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/virtio-crypto.h"
15 #include "hw/virtio/vhost-user.h"
16 #include "hw/virtio/vhost-backend.h"
17 #include "hw/virtio/virtio.h"
18 #include "hw/virtio/virtio-net.h"
19 #include "chardev/char-fe.h"
20 #include "io/channel-socket.h"
21 #include "sysemu/kvm.h"
22 #include "qemu/error-report.h"
23 #include "qemu/main-loop.h"
24 #include "qemu/sockets.h"
25 #include "sysemu/runstate.h"
26 #include "sysemu/cryptodev.h"
27 #include "migration/migration.h"
28 #include "migration/postcopy-ram.h"
30 #include "exec/ramblock.h"
32 #include <sys/ioctl.h>
33 #include <sys/socket.h>
36 #include "standard-headers/linux/vhost_types.h"
39 #include <linux/userfaultfd.h>
42 #define VHOST_MEMORY_BASELINE_NREGIONS 8
43 #define VHOST_USER_F_PROTOCOL_FEATURES 30
44 #define VHOST_USER_BACKEND_MAX_FDS 8
46 #if defined(TARGET_PPC) || defined(TARGET_PPC64)
47 #include "hw/ppc/spapr.h"
48 #define VHOST_USER_MAX_RAM_SLOTS SPAPR_MAX_RAM_SLOTS
51 #define VHOST_USER_MAX_RAM_SLOTS 512
55 * Maximum size of virtio device config space
57 #define VHOST_USER_MAX_CONFIG_SIZE 256
59 enum VhostUserProtocolFeature
{
60 VHOST_USER_PROTOCOL_F_MQ
= 0,
61 VHOST_USER_PROTOCOL_F_LOG_SHMFD
= 1,
62 VHOST_USER_PROTOCOL_F_RARP
= 2,
63 VHOST_USER_PROTOCOL_F_REPLY_ACK
= 3,
64 VHOST_USER_PROTOCOL_F_NET_MTU
= 4,
65 VHOST_USER_PROTOCOL_F_BACKEND_REQ
= 5,
66 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
= 6,
67 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
= 7,
68 VHOST_USER_PROTOCOL_F_PAGEFAULT
= 8,
69 VHOST_USER_PROTOCOL_F_CONFIG
= 9,
70 VHOST_USER_PROTOCOL_F_BACKEND_SEND_FD
= 10,
71 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
= 11,
72 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
= 12,
73 VHOST_USER_PROTOCOL_F_RESET_DEVICE
= 13,
74 /* Feature 14 reserved for VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS. */
75 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
= 15,
76 VHOST_USER_PROTOCOL_F_STATUS
= 16,
77 VHOST_USER_PROTOCOL_F_MAX
80 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
82 typedef enum VhostUserRequest
{
84 VHOST_USER_GET_FEATURES
= 1,
85 VHOST_USER_SET_FEATURES
= 2,
86 VHOST_USER_SET_OWNER
= 3,
87 VHOST_USER_RESET_OWNER
= 4,
88 VHOST_USER_SET_MEM_TABLE
= 5,
89 VHOST_USER_SET_LOG_BASE
= 6,
90 VHOST_USER_SET_LOG_FD
= 7,
91 VHOST_USER_SET_VRING_NUM
= 8,
92 VHOST_USER_SET_VRING_ADDR
= 9,
93 VHOST_USER_SET_VRING_BASE
= 10,
94 VHOST_USER_GET_VRING_BASE
= 11,
95 VHOST_USER_SET_VRING_KICK
= 12,
96 VHOST_USER_SET_VRING_CALL
= 13,
97 VHOST_USER_SET_VRING_ERR
= 14,
98 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
99 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
100 VHOST_USER_GET_QUEUE_NUM
= 17,
101 VHOST_USER_SET_VRING_ENABLE
= 18,
102 VHOST_USER_SEND_RARP
= 19,
103 VHOST_USER_NET_SET_MTU
= 20,
104 VHOST_USER_SET_BACKEND_REQ_FD
= 21,
105 VHOST_USER_IOTLB_MSG
= 22,
106 VHOST_USER_SET_VRING_ENDIAN
= 23,
107 VHOST_USER_GET_CONFIG
= 24,
108 VHOST_USER_SET_CONFIG
= 25,
109 VHOST_USER_CREATE_CRYPTO_SESSION
= 26,
110 VHOST_USER_CLOSE_CRYPTO_SESSION
= 27,
111 VHOST_USER_POSTCOPY_ADVISE
= 28,
112 VHOST_USER_POSTCOPY_LISTEN
= 29,
113 VHOST_USER_POSTCOPY_END
= 30,
114 VHOST_USER_GET_INFLIGHT_FD
= 31,
115 VHOST_USER_SET_INFLIGHT_FD
= 32,
116 VHOST_USER_GPU_SET_SOCKET
= 33,
117 VHOST_USER_RESET_DEVICE
= 34,
118 /* Message number 35 reserved for VHOST_USER_VRING_KICK. */
119 VHOST_USER_GET_MAX_MEM_SLOTS
= 36,
120 VHOST_USER_ADD_MEM_REG
= 37,
121 VHOST_USER_REM_MEM_REG
= 38,
122 VHOST_USER_SET_STATUS
= 39,
123 VHOST_USER_GET_STATUS
= 40,
127 typedef enum VhostUserBackendRequest
{
128 VHOST_USER_BACKEND_NONE
= 0,
129 VHOST_USER_BACKEND_IOTLB_MSG
= 1,
130 VHOST_USER_BACKEND_CONFIG_CHANGE_MSG
= 2,
131 VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG
= 3,
132 VHOST_USER_BACKEND_MAX
133 } VhostUserBackendRequest
;
135 typedef struct VhostUserMemoryRegion
{
136 uint64_t guest_phys_addr
;
137 uint64_t memory_size
;
138 uint64_t userspace_addr
;
139 uint64_t mmap_offset
;
140 } VhostUserMemoryRegion
;
142 typedef struct VhostUserMemory
{
145 VhostUserMemoryRegion regions
[VHOST_MEMORY_BASELINE_NREGIONS
];
148 typedef struct VhostUserMemRegMsg
{
150 VhostUserMemoryRegion region
;
151 } VhostUserMemRegMsg
;
153 typedef struct VhostUserLog
{
155 uint64_t mmap_offset
;
158 typedef struct VhostUserConfig
{
162 uint8_t region
[VHOST_USER_MAX_CONFIG_SIZE
];
165 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN 512
166 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN 64
167 #define VHOST_CRYPTO_ASYM_MAX_KEY_LEN 1024
169 typedef struct VhostUserCryptoSession
{
173 CryptoDevBackendSymSessionInfo session_setup_data
;
174 uint8_t key
[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN
];
175 uint8_t auth_key
[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN
];
178 CryptoDevBackendAsymSessionInfo session_setup_data
;
179 uint8_t key
[VHOST_CRYPTO_ASYM_MAX_KEY_LEN
];
183 /* session id for success, -1 on errors */
185 } VhostUserCryptoSession
;
187 static VhostUserConfig c
__attribute__ ((unused
));
188 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
192 typedef struct VhostUserVringArea
{
196 } VhostUserVringArea
;
198 typedef struct VhostUserInflight
{
200 uint64_t mmap_offset
;
206 VhostUserRequest request
;
208 #define VHOST_USER_VERSION_MASK (0x3)
209 #define VHOST_USER_REPLY_MASK (0x1 << 2)
210 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
212 uint32_t size
; /* the following payload size */
213 } QEMU_PACKED VhostUserHeader
;
216 #define VHOST_USER_VRING_IDX_MASK (0xff)
217 #define VHOST_USER_VRING_NOFD_MASK (0x1 << 8)
219 struct vhost_vring_state state
;
220 struct vhost_vring_addr addr
;
221 VhostUserMemory memory
;
222 VhostUserMemRegMsg mem_reg
;
224 struct vhost_iotlb_msg iotlb
;
225 VhostUserConfig config
;
226 VhostUserCryptoSession session
;
227 VhostUserVringArea area
;
228 VhostUserInflight inflight
;
231 typedef struct VhostUserMsg
{
233 VhostUserPayload payload
;
234 } QEMU_PACKED VhostUserMsg
;
236 static VhostUserMsg m
__attribute__ ((unused
));
237 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
239 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
241 /* The version of the protocol we support */
242 #define VHOST_USER_VERSION (0x1)
245 struct vhost_dev
*dev
;
246 /* Shared between vhost devs of the same virtio device */
247 VhostUserState
*user
;
248 QIOChannel
*backend_ioc
;
249 GSource
*backend_src
;
250 NotifierWithReturn postcopy_notifier
;
251 struct PostCopyFD postcopy_fd
;
252 uint64_t postcopy_client_bases
[VHOST_USER_MAX_RAM_SLOTS
];
253 /* Length of the region_rb and region_rb_offset arrays */
254 size_t region_rb_len
;
255 /* RAMBlock associated with a given region */
256 RAMBlock
**region_rb
;
258 * The offset from the start of the RAMBlock to the start of the
261 ram_addr_t
*region_rb_offset
;
263 /* True once we've entered postcopy_listen */
264 bool postcopy_listen
;
266 /* Our current regions */
267 int num_shadow_regions
;
268 struct vhost_memory_region shadow_regions
[VHOST_USER_MAX_RAM_SLOTS
];
271 struct scrub_regions
{
272 struct vhost_memory_region
*region
;
277 static bool ioeventfd_enabled(void)
279 return !kvm_enabled() || kvm_eventfds_enabled();
282 static int vhost_user_read_header(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
284 struct vhost_user
*u
= dev
->opaque
;
285 CharBackend
*chr
= u
->user
->chr
;
286 uint8_t *p
= (uint8_t *) msg
;
287 int r
, size
= VHOST_USER_HDR_SIZE
;
289 r
= qemu_chr_fe_read_all(chr
, p
, size
);
291 int saved_errno
= errno
;
292 error_report("Failed to read msg header. Read %d instead of %d."
293 " Original request %d.", r
, size
, msg
->hdr
.request
);
294 return r
< 0 ? -saved_errno
: -EIO
;
297 /* validate received flags */
298 if (msg
->hdr
.flags
!= (VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
)) {
299 error_report("Failed to read msg header."
300 " Flags 0x%x instead of 0x%x.", msg
->hdr
.flags
,
301 VHOST_USER_REPLY_MASK
| VHOST_USER_VERSION
);
305 trace_vhost_user_read(msg
->hdr
.request
, msg
->hdr
.flags
);
310 static int vhost_user_read(struct vhost_dev
*dev
, VhostUserMsg
*msg
)
312 struct vhost_user
*u
= dev
->opaque
;
313 CharBackend
*chr
= u
->user
->chr
;
314 uint8_t *p
= (uint8_t *) msg
;
317 r
= vhost_user_read_header(dev
, msg
);
322 /* validate message size is sane */
323 if (msg
->hdr
.size
> VHOST_USER_PAYLOAD_SIZE
) {
324 error_report("Failed to read msg header."
325 " Size %d exceeds the maximum %zu.", msg
->hdr
.size
,
326 VHOST_USER_PAYLOAD_SIZE
);
331 p
+= VHOST_USER_HDR_SIZE
;
332 size
= msg
->hdr
.size
;
333 r
= qemu_chr_fe_read_all(chr
, p
, size
);
335 int saved_errno
= errno
;
336 error_report("Failed to read msg payload."
337 " Read %d instead of %d.", r
, msg
->hdr
.size
);
338 return r
< 0 ? -saved_errno
: -EIO
;
345 static int process_message_reply(struct vhost_dev
*dev
,
346 const VhostUserMsg
*msg
)
349 VhostUserMsg msg_reply
;
351 if ((msg
->hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) == 0) {
355 ret
= vhost_user_read(dev
, &msg_reply
);
360 if (msg_reply
.hdr
.request
!= msg
->hdr
.request
) {
361 error_report("Received unexpected msg type. "
362 "Expected %d received %d",
363 msg
->hdr
.request
, msg_reply
.hdr
.request
);
367 return msg_reply
.payload
.u64
? -EIO
: 0;
370 static bool vhost_user_one_time_request(VhostUserRequest request
)
373 case VHOST_USER_SET_OWNER
:
374 case VHOST_USER_RESET_OWNER
:
375 case VHOST_USER_SET_MEM_TABLE
:
376 case VHOST_USER_GET_QUEUE_NUM
:
377 case VHOST_USER_NET_SET_MTU
:
378 case VHOST_USER_ADD_MEM_REG
:
379 case VHOST_USER_REM_MEM_REG
:
386 /* most non-init callers ignore the error */
387 static int vhost_user_write(struct vhost_dev
*dev
, VhostUserMsg
*msg
,
388 int *fds
, int fd_num
)
390 struct vhost_user
*u
= dev
->opaque
;
391 CharBackend
*chr
= u
->user
->chr
;
392 int ret
, size
= VHOST_USER_HDR_SIZE
+ msg
->hdr
.size
;
395 * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
396 * we just need send it once in the first time. For later such
397 * request, we just ignore it.
399 if (vhost_user_one_time_request(msg
->hdr
.request
) && dev
->vq_index
!= 0) {
400 msg
->hdr
.flags
&= ~VHOST_USER_NEED_REPLY_MASK
;
404 if (qemu_chr_fe_set_msgfds(chr
, fds
, fd_num
) < 0) {
405 error_report("Failed to set msg fds.");
409 ret
= qemu_chr_fe_write_all(chr
, (const uint8_t *) msg
, size
);
411 int saved_errno
= errno
;
412 error_report("Failed to write msg."
413 " Wrote %d instead of %d.", ret
, size
);
414 return ret
< 0 ? -saved_errno
: -EIO
;
417 trace_vhost_user_write(msg
->hdr
.request
, msg
->hdr
.flags
);
422 int vhost_user_gpu_set_socket(struct vhost_dev
*dev
, int fd
)
425 .hdr
.request
= VHOST_USER_GPU_SET_SOCKET
,
426 .hdr
.flags
= VHOST_USER_VERSION
,
429 return vhost_user_write(dev
, &msg
, &fd
, 1);
432 static int vhost_user_set_log_base(struct vhost_dev
*dev
, uint64_t base
,
433 struct vhost_log
*log
)
435 int fds
[VHOST_USER_MAX_RAM_SLOTS
];
437 bool shmfd
= virtio_has_feature(dev
->protocol_features
,
438 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
441 .hdr
.request
= VHOST_USER_SET_LOG_BASE
,
442 .hdr
.flags
= VHOST_USER_VERSION
,
443 .payload
.log
.mmap_size
= log
->size
* sizeof(*(log
->log
)),
444 .payload
.log
.mmap_offset
= 0,
445 .hdr
.size
= sizeof(msg
.payload
.log
),
448 /* Send only once with first queue pair */
449 if (dev
->vq_index
!= 0) {
453 if (shmfd
&& log
->fd
!= -1) {
454 fds
[fd_num
++] = log
->fd
;
457 ret
= vhost_user_write(dev
, &msg
, fds
, fd_num
);
464 ret
= vhost_user_read(dev
, &msg
);
469 if (msg
.hdr
.request
!= VHOST_USER_SET_LOG_BASE
) {
470 error_report("Received unexpected msg type. "
471 "Expected %d received %d",
472 VHOST_USER_SET_LOG_BASE
, msg
.hdr
.request
);
480 static MemoryRegion
*vhost_user_get_mr_data(uint64_t addr
, ram_addr_t
*offset
,
485 assert((uintptr_t)addr
== addr
);
486 mr
= memory_region_from_host((void *)(uintptr_t)addr
, offset
);
487 *fd
= memory_region_get_fd(mr
);
488 *offset
+= mr
->ram_block
->fd_offset
;
493 static void vhost_user_fill_msg_region(VhostUserMemoryRegion
*dst
,
494 struct vhost_memory_region
*src
,
495 uint64_t mmap_offset
)
497 assert(src
!= NULL
&& dst
!= NULL
);
498 dst
->userspace_addr
= src
->userspace_addr
;
499 dst
->memory_size
= src
->memory_size
;
500 dst
->guest_phys_addr
= src
->guest_phys_addr
;
501 dst
->mmap_offset
= mmap_offset
;
504 static int vhost_user_fill_set_mem_table_msg(struct vhost_user
*u
,
505 struct vhost_dev
*dev
,
507 int *fds
, size_t *fd_num
,
508 bool track_ramblocks
)
513 struct vhost_memory_region
*reg
;
514 VhostUserMemoryRegion region_buffer
;
516 msg
->hdr
.request
= VHOST_USER_SET_MEM_TABLE
;
518 for (i
= 0; i
< dev
->mem
->nregions
; ++i
) {
519 reg
= dev
->mem
->regions
+ i
;
521 mr
= vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
523 if (track_ramblocks
) {
524 assert(*fd_num
< VHOST_MEMORY_BASELINE_NREGIONS
);
525 trace_vhost_user_set_mem_table_withfd(*fd_num
, mr
->name
,
527 reg
->guest_phys_addr
,
530 u
->region_rb_offset
[i
] = offset
;
531 u
->region_rb
[i
] = mr
->ram_block
;
532 } else if (*fd_num
== VHOST_MEMORY_BASELINE_NREGIONS
) {
533 error_report("Failed preparing vhost-user memory table msg");
536 vhost_user_fill_msg_region(®ion_buffer
, reg
, offset
);
537 msg
->payload
.memory
.regions
[*fd_num
] = region_buffer
;
538 fds
[(*fd_num
)++] = fd
;
539 } else if (track_ramblocks
) {
540 u
->region_rb_offset
[i
] = 0;
541 u
->region_rb
[i
] = NULL
;
545 msg
->payload
.memory
.nregions
= *fd_num
;
548 error_report("Failed initializing vhost-user memory map, "
549 "consider using -object memory-backend-file share=on");
553 msg
->hdr
.size
= sizeof(msg
->payload
.memory
.nregions
);
554 msg
->hdr
.size
+= sizeof(msg
->payload
.memory
.padding
);
555 msg
->hdr
.size
+= *fd_num
* sizeof(VhostUserMemoryRegion
);
560 static inline bool reg_equal(struct vhost_memory_region
*shadow_reg
,
561 struct vhost_memory_region
*vdev_reg
)
563 return shadow_reg
->guest_phys_addr
== vdev_reg
->guest_phys_addr
&&
564 shadow_reg
->userspace_addr
== vdev_reg
->userspace_addr
&&
565 shadow_reg
->memory_size
== vdev_reg
->memory_size
;
568 static void scrub_shadow_regions(struct vhost_dev
*dev
,
569 struct scrub_regions
*add_reg
,
571 struct scrub_regions
*rem_reg
,
572 int *nr_rem_reg
, uint64_t *shadow_pcb
,
573 bool track_ramblocks
)
575 struct vhost_user
*u
= dev
->opaque
;
576 bool found
[VHOST_USER_MAX_RAM_SLOTS
] = {};
577 struct vhost_memory_region
*reg
, *shadow_reg
;
578 int i
, j
, fd
, add_idx
= 0, rm_idx
= 0, fd_num
= 0;
584 * Find memory regions present in our shadow state which are not in
585 * the device's current memory state.
587 * Mark regions in both the shadow and device state as "found".
589 for (i
= 0; i
< u
->num_shadow_regions
; i
++) {
590 shadow_reg
= &u
->shadow_regions
[i
];
593 for (j
= 0; j
< dev
->mem
->nregions
; j
++) {
594 reg
= &dev
->mem
->regions
[j
];
596 mr
= vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
598 if (reg_equal(shadow_reg
, reg
)) {
601 if (track_ramblocks
) {
603 * Reset postcopy client bases, region_rb, and
604 * region_rb_offset in case regions are removed.
607 u
->region_rb_offset
[j
] = offset
;
608 u
->region_rb
[j
] = mr
->ram_block
;
609 shadow_pcb
[j
] = u
->postcopy_client_bases
[i
];
611 u
->region_rb_offset
[j
] = 0;
612 u
->region_rb
[j
] = NULL
;
620 * If the region was not found in the current device memory state
621 * create an entry for it in the removed list.
624 rem_reg
[rm_idx
].region
= shadow_reg
;
625 rem_reg
[rm_idx
++].reg_idx
= i
;
630 * For regions not marked "found", create entries in the added list.
632 * Note their indexes in the device memory state and the indexes of their
635 for (i
= 0; i
< dev
->mem
->nregions
; i
++) {
636 reg
= &dev
->mem
->regions
[i
];
637 vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
643 * If the region was in both the shadow and device state we don't
644 * need to send a VHOST_USER_ADD_MEM_REG message for it.
650 add_reg
[add_idx
].region
= reg
;
651 add_reg
[add_idx
].reg_idx
= i
;
652 add_reg
[add_idx
++].fd_idx
= fd_num
;
654 *nr_rem_reg
= rm_idx
;
655 *nr_add_reg
= add_idx
;
660 static int send_remove_regions(struct vhost_dev
*dev
,
661 struct scrub_regions
*remove_reg
,
662 int nr_rem_reg
, VhostUserMsg
*msg
,
663 bool reply_supported
)
665 struct vhost_user
*u
= dev
->opaque
;
666 struct vhost_memory_region
*shadow_reg
;
667 int i
, fd
, shadow_reg_idx
, ret
;
669 VhostUserMemoryRegion region_buffer
;
672 * The regions in remove_reg appear in the same order they do in the
673 * shadow table. Therefore we can minimize memory copies by iterating
674 * through remove_reg backwards.
676 for (i
= nr_rem_reg
- 1; i
>= 0; i
--) {
677 shadow_reg
= remove_reg
[i
].region
;
678 shadow_reg_idx
= remove_reg
[i
].reg_idx
;
680 vhost_user_get_mr_data(shadow_reg
->userspace_addr
, &offset
, &fd
);
683 msg
->hdr
.request
= VHOST_USER_REM_MEM_REG
;
684 vhost_user_fill_msg_region(®ion_buffer
, shadow_reg
, 0);
685 msg
->payload
.mem_reg
.region
= region_buffer
;
687 ret
= vhost_user_write(dev
, msg
, NULL
, 0);
692 if (reply_supported
) {
693 ret
= process_message_reply(dev
, msg
);
701 * At this point we know the backend has unmapped the region. It is now
702 * safe to remove it from the shadow table.
704 memmove(&u
->shadow_regions
[shadow_reg_idx
],
705 &u
->shadow_regions
[shadow_reg_idx
+ 1],
706 sizeof(struct vhost_memory_region
) *
707 (u
->num_shadow_regions
- shadow_reg_idx
- 1));
708 u
->num_shadow_regions
--;
714 static int send_add_regions(struct vhost_dev
*dev
,
715 struct scrub_regions
*add_reg
, int nr_add_reg
,
716 VhostUserMsg
*msg
, uint64_t *shadow_pcb
,
717 bool reply_supported
, bool track_ramblocks
)
719 struct vhost_user
*u
= dev
->opaque
;
720 int i
, fd
, ret
, reg_idx
, reg_fd_idx
;
721 struct vhost_memory_region
*reg
;
724 VhostUserMsg msg_reply
;
725 VhostUserMemoryRegion region_buffer
;
727 for (i
= 0; i
< nr_add_reg
; i
++) {
728 reg
= add_reg
[i
].region
;
729 reg_idx
= add_reg
[i
].reg_idx
;
730 reg_fd_idx
= add_reg
[i
].fd_idx
;
732 mr
= vhost_user_get_mr_data(reg
->userspace_addr
, &offset
, &fd
);
735 if (track_ramblocks
) {
736 trace_vhost_user_set_mem_table_withfd(reg_fd_idx
, mr
->name
,
738 reg
->guest_phys_addr
,
741 u
->region_rb_offset
[reg_idx
] = offset
;
742 u
->region_rb
[reg_idx
] = mr
->ram_block
;
744 msg
->hdr
.request
= VHOST_USER_ADD_MEM_REG
;
745 vhost_user_fill_msg_region(®ion_buffer
, reg
, offset
);
746 msg
->payload
.mem_reg
.region
= region_buffer
;
748 ret
= vhost_user_write(dev
, msg
, &fd
, 1);
753 if (track_ramblocks
) {
756 ret
= vhost_user_read(dev
, &msg_reply
);
761 reply_gpa
= msg_reply
.payload
.mem_reg
.region
.guest_phys_addr
;
763 if (msg_reply
.hdr
.request
!= VHOST_USER_ADD_MEM_REG
) {
764 error_report("%s: Received unexpected msg type."
765 "Expected %d received %d", __func__
,
766 VHOST_USER_ADD_MEM_REG
,
767 msg_reply
.hdr
.request
);
772 * We're using the same structure, just reusing one of the
773 * fields, so it should be the same size.
775 if (msg_reply
.hdr
.size
!= msg
->hdr
.size
) {
776 error_report("%s: Unexpected size for postcopy reply "
777 "%d vs %d", __func__
, msg_reply
.hdr
.size
,
782 /* Get the postcopy client base from the backend's reply. */
783 if (reply_gpa
== dev
->mem
->regions
[reg_idx
].guest_phys_addr
) {
784 shadow_pcb
[reg_idx
] =
785 msg_reply
.payload
.mem_reg
.region
.userspace_addr
;
786 trace_vhost_user_set_mem_table_postcopy(
787 msg_reply
.payload
.mem_reg
.region
.userspace_addr
,
788 msg
->payload
.mem_reg
.region
.userspace_addr
,
789 reg_fd_idx
, reg_idx
);
791 error_report("%s: invalid postcopy reply for region. "
792 "Got guest physical address %" PRIX64
", expected "
793 "%" PRIX64
, __func__
, reply_gpa
,
794 dev
->mem
->regions
[reg_idx
].guest_phys_addr
);
797 } else if (reply_supported
) {
798 ret
= process_message_reply(dev
, msg
);
803 } else if (track_ramblocks
) {
804 u
->region_rb_offset
[reg_idx
] = 0;
805 u
->region_rb
[reg_idx
] = NULL
;
809 * At this point, we know the backend has mapped in the new
810 * region, if the region has a valid file descriptor.
812 * The region should now be added to the shadow table.
814 u
->shadow_regions
[u
->num_shadow_regions
].guest_phys_addr
=
815 reg
->guest_phys_addr
;
816 u
->shadow_regions
[u
->num_shadow_regions
].userspace_addr
=
818 u
->shadow_regions
[u
->num_shadow_regions
].memory_size
=
820 u
->num_shadow_regions
++;
826 static int vhost_user_add_remove_regions(struct vhost_dev
*dev
,
828 bool reply_supported
,
829 bool track_ramblocks
)
831 struct vhost_user
*u
= dev
->opaque
;
832 struct scrub_regions add_reg
[VHOST_USER_MAX_RAM_SLOTS
];
833 struct scrub_regions rem_reg
[VHOST_USER_MAX_RAM_SLOTS
];
834 uint64_t shadow_pcb
[VHOST_USER_MAX_RAM_SLOTS
] = {};
835 int nr_add_reg
, nr_rem_reg
;
838 msg
->hdr
.size
= sizeof(msg
->payload
.mem_reg
);
840 /* Find the regions which need to be removed or added. */
841 scrub_shadow_regions(dev
, add_reg
, &nr_add_reg
, rem_reg
, &nr_rem_reg
,
842 shadow_pcb
, track_ramblocks
);
845 ret
= send_remove_regions(dev
, rem_reg
, nr_rem_reg
, msg
,
853 ret
= send_add_regions(dev
, add_reg
, nr_add_reg
, msg
, shadow_pcb
,
854 reply_supported
, track_ramblocks
);
860 if (track_ramblocks
) {
861 memcpy(u
->postcopy_client_bases
, shadow_pcb
,
862 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS
);
864 * Now we've registered this with the postcopy code, we ack to the
865 * client, because now we're in the position to be able to deal with
866 * any faults it generates.
868 /* TODO: Use this for failure cases as well with a bad value. */
869 msg
->hdr
.size
= sizeof(msg
->payload
.u64
);
870 msg
->payload
.u64
= 0; /* OK */
872 ret
= vhost_user_write(dev
, msg
, NULL
, 0);
881 if (track_ramblocks
) {
882 memcpy(u
->postcopy_client_bases
, shadow_pcb
,
883 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS
);
889 static int vhost_user_set_mem_table_postcopy(struct vhost_dev
*dev
,
890 struct vhost_memory
*mem
,
891 bool reply_supported
,
892 bool config_mem_slots
)
894 struct vhost_user
*u
= dev
->opaque
;
895 int fds
[VHOST_MEMORY_BASELINE_NREGIONS
];
897 VhostUserMsg msg_reply
;
902 .hdr
.flags
= VHOST_USER_VERSION
,
905 if (u
->region_rb_len
< dev
->mem
->nregions
) {
906 u
->region_rb
= g_renew(RAMBlock
*, u
->region_rb
, dev
->mem
->nregions
);
907 u
->region_rb_offset
= g_renew(ram_addr_t
, u
->region_rb_offset
,
909 memset(&(u
->region_rb
[u
->region_rb_len
]), '\0',
910 sizeof(RAMBlock
*) * (dev
->mem
->nregions
- u
->region_rb_len
));
911 memset(&(u
->region_rb_offset
[u
->region_rb_len
]), '\0',
912 sizeof(ram_addr_t
) * (dev
->mem
->nregions
- u
->region_rb_len
));
913 u
->region_rb_len
= dev
->mem
->nregions
;
916 if (config_mem_slots
) {
917 ret
= vhost_user_add_remove_regions(dev
, &msg
, reply_supported
, true);
922 ret
= vhost_user_fill_set_mem_table_msg(u
, dev
, &msg
, fds
, &fd_num
,
928 ret
= vhost_user_write(dev
, &msg
, fds
, fd_num
);
933 ret
= vhost_user_read(dev
, &msg_reply
);
938 if (msg_reply
.hdr
.request
!= VHOST_USER_SET_MEM_TABLE
) {
939 error_report("%s: Received unexpected msg type."
940 "Expected %d received %d", __func__
,
941 VHOST_USER_SET_MEM_TABLE
, msg_reply
.hdr
.request
);
946 * We're using the same structure, just reusing one of the
947 * fields, so it should be the same size.
949 if (msg_reply
.hdr
.size
!= msg
.hdr
.size
) {
950 error_report("%s: Unexpected size for postcopy reply "
951 "%d vs %d", __func__
, msg_reply
.hdr
.size
,
956 memset(u
->postcopy_client_bases
, 0,
957 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS
);
960 * They're in the same order as the regions that were sent
961 * but some of the regions were skipped (above) if they
964 for (msg_i
= 0, region_i
= 0;
965 region_i
< dev
->mem
->nregions
;
967 if (msg_i
< fd_num
&&
968 msg_reply
.payload
.memory
.regions
[msg_i
].guest_phys_addr
==
969 dev
->mem
->regions
[region_i
].guest_phys_addr
) {
970 u
->postcopy_client_bases
[region_i
] =
971 msg_reply
.payload
.memory
.regions
[msg_i
].userspace_addr
;
972 trace_vhost_user_set_mem_table_postcopy(
973 msg_reply
.payload
.memory
.regions
[msg_i
].userspace_addr
,
974 msg
.payload
.memory
.regions
[msg_i
].userspace_addr
,
979 if (msg_i
!= fd_num
) {
980 error_report("%s: postcopy reply not fully consumed "
982 __func__
, msg_i
, fd_num
);
987 * Now we've registered this with the postcopy code, we ack to the
988 * client, because now we're in the position to be able to deal
989 * with any faults it generates.
991 /* TODO: Use this for failure cases as well with a bad value. */
992 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
993 msg
.payload
.u64
= 0; /* OK */
994 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1003 static int vhost_user_set_mem_table(struct vhost_dev
*dev
,
1004 struct vhost_memory
*mem
)
1006 struct vhost_user
*u
= dev
->opaque
;
1007 int fds
[VHOST_MEMORY_BASELINE_NREGIONS
];
1009 bool do_postcopy
= u
->postcopy_listen
&& u
->postcopy_fd
.handler
;
1010 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1011 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1012 bool config_mem_slots
=
1013 virtio_has_feature(dev
->protocol_features
,
1014 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
);
1019 * Postcopy has enough differences that it's best done in it's own
1022 return vhost_user_set_mem_table_postcopy(dev
, mem
, reply_supported
,
1026 VhostUserMsg msg
= {
1027 .hdr
.flags
= VHOST_USER_VERSION
,
1030 if (reply_supported
) {
1031 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1034 if (config_mem_slots
) {
1035 ret
= vhost_user_add_remove_regions(dev
, &msg
, reply_supported
, false);
1040 ret
= vhost_user_fill_set_mem_table_msg(u
, dev
, &msg
, fds
, &fd_num
,
1046 ret
= vhost_user_write(dev
, &msg
, fds
, fd_num
);
1051 if (reply_supported
) {
1052 return process_message_reply(dev
, &msg
);
1059 static int vhost_user_set_vring_endian(struct vhost_dev
*dev
,
1060 struct vhost_vring_state
*ring
)
1062 bool cross_endian
= virtio_has_feature(dev
->protocol_features
,
1063 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN
);
1064 VhostUserMsg msg
= {
1065 .hdr
.request
= VHOST_USER_SET_VRING_ENDIAN
,
1066 .hdr
.flags
= VHOST_USER_VERSION
,
1067 .payload
.state
= *ring
,
1068 .hdr
.size
= sizeof(msg
.payload
.state
),
1071 if (!cross_endian
) {
1072 error_report("vhost-user trying to send unhandled ioctl");
1076 return vhost_user_write(dev
, &msg
, NULL
, 0);
1079 static int vhost_set_vring(struct vhost_dev
*dev
,
1080 unsigned long int request
,
1081 struct vhost_vring_state
*ring
)
1083 VhostUserMsg msg
= {
1084 .hdr
.request
= request
,
1085 .hdr
.flags
= VHOST_USER_VERSION
,
1086 .payload
.state
= *ring
,
1087 .hdr
.size
= sizeof(msg
.payload
.state
),
1090 return vhost_user_write(dev
, &msg
, NULL
, 0);
1093 static int vhost_user_set_vring_num(struct vhost_dev
*dev
,
1094 struct vhost_vring_state
*ring
)
1096 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_NUM
, ring
);
1099 static void vhost_user_host_notifier_free(VhostUserHostNotifier
*n
)
1101 assert(n
&& n
->unmap_addr
);
1102 munmap(n
->unmap_addr
, qemu_real_host_page_size());
1103 n
->unmap_addr
= NULL
;
1107 * clean-up function for notifier, will finally free the structure
1110 static void vhost_user_host_notifier_remove(VhostUserHostNotifier
*n
,
1115 virtio_queue_set_host_notifier_mr(vdev
, n
->idx
, &n
->mr
, false);
1117 assert(!n
->unmap_addr
);
1118 n
->unmap_addr
= n
->addr
;
1120 call_rcu(n
, vhost_user_host_notifier_free
, rcu
);
1124 static int vhost_user_set_vring_base(struct vhost_dev
*dev
,
1125 struct vhost_vring_state
*ring
)
1127 return vhost_set_vring(dev
, VHOST_USER_SET_VRING_BASE
, ring
);
1130 static int vhost_user_set_vring_enable(struct vhost_dev
*dev
, int enable
)
1134 if (!virtio_has_feature(dev
->features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
1138 for (i
= 0; i
< dev
->nvqs
; ++i
) {
1140 struct vhost_vring_state state
= {
1141 .index
= dev
->vq_index
+ i
,
1145 ret
= vhost_set_vring(dev
, VHOST_USER_SET_VRING_ENABLE
, &state
);
1148 * Restoring the previous state is likely infeasible, as well as
1149 * proceeding regardless the error, so just bail out and hope for
1150 * the device-level recovery.
1159 static VhostUserHostNotifier
*fetch_notifier(VhostUserState
*u
,
1162 if (idx
>= u
->notifiers
->len
) {
1165 return g_ptr_array_index(u
->notifiers
, idx
);
1168 static int vhost_user_get_vring_base(struct vhost_dev
*dev
,
1169 struct vhost_vring_state
*ring
)
1172 VhostUserMsg msg
= {
1173 .hdr
.request
= VHOST_USER_GET_VRING_BASE
,
1174 .hdr
.flags
= VHOST_USER_VERSION
,
1175 .payload
.state
= *ring
,
1176 .hdr
.size
= sizeof(msg
.payload
.state
),
1178 struct vhost_user
*u
= dev
->opaque
;
1180 VhostUserHostNotifier
*n
= fetch_notifier(u
->user
, ring
->index
);
1182 vhost_user_host_notifier_remove(n
, dev
->vdev
);
1185 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1190 ret
= vhost_user_read(dev
, &msg
);
1195 if (msg
.hdr
.request
!= VHOST_USER_GET_VRING_BASE
) {
1196 error_report("Received unexpected msg type. Expected %d received %d",
1197 VHOST_USER_GET_VRING_BASE
, msg
.hdr
.request
);
1201 if (msg
.hdr
.size
!= sizeof(msg
.payload
.state
)) {
1202 error_report("Received bad msg size.");
1206 *ring
= msg
.payload
.state
;
1211 static int vhost_set_vring_file(struct vhost_dev
*dev
,
1212 VhostUserRequest request
,
1213 struct vhost_vring_file
*file
)
1215 int fds
[VHOST_USER_MAX_RAM_SLOTS
];
1217 VhostUserMsg msg
= {
1218 .hdr
.request
= request
,
1219 .hdr
.flags
= VHOST_USER_VERSION
,
1220 .payload
.u64
= file
->index
& VHOST_USER_VRING_IDX_MASK
,
1221 .hdr
.size
= sizeof(msg
.payload
.u64
),
1224 if (ioeventfd_enabled() && file
->fd
> 0) {
1225 fds
[fd_num
++] = file
->fd
;
1227 msg
.payload
.u64
|= VHOST_USER_VRING_NOFD_MASK
;
1230 return vhost_user_write(dev
, &msg
, fds
, fd_num
);
1233 static int vhost_user_set_vring_kick(struct vhost_dev
*dev
,
1234 struct vhost_vring_file
*file
)
1236 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_KICK
, file
);
1239 static int vhost_user_set_vring_call(struct vhost_dev
*dev
,
1240 struct vhost_vring_file
*file
)
1242 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_CALL
, file
);
1245 static int vhost_user_set_vring_err(struct vhost_dev
*dev
,
1246 struct vhost_vring_file
*file
)
1248 return vhost_set_vring_file(dev
, VHOST_USER_SET_VRING_ERR
, file
);
1251 static int vhost_user_get_u64(struct vhost_dev
*dev
, int request
, uint64_t *u64
)
1254 VhostUserMsg msg
= {
1255 .hdr
.request
= request
,
1256 .hdr
.flags
= VHOST_USER_VERSION
,
1259 if (vhost_user_one_time_request(request
) && dev
->vq_index
!= 0) {
1263 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1268 ret
= vhost_user_read(dev
, &msg
);
1273 if (msg
.hdr
.request
!= request
) {
1274 error_report("Received unexpected msg type. Expected %d received %d",
1275 request
, msg
.hdr
.request
);
1279 if (msg
.hdr
.size
!= sizeof(msg
.payload
.u64
)) {
1280 error_report("Received bad msg size.");
1284 *u64
= msg
.payload
.u64
;
1289 static int vhost_user_get_features(struct vhost_dev
*dev
, uint64_t *features
)
1291 if (vhost_user_get_u64(dev
, VHOST_USER_GET_FEATURES
, features
) < 0) {
1298 static int enforce_reply(struct vhost_dev
*dev
,
1299 const VhostUserMsg
*msg
)
1303 if (msg
->hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) {
1304 return process_message_reply(dev
, msg
);
1308 * We need to wait for a reply but the backend does not
1309 * support replies for the command we just sent.
1310 * Send VHOST_USER_GET_FEATURES which makes all backends
1313 return vhost_user_get_features(dev
, &dummy
);
1316 static int vhost_user_set_vring_addr(struct vhost_dev
*dev
,
1317 struct vhost_vring_addr
*addr
)
1320 VhostUserMsg msg
= {
1321 .hdr
.request
= VHOST_USER_SET_VRING_ADDR
,
1322 .hdr
.flags
= VHOST_USER_VERSION
,
1323 .payload
.addr
= *addr
,
1324 .hdr
.size
= sizeof(msg
.payload
.addr
),
1327 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1328 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1331 * wait for a reply if logging is enabled to make sure
1332 * backend is actually logging changes
1334 bool wait_for_reply
= addr
->flags
& (1 << VHOST_VRING_F_LOG
);
1336 if (reply_supported
&& wait_for_reply
) {
1337 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1340 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1345 if (wait_for_reply
) {
1346 return enforce_reply(dev
, &msg
);
1352 static int vhost_user_set_u64(struct vhost_dev
*dev
, int request
, uint64_t u64
,
1353 bool wait_for_reply
)
1355 VhostUserMsg msg
= {
1356 .hdr
.request
= request
,
1357 .hdr
.flags
= VHOST_USER_VERSION
,
1359 .hdr
.size
= sizeof(msg
.payload
.u64
),
1363 if (wait_for_reply
) {
1364 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1365 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1366 if (reply_supported
) {
1367 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1371 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1376 if (wait_for_reply
) {
1377 return enforce_reply(dev
, &msg
);
1383 static int vhost_user_set_status(struct vhost_dev
*dev
, uint8_t status
)
1385 return vhost_user_set_u64(dev
, VHOST_USER_SET_STATUS
, status
, false);
1388 static int vhost_user_get_status(struct vhost_dev
*dev
, uint8_t *status
)
1393 ret
= vhost_user_get_u64(dev
, VHOST_USER_GET_STATUS
, &value
);
1402 static int vhost_user_add_status(struct vhost_dev
*dev
, uint8_t status
)
1407 ret
= vhost_user_get_status(dev
, &s
);
1412 if ((s
& status
) == status
) {
1417 return vhost_user_set_status(dev
, s
);
1420 static int vhost_user_set_features(struct vhost_dev
*dev
,
1424 * wait for a reply if logging is enabled to make sure
1425 * backend is actually logging changes
1427 bool log_enabled
= features
& (0x1ULL
<< VHOST_F_LOG_ALL
);
1431 * We need to include any extra backend only feature bits that
1432 * might be needed by our device. Currently this includes the
1433 * VHOST_USER_F_PROTOCOL_FEATURES bit for enabling protocol
1436 ret
= vhost_user_set_u64(dev
, VHOST_USER_SET_FEATURES
,
1437 features
| dev
->backend_features
,
1440 if (virtio_has_feature(dev
->protocol_features
,
1441 VHOST_USER_PROTOCOL_F_STATUS
)) {
1443 return vhost_user_add_status(dev
, VIRTIO_CONFIG_S_FEATURES_OK
);
1450 static int vhost_user_set_protocol_features(struct vhost_dev
*dev
,
1453 return vhost_user_set_u64(dev
, VHOST_USER_SET_PROTOCOL_FEATURES
, features
,
1457 static int vhost_user_set_owner(struct vhost_dev
*dev
)
1459 VhostUserMsg msg
= {
1460 .hdr
.request
= VHOST_USER_SET_OWNER
,
1461 .hdr
.flags
= VHOST_USER_VERSION
,
1464 return vhost_user_write(dev
, &msg
, NULL
, 0);
1467 static int vhost_user_get_max_memslots(struct vhost_dev
*dev
,
1468 uint64_t *max_memslots
)
1470 uint64_t backend_max_memslots
;
1473 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_MAX_MEM_SLOTS
,
1474 &backend_max_memslots
);
1479 *max_memslots
= backend_max_memslots
;
1484 static int vhost_user_reset_device(struct vhost_dev
*dev
)
1486 VhostUserMsg msg
= {
1487 .hdr
.flags
= VHOST_USER_VERSION
,
1490 msg
.hdr
.request
= virtio_has_feature(dev
->protocol_features
,
1491 VHOST_USER_PROTOCOL_F_RESET_DEVICE
)
1492 ? VHOST_USER_RESET_DEVICE
1493 : VHOST_USER_RESET_OWNER
;
1495 return vhost_user_write(dev
, &msg
, NULL
, 0);
1498 static int vhost_user_backend_handle_config_change(struct vhost_dev
*dev
)
1500 if (!dev
->config_ops
|| !dev
->config_ops
->vhost_dev_config_notifier
) {
1504 return dev
->config_ops
->vhost_dev_config_notifier(dev
);
1508 * Fetch or create the notifier for a given idx. Newly created
1509 * notifiers are added to the pointer array that tracks them.
1511 static VhostUserHostNotifier
*fetch_or_create_notifier(VhostUserState
*u
,
1514 VhostUserHostNotifier
*n
= NULL
;
1515 if (idx
>= u
->notifiers
->len
) {
1516 g_ptr_array_set_size(u
->notifiers
, idx
+ 1);
1519 n
= g_ptr_array_index(u
->notifiers
, idx
);
1522 * In case notification arrive out-of-order,
1523 * make room for current index.
1525 g_ptr_array_remove_index(u
->notifiers
, idx
);
1526 n
= g_new0(VhostUserHostNotifier
, 1);
1528 g_ptr_array_insert(u
->notifiers
, idx
, n
);
1529 trace_vhost_user_create_notifier(idx
, n
);
1535 static int vhost_user_backend_handle_vring_host_notifier(struct vhost_dev
*dev
,
1536 VhostUserVringArea
*area
,
1539 int queue_idx
= area
->u64
& VHOST_USER_VRING_IDX_MASK
;
1540 size_t page_size
= qemu_real_host_page_size();
1541 struct vhost_user
*u
= dev
->opaque
;
1542 VhostUserState
*user
= u
->user
;
1543 VirtIODevice
*vdev
= dev
->vdev
;
1544 VhostUserHostNotifier
*n
;
1548 if (!virtio_has_feature(dev
->protocol_features
,
1549 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER
) ||
1550 vdev
== NULL
|| queue_idx
>= virtio_get_num_queues(vdev
)) {
1555 * Fetch notifier and invalidate any old data before setting up
1556 * new mapped address.
1558 n
= fetch_or_create_notifier(user
, queue_idx
);
1559 vhost_user_host_notifier_remove(n
, vdev
);
1561 if (area
->u64
& VHOST_USER_VRING_NOFD_MASK
) {
1566 if (area
->size
!= page_size
) {
1570 addr
= mmap(NULL
, page_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
,
1572 if (addr
== MAP_FAILED
) {
1576 name
= g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1578 if (!n
->mr
.ram
) { /* Don't init again after suspend. */
1579 memory_region_init_ram_device_ptr(&n
->mr
, OBJECT(vdev
), name
,
1582 n
->mr
.ram_block
->host
= addr
;
1586 if (virtio_queue_set_host_notifier_mr(vdev
, queue_idx
, &n
->mr
, true)) {
1587 object_unparent(OBJECT(&n
->mr
));
1588 munmap(addr
, page_size
);
1597 static void close_backend_channel(struct vhost_user
*u
)
1599 g_source_destroy(u
->backend_src
);
1600 g_source_unref(u
->backend_src
);
1601 u
->backend_src
= NULL
;
1602 object_unref(OBJECT(u
->backend_ioc
));
1603 u
->backend_ioc
= NULL
;
1606 static gboolean
backend_read(QIOChannel
*ioc
, GIOCondition condition
,
1609 struct vhost_dev
*dev
= opaque
;
1610 struct vhost_user
*u
= dev
->opaque
;
1611 VhostUserHeader hdr
= { 0, };
1612 VhostUserPayload payload
= { 0, };
1613 Error
*local_err
= NULL
;
1614 gboolean rc
= G_SOURCE_CONTINUE
;
1617 g_autofree
int *fd
= NULL
;
1622 iov
.iov_base
= &hdr
;
1623 iov
.iov_len
= VHOST_USER_HDR_SIZE
;
1625 if (qio_channel_readv_full_all(ioc
, &iov
, 1, &fd
, &fdsize
, &local_err
)) {
1626 error_report_err(local_err
);
1630 if (hdr
.size
> VHOST_USER_PAYLOAD_SIZE
) {
1631 error_report("Failed to read msg header."
1632 " Size %d exceeds the maximum %zu.", hdr
.size
,
1633 VHOST_USER_PAYLOAD_SIZE
);
1638 if (qio_channel_read_all(ioc
, (char *) &payload
, hdr
.size
, &local_err
)) {
1639 error_report_err(local_err
);
1643 switch (hdr
.request
) {
1644 case VHOST_USER_BACKEND_IOTLB_MSG
:
1645 ret
= vhost_backend_handle_iotlb_msg(dev
, &payload
.iotlb
);
1647 case VHOST_USER_BACKEND_CONFIG_CHANGE_MSG
:
1648 ret
= vhost_user_backend_handle_config_change(dev
);
1650 case VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG
:
1651 ret
= vhost_user_backend_handle_vring_host_notifier(dev
, &payload
.area
,
1655 error_report("Received unexpected msg type: %d.", hdr
.request
);
1660 * REPLY_ACK feature handling. Other reply types has to be managed
1661 * directly in their request handlers.
1663 if (hdr
.flags
& VHOST_USER_NEED_REPLY_MASK
) {
1664 struct iovec iovec
[2];
1667 hdr
.flags
&= ~VHOST_USER_NEED_REPLY_MASK
;
1668 hdr
.flags
|= VHOST_USER_REPLY_MASK
;
1670 payload
.u64
= !!ret
;
1671 hdr
.size
= sizeof(payload
.u64
);
1673 iovec
[0].iov_base
= &hdr
;
1674 iovec
[0].iov_len
= VHOST_USER_HDR_SIZE
;
1675 iovec
[1].iov_base
= &payload
;
1676 iovec
[1].iov_len
= hdr
.size
;
1678 if (qio_channel_writev_all(ioc
, iovec
, ARRAY_SIZE(iovec
), &local_err
)) {
1679 error_report_err(local_err
);
1687 close_backend_channel(u
);
1688 rc
= G_SOURCE_REMOVE
;
1692 for (i
= 0; i
< fdsize
; i
++) {
1699 static int vhost_setup_backend_channel(struct vhost_dev
*dev
)
1701 VhostUserMsg msg
= {
1702 .hdr
.request
= VHOST_USER_SET_BACKEND_REQ_FD
,
1703 .hdr
.flags
= VHOST_USER_VERSION
,
1705 struct vhost_user
*u
= dev
->opaque
;
1707 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
1708 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
1709 Error
*local_err
= NULL
;
1712 if (!virtio_has_feature(dev
->protocol_features
,
1713 VHOST_USER_PROTOCOL_F_BACKEND_REQ
)) {
1717 if (qemu_socketpair(PF_UNIX
, SOCK_STREAM
, 0, sv
) == -1) {
1718 int saved_errno
= errno
;
1719 error_report("socketpair() failed");
1720 return -saved_errno
;
1723 ioc
= QIO_CHANNEL(qio_channel_socket_new_fd(sv
[0], &local_err
));
1725 error_report_err(local_err
);
1726 return -ECONNREFUSED
;
1728 u
->backend_ioc
= ioc
;
1729 u
->backend_src
= qio_channel_add_watch_source(u
->backend_ioc
,
1731 backend_read
, dev
, NULL
, NULL
);
1733 if (reply_supported
) {
1734 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
1737 ret
= vhost_user_write(dev
, &msg
, &sv
[1], 1);
1742 if (reply_supported
) {
1743 ret
= process_message_reply(dev
, &msg
);
1749 close_backend_channel(u
);
1757 * Called back from the postcopy fault thread when a fault is received on our
1759 * TODO: This is Linux specific
1761 static int vhost_user_postcopy_fault_handler(struct PostCopyFD
*pcfd
,
1764 struct vhost_dev
*dev
= pcfd
->data
;
1765 struct vhost_user
*u
= dev
->opaque
;
1766 struct uffd_msg
*msg
= ufd
;
1767 uint64_t faultaddr
= msg
->arg
.pagefault
.address
;
1768 RAMBlock
*rb
= NULL
;
1772 trace_vhost_user_postcopy_fault_handler(pcfd
->idstr
, faultaddr
,
1773 dev
->mem
->nregions
);
1774 for (i
= 0; i
< MIN(dev
->mem
->nregions
, u
->region_rb_len
); i
++) {
1775 trace_vhost_user_postcopy_fault_handler_loop(i
,
1776 u
->postcopy_client_bases
[i
], dev
->mem
->regions
[i
].memory_size
);
1777 if (faultaddr
>= u
->postcopy_client_bases
[i
]) {
1778 /* Ofset of the fault address in the vhost region */
1779 uint64_t region_offset
= faultaddr
- u
->postcopy_client_bases
[i
];
1780 if (region_offset
< dev
->mem
->regions
[i
].memory_size
) {
1781 rb_offset
= region_offset
+ u
->region_rb_offset
[i
];
1782 trace_vhost_user_postcopy_fault_handler_found(i
,
1783 region_offset
, rb_offset
);
1784 rb
= u
->region_rb
[i
];
1785 return postcopy_request_shared_page(pcfd
, rb
, faultaddr
,
1790 error_report("%s: Failed to find region for fault %" PRIx64
,
1791 __func__
, faultaddr
);
1795 static int vhost_user_postcopy_waker(struct PostCopyFD
*pcfd
, RAMBlock
*rb
,
1798 struct vhost_dev
*dev
= pcfd
->data
;
1799 struct vhost_user
*u
= dev
->opaque
;
1802 trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb
), offset
);
1807 /* Translate the offset into an address in the clients address space */
1808 for (i
= 0; i
< MIN(dev
->mem
->nregions
, u
->region_rb_len
); i
++) {
1809 if (u
->region_rb
[i
] == rb
&&
1810 offset
>= u
->region_rb_offset
[i
] &&
1811 offset
< (u
->region_rb_offset
[i
] +
1812 dev
->mem
->regions
[i
].memory_size
)) {
1813 uint64_t client_addr
= (offset
- u
->region_rb_offset
[i
]) +
1814 u
->postcopy_client_bases
[i
];
1815 trace_vhost_user_postcopy_waker_found(client_addr
);
1816 return postcopy_wake_shared(pcfd
, client_addr
, rb
);
1820 trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb
), offset
);
1826 * Called at the start of an inbound postcopy on reception of the
1829 static int vhost_user_postcopy_advise(struct vhost_dev
*dev
, Error
**errp
)
1832 struct vhost_user
*u
= dev
->opaque
;
1833 CharBackend
*chr
= u
->user
->chr
;
1836 VhostUserMsg msg
= {
1837 .hdr
.request
= VHOST_USER_POSTCOPY_ADVISE
,
1838 .hdr
.flags
= VHOST_USER_VERSION
,
1841 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1843 error_setg(errp
, "Failed to send postcopy_advise to vhost");
1847 ret
= vhost_user_read(dev
, &msg
);
1849 error_setg(errp
, "Failed to get postcopy_advise reply from vhost");
1853 if (msg
.hdr
.request
!= VHOST_USER_POSTCOPY_ADVISE
) {
1854 error_setg(errp
, "Unexpected msg type. Expected %d received %d",
1855 VHOST_USER_POSTCOPY_ADVISE
, msg
.hdr
.request
);
1860 error_setg(errp
, "Received bad msg size.");
1863 ufd
= qemu_chr_fe_get_msgfd(chr
);
1865 error_setg(errp
, "%s: Failed to get ufd", __func__
);
1868 qemu_socket_set_nonblock(ufd
);
1870 /* register ufd with userfault thread */
1871 u
->postcopy_fd
.fd
= ufd
;
1872 u
->postcopy_fd
.data
= dev
;
1873 u
->postcopy_fd
.handler
= vhost_user_postcopy_fault_handler
;
1874 u
->postcopy_fd
.waker
= vhost_user_postcopy_waker
;
1875 u
->postcopy_fd
.idstr
= "vhost-user"; /* Need to find unique name */
1876 postcopy_register_shared_ufd(&u
->postcopy_fd
);
1879 error_setg(errp
, "Postcopy not supported on non-Linux systems");
1885 * Called at the switch to postcopy on reception of the 'listen' command.
1887 static int vhost_user_postcopy_listen(struct vhost_dev
*dev
, Error
**errp
)
1889 struct vhost_user
*u
= dev
->opaque
;
1891 VhostUserMsg msg
= {
1892 .hdr
.request
= VHOST_USER_POSTCOPY_LISTEN
,
1893 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
1895 u
->postcopy_listen
= true;
1897 trace_vhost_user_postcopy_listen();
1899 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1901 error_setg(errp
, "Failed to send postcopy_listen to vhost");
1905 ret
= process_message_reply(dev
, &msg
);
1907 error_setg(errp
, "Failed to receive reply to postcopy_listen");
1915 * Called at the end of postcopy
1917 static int vhost_user_postcopy_end(struct vhost_dev
*dev
, Error
**errp
)
1919 VhostUserMsg msg
= {
1920 .hdr
.request
= VHOST_USER_POSTCOPY_END
,
1921 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
1924 struct vhost_user
*u
= dev
->opaque
;
1926 trace_vhost_user_postcopy_end_entry();
1928 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
1930 error_setg(errp
, "Failed to send postcopy_end to vhost");
1934 ret
= process_message_reply(dev
, &msg
);
1936 error_setg(errp
, "Failed to receive reply to postcopy_end");
1939 postcopy_unregister_shared_ufd(&u
->postcopy_fd
);
1940 close(u
->postcopy_fd
.fd
);
1941 u
->postcopy_fd
.handler
= NULL
;
1943 trace_vhost_user_postcopy_end_exit();
1948 static int vhost_user_postcopy_notifier(NotifierWithReturn
*notifier
,
1951 struct PostcopyNotifyData
*pnd
= opaque
;
1952 struct vhost_user
*u
= container_of(notifier
, struct vhost_user
,
1954 struct vhost_dev
*dev
= u
->dev
;
1956 switch (pnd
->reason
) {
1957 case POSTCOPY_NOTIFY_PROBE
:
1958 if (!virtio_has_feature(dev
->protocol_features
,
1959 VHOST_USER_PROTOCOL_F_PAGEFAULT
)) {
1960 /* TODO: Get the device name into this error somehow */
1961 error_setg(pnd
->errp
,
1962 "vhost-user backend not capable of postcopy");
1967 case POSTCOPY_NOTIFY_INBOUND_ADVISE
:
1968 return vhost_user_postcopy_advise(dev
, pnd
->errp
);
1970 case POSTCOPY_NOTIFY_INBOUND_LISTEN
:
1971 return vhost_user_postcopy_listen(dev
, pnd
->errp
);
1973 case POSTCOPY_NOTIFY_INBOUND_END
:
1974 return vhost_user_postcopy_end(dev
, pnd
->errp
);
1977 /* We ignore notifications we don't know */
1984 static int vhost_user_backend_init(struct vhost_dev
*dev
, void *opaque
,
1987 uint64_t features
, ram_slots
;
1988 struct vhost_user
*u
;
1989 VhostUserState
*vus
= (VhostUserState
*) opaque
;
1992 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
1994 u
= g_new0(struct vhost_user
, 1);
1999 err
= vhost_user_get_features(dev
, &features
);
2001 error_setg_errno(errp
, -err
, "vhost_backend_init failed");
2005 if (virtio_has_feature(features
, VHOST_USER_F_PROTOCOL_FEATURES
)) {
2006 bool supports_f_config
= vus
->supports_config
||
2007 (dev
->config_ops
&& dev
->config_ops
->vhost_dev_config_notifier
);
2008 uint64_t protocol_features
;
2010 dev
->backend_features
|= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES
;
2012 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_PROTOCOL_FEATURES
,
2013 &protocol_features
);
2015 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2020 * We will use all the protocol features we support - although
2021 * we suppress F_CONFIG if we know QEMUs internal code can not support
2024 protocol_features
&= VHOST_USER_PROTOCOL_FEATURE_MASK
;
2026 if (supports_f_config
) {
2027 if (!virtio_has_feature(protocol_features
,
2028 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2029 error_setg(errp
, "vhost-user device expecting "
2030 "VHOST_USER_PROTOCOL_F_CONFIG but the vhost-user backend does "
2035 if (virtio_has_feature(protocol_features
,
2036 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2037 warn_report("vhost-user backend supports "
2038 "VHOST_USER_PROTOCOL_F_CONFIG but QEMU does not.");
2039 protocol_features
&= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG
);
2043 /* final set of protocol features */
2044 dev
->protocol_features
= protocol_features
;
2045 err
= vhost_user_set_protocol_features(dev
, dev
->protocol_features
);
2047 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2051 /* query the max queues we support if backend supports Multiple Queue */
2052 if (dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_MQ
)) {
2053 err
= vhost_user_get_u64(dev
, VHOST_USER_GET_QUEUE_NUM
,
2056 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2060 dev
->max_queues
= 1;
2063 if (dev
->num_queues
&& dev
->max_queues
< dev
->num_queues
) {
2064 error_setg(errp
, "The maximum number of queues supported by the "
2065 "backend is %" PRIu64
, dev
->max_queues
);
2069 if (virtio_has_feature(features
, VIRTIO_F_IOMMU_PLATFORM
) &&
2070 !(virtio_has_feature(dev
->protocol_features
,
2071 VHOST_USER_PROTOCOL_F_BACKEND_REQ
) &&
2072 virtio_has_feature(dev
->protocol_features
,
2073 VHOST_USER_PROTOCOL_F_REPLY_ACK
))) {
2074 error_setg(errp
, "IOMMU support requires reply-ack and "
2075 "backend-req protocol features.");
2079 /* get max memory regions if backend supports configurable RAM slots */
2080 if (!virtio_has_feature(dev
->protocol_features
,
2081 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS
)) {
2082 u
->user
->memory_slots
= VHOST_MEMORY_BASELINE_NREGIONS
;
2084 err
= vhost_user_get_max_memslots(dev
, &ram_slots
);
2086 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2090 if (ram_slots
< u
->user
->memory_slots
) {
2091 error_setg(errp
, "The backend specified a max ram slots limit "
2092 "of %" PRIu64
", when the prior validated limit was "
2093 "%d. This limit should never decrease.", ram_slots
,
2094 u
->user
->memory_slots
);
2098 u
->user
->memory_slots
= MIN(ram_slots
, VHOST_USER_MAX_RAM_SLOTS
);
2102 if (dev
->migration_blocker
== NULL
&&
2103 !virtio_has_feature(dev
->protocol_features
,
2104 VHOST_USER_PROTOCOL_F_LOG_SHMFD
)) {
2105 error_setg(&dev
->migration_blocker
,
2106 "Migration disabled: vhost-user backend lacks "
2107 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
2110 if (dev
->vq_index
== 0) {
2111 err
= vhost_setup_backend_channel(dev
);
2113 error_setg_errno(errp
, EPROTO
, "vhost_backend_init failed");
2118 u
->postcopy_notifier
.notify
= vhost_user_postcopy_notifier
;
2119 postcopy_add_notifier(&u
->postcopy_notifier
);
2124 static int vhost_user_backend_cleanup(struct vhost_dev
*dev
)
2126 struct vhost_user
*u
;
2128 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2131 if (u
->postcopy_notifier
.notify
) {
2132 postcopy_remove_notifier(&u
->postcopy_notifier
);
2133 u
->postcopy_notifier
.notify
= NULL
;
2135 u
->postcopy_listen
= false;
2136 if (u
->postcopy_fd
.handler
) {
2137 postcopy_unregister_shared_ufd(&u
->postcopy_fd
);
2138 close(u
->postcopy_fd
.fd
);
2139 u
->postcopy_fd
.handler
= NULL
;
2141 if (u
->backend_ioc
) {
2142 close_backend_channel(u
);
2144 g_free(u
->region_rb
);
2145 u
->region_rb
= NULL
;
2146 g_free(u
->region_rb_offset
);
2147 u
->region_rb_offset
= NULL
;
2148 u
->region_rb_len
= 0;
2155 static int vhost_user_get_vq_index(struct vhost_dev
*dev
, int idx
)
2157 assert(idx
>= dev
->vq_index
&& idx
< dev
->vq_index
+ dev
->nvqs
);
2162 static int vhost_user_memslots_limit(struct vhost_dev
*dev
)
2164 struct vhost_user
*u
= dev
->opaque
;
2166 return u
->user
->memory_slots
;
2169 static bool vhost_user_requires_shm_log(struct vhost_dev
*dev
)
2171 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2173 return virtio_has_feature(dev
->protocol_features
,
2174 VHOST_USER_PROTOCOL_F_LOG_SHMFD
);
2177 static int vhost_user_migration_done(struct vhost_dev
*dev
, char* mac_addr
)
2179 VhostUserMsg msg
= { };
2181 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2183 /* If guest supports GUEST_ANNOUNCE do nothing */
2184 if (virtio_has_feature(dev
->acked_features
, VIRTIO_NET_F_GUEST_ANNOUNCE
)) {
2188 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
2189 if (virtio_has_feature(dev
->protocol_features
,
2190 VHOST_USER_PROTOCOL_F_RARP
)) {
2191 msg
.hdr
.request
= VHOST_USER_SEND_RARP
;
2192 msg
.hdr
.flags
= VHOST_USER_VERSION
;
2193 memcpy((char *)&msg
.payload
.u64
, mac_addr
, 6);
2194 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
2196 return vhost_user_write(dev
, &msg
, NULL
, 0);
2201 static bool vhost_user_can_merge(struct vhost_dev
*dev
,
2202 uint64_t start1
, uint64_t size1
,
2203 uint64_t start2
, uint64_t size2
)
2208 (void)vhost_user_get_mr_data(start1
, &offset
, &mfd
);
2209 (void)vhost_user_get_mr_data(start2
, &offset
, &rfd
);
2214 static int vhost_user_net_set_mtu(struct vhost_dev
*dev
, uint16_t mtu
)
2217 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
2218 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
2221 if (!(dev
->protocol_features
& (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU
))) {
2225 msg
.hdr
.request
= VHOST_USER_NET_SET_MTU
;
2226 msg
.payload
.u64
= mtu
;
2227 msg
.hdr
.size
= sizeof(msg
.payload
.u64
);
2228 msg
.hdr
.flags
= VHOST_USER_VERSION
;
2229 if (reply_supported
) {
2230 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
2233 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2238 /* If reply_ack supported, backend has to ack specified MTU is valid */
2239 if (reply_supported
) {
2240 return process_message_reply(dev
, &msg
);
2246 static int vhost_user_send_device_iotlb_msg(struct vhost_dev
*dev
,
2247 struct vhost_iotlb_msg
*imsg
)
2250 VhostUserMsg msg
= {
2251 .hdr
.request
= VHOST_USER_IOTLB_MSG
,
2252 .hdr
.size
= sizeof(msg
.payload
.iotlb
),
2253 .hdr
.flags
= VHOST_USER_VERSION
| VHOST_USER_NEED_REPLY_MASK
,
2254 .payload
.iotlb
= *imsg
,
2257 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2262 return process_message_reply(dev
, &msg
);
2266 static void vhost_user_set_iotlb_callback(struct vhost_dev
*dev
, int enabled
)
2268 /* No-op as the receive channel is not dedicated to IOTLB messages. */
2271 static int vhost_user_get_config(struct vhost_dev
*dev
, uint8_t *config
,
2272 uint32_t config_len
, Error
**errp
)
2275 VhostUserMsg msg
= {
2276 .hdr
.request
= VHOST_USER_GET_CONFIG
,
2277 .hdr
.flags
= VHOST_USER_VERSION
,
2278 .hdr
.size
= VHOST_USER_CONFIG_HDR_SIZE
+ config_len
,
2281 if (!virtio_has_feature(dev
->protocol_features
,
2282 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2283 error_setg(errp
, "VHOST_USER_PROTOCOL_F_CONFIG not supported");
2287 assert(config_len
<= VHOST_USER_MAX_CONFIG_SIZE
);
2289 msg
.payload
.config
.offset
= 0;
2290 msg
.payload
.config
.size
= config_len
;
2291 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2293 error_setg_errno(errp
, -ret
, "vhost_get_config failed");
2297 ret
= vhost_user_read(dev
, &msg
);
2299 error_setg_errno(errp
, -ret
, "vhost_get_config failed");
2303 if (msg
.hdr
.request
!= VHOST_USER_GET_CONFIG
) {
2305 "Received unexpected msg type. Expected %d received %d",
2306 VHOST_USER_GET_CONFIG
, msg
.hdr
.request
);
2310 if (msg
.hdr
.size
!= VHOST_USER_CONFIG_HDR_SIZE
+ config_len
) {
2311 error_setg(errp
, "Received bad msg size.");
2315 memcpy(config
, msg
.payload
.config
.region
, config_len
);
2320 static int vhost_user_set_config(struct vhost_dev
*dev
, const uint8_t *data
,
2321 uint32_t offset
, uint32_t size
, uint32_t flags
)
2325 bool reply_supported
= virtio_has_feature(dev
->protocol_features
,
2326 VHOST_USER_PROTOCOL_F_REPLY_ACK
);
2328 VhostUserMsg msg
= {
2329 .hdr
.request
= VHOST_USER_SET_CONFIG
,
2330 .hdr
.flags
= VHOST_USER_VERSION
,
2331 .hdr
.size
= VHOST_USER_CONFIG_HDR_SIZE
+ size
,
2334 if (!virtio_has_feature(dev
->protocol_features
,
2335 VHOST_USER_PROTOCOL_F_CONFIG
)) {
2339 if (reply_supported
) {
2340 msg
.hdr
.flags
|= VHOST_USER_NEED_REPLY_MASK
;
2343 if (size
> VHOST_USER_MAX_CONFIG_SIZE
) {
2347 msg
.payload
.config
.offset
= offset
,
2348 msg
.payload
.config
.size
= size
,
2349 msg
.payload
.config
.flags
= flags
,
2350 p
= msg
.payload
.config
.region
;
2351 memcpy(p
, data
, size
);
2353 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2358 if (reply_supported
) {
2359 return process_message_reply(dev
, &msg
);
2365 static int vhost_user_crypto_create_session(struct vhost_dev
*dev
,
2367 uint64_t *session_id
)
2370 bool crypto_session
= virtio_has_feature(dev
->protocol_features
,
2371 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
);
2372 CryptoDevBackendSessionInfo
*backend_info
= session_info
;
2373 VhostUserMsg msg
= {
2374 .hdr
.request
= VHOST_USER_CREATE_CRYPTO_SESSION
,
2375 .hdr
.flags
= VHOST_USER_VERSION
,
2376 .hdr
.size
= sizeof(msg
.payload
.session
),
2379 assert(dev
->vhost_ops
->backend_type
== VHOST_BACKEND_TYPE_USER
);
2381 if (!crypto_session
) {
2382 error_report("vhost-user trying to send unhandled ioctl");
2386 if (backend_info
->op_code
== VIRTIO_CRYPTO_AKCIPHER_CREATE_SESSION
) {
2387 CryptoDevBackendAsymSessionInfo
*sess
= &backend_info
->u
.asym_sess_info
;
2390 memcpy(&msg
.payload
.session
.u
.asym
.session_setup_data
, sess
,
2391 sizeof(CryptoDevBackendAsymSessionInfo
));
2393 keylen
= sizeof(msg
.payload
.session
.u
.asym
.key
);
2394 if (sess
->keylen
> keylen
) {
2395 error_report("Unsupported asymmetric key size");
2399 memcpy(&msg
.payload
.session
.u
.asym
.key
, sess
->key
,
2403 CryptoDevBackendSymSessionInfo
*sess
= &backend_info
->u
.sym_sess_info
;
2406 memcpy(&msg
.payload
.session
.u
.sym
.session_setup_data
, sess
,
2407 sizeof(CryptoDevBackendSymSessionInfo
));
2408 if (sess
->key_len
) {
2409 keylen
= sizeof(msg
.payload
.session
.u
.sym
.key
);
2410 if (sess
->key_len
> keylen
) {
2411 error_report("Unsupported cipher key size");
2415 memcpy(&msg
.payload
.session
.u
.sym
.key
, sess
->cipher_key
,
2419 if (sess
->auth_key_len
> 0) {
2420 keylen
= sizeof(msg
.payload
.session
.u
.sym
.auth_key
);
2421 if (sess
->auth_key_len
> keylen
) {
2422 error_report("Unsupported auth key size");
2426 memcpy(&msg
.payload
.session
.u
.sym
.auth_key
, sess
->auth_key
,
2427 sess
->auth_key_len
);
2431 msg
.payload
.session
.op_code
= backend_info
->op_code
;
2432 msg
.payload
.session
.session_id
= backend_info
->session_id
;
2433 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2435 error_report("vhost_user_write() return %d, create session failed",
2440 ret
= vhost_user_read(dev
, &msg
);
2442 error_report("vhost_user_read() return %d, create session failed",
2447 if (msg
.hdr
.request
!= VHOST_USER_CREATE_CRYPTO_SESSION
) {
2448 error_report("Received unexpected msg type. Expected %d received %d",
2449 VHOST_USER_CREATE_CRYPTO_SESSION
, msg
.hdr
.request
);
2453 if (msg
.hdr
.size
!= sizeof(msg
.payload
.session
)) {
2454 error_report("Received bad msg size.");
2458 if (msg
.payload
.session
.session_id
< 0) {
2459 error_report("Bad session id: %" PRId64
"",
2460 msg
.payload
.session
.session_id
);
2463 *session_id
= msg
.payload
.session
.session_id
;
2469 vhost_user_crypto_close_session(struct vhost_dev
*dev
, uint64_t session_id
)
2472 bool crypto_session
= virtio_has_feature(dev
->protocol_features
,
2473 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION
);
2474 VhostUserMsg msg
= {
2475 .hdr
.request
= VHOST_USER_CLOSE_CRYPTO_SESSION
,
2476 .hdr
.flags
= VHOST_USER_VERSION
,
2477 .hdr
.size
= sizeof(msg
.payload
.u64
),
2479 msg
.payload
.u64
= session_id
;
2481 if (!crypto_session
) {
2482 error_report("vhost-user trying to send unhandled ioctl");
2486 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2488 error_report("vhost_user_write() return %d, close session failed",
2496 static bool vhost_user_mem_section_filter(struct vhost_dev
*dev
,
2497 MemoryRegionSection
*section
)
2499 return memory_region_get_fd(section
->mr
) >= 0;
2502 static int vhost_user_get_inflight_fd(struct vhost_dev
*dev
,
2503 uint16_t queue_size
,
2504 struct vhost_inflight
*inflight
)
2509 struct vhost_user
*u
= dev
->opaque
;
2510 CharBackend
*chr
= u
->user
->chr
;
2511 VhostUserMsg msg
= {
2512 .hdr
.request
= VHOST_USER_GET_INFLIGHT_FD
,
2513 .hdr
.flags
= VHOST_USER_VERSION
,
2514 .payload
.inflight
.num_queues
= dev
->nvqs
,
2515 .payload
.inflight
.queue_size
= queue_size
,
2516 .hdr
.size
= sizeof(msg
.payload
.inflight
),
2519 if (!virtio_has_feature(dev
->protocol_features
,
2520 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
)) {
2524 ret
= vhost_user_write(dev
, &msg
, NULL
, 0);
2529 ret
= vhost_user_read(dev
, &msg
);
2534 if (msg
.hdr
.request
!= VHOST_USER_GET_INFLIGHT_FD
) {
2535 error_report("Received unexpected msg type. "
2536 "Expected %d received %d",
2537 VHOST_USER_GET_INFLIGHT_FD
, msg
.hdr
.request
);
2541 if (msg
.hdr
.size
!= sizeof(msg
.payload
.inflight
)) {
2542 error_report("Received bad msg size.");
2546 if (!msg
.payload
.inflight
.mmap_size
) {
2550 fd
= qemu_chr_fe_get_msgfd(chr
);
2552 error_report("Failed to get mem fd");
2556 addr
= mmap(0, msg
.payload
.inflight
.mmap_size
, PROT_READ
| PROT_WRITE
,
2557 MAP_SHARED
, fd
, msg
.payload
.inflight
.mmap_offset
);
2559 if (addr
== MAP_FAILED
) {
2560 error_report("Failed to mmap mem fd");
2565 inflight
->addr
= addr
;
2567 inflight
->size
= msg
.payload
.inflight
.mmap_size
;
2568 inflight
->offset
= msg
.payload
.inflight
.mmap_offset
;
2569 inflight
->queue_size
= queue_size
;
2574 static int vhost_user_set_inflight_fd(struct vhost_dev
*dev
,
2575 struct vhost_inflight
*inflight
)
2577 VhostUserMsg msg
= {
2578 .hdr
.request
= VHOST_USER_SET_INFLIGHT_FD
,
2579 .hdr
.flags
= VHOST_USER_VERSION
,
2580 .payload
.inflight
.mmap_size
= inflight
->size
,
2581 .payload
.inflight
.mmap_offset
= inflight
->offset
,
2582 .payload
.inflight
.num_queues
= dev
->nvqs
,
2583 .payload
.inflight
.queue_size
= inflight
->queue_size
,
2584 .hdr
.size
= sizeof(msg
.payload
.inflight
),
2587 if (!virtio_has_feature(dev
->protocol_features
,
2588 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD
)) {
2592 return vhost_user_write(dev
, &msg
, &inflight
->fd
, 1);
2595 static void vhost_user_state_destroy(gpointer data
)
2597 VhostUserHostNotifier
*n
= (VhostUserHostNotifier
*) data
;
2599 vhost_user_host_notifier_remove(n
, NULL
);
2600 object_unparent(OBJECT(&n
->mr
));
2602 * We can't free until vhost_user_host_notifier_remove has
2603 * done it's thing so schedule the free with RCU.
2609 bool vhost_user_init(VhostUserState
*user
, CharBackend
*chr
, Error
**errp
)
2612 error_setg(errp
, "Cannot initialize vhost-user state");
2616 user
->memory_slots
= 0;
2617 user
->notifiers
= g_ptr_array_new_full(VIRTIO_QUEUE_MAX
/ 4,
2618 &vhost_user_state_destroy
);
2622 void vhost_user_cleanup(VhostUserState
*user
)
2627 memory_region_transaction_begin();
2628 user
->notifiers
= (GPtrArray
*) g_ptr_array_free(user
->notifiers
, true);
2629 memory_region_transaction_commit();
2635 vu_async_close_fn cb
;
2638 struct vhost_dev
*vhost
;
2639 } VhostAsyncCallback
;
2641 static void vhost_user_async_close_bh(void *opaque
)
2643 VhostAsyncCallback
*data
= opaque
;
2644 struct vhost_dev
*vhost
= data
->vhost
;
2647 * If the vhost_dev has been cleared in the meantime there is
2648 * nothing left to do as some other path has completed the
2652 data
->cb(data
->dev
);
2659 * We only schedule the work if the machine is running. If suspended
2660 * we want to keep all the in-flight data as is for migration
2663 void vhost_user_async_close(DeviceState
*d
,
2664 CharBackend
*chardev
, struct vhost_dev
*vhost
,
2665 vu_async_close_fn cb
)
2667 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2669 * A close event may happen during a read/write, but vhost
2670 * code assumes the vhost_dev remains setup, so delay the
2673 AioContext
*ctx
= qemu_get_current_aio_context();
2674 VhostAsyncCallback
*data
= g_new0(VhostAsyncCallback
, 1);
2676 /* Save data for the callback */
2680 data
->vhost
= vhost
;
2682 /* Disable any further notifications on the chardev */
2683 qemu_chr_fe_set_handlers(chardev
,
2684 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
2687 aio_bh_schedule_oneshot(ctx
, vhost_user_async_close_bh
, data
);
2690 * Move vhost device to the stopped state. The vhost-user device
2691 * will be clean up and disconnected in BH. This can be useful in
2692 * the vhost migration code. If disconnect was caught there is an
2693 * option for the general vhost code to get the dev state without
2694 * knowing its type (in this case vhost-user).
2696 * Note if the vhost device is fully cleared by the time we
2697 * execute the bottom half we won't continue with the cleanup.
2699 vhost
->started
= false;
2703 static int vhost_user_dev_start(struct vhost_dev
*dev
, bool started
)
2705 if (!virtio_has_feature(dev
->protocol_features
,
2706 VHOST_USER_PROTOCOL_F_STATUS
)) {
2710 /* Set device status only for last queue pair */
2711 if (dev
->vq_index
+ dev
->nvqs
!= dev
->vq_index_end
) {
2716 return vhost_user_add_status(dev
, VIRTIO_CONFIG_S_ACKNOWLEDGE
|
2717 VIRTIO_CONFIG_S_DRIVER
|
2718 VIRTIO_CONFIG_S_DRIVER_OK
);
2724 static void vhost_user_reset_status(struct vhost_dev
*dev
)
2726 /* Set device status only for last queue pair */
2727 if (dev
->vq_index
+ dev
->nvqs
!= dev
->vq_index_end
) {
2731 if (virtio_has_feature(dev
->protocol_features
,
2732 VHOST_USER_PROTOCOL_F_STATUS
)) {
2733 vhost_user_set_status(dev
, 0);
2737 const VhostOps user_ops
= {
2738 .backend_type
= VHOST_BACKEND_TYPE_USER
,
2739 .vhost_backend_init
= vhost_user_backend_init
,
2740 .vhost_backend_cleanup
= vhost_user_backend_cleanup
,
2741 .vhost_backend_memslots_limit
= vhost_user_memslots_limit
,
2742 .vhost_set_log_base
= vhost_user_set_log_base
,
2743 .vhost_set_mem_table
= vhost_user_set_mem_table
,
2744 .vhost_set_vring_addr
= vhost_user_set_vring_addr
,
2745 .vhost_set_vring_endian
= vhost_user_set_vring_endian
,
2746 .vhost_set_vring_num
= vhost_user_set_vring_num
,
2747 .vhost_set_vring_base
= vhost_user_set_vring_base
,
2748 .vhost_get_vring_base
= vhost_user_get_vring_base
,
2749 .vhost_set_vring_kick
= vhost_user_set_vring_kick
,
2750 .vhost_set_vring_call
= vhost_user_set_vring_call
,
2751 .vhost_set_vring_err
= vhost_user_set_vring_err
,
2752 .vhost_set_features
= vhost_user_set_features
,
2753 .vhost_get_features
= vhost_user_get_features
,
2754 .vhost_set_owner
= vhost_user_set_owner
,
2755 .vhost_reset_device
= vhost_user_reset_device
,
2756 .vhost_get_vq_index
= vhost_user_get_vq_index
,
2757 .vhost_set_vring_enable
= vhost_user_set_vring_enable
,
2758 .vhost_requires_shm_log
= vhost_user_requires_shm_log
,
2759 .vhost_migration_done
= vhost_user_migration_done
,
2760 .vhost_backend_can_merge
= vhost_user_can_merge
,
2761 .vhost_net_set_mtu
= vhost_user_net_set_mtu
,
2762 .vhost_set_iotlb_callback
= vhost_user_set_iotlb_callback
,
2763 .vhost_send_device_iotlb_msg
= vhost_user_send_device_iotlb_msg
,
2764 .vhost_get_config
= vhost_user_get_config
,
2765 .vhost_set_config
= vhost_user_set_config
,
2766 .vhost_crypto_create_session
= vhost_user_crypto_create_session
,
2767 .vhost_crypto_close_session
= vhost_user_crypto_close_session
,
2768 .vhost_backend_mem_section_filter
= vhost_user_mem_section_filter
,
2769 .vhost_get_inflight_fd
= vhost_user_get_inflight_fd
,
2770 .vhost_set_inflight_fd
= vhost_user_set_inflight_fd
,
2771 .vhost_dev_start
= vhost_user_dev_start
,
2772 .vhost_reset_status
= vhost_user_reset_status
,