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/virtio-dmabuf.h"
14 #include "hw/virtio/vhost.h"
15 #include "hw/virtio/virtio-crypto.h"
16 #include "hw/virtio/vhost-user.h"
17 #include "hw/virtio/vhost-backend.h"
18 #include "hw/virtio/virtio.h"
19 #include "hw/virtio/virtio-net.h"
20 #include "chardev/char-fe.h"
21 #include "io/channel-socket.h"
22 #include "system/kvm.h"
23 #include "qemu/error-report.h"
24 #include "qemu/main-loop.h"
25 #include "qemu/uuid.h"
26 #include "qemu/sockets.h"
27 #include "system/runstate.h"
28 #include "system/cryptodev.h"
29 #include "migration/postcopy-ram.h"
30 #include "trace.h"
31 #include "system/ramblock.h"
32
33 #include <sys/ioctl.h>
34 #include <sys/socket.h>
35 #include <sys/un.h>
36
37 #include "standard-headers/linux/vhost_types.h"
38
39 #ifdef CONFIG_LINUX
40 #include <linux/userfaultfd.h>
41 #endif
42
43 #define VHOST_MEMORY_BASELINE_NREGIONS 8
44 #define VHOST_USER_F_PROTOCOL_FEATURES 30
45 #define VHOST_USER_BACKEND_MAX_FDS 8
46
47 #if defined(TARGET_PPC) || defined(TARGET_PPC64)
48 #include "hw/ppc/spapr.h"
49 #define VHOST_USER_MAX_RAM_SLOTS SPAPR_MAX_RAM_SLOTS
50
51 #else
52 #define VHOST_USER_MAX_RAM_SLOTS 512
53 #endif
54
55 /*
56 * Maximum size of virtio device config space
57 */
58 #define VHOST_USER_MAX_CONFIG_SIZE 256
59
60 #define VHOST_USER_PROTOCOL_FEATURE_MASK ((1 << VHOST_USER_PROTOCOL_F_MAX) - 1)
61
62 typedef enum VhostUserRequest {
63 VHOST_USER_NONE = 0,
64 VHOST_USER_GET_FEATURES = 1,
65 VHOST_USER_SET_FEATURES = 2,
66 VHOST_USER_SET_OWNER = 3,
67 VHOST_USER_RESET_OWNER = 4,
68 VHOST_USER_SET_MEM_TABLE = 5,
69 VHOST_USER_SET_LOG_BASE = 6,
70 VHOST_USER_SET_LOG_FD = 7,
71 VHOST_USER_SET_VRING_NUM = 8,
72 VHOST_USER_SET_VRING_ADDR = 9,
73 VHOST_USER_SET_VRING_BASE = 10,
74 VHOST_USER_GET_VRING_BASE = 11,
75 VHOST_USER_SET_VRING_KICK = 12,
76 VHOST_USER_SET_VRING_CALL = 13,
77 VHOST_USER_SET_VRING_ERR = 14,
78 VHOST_USER_GET_PROTOCOL_FEATURES = 15,
79 VHOST_USER_SET_PROTOCOL_FEATURES = 16,
80 VHOST_USER_GET_QUEUE_NUM = 17,
81 VHOST_USER_SET_VRING_ENABLE = 18,
82 VHOST_USER_SEND_RARP = 19,
83 VHOST_USER_NET_SET_MTU = 20,
84 VHOST_USER_SET_BACKEND_REQ_FD = 21,
85 VHOST_USER_IOTLB_MSG = 22,
86 VHOST_USER_SET_VRING_ENDIAN = 23,
87 VHOST_USER_GET_CONFIG = 24,
88 VHOST_USER_SET_CONFIG = 25,
89 VHOST_USER_CREATE_CRYPTO_SESSION = 26,
90 VHOST_USER_CLOSE_CRYPTO_SESSION = 27,
91 VHOST_USER_POSTCOPY_ADVISE = 28,
92 VHOST_USER_POSTCOPY_LISTEN = 29,
93 VHOST_USER_POSTCOPY_END = 30,
94 VHOST_USER_GET_INFLIGHT_FD = 31,
95 VHOST_USER_SET_INFLIGHT_FD = 32,
96 VHOST_USER_GPU_SET_SOCKET = 33,
97 VHOST_USER_RESET_DEVICE = 34,
98 /* Message number 35 reserved for VHOST_USER_VRING_KICK. */
99 VHOST_USER_GET_MAX_MEM_SLOTS = 36,
100 VHOST_USER_ADD_MEM_REG = 37,
101 VHOST_USER_REM_MEM_REG = 38,
102 VHOST_USER_SET_STATUS = 39,
103 VHOST_USER_GET_STATUS = 40,
104 VHOST_USER_GET_SHARED_OBJECT = 41,
105 VHOST_USER_SET_DEVICE_STATE_FD = 42,
106 VHOST_USER_CHECK_DEVICE_STATE = 43,
107 VHOST_USER_MAX
108 } VhostUserRequest;
109
110 typedef enum VhostUserBackendRequest {
111 VHOST_USER_BACKEND_NONE = 0,
112 VHOST_USER_BACKEND_IOTLB_MSG = 1,
113 VHOST_USER_BACKEND_CONFIG_CHANGE_MSG = 2,
114 VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG = 3,
115 VHOST_USER_BACKEND_SHARED_OBJECT_ADD = 6,
116 VHOST_USER_BACKEND_SHARED_OBJECT_REMOVE = 7,
117 VHOST_USER_BACKEND_SHARED_OBJECT_LOOKUP = 8,
118 VHOST_USER_BACKEND_MAX
119 } VhostUserBackendRequest;
120
121 typedef struct VhostUserMemoryRegion {
122 uint64_t guest_phys_addr;
123 uint64_t memory_size;
124 uint64_t userspace_addr;
125 uint64_t mmap_offset;
126 } VhostUserMemoryRegion;
127
128 typedef struct VhostUserMemory {
129 uint32_t nregions;
130 uint32_t padding;
131 VhostUserMemoryRegion regions[VHOST_MEMORY_BASELINE_NREGIONS];
132 } VhostUserMemory;
133
134 typedef struct VhostUserMemRegMsg {
135 uint64_t padding;
136 VhostUserMemoryRegion region;
137 } VhostUserMemRegMsg;
138
139 typedef struct VhostUserLog {
140 uint64_t mmap_size;
141 uint64_t mmap_offset;
142 } VhostUserLog;
143
144 typedef struct VhostUserConfig {
145 uint32_t offset;
146 uint32_t size;
147 uint32_t flags;
148 uint8_t region[VHOST_USER_MAX_CONFIG_SIZE];
149 } VhostUserConfig;
150
151 #define VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN 512
152 #define VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN 64
153 #define VHOST_CRYPTO_ASYM_MAX_KEY_LEN 1024
154
155 typedef struct VhostUserCryptoSession {
156 uint64_t op_code;
157 union {
158 struct {
159 CryptoDevBackendSymSessionInfo session_setup_data;
160 uint8_t key[VHOST_CRYPTO_SYM_CIPHER_MAX_KEY_LEN];
161 uint8_t auth_key[VHOST_CRYPTO_SYM_HMAC_MAX_KEY_LEN];
162 } sym;
163 struct {
164 CryptoDevBackendAsymSessionInfo session_setup_data;
165 uint8_t key[VHOST_CRYPTO_ASYM_MAX_KEY_LEN];
166 } asym;
167 } u;
168
169 /* session id for success, -1 on errors */
170 int64_t session_id;
171 } VhostUserCryptoSession;
172
173 static VhostUserConfig c __attribute__ ((unused));
174 #define VHOST_USER_CONFIG_HDR_SIZE (sizeof(c.offset) \
175 + sizeof(c.size) \
176 + sizeof(c.flags))
177
178 typedef struct VhostUserVringArea {
179 uint64_t u64;
180 uint64_t size;
181 uint64_t offset;
182 } VhostUserVringArea;
183
184 typedef struct VhostUserInflight {
185 uint64_t mmap_size;
186 uint64_t mmap_offset;
187 uint16_t num_queues;
188 uint16_t queue_size;
189 } VhostUserInflight;
190
191 typedef struct VhostUserShared {
192 unsigned char uuid[16];
193 } VhostUserShared;
194
195 typedef struct {
196 VhostUserRequest request;
197
198 #define VHOST_USER_VERSION_MASK (0x3)
199 #define VHOST_USER_REPLY_MASK (0x1 << 2)
200 #define VHOST_USER_NEED_REPLY_MASK (0x1 << 3)
201 uint32_t flags;
202 uint32_t size; /* the following payload size */
203 } QEMU_PACKED VhostUserHeader;
204
205 /* Request payload of VHOST_USER_SET_DEVICE_STATE_FD */
206 typedef struct VhostUserTransferDeviceState {
207 uint32_t direction;
208 uint32_t phase;
209 } VhostUserTransferDeviceState;
210
211 typedef union {
212 #define VHOST_USER_VRING_IDX_MASK (0xff)
213 #define VHOST_USER_VRING_NOFD_MASK (0x1 << 8)
214 uint64_t u64;
215 struct vhost_vring_state state;
216 struct vhost_vring_addr addr;
217 VhostUserMemory memory;
218 VhostUserMemRegMsg mem_reg;
219 VhostUserLog log;
220 struct vhost_iotlb_msg iotlb;
221 VhostUserConfig config;
222 VhostUserCryptoSession session;
223 VhostUserVringArea area;
224 VhostUserInflight inflight;
225 VhostUserShared object;
226 VhostUserTransferDeviceState transfer_state;
227 } VhostUserPayload;
228
229 typedef struct VhostUserMsg {
230 VhostUserHeader hdr;
231 VhostUserPayload payload;
232 } QEMU_PACKED VhostUserMsg;
233
234 static VhostUserMsg m __attribute__ ((unused));
235 #define VHOST_USER_HDR_SIZE (sizeof(VhostUserHeader))
236
237 #define VHOST_USER_PAYLOAD_SIZE (sizeof(VhostUserPayload))
238
239 /* The version of the protocol we support */
240 #define VHOST_USER_VERSION (0x1)
241
242 struct vhost_user {
243 struct vhost_dev *dev;
244 /* Shared between vhost devs of the same virtio device */
245 VhostUserState *user;
246 QIOChannel *backend_ioc;
247 GSource *backend_src;
248 NotifierWithReturn postcopy_notifier;
249 struct PostCopyFD postcopy_fd;
250 uint64_t postcopy_client_bases[VHOST_USER_MAX_RAM_SLOTS];
251 /* Length of the region_rb and region_rb_offset arrays */
252 size_t region_rb_len;
253 /* RAMBlock associated with a given region */
254 RAMBlock **region_rb;
255 /*
256 * The offset from the start of the RAMBlock to the start of the
257 * vhost region.
258 */
259 ram_addr_t *region_rb_offset;
260
261 /* True once we've entered postcopy_listen */
262 bool postcopy_listen;
263
264 /* Our current regions */
265 int num_shadow_regions;
266 struct vhost_memory_region shadow_regions[VHOST_USER_MAX_RAM_SLOTS];
267 };
268
269 struct scrub_regions {
270 struct vhost_memory_region *region;
271 int reg_idx;
272 int fd_idx;
273 };
274
vhost_user_read_header(struct vhost_dev * dev,VhostUserMsg * msg)275 static int vhost_user_read_header(struct vhost_dev *dev, VhostUserMsg *msg)
276 {
277 struct vhost_user *u = dev->opaque;
278 CharBackend *chr = u->user->chr;
279 uint8_t *p = (uint8_t *) msg;
280 int r, size = VHOST_USER_HDR_SIZE;
281
282 r = qemu_chr_fe_read_all(chr, p, size);
283 if (r != size) {
284 int saved_errno = errno;
285 error_report("Failed to read msg header. Read %d instead of %d."
286 " Original request %d.", r, size, msg->hdr.request);
287 return r < 0 ? -saved_errno : -EIO;
288 }
289
290 /* validate received flags */
291 if (msg->hdr.flags != (VHOST_USER_REPLY_MASK | VHOST_USER_VERSION)) {
292 error_report("Failed to read msg header."
293 " Flags 0x%x instead of 0x%x.", msg->hdr.flags,
294 VHOST_USER_REPLY_MASK | VHOST_USER_VERSION);
295 return -EPROTO;
296 }
297
298 trace_vhost_user_read(msg->hdr.request, msg->hdr.flags);
299
300 return 0;
301 }
302
vhost_user_read(struct vhost_dev * dev,VhostUserMsg * msg)303 static int vhost_user_read(struct vhost_dev *dev, VhostUserMsg *msg)
304 {
305 struct vhost_user *u = dev->opaque;
306 CharBackend *chr = u->user->chr;
307 uint8_t *p = (uint8_t *) msg;
308 int r, size;
309
310 r = vhost_user_read_header(dev, msg);
311 if (r < 0) {
312 return r;
313 }
314
315 /* validate message size is sane */
316 if (msg->hdr.size > VHOST_USER_PAYLOAD_SIZE) {
317 error_report("Failed to read msg header."
318 " Size %d exceeds the maximum %zu.", msg->hdr.size,
319 VHOST_USER_PAYLOAD_SIZE);
320 return -EPROTO;
321 }
322
323 if (msg->hdr.size) {
324 p += VHOST_USER_HDR_SIZE;
325 size = msg->hdr.size;
326 r = qemu_chr_fe_read_all(chr, p, size);
327 if (r != size) {
328 int saved_errno = errno;
329 error_report("Failed to read msg payload."
330 " Read %d instead of %d.", r, msg->hdr.size);
331 return r < 0 ? -saved_errno : -EIO;
332 }
333 }
334
335 return 0;
336 }
337
process_message_reply(struct vhost_dev * dev,const VhostUserMsg * msg)338 static int process_message_reply(struct vhost_dev *dev,
339 const VhostUserMsg *msg)
340 {
341 int ret;
342 VhostUserMsg msg_reply;
343
344 if ((msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) == 0) {
345 return 0;
346 }
347
348 ret = vhost_user_read(dev, &msg_reply);
349 if (ret < 0) {
350 return ret;
351 }
352
353 if (msg_reply.hdr.request != msg->hdr.request) {
354 error_report("Received unexpected msg type. "
355 "Expected %d received %d",
356 msg->hdr.request, msg_reply.hdr.request);
357 return -EPROTO;
358 }
359
360 return msg_reply.payload.u64 ? -EIO : 0;
361 }
362
vhost_user_per_device_request(VhostUserRequest request)363 static bool vhost_user_per_device_request(VhostUserRequest request)
364 {
365 switch (request) {
366 case VHOST_USER_SET_OWNER:
367 case VHOST_USER_RESET_OWNER:
368 case VHOST_USER_SET_MEM_TABLE:
369 case VHOST_USER_GET_QUEUE_NUM:
370 case VHOST_USER_NET_SET_MTU:
371 case VHOST_USER_RESET_DEVICE:
372 case VHOST_USER_ADD_MEM_REG:
373 case VHOST_USER_REM_MEM_REG:
374 case VHOST_USER_SET_LOG_BASE:
375 return true;
376 default:
377 return false;
378 }
379 }
380
381 /* most non-init callers ignore the error */
vhost_user_write(struct vhost_dev * dev,VhostUserMsg * msg,int * fds,int fd_num)382 static int vhost_user_write(struct vhost_dev *dev, VhostUserMsg *msg,
383 int *fds, int fd_num)
384 {
385 struct vhost_user *u = dev->opaque;
386 CharBackend *chr = u->user->chr;
387 int ret, size = VHOST_USER_HDR_SIZE + msg->hdr.size;
388
389 /*
390 * Some devices, like virtio-scsi, are implemented as a single vhost_dev,
391 * while others, like virtio-net, contain multiple vhost_devs. For
392 * operations such as configuring device memory mappings or issuing device
393 * resets, which affect the whole device instead of individual VQs,
394 * vhost-user messages should only be sent once.
395 *
396 * Devices with multiple vhost_devs are given an associated dev->vq_index
397 * so per_device requests are only sent if vq_index is 0.
398 */
399 if (vhost_user_per_device_request(msg->hdr.request)
400 && dev->vq_index != 0) {
401 msg->hdr.flags &= ~VHOST_USER_NEED_REPLY_MASK;
402 return 0;
403 }
404
405 if (qemu_chr_fe_set_msgfds(chr, fds, fd_num) < 0) {
406 error_report("Failed to set msg fds.");
407 return -EINVAL;
408 }
409
410 ret = qemu_chr_fe_write_all(chr, (const uint8_t *) msg, size);
411 if (ret != size) {
412 int saved_errno = errno;
413 error_report("Failed to write msg."
414 " Wrote %d instead of %d.", ret, size);
415 return ret < 0 ? -saved_errno : -EIO;
416 }
417
418 trace_vhost_user_write(msg->hdr.request, msg->hdr.flags);
419
420 return 0;
421 }
422
vhost_user_gpu_set_socket(struct vhost_dev * dev,int fd)423 int vhost_user_gpu_set_socket(struct vhost_dev *dev, int fd)
424 {
425 VhostUserMsg msg = {
426 .hdr.request = VHOST_USER_GPU_SET_SOCKET,
427 .hdr.flags = VHOST_USER_VERSION,
428 };
429
430 return vhost_user_write(dev, &msg, &fd, 1);
431 }
432
vhost_user_set_log_base(struct vhost_dev * dev,uint64_t base,struct vhost_log * log)433 static int vhost_user_set_log_base(struct vhost_dev *dev, uint64_t base,
434 struct vhost_log *log)
435 {
436 int fds[VHOST_USER_MAX_RAM_SLOTS];
437 size_t fd_num = 0;
438 bool shmfd = virtio_has_feature(dev->protocol_features,
439 VHOST_USER_PROTOCOL_F_LOG_SHMFD);
440 int ret;
441 VhostUserMsg msg = {
442 .hdr.request = VHOST_USER_SET_LOG_BASE,
443 .hdr.flags = VHOST_USER_VERSION,
444 .payload.log.mmap_size = log->size * sizeof(*(log->log)),
445 .payload.log.mmap_offset = 0,
446 .hdr.size = sizeof(msg.payload.log),
447 };
448
449 /* Send only once with first queue pair */
450 if (dev->vq_index != 0) {
451 return 0;
452 }
453
454 if (shmfd && log->fd != -1) {
455 fds[fd_num++] = log->fd;
456 }
457
458 ret = vhost_user_write(dev, &msg, fds, fd_num);
459 if (ret < 0) {
460 return ret;
461 }
462
463 if (shmfd) {
464 msg.hdr.size = 0;
465 ret = vhost_user_read(dev, &msg);
466 if (ret < 0) {
467 return ret;
468 }
469
470 if (msg.hdr.request != VHOST_USER_SET_LOG_BASE) {
471 error_report("Received unexpected msg type. "
472 "Expected %d received %d",
473 VHOST_USER_SET_LOG_BASE, msg.hdr.request);
474 return -EPROTO;
475 }
476 }
477
478 return 0;
479 }
480
vhost_user_get_mr_data(uint64_t addr,ram_addr_t * offset,int * fd)481 static MemoryRegion *vhost_user_get_mr_data(uint64_t addr, ram_addr_t *offset,
482 int *fd)
483 {
484 MemoryRegion *mr;
485
486 assert((uintptr_t)addr == addr);
487 mr = memory_region_from_host((void *)(uintptr_t)addr, offset);
488 *fd = memory_region_get_fd(mr);
489 *offset += mr->ram_block->fd_offset;
490
491 return mr;
492 }
493
vhost_user_fill_msg_region(VhostUserMemoryRegion * dst,struct vhost_memory_region * src,uint64_t mmap_offset)494 static void vhost_user_fill_msg_region(VhostUserMemoryRegion *dst,
495 struct vhost_memory_region *src,
496 uint64_t mmap_offset)
497 {
498 assert(src != NULL && dst != NULL);
499 dst->userspace_addr = src->userspace_addr;
500 dst->memory_size = src->memory_size;
501 dst->guest_phys_addr = src->guest_phys_addr;
502 dst->mmap_offset = mmap_offset;
503 }
504
vhost_user_fill_set_mem_table_msg(struct vhost_user * u,struct vhost_dev * dev,VhostUserMsg * msg,int * fds,size_t * fd_num,bool track_ramblocks)505 static int vhost_user_fill_set_mem_table_msg(struct vhost_user *u,
506 struct vhost_dev *dev,
507 VhostUserMsg *msg,
508 int *fds, size_t *fd_num,
509 bool track_ramblocks)
510 {
511 int i, fd;
512 ram_addr_t offset;
513 MemoryRegion *mr;
514 struct vhost_memory_region *reg;
515 VhostUserMemoryRegion region_buffer;
516
517 msg->hdr.request = VHOST_USER_SET_MEM_TABLE;
518
519 for (i = 0; i < dev->mem->nregions; ++i) {
520 reg = dev->mem->regions + i;
521
522 mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
523 if (fd > 0) {
524 if (track_ramblocks) {
525 assert(*fd_num < VHOST_MEMORY_BASELINE_NREGIONS);
526 trace_vhost_user_set_mem_table_withfd(*fd_num, mr->name,
527 reg->memory_size,
528 reg->guest_phys_addr,
529 reg->userspace_addr,
530 offset);
531 u->region_rb_offset[i] = offset;
532 u->region_rb[i] = mr->ram_block;
533 } else if (*fd_num == VHOST_MEMORY_BASELINE_NREGIONS) {
534 error_report("Failed preparing vhost-user memory table msg");
535 return -ENOBUFS;
536 }
537 vhost_user_fill_msg_region(®ion_buffer, reg, offset);
538 msg->payload.memory.regions[*fd_num] = region_buffer;
539 fds[(*fd_num)++] = fd;
540 } else if (track_ramblocks) {
541 u->region_rb_offset[i] = 0;
542 u->region_rb[i] = NULL;
543 }
544 }
545
546 msg->payload.memory.nregions = *fd_num;
547
548 if (!*fd_num) {
549 error_report("Failed initializing vhost-user memory map, "
550 "consider using -object memory-backend-file share=on");
551 return -EINVAL;
552 }
553
554 msg->hdr.size = sizeof(msg->payload.memory.nregions);
555 msg->hdr.size += sizeof(msg->payload.memory.padding);
556 msg->hdr.size += *fd_num * sizeof(VhostUserMemoryRegion);
557
558 return 0;
559 }
560
reg_equal(struct vhost_memory_region * shadow_reg,struct vhost_memory_region * vdev_reg)561 static inline bool reg_equal(struct vhost_memory_region *shadow_reg,
562 struct vhost_memory_region *vdev_reg)
563 {
564 return shadow_reg->guest_phys_addr == vdev_reg->guest_phys_addr &&
565 shadow_reg->userspace_addr == vdev_reg->userspace_addr &&
566 shadow_reg->memory_size == vdev_reg->memory_size;
567 }
568
scrub_shadow_regions(struct vhost_dev * dev,struct scrub_regions * add_reg,int * nr_add_reg,struct scrub_regions * rem_reg,int * nr_rem_reg,uint64_t * shadow_pcb,bool track_ramblocks)569 static void scrub_shadow_regions(struct vhost_dev *dev,
570 struct scrub_regions *add_reg,
571 int *nr_add_reg,
572 struct scrub_regions *rem_reg,
573 int *nr_rem_reg, uint64_t *shadow_pcb,
574 bool track_ramblocks)
575 {
576 struct vhost_user *u = dev->opaque;
577 bool found[VHOST_USER_MAX_RAM_SLOTS] = {};
578 struct vhost_memory_region *reg, *shadow_reg;
579 int i, j, fd, add_idx = 0, rm_idx = 0, fd_num = 0;
580 ram_addr_t offset;
581 MemoryRegion *mr;
582 bool matching;
583
584 /*
585 * Find memory regions present in our shadow state which are not in
586 * the device's current memory state.
587 *
588 * Mark regions in both the shadow and device state as "found".
589 */
590 for (i = 0; i < u->num_shadow_regions; i++) {
591 shadow_reg = &u->shadow_regions[i];
592 matching = false;
593
594 for (j = 0; j < dev->mem->nregions; j++) {
595 reg = &dev->mem->regions[j];
596
597 mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
598
599 if (reg_equal(shadow_reg, reg)) {
600 matching = true;
601 found[j] = true;
602 if (track_ramblocks) {
603 /*
604 * Reset postcopy client bases, region_rb, and
605 * region_rb_offset in case regions are removed.
606 */
607 if (fd > 0) {
608 u->region_rb_offset[j] = offset;
609 u->region_rb[j] = mr->ram_block;
610 shadow_pcb[j] = u->postcopy_client_bases[i];
611 } else {
612 u->region_rb_offset[j] = 0;
613 u->region_rb[j] = NULL;
614 }
615 }
616 break;
617 }
618 }
619
620 /*
621 * If the region was not found in the current device memory state
622 * create an entry for it in the removed list.
623 */
624 if (!matching) {
625 rem_reg[rm_idx].region = shadow_reg;
626 rem_reg[rm_idx++].reg_idx = i;
627 }
628 }
629
630 /*
631 * For regions not marked "found", create entries in the added list.
632 *
633 * Note their indexes in the device memory state and the indexes of their
634 * file descriptors.
635 */
636 for (i = 0; i < dev->mem->nregions; i++) {
637 reg = &dev->mem->regions[i];
638 vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
639 if (fd > 0) {
640 ++fd_num;
641 }
642
643 /*
644 * If the region was in both the shadow and device state we don't
645 * need to send a VHOST_USER_ADD_MEM_REG message for it.
646 */
647 if (found[i]) {
648 continue;
649 }
650
651 add_reg[add_idx].region = reg;
652 add_reg[add_idx].reg_idx = i;
653 add_reg[add_idx++].fd_idx = fd_num;
654 }
655 *nr_rem_reg = rm_idx;
656 *nr_add_reg = add_idx;
657 }
658
send_remove_regions(struct vhost_dev * dev,struct scrub_regions * remove_reg,int nr_rem_reg,VhostUserMsg * msg,bool reply_supported)659 static int send_remove_regions(struct vhost_dev *dev,
660 struct scrub_regions *remove_reg,
661 int nr_rem_reg, VhostUserMsg *msg,
662 bool reply_supported)
663 {
664 struct vhost_user *u = dev->opaque;
665 struct vhost_memory_region *shadow_reg;
666 int i, fd, shadow_reg_idx, ret;
667 ram_addr_t offset;
668 VhostUserMemoryRegion region_buffer;
669
670 /*
671 * The regions in remove_reg appear in the same order they do in the
672 * shadow table. Therefore we can minimize memory copies by iterating
673 * through remove_reg backwards.
674 */
675 for (i = nr_rem_reg - 1; i >= 0; i--) {
676 shadow_reg = remove_reg[i].region;
677 shadow_reg_idx = remove_reg[i].reg_idx;
678
679 vhost_user_get_mr_data(shadow_reg->userspace_addr, &offset, &fd);
680
681 if (fd > 0) {
682 msg->hdr.request = VHOST_USER_REM_MEM_REG;
683 vhost_user_fill_msg_region(®ion_buffer, shadow_reg, 0);
684 msg->payload.mem_reg.region = region_buffer;
685
686 ret = vhost_user_write(dev, msg, NULL, 0);
687 if (ret < 0) {
688 return ret;
689 }
690
691 if (reply_supported) {
692 ret = process_message_reply(dev, msg);
693 if (ret) {
694 return ret;
695 }
696 }
697 }
698
699 /*
700 * At this point we know the backend has unmapped the region. It is now
701 * safe to remove it from the shadow table.
702 */
703 memmove(&u->shadow_regions[shadow_reg_idx],
704 &u->shadow_regions[shadow_reg_idx + 1],
705 sizeof(struct vhost_memory_region) *
706 (u->num_shadow_regions - shadow_reg_idx - 1));
707 u->num_shadow_regions--;
708 }
709
710 return 0;
711 }
712
send_add_regions(struct vhost_dev * dev,struct scrub_regions * add_reg,int nr_add_reg,VhostUserMsg * msg,uint64_t * shadow_pcb,bool reply_supported,bool track_ramblocks)713 static int send_add_regions(struct vhost_dev *dev,
714 struct scrub_regions *add_reg, int nr_add_reg,
715 VhostUserMsg *msg, uint64_t *shadow_pcb,
716 bool reply_supported, bool track_ramblocks)
717 {
718 struct vhost_user *u = dev->opaque;
719 int i, fd, ret, reg_idx, reg_fd_idx;
720 struct vhost_memory_region *reg;
721 MemoryRegion *mr;
722 ram_addr_t offset;
723 VhostUserMsg msg_reply;
724 VhostUserMemoryRegion region_buffer;
725
726 for (i = 0; i < nr_add_reg; i++) {
727 reg = add_reg[i].region;
728 reg_idx = add_reg[i].reg_idx;
729 reg_fd_idx = add_reg[i].fd_idx;
730
731 mr = vhost_user_get_mr_data(reg->userspace_addr, &offset, &fd);
732
733 if (fd > 0) {
734 if (track_ramblocks) {
735 trace_vhost_user_set_mem_table_withfd(reg_fd_idx, mr->name,
736 reg->memory_size,
737 reg->guest_phys_addr,
738 reg->userspace_addr,
739 offset);
740 u->region_rb_offset[reg_idx] = offset;
741 u->region_rb[reg_idx] = mr->ram_block;
742 }
743 msg->hdr.request = VHOST_USER_ADD_MEM_REG;
744 vhost_user_fill_msg_region(®ion_buffer, reg, offset);
745 msg->payload.mem_reg.region = region_buffer;
746
747 ret = vhost_user_write(dev, msg, &fd, 1);
748 if (ret < 0) {
749 return ret;
750 }
751
752 if (track_ramblocks) {
753 uint64_t reply_gpa;
754
755 ret = vhost_user_read(dev, &msg_reply);
756 if (ret < 0) {
757 return ret;
758 }
759
760 reply_gpa = msg_reply.payload.mem_reg.region.guest_phys_addr;
761
762 if (msg_reply.hdr.request != VHOST_USER_ADD_MEM_REG) {
763 error_report("%s: Received unexpected msg type."
764 "Expected %d received %d", __func__,
765 VHOST_USER_ADD_MEM_REG,
766 msg_reply.hdr.request);
767 return -EPROTO;
768 }
769
770 /*
771 * We're using the same structure, just reusing one of the
772 * fields, so it should be the same size.
773 */
774 if (msg_reply.hdr.size != msg->hdr.size) {
775 error_report("%s: Unexpected size for postcopy reply "
776 "%d vs %d", __func__, msg_reply.hdr.size,
777 msg->hdr.size);
778 return -EPROTO;
779 }
780
781 /* Get the postcopy client base from the backend's reply. */
782 if (reply_gpa == dev->mem->regions[reg_idx].guest_phys_addr) {
783 shadow_pcb[reg_idx] =
784 msg_reply.payload.mem_reg.region.userspace_addr;
785 trace_vhost_user_set_mem_table_postcopy(
786 msg_reply.payload.mem_reg.region.userspace_addr,
787 msg->payload.mem_reg.region.userspace_addr,
788 reg_fd_idx, reg_idx);
789 } else {
790 error_report("%s: invalid postcopy reply for region. "
791 "Got guest physical address %" PRIX64 ", expected "
792 "%" PRIX64, __func__, reply_gpa,
793 dev->mem->regions[reg_idx].guest_phys_addr);
794 return -EPROTO;
795 }
796 } else if (reply_supported) {
797 ret = process_message_reply(dev, msg);
798 if (ret) {
799 return ret;
800 }
801 }
802 } else if (track_ramblocks) {
803 u->region_rb_offset[reg_idx] = 0;
804 u->region_rb[reg_idx] = NULL;
805 }
806
807 /*
808 * At this point, we know the backend has mapped in the new
809 * region, if the region has a valid file descriptor.
810 *
811 * The region should now be added to the shadow table.
812 */
813 u->shadow_regions[u->num_shadow_regions].guest_phys_addr =
814 reg->guest_phys_addr;
815 u->shadow_regions[u->num_shadow_regions].userspace_addr =
816 reg->userspace_addr;
817 u->shadow_regions[u->num_shadow_regions].memory_size =
818 reg->memory_size;
819 u->num_shadow_regions++;
820 }
821
822 return 0;
823 }
824
vhost_user_add_remove_regions(struct vhost_dev * dev,VhostUserMsg * msg,bool reply_supported,bool track_ramblocks)825 static int vhost_user_add_remove_regions(struct vhost_dev *dev,
826 VhostUserMsg *msg,
827 bool reply_supported,
828 bool track_ramblocks)
829 {
830 struct vhost_user *u = dev->opaque;
831 struct scrub_regions add_reg[VHOST_USER_MAX_RAM_SLOTS];
832 struct scrub_regions rem_reg[VHOST_USER_MAX_RAM_SLOTS];
833 uint64_t shadow_pcb[VHOST_USER_MAX_RAM_SLOTS] = {};
834 int nr_add_reg, nr_rem_reg;
835 int ret;
836
837 msg->hdr.size = sizeof(msg->payload.mem_reg);
838
839 /* Find the regions which need to be removed or added. */
840 scrub_shadow_regions(dev, add_reg, &nr_add_reg, rem_reg, &nr_rem_reg,
841 shadow_pcb, track_ramblocks);
842
843 if (nr_rem_reg) {
844 ret = send_remove_regions(dev, rem_reg, nr_rem_reg, msg,
845 reply_supported);
846 if (ret < 0) {
847 goto err;
848 }
849 }
850
851 if (nr_add_reg) {
852 ret = send_add_regions(dev, add_reg, nr_add_reg, msg, shadow_pcb,
853 reply_supported, track_ramblocks);
854 if (ret < 0) {
855 goto err;
856 }
857 }
858
859 if (track_ramblocks) {
860 memcpy(u->postcopy_client_bases, shadow_pcb,
861 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
862 /*
863 * Now we've registered this with the postcopy code, we ack to the
864 * client, because now we're in the position to be able to deal with
865 * any faults it generates.
866 */
867 /* TODO: Use this for failure cases as well with a bad value. */
868 msg->hdr.size = sizeof(msg->payload.u64);
869 msg->payload.u64 = 0; /* OK */
870
871 ret = vhost_user_write(dev, msg, NULL, 0);
872 if (ret < 0) {
873 return ret;
874 }
875 }
876
877 return 0;
878
879 err:
880 if (track_ramblocks) {
881 memcpy(u->postcopy_client_bases, shadow_pcb,
882 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
883 }
884
885 return ret;
886 }
887
vhost_user_set_mem_table_postcopy(struct vhost_dev * dev,struct vhost_memory * mem,bool reply_supported,bool config_mem_slots)888 static int vhost_user_set_mem_table_postcopy(struct vhost_dev *dev,
889 struct vhost_memory *mem,
890 bool reply_supported,
891 bool config_mem_slots)
892 {
893 struct vhost_user *u = dev->opaque;
894 int fds[VHOST_MEMORY_BASELINE_NREGIONS];
895 size_t fd_num = 0;
896 VhostUserMsg msg_reply;
897 int region_i, msg_i;
898 int ret;
899
900 VhostUserMsg msg = {
901 .hdr.flags = VHOST_USER_VERSION,
902 };
903
904 if (u->region_rb_len < dev->mem->nregions) {
905 u->region_rb = g_renew(RAMBlock*, u->region_rb, dev->mem->nregions);
906 u->region_rb_offset = g_renew(ram_addr_t, u->region_rb_offset,
907 dev->mem->nregions);
908 memset(&(u->region_rb[u->region_rb_len]), '\0',
909 sizeof(RAMBlock *) * (dev->mem->nregions - u->region_rb_len));
910 memset(&(u->region_rb_offset[u->region_rb_len]), '\0',
911 sizeof(ram_addr_t) * (dev->mem->nregions - u->region_rb_len));
912 u->region_rb_len = dev->mem->nregions;
913 }
914
915 if (config_mem_slots) {
916 ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, true);
917 if (ret < 0) {
918 return ret;
919 }
920 } else {
921 ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
922 true);
923 if (ret < 0) {
924 return ret;
925 }
926
927 ret = vhost_user_write(dev, &msg, fds, fd_num);
928 if (ret < 0) {
929 return ret;
930 }
931
932 ret = vhost_user_read(dev, &msg_reply);
933 if (ret < 0) {
934 return ret;
935 }
936
937 if (msg_reply.hdr.request != VHOST_USER_SET_MEM_TABLE) {
938 error_report("%s: Received unexpected msg type."
939 "Expected %d received %d", __func__,
940 VHOST_USER_SET_MEM_TABLE, msg_reply.hdr.request);
941 return -EPROTO;
942 }
943
944 /*
945 * We're using the same structure, just reusing one of the
946 * fields, so it should be the same size.
947 */
948 if (msg_reply.hdr.size != msg.hdr.size) {
949 error_report("%s: Unexpected size for postcopy reply "
950 "%d vs %d", __func__, msg_reply.hdr.size,
951 msg.hdr.size);
952 return -EPROTO;
953 }
954
955 memset(u->postcopy_client_bases, 0,
956 sizeof(uint64_t) * VHOST_USER_MAX_RAM_SLOTS);
957
958 /*
959 * They're in the same order as the regions that were sent
960 * but some of the regions were skipped (above) if they
961 * didn't have fd's
962 */
963 for (msg_i = 0, region_i = 0;
964 region_i < dev->mem->nregions;
965 region_i++) {
966 if (msg_i < fd_num &&
967 msg_reply.payload.memory.regions[msg_i].guest_phys_addr ==
968 dev->mem->regions[region_i].guest_phys_addr) {
969 u->postcopy_client_bases[region_i] =
970 msg_reply.payload.memory.regions[msg_i].userspace_addr;
971 trace_vhost_user_set_mem_table_postcopy(
972 msg_reply.payload.memory.regions[msg_i].userspace_addr,
973 msg.payload.memory.regions[msg_i].userspace_addr,
974 msg_i, region_i);
975 msg_i++;
976 }
977 }
978 if (msg_i != fd_num) {
979 error_report("%s: postcopy reply not fully consumed "
980 "%d vs %zd",
981 __func__, msg_i, fd_num);
982 return -EIO;
983 }
984
985 /*
986 * Now we've registered this with the postcopy code, we ack to the
987 * client, because now we're in the position to be able to deal
988 * with any faults it generates.
989 */
990 /* TODO: Use this for failure cases as well with a bad value. */
991 msg.hdr.size = sizeof(msg.payload.u64);
992 msg.payload.u64 = 0; /* OK */
993 ret = vhost_user_write(dev, &msg, NULL, 0);
994 if (ret < 0) {
995 return ret;
996 }
997 }
998
999 return 0;
1000 }
1001
vhost_user_set_mem_table(struct vhost_dev * dev,struct vhost_memory * mem)1002 static int vhost_user_set_mem_table(struct vhost_dev *dev,
1003 struct vhost_memory *mem)
1004 {
1005 struct vhost_user *u = dev->opaque;
1006 int fds[VHOST_MEMORY_BASELINE_NREGIONS];
1007 size_t fd_num = 0;
1008 bool do_postcopy = u->postcopy_listen && u->postcopy_fd.handler;
1009 bool reply_supported = virtio_has_feature(dev->protocol_features,
1010 VHOST_USER_PROTOCOL_F_REPLY_ACK);
1011 bool config_mem_slots =
1012 virtio_has_feature(dev->protocol_features,
1013 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS);
1014 int ret;
1015
1016 if (do_postcopy) {
1017 /*
1018 * Postcopy has enough differences that it's best done in it's own
1019 * version
1020 */
1021 return vhost_user_set_mem_table_postcopy(dev, mem, reply_supported,
1022 config_mem_slots);
1023 }
1024
1025 VhostUserMsg msg = {
1026 .hdr.flags = VHOST_USER_VERSION,
1027 };
1028
1029 if (reply_supported) {
1030 msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1031 }
1032
1033 if (config_mem_slots) {
1034 ret = vhost_user_add_remove_regions(dev, &msg, reply_supported, false);
1035 if (ret < 0) {
1036 return ret;
1037 }
1038 } else {
1039 ret = vhost_user_fill_set_mem_table_msg(u, dev, &msg, fds, &fd_num,
1040 false);
1041 if (ret < 0) {
1042 return ret;
1043 }
1044
1045 ret = vhost_user_write(dev, &msg, fds, fd_num);
1046 if (ret < 0) {
1047 return ret;
1048 }
1049
1050 if (reply_supported) {
1051 return process_message_reply(dev, &msg);
1052 }
1053 }
1054
1055 return 0;
1056 }
1057
vhost_user_set_vring_endian(struct vhost_dev * dev,struct vhost_vring_state * ring)1058 static int vhost_user_set_vring_endian(struct vhost_dev *dev,
1059 struct vhost_vring_state *ring)
1060 {
1061 bool cross_endian = virtio_has_feature(dev->protocol_features,
1062 VHOST_USER_PROTOCOL_F_CROSS_ENDIAN);
1063 VhostUserMsg msg = {
1064 .hdr.request = VHOST_USER_SET_VRING_ENDIAN,
1065 .hdr.flags = VHOST_USER_VERSION,
1066 .payload.state = *ring,
1067 .hdr.size = sizeof(msg.payload.state),
1068 };
1069
1070 if (!cross_endian) {
1071 error_report("vhost-user trying to send unhandled ioctl");
1072 return -ENOTSUP;
1073 }
1074
1075 return vhost_user_write(dev, &msg, NULL, 0);
1076 }
1077
vhost_user_get_u64(struct vhost_dev * dev,int request,uint64_t * u64)1078 static int vhost_user_get_u64(struct vhost_dev *dev, int request, uint64_t *u64)
1079 {
1080 int ret;
1081 VhostUserMsg msg = {
1082 .hdr.request = request,
1083 .hdr.flags = VHOST_USER_VERSION,
1084 };
1085
1086 if (vhost_user_per_device_request(request) && dev->vq_index != 0) {
1087 return 0;
1088 }
1089
1090 ret = vhost_user_write(dev, &msg, NULL, 0);
1091 if (ret < 0) {
1092 return ret;
1093 }
1094
1095 ret = vhost_user_read(dev, &msg);
1096 if (ret < 0) {
1097 return ret;
1098 }
1099
1100 if (msg.hdr.request != request) {
1101 error_report("Received unexpected msg type. Expected %d received %d",
1102 request, msg.hdr.request);
1103 return -EPROTO;
1104 }
1105
1106 if (msg.hdr.size != sizeof(msg.payload.u64)) {
1107 error_report("Received bad msg size.");
1108 return -EPROTO;
1109 }
1110
1111 *u64 = msg.payload.u64;
1112
1113 return 0;
1114 }
1115
vhost_user_get_features(struct vhost_dev * dev,uint64_t * features)1116 static int vhost_user_get_features(struct vhost_dev *dev, uint64_t *features)
1117 {
1118 if (vhost_user_get_u64(dev, VHOST_USER_GET_FEATURES, features) < 0) {
1119 return -EPROTO;
1120 }
1121
1122 return 0;
1123 }
1124
1125 /* Note: "msg->hdr.flags" may be modified. */
vhost_user_write_sync(struct vhost_dev * dev,VhostUserMsg * msg,bool wait_for_reply)1126 static int vhost_user_write_sync(struct vhost_dev *dev, VhostUserMsg *msg,
1127 bool wait_for_reply)
1128 {
1129 int ret;
1130
1131 if (wait_for_reply) {
1132 bool reply_supported = virtio_has_feature(dev->protocol_features,
1133 VHOST_USER_PROTOCOL_F_REPLY_ACK);
1134 if (reply_supported) {
1135 msg->hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1136 }
1137 }
1138
1139 ret = vhost_user_write(dev, msg, NULL, 0);
1140 if (ret < 0) {
1141 return ret;
1142 }
1143
1144 if (wait_for_reply) {
1145 uint64_t dummy;
1146
1147 if (msg->hdr.flags & VHOST_USER_NEED_REPLY_MASK) {
1148 return process_message_reply(dev, msg);
1149 }
1150
1151 /*
1152 * We need to wait for a reply but the backend does not
1153 * support replies for the command we just sent.
1154 * Send VHOST_USER_GET_FEATURES which makes all backends
1155 * send a reply.
1156 */
1157 return vhost_user_get_features(dev, &dummy);
1158 }
1159
1160 return 0;
1161 }
1162
vhost_set_vring(struct vhost_dev * dev,unsigned long int request,struct vhost_vring_state * ring,bool wait_for_reply)1163 static int vhost_set_vring(struct vhost_dev *dev,
1164 unsigned long int request,
1165 struct vhost_vring_state *ring,
1166 bool wait_for_reply)
1167 {
1168 VhostUserMsg msg = {
1169 .hdr.request = request,
1170 .hdr.flags = VHOST_USER_VERSION,
1171 .payload.state = *ring,
1172 .hdr.size = sizeof(msg.payload.state),
1173 };
1174
1175 return vhost_user_write_sync(dev, &msg, wait_for_reply);
1176 }
1177
vhost_user_set_vring_num(struct vhost_dev * dev,struct vhost_vring_state * ring)1178 static int vhost_user_set_vring_num(struct vhost_dev *dev,
1179 struct vhost_vring_state *ring)
1180 {
1181 return vhost_set_vring(dev, VHOST_USER_SET_VRING_NUM, ring, false);
1182 }
1183
vhost_user_host_notifier_free(VhostUserHostNotifier * n)1184 static void vhost_user_host_notifier_free(VhostUserHostNotifier *n)
1185 {
1186 if (n->unmap_addr) {
1187 munmap(n->unmap_addr, qemu_real_host_page_size());
1188 n->unmap_addr = NULL;
1189 }
1190 if (n->destroy) {
1191 memory_region_transaction_begin();
1192 object_unparent(OBJECT(&n->mr));
1193 memory_region_transaction_commit();
1194 g_free(n);
1195 }
1196 }
1197
1198 /*
1199 * clean-up function for notifier, will finally free the structure
1200 * under rcu.
1201 */
vhost_user_host_notifier_remove(VhostUserHostNotifier * n,VirtIODevice * vdev,bool destroy)1202 static void vhost_user_host_notifier_remove(VhostUserHostNotifier *n,
1203 VirtIODevice *vdev, bool destroy)
1204 {
1205 /*
1206 * if destroy == false and n->addr == NULL, we have nothing to do.
1207 * so, just return.
1208 */
1209 if (!n || (!destroy && !n->addr)) {
1210 return;
1211 }
1212
1213 if (n->addr) {
1214 if (vdev) {
1215 memory_region_transaction_begin();
1216 virtio_queue_set_host_notifier_mr(vdev, n->idx, &n->mr, false);
1217 memory_region_transaction_commit();
1218 }
1219 assert(!n->unmap_addr);
1220 n->unmap_addr = n->addr;
1221 n->addr = NULL;
1222 }
1223 n->destroy = destroy;
1224 call_rcu(n, vhost_user_host_notifier_free, rcu);
1225 }
1226
vhost_user_set_vring_base(struct vhost_dev * dev,struct vhost_vring_state * ring)1227 static int vhost_user_set_vring_base(struct vhost_dev *dev,
1228 struct vhost_vring_state *ring)
1229 {
1230 return vhost_set_vring(dev, VHOST_USER_SET_VRING_BASE, ring, false);
1231 }
1232
vhost_user_set_vring_enable(struct vhost_dev * dev,int enable)1233 static int vhost_user_set_vring_enable(struct vhost_dev *dev, int enable)
1234 {
1235 int i;
1236
1237 if (!virtio_has_feature(dev->features, VHOST_USER_F_PROTOCOL_FEATURES)) {
1238 return -EINVAL;
1239 }
1240
1241 for (i = 0; i < dev->nvqs; ++i) {
1242 int ret;
1243 struct vhost_vring_state state = {
1244 .index = dev->vq_index + i,
1245 .num = enable,
1246 };
1247
1248 /*
1249 * SET_VRING_ENABLE travels from guest to QEMU to vhost-user backend /
1250 * control plane thread via unix domain socket. Virtio requests travel
1251 * from guest to vhost-user backend / data plane thread via eventfd.
1252 * Even if the guest enables the ring first, and pushes its first virtio
1253 * request second (conforming to the virtio spec), the data plane thread
1254 * in the backend may see the virtio request before the control plane
1255 * thread sees the queue enablement. This causes (in fact, requires) the
1256 * data plane thread to discard the virtio request (it arrived on a
1257 * seemingly disabled queue). To prevent this out-of-order delivery,
1258 * don't let the guest proceed to pushing the virtio request until the
1259 * backend control plane acknowledges enabling the queue -- IOW, pass
1260 * wait_for_reply=true below.
1261 */
1262 ret = vhost_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, &state, true);
1263 if (ret < 0) {
1264 /*
1265 * Restoring the previous state is likely infeasible, as well as
1266 * proceeding regardless the error, so just bail out and hope for
1267 * the device-level recovery.
1268 */
1269 return ret;
1270 }
1271 }
1272
1273 return 0;
1274 }
1275
fetch_notifier(VhostUserState * u,int idx)1276 static VhostUserHostNotifier *fetch_notifier(VhostUserState *u,
1277 int idx)
1278 {
1279 if (idx >= u->notifiers->len) {
1280 return NULL;
1281 }
1282 return g_ptr_array_index(u->notifiers, idx);
1283 }
1284
vhost_user_get_vring_base(struct vhost_dev * dev,struct vhost_vring_state * ring)1285 static int vhost_user_get_vring_base(struct vhost_dev *dev,
1286 struct vhost_vring_state *ring)
1287 {
1288 int ret;
1289 VhostUserMsg msg = {
1290 .hdr.request = VHOST_USER_GET_VRING_BASE,
1291 .hdr.flags = VHOST_USER_VERSION,
1292 .payload.state = *ring,
1293 .hdr.size = sizeof(msg.payload.state),
1294 };
1295 struct vhost_user *u = dev->opaque;
1296
1297 VhostUserHostNotifier *n = fetch_notifier(u->user, ring->index);
1298 vhost_user_host_notifier_remove(n, dev->vdev, false);
1299
1300 ret = vhost_user_write(dev, &msg, NULL, 0);
1301 if (ret < 0) {
1302 return ret;
1303 }
1304
1305 ret = vhost_user_read(dev, &msg);
1306 if (ret < 0) {
1307 return ret;
1308 }
1309
1310 if (msg.hdr.request != VHOST_USER_GET_VRING_BASE) {
1311 error_report("Received unexpected msg type. Expected %d received %d",
1312 VHOST_USER_GET_VRING_BASE, msg.hdr.request);
1313 return -EPROTO;
1314 }
1315
1316 if (msg.hdr.size != sizeof(msg.payload.state)) {
1317 error_report("Received bad msg size.");
1318 return -EPROTO;
1319 }
1320
1321 *ring = msg.payload.state;
1322
1323 return 0;
1324 }
1325
vhost_set_vring_file(struct vhost_dev * dev,VhostUserRequest request,struct vhost_vring_file * file)1326 static int vhost_set_vring_file(struct vhost_dev *dev,
1327 VhostUserRequest request,
1328 struct vhost_vring_file *file)
1329 {
1330 int fds[VHOST_USER_MAX_RAM_SLOTS];
1331 size_t fd_num = 0;
1332 VhostUserMsg msg = {
1333 .hdr.request = request,
1334 .hdr.flags = VHOST_USER_VERSION,
1335 .payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK,
1336 .hdr.size = sizeof(msg.payload.u64),
1337 };
1338
1339 if (file->fd > 0) {
1340 fds[fd_num++] = file->fd;
1341 } else {
1342 msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
1343 }
1344
1345 return vhost_user_write(dev, &msg, fds, fd_num);
1346 }
1347
vhost_user_set_vring_kick(struct vhost_dev * dev,struct vhost_vring_file * file)1348 static int vhost_user_set_vring_kick(struct vhost_dev *dev,
1349 struct vhost_vring_file *file)
1350 {
1351 return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_KICK, file);
1352 }
1353
vhost_user_set_vring_call(struct vhost_dev * dev,struct vhost_vring_file * file)1354 static int vhost_user_set_vring_call(struct vhost_dev *dev,
1355 struct vhost_vring_file *file)
1356 {
1357 return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_CALL, file);
1358 }
1359
vhost_user_set_vring_err(struct vhost_dev * dev,struct vhost_vring_file * file)1360 static int vhost_user_set_vring_err(struct vhost_dev *dev,
1361 struct vhost_vring_file *file)
1362 {
1363 return vhost_set_vring_file(dev, VHOST_USER_SET_VRING_ERR, file);
1364 }
1365
vhost_user_set_vring_addr(struct vhost_dev * dev,struct vhost_vring_addr * addr)1366 static int vhost_user_set_vring_addr(struct vhost_dev *dev,
1367 struct vhost_vring_addr *addr)
1368 {
1369 VhostUserMsg msg = {
1370 .hdr.request = VHOST_USER_SET_VRING_ADDR,
1371 .hdr.flags = VHOST_USER_VERSION,
1372 .payload.addr = *addr,
1373 .hdr.size = sizeof(msg.payload.addr),
1374 };
1375
1376 /*
1377 * wait for a reply if logging is enabled to make sure
1378 * backend is actually logging changes
1379 */
1380 bool wait_for_reply = addr->flags & (1 << VHOST_VRING_F_LOG);
1381
1382 return vhost_user_write_sync(dev, &msg, wait_for_reply);
1383 }
1384
vhost_user_set_u64(struct vhost_dev * dev,int request,uint64_t u64,bool wait_for_reply)1385 static int vhost_user_set_u64(struct vhost_dev *dev, int request, uint64_t u64,
1386 bool wait_for_reply)
1387 {
1388 VhostUserMsg msg = {
1389 .hdr.request = request,
1390 .hdr.flags = VHOST_USER_VERSION,
1391 .payload.u64 = u64,
1392 .hdr.size = sizeof(msg.payload.u64),
1393 };
1394
1395 return vhost_user_write_sync(dev, &msg, wait_for_reply);
1396 }
1397
vhost_user_set_status(struct vhost_dev * dev,uint8_t status)1398 static int vhost_user_set_status(struct vhost_dev *dev, uint8_t status)
1399 {
1400 return vhost_user_set_u64(dev, VHOST_USER_SET_STATUS, status, false);
1401 }
1402
vhost_user_get_status(struct vhost_dev * dev,uint8_t * status)1403 static int vhost_user_get_status(struct vhost_dev *dev, uint8_t *status)
1404 {
1405 uint64_t value;
1406 int ret;
1407
1408 ret = vhost_user_get_u64(dev, VHOST_USER_GET_STATUS, &value);
1409 if (ret < 0) {
1410 return ret;
1411 }
1412 *status = value;
1413
1414 return 0;
1415 }
1416
vhost_user_add_status(struct vhost_dev * dev,uint8_t status)1417 static int vhost_user_add_status(struct vhost_dev *dev, uint8_t status)
1418 {
1419 uint8_t s;
1420 int ret;
1421
1422 ret = vhost_user_get_status(dev, &s);
1423 if (ret < 0) {
1424 return ret;
1425 }
1426
1427 if ((s & status) == status) {
1428 return 0;
1429 }
1430 s |= status;
1431
1432 return vhost_user_set_status(dev, s);
1433 }
1434
vhost_user_set_features(struct vhost_dev * dev,uint64_t features)1435 static int vhost_user_set_features(struct vhost_dev *dev,
1436 uint64_t features)
1437 {
1438 /*
1439 * wait for a reply if logging is enabled to make sure
1440 * backend is actually logging changes
1441 */
1442 bool log_enabled = features & (0x1ULL << VHOST_F_LOG_ALL);
1443 int ret;
1444
1445 /*
1446 * We need to include any extra backend only feature bits that
1447 * might be needed by our device. Currently this includes the
1448 * VHOST_USER_F_PROTOCOL_FEATURES bit for enabling protocol
1449 * features.
1450 */
1451 ret = vhost_user_set_u64(dev, VHOST_USER_SET_FEATURES,
1452 features | dev->backend_features,
1453 log_enabled);
1454
1455 if (virtio_has_feature(dev->protocol_features,
1456 VHOST_USER_PROTOCOL_F_STATUS)) {
1457 if (!ret) {
1458 return vhost_user_add_status(dev, VIRTIO_CONFIG_S_FEATURES_OK);
1459 }
1460 }
1461
1462 return ret;
1463 }
1464
vhost_user_set_protocol_features(struct vhost_dev * dev,uint64_t features)1465 static int vhost_user_set_protocol_features(struct vhost_dev *dev,
1466 uint64_t features)
1467 {
1468 return vhost_user_set_u64(dev, VHOST_USER_SET_PROTOCOL_FEATURES, features,
1469 false);
1470 }
1471
vhost_user_set_owner(struct vhost_dev * dev)1472 static int vhost_user_set_owner(struct vhost_dev *dev)
1473 {
1474 VhostUserMsg msg = {
1475 .hdr.request = VHOST_USER_SET_OWNER,
1476 .hdr.flags = VHOST_USER_VERSION,
1477 };
1478
1479 return vhost_user_write(dev, &msg, NULL, 0);
1480 }
1481
vhost_user_get_max_memslots(struct vhost_dev * dev,uint64_t * max_memslots)1482 static int vhost_user_get_max_memslots(struct vhost_dev *dev,
1483 uint64_t *max_memslots)
1484 {
1485 uint64_t backend_max_memslots;
1486 int err;
1487
1488 err = vhost_user_get_u64(dev, VHOST_USER_GET_MAX_MEM_SLOTS,
1489 &backend_max_memslots);
1490 if (err < 0) {
1491 return err;
1492 }
1493
1494 *max_memslots = backend_max_memslots;
1495
1496 return 0;
1497 }
1498
vhost_user_reset_device(struct vhost_dev * dev)1499 static int vhost_user_reset_device(struct vhost_dev *dev)
1500 {
1501 VhostUserMsg msg = {
1502 .hdr.flags = VHOST_USER_VERSION,
1503 .hdr.request = VHOST_USER_RESET_DEVICE,
1504 };
1505
1506 /*
1507 * Historically, reset was not implemented so only reset devices
1508 * that are expecting it.
1509 */
1510 if (!virtio_has_feature(dev->protocol_features,
1511 VHOST_USER_PROTOCOL_F_RESET_DEVICE)) {
1512 return -ENOSYS;
1513 }
1514
1515 return vhost_user_write(dev, &msg, NULL, 0);
1516 }
1517
vhost_user_backend_handle_config_change(struct vhost_dev * dev)1518 static int vhost_user_backend_handle_config_change(struct vhost_dev *dev)
1519 {
1520 if (!dev->config_ops || !dev->config_ops->vhost_dev_config_notifier) {
1521 return -ENOSYS;
1522 }
1523
1524 return dev->config_ops->vhost_dev_config_notifier(dev);
1525 }
1526
1527 /*
1528 * Fetch or create the notifier for a given idx. Newly created
1529 * notifiers are added to the pointer array that tracks them.
1530 */
fetch_or_create_notifier(VhostUserState * u,int idx)1531 static VhostUserHostNotifier *fetch_or_create_notifier(VhostUserState *u,
1532 int idx)
1533 {
1534 VhostUserHostNotifier *n = NULL;
1535 if (idx >= u->notifiers->len) {
1536 g_ptr_array_set_size(u->notifiers, idx + 1);
1537 }
1538
1539 n = g_ptr_array_index(u->notifiers, idx);
1540 if (!n) {
1541 /*
1542 * In case notification arrive out-of-order,
1543 * make room for current index.
1544 */
1545 g_ptr_array_remove_index(u->notifiers, idx);
1546 n = g_new0(VhostUserHostNotifier, 1);
1547 n->idx = idx;
1548 g_ptr_array_insert(u->notifiers, idx, n);
1549 trace_vhost_user_create_notifier(idx, n);
1550 }
1551
1552 return n;
1553 }
1554
vhost_user_backend_handle_vring_host_notifier(struct vhost_dev * dev,VhostUserVringArea * area,int fd)1555 static int vhost_user_backend_handle_vring_host_notifier(struct vhost_dev *dev,
1556 VhostUserVringArea *area,
1557 int fd)
1558 {
1559 int queue_idx = area->u64 & VHOST_USER_VRING_IDX_MASK;
1560 size_t page_size = qemu_real_host_page_size();
1561 struct vhost_user *u = dev->opaque;
1562 VhostUserState *user = u->user;
1563 VirtIODevice *vdev = dev->vdev;
1564 VhostUserHostNotifier *n;
1565 void *addr;
1566 char *name;
1567
1568 if (!virtio_has_feature(dev->protocol_features,
1569 VHOST_USER_PROTOCOL_F_HOST_NOTIFIER) ||
1570 vdev == NULL || queue_idx >= virtio_get_num_queues(vdev)) {
1571 return -EINVAL;
1572 }
1573
1574 /*
1575 * Fetch notifier and invalidate any old data before setting up
1576 * new mapped address.
1577 */
1578 n = fetch_or_create_notifier(user, queue_idx);
1579 vhost_user_host_notifier_remove(n, vdev, false);
1580
1581 if (area->u64 & VHOST_USER_VRING_NOFD_MASK) {
1582 return 0;
1583 }
1584
1585 /* Sanity check. */
1586 if (area->size != page_size) {
1587 return -EINVAL;
1588 }
1589
1590 addr = mmap(NULL, page_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1591 fd, area->offset);
1592 if (addr == MAP_FAILED) {
1593 return -EFAULT;
1594 }
1595
1596 name = g_strdup_printf("vhost-user/host-notifier@%p mmaps[%d]",
1597 user, queue_idx);
1598 if (!n->mr.ram) { /* Don't init again after suspend. */
1599 memory_region_init_ram_device_ptr(&n->mr, OBJECT(vdev), name,
1600 page_size, addr);
1601 } else {
1602 n->mr.ram_block->host = addr;
1603 }
1604 g_free(name);
1605
1606 if (virtio_queue_set_host_notifier_mr(vdev, queue_idx, &n->mr, true)) {
1607 object_unparent(OBJECT(&n->mr));
1608 munmap(addr, page_size);
1609 return -ENXIO;
1610 }
1611
1612 n->addr = addr;
1613
1614 return 0;
1615 }
1616
1617 static int
vhost_user_backend_handle_shared_object_add(struct vhost_dev * dev,VhostUserShared * object)1618 vhost_user_backend_handle_shared_object_add(struct vhost_dev *dev,
1619 VhostUserShared *object)
1620 {
1621 QemuUUID uuid;
1622
1623 memcpy(uuid.data, object->uuid, sizeof(object->uuid));
1624 return !virtio_add_vhost_device(&uuid, dev);
1625 }
1626
1627 /*
1628 * Handle VHOST_USER_BACKEND_SHARED_OBJECT_REMOVE backend requests.
1629 *
1630 * Return: 0 on success, 1 on error.
1631 */
1632 static int
vhost_user_backend_handle_shared_object_remove(struct vhost_dev * dev,VhostUserShared * object)1633 vhost_user_backend_handle_shared_object_remove(struct vhost_dev *dev,
1634 VhostUserShared *object)
1635 {
1636 QemuUUID uuid;
1637
1638 memcpy(uuid.data, object->uuid, sizeof(object->uuid));
1639 switch (virtio_object_type(&uuid)) {
1640 case TYPE_VHOST_DEV:
1641 {
1642 struct vhost_dev *owner = virtio_lookup_vhost_device(&uuid);
1643 if (dev != owner) {
1644 /* Not allowed to remove non-owned entries */
1645 return 1;
1646 }
1647 break;
1648 }
1649 default:
1650 /* Not allowed to remove non-owned entries */
1651 return 1;
1652 }
1653
1654 return !virtio_remove_resource(&uuid);
1655 }
1656
vhost_user_send_resp(QIOChannel * ioc,VhostUserHeader * hdr,VhostUserPayload * payload,Error ** errp)1657 static bool vhost_user_send_resp(QIOChannel *ioc, VhostUserHeader *hdr,
1658 VhostUserPayload *payload, Error **errp)
1659 {
1660 struct iovec iov[] = {
1661 { .iov_base = hdr, .iov_len = VHOST_USER_HDR_SIZE },
1662 { .iov_base = payload, .iov_len = hdr->size },
1663 };
1664
1665 hdr->flags &= ~VHOST_USER_NEED_REPLY_MASK;
1666 hdr->flags |= VHOST_USER_REPLY_MASK;
1667
1668 return !qio_channel_writev_all(ioc, iov, ARRAY_SIZE(iov), errp);
1669 }
1670
vhost_user_get_shared_object(struct vhost_dev * dev,unsigned char * uuid,int * dmabuf_fd)1671 int vhost_user_get_shared_object(struct vhost_dev *dev, unsigned char *uuid,
1672 int *dmabuf_fd)
1673 {
1674 struct vhost_user *u = dev->opaque;
1675 CharBackend *chr = u->user->chr;
1676 int ret;
1677 VhostUserMsg msg = {
1678 .hdr.request = VHOST_USER_GET_SHARED_OBJECT,
1679 .hdr.flags = VHOST_USER_VERSION,
1680 };
1681 memcpy(msg.payload.object.uuid, uuid, sizeof(msg.payload.object.uuid));
1682
1683 ret = vhost_user_write(dev, &msg, NULL, 0);
1684 if (ret < 0) {
1685 return ret;
1686 }
1687
1688 ret = vhost_user_read(dev, &msg);
1689 if (ret < 0) {
1690 return ret;
1691 }
1692
1693 if (msg.hdr.request != VHOST_USER_GET_SHARED_OBJECT) {
1694 error_report("Received unexpected msg type. "
1695 "Expected %d received %d",
1696 VHOST_USER_GET_SHARED_OBJECT, msg.hdr.request);
1697 return -EPROTO;
1698 }
1699
1700 *dmabuf_fd = qemu_chr_fe_get_msgfd(chr);
1701 if (*dmabuf_fd < 0) {
1702 error_report("Failed to get dmabuf fd");
1703 return -EIO;
1704 }
1705
1706 return 0;
1707 }
1708
1709 static int
vhost_user_backend_handle_shared_object_lookup(struct vhost_user * u,VhostUserShared * object)1710 vhost_user_backend_handle_shared_object_lookup(struct vhost_user *u,
1711 VhostUserShared *object)
1712 {
1713 QemuUUID uuid;
1714 CharBackend *chr = u->user->chr;
1715 int dmabuf_fd = -1;
1716 int fd_num = 0;
1717
1718 memcpy(uuid.data, object->uuid, sizeof(object->uuid));
1719
1720 switch (virtio_object_type(&uuid)) {
1721 case TYPE_DMABUF:
1722 dmabuf_fd = virtio_lookup_dmabuf(&uuid);
1723 break;
1724 case TYPE_VHOST_DEV:
1725 {
1726 struct vhost_dev *dev = virtio_lookup_vhost_device(&uuid);
1727 if (dev == NULL) {
1728 return -EINVAL;
1729 }
1730 int ret = vhost_user_get_shared_object(dev, uuid.data, &dmabuf_fd);
1731 if (ret < 0) {
1732 return ret;
1733 }
1734 break;
1735 }
1736 case TYPE_INVALID:
1737 return -EINVAL;
1738 }
1739
1740 if (dmabuf_fd != -1) {
1741 fd_num++;
1742 }
1743
1744 if (qemu_chr_fe_set_msgfds(chr, &dmabuf_fd, fd_num) < 0) {
1745 error_report("Failed to set msg fds.");
1746 return -EINVAL;
1747 }
1748
1749 return 0;
1750 }
1751
close_backend_channel(struct vhost_user * u)1752 static void close_backend_channel(struct vhost_user *u)
1753 {
1754 g_source_destroy(u->backend_src);
1755 g_source_unref(u->backend_src);
1756 u->backend_src = NULL;
1757 object_unref(OBJECT(u->backend_ioc));
1758 u->backend_ioc = NULL;
1759 }
1760
backend_read(QIOChannel * ioc,GIOCondition condition,gpointer opaque)1761 static gboolean backend_read(QIOChannel *ioc, GIOCondition condition,
1762 gpointer opaque)
1763 {
1764 struct vhost_dev *dev = opaque;
1765 struct vhost_user *u = dev->opaque;
1766 VhostUserHeader hdr = { 0, };
1767 VhostUserPayload payload = { 0, };
1768 Error *local_err = NULL;
1769 gboolean rc = G_SOURCE_CONTINUE;
1770 int ret = 0;
1771 struct iovec iov;
1772 g_autofree int *fd = NULL;
1773 size_t fdsize = 0;
1774 bool reply_ack;
1775 int i;
1776
1777 /* Read header */
1778 iov.iov_base = &hdr;
1779 iov.iov_len = VHOST_USER_HDR_SIZE;
1780
1781 if (qio_channel_readv_full_all(ioc, &iov, 1, &fd, &fdsize, &local_err)) {
1782 error_report_err(local_err);
1783 goto err;
1784 }
1785
1786 if (hdr.size > VHOST_USER_PAYLOAD_SIZE) {
1787 error_report("Failed to read msg header."
1788 " Size %d exceeds the maximum %zu.", hdr.size,
1789 VHOST_USER_PAYLOAD_SIZE);
1790 goto err;
1791 }
1792
1793 reply_ack = hdr.flags & VHOST_USER_NEED_REPLY_MASK;
1794
1795 /* Read payload */
1796 if (qio_channel_read_all(ioc, (char *) &payload, hdr.size, &local_err)) {
1797 error_report_err(local_err);
1798 goto err;
1799 }
1800
1801 switch (hdr.request) {
1802 case VHOST_USER_BACKEND_IOTLB_MSG:
1803 ret = vhost_backend_handle_iotlb_msg(dev, &payload.iotlb);
1804 break;
1805 case VHOST_USER_BACKEND_CONFIG_CHANGE_MSG:
1806 ret = vhost_user_backend_handle_config_change(dev);
1807 break;
1808 case VHOST_USER_BACKEND_VRING_HOST_NOTIFIER_MSG:
1809 ret = vhost_user_backend_handle_vring_host_notifier(dev, &payload.area,
1810 fd ? fd[0] : -1);
1811 break;
1812 case VHOST_USER_BACKEND_SHARED_OBJECT_ADD:
1813 ret = vhost_user_backend_handle_shared_object_add(dev, &payload.object);
1814 break;
1815 case VHOST_USER_BACKEND_SHARED_OBJECT_REMOVE:
1816 ret = vhost_user_backend_handle_shared_object_remove(dev,
1817 &payload.object);
1818 break;
1819 case VHOST_USER_BACKEND_SHARED_OBJECT_LOOKUP:
1820 /* The backend always expects a response */
1821 reply_ack = true;
1822 ret = vhost_user_backend_handle_shared_object_lookup(dev->opaque,
1823 &payload.object);
1824 break;
1825 default:
1826 error_report("Received unexpected msg type: %d.", hdr.request);
1827 ret = -EINVAL;
1828 }
1829
1830 /*
1831 * REPLY_ACK feature handling. Other reply types has to be managed
1832 * directly in their request handlers.
1833 */
1834 if (reply_ack) {
1835 payload.u64 = !!ret;
1836 hdr.size = sizeof(payload.u64);
1837
1838 if (!vhost_user_send_resp(ioc, &hdr, &payload, &local_err)) {
1839 error_report_err(local_err);
1840 goto err;
1841 }
1842 }
1843
1844 goto fdcleanup;
1845
1846 err:
1847 close_backend_channel(u);
1848 rc = G_SOURCE_REMOVE;
1849
1850 fdcleanup:
1851 if (fd) {
1852 for (i = 0; i < fdsize; i++) {
1853 close(fd[i]);
1854 }
1855 }
1856 return rc;
1857 }
1858
vhost_setup_backend_channel(struct vhost_dev * dev)1859 static int vhost_setup_backend_channel(struct vhost_dev *dev)
1860 {
1861 VhostUserMsg msg = {
1862 .hdr.request = VHOST_USER_SET_BACKEND_REQ_FD,
1863 .hdr.flags = VHOST_USER_VERSION,
1864 };
1865 struct vhost_user *u = dev->opaque;
1866 int sv[2], ret = 0;
1867 bool reply_supported = virtio_has_feature(dev->protocol_features,
1868 VHOST_USER_PROTOCOL_F_REPLY_ACK);
1869 Error *local_err = NULL;
1870 QIOChannel *ioc;
1871
1872 if (!virtio_has_feature(dev->protocol_features,
1873 VHOST_USER_PROTOCOL_F_BACKEND_REQ)) {
1874 return 0;
1875 }
1876
1877 if (qemu_socketpair(PF_UNIX, SOCK_STREAM, 0, sv) == -1) {
1878 int saved_errno = errno;
1879 error_report("socketpair() failed");
1880 return -saved_errno;
1881 }
1882
1883 ioc = QIO_CHANNEL(qio_channel_socket_new_fd(sv[0], &local_err));
1884 if (!ioc) {
1885 error_report_err(local_err);
1886 return -ECONNREFUSED;
1887 }
1888 u->backend_ioc = ioc;
1889 u->backend_src = qio_channel_add_watch_source(u->backend_ioc,
1890 G_IO_IN | G_IO_HUP,
1891 backend_read, dev, NULL, NULL);
1892
1893 if (reply_supported) {
1894 msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
1895 }
1896
1897 ret = vhost_user_write(dev, &msg, &sv[1], 1);
1898 if (ret) {
1899 goto out;
1900 }
1901
1902 if (reply_supported) {
1903 ret = process_message_reply(dev, &msg);
1904 }
1905
1906 out:
1907 close(sv[1]);
1908 if (ret) {
1909 close_backend_channel(u);
1910 }
1911
1912 return ret;
1913 }
1914
1915 #ifdef CONFIG_LINUX
1916 /*
1917 * Called back from the postcopy fault thread when a fault is received on our
1918 * ufd.
1919 * TODO: This is Linux specific
1920 */
vhost_user_postcopy_fault_handler(struct PostCopyFD * pcfd,void * ufd)1921 static int vhost_user_postcopy_fault_handler(struct PostCopyFD *pcfd,
1922 void *ufd)
1923 {
1924 struct vhost_dev *dev = pcfd->data;
1925 struct vhost_user *u = dev->opaque;
1926 struct uffd_msg *msg = ufd;
1927 uint64_t faultaddr = msg->arg.pagefault.address;
1928 RAMBlock *rb = NULL;
1929 uint64_t rb_offset;
1930 int i;
1931
1932 trace_vhost_user_postcopy_fault_handler(pcfd->idstr, faultaddr,
1933 dev->mem->nregions);
1934 for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1935 trace_vhost_user_postcopy_fault_handler_loop(i,
1936 u->postcopy_client_bases[i], dev->mem->regions[i].memory_size);
1937 if (faultaddr >= u->postcopy_client_bases[i]) {
1938 /* Ofset of the fault address in the vhost region */
1939 uint64_t region_offset = faultaddr - u->postcopy_client_bases[i];
1940 if (region_offset < dev->mem->regions[i].memory_size) {
1941 rb_offset = region_offset + u->region_rb_offset[i];
1942 trace_vhost_user_postcopy_fault_handler_found(i,
1943 region_offset, rb_offset);
1944 rb = u->region_rb[i];
1945 return postcopy_request_shared_page(pcfd, rb, faultaddr,
1946 rb_offset);
1947 }
1948 }
1949 }
1950 error_report("%s: Failed to find region for fault %" PRIx64,
1951 __func__, faultaddr);
1952 return -1;
1953 }
1954
vhost_user_postcopy_waker(struct PostCopyFD * pcfd,RAMBlock * rb,uint64_t offset)1955 static int vhost_user_postcopy_waker(struct PostCopyFD *pcfd, RAMBlock *rb,
1956 uint64_t offset)
1957 {
1958 struct vhost_dev *dev = pcfd->data;
1959 struct vhost_user *u = dev->opaque;
1960 int i;
1961
1962 trace_vhost_user_postcopy_waker(qemu_ram_get_idstr(rb), offset);
1963
1964 if (!u) {
1965 return 0;
1966 }
1967 /* Translate the offset into an address in the clients address space */
1968 for (i = 0; i < MIN(dev->mem->nregions, u->region_rb_len); i++) {
1969 if (u->region_rb[i] == rb &&
1970 offset >= u->region_rb_offset[i] &&
1971 offset < (u->region_rb_offset[i] +
1972 dev->mem->regions[i].memory_size)) {
1973 uint64_t client_addr = (offset - u->region_rb_offset[i]) +
1974 u->postcopy_client_bases[i];
1975 trace_vhost_user_postcopy_waker_found(client_addr);
1976 return postcopy_wake_shared(pcfd, client_addr, rb);
1977 }
1978 }
1979
1980 trace_vhost_user_postcopy_waker_nomatch(qemu_ram_get_idstr(rb), offset);
1981 return 0;
1982 }
1983 #endif
1984
1985 /*
1986 * Called at the start of an inbound postcopy on reception of the
1987 * 'advise' command.
1988 */
vhost_user_postcopy_advise(struct vhost_dev * dev,Error ** errp)1989 static int vhost_user_postcopy_advise(struct vhost_dev *dev, Error **errp)
1990 {
1991 #ifdef CONFIG_LINUX
1992 struct vhost_user *u = dev->opaque;
1993 CharBackend *chr = u->user->chr;
1994 int ufd;
1995 int ret;
1996 VhostUserMsg msg = {
1997 .hdr.request = VHOST_USER_POSTCOPY_ADVISE,
1998 .hdr.flags = VHOST_USER_VERSION,
1999 };
2000
2001 ret = vhost_user_write(dev, &msg, NULL, 0);
2002 if (ret < 0) {
2003 error_setg(errp, "Failed to send postcopy_advise to vhost");
2004 return ret;
2005 }
2006
2007 ret = vhost_user_read(dev, &msg);
2008 if (ret < 0) {
2009 error_setg(errp, "Failed to get postcopy_advise reply from vhost");
2010 return ret;
2011 }
2012
2013 if (msg.hdr.request != VHOST_USER_POSTCOPY_ADVISE) {
2014 error_setg(errp, "Unexpected msg type. Expected %d received %d",
2015 VHOST_USER_POSTCOPY_ADVISE, msg.hdr.request);
2016 return -EPROTO;
2017 }
2018
2019 if (msg.hdr.size) {
2020 error_setg(errp, "Received bad msg size.");
2021 return -EPROTO;
2022 }
2023 ufd = qemu_chr_fe_get_msgfd(chr);
2024 if (ufd < 0) {
2025 error_setg(errp, "%s: Failed to get ufd", __func__);
2026 return -EIO;
2027 }
2028 qemu_socket_set_nonblock(ufd);
2029
2030 /* register ufd with userfault thread */
2031 u->postcopy_fd.fd = ufd;
2032 u->postcopy_fd.data = dev;
2033 u->postcopy_fd.handler = vhost_user_postcopy_fault_handler;
2034 u->postcopy_fd.waker = vhost_user_postcopy_waker;
2035 u->postcopy_fd.idstr = "vhost-user"; /* Need to find unique name */
2036 postcopy_register_shared_ufd(&u->postcopy_fd);
2037 return 0;
2038 #else
2039 error_setg(errp, "Postcopy not supported on non-Linux systems");
2040 return -ENOSYS;
2041 #endif
2042 }
2043
2044 /*
2045 * Called at the switch to postcopy on reception of the 'listen' command.
2046 */
vhost_user_postcopy_listen(struct vhost_dev * dev,Error ** errp)2047 static int vhost_user_postcopy_listen(struct vhost_dev *dev, Error **errp)
2048 {
2049 struct vhost_user *u = dev->opaque;
2050 int ret;
2051 VhostUserMsg msg = {
2052 .hdr.request = VHOST_USER_POSTCOPY_LISTEN,
2053 .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2054 };
2055 u->postcopy_listen = true;
2056
2057 trace_vhost_user_postcopy_listen();
2058
2059 ret = vhost_user_write(dev, &msg, NULL, 0);
2060 if (ret < 0) {
2061 error_setg(errp, "Failed to send postcopy_listen to vhost");
2062 return ret;
2063 }
2064
2065 ret = process_message_reply(dev, &msg);
2066 if (ret) {
2067 error_setg(errp, "Failed to receive reply to postcopy_listen");
2068 return ret;
2069 }
2070
2071 return 0;
2072 }
2073
2074 /*
2075 * Called at the end of postcopy
2076 */
vhost_user_postcopy_end(struct vhost_dev * dev,Error ** errp)2077 static int vhost_user_postcopy_end(struct vhost_dev *dev, Error **errp)
2078 {
2079 VhostUserMsg msg = {
2080 .hdr.request = VHOST_USER_POSTCOPY_END,
2081 .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2082 };
2083 int ret;
2084 struct vhost_user *u = dev->opaque;
2085
2086 trace_vhost_user_postcopy_end_entry();
2087
2088 ret = vhost_user_write(dev, &msg, NULL, 0);
2089 if (ret < 0) {
2090 error_setg(errp, "Failed to send postcopy_end to vhost");
2091 return ret;
2092 }
2093
2094 ret = process_message_reply(dev, &msg);
2095 if (ret) {
2096 error_setg(errp, "Failed to receive reply to postcopy_end");
2097 return ret;
2098 }
2099 postcopy_unregister_shared_ufd(&u->postcopy_fd);
2100 close(u->postcopy_fd.fd);
2101 u->postcopy_fd.handler = NULL;
2102
2103 trace_vhost_user_postcopy_end_exit();
2104
2105 return 0;
2106 }
2107
vhost_user_postcopy_notifier(NotifierWithReturn * notifier,void * opaque,Error ** errp)2108 static int vhost_user_postcopy_notifier(NotifierWithReturn *notifier,
2109 void *opaque, Error **errp)
2110 {
2111 struct PostcopyNotifyData *pnd = opaque;
2112 struct vhost_user *u = container_of(notifier, struct vhost_user,
2113 postcopy_notifier);
2114 struct vhost_dev *dev = u->dev;
2115
2116 switch (pnd->reason) {
2117 case POSTCOPY_NOTIFY_PROBE:
2118 if (!virtio_has_feature(dev->protocol_features,
2119 VHOST_USER_PROTOCOL_F_PAGEFAULT)) {
2120 /* TODO: Get the device name into this error somehow */
2121 error_setg(errp,
2122 "vhost-user backend not capable of postcopy");
2123 return -ENOENT;
2124 }
2125 break;
2126
2127 case POSTCOPY_NOTIFY_INBOUND_ADVISE:
2128 return vhost_user_postcopy_advise(dev, errp);
2129
2130 case POSTCOPY_NOTIFY_INBOUND_LISTEN:
2131 return vhost_user_postcopy_listen(dev, errp);
2132
2133 case POSTCOPY_NOTIFY_INBOUND_END:
2134 return vhost_user_postcopy_end(dev, errp);
2135
2136 default:
2137 /* We ignore notifications we don't know */
2138 break;
2139 }
2140
2141 return 0;
2142 }
2143
vhost_user_backend_init(struct vhost_dev * dev,void * opaque,Error ** errp)2144 static int vhost_user_backend_init(struct vhost_dev *dev, void *opaque,
2145 Error **errp)
2146 {
2147 uint64_t features, ram_slots;
2148 struct vhost_user *u;
2149 VhostUserState *vus = (VhostUserState *) opaque;
2150 int err;
2151
2152 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2153
2154 u = g_new0(struct vhost_user, 1);
2155 u->user = vus;
2156 u->dev = dev;
2157 dev->opaque = u;
2158
2159 err = vhost_user_get_features(dev, &features);
2160 if (err < 0) {
2161 error_setg_errno(errp, -err, "vhost_backend_init failed");
2162 return err;
2163 }
2164
2165 if (virtio_has_feature(features, VHOST_USER_F_PROTOCOL_FEATURES)) {
2166 bool supports_f_config = vus->supports_config ||
2167 (dev->config_ops && dev->config_ops->vhost_dev_config_notifier);
2168 uint64_t protocol_features;
2169
2170 dev->backend_features |= 1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
2171
2172 err = vhost_user_get_u64(dev, VHOST_USER_GET_PROTOCOL_FEATURES,
2173 &protocol_features);
2174 if (err < 0) {
2175 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2176 return -EPROTO;
2177 }
2178
2179 /*
2180 * We will use all the protocol features we support - although
2181 * we suppress F_CONFIG if we know QEMUs internal code can not support
2182 * it.
2183 */
2184 protocol_features &= VHOST_USER_PROTOCOL_FEATURE_MASK;
2185
2186 if (supports_f_config) {
2187 if (!virtio_has_feature(protocol_features,
2188 VHOST_USER_PROTOCOL_F_CONFIG)) {
2189 error_setg(errp, "vhost-user device expecting "
2190 "VHOST_USER_PROTOCOL_F_CONFIG but the vhost-user backend does "
2191 "not support it.");
2192 return -EPROTO;
2193 }
2194 } else {
2195 if (virtio_has_feature(protocol_features,
2196 VHOST_USER_PROTOCOL_F_CONFIG)) {
2197 warn_report("vhost-user backend supports "
2198 "VHOST_USER_PROTOCOL_F_CONFIG but QEMU does not.");
2199 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_CONFIG);
2200 }
2201 }
2202
2203 /* final set of protocol features */
2204 dev->protocol_features = protocol_features;
2205 err = vhost_user_set_protocol_features(dev, dev->protocol_features);
2206 if (err < 0) {
2207 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2208 return -EPROTO;
2209 }
2210
2211 /* query the max queues we support if backend supports Multiple Queue */
2212 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_MQ)) {
2213 err = vhost_user_get_u64(dev, VHOST_USER_GET_QUEUE_NUM,
2214 &dev->max_queues);
2215 if (err < 0) {
2216 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2217 return -EPROTO;
2218 }
2219 } else {
2220 dev->max_queues = 1;
2221 }
2222
2223 if (dev->num_queues && dev->max_queues < dev->num_queues) {
2224 error_setg(errp, "The maximum number of queues supported by the "
2225 "backend is %" PRIu64, dev->max_queues);
2226 return -EINVAL;
2227 }
2228
2229 if (virtio_has_feature(features, VIRTIO_F_IOMMU_PLATFORM) &&
2230 !(virtio_has_feature(dev->protocol_features,
2231 VHOST_USER_PROTOCOL_F_BACKEND_REQ) &&
2232 virtio_has_feature(dev->protocol_features,
2233 VHOST_USER_PROTOCOL_F_REPLY_ACK))) {
2234 error_setg(errp, "IOMMU support requires reply-ack and "
2235 "backend-req protocol features.");
2236 return -EINVAL;
2237 }
2238
2239 /* get max memory regions if backend supports configurable RAM slots */
2240 if (!virtio_has_feature(dev->protocol_features,
2241 VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS)) {
2242 u->user->memory_slots = VHOST_MEMORY_BASELINE_NREGIONS;
2243 } else {
2244 err = vhost_user_get_max_memslots(dev, &ram_slots);
2245 if (err < 0) {
2246 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2247 return -EPROTO;
2248 }
2249
2250 if (ram_slots < u->user->memory_slots) {
2251 error_setg(errp, "The backend specified a max ram slots limit "
2252 "of %" PRIu64", when the prior validated limit was "
2253 "%d. This limit should never decrease.", ram_slots,
2254 u->user->memory_slots);
2255 return -EINVAL;
2256 }
2257
2258 u->user->memory_slots = MIN(ram_slots, VHOST_USER_MAX_RAM_SLOTS);
2259 }
2260 }
2261
2262 if (dev->migration_blocker == NULL &&
2263 !virtio_has_feature(dev->protocol_features,
2264 VHOST_USER_PROTOCOL_F_LOG_SHMFD)) {
2265 error_setg(&dev->migration_blocker,
2266 "Migration disabled: vhost-user backend lacks "
2267 "VHOST_USER_PROTOCOL_F_LOG_SHMFD feature.");
2268 }
2269
2270 if (dev->vq_index == 0) {
2271 err = vhost_setup_backend_channel(dev);
2272 if (err < 0) {
2273 error_setg_errno(errp, EPROTO, "vhost_backend_init failed");
2274 return -EPROTO;
2275 }
2276 }
2277
2278 u->postcopy_notifier.notify = vhost_user_postcopy_notifier;
2279 postcopy_add_notifier(&u->postcopy_notifier);
2280
2281 return 0;
2282 }
2283
vhost_user_backend_cleanup(struct vhost_dev * dev)2284 static int vhost_user_backend_cleanup(struct vhost_dev *dev)
2285 {
2286 struct vhost_user *u;
2287
2288 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2289
2290 u = dev->opaque;
2291 if (u->postcopy_notifier.notify) {
2292 postcopy_remove_notifier(&u->postcopy_notifier);
2293 u->postcopy_notifier.notify = NULL;
2294 }
2295 u->postcopy_listen = false;
2296 if (u->postcopy_fd.handler) {
2297 postcopy_unregister_shared_ufd(&u->postcopy_fd);
2298 close(u->postcopy_fd.fd);
2299 u->postcopy_fd.handler = NULL;
2300 }
2301 if (u->backend_ioc) {
2302 close_backend_channel(u);
2303 }
2304 g_free(u->region_rb);
2305 u->region_rb = NULL;
2306 g_free(u->region_rb_offset);
2307 u->region_rb_offset = NULL;
2308 u->region_rb_len = 0;
2309 g_free(u);
2310 dev->opaque = 0;
2311
2312 return 0;
2313 }
2314
vhost_user_get_vq_index(struct vhost_dev * dev,int idx)2315 static int vhost_user_get_vq_index(struct vhost_dev *dev, int idx)
2316 {
2317 assert(idx >= dev->vq_index && idx < dev->vq_index + dev->nvqs);
2318
2319 return idx;
2320 }
2321
vhost_user_memslots_limit(struct vhost_dev * dev)2322 static int vhost_user_memslots_limit(struct vhost_dev *dev)
2323 {
2324 struct vhost_user *u = dev->opaque;
2325
2326 return u->user->memory_slots;
2327 }
2328
vhost_user_requires_shm_log(struct vhost_dev * dev)2329 static bool vhost_user_requires_shm_log(struct vhost_dev *dev)
2330 {
2331 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2332
2333 return virtio_has_feature(dev->protocol_features,
2334 VHOST_USER_PROTOCOL_F_LOG_SHMFD);
2335 }
2336
vhost_user_migration_done(struct vhost_dev * dev,char * mac_addr)2337 static int vhost_user_migration_done(struct vhost_dev *dev, char* mac_addr)
2338 {
2339 VhostUserMsg msg = { };
2340
2341 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2342
2343 /* If guest supports GUEST_ANNOUNCE do nothing */
2344 if (virtio_has_feature(dev->acked_features, VIRTIO_NET_F_GUEST_ANNOUNCE)) {
2345 return 0;
2346 }
2347
2348 /* if backend supports VHOST_USER_PROTOCOL_F_RARP ask it to send the RARP */
2349 if (virtio_has_feature(dev->protocol_features,
2350 VHOST_USER_PROTOCOL_F_RARP)) {
2351 msg.hdr.request = VHOST_USER_SEND_RARP;
2352 msg.hdr.flags = VHOST_USER_VERSION;
2353 memcpy((char *)&msg.payload.u64, mac_addr, 6);
2354 msg.hdr.size = sizeof(msg.payload.u64);
2355
2356 return vhost_user_write(dev, &msg, NULL, 0);
2357 }
2358 return -ENOTSUP;
2359 }
2360
vhost_user_net_set_mtu(struct vhost_dev * dev,uint16_t mtu)2361 static int vhost_user_net_set_mtu(struct vhost_dev *dev, uint16_t mtu)
2362 {
2363 VhostUserMsg msg;
2364 bool reply_supported = virtio_has_feature(dev->protocol_features,
2365 VHOST_USER_PROTOCOL_F_REPLY_ACK);
2366 int ret;
2367
2368 if (!(dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_NET_MTU))) {
2369 return 0;
2370 }
2371
2372 msg.hdr.request = VHOST_USER_NET_SET_MTU;
2373 msg.payload.u64 = mtu;
2374 msg.hdr.size = sizeof(msg.payload.u64);
2375 msg.hdr.flags = VHOST_USER_VERSION;
2376 if (reply_supported) {
2377 msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2378 }
2379
2380 ret = vhost_user_write(dev, &msg, NULL, 0);
2381 if (ret < 0) {
2382 return ret;
2383 }
2384
2385 /* If reply_ack supported, backend has to ack specified MTU is valid */
2386 if (reply_supported) {
2387 return process_message_reply(dev, &msg);
2388 }
2389
2390 return 0;
2391 }
2392
vhost_user_send_device_iotlb_msg(struct vhost_dev * dev,struct vhost_iotlb_msg * imsg)2393 static int vhost_user_send_device_iotlb_msg(struct vhost_dev *dev,
2394 struct vhost_iotlb_msg *imsg)
2395 {
2396 int ret;
2397 VhostUserMsg msg = {
2398 .hdr.request = VHOST_USER_IOTLB_MSG,
2399 .hdr.size = sizeof(msg.payload.iotlb),
2400 .hdr.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY_MASK,
2401 .payload.iotlb = *imsg,
2402 };
2403
2404 ret = vhost_user_write(dev, &msg, NULL, 0);
2405 if (ret < 0) {
2406 return ret;
2407 }
2408
2409 return process_message_reply(dev, &msg);
2410 }
2411
2412
vhost_user_set_iotlb_callback(struct vhost_dev * dev,int enabled)2413 static void vhost_user_set_iotlb_callback(struct vhost_dev *dev, int enabled)
2414 {
2415 /* No-op as the receive channel is not dedicated to IOTLB messages. */
2416 }
2417
vhost_user_get_config(struct vhost_dev * dev,uint8_t * config,uint32_t config_len,Error ** errp)2418 static int vhost_user_get_config(struct vhost_dev *dev, uint8_t *config,
2419 uint32_t config_len, Error **errp)
2420 {
2421 int ret;
2422 VhostUserMsg msg = {
2423 .hdr.request = VHOST_USER_GET_CONFIG,
2424 .hdr.flags = VHOST_USER_VERSION,
2425 .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + config_len,
2426 };
2427
2428 if (!virtio_has_feature(dev->protocol_features,
2429 VHOST_USER_PROTOCOL_F_CONFIG)) {
2430 error_setg(errp, "VHOST_USER_PROTOCOL_F_CONFIG not supported");
2431 return -EINVAL;
2432 }
2433
2434 assert(config_len <= VHOST_USER_MAX_CONFIG_SIZE);
2435
2436 msg.payload.config.offset = 0;
2437 msg.payload.config.size = config_len;
2438 ret = vhost_user_write(dev, &msg, NULL, 0);
2439 if (ret < 0) {
2440 error_setg_errno(errp, -ret, "vhost_get_config failed");
2441 return ret;
2442 }
2443
2444 ret = vhost_user_read(dev, &msg);
2445 if (ret < 0) {
2446 error_setg_errno(errp, -ret, "vhost_get_config failed");
2447 return ret;
2448 }
2449
2450 if (msg.hdr.request != VHOST_USER_GET_CONFIG) {
2451 error_setg(errp,
2452 "Received unexpected msg type. Expected %d received %d",
2453 VHOST_USER_GET_CONFIG, msg.hdr.request);
2454 return -EPROTO;
2455 }
2456
2457 if (msg.hdr.size != VHOST_USER_CONFIG_HDR_SIZE + config_len) {
2458 error_setg(errp, "Received bad msg size.");
2459 return -EPROTO;
2460 }
2461
2462 memcpy(config, msg.payload.config.region, config_len);
2463
2464 return 0;
2465 }
2466
vhost_user_set_config(struct vhost_dev * dev,const uint8_t * data,uint32_t offset,uint32_t size,uint32_t flags)2467 static int vhost_user_set_config(struct vhost_dev *dev, const uint8_t *data,
2468 uint32_t offset, uint32_t size, uint32_t flags)
2469 {
2470 int ret;
2471 uint8_t *p;
2472 bool reply_supported = virtio_has_feature(dev->protocol_features,
2473 VHOST_USER_PROTOCOL_F_REPLY_ACK);
2474
2475 VhostUserMsg msg = {
2476 .hdr.request = VHOST_USER_SET_CONFIG,
2477 .hdr.flags = VHOST_USER_VERSION,
2478 .hdr.size = VHOST_USER_CONFIG_HDR_SIZE + size,
2479 };
2480
2481 if (!virtio_has_feature(dev->protocol_features,
2482 VHOST_USER_PROTOCOL_F_CONFIG)) {
2483 return -ENOTSUP;
2484 }
2485
2486 if (reply_supported) {
2487 msg.hdr.flags |= VHOST_USER_NEED_REPLY_MASK;
2488 }
2489
2490 if (size > VHOST_USER_MAX_CONFIG_SIZE) {
2491 return -EINVAL;
2492 }
2493
2494 msg.payload.config.offset = offset,
2495 msg.payload.config.size = size,
2496 msg.payload.config.flags = flags,
2497 p = msg.payload.config.region;
2498 memcpy(p, data, size);
2499
2500 ret = vhost_user_write(dev, &msg, NULL, 0);
2501 if (ret < 0) {
2502 return ret;
2503 }
2504
2505 if (reply_supported) {
2506 return process_message_reply(dev, &msg);
2507 }
2508
2509 return 0;
2510 }
2511
vhost_user_crypto_create_session(struct vhost_dev * dev,void * session_info,uint64_t * session_id)2512 static int vhost_user_crypto_create_session(struct vhost_dev *dev,
2513 void *session_info,
2514 uint64_t *session_id)
2515 {
2516 int ret;
2517 bool crypto_session = virtio_has_feature(dev->protocol_features,
2518 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2519 CryptoDevBackendSessionInfo *backend_info = session_info;
2520 VhostUserMsg msg = {
2521 .hdr.request = VHOST_USER_CREATE_CRYPTO_SESSION,
2522 .hdr.flags = VHOST_USER_VERSION,
2523 .hdr.size = sizeof(msg.payload.session),
2524 };
2525
2526 assert(dev->vhost_ops->backend_type == VHOST_BACKEND_TYPE_USER);
2527
2528 if (!crypto_session) {
2529 error_report("vhost-user trying to send unhandled ioctl");
2530 return -ENOTSUP;
2531 }
2532
2533 if (backend_info->op_code == VIRTIO_CRYPTO_AKCIPHER_CREATE_SESSION) {
2534 CryptoDevBackendAsymSessionInfo *sess = &backend_info->u.asym_sess_info;
2535 size_t keylen;
2536
2537 memcpy(&msg.payload.session.u.asym.session_setup_data, sess,
2538 sizeof(CryptoDevBackendAsymSessionInfo));
2539 if (sess->keylen) {
2540 keylen = sizeof(msg.payload.session.u.asym.key);
2541 if (sess->keylen > keylen) {
2542 error_report("Unsupported asymmetric key size");
2543 return -ENOTSUP;
2544 }
2545
2546 memcpy(&msg.payload.session.u.asym.key, sess->key,
2547 sess->keylen);
2548 }
2549 } else {
2550 CryptoDevBackendSymSessionInfo *sess = &backend_info->u.sym_sess_info;
2551 size_t keylen;
2552
2553 memcpy(&msg.payload.session.u.sym.session_setup_data, sess,
2554 sizeof(CryptoDevBackendSymSessionInfo));
2555 if (sess->key_len) {
2556 keylen = sizeof(msg.payload.session.u.sym.key);
2557 if (sess->key_len > keylen) {
2558 error_report("Unsupported cipher key size");
2559 return -ENOTSUP;
2560 }
2561
2562 memcpy(&msg.payload.session.u.sym.key, sess->cipher_key,
2563 sess->key_len);
2564 }
2565
2566 if (sess->auth_key_len > 0) {
2567 keylen = sizeof(msg.payload.session.u.sym.auth_key);
2568 if (sess->auth_key_len > keylen) {
2569 error_report("Unsupported auth key size");
2570 return -ENOTSUP;
2571 }
2572
2573 memcpy(&msg.payload.session.u.sym.auth_key, sess->auth_key,
2574 sess->auth_key_len);
2575 }
2576 }
2577
2578 msg.payload.session.op_code = backend_info->op_code;
2579 msg.payload.session.session_id = backend_info->session_id;
2580 ret = vhost_user_write(dev, &msg, NULL, 0);
2581 if (ret < 0) {
2582 error_report("vhost_user_write() return %d, create session failed",
2583 ret);
2584 return ret;
2585 }
2586
2587 ret = vhost_user_read(dev, &msg);
2588 if (ret < 0) {
2589 error_report("vhost_user_read() return %d, create session failed",
2590 ret);
2591 return ret;
2592 }
2593
2594 if (msg.hdr.request != VHOST_USER_CREATE_CRYPTO_SESSION) {
2595 error_report("Received unexpected msg type. Expected %d received %d",
2596 VHOST_USER_CREATE_CRYPTO_SESSION, msg.hdr.request);
2597 return -EPROTO;
2598 }
2599
2600 if (msg.hdr.size != sizeof(msg.payload.session)) {
2601 error_report("Received bad msg size.");
2602 return -EPROTO;
2603 }
2604
2605 if (msg.payload.session.session_id < 0) {
2606 error_report("Bad session id: %" PRId64 "",
2607 msg.payload.session.session_id);
2608 return -EINVAL;
2609 }
2610 *session_id = msg.payload.session.session_id;
2611
2612 return 0;
2613 }
2614
2615 static int
vhost_user_crypto_close_session(struct vhost_dev * dev,uint64_t session_id)2616 vhost_user_crypto_close_session(struct vhost_dev *dev, uint64_t session_id)
2617 {
2618 int ret;
2619 bool crypto_session = virtio_has_feature(dev->protocol_features,
2620 VHOST_USER_PROTOCOL_F_CRYPTO_SESSION);
2621 VhostUserMsg msg = {
2622 .hdr.request = VHOST_USER_CLOSE_CRYPTO_SESSION,
2623 .hdr.flags = VHOST_USER_VERSION,
2624 .hdr.size = sizeof(msg.payload.u64),
2625 };
2626 msg.payload.u64 = session_id;
2627
2628 if (!crypto_session) {
2629 error_report("vhost-user trying to send unhandled ioctl");
2630 return -ENOTSUP;
2631 }
2632
2633 ret = vhost_user_write(dev, &msg, NULL, 0);
2634 if (ret < 0) {
2635 error_report("vhost_user_write() return %d, close session failed",
2636 ret);
2637 return ret;
2638 }
2639
2640 return 0;
2641 }
2642
vhost_user_no_private_memslots(struct vhost_dev * dev)2643 static bool vhost_user_no_private_memslots(struct vhost_dev *dev)
2644 {
2645 return true;
2646 }
2647
vhost_user_get_inflight_fd(struct vhost_dev * dev,uint16_t queue_size,struct vhost_inflight * inflight)2648 static int vhost_user_get_inflight_fd(struct vhost_dev *dev,
2649 uint16_t queue_size,
2650 struct vhost_inflight *inflight)
2651 {
2652 void *addr;
2653 int fd;
2654 int ret;
2655 struct vhost_user *u = dev->opaque;
2656 CharBackend *chr = u->user->chr;
2657 VhostUserMsg msg = {
2658 .hdr.request = VHOST_USER_GET_INFLIGHT_FD,
2659 .hdr.flags = VHOST_USER_VERSION,
2660 .payload.inflight.num_queues = dev->nvqs,
2661 .payload.inflight.queue_size = queue_size,
2662 .hdr.size = sizeof(msg.payload.inflight),
2663 };
2664
2665 if (!virtio_has_feature(dev->protocol_features,
2666 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2667 return 0;
2668 }
2669
2670 ret = vhost_user_write(dev, &msg, NULL, 0);
2671 if (ret < 0) {
2672 return ret;
2673 }
2674
2675 ret = vhost_user_read(dev, &msg);
2676 if (ret < 0) {
2677 return ret;
2678 }
2679
2680 if (msg.hdr.request != VHOST_USER_GET_INFLIGHT_FD) {
2681 error_report("Received unexpected msg type. "
2682 "Expected %d received %d",
2683 VHOST_USER_GET_INFLIGHT_FD, msg.hdr.request);
2684 return -EPROTO;
2685 }
2686
2687 if (msg.hdr.size != sizeof(msg.payload.inflight)) {
2688 error_report("Received bad msg size.");
2689 return -EPROTO;
2690 }
2691
2692 if (!msg.payload.inflight.mmap_size) {
2693 return 0;
2694 }
2695
2696 fd = qemu_chr_fe_get_msgfd(chr);
2697 if (fd < 0) {
2698 error_report("Failed to get mem fd");
2699 return -EIO;
2700 }
2701
2702 addr = mmap(0, msg.payload.inflight.mmap_size, PROT_READ | PROT_WRITE,
2703 MAP_SHARED, fd, msg.payload.inflight.mmap_offset);
2704
2705 if (addr == MAP_FAILED) {
2706 error_report("Failed to mmap mem fd");
2707 close(fd);
2708 return -EFAULT;
2709 }
2710
2711 inflight->addr = addr;
2712 inflight->fd = fd;
2713 inflight->size = msg.payload.inflight.mmap_size;
2714 inflight->offset = msg.payload.inflight.mmap_offset;
2715 inflight->queue_size = queue_size;
2716
2717 return 0;
2718 }
2719
vhost_user_set_inflight_fd(struct vhost_dev * dev,struct vhost_inflight * inflight)2720 static int vhost_user_set_inflight_fd(struct vhost_dev *dev,
2721 struct vhost_inflight *inflight)
2722 {
2723 VhostUserMsg msg = {
2724 .hdr.request = VHOST_USER_SET_INFLIGHT_FD,
2725 .hdr.flags = VHOST_USER_VERSION,
2726 .payload.inflight.mmap_size = inflight->size,
2727 .payload.inflight.mmap_offset = inflight->offset,
2728 .payload.inflight.num_queues = dev->nvqs,
2729 .payload.inflight.queue_size = inflight->queue_size,
2730 .hdr.size = sizeof(msg.payload.inflight),
2731 };
2732
2733 if (!virtio_has_feature(dev->protocol_features,
2734 VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
2735 return 0;
2736 }
2737
2738 return vhost_user_write(dev, &msg, &inflight->fd, 1);
2739 }
2740
vhost_user_state_destroy(gpointer data)2741 static void vhost_user_state_destroy(gpointer data)
2742 {
2743 VhostUserHostNotifier *n = (VhostUserHostNotifier *) data;
2744 vhost_user_host_notifier_remove(n, NULL, true);
2745 }
2746
vhost_user_init(VhostUserState * user,CharBackend * chr,Error ** errp)2747 bool vhost_user_init(VhostUserState *user, CharBackend *chr, Error **errp)
2748 {
2749 if (user->chr) {
2750 error_setg(errp, "Cannot initialize vhost-user state");
2751 return false;
2752 }
2753 user->chr = chr;
2754 user->memory_slots = 0;
2755 user->notifiers = g_ptr_array_new_full(VIRTIO_QUEUE_MAX / 4,
2756 &vhost_user_state_destroy);
2757 return true;
2758 }
2759
vhost_user_cleanup(VhostUserState * user)2760 void vhost_user_cleanup(VhostUserState *user)
2761 {
2762 if (!user->chr) {
2763 return;
2764 }
2765 user->notifiers = (GPtrArray *) g_ptr_array_free(user->notifiers, true);
2766 user->chr = NULL;
2767 }
2768
2769
2770 typedef struct {
2771 vu_async_close_fn cb;
2772 DeviceState *dev;
2773 CharBackend *cd;
2774 struct vhost_dev *vhost;
2775 } VhostAsyncCallback;
2776
vhost_user_async_close_bh(void * opaque)2777 static void vhost_user_async_close_bh(void *opaque)
2778 {
2779 VhostAsyncCallback *data = opaque;
2780
2781 data->cb(data->dev);
2782
2783 g_free(data);
2784 }
2785
2786 /*
2787 * We only schedule the work if the machine is running. If suspended
2788 * we want to keep all the in-flight data as is for migration
2789 * purposes.
2790 */
vhost_user_async_close(DeviceState * d,CharBackend * chardev,struct vhost_dev * vhost,vu_async_close_fn cb)2791 void vhost_user_async_close(DeviceState *d,
2792 CharBackend *chardev, struct vhost_dev *vhost,
2793 vu_async_close_fn cb)
2794 {
2795 if (!runstate_check(RUN_STATE_SHUTDOWN)) {
2796 /*
2797 * A close event may happen during a read/write, but vhost
2798 * code assumes the vhost_dev remains setup, so delay the
2799 * stop & clear.
2800 */
2801 AioContext *ctx = qemu_get_current_aio_context();
2802 VhostAsyncCallback *data = g_new0(VhostAsyncCallback, 1);
2803
2804 /* Save data for the callback */
2805 data->cb = cb;
2806 data->dev = d;
2807 data->cd = chardev;
2808 data->vhost = vhost;
2809
2810 /* Disable any further notifications on the chardev */
2811 qemu_chr_fe_set_handlers(chardev,
2812 NULL, NULL, NULL, NULL, NULL, NULL,
2813 false);
2814
2815 aio_bh_schedule_oneshot(ctx, vhost_user_async_close_bh, data);
2816
2817 /*
2818 * Move vhost device to the stopped state. The vhost-user device
2819 * will be clean up and disconnected in BH. This can be useful in
2820 * the vhost migration code. If disconnect was caught there is an
2821 * option for the general vhost code to get the dev state without
2822 * knowing its type (in this case vhost-user).
2823 *
2824 * Note if the vhost device is fully cleared by the time we
2825 * execute the bottom half we won't continue with the cleanup.
2826 */
2827 vhost->started = false;
2828 }
2829 }
2830
vhost_user_dev_start(struct vhost_dev * dev,bool started)2831 static int vhost_user_dev_start(struct vhost_dev *dev, bool started)
2832 {
2833 if (!virtio_has_feature(dev->protocol_features,
2834 VHOST_USER_PROTOCOL_F_STATUS)) {
2835 return 0;
2836 }
2837
2838 /* Set device status only for last queue pair */
2839 if (dev->vq_index + dev->nvqs != dev->vq_index_end) {
2840 return 0;
2841 }
2842
2843 if (started) {
2844 return vhost_user_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE |
2845 VIRTIO_CONFIG_S_DRIVER |
2846 VIRTIO_CONFIG_S_DRIVER_OK);
2847 } else {
2848 return 0;
2849 }
2850 }
2851
vhost_user_reset_status(struct vhost_dev * dev)2852 static void vhost_user_reset_status(struct vhost_dev *dev)
2853 {
2854 /* Set device status only for last queue pair */
2855 if (dev->vq_index + dev->nvqs != dev->vq_index_end) {
2856 return;
2857 }
2858
2859 if (virtio_has_feature(dev->protocol_features,
2860 VHOST_USER_PROTOCOL_F_STATUS)) {
2861 vhost_user_set_status(dev, 0);
2862 }
2863 }
2864
vhost_user_supports_device_state(struct vhost_dev * dev)2865 static bool vhost_user_supports_device_state(struct vhost_dev *dev)
2866 {
2867 return virtio_has_feature(dev->protocol_features,
2868 VHOST_USER_PROTOCOL_F_DEVICE_STATE);
2869 }
2870
vhost_user_set_device_state_fd(struct vhost_dev * dev,VhostDeviceStateDirection direction,VhostDeviceStatePhase phase,int fd,int * reply_fd,Error ** errp)2871 static int vhost_user_set_device_state_fd(struct vhost_dev *dev,
2872 VhostDeviceStateDirection direction,
2873 VhostDeviceStatePhase phase,
2874 int fd,
2875 int *reply_fd,
2876 Error **errp)
2877 {
2878 int ret;
2879 struct vhost_user *vu = dev->opaque;
2880 VhostUserMsg msg = {
2881 .hdr = {
2882 .request = VHOST_USER_SET_DEVICE_STATE_FD,
2883 .flags = VHOST_USER_VERSION,
2884 .size = sizeof(msg.payload.transfer_state),
2885 },
2886 .payload.transfer_state = {
2887 .direction = direction,
2888 .phase = phase,
2889 },
2890 };
2891
2892 *reply_fd = -1;
2893
2894 if (!vhost_user_supports_device_state(dev)) {
2895 close(fd);
2896 error_setg(errp, "Back-end does not support migration state transfer");
2897 return -ENOTSUP;
2898 }
2899
2900 ret = vhost_user_write(dev, &msg, &fd, 1);
2901 close(fd);
2902 if (ret < 0) {
2903 error_setg_errno(errp, -ret,
2904 "Failed to send SET_DEVICE_STATE_FD message");
2905 return ret;
2906 }
2907
2908 ret = vhost_user_read(dev, &msg);
2909 if (ret < 0) {
2910 error_setg_errno(errp, -ret,
2911 "Failed to receive SET_DEVICE_STATE_FD reply");
2912 return ret;
2913 }
2914
2915 if (msg.hdr.request != VHOST_USER_SET_DEVICE_STATE_FD) {
2916 error_setg(errp,
2917 "Received unexpected message type, expected %d, received %d",
2918 VHOST_USER_SET_DEVICE_STATE_FD, msg.hdr.request);
2919 return -EPROTO;
2920 }
2921
2922 if (msg.hdr.size != sizeof(msg.payload.u64)) {
2923 error_setg(errp,
2924 "Received bad message size, expected %zu, received %" PRIu32,
2925 sizeof(msg.payload.u64), msg.hdr.size);
2926 return -EPROTO;
2927 }
2928
2929 if ((msg.payload.u64 & 0xff) != 0) {
2930 error_setg(errp, "Back-end did not accept migration state transfer");
2931 return -EIO;
2932 }
2933
2934 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK)) {
2935 *reply_fd = qemu_chr_fe_get_msgfd(vu->user->chr);
2936 if (*reply_fd < 0) {
2937 error_setg(errp,
2938 "Failed to get back-end-provided transfer pipe FD");
2939 *reply_fd = -1;
2940 return -EIO;
2941 }
2942 }
2943
2944 return 0;
2945 }
2946
vhost_user_check_device_state(struct vhost_dev * dev,Error ** errp)2947 static int vhost_user_check_device_state(struct vhost_dev *dev, Error **errp)
2948 {
2949 int ret;
2950 VhostUserMsg msg = {
2951 .hdr = {
2952 .request = VHOST_USER_CHECK_DEVICE_STATE,
2953 .flags = VHOST_USER_VERSION,
2954 .size = 0,
2955 },
2956 };
2957
2958 if (!vhost_user_supports_device_state(dev)) {
2959 error_setg(errp, "Back-end does not support migration state transfer");
2960 return -ENOTSUP;
2961 }
2962
2963 ret = vhost_user_write(dev, &msg, NULL, 0);
2964 if (ret < 0) {
2965 error_setg_errno(errp, -ret,
2966 "Failed to send CHECK_DEVICE_STATE message");
2967 return ret;
2968 }
2969
2970 ret = vhost_user_read(dev, &msg);
2971 if (ret < 0) {
2972 error_setg_errno(errp, -ret,
2973 "Failed to receive CHECK_DEVICE_STATE reply");
2974 return ret;
2975 }
2976
2977 if (msg.hdr.request != VHOST_USER_CHECK_DEVICE_STATE) {
2978 error_setg(errp,
2979 "Received unexpected message type, expected %d, received %d",
2980 VHOST_USER_CHECK_DEVICE_STATE, msg.hdr.request);
2981 return -EPROTO;
2982 }
2983
2984 if (msg.hdr.size != sizeof(msg.payload.u64)) {
2985 error_setg(errp,
2986 "Received bad message size, expected %zu, received %" PRIu32,
2987 sizeof(msg.payload.u64), msg.hdr.size);
2988 return -EPROTO;
2989 }
2990
2991 if (msg.payload.u64 != 0) {
2992 error_setg(errp, "Back-end failed to process its internal state");
2993 return -EIO;
2994 }
2995
2996 return 0;
2997 }
2998
2999 const VhostOps user_ops = {
3000 .backend_type = VHOST_BACKEND_TYPE_USER,
3001 .vhost_backend_init = vhost_user_backend_init,
3002 .vhost_backend_cleanup = vhost_user_backend_cleanup,
3003 .vhost_backend_memslots_limit = vhost_user_memslots_limit,
3004 .vhost_backend_no_private_memslots = vhost_user_no_private_memslots,
3005 .vhost_set_log_base = vhost_user_set_log_base,
3006 .vhost_set_mem_table = vhost_user_set_mem_table,
3007 .vhost_set_vring_addr = vhost_user_set_vring_addr,
3008 .vhost_set_vring_endian = vhost_user_set_vring_endian,
3009 .vhost_set_vring_num = vhost_user_set_vring_num,
3010 .vhost_set_vring_base = vhost_user_set_vring_base,
3011 .vhost_get_vring_base = vhost_user_get_vring_base,
3012 .vhost_set_vring_kick = vhost_user_set_vring_kick,
3013 .vhost_set_vring_call = vhost_user_set_vring_call,
3014 .vhost_set_vring_err = vhost_user_set_vring_err,
3015 .vhost_set_features = vhost_user_set_features,
3016 .vhost_get_features = vhost_user_get_features,
3017 .vhost_set_owner = vhost_user_set_owner,
3018 .vhost_reset_device = vhost_user_reset_device,
3019 .vhost_get_vq_index = vhost_user_get_vq_index,
3020 .vhost_set_vring_enable = vhost_user_set_vring_enable,
3021 .vhost_requires_shm_log = vhost_user_requires_shm_log,
3022 .vhost_migration_done = vhost_user_migration_done,
3023 .vhost_net_set_mtu = vhost_user_net_set_mtu,
3024 .vhost_set_iotlb_callback = vhost_user_set_iotlb_callback,
3025 .vhost_send_device_iotlb_msg = vhost_user_send_device_iotlb_msg,
3026 .vhost_get_config = vhost_user_get_config,
3027 .vhost_set_config = vhost_user_set_config,
3028 .vhost_crypto_create_session = vhost_user_crypto_create_session,
3029 .vhost_crypto_close_session = vhost_user_crypto_close_session,
3030 .vhost_get_inflight_fd = vhost_user_get_inflight_fd,
3031 .vhost_set_inflight_fd = vhost_user_set_inflight_fd,
3032 .vhost_dev_start = vhost_user_dev_start,
3033 .vhost_reset_status = vhost_user_reset_status,
3034 .vhost_supports_device_state = vhost_user_supports_device_state,
3035 .vhost_set_device_state_fd = vhost_user_set_device_state_fd,
3036 .vhost_check_device_state = vhost_user_check_device_state,
3037 };
3038