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 "hw/virtio/vhost.h"
13 #include "hw/virtio/vhost-backend.h"
14 #include "hw/virtio/virtio-net.h"
15 #include "sysemu/char.h"
16 #include "sysemu/kvm.h"
17 #include "qemu/error-report.h"
18 #include "qemu/sockets.h"
19 #include "exec/ram_addr.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) {
191 qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
);
194 return qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
) == size
?
198 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
199 struct vhost_log
*log
)
201 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
203 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
204 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
206 .request
= VHOST_USER_SET_LOG_BASE
,
207 .flags
= VHOST_USER_VERSION
,
208 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
209 .payload
.log
.mmap_offset
= 0,
210 .size
= sizeof(msg
.payload
.log
),
213 if (shmfd
&& log
->fd
!= -1) {
214 fds
[fd_num
++] = log
->fd
;
217 vhost_user_write(dev
, &msg
, fds
, fd_num
);
221 if (vhost_user_read(dev
, &msg
) < 0) {
225 if (msg
.request
!= VHOST_USER_SET_LOG_BASE
) {
226 error_report("Received unexpected msg type. "
227 "Expected %d received %d",
228 VHOST_USER_SET_LOG_BASE
, msg
.request
);
236 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
237 struct vhost_memory
*mem
)
239 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
243 .request
= VHOST_USER_SET_MEM_TABLE
,
244 .flags
= VHOST_USER_VERSION
,
247 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
248 struct vhost_memory_region
*reg
= dev
->mem
->regions
+ i
;
251 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
252 qemu_ram_addr_from_host((void *)(uintptr_t)reg
->userspace_addr
,
254 fd
= qemu_get_ram_fd(ram_addr
);
256 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
= reg
->userspace_addr
;
257 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
258 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
= reg
->guest_phys_addr
;
259 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= reg
->userspace_addr
-
260 (uintptr_t) qemu_get_ram_block_host_ptr(ram_addr
);
261 assert(fd_num
< VHOST_MEMORY_MAX_NREGIONS
);
266 msg
.payload
.memory
.nregions
= fd_num
;
269 error_report("Failed initializing vhost-user memory map, "
270 "consider using -object memory-backend-file share=on");
274 msg
.size
= sizeof(msg
.payload
.memory
.nregions
);
275 msg
.size
+= sizeof(msg
.payload
.memory
.padding
);
276 msg
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
278 vhost_user_write(dev
, &msg
, fds
, fd_num
);
283 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
284 struct vhost_vring_addr
*addr
)
287 .request
= VHOST_USER_SET_VRING_ADDR
,
288 .flags
= VHOST_USER_VERSION
,
289 .payload
.addr
= *addr
,
290 .size
= sizeof(msg
.payload
.addr
),
293 vhost_user_write(dev
, &msg
, NULL
, 0);
298 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
299 struct vhost_vring_state
*ring
)
301 error_report("vhost-user trying to send unhandled ioctl");
305 static int vhost_set_vring(struct vhost_dev
*dev
,
306 unsigned long int request
,
307 struct vhost_vring_state
*ring
)
311 .flags
= VHOST_USER_VERSION
,
312 .payload
.state
= *ring
,
313 .size
= sizeof(msg
.payload
.state
),
316 vhost_user_write(dev
, &msg
, NULL
, 0);
321 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
322 struct vhost_vring_state
*ring
)
324 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
327 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
328 struct vhost_vring_state
*ring
)
330 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
333 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
337 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
341 for (i
= 0; i
< dev
->nvqs
; ++i
) {
342 struct vhost_vring_state state
= {
343 .index
= dev
->vq_index
+ i
,
347 vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
353 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
354 struct vhost_vring_state
*ring
)
357 .request
= VHOST_USER_GET_VRING_BASE
,
358 .flags
= VHOST_USER_VERSION
,
359 .payload
.state
= *ring
,
360 .size
= sizeof(msg
.payload
.state
),
363 vhost_user_write(dev
, &msg
, NULL
, 0);
365 if (vhost_user_read(dev
, &msg
) < 0) {
369 if (msg
.request
!= VHOST_USER_GET_VRING_BASE
) {
370 error_report("Received unexpected msg type. Expected %d received %d",
371 VHOST_USER_GET_VRING_BASE
, msg
.request
);
375 if (msg
.size
!= sizeof(msg
.payload
.state
)) {
376 error_report("Received bad msg size.");
380 *ring
= msg
.payload
.state
;
385 static int vhost_set_vring_file(struct vhost_dev
*dev
,
386 VhostUserRequest request
,
387 struct vhost_vring_file
*file
)
389 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
393 .flags
= VHOST_USER_VERSION
,
394 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
395 .size
= sizeof(msg
.payload
.u64
),
398 if (ioeventfd_enabled() && file
->fd
> 0) {
399 fds
[fd_num
++] = file
->fd
;
401 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
404 vhost_user_write(dev
, &msg
, fds
, fd_num
);
409 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
410 struct vhost_vring_file
*file
)
412 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
415 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
416 struct vhost_vring_file
*file
)
418 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
421 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
)
425 .flags
= VHOST_USER_VERSION
,
427 .size
= sizeof(msg
.payload
.u64
),
430 vhost_user_write(dev
, &msg
, NULL
, 0);
435 static int vhost_user_set_features(struct vhost_dev
*dev
,
438 return vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
, features
);
441 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
444 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
);
447 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
451 .flags
= VHOST_USER_VERSION
,
454 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
458 vhost_user_write(dev
, &msg
, NULL
, 0);
460 if (vhost_user_read(dev
, &msg
) < 0) {
464 if (msg
.request
!= request
) {
465 error_report("Received unexpected msg type. Expected %d received %d",
466 request
, msg
.request
);
470 if (msg
.size
!= sizeof(msg
.payload
.u64
)) {
471 error_report("Received bad msg size.");
475 *u64
= msg
.payload
.u64
;
480 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
482 return vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
);
485 static int vhost_user_set_owner(struct vhost_dev
*dev
)
488 .request
= VHOST_USER_SET_OWNER
,
489 .flags
= VHOST_USER_VERSION
,
492 vhost_user_write(dev
, &msg
, NULL
, 0);
497 static int vhost_user_reset_device(struct vhost_dev
*dev
)
500 .request
= VHOST_USER_RESET_OWNER
,
501 .flags
= VHOST_USER_VERSION
,
504 vhost_user_write(dev
, &msg
, NULL
, 0);
509 static int vhost_user_init(struct vhost_dev
*dev
, void *opaque
)
514 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
516 dev
->opaque
= opaque
;
518 err
= vhost_user_get_features(dev
, &features
);
523 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
524 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
526 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
532 dev
->protocol_features
= features
& VHOST_USER_PROTOCOL_FEATURE_MASK
;
533 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
538 /* query the max queues we support if backend supports Multiple Queue */
539 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
540 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
548 if (dev
->migration_blocker
== NULL
&&
549 !virtio_has_feature(dev
->protocol_features
,
550 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
551 error_setg(&dev
->migration_blocker
,
552 "Migration disabled: vhost-user backend lacks "
553 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
559 static int vhost_user_cleanup(struct vhost_dev
*dev
)
561 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
568 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
570 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
575 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
577 return VHOST_MEMORY_MAX_NREGIONS
;
580 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
582 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
584 return virtio_has_feature(dev
->protocol_features
,
585 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
588 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
590 VhostUserMsg msg
= { 0 };
593 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
595 /* If guest supports GUEST_ANNOUNCE do nothing */
596 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
600 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
601 if (virtio_has_feature(dev
->protocol_features
,
602 VHOST_USER_PROTOCOL_F_RARP
)) {
603 msg
.request
= VHOST_USER_SEND_RARP
;
604 msg
.flags
= VHOST_USER_VERSION
;
605 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
606 msg
.size
= sizeof(msg
.payload
.u64
);
608 err
= vhost_user_write(dev
, &msg
, NULL
, 0);
614 static bool vhost_user_can_merge(struct vhost_dev
*dev
,
615 uint64_t start1
, uint64_t size1
,
616 uint64_t start2
, uint64_t size2
)
622 mr
= qemu_ram_addr_from_host((void *)(uintptr_t)start1
, &ram_addr
);
624 mfd
= qemu_get_ram_fd(ram_addr
);
626 mr
= qemu_ram_addr_from_host((void *)(uintptr_t)start2
, &ram_addr
);
628 rfd
= qemu_get_ram_fd(ram_addr
);
633 const VhostOps user_ops
= {
634 .backend_type
= VHOST_BACKEND_TYPE_USER
,
635 .vhost_backend_init
= vhost_user_init
,
636 .vhost_backend_cleanup
= vhost_user_cleanup
,
637 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
638 .vhost_set_log_base
= vhost_user_set_log_base
,
639 .vhost_set_mem_table
= vhost_user_set_mem_table
,
640 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
641 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
642 .vhost_set_vring_num
= vhost_user_set_vring_num
,
643 .vhost_set_vring_base
= vhost_user_set_vring_base
,
644 .vhost_get_vring_base
= vhost_user_get_vring_base
,
645 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
646 .vhost_set_vring_call
= vhost_user_set_vring_call
,
647 .vhost_set_features
= vhost_user_set_features
,
648 .vhost_get_features
= vhost_user_get_features
,
649 .vhost_set_owner
= vhost_user_set_owner
,
650 .vhost_reset_device
= vhost_user_reset_device
,
651 .vhost_get_vq_index
= vhost_user_get_vq_index
,
652 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
653 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
654 .vhost_migration_done
= vhost_user_migration_done
,
655 .vhost_backend_can_merge
= vhost_user_can_merge
,