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,
34 VHOST_USER_PROTOCOL_F_REPLY_ACK
= 3,
35 VHOST_USER_PROTOCOL_F_NET_MTU
= 4,
37 VHOST_USER_PROTOCOL_F_MAX
40 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
42 typedef enum VhostUserRequest
{
44 VHOST_USER_GET_FEATURES
= 1,
45 VHOST_USER_SET_FEATURES
= 2,
46 VHOST_USER_SET_OWNER
= 3,
47 VHOST_USER_RESET_OWNER
= 4,
48 VHOST_USER_SET_MEM_TABLE
= 5,
49 VHOST_USER_SET_LOG_BASE
= 6,
50 VHOST_USER_SET_LOG_FD
= 7,
51 VHOST_USER_SET_VRING_NUM
= 8,
52 VHOST_USER_SET_VRING_ADDR
= 9,
53 VHOST_USER_SET_VRING_BASE
= 10,
54 VHOST_USER_GET_VRING_BASE
= 11,
55 VHOST_USER_SET_VRING_KICK
= 12,
56 VHOST_USER_SET_VRING_CALL
= 13,
57 VHOST_USER_SET_VRING_ERR
= 14,
58 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
59 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
60 VHOST_USER_GET_QUEUE_NUM
= 17,
61 VHOST_USER_SET_VRING_ENABLE
= 18,
62 VHOST_USER_SEND_RARP
= 19,
63 VHOST_USER_NET_SET_MTU
= 20,
67 typedef struct VhostUserMemoryRegion
{
68 uint64_t guest_phys_addr
;
70 uint64_t userspace_addr
;
72 } VhostUserMemoryRegion
;
74 typedef struct VhostUserMemory
{
77 VhostUserMemoryRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
80 typedef struct VhostUserLog
{
85 typedef struct VhostUserMsg
{
86 VhostUserRequest request
;
88 #define VHOST_USER_VERSION_MASK (0x3)
89 #define VHOST_USER_REPLY_MASK (0x1<<2)
90 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
92 uint32_t size
; /* the following payload size */
94 #define VHOST_USER_VRING_IDX_MASK (0xff)
95 #define VHOST_USER_VRING_NOFD_MASK (0x1<<8)
97 struct vhost_vring_state state
;
98 struct vhost_vring_addr addr
;
99 VhostUserMemory memory
;
102 } QEMU_PACKED VhostUserMsg
;
104 static VhostUserMsg m
__attribute__ ((unused
));
105 #define VHOST_USER_HDR_SIZE (sizeof(m.request) \
109 #define VHOST_USER_PAYLOAD_SIZE (sizeof(m) - VHOST_USER_HDR_SIZE)
111 /* The version of the protocol we support */
112 #define VHOST_USER_VERSION (0x1)
114 static bool ioeventfd_enabled(void)
116 return kvm_enabled() && kvm_eventfds_enabled();
119 static int vhost_user_read(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
121 CharBackend
*chr
= dev
->opaque
;
122 uint8_t *p
= (uint8_t *) msg
;
123 int r
, size
= VHOST_USER_HDR_SIZE
;
125 r
= qemu_chr_fe_read_all(chr
, p
, size
);
127 error_report("Failed to read msg header. Read %d instead of %d."
128 " Original request %d.", r
, size
, msg
->request
);
132 /* validate received flags */
133 if (msg
->flags
!= (VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
)) {
134 error_report("Failed to read msg header."
135 " Flags 0x%x instead of 0x%x.", msg
->flags
,
136 VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
);
140 /* validate message size is sane */
141 if (msg
->size
> VHOST_USER_PAYLOAD_SIZE
) {
142 error_report("Failed to read msg header."
143 " Size %d exceeds the maximum %zu.", msg
->size
,
144 VHOST_USER_PAYLOAD_SIZE
);
149 p
+= VHOST_USER_HDR_SIZE
;
151 r
= qemu_chr_fe_read_all(chr
, p
, size
);
153 error_report("Failed to read msg payload."
154 " Read %d instead of %d.", r
, msg
->size
);
165 static int process_message_reply(struct vhost_dev
*dev
,
166 VhostUserRequest request
)
170 if (vhost_user_read(dev
, &msg
) < 0) {
174 if (msg
.request
!= request
) {
175 error_report("Received unexpected msg type."
176 "Expected %d received %d",
177 request
, msg
.request
);
181 return msg
.payload
.u64
? -1 : 0;
184 static bool vhost_user_one_time_request(VhostUserRequest request
)
187 case VHOST_USER_SET_OWNER
:
188 case VHOST_USER_RESET_OWNER
:
189 case VHOST_USER_SET_MEM_TABLE
:
190 case VHOST_USER_GET_QUEUE_NUM
:
191 case VHOST_USER_NET_SET_MTU
:
198 /* most non-init callers ignore the error */
199 static int vhost_user_write(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
200 int *fds
, int fd_num
)
202 CharBackend
*chr
= dev
->opaque
;
203 int ret
, size
= VHOST_USER_HDR_SIZE
+ msg
->size
;
206 * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
207 * we just need send it once in the first time. For later such
208 * request, we just ignore it.
210 if (vhost_user_one_time_request(msg
->request
) && dev
->vq_index
!= 0) {
214 if (qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
) < 0) {
215 error_report("Failed to set msg fds.");
219 ret
= qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
);
221 error_report("Failed to write msg."
222 " Wrote %d instead of %d.", ret
, size
);
229 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
230 struct vhost_log
*log
)
232 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
234 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
235 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
237 .request
= VHOST_USER_SET_LOG_BASE
,
238 .flags
= VHOST_USER_VERSION
,
239 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
240 .payload
.log
.mmap_offset
= 0,
241 .size
= sizeof(msg
.payload
.log
),
244 if (shmfd
&& log
->fd
!= -1) {
245 fds
[fd_num
++] = log
->fd
;
248 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
254 if (vhost_user_read(dev
, &msg
) < 0) {
258 if (msg
.request
!= VHOST_USER_SET_LOG_BASE
) {
259 error_report("Received unexpected msg type. "
260 "Expected %d received %d",
261 VHOST_USER_SET_LOG_BASE
, msg
.request
);
269 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
270 struct vhost_memory
*mem
)
272 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
275 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
276 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
279 .request
= VHOST_USER_SET_MEM_TABLE
,
280 .flags
= VHOST_USER_VERSION
,
283 if (reply_supported
) {
284 msg
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
287 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
288 struct vhost_memory_region
*reg
= dev
->mem
->regions
+ i
;
292 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
293 mr
= memory_region_from_host((void *)(uintptr_t)reg
->userspace_addr
,
295 fd
= memory_region_get_fd(mr
);
297 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
= reg
->userspace_addr
;
298 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
299 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
= reg
->guest_phys_addr
;
300 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= offset
;
301 assert(fd_num
< VHOST_MEMORY_MAX_NREGIONS
);
306 msg
.payload
.memory
.nregions
= fd_num
;
309 error_report("Failed initializing vhost-user memory map, "
310 "consider using -object memory-backend-file share=on");
314 msg
.size
= sizeof(msg
.payload
.memory
.nregions
);
315 msg
.size
+= sizeof(msg
.payload
.memory
.padding
);
316 msg
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
318 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
322 if (reply_supported
) {
323 return process_message_reply(dev
, msg
.request
);
329 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
330 struct vhost_vring_addr
*addr
)
333 .request
= VHOST_USER_SET_VRING_ADDR
,
334 .flags
= VHOST_USER_VERSION
,
335 .payload
.addr
= *addr
,
336 .size
= sizeof(msg
.payload
.addr
),
339 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
346 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
347 struct vhost_vring_state
*ring
)
349 error_report("vhost-user trying to send unhandled ioctl");
353 static int vhost_set_vring(struct vhost_dev
*dev
,
354 unsigned long int request
,
355 struct vhost_vring_state
*ring
)
359 .flags
= VHOST_USER_VERSION
,
360 .payload
.state
= *ring
,
361 .size
= sizeof(msg
.payload
.state
),
364 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
371 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
372 struct vhost_vring_state
*ring
)
374 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
377 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
378 struct vhost_vring_state
*ring
)
380 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
383 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
387 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
391 for (i
= 0; i
< dev
->nvqs
; ++i
) {
392 struct vhost_vring_state state
= {
393 .index
= dev
->vq_index
+ i
,
397 vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
403 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
404 struct vhost_vring_state
*ring
)
407 .request
= VHOST_USER_GET_VRING_BASE
,
408 .flags
= VHOST_USER_VERSION
,
409 .payload
.state
= *ring
,
410 .size
= sizeof(msg
.payload
.state
),
413 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
417 if (vhost_user_read(dev
, &msg
) < 0) {
421 if (msg
.request
!= VHOST_USER_GET_VRING_BASE
) {
422 error_report("Received unexpected msg type. Expected %d received %d",
423 VHOST_USER_GET_VRING_BASE
, msg
.request
);
427 if (msg
.size
!= sizeof(msg
.payload
.state
)) {
428 error_report("Received bad msg size.");
432 *ring
= msg
.payload
.state
;
437 static int vhost_set_vring_file(struct vhost_dev
*dev
,
438 VhostUserRequest request
,
439 struct vhost_vring_file
*file
)
441 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
445 .flags
= VHOST_USER_VERSION
,
446 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
447 .size
= sizeof(msg
.payload
.u64
),
450 if (ioeventfd_enabled() && file
->fd
> 0) {
451 fds
[fd_num
++] = file
->fd
;
453 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
456 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
463 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
464 struct vhost_vring_file
*file
)
466 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
469 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
470 struct vhost_vring_file
*file
)
472 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
475 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
)
479 .flags
= VHOST_USER_VERSION
,
481 .size
= sizeof(msg
.payload
.u64
),
484 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
491 static int vhost_user_set_features(struct vhost_dev
*dev
,
494 return vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
, features
);
497 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
500 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
);
503 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
507 .flags
= VHOST_USER_VERSION
,
510 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
514 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
518 if (vhost_user_read(dev
, &msg
) < 0) {
522 if (msg
.request
!= request
) {
523 error_report("Received unexpected msg type. Expected %d received %d",
524 request
, msg
.request
);
528 if (msg
.size
!= sizeof(msg
.payload
.u64
)) {
529 error_report("Received bad msg size.");
533 *u64
= msg
.payload
.u64
;
538 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
540 return vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
);
543 static int vhost_user_set_owner(struct vhost_dev
*dev
)
546 .request
= VHOST_USER_SET_OWNER
,
547 .flags
= VHOST_USER_VERSION
,
550 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
557 static int vhost_user_reset_device(struct vhost_dev
*dev
)
560 .request
= VHOST_USER_RESET_OWNER
,
561 .flags
= VHOST_USER_VERSION
,
564 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
571 static int vhost_user_init(struct vhost_dev
*dev
, void *opaque
)
576 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
578 dev
->opaque
= opaque
;
580 err
= vhost_user_get_features(dev
, &features
);
585 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
586 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
588 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
594 dev
->protocol_features
= features
& VHOST_USER_PROTOCOL_FEATURE_MASK
;
595 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
600 /* query the max queues we support if backend supports Multiple Queue */
601 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
602 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
610 if (dev
->migration_blocker
== NULL
&&
611 !virtio_has_feature(dev
->protocol_features
,
612 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
613 error_setg(&dev
->migration_blocker
,
614 "Migration disabled: vhost-user backend lacks "
615 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
621 static int vhost_user_cleanup(struct vhost_dev
*dev
)
623 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
630 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
632 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
637 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
639 return VHOST_MEMORY_MAX_NREGIONS
;
642 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
644 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
646 return virtio_has_feature(dev
->protocol_features
,
647 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
650 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
652 VhostUserMsg msg
= { 0 };
654 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
656 /* If guest supports GUEST_ANNOUNCE do nothing */
657 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
661 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
662 if (virtio_has_feature(dev
->protocol_features
,
663 VHOST_USER_PROTOCOL_F_RARP
)) {
664 msg
.request
= VHOST_USER_SEND_RARP
;
665 msg
.flags
= VHOST_USER_VERSION
;
666 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
667 msg
.size
= sizeof(msg
.payload
.u64
);
669 return vhost_user_write(dev
, &msg
, NULL
, 0);
674 static bool vhost_user_can_merge(struct vhost_dev
*dev
,
675 uint64_t start1
, uint64_t size1
,
676 uint64_t start2
, uint64_t size2
)
682 mr
= memory_region_from_host((void *)(uintptr_t)start1
, &offset
);
683 mfd
= memory_region_get_fd(mr
);
685 mr
= memory_region_from_host((void *)(uintptr_t)start2
, &offset
);
686 rfd
= memory_region_get_fd(mr
);
691 static int vhost_user_net_set_mtu(struct vhost_dev
*dev
, uint16_t mtu
)
694 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
695 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
697 if (!(dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU
))) {
701 msg
.request
= VHOST_USER_NET_SET_MTU
;
702 msg
.payload
.u64
= mtu
;
703 msg
.size
= sizeof(msg
.payload
.u64
);
704 msg
.flags
= VHOST_USER_VERSION
;
705 if (reply_supported
) {
706 msg
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
709 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
713 /* If reply_ack supported, slave has to ack specified MTU is valid */
714 if (reply_supported
) {
715 return process_message_reply(dev
, msg
.request
);
721 const VhostOps user_ops
= {
722 .backend_type
= VHOST_BACKEND_TYPE_USER
,
723 .vhost_backend_init
= vhost_user_init
,
724 .vhost_backend_cleanup
= vhost_user_cleanup
,
725 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
726 .vhost_set_log_base
= vhost_user_set_log_base
,
727 .vhost_set_mem_table
= vhost_user_set_mem_table
,
728 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
729 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
730 .vhost_set_vring_num
= vhost_user_set_vring_num
,
731 .vhost_set_vring_base
= vhost_user_set_vring_base
,
732 .vhost_get_vring_base
= vhost_user_get_vring_base
,
733 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
734 .vhost_set_vring_call
= vhost_user_set_vring_call
,
735 .vhost_set_features
= vhost_user_set_features
,
736 .vhost_get_features
= vhost_user_get_features
,
737 .vhost_set_owner
= vhost_user_set_owner
,
738 .vhost_reset_device
= vhost_user_reset_device
,
739 .vhost_get_vq_index
= vhost_user_get_vq_index
,
740 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
741 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
742 .vhost_migration_done
= vhost_user_migration_done
,
743 .vhost_backend_can_merge
= vhost_user_can_merge
,
744 .vhost_net_set_mtu
= vhost_user_net_set_mtu
,