]> git.ipfire.org Git - thirdparty/qemu.git/blob - hw/virtio/vhost-user.c
include/qemu/osdep.h: Don't include qapi/error.h
[thirdparty/qemu.git] / hw / virtio / vhost-user.c
1 /*
2 * vhost-user
3 *
4 * Copyright (c) 2013 Virtual Open Systems Sarl.
5 *
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.
8 *
9 */
10
11 #include "qemu/osdep.h"
12 #include "qapi/error.h"
13 #include "hw/virtio/vhost.h"
14 #include "hw/virtio/vhost-backend.h"
15 #include "hw/virtio/virtio-net.h"
16 #include "sysemu/char.h"
17 #include "sysemu/kvm.h"
18 #include "qemu/error-report.h"
19 #include "qemu/sockets.h"
20 #include "exec/ram_addr.h"
21 #include "migration/migration.h"
22
23 #include <sys/ioctl.h>
24 #include <sys/socket.h>
25 #include <sys/un.h>
26 #include <linux/vhost.h>
27
28 #define VHOST_MEMORY_MAX_NREGIONS 8
29 #define VHOST_USER_F_PROTOCOL_FEATURES 30
30
31 enum VhostUserProtocolFeature {
32 VHOST_USER_PROTOCOL_F_MQ = 0,
33 VHOST_USER_PROTOCOL_F_LOG_SHMFD = 1,
34 VHOST_USER_PROTOCOL_F_RARP = 2,
35
36 VHOST_USER_PROTOCOL_F_MAX
37 };
38
39 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
40
41 typedef enum VhostUserRequest {
42 VHOST_USER_NONE = 0,
43 VHOST_USER_GET_FEATURES = 1,
44 VHOST_USER_SET_FEATURES = 2,
45 VHOST_USER_SET_OWNER = 3,
46 VHOST_USER_RESET_OWNER = 4,
47 VHOST_USER_SET_MEM_TABLE = 5,
48 VHOST_USER_SET_LOG_BASE = 6,
49 VHOST_USER_SET_LOG_FD = 7,
50 VHOST_USER_SET_VRING_NUM = 8,
51 VHOST_USER_SET_VRING_ADDR = 9,
52 VHOST_USER_SET_VRING_BASE = 10,
53 VHOST_USER_GET_VRING_BASE = 11,
54 VHOST_USER_SET_VRING_KICK = 12,
55 VHOST_USER_SET_VRING_CALL = 13,
56 VHOST_USER_SET_VRING_ERR = 14,
57 VHOST_USER_GET_PROTOCOL_FEATURES = 15,
58 VHOST_USER_SET_PROTOCOL_FEATURES = 16,
59 VHOST_USER_GET_QUEUE_NUM = 17,
60 VHOST_USER_SET_VRING_ENABLE = 18,
61 VHOST_USER_SEND_RARP = 19,
62 VHOST_USER_MAX
63 } VhostUserRequest;
64
65 typedef struct VhostUserMemoryRegion {
66 uint64_t guest_phys_addr;
67 uint64_t memory_size;
68 uint64_t userspace_addr;
69 uint64_t mmap_offset;
70 } VhostUserMemoryRegion;
71
72 typedef struct VhostUserMemory {
73 uint32_t nregions;
74 uint32_t padding;
75 VhostUserMemoryRegion regions[VHOST_MEMORY_MAX_NREGIONS];
76 } VhostUserMemory;
77
78 typedef struct VhostUserLog {
79 uint64_t mmap_size;
80 uint64_t mmap_offset;
81 } VhostUserLog;
82
83 typedef struct VhostUserMsg {
84 VhostUserRequest request;
85
86 #define VHOST_USER_VERSION_MASK (0x3)
87 #define VHOST_USER_REPLY_MASK (0x1<<2)
88 uint32_t flags;
89 uint32_t size; /* the following payload size */
90 union {
91 #define VHOST_USER_VRING_IDX_MASK (0xff)
92 #define VHOST_USER_VRING_NOFD_MASK (0x1<<8)
93 uint64_t u64;
94 struct vhost_vring_state state;
95 struct vhost_vring_addr addr;
96 VhostUserMemory memory;
97 VhostUserLog log;
98 } payload;
99 } QEMU_PACKED VhostUserMsg;
100
101 static VhostUserMsg m __attribute__ ((unused));
102 #define VHOST_USER_HDR_SIZE (sizeof(m.request) \
103 + sizeof(m.flags) \
104 + sizeof(m.size))
105
106 #define VHOST_USER_PAYLOAD_SIZE (sizeof(m) - VHOST_USER_HDR_SIZE)
107
108 /* The version of the protocol we support */
109 #define VHOST_USER_VERSION (0x1)
110
111 static bool ioeventfd_enabled(void)
112 {
113 return kvm_enabled() && kvm_eventfds_enabled();
114 }
115
116 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
117 {
118 CharDriverState *chr = dev->opaque;
119 uint8_t *p = (uint8_t *) msg;
120 int r, size = VHOST_USER_HDR_SIZE;
121
122 r = qemu_chr_fe_read_all(chr, p, size);
123 if (r != size) {
124 error_report("Failed to read msg header. Read %d instead of %d."
125 " Original request %d.", r, size, msg->request);
126 goto fail;
127 }
128
129 /* validate received flags */
130 if (msg->flags != (VHOST_USER_REPLY_MASK | VHOST_USER_VERSION)) {
131 error_report("Failed to read msg header."
132 " Flags 0x%x instead of 0x%x.", msg->flags,
133 VHOST_USER_REPLY_MASK | VHOST_USER_VERSION);
134 goto fail;
135 }
136
137 /* validate message size is sane */
138 if (msg->size > VHOST_USER_PAYLOAD_SIZE) {
139 error_report("Failed to read msg header."
140 " Size %d exceeds the maximum %zu.", msg->size,
141 VHOST_USER_PAYLOAD_SIZE);
142 goto fail;
143 }
144
145 if (msg->size) {
146 p += VHOST_USER_HDR_SIZE;
147 size = msg->size;
148 r = qemu_chr_fe_read_all(chr, p, size);
149 if (r != size) {
150 error_report("Failed to read msg payload."
151 " Read %d instead of %d.", r, msg->size);
152 goto fail;
153 }
154 }
155
156 return 0;
157
158 fail:
159 return -1;
160 }
161
162 static bool vhost_user_one_time_request(VhostUserRequest request)
163 {
164 switch (request) {
165 case VHOST_USER_SET_OWNER:
166 case VHOST_USER_RESET_OWNER:
167 case VHOST_USER_SET_MEM_TABLE:
168 case VHOST_USER_GET_QUEUE_NUM:
169 return true;
170 default:
171 return false;
172 }
173 }
174
175 /* most non-init callers ignore the error */
176 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
177 int *fds, int fd_num)
178 {
179 CharDriverState *chr = dev->opaque;
180 int size = VHOST_USER_HDR_SIZE + msg->size;
181
182 /*
183 * For non-vring specific requests, like VHOST_USER_SET_MEM_TABLE,
184 * we just need send it once in the first time. For later such
185 * request, we just ignore it.
186 */
187 if (vhost_user_one_time_request(msg->request) && dev->vq_index != 0) {
188 return 0;
189 }
190
191 if (fd_num) {
192 qemu_chr_fe_set_msgfds(chr, fds, fd_num);
193 }
194
195 return qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size) == size ?
196 0 : -1;
197 }
198
199 static int vhost_user_set_log_base(struct vhost_dev *dev, uint64_t base,
200 struct vhost_log *log)
201 {
202 int fds[VHOST_MEMORY_MAX_NREGIONS];
203 size_t fd_num = 0;
204 bool shmfd = virtio_has_feature(dev->protocol_features,
205 VHOST_USER_PROTOCOL_F_LOG_SHMFD);
206 VhostUserMsg msg = {
207 .request = VHOST_USER_SET_LOG_BASE,
208 .flags = VHOST_USER_VERSION,
209 .payload.log.mmap_size = log->size * sizeof(*(log->log)),
210 .payload.log.mmap_offset = 0,
211 .size = sizeof(msg.payload.log),
212 };
213
214 if (shmfd && log->fd != -1) {
215 fds[fd_num++] = log->fd;
216 }
217
218 vhost_user_write(dev, &msg, fds, fd_num);
219
220 if (shmfd) {
221 msg.size = 0;
222 if (vhost_user_read(dev, &msg) < 0) {
223 return 0;
224 }
225
226 if (msg.request != VHOST_USER_SET_LOG_BASE) {
227 error_report("Received unexpected msg type. "
228 "Expected %d received %d",
229 VHOST_USER_SET_LOG_BASE, msg.request);
230 return -1;
231 }
232 }
233
234 return 0;
235 }
236
237 static int vhost_user_set_mem_table(struct vhost_dev *dev,
238 struct vhost_memory *mem)
239 {
240 int fds[VHOST_MEMORY_MAX_NREGIONS];
241 int i, fd;
242 size_t fd_num = 0;
243 VhostUserMsg msg = {
244 .request = VHOST_USER_SET_MEM_TABLE,
245 .flags = VHOST_USER_VERSION,
246 };
247
248 for (i = 0; i < dev->mem->nregions; ++i) {
249 struct vhost_memory_region *reg = dev->mem->regions + i;
250 ram_addr_t ram_addr;
251
252 assert((uintptr_t)reg->userspace_addr == reg->userspace_addr);
253 qemu_ram_addr_from_host((void *)(uintptr_t)reg->userspace_addr,
254 &ram_addr);
255 fd = qemu_get_ram_fd(ram_addr);
256 if (fd > 0) {
257 msg.payload.memory.regions[fd_num].userspace_addr = reg->userspace_addr;
258 msg.payload.memory.regions[fd_num].memory_size = reg->memory_size;
259 msg.payload.memory.regions[fd_num].guest_phys_addr = reg->guest_phys_addr;
260 msg.payload.memory.regions[fd_num].mmap_offset = reg->userspace_addr -
261 (uintptr_t) qemu_get_ram_block_host_ptr(ram_addr);
262 assert(fd_num < VHOST_MEMORY_MAX_NREGIONS);
263 fds[fd_num++] = fd;
264 }
265 }
266
267 msg.payload.memory.nregions = fd_num;
268
269 if (!fd_num) {
270 error_report("Failed initializing vhost-user memory map, "
271 "consider using -object memory-backend-file share=on");
272 return -1;
273 }
274
275 msg.size = sizeof(msg.payload.memory.nregions);
276 msg.size += sizeof(msg.payload.memory.padding);
277 msg.size += fd_num * sizeof(VhostUserMemoryRegion);
278
279 vhost_user_write(dev, &msg, fds, fd_num);
280
281 return 0;
282 }
283
284 static int vhost_user_set_vring_addr(struct vhost_dev *dev,
285 struct vhost_vring_addr *addr)
286 {
287 VhostUserMsg msg = {
288 .request = VHOST_USER_SET_VRING_ADDR,
289 .flags = VHOST_USER_VERSION,
290 .payload.addr = *addr,
291 .size = sizeof(msg.payload.addr),
292 };
293
294 vhost_user_write(dev, &msg, NULL, 0);
295
296 return 0;
297 }
298
299 static int vhost_user_set_vring_endian(struct vhost_dev *dev,
300 struct vhost_vring_state *ring)
301 {
302 error_report("vhost-user trying to send unhandled ioctl");
303 return -1;
304 }
305
306 static int vhost_set_vring(struct vhost_dev *dev,
307 unsigned long int request,
308 struct vhost_vring_state *ring)
309 {
310 VhostUserMsg msg = {
311 .request = request,
312 .flags = VHOST_USER_VERSION,
313 .payload.state = *ring,
314 .size = sizeof(msg.payload.state),
315 };
316
317 vhost_user_write(dev, &msg, NULL, 0);
318
319 return 0;
320 }
321
322 static int vhost_user_set_vring_num(struct vhost_dev *dev,
323 struct vhost_vring_state *ring)
324 {
325 return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring);
326 }
327
328 static int vhost_user_set_vring_base(struct vhost_dev *dev,
329 struct vhost_vring_state *ring)
330 {
331 return vhost_set_vring(dev, VHOST_USER_SET_VRING_BASE, ring);
332 }
333
334 static int vhost_user_set_vring_enable(struct vhost_dev *dev, int enable)
335 {
336 int i;
337
338 if (!virtio_has_feature(dev->features, VHOST_USER_F_PROTOCOL_FEATURES)) {
339 return -1;
340 }
341
342 for (i = 0; i < dev->nvqs; ++i) {
343 struct vhost_vring_state state = {
344 .index = dev->vq_index + i,
345 .num = enable,
346 };
347
348 vhost_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, &state);
349 }
350
351 return 0;
352 }
353
354 static int vhost_user_get_vring_base(struct vhost_dev *dev,
355 struct vhost_vring_state *ring)
356 {
357 VhostUserMsg msg = {
358 .request = VHOST_USER_GET_VRING_BASE,
359 .flags = VHOST_USER_VERSION,
360 .payload.state = *ring,
361 .size = sizeof(msg.payload.state),
362 };
363
364 vhost_user_write(dev, &msg, NULL, 0);
365
366 if (vhost_user_read(dev, &msg) < 0) {
367 return 0;
368 }
369
370 if (msg.request != VHOST_USER_GET_VRING_BASE) {
371 error_report("Received unexpected msg type. Expected %d received %d",
372 VHOST_USER_GET_VRING_BASE, msg.request);
373 return -1;
374 }
375
376 if (msg.size != sizeof(msg.payload.state)) {
377 error_report("Received bad msg size.");
378 return -1;
379 }
380
381 *ring = msg.payload.state;
382
383 return 0;
384 }
385
386 static int vhost_set_vring_file(struct vhost_dev *dev,
387 VhostUserRequest request,
388 struct vhost_vring_file *file)
389 {
390 int fds[VHOST_MEMORY_MAX_NREGIONS];
391 size_t fd_num = 0;
392 VhostUserMsg msg = {
393 .request = request,
394 .flags = VHOST_USER_VERSION,
395 .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
396 .size = sizeof(msg.payload.u64),
397 };
398
399 if (ioeventfd_enabled() && file->fd > 0) {
400 fds[fd_num++] = file->fd;
401 } else {
402 msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
403 }
404
405 vhost_user_write(dev, &msg, fds, fd_num);
406
407 return 0;
408 }
409
410 static int vhost_user_set_vring_kick(struct vhost_dev *dev,
411 struct vhost_vring_file *file)
412 {
413 return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
414 }
415
416 static int vhost_user_set_vring_call(struct vhost_dev *dev,
417 struct vhost_vring_file *file)
418 {
419 return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
420 }
421
422 static int vhost_user_set_u64(struct vhost_dev *dev, int request, uint64_t u64)
423 {
424 VhostUserMsg msg = {
425 .request = request,
426 .flags = VHOST_USER_VERSION,
427 .payload.u64 = u64,
428 .size = sizeof(msg.payload.u64),
429 };
430
431 vhost_user_write(dev, &msg, NULL, 0);
432
433 return 0;
434 }
435
436 static int vhost_user_set_features(struct vhost_dev *dev,
437 uint64_t features)
438 {
439 return vhost_user_set_u64(dev, VHOST_USER_SET_FEATURES, features);
440 }
441
442 static int vhost_user_set_protocol_features(struct vhost_dev *dev,
443 uint64_t features)
444 {
445 return vhost_user_set_u64(dev, VHOST_USER_SET_PROTOCOL_FEATURES, features);
446 }
447
448 static int vhost_user_get_u64(struct vhost_dev *dev, int request, uint64_t *u64)
449 {
450 VhostUserMsg msg = {
451 .request = request,
452 .flags = VHOST_USER_VERSION,
453 };
454
455 if (vhost_user_one_time_request(request) && dev->vq_index != 0) {
456 return 0;
457 }
458
459 vhost_user_write(dev, &msg, NULL, 0);
460
461 if (vhost_user_read(dev, &msg) < 0) {
462 return 0;
463 }
464
465 if (msg.request != request) {
466 error_report("Received unexpected msg type. Expected %d received %d",
467 request, msg.request);
468 return -1;
469 }
470
471 if (msg.size != sizeof(msg.payload.u64)) {
472 error_report("Received bad msg size.");
473 return -1;
474 }
475
476 *u64 = msg.payload.u64;
477
478 return 0;
479 }
480
481 static int vhost_user_get_features(struct vhost_dev *dev, uint64_t *features)
482 {
483 return vhost_user_get_u64(dev, VHOST_USER_GET_FEATURES, features);
484 }
485
486 static int vhost_user_set_owner(struct vhost_dev *dev)
487 {
488 VhostUserMsg msg = {
489 .request = VHOST_USER_SET_OWNER,
490 .flags = VHOST_USER_VERSION,
491 };
492
493 vhost_user_write(dev, &msg, NULL, 0);
494
495 return 0;
496 }
497
498 static int vhost_user_reset_device(struct vhost_dev *dev)
499 {
500 VhostUserMsg msg = {
501 .request = VHOST_USER_RESET_OWNER,
502 .flags = VHOST_USER_VERSION,
503 };
504
505 vhost_user_write(dev, &msg, NULL, 0);
506
507 return 0;
508 }
509
510 static int vhost_user_init(struct vhost_dev *dev, void *opaque)
511 {
512 uint64_t features;
513 int err;
514
515 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
516
517 dev->opaque = opaque;
518
519 err = vhost_user_get_features(dev, &features);
520 if (err < 0) {
521 return err;
522 }
523
524 if (virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
525 dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
526
527 err = vhost_user_get_u64(dev, VHOST_USER_GET_PROTOCOL_FEATURES,
528 &features);
529 if (err < 0) {
530 return err;
531 }
532
533 dev->protocol_features = features & VHOST_USER_PROTOCOL_FEATURE_MASK;
534 err = vhost_user_set_protocol_features(dev, dev->protocol_features);
535 if (err < 0) {
536 return err;
537 }
538
539 /* query the max queues we support if backend supports Multiple Queue */
540 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
541 err = vhost_user_get_u64(dev, VHOST_USER_GET_QUEUE_NUM,
542 &dev->max_queues);
543 if (err < 0) {
544 return err;
545 }
546 }
547 }
548
549 if (dev->migration_blocker == NULL &&
550 !virtio_has_feature(dev->protocol_features,
551 VHOST_USER_PROTOCOL_F_LOG_SHMFD)) {
552 error_setg(&dev->migration_blocker,
553 "Migration disabled: vhost-user backend lacks "
554 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
555 }
556
557 return 0;
558 }
559
560 static int vhost_user_cleanup(struct vhost_dev *dev)
561 {
562 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
563
564 dev->opaque = 0;
565
566 return 0;
567 }
568
569 static int vhost_user_get_vq_index(struct vhost_dev *dev, int idx)
570 {
571 assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
572
573 return idx;
574 }
575
576 static int vhost_user_memslots_limit(struct vhost_dev *dev)
577 {
578 return VHOST_MEMORY_MAX_NREGIONS;
579 }
580
581 static bool vhost_user_requires_shm_log(struct vhost_dev *dev)
582 {
583 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
584
585 return virtio_has_feature(dev->protocol_features,
586 VHOST_USER_PROTOCOL_F_LOG_SHMFD);
587 }
588
589 static int vhost_user_migration_done(struct vhost_dev *dev, char* mac_addr)
590 {
591 VhostUserMsg msg = { 0 };
592 int err;
593
594 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
595
596 /* If guest supports GUEST_ANNOUNCE do nothing */
597 if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) {
598 return 0;
599 }
600
601 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
602 if (virtio_has_feature(dev->protocol_features,
603 VHOST_USER_PROTOCOL_F_RARP)) {
604 msg.request = VHOST_USER_SEND_RARP;
605 msg.flags = VHOST_USER_VERSION;
606 memcpy((char *)&msg.payload.u64, mac_addr, 6);
607 msg.size = sizeof(msg.payload.u64);
608
609 err = vhost_user_write(dev, &msg, NULL, 0);
610 return err;
611 }
612 return -1;
613 }
614
615 static bool vhost_user_can_merge(struct vhost_dev *dev,
616 uint64_t start1, uint64_t size1,
617 uint64_t start2, uint64_t size2)
618 {
619 ram_addr_t ram_addr;
620 int mfd, rfd;
621 MemoryRegion *mr;
622
623 mr = qemu_ram_addr_from_host((void *)(uintptr_t)start1, &ram_addr);
624 assert(mr);
625 mfd = qemu_get_ram_fd(ram_addr);
626
627 mr = qemu_ram_addr_from_host((void *)(uintptr_t)start2, &ram_addr);
628 assert(mr);
629 rfd = qemu_get_ram_fd(ram_addr);
630
631 return mfd == rfd;
632 }
633
634 const VhostOps user_ops = {
635 .backend_type = VHOST_BACKEND_TYPE_USER,
636 .vhost_backend_init = vhost_user_init,
637 .vhost_backend_cleanup = vhost_user_cleanup,
638 .vhost_backend_memslots_limit = vhost_user_memslots_limit,
639 .vhost_set_log_base = vhost_user_set_log_base,
640 .vhost_set_mem_table = vhost_user_set_mem_table,
641 .vhost_set_vring_addr = vhost_user_set_vring_addr,
642 .vhost_set_vring_endian = vhost_user_set_vring_endian,
643 .vhost_set_vring_num = vhost_user_set_vring_num,
644 .vhost_set_vring_base = vhost_user_set_vring_base,
645 .vhost_get_vring_base = vhost_user_get_vring_base,
646 .vhost_set_vring_kick = vhost_user_set_vring_kick,
647 .vhost_set_vring_call = vhost_user_set_vring_call,
648 .vhost_set_features = vhost_user_set_features,
649 .vhost_get_features = vhost_user_get_features,
650 .vhost_set_owner = vhost_user_set_owner,
651 .vhost_reset_device = vhost_user_reset_device,
652 .vhost_get_vq_index = vhost_user_get_vq_index,
653 .vhost_set_vring_enable = vhost_user_set_vring_enable,
654 .vhost_requires_shm_log = vhost_user_requires_shm_log,
655 .vhost_migration_done = vhost_user_migration_done,
656 .vhost_backend_can_merge = vhost_user_can_merge,
657 };