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 "exec/ram_addr.h"
21 #include "migration/migration.h"
23 #include <sys/ioctl.h>
24 #include <sys/socket.h>
26 #include <linux/vhost.h>
28 #define VHOST_MEMORY_MAX_NREGIONS 8
29 #define VHOST_USER_F_PROTOCOL_FEATURES 30
31 enum VhostUserProtocolFeature
{
32 VHOST_USER_PROTOCOL_F_MQ
= 0,
33 VHOST_USER_PROTOCOL_F_LOG_SHMFD
= 1,
34 VHOST_USER_PROTOCOL_F_RARP
= 2,
36 VHOST_USER_PROTOCOL_F_MAX
39 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
41 typedef enum VhostUserRequest
{
43 VHOST_USER_GET_FEATURES
= 1,
44 VHOST_USER_SET_FEATURES
= 2,
45 VHOST_USER_SET_OWNER
= 3,
46 VHOST_USER_RESET_OWNER
= 4,
47 VHOST_USER_SET_MEM_TABLE
= 5,
48 VHOST_USER_SET_LOG_BASE
= 6,
49 VHOST_USER_SET_LOG_FD
= 7,
50 VHOST_USER_SET_VRING_NUM
= 8,
51 VHOST_USER_SET_VRING_ADDR
= 9,
52 VHOST_USER_SET_VRING_BASE
= 10,
53 VHOST_USER_GET_VRING_BASE
= 11,
54 VHOST_USER_SET_VRING_KICK
= 12,
55 VHOST_USER_SET_VRING_CALL
= 13,
56 VHOST_USER_SET_VRING_ERR
= 14,
57 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
58 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
59 VHOST_USER_GET_QUEUE_NUM
= 17,
60 VHOST_USER_SET_VRING_ENABLE
= 18,
61 VHOST_USER_SEND_RARP
= 19,
65 typedef struct VhostUserMemoryRegion
{
66 uint64_t guest_phys_addr
;
68 uint64_t userspace_addr
;
70 } VhostUserMemoryRegion
;
72 typedef struct VhostUserMemory
{
75 VhostUserMemoryRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
78 typedef struct VhostUserLog
{
83 typedef struct VhostUserMsg
{
84 VhostUserRequest request
;
86 #define VHOST_USER_VERSION_MASK (0x3)
87 #define VHOST_USER_REPLY_MASK (0x1<<2)
89 uint32_t size
; /* the following payload size */
91 #define VHOST_USER_VRING_IDX_MASK (0xff)
92 #define VHOST_USER_VRING_NOFD_MASK (0x1<<8)
94 struct vhost_vring_state state
;
95 struct vhost_vring_addr addr
;
96 VhostUserMemory memory
;
99 } QEMU_PACKED VhostUserMsg
;
101 static VhostUserMsg m
__attribute__ ((unused
));
102 #define VHOST_USER_HDR_SIZE (sizeof(m.request) \
106 #define VHOST_USER_PAYLOAD_SIZE (sizeof(m) - VHOST_USER_HDR_SIZE)
108 /* The version of the protocol we support */
109 #define VHOST_USER_VERSION (0x1)
111 static bool ioeventfd_enabled(void)
113 return kvm_enabled() && kvm_eventfds_enabled();
116 static int vhost_user_read(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
118 CharDriverState
*chr
= dev
->opaque
;
119 uint8_t *p
= (uint8_t *) msg
;
120 int r
, size
= VHOST_USER_HDR_SIZE
;
122 r
= qemu_chr_fe_read_all(chr
, p
, size
);
124 error_report("Failed to read msg header. Read %d instead of %d."
125 " Original request %d.", r
, size
, msg
->request
);
129 /* validate received flags */
130 if (msg
->flags
!= (VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
)) {
131 error_report("Failed to read msg header."
132 " Flags 0x%x instead of 0x%x.", msg
->flags
,
133 VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
);
137 /* validate message size is sane */
138 if (msg
->size
> VHOST_USER_PAYLOAD_SIZE
) {
139 error_report("Failed to read msg header."
140 " Size %d exceeds the maximum %zu.", msg
->size
,
141 VHOST_USER_PAYLOAD_SIZE
);
146 p
+= VHOST_USER_HDR_SIZE
;
148 r
= qemu_chr_fe_read_all(chr
, p
, size
);
150 error_report("Failed to read msg payload."
151 " Read %d instead of %d.", r
, msg
->size
);
162 static bool vhost_user_one_time_request(VhostUserRequest request
)
165 case VHOST_USER_SET_OWNER
:
166 case VHOST_USER_RESET_OWNER
:
167 case VHOST_USER_SET_MEM_TABLE
:
168 case VHOST_USER_GET_QUEUE_NUM
:
175 /* most non-init callers ignore the error */
176 static int vhost_user_write(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
177 int *fds
, int fd_num
)
179 CharDriverState
*chr
= dev
->opaque
;
180 int size
= VHOST_USER_HDR_SIZE
+ msg
->size
;
183 * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
184 * we just need send it once in the first time. For later such
185 * request, we just ignore it.
187 if (vhost_user_one_time_request(msg
->request
) && dev
->vq_index
!= 0) {
192 qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
);
195 return qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
) == size
?
199 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
200 struct vhost_log
*log
)
202 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
204 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
205 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
207 .request
= VHOST_USER_SET_LOG_BASE
,
208 .flags
= VHOST_USER_VERSION
,
209 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
210 .payload
.log
.mmap_offset
= 0,
211 .size
= sizeof(msg
.payload
.log
),
214 if (shmfd
&& log
->fd
!= -1) {
215 fds
[fd_num
++] = log
->fd
;
218 vhost_user_write(dev
, &msg
, fds
, fd_num
);
222 if (vhost_user_read(dev
, &msg
) < 0) {
226 if (msg
.request
!= VHOST_USER_SET_LOG_BASE
) {
227 error_report("Received unexpected msg type. "
228 "Expected %d received %d",
229 VHOST_USER_SET_LOG_BASE
, msg
.request
);
237 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
238 struct vhost_memory
*mem
)
240 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
244 .request
= VHOST_USER_SET_MEM_TABLE
,
245 .flags
= VHOST_USER_VERSION
,
248 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
249 struct vhost_memory_region
*reg
= dev
->mem
->regions
+ i
;
253 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
254 mr
= qemu_ram_addr_from_host((void *)(uintptr_t)reg
->userspace_addr
,
256 fd
= memory_region_get_fd(mr
);
258 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
= reg
->userspace_addr
;
259 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
260 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
= reg
->guest_phys_addr
;
261 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= reg
->userspace_addr
-
262 (uintptr_t) memory_region_get_ram_ptr(mr
);
263 assert(fd_num
< VHOST_MEMORY_MAX_NREGIONS
);
268 msg
.payload
.memory
.nregions
= fd_num
;
271 error_report("Failed initializing vhost-user memory map, "
272 "consider using -object memory-backend-file share=on");
276 msg
.size
= sizeof(msg
.payload
.memory
.nregions
);
277 msg
.size
+= sizeof(msg
.payload
.memory
.padding
);
278 msg
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
280 vhost_user_write(dev
, &msg
, fds
, fd_num
);
285 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
286 struct vhost_vring_addr
*addr
)
289 .request
= VHOST_USER_SET_VRING_ADDR
,
290 .flags
= VHOST_USER_VERSION
,
291 .payload
.addr
= *addr
,
292 .size
= sizeof(msg
.payload
.addr
),
295 vhost_user_write(dev
, &msg
, NULL
, 0);
300 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
301 struct vhost_vring_state
*ring
)
303 error_report("vhost-user trying to send unhandled ioctl");
307 static int vhost_set_vring(struct vhost_dev
*dev
,
308 unsigned long int request
,
309 struct vhost_vring_state
*ring
)
313 .flags
= VHOST_USER_VERSION
,
314 .payload
.state
= *ring
,
315 .size
= sizeof(msg
.payload
.state
),
318 vhost_user_write(dev
, &msg
, NULL
, 0);
323 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
324 struct vhost_vring_state
*ring
)
326 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
329 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
330 struct vhost_vring_state
*ring
)
332 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
335 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
339 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
343 for (i
= 0; i
< dev
->nvqs
; ++i
) {
344 struct vhost_vring_state state
= {
345 .index
= dev
->vq_index
+ i
,
349 vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
355 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
356 struct vhost_vring_state
*ring
)
359 .request
= VHOST_USER_GET_VRING_BASE
,
360 .flags
= VHOST_USER_VERSION
,
361 .payload
.state
= *ring
,
362 .size
= sizeof(msg
.payload
.state
),
365 vhost_user_write(dev
, &msg
, NULL
, 0);
367 if (vhost_user_read(dev
, &msg
) < 0) {
371 if (msg
.request
!= VHOST_USER_GET_VRING_BASE
) {
372 error_report("Received unexpected msg type. Expected %d received %d",
373 VHOST_USER_GET_VRING_BASE
, msg
.request
);
377 if (msg
.size
!= sizeof(msg
.payload
.state
)) {
378 error_report("Received bad msg size.");
382 *ring
= msg
.payload
.state
;
387 static int vhost_set_vring_file(struct vhost_dev
*dev
,
388 VhostUserRequest request
,
389 struct vhost_vring_file
*file
)
391 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
395 .flags
= VHOST_USER_VERSION
,
396 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
397 .size
= sizeof(msg
.payload
.u64
),
400 if (ioeventfd_enabled() && file
->fd
> 0) {
401 fds
[fd_num
++] = file
->fd
;
403 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
406 vhost_user_write(dev
, &msg
, fds
, fd_num
);
411 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
412 struct vhost_vring_file
*file
)
414 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
417 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
418 struct vhost_vring_file
*file
)
420 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
423 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
)
427 .flags
= VHOST_USER_VERSION
,
429 .size
= sizeof(msg
.payload
.u64
),
432 vhost_user_write(dev
, &msg
, NULL
, 0);
437 static int vhost_user_set_features(struct vhost_dev
*dev
,
440 return vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
, features
);
443 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
446 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
);
449 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
453 .flags
= VHOST_USER_VERSION
,
456 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
460 vhost_user_write(dev
, &msg
, NULL
, 0);
462 if (vhost_user_read(dev
, &msg
) < 0) {
466 if (msg
.request
!= request
) {
467 error_report("Received unexpected msg type. Expected %d received %d",
468 request
, msg
.request
);
472 if (msg
.size
!= sizeof(msg
.payload
.u64
)) {
473 error_report("Received bad msg size.");
477 *u64
= msg
.payload
.u64
;
482 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
484 return vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
);
487 static int vhost_user_set_owner(struct vhost_dev
*dev
)
490 .request
= VHOST_USER_SET_OWNER
,
491 .flags
= VHOST_USER_VERSION
,
494 vhost_user_write(dev
, &msg
, NULL
, 0);
499 static int vhost_user_reset_device(struct vhost_dev
*dev
)
502 .request
= VHOST_USER_RESET_OWNER
,
503 .flags
= VHOST_USER_VERSION
,
506 vhost_user_write(dev
, &msg
, NULL
, 0);
511 static int vhost_user_init(struct vhost_dev
*dev
, void *opaque
)
516 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
518 dev
->opaque
= opaque
;
520 err
= vhost_user_get_features(dev
, &features
);
525 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
526 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
528 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
534 dev
->protocol_features
= features
& VHOST_USER_PROTOCOL_FEATURE_MASK
;
535 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
540 /* query the max queues we support if backend supports Multiple Queue */
541 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
542 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
550 if (dev
->migration_blocker
== NULL
&&
551 !virtio_has_feature(dev
->protocol_features
,
552 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
553 error_setg(&dev
->migration_blocker
,
554 "Migration disabled: vhost-user backend lacks "
555 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
561 static int vhost_user_cleanup(struct vhost_dev
*dev
)
563 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
570 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
572 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
577 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
579 return VHOST_MEMORY_MAX_NREGIONS
;
582 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
584 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
586 return virtio_has_feature(dev
->protocol_features
,
587 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
590 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
592 VhostUserMsg msg
= { 0 };
595 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
597 /* If guest supports GUEST_ANNOUNCE do nothing */
598 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
602 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
603 if (virtio_has_feature(dev
->protocol_features
,
604 VHOST_USER_PROTOCOL_F_RARP
)) {
605 msg
.request
= VHOST_USER_SEND_RARP
;
606 msg
.flags
= VHOST_USER_VERSION
;
607 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
608 msg
.size
= sizeof(msg
.payload
.u64
);
610 err
= vhost_user_write(dev
, &msg
, NULL
, 0);
616 static bool vhost_user_can_merge(struct vhost_dev
*dev
,
617 uint64_t start1
, uint64_t size1
,
618 uint64_t start2
, uint64_t size2
)
624 mr
= qemu_ram_addr_from_host((void *)(uintptr_t)start1
, &ram_addr
);
625 mfd
= memory_region_get_fd(mr
);
627 mr
= qemu_ram_addr_from_host((void *)(uintptr_t)start2
, &ram_addr
);
628 rfd
= memory_region_get_fd(mr
);
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
,