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
,
168 VhostUserMsg msg_reply
;
170 if ((msg
.flags
& VHOST_USER_NEED_REPLY_MASK
) == 0) {
174 if (vhost_user_read(dev
, &msg_reply
) < 0) {
178 if (msg_reply
.request
!= msg
.request
) {
179 error_report("Received unexpected msg type."
180 "Expected %d received %d",
181 msg
.request
, msg_reply
.request
);
185 return msg_reply
.payload
.u64
? -1 : 0;
188 static bool vhost_user_one_time_request(VhostUserRequest request
)
191 case VHOST_USER_SET_OWNER
:
192 case VHOST_USER_RESET_OWNER
:
193 case VHOST_USER_SET_MEM_TABLE
:
194 case VHOST_USER_GET_QUEUE_NUM
:
195 case VHOST_USER_NET_SET_MTU
:
202 /* most non-init callers ignore the error */
203 static int vhost_user_write(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
204 int *fds
, int fd_num
)
206 CharBackend
*chr
= dev
->opaque
;
207 int ret
, size
= VHOST_USER_HDR_SIZE
+ msg
->size
;
210 * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
211 * we just need send it once in the first time. For later such
212 * request, we just ignore it.
214 if (vhost_user_one_time_request(msg
->request
) && dev
->vq_index
!= 0) {
215 msg
->flags
&= ~VHOST_USER_NEED_REPLY_MASK
;
219 if (qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
) < 0) {
220 error_report("Failed to set msg fds.");
224 ret
= qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
);
226 error_report("Failed to write msg."
227 " Wrote %d instead of %d.", ret
, size
);
234 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
235 struct vhost_log
*log
)
237 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
239 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
240 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
242 .request
= VHOST_USER_SET_LOG_BASE
,
243 .flags
= VHOST_USER_VERSION
,
244 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
245 .payload
.log
.mmap_offset
= 0,
246 .size
= sizeof(msg
.payload
.log
),
249 if (shmfd
&& log
->fd
!= -1) {
250 fds
[fd_num
++] = log
->fd
;
253 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
259 if (vhost_user_read(dev
, &msg
) < 0) {
263 if (msg
.request
!= VHOST_USER_SET_LOG_BASE
) {
264 error_report("Received unexpected msg type. "
265 "Expected %d received %d",
266 VHOST_USER_SET_LOG_BASE
, msg
.request
);
274 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
275 struct vhost_memory
*mem
)
277 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
280 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
281 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
284 .request
= VHOST_USER_SET_MEM_TABLE
,
285 .flags
= VHOST_USER_VERSION
,
288 if (reply_supported
) {
289 msg
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
292 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
293 struct vhost_memory_region
*reg
= dev
->mem
->regions
+ i
;
297 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
298 mr
= memory_region_from_host((void *)(uintptr_t)reg
->userspace_addr
,
300 fd
= memory_region_get_fd(mr
);
302 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
= reg
->userspace_addr
;
303 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
304 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
= reg
->guest_phys_addr
;
305 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= offset
;
306 assert(fd_num
< VHOST_MEMORY_MAX_NREGIONS
);
311 msg
.payload
.memory
.nregions
= fd_num
;
314 error_report("Failed initializing vhost-user memory map, "
315 "consider using -object memory-backend-file share=on");
319 msg
.size
= sizeof(msg
.payload
.memory
.nregions
);
320 msg
.size
+= sizeof(msg
.payload
.memory
.padding
);
321 msg
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
323 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
327 if (reply_supported
) {
328 return process_message_reply(dev
, msg
);
334 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
335 struct vhost_vring_addr
*addr
)
338 .request
= VHOST_USER_SET_VRING_ADDR
,
339 .flags
= VHOST_USER_VERSION
,
340 .payload
.addr
= *addr
,
341 .size
= sizeof(msg
.payload
.addr
),
344 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
351 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
352 struct vhost_vring_state
*ring
)
354 error_report("vhost-user trying to send unhandled ioctl");
358 static int vhost_set_vring(struct vhost_dev
*dev
,
359 unsigned long int request
,
360 struct vhost_vring_state
*ring
)
364 .flags
= VHOST_USER_VERSION
,
365 .payload
.state
= *ring
,
366 .size
= sizeof(msg
.payload
.state
),
369 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
376 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
377 struct vhost_vring_state
*ring
)
379 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
382 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
383 struct vhost_vring_state
*ring
)
385 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
388 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
392 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
396 for (i
= 0; i
< dev
->nvqs
; ++i
) {
397 struct vhost_vring_state state
= {
398 .index
= dev
->vq_index
+ i
,
402 vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
408 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
409 struct vhost_vring_state
*ring
)
412 .request
= VHOST_USER_GET_VRING_BASE
,
413 .flags
= VHOST_USER_VERSION
,
414 .payload
.state
= *ring
,
415 .size
= sizeof(msg
.payload
.state
),
418 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
422 if (vhost_user_read(dev
, &msg
) < 0) {
426 if (msg
.request
!= VHOST_USER_GET_VRING_BASE
) {
427 error_report("Received unexpected msg type. Expected %d received %d",
428 VHOST_USER_GET_VRING_BASE
, msg
.request
);
432 if (msg
.size
!= sizeof(msg
.payload
.state
)) {
433 error_report("Received bad msg size.");
437 *ring
= msg
.payload
.state
;
442 static int vhost_set_vring_file(struct vhost_dev
*dev
,
443 VhostUserRequest request
,
444 struct vhost_vring_file
*file
)
446 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
450 .flags
= VHOST_USER_VERSION
,
451 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
452 .size
= sizeof(msg
.payload
.u64
),
455 if (ioeventfd_enabled() && file
->fd
> 0) {
456 fds
[fd_num
++] = file
->fd
;
458 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
461 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
468 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
469 struct vhost_vring_file
*file
)
471 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
474 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
475 struct vhost_vring_file
*file
)
477 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
480 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
)
484 .flags
= VHOST_USER_VERSION
,
486 .size
= sizeof(msg
.payload
.u64
),
489 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
496 static int vhost_user_set_features(struct vhost_dev
*dev
,
499 return vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
, features
);
502 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
505 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
);
508 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
512 .flags
= VHOST_USER_VERSION
,
515 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
519 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
523 if (vhost_user_read(dev
, &msg
) < 0) {
527 if (msg
.request
!= request
) {
528 error_report("Received unexpected msg type. Expected %d received %d",
529 request
, msg
.request
);
533 if (msg
.size
!= sizeof(msg
.payload
.u64
)) {
534 error_report("Received bad msg size.");
538 *u64
= msg
.payload
.u64
;
543 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
545 return vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
);
548 static int vhost_user_set_owner(struct vhost_dev
*dev
)
551 .request
= VHOST_USER_SET_OWNER
,
552 .flags
= VHOST_USER_VERSION
,
555 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
562 static int vhost_user_reset_device(struct vhost_dev
*dev
)
565 .request
= VHOST_USER_RESET_OWNER
,
566 .flags
= VHOST_USER_VERSION
,
569 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
576 static int vhost_user_init(struct vhost_dev
*dev
, void *opaque
)
581 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
583 dev
->opaque
= opaque
;
585 err
= vhost_user_get_features(dev
, &features
);
590 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
591 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
593 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
599 dev
->protocol_features
= features
& VHOST_USER_PROTOCOL_FEATURE_MASK
;
600 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
605 /* query the max queues we support if backend supports Multiple Queue */
606 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
607 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
615 if (dev
->migration_blocker
== NULL
&&
616 !virtio_has_feature(dev
->protocol_features
,
617 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
618 error_setg(&dev
->migration_blocker
,
619 "Migration disabled: vhost-user backend lacks "
620 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
626 static int vhost_user_cleanup(struct vhost_dev
*dev
)
628 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
635 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
637 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
642 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
644 return VHOST_MEMORY_MAX_NREGIONS
;
647 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
649 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
651 return virtio_has_feature(dev
->protocol_features
,
652 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
655 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
657 VhostUserMsg msg
= { 0 };
659 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
661 /* If guest supports GUEST_ANNOUNCE do nothing */
662 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
666 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
667 if (virtio_has_feature(dev
->protocol_features
,
668 VHOST_USER_PROTOCOL_F_RARP
)) {
669 msg
.request
= VHOST_USER_SEND_RARP
;
670 msg
.flags
= VHOST_USER_VERSION
;
671 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
672 msg
.size
= sizeof(msg
.payload
.u64
);
674 return vhost_user_write(dev
, &msg
, NULL
, 0);
679 static bool vhost_user_can_merge(struct vhost_dev
*dev
,
680 uint64_t start1
, uint64_t size1
,
681 uint64_t start2
, uint64_t size2
)
687 mr
= memory_region_from_host((void *)(uintptr_t)start1
, &offset
);
688 mfd
= memory_region_get_fd(mr
);
690 mr
= memory_region_from_host((void *)(uintptr_t)start2
, &offset
);
691 rfd
= memory_region_get_fd(mr
);
696 static int vhost_user_net_set_mtu(struct vhost_dev
*dev
, uint16_t mtu
)
699 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
700 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
702 if (!(dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU
))) {
706 msg
.request
= VHOST_USER_NET_SET_MTU
;
707 msg
.payload
.u64
= mtu
;
708 msg
.size
= sizeof(msg
.payload
.u64
);
709 msg
.flags
= VHOST_USER_VERSION
;
710 if (reply_supported
) {
711 msg
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
714 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
718 /* If reply_ack supported, slave has to ack specified MTU is valid */
719 if (reply_supported
) {
720 return process_message_reply(dev
, msg
);
726 const VhostOps user_ops
= {
727 .backend_type
= VHOST_BACKEND_TYPE_USER
,
728 .vhost_backend_init
= vhost_user_init
,
729 .vhost_backend_cleanup
= vhost_user_cleanup
,
730 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
731 .vhost_set_log_base
= vhost_user_set_log_base
,
732 .vhost_set_mem_table
= vhost_user_set_mem_table
,
733 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
734 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
735 .vhost_set_vring_num
= vhost_user_set_vring_num
,
736 .vhost_set_vring_base
= vhost_user_set_vring_base
,
737 .vhost_get_vring_base
= vhost_user_get_vring_base
,
738 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
739 .vhost_set_vring_call
= vhost_user_set_vring_call
,
740 .vhost_set_features
= vhost_user_set_features
,
741 .vhost_get_features
= vhost_user_get_features
,
742 .vhost_set_owner
= vhost_user_set_owner
,
743 .vhost_reset_device
= vhost_user_reset_device
,
744 .vhost_get_vq_index
= vhost_user_get_vq_index
,
745 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
746 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
747 .vhost_migration_done
= vhost_user_migration_done
,
748 .vhost_backend_can_merge
= vhost_user_can_merge
,
749 .vhost_net_set_mtu
= vhost_user_net_set_mtu
,