4 * Copyright (c) 2015 Red Hat, Inc.
7 * Victor Kaplansky <victork@redhat.com>
9 * This work is licensed under the terms of the GNU GPL, version 2 or
10 * later. See the COPYING file in the top-level directory.
15 * - main should get parameters from the command line.
16 * - implement all request handlers. Still not implemented:
17 * vubr_get_queue_num_exec()
18 * vubr_send_rarp_exec()
19 * - test for broken requests and virtqueue.
20 * - implement features defined by Virtio 1.0 spec.
21 * - support mergeable buffers and indirect descriptors.
22 * - implement clean shutdown.
23 * - implement non-blocking writes to UDP backend.
24 * - implement polling strategy.
25 * - implement clean starting/stopping of vq processing
26 * - implement clean starting/stopping of used and buffers
30 #define _FILE_OFFSET_BITS 64
41 #include <sys/types.h>
42 #include <sys/socket.h>
44 #include <sys/unistd.h>
46 #include <sys/eventfd.h>
47 #include <arpa/inet.h>
51 #include <linux/vhost.h>
53 #include "qemu/atomic.h"
54 #include "standard-headers/linux/virtio_net.h"
55 #include "standard-headers/linux/virtio_ring.h"
57 #define VHOST_USER_BRIDGE_DEBUG 1
61 if (VHOST_USER_BRIDGE_DEBUG) { \
62 printf(__VA_ARGS__); \
66 typedef void (*CallbackFunc
)(int sock
, void *ctx
);
68 typedef struct Event
{
70 CallbackFunc callback
;
73 typedef struct Dispatcher
{
76 Event events
[FD_SETSIZE
];
80 vubr_die(const char *s
)
87 dispatcher_init(Dispatcher
*dispr
)
89 FD_ZERO(&dispr
->fdset
);
95 dispatcher_add(Dispatcher
*dispr
, int sock
, void *ctx
, CallbackFunc cb
)
97 if (sock
>= FD_SETSIZE
) {
99 "Error: Failed to add new event. sock %d should be less than %d\n",
104 dispr
->events
[sock
].ctx
= ctx
;
105 dispr
->events
[sock
].callback
= cb
;
107 FD_SET(sock
, &dispr
->fdset
);
108 if (sock
> dispr
->max_sock
) {
109 dispr
->max_sock
= sock
;
111 DPRINT("Added sock %d for watching. max_sock: %d\n",
112 sock
, dispr
->max_sock
);
117 /* dispatcher_remove() is not currently in use but may be useful
120 dispatcher_remove(Dispatcher
*dispr
, int sock
)
122 if (sock
>= FD_SETSIZE
) {
124 "Error: Failed to remove event. sock %d should be less than %d\n",
129 FD_CLR(sock
, &dispr
->fdset
);
136 dispatcher_wait(Dispatcher
*dispr
, uint32_t timeout
)
139 tv
.tv_sec
= timeout
/ 1000000;
140 tv
.tv_usec
= timeout
% 1000000;
142 fd_set fdset
= dispr
->fdset
;
144 /* wait until some of sockets become readable. */
145 int rc
= select(dispr
->max_sock
+ 1, &fdset
, 0, 0, &tv
);
156 /* Now call callback for every ready socket. */
159 for (sock
= 0; sock
< dispr
->max_sock
+ 1; sock
++)
160 if (FD_ISSET(sock
, &fdset
)) {
161 Event
*e
= &dispr
->events
[sock
];
162 e
->callback(sock
, e
->ctx
);
168 typedef struct VubrVirtq
{
172 uint16_t last_avail_index
;
173 uint16_t last_used_index
;
174 struct vring_desc
*desc
;
175 struct vring_avail
*avail
;
176 struct vring_used
*used
;
177 uint64_t log_guest_addr
;
181 /* Based on qemu/hw/virtio/vhost-user.c */
183 #define VHOST_MEMORY_MAX_NREGIONS 8
184 #define VHOST_USER_F_PROTOCOL_FEATURES 30
186 #define VHOST_LOG_PAGE 4096
188 enum VhostUserProtocolFeature
{
189 VHOST_USER_PROTOCOL_F_MQ
= 0,
190 VHOST_USER_PROTOCOL_F_LOG_SHMFD
= 1,
191 VHOST_USER_PROTOCOL_F_RARP
= 2,
193 VHOST_USER_PROTOCOL_F_MAX
196 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
198 typedef enum VhostUserRequest
{
200 VHOST_USER_GET_FEATURES
= 1,
201 VHOST_USER_SET_FEATURES
= 2,
202 VHOST_USER_SET_OWNER
= 3,
203 VHOST_USER_RESET_OWNER
= 4,
204 VHOST_USER_SET_MEM_TABLE
= 5,
205 VHOST_USER_SET_LOG_BASE
= 6,
206 VHOST_USER_SET_LOG_FD
= 7,
207 VHOST_USER_SET_VRING_NUM
= 8,
208 VHOST_USER_SET_VRING_ADDR
= 9,
209 VHOST_USER_SET_VRING_BASE
= 10,
210 VHOST_USER_GET_VRING_BASE
= 11,
211 VHOST_USER_SET_VRING_KICK
= 12,
212 VHOST_USER_SET_VRING_CALL
= 13,
213 VHOST_USER_SET_VRING_ERR
= 14,
214 VHOST_USER_GET_PROTOCOL_FEATURES
= 15,
215 VHOST_USER_SET_PROTOCOL_FEATURES
= 16,
216 VHOST_USER_GET_QUEUE_NUM
= 17,
217 VHOST_USER_SET_VRING_ENABLE
= 18,
218 VHOST_USER_SEND_RARP
= 19,
222 typedef struct VhostUserMemoryRegion
{
223 uint64_t guest_phys_addr
;
224 uint64_t memory_size
;
225 uint64_t userspace_addr
;
226 uint64_t mmap_offset
;
227 } VhostUserMemoryRegion
;
229 typedef struct VhostUserMemory
{
232 VhostUserMemoryRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
235 typedef struct VhostUserLog
{
237 uint64_t mmap_offset
;
240 typedef struct VhostUserMsg
{
241 VhostUserRequest request
;
243 #define VHOST_USER_VERSION_MASK (0x3)
244 #define VHOST_USER_REPLY_MASK (0x1<<2)
246 uint32_t size
; /* the following payload size */
248 #define VHOST_USER_VRING_IDX_MASK (0xff)
249 #define VHOST_USER_VRING_NOFD_MASK (0x1<<8)
251 struct vhost_vring_state state
;
252 struct vhost_vring_addr addr
;
253 VhostUserMemory memory
;
256 int fds
[VHOST_MEMORY_MAX_NREGIONS
];
258 } QEMU_PACKED VhostUserMsg
;
260 #define VHOST_USER_HDR_SIZE offsetof(VhostUserMsg, payload.u64)
262 /* The version of the protocol we support */
263 #define VHOST_USER_VERSION (0x1)
265 #define MAX_NR_VIRTQUEUE (8)
267 typedef struct VubrDevRegion
{
268 /* Guest Physical address. */
270 /* Memory region size. */
272 /* QEMU virtual address (userspace). */
274 /* Starting offset in our mmaped space. */
275 uint64_t mmap_offset
;
276 /* Start address of mmaped space. */
280 typedef struct VubrDev
{
282 Dispatcher dispatcher
;
284 VubrDevRegion regions
[VHOST_MEMORY_MAX_NREGIONS
];
285 VubrVirtq vq
[MAX_NR_VIRTQUEUE
];
289 int backend_udp_sock
;
290 struct sockaddr_in backend_udp_dest
;
295 static const char *vubr_request_str
[] = {
296 [VHOST_USER_NONE
] = "VHOST_USER_NONE",
297 [VHOST_USER_GET_FEATURES
] = "VHOST_USER_GET_FEATURES",
298 [VHOST_USER_SET_FEATURES
] = "VHOST_USER_SET_FEATURES",
299 [VHOST_USER_SET_OWNER
] = "VHOST_USER_SET_OWNER",
300 [VHOST_USER_RESET_OWNER
] = "VHOST_USER_RESET_OWNER",
301 [VHOST_USER_SET_MEM_TABLE
] = "VHOST_USER_SET_MEM_TABLE",
302 [VHOST_USER_SET_LOG_BASE
] = "VHOST_USER_SET_LOG_BASE",
303 [VHOST_USER_SET_LOG_FD
] = "VHOST_USER_SET_LOG_FD",
304 [VHOST_USER_SET_VRING_NUM
] = "VHOST_USER_SET_VRING_NUM",
305 [VHOST_USER_SET_VRING_ADDR
] = "VHOST_USER_SET_VRING_ADDR",
306 [VHOST_USER_SET_VRING_BASE
] = "VHOST_USER_SET_VRING_BASE",
307 [VHOST_USER_GET_VRING_BASE
] = "VHOST_USER_GET_VRING_BASE",
308 [VHOST_USER_SET_VRING_KICK
] = "VHOST_USER_SET_VRING_KICK",
309 [VHOST_USER_SET_VRING_CALL
] = "VHOST_USER_SET_VRING_CALL",
310 [VHOST_USER_SET_VRING_ERR
] = "VHOST_USER_SET_VRING_ERR",
311 [VHOST_USER_GET_PROTOCOL_FEATURES
] = "VHOST_USER_GET_PROTOCOL_FEATURES",
312 [VHOST_USER_SET_PROTOCOL_FEATURES
] = "VHOST_USER_SET_PROTOCOL_FEATURES",
313 [VHOST_USER_GET_QUEUE_NUM
] = "VHOST_USER_GET_QUEUE_NUM",
314 [VHOST_USER_SET_VRING_ENABLE
] = "VHOST_USER_SET_VRING_ENABLE",
315 [VHOST_USER_SEND_RARP
] = "VHOST_USER_SEND_RARP",
316 [VHOST_USER_MAX
] = "VHOST_USER_MAX",
320 print_buffer(uint8_t *buf
, size_t len
)
323 printf("Raw buffer:\n");
324 for (i
= 0; i
< len
; i
++) {
331 printf("%02x ", buf
[i
]);
333 printf("\n............................................................\n");
336 /* Translate guest physical address to our virtual address. */
338 gpa_to_va(VubrDev
*dev
, uint64_t guest_addr
)
342 /* Find matching memory region. */
343 for (i
= 0; i
< dev
->nregions
; i
++) {
344 VubrDevRegion
*r
= &dev
->regions
[i
];
346 if ((guest_addr
>= r
->gpa
) && (guest_addr
< (r
->gpa
+ r
->size
))) {
347 return guest_addr
- r
->gpa
+ r
->mmap_addr
+ r
->mmap_offset
;
351 assert(!"address not found in regions");
355 /* Translate qemu virtual address to our virtual address. */
357 qva_to_va(VubrDev
*dev
, uint64_t qemu_addr
)
361 /* Find matching memory region. */
362 for (i
= 0; i
< dev
->nregions
; i
++) {
363 VubrDevRegion
*r
= &dev
->regions
[i
];
365 if ((qemu_addr
>= r
->qva
) && (qemu_addr
< (r
->qva
+ r
->size
))) {
366 return qemu_addr
- r
->qva
+ r
->mmap_addr
+ r
->mmap_offset
;
370 assert(!"address not found in regions");
375 vubr_message_read(int conn_fd
, VhostUserMsg
*vmsg
)
377 char control
[CMSG_SPACE(VHOST_MEMORY_MAX_NREGIONS
* sizeof(int))] = { };
379 .iov_base
= (char *)vmsg
,
380 .iov_len
= VHOST_USER_HDR_SIZE
,
382 struct msghdr msg
= {
385 .msg_control
= control
,
386 .msg_controllen
= sizeof(control
),
389 struct cmsghdr
*cmsg
;
392 rc
= recvmsg(conn_fd
, &msg
, 0);
396 fprintf(stderr
, "Peer disconnected.\n");
404 for (cmsg
= CMSG_FIRSTHDR(&msg
);
406 cmsg
= CMSG_NXTHDR(&msg
, cmsg
))
408 if (cmsg
->cmsg_level
== SOL_SOCKET
&& cmsg
->cmsg_type
== SCM_RIGHTS
) {
409 fd_size
= cmsg
->cmsg_len
- CMSG_LEN(0);
410 vmsg
->fd_num
= fd_size
/ sizeof(int);
411 memcpy(vmsg
->fds
, CMSG_DATA(cmsg
), fd_size
);
416 if (vmsg
->size
> sizeof(vmsg
->payload
)) {
418 "Error: too big message request: %d, size: vmsg->size: %u, "
419 "while sizeof(vmsg->payload) = %lu\n",
420 vmsg
->request
, vmsg
->size
, sizeof(vmsg
->payload
));
425 rc
= read(conn_fd
, &vmsg
->payload
, vmsg
->size
);
428 fprintf(stderr
, "Peer disconnected.\n");
435 assert(rc
== vmsg
->size
);
440 vubr_message_write(int conn_fd
, VhostUserMsg
*vmsg
)
445 rc
= write(conn_fd
, vmsg
, VHOST_USER_HDR_SIZE
+ vmsg
->size
);
446 } while (rc
< 0 && errno
== EINTR
);
454 vubr_backend_udp_sendbuf(VubrDev
*dev
, uint8_t *buf
, size_t len
)
456 int slen
= sizeof(struct sockaddr_in
);
458 if (sendto(dev
->backend_udp_sock
, buf
, len
, 0,
459 (struct sockaddr
*) &dev
->backend_udp_dest
, slen
) == -1) {
460 vubr_die("sendto()");
465 vubr_backend_udp_recvbuf(VubrDev
*dev
, uint8_t *buf
, size_t buflen
)
467 int slen
= sizeof(struct sockaddr_in
);
470 rc
= recvfrom(dev
->backend_udp_sock
, buf
, buflen
, 0,
471 (struct sockaddr
*) &dev
->backend_udp_dest
,
474 vubr_die("recvfrom()");
481 vubr_consume_raw_packet(VubrDev
*dev
, uint8_t *buf
, uint32_t len
)
483 int hdrlen
= sizeof(struct virtio_net_hdr_v1
);
485 if (VHOST_USER_BRIDGE_DEBUG
) {
486 print_buffer(buf
, len
);
488 vubr_backend_udp_sendbuf(dev
, buf
+ hdrlen
, len
- hdrlen
);
491 /* Kick the log_call_fd if required. */
493 vubr_log_kick(VubrDev
*dev
)
495 if (dev
->log_call_fd
!= -1) {
496 DPRINT("Kicking the QEMU's log...\n");
497 eventfd_write(dev
->log_call_fd
, 1);
501 /* Kick the guest if necessary. */
503 vubr_virtqueue_kick(VubrVirtq
*vq
)
505 if (!(vq
->avail
->flags
& VRING_AVAIL_F_NO_INTERRUPT
)) {
506 DPRINT("Kicking the guest...\n");
507 eventfd_write(vq
->call_fd
, 1);
512 vubr_log_page(uint8_t *log_table
, uint64_t page
)
514 DPRINT("Logged dirty guest page: %"PRId64
"\n", page
);
515 atomic_or(&log_table
[page
/ 8], 1 << (page
% 8));
519 vubr_log_write(VubrDev
*dev
, uint64_t address
, uint64_t length
)
523 if (!(dev
->features
& (1ULL << VHOST_F_LOG_ALL
)) ||
524 !dev
->log_table
|| !length
) {
528 assert(dev
->log_size
> ((address
+ length
- 1) / VHOST_LOG_PAGE
/ 8));
530 page
= address
/ VHOST_LOG_PAGE
;
531 while (page
* VHOST_LOG_PAGE
< address
+ length
) {
532 vubr_log_page(dev
->log_table
, page
);
533 page
+= VHOST_LOG_PAGE
;
539 vubr_post_buffer(VubrDev
*dev
, VubrVirtq
*vq
, uint8_t *buf
, int32_t len
)
541 struct vring_desc
*desc
= vq
->desc
;
542 struct vring_avail
*avail
= vq
->avail
;
543 struct vring_used
*used
= vq
->used
;
544 uint64_t log_guest_addr
= vq
->log_guest_addr
;
546 unsigned int size
= vq
->size
;
548 uint16_t avail_index
= atomic_mb_read(&avail
->idx
);
550 /* We check the available descriptors before posting the
551 * buffer, so here we assume that enough available
553 assert(vq
->last_avail_index
!= avail_index
);
554 uint16_t a_index
= vq
->last_avail_index
% size
;
555 uint16_t u_index
= vq
->last_used_index
% size
;
556 uint16_t d_index
= avail
->ring
[a_index
];
560 DPRINT("Post packet to guest on vq:\n");
561 DPRINT(" size = %d\n", vq
->size
);
562 DPRINT(" last_avail_index = %d\n", vq
->last_avail_index
);
563 DPRINT(" last_used_index = %d\n", vq
->last_used_index
);
564 DPRINT(" a_index = %d\n", a_index
);
565 DPRINT(" u_index = %d\n", u_index
);
566 DPRINT(" d_index = %d\n", d_index
);
567 DPRINT(" desc[%d].addr = 0x%016"PRIx64
"\n", i
, desc
[i
].addr
);
568 DPRINT(" desc[%d].len = %d\n", i
, desc
[i
].len
);
569 DPRINT(" desc[%d].flags = %d\n", i
, desc
[i
].flags
);
570 DPRINT(" avail->idx = %d\n", avail_index
);
571 DPRINT(" used->idx = %d\n", used
->idx
);
573 if (!(desc
[i
].flags
& VRING_DESC_F_WRITE
)) {
574 /* FIXME: we should find writable descriptor. */
575 fprintf(stderr
, "Error: descriptor is not writable. Exiting.\n");
579 void *chunk_start
= (void *)gpa_to_va(dev
, desc
[i
].addr
);
580 uint32_t chunk_len
= desc
[i
].len
;
582 if (len
<= chunk_len
) {
583 memcpy(chunk_start
, buf
, len
);
584 vubr_log_write(dev
, desc
[i
].addr
, len
);
587 "Received too long packet from the backend. Dropping...\n");
591 /* Add descriptor to the used ring. */
592 used
->ring
[u_index
].id
= d_index
;
593 used
->ring
[u_index
].len
= len
;
595 log_guest_addr
+ offsetof(struct vring_used
, ring
[u_index
]),
596 sizeof(used
->ring
[u_index
]));
598 vq
->last_avail_index
++;
599 vq
->last_used_index
++;
601 atomic_mb_set(&used
->idx
, vq
->last_used_index
);
603 log_guest_addr
+ offsetof(struct vring_used
, idx
),
606 /* Kick the guest if necessary. */
607 vubr_virtqueue_kick(vq
);
611 vubr_process_desc(VubrDev
*dev
, VubrVirtq
*vq
)
613 struct vring_desc
*desc
= vq
->desc
;
614 struct vring_avail
*avail
= vq
->avail
;
615 struct vring_used
*used
= vq
->used
;
616 uint64_t log_guest_addr
= vq
->log_guest_addr
;
618 unsigned int size
= vq
->size
;
620 uint16_t a_index
= vq
->last_avail_index
% size
;
621 uint16_t u_index
= vq
->last_used_index
% size
;
622 uint16_t d_index
= avail
->ring
[a_index
];
625 size_t buf_size
= 4096;
631 void *chunk_start
= (void *)gpa_to_va(dev
, desc
[i
].addr
);
632 uint32_t chunk_len
= desc
[i
].len
;
634 assert(!(desc
[i
].flags
& VRING_DESC_F_WRITE
));
636 if (len
+ chunk_len
< buf_size
) {
637 memcpy(buf
+ len
, chunk_start
, chunk_len
);
638 DPRINT("%d ", chunk_len
);
640 fprintf(stderr
, "Error: too long packet. Dropping...\n");
646 if (!(desc
[i
].flags
& VRING_DESC_F_NEXT
)) {
658 /* Add descriptor to the used ring. */
659 used
->ring
[u_index
].id
= d_index
;
660 used
->ring
[u_index
].len
= len
;
662 log_guest_addr
+ offsetof(struct vring_used
, ring
[u_index
]),
663 sizeof(used
->ring
[u_index
]));
665 vubr_consume_raw_packet(dev
, buf
, len
);
671 vubr_process_avail(VubrDev
*dev
, VubrVirtq
*vq
)
673 struct vring_avail
*avail
= vq
->avail
;
674 struct vring_used
*used
= vq
->used
;
675 uint64_t log_guest_addr
= vq
->log_guest_addr
;
677 while (vq
->last_avail_index
!= atomic_mb_read(&avail
->idx
)) {
678 vubr_process_desc(dev
, vq
);
679 vq
->last_avail_index
++;
680 vq
->last_used_index
++;
683 atomic_mb_set(&used
->idx
, vq
->last_used_index
);
685 log_guest_addr
+ offsetof(struct vring_used
, idx
),
690 vubr_backend_recv_cb(int sock
, void *ctx
)
692 VubrDev
*dev
= (VubrDev
*) ctx
;
693 VubrVirtq
*rx_vq
= &dev
->vq
[0];
695 struct virtio_net_hdr_v1
*hdr
= (struct virtio_net_hdr_v1
*)buf
;
696 int hdrlen
= sizeof(struct virtio_net_hdr_v1
);
697 int buflen
= sizeof(buf
);
704 DPRINT("\n\n *** IN UDP RECEIVE CALLBACK ***\n\n");
706 uint16_t avail_index
= atomic_mb_read(&rx_vq
->avail
->idx
);
708 /* If there is no available descriptors, just do nothing.
709 * The buffer will be handled by next arrived UDP packet,
710 * or next kick on receive virtq. */
711 if (rx_vq
->last_avail_index
== avail_index
) {
712 DPRINT("Got UDP packet, but no available descriptors on RX virtq.\n");
716 len
= vubr_backend_udp_recvbuf(dev
, buf
+ hdrlen
, buflen
- hdrlen
);
718 *hdr
= (struct virtio_net_hdr_v1
) { };
719 hdr
->num_buffers
= 1;
720 vubr_post_buffer(dev
, rx_vq
, buf
, len
+ hdrlen
);
724 vubr_kick_cb(int sock
, void *ctx
)
726 VubrDev
*dev
= (VubrDev
*) ctx
;
730 rc
= eventfd_read(sock
, &kick_data
);
732 vubr_die("eventfd_read()");
734 DPRINT("Got kick_data: %016"PRIx64
"\n", kick_data
);
735 vubr_process_avail(dev
, &dev
->vq
[1]);
740 vubr_none_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
742 DPRINT("Function %s() not implemented yet.\n", __func__
);
747 vubr_get_features_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
750 ((1ULL << VIRTIO_NET_F_MRG_RXBUF
) |
751 (1ULL << VHOST_F_LOG_ALL
) |
752 (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE
) |
753 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES
));
755 vmsg
->size
= sizeof(vmsg
->payload
.u64
);
757 DPRINT("Sending back to guest u64: 0x%016"PRIx64
"\n", vmsg
->payload
.u64
);
764 vubr_set_features_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
766 DPRINT("u64: 0x%016"PRIx64
"\n", vmsg
->payload
.u64
);
767 dev
->features
= vmsg
->payload
.u64
;
772 vubr_set_owner_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
778 vubr_close_log(VubrDev
*dev
)
780 if (dev
->log_table
) {
781 if (munmap(dev
->log_table
, dev
->log_size
) != 0) {
782 vubr_die("munmap()");
787 if (dev
->log_call_fd
!= -1) {
788 close(dev
->log_call_fd
);
789 dev
->log_call_fd
= -1;
794 vubr_reset_device_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
803 vubr_set_mem_table_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
806 VhostUserMemory
*memory
= &vmsg
->payload
.memory
;
807 dev
->nregions
= memory
->nregions
;
809 DPRINT("Nregions: %d\n", memory
->nregions
);
810 for (i
= 0; i
< dev
->nregions
; i
++) {
812 VhostUserMemoryRegion
*msg_region
= &memory
->regions
[i
];
813 VubrDevRegion
*dev_region
= &dev
->regions
[i
];
815 DPRINT("Region %d\n", i
);
816 DPRINT(" guest_phys_addr: 0x%016"PRIx64
"\n",
817 msg_region
->guest_phys_addr
);
818 DPRINT(" memory_size: 0x%016"PRIx64
"\n",
819 msg_region
->memory_size
);
820 DPRINT(" userspace_addr 0x%016"PRIx64
"\n",
821 msg_region
->userspace_addr
);
822 DPRINT(" mmap_offset 0x%016"PRIx64
"\n",
823 msg_region
->mmap_offset
);
825 dev_region
->gpa
= msg_region
->guest_phys_addr
;
826 dev_region
->size
= msg_region
->memory_size
;
827 dev_region
->qva
= msg_region
->userspace_addr
;
828 dev_region
->mmap_offset
= msg_region
->mmap_offset
;
830 /* We don't use offset argument of mmap() since the
831 * mapped address has to be page aligned, and we use huge
833 mmap_addr
= mmap(0, dev_region
->size
+ dev_region
->mmap_offset
,
834 PROT_READ
| PROT_WRITE
, MAP_SHARED
,
837 if (mmap_addr
== MAP_FAILED
) {
841 dev_region
->mmap_addr
= (uint64_t) mmap_addr
;
842 DPRINT(" mmap_addr: 0x%016"PRIx64
"\n", dev_region
->mmap_addr
);
849 vubr_set_log_base_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
852 uint64_t log_mmap_size
, log_mmap_offset
;
855 assert(vmsg
->fd_num
== 1);
858 assert(vmsg
->size
== sizeof(vmsg
->payload
.log
));
859 log_mmap_offset
= vmsg
->payload
.log
.mmap_offset
;
860 log_mmap_size
= vmsg
->payload
.log
.mmap_size
;
861 DPRINT("Log mmap_offset: %"PRId64
"\n", log_mmap_offset
);
862 DPRINT("Log mmap_size: %"PRId64
"\n", log_mmap_size
);
864 rc
= mmap(0, log_mmap_size
, PROT_READ
| PROT_WRITE
, MAP_SHARED
, fd
,
866 if (rc
== MAP_FAILED
) {
870 dev
->log_size
= log_mmap_size
;
872 vmsg
->size
= sizeof(vmsg
->payload
.u64
);
878 vubr_set_log_fd_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
880 assert(vmsg
->fd_num
== 1);
881 dev
->log_call_fd
= vmsg
->fds
[0];
882 DPRINT("Got log_call_fd: %d\n", vmsg
->fds
[0]);
887 vubr_set_vring_num_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
889 unsigned int index
= vmsg
->payload
.state
.index
;
890 unsigned int num
= vmsg
->payload
.state
.num
;
892 DPRINT("State.index: %d\n", index
);
893 DPRINT("State.num: %d\n", num
);
894 dev
->vq
[index
].size
= num
;
899 vubr_set_vring_addr_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
901 struct vhost_vring_addr
*vra
= &vmsg
->payload
.addr
;
902 unsigned int index
= vra
->index
;
903 VubrVirtq
*vq
= &dev
->vq
[index
];
905 DPRINT("vhost_vring_addr:\n");
906 DPRINT(" index: %d\n", vra
->index
);
907 DPRINT(" flags: %d\n", vra
->flags
);
908 DPRINT(" desc_user_addr: 0x%016llx\n", vra
->desc_user_addr
);
909 DPRINT(" used_user_addr: 0x%016llx\n", vra
->used_user_addr
);
910 DPRINT(" avail_user_addr: 0x%016llx\n", vra
->avail_user_addr
);
911 DPRINT(" log_guest_addr: 0x%016llx\n", vra
->log_guest_addr
);
913 vq
->desc
= (struct vring_desc
*)qva_to_va(dev
, vra
->desc_user_addr
);
914 vq
->used
= (struct vring_used
*)qva_to_va(dev
, vra
->used_user_addr
);
915 vq
->avail
= (struct vring_avail
*)qva_to_va(dev
, vra
->avail_user_addr
);
916 vq
->log_guest_addr
= vra
->log_guest_addr
;
918 DPRINT("Setting virtq addresses:\n");
919 DPRINT(" vring_desc at %p\n", vq
->desc
);
920 DPRINT(" vring_used at %p\n", vq
->used
);
921 DPRINT(" vring_avail at %p\n", vq
->avail
);
923 vq
->last_used_index
= vq
->used
->idx
;
928 vubr_set_vring_base_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
930 unsigned int index
= vmsg
->payload
.state
.index
;
931 unsigned int num
= vmsg
->payload
.state
.num
;
933 DPRINT("State.index: %d\n", index
);
934 DPRINT("State.num: %d\n", num
);
935 dev
->vq
[index
].last_avail_index
= num
;
941 vubr_get_vring_base_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
943 unsigned int index
= vmsg
->payload
.state
.index
;
945 DPRINT("State.index: %d\n", index
);
946 vmsg
->payload
.state
.num
= dev
->vq
[index
].last_avail_index
;
947 vmsg
->size
= sizeof(vmsg
->payload
.state
);
948 /* FIXME: this is a work-around for a bug in QEMU enabling
949 * too early vrings. When protocol features are enabled,
950 * we have to respect * VHOST_USER_SET_VRING_ENABLE request. */
958 vubr_set_vring_kick_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
960 uint64_t u64_arg
= vmsg
->payload
.u64
;
961 int index
= u64_arg
& VHOST_USER_VRING_IDX_MASK
;
963 DPRINT("u64: 0x%016"PRIx64
"\n", vmsg
->payload
.u64
);
965 assert((u64_arg
& VHOST_USER_VRING_NOFD_MASK
) == 0);
966 assert(vmsg
->fd_num
== 1);
968 dev
->vq
[index
].kick_fd
= vmsg
->fds
[0];
969 DPRINT("Got kick_fd: %d for vq: %d\n", vmsg
->fds
[0], index
);
971 if (index
% 2 == 1) {
973 dispatcher_add(&dev
->dispatcher
, dev
->vq
[index
].kick_fd
,
976 DPRINT("Waiting for kicks on fd: %d for vq: %d\n",
977 dev
->vq
[index
].kick_fd
, index
);
979 /* We temporarily use this hack to determine that both TX and RX
980 * queues are set up and ready for processing.
981 * FIXME: we need to rely in VHOST_USER_SET_VRING_ENABLE and
983 if (dev
->vq
[0].kick_fd
!= -1 &&
984 dev
->vq
[1].kick_fd
!= -1) {
986 DPRINT("vhost-user-bridge is ready for processing queues.\n");
993 vubr_set_vring_call_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
995 uint64_t u64_arg
= vmsg
->payload
.u64
;
996 int index
= u64_arg
& VHOST_USER_VRING_IDX_MASK
;
998 DPRINT("u64: 0x%016"PRIx64
"\n", vmsg
->payload
.u64
);
999 assert((u64_arg
& VHOST_USER_VRING_NOFD_MASK
) == 0);
1000 assert(vmsg
->fd_num
== 1);
1002 dev
->vq
[index
].call_fd
= vmsg
->fds
[0];
1003 DPRINT("Got call_fd: %d for vq: %d\n", vmsg
->fds
[0], index
);
1009 vubr_set_vring_err_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
1011 DPRINT("u64: 0x%016"PRIx64
"\n", vmsg
->payload
.u64
);
1016 vubr_get_protocol_features_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
1018 vmsg
->payload
.u64
= 1ULL << VHOST_USER_PROTOCOL_F_LOG_SHMFD
;
1019 DPRINT("u64: 0x%016"PRIx64
"\n", vmsg
->payload
.u64
);
1020 vmsg
->size
= sizeof(vmsg
->payload
.u64
);
1027 vubr_set_protocol_features_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
1029 /* FIXME: unimplented */
1030 DPRINT("u64: 0x%016"PRIx64
"\n", vmsg
->payload
.u64
);
1035 vubr_get_queue_num_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
1037 DPRINT("Function %s() not implemented yet.\n", __func__
);
1042 vubr_set_vring_enable_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
1044 unsigned int index
= vmsg
->payload
.state
.index
;
1045 unsigned int enable
= vmsg
->payload
.state
.num
;
1047 DPRINT("State.index: %d\n", index
);
1048 DPRINT("State.enable: %d\n", enable
);
1049 dev
->vq
[index
].enable
= enable
;
1054 vubr_send_rarp_exec(VubrDev
*dev
, VhostUserMsg
*vmsg
)
1056 DPRINT("Function %s() not implemented yet.\n", __func__
);
1061 vubr_execute_request(VubrDev
*dev
, VhostUserMsg
*vmsg
)
1063 /* Print out generic part of the request. */
1065 "================== Vhost user message from QEMU ==================\n");
1066 DPRINT("Request: %s (%d)\n", vubr_request_str
[vmsg
->request
],
1068 DPRINT("Flags: 0x%x\n", vmsg
->flags
);
1069 DPRINT("Size: %d\n", vmsg
->size
);
1074 for (i
= 0; i
< vmsg
->fd_num
; i
++) {
1075 DPRINT(" %d", vmsg
->fds
[i
]);
1080 switch (vmsg
->request
) {
1081 case VHOST_USER_NONE
:
1082 return vubr_none_exec(dev
, vmsg
);
1083 case VHOST_USER_GET_FEATURES
:
1084 return vubr_get_features_exec(dev
, vmsg
);
1085 case VHOST_USER_SET_FEATURES
:
1086 return vubr_set_features_exec(dev
, vmsg
);
1087 case VHOST_USER_SET_OWNER
:
1088 return vubr_set_owner_exec(dev
, vmsg
);
1089 case VHOST_USER_RESET_OWNER
:
1090 return vubr_reset_device_exec(dev
, vmsg
);
1091 case VHOST_USER_SET_MEM_TABLE
:
1092 return vubr_set_mem_table_exec(dev
, vmsg
);
1093 case VHOST_USER_SET_LOG_BASE
:
1094 return vubr_set_log_base_exec(dev
, vmsg
);
1095 case VHOST_USER_SET_LOG_FD
:
1096 return vubr_set_log_fd_exec(dev
, vmsg
);
1097 case VHOST_USER_SET_VRING_NUM
:
1098 return vubr_set_vring_num_exec(dev
, vmsg
);
1099 case VHOST_USER_SET_VRING_ADDR
:
1100 return vubr_set_vring_addr_exec(dev
, vmsg
);
1101 case VHOST_USER_SET_VRING_BASE
:
1102 return vubr_set_vring_base_exec(dev
, vmsg
);
1103 case VHOST_USER_GET_VRING_BASE
:
1104 return vubr_get_vring_base_exec(dev
, vmsg
);
1105 case VHOST_USER_SET_VRING_KICK
:
1106 return vubr_set_vring_kick_exec(dev
, vmsg
);
1107 case VHOST_USER_SET_VRING_CALL
:
1108 return vubr_set_vring_call_exec(dev
, vmsg
);
1109 case VHOST_USER_SET_VRING_ERR
:
1110 return vubr_set_vring_err_exec(dev
, vmsg
);
1111 case VHOST_USER_GET_PROTOCOL_FEATURES
:
1112 return vubr_get_protocol_features_exec(dev
, vmsg
);
1113 case VHOST_USER_SET_PROTOCOL_FEATURES
:
1114 return vubr_set_protocol_features_exec(dev
, vmsg
);
1115 case VHOST_USER_GET_QUEUE_NUM
:
1116 return vubr_get_queue_num_exec(dev
, vmsg
);
1117 case VHOST_USER_SET_VRING_ENABLE
:
1118 return vubr_set_vring_enable_exec(dev
, vmsg
);
1119 case VHOST_USER_SEND_RARP
:
1120 return vubr_send_rarp_exec(dev
, vmsg
);
1122 case VHOST_USER_MAX
:
1123 assert(vmsg
->request
!= VHOST_USER_MAX
);
1129 vubr_receive_cb(int sock
, void *ctx
)
1131 VubrDev
*dev
= (VubrDev
*) ctx
;
1133 int reply_requested
;
1135 vubr_message_read(sock
, &vmsg
);
1136 reply_requested
= vubr_execute_request(dev
, &vmsg
);
1137 if (reply_requested
) {
1138 /* Set the version in the flags when sending the reply */
1139 vmsg
.flags
&= ~VHOST_USER_VERSION_MASK
;
1140 vmsg
.flags
|= VHOST_USER_VERSION
;
1141 vmsg
.flags
|= VHOST_USER_REPLY_MASK
;
1142 vubr_message_write(sock
, &vmsg
);
1147 vubr_accept_cb(int sock
, void *ctx
)
1149 VubrDev
*dev
= (VubrDev
*)ctx
;
1151 struct sockaddr_un un
;
1152 socklen_t len
= sizeof(un
);
1154 conn_fd
= accept(sock
, (struct sockaddr
*) &un
, &len
);
1155 if (conn_fd
== -1) {
1156 vubr_die("accept()");
1158 DPRINT("Got connection from remote peer on sock %d\n", conn_fd
);
1159 dispatcher_add(&dev
->dispatcher
, conn_fd
, ctx
, vubr_receive_cb
);
1163 vubr_new(const char *path
)
1165 VubrDev
*dev
= (VubrDev
*) calloc(1, sizeof(VubrDev
));
1168 struct sockaddr_un un
;
1171 for (i
= 0; i
< MAX_NR_VIRTQUEUE
; i
++) {
1172 dev
->vq
[i
] = (VubrVirtq
) {
1173 .call_fd
= -1, .kick_fd
= -1,
1175 .last_avail_index
= 0, .last_used_index
= 0,
1176 .desc
= 0, .avail
= 0, .used
= 0,
1182 dev
->log_call_fd
= -1;
1188 /* Get a UNIX socket. */
1189 dev
->sock
= socket(AF_UNIX
, SOCK_STREAM
, 0);
1190 if (dev
->sock
== -1) {
1194 un
.sun_family
= AF_UNIX
;
1195 strcpy(un
.sun_path
, path
);
1196 len
= sizeof(un
.sun_family
) + strlen(path
);
1199 if (bind(dev
->sock
, (struct sockaddr
*) &un
, len
) == -1) {
1203 if (listen(dev
->sock
, 1) == -1) {
1207 dispatcher_init(&dev
->dispatcher
);
1208 dispatcher_add(&dev
->dispatcher
, dev
->sock
, (void *)dev
,
1211 DPRINT("Waiting for connections on UNIX socket %s ...\n", path
);
1216 vubr_set_host(struct sockaddr_in
*saddr
, const char *host
)
1218 if (isdigit(host
[0])) {
1219 if (!inet_aton(host
, &saddr
->sin_addr
)) {
1220 fprintf(stderr
, "inet_aton() failed.\n");
1224 struct hostent
*he
= gethostbyname(host
);
1227 fprintf(stderr
, "gethostbyname() failed.\n");
1230 saddr
->sin_addr
= *(struct in_addr
*)he
->h_addr
;
1235 vubr_backend_udp_setup(VubrDev
*dev
,
1236 const char *local_host
,
1237 const char *local_port
,
1238 const char *remote_host
,
1239 const char *remote_port
)
1246 lport
= strtol(local_port
, (char **)&r
, 0);
1247 if (r
== local_port
) {
1248 fprintf(stderr
, "lport parsing failed.\n");
1252 rport
= strtol(remote_port
, (char **)&r
, 0);
1253 if (r
== remote_port
) {
1254 fprintf(stderr
, "rport parsing failed.\n");
1258 struct sockaddr_in si_local
= {
1259 .sin_family
= AF_INET
,
1260 .sin_port
= htons(lport
),
1263 vubr_set_host(&si_local
, local_host
);
1265 /* setup destination for sends */
1266 dev
->backend_udp_dest
= (struct sockaddr_in
) {
1267 .sin_family
= AF_INET
,
1268 .sin_port
= htons(rport
),
1270 vubr_set_host(&dev
->backend_udp_dest
, remote_host
);
1272 sock
= socket(AF_INET
, SOCK_DGRAM
, IPPROTO_UDP
);
1277 if (bind(sock
, (struct sockaddr
*)&si_local
, sizeof(si_local
)) == -1) {
1281 dev
->backend_udp_sock
= sock
;
1282 dispatcher_add(&dev
->dispatcher
, sock
, dev
, vubr_backend_recv_cb
);
1283 DPRINT("Waiting for data from udp backend on %s:%d...\n",
1288 vubr_run(VubrDev
*dev
)
1292 dispatcher_wait(&dev
->dispatcher
, 200000);
1293 /* Here one can try polling strategy. */
1298 vubr_parse_host_port(const char **host
, const char **port
, const char *buf
)
1300 char *p
= strchr(buf
, ':');
1306 *host
= strdup(buf
);
1307 *port
= strdup(p
+ 1);
1311 #define DEFAULT_UD_SOCKET "/tmp/vubr.sock"
1312 #define DEFAULT_LHOST "127.0.0.1"
1313 #define DEFAULT_LPORT "4444"
1314 #define DEFAULT_RHOST "127.0.0.1"
1315 #define DEFAULT_RPORT "5555"
1317 static const char *ud_socket_path
= DEFAULT_UD_SOCKET
;
1318 static const char *lhost
= DEFAULT_LHOST
;
1319 static const char *lport
= DEFAULT_LPORT
;
1320 static const char *rhost
= DEFAULT_RHOST
;
1321 static const char *rport
= DEFAULT_RPORT
;
1324 main(int argc
, char *argv
[])
1329 while ((opt
= getopt(argc
, argv
, "l:r:u:")) != -1) {
1333 if (vubr_parse_host_port(&lhost
, &lport
, optarg
) < 0) {
1338 if (vubr_parse_host_port(&rhost
, &rport
, optarg
) < 0) {
1343 ud_socket_path
= strdup(optarg
);
1350 DPRINT("ud socket: %s\n", ud_socket_path
);
1351 DPRINT("local: %s:%s\n", lhost
, lport
);
1352 DPRINT("remote: %s:%s\n", rhost
, rport
);
1354 dev
= vubr_new(ud_socket_path
);
1359 vubr_backend_udp_setup(dev
, lhost
, lport
, rhost
, rport
);
1364 fprintf(stderr
, "Usage: %s ", argv
[0]);
1365 fprintf(stderr
, "[-u ud_socket_path] [-l lhost:lport] [-r rhost:rport]\n");
1366 fprintf(stderr
, "\t-u path to unix doman socket. default: %s\n",
1368 fprintf(stderr
, "\t-l local host and port. default: %s:%s\n",
1369 DEFAULT_LHOST
, DEFAULT_LPORT
);
1370 fprintf(stderr
, "\t-r remote host and port. default: %s:%s\n",
1371 DEFAULT_RHOST
, DEFAULT_RPORT
);