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-user.h"
15 #include "hw/virtio/vhost-backend.h"
16 #include "hw/virtio/virtio.h"
17 #include "hw/virtio/virtio-net.h"
18 #include "chardev/char-fe.h"
19 #include "sysemu/kvm.h"
20 #include "qemu/error-report.h"
21 #include "qemu/sockets.h"
22 #include "sysemu/cryptodev.h"
23 #include "migration/migration.h"
24 #include "migration/postcopy-ram.h"
27 #include <sys/ioctl.h>
28 #include <sys/socket.h>
31 #include "standard-headers/linux/vhost_types.h"
34 #include <linux/userfaultfd.h>
37 #define VHOST_MEMORY_MAX_NREGIONS 8
38 #define VHOST_USER_F_PROTOCOL_FEATURES 30
39 #define VHOST_USER_SLAVE_MAX_FDS 8
42 * Maximum size of virtio device config space
44 #define VHOST_USER_MAX_CONFIG_SIZE 256
46 enum VhostUserProtocolFeature
{
47 VHOST_USER_PROTOCOL_F_MQ
= 0,
48 VHOST_USER_PROTOCOL_F_LOG_SHMFD
= 1,
49 VHOST_USER_PROTOCOL_F_RARP
= 2,
50 VHOST_USER_PROTOCOL_F_REPLY_ACK
= 3,
51 VHOST_USER_PROTOCOL_F_NET_MTU
= 4,
52 VHOST_USER_PROTOCOL_F_SLAVE_REQ
= 5,
53 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
= 6,
54 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
= 7,
55 VHOST_USER_PROTOCOL_F_PAGEFAULT
= 8,
56 VHOST_USER_PROTOCOL_F_CONFIG
= 9,
57 VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD
= 10,
58 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
= 11,
59 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
= 12,
60 VHOST_USER_PROTOCOL_F_MAX
63 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
65 typedef enum VhostUserRequest
{
67 VHOST_USER_GET_FEATURES
= 1,
68 VHOST_USER_SET_FEATURES
= 2,
69 VHOST_USER_SET_OWNER
= 3,
70 VHOST_USER_RESET_OWNER
= 4,
71 VHOST_USER_SET_MEM_TABLE
= 5,
72 VHOST_USER_SET_LOG_BASE
= 6,
73 VHOST_USER_SET_LOG_FD
= 7,
74 VHOST_USER_SET_VRING_NUM
= 8,
75 VHOST_USER_SET_VRING_ADDR
= 9,
76 VHOST_USER_SET_VRING_BASE
= 10,
77 VHOST_USER_GET_VRING_BASE
= 11,
78 VHOST_USER_SET_VRING_KICK
= 12,
79 VHOST_USER_SET_VRING_CALL
= 13,
80 VHOST_USER_SET_VRING_ERR
= 14,
81 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
82 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
83 VHOST_USER_GET_QUEUE_NUM
= 17,
84 VHOST_USER_SET_VRING_ENABLE
= 18,
85 VHOST_USER_SEND_RARP
= 19,
86 VHOST_USER_NET_SET_MTU
= 20,
87 VHOST_USER_SET_SLAVE_REQ_FD
= 21,
88 VHOST_USER_IOTLB_MSG
= 22,
89 VHOST_USER_SET_VRING_ENDIAN
= 23,
90 VHOST_USER_GET_CONFIG
= 24,
91 VHOST_USER_SET_CONFIG
= 25,
92 VHOST_USER_CREATE_CRYPTO_SESSION
= 26,
93 VHOST_USER_CLOSE_CRYPTO_SESSION
= 27,
94 VHOST_USER_POSTCOPY_ADVISE
= 28,
95 VHOST_USER_POSTCOPY_LISTEN
= 29,
96 VHOST_USER_POSTCOPY_END
= 30,
97 VHOST_USER_GET_INFLIGHT_FD
= 31,
98 VHOST_USER_SET_INFLIGHT_FD
= 32,
102 typedef enum VhostUserSlaveRequest
{
103 VHOST_USER_SLAVE_NONE
= 0,
104 VHOST_USER_SLAVE_IOTLB_MSG
= 1,
105 VHOST_USER_SLAVE_CONFIG_CHANGE_MSG
= 2,
106 VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG
= 3,
108 } VhostUserSlaveRequest
;
110 typedef struct VhostUserMemoryRegion
{
111 uint64_t guest_phys_addr
;
112 uint64_t memory_size
;
113 uint64_t userspace_addr
;
114 uint64_t mmap_offset
;
115 } VhostUserMemoryRegion
;
117 typedef struct VhostUserMemory
{
120 VhostUserMemoryRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
123 typedef struct VhostUserLog
{
125 uint64_t mmap_offset
;
128 typedef struct VhostUserConfig
{
132 uint8_t region
[VHOST_USER_MAX_CONFIG_SIZE
];
135 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN 512
136 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN 64
138 typedef struct VhostUserCryptoSession
{
139 /* session id for success, -1 on errors */
141 CryptoDevBackendSymSessionInfo session_setup_data
;
142 uint8_t key
[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN
];
143 uint8_t auth_key
[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN
];
144 } VhostUserCryptoSession
;
146 static VhostUserConfig c
__attribute__ ((unused
));
147 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
151 typedef struct VhostUserVringArea
{
155 } VhostUserVringArea
;
157 typedef struct VhostUserInflight
{
159 uint64_t mmap_offset
;
165 VhostUserRequest request
;
167 #define VHOST_USER_VERSION_MASK (0x3)
168 #define VHOST_USER_REPLY_MASK (0x1<<2)
169 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
171 uint32_t size
; /* the following payload size */
172 } QEMU_PACKED VhostUserHeader
;
175 #define VHOST_USER_VRING_IDX_MASK (0xff)
176 #define VHOST_USER_VRING_NOFD_MASK (0x1<<8)
178 struct vhost_vring_state state
;
179 struct vhost_vring_addr addr
;
180 VhostUserMemory memory
;
182 struct vhost_iotlb_msg iotlb
;
183 VhostUserConfig config
;
184 VhostUserCryptoSession session
;
185 VhostUserVringArea area
;
186 VhostUserInflight inflight
;
189 typedef struct VhostUserMsg
{
191 VhostUserPayload payload
;
192 } QEMU_PACKED VhostUserMsg
;
194 static VhostUserMsg m
__attribute__ ((unused
));
195 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
197 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
199 /* The version of the protocol we support */
200 #define VHOST_USER_VERSION (0x1)
203 struct vhost_dev
*dev
;
204 /* Shared between vhost devs of the same virtio device */
205 VhostUserState
*user
;
207 NotifierWithReturn postcopy_notifier
;
208 struct PostCopyFD postcopy_fd
;
209 uint64_t postcopy_client_bases
[VHOST_MEMORY_MAX_NREGIONS
];
210 /* Length of the region_rb and region_rb_offset arrays */
211 size_t region_rb_len
;
212 /* RAMBlock associated with a given region */
213 RAMBlock
**region_rb
;
214 /* The offset from the start of the RAMBlock to the start of the
217 ram_addr_t
*region_rb_offset
;
219 /* True once we've entered postcopy_listen */
220 bool postcopy_listen
;
223 static bool ioeventfd_enabled(void)
225 return !kvm_enabled() || kvm_eventfds_enabled();
228 static int vhost_user_read_header(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
230 struct vhost_user
*u
= dev
->opaque
;
231 CharBackend
*chr
= u
->user
->chr
;
232 uint8_t *p
= (uint8_t *) msg
;
233 int r
, size
= VHOST_USER_HDR_SIZE
;
235 r
= qemu_chr_fe_read_all(chr
, p
, size
);
237 error_report("Failed to read msg header. Read %d instead of %d."
238 " Original request %d.", r
, size
, msg
->hdr
.request
);
242 /* validate received flags */
243 if (msg
->hdr
.flags
!= (VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
)) {
244 error_report("Failed to read msg header."
245 " Flags 0x%x instead of 0x%x.", msg
->hdr
.flags
,
246 VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
);
253 static int vhost_user_read(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
255 struct vhost_user
*u
= dev
->opaque
;
256 CharBackend
*chr
= u
->user
->chr
;
257 uint8_t *p
= (uint8_t *) msg
;
260 if (vhost_user_read_header(dev
, msg
) < 0) {
264 /* validate message size is sane */
265 if (msg
->hdr
.size
> VHOST_USER_PAYLOAD_SIZE
) {
266 error_report("Failed to read msg header."
267 " Size %d exceeds the maximum %zu.", msg
->hdr
.size
,
268 VHOST_USER_PAYLOAD_SIZE
);
273 p
+= VHOST_USER_HDR_SIZE
;
274 size
= msg
->hdr
.size
;
275 r
= qemu_chr_fe_read_all(chr
, p
, size
);
277 error_report("Failed to read msg payload."
278 " Read %d instead of %d.", r
, msg
->hdr
.size
);
286 static int process_message_reply(struct vhost_dev
*dev
,
287 const VhostUserMsg
*msg
)
289 VhostUserMsg msg_reply
;
291 if ((msg
->hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) == 0) {
295 if (vhost_user_read(dev
, &msg_reply
) < 0) {
299 if (msg_reply
.hdr
.request
!= msg
->hdr
.request
) {
300 error_report("Received unexpected msg type."
301 "Expected %d received %d",
302 msg
->hdr
.request
, msg_reply
.hdr
.request
);
306 return msg_reply
.payload
.u64
? -1 : 0;
309 static bool vhost_user_one_time_request(VhostUserRequest request
)
312 case VHOST_USER_SET_OWNER
:
313 case VHOST_USER_RESET_OWNER
:
314 case VHOST_USER_SET_MEM_TABLE
:
315 case VHOST_USER_GET_QUEUE_NUM
:
316 case VHOST_USER_NET_SET_MTU
:
323 /* most non-init callers ignore the error */
324 static int vhost_user_write(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
325 int *fds
, int fd_num
)
327 struct vhost_user
*u
= dev
->opaque
;
328 CharBackend
*chr
= u
->user
->chr
;
329 int ret
, size
= VHOST_USER_HDR_SIZE
+ msg
->hdr
.size
;
332 * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
333 * we just need send it once in the first time. For later such
334 * request, we just ignore it.
336 if (vhost_user_one_time_request(msg
->hdr
.request
) && dev
->vq_index
!= 0) {
337 msg
->hdr
.flags
&= ~VHOST_USER_NEED_REPLY_MASK
;
341 if (qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
) < 0) {
342 error_report("Failed to set msg fds.");
346 ret
= qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
);
348 error_report("Failed to write msg."
349 " Wrote %d instead of %d.", ret
, size
);
356 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
357 struct vhost_log
*log
)
359 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
361 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
362 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
364 .hdr
.request
= VHOST_USER_SET_LOG_BASE
,
365 .hdr
.flags
= VHOST_USER_VERSION
,
366 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
367 .payload
.log
.mmap_offset
= 0,
368 .hdr
.size
= sizeof(msg
.payload
.log
),
371 if (shmfd
&& log
->fd
!= -1) {
372 fds
[fd_num
++] = log
->fd
;
375 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
381 if (vhost_user_read(dev
, &msg
) < 0) {
385 if (msg
.hdr
.request
!= VHOST_USER_SET_LOG_BASE
) {
386 error_report("Received unexpected msg type. "
387 "Expected %d received %d",
388 VHOST_USER_SET_LOG_BASE
, msg
.hdr
.request
);
396 static int vhost_user_set_mem_table_postcopy(struct vhost_dev
*dev
,
397 struct vhost_memory
*mem
)
399 struct vhost_user
*u
= dev
->opaque
;
400 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
403 VhostUserMsg msg_reply
;
407 .hdr
.request
= VHOST_USER_SET_MEM_TABLE
,
408 .hdr
.flags
= VHOST_USER_VERSION
,
411 if (u
->region_rb_len
< dev
->mem
->nregions
) {
412 u
->region_rb
= g_renew(RAMBlock
*, u
->region_rb
, dev
->mem
->nregions
);
413 u
->region_rb_offset
= g_renew(ram_addr_t
, u
->region_rb_offset
,
415 memset(&(u
->region_rb
[u
->region_rb_len
]), '\0',
416 sizeof(RAMBlock
*) * (dev
->mem
->nregions
- u
->region_rb_len
));
417 memset(&(u
->region_rb_offset
[u
->region_rb_len
]), '\0',
418 sizeof(ram_addr_t
) * (dev
->mem
->nregions
- u
->region_rb_len
));
419 u
->region_rb_len
= dev
->mem
->nregions
;
422 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
423 struct vhost_memory_region
*reg
= dev
->mem
->regions
+ i
;
427 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
428 mr
= memory_region_from_host((void *)(uintptr_t)reg
->userspace_addr
,
430 fd
= memory_region_get_fd(mr
);
432 trace_vhost_user_set_mem_table_withfd(fd_num
, mr
->name
,
434 reg
->guest_phys_addr
,
435 reg
->userspace_addr
, offset
);
436 u
->region_rb_offset
[i
] = offset
;
437 u
->region_rb
[i
] = mr
->ram_block
;
438 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
=
440 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
441 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
=
442 reg
->guest_phys_addr
;
443 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= offset
;
444 assert(fd_num
< VHOST_MEMORY_MAX_NREGIONS
);
447 u
->region_rb_offset
[i
] = 0;
448 u
->region_rb
[i
] = NULL
;
452 msg
.payload
.memory
.nregions
= fd_num
;
455 error_report("Failed initializing vhost-user memory map, "
456 "consider using -object memory-backend-file share=on");
460 msg
.hdr
.size
= sizeof(msg
.payload
.memory
.nregions
);
461 msg
.hdr
.size
+= sizeof(msg
.payload
.memory
.padding
);
462 msg
.hdr
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
464 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
468 if (vhost_user_read(dev
, &msg_reply
) < 0) {
472 if (msg_reply
.hdr
.request
!= VHOST_USER_SET_MEM_TABLE
) {
473 error_report("%s: Received unexpected msg type."
474 "Expected %d received %d", __func__
,
475 VHOST_USER_SET_MEM_TABLE
, msg_reply
.hdr
.request
);
478 /* We're using the same structure, just reusing one of the
479 * fields, so it should be the same size.
481 if (msg_reply
.hdr
.size
!= msg
.hdr
.size
) {
482 error_report("%s: Unexpected size for postcopy reply "
483 "%d vs %d", __func__
, msg_reply
.hdr
.size
, msg
.hdr
.size
);
487 memset(u
->postcopy_client_bases
, 0,
488 sizeof(uint64_t) * VHOST_MEMORY_MAX_NREGIONS
);
490 /* They're in the same order as the regions that were sent
491 * but some of the regions were skipped (above) if they
494 for (msg_i
= 0, region_i
= 0;
495 region_i
< dev
->mem
->nregions
;
497 if (msg_i
< fd_num
&&
498 msg_reply
.payload
.memory
.regions
[msg_i
].guest_phys_addr
==
499 dev
->mem
->regions
[region_i
].guest_phys_addr
) {
500 u
->postcopy_client_bases
[region_i
] =
501 msg_reply
.payload
.memory
.regions
[msg_i
].userspace_addr
;
502 trace_vhost_user_set_mem_table_postcopy(
503 msg_reply
.payload
.memory
.regions
[msg_i
].userspace_addr
,
504 msg
.payload
.memory
.regions
[msg_i
].userspace_addr
,
509 if (msg_i
!= fd_num
) {
510 error_report("%s: postcopy reply not fully consumed "
512 __func__
, msg_i
, fd_num
);
515 /* Now we've registered this with the postcopy code, we ack to the client,
516 * because now we're in the position to be able to deal with any faults
519 /* TODO: Use this for failure cases as well with a bad value */
520 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
521 msg
.payload
.u64
= 0; /* OK */
522 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
529 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
530 struct vhost_memory
*mem
)
532 struct vhost_user
*u
= dev
->opaque
;
533 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
536 bool do_postcopy
= u
->postcopy_listen
&& u
->postcopy_fd
.handler
;
537 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
538 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
541 /* Postcopy has enough differences that it's best done in it's own
544 return vhost_user_set_mem_table_postcopy(dev
, mem
);
548 .hdr
.request
= VHOST_USER_SET_MEM_TABLE
,
549 .hdr
.flags
= VHOST_USER_VERSION
,
552 if (reply_supported
) {
553 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
556 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
557 struct vhost_memory_region
*reg
= dev
->mem
->regions
+ i
;
561 assert((uintptr_t)reg
->userspace_addr
== reg
->userspace_addr
);
562 mr
= memory_region_from_host((void *)(uintptr_t)reg
->userspace_addr
,
564 fd
= memory_region_get_fd(mr
);
566 if (fd_num
== VHOST_MEMORY_MAX_NREGIONS
) {
567 error_report("Failed preparing vhost-user memory table msg");
570 msg
.payload
.memory
.regions
[fd_num
].userspace_addr
=
572 msg
.payload
.memory
.regions
[fd_num
].memory_size
= reg
->memory_size
;
573 msg
.payload
.memory
.regions
[fd_num
].guest_phys_addr
=
574 reg
->guest_phys_addr
;
575 msg
.payload
.memory
.regions
[fd_num
].mmap_offset
= offset
;
580 msg
.payload
.memory
.nregions
= fd_num
;
583 error_report("Failed initializing vhost-user memory map, "
584 "consider using -object memory-backend-file share=on");
588 msg
.hdr
.size
= sizeof(msg
.payload
.memory
.nregions
);
589 msg
.hdr
.size
+= sizeof(msg
.payload
.memory
.padding
);
590 msg
.hdr
.size
+= fd_num
* sizeof(VhostUserMemoryRegion
);
592 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
596 if (reply_supported
) {
597 return process_message_reply(dev
, &msg
);
603 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
604 struct vhost_vring_addr
*addr
)
607 .hdr
.request
= VHOST_USER_SET_VRING_ADDR
,
608 .hdr
.flags
= VHOST_USER_VERSION
,
609 .payload
.addr
= *addr
,
610 .hdr
.size
= sizeof(msg
.payload
.addr
),
613 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
620 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
621 struct vhost_vring_state
*ring
)
623 bool cross_endian
= virtio_has_feature(dev
->protocol_features
,
624 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
);
626 .hdr
.request
= VHOST_USER_SET_VRING_ENDIAN
,
627 .hdr
.flags
= VHOST_USER_VERSION
,
628 .payload
.state
= *ring
,
629 .hdr
.size
= sizeof(msg
.payload
.state
),
633 error_report("vhost-user trying to send unhandled ioctl");
637 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
644 static int vhost_set_vring(struct vhost_dev
*dev
,
645 unsigned long int request
,
646 struct vhost_vring_state
*ring
)
649 .hdr
.request
= request
,
650 .hdr
.flags
= VHOST_USER_VERSION
,
651 .payload
.state
= *ring
,
652 .hdr
.size
= sizeof(msg
.payload
.state
),
655 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
662 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
663 struct vhost_vring_state
*ring
)
665 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
668 static void vhost_user_host_notifier_restore(struct vhost_dev
*dev
,
671 struct vhost_user
*u
= dev
->opaque
;
672 VhostUserHostNotifier
*n
= &u
->user
->notifier
[queue_idx
];
673 VirtIODevice
*vdev
= dev
->vdev
;
675 if (n
->addr
&& !n
->set
) {
676 virtio_queue_set_host_notifier_mr(vdev
, queue_idx
, &n
->mr
, true);
681 static void vhost_user_host_notifier_remove(struct vhost_dev
*dev
,
684 struct vhost_user
*u
= dev
->opaque
;
685 VhostUserHostNotifier
*n
= &u
->user
->notifier
[queue_idx
];
686 VirtIODevice
*vdev
= dev
->vdev
;
688 if (n
->addr
&& n
->set
) {
689 virtio_queue_set_host_notifier_mr(vdev
, queue_idx
, &n
->mr
, false);
694 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
695 struct vhost_vring_state
*ring
)
697 vhost_user_host_notifier_restore(dev
, ring
->index
);
699 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
702 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
706 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
710 for (i
= 0; i
< dev
->nvqs
; ++i
) {
711 struct vhost_vring_state state
= {
712 .index
= dev
->vq_index
+ i
,
716 vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
722 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
723 struct vhost_vring_state
*ring
)
726 .hdr
.request
= VHOST_USER_GET_VRING_BASE
,
727 .hdr
.flags
= VHOST_USER_VERSION
,
728 .payload
.state
= *ring
,
729 .hdr
.size
= sizeof(msg
.payload
.state
),
732 vhost_user_host_notifier_remove(dev
, ring
->index
);
734 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
738 if (vhost_user_read(dev
, &msg
) < 0) {
742 if (msg
.hdr
.request
!= VHOST_USER_GET_VRING_BASE
) {
743 error_report("Received unexpected msg type. Expected %d received %d",
744 VHOST_USER_GET_VRING_BASE
, msg
.hdr
.request
);
748 if (msg
.hdr
.size
!= sizeof(msg
.payload
.state
)) {
749 error_report("Received bad msg size.");
753 *ring
= msg
.payload
.state
;
758 static int vhost_set_vring_file(struct vhost_dev
*dev
,
759 VhostUserRequest request
,
760 struct vhost_vring_file
*file
)
762 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
765 .hdr
.request
= request
,
766 .hdr
.flags
= VHOST_USER_VERSION
,
767 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
768 .hdr
.size
= sizeof(msg
.payload
.u64
),
771 if (ioeventfd_enabled() && file
->fd
> 0) {
772 fds
[fd_num
++] = file
->fd
;
774 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
777 if (vhost_user_write(dev
, &msg
, fds
, fd_num
) < 0) {
784 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
785 struct vhost_vring_file
*file
)
787 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
790 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
791 struct vhost_vring_file
*file
)
793 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
796 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
)
799 .hdr
.request
= request
,
800 .hdr
.flags
= VHOST_USER_VERSION
,
802 .hdr
.size
= sizeof(msg
.payload
.u64
),
805 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
812 static int vhost_user_set_features(struct vhost_dev
*dev
,
815 return vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
, features
);
818 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
821 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
);
824 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
827 .hdr
.request
= request
,
828 .hdr
.flags
= VHOST_USER_VERSION
,
831 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
835 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
839 if (vhost_user_read(dev
, &msg
) < 0) {
843 if (msg
.hdr
.request
!= request
) {
844 error_report("Received unexpected msg type. Expected %d received %d",
845 request
, msg
.hdr
.request
);
849 if (msg
.hdr
.size
!= sizeof(msg
.payload
.u64
)) {
850 error_report("Received bad msg size.");
854 *u64
= msg
.payload
.u64
;
859 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
861 return vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
);
864 static int vhost_user_set_owner(struct vhost_dev
*dev
)
867 .hdr
.request
= VHOST_USER_SET_OWNER
,
868 .hdr
.flags
= VHOST_USER_VERSION
,
871 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
878 static int vhost_user_reset_device(struct vhost_dev
*dev
)
881 .hdr
.request
= VHOST_USER_RESET_OWNER
,
882 .hdr
.flags
= VHOST_USER_VERSION
,
885 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
892 static int vhost_user_slave_handle_config_change(struct vhost_dev
*dev
)
896 if (!dev
->config_ops
) {
900 if (dev
->config_ops
->vhost_dev_config_notifier
) {
901 ret
= dev
->config_ops
->vhost_dev_config_notifier(dev
);
907 static int vhost_user_slave_handle_vring_host_notifier(struct vhost_dev
*dev
,
908 VhostUserVringArea
*area
,
911 int queue_idx
= area
->u64
& VHOST_USER_VRING_IDX_MASK
;
912 size_t page_size
= qemu_real_host_page_size
;
913 struct vhost_user
*u
= dev
->opaque
;
914 VhostUserState
*user
= u
->user
;
915 VirtIODevice
*vdev
= dev
->vdev
;
916 VhostUserHostNotifier
*n
;
920 if (!virtio_has_feature(dev
->protocol_features
,
921 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
) ||
922 vdev
== NULL
|| queue_idx
>= virtio_get_num_queues(vdev
)) {
926 n
= &user
->notifier
[queue_idx
];
929 virtio_queue_set_host_notifier_mr(vdev
, queue_idx
, &n
->mr
, false);
930 object_unparent(OBJECT(&n
->mr
));
931 munmap(n
->addr
, page_size
);
935 if (area
->u64
& VHOST_USER_VRING_NOFD_MASK
) {
940 if (area
->size
!= page_size
) {
944 addr
= mmap(NULL
, page_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
946 if (addr
== MAP_FAILED
) {
950 name
= g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
952 memory_region_init_ram_device_ptr(&n
->mr
, OBJECT(vdev
), name
,
956 if (virtio_queue_set_host_notifier_mr(vdev
, queue_idx
, &n
->mr
, true)) {
957 munmap(addr
, page_size
);
967 static void slave_read(void *opaque
)
969 struct vhost_dev
*dev
= opaque
;
970 struct vhost_user
*u
= dev
->opaque
;
971 VhostUserHeader hdr
= { 0, };
972 VhostUserPayload payload
= { 0, };
976 int fd
[VHOST_USER_SLAVE_MAX_FDS
];
977 char control
[CMSG_SPACE(sizeof(fd
))];
978 struct cmsghdr
*cmsg
;
981 memset(&msgh
, 0, sizeof(msgh
));
984 msgh
.msg_control
= control
;
985 msgh
.msg_controllen
= sizeof(control
);
987 memset(fd
, -1, sizeof(fd
));
991 iov
.iov_len
= VHOST_USER_HDR_SIZE
;
994 size
= recvmsg(u
->slave_fd
, &msgh
, 0);
995 } while (size
< 0 && (errno
== EINTR
|| errno
== EAGAIN
));
997 if (size
!= VHOST_USER_HDR_SIZE
) {
998 error_report("Failed to read from slave.");
1002 if (msgh
.msg_flags
& MSG_CTRUNC
) {
1003 error_report("Truncated message.");
1007 for (cmsg
= CMSG_FIRSTHDR(&msgh
); cmsg
!= NULL
;
1008 cmsg
= CMSG_NXTHDR(&msgh
, cmsg
)) {
1009 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1010 cmsg
->cmsg_type
== SCM_RIGHTS
) {
1011 fdsize
= cmsg
->cmsg_len
- CMSG_LEN(0);
1012 memcpy(fd
, CMSG_DATA(cmsg
), fdsize
);
1017 if (hdr
.size
> VHOST_USER_PAYLOAD_SIZE
) {
1018 error_report("Failed to read msg header."
1019 " Size %d exceeds the maximum %zu.", hdr
.size
,
1020 VHOST_USER_PAYLOAD_SIZE
);
1026 size
= read(u
->slave_fd
, &payload
, hdr
.size
);
1027 } while (size
< 0 && (errno
== EINTR
|| errno
== EAGAIN
));
1029 if (size
!= hdr
.size
) {
1030 error_report("Failed to read payload from slave.");
1034 switch (hdr
.request
) {
1035 case VHOST_USER_SLAVE_IOTLB_MSG
:
1036 ret
= vhost_backend_handle_iotlb_msg(dev
, &payload
.iotlb
);
1038 case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG
:
1039 ret
= vhost_user_slave_handle_config_change(dev
);
1041 case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG
:
1042 ret
= vhost_user_slave_handle_vring_host_notifier(dev
, &payload
.area
,
1046 error_report("Received unexpected msg type.");
1050 /* Close the remaining file descriptors. */
1051 for (i
= 0; i
< fdsize
; i
++) {
1058 * REPLY_ACK feature handling. Other reply types has to be managed
1059 * directly in their request handlers.
1061 if (hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) {
1062 struct iovec iovec
[2];
1065 hdr
.flags
&= ~VHOST_USER_NEED_REPLY_MASK
;
1066 hdr
.flags
|= VHOST_USER_REPLY_MASK
;
1068 payload
.u64
= !!ret
;
1069 hdr
.size
= sizeof(payload
.u64
);
1071 iovec
[0].iov_base
= &hdr
;
1072 iovec
[0].iov_len
= VHOST_USER_HDR_SIZE
;
1073 iovec
[1].iov_base
= &payload
;
1074 iovec
[1].iov_len
= hdr
.size
;
1077 size
= writev(u
->slave_fd
, iovec
, ARRAY_SIZE(iovec
));
1078 } while (size
< 0 && (errno
== EINTR
|| errno
== EAGAIN
));
1080 if (size
!= VHOST_USER_HDR_SIZE
+ hdr
.size
) {
1081 error_report("Failed to send msg reply to slave.");
1089 qemu_set_fd_handler(u
->slave_fd
, NULL
, NULL
, NULL
);
1092 for (i
= 0; i
< fdsize
; i
++) {
1100 static int vhost_setup_slave_channel(struct vhost_dev
*dev
)
1102 VhostUserMsg msg
= {
1103 .hdr
.request
= VHOST_USER_SET_SLAVE_REQ_FD
,
1104 .hdr
.flags
= VHOST_USER_VERSION
,
1106 struct vhost_user
*u
= dev
->opaque
;
1108 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1109 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1111 if (!virtio_has_feature(dev
->protocol_features
,
1112 VHOST_USER_PROTOCOL_F_SLAVE_REQ
)) {
1116 if (socketpair(PF_UNIX
, SOCK_STREAM
, 0, sv
) == -1) {
1117 error_report("socketpair() failed");
1121 u
->slave_fd
= sv
[0];
1122 qemu_set_fd_handler(u
->slave_fd
, slave_read
, NULL
, dev
);
1124 if (reply_supported
) {
1125 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1128 ret
= vhost_user_write(dev
, &msg
, &sv
[1], 1);
1133 if (reply_supported
) {
1134 ret
= process_message_reply(dev
, &msg
);
1140 qemu_set_fd_handler(u
->slave_fd
, NULL
, NULL
, NULL
);
1150 * Called back from the postcopy fault thread when a fault is received on our
1152 * TODO: This is Linux specific
1154 static int vhost_user_postcopy_fault_handler(struct PostCopyFD
*pcfd
,
1157 struct vhost_dev
*dev
= pcfd
->data
;
1158 struct vhost_user
*u
= dev
->opaque
;
1159 struct uffd_msg
*msg
= ufd
;
1160 uint64_t faultaddr
= msg
->arg
.pagefault
.address
;
1161 RAMBlock
*rb
= NULL
;
1165 trace_vhost_user_postcopy_fault_handler(pcfd
->idstr
, faultaddr
,
1166 dev
->mem
->nregions
);
1167 for (i
= 0; i
< MIN(dev
->mem
->nregions
, u
->region_rb_len
); i
++) {
1168 trace_vhost_user_postcopy_fault_handler_loop(i
,
1169 u
->postcopy_client_bases
[i
], dev
->mem
->regions
[i
].memory_size
);
1170 if (faultaddr
>= u
->postcopy_client_bases
[i
]) {
1171 /* Ofset of the fault address in the vhost region */
1172 uint64_t region_offset
= faultaddr
- u
->postcopy_client_bases
[i
];
1173 if (region_offset
< dev
->mem
->regions
[i
].memory_size
) {
1174 rb_offset
= region_offset
+ u
->region_rb_offset
[i
];
1175 trace_vhost_user_postcopy_fault_handler_found(i
,
1176 region_offset
, rb_offset
);
1177 rb
= u
->region_rb
[i
];
1178 return postcopy_request_shared_page(pcfd
, rb
, faultaddr
,
1183 error_report("%s: Failed to find region for fault %" PRIx64
,
1184 __func__
, faultaddr
);
1188 static int vhost_user_postcopy_waker(struct PostCopyFD
*pcfd
, RAMBlock
*rb
,
1191 struct vhost_dev
*dev
= pcfd
->data
;
1192 struct vhost_user
*u
= dev
->opaque
;
1195 trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb
), offset
);
1200 /* Translate the offset into an address in the clients address space */
1201 for (i
= 0; i
< MIN(dev
->mem
->nregions
, u
->region_rb_len
); i
++) {
1202 if (u
->region_rb
[i
] == rb
&&
1203 offset
>= u
->region_rb_offset
[i
] &&
1204 offset
< (u
->region_rb_offset
[i
] +
1205 dev
->mem
->regions
[i
].memory_size
)) {
1206 uint64_t client_addr
= (offset
- u
->region_rb_offset
[i
]) +
1207 u
->postcopy_client_bases
[i
];
1208 trace_vhost_user_postcopy_waker_found(client_addr
);
1209 return postcopy_wake_shared(pcfd
, client_addr
, rb
);
1213 trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb
), offset
);
1219 * Called at the start of an inbound postcopy on reception of the
1222 static int vhost_user_postcopy_advise(struct vhost_dev
*dev
, Error
**errp
)
1225 struct vhost_user
*u
= dev
->opaque
;
1226 CharBackend
*chr
= u
->user
->chr
;
1228 VhostUserMsg msg
= {
1229 .hdr
.request
= VHOST_USER_POSTCOPY_ADVISE
,
1230 .hdr
.flags
= VHOST_USER_VERSION
,
1233 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1234 error_setg(errp
, "Failed to send postcopy_advise to vhost");
1238 if (vhost_user_read(dev
, &msg
) < 0) {
1239 error_setg(errp
, "Failed to get postcopy_advise reply from vhost");
1243 if (msg
.hdr
.request
!= VHOST_USER_POSTCOPY_ADVISE
) {
1244 error_setg(errp
, "Unexpected msg type. Expected %d received %d",
1245 VHOST_USER_POSTCOPY_ADVISE
, msg
.hdr
.request
);
1250 error_setg(errp
, "Received bad msg size.");
1253 ufd
= qemu_chr_fe_get_msgfd(chr
);
1255 error_setg(errp
, "%s: Failed to get ufd", __func__
);
1258 qemu_set_nonblock(ufd
);
1260 /* register ufd with userfault thread */
1261 u
->postcopy_fd
.fd
= ufd
;
1262 u
->postcopy_fd
.data
= dev
;
1263 u
->postcopy_fd
.handler
= vhost_user_postcopy_fault_handler
;
1264 u
->postcopy_fd
.waker
= vhost_user_postcopy_waker
;
1265 u
->postcopy_fd
.idstr
= "vhost-user"; /* Need to find unique name */
1266 postcopy_register_shared_ufd(&u
->postcopy_fd
);
1269 error_setg(errp
, "Postcopy not supported on non-Linux systems");
1275 * Called at the switch to postcopy on reception of the 'listen' command.
1277 static int vhost_user_postcopy_listen(struct vhost_dev
*dev
, Error
**errp
)
1279 struct vhost_user
*u
= dev
->opaque
;
1281 VhostUserMsg msg
= {
1282 .hdr
.request
= VHOST_USER_POSTCOPY_LISTEN
,
1283 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
1285 u
->postcopy_listen
= true;
1286 trace_vhost_user_postcopy_listen();
1287 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1288 error_setg(errp
, "Failed to send postcopy_listen to vhost");
1292 ret
= process_message_reply(dev
, &msg
);
1294 error_setg(errp
, "Failed to receive reply to postcopy_listen");
1302 * Called at the end of postcopy
1304 static int vhost_user_postcopy_end(struct vhost_dev
*dev
, Error
**errp
)
1306 VhostUserMsg msg
= {
1307 .hdr
.request
= VHOST_USER_POSTCOPY_END
,
1308 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
1311 struct vhost_user
*u
= dev
->opaque
;
1313 trace_vhost_user_postcopy_end_entry();
1314 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1315 error_setg(errp
, "Failed to send postcopy_end to vhost");
1319 ret
= process_message_reply(dev
, &msg
);
1321 error_setg(errp
, "Failed to receive reply to postcopy_end");
1324 postcopy_unregister_shared_ufd(&u
->postcopy_fd
);
1325 close(u
->postcopy_fd
.fd
);
1326 u
->postcopy_fd
.handler
= NULL
;
1328 trace_vhost_user_postcopy_end_exit();
1333 static int vhost_user_postcopy_notifier(NotifierWithReturn
*notifier
,
1336 struct PostcopyNotifyData
*pnd
= opaque
;
1337 struct vhost_user
*u
= container_of(notifier
, struct vhost_user
,
1339 struct vhost_dev
*dev
= u
->dev
;
1341 switch (pnd
->reason
) {
1342 case POSTCOPY_NOTIFY_PROBE
:
1343 if (!virtio_has_feature(dev
->protocol_features
,
1344 VHOST_USER_PROTOCOL_F_PAGEFAULT
)) {
1345 /* TODO: Get the device name into this error somehow */
1346 error_setg(pnd
->errp
,
1347 "vhost-user backend not capable of postcopy");
1352 case POSTCOPY_NOTIFY_INBOUND_ADVISE
:
1353 return vhost_user_postcopy_advise(dev
, pnd
->errp
);
1355 case POSTCOPY_NOTIFY_INBOUND_LISTEN
:
1356 return vhost_user_postcopy_listen(dev
, pnd
->errp
);
1358 case POSTCOPY_NOTIFY_INBOUND_END
:
1359 return vhost_user_postcopy_end(dev
, pnd
->errp
);
1362 /* We ignore notifications we don't know */
1369 static int vhost_user_backend_init(struct vhost_dev
*dev
, void *opaque
)
1371 uint64_t features
, protocol_features
;
1372 struct vhost_user
*u
;
1375 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
1377 u
= g_new0(struct vhost_user
, 1);
1383 err
= vhost_user_get_features(dev
, &features
);
1388 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
1389 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
1391 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
1392 &protocol_features
);
1397 dev
->protocol_features
=
1398 protocol_features
& VHOST_USER_PROTOCOL_FEATURE_MASK
;
1400 if (!dev
->config_ops
|| !dev
->config_ops
->vhost_dev_config_notifier
) {
1401 /* Don't acknowledge CONFIG feature if device doesn't support it */
1402 dev
->protocol_features
&= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG
);
1403 } else if (!(protocol_features
&
1404 (1ULL << VHOST_USER_PROTOCOL_F_CONFIG
))) {
1405 error_report("Device expects VHOST_USER_PROTOCOL_F_CONFIG "
1406 "but backend does not support it.");
1410 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
1415 /* query the max queues we support if backend supports Multiple Queue */
1416 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
1417 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
1424 if (virtio_has_feature(features
, VIRTIO_F_IOMMU_PLATFORM
) &&
1425 !(virtio_has_feature(dev
->protocol_features
,
1426 VHOST_USER_PROTOCOL_F_SLAVE_REQ
) &&
1427 virtio_has_feature(dev
->protocol_features
,
1428 VHOST_USER_PROTOCOL_F_REPLY_ACK
))) {
1429 error_report("IOMMU support requires reply-ack and "
1430 "slave-req protocol features.");
1435 if (dev
->migration_blocker
== NULL
&&
1436 !virtio_has_feature(dev
->protocol_features
,
1437 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
1438 error_setg(&dev
->migration_blocker
,
1439 "Migration disabled: vhost-user backend lacks "
1440 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
1443 err
= vhost_setup_slave_channel(dev
);
1448 u
->postcopy_notifier
.notify
= vhost_user_postcopy_notifier
;
1449 postcopy_add_notifier(&u
->postcopy_notifier
);
1454 static int vhost_user_backend_cleanup(struct vhost_dev
*dev
)
1456 struct vhost_user
*u
;
1458 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
1461 if (u
->postcopy_notifier
.notify
) {
1462 postcopy_remove_notifier(&u
->postcopy_notifier
);
1463 u
->postcopy_notifier
.notify
= NULL
;
1465 u
->postcopy_listen
= false;
1466 if (u
->postcopy_fd
.handler
) {
1467 postcopy_unregister_shared_ufd(&u
->postcopy_fd
);
1468 close(u
->postcopy_fd
.fd
);
1469 u
->postcopy_fd
.handler
= NULL
;
1471 if (u
->slave_fd
>= 0) {
1472 qemu_set_fd_handler(u
->slave_fd
, NULL
, NULL
, NULL
);
1476 g_free(u
->region_rb
);
1477 u
->region_rb
= NULL
;
1478 g_free(u
->region_rb_offset
);
1479 u
->region_rb_offset
= NULL
;
1480 u
->region_rb_len
= 0;
1487 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
1489 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
1494 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
1496 return VHOST_MEMORY_MAX_NREGIONS
;
1499 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
1501 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
1503 return virtio_has_feature(dev
->protocol_features
,
1504 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
1507 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
1509 VhostUserMsg msg
= { };
1511 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
1513 /* If guest supports GUEST_ANNOUNCE do nothing */
1514 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
1518 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
1519 if (virtio_has_feature(dev
->protocol_features
,
1520 VHOST_USER_PROTOCOL_F_RARP
)) {
1521 msg
.hdr
.request
= VHOST_USER_SEND_RARP
;
1522 msg
.hdr
.flags
= VHOST_USER_VERSION
;
1523 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
1524 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
1526 return vhost_user_write(dev
, &msg
, NULL
, 0);
1531 static bool vhost_user_can_merge(struct vhost_dev
*dev
,
1532 uint64_t start1
, uint64_t size1
,
1533 uint64_t start2
, uint64_t size2
)
1539 mr
= memory_region_from_host((void *)(uintptr_t)start1
, &offset
);
1540 mfd
= memory_region_get_fd(mr
);
1542 mr
= memory_region_from_host((void *)(uintptr_t)start2
, &offset
);
1543 rfd
= memory_region_get_fd(mr
);
1548 static int vhost_user_net_set_mtu(struct vhost_dev
*dev
, uint16_t mtu
)
1551 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1552 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1554 if (!(dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU
))) {
1558 msg
.hdr
.request
= VHOST_USER_NET_SET_MTU
;
1559 msg
.payload
.u64
= mtu
;
1560 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
1561 msg
.hdr
.flags
= VHOST_USER_VERSION
;
1562 if (reply_supported
) {
1563 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1566 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1570 /* If reply_ack supported, slave has to ack specified MTU is valid */
1571 if (reply_supported
) {
1572 return process_message_reply(dev
, &msg
);
1578 static int vhost_user_send_device_iotlb_msg(struct vhost_dev
*dev
,
1579 struct vhost_iotlb_msg
*imsg
)
1581 VhostUserMsg msg
= {
1582 .hdr
.request
= VHOST_USER_IOTLB_MSG
,
1583 .hdr
.size
= sizeof(msg
.payload
.iotlb
),
1584 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
1585 .payload
.iotlb
= *imsg
,
1588 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1592 return process_message_reply(dev
, &msg
);
1596 static void vhost_user_set_iotlb_callback(struct vhost_dev
*dev
, int enabled
)
1598 /* No-op as the receive channel is not dedicated to IOTLB messages. */
1601 static int vhost_user_get_config(struct vhost_dev
*dev
, uint8_t *config
,
1602 uint32_t config_len
)
1604 VhostUserMsg msg
= {
1605 .hdr
.request
= VHOST_USER_GET_CONFIG
,
1606 .hdr
.flags
= VHOST_USER_VERSION
,
1607 .hdr
.size
= VHOST_USER_CONFIG_HDR_SIZE
+ config_len
,
1610 if (!virtio_has_feature(dev
->protocol_features
,
1611 VHOST_USER_PROTOCOL_F_CONFIG
)) {
1615 if (config_len
> VHOST_USER_MAX_CONFIG_SIZE
) {
1619 msg
.payload
.config
.offset
= 0;
1620 msg
.payload
.config
.size
= config_len
;
1621 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1625 if (vhost_user_read(dev
, &msg
) < 0) {
1629 if (msg
.hdr
.request
!= VHOST_USER_GET_CONFIG
) {
1630 error_report("Received unexpected msg type. Expected %d received %d",
1631 VHOST_USER_GET_CONFIG
, msg
.hdr
.request
);
1635 if (msg
.hdr
.size
!= VHOST_USER_CONFIG_HDR_SIZE
+ config_len
) {
1636 error_report("Received bad msg size.");
1640 memcpy(config
, msg
.payload
.config
.region
, config_len
);
1645 static int vhost_user_set_config(struct vhost_dev
*dev
, const uint8_t *data
,
1646 uint32_t offset
, uint32_t size
, uint32_t flags
)
1649 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1650 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1652 VhostUserMsg msg
= {
1653 .hdr
.request
= VHOST_USER_SET_CONFIG
,
1654 .hdr
.flags
= VHOST_USER_VERSION
,
1655 .hdr
.size
= VHOST_USER_CONFIG_HDR_SIZE
+ size
,
1658 if (!virtio_has_feature(dev
->protocol_features
,
1659 VHOST_USER_PROTOCOL_F_CONFIG
)) {
1663 if (reply_supported
) {
1664 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1667 if (size
> VHOST_USER_MAX_CONFIG_SIZE
) {
1671 msg
.payload
.config
.offset
= offset
,
1672 msg
.payload
.config
.size
= size
,
1673 msg
.payload
.config
.flags
= flags
,
1674 p
= msg
.payload
.config
.region
;
1675 memcpy(p
, data
, size
);
1677 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1681 if (reply_supported
) {
1682 return process_message_reply(dev
, &msg
);
1688 static int vhost_user_crypto_create_session(struct vhost_dev
*dev
,
1690 uint64_t *session_id
)
1692 bool crypto_session
= virtio_has_feature(dev
->protocol_features
,
1693 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
);
1694 CryptoDevBackendSymSessionInfo
*sess_info
= session_info
;
1695 VhostUserMsg msg
= {
1696 .hdr
.request
= VHOST_USER_CREATE_CRYPTO_SESSION
,
1697 .hdr
.flags
= VHOST_USER_VERSION
,
1698 .hdr
.size
= sizeof(msg
.payload
.session
),
1701 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
1703 if (!crypto_session
) {
1704 error_report("vhost-user trying to send unhandled ioctl");
1708 memcpy(&msg
.payload
.session
.session_setup_data
, sess_info
,
1709 sizeof(CryptoDevBackendSymSessionInfo
));
1710 if (sess_info
->key_len
) {
1711 memcpy(&msg
.payload
.session
.key
, sess_info
->cipher_key
,
1712 sess_info
->key_len
);
1714 if (sess_info
->auth_key_len
> 0) {
1715 memcpy(&msg
.payload
.session
.auth_key
, sess_info
->auth_key
,
1716 sess_info
->auth_key_len
);
1718 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1719 error_report("vhost_user_write() return -1, create session failed");
1723 if (vhost_user_read(dev
, &msg
) < 0) {
1724 error_report("vhost_user_read() return -1, create session failed");
1728 if (msg
.hdr
.request
!= VHOST_USER_CREATE_CRYPTO_SESSION
) {
1729 error_report("Received unexpected msg type. Expected %d received %d",
1730 VHOST_USER_CREATE_CRYPTO_SESSION
, msg
.hdr
.request
);
1734 if (msg
.hdr
.size
!= sizeof(msg
.payload
.session
)) {
1735 error_report("Received bad msg size.");
1739 if (msg
.payload
.session
.session_id
< 0) {
1740 error_report("Bad session id: %" PRId64
"",
1741 msg
.payload
.session
.session_id
);
1744 *session_id
= msg
.payload
.session
.session_id
;
1750 vhost_user_crypto_close_session(struct vhost_dev
*dev
, uint64_t session_id
)
1752 bool crypto_session
= virtio_has_feature(dev
->protocol_features
,
1753 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
);
1754 VhostUserMsg msg
= {
1755 .hdr
.request
= VHOST_USER_CLOSE_CRYPTO_SESSION
,
1756 .hdr
.flags
= VHOST_USER_VERSION
,
1757 .hdr
.size
= sizeof(msg
.payload
.u64
),
1759 msg
.payload
.u64
= session_id
;
1761 if (!crypto_session
) {
1762 error_report("vhost-user trying to send unhandled ioctl");
1766 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1767 error_report("vhost_user_write() return -1, close session failed");
1774 static bool vhost_user_mem_section_filter(struct vhost_dev
*dev
,
1775 MemoryRegionSection
*section
)
1779 result
= memory_region_get_fd(section
->mr
) >= 0;
1784 static int vhost_user_get_inflight_fd(struct vhost_dev
*dev
,
1785 uint16_t queue_size
,
1786 struct vhost_inflight
*inflight
)
1790 struct vhost_user
*u
= dev
->opaque
;
1791 CharBackend
*chr
= u
->user
->chr
;
1792 VhostUserMsg msg
= {
1793 .hdr
.request
= VHOST_USER_GET_INFLIGHT_FD
,
1794 .hdr
.flags
= VHOST_USER_VERSION
,
1795 .payload
.inflight
.num_queues
= dev
->nvqs
,
1796 .payload
.inflight
.queue_size
= queue_size
,
1797 .hdr
.size
= sizeof(msg
.payload
.inflight
),
1800 if (!virtio_has_feature(dev
->protocol_features
,
1801 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
)) {
1805 if (vhost_user_write(dev
, &msg
, NULL
, 0) < 0) {
1809 if (vhost_user_read(dev
, &msg
) < 0) {
1813 if (msg
.hdr
.request
!= VHOST_USER_GET_INFLIGHT_FD
) {
1814 error_report("Received unexpected msg type. "
1815 "Expected %d received %d",
1816 VHOST_USER_GET_INFLIGHT_FD
, msg
.hdr
.request
);
1820 if (msg
.hdr
.size
!= sizeof(msg
.payload
.inflight
)) {
1821 error_report("Received bad msg size.");
1825 if (!msg
.payload
.inflight
.mmap_size
) {
1829 fd
= qemu_chr_fe_get_msgfd(chr
);
1831 error_report("Failed to get mem fd");
1835 addr
= mmap(0, msg
.payload
.inflight
.mmap_size
, PROT_READ
| PROT_WRITE
,
1836 MAP_SHARED
, fd
, msg
.payload
.inflight
.mmap_offset
);
1838 if (addr
== MAP_FAILED
) {
1839 error_report("Failed to mmap mem fd");
1844 inflight
->addr
= addr
;
1846 inflight
->size
= msg
.payload
.inflight
.mmap_size
;
1847 inflight
->offset
= msg
.payload
.inflight
.mmap_offset
;
1848 inflight
->queue_size
= queue_size
;
1853 static int vhost_user_set_inflight_fd(struct vhost_dev
*dev
,
1854 struct vhost_inflight
*inflight
)
1856 VhostUserMsg msg
= {
1857 .hdr
.request
= VHOST_USER_SET_INFLIGHT_FD
,
1858 .hdr
.flags
= VHOST_USER_VERSION
,
1859 .payload
.inflight
.mmap_size
= inflight
->size
,
1860 .payload
.inflight
.mmap_offset
= inflight
->offset
,
1861 .payload
.inflight
.num_queues
= dev
->nvqs
,
1862 .payload
.inflight
.queue_size
= inflight
->queue_size
,
1863 .hdr
.size
= sizeof(msg
.payload
.inflight
),
1866 if (!virtio_has_feature(dev
->protocol_features
,
1867 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
)) {
1871 if (vhost_user_write(dev
, &msg
, &inflight
->fd
, 1) < 0) {
1878 bool vhost_user_init(VhostUserState
*user
, CharBackend
*chr
, Error
**errp
)
1881 error_setg(errp
, "Cannot initialize vhost-user state");
1888 void vhost_user_cleanup(VhostUserState
*user
)
1896 for (i
= 0; i
< VIRTIO_QUEUE_MAX
; i
++) {
1897 if (user
->notifier
[i
].addr
) {
1898 object_unparent(OBJECT(&user
->notifier
[i
].mr
));
1899 munmap(user
->notifier
[i
].addr
, qemu_real_host_page_size
);
1900 user
->notifier
[i
].addr
= NULL
;
1906 const VhostOps user_ops
= {
1907 .backend_type
= VHOST_BACKEND_TYPE_USER
,
1908 .vhost_backend_init
= vhost_user_backend_init
,
1909 .vhost_backend_cleanup
= vhost_user_backend_cleanup
,
1910 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
1911 .vhost_set_log_base
= vhost_user_set_log_base
,
1912 .vhost_set_mem_table
= vhost_user_set_mem_table
,
1913 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
1914 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
1915 .vhost_set_vring_num
= vhost_user_set_vring_num
,
1916 .vhost_set_vring_base
= vhost_user_set_vring_base
,
1917 .vhost_get_vring_base
= vhost_user_get_vring_base
,
1918 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
1919 .vhost_set_vring_call
= vhost_user_set_vring_call
,
1920 .vhost_set_features
= vhost_user_set_features
,
1921 .vhost_get_features
= vhost_user_get_features
,
1922 .vhost_set_owner
= vhost_user_set_owner
,
1923 .vhost_reset_device
= vhost_user_reset_device
,
1924 .vhost_get_vq_index
= vhost_user_get_vq_index
,
1925 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
1926 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
1927 .vhost_migration_done
= vhost_user_migration_done
,
1928 .vhost_backend_can_merge
= vhost_user_can_merge
,
1929 .vhost_net_set_mtu
= vhost_user_net_set_mtu
,
1930 .vhost_set_iotlb_callback
= vhost_user_set_iotlb_callback
,
1931 .vhost_send_device_iotlb_msg
= vhost_user_send_device_iotlb_msg
,
1932 .vhost_get_config
= vhost_user_get_config
,
1933 .vhost_set_config
= vhost_user_set_config
,
1934 .vhost_crypto_create_session
= vhost_user_crypto_create_session
,
1935 .vhost_crypto_close_session
= vhost_user_crypto_close_session
,
1936 .vhost_backend_mem_section_filter
= vhost_user_mem_section_filter
,
1937 .vhost_get_inflight_fd
= vhost_user_get_inflight_fd
,
1938 .vhost_set_inflight_fd
= vhost_user_set_inflight_fd
,