1040 ret
= vhost_user_fill_set_mem_table_msg(u
, dev
, &msg
, fds
, &fd_num
,
1046 ret
= vhost_user_write(dev
, &msg
, fds
, fd_num
);
1051 if (reply_supported
) {
1052 return process_message_reply(dev
, &msg
);
1059 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
1060 struct vhost_vring_state
*ring
)
1062 bool cross_endian
= virtio_has_feature(dev
->protocol_features
,
1063 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
);
1064 VhostUserMsg msg
= {
1065 .hdr
.request
= VHOST_USER_SET_VRING_ENDIAN
,
1066 .hdr
.flags
= VHOST_USER_VERSION
,
1067 .payload
.state
= *ring
,
1068 .hdr
.size
= sizeof(msg
.payload
.state
),
1071 if (!cross_endian
) {
1072 error_report("vhost-user trying to send unhandled ioctl");
1076 return vhost_user_write(dev
, &msg
, NULL
, 0);
1079 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
1082 VhostUserMsg msg
= {
1083 .hdr
.request
= request
,
1084 .hdr
.flags
= VHOST_USER_VERSION
,
1087 if (vhost_user_per_device_request(request
) && dev
->vq_index
!= 0) {
1091 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1096 ret
= vhost_user_read(dev
, &msg
);
1101 if (msg
.hdr
.request
!= request
) {
1102 error_report("Received unexpected msg type. Expected %d received %d",
1103 request
, msg
.hdr
.request
);
1107 if (msg
.hdr
.size
!= sizeof(msg
.payload
.u64
)) {
1108 error_report("Received bad msg size.");
1112 *u64
= msg
.payload
.u64
;
1117 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
1119 if (vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
) < 0) {
1126 /* Note: "msg->hdr.flags" may be modified. */
1127 static int vhost_user_write_sync(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
1128 bool wait_for_reply
)
1132 if (wait_for_reply
) {
1133 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1134 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1135 if (reply_supported
) {
1136 msg
->hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1140 ret
= vhost_user_write(dev
, msg
, NULL
, 0);
1145 if (wait_for_reply
) {
1148 if (msg
->hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) {
1149 return process_message_reply(dev
, msg
);
1153 * We need to wait for a reply but the backend does not
1154 * support replies for the command we just sent.
1155 * Send VHOST_USER_GET_FEATURES which makes all backends
1158 return vhost_user_get_features(dev
, &dummy
);
1164 static int vhost_set_vring(struct vhost_dev
*dev
,
1165 unsigned long int request
,
1166 struct vhost_vring_state
*ring
,
1167 bool wait_for_reply
)
1169 VhostUserMsg msg
= {
1170 .hdr
.request
= request
,
1171 .hdr
.flags
= VHOST_USER_VERSION
,
1172 .payload
.state
= *ring
,
1173 .hdr
.size
= sizeof(msg
.payload
.state
),
1176 return vhost_user_write_sync(dev
, &msg
, wait_for_reply
);
1179 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
1180 struct vhost_vring_state
*ring
)
1182 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
, false);
1185 static void vhost_user_host_notifier_free(VhostUserHostNotifier
*n
)
1187 assert(n
&& n
->unmap_addr
);
1188 munmap(n
->unmap_addr
, qemu_real_host_page_size());
1189 n
->unmap_addr
= NULL
;
1193 * clean-up function for notifier, will finally free the structure
1196 static void vhost_user_host_notifier_remove(VhostUserHostNotifier
*n
,
1201 virtio_queue_set_host_notifier_mr(vdev
, n
->idx
, &n
->mr
, false);
1203 assert(!n
->unmap_addr
);
1204 n
->unmap_addr
= n
->addr
;
1206 call_rcu(n
, vhost_user_host_notifier_free
, rcu
);
1210 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
1211 struct vhost_vring_state
*ring
)
1213 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
, false);
1216 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
1220 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
1224 for (i
= 0; i
< dev
->nvqs
; ++i
) {
1226 struct vhost_vring_state state
= {
1227 .index
= dev
->vq_index
+ i
,
1232 * SET_VRING_ENABLE travels from guest to QEMU to vhost-user backend /
1233 * control plane thread via unix domain socket. Virtio requests travel
1234 * from guest to vhost-user backend / data plane thread via eventfd.
1235 * Even if the guest enables the ring first, and pushes its first virtio
1236 * request second (conforming to the virtio spec), the data plane thread
1237 * in the backend may see the virtio request before the control plane
1238 * thread sees the queue enablement. This causes (in fact, requires) the
1239 * data plane thread to discard the virtio request (it arrived on a
1240 * seemingly disabled queue). To prevent this out-of-order delivery,
1241 * don't let the guest proceed to pushing the virtio request until the
1242 * backend control plane acknowledges enabling the queue -- IOW, pass
1243 * wait_for_reply=true below.
1245 ret
= vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
, true);
1248 * Restoring the previous state is likely infeasible, as well as
1249 * proceeding regardless the error, so just bail out and hope for
1250 * the device-level recovery.
1259 static VhostUserHostNotifier
*fetch_notifier(VhostUserState
*u
,
1262 if (idx
>= u
->notifiers
->len
) {
1265 return g_ptr_array_index(u
->notifiers
, idx
);
1268 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
1269 struct vhost_vring_state
*ring
)
1272 VhostUserMsg msg
= {
1273 .hdr
.request
= VHOST_USER_GET_VRING_BASE
,
1274 .hdr
.flags
= VHOST_USER_VERSION
,
1275 .payload
.state
= *ring
,
1276 .hdr
.size
= sizeof(msg
.payload
.state
),
1278 struct vhost_user
*u
= dev
->opaque
;
1280 VhostUserHostNotifier
*n
= fetch_notifier(u
->user
, ring
->index
);
1282 vhost_user_host_notifier_remove(n
, dev
->vdev
);
1285 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1290 ret
= vhost_user_read(dev
, &msg
);
1295 if (msg
.hdr
.request
!= VHOST_USER_GET_VRING_BASE
) {
1296 error_report("Received unexpected msg type. Expected %d received %d",
1297 VHOST_USER_GET_VRING_BASE
, msg
.hdr
.request
);
1301 if (msg
.hdr
.size
!= sizeof(msg
.payload
.state
)) {
1302 error_report("Received bad msg size.");
1306 *ring
= msg
.payload
.state
;
1311 static int vhost_set_vring_file(struct vhost_dev
*dev
,
1312 VhostUserRequest request
,
1313 struct vhost_vring_file
*file
)
1315 int fds
[VHOST_USER_MAX_RAM_SLOTS
];
1317 VhostUserMsg msg
= {
1318 .hdr
.request
= request
,
1319 .hdr
.flags
= VHOST_USER_VERSION
,
1320 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
1321 .hdr
.size
= sizeof(msg
.payload
.u64
),
1325 fds
[fd_num
++] = file
->fd
;
1327 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
1330 return vhost_user_write(dev
, &msg
, fds
, fd_num
);
1333 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
1334 struct vhost_vring_file
*file
)
1336 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
1339 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
1340 struct vhost_vring_file
*file
)
1342 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
1345 static int vhost_user_set_vring_err(struct vhost_dev
*dev
,
1346 struct vhost_vring_file
*file
)
1348 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_ERR
, file
);
1351 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
1352 struct vhost_vring_addr
*addr
)
1354 VhostUserMsg msg
= {
1355 .hdr
.request
= VHOST_USER_SET_VRING_ADDR
,
1356 .hdr
.flags
= VHOST_USER_VERSION
,
1357 .payload
.addr
= *addr
,
1358 .hdr
.size
= sizeof(msg
.payload
.addr
),
1362 * wait for a reply if logging is enabled to make sure
1363 * backend is actually logging changes
1365 bool wait_for_reply
= addr
->flags
& (1 << VHOST_VRING_F_LOG
);
1367 return vhost_user_write_sync(dev
, &msg
, wait_for_reply
);
1370 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
,
1371 bool wait_for_reply
)
1373 VhostUserMsg msg
= {
1374 .hdr
.request
= request
,
1375 .hdr
.flags
= VHOST_USER_VERSION
,
1377 .hdr
.size
= sizeof(msg
.payload
.u64
),
1380 return vhost_user_write_sync(dev
, &msg
, wait_for_reply
);
1383 static int vhost_user_set_status(struct vhost_dev
*dev
, uint8_t status
)
1385 return vhost_user_set_u64(dev
, VHOST_USER_SET_STATUS
, status
, false);
1388 static int vhost_user_get_status(struct vhost_dev
*dev
, uint8_t *status
)
1393 ret
= vhost_user_get_u64(dev
, VHOST_USER_GET_STATUS
, &value
);
1402 static int vhost_user_add_status(struct vhost_dev
*dev
, uint8_t status
)
1407 ret
= vhost_user_get_status(dev
, &s
);
1412 if ((s
& status
) == status
) {
1417 return vhost_user_set_status(dev
, s
);
1420 static int vhost_user_set_features(struct vhost_dev
*dev
,
1424 * wait for a reply if logging is enabled to make sure
1425 * backend is actually logging changes
1427 bool log_enabled
= features
& (0x1ULL
<< VHOST_F_LOG_ALL
);
1431 * We need to include any extra backend only feature bits that
1432 * might be needed by our device. Currently this includes the
1433 * VHOST_USER_F_PROTOCOL_FEATURES bit for enabling protocol
1436 ret
= vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
,
1437 features
| dev
->backend_features
,
1440 if (virtio_has_feature(dev
->protocol_features
,
1441 VHOST_USER_PROTOCOL_F_STATUS
)) {
1443 return vhost_user_add_status(dev
, VIRTIO_CONFIG_S_FEATURES_OK
);
1450 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
1453 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
,
1457 static int vhost_user_set_owner(struct vhost_dev
*dev
)
1459 VhostUserMsg msg
= {
1460 .hdr
.request
= VHOST_USER_SET_OWNER
,
1461 .hdr
.flags
= VHOST_USER_VERSION
,
1464 return vhost_user_write(dev
, &msg
, NULL
, 0);
1467 static int vhost_user_get_max_memslots(struct vhost_dev
*dev
,
1468 uint64_t *max_memslots
)
1470 uint64_t backend_max_memslots
;
1473 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_MAX_MEM_SLOTS
,
1474 &backend_max_memslots
);
1479 *max_memslots
= backend_max_memslots
;
1484 static int vhost_user_reset_device(struct vhost_dev
*dev
)
1486 VhostUserMsg msg
= {
1487 .hdr
.flags
= VHOST_USER_VERSION
,
1488 .hdr
.request
= VHOST_USER_RESET_DEVICE
,
1492 * Historically, reset was not implemented so only reset devices
1493 * that are expecting it.
1495 if (!virtio_has_feature(dev
->protocol_features
,
1496 VHOST_USER_PROTOCOL_F_RESET_DEVICE
)) {
1500 return vhost_user_write(dev
, &msg
, NULL
, 0);
1503 static int vhost_user_backend_handle_config_change(struct vhost_dev
*dev
)
1505 if (!dev
->config_ops
|| !dev
->config_ops
->vhost_dev_config_notifier
) {
1509 return dev
->config_ops
->vhost_dev_config_notifier(dev
);
1513 * Fetch or create the notifier for a given idx. Newly created
1514 * notifiers are added to the pointer array that tracks them.
1516 static VhostUserHostNotifier
*fetch_or_create_notifier(VhostUserState
*u
,
1519 VhostUserHostNotifier
*n
= NULL
;
1520 if (idx
>= u
->notifiers
->len
) {
1521 g_ptr_array_set_size(u
->notifiers
, idx
+ 1);
1524 n
= g_ptr_array_index(u
->notifiers
, idx
);
1527 * In case notification arrive out-of-order,
1528 * make room for current index.
1530 g_ptr_array_remove_index(u
->notifiers
, idx
);
1531 n
= g_new0(VhostUserHostNotifier
, 1);
1533 g_ptr_array_insert(u
->notifiers
, idx
, n
);
1534 trace_vhost_user_create_notifier(idx
, n
);
1540 static int vhost_user_backend_handle_vring_host_notifier(struct vhost_dev
*dev
,
1541 VhostUserVringArea
*area
,
1544 int queue_idx
= area
->u64
& VHOST_USER_VRING_IDX_MASK
;
1545 size_t page_size
= qemu_real_host_page_size();
1546 struct vhost_user
*u
= dev
->opaque
;
1547 VhostUserState
*user
= u
->user
;
1548 VirtIODevice
*vdev
= dev
->vdev
;
1549 VhostUserHostNotifier
*n
;
1553 if (!virtio_has_feature(dev
->protocol_features
,
1554 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
) ||
1555 vdev
== NULL
|| queue_idx
>= virtio_get_num_queues(vdev
)) {
1560 * Fetch notifier and invalidate any old data before setting up
1561 * new mapped address.
1563 n
= fetch_or_create_notifier(user
, queue_idx
);
1564 vhost_user_host_notifier_remove(n
, vdev
);
1566 if (area
->u64
& VHOST_USER_VRING_NOFD_MASK
) {
1571 if (area
->size
!= page_size
) {
1575 addr
= mmap(NULL
, page_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
1577 if (addr
== MAP_FAILED
) {
1581 name
= g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1583 if (!n
->mr
.ram
) { /* Don't init again after suspend. */
1584 memory_region_init_ram_device_ptr(&n
->mr
, OBJECT(vdev
), name
,
1587 n
->mr
.ram_block
->host
= addr
;
1591 if (virtio_queue_set_host_notifier_mr(vdev
, queue_idx
, &n
->mr
, true)) {
1592 object_unparent(OBJECT(&n
->mr
));
1593 munmap(addr
, page_size
);
1603 vhost_user_backend_handle_shared_object_add(struct vhost_dev
*dev
,
1604 VhostUserShared
*object
)
1608 memcpy(uuid
.data
, object
->uuid
, sizeof(object
->uuid
));
1609 return virtio_add_vhost_device(&uuid
, dev
);
1613 vhost_user_backend_handle_shared_object_remove(struct vhost_dev
*dev
,
1614 VhostUserShared
*object
)
1618 memcpy(uuid
.data
, object
->uuid
, sizeof(object
->uuid
));
1619 switch (virtio_object_type(&uuid
)) {
1620 case TYPE_VHOST_DEV
:
1622 struct vhost_dev
*owner
= virtio_lookup_vhost_device(&uuid
);
1624 /* Not allowed to remove non-owned entries */
1630 /* Not allowed to remove non-owned entries */
1634 return virtio_remove_resource(&uuid
);
1637 static bool vhost_user_send_resp(QIOChannel
*ioc
, VhostUserHeader
*hdr
,
1638 VhostUserPayload
*payload
, Error
**errp
)
1640 struct iovec iov
[] = {
1641 { .iov_base
= hdr
, .iov_len
= VHOST_USER_HDR_SIZE
},
1642 { .iov_base
= payload
, .iov_len
= hdr
->size
},
1645 hdr
->flags
&= ~VHOST_USER_NEED_REPLY_MASK
;
1646 hdr
->flags
|= VHOST_USER_REPLY_MASK
;
1648 return !qio_channel_writev_all(ioc
, iov
, ARRAY_SIZE(iov
), errp
);
1652 vhost_user_backend_send_dmabuf_fd(QIOChannel
*ioc
, VhostUserHeader
*hdr
,
1653 VhostUserPayload
*payload
, Error
**errp
)
1655 hdr
->size
= sizeof(payload
->u64
);
1656 return vhost_user_send_resp(ioc
, hdr
, payload
, errp
);
1659 int vhost_user_get_shared_object(struct vhost_dev
*dev
, unsigned char *uuid
,
1662 struct vhost_user
*u
= dev
->opaque
;
1663 CharBackend
*chr
= u
->user
->chr
;
1665 VhostUserMsg msg
= {
1666 .hdr
.request
= VHOST_USER_GET_SHARED_OBJECT
,
1667 .hdr
.flags
= VHOST_USER_VERSION
,
1669 memcpy(msg
.payload
.object
.uuid
, uuid
, sizeof(msg
.payload
.object
.uuid
));
1671 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1676 ret
= vhost_user_read(dev
, &msg
);
1681 if (msg
.hdr
.request
!= VHOST_USER_GET_SHARED_OBJECT
) {
1682 error_report("Received unexpected msg type. "
1683 "Expected %d received %d",
1684 VHOST_USER_GET_SHARED_OBJECT
, msg
.hdr
.request
);
1688 *dmabuf_fd
= qemu_chr_fe_get_msgfd(chr
);
1689 if (*dmabuf_fd
< 0) {
1690 error_report("Failed to get dmabuf fd");
1698 vhost_user_backend_handle_shared_object_lookup(struct vhost_user
*u
,
1700 VhostUserHeader
*hdr
,
1701 VhostUserPayload
*payload
)
1704 CharBackend
*chr
= u
->user
->chr
;
1705 Error
*local_err
= NULL
;
1709 memcpy(uuid
.data
, payload
->object
.uuid
, sizeof(payload
->object
.uuid
));
1712 switch (virtio_object_type(&uuid
)) {
1714 dmabuf_fd
= virtio_lookup_dmabuf(&uuid
);
1716 case TYPE_VHOST_DEV
:
1718 struct vhost_dev
*dev
= virtio_lookup_vhost_device(&uuid
);
1720 payload
->u64
= -EINVAL
;
1723 int ret
= vhost_user_get_shared_object(dev
, uuid
.data
, &dmabuf_fd
);
1730 payload
->u64
= -EINVAL
;
1734 if (dmabuf_fd
!= -1) {
1738 if (qemu_chr_fe_set_msgfds(chr
, &dmabuf_fd
, fd_num
) < 0) {
1739 error_report("Failed to set msg fds.");
1740 payload
->u64
= -EINVAL
;
1743 if (!vhost_user_backend_send_dmabuf_fd(ioc
, hdr
, payload
, &local_err
)) {
1744 error_report_err(local_err
);
1751 static void close_backend_channel(struct vhost_user
*u
)
1753 g_source_destroy(u
->backend_src
);
1754 g_source_unref(u
->backend_src
);
1755 u
->backend_src
= NULL
;
1756 object_unref(OBJECT(u
->backend_ioc
));
1757 u
->backend_ioc
= NULL
;
1760 static gboolean
backend_read(QIOChannel
*ioc
, GIOCondition condition
,
1763 struct vhost_dev
*dev
= opaque
;
1764 struct vhost_user
*u
= dev
->opaque
;
1765 VhostUserHeader hdr
= { 0, };
1766 VhostUserPayload payload
= { 0, };
1767 Error
*local_err
= NULL
;
1768 gboolean rc
= G_SOURCE_CONTINUE
;
1771 g_autofree
int *fd
= NULL
;
1776 iov
.iov_base
= &hdr
;
1777 iov
.iov_len
= VHOST_USER_HDR_SIZE
;
1779 if (qio_channel_readv_full_all(ioc
, &iov
, 1, &fd
, &fdsize
, &local_err
)) {
1780 error_report_err(local_err
);
1784 if (hdr
.size
> VHOST_USER_PAYLOAD_SIZE
) {
1785 error_report("Failed to read msg header."
1786 " Size %d exceeds the maximum %zu.", hdr
.size
,
1787 VHOST_USER_PAYLOAD_SIZE
);
1792 if (qio_channel_read_all(ioc
, (char *) &payload
, hdr
.size
, &local_err
)) {
1793 error_report_err(local_err
);
1797 switch (hdr
.request
) {
1798 case VHOST_USER_BACKEND_IOTLB_MSG
:
1799 ret
= vhost_backend_handle_iotlb_msg(dev
, &payload
.iotlb
);
1801 case VHOST_USER_BACKEND_CONFIG_CHANGE_MSG
:
1802 ret
= vhost_user_backend_handle_config_change(dev
);
1804 case VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG
:
1805 ret
= vhost_user_backend_handle_vring_host_notifier(dev
, &payload
.area
,
1808 case VHOST_USER_BACKEND_SHARED_OBJECT_ADD
:
1809 ret
= vhost_user_backend_handle_shared_object_add(dev
, &payload
.object
);
1811 case VHOST_USER_BACKEND_SHARED_OBJECT_REMOVE
:
1812 ret
= vhost_user_backend_handle_shared_object_remove(dev
,
1815 case VHOST_USER_BACKEND_SHARED_OBJECT_LOOKUP
:
1816 ret
= vhost_user_backend_handle_shared_object_lookup(dev
->opaque
, ioc
,
1820 error_report("Received unexpected msg type: %d.", hdr
.request
);
1825 * REPLY_ACK feature handling. Other reply types has to be managed
1826 * directly in their request handlers.
1828 if (hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) {
1829 payload
.u64
= !!ret
;
1830 hdr
.size
= sizeof(payload
.u64
);
1832 if (!vhost_user_send_resp(ioc
, &hdr
, &payload
, &local_err
)) {
1833 error_report_err(local_err
);
1841 close_backend_channel(u
);
1842 rc
= G_SOURCE_REMOVE
;
1846 for (i
= 0; i
< fdsize
; i
++) {
1853 static int vhost_setup_backend_channel(struct vhost_dev
*dev
)
1855 VhostUserMsg msg
= {
1856 .hdr
.request
= VHOST_USER_SET_BACKEND_REQ_FD
,
1857 .hdr
.flags
= VHOST_USER_VERSION
,
1859 struct vhost_user
*u
= dev
->opaque
;
1861 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1862 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1863 Error
*local_err
= NULL
;
1866 if (!virtio_has_feature(dev
->protocol_features
,
1867 VHOST_USER_PROTOCOL_F_BACKEND_REQ
)) {
1871 if (qemu_socketpair(PF_UNIX
, SOCK_STREAM
, 0, sv
) == -1) {
1872 int saved_errno
= errno
;
1873 error_report("socketpair() failed");
1874 return -saved_errno
;
1877 ioc
= QIO_CHANNEL(qio_channel_socket_new_fd(sv
[0], &local_err
));
1879 error_report_err(local_err
);
1880 return -ECONNREFUSED
;
1882 u
->backend_ioc
= ioc
;
1883 u
->backend_src
= qio_channel_add_watch_source(u
->backend_ioc
,
1885 backend_read
, dev
, NULL
, NULL
);
1887 if (reply_supported
) {
1888 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1891 ret
= vhost_user_write(dev
, &msg
, &sv
[1], 1);
1896 if (reply_supported
) {
1897 ret
= process_message_reply(dev
, &msg
);
1903 close_backend_channel(u
);
1911 * Called back from the postcopy fault thread when a fault is received on our
1913 * TODO: This is Linux specific
1915 static int vhost_user_postcopy_fault_handler(struct PostCopyFD
*pcfd
,
1918 struct vhost_dev
*dev
= pcfd
->data
;
1919 struct vhost_user
*u
= dev
->opaque
;
1920 struct uffd_msg
*msg
= ufd
;
1921 uint64_t faultaddr
= msg
->arg
.pagefault
.address
;
1922 RAMBlock
*rb
= NULL
;
1926 trace_vhost_user_postcopy_fault_handler(pcfd
->idstr
, faultaddr
,
1927 dev
->mem
->nregions
);
1928 for (i
= 0; i
< MIN(dev
->mem
->nregions
, u
->region_rb_len
); i
++) {
1929 trace_vhost_user_postcopy_fault_handler_loop(i
,
1930 u
->postcopy_client_bases
[i
], dev
->mem
->regions
[i
].memory_size
);
1931 if (faultaddr
>= u
->postcopy_client_bases
[i
]) {
1932 /* Ofset of the fault address in the vhost region */
1933 uint64_t region_offset
= faultaddr
- u
->postcopy_client_bases
[i
];
1934 if (region_offset
< dev
->mem
->regions
[i
].memory_size
) {
1935 rb_offset
= region_offset
+ u
->region_rb_offset
[i
];
1936 trace_vhost_user_postcopy_fault_handler_found(i
,
1937 region_offset
, rb_offset
);
1938 rb
= u
->region_rb
[i
];
1939 return postcopy_request_shared_page(pcfd
, rb
, faultaddr
,
1944 error_report("%s: Failed to find region for fault %" PRIx64
,
1945 __func__
, faultaddr
);
1949 static int vhost_user_postcopy_waker(struct PostCopyFD
*pcfd
, RAMBlock
*rb
,
1952 struct vhost_dev
*dev
= pcfd
->data
;
1953 struct vhost_user
*u
= dev
->opaque
;
1956 trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb
), offset
);
1961 /* Translate the offset into an address in the clients address space */
1962 for (i
= 0; i
< MIN(dev
->mem
->nregions
, u
->region_rb_len
); i
++) {
1963 if (u
->region_rb
[i
] == rb
&&
1964 offset
>= u
->region_rb_offset
[i
] &&
1965 offset
< (u
->region_rb_offset
[i
] +
1966 dev
->mem
->regions
[i
].memory_size
)) {
1967 uint64_t client_addr
= (offset
- u
->region_rb_offset
[i
]) +
1968 u
->postcopy_client_bases
[i
];
1969 trace_vhost_user_postcopy_waker_found(client_addr
);
1970 return postcopy_wake_shared(pcfd
, client_addr
, rb
);
1974 trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb
), offset
);
1980 * Called at the start of an inbound postcopy on reception of the
1983 static int vhost_user_postcopy_advise(struct vhost_dev
*dev
, Error
**errp
)
1986 struct vhost_user
*u
= dev
->opaque
;
1987 CharBackend
*chr
= u
->user
->chr
;
1990 VhostUserMsg msg
= {
1991 .hdr
.request
= VHOST_USER_POSTCOPY_ADVISE
,
1992 .hdr
.flags
= VHOST_USER_VERSION
,
1995 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1997 error_setg(errp
, "Failed to send postcopy_advise to vhost");
2001 ret
= vhost_user_read(dev
, &msg
);
2003 error_setg(errp
, "Failed to get postcopy_advise reply from vhost");
2007 if (msg
.hdr
.request
!= VHOST_USER_POSTCOPY_ADVISE
) {
2008 error_setg(errp
, "Unexpected msg type. Expected %d received %d",
2009 VHOST_USER_POSTCOPY_ADVISE
, msg
.hdr
.request
);
2014 error_setg(errp
, "Received bad msg size.");
2017 ufd
= qemu_chr_fe_get_msgfd(chr
);
2019 error_setg(errp
, "%s: Failed to get ufd", __func__
);
2022 qemu_socket_set_nonblock(ufd
);
2024 /* register ufd with userfault thread */
2025 u
->postcopy_fd
.fd
= ufd
;
2026 u
->postcopy_fd
.data
= dev
;
2027 u
->postcopy_fd
.handler
= vhost_user_postcopy_fault_handler
;
2028 u
->postcopy_fd
.waker
= vhost_user_postcopy_waker
;
2029 u
->postcopy_fd
.idstr
= "vhost-user"; /* Need to find unique name */
2030 postcopy_register_shared_ufd(&u
->postcopy_fd
);
2033 error_setg(errp
, "Postcopy not supported on non-Linux systems");
2039 * Called at the switch to postcopy on reception of the 'listen' command.
2041 static int vhost_user_postcopy_listen(struct vhost_dev
*dev
, Error
**errp
)
2043 struct vhost_user
*u
= dev
->opaque
;
2045 VhostUserMsg msg
= {
2046 .hdr
.request
= VHOST_USER_POSTCOPY_LISTEN
,
2047 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
2049 u
->postcopy_listen
= true;
2051 trace_vhost_user_postcopy_listen();
2053 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2055 error_setg(errp
, "Failed to send postcopy_listen to vhost");
2059 ret
= process_message_reply(dev
, &msg
);
2061 error_setg(errp
, "Failed to receive reply to postcopy_listen");
2069 * Called at the end of postcopy
2071 static int vhost_user_postcopy_end(struct vhost_dev
*dev
, Error
**errp
)
2073 VhostUserMsg msg
= {
2074 .hdr
.request
= VHOST_USER_POSTCOPY_END
,
2075 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
2078 struct vhost_user
*u
= dev
->opaque
;
2080 trace_vhost_user_postcopy_end_entry();
2082 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2084 error_setg(errp
, "Failed to send postcopy_end to vhost");
2088 ret
= process_message_reply(dev
, &msg
);
2090 error_setg(errp
, "Failed to receive reply to postcopy_end");
2093 postcopy_unregister_shared_ufd(&u
->postcopy_fd
);
2094 close(u
->postcopy_fd
.fd
);
2095 u
->postcopy_fd
.handler
= NULL
;
2097 trace_vhost_user_postcopy_end_exit();
2102 static int vhost_user_postcopy_notifier(NotifierWithReturn
*notifier
,
2103 void *opaque
, Error
**errp
)
2105 struct PostcopyNotifyData
*pnd
= opaque
;
2106 struct vhost_user
*u
= container_of(notifier
, struct vhost_user
,
2108 struct vhost_dev
*dev
= u
->dev
;
2110 switch (pnd
->reason
) {
2111 case POSTCOPY_NOTIFY_PROBE
:
2112 if (!virtio_has_feature(dev
->protocol_features
,
2113 VHOST_USER_PROTOCOL_F_PAGEFAULT
)) {
2114 /* TODO: Get the device name into this error somehow */
2116 "vhost-user backend not capable of postcopy");
2121 case POSTCOPY_NOTIFY_INBOUND_ADVISE
:
2122 return vhost_user_postcopy_advise(dev
, errp
);
2124 case POSTCOPY_NOTIFY_INBOUND_LISTEN
:
2125 return vhost_user_postcopy_listen(dev
, errp
);
2127 case POSTCOPY_NOTIFY_INBOUND_END
:
2128 return vhost_user_postcopy_end(dev
, errp
);
2131 /* We ignore notifications we don't know */
2138 static int vhost_user_backend_init(struct vhost_dev
*dev
, void *opaque
,
2141 uint64_t features
, ram_slots
;
2142 struct vhost_user
*u
;
2143 VhostUserState
*vus
= (VhostUserState
*) opaque
;
2146 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2148 u
= g_new0(struct vhost_user
, 1);
2153 err
= vhost_user_get_features(dev
, &features
);
2155 error_setg_errno(errp
, -err
, "vhost_backend_init failed");
2159 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
2160 bool supports_f_config
= vus
->supports_config
||
2161 (dev
->config_ops
&& dev
->config_ops
->vhost_dev_config_notifier
);
2162 uint64_t protocol_features
;
2164 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
2166 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
2167 &protocol_features
);
2169 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2174 * We will use all the protocol features we support - although
2175 * we suppress F_CONFIG if we know QEMUs internal code can not support
2178 protocol_features
&= VHOST_USER_PROTOCOL_FEATURE_MASK
;
2180 if (supports_f_config
) {
2181 if (!virtio_has_feature(protocol_features
,
2182 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2183 error_setg(errp
, "vhost-user device expecting "
2184 "VHOST_USER_PROTOCOL_F_CONFIG but the vhost-user backend does "
2189 if (virtio_has_feature(protocol_features
,
2190 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2191 warn_report("vhost-user backend supports "
2192 "VHOST_USER_PROTOCOL_F_CONFIG but QEMU does not.");
2193 protocol_features
&= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG
);
2197 /* final set of protocol features */
2198 dev
->protocol_features
= protocol_features
;
2199 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
2201 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2205 /* query the max queues we support if backend supports Multiple Queue */
2206 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
2207 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
2210 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2214 dev
->max_queues
= 1;
2217 if (dev
->num_queues
&& dev
->max_queues
< dev
->num_queues
) {
2218 error_setg(errp
, "The maximum number of queues supported by the "
2219 "backend is %" PRIu64
, dev
->max_queues
);
2223 if (virtio_has_feature(features
, VIRTIO_F_IOMMU_PLATFORM
) &&
2224 !(virtio_has_feature(dev
->protocol_features
,
2225 VHOST_USER_PROTOCOL_F_BACKEND_REQ
) &&
2226 virtio_has_feature(dev
->protocol_features
,
2227 VHOST_USER_PROTOCOL_F_REPLY_ACK
))) {
2228 error_setg(errp
, "IOMMU support requires reply-ack and "
2229 "backend-req protocol features.");
2233 /* get max memory regions if backend supports configurable RAM slots */
2234 if (!virtio_has_feature(dev
->protocol_features
,
2235 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
)) {
2236 u
->user
->memory_slots
= VHOST_MEMORY_BASELINE_NREGIONS
;
2238 err
= vhost_user_get_max_memslots(dev
, &ram_slots
);
2240 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2244 if (ram_slots
< u
->user
->memory_slots
) {
2245 error_setg(errp
, "The backend specified a max ram slots limit "
2246 "of %" PRIu64
", when the prior validated limit was "
2247 "%d. This limit should never decrease.", ram_slots
,
2248 u
->user
->memory_slots
);
2252 u
->user
->memory_slots
= MIN(ram_slots
, VHOST_USER_MAX_RAM_SLOTS
);
2256 if (dev
->migration_blocker
== NULL
&&
2257 !virtio_has_feature(dev
->protocol_features
,
2258 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
2259 error_setg(&dev
->migration_blocker
,
2260 "Migration disabled: vhost-user backend lacks "
2261 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
2264 if (dev
->vq_index
== 0) {
2265 err
= vhost_setup_backend_channel(dev
);
2267 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2272 u
->postcopy_notifier
.notify
= vhost_user_postcopy_notifier
;
2273 postcopy_add_notifier(&u
->postcopy_notifier
);
2278 static int vhost_user_backend_cleanup(struct vhost_dev
*dev
)
2280 struct vhost_user
*u
;
2282 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2285 if (u
->postcopy_notifier
.notify
) {
2286 postcopy_remove_notifier(&u
->postcopy_notifier
);
2287 u
->postcopy_notifier
.notify
= NULL
;
2289 u
->postcopy_listen
= false;
2290 if (u
->postcopy_fd
.handler
) {
2291 postcopy_unregister_shared_ufd(&u
->postcopy_fd
);
2292 close(u
->postcopy_fd
.fd
);
2293 u
->postcopy_fd
.handler
= NULL
;
2295 if (u
->backend_ioc
) {
2296 close_backend_channel(u
);
2298 g_free(u
->region_rb
);
2299 u
->region_rb
= NULL
;
2300 g_free(u
->region_rb_offset
);
2301 u
->region_rb_offset
= NULL
;
2302 u
->region_rb_len
= 0;
2309 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
2311 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
2316 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
2318 struct vhost_user
*u
= dev
->opaque
;
2320 return u
->user
->memory_slots
;
2323 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
2325 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2327 return virtio_has_feature(dev
->protocol_features
,
2328 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
2331 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
2333 VhostUserMsg msg
= { };
2335 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2337 /* If guest supports GUEST_ANNOUNCE do nothing */
2338 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
2342 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
2343 if (virtio_has_feature(dev
->protocol_features
,
2344 VHOST_USER_PROTOCOL_F_RARP
)) {
2345 msg
.hdr
.request
= VHOST_USER_SEND_RARP
;
2346 msg
.hdr
.flags
= VHOST_USER_VERSION
;
2347 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
2348 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
2350 return vhost_user_write(dev
, &msg
, NULL
, 0);
2355 static int vhost_user_net_set_mtu(struct vhost_dev
*dev
, uint16_t mtu
)
2358 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
2359 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
2362 if (!(dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU
))) {
2366 msg
.hdr
.request
= VHOST_USER_NET_SET_MTU
;
2367 msg
.payload
.u64
= mtu
;
2368 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
2369 msg
.hdr
.flags
= VHOST_USER_VERSION
;
2370 if (reply_supported
) {
2371 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
2374 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2379 /* If reply_ack supported, backend has to ack specified MTU is valid */
2380 if (reply_supported
) {
2381 return process_message_reply(dev
, &msg
);
2387 static int vhost_user_send_device_iotlb_msg(struct vhost_dev
*dev
,
2388 struct vhost_iotlb_msg
*imsg
)
2391 VhostUserMsg msg
= {
2392 .hdr
.request
= VHOST_USER_IOTLB_MSG
,
2393 .hdr
.size
= sizeof(msg
.payload
.iotlb
),
2394 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
2395 .payload
.iotlb
= *imsg
,
2398 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2403 return process_message_reply(dev
, &msg
);
2407 static void vhost_user_set_iotlb_callback(struct vhost_dev
*dev
, int enabled
)
2409 /* No-op as the receive channel is not dedicated to IOTLB messages. */
2412 static int vhost_user_get_config(struct vhost_dev
*dev
, uint8_t *config
,
2413 uint32_t config_len
, Error
**errp
)
2416 VhostUserMsg msg
= {
2417 .hdr
.request
= VHOST_USER_GET_CONFIG
,
2418 .hdr
.flags
= VHOST_USER_VERSION
,
2419 .hdr
.size
= VHOST_USER_CONFIG_HDR_SIZE
+ config_len
,
2422 if (!virtio_has_feature(dev
->protocol_features
,
2423 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2424 error_setg(errp
, "VHOST_USER_PROTOCOL_F_CONFIG not supported");
2428 assert(config_len
<= VHOST_USER_MAX_CONFIG_SIZE
);
2430 msg
.payload
.config
.offset
= 0;
2431 msg
.payload
.config
.size
= config_len
;
2432 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2434 error_setg_errno(errp
, -ret
, "vhost_get_config failed");
2438 ret
= vhost_user_read(dev
, &msg
);
2440 error_setg_errno(errp
, -ret
, "vhost_get_config failed");
2444 if (msg
.hdr
.request
!= VHOST_USER_GET_CONFIG
) {
2446 "Received unexpected msg type. Expected %d received %d",
2447 VHOST_USER_GET_CONFIG
, msg
.hdr
.request
);
2451 if (msg
.hdr
.size
!= VHOST_USER_CONFIG_HDR_SIZE
+ config_len
) {
2452 error_setg(errp
, "Received bad msg size.");
2456 memcpy(config
, msg
.payload
.config
.region
, config_len
);
2461 static int vhost_user_set_config(struct vhost_dev
*dev
, const uint8_t *data
,
2462 uint32_t offset
, uint32_t size
, uint32_t flags
)
2466 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
2467 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
2469 VhostUserMsg msg
= {
2470 .hdr
.request
= VHOST_USER_SET_CONFIG
,
2471 .hdr
.flags
= VHOST_USER_VERSION
,
2472 .hdr
.size
= VHOST_USER_CONFIG_HDR_SIZE
+ size
,
2475 if (!virtio_has_feature(dev
->protocol_features
,
2476 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2480 if (reply_supported
) {
2481 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
2484 if (size
> VHOST_USER_MAX_CONFIG_SIZE
) {
2488 msg
.payload
.config
.offset
= offset
,
2489 msg
.payload
.config
.size
= size
,
2490 msg
.payload
.config
.flags
= flags
,
2491 p
= msg
.payload
.config
.region
;
2492 memcpy(p
, data
, size
);
2494 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2499 if (reply_supported
) {
2500 return process_message_reply(dev
, &msg
);
2506 static int vhost_user_crypto_create_session(struct vhost_dev
*dev
,
2508 uint64_t *session_id
)
2511 bool crypto_session
= virtio_has_feature(dev
->protocol_features
,
2512 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
);
2513 CryptoDevBackendSessionInfo
*backend_info
= session_info
;
2514 VhostUserMsg msg
= {
2515 .hdr
.request
= VHOST_USER_CREATE_CRYPTO_SESSION
,
2516 .hdr
.flags
= VHOST_USER_VERSION
,
2517 .hdr
.size
= sizeof(msg
.payload
.session
),
2520 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2522 if (!crypto_session
) {
2523 error_report("vhost-user trying to send unhandled ioctl");
2527 if (backend_info
->op_code
== VIRTIO_CRYPTO_AKCIPHER_CREATE_SESSION
) {
2528 CryptoDevBackendAsymSessionInfo
*sess
= &backend_info
->u
.asym_sess_info
;
2531 memcpy(&msg
.payload
.session
.u
.asym
.session_setup_data
, sess
,
2532 sizeof(CryptoDevBackendAsymSessionInfo
));
2534 keylen
= sizeof(msg
.payload
.session
.u
.asym
.key
);
2535 if (sess
->keylen
> keylen
) {
2536 error_report("Unsupported asymmetric key size");
2540 memcpy(&msg
.payload
.session
.u
.asym
.key
, sess
->key
,
2544 CryptoDevBackendSymSessionInfo
*sess
= &backend_info
->u
.sym_sess_info
;
2547 memcpy(&msg
.payload
.session
.u
.sym
.session_setup_data
, sess
,
2548 sizeof(CryptoDevBackendSymSessionInfo
));
2549 if (sess
->key_len
) {
2550 keylen
= sizeof(msg
.payload
.session
.u
.sym
.key
);
2551 if (sess
->key_len
> keylen
) {
2552 error_report("Unsupported cipher key size");
2556 memcpy(&msg
.payload
.session
.u
.sym
.key
, sess
->cipher_key
,
2560 if (sess
->auth_key_len
> 0) {
2561 keylen
= sizeof(msg
.payload
.session
.u
.sym
.auth_key
);
2562 if (sess
->auth_key_len
> keylen
) {
2563 error_report("Unsupported auth key size");
2567 memcpy(&msg
.payload
.session
.u
.sym
.auth_key
, sess
->auth_key
,
2568 sess
->auth_key_len
);
2572 msg
.payload
.session
.op_code
= backend_info
->op_code
;
2573 msg
.payload
.session
.session_id
= backend_info
->session_id
;
2574 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2576 error_report("vhost_user_write() return %d, create session failed",
2581 ret
= vhost_user_read(dev
, &msg
);
2583 error_report("vhost_user_read() return %d, create session failed",
2588 if (msg
.hdr
.request
!= VHOST_USER_CREATE_CRYPTO_SESSION
) {
2589 error_report("Received unexpected msg type. Expected %d received %d",
2590 VHOST_USER_CREATE_CRYPTO_SESSION
, msg
.hdr
.request
);
2594 if (msg
.hdr
.size
!= sizeof(msg
.payload
.session
)) {
2595 error_report("Received bad msg size.");
2599 if (msg
.payload
.session
.session_id
< 0) {
2600 error_report("Bad session id: %" PRId64
"",
2601 msg
.payload
.session
.session_id
);
2604 *session_id
= msg
.payload
.session
.session_id
;
2610 vhost_user_crypto_close_session(struct vhost_dev
*dev
, uint64_t session_id
)
2613 bool crypto_session
= virtio_has_feature(dev
->protocol_features
,
2614 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
);
2615 VhostUserMsg msg
= {
2616 .hdr
.request
= VHOST_USER_CLOSE_CRYPTO_SESSION
,
2617 .hdr
.flags
= VHOST_USER_VERSION
,
2618 .hdr
.size
= sizeof(msg
.payload
.u64
),
2620 msg
.payload
.u64
= session_id
;
2622 if (!crypto_session
) {
2623 error_report("vhost-user trying to send unhandled ioctl");
2627 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2629 error_report("vhost_user_write() return %d, close session failed",
2637 static bool vhost_user_no_private_memslots(struct vhost_dev
*dev
)
2642 static int vhost_user_get_inflight_fd(struct vhost_dev
*dev
,
2643 uint16_t queue_size
,
2644 struct vhost_inflight
*inflight
)
2649 struct vhost_user
*u
= dev
->opaque
;
2650 CharBackend
*chr
= u
->user
->chr
;
2651 VhostUserMsg msg
= {
2652 .hdr
.request
= VHOST_USER_GET_INFLIGHT_FD
,
2653 .hdr
.flags
= VHOST_USER_VERSION
,
2654 .payload
.inflight
.num_queues
= dev
->nvqs
,
2655 .payload
.inflight
.queue_size
= queue_size
,
2656 .hdr
.size
= sizeof(msg
.payload
.inflight
),
2659 if (!virtio_has_feature(dev
->protocol_features
,
2660 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
)) {
2664 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2669 ret
= vhost_user_read(dev
, &msg
);
2674 if (msg
.hdr
.request
!= VHOST_USER_GET_INFLIGHT_FD
) {
2675 error_report("Received unexpected msg type. "
2676 "Expected %d received %d",
2677 VHOST_USER_GET_INFLIGHT_FD
, msg
.hdr
.request
);
2681 if (msg
.hdr
.size
!= sizeof(msg
.payload
.inflight
)) {
2682 error_report("Received bad msg size.");
2686 if (!msg
.payload
.inflight
.mmap_size
) {
2690 fd
= qemu_chr_fe_get_msgfd(chr
);
2692 error_report("Failed to get mem fd");
2696 addr
= mmap(0, msg
.payload
.inflight
.mmap_size
, PROT_READ
| PROT_WRITE
,
2697 MAP_SHARED
, fd
, msg
.payload
.inflight
.mmap_offset
);
2699 if (addr
== MAP_FAILED
) {
2700 error_report("Failed to mmap mem fd");
2705 inflight
->addr
= addr
;
2707 inflight
->size
= msg
.payload
.inflight
.mmap_size
;
2708 inflight
->offset
= msg
.payload
.inflight
.mmap_offset
;
2709 inflight
->queue_size
= queue_size
;
2714 static int vhost_user_set_inflight_fd(struct vhost_dev
*dev
,
2715 struct vhost_inflight
*inflight
)
2717 VhostUserMsg msg
= {
2718 .hdr
.request
= VHOST_USER_SET_INFLIGHT_FD
,
2719 .hdr
.flags
= VHOST_USER_VERSION
,
2720 .payload
.inflight
.mmap_size
= inflight
->size
,
2721 .payload
.inflight
.mmap_offset
= inflight
->offset
,
2722 .payload
.inflight
.num_queues
= dev
->nvqs
,
2723 .payload
.inflight
.queue_size
= inflight
->queue_size
,
2724 .hdr
.size
= sizeof(msg
.payload
.inflight
),
2727 if (!virtio_has_feature(dev
->protocol_features
,
2728 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
)) {
2732 return vhost_user_write(dev
, &msg
, &inflight
->fd
, 1);
2735 static void vhost_user_state_destroy(gpointer data
)
2737 VhostUserHostNotifier
*n
= (VhostUserHostNotifier
*) data
;
2739 vhost_user_host_notifier_remove(n
, NULL
);
2740 object_unparent(OBJECT(&n
->mr
));
2742 * We can't free until vhost_user_host_notifier_remove has
2743 * done it's thing so schedule the free with RCU.
2749 bool vhost_user_init(VhostUserState
*user
, CharBackend
*chr
, Error
**errp
)
2752 error_setg(errp
, "Cannot initialize vhost-user state");
2756 user
->memory_slots
= 0;
2757 user
->notifiers
= g_ptr_array_new_full(VIRTIO_QUEUE_MAX
/ 4,
2758 &vhost_user_state_destroy
);
2762 void vhost_user_cleanup(VhostUserState
*user
)
2767 memory_region_transaction_begin();
2768 user
->notifiers
= (GPtrArray
*) g_ptr_array_free(user
->notifiers
, true);
2769 memory_region_transaction_commit();
2775 vu_async_close_fn cb
;
2778 struct vhost_dev
*vhost
;
2779 IOEventHandler
*event_cb
;
2780 } VhostAsyncCallback
;
2782 static void vhost_user_async_close_bh(void *opaque
)
2784 VhostAsyncCallback
*data
= opaque
;
2785 struct vhost_dev
*vhost
= data
->vhost
;
2788 * If the vhost_dev has been cleared in the meantime there is
2789 * nothing left to do as some other path has completed the
2793 data
->cb(data
->dev
);
2794 } else if (data
->event_cb
) {
2795 qemu_chr_fe_set_handlers(data
->cd
, NULL
, NULL
, data
->event_cb
,
2796 NULL
, data
->dev
, NULL
, true);
2803 * We only schedule the work if the machine is running. If suspended
2804 * we want to keep all the in-flight data as is for migration
2807 void vhost_user_async_close(DeviceState
*d
,
2808 CharBackend
*chardev
, struct vhost_dev
*vhost
,
2809 vu_async_close_fn cb
,
2810 IOEventHandler
*event_cb
)
2812 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2814 * A close event may happen during a read/write, but vhost
2815 * code assumes the vhost_dev remains setup, so delay the
2818 AioContext
*ctx
= qemu_get_current_aio_context();
2819 VhostAsyncCallback
*data
= g_new0(VhostAsyncCallback
, 1);
2821 /* Save data for the callback */
2825 data
->vhost
= vhost
;
2826 data
->event_cb
= event_cb
;
2828 /* Disable any further notifications on the chardev */
2829 qemu_chr_fe_set_handlers(chardev
,
2830 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
2833 aio_bh_schedule_oneshot(ctx
, vhost_user_async_close_bh
, data
);
2836 * Move vhost device to the stopped state. The vhost-user device
2837 * will be clean up and disconnected in BH. This can be useful in
2838 * the vhost migration code. If disconnect was caught there is an
2839 * option for the general vhost code to get the dev state without
2840 * knowing its type (in this case vhost-user).
2842 * Note if the vhost device is fully cleared by the time we
2843 * execute the bottom half we won't continue with the cleanup.
2845 vhost
->started
= false;
2849 static int vhost_user_dev_start(struct vhost_dev
*dev
, bool started
)
2851 if (!virtio_has_feature(dev
->protocol_features
,
2852 VHOST_USER_PROTOCOL_F_STATUS
)) {
2856 /* Set device status only for last queue pair */
2857 if (dev
->vq_index
+ dev
->nvqs
!= dev
->vq_index_end
) {
2862 return vhost_user_add_status(dev
, VIRTIO_CONFIG_S_ACKNOWLEDGE
|
2863 VIRTIO_CONFIG_S_DRIVER
|
2864 VIRTIO_CONFIG_S_DRIVER_OK
);
2870 static void vhost_user_reset_status(struct vhost_dev
*dev
)
2872 /* Set device status only for last queue pair */
2873 if (dev
->vq_index
+ dev
->nvqs
!= dev
->vq_index_end
) {
2877 if (virtio_has_feature(dev
->protocol_features
,
2878 VHOST_USER_PROTOCOL_F_STATUS
)) {
2879 vhost_user_set_status(dev
, 0);
2883 static bool vhost_user_supports_device_state(struct vhost_dev
*dev
)
2885 return virtio_has_feature(dev
->protocol_features
,
2886 VHOST_USER_PROTOCOL_F_DEVICE_STATE
);
2889 static int vhost_user_set_device_state_fd(struct vhost_dev
*dev
,
2890 VhostDeviceStateDirection direction
,
2891 VhostDeviceStatePhase phase
,
2897 struct vhost_user
*vu
= dev
->opaque
;
2898 VhostUserMsg msg
= {
2900 .request
= VHOST_USER_SET_DEVICE_STATE_FD
,
2901 .flags
= VHOST_USER_VERSION
,
2902 .size
= sizeof(msg
.payload
.transfer_state
),
2904 .payload
.transfer_state
= {
2905 .direction
= direction
,
2912 if (!vhost_user_supports_device_state(dev
)) {
2914 error_setg(errp
, "Back-end does not support migration state transfer");
2918 ret
= vhost_user_write(dev
, &msg
, &fd
, 1);
2921 error_setg_errno(errp
, -ret
,
2922 "Failed to send SET_DEVICE_STATE_FD message");
2926 ret
= vhost_user_read(dev
, &msg
);
2928 error_setg_errno(errp
, -ret
,
2929 "Failed to receive SET_DEVICE_STATE_FD reply");
2933 if (msg
.hdr
.request
!= VHOST_USER_SET_DEVICE_STATE_FD
) {
2935 "Received unexpected message type, expected %d, received %d",
2936 VHOST_USER_SET_DEVICE_STATE_FD
, msg
.hdr
.request
);
2940 if (msg
.hdr
.size
!= sizeof(msg
.payload
.u64
)) {
2942 "Received bad message size, expected %zu, received %" PRIu32
,
2943 sizeof(msg
.payload
.u64
), msg
.hdr
.size
);
2947 if ((msg
.payload
.u64
& 0xff) != 0) {
2948 error_setg(errp
, "Back-end did not accept migration state transfer");
2952 if (!(msg
.payload
.u64
& VHOST_USER_VRING_NOFD_MASK
)) {
2953 *reply_fd
= qemu_chr_fe_get_msgfd(vu
->user
->chr
);
2954 if (*reply_fd
< 0) {
2956 "Failed to get back-end-provided transfer pipe FD");
2965 static int vhost_user_check_device_state(struct vhost_dev
*dev
, Error
**errp
)
2968 VhostUserMsg msg
= {
2970 .request
= VHOST_USER_CHECK_DEVICE_STATE
,
2971 .flags
= VHOST_USER_VERSION
,
2976 if (!vhost_user_supports_device_state(dev
)) {
2977 error_setg(errp
, "Back-end does not support migration state transfer");
2981 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2983 error_setg_errno(errp
, -ret
,
2984 "Failed to send CHECK_DEVICE_STATE message");
2988 ret
= vhost_user_read(dev
, &msg
);
2990 error_setg_errno(errp
, -ret
,
2991 "Failed to receive CHECK_DEVICE_STATE reply");
2995 if (msg
.hdr
.request
!= VHOST_USER_CHECK_DEVICE_STATE
) {
2997 "Received unexpected message type, expected %d, received %d",
2998 VHOST_USER_CHECK_DEVICE_STATE
, msg
.hdr
.request
);
3002 if (msg
.hdr
.size
!= sizeof(msg
.payload
.u64
)) {
3004 "Received bad message size, expected %zu, received %" PRIu32
,
3005 sizeof(msg
.payload
.u64
), msg
.hdr
.size
);
3009 if (msg
.payload
.u64
!= 0) {
3010 error_setg(errp
, "Back-end failed to process its internal state");
3017 const VhostOps user_ops
= {
3018 .backend_type
= VHOST_BACKEND_TYPE_USER
,
3019 .vhost_backend_init
= vhost_user_backend_init
,
3020 .vhost_backend_cleanup
= vhost_user_backend_cleanup
,
3021 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
3022 .vhost_backend_no_private_memslots
= vhost_user_no_private_memslots
,
3023 .vhost_set_log_base
= vhost_user_set_log_base
,
3024 .vhost_set_mem_table
= vhost_user_set_mem_table
,
3025 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
3026 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
3027 .vhost_set_vring_num
= vhost_user_set_vring_num
,
3028 .vhost_set_vring_base
= vhost_user_set_vring_base
,
3029 .vhost_get_vring_base
= vhost_user_get_vring_base
,
3030 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
3031 .vhost_set_vring_call
= vhost_user_set_vring_call
,
3032 .vhost_set_vring_err
= vhost_user_set_vring_err
,
3033 .vhost_set_features
= vhost_user_set_features
,
3034 .vhost_get_features
= vhost_user_get_features
,
3035 .vhost_set_owner
= vhost_user_set_owner
,
3036 .vhost_reset_device
= vhost_user_reset_device
,
3037 .vhost_get_vq_index
= vhost_user_get_vq_index
,
3038 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
3039 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
3040 .vhost_migration_done
= vhost_user_migration_done
,
3041 .vhost_net_set_mtu
= vhost_user_net_set_mtu
,
3042 .vhost_set_iotlb_callback
= vhost_user_set_iotlb_callback
,
3043 .vhost_send_device_iotlb_msg
= vhost_user_send_device_iotlb_msg
,
3044 .vhost_get_config
= vhost_user_get_config
,
3045 .vhost_set_config
= vhost_user_set_config
,
3046 .vhost_crypto_create_session
= vhost_user_crypto_create_session
,
3047 .vhost_crypto_close_session
= vhost_user_crypto_close_session
,
3048 .vhost_get_inflight_fd
= vhost_user_get_inflight_fd
,
3049 .vhost_set_inflight_fd
= vhost_user_set_inflight_fd
,
3050 .vhost_dev_start
= vhost_user_dev_start
,
3051 .vhost_reset_status
= vhost_user_reset_status
,
3052 .vhost_supports_device_state
= vhost_user_supports_device_state
,
3053 .vhost_set_device_state_fd
= vhost_user_set_device_state_fd
,
3054 .vhost_check_device_state
= vhost_user_check_device_state
,