xref: /dpdk/lib/vhost/vhost_user.c (revision 30a1de105a5f40d77b344a891c4a68f79e815c43)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2018 Intel Corporation
3  */
4 
5 /* Security model
6  * --------------
7  * The vhost-user protocol connection is an external interface, so it must be
8  * robust against invalid inputs.
9  *
10  * This is important because the vhost-user master is only one step removed
11  * from the guest.  Malicious guests that have escaped will then launch further
12  * attacks from the vhost-user master.
13  *
14  * Even in deployments where guests are trusted, a bug in the vhost-user master
15  * can still cause invalid messages to be sent.  Such messages must not
16  * compromise the stability of the DPDK application by causing crashes, memory
17  * corruption, or other problematic behavior.
18  *
19  * Do not assume received VhostUserMsg fields contain sensible values!
20  */
21 
22 #include <stdint.h>
23 #include <stdio.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <unistd.h>
27 #include <fcntl.h>
28 #include <sys/ioctl.h>
29 #include <sys/mman.h>
30 #include <sys/stat.h>
31 #include <sys/syscall.h>
32 #ifdef RTE_LIBRTE_VHOST_NUMA
33 #include <numaif.h>
34 #endif
35 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
36 #include <linux/userfaultfd.h>
37 #endif
38 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
39 #include <linux/memfd.h>
40 #define MEMFD_SUPPORTED
41 #endif
42 
43 #include <rte_common.h>
44 #include <rte_malloc.h>
45 #include <rte_log.h>
46 #include <rte_vfio.h>
47 #include <rte_errno.h>
48 
49 #include "iotlb.h"
50 #include "vhost.h"
51 #include "vhost_user.h"
52 
53 #define VIRTIO_MIN_MTU 68
54 #define VIRTIO_MAX_MTU 65535
55 
56 #define INFLIGHT_ALIGNMENT	64
57 #define INFLIGHT_VERSION	0x1
58 
59 static const char *vhost_message_str[VHOST_USER_MAX] = {
60 	[VHOST_USER_NONE] = "VHOST_USER_NONE",
61 	[VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
62 	[VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
63 	[VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
64 	[VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
65 	[VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
66 	[VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
67 	[VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
68 	[VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
69 	[VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
70 	[VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
71 	[VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
72 	[VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
73 	[VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
74 	[VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
75 	[VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
76 	[VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
77 	[VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
78 	[VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
79 	[VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
80 	[VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
81 	[VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
82 	[VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
83 	[VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
84 	[VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
85 	[VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
86 	[VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
87 	[VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
88 	[VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
89 	[VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
90 	[VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
91 	[VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
92 };
93 
94 static int send_vhost_reply(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx);
95 static int read_vhost_message(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx);
96 
97 static void
98 close_msg_fds(struct vhu_msg_context *ctx)
99 {
100 	int i;
101 
102 	for (i = 0; i < ctx->fd_num; i++) {
103 		int fd = ctx->fds[i];
104 
105 		if (fd == -1)
106 			continue;
107 
108 		ctx->fds[i] = -1;
109 		close(fd);
110 	}
111 }
112 
113 /*
114  * Ensure the expected number of FDs is received,
115  * close all FDs and return an error if this is not the case.
116  */
117 static int
118 validate_msg_fds(struct virtio_net *dev, struct vhu_msg_context *ctx, int expected_fds)
119 {
120 	if (ctx->fd_num == expected_fds)
121 		return 0;
122 
123 	VHOST_LOG_CONFIG(ERR, "(%s) expect %d FDs for request %s, received %d\n",
124 		dev->ifname, expected_fds,
125 		vhost_message_str[ctx->msg.request.master],
126 		ctx->fd_num);
127 
128 	close_msg_fds(ctx);
129 
130 	return -1;
131 }
132 
133 static uint64_t
134 get_blk_size(int fd)
135 {
136 	struct stat stat;
137 	int ret;
138 
139 	ret = fstat(fd, &stat);
140 	return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
141 }
142 
143 static int
144 async_dma_map(struct virtio_net *dev, struct rte_vhost_mem_region *region, bool do_map)
145 {
146 	uint64_t host_iova;
147 	int ret = 0;
148 
149 	host_iova = rte_mem_virt2iova((void *)(uintptr_t)region->host_user_addr);
150 	if (do_map) {
151 		/* Add mapped region into the default container of DPDK. */
152 		ret = rte_vfio_container_dma_map(RTE_VFIO_DEFAULT_CONTAINER_FD,
153 						 region->host_user_addr,
154 						 host_iova,
155 						 region->size);
156 		if (ret) {
157 			/*
158 			 * DMA device may bind with kernel driver, in this case,
159 			 * we don't need to program IOMMU manually. However, if no
160 			 * device is bound with vfio/uio in DPDK, and vfio kernel
161 			 * module is loaded, the API will still be called and return
162 			 * with ENODEV/ENOSUP.
163 			 *
164 			 * DPDK vfio only returns ENODEV/ENOSUP in very similar
165 			 * situations(vfio either unsupported, or supported
166 			 * but no devices found). Either way, no mappings could be
167 			 * performed. We treat it as normal case in async path.
168 			 */
169 			if (rte_errno == ENODEV || rte_errno == ENOTSUP)
170 				return 0;
171 
172 			VHOST_LOG_CONFIG(ERR, "(%s) DMA engine map failed\n", dev->ifname);
173 			/* DMA mapping errors won't stop VHST_USER_SET_MEM_TABLE. */
174 			return 0;
175 		}
176 
177 	} else {
178 		/* Remove mapped region from the default container of DPDK. */
179 		ret = rte_vfio_container_dma_unmap(RTE_VFIO_DEFAULT_CONTAINER_FD,
180 						   region->host_user_addr,
181 						   host_iova,
182 						   region->size);
183 		if (ret) {
184 			/* like DMA map, ignore the kernel driver case when unmap. */
185 			if (rte_errno == EINVAL)
186 				return 0;
187 
188 			VHOST_LOG_CONFIG(ERR, "(%s) DMA engine unmap failed\n", dev->ifname);
189 			return ret;
190 		}
191 	}
192 
193 	return ret;
194 }
195 
196 static void
197 free_mem_region(struct virtio_net *dev)
198 {
199 	uint32_t i;
200 	struct rte_vhost_mem_region *reg;
201 
202 	if (!dev || !dev->mem)
203 		return;
204 
205 	for (i = 0; i < dev->mem->nregions; i++) {
206 		reg = &dev->mem->regions[i];
207 		if (reg->host_user_addr) {
208 			if (dev->async_copy && rte_vfio_is_enabled("vfio"))
209 				async_dma_map(dev, reg, false);
210 
211 			munmap(reg->mmap_addr, reg->mmap_size);
212 			close(reg->fd);
213 		}
214 	}
215 }
216 
217 void
218 vhost_backend_cleanup(struct virtio_net *dev)
219 {
220 	struct rte_vdpa_device *vdpa_dev;
221 
222 	vdpa_dev = dev->vdpa_dev;
223 	if (vdpa_dev && vdpa_dev->ops->dev_cleanup != NULL)
224 		vdpa_dev->ops->dev_cleanup(dev->vid);
225 
226 	if (dev->mem) {
227 		free_mem_region(dev);
228 		rte_free(dev->mem);
229 		dev->mem = NULL;
230 	}
231 
232 	rte_free(dev->guest_pages);
233 	dev->guest_pages = NULL;
234 
235 	if (dev->log_addr) {
236 		munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
237 		dev->log_addr = 0;
238 	}
239 
240 	if (dev->inflight_info) {
241 		if (dev->inflight_info->addr) {
242 			munmap(dev->inflight_info->addr,
243 			       dev->inflight_info->size);
244 			dev->inflight_info->addr = NULL;
245 		}
246 
247 		if (dev->inflight_info->fd >= 0) {
248 			close(dev->inflight_info->fd);
249 			dev->inflight_info->fd = -1;
250 		}
251 
252 		rte_free(dev->inflight_info);
253 		dev->inflight_info = NULL;
254 	}
255 
256 	if (dev->slave_req_fd >= 0) {
257 		close(dev->slave_req_fd);
258 		dev->slave_req_fd = -1;
259 	}
260 
261 	if (dev->postcopy_ufd >= 0) {
262 		close(dev->postcopy_ufd);
263 		dev->postcopy_ufd = -1;
264 	}
265 
266 	dev->postcopy_listening = 0;
267 }
268 
269 static void
270 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
271 			      int enable)
272 {
273 	struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
274 	struct vhost_virtqueue *vq = dev->virtqueue[index];
275 
276 	/* Configure guest notifications on enable */
277 	if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
278 		vhost_enable_guest_notification(dev, vq, vq->notif_enable);
279 
280 	if (vdpa_dev && vdpa_dev->ops->set_vring_state)
281 		vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
282 
283 	if (dev->notify_ops->vring_state_changed)
284 		dev->notify_ops->vring_state_changed(dev->vid,
285 				index, enable);
286 }
287 
288 /*
289  * This function just returns success at the moment unless
290  * the device hasn't been initialised.
291  */
292 static int
293 vhost_user_set_owner(struct virtio_net **pdev,
294 			struct vhu_msg_context *ctx,
295 			int main_fd __rte_unused)
296 {
297 	struct virtio_net *dev = *pdev;
298 
299 	if (validate_msg_fds(dev, ctx, 0) != 0)
300 		return RTE_VHOST_MSG_RESULT_ERR;
301 
302 	return RTE_VHOST_MSG_RESULT_OK;
303 }
304 
305 static int
306 vhost_user_reset_owner(struct virtio_net **pdev,
307 			struct vhu_msg_context *ctx,
308 			int main_fd __rte_unused)
309 {
310 	struct virtio_net *dev = *pdev;
311 
312 	if (validate_msg_fds(dev, ctx, 0) != 0)
313 		return RTE_VHOST_MSG_RESULT_ERR;
314 
315 	vhost_destroy_device_notify(dev);
316 
317 	cleanup_device(dev, 0);
318 	reset_device(dev);
319 	return RTE_VHOST_MSG_RESULT_OK;
320 }
321 
322 /*
323  * The features that we support are requested.
324  */
325 static int
326 vhost_user_get_features(struct virtio_net **pdev,
327 			struct vhu_msg_context *ctx,
328 			int main_fd __rte_unused)
329 {
330 	struct virtio_net *dev = *pdev;
331 	uint64_t features = 0;
332 
333 	if (validate_msg_fds(dev, ctx, 0) != 0)
334 		return RTE_VHOST_MSG_RESULT_ERR;
335 
336 	rte_vhost_driver_get_features(dev->ifname, &features);
337 
338 	ctx->msg.payload.u64 = features;
339 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
340 	ctx->fd_num = 0;
341 
342 	return RTE_VHOST_MSG_RESULT_REPLY;
343 }
344 
345 /*
346  * The queue number that we support are requested.
347  */
348 static int
349 vhost_user_get_queue_num(struct virtio_net **pdev,
350 			struct vhu_msg_context *ctx,
351 			int main_fd __rte_unused)
352 {
353 	struct virtio_net *dev = *pdev;
354 	uint32_t queue_num = 0;
355 
356 	if (validate_msg_fds(dev, ctx, 0) != 0)
357 		return RTE_VHOST_MSG_RESULT_ERR;
358 
359 	rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
360 
361 	ctx->msg.payload.u64 = (uint64_t)queue_num;
362 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
363 	ctx->fd_num = 0;
364 
365 	return RTE_VHOST_MSG_RESULT_REPLY;
366 }
367 
368 /*
369  * We receive the negotiated features supported by us and the virtio device.
370  */
371 static int
372 vhost_user_set_features(struct virtio_net **pdev,
373 			struct vhu_msg_context *ctx,
374 			int main_fd __rte_unused)
375 {
376 	struct virtio_net *dev = *pdev;
377 	uint64_t features = ctx->msg.payload.u64;
378 	uint64_t vhost_features = 0;
379 	struct rte_vdpa_device *vdpa_dev;
380 
381 	if (validate_msg_fds(dev, ctx, 0) != 0)
382 		return RTE_VHOST_MSG_RESULT_ERR;
383 
384 	rte_vhost_driver_get_features(dev->ifname, &vhost_features);
385 	if (features & ~vhost_features) {
386 		VHOST_LOG_CONFIG(ERR, "(%s) received invalid negotiated features.\n",
387 			dev->ifname);
388 		dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
389 		dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
390 
391 		return RTE_VHOST_MSG_RESULT_ERR;
392 	}
393 
394 	if (dev->flags & VIRTIO_DEV_RUNNING) {
395 		if (dev->features == features)
396 			return RTE_VHOST_MSG_RESULT_OK;
397 
398 		/*
399 		 * Error out if master tries to change features while device is
400 		 * in running state. The exception being VHOST_F_LOG_ALL, which
401 		 * is enabled when the live-migration starts.
402 		 */
403 		if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
404 			VHOST_LOG_CONFIG(ERR, "(%s) features changed while device is running.\n",
405 				dev->ifname);
406 			return RTE_VHOST_MSG_RESULT_ERR;
407 		}
408 
409 		if (dev->notify_ops->features_changed)
410 			dev->notify_ops->features_changed(dev->vid, features);
411 	}
412 
413 	dev->features = features;
414 	if (dev->features &
415 		((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
416 		 (1ULL << VIRTIO_F_VERSION_1) |
417 		 (1ULL << VIRTIO_F_RING_PACKED))) {
418 		dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
419 	} else {
420 		dev->vhost_hlen = sizeof(struct virtio_net_hdr);
421 	}
422 	VHOST_LOG_CONFIG(INFO, "(%s) negotiated Virtio features: 0x%" PRIx64 "\n",
423 			dev->ifname, dev->features);
424 	VHOST_LOG_CONFIG(DEBUG, "(%s) mergeable RX buffers %s, virtio 1 %s\n",
425 		dev->ifname,
426 		(dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
427 		(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
428 
429 	if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
430 	    !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
431 		/*
432 		 * Remove all but first queue pair if MQ hasn't been
433 		 * negotiated. This is safe because the device is not
434 		 * running at this stage.
435 		 */
436 		while (dev->nr_vring > 2) {
437 			struct vhost_virtqueue *vq;
438 
439 			vq = dev->virtqueue[--dev->nr_vring];
440 			if (!vq)
441 				continue;
442 
443 			dev->virtqueue[dev->nr_vring] = NULL;
444 			cleanup_vq(vq, 1);
445 			cleanup_vq_inflight(dev, vq);
446 			free_vq(dev, vq);
447 		}
448 	}
449 
450 	vdpa_dev = dev->vdpa_dev;
451 	if (vdpa_dev)
452 		vdpa_dev->ops->set_features(dev->vid);
453 
454 	dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
455 	return RTE_VHOST_MSG_RESULT_OK;
456 }
457 
458 /*
459  * The virtio device sends us the size of the descriptor ring.
460  */
461 static int
462 vhost_user_set_vring_num(struct virtio_net **pdev,
463 			struct vhu_msg_context *ctx,
464 			int main_fd __rte_unused)
465 {
466 	struct virtio_net *dev = *pdev;
467 	struct vhost_virtqueue *vq = dev->virtqueue[ctx->msg.payload.state.index];
468 
469 	if (validate_msg_fds(dev, ctx, 0) != 0)
470 		return RTE_VHOST_MSG_RESULT_ERR;
471 
472 	if (ctx->msg.payload.state.num > 32768) {
473 		VHOST_LOG_CONFIG(ERR, "(%s) invalid virtqueue size %u\n",
474 				dev->ifname, ctx->msg.payload.state.num);
475 		return RTE_VHOST_MSG_RESULT_ERR;
476 	}
477 
478 	vq->size = ctx->msg.payload.state.num;
479 
480 	/* VIRTIO 1.0, 2.4 Virtqueues says:
481 	 *
482 	 *   Queue Size value is always a power of 2. The maximum Queue Size
483 	 *   value is 32768.
484 	 *
485 	 * VIRTIO 1.1 2.7 Virtqueues says:
486 	 *
487 	 *   Packed virtqueues support up to 2^15 entries each.
488 	 */
489 	if (!vq_is_packed(dev)) {
490 		if (vq->size & (vq->size - 1)) {
491 			VHOST_LOG_CONFIG(ERR, "(%s) invalid virtqueue size %u\n",
492 					dev->ifname, vq->size);
493 			return RTE_VHOST_MSG_RESULT_ERR;
494 		}
495 	}
496 
497 	if (vq_is_packed(dev)) {
498 		rte_free(vq->shadow_used_packed);
499 		vq->shadow_used_packed = rte_malloc_socket(NULL,
500 				vq->size *
501 				sizeof(struct vring_used_elem_packed),
502 				RTE_CACHE_LINE_SIZE, vq->numa_node);
503 		if (!vq->shadow_used_packed) {
504 			VHOST_LOG_CONFIG(ERR,
505 				"(%s) failed to allocate memory for shadow used ring.\n",
506 				dev->ifname);
507 			return RTE_VHOST_MSG_RESULT_ERR;
508 		}
509 
510 	} else {
511 		rte_free(vq->shadow_used_split);
512 
513 		vq->shadow_used_split = rte_malloc_socket(NULL,
514 				vq->size * sizeof(struct vring_used_elem),
515 				RTE_CACHE_LINE_SIZE, vq->numa_node);
516 
517 		if (!vq->shadow_used_split) {
518 			VHOST_LOG_CONFIG(ERR,
519 				"(%s) failed to allocate memory for vq internal data.\n",
520 				dev->ifname);
521 			return RTE_VHOST_MSG_RESULT_ERR;
522 		}
523 	}
524 
525 	rte_free(vq->batch_copy_elems);
526 	vq->batch_copy_elems = rte_malloc_socket(NULL,
527 				vq->size * sizeof(struct batch_copy_elem),
528 				RTE_CACHE_LINE_SIZE, vq->numa_node);
529 	if (!vq->batch_copy_elems) {
530 		VHOST_LOG_CONFIG(ERR, "(%s) failed to allocate memory for batching copy.\n",
531 			dev->ifname);
532 		return RTE_VHOST_MSG_RESULT_ERR;
533 	}
534 
535 	return RTE_VHOST_MSG_RESULT_OK;
536 }
537 
538 /*
539  * Reallocate virtio_dev, vhost_virtqueue and related data structures to
540  * make them on the same numa node as the memory of vring descriptor.
541  */
542 #ifdef RTE_LIBRTE_VHOST_NUMA
543 static struct virtio_net*
544 numa_realloc(struct virtio_net *dev, int index)
545 {
546 	int node, dev_node;
547 	struct virtio_net *old_dev;
548 	struct vhost_virtqueue *vq;
549 	struct batch_copy_elem *bce;
550 	struct guest_page *gp;
551 	struct rte_vhost_memory *mem;
552 	size_t mem_size;
553 	int ret;
554 
555 	old_dev = dev;
556 	vq = dev->virtqueue[index];
557 
558 	/*
559 	 * If VQ is ready, it is too late to reallocate, it certainly already
560 	 * happened anyway on VHOST_USER_SET_VRING_ADRR.
561 	 */
562 	if (vq->ready)
563 		return dev;
564 
565 	ret = get_mempolicy(&node, NULL, 0, vq->desc, MPOL_F_NODE | MPOL_F_ADDR);
566 	if (ret) {
567 		VHOST_LOG_CONFIG(ERR, "(%s) unable to get virtqueue %d numa information.\n",
568 				dev->ifname, index);
569 		return dev;
570 	}
571 
572 	if (node == vq->numa_node)
573 		goto out_dev_realloc;
574 
575 	vq = rte_realloc_socket(vq, sizeof(*vq), 0, node);
576 	if (!vq) {
577 		VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc virtqueue %d on node %d\n",
578 				dev->ifname, index, node);
579 		return dev;
580 	}
581 
582 	if (vq != dev->virtqueue[index]) {
583 		VHOST_LOG_CONFIG(INFO, "(%s) reallocated virtqueue on node %d\n",
584 				dev->ifname, node);
585 		dev->virtqueue[index] = vq;
586 		vhost_user_iotlb_init(dev, index);
587 	}
588 
589 	if (vq_is_packed(dev)) {
590 		struct vring_used_elem_packed *sup;
591 
592 		sup = rte_realloc_socket(vq->shadow_used_packed, vq->size * sizeof(*sup),
593 				RTE_CACHE_LINE_SIZE, node);
594 		if (!sup) {
595 			VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc shadow packed on node %d\n",
596 					dev->ifname, node);
597 			return dev;
598 		}
599 		vq->shadow_used_packed = sup;
600 	} else {
601 		struct vring_used_elem *sus;
602 
603 		sus = rte_realloc_socket(vq->shadow_used_split, vq->size * sizeof(*sus),
604 				RTE_CACHE_LINE_SIZE, node);
605 		if (!sus) {
606 			VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc shadow split on node %d\n",
607 					dev->ifname, node);
608 			return dev;
609 		}
610 		vq->shadow_used_split = sus;
611 	}
612 
613 	bce = rte_realloc_socket(vq->batch_copy_elems, vq->size * sizeof(*bce),
614 			RTE_CACHE_LINE_SIZE, node);
615 	if (!bce) {
616 		VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc batch copy elem on node %d\n",
617 				dev->ifname, node);
618 		return dev;
619 	}
620 	vq->batch_copy_elems = bce;
621 
622 	if (vq->log_cache) {
623 		struct log_cache_entry *lc;
624 
625 		lc = rte_realloc_socket(vq->log_cache, sizeof(*lc) * VHOST_LOG_CACHE_NR, 0, node);
626 		if (!lc) {
627 			VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc log cache on node %d\n",
628 					dev->ifname, node);
629 			return dev;
630 		}
631 		vq->log_cache = lc;
632 	}
633 
634 	if (vq->resubmit_inflight) {
635 		struct rte_vhost_resubmit_info *ri;
636 
637 		ri = rte_realloc_socket(vq->resubmit_inflight, sizeof(*ri), 0, node);
638 		if (!ri) {
639 			VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc resubmit inflight on node %d\n",
640 					dev->ifname, node);
641 			return dev;
642 		}
643 		vq->resubmit_inflight = ri;
644 
645 		if (ri->resubmit_list) {
646 			struct rte_vhost_resubmit_desc *rd;
647 
648 			rd = rte_realloc_socket(ri->resubmit_list, sizeof(*rd) * ri->resubmit_num,
649 					0, node);
650 			if (!rd) {
651 				VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc resubmit list on node %d\n",
652 						dev->ifname, node);
653 				return dev;
654 			}
655 			ri->resubmit_list = rd;
656 		}
657 	}
658 
659 	vq->numa_node = node;
660 
661 out_dev_realloc:
662 
663 	if (dev->flags & VIRTIO_DEV_RUNNING)
664 		return dev;
665 
666 	ret = get_mempolicy(&dev_node, NULL, 0, dev, MPOL_F_NODE | MPOL_F_ADDR);
667 	if (ret) {
668 		VHOST_LOG_CONFIG(ERR, "(%s) unable to get numa information.\n", dev->ifname);
669 		return dev;
670 	}
671 
672 	if (dev_node == node)
673 		return dev;
674 
675 	dev = rte_realloc_socket(old_dev, sizeof(*dev), 0, node);
676 	if (!dev) {
677 		VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc dev on node %d\n",
678 				old_dev->ifname, node);
679 		return old_dev;
680 	}
681 
682 	VHOST_LOG_CONFIG(INFO, "(%s) reallocated device on node %d\n", dev->ifname, node);
683 	vhost_devices[dev->vid] = dev;
684 
685 	mem_size = sizeof(struct rte_vhost_memory) +
686 		sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
687 	mem = rte_realloc_socket(dev->mem, mem_size, 0, node);
688 	if (!mem) {
689 		VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc mem table on node %d\n",
690 				dev->ifname, node);
691 		return dev;
692 	}
693 	dev->mem = mem;
694 
695 	gp = rte_realloc_socket(dev->guest_pages, dev->max_guest_pages * sizeof(*gp),
696 			RTE_CACHE_LINE_SIZE, node);
697 	if (!gp) {
698 		VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc guest pages on node %d\n",
699 				dev->ifname, node);
700 		return dev;
701 	}
702 	dev->guest_pages = gp;
703 
704 	return dev;
705 }
706 #else
707 static struct virtio_net*
708 numa_realloc(struct virtio_net *dev, int index __rte_unused)
709 {
710 	return dev;
711 }
712 #endif
713 
714 /* Converts QEMU virtual address to Vhost virtual address. */
715 static uint64_t
716 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
717 {
718 	struct rte_vhost_mem_region *r;
719 	uint32_t i;
720 
721 	if (unlikely(!dev || !dev->mem))
722 		goto out_error;
723 
724 	/* Find the region where the address lives. */
725 	for (i = 0; i < dev->mem->nregions; i++) {
726 		r = &dev->mem->regions[i];
727 
728 		if (qva >= r->guest_user_addr &&
729 		    qva <  r->guest_user_addr + r->size) {
730 
731 			if (unlikely(*len > r->guest_user_addr + r->size - qva))
732 				*len = r->guest_user_addr + r->size - qva;
733 
734 			return qva - r->guest_user_addr +
735 			       r->host_user_addr;
736 		}
737 	}
738 out_error:
739 	*len = 0;
740 
741 	return 0;
742 }
743 
744 
745 /*
746  * Converts ring address to Vhost virtual address.
747  * If IOMMU is enabled, the ring address is a guest IO virtual address,
748  * else it is a QEMU virtual address.
749  */
750 static uint64_t
751 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
752 		uint64_t ra, uint64_t *size)
753 {
754 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
755 		uint64_t vva;
756 
757 		vhost_user_iotlb_rd_lock(vq);
758 		vva = vhost_iova_to_vva(dev, vq, ra,
759 					size, VHOST_ACCESS_RW);
760 		vhost_user_iotlb_rd_unlock(vq);
761 
762 		return vva;
763 	}
764 
765 	return qva_to_vva(dev, ra, size);
766 }
767 
768 static uint64_t
769 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
770 {
771 	uint64_t log_gpa;
772 
773 	vhost_user_iotlb_rd_lock(vq);
774 	log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
775 	vhost_user_iotlb_rd_unlock(vq);
776 
777 	return log_gpa;
778 }
779 
780 static struct virtio_net *
781 translate_ring_addresses(struct virtio_net *dev, int vq_index)
782 {
783 	struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
784 	struct vhost_vring_addr *addr = &vq->ring_addrs;
785 	uint64_t len, expected_len;
786 
787 	if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
788 		vq->log_guest_addr =
789 			log_addr_to_gpa(dev, vq);
790 		if (vq->log_guest_addr == 0) {
791 			VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map log_guest_addr.\n",
792 				dev->ifname);
793 			return dev;
794 		}
795 	}
796 
797 	if (vq_is_packed(dev)) {
798 		len = sizeof(struct vring_packed_desc) * vq->size;
799 		vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
800 			ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
801 		if (vq->desc_packed == NULL ||
802 				len != sizeof(struct vring_packed_desc) *
803 				vq->size) {
804 			VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map desc_packed ring.\n",
805 				dev->ifname);
806 			return dev;
807 		}
808 
809 		dev = numa_realloc(dev, vq_index);
810 		vq = dev->virtqueue[vq_index];
811 		addr = &vq->ring_addrs;
812 
813 		len = sizeof(struct vring_packed_desc_event);
814 		vq->driver_event = (struct vring_packed_desc_event *)
815 					(uintptr_t)ring_addr_to_vva(dev,
816 					vq, addr->avail_user_addr, &len);
817 		if (vq->driver_event == NULL ||
818 				len != sizeof(struct vring_packed_desc_event)) {
819 			VHOST_LOG_CONFIG(DEBUG, "(%s) failed to find driver area address.\n",
820 				dev->ifname);
821 			return dev;
822 		}
823 
824 		len = sizeof(struct vring_packed_desc_event);
825 		vq->device_event = (struct vring_packed_desc_event *)
826 					(uintptr_t)ring_addr_to_vva(dev,
827 					vq, addr->used_user_addr, &len);
828 		if (vq->device_event == NULL ||
829 				len != sizeof(struct vring_packed_desc_event)) {
830 			VHOST_LOG_CONFIG(DEBUG, "(%s) failed to find device area address.\n",
831 				dev->ifname);
832 			return dev;
833 		}
834 
835 		vq->access_ok = true;
836 		return dev;
837 	}
838 
839 	/* The addresses are converted from QEMU virtual to Vhost virtual. */
840 	if (vq->desc && vq->avail && vq->used)
841 		return dev;
842 
843 	len = sizeof(struct vring_desc) * vq->size;
844 	vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
845 			vq, addr->desc_user_addr, &len);
846 	if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
847 		VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map desc ring.\n", dev->ifname);
848 		return dev;
849 	}
850 
851 	dev = numa_realloc(dev, vq_index);
852 	vq = dev->virtqueue[vq_index];
853 	addr = &vq->ring_addrs;
854 
855 	len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
856 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
857 		len += sizeof(uint16_t);
858 	expected_len = len;
859 	vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
860 			vq, addr->avail_user_addr, &len);
861 	if (vq->avail == 0 || len != expected_len) {
862 		VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map avail ring.\n", dev->ifname);
863 		return dev;
864 	}
865 
866 	len = sizeof(struct vring_used) +
867 		sizeof(struct vring_used_elem) * vq->size;
868 	if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
869 		len += sizeof(uint16_t);
870 	expected_len = len;
871 	vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
872 			vq, addr->used_user_addr, &len);
873 	if (vq->used == 0 || len != expected_len) {
874 		VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map used ring.\n", dev->ifname);
875 		return dev;
876 	}
877 
878 	if (vq->last_used_idx != vq->used->idx) {
879 		VHOST_LOG_CONFIG(WARNING, "(%s) last_used_idx (%u) and vq->used->idx (%u) mismatches;\n",
880 			dev->ifname,
881 			vq->last_used_idx, vq->used->idx);
882 		vq->last_used_idx  = vq->used->idx;
883 		vq->last_avail_idx = vq->used->idx;
884 		VHOST_LOG_CONFIG(WARNING, "(%s) some packets maybe resent for Tx and dropped for Rx\n",
885 			dev->ifname);
886 	}
887 
888 	vq->access_ok = true;
889 
890 	VHOST_LOG_CONFIG(DEBUG, "(%s) mapped address desc: %p\n", dev->ifname, vq->desc);
891 	VHOST_LOG_CONFIG(DEBUG, "(%s) mapped address avail: %p\n", dev->ifname, vq->avail);
892 	VHOST_LOG_CONFIG(DEBUG, "(%s) mapped address used: %p\n", dev->ifname, vq->used);
893 	VHOST_LOG_CONFIG(DEBUG, "(%s) log_guest_addr: %" PRIx64 "\n",
894 			dev->ifname, vq->log_guest_addr);
895 
896 	return dev;
897 }
898 
899 /*
900  * The virtio device sends us the desc, used and avail ring addresses.
901  * This function then converts these to our address space.
902  */
903 static int
904 vhost_user_set_vring_addr(struct virtio_net **pdev,
905 			struct vhu_msg_context *ctx,
906 			int main_fd __rte_unused)
907 {
908 	struct virtio_net *dev = *pdev;
909 	struct vhost_virtqueue *vq;
910 	struct vhost_vring_addr *addr = &ctx->msg.payload.addr;
911 	bool access_ok;
912 
913 	if (validate_msg_fds(dev, ctx, 0) != 0)
914 		return RTE_VHOST_MSG_RESULT_ERR;
915 
916 	if (dev->mem == NULL)
917 		return RTE_VHOST_MSG_RESULT_ERR;
918 
919 	/* addr->index refers to the queue index. The txq 1, rxq is 0. */
920 	vq = dev->virtqueue[ctx->msg.payload.addr.index];
921 
922 	access_ok = vq->access_ok;
923 
924 	/*
925 	 * Rings addresses should not be interpreted as long as the ring is not
926 	 * started and enabled
927 	 */
928 	memcpy(&vq->ring_addrs, addr, sizeof(*addr));
929 
930 	vring_invalidate(dev, vq);
931 
932 	if ((vq->enabled && (dev->features &
933 				(1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
934 			access_ok) {
935 		dev = translate_ring_addresses(dev, ctx->msg.payload.addr.index);
936 		if (!dev)
937 			return RTE_VHOST_MSG_RESULT_ERR;
938 
939 		*pdev = dev;
940 	}
941 
942 	return RTE_VHOST_MSG_RESULT_OK;
943 }
944 
945 /*
946  * The virtio device sends us the available ring last used index.
947  */
948 static int
949 vhost_user_set_vring_base(struct virtio_net **pdev,
950 			struct vhu_msg_context *ctx,
951 			int main_fd __rte_unused)
952 {
953 	struct virtio_net *dev = *pdev;
954 	struct vhost_virtqueue *vq = dev->virtqueue[ctx->msg.payload.state.index];
955 	uint64_t val = ctx->msg.payload.state.num;
956 
957 	if (validate_msg_fds(dev, ctx, 0) != 0)
958 		return RTE_VHOST_MSG_RESULT_ERR;
959 
960 	if (vq_is_packed(dev)) {
961 		/*
962 		 * Bit[0:14]: avail index
963 		 * Bit[15]: avail wrap counter
964 		 */
965 		vq->last_avail_idx = val & 0x7fff;
966 		vq->avail_wrap_counter = !!(val & (0x1 << 15));
967 		/*
968 		 * Set used index to same value as available one, as
969 		 * their values should be the same since ring processing
970 		 * was stopped at get time.
971 		 */
972 		vq->last_used_idx = vq->last_avail_idx;
973 		vq->used_wrap_counter = vq->avail_wrap_counter;
974 	} else {
975 		vq->last_used_idx = ctx->msg.payload.state.num;
976 		vq->last_avail_idx = ctx->msg.payload.state.num;
977 	}
978 
979 	VHOST_LOG_CONFIG(INFO,
980 		"(%s) vring base idx:%u last_used_idx:%u last_avail_idx:%u.\n",
981 		dev->ifname, ctx->msg.payload.state.index, vq->last_used_idx,
982 		vq->last_avail_idx);
983 
984 	return RTE_VHOST_MSG_RESULT_OK;
985 }
986 
987 static int
988 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
989 		   uint64_t host_phys_addr, uint64_t size)
990 {
991 	struct guest_page *page, *last_page;
992 	struct guest_page *old_pages;
993 
994 	if (dev->nr_guest_pages == dev->max_guest_pages) {
995 		dev->max_guest_pages *= 2;
996 		old_pages = dev->guest_pages;
997 		dev->guest_pages = rte_realloc(dev->guest_pages,
998 					dev->max_guest_pages * sizeof(*page),
999 					RTE_CACHE_LINE_SIZE);
1000 		if (dev->guest_pages == NULL) {
1001 			VHOST_LOG_CONFIG(ERR, "(%s) cannot realloc guest_pages\n", dev->ifname);
1002 			rte_free(old_pages);
1003 			return -1;
1004 		}
1005 	}
1006 
1007 	if (dev->nr_guest_pages > 0) {
1008 		last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
1009 		/* merge if the two pages are continuous */
1010 		if (host_phys_addr == last_page->host_phys_addr +
1011 				      last_page->size) {
1012 			last_page->size += size;
1013 			return 0;
1014 		}
1015 	}
1016 
1017 	page = &dev->guest_pages[dev->nr_guest_pages++];
1018 	page->guest_phys_addr = guest_phys_addr;
1019 	page->host_phys_addr  = host_phys_addr;
1020 	page->size = size;
1021 
1022 	return 0;
1023 }
1024 
1025 static int
1026 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
1027 		uint64_t page_size)
1028 {
1029 	uint64_t reg_size = reg->size;
1030 	uint64_t host_user_addr  = reg->host_user_addr;
1031 	uint64_t guest_phys_addr = reg->guest_phys_addr;
1032 	uint64_t host_phys_addr;
1033 	uint64_t size;
1034 
1035 	host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
1036 	size = page_size - (guest_phys_addr & (page_size - 1));
1037 	size = RTE_MIN(size, reg_size);
1038 
1039 	if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
1040 		return -1;
1041 
1042 	host_user_addr  += size;
1043 	guest_phys_addr += size;
1044 	reg_size -= size;
1045 
1046 	while (reg_size > 0) {
1047 		size = RTE_MIN(reg_size, page_size);
1048 		host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
1049 						  host_user_addr);
1050 		if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
1051 				size) < 0)
1052 			return -1;
1053 
1054 		host_user_addr  += size;
1055 		guest_phys_addr += size;
1056 		reg_size -= size;
1057 	}
1058 
1059 	/* sort guest page array if over binary search threshold */
1060 	if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
1061 		qsort((void *)dev->guest_pages, dev->nr_guest_pages,
1062 			sizeof(struct guest_page), guest_page_addrcmp);
1063 	}
1064 
1065 	return 0;
1066 }
1067 
1068 #ifdef RTE_LIBRTE_VHOST_DEBUG
1069 /* TODO: enable it only in debug mode? */
1070 static void
1071 dump_guest_pages(struct virtio_net *dev)
1072 {
1073 	uint32_t i;
1074 	struct guest_page *page;
1075 
1076 	for (i = 0; i < dev->nr_guest_pages; i++) {
1077 		page = &dev->guest_pages[i];
1078 
1079 		VHOST_LOG_CONFIG(INFO, "(%s) guest physical page region %u\n",
1080 				dev->ifname, i);
1081 		VHOST_LOG_CONFIG(INFO, "(%s)\tguest_phys_addr: %" PRIx64 "\n",
1082 				dev->ifname, page->guest_phys_addr);
1083 		VHOST_LOG_CONFIG(INFO, "(%s)\thost_phys_addr : %" PRIx64 "\n",
1084 				dev->ifname, page->host_phys_addr);
1085 		VHOST_LOG_CONFIG(INFO, "(%s)\tsize           : %" PRIx64 "\n",
1086 				dev->ifname, page->size);
1087 	}
1088 }
1089 #else
1090 #define dump_guest_pages(dev)
1091 #endif
1092 
1093 static bool
1094 vhost_memory_changed(struct VhostUserMemory *new,
1095 		     struct rte_vhost_memory *old)
1096 {
1097 	uint32_t i;
1098 
1099 	if (new->nregions != old->nregions)
1100 		return true;
1101 
1102 	for (i = 0; i < new->nregions; ++i) {
1103 		VhostUserMemoryRegion *new_r = &new->regions[i];
1104 		struct rte_vhost_mem_region *old_r = &old->regions[i];
1105 
1106 		if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1107 			return true;
1108 		if (new_r->memory_size != old_r->size)
1109 			return true;
1110 		if (new_r->userspace_addr != old_r->guest_user_addr)
1111 			return true;
1112 	}
1113 
1114 	return false;
1115 }
1116 
1117 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1118 static int
1119 vhost_user_postcopy_region_register(struct virtio_net *dev,
1120 		struct rte_vhost_mem_region *reg)
1121 {
1122 	struct uffdio_register reg_struct;
1123 
1124 	/*
1125 	 * Let's register all the mmapped area to ensure
1126 	 * alignment on page boundary.
1127 	 */
1128 	reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1129 	reg_struct.range.len = reg->mmap_size;
1130 	reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1131 
1132 	if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1133 				&reg_struct)) {
1134 		VHOST_LOG_CONFIG(ERR, "(%s) failed to register ufd for region "
1135 				"%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1136 				dev->ifname,
1137 				(uint64_t)reg_struct.range.start,
1138 				(uint64_t)reg_struct.range.start +
1139 				(uint64_t)reg_struct.range.len - 1,
1140 				dev->postcopy_ufd,
1141 				strerror(errno));
1142 		return -1;
1143 	}
1144 
1145 	VHOST_LOG_CONFIG(INFO,
1146 			"(%s)\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1147 			dev->ifname,
1148 			(uint64_t)reg_struct.range.start,
1149 			(uint64_t)reg_struct.range.start +
1150 			(uint64_t)reg_struct.range.len - 1);
1151 
1152 	return 0;
1153 }
1154 #else
1155 static int
1156 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1157 		struct rte_vhost_mem_region *reg __rte_unused)
1158 {
1159 	return -1;
1160 }
1161 #endif
1162 
1163 static int
1164 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1165 		struct vhu_msg_context *ctx)
1166 {
1167 	struct VhostUserMemory *memory;
1168 	struct rte_vhost_mem_region *reg;
1169 	struct vhu_msg_context ack_ctx;
1170 	uint32_t i;
1171 
1172 	if (!dev->postcopy_listening)
1173 		return 0;
1174 
1175 	/*
1176 	 * We haven't a better way right now than sharing
1177 	 * DPDK's virtual address with Qemu, so that Qemu can
1178 	 * retrieve the region offset when handling userfaults.
1179 	 */
1180 	memory = &ctx->msg.payload.memory;
1181 	for (i = 0; i < memory->nregions; i++) {
1182 		reg = &dev->mem->regions[i];
1183 		memory->regions[i].userspace_addr = reg->host_user_addr;
1184 	}
1185 
1186 	/* Send the addresses back to qemu */
1187 	ctx->fd_num = 0;
1188 	send_vhost_reply(dev, main_fd, ctx);
1189 
1190 	/* Wait for qemu to acknowledge it got the addresses
1191 	 * we've got to wait before we're allowed to generate faults.
1192 	 */
1193 	if (read_vhost_message(dev, main_fd, &ack_ctx) <= 0) {
1194 		VHOST_LOG_CONFIG(ERR, "(%s) failed to read qemu ack on postcopy set-mem-table\n",
1195 				dev->ifname);
1196 		return -1;
1197 	}
1198 
1199 	if (validate_msg_fds(dev, &ack_ctx, 0) != 0)
1200 		return -1;
1201 
1202 	if (ack_ctx.msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1203 		VHOST_LOG_CONFIG(ERR, "(%s) bad qemu ack on postcopy set-mem-table (%d)\n",
1204 				dev->ifname, ack_ctx.msg.request.master);
1205 		return -1;
1206 	}
1207 
1208 	/* Now userfault register and we can use the memory */
1209 	for (i = 0; i < memory->nregions; i++) {
1210 		reg = &dev->mem->regions[i];
1211 		if (vhost_user_postcopy_region_register(dev, reg) < 0)
1212 			return -1;
1213 	}
1214 
1215 	return 0;
1216 }
1217 
1218 static int
1219 vhost_user_mmap_region(struct virtio_net *dev,
1220 		struct rte_vhost_mem_region *region,
1221 		uint64_t mmap_offset)
1222 {
1223 	void *mmap_addr;
1224 	uint64_t mmap_size;
1225 	uint64_t alignment;
1226 	int populate;
1227 	int ret;
1228 
1229 	/* Check for memory_size + mmap_offset overflow */
1230 	if (mmap_offset >= -region->size) {
1231 		VHOST_LOG_CONFIG(ERR, "(%s) mmap_offset (%#"PRIx64") and memory_size (%#"PRIx64") overflow\n",
1232 				dev->ifname, mmap_offset, region->size);
1233 		return -1;
1234 	}
1235 
1236 	mmap_size = region->size + mmap_offset;
1237 
1238 	/* mmap() without flag of MAP_ANONYMOUS, should be called with length
1239 	 * argument aligned with hugepagesz at older longterm version Linux,
1240 	 * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1241 	 *
1242 	 * To avoid failure, make sure in caller to keep length aligned.
1243 	 */
1244 	alignment = get_blk_size(region->fd);
1245 	if (alignment == (uint64_t)-1) {
1246 		VHOST_LOG_CONFIG(ERR, "(%s) couldn't get hugepage size through fstat\n",
1247 				dev->ifname);
1248 		return -1;
1249 	}
1250 	mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1251 	if (mmap_size == 0) {
1252 		/*
1253 		 * It could happen if initial mmap_size + alignment overflows
1254 		 * the sizeof uint64, which could happen if either mmap_size or
1255 		 * alignment value is wrong.
1256 		 *
1257 		 * mmap() kernel implementation would return an error, but
1258 		 * better catch it before and provide useful info in the logs.
1259 		 */
1260 		VHOST_LOG_CONFIG(ERR, "(%s) mmap size (0x%" PRIx64 ") or alignment (0x%" PRIx64 ") is invalid\n",
1261 				dev->ifname, region->size + mmap_offset, alignment);
1262 		return -1;
1263 	}
1264 
1265 	populate = dev->async_copy ? MAP_POPULATE : 0;
1266 	mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1267 			MAP_SHARED | populate, region->fd, 0);
1268 
1269 	if (mmap_addr == MAP_FAILED) {
1270 		VHOST_LOG_CONFIG(ERR, "(%s) mmap failed (%s).\n", dev->ifname, strerror(errno));
1271 		return -1;
1272 	}
1273 
1274 	region->mmap_addr = mmap_addr;
1275 	region->mmap_size = mmap_size;
1276 	region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1277 
1278 	if (dev->async_copy) {
1279 		if (add_guest_pages(dev, region, alignment) < 0) {
1280 			VHOST_LOG_CONFIG(ERR, "(%s) adding guest pages to region failed.\n",
1281 					dev->ifname);
1282 			return -1;
1283 		}
1284 
1285 		if (rte_vfio_is_enabled("vfio")) {
1286 			ret = async_dma_map(dev, region, true);
1287 			if (ret) {
1288 				VHOST_LOG_CONFIG(ERR,
1289 					"(%s) configure IOMMU for DMA engine failed\n",
1290 					dev->ifname);
1291 				return -1;
1292 			}
1293 		}
1294 	}
1295 
1296 	VHOST_LOG_CONFIG(INFO, "(%s) guest memory region size: 0x%" PRIx64 "\n",
1297 			dev->ifname, region->size);
1298 	VHOST_LOG_CONFIG(INFO, "(%s)\t guest physical addr: 0x%" PRIx64 "\n",
1299 			dev->ifname, region->guest_phys_addr);
1300 	VHOST_LOG_CONFIG(INFO, "(%s)\t guest virtual  addr: 0x%" PRIx64 "\n",
1301 			dev->ifname, region->guest_user_addr);
1302 	VHOST_LOG_CONFIG(INFO, "(%s)\t host  virtual  addr: 0x%" PRIx64 "\n",
1303 			dev->ifname, region->host_user_addr);
1304 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap addr : 0x%" PRIx64 "\n",
1305 			dev->ifname, (uint64_t)(uintptr_t)mmap_addr);
1306 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap size : 0x%" PRIx64 "\n",
1307 			dev->ifname, mmap_size);
1308 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap align: 0x%" PRIx64 "\n",
1309 			dev->ifname, alignment);
1310 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap off  : 0x%" PRIx64 "\n",
1311 			dev->ifname, mmap_offset);
1312 
1313 	return 0;
1314 }
1315 
1316 static int
1317 vhost_user_set_mem_table(struct virtio_net **pdev,
1318 			struct vhu_msg_context *ctx,
1319 			int main_fd)
1320 {
1321 	struct virtio_net *dev = *pdev;
1322 	struct VhostUserMemory *memory = &ctx->msg.payload.memory;
1323 	struct rte_vhost_mem_region *reg;
1324 	int numa_node = SOCKET_ID_ANY;
1325 	uint64_t mmap_offset;
1326 	uint32_t i;
1327 	bool async_notify = false;
1328 
1329 	if (validate_msg_fds(dev, ctx, memory->nregions) != 0)
1330 		return RTE_VHOST_MSG_RESULT_ERR;
1331 
1332 	if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1333 		VHOST_LOG_CONFIG(ERR, "(%s) too many memory regions (%u)\n",
1334 				dev->ifname, memory->nregions);
1335 		goto close_msg_fds;
1336 	}
1337 
1338 	if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1339 		VHOST_LOG_CONFIG(INFO, "(%s) memory regions not changed\n", dev->ifname);
1340 
1341 		close_msg_fds(ctx);
1342 
1343 		return RTE_VHOST_MSG_RESULT_OK;
1344 	}
1345 
1346 	if (dev->mem) {
1347 		if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1348 			struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1349 
1350 			if (vdpa_dev && vdpa_dev->ops->dev_close)
1351 				vdpa_dev->ops->dev_close(dev->vid);
1352 			dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1353 		}
1354 
1355 		/* notify the vhost application to stop DMA transfers */
1356 		if (dev->async_copy && dev->notify_ops->vring_state_changed) {
1357 			for (i = 0; i < dev->nr_vring; i++) {
1358 				dev->notify_ops->vring_state_changed(dev->vid,
1359 						i, 0);
1360 			}
1361 			async_notify = true;
1362 		}
1363 
1364 		free_mem_region(dev);
1365 		rte_free(dev->mem);
1366 		dev->mem = NULL;
1367 	}
1368 
1369 	/* Flush IOTLB cache as previous HVAs are now invalid */
1370 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1371 		for (i = 0; i < dev->nr_vring; i++)
1372 			vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1373 
1374 	/*
1375 	 * If VQ 0 has already been allocated, try to allocate on the same
1376 	 * NUMA node. It can be reallocated later in numa_realloc().
1377 	 */
1378 	if (dev->nr_vring > 0)
1379 		numa_node = dev->virtqueue[0]->numa_node;
1380 
1381 	dev->nr_guest_pages = 0;
1382 	if (dev->guest_pages == NULL) {
1383 		dev->max_guest_pages = 8;
1384 		dev->guest_pages = rte_zmalloc_socket(NULL,
1385 					dev->max_guest_pages *
1386 					sizeof(struct guest_page),
1387 					RTE_CACHE_LINE_SIZE,
1388 					numa_node);
1389 		if (dev->guest_pages == NULL) {
1390 			VHOST_LOG_CONFIG(ERR,
1391 				"(%s) failed to allocate memory for dev->guest_pages\n",
1392 				dev->ifname);
1393 			goto close_msg_fds;
1394 		}
1395 	}
1396 
1397 	dev->mem = rte_zmalloc_socket("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1398 		sizeof(struct rte_vhost_mem_region) * memory->nregions, 0, numa_node);
1399 	if (dev->mem == NULL) {
1400 		VHOST_LOG_CONFIG(ERR,
1401 			"(%s) failed to allocate memory for dev->mem\n",
1402 			dev->ifname);
1403 		goto free_guest_pages;
1404 	}
1405 
1406 	for (i = 0; i < memory->nregions; i++) {
1407 		reg = &dev->mem->regions[i];
1408 
1409 		reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1410 		reg->guest_user_addr = memory->regions[i].userspace_addr;
1411 		reg->size            = memory->regions[i].memory_size;
1412 		reg->fd              = ctx->fds[i];
1413 
1414 		/*
1415 		 * Assign invalid file descriptor value to avoid double
1416 		 * closing on error path.
1417 		 */
1418 		ctx->fds[i] = -1;
1419 
1420 		mmap_offset = memory->regions[i].mmap_offset;
1421 
1422 		if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1423 			VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap region %u\n", dev->ifname, i);
1424 			goto free_mem_table;
1425 		}
1426 
1427 		dev->mem->nregions++;
1428 	}
1429 
1430 	if (vhost_user_postcopy_register(dev, main_fd, ctx) < 0)
1431 		goto free_mem_table;
1432 
1433 	for (i = 0; i < dev->nr_vring; i++) {
1434 		struct vhost_virtqueue *vq = dev->virtqueue[i];
1435 
1436 		if (!vq)
1437 			continue;
1438 
1439 		if (vq->desc || vq->avail || vq->used) {
1440 			/*
1441 			 * If the memory table got updated, the ring addresses
1442 			 * need to be translated again as virtual addresses have
1443 			 * changed.
1444 			 */
1445 			vring_invalidate(dev, vq);
1446 
1447 			dev = translate_ring_addresses(dev, i);
1448 			if (!dev) {
1449 				dev = *pdev;
1450 				goto free_mem_table;
1451 			}
1452 
1453 			*pdev = dev;
1454 		}
1455 	}
1456 
1457 	dump_guest_pages(dev);
1458 
1459 	if (async_notify) {
1460 		for (i = 0; i < dev->nr_vring; i++)
1461 			dev->notify_ops->vring_state_changed(dev->vid, i, 1);
1462 	}
1463 
1464 	return RTE_VHOST_MSG_RESULT_OK;
1465 
1466 free_mem_table:
1467 	free_mem_region(dev);
1468 	rte_free(dev->mem);
1469 	dev->mem = NULL;
1470 
1471 free_guest_pages:
1472 	rte_free(dev->guest_pages);
1473 	dev->guest_pages = NULL;
1474 close_msg_fds:
1475 	close_msg_fds(ctx);
1476 	return RTE_VHOST_MSG_RESULT_ERR;
1477 }
1478 
1479 static bool
1480 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1481 {
1482 	bool rings_ok;
1483 
1484 	if (!vq)
1485 		return false;
1486 
1487 	if (vq_is_packed(dev))
1488 		rings_ok = vq->desc_packed && vq->driver_event &&
1489 			vq->device_event;
1490 	else
1491 		rings_ok = vq->desc && vq->avail && vq->used;
1492 
1493 	return rings_ok &&
1494 	       vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1495 	       vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1496 	       vq->enabled;
1497 }
1498 
1499 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1500 
1501 static int
1502 virtio_is_ready(struct virtio_net *dev)
1503 {
1504 	struct vhost_virtqueue *vq;
1505 	uint32_t i, nr_vring = dev->nr_vring;
1506 
1507 	if (dev->flags & VIRTIO_DEV_READY)
1508 		return 1;
1509 
1510 	if (!dev->nr_vring)
1511 		return 0;
1512 
1513 	if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1514 		nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1515 
1516 		if (dev->nr_vring < nr_vring)
1517 			return 0;
1518 	}
1519 
1520 	for (i = 0; i < nr_vring; i++) {
1521 		vq = dev->virtqueue[i];
1522 
1523 		if (!vq_is_ready(dev, vq))
1524 			return 0;
1525 	}
1526 
1527 	/* If supported, ensure the frontend is really done with config */
1528 	if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1529 		if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1530 			return 0;
1531 
1532 	dev->flags |= VIRTIO_DEV_READY;
1533 
1534 	if (!(dev->flags & VIRTIO_DEV_RUNNING))
1535 		VHOST_LOG_CONFIG(INFO, "(%s) virtio is now ready for processing.\n", dev->ifname);
1536 	return 1;
1537 }
1538 
1539 static void *
1540 inflight_mem_alloc(struct virtio_net *dev, const char *name, size_t size, int *fd)
1541 {
1542 	void *ptr;
1543 	int mfd = -1;
1544 	char fname[20] = "/tmp/memfd-XXXXXX";
1545 
1546 	*fd = -1;
1547 #ifdef MEMFD_SUPPORTED
1548 	mfd = memfd_create(name, MFD_CLOEXEC);
1549 #else
1550 	RTE_SET_USED(name);
1551 #endif
1552 	if (mfd == -1) {
1553 		mfd = mkstemp(fname);
1554 		if (mfd == -1) {
1555 			VHOST_LOG_CONFIG(ERR, "(%s) failed to get inflight buffer fd\n",
1556 					dev->ifname);
1557 			return NULL;
1558 		}
1559 
1560 		unlink(fname);
1561 	}
1562 
1563 	if (ftruncate(mfd, size) == -1) {
1564 		VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc inflight buffer\n", dev->ifname);
1565 		close(mfd);
1566 		return NULL;
1567 	}
1568 
1569 	ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1570 	if (ptr == MAP_FAILED) {
1571 		VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap inflight buffer\n", dev->ifname);
1572 		close(mfd);
1573 		return NULL;
1574 	}
1575 
1576 	*fd = mfd;
1577 	return ptr;
1578 }
1579 
1580 static uint32_t
1581 get_pervq_shm_size_split(uint16_t queue_size)
1582 {
1583 	return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1584 				  queue_size + sizeof(uint64_t) +
1585 				  sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1586 }
1587 
1588 static uint32_t
1589 get_pervq_shm_size_packed(uint16_t queue_size)
1590 {
1591 	return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1592 				  * queue_size + sizeof(uint64_t) +
1593 				  sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1594 				  INFLIGHT_ALIGNMENT);
1595 }
1596 
1597 static int
1598 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1599 			   struct vhu_msg_context *ctx,
1600 			   int main_fd __rte_unused)
1601 {
1602 	struct rte_vhost_inflight_info_packed *inflight_packed;
1603 	uint64_t pervq_inflight_size, mmap_size;
1604 	uint16_t num_queues, queue_size;
1605 	struct virtio_net *dev = *pdev;
1606 	int fd, i, j;
1607 	int numa_node = SOCKET_ID_ANY;
1608 	void *addr;
1609 
1610 	if (ctx->msg.size != sizeof(ctx->msg.payload.inflight)) {
1611 		VHOST_LOG_CONFIG(ERR, "(%s) invalid get_inflight_fd message size is %d\n",
1612 			dev->ifname, ctx->msg.size);
1613 		return RTE_VHOST_MSG_RESULT_ERR;
1614 	}
1615 
1616 	/*
1617 	 * If VQ 0 has already been allocated, try to allocate on the same
1618 	 * NUMA node. It can be reallocated later in numa_realloc().
1619 	 */
1620 	if (dev->nr_vring > 0)
1621 		numa_node = dev->virtqueue[0]->numa_node;
1622 
1623 	if (dev->inflight_info == NULL) {
1624 		dev->inflight_info = rte_zmalloc_socket("inflight_info",
1625 				sizeof(struct inflight_mem_info), 0, numa_node);
1626 		if (!dev->inflight_info) {
1627 			VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc dev inflight area\n",
1628 					dev->ifname);
1629 			return RTE_VHOST_MSG_RESULT_ERR;
1630 		}
1631 		dev->inflight_info->fd = -1;
1632 	}
1633 
1634 	num_queues = ctx->msg.payload.inflight.num_queues;
1635 	queue_size = ctx->msg.payload.inflight.queue_size;
1636 
1637 	VHOST_LOG_CONFIG(INFO, "(%s) get_inflight_fd num_queues: %u\n",
1638 		dev->ifname, ctx->msg.payload.inflight.num_queues);
1639 	VHOST_LOG_CONFIG(INFO, "(%s) get_inflight_fd queue_size: %u\n",
1640 		dev->ifname, ctx->msg.payload.inflight.queue_size);
1641 
1642 	if (vq_is_packed(dev))
1643 		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1644 	else
1645 		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1646 
1647 	mmap_size = num_queues * pervq_inflight_size;
1648 	addr = inflight_mem_alloc(dev, "vhost-inflight", mmap_size, &fd);
1649 	if (!addr) {
1650 		VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc vhost inflight area\n", dev->ifname);
1651 			ctx->msg.payload.inflight.mmap_size = 0;
1652 		return RTE_VHOST_MSG_RESULT_ERR;
1653 	}
1654 	memset(addr, 0, mmap_size);
1655 
1656 	if (dev->inflight_info->addr) {
1657 		munmap(dev->inflight_info->addr, dev->inflight_info->size);
1658 		dev->inflight_info->addr = NULL;
1659 	}
1660 
1661 	if (dev->inflight_info->fd >= 0) {
1662 		close(dev->inflight_info->fd);
1663 		dev->inflight_info->fd = -1;
1664 	}
1665 
1666 	dev->inflight_info->addr = addr;
1667 	dev->inflight_info->size = ctx->msg.payload.inflight.mmap_size = mmap_size;
1668 	dev->inflight_info->fd = ctx->fds[0] = fd;
1669 	ctx->msg.payload.inflight.mmap_offset = 0;
1670 	ctx->fd_num = 1;
1671 
1672 	if (vq_is_packed(dev)) {
1673 		for (i = 0; i < num_queues; i++) {
1674 			inflight_packed =
1675 				(struct rte_vhost_inflight_info_packed *)addr;
1676 			inflight_packed->used_wrap_counter = 1;
1677 			inflight_packed->old_used_wrap_counter = 1;
1678 			for (j = 0; j < queue_size; j++)
1679 				inflight_packed->desc[j].next = j + 1;
1680 			addr = (void *)((char *)addr + pervq_inflight_size);
1681 		}
1682 	}
1683 
1684 	VHOST_LOG_CONFIG(INFO, "(%s) send inflight mmap_size: %"PRIu64"\n",
1685 			dev->ifname, ctx->msg.payload.inflight.mmap_size);
1686 	VHOST_LOG_CONFIG(INFO, "(%s) send inflight mmap_offset: %"PRIu64"\n",
1687 			dev->ifname, ctx->msg.payload.inflight.mmap_offset);
1688 	VHOST_LOG_CONFIG(INFO, "(%s) send inflight fd: %d\n", dev->ifname, ctx->fds[0]);
1689 
1690 	return RTE_VHOST_MSG_RESULT_REPLY;
1691 }
1692 
1693 static int
1694 vhost_user_set_inflight_fd(struct virtio_net **pdev,
1695 			   struct vhu_msg_context *ctx,
1696 			   int main_fd __rte_unused)
1697 {
1698 	uint64_t mmap_size, mmap_offset;
1699 	uint16_t num_queues, queue_size;
1700 	struct virtio_net *dev = *pdev;
1701 	uint32_t pervq_inflight_size;
1702 	struct vhost_virtqueue *vq;
1703 	void *addr;
1704 	int fd, i;
1705 	int numa_node = SOCKET_ID_ANY;
1706 
1707 	fd = ctx->fds[0];
1708 	if (ctx->msg.size != sizeof(ctx->msg.payload.inflight) || fd < 0) {
1709 		VHOST_LOG_CONFIG(ERR, "(%s) invalid set_inflight_fd message size is %d,fd is %d\n",
1710 			dev->ifname, ctx->msg.size, fd);
1711 		return RTE_VHOST_MSG_RESULT_ERR;
1712 	}
1713 
1714 	mmap_size = ctx->msg.payload.inflight.mmap_size;
1715 	mmap_offset = ctx->msg.payload.inflight.mmap_offset;
1716 	num_queues = ctx->msg.payload.inflight.num_queues;
1717 	queue_size = ctx->msg.payload.inflight.queue_size;
1718 
1719 	if (vq_is_packed(dev))
1720 		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1721 	else
1722 		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1723 
1724 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd mmap_size: %"PRIu64"\n",
1725 			dev->ifname, mmap_size);
1726 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd mmap_offset: %"PRIu64"\n",
1727 			dev->ifname, mmap_offset);
1728 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd num_queues: %u\n", dev->ifname, num_queues);
1729 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd queue_size: %u\n", dev->ifname, queue_size);
1730 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd fd: %d\n", dev->ifname, fd);
1731 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd pervq_inflight_size: %d\n",
1732 			dev->ifname, pervq_inflight_size);
1733 
1734 	/*
1735 	 * If VQ 0 has already been allocated, try to allocate on the same
1736 	 * NUMA node. It can be reallocated later in numa_realloc().
1737 	 */
1738 	if (dev->nr_vring > 0)
1739 		numa_node = dev->virtqueue[0]->numa_node;
1740 
1741 	if (!dev->inflight_info) {
1742 		dev->inflight_info = rte_zmalloc_socket("inflight_info",
1743 				sizeof(struct inflight_mem_info), 0, numa_node);
1744 		if (dev->inflight_info == NULL) {
1745 			VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc dev inflight area\n",
1746 					dev->ifname);
1747 			return RTE_VHOST_MSG_RESULT_ERR;
1748 		}
1749 		dev->inflight_info->fd = -1;
1750 	}
1751 
1752 	if (dev->inflight_info->addr) {
1753 		munmap(dev->inflight_info->addr, dev->inflight_info->size);
1754 		dev->inflight_info->addr = NULL;
1755 	}
1756 
1757 	addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1758 		    fd, mmap_offset);
1759 	if (addr == MAP_FAILED) {
1760 		VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap share memory.\n", dev->ifname);
1761 		return RTE_VHOST_MSG_RESULT_ERR;
1762 	}
1763 
1764 	if (dev->inflight_info->fd >= 0) {
1765 		close(dev->inflight_info->fd);
1766 		dev->inflight_info->fd = -1;
1767 	}
1768 
1769 	dev->inflight_info->fd = fd;
1770 	dev->inflight_info->addr = addr;
1771 	dev->inflight_info->size = mmap_size;
1772 
1773 	for (i = 0; i < num_queues; i++) {
1774 		vq = dev->virtqueue[i];
1775 		if (!vq)
1776 			continue;
1777 
1778 		if (vq_is_packed(dev)) {
1779 			vq->inflight_packed = addr;
1780 			vq->inflight_packed->desc_num = queue_size;
1781 		} else {
1782 			vq->inflight_split = addr;
1783 			vq->inflight_split->desc_num = queue_size;
1784 		}
1785 		addr = (void *)((char *)addr + pervq_inflight_size);
1786 	}
1787 
1788 	return RTE_VHOST_MSG_RESULT_OK;
1789 }
1790 
1791 static int
1792 vhost_user_set_vring_call(struct virtio_net **pdev,
1793 			struct vhu_msg_context *ctx,
1794 			int main_fd __rte_unused)
1795 {
1796 	struct virtio_net *dev = *pdev;
1797 	struct vhost_vring_file file;
1798 	struct vhost_virtqueue *vq;
1799 	int expected_fds;
1800 
1801 	expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1802 	if (validate_msg_fds(dev, ctx, expected_fds) != 0)
1803 		return RTE_VHOST_MSG_RESULT_ERR;
1804 
1805 	file.index = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
1806 	if (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1807 		file.fd = VIRTIO_INVALID_EVENTFD;
1808 	else
1809 		file.fd = ctx->fds[0];
1810 	VHOST_LOG_CONFIG(INFO, "(%s) vring call idx:%d file:%d\n",
1811 			dev->ifname, file.index, file.fd);
1812 
1813 	vq = dev->virtqueue[file.index];
1814 
1815 	if (vq->ready) {
1816 		vq->ready = false;
1817 		vhost_user_notify_queue_state(dev, file.index, 0);
1818 	}
1819 
1820 	if (vq->callfd >= 0)
1821 		close(vq->callfd);
1822 
1823 	vq->callfd = file.fd;
1824 
1825 	return RTE_VHOST_MSG_RESULT_OK;
1826 }
1827 
1828 static int vhost_user_set_vring_err(struct virtio_net **pdev,
1829 			struct vhu_msg_context *ctx,
1830 			int main_fd __rte_unused)
1831 {
1832 	struct virtio_net *dev = *pdev;
1833 	int expected_fds;
1834 
1835 	expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1836 	if (validate_msg_fds(dev, ctx, expected_fds) != 0)
1837 		return RTE_VHOST_MSG_RESULT_ERR;
1838 
1839 	if (!(ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1840 		close(ctx->fds[0]);
1841 	VHOST_LOG_CONFIG(INFO, "(%s) not implemented\n", dev->ifname);
1842 
1843 	return RTE_VHOST_MSG_RESULT_OK;
1844 }
1845 
1846 static int
1847 resubmit_desc_compare(const void *a, const void *b)
1848 {
1849 	const struct rte_vhost_resubmit_desc *desc0 = a;
1850 	const struct rte_vhost_resubmit_desc *desc1 = b;
1851 
1852 	if (desc1->counter > desc0->counter)
1853 		return 1;
1854 
1855 	return -1;
1856 }
1857 
1858 static int
1859 vhost_check_queue_inflights_split(struct virtio_net *dev,
1860 				  struct vhost_virtqueue *vq)
1861 {
1862 	uint16_t i;
1863 	uint16_t resubmit_num = 0, last_io, num;
1864 	struct vring_used *used = vq->used;
1865 	struct rte_vhost_resubmit_info *resubmit;
1866 	struct rte_vhost_inflight_info_split *inflight_split;
1867 
1868 	if (!(dev->protocol_features &
1869 	    (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1870 		return RTE_VHOST_MSG_RESULT_OK;
1871 
1872 	/* The frontend may still not support the inflight feature
1873 	 * although we negotiate the protocol feature.
1874 	 */
1875 	if ((!vq->inflight_split))
1876 		return RTE_VHOST_MSG_RESULT_OK;
1877 
1878 	if (!vq->inflight_split->version) {
1879 		vq->inflight_split->version = INFLIGHT_VERSION;
1880 		return RTE_VHOST_MSG_RESULT_OK;
1881 	}
1882 
1883 	if (vq->resubmit_inflight)
1884 		return RTE_VHOST_MSG_RESULT_OK;
1885 
1886 	inflight_split = vq->inflight_split;
1887 	vq->global_counter = 0;
1888 	last_io = inflight_split->last_inflight_io;
1889 
1890 	if (inflight_split->used_idx != used->idx) {
1891 		inflight_split->desc[last_io].inflight = 0;
1892 		rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1893 		inflight_split->used_idx = used->idx;
1894 	}
1895 
1896 	for (i = 0; i < inflight_split->desc_num; i++) {
1897 		if (inflight_split->desc[i].inflight == 1)
1898 			resubmit_num++;
1899 	}
1900 
1901 	vq->last_avail_idx += resubmit_num;
1902 
1903 	if (resubmit_num) {
1904 		resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1905 				0, vq->numa_node);
1906 		if (!resubmit) {
1907 			VHOST_LOG_CONFIG(ERR,
1908 					"(%s) failed to allocate memory for resubmit info.\n",
1909 					dev->ifname);
1910 			return RTE_VHOST_MSG_RESULT_ERR;
1911 		}
1912 
1913 		resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
1914 				resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
1915 				0, vq->numa_node);
1916 		if (!resubmit->resubmit_list) {
1917 			VHOST_LOG_CONFIG(ERR,
1918 					"(%s) failed to allocate memory for inflight desc.\n",
1919 					dev->ifname);
1920 			rte_free(resubmit);
1921 			return RTE_VHOST_MSG_RESULT_ERR;
1922 		}
1923 
1924 		num = 0;
1925 		for (i = 0; i < vq->inflight_split->desc_num; i++) {
1926 			if (vq->inflight_split->desc[i].inflight == 1) {
1927 				resubmit->resubmit_list[num].index = i;
1928 				resubmit->resubmit_list[num].counter =
1929 					inflight_split->desc[i].counter;
1930 				num++;
1931 			}
1932 		}
1933 		resubmit->resubmit_num = num;
1934 
1935 		if (resubmit->resubmit_num > 1)
1936 			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1937 			      sizeof(struct rte_vhost_resubmit_desc),
1938 			      resubmit_desc_compare);
1939 
1940 		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1941 		vq->resubmit_inflight = resubmit;
1942 	}
1943 
1944 	return RTE_VHOST_MSG_RESULT_OK;
1945 }
1946 
1947 static int
1948 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1949 				   struct vhost_virtqueue *vq)
1950 {
1951 	uint16_t i;
1952 	uint16_t resubmit_num = 0, old_used_idx, num;
1953 	struct rte_vhost_resubmit_info *resubmit;
1954 	struct rte_vhost_inflight_info_packed *inflight_packed;
1955 
1956 	if (!(dev->protocol_features &
1957 	    (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1958 		return RTE_VHOST_MSG_RESULT_OK;
1959 
1960 	/* The frontend may still not support the inflight feature
1961 	 * although we negotiate the protocol feature.
1962 	 */
1963 	if ((!vq->inflight_packed))
1964 		return RTE_VHOST_MSG_RESULT_OK;
1965 
1966 	if (!vq->inflight_packed->version) {
1967 		vq->inflight_packed->version = INFLIGHT_VERSION;
1968 		return RTE_VHOST_MSG_RESULT_OK;
1969 	}
1970 
1971 	if (vq->resubmit_inflight)
1972 		return RTE_VHOST_MSG_RESULT_OK;
1973 
1974 	inflight_packed = vq->inflight_packed;
1975 	vq->global_counter = 0;
1976 	old_used_idx = inflight_packed->old_used_idx;
1977 
1978 	if (inflight_packed->used_idx != old_used_idx) {
1979 		if (inflight_packed->desc[old_used_idx].inflight == 0) {
1980 			inflight_packed->old_used_idx =
1981 				inflight_packed->used_idx;
1982 			inflight_packed->old_used_wrap_counter =
1983 				inflight_packed->used_wrap_counter;
1984 			inflight_packed->old_free_head =
1985 				inflight_packed->free_head;
1986 		} else {
1987 			inflight_packed->used_idx =
1988 				inflight_packed->old_used_idx;
1989 			inflight_packed->used_wrap_counter =
1990 				inflight_packed->old_used_wrap_counter;
1991 			inflight_packed->free_head =
1992 				inflight_packed->old_free_head;
1993 		}
1994 	}
1995 
1996 	for (i = 0; i < inflight_packed->desc_num; i++) {
1997 		if (inflight_packed->desc[i].inflight == 1)
1998 			resubmit_num++;
1999 	}
2000 
2001 	if (resubmit_num) {
2002 		resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
2003 				0, vq->numa_node);
2004 		if (resubmit == NULL) {
2005 			VHOST_LOG_CONFIG(ERR,
2006 					"(%s) failed to allocate memory for resubmit info.\n",
2007 					dev->ifname);
2008 			return RTE_VHOST_MSG_RESULT_ERR;
2009 		}
2010 
2011 		resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
2012 				resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
2013 				0, vq->numa_node);
2014 		if (resubmit->resubmit_list == NULL) {
2015 			VHOST_LOG_CONFIG(ERR,
2016 					"(%s) failed to allocate memory for resubmit desc.\n",
2017 					dev->ifname);
2018 			rte_free(resubmit);
2019 			return RTE_VHOST_MSG_RESULT_ERR;
2020 		}
2021 
2022 		num = 0;
2023 		for (i = 0; i < inflight_packed->desc_num; i++) {
2024 			if (vq->inflight_packed->desc[i].inflight == 1) {
2025 				resubmit->resubmit_list[num].index = i;
2026 				resubmit->resubmit_list[num].counter =
2027 					inflight_packed->desc[i].counter;
2028 				num++;
2029 			}
2030 		}
2031 		resubmit->resubmit_num = num;
2032 
2033 		if (resubmit->resubmit_num > 1)
2034 			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
2035 			      sizeof(struct rte_vhost_resubmit_desc),
2036 			      resubmit_desc_compare);
2037 
2038 		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
2039 		vq->resubmit_inflight = resubmit;
2040 	}
2041 
2042 	return RTE_VHOST_MSG_RESULT_OK;
2043 }
2044 
2045 static int
2046 vhost_user_set_vring_kick(struct virtio_net **pdev,
2047 			struct vhu_msg_context *ctx,
2048 			int main_fd __rte_unused)
2049 {
2050 	struct virtio_net *dev = *pdev;
2051 	struct vhost_vring_file file;
2052 	struct vhost_virtqueue *vq;
2053 	int expected_fds;
2054 
2055 	expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
2056 	if (validate_msg_fds(dev, ctx, expected_fds) != 0)
2057 		return RTE_VHOST_MSG_RESULT_ERR;
2058 
2059 	file.index = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
2060 	if (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK)
2061 		file.fd = VIRTIO_INVALID_EVENTFD;
2062 	else
2063 		file.fd = ctx->fds[0];
2064 	VHOST_LOG_CONFIG(INFO, "(%s) vring kick idx:%d file:%d\n",
2065 			dev->ifname, file.index, file.fd);
2066 
2067 	/* Interpret ring addresses only when ring is started. */
2068 	dev = translate_ring_addresses(dev, file.index);
2069 	if (!dev) {
2070 		if (file.fd != VIRTIO_INVALID_EVENTFD)
2071 			close(file.fd);
2072 
2073 		return RTE_VHOST_MSG_RESULT_ERR;
2074 	}
2075 
2076 	*pdev = dev;
2077 
2078 	vq = dev->virtqueue[file.index];
2079 
2080 	/*
2081 	 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
2082 	 * the ring starts already enabled. Otherwise, it is enabled via
2083 	 * the SET_VRING_ENABLE message.
2084 	 */
2085 	if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
2086 		vq->enabled = true;
2087 	}
2088 
2089 	if (vq->ready) {
2090 		vq->ready = false;
2091 		vhost_user_notify_queue_state(dev, file.index, 0);
2092 	}
2093 
2094 	if (vq->kickfd >= 0)
2095 		close(vq->kickfd);
2096 	vq->kickfd = file.fd;
2097 
2098 	if (vq_is_packed(dev)) {
2099 		if (vhost_check_queue_inflights_packed(dev, vq)) {
2100 			VHOST_LOG_CONFIG(ERR, "(%s) failed to inflights for vq: %d\n",
2101 					dev->ifname, file.index);
2102 			return RTE_VHOST_MSG_RESULT_ERR;
2103 		}
2104 	} else {
2105 		if (vhost_check_queue_inflights_split(dev, vq)) {
2106 			VHOST_LOG_CONFIG(ERR, "(%s) failed to inflights for vq: %d\n",
2107 					dev->ifname, file.index);
2108 			return RTE_VHOST_MSG_RESULT_ERR;
2109 		}
2110 	}
2111 
2112 	return RTE_VHOST_MSG_RESULT_OK;
2113 }
2114 
2115 /*
2116  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
2117  */
2118 static int
2119 vhost_user_get_vring_base(struct virtio_net **pdev,
2120 			struct vhu_msg_context *ctx,
2121 			int main_fd __rte_unused)
2122 {
2123 	struct virtio_net *dev = *pdev;
2124 	struct vhost_virtqueue *vq = dev->virtqueue[ctx->msg.payload.state.index];
2125 	uint64_t val;
2126 
2127 	if (validate_msg_fds(dev, ctx, 0) != 0)
2128 		return RTE_VHOST_MSG_RESULT_ERR;
2129 
2130 	/* We have to stop the queue (virtio) if it is running. */
2131 	vhost_destroy_device_notify(dev);
2132 
2133 	dev->flags &= ~VIRTIO_DEV_READY;
2134 	dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
2135 
2136 	/* Here we are safe to get the indexes */
2137 	if (vq_is_packed(dev)) {
2138 		/*
2139 		 * Bit[0:14]: avail index
2140 		 * Bit[15]: avail wrap counter
2141 		 */
2142 		val = vq->last_avail_idx & 0x7fff;
2143 		val |= vq->avail_wrap_counter << 15;
2144 		ctx->msg.payload.state.num = val;
2145 	} else {
2146 		ctx->msg.payload.state.num = vq->last_avail_idx;
2147 	}
2148 
2149 	VHOST_LOG_CONFIG(INFO, "(%s) vring base idx:%d file:%d\n",
2150 			dev->ifname, ctx->msg.payload.state.index,
2151 			ctx->msg.payload.state.num);
2152 	/*
2153 	 * Based on current qemu vhost-user implementation, this message is
2154 	 * sent and only sent in vhost_vring_stop.
2155 	 * TODO: cleanup the vring, it isn't usable since here.
2156 	 */
2157 	if (vq->kickfd >= 0)
2158 		close(vq->kickfd);
2159 
2160 	vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2161 
2162 	if (vq->callfd >= 0)
2163 		close(vq->callfd);
2164 
2165 	vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2166 
2167 	vq->signalled_used_valid = false;
2168 
2169 	if (vq_is_packed(dev)) {
2170 		rte_free(vq->shadow_used_packed);
2171 		vq->shadow_used_packed = NULL;
2172 	} else {
2173 		rte_free(vq->shadow_used_split);
2174 		vq->shadow_used_split = NULL;
2175 	}
2176 
2177 	rte_free(vq->batch_copy_elems);
2178 	vq->batch_copy_elems = NULL;
2179 
2180 	rte_free(vq->log_cache);
2181 	vq->log_cache = NULL;
2182 
2183 	ctx->msg.size = sizeof(ctx->msg.payload.state);
2184 	ctx->fd_num = 0;
2185 
2186 	vhost_user_iotlb_flush_all(vq);
2187 
2188 	vring_invalidate(dev, vq);
2189 
2190 	return RTE_VHOST_MSG_RESULT_REPLY;
2191 }
2192 
2193 /*
2194  * when virtio queues are ready to work, qemu will send us to
2195  * enable the virtio queue pair.
2196  */
2197 static int
2198 vhost_user_set_vring_enable(struct virtio_net **pdev,
2199 			struct vhu_msg_context *ctx,
2200 			int main_fd __rte_unused)
2201 {
2202 	struct virtio_net *dev = *pdev;
2203 	bool enable = !!ctx->msg.payload.state.num;
2204 	int index = (int)ctx->msg.payload.state.index;
2205 
2206 	if (validate_msg_fds(dev, ctx, 0) != 0)
2207 		return RTE_VHOST_MSG_RESULT_ERR;
2208 
2209 	VHOST_LOG_CONFIG(INFO, "(%s) set queue enable: %d to qp idx: %d\n",
2210 			dev->ifname, enable, index);
2211 
2212 	if (enable && dev->virtqueue[index]->async) {
2213 		if (dev->virtqueue[index]->async->pkts_inflight_n) {
2214 			VHOST_LOG_CONFIG(ERR,
2215 				"(%s) failed to enable vring. Inflight packets must be completed first\n",
2216 				dev->ifname);
2217 			return RTE_VHOST_MSG_RESULT_ERR;
2218 		}
2219 	}
2220 
2221 	dev->virtqueue[index]->enabled = enable;
2222 
2223 	return RTE_VHOST_MSG_RESULT_OK;
2224 }
2225 
2226 static int
2227 vhost_user_get_protocol_features(struct virtio_net **pdev,
2228 			struct vhu_msg_context *ctx,
2229 			int main_fd __rte_unused)
2230 {
2231 	struct virtio_net *dev = *pdev;
2232 	uint64_t features, protocol_features;
2233 
2234 	if (validate_msg_fds(dev, ctx, 0) != 0)
2235 		return RTE_VHOST_MSG_RESULT_ERR;
2236 
2237 	rte_vhost_driver_get_features(dev->ifname, &features);
2238 	rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2239 
2240 	ctx->msg.payload.u64 = protocol_features;
2241 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
2242 	ctx->fd_num = 0;
2243 
2244 	return RTE_VHOST_MSG_RESULT_REPLY;
2245 }
2246 
2247 static int
2248 vhost_user_set_protocol_features(struct virtio_net **pdev,
2249 			struct vhu_msg_context *ctx,
2250 			int main_fd __rte_unused)
2251 {
2252 	struct virtio_net *dev = *pdev;
2253 	uint64_t protocol_features = ctx->msg.payload.u64;
2254 	uint64_t slave_protocol_features = 0;
2255 
2256 	if (validate_msg_fds(dev, ctx, 0) != 0)
2257 		return RTE_VHOST_MSG_RESULT_ERR;
2258 
2259 	rte_vhost_driver_get_protocol_features(dev->ifname,
2260 			&slave_protocol_features);
2261 	if (protocol_features & ~slave_protocol_features) {
2262 		VHOST_LOG_CONFIG(ERR, "(%s) received invalid protocol features.\n", dev->ifname);
2263 		return RTE_VHOST_MSG_RESULT_ERR;
2264 	}
2265 
2266 	dev->protocol_features = protocol_features;
2267 	VHOST_LOG_CONFIG(INFO, "(%s) negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2268 		dev->ifname, dev->protocol_features);
2269 
2270 	return RTE_VHOST_MSG_RESULT_OK;
2271 }
2272 
2273 static int
2274 vhost_user_set_log_base(struct virtio_net **pdev,
2275 			struct vhu_msg_context *ctx,
2276 			int main_fd __rte_unused)
2277 {
2278 	struct virtio_net *dev = *pdev;
2279 	int fd = ctx->fds[0];
2280 	uint64_t size, off;
2281 	void *addr;
2282 	uint32_t i;
2283 
2284 	if (validate_msg_fds(dev, ctx, 1) != 0)
2285 		return RTE_VHOST_MSG_RESULT_ERR;
2286 
2287 	if (fd < 0) {
2288 		VHOST_LOG_CONFIG(ERR, "(%s) invalid log fd: %d\n", dev->ifname, fd);
2289 		return RTE_VHOST_MSG_RESULT_ERR;
2290 	}
2291 
2292 	if (ctx->msg.size != sizeof(VhostUserLog)) {
2293 		VHOST_LOG_CONFIG(ERR, "(%s) invalid log base msg size: %"PRId32" != %d\n",
2294 			dev->ifname, ctx->msg.size, (int)sizeof(VhostUserLog));
2295 		goto close_msg_fds;
2296 	}
2297 
2298 	size = ctx->msg.payload.log.mmap_size;
2299 	off  = ctx->msg.payload.log.mmap_offset;
2300 
2301 	/* Check for mmap size and offset overflow. */
2302 	if (off >= -size) {
2303 		VHOST_LOG_CONFIG(ERR,
2304 				"(%s) log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2305 				dev->ifname, off, size);
2306 		goto close_msg_fds;
2307 	}
2308 
2309 	VHOST_LOG_CONFIG(INFO, "(%s) log mmap size: %"PRId64", offset: %"PRId64"\n",
2310 			dev->ifname, size, off);
2311 
2312 	/*
2313 	 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2314 	 * fail when offset is not page size aligned.
2315 	 */
2316 	addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2317 	close(fd);
2318 	if (addr == MAP_FAILED) {
2319 		VHOST_LOG_CONFIG(ERR, "(%s) mmap log base failed!\n", dev->ifname);
2320 		return RTE_VHOST_MSG_RESULT_ERR;
2321 	}
2322 
2323 	/*
2324 	 * Free previously mapped log memory on occasionally
2325 	 * multiple VHOST_USER_SET_LOG_BASE.
2326 	 */
2327 	if (dev->log_addr) {
2328 		munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2329 	}
2330 	dev->log_addr = (uint64_t)(uintptr_t)addr;
2331 	dev->log_base = dev->log_addr + off;
2332 	dev->log_size = size;
2333 
2334 	for (i = 0; i < dev->nr_vring; i++) {
2335 		struct vhost_virtqueue *vq = dev->virtqueue[i];
2336 
2337 		rte_free(vq->log_cache);
2338 		vq->log_cache = NULL;
2339 		vq->log_cache_nb_elem = 0;
2340 		vq->log_cache = rte_malloc_socket("vq log cache",
2341 				sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2342 				0, vq->numa_node);
2343 		/*
2344 		 * If log cache alloc fail, don't fail migration, but no
2345 		 * caching will be done, which will impact performance
2346 		 */
2347 		if (!vq->log_cache)
2348 			VHOST_LOG_CONFIG(ERR, "(%s) failed to allocate VQ logging cache\n",
2349 					dev->ifname);
2350 	}
2351 
2352 	/*
2353 	 * The spec is not clear about it (yet), but QEMU doesn't expect
2354 	 * any payload in the reply.
2355 	 */
2356 	ctx->msg.size = 0;
2357 	ctx->fd_num = 0;
2358 
2359 	return RTE_VHOST_MSG_RESULT_REPLY;
2360 
2361 close_msg_fds:
2362 	close_msg_fds(ctx);
2363 	return RTE_VHOST_MSG_RESULT_ERR;
2364 }
2365 
2366 static int vhost_user_set_log_fd(struct virtio_net **pdev,
2367 			struct vhu_msg_context *ctx,
2368 			int main_fd __rte_unused)
2369 {
2370 	struct virtio_net *dev = *pdev;
2371 
2372 	if (validate_msg_fds(dev, ctx, 1) != 0)
2373 		return RTE_VHOST_MSG_RESULT_ERR;
2374 
2375 	close(ctx->fds[0]);
2376 	VHOST_LOG_CONFIG(INFO, "(%s) not implemented.\n", dev->ifname);
2377 
2378 	return RTE_VHOST_MSG_RESULT_OK;
2379 }
2380 
2381 /*
2382  * An rarp packet is constructed and broadcasted to notify switches about
2383  * the new location of the migrated VM, so that packets from outside will
2384  * not be lost after migration.
2385  *
2386  * However, we don't actually "send" a rarp packet here, instead, we set
2387  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2388  */
2389 static int
2390 vhost_user_send_rarp(struct virtio_net **pdev,
2391 			struct vhu_msg_context *ctx,
2392 			int main_fd __rte_unused)
2393 {
2394 	struct virtio_net *dev = *pdev;
2395 	uint8_t *mac = (uint8_t *)&ctx->msg.payload.u64;
2396 	struct rte_vdpa_device *vdpa_dev;
2397 
2398 	if (validate_msg_fds(dev, ctx, 0) != 0)
2399 		return RTE_VHOST_MSG_RESULT_ERR;
2400 
2401 	VHOST_LOG_CONFIG(DEBUG, "(%s) MAC: " RTE_ETHER_ADDR_PRT_FMT "\n",
2402 		dev->ifname, mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2403 	memcpy(dev->mac.addr_bytes, mac, 6);
2404 
2405 	/*
2406 	 * Set the flag to inject a RARP broadcast packet at
2407 	 * rte_vhost_dequeue_burst().
2408 	 *
2409 	 * __ATOMIC_RELEASE ordering is for making sure the mac is
2410 	 * copied before the flag is set.
2411 	 */
2412 	__atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2413 	vdpa_dev = dev->vdpa_dev;
2414 	if (vdpa_dev && vdpa_dev->ops->migration_done)
2415 		vdpa_dev->ops->migration_done(dev->vid);
2416 
2417 	return RTE_VHOST_MSG_RESULT_OK;
2418 }
2419 
2420 static int
2421 vhost_user_net_set_mtu(struct virtio_net **pdev,
2422 			struct vhu_msg_context *ctx,
2423 			int main_fd __rte_unused)
2424 {
2425 	struct virtio_net *dev = *pdev;
2426 
2427 	if (validate_msg_fds(dev, ctx, 0) != 0)
2428 		return RTE_VHOST_MSG_RESULT_ERR;
2429 
2430 	if (ctx->msg.payload.u64 < VIRTIO_MIN_MTU ||
2431 			ctx->msg.payload.u64 > VIRTIO_MAX_MTU) {
2432 		VHOST_LOG_CONFIG(ERR, "(%s) invalid MTU size (%"PRIu64")\n",
2433 				dev->ifname, ctx->msg.payload.u64);
2434 
2435 		return RTE_VHOST_MSG_RESULT_ERR;
2436 	}
2437 
2438 	dev->mtu = ctx->msg.payload.u64;
2439 
2440 	return RTE_VHOST_MSG_RESULT_OK;
2441 }
2442 
2443 static int
2444 vhost_user_set_req_fd(struct virtio_net **pdev,
2445 			struct vhu_msg_context *ctx,
2446 			int main_fd __rte_unused)
2447 {
2448 	struct virtio_net *dev = *pdev;
2449 	int fd = ctx->fds[0];
2450 
2451 	if (validate_msg_fds(dev, ctx, 1) != 0)
2452 		return RTE_VHOST_MSG_RESULT_ERR;
2453 
2454 	if (fd < 0) {
2455 		VHOST_LOG_CONFIG(ERR, "(%s) invalid file descriptor for slave channel (%d)\n",
2456 				dev->ifname, fd);
2457 		return RTE_VHOST_MSG_RESULT_ERR;
2458 	}
2459 
2460 	if (dev->slave_req_fd >= 0)
2461 		close(dev->slave_req_fd);
2462 
2463 	dev->slave_req_fd = fd;
2464 
2465 	return RTE_VHOST_MSG_RESULT_OK;
2466 }
2467 
2468 static int
2469 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2470 {
2471 	struct vhost_vring_addr *ra;
2472 	uint64_t start, end, len;
2473 
2474 	start = imsg->iova;
2475 	end = start + imsg->size;
2476 
2477 	ra = &vq->ring_addrs;
2478 	len = sizeof(struct vring_desc) * vq->size;
2479 	if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2480 		return 1;
2481 
2482 	len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2483 	if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2484 		return 1;
2485 
2486 	len = sizeof(struct vring_used) +
2487 	       sizeof(struct vring_used_elem) * vq->size;
2488 	if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2489 		return 1;
2490 
2491 	if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2492 		len = sizeof(uint64_t);
2493 		if (ra->log_guest_addr < end &&
2494 		    (ra->log_guest_addr + len) > start)
2495 			return 1;
2496 	}
2497 
2498 	return 0;
2499 }
2500 
2501 static int
2502 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2503 {
2504 	struct vhost_vring_addr *ra;
2505 	uint64_t start, end, len;
2506 
2507 	start = imsg->iova;
2508 	end = start + imsg->size;
2509 
2510 	ra = &vq->ring_addrs;
2511 	len = sizeof(struct vring_packed_desc) * vq->size;
2512 	if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2513 		return 1;
2514 
2515 	len = sizeof(struct vring_packed_desc_event);
2516 	if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2517 		return 1;
2518 
2519 	len = sizeof(struct vring_packed_desc_event);
2520 	if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2521 		return 1;
2522 
2523 	if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2524 		len = sizeof(uint64_t);
2525 		if (ra->log_guest_addr < end &&
2526 		    (ra->log_guest_addr + len) > start)
2527 			return 1;
2528 	}
2529 
2530 	return 0;
2531 }
2532 
2533 static int is_vring_iotlb(struct virtio_net *dev,
2534 			  struct vhost_virtqueue *vq,
2535 			  struct vhost_iotlb_msg *imsg)
2536 {
2537 	if (vq_is_packed(dev))
2538 		return is_vring_iotlb_packed(vq, imsg);
2539 	else
2540 		return is_vring_iotlb_split(vq, imsg);
2541 }
2542 
2543 static int
2544 vhost_user_iotlb_msg(struct virtio_net **pdev,
2545 			struct vhu_msg_context *ctx,
2546 			int main_fd __rte_unused)
2547 {
2548 	struct virtio_net *dev = *pdev;
2549 	struct vhost_iotlb_msg *imsg = &ctx->msg.payload.iotlb;
2550 	uint16_t i;
2551 	uint64_t vva, len;
2552 
2553 	if (validate_msg_fds(dev, ctx, 0) != 0)
2554 		return RTE_VHOST_MSG_RESULT_ERR;
2555 
2556 	switch (imsg->type) {
2557 	case VHOST_IOTLB_UPDATE:
2558 		len = imsg->size;
2559 		vva = qva_to_vva(dev, imsg->uaddr, &len);
2560 		if (!vva)
2561 			return RTE_VHOST_MSG_RESULT_ERR;
2562 
2563 		for (i = 0; i < dev->nr_vring; i++) {
2564 			struct vhost_virtqueue *vq = dev->virtqueue[i];
2565 
2566 			if (!vq)
2567 				continue;
2568 
2569 			vhost_user_iotlb_cache_insert(dev, vq, imsg->iova, vva,
2570 					len, imsg->perm);
2571 
2572 			if (is_vring_iotlb(dev, vq, imsg))
2573 				*pdev = dev = translate_ring_addresses(dev, i);
2574 		}
2575 		break;
2576 	case VHOST_IOTLB_INVALIDATE:
2577 		for (i = 0; i < dev->nr_vring; i++) {
2578 			struct vhost_virtqueue *vq = dev->virtqueue[i];
2579 
2580 			if (!vq)
2581 				continue;
2582 
2583 			vhost_user_iotlb_cache_remove(vq, imsg->iova,
2584 					imsg->size);
2585 
2586 			if (is_vring_iotlb(dev, vq, imsg))
2587 				vring_invalidate(dev, vq);
2588 		}
2589 		break;
2590 	default:
2591 		VHOST_LOG_CONFIG(ERR, "(%s) invalid IOTLB message type (%d)\n",
2592 				dev->ifname, imsg->type);
2593 		return RTE_VHOST_MSG_RESULT_ERR;
2594 	}
2595 
2596 	return RTE_VHOST_MSG_RESULT_OK;
2597 }
2598 
2599 static int
2600 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2601 			struct vhu_msg_context *ctx,
2602 			int main_fd __rte_unused)
2603 {
2604 	struct virtio_net *dev = *pdev;
2605 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2606 	struct uffdio_api api_struct;
2607 
2608 	if (validate_msg_fds(dev, ctx, 0) != 0)
2609 		return RTE_VHOST_MSG_RESULT_ERR;
2610 
2611 	dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2612 
2613 	if (dev->postcopy_ufd == -1) {
2614 		VHOST_LOG_CONFIG(ERR, "(%s) userfaultfd not available: %s\n",
2615 			dev->ifname, strerror(errno));
2616 		return RTE_VHOST_MSG_RESULT_ERR;
2617 	}
2618 	api_struct.api = UFFD_API;
2619 	api_struct.features = 0;
2620 	if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2621 		VHOST_LOG_CONFIG(ERR, "(%s) UFFDIO_API ioctl failure: %s\n",
2622 			dev->ifname, strerror(errno));
2623 		close(dev->postcopy_ufd);
2624 		dev->postcopy_ufd = -1;
2625 		return RTE_VHOST_MSG_RESULT_ERR;
2626 	}
2627 	ctx->fds[0] = dev->postcopy_ufd;
2628 	ctx->fd_num = 1;
2629 
2630 	return RTE_VHOST_MSG_RESULT_REPLY;
2631 #else
2632 	dev->postcopy_ufd = -1;
2633 	ctx->fd_num = 0;
2634 
2635 	return RTE_VHOST_MSG_RESULT_ERR;
2636 #endif
2637 }
2638 
2639 static int
2640 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2641 			struct vhu_msg_context *ctx __rte_unused,
2642 			int main_fd __rte_unused)
2643 {
2644 	struct virtio_net *dev = *pdev;
2645 
2646 	if (validate_msg_fds(dev, ctx, 0) != 0)
2647 		return RTE_VHOST_MSG_RESULT_ERR;
2648 
2649 	if (dev->mem && dev->mem->nregions) {
2650 		VHOST_LOG_CONFIG(ERR, "(%s) regions already registered at postcopy-listen\n",
2651 				dev->ifname);
2652 		return RTE_VHOST_MSG_RESULT_ERR;
2653 	}
2654 	dev->postcopy_listening = 1;
2655 
2656 	return RTE_VHOST_MSG_RESULT_OK;
2657 }
2658 
2659 static int
2660 vhost_user_postcopy_end(struct virtio_net **pdev,
2661 			struct vhu_msg_context *ctx,
2662 			int main_fd __rte_unused)
2663 {
2664 	struct virtio_net *dev = *pdev;
2665 
2666 	if (validate_msg_fds(dev, ctx, 0) != 0)
2667 		return RTE_VHOST_MSG_RESULT_ERR;
2668 
2669 	dev->postcopy_listening = 0;
2670 	if (dev->postcopy_ufd >= 0) {
2671 		close(dev->postcopy_ufd);
2672 		dev->postcopy_ufd = -1;
2673 	}
2674 
2675 	ctx->msg.payload.u64 = 0;
2676 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
2677 	ctx->fd_num = 0;
2678 
2679 	return RTE_VHOST_MSG_RESULT_REPLY;
2680 }
2681 
2682 static int
2683 vhost_user_get_status(struct virtio_net **pdev,
2684 		      struct vhu_msg_context *ctx,
2685 		      int main_fd __rte_unused)
2686 {
2687 	struct virtio_net *dev = *pdev;
2688 
2689 	if (validate_msg_fds(dev, ctx, 0) != 0)
2690 		return RTE_VHOST_MSG_RESULT_ERR;
2691 
2692 	ctx->msg.payload.u64 = dev->status;
2693 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
2694 	ctx->fd_num = 0;
2695 
2696 	return RTE_VHOST_MSG_RESULT_REPLY;
2697 }
2698 
2699 static int
2700 vhost_user_set_status(struct virtio_net **pdev,
2701 			struct vhu_msg_context *ctx,
2702 			int main_fd __rte_unused)
2703 {
2704 	struct virtio_net *dev = *pdev;
2705 
2706 	if (validate_msg_fds(dev, ctx, 0) != 0)
2707 		return RTE_VHOST_MSG_RESULT_ERR;
2708 
2709 	/* As per Virtio specification, the device status is 8bits long */
2710 	if (ctx->msg.payload.u64 > UINT8_MAX) {
2711 		VHOST_LOG_CONFIG(ERR, "(%s) invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2712 				dev->ifname, ctx->msg.payload.u64);
2713 		return RTE_VHOST_MSG_RESULT_ERR;
2714 	}
2715 
2716 	dev->status = ctx->msg.payload.u64;
2717 
2718 	if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2719 	    (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2720 		VHOST_LOG_CONFIG(ERR,
2721 				"(%s) FEATURES_OK bit is set but feature negotiation failed\n",
2722 				dev->ifname);
2723 		/*
2724 		 * Clear the bit to let the driver know about the feature
2725 		 * negotiation failure
2726 		 */
2727 		dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2728 	}
2729 
2730 	VHOST_LOG_CONFIG(INFO, "(%s) new device status(0x%08x):\n", dev->ifname,
2731 			dev->status);
2732 	VHOST_LOG_CONFIG(INFO, "(%s)\t-RESET: %u\n", dev->ifname,
2733 			(dev->status == VIRTIO_DEVICE_STATUS_RESET));
2734 	VHOST_LOG_CONFIG(INFO, "(%s)\t-ACKNOWLEDGE: %u\n", dev->ifname,
2735 			!!(dev->status & VIRTIO_DEVICE_STATUS_ACK));
2736 	VHOST_LOG_CONFIG(INFO, "(%s)\t-DRIVER: %u\n", dev->ifname,
2737 			!!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER));
2738 	VHOST_LOG_CONFIG(INFO, "(%s)\t-FEATURES_OK: %u\n", dev->ifname,
2739 			!!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK));
2740 	VHOST_LOG_CONFIG(INFO, "(%s)\t-DRIVER_OK: %u\n", dev->ifname,
2741 			!!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK));
2742 	VHOST_LOG_CONFIG(INFO, "(%s)\t-DEVICE_NEED_RESET: %u\n", dev->ifname,
2743 			!!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET));
2744 	VHOST_LOG_CONFIG(INFO, "(%s)\t-FAILED: %u\n", dev->ifname,
2745 			!!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2746 
2747 	return RTE_VHOST_MSG_RESULT_OK;
2748 }
2749 
2750 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2751 					struct vhu_msg_context *ctx,
2752 					int main_fd);
2753 
2754 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2755 	[VHOST_USER_NONE] = NULL,
2756 	[VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2757 	[VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2758 	[VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2759 	[VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2760 	[VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2761 	[VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2762 	[VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2763 	[VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2764 	[VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2765 	[VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2766 	[VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2767 	[VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2768 	[VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2769 	[VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2770 	[VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2771 	[VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2772 	[VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2773 	[VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2774 	[VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2775 	[VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2776 	[VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2777 	[VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2778 	[VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2779 	[VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2780 	[VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2781 	[VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2782 	[VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2783 	[VHOST_USER_SET_STATUS] = vhost_user_set_status,
2784 	[VHOST_USER_GET_STATUS] = vhost_user_get_status,
2785 };
2786 
2787 /* return bytes# of read on success or negative val on failure. */
2788 static int
2789 read_vhost_message(struct virtio_net *dev, int sockfd, struct  vhu_msg_context *ctx)
2790 {
2791 	int ret;
2792 
2793 	ret = read_fd_message(dev->ifname, sockfd, (char *)&ctx->msg, VHOST_USER_HDR_SIZE,
2794 		ctx->fds, VHOST_MEMORY_MAX_NREGIONS, &ctx->fd_num);
2795 	if (ret <= 0) {
2796 		return ret;
2797 	} else if (ret != VHOST_USER_HDR_SIZE) {
2798 		VHOST_LOG_CONFIG(ERR, "(%s) Unexpected header size read\n", dev->ifname);
2799 		close_msg_fds(ctx);
2800 		return -1;
2801 	}
2802 
2803 	if (ctx->msg.size) {
2804 		if (ctx->msg.size > sizeof(ctx->msg.payload)) {
2805 			VHOST_LOG_CONFIG(ERR, "(%s) invalid msg size: %d\n",
2806 					dev->ifname, ctx->msg.size);
2807 			return -1;
2808 		}
2809 		ret = read(sockfd, &ctx->msg.payload, ctx->msg.size);
2810 		if (ret <= 0)
2811 			return ret;
2812 		if (ret != (int)ctx->msg.size) {
2813 			VHOST_LOG_CONFIG(ERR, "(%s) read control message failed\n", dev->ifname);
2814 			return -1;
2815 		}
2816 	}
2817 
2818 	return ret;
2819 }
2820 
2821 static int
2822 send_vhost_message(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx)
2823 {
2824 	if (!ctx)
2825 		return 0;
2826 
2827 	return send_fd_message(dev->ifname, sockfd, (char *)&ctx->msg,
2828 		VHOST_USER_HDR_SIZE + ctx->msg.size, ctx->fds, ctx->fd_num);
2829 }
2830 
2831 static int
2832 send_vhost_reply(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx)
2833 {
2834 	if (!ctx)
2835 		return 0;
2836 
2837 	ctx->msg.flags &= ~VHOST_USER_VERSION_MASK;
2838 	ctx->msg.flags &= ~VHOST_USER_NEED_REPLY;
2839 	ctx->msg.flags |= VHOST_USER_VERSION;
2840 	ctx->msg.flags |= VHOST_USER_REPLY_MASK;
2841 
2842 	return send_vhost_message(dev, sockfd, ctx);
2843 }
2844 
2845 static int
2846 send_vhost_slave_message(struct virtio_net *dev,
2847 		struct vhu_msg_context *ctx)
2848 {
2849 	int ret;
2850 
2851 	if (ctx->msg.flags & VHOST_USER_NEED_REPLY)
2852 		rte_spinlock_lock(&dev->slave_req_lock);
2853 
2854 	ret = send_vhost_message(dev, dev->slave_req_fd, ctx);
2855 	if (ret < 0 && (ctx->msg.flags & VHOST_USER_NEED_REPLY))
2856 		rte_spinlock_unlock(&dev->slave_req_lock);
2857 
2858 	return ret;
2859 }
2860 
2861 /*
2862  * Allocate a queue pair if it hasn't been allocated yet
2863  */
2864 static int
2865 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2866 			struct vhu_msg_context *ctx)
2867 {
2868 	uint32_t vring_idx;
2869 
2870 	switch (ctx->msg.request.master) {
2871 	case VHOST_USER_SET_VRING_KICK:
2872 	case VHOST_USER_SET_VRING_CALL:
2873 	case VHOST_USER_SET_VRING_ERR:
2874 		vring_idx = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
2875 		break;
2876 	case VHOST_USER_SET_VRING_NUM:
2877 	case VHOST_USER_SET_VRING_BASE:
2878 	case VHOST_USER_GET_VRING_BASE:
2879 	case VHOST_USER_SET_VRING_ENABLE:
2880 		vring_idx = ctx->msg.payload.state.index;
2881 		break;
2882 	case VHOST_USER_SET_VRING_ADDR:
2883 		vring_idx = ctx->msg.payload.addr.index;
2884 		break;
2885 	default:
2886 		return 0;
2887 	}
2888 
2889 	if (vring_idx >= VHOST_MAX_VRING) {
2890 		VHOST_LOG_CONFIG(ERR, "(%s) invalid vring index: %u\n", dev->ifname, vring_idx);
2891 		return -1;
2892 	}
2893 
2894 	if (dev->virtqueue[vring_idx])
2895 		return 0;
2896 
2897 	return alloc_vring_queue(dev, vring_idx);
2898 }
2899 
2900 static void
2901 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2902 {
2903 	unsigned int i = 0;
2904 	unsigned int vq_num = 0;
2905 
2906 	while (vq_num < dev->nr_vring) {
2907 		struct vhost_virtqueue *vq = dev->virtqueue[i];
2908 
2909 		if (vq) {
2910 			rte_spinlock_lock(&vq->access_lock);
2911 			vq_num++;
2912 		}
2913 		i++;
2914 	}
2915 }
2916 
2917 static void
2918 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2919 {
2920 	unsigned int i = 0;
2921 	unsigned int vq_num = 0;
2922 
2923 	while (vq_num < dev->nr_vring) {
2924 		struct vhost_virtqueue *vq = dev->virtqueue[i];
2925 
2926 		if (vq) {
2927 			rte_spinlock_unlock(&vq->access_lock);
2928 			vq_num++;
2929 		}
2930 		i++;
2931 	}
2932 }
2933 
2934 int
2935 vhost_user_msg_handler(int vid, int fd)
2936 {
2937 	struct virtio_net *dev;
2938 	struct vhu_msg_context ctx;
2939 	struct rte_vdpa_device *vdpa_dev;
2940 	int ret;
2941 	int unlock_required = 0;
2942 	bool handled;
2943 	int request;
2944 	uint32_t i;
2945 
2946 	dev = get_device(vid);
2947 	if (dev == NULL)
2948 		return -1;
2949 
2950 	if (!dev->notify_ops) {
2951 		dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2952 		if (!dev->notify_ops) {
2953 			VHOST_LOG_CONFIG(ERR, "(%s) failed to get callback ops for driver\n",
2954 				dev->ifname);
2955 			return -1;
2956 		}
2957 	}
2958 
2959 	ret = read_vhost_message(dev, fd, &ctx);
2960 	if (ret <= 0) {
2961 		if (ret < 0)
2962 			VHOST_LOG_CONFIG(ERR, "(%s) vhost read message failed\n", dev->ifname);
2963 		else
2964 			VHOST_LOG_CONFIG(INFO, "(%s) vhost peer closed\n", dev->ifname);
2965 
2966 		return -1;
2967 	}
2968 
2969 	ret = 0;
2970 	request = ctx.msg.request.master;
2971 	if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2972 			vhost_message_str[request]) {
2973 		if (request != VHOST_USER_IOTLB_MSG)
2974 			VHOST_LOG_CONFIG(INFO, "(%s) read message %s\n",
2975 				dev->ifname, vhost_message_str[request]);
2976 		else
2977 			VHOST_LOG_CONFIG(DEBUG, "(%s) read message %s\n",
2978 				dev->ifname, vhost_message_str[request]);
2979 	} else {
2980 		VHOST_LOG_CONFIG(DEBUG, "(%s) external request %d\n", dev->ifname, request);
2981 	}
2982 
2983 	ret = vhost_user_check_and_alloc_queue_pair(dev, &ctx);
2984 	if (ret < 0) {
2985 		VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc queue\n", dev->ifname);
2986 		return -1;
2987 	}
2988 
2989 	/*
2990 	 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2991 	 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2992 	 * and device is destroyed. destroy_device waits for queues to be
2993 	 * inactive, so it is safe. Otherwise taking the access_lock
2994 	 * would cause a dead lock.
2995 	 */
2996 	switch (request) {
2997 	case VHOST_USER_SET_FEATURES:
2998 	case VHOST_USER_SET_PROTOCOL_FEATURES:
2999 	case VHOST_USER_SET_OWNER:
3000 	case VHOST_USER_SET_MEM_TABLE:
3001 	case VHOST_USER_SET_LOG_BASE:
3002 	case VHOST_USER_SET_LOG_FD:
3003 	case VHOST_USER_SET_VRING_NUM:
3004 	case VHOST_USER_SET_VRING_ADDR:
3005 	case VHOST_USER_SET_VRING_BASE:
3006 	case VHOST_USER_SET_VRING_KICK:
3007 	case VHOST_USER_SET_VRING_CALL:
3008 	case VHOST_USER_SET_VRING_ERR:
3009 	case VHOST_USER_SET_VRING_ENABLE:
3010 	case VHOST_USER_SEND_RARP:
3011 	case VHOST_USER_NET_SET_MTU:
3012 	case VHOST_USER_SET_SLAVE_REQ_FD:
3013 		if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3014 			vhost_user_lock_all_queue_pairs(dev);
3015 			unlock_required = 1;
3016 		}
3017 		break;
3018 	default:
3019 		break;
3020 
3021 	}
3022 
3023 	handled = false;
3024 	if (dev->extern_ops.pre_msg_handle) {
3025 		ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
3026 				(void *)&ctx.msg);
3027 		switch (ret) {
3028 		case RTE_VHOST_MSG_RESULT_REPLY:
3029 			send_vhost_reply(dev, fd, &ctx);
3030 			/* Fall-through */
3031 		case RTE_VHOST_MSG_RESULT_ERR:
3032 		case RTE_VHOST_MSG_RESULT_OK:
3033 			handled = true;
3034 			goto skip_to_post_handle;
3035 		case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3036 		default:
3037 			break;
3038 		}
3039 	}
3040 
3041 	if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
3042 		if (!vhost_message_handlers[request])
3043 			goto skip_to_post_handle;
3044 		ret = vhost_message_handlers[request](&dev, &ctx, fd);
3045 
3046 		switch (ret) {
3047 		case RTE_VHOST_MSG_RESULT_ERR:
3048 			VHOST_LOG_CONFIG(ERR, "(%s) processing %s failed.\n",
3049 					dev->ifname, vhost_message_str[request]);
3050 			handled = true;
3051 			break;
3052 		case RTE_VHOST_MSG_RESULT_OK:
3053 			VHOST_LOG_CONFIG(DEBUG, "(%s) processing %s succeeded.\n",
3054 					dev->ifname, vhost_message_str[request]);
3055 			handled = true;
3056 			break;
3057 		case RTE_VHOST_MSG_RESULT_REPLY:
3058 			VHOST_LOG_CONFIG(DEBUG, "(%s) processing %s succeeded and needs reply.\n",
3059 					dev->ifname, vhost_message_str[request]);
3060 			send_vhost_reply(dev, fd, &ctx);
3061 			handled = true;
3062 			break;
3063 		default:
3064 			break;
3065 		}
3066 	}
3067 
3068 skip_to_post_handle:
3069 	if (ret != RTE_VHOST_MSG_RESULT_ERR &&
3070 			dev->extern_ops.post_msg_handle) {
3071 		ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
3072 				(void *)&ctx.msg);
3073 		switch (ret) {
3074 		case RTE_VHOST_MSG_RESULT_REPLY:
3075 			send_vhost_reply(dev, fd, &ctx);
3076 			/* Fall-through */
3077 		case RTE_VHOST_MSG_RESULT_ERR:
3078 		case RTE_VHOST_MSG_RESULT_OK:
3079 			handled = true;
3080 		case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3081 		default:
3082 			break;
3083 		}
3084 	}
3085 
3086 	/* If message was not handled at this stage, treat it as an error */
3087 	if (!handled) {
3088 		VHOST_LOG_CONFIG(ERR, "(%s) vhost message (req: %d) was not handled.\n",
3089 				dev->ifname, request);
3090 		close_msg_fds(&ctx);
3091 		ret = RTE_VHOST_MSG_RESULT_ERR;
3092 	}
3093 
3094 	/*
3095 	 * If the request required a reply that was already sent,
3096 	 * this optional reply-ack won't be sent as the
3097 	 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
3098 	 */
3099 	if (ctx.msg.flags & VHOST_USER_NEED_REPLY) {
3100 		ctx.msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
3101 		ctx.msg.size = sizeof(ctx.msg.payload.u64);
3102 		ctx.fd_num = 0;
3103 		send_vhost_reply(dev, fd, &ctx);
3104 	} else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
3105 		VHOST_LOG_CONFIG(ERR, "(%s) vhost message handling failed.\n", dev->ifname);
3106 		return -1;
3107 	}
3108 
3109 	for (i = 0; i < dev->nr_vring; i++) {
3110 		struct vhost_virtqueue *vq = dev->virtqueue[i];
3111 		bool cur_ready = vq_is_ready(dev, vq);
3112 
3113 		if (cur_ready != (vq && vq->ready)) {
3114 			vq->ready = cur_ready;
3115 			vhost_user_notify_queue_state(dev, i, cur_ready);
3116 		}
3117 	}
3118 
3119 	if (unlock_required)
3120 		vhost_user_unlock_all_queue_pairs(dev);
3121 
3122 	if (!virtio_is_ready(dev))
3123 		goto out;
3124 
3125 	/*
3126 	 * Virtio is now ready. If not done already, it is time
3127 	 * to notify the application it can process the rings and
3128 	 * configure the vDPA device if present.
3129 	 */
3130 
3131 	if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
3132 		if (dev->notify_ops->new_device(dev->vid) == 0)
3133 			dev->flags |= VIRTIO_DEV_RUNNING;
3134 	}
3135 
3136 	vdpa_dev = dev->vdpa_dev;
3137 	if (!vdpa_dev)
3138 		goto out;
3139 
3140 	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3141 		if (vdpa_dev->ops->dev_conf(dev->vid))
3142 			VHOST_LOG_CONFIG(ERR, "(%s) failed to configure vDPA device\n",
3143 					dev->ifname);
3144 		else
3145 			dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3146 	}
3147 
3148 out:
3149 	return 0;
3150 }
3151 
3152 static int process_slave_message_reply(struct virtio_net *dev,
3153 				       const struct vhu_msg_context *ctx)
3154 {
3155 	struct vhu_msg_context msg_reply;
3156 	int ret;
3157 
3158 	if ((ctx->msg.flags & VHOST_USER_NEED_REPLY) == 0)
3159 		return 0;
3160 
3161 	ret = read_vhost_message(dev, dev->slave_req_fd, &msg_reply);
3162 	if (ret <= 0) {
3163 		if (ret < 0)
3164 			VHOST_LOG_CONFIG(ERR, "(%s) vhost read slave message reply failed\n",
3165 					dev->ifname);
3166 		else
3167 			VHOST_LOG_CONFIG(INFO, "(%s) vhost peer closed\n", dev->ifname);
3168 		ret = -1;
3169 		goto out;
3170 	}
3171 
3172 	ret = 0;
3173 	if (msg_reply.msg.request.slave != ctx->msg.request.slave) {
3174 		VHOST_LOG_CONFIG(ERR, "(%s) received unexpected msg type (%u), expected %u\n",
3175 				dev->ifname, msg_reply.msg.request.slave, ctx->msg.request.slave);
3176 		ret = -1;
3177 		goto out;
3178 	}
3179 
3180 	ret = msg_reply.msg.payload.u64 ? -1 : 0;
3181 
3182 out:
3183 	rte_spinlock_unlock(&dev->slave_req_lock);
3184 	return ret;
3185 }
3186 
3187 int
3188 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3189 {
3190 	int ret;
3191 	struct vhu_msg_context ctx = {
3192 		.msg = {
3193 			.request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3194 			.flags = VHOST_USER_VERSION,
3195 			.size = sizeof(ctx.msg.payload.iotlb),
3196 			.payload.iotlb = {
3197 				.iova = iova,
3198 				.perm = perm,
3199 				.type = VHOST_IOTLB_MISS,
3200 			},
3201 		},
3202 	};
3203 
3204 	ret = send_vhost_message(dev, dev->slave_req_fd, &ctx);
3205 	if (ret < 0) {
3206 		VHOST_LOG_CONFIG(ERR, "(%s) failed to send IOTLB miss message (%d)\n",
3207 				dev->ifname, ret);
3208 		return ret;
3209 	}
3210 
3211 	return 0;
3212 }
3213 
3214 static int
3215 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3216 {
3217 	int ret;
3218 	struct vhu_msg_context ctx = {
3219 		.msg = {
3220 			.request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3221 			.flags = VHOST_USER_VERSION,
3222 			.size = 0,
3223 		}
3224 	};
3225 
3226 	if (need_reply)
3227 		ctx.msg.flags |= VHOST_USER_NEED_REPLY;
3228 
3229 	ret = send_vhost_slave_message(dev, &ctx);
3230 	if (ret < 0) {
3231 		VHOST_LOG_CONFIG(ERR, "(%s) failed to send config change (%d)\n",
3232 				dev->ifname, ret);
3233 		return ret;
3234 	}
3235 
3236 	return process_slave_message_reply(dev, &ctx);
3237 }
3238 
3239 int
3240 rte_vhost_slave_config_change(int vid, bool need_reply)
3241 {
3242 	struct virtio_net *dev;
3243 
3244 	dev = get_device(vid);
3245 	if (!dev)
3246 		return -ENODEV;
3247 
3248 	return vhost_user_slave_config_change(dev, need_reply);
3249 }
3250 
3251 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3252 						    int index, int fd,
3253 						    uint64_t offset,
3254 						    uint64_t size)
3255 {
3256 	int ret;
3257 	struct vhu_msg_context ctx = {
3258 		.msg = {
3259 			.request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3260 			.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3261 			.size = sizeof(ctx.msg.payload.area),
3262 			.payload.area = {
3263 				.u64 = index & VHOST_USER_VRING_IDX_MASK,
3264 				.size = size,
3265 				.offset = offset,
3266 			},
3267 		},
3268 	};
3269 
3270 	if (fd < 0)
3271 		ctx.msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3272 	else {
3273 		ctx.fds[0] = fd;
3274 		ctx.fd_num = 1;
3275 	}
3276 
3277 	ret = send_vhost_slave_message(dev, &ctx);
3278 	if (ret < 0) {
3279 		VHOST_LOG_CONFIG(ERR, "(%s) failed to set host notifier (%d)\n",
3280 				dev->ifname, ret);
3281 		return ret;
3282 	}
3283 
3284 	return process_slave_message_reply(dev, &ctx);
3285 }
3286 
3287 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3288 {
3289 	struct virtio_net *dev;
3290 	struct rte_vdpa_device *vdpa_dev;
3291 	int vfio_device_fd, ret = 0;
3292 	uint64_t offset, size;
3293 	unsigned int i, q_start, q_last;
3294 
3295 	dev = get_device(vid);
3296 	if (!dev)
3297 		return -ENODEV;
3298 
3299 	vdpa_dev = dev->vdpa_dev;
3300 	if (vdpa_dev == NULL)
3301 		return -ENODEV;
3302 
3303 	if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3304 	    !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3305 	    !(dev->protocol_features &
3306 			(1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3307 	    !(dev->protocol_features &
3308 			(1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3309 	    !(dev->protocol_features &
3310 			(1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3311 		return -ENOTSUP;
3312 
3313 	if (qid == RTE_VHOST_QUEUE_ALL) {
3314 		q_start = 0;
3315 		q_last = dev->nr_vring - 1;
3316 	} else {
3317 		if (qid >= dev->nr_vring)
3318 			return -EINVAL;
3319 		q_start = qid;
3320 		q_last = qid;
3321 	}
3322 
3323 	RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3324 	RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3325 
3326 	vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3327 	if (vfio_device_fd < 0)
3328 		return -ENOTSUP;
3329 
3330 	if (enable) {
3331 		for (i = q_start; i <= q_last; i++) {
3332 			if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3333 					&size) < 0) {
3334 				ret = -ENOTSUP;
3335 				goto disable;
3336 			}
3337 
3338 			if (vhost_user_slave_set_vring_host_notifier(dev, i,
3339 					vfio_device_fd, offset, size) < 0) {
3340 				ret = -EFAULT;
3341 				goto disable;
3342 			}
3343 		}
3344 	} else {
3345 disable:
3346 		for (i = q_start; i <= q_last; i++) {
3347 			vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3348 					0, 0);
3349 		}
3350 	}
3351 
3352 	return ret;
3353 }
3354