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 "hw/virtio/vhost.h"
12 #include "hw/virtio/vhost-backend.h"
13 #include "hw/virtio/virtio-net.h"
14 #include "sysemu/char.h"
15 #include "sysemu/kvm.h"
16 #include "qemu/error-report.h"
17 #include "qemu/sockets.h"
18 #include "exec/ram_addr.h"
19 #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.", r
,
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
,
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
;
252 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
253 qemu_ram_addr_from_host((void *)(uintptr_t)reg
->userspace_addr
,
255 fd
= qemu_get_ram_fd(ram_addr
);
257 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
= reg
->userspace_addr
;
258 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
259 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
= reg
->guest_phys_addr
;
260 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= reg
->userspace_addr
-
261 (uintptr_t) qemu_get_ram_block_host_ptr(ram_addr
);
262 assert(fd_num
< VHOST_MEMORY_MAX_NREGIONS
);
267 msg
.payload
.memory
.nregions
= fd_num
;
270 error_report("Failed initializing vhost-user memory map, "
271 "consider using -object memory-backend-file share=on");
275 msg
.size
= sizeof(msg
.payload
.memory
.nregions
);
276 msg
.size
+= sizeof(msg
.payload
.memory
.padding
);
277 msg
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
279 vhost_user_write(dev
, &msg
, fds
, fd_num
);
284 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
285 struct vhost_vring_addr
*addr
)
288 .request
= VHOST_USER_SET_VRING_ADDR
,
289 .flags
= VHOST_USER_VERSION
,
290 .payload
.addr
= *addr
,
291 .size
= sizeof(msg
.payload
.addr
),
294 vhost_user_write(dev
, &msg
, NULL
, 0);
299 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
300 struct vhost_vring_state
*ring
)
302 error_report("vhost-user trying to send unhandled ioctl");
306 static int vhost_set_vring(struct vhost_dev
*dev
,
307 unsigned long int request
,
308 struct vhost_vring_state
*ring
)
312 .flags
= VHOST_USER_VERSION
,
313 .payload
.state
= *ring
,
314 .size
= sizeof(msg
.payload
.state
),
317 vhost_user_write(dev
, &msg
, NULL
, 0);
322 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
323 struct vhost_vring_state
*ring
)
325 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
328 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
329 struct vhost_vring_state
*ring
)
331 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
334 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
336 struct vhost_vring_state state
= {
337 .index
= dev
->vq_index
,
341 if (!(dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
))) {
345 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
349 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
350 struct vhost_vring_state
*ring
)
353 .request
= VHOST_USER_GET_VRING_BASE
,
354 .flags
= VHOST_USER_VERSION
,
355 .payload
.state
= *ring
,
356 .size
= sizeof(msg
.payload
.state
),
359 vhost_user_write(dev
, &msg
, NULL
, 0);
361 if (vhost_user_read(dev
, &msg
) < 0) {
365 if (msg
.request
!= VHOST_USER_GET_VRING_BASE
) {
366 error_report("Received unexpected msg type. Expected %d received %d",
367 VHOST_USER_GET_VRING_BASE
, msg
.request
);
371 if (msg
.size
!= sizeof(msg
.payload
.state
)) {
372 error_report("Received bad msg size.");
376 *ring
= msg
.payload
.state
;
381 static int vhost_set_vring_file(struct vhost_dev
*dev
,
382 VhostUserRequest request
,
383 struct vhost_vring_file
*file
)
385 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
389 .flags
= VHOST_USER_VERSION
,
390 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
391 .size
= sizeof(msg
.payload
.u64
),
394 if (ioeventfd_enabled() && file
->fd
> 0) {
395 fds
[fd_num
++] = file
->fd
;
397 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
400 vhost_user_write(dev
, &msg
, fds
, fd_num
);
405 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
406 struct vhost_vring_file
*file
)
408 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
411 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
412 struct vhost_vring_file
*file
)
414 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
417 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
)
421 .flags
= VHOST_USER_VERSION
,
423 .size
= sizeof(msg
.payload
.u64
),
426 vhost_user_write(dev
, &msg
, NULL
, 0);
431 static int vhost_user_set_features(struct vhost_dev
*dev
,
434 return vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
, features
);
437 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
440 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
);
443 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
447 .flags
= VHOST_USER_VERSION
,
450 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
454 vhost_user_write(dev
, &msg
, NULL
, 0);
456 if (vhost_user_read(dev
, &msg
) < 0) {
460 if (msg
.request
!= request
) {
461 error_report("Received unexpected msg type. Expected %d received %d",
462 request
, msg
.request
);
466 if (msg
.size
!= sizeof(msg
.payload
.u64
)) {
467 error_report("Received bad msg size.");
471 *u64
= msg
.payload
.u64
;
476 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
478 return vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
);
481 static int vhost_user_set_owner(struct vhost_dev
*dev
)
484 .request
= VHOST_USER_SET_OWNER
,
485 .flags
= VHOST_USER_VERSION
,
488 vhost_user_write(dev
, &msg
, NULL
, 0);
493 static int vhost_user_reset_device(struct vhost_dev
*dev
)
496 .request
= VHOST_USER_RESET_OWNER
,
497 .flags
= VHOST_USER_VERSION
,
500 vhost_user_write(dev
, &msg
, NULL
, 0);
505 static int vhost_user_init(struct vhost_dev
*dev
, void *opaque
)
510 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
512 dev
->opaque
= opaque
;
514 err
= vhost_user_get_features(dev
, &features
);
519 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
520 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
522 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
528 dev
->protocol_features
= features
& VHOST_USER_PROTOCOL_FEATURE_MASK
;
529 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
534 /* query the max queues we support if backend supports Multiple Queue */
535 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
536 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
544 if (dev
->migration_blocker
== NULL
&&
545 !virtio_has_feature(dev
->protocol_features
,
546 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
547 error_setg(&dev
->migration_blocker
,
548 "Migration disabled: vhost-user backend lacks "
549 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
555 static int vhost_user_cleanup(struct vhost_dev
*dev
)
557 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
564 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
566 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
571 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
573 return VHOST_MEMORY_MAX_NREGIONS
;
576 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
578 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
580 return virtio_has_feature(dev
->protocol_features
,
581 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
584 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
586 VhostUserMsg msg
= { 0 };
589 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
591 /* If guest supports GUEST_ANNOUNCE do nothing */
592 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
596 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
597 if (virtio_has_feature(dev
->protocol_features
,
598 VHOST_USER_PROTOCOL_F_RARP
)) {
599 msg
.request
= VHOST_USER_SEND_RARP
;
600 msg
.flags
= VHOST_USER_VERSION
;
601 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
602 msg
.size
= sizeof(msg
.payload
.u64
);
604 err
= vhost_user_write(dev
, &msg
, NULL
, 0);
610 const VhostOps user_ops
= {
611 .backend_type
= VHOST_BACKEND_TYPE_USER
,
612 .vhost_backend_init
= vhost_user_init
,
613 .vhost_backend_cleanup
= vhost_user_cleanup
,
614 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
615 .vhost_set_log_base
= vhost_user_set_log_base
,
616 .vhost_set_mem_table
= vhost_user_set_mem_table
,
617 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
618 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
619 .vhost_set_vring_num
= vhost_user_set_vring_num
,
620 .vhost_set_vring_base
= vhost_user_set_vring_base
,
621 .vhost_get_vring_base
= vhost_user_get_vring_base
,
622 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
623 .vhost_set_vring_call
= vhost_user_set_vring_call
,
624 .vhost_set_features
= vhost_user_set_features
,
625 .vhost_get_features
= vhost_user_get_features
,
626 .vhost_set_owner
= vhost_user_set_owner
,
627 .vhost_reset_device
= vhost_user_reset_device
,
628 .vhost_get_vq_index
= vhost_user_get_vq_index
,
629 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
630 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
631 .vhost_migration_done
= vhost_user_migration_done
,