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/vhost.h"
14 #include "hw/virtio/vhost-backend.h"
15 #include "hw/virtio/virtio-net.h"
16 #include "sysemu/char.h"
17 #include "sysemu/kvm.h"
18 #include "qemu/error-report.h"
19 #include "qemu/sockets.h"
20 #include "migration/migration.h"
22 #include <sys/ioctl.h>
23 #include <sys/socket.h>
25 #include <linux/vhost.h>
27 #define VHOST_MEMORY_MAX_NREGIONS 8
28 #define VHOST_USER_F_PROTOCOL_FEATURES 30
30 enum VhostUserProtocolFeature
{
31 VHOST_USER_PROTOCOL_F_MQ
= 0,
32 VHOST_USER_PROTOCOL_F_LOG_SHMFD
= 1,
33 VHOST_USER_PROTOCOL_F_RARP
= 2,
35 VHOST_USER_PROTOCOL_F_MAX
38 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
40 typedef enum VhostUserRequest
{
42 VHOST_USER_GET_FEATURES
= 1,
43 VHOST_USER_SET_FEATURES
= 2,
44 VHOST_USER_SET_OWNER
= 3,
45 VHOST_USER_RESET_OWNER
= 4,
46 VHOST_USER_SET_MEM_TABLE
= 5,
47 VHOST_USER_SET_LOG_BASE
= 6,
48 VHOST_USER_SET_LOG_FD
= 7,
49 VHOST_USER_SET_VRING_NUM
= 8,
50 VHOST_USER_SET_VRING_ADDR
= 9,
51 VHOST_USER_SET_VRING_BASE
= 10,
52 VHOST_USER_GET_VRING_BASE
= 11,
53 VHOST_USER_SET_VRING_KICK
= 12,
54 VHOST_USER_SET_VRING_CALL
= 13,
55 VHOST_USER_SET_VRING_ERR
= 14,
56 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
57 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
58 VHOST_USER_GET_QUEUE_NUM
= 17,
59 VHOST_USER_SET_VRING_ENABLE
= 18,
60 VHOST_USER_SEND_RARP
= 19,
64 typedef struct VhostUserMemoryRegion
{
65 uint64_t guest_phys_addr
;
67 uint64_t userspace_addr
;
69 } VhostUserMemoryRegion
;
71 typedef struct VhostUserMemory
{
74 VhostUserMemoryRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
77 typedef struct VhostUserLog
{
82 typedef struct VhostUserMsg
{
83 VhostUserRequest request
;
85 #define VHOST_USER_VERSION_MASK (0x3)
86 #define VHOST_USER_REPLY_MASK (0x1<<2)
88 uint32_t size
; /* the following payload size */
90 #define VHOST_USER_VRING_IDX_MASK (0xff)
91 #define VHOST_USER_VRING_NOFD_MASK (0x1<<8)
93 struct vhost_vring_state state
;
94 struct vhost_vring_addr addr
;
95 VhostUserMemory memory
;
98 } QEMU_PACKED VhostUserMsg
;
100 static VhostUserMsg m
__attribute__ ((unused
));
101 #define VHOST_USER_HDR_SIZE (sizeof(m.request) \
105 #define VHOST_USER_PAYLOAD_SIZE (sizeof(m) - VHOST_USER_HDR_SIZE)
107 /* The version of the protocol we support */
108 #define VHOST_USER_VERSION (0x1)
110 static bool ioeventfd_enabled(void)
112 return kvm_enabled() && kvm_eventfds_enabled();
115 static int vhost_user_read(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
117 CharDriverState
*chr
= dev
->opaque
;
118 uint8_t *p
= (uint8_t *) msg
;
119 int r
, size
= VHOST_USER_HDR_SIZE
;
121 r
= qemu_chr_fe_read_all(chr
, p
, size
);
123 error_report("Failed to read msg header. Read %d instead of %d."
124 " Original request %d.", r
, size
, msg
->request
);
128 /* validate received flags */
129 if (msg
->flags
!= (VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
)) {
130 error_report("Failed to read msg header."
131 " Flags 0x%x instead of 0x%x.", msg
->flags
,
132 VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
);
136 /* validate message size is sane */
137 if (msg
->size
> VHOST_USER_PAYLOAD_SIZE
) {
138 error_report("Failed to read msg header."
139 " Size %d exceeds the maximum %zu.", msg
->size
,
140 VHOST_USER_PAYLOAD_SIZE
);
145 p
+= VHOST_USER_HDR_SIZE
;
147 r
= qemu_chr_fe_read_all(chr
, p
, size
);
149 error_report("Failed to read msg payload."
150 " Read %d instead of %d.", r
, msg
->size
);
161 static bool vhost_user_one_time_request(VhostUserRequest request
)
164 case VHOST_USER_SET_OWNER
:
165 case VHOST_USER_RESET_OWNER
:
166 case VHOST_USER_SET_MEM_TABLE
:
167 case VHOST_USER_GET_QUEUE_NUM
:
174 /* most non-init callers ignore the error */
175 static int vhost_user_write(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
176 int *fds
, int fd_num
)
178 CharDriverState
*chr
= dev
->opaque
;
179 int size
= VHOST_USER_HDR_SIZE
+ msg
->size
;
182 * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
183 * we just need send it once in the first time. For later such
184 * request, we just ignore it.
186 if (vhost_user_one_time_request(msg
->request
) && dev
->vq_index
!= 0) {
190 qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
);
192 return qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
) == size
?
196 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
197 struct vhost_log
*log
)
199 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
201 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
202 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
204 .request
= VHOST_USER_SET_LOG_BASE
,
205 .flags
= VHOST_USER_VERSION
,
206 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
207 .payload
.log
.mmap_offset
= 0,
208 .size
= sizeof(msg
.payload
.log
),
211 if (shmfd
&& log
->fd
!= -1) {
212 fds
[fd_num
++] = log
->fd
;
215 vhost_user_write(dev
, &msg
, fds
, fd_num
);
219 if (vhost_user_read(dev
, &msg
) < 0) {
223 if (msg
.request
!= VHOST_USER_SET_LOG_BASE
) {
224 error_report("Received unexpected msg type. "
225 "Expected %d received %d",
226 VHOST_USER_SET_LOG_BASE
, msg
.request
);
234 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
235 struct vhost_memory
*mem
)
237 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
241 .request
= VHOST_USER_SET_MEM_TABLE
,
242 .flags
= VHOST_USER_VERSION
,
245 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
246 struct vhost_memory_region
*reg
= dev
->mem
->regions
+ i
;
250 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
251 mr
= memory_region_from_host((void *)(uintptr_t)reg
->userspace_addr
,
253 fd
= memory_region_get_fd(mr
);
255 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
= reg
->userspace_addr
;
256 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
257 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
= reg
->guest_phys_addr
;
258 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= offset
;
259 assert(fd_num
< VHOST_MEMORY_MAX_NREGIONS
);
264 msg
.payload
.memory
.nregions
= fd_num
;
267 error_report("Failed initializing vhost-user memory map, "
268 "consider using -object memory-backend-file share=on");
272 msg
.size
= sizeof(msg
.payload
.memory
.nregions
);
273 msg
.size
+= sizeof(msg
.payload
.memory
.padding
);
274 msg
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
276 vhost_user_write(dev
, &msg
, fds
, fd_num
);
281 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
282 struct vhost_vring_addr
*addr
)
285 .request
= VHOST_USER_SET_VRING_ADDR
,
286 .flags
= VHOST_USER_VERSION
,
287 .payload
.addr
= *addr
,
288 .size
= sizeof(msg
.payload
.addr
),
291 vhost_user_write(dev
, &msg
, NULL
, 0);
296 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
297 struct vhost_vring_state
*ring
)
299 error_report("vhost-user trying to send unhandled ioctl");
303 static int vhost_set_vring(struct vhost_dev
*dev
,
304 unsigned long int request
,
305 struct vhost_vring_state
*ring
)
309 .flags
= VHOST_USER_VERSION
,
310 .payload
.state
= *ring
,
311 .size
= sizeof(msg
.payload
.state
),
314 vhost_user_write(dev
, &msg
, NULL
, 0);
319 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
320 struct vhost_vring_state
*ring
)
322 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
325 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
326 struct vhost_vring_state
*ring
)
328 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
331 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
335 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
339 for (i
= 0; i
< dev
->nvqs
; ++i
) {
340 struct vhost_vring_state state
= {
341 .index
= dev
->vq_index
+ i
,
345 vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
351 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
352 struct vhost_vring_state
*ring
)
355 .request
= VHOST_USER_GET_VRING_BASE
,
356 .flags
= VHOST_USER_VERSION
,
357 .payload
.state
= *ring
,
358 .size
= sizeof(msg
.payload
.state
),
361 vhost_user_write(dev
, &msg
, NULL
, 0);
363 if (vhost_user_read(dev
, &msg
) < 0) {
367 if (msg
.request
!= VHOST_USER_GET_VRING_BASE
) {
368 error_report("Received unexpected msg type. Expected %d received %d",
369 VHOST_USER_GET_VRING_BASE
, msg
.request
);
373 if (msg
.size
!= sizeof(msg
.payload
.state
)) {
374 error_report("Received bad msg size.");
378 *ring
= msg
.payload
.state
;
383 static int vhost_set_vring_file(struct vhost_dev
*dev
,
384 VhostUserRequest request
,
385 struct vhost_vring_file
*file
)
387 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
391 .flags
= VHOST_USER_VERSION
,
392 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
393 .size
= sizeof(msg
.payload
.u64
),
396 if (ioeventfd_enabled() && file
->fd
> 0) {
397 fds
[fd_num
++] = file
->fd
;
399 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
402 vhost_user_write(dev
, &msg
, fds
, fd_num
);
407 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
408 struct vhost_vring_file
*file
)
410 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
413 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
414 struct vhost_vring_file
*file
)
416 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
419 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
)
423 .flags
= VHOST_USER_VERSION
,
425 .size
= sizeof(msg
.payload
.u64
),
428 vhost_user_write(dev
, &msg
, NULL
, 0);
433 static int vhost_user_set_features(struct vhost_dev
*dev
,
436 return vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
, features
);
439 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
442 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
);
445 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
449 .flags
= VHOST_USER_VERSION
,
452 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
456 vhost_user_write(dev
, &msg
, NULL
, 0);
458 if (vhost_user_read(dev
, &msg
) < 0) {
462 if (msg
.request
!= request
) {
463 error_report("Received unexpected msg type. Expected %d received %d",
464 request
, msg
.request
);
468 if (msg
.size
!= sizeof(msg
.payload
.u64
)) {
469 error_report("Received bad msg size.");
473 *u64
= msg
.payload
.u64
;
478 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
480 return vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
);
483 static int vhost_user_set_owner(struct vhost_dev
*dev
)
486 .request
= VHOST_USER_SET_OWNER
,
487 .flags
= VHOST_USER_VERSION
,
490 vhost_user_write(dev
, &msg
, NULL
, 0);
495 static int vhost_user_reset_device(struct vhost_dev
*dev
)
498 .request
= VHOST_USER_RESET_OWNER
,
499 .flags
= VHOST_USER_VERSION
,
502 vhost_user_write(dev
, &msg
, NULL
, 0);
507 static int vhost_user_init(struct vhost_dev
*dev
, void *opaque
)
512 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
514 dev
->opaque
= opaque
;
516 err
= vhost_user_get_features(dev
, &features
);
521 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
522 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
524 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
530 dev
->protocol_features
= features
& VHOST_USER_PROTOCOL_FEATURE_MASK
;
531 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
536 /* query the max queues we support if backend supports Multiple Queue */
537 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
538 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
546 if (dev
->migration_blocker
== NULL
&&
547 !virtio_has_feature(dev
->protocol_features
,
548 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
549 error_setg(&dev
->migration_blocker
,
550 "Migration disabled: vhost-user backend lacks "
551 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
557 static int vhost_user_cleanup(struct vhost_dev
*dev
)
559 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
566 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
568 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
573 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
575 return VHOST_MEMORY_MAX_NREGIONS
;
578 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
580 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
582 return virtio_has_feature(dev
->protocol_features
,
583 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
586 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
588 VhostUserMsg msg
= { 0 };
591 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
593 /* If guest supports GUEST_ANNOUNCE do nothing */
594 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
598 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
599 if (virtio_has_feature(dev
->protocol_features
,
600 VHOST_USER_PROTOCOL_F_RARP
)) {
601 msg
.request
= VHOST_USER_SEND_RARP
;
602 msg
.flags
= VHOST_USER_VERSION
;
603 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
604 msg
.size
= sizeof(msg
.payload
.u64
);
606 err
= vhost_user_write(dev
, &msg
, NULL
, 0);
612 static bool vhost_user_can_merge(struct vhost_dev
*dev
,
613 uint64_t start1
, uint64_t size1
,
614 uint64_t start2
, uint64_t size2
)
620 mr
= memory_region_from_host((void *)(uintptr_t)start1
, &offset
);
621 mfd
= memory_region_get_fd(mr
);
623 mr
= memory_region_from_host((void *)(uintptr_t)start2
, &offset
);
624 rfd
= memory_region_get_fd(mr
);
629 const VhostOps user_ops
= {
630 .backend_type
= VHOST_BACKEND_TYPE_USER
,
631 .vhost_backend_init
= vhost_user_init
,
632 .vhost_backend_cleanup
= vhost_user_cleanup
,
633 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
634 .vhost_set_log_base
= vhost_user_set_log_base
,
635 .vhost_set_mem_table
= vhost_user_set_mem_table
,
636 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
637 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
638 .vhost_set_vring_num
= vhost_user_set_vring_num
,
639 .vhost_set_vring_base
= vhost_user_set_vring_base
,
640 .vhost_get_vring_base
= vhost_user_get_vring_base
,
641 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
642 .vhost_set_vring_call
= vhost_user_set_vring_call
,
643 .vhost_set_features
= vhost_user_set_features
,
644 .vhost_get_features
= vhost_user_get_features
,
645 .vhost_set_owner
= vhost_user_set_owner
,
646 .vhost_reset_device
= vhost_user_reset_device
,
647 .vhost_get_vq_index
= vhost_user_get_vq_index
,
648 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
649 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
650 .vhost_migration_done
= vhost_user_migration_done
,
651 .vhost_backend_can_merge
= vhost_user_can_merge
,