xref: /dpdk/lib/vhost/vhost_user.c (revision 9ad3a41ab2a10db0059e1decdbf3ec038f348e08)
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 void
144 async_dma_map(struct virtio_net *dev, bool do_map)
145 {
146 	int ret = 0;
147 	uint32_t i;
148 	struct guest_page *page;
149 
150 	if (do_map) {
151 		for (i = 0; i < dev->nr_guest_pages; i++) {
152 			page = &dev->guest_pages[i];
153 			ret = rte_vfio_container_dma_map(RTE_VFIO_DEFAULT_CONTAINER_FD,
154 							 page->host_user_addr,
155 							 page->host_iova,
156 							 page->size);
157 			if (ret) {
158 				/*
159 				 * DMA device may bind with kernel driver, in this case,
160 				 * we don't need to program IOMMU manually. However, if no
161 				 * device is bound with vfio/uio in DPDK, and vfio kernel
162 				 * module is loaded, the API will still be called and return
163 				 * with ENODEV.
164 				 *
165 				 * DPDK vfio only returns ENODEV in very similar situations
166 				 * (vfio either unsupported, or supported but no devices found).
167 				 * Either way, no mappings could be performed. We treat it as
168 				 * normal case in async path. This is a workaround.
169 				 */
170 				if (rte_errno == ENODEV)
171 					return;
172 
173 				/* DMA mapping errors won't stop VHOST_USER_SET_MEM_TABLE. */
174 				VHOST_LOG_CONFIG(ERR, "DMA engine map failed\n");
175 			}
176 		}
177 
178 	} else {
179 		for (i = 0; i < dev->nr_guest_pages; i++) {
180 			page = &dev->guest_pages[i];
181 			ret = rte_vfio_container_dma_unmap(RTE_VFIO_DEFAULT_CONTAINER_FD,
182 							   page->host_user_addr,
183 							   page->host_iova,
184 							   page->size);
185 			if (ret) {
186 				/* like DMA map, ignore the kernel driver case when unmap. */
187 				if (rte_errno == EINVAL)
188 					return;
189 
190 				VHOST_LOG_CONFIG(ERR, "DMA engine unmap failed\n");
191 			}
192 		}
193 	}
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 	if (dev->async_copy && rte_vfio_is_enabled("vfio"))
206 		async_dma_map(dev, false);
207 
208 	for (i = 0; i < dev->mem->nregions; i++) {
209 		reg = &dev->mem->regions[i];
210 		if (reg->host_user_addr) {
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_iova, uint64_t host_user_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, "cannot realloc guest_pages\n");
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_iova == last_page->host_iova + last_page->size &&
1011 		    guest_phys_addr == last_page->guest_phys_addr + last_page->size &&
1012 		    host_user_addr == last_page->host_user_addr + last_page->size) {
1013 			last_page->size += size;
1014 			return 0;
1015 		}
1016 	}
1017 
1018 	page = &dev->guest_pages[dev->nr_guest_pages++];
1019 	page->guest_phys_addr = guest_phys_addr;
1020 	page->host_iova  = host_iova;
1021 	page->host_user_addr = host_user_addr;
1022 	page->size = size;
1023 
1024 	return 0;
1025 }
1026 
1027 static int
1028 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
1029 		uint64_t page_size)
1030 {
1031 	uint64_t reg_size = reg->size;
1032 	uint64_t host_user_addr  = reg->host_user_addr;
1033 	uint64_t guest_phys_addr = reg->guest_phys_addr;
1034 	uint64_t host_iova;
1035 	uint64_t size;
1036 
1037 	host_iova = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
1038 	size = page_size - (guest_phys_addr & (page_size - 1));
1039 	size = RTE_MIN(size, reg_size);
1040 
1041 	if (add_one_guest_page(dev, guest_phys_addr, host_iova,
1042 			       host_user_addr, size) < 0)
1043 		return -1;
1044 
1045 	host_user_addr  += size;
1046 	guest_phys_addr += size;
1047 	reg_size -= size;
1048 
1049 	while (reg_size > 0) {
1050 		size = RTE_MIN(reg_size, page_size);
1051 		host_iova = rte_mem_virt2iova((void *)(uintptr_t)
1052 						  host_user_addr);
1053 		if (add_one_guest_page(dev, guest_phys_addr, host_iova,
1054 				       host_user_addr, size) < 0)
1055 			return -1;
1056 
1057 		host_user_addr  += size;
1058 		guest_phys_addr += size;
1059 		reg_size -= size;
1060 	}
1061 
1062 	/* sort guest page array if over binary search threshold */
1063 	if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
1064 		qsort((void *)dev->guest_pages, dev->nr_guest_pages,
1065 			sizeof(struct guest_page), guest_page_addrcmp);
1066 	}
1067 
1068 	return 0;
1069 }
1070 
1071 #ifdef RTE_LIBRTE_VHOST_DEBUG
1072 /* TODO: enable it only in debug mode? */
1073 static void
1074 dump_guest_pages(struct virtio_net *dev)
1075 {
1076 	uint32_t i;
1077 	struct guest_page *page;
1078 
1079 	for (i = 0; i < dev->nr_guest_pages; i++) {
1080 		page = &dev->guest_pages[i];
1081 
1082 		VHOST_LOG_CONFIG(INFO, "(%s) guest physical page region %u\n",
1083 				dev->ifname, i);
1084 		VHOST_LOG_CONFIG(INFO, "(%s)\tguest_phys_addr: %" PRIx64 "\n",
1085 				dev->ifname, page->guest_phys_addr);
1086 		VHOST_LOG_CONFIG(INFO, "(%s)\thost_iova : %" PRIx64 "\n",
1087 				dev->ifname, page->host_iova);
1088 		VHOST_LOG_CONFIG(INFO, "(%s)\tsize           : %" PRIx64 "\n",
1089 				dev->ifname, page->size);
1090 	}
1091 }
1092 #else
1093 #define dump_guest_pages(dev)
1094 #endif
1095 
1096 static bool
1097 vhost_memory_changed(struct VhostUserMemory *new,
1098 		     struct rte_vhost_memory *old)
1099 {
1100 	uint32_t i;
1101 
1102 	if (new->nregions != old->nregions)
1103 		return true;
1104 
1105 	for (i = 0; i < new->nregions; ++i) {
1106 		VhostUserMemoryRegion *new_r = &new->regions[i];
1107 		struct rte_vhost_mem_region *old_r = &old->regions[i];
1108 
1109 		if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1110 			return true;
1111 		if (new_r->memory_size != old_r->size)
1112 			return true;
1113 		if (new_r->userspace_addr != old_r->guest_user_addr)
1114 			return true;
1115 	}
1116 
1117 	return false;
1118 }
1119 
1120 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1121 static int
1122 vhost_user_postcopy_region_register(struct virtio_net *dev,
1123 		struct rte_vhost_mem_region *reg)
1124 {
1125 	struct uffdio_register reg_struct;
1126 
1127 	/*
1128 	 * Let's register all the mmapped area to ensure
1129 	 * alignment on page boundary.
1130 	 */
1131 	reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1132 	reg_struct.range.len = reg->mmap_size;
1133 	reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1134 
1135 	if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1136 				&reg_struct)) {
1137 		VHOST_LOG_CONFIG(ERR, "(%s) failed to register ufd for region "
1138 				"%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1139 				dev->ifname,
1140 				(uint64_t)reg_struct.range.start,
1141 				(uint64_t)reg_struct.range.start +
1142 				(uint64_t)reg_struct.range.len - 1,
1143 				dev->postcopy_ufd,
1144 				strerror(errno));
1145 		return -1;
1146 	}
1147 
1148 	VHOST_LOG_CONFIG(INFO,
1149 			"(%s)\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1150 			dev->ifname,
1151 			(uint64_t)reg_struct.range.start,
1152 			(uint64_t)reg_struct.range.start +
1153 			(uint64_t)reg_struct.range.len - 1);
1154 
1155 	return 0;
1156 }
1157 #else
1158 static int
1159 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1160 		struct rte_vhost_mem_region *reg __rte_unused)
1161 {
1162 	return -1;
1163 }
1164 #endif
1165 
1166 static int
1167 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1168 		struct vhu_msg_context *ctx)
1169 {
1170 	struct VhostUserMemory *memory;
1171 	struct rte_vhost_mem_region *reg;
1172 	struct vhu_msg_context ack_ctx;
1173 	uint32_t i;
1174 
1175 	if (!dev->postcopy_listening)
1176 		return 0;
1177 
1178 	/*
1179 	 * We haven't a better way right now than sharing
1180 	 * DPDK's virtual address with Qemu, so that Qemu can
1181 	 * retrieve the region offset when handling userfaults.
1182 	 */
1183 	memory = &ctx->msg.payload.memory;
1184 	for (i = 0; i < memory->nregions; i++) {
1185 		reg = &dev->mem->regions[i];
1186 		memory->regions[i].userspace_addr = reg->host_user_addr;
1187 	}
1188 
1189 	/* Send the addresses back to qemu */
1190 	ctx->fd_num = 0;
1191 	send_vhost_reply(dev, main_fd, ctx);
1192 
1193 	/* Wait for qemu to acknowledge it got the addresses
1194 	 * we've got to wait before we're allowed to generate faults.
1195 	 */
1196 	if (read_vhost_message(dev, main_fd, &ack_ctx) <= 0) {
1197 		VHOST_LOG_CONFIG(ERR, "(%s) failed to read qemu ack on postcopy set-mem-table\n",
1198 				dev->ifname);
1199 		return -1;
1200 	}
1201 
1202 	if (validate_msg_fds(dev, &ack_ctx, 0) != 0)
1203 		return -1;
1204 
1205 	if (ack_ctx.msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1206 		VHOST_LOG_CONFIG(ERR, "(%s) bad qemu ack on postcopy set-mem-table (%d)\n",
1207 				dev->ifname, ack_ctx.msg.request.master);
1208 		return -1;
1209 	}
1210 
1211 	/* Now userfault register and we can use the memory */
1212 	for (i = 0; i < memory->nregions; i++) {
1213 		reg = &dev->mem->regions[i];
1214 		if (vhost_user_postcopy_region_register(dev, reg) < 0)
1215 			return -1;
1216 	}
1217 
1218 	return 0;
1219 }
1220 
1221 static int
1222 vhost_user_mmap_region(struct virtio_net *dev,
1223 		struct rte_vhost_mem_region *region,
1224 		uint64_t mmap_offset)
1225 {
1226 	void *mmap_addr;
1227 	uint64_t mmap_size;
1228 	uint64_t alignment;
1229 	int populate;
1230 
1231 	/* Check for memory_size + mmap_offset overflow */
1232 	if (mmap_offset >= -region->size) {
1233 		VHOST_LOG_CONFIG(ERR, "(%s) mmap_offset (%#"PRIx64") and memory_size (%#"PRIx64") overflow\n",
1234 				dev->ifname, mmap_offset, region->size);
1235 		return -1;
1236 	}
1237 
1238 	mmap_size = region->size + mmap_offset;
1239 
1240 	/* mmap() without flag of MAP_ANONYMOUS, should be called with length
1241 	 * argument aligned with hugepagesz at older longterm version Linux,
1242 	 * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1243 	 *
1244 	 * To avoid failure, make sure in caller to keep length aligned.
1245 	 */
1246 	alignment = get_blk_size(region->fd);
1247 	if (alignment == (uint64_t)-1) {
1248 		VHOST_LOG_CONFIG(ERR, "(%s) couldn't get hugepage size through fstat\n",
1249 				dev->ifname);
1250 		return -1;
1251 	}
1252 	mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1253 	if (mmap_size == 0) {
1254 		/*
1255 		 * It could happen if initial mmap_size + alignment overflows
1256 		 * the sizeof uint64, which could happen if either mmap_size or
1257 		 * alignment value is wrong.
1258 		 *
1259 		 * mmap() kernel implementation would return an error, but
1260 		 * better catch it before and provide useful info in the logs.
1261 		 */
1262 		VHOST_LOG_CONFIG(ERR, "(%s) mmap size (0x%" PRIx64 ") or alignment (0x%" PRIx64 ") is invalid\n",
1263 				dev->ifname, region->size + mmap_offset, alignment);
1264 		return -1;
1265 	}
1266 
1267 	populate = dev->async_copy ? MAP_POPULATE : 0;
1268 	mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1269 			MAP_SHARED | populate, region->fd, 0);
1270 
1271 	if (mmap_addr == MAP_FAILED) {
1272 		VHOST_LOG_CONFIG(ERR, "(%s) mmap failed (%s).\n", dev->ifname, strerror(errno));
1273 		return -1;
1274 	}
1275 
1276 	region->mmap_addr = mmap_addr;
1277 	region->mmap_size = mmap_size;
1278 	region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1279 
1280 	if (dev->async_copy) {
1281 		if (add_guest_pages(dev, region, alignment) < 0) {
1282 			VHOST_LOG_CONFIG(ERR, "(%s) adding guest pages to region failed.\n",
1283 					dev->ifname);
1284 			return -1;
1285 		}
1286 	}
1287 
1288 	VHOST_LOG_CONFIG(INFO, "(%s) guest memory region size: 0x%" PRIx64 "\n",
1289 			dev->ifname, region->size);
1290 	VHOST_LOG_CONFIG(INFO, "(%s)\t guest physical addr: 0x%" PRIx64 "\n",
1291 			dev->ifname, region->guest_phys_addr);
1292 	VHOST_LOG_CONFIG(INFO, "(%s)\t guest virtual  addr: 0x%" PRIx64 "\n",
1293 			dev->ifname, region->guest_user_addr);
1294 	VHOST_LOG_CONFIG(INFO, "(%s)\t host  virtual  addr: 0x%" PRIx64 "\n",
1295 			dev->ifname, region->host_user_addr);
1296 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap addr : 0x%" PRIx64 "\n",
1297 			dev->ifname, (uint64_t)(uintptr_t)mmap_addr);
1298 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap size : 0x%" PRIx64 "\n",
1299 			dev->ifname, mmap_size);
1300 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap align: 0x%" PRIx64 "\n",
1301 			dev->ifname, alignment);
1302 	VHOST_LOG_CONFIG(INFO, "(%s)\t mmap off  : 0x%" PRIx64 "\n",
1303 			dev->ifname, mmap_offset);
1304 
1305 	return 0;
1306 }
1307 
1308 static int
1309 vhost_user_set_mem_table(struct virtio_net **pdev,
1310 			struct vhu_msg_context *ctx,
1311 			int main_fd)
1312 {
1313 	struct virtio_net *dev = *pdev;
1314 	struct VhostUserMemory *memory = &ctx->msg.payload.memory;
1315 	struct rte_vhost_mem_region *reg;
1316 	int numa_node = SOCKET_ID_ANY;
1317 	uint64_t mmap_offset;
1318 	uint32_t i;
1319 	bool async_notify = false;
1320 
1321 	if (validate_msg_fds(dev, ctx, memory->nregions) != 0)
1322 		return RTE_VHOST_MSG_RESULT_ERR;
1323 
1324 	if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1325 		VHOST_LOG_CONFIG(ERR, "(%s) too many memory regions (%u)\n",
1326 				dev->ifname, memory->nregions);
1327 		goto close_msg_fds;
1328 	}
1329 
1330 	if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1331 		VHOST_LOG_CONFIG(INFO, "(%s) memory regions not changed\n", dev->ifname);
1332 
1333 		close_msg_fds(ctx);
1334 
1335 		return RTE_VHOST_MSG_RESULT_OK;
1336 	}
1337 
1338 	if (dev->mem) {
1339 		if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1340 			struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1341 
1342 			if (vdpa_dev && vdpa_dev->ops->dev_close)
1343 				vdpa_dev->ops->dev_close(dev->vid);
1344 			dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1345 		}
1346 
1347 		/* notify the vhost application to stop DMA transfers */
1348 		if (dev->async_copy && dev->notify_ops->vring_state_changed) {
1349 			for (i = 0; i < dev->nr_vring; i++) {
1350 				dev->notify_ops->vring_state_changed(dev->vid,
1351 						i, 0);
1352 			}
1353 			async_notify = true;
1354 		}
1355 
1356 		free_mem_region(dev);
1357 		rte_free(dev->mem);
1358 		dev->mem = NULL;
1359 	}
1360 
1361 	/* Flush IOTLB cache as previous HVAs are now invalid */
1362 	if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1363 		for (i = 0; i < dev->nr_vring; i++)
1364 			vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1365 
1366 	/*
1367 	 * If VQ 0 has already been allocated, try to allocate on the same
1368 	 * NUMA node. It can be reallocated later in numa_realloc().
1369 	 */
1370 	if (dev->nr_vring > 0)
1371 		numa_node = dev->virtqueue[0]->numa_node;
1372 
1373 	dev->nr_guest_pages = 0;
1374 	if (dev->guest_pages == NULL) {
1375 		dev->max_guest_pages = 8;
1376 		dev->guest_pages = rte_zmalloc_socket(NULL,
1377 					dev->max_guest_pages *
1378 					sizeof(struct guest_page),
1379 					RTE_CACHE_LINE_SIZE,
1380 					numa_node);
1381 		if (dev->guest_pages == NULL) {
1382 			VHOST_LOG_CONFIG(ERR,
1383 				"(%s) failed to allocate memory for dev->guest_pages\n",
1384 				dev->ifname);
1385 			goto close_msg_fds;
1386 		}
1387 	}
1388 
1389 	dev->mem = rte_zmalloc_socket("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1390 		sizeof(struct rte_vhost_mem_region) * memory->nregions, 0, numa_node);
1391 	if (dev->mem == NULL) {
1392 		VHOST_LOG_CONFIG(ERR,
1393 			"(%s) failed to allocate memory for dev->mem\n",
1394 			dev->ifname);
1395 		goto free_guest_pages;
1396 	}
1397 
1398 	for (i = 0; i < memory->nregions; i++) {
1399 		reg = &dev->mem->regions[i];
1400 
1401 		reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1402 		reg->guest_user_addr = memory->regions[i].userspace_addr;
1403 		reg->size            = memory->regions[i].memory_size;
1404 		reg->fd              = ctx->fds[i];
1405 
1406 		/*
1407 		 * Assign invalid file descriptor value to avoid double
1408 		 * closing on error path.
1409 		 */
1410 		ctx->fds[i] = -1;
1411 
1412 		mmap_offset = memory->regions[i].mmap_offset;
1413 
1414 		if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1415 			VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap region %u\n", dev->ifname, i);
1416 			goto free_mem_table;
1417 		}
1418 
1419 		dev->mem->nregions++;
1420 	}
1421 
1422 	if (dev->async_copy && rte_vfio_is_enabled("vfio"))
1423 		async_dma_map(dev, true);
1424 
1425 	if (vhost_user_postcopy_register(dev, main_fd, ctx) < 0)
1426 		goto free_mem_table;
1427 
1428 	for (i = 0; i < dev->nr_vring; i++) {
1429 		struct vhost_virtqueue *vq = dev->virtqueue[i];
1430 
1431 		if (!vq)
1432 			continue;
1433 
1434 		if (vq->desc || vq->avail || vq->used) {
1435 			/*
1436 			 * If the memory table got updated, the ring addresses
1437 			 * need to be translated again as virtual addresses have
1438 			 * changed.
1439 			 */
1440 			vring_invalidate(dev, vq);
1441 
1442 			dev = translate_ring_addresses(dev, i);
1443 			if (!dev) {
1444 				dev = *pdev;
1445 				goto free_mem_table;
1446 			}
1447 
1448 			*pdev = dev;
1449 		}
1450 	}
1451 
1452 	dump_guest_pages(dev);
1453 
1454 	if (async_notify) {
1455 		for (i = 0; i < dev->nr_vring; i++)
1456 			dev->notify_ops->vring_state_changed(dev->vid, i, 1);
1457 	}
1458 
1459 	return RTE_VHOST_MSG_RESULT_OK;
1460 
1461 free_mem_table:
1462 	free_mem_region(dev);
1463 	rte_free(dev->mem);
1464 	dev->mem = NULL;
1465 
1466 free_guest_pages:
1467 	rte_free(dev->guest_pages);
1468 	dev->guest_pages = NULL;
1469 close_msg_fds:
1470 	close_msg_fds(ctx);
1471 	return RTE_VHOST_MSG_RESULT_ERR;
1472 }
1473 
1474 static bool
1475 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1476 {
1477 	bool rings_ok;
1478 
1479 	if (!vq)
1480 		return false;
1481 
1482 	if (vq_is_packed(dev))
1483 		rings_ok = vq->desc_packed && vq->driver_event &&
1484 			vq->device_event;
1485 	else
1486 		rings_ok = vq->desc && vq->avail && vq->used;
1487 
1488 	return rings_ok &&
1489 	       vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1490 	       vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1491 	       vq->enabled;
1492 }
1493 
1494 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1495 
1496 static int
1497 virtio_is_ready(struct virtio_net *dev)
1498 {
1499 	struct vhost_virtqueue *vq;
1500 	uint32_t i, nr_vring = dev->nr_vring;
1501 
1502 	if (dev->flags & VIRTIO_DEV_READY)
1503 		return 1;
1504 
1505 	if (!dev->nr_vring)
1506 		return 0;
1507 
1508 	if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1509 		nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1510 
1511 		if (dev->nr_vring < nr_vring)
1512 			return 0;
1513 	}
1514 
1515 	for (i = 0; i < nr_vring; i++) {
1516 		vq = dev->virtqueue[i];
1517 
1518 		if (!vq_is_ready(dev, vq))
1519 			return 0;
1520 	}
1521 
1522 	/* If supported, ensure the frontend is really done with config */
1523 	if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1524 		if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1525 			return 0;
1526 
1527 	dev->flags |= VIRTIO_DEV_READY;
1528 
1529 	if (!(dev->flags & VIRTIO_DEV_RUNNING))
1530 		VHOST_LOG_CONFIG(INFO, "(%s) virtio is now ready for processing.\n", dev->ifname);
1531 	return 1;
1532 }
1533 
1534 static void *
1535 inflight_mem_alloc(struct virtio_net *dev, const char *name, size_t size, int *fd)
1536 {
1537 	void *ptr;
1538 	int mfd = -1;
1539 	char fname[20] = "/tmp/memfd-XXXXXX";
1540 
1541 	*fd = -1;
1542 #ifdef MEMFD_SUPPORTED
1543 	mfd = memfd_create(name, MFD_CLOEXEC);
1544 #else
1545 	RTE_SET_USED(name);
1546 #endif
1547 	if (mfd == -1) {
1548 		mfd = mkstemp(fname);
1549 		if (mfd == -1) {
1550 			VHOST_LOG_CONFIG(ERR, "(%s) failed to get inflight buffer fd\n",
1551 					dev->ifname);
1552 			return NULL;
1553 		}
1554 
1555 		unlink(fname);
1556 	}
1557 
1558 	if (ftruncate(mfd, size) == -1) {
1559 		VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc inflight buffer\n", dev->ifname);
1560 		close(mfd);
1561 		return NULL;
1562 	}
1563 
1564 	ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1565 	if (ptr == MAP_FAILED) {
1566 		VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap inflight buffer\n", dev->ifname);
1567 		close(mfd);
1568 		return NULL;
1569 	}
1570 
1571 	*fd = mfd;
1572 	return ptr;
1573 }
1574 
1575 static uint32_t
1576 get_pervq_shm_size_split(uint16_t queue_size)
1577 {
1578 	return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1579 				  queue_size + sizeof(uint64_t) +
1580 				  sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1581 }
1582 
1583 static uint32_t
1584 get_pervq_shm_size_packed(uint16_t queue_size)
1585 {
1586 	return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1587 				  * queue_size + sizeof(uint64_t) +
1588 				  sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1589 				  INFLIGHT_ALIGNMENT);
1590 }
1591 
1592 static int
1593 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1594 			   struct vhu_msg_context *ctx,
1595 			   int main_fd __rte_unused)
1596 {
1597 	struct rte_vhost_inflight_info_packed *inflight_packed;
1598 	uint64_t pervq_inflight_size, mmap_size;
1599 	uint16_t num_queues, queue_size;
1600 	struct virtio_net *dev = *pdev;
1601 	int fd, i, j;
1602 	int numa_node = SOCKET_ID_ANY;
1603 	void *addr;
1604 
1605 	if (ctx->msg.size != sizeof(ctx->msg.payload.inflight)) {
1606 		VHOST_LOG_CONFIG(ERR, "(%s) invalid get_inflight_fd message size is %d\n",
1607 			dev->ifname, ctx->msg.size);
1608 		return RTE_VHOST_MSG_RESULT_ERR;
1609 	}
1610 
1611 	/*
1612 	 * If VQ 0 has already been allocated, try to allocate on the same
1613 	 * NUMA node. It can be reallocated later in numa_realloc().
1614 	 */
1615 	if (dev->nr_vring > 0)
1616 		numa_node = dev->virtqueue[0]->numa_node;
1617 
1618 	if (dev->inflight_info == NULL) {
1619 		dev->inflight_info = rte_zmalloc_socket("inflight_info",
1620 				sizeof(struct inflight_mem_info), 0, numa_node);
1621 		if (!dev->inflight_info) {
1622 			VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc dev inflight area\n",
1623 					dev->ifname);
1624 			return RTE_VHOST_MSG_RESULT_ERR;
1625 		}
1626 		dev->inflight_info->fd = -1;
1627 	}
1628 
1629 	num_queues = ctx->msg.payload.inflight.num_queues;
1630 	queue_size = ctx->msg.payload.inflight.queue_size;
1631 
1632 	VHOST_LOG_CONFIG(INFO, "(%s) get_inflight_fd num_queues: %u\n",
1633 		dev->ifname, ctx->msg.payload.inflight.num_queues);
1634 	VHOST_LOG_CONFIG(INFO, "(%s) get_inflight_fd queue_size: %u\n",
1635 		dev->ifname, ctx->msg.payload.inflight.queue_size);
1636 
1637 	if (vq_is_packed(dev))
1638 		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1639 	else
1640 		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1641 
1642 	mmap_size = num_queues * pervq_inflight_size;
1643 	addr = inflight_mem_alloc(dev, "vhost-inflight", mmap_size, &fd);
1644 	if (!addr) {
1645 		VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc vhost inflight area\n", dev->ifname);
1646 			ctx->msg.payload.inflight.mmap_size = 0;
1647 		return RTE_VHOST_MSG_RESULT_ERR;
1648 	}
1649 	memset(addr, 0, mmap_size);
1650 
1651 	if (dev->inflight_info->addr) {
1652 		munmap(dev->inflight_info->addr, dev->inflight_info->size);
1653 		dev->inflight_info->addr = NULL;
1654 	}
1655 
1656 	if (dev->inflight_info->fd >= 0) {
1657 		close(dev->inflight_info->fd);
1658 		dev->inflight_info->fd = -1;
1659 	}
1660 
1661 	dev->inflight_info->addr = addr;
1662 	dev->inflight_info->size = ctx->msg.payload.inflight.mmap_size = mmap_size;
1663 	dev->inflight_info->fd = ctx->fds[0] = fd;
1664 	ctx->msg.payload.inflight.mmap_offset = 0;
1665 	ctx->fd_num = 1;
1666 
1667 	if (vq_is_packed(dev)) {
1668 		for (i = 0; i < num_queues; i++) {
1669 			inflight_packed =
1670 				(struct rte_vhost_inflight_info_packed *)addr;
1671 			inflight_packed->used_wrap_counter = 1;
1672 			inflight_packed->old_used_wrap_counter = 1;
1673 			for (j = 0; j < queue_size; j++)
1674 				inflight_packed->desc[j].next = j + 1;
1675 			addr = (void *)((char *)addr + pervq_inflight_size);
1676 		}
1677 	}
1678 
1679 	VHOST_LOG_CONFIG(INFO, "(%s) send inflight mmap_size: %"PRIu64"\n",
1680 			dev->ifname, ctx->msg.payload.inflight.mmap_size);
1681 	VHOST_LOG_CONFIG(INFO, "(%s) send inflight mmap_offset: %"PRIu64"\n",
1682 			dev->ifname, ctx->msg.payload.inflight.mmap_offset);
1683 	VHOST_LOG_CONFIG(INFO, "(%s) send inflight fd: %d\n", dev->ifname, ctx->fds[0]);
1684 
1685 	return RTE_VHOST_MSG_RESULT_REPLY;
1686 }
1687 
1688 static int
1689 vhost_user_set_inflight_fd(struct virtio_net **pdev,
1690 			   struct vhu_msg_context *ctx,
1691 			   int main_fd __rte_unused)
1692 {
1693 	uint64_t mmap_size, mmap_offset;
1694 	uint16_t num_queues, queue_size;
1695 	struct virtio_net *dev = *pdev;
1696 	uint32_t pervq_inflight_size;
1697 	struct vhost_virtqueue *vq;
1698 	void *addr;
1699 	int fd, i;
1700 	int numa_node = SOCKET_ID_ANY;
1701 
1702 	fd = ctx->fds[0];
1703 	if (ctx->msg.size != sizeof(ctx->msg.payload.inflight) || fd < 0) {
1704 		VHOST_LOG_CONFIG(ERR, "(%s) invalid set_inflight_fd message size is %d,fd is %d\n",
1705 			dev->ifname, ctx->msg.size, fd);
1706 		return RTE_VHOST_MSG_RESULT_ERR;
1707 	}
1708 
1709 	mmap_size = ctx->msg.payload.inflight.mmap_size;
1710 	mmap_offset = ctx->msg.payload.inflight.mmap_offset;
1711 	num_queues = ctx->msg.payload.inflight.num_queues;
1712 	queue_size = ctx->msg.payload.inflight.queue_size;
1713 
1714 	if (vq_is_packed(dev))
1715 		pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1716 	else
1717 		pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1718 
1719 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd mmap_size: %"PRIu64"\n",
1720 			dev->ifname, mmap_size);
1721 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd mmap_offset: %"PRIu64"\n",
1722 			dev->ifname, mmap_offset);
1723 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd num_queues: %u\n", dev->ifname, num_queues);
1724 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd queue_size: %u\n", dev->ifname, queue_size);
1725 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd fd: %d\n", dev->ifname, fd);
1726 	VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd pervq_inflight_size: %d\n",
1727 			dev->ifname, pervq_inflight_size);
1728 
1729 	/*
1730 	 * If VQ 0 has already been allocated, try to allocate on the same
1731 	 * NUMA node. It can be reallocated later in numa_realloc().
1732 	 */
1733 	if (dev->nr_vring > 0)
1734 		numa_node = dev->virtqueue[0]->numa_node;
1735 
1736 	if (!dev->inflight_info) {
1737 		dev->inflight_info = rte_zmalloc_socket("inflight_info",
1738 				sizeof(struct inflight_mem_info), 0, numa_node);
1739 		if (dev->inflight_info == NULL) {
1740 			VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc dev inflight area\n",
1741 					dev->ifname);
1742 			return RTE_VHOST_MSG_RESULT_ERR;
1743 		}
1744 		dev->inflight_info->fd = -1;
1745 	}
1746 
1747 	if (dev->inflight_info->addr) {
1748 		munmap(dev->inflight_info->addr, dev->inflight_info->size);
1749 		dev->inflight_info->addr = NULL;
1750 	}
1751 
1752 	addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1753 		    fd, mmap_offset);
1754 	if (addr == MAP_FAILED) {
1755 		VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap share memory.\n", dev->ifname);
1756 		return RTE_VHOST_MSG_RESULT_ERR;
1757 	}
1758 
1759 	if (dev->inflight_info->fd >= 0) {
1760 		close(dev->inflight_info->fd);
1761 		dev->inflight_info->fd = -1;
1762 	}
1763 
1764 	dev->inflight_info->fd = fd;
1765 	dev->inflight_info->addr = addr;
1766 	dev->inflight_info->size = mmap_size;
1767 
1768 	for (i = 0; i < num_queues; i++) {
1769 		vq = dev->virtqueue[i];
1770 		if (!vq)
1771 			continue;
1772 
1773 		if (vq_is_packed(dev)) {
1774 			vq->inflight_packed = addr;
1775 			vq->inflight_packed->desc_num = queue_size;
1776 		} else {
1777 			vq->inflight_split = addr;
1778 			vq->inflight_split->desc_num = queue_size;
1779 		}
1780 		addr = (void *)((char *)addr + pervq_inflight_size);
1781 	}
1782 
1783 	return RTE_VHOST_MSG_RESULT_OK;
1784 }
1785 
1786 static int
1787 vhost_user_set_vring_call(struct virtio_net **pdev,
1788 			struct vhu_msg_context *ctx,
1789 			int main_fd __rte_unused)
1790 {
1791 	struct virtio_net *dev = *pdev;
1792 	struct vhost_vring_file file;
1793 	struct vhost_virtqueue *vq;
1794 	int expected_fds;
1795 
1796 	expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1797 	if (validate_msg_fds(dev, ctx, expected_fds) != 0)
1798 		return RTE_VHOST_MSG_RESULT_ERR;
1799 
1800 	file.index = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
1801 	if (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1802 		file.fd = VIRTIO_INVALID_EVENTFD;
1803 	else
1804 		file.fd = ctx->fds[0];
1805 	VHOST_LOG_CONFIG(INFO, "(%s) vring call idx:%d file:%d\n",
1806 			dev->ifname, file.index, file.fd);
1807 
1808 	vq = dev->virtqueue[file.index];
1809 
1810 	if (vq->ready) {
1811 		vq->ready = false;
1812 		vhost_user_notify_queue_state(dev, file.index, 0);
1813 	}
1814 
1815 	if (vq->callfd >= 0)
1816 		close(vq->callfd);
1817 
1818 	vq->callfd = file.fd;
1819 
1820 	return RTE_VHOST_MSG_RESULT_OK;
1821 }
1822 
1823 static int vhost_user_set_vring_err(struct virtio_net **pdev,
1824 			struct vhu_msg_context *ctx,
1825 			int main_fd __rte_unused)
1826 {
1827 	struct virtio_net *dev = *pdev;
1828 	int expected_fds;
1829 
1830 	expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1831 	if (validate_msg_fds(dev, ctx, expected_fds) != 0)
1832 		return RTE_VHOST_MSG_RESULT_ERR;
1833 
1834 	if (!(ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1835 		close(ctx->fds[0]);
1836 	VHOST_LOG_CONFIG(INFO, "(%s) not implemented\n", dev->ifname);
1837 
1838 	return RTE_VHOST_MSG_RESULT_OK;
1839 }
1840 
1841 static int
1842 resubmit_desc_compare(const void *a, const void *b)
1843 {
1844 	const struct rte_vhost_resubmit_desc *desc0 = a;
1845 	const struct rte_vhost_resubmit_desc *desc1 = b;
1846 
1847 	if (desc1->counter > desc0->counter)
1848 		return 1;
1849 
1850 	return -1;
1851 }
1852 
1853 static int
1854 vhost_check_queue_inflights_split(struct virtio_net *dev,
1855 				  struct vhost_virtqueue *vq)
1856 {
1857 	uint16_t i;
1858 	uint16_t resubmit_num = 0, last_io, num;
1859 	struct vring_used *used = vq->used;
1860 	struct rte_vhost_resubmit_info *resubmit;
1861 	struct rte_vhost_inflight_info_split *inflight_split;
1862 
1863 	if (!(dev->protocol_features &
1864 	    (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1865 		return RTE_VHOST_MSG_RESULT_OK;
1866 
1867 	/* The frontend may still not support the inflight feature
1868 	 * although we negotiate the protocol feature.
1869 	 */
1870 	if ((!vq->inflight_split))
1871 		return RTE_VHOST_MSG_RESULT_OK;
1872 
1873 	if (!vq->inflight_split->version) {
1874 		vq->inflight_split->version = INFLIGHT_VERSION;
1875 		return RTE_VHOST_MSG_RESULT_OK;
1876 	}
1877 
1878 	if (vq->resubmit_inflight)
1879 		return RTE_VHOST_MSG_RESULT_OK;
1880 
1881 	inflight_split = vq->inflight_split;
1882 	vq->global_counter = 0;
1883 	last_io = inflight_split->last_inflight_io;
1884 
1885 	if (inflight_split->used_idx != used->idx) {
1886 		inflight_split->desc[last_io].inflight = 0;
1887 		rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1888 		inflight_split->used_idx = used->idx;
1889 	}
1890 
1891 	for (i = 0; i < inflight_split->desc_num; i++) {
1892 		if (inflight_split->desc[i].inflight == 1)
1893 			resubmit_num++;
1894 	}
1895 
1896 	vq->last_avail_idx += resubmit_num;
1897 
1898 	if (resubmit_num) {
1899 		resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1900 				0, vq->numa_node);
1901 		if (!resubmit) {
1902 			VHOST_LOG_CONFIG(ERR,
1903 					"(%s) failed to allocate memory for resubmit info.\n",
1904 					dev->ifname);
1905 			return RTE_VHOST_MSG_RESULT_ERR;
1906 		}
1907 
1908 		resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
1909 				resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
1910 				0, vq->numa_node);
1911 		if (!resubmit->resubmit_list) {
1912 			VHOST_LOG_CONFIG(ERR,
1913 					"(%s) failed to allocate memory for inflight desc.\n",
1914 					dev->ifname);
1915 			rte_free(resubmit);
1916 			return RTE_VHOST_MSG_RESULT_ERR;
1917 		}
1918 
1919 		num = 0;
1920 		for (i = 0; i < vq->inflight_split->desc_num; i++) {
1921 			if (vq->inflight_split->desc[i].inflight == 1) {
1922 				resubmit->resubmit_list[num].index = i;
1923 				resubmit->resubmit_list[num].counter =
1924 					inflight_split->desc[i].counter;
1925 				num++;
1926 			}
1927 		}
1928 		resubmit->resubmit_num = num;
1929 
1930 		if (resubmit->resubmit_num > 1)
1931 			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1932 			      sizeof(struct rte_vhost_resubmit_desc),
1933 			      resubmit_desc_compare);
1934 
1935 		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1936 		vq->resubmit_inflight = resubmit;
1937 	}
1938 
1939 	return RTE_VHOST_MSG_RESULT_OK;
1940 }
1941 
1942 static int
1943 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1944 				   struct vhost_virtqueue *vq)
1945 {
1946 	uint16_t i;
1947 	uint16_t resubmit_num = 0, old_used_idx, num;
1948 	struct rte_vhost_resubmit_info *resubmit;
1949 	struct rte_vhost_inflight_info_packed *inflight_packed;
1950 
1951 	if (!(dev->protocol_features &
1952 	    (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1953 		return RTE_VHOST_MSG_RESULT_OK;
1954 
1955 	/* The frontend may still not support the inflight feature
1956 	 * although we negotiate the protocol feature.
1957 	 */
1958 	if ((!vq->inflight_packed))
1959 		return RTE_VHOST_MSG_RESULT_OK;
1960 
1961 	if (!vq->inflight_packed->version) {
1962 		vq->inflight_packed->version = INFLIGHT_VERSION;
1963 		return RTE_VHOST_MSG_RESULT_OK;
1964 	}
1965 
1966 	if (vq->resubmit_inflight)
1967 		return RTE_VHOST_MSG_RESULT_OK;
1968 
1969 	inflight_packed = vq->inflight_packed;
1970 	vq->global_counter = 0;
1971 	old_used_idx = inflight_packed->old_used_idx;
1972 
1973 	if (inflight_packed->used_idx != old_used_idx) {
1974 		if (inflight_packed->desc[old_used_idx].inflight == 0) {
1975 			inflight_packed->old_used_idx =
1976 				inflight_packed->used_idx;
1977 			inflight_packed->old_used_wrap_counter =
1978 				inflight_packed->used_wrap_counter;
1979 			inflight_packed->old_free_head =
1980 				inflight_packed->free_head;
1981 		} else {
1982 			inflight_packed->used_idx =
1983 				inflight_packed->old_used_idx;
1984 			inflight_packed->used_wrap_counter =
1985 				inflight_packed->old_used_wrap_counter;
1986 			inflight_packed->free_head =
1987 				inflight_packed->old_free_head;
1988 		}
1989 	}
1990 
1991 	for (i = 0; i < inflight_packed->desc_num; i++) {
1992 		if (inflight_packed->desc[i].inflight == 1)
1993 			resubmit_num++;
1994 	}
1995 
1996 	if (resubmit_num) {
1997 		resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1998 				0, vq->numa_node);
1999 		if (resubmit == NULL) {
2000 			VHOST_LOG_CONFIG(ERR,
2001 					"(%s) failed to allocate memory for resubmit info.\n",
2002 					dev->ifname);
2003 			return RTE_VHOST_MSG_RESULT_ERR;
2004 		}
2005 
2006 		resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
2007 				resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
2008 				0, vq->numa_node);
2009 		if (resubmit->resubmit_list == NULL) {
2010 			VHOST_LOG_CONFIG(ERR,
2011 					"(%s) failed to allocate memory for resubmit desc.\n",
2012 					dev->ifname);
2013 			rte_free(resubmit);
2014 			return RTE_VHOST_MSG_RESULT_ERR;
2015 		}
2016 
2017 		num = 0;
2018 		for (i = 0; i < inflight_packed->desc_num; i++) {
2019 			if (vq->inflight_packed->desc[i].inflight == 1) {
2020 				resubmit->resubmit_list[num].index = i;
2021 				resubmit->resubmit_list[num].counter =
2022 					inflight_packed->desc[i].counter;
2023 				num++;
2024 			}
2025 		}
2026 		resubmit->resubmit_num = num;
2027 
2028 		if (resubmit->resubmit_num > 1)
2029 			qsort(resubmit->resubmit_list, resubmit->resubmit_num,
2030 			      sizeof(struct rte_vhost_resubmit_desc),
2031 			      resubmit_desc_compare);
2032 
2033 		vq->global_counter = resubmit->resubmit_list[0].counter + 1;
2034 		vq->resubmit_inflight = resubmit;
2035 	}
2036 
2037 	return RTE_VHOST_MSG_RESULT_OK;
2038 }
2039 
2040 static int
2041 vhost_user_set_vring_kick(struct virtio_net **pdev,
2042 			struct vhu_msg_context *ctx,
2043 			int main_fd __rte_unused)
2044 {
2045 	struct virtio_net *dev = *pdev;
2046 	struct vhost_vring_file file;
2047 	struct vhost_virtqueue *vq;
2048 	int expected_fds;
2049 
2050 	expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
2051 	if (validate_msg_fds(dev, ctx, expected_fds) != 0)
2052 		return RTE_VHOST_MSG_RESULT_ERR;
2053 
2054 	file.index = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
2055 	if (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK)
2056 		file.fd = VIRTIO_INVALID_EVENTFD;
2057 	else
2058 		file.fd = ctx->fds[0];
2059 	VHOST_LOG_CONFIG(INFO, "(%s) vring kick idx:%d file:%d\n",
2060 			dev->ifname, file.index, file.fd);
2061 
2062 	/* Interpret ring addresses only when ring is started. */
2063 	dev = translate_ring_addresses(dev, file.index);
2064 	if (!dev) {
2065 		if (file.fd != VIRTIO_INVALID_EVENTFD)
2066 			close(file.fd);
2067 
2068 		return RTE_VHOST_MSG_RESULT_ERR;
2069 	}
2070 
2071 	*pdev = dev;
2072 
2073 	vq = dev->virtqueue[file.index];
2074 
2075 	/*
2076 	 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
2077 	 * the ring starts already enabled. Otherwise, it is enabled via
2078 	 * the SET_VRING_ENABLE message.
2079 	 */
2080 	if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
2081 		vq->enabled = true;
2082 	}
2083 
2084 	if (vq->ready) {
2085 		vq->ready = false;
2086 		vhost_user_notify_queue_state(dev, file.index, 0);
2087 	}
2088 
2089 	if (vq->kickfd >= 0)
2090 		close(vq->kickfd);
2091 	vq->kickfd = file.fd;
2092 
2093 	if (vq_is_packed(dev)) {
2094 		if (vhost_check_queue_inflights_packed(dev, vq)) {
2095 			VHOST_LOG_CONFIG(ERR, "(%s) failed to inflights for vq: %d\n",
2096 					dev->ifname, file.index);
2097 			return RTE_VHOST_MSG_RESULT_ERR;
2098 		}
2099 	} else {
2100 		if (vhost_check_queue_inflights_split(dev, vq)) {
2101 			VHOST_LOG_CONFIG(ERR, "(%s) failed to inflights for vq: %d\n",
2102 					dev->ifname, file.index);
2103 			return RTE_VHOST_MSG_RESULT_ERR;
2104 		}
2105 	}
2106 
2107 	return RTE_VHOST_MSG_RESULT_OK;
2108 }
2109 
2110 /*
2111  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
2112  */
2113 static int
2114 vhost_user_get_vring_base(struct virtio_net **pdev,
2115 			struct vhu_msg_context *ctx,
2116 			int main_fd __rte_unused)
2117 {
2118 	struct virtio_net *dev = *pdev;
2119 	struct vhost_virtqueue *vq = dev->virtqueue[ctx->msg.payload.state.index];
2120 	uint64_t val;
2121 
2122 	if (validate_msg_fds(dev, ctx, 0) != 0)
2123 		return RTE_VHOST_MSG_RESULT_ERR;
2124 
2125 	/* We have to stop the queue (virtio) if it is running. */
2126 	vhost_destroy_device_notify(dev);
2127 
2128 	dev->flags &= ~VIRTIO_DEV_READY;
2129 	dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
2130 
2131 	/* Here we are safe to get the indexes */
2132 	if (vq_is_packed(dev)) {
2133 		/*
2134 		 * Bit[0:14]: avail index
2135 		 * Bit[15]: avail wrap counter
2136 		 */
2137 		val = vq->last_avail_idx & 0x7fff;
2138 		val |= vq->avail_wrap_counter << 15;
2139 		ctx->msg.payload.state.num = val;
2140 	} else {
2141 		ctx->msg.payload.state.num = vq->last_avail_idx;
2142 	}
2143 
2144 	VHOST_LOG_CONFIG(INFO, "(%s) vring base idx:%d file:%d\n",
2145 			dev->ifname, ctx->msg.payload.state.index,
2146 			ctx->msg.payload.state.num);
2147 	/*
2148 	 * Based on current qemu vhost-user implementation, this message is
2149 	 * sent and only sent in vhost_vring_stop.
2150 	 * TODO: cleanup the vring, it isn't usable since here.
2151 	 */
2152 	if (vq->kickfd >= 0)
2153 		close(vq->kickfd);
2154 
2155 	vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2156 
2157 	if (vq->callfd >= 0)
2158 		close(vq->callfd);
2159 
2160 	vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2161 
2162 	vq->signalled_used_valid = false;
2163 
2164 	if (vq_is_packed(dev)) {
2165 		rte_free(vq->shadow_used_packed);
2166 		vq->shadow_used_packed = NULL;
2167 	} else {
2168 		rte_free(vq->shadow_used_split);
2169 		vq->shadow_used_split = NULL;
2170 	}
2171 
2172 	rte_free(vq->batch_copy_elems);
2173 	vq->batch_copy_elems = NULL;
2174 
2175 	rte_free(vq->log_cache);
2176 	vq->log_cache = NULL;
2177 
2178 	ctx->msg.size = sizeof(ctx->msg.payload.state);
2179 	ctx->fd_num = 0;
2180 
2181 	vhost_user_iotlb_flush_all(vq);
2182 
2183 	vring_invalidate(dev, vq);
2184 
2185 	return RTE_VHOST_MSG_RESULT_REPLY;
2186 }
2187 
2188 /*
2189  * when virtio queues are ready to work, qemu will send us to
2190  * enable the virtio queue pair.
2191  */
2192 static int
2193 vhost_user_set_vring_enable(struct virtio_net **pdev,
2194 			struct vhu_msg_context *ctx,
2195 			int main_fd __rte_unused)
2196 {
2197 	struct virtio_net *dev = *pdev;
2198 	bool enable = !!ctx->msg.payload.state.num;
2199 	int index = (int)ctx->msg.payload.state.index;
2200 
2201 	if (validate_msg_fds(dev, ctx, 0) != 0)
2202 		return RTE_VHOST_MSG_RESULT_ERR;
2203 
2204 	VHOST_LOG_CONFIG(INFO, "(%s) set queue enable: %d to qp idx: %d\n",
2205 			dev->ifname, enable, index);
2206 
2207 	if (enable && dev->virtqueue[index]->async) {
2208 		if (dev->virtqueue[index]->async->pkts_inflight_n) {
2209 			VHOST_LOG_CONFIG(ERR,
2210 				"(%s) failed to enable vring. Inflight packets must be completed first\n",
2211 				dev->ifname);
2212 			return RTE_VHOST_MSG_RESULT_ERR;
2213 		}
2214 	}
2215 
2216 	dev->virtqueue[index]->enabled = enable;
2217 
2218 	return RTE_VHOST_MSG_RESULT_OK;
2219 }
2220 
2221 static int
2222 vhost_user_get_protocol_features(struct virtio_net **pdev,
2223 			struct vhu_msg_context *ctx,
2224 			int main_fd __rte_unused)
2225 {
2226 	struct virtio_net *dev = *pdev;
2227 	uint64_t features, protocol_features;
2228 
2229 	if (validate_msg_fds(dev, ctx, 0) != 0)
2230 		return RTE_VHOST_MSG_RESULT_ERR;
2231 
2232 	rte_vhost_driver_get_features(dev->ifname, &features);
2233 	rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2234 
2235 	ctx->msg.payload.u64 = protocol_features;
2236 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
2237 	ctx->fd_num = 0;
2238 
2239 	return RTE_VHOST_MSG_RESULT_REPLY;
2240 }
2241 
2242 static int
2243 vhost_user_set_protocol_features(struct virtio_net **pdev,
2244 			struct vhu_msg_context *ctx,
2245 			int main_fd __rte_unused)
2246 {
2247 	struct virtio_net *dev = *pdev;
2248 	uint64_t protocol_features = ctx->msg.payload.u64;
2249 	uint64_t slave_protocol_features = 0;
2250 
2251 	if (validate_msg_fds(dev, ctx, 0) != 0)
2252 		return RTE_VHOST_MSG_RESULT_ERR;
2253 
2254 	rte_vhost_driver_get_protocol_features(dev->ifname,
2255 			&slave_protocol_features);
2256 	if (protocol_features & ~slave_protocol_features) {
2257 		VHOST_LOG_CONFIG(ERR, "(%s) received invalid protocol features.\n", dev->ifname);
2258 		return RTE_VHOST_MSG_RESULT_ERR;
2259 	}
2260 
2261 	dev->protocol_features = protocol_features;
2262 	VHOST_LOG_CONFIG(INFO, "(%s) negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2263 		dev->ifname, dev->protocol_features);
2264 
2265 	return RTE_VHOST_MSG_RESULT_OK;
2266 }
2267 
2268 static int
2269 vhost_user_set_log_base(struct virtio_net **pdev,
2270 			struct vhu_msg_context *ctx,
2271 			int main_fd __rte_unused)
2272 {
2273 	struct virtio_net *dev = *pdev;
2274 	int fd = ctx->fds[0];
2275 	uint64_t size, off;
2276 	void *addr;
2277 	uint32_t i;
2278 
2279 	if (validate_msg_fds(dev, ctx, 1) != 0)
2280 		return RTE_VHOST_MSG_RESULT_ERR;
2281 
2282 	if (fd < 0) {
2283 		VHOST_LOG_CONFIG(ERR, "(%s) invalid log fd: %d\n", dev->ifname, fd);
2284 		return RTE_VHOST_MSG_RESULT_ERR;
2285 	}
2286 
2287 	if (ctx->msg.size != sizeof(VhostUserLog)) {
2288 		VHOST_LOG_CONFIG(ERR, "(%s) invalid log base msg size: %"PRId32" != %d\n",
2289 			dev->ifname, ctx->msg.size, (int)sizeof(VhostUserLog));
2290 		goto close_msg_fds;
2291 	}
2292 
2293 	size = ctx->msg.payload.log.mmap_size;
2294 	off  = ctx->msg.payload.log.mmap_offset;
2295 
2296 	/* Check for mmap size and offset overflow. */
2297 	if (off >= -size) {
2298 		VHOST_LOG_CONFIG(ERR,
2299 				"(%s) log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2300 				dev->ifname, off, size);
2301 		goto close_msg_fds;
2302 	}
2303 
2304 	VHOST_LOG_CONFIG(INFO, "(%s) log mmap size: %"PRId64", offset: %"PRId64"\n",
2305 			dev->ifname, size, off);
2306 
2307 	/*
2308 	 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2309 	 * fail when offset is not page size aligned.
2310 	 */
2311 	addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2312 	close(fd);
2313 	if (addr == MAP_FAILED) {
2314 		VHOST_LOG_CONFIG(ERR, "(%s) mmap log base failed!\n", dev->ifname);
2315 		return RTE_VHOST_MSG_RESULT_ERR;
2316 	}
2317 
2318 	/*
2319 	 * Free previously mapped log memory on occasionally
2320 	 * multiple VHOST_USER_SET_LOG_BASE.
2321 	 */
2322 	if (dev->log_addr) {
2323 		munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2324 	}
2325 	dev->log_addr = (uint64_t)(uintptr_t)addr;
2326 	dev->log_base = dev->log_addr + off;
2327 	dev->log_size = size;
2328 
2329 	for (i = 0; i < dev->nr_vring; i++) {
2330 		struct vhost_virtqueue *vq = dev->virtqueue[i];
2331 
2332 		rte_free(vq->log_cache);
2333 		vq->log_cache = NULL;
2334 		vq->log_cache_nb_elem = 0;
2335 		vq->log_cache = rte_malloc_socket("vq log cache",
2336 				sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2337 				0, vq->numa_node);
2338 		/*
2339 		 * If log cache alloc fail, don't fail migration, but no
2340 		 * caching will be done, which will impact performance
2341 		 */
2342 		if (!vq->log_cache)
2343 			VHOST_LOG_CONFIG(ERR, "(%s) failed to allocate VQ logging cache\n",
2344 					dev->ifname);
2345 	}
2346 
2347 	/*
2348 	 * The spec is not clear about it (yet), but QEMU doesn't expect
2349 	 * any payload in the reply.
2350 	 */
2351 	ctx->msg.size = 0;
2352 	ctx->fd_num = 0;
2353 
2354 	return RTE_VHOST_MSG_RESULT_REPLY;
2355 
2356 close_msg_fds:
2357 	close_msg_fds(ctx);
2358 	return RTE_VHOST_MSG_RESULT_ERR;
2359 }
2360 
2361 static int vhost_user_set_log_fd(struct virtio_net **pdev,
2362 			struct vhu_msg_context *ctx,
2363 			int main_fd __rte_unused)
2364 {
2365 	struct virtio_net *dev = *pdev;
2366 
2367 	if (validate_msg_fds(dev, ctx, 1) != 0)
2368 		return RTE_VHOST_MSG_RESULT_ERR;
2369 
2370 	close(ctx->fds[0]);
2371 	VHOST_LOG_CONFIG(INFO, "(%s) not implemented.\n", dev->ifname);
2372 
2373 	return RTE_VHOST_MSG_RESULT_OK;
2374 }
2375 
2376 /*
2377  * An rarp packet is constructed and broadcasted to notify switches about
2378  * the new location of the migrated VM, so that packets from outside will
2379  * not be lost after migration.
2380  *
2381  * However, we don't actually "send" a rarp packet here, instead, we set
2382  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2383  */
2384 static int
2385 vhost_user_send_rarp(struct virtio_net **pdev,
2386 			struct vhu_msg_context *ctx,
2387 			int main_fd __rte_unused)
2388 {
2389 	struct virtio_net *dev = *pdev;
2390 	uint8_t *mac = (uint8_t *)&ctx->msg.payload.u64;
2391 	struct rte_vdpa_device *vdpa_dev;
2392 
2393 	if (validate_msg_fds(dev, ctx, 0) != 0)
2394 		return RTE_VHOST_MSG_RESULT_ERR;
2395 
2396 	VHOST_LOG_CONFIG(DEBUG, "(%s) MAC: " RTE_ETHER_ADDR_PRT_FMT "\n",
2397 		dev->ifname, mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2398 	memcpy(dev->mac.addr_bytes, mac, 6);
2399 
2400 	/*
2401 	 * Set the flag to inject a RARP broadcast packet at
2402 	 * rte_vhost_dequeue_burst().
2403 	 *
2404 	 * __ATOMIC_RELEASE ordering is for making sure the mac is
2405 	 * copied before the flag is set.
2406 	 */
2407 	__atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2408 	vdpa_dev = dev->vdpa_dev;
2409 	if (vdpa_dev && vdpa_dev->ops->migration_done)
2410 		vdpa_dev->ops->migration_done(dev->vid);
2411 
2412 	return RTE_VHOST_MSG_RESULT_OK;
2413 }
2414 
2415 static int
2416 vhost_user_net_set_mtu(struct virtio_net **pdev,
2417 			struct vhu_msg_context *ctx,
2418 			int main_fd __rte_unused)
2419 {
2420 	struct virtio_net *dev = *pdev;
2421 
2422 	if (validate_msg_fds(dev, ctx, 0) != 0)
2423 		return RTE_VHOST_MSG_RESULT_ERR;
2424 
2425 	if (ctx->msg.payload.u64 < VIRTIO_MIN_MTU ||
2426 			ctx->msg.payload.u64 > VIRTIO_MAX_MTU) {
2427 		VHOST_LOG_CONFIG(ERR, "(%s) invalid MTU size (%"PRIu64")\n",
2428 				dev->ifname, ctx->msg.payload.u64);
2429 
2430 		return RTE_VHOST_MSG_RESULT_ERR;
2431 	}
2432 
2433 	dev->mtu = ctx->msg.payload.u64;
2434 
2435 	return RTE_VHOST_MSG_RESULT_OK;
2436 }
2437 
2438 static int
2439 vhost_user_set_req_fd(struct virtio_net **pdev,
2440 			struct vhu_msg_context *ctx,
2441 			int main_fd __rte_unused)
2442 {
2443 	struct virtio_net *dev = *pdev;
2444 	int fd = ctx->fds[0];
2445 
2446 	if (validate_msg_fds(dev, ctx, 1) != 0)
2447 		return RTE_VHOST_MSG_RESULT_ERR;
2448 
2449 	if (fd < 0) {
2450 		VHOST_LOG_CONFIG(ERR, "(%s) invalid file descriptor for slave channel (%d)\n",
2451 				dev->ifname, fd);
2452 		return RTE_VHOST_MSG_RESULT_ERR;
2453 	}
2454 
2455 	if (dev->slave_req_fd >= 0)
2456 		close(dev->slave_req_fd);
2457 
2458 	dev->slave_req_fd = fd;
2459 
2460 	return RTE_VHOST_MSG_RESULT_OK;
2461 }
2462 
2463 static int
2464 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2465 {
2466 	struct vhost_vring_addr *ra;
2467 	uint64_t start, end, len;
2468 
2469 	start = imsg->iova;
2470 	end = start + imsg->size;
2471 
2472 	ra = &vq->ring_addrs;
2473 	len = sizeof(struct vring_desc) * vq->size;
2474 	if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2475 		return 1;
2476 
2477 	len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2478 	if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2479 		return 1;
2480 
2481 	len = sizeof(struct vring_used) +
2482 	       sizeof(struct vring_used_elem) * vq->size;
2483 	if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2484 		return 1;
2485 
2486 	if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2487 		len = sizeof(uint64_t);
2488 		if (ra->log_guest_addr < end &&
2489 		    (ra->log_guest_addr + len) > start)
2490 			return 1;
2491 	}
2492 
2493 	return 0;
2494 }
2495 
2496 static int
2497 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2498 {
2499 	struct vhost_vring_addr *ra;
2500 	uint64_t start, end, len;
2501 
2502 	start = imsg->iova;
2503 	end = start + imsg->size;
2504 
2505 	ra = &vq->ring_addrs;
2506 	len = sizeof(struct vring_packed_desc) * vq->size;
2507 	if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2508 		return 1;
2509 
2510 	len = sizeof(struct vring_packed_desc_event);
2511 	if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2512 		return 1;
2513 
2514 	len = sizeof(struct vring_packed_desc_event);
2515 	if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2516 		return 1;
2517 
2518 	if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2519 		len = sizeof(uint64_t);
2520 		if (ra->log_guest_addr < end &&
2521 		    (ra->log_guest_addr + len) > start)
2522 			return 1;
2523 	}
2524 
2525 	return 0;
2526 }
2527 
2528 static int is_vring_iotlb(struct virtio_net *dev,
2529 			  struct vhost_virtqueue *vq,
2530 			  struct vhost_iotlb_msg *imsg)
2531 {
2532 	if (vq_is_packed(dev))
2533 		return is_vring_iotlb_packed(vq, imsg);
2534 	else
2535 		return is_vring_iotlb_split(vq, imsg);
2536 }
2537 
2538 static int
2539 vhost_user_iotlb_msg(struct virtio_net **pdev,
2540 			struct vhu_msg_context *ctx,
2541 			int main_fd __rte_unused)
2542 {
2543 	struct virtio_net *dev = *pdev;
2544 	struct vhost_iotlb_msg *imsg = &ctx->msg.payload.iotlb;
2545 	uint16_t i;
2546 	uint64_t vva, len;
2547 
2548 	if (validate_msg_fds(dev, ctx, 0) != 0)
2549 		return RTE_VHOST_MSG_RESULT_ERR;
2550 
2551 	switch (imsg->type) {
2552 	case VHOST_IOTLB_UPDATE:
2553 		len = imsg->size;
2554 		vva = qva_to_vva(dev, imsg->uaddr, &len);
2555 		if (!vva)
2556 			return RTE_VHOST_MSG_RESULT_ERR;
2557 
2558 		for (i = 0; i < dev->nr_vring; i++) {
2559 			struct vhost_virtqueue *vq = dev->virtqueue[i];
2560 
2561 			if (!vq)
2562 				continue;
2563 
2564 			vhost_user_iotlb_cache_insert(dev, vq, imsg->iova, vva,
2565 					len, imsg->perm);
2566 
2567 			if (is_vring_iotlb(dev, vq, imsg)) {
2568 				rte_spinlock_lock(&vq->access_lock);
2569 				*pdev = dev = translate_ring_addresses(dev, i);
2570 				rte_spinlock_unlock(&vq->access_lock);
2571 			}
2572 		}
2573 		break;
2574 	case VHOST_IOTLB_INVALIDATE:
2575 		for (i = 0; i < dev->nr_vring; i++) {
2576 			struct vhost_virtqueue *vq = dev->virtqueue[i];
2577 
2578 			if (!vq)
2579 				continue;
2580 
2581 			vhost_user_iotlb_cache_remove(vq, imsg->iova,
2582 					imsg->size);
2583 
2584 			if (is_vring_iotlb(dev, vq, imsg)) {
2585 				rte_spinlock_lock(&vq->access_lock);
2586 				vring_invalidate(dev, vq);
2587 				rte_spinlock_unlock(&vq->access_lock);
2588 			}
2589 		}
2590 		break;
2591 	default:
2592 		VHOST_LOG_CONFIG(ERR, "(%s) invalid IOTLB message type (%d)\n",
2593 				dev->ifname, imsg->type);
2594 		return RTE_VHOST_MSG_RESULT_ERR;
2595 	}
2596 
2597 	return RTE_VHOST_MSG_RESULT_OK;
2598 }
2599 
2600 static int
2601 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2602 			struct vhu_msg_context *ctx,
2603 			int main_fd __rte_unused)
2604 {
2605 	struct virtio_net *dev = *pdev;
2606 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2607 	struct uffdio_api api_struct;
2608 
2609 	if (validate_msg_fds(dev, ctx, 0) != 0)
2610 		return RTE_VHOST_MSG_RESULT_ERR;
2611 
2612 	dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2613 
2614 	if (dev->postcopy_ufd == -1) {
2615 		VHOST_LOG_CONFIG(ERR, "(%s) userfaultfd not available: %s\n",
2616 			dev->ifname, strerror(errno));
2617 		return RTE_VHOST_MSG_RESULT_ERR;
2618 	}
2619 	api_struct.api = UFFD_API;
2620 	api_struct.features = 0;
2621 	if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2622 		VHOST_LOG_CONFIG(ERR, "(%s) UFFDIO_API ioctl failure: %s\n",
2623 			dev->ifname, strerror(errno));
2624 		close(dev->postcopy_ufd);
2625 		dev->postcopy_ufd = -1;
2626 		return RTE_VHOST_MSG_RESULT_ERR;
2627 	}
2628 	ctx->fds[0] = dev->postcopy_ufd;
2629 	ctx->fd_num = 1;
2630 
2631 	return RTE_VHOST_MSG_RESULT_REPLY;
2632 #else
2633 	dev->postcopy_ufd = -1;
2634 	ctx->fd_num = 0;
2635 
2636 	return RTE_VHOST_MSG_RESULT_ERR;
2637 #endif
2638 }
2639 
2640 static int
2641 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2642 			struct vhu_msg_context *ctx __rte_unused,
2643 			int main_fd __rte_unused)
2644 {
2645 	struct virtio_net *dev = *pdev;
2646 
2647 	if (validate_msg_fds(dev, ctx, 0) != 0)
2648 		return RTE_VHOST_MSG_RESULT_ERR;
2649 
2650 	if (dev->mem && dev->mem->nregions) {
2651 		VHOST_LOG_CONFIG(ERR, "(%s) regions already registered at postcopy-listen\n",
2652 				dev->ifname);
2653 		return RTE_VHOST_MSG_RESULT_ERR;
2654 	}
2655 	dev->postcopy_listening = 1;
2656 
2657 	return RTE_VHOST_MSG_RESULT_OK;
2658 }
2659 
2660 static int
2661 vhost_user_postcopy_end(struct virtio_net **pdev,
2662 			struct vhu_msg_context *ctx,
2663 			int main_fd __rte_unused)
2664 {
2665 	struct virtio_net *dev = *pdev;
2666 
2667 	if (validate_msg_fds(dev, ctx, 0) != 0)
2668 		return RTE_VHOST_MSG_RESULT_ERR;
2669 
2670 	dev->postcopy_listening = 0;
2671 	if (dev->postcopy_ufd >= 0) {
2672 		close(dev->postcopy_ufd);
2673 		dev->postcopy_ufd = -1;
2674 	}
2675 
2676 	ctx->msg.payload.u64 = 0;
2677 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
2678 	ctx->fd_num = 0;
2679 
2680 	return RTE_VHOST_MSG_RESULT_REPLY;
2681 }
2682 
2683 static int
2684 vhost_user_get_status(struct virtio_net **pdev,
2685 		      struct vhu_msg_context *ctx,
2686 		      int main_fd __rte_unused)
2687 {
2688 	struct virtio_net *dev = *pdev;
2689 
2690 	if (validate_msg_fds(dev, ctx, 0) != 0)
2691 		return RTE_VHOST_MSG_RESULT_ERR;
2692 
2693 	ctx->msg.payload.u64 = dev->status;
2694 	ctx->msg.size = sizeof(ctx->msg.payload.u64);
2695 	ctx->fd_num = 0;
2696 
2697 	return RTE_VHOST_MSG_RESULT_REPLY;
2698 }
2699 
2700 static int
2701 vhost_user_set_status(struct virtio_net **pdev,
2702 			struct vhu_msg_context *ctx,
2703 			int main_fd __rte_unused)
2704 {
2705 	struct virtio_net *dev = *pdev;
2706 
2707 	if (validate_msg_fds(dev, ctx, 0) != 0)
2708 		return RTE_VHOST_MSG_RESULT_ERR;
2709 
2710 	/* As per Virtio specification, the device status is 8bits long */
2711 	if (ctx->msg.payload.u64 > UINT8_MAX) {
2712 		VHOST_LOG_CONFIG(ERR, "(%s) invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2713 				dev->ifname, ctx->msg.payload.u64);
2714 		return RTE_VHOST_MSG_RESULT_ERR;
2715 	}
2716 
2717 	dev->status = ctx->msg.payload.u64;
2718 
2719 	if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2720 	    (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2721 		VHOST_LOG_CONFIG(ERR,
2722 				"(%s) FEATURES_OK bit is set but feature negotiation failed\n",
2723 				dev->ifname);
2724 		/*
2725 		 * Clear the bit to let the driver know about the feature
2726 		 * negotiation failure
2727 		 */
2728 		dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2729 	}
2730 
2731 	VHOST_LOG_CONFIG(INFO, "(%s) new device status(0x%08x):\n", dev->ifname,
2732 			dev->status);
2733 	VHOST_LOG_CONFIG(INFO, "(%s)\t-RESET: %u\n", dev->ifname,
2734 			(dev->status == VIRTIO_DEVICE_STATUS_RESET));
2735 	VHOST_LOG_CONFIG(INFO, "(%s)\t-ACKNOWLEDGE: %u\n", dev->ifname,
2736 			!!(dev->status & VIRTIO_DEVICE_STATUS_ACK));
2737 	VHOST_LOG_CONFIG(INFO, "(%s)\t-DRIVER: %u\n", dev->ifname,
2738 			!!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER));
2739 	VHOST_LOG_CONFIG(INFO, "(%s)\t-FEATURES_OK: %u\n", dev->ifname,
2740 			!!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK));
2741 	VHOST_LOG_CONFIG(INFO, "(%s)\t-DRIVER_OK: %u\n", dev->ifname,
2742 			!!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK));
2743 	VHOST_LOG_CONFIG(INFO, "(%s)\t-DEVICE_NEED_RESET: %u\n", dev->ifname,
2744 			!!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET));
2745 	VHOST_LOG_CONFIG(INFO, "(%s)\t-FAILED: %u\n", dev->ifname,
2746 			!!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2747 
2748 	return RTE_VHOST_MSG_RESULT_OK;
2749 }
2750 
2751 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2752 					struct vhu_msg_context *ctx,
2753 					int main_fd);
2754 
2755 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2756 	[VHOST_USER_NONE] = NULL,
2757 	[VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2758 	[VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2759 	[VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2760 	[VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2761 	[VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2762 	[VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2763 	[VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2764 	[VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2765 	[VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2766 	[VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2767 	[VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2768 	[VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2769 	[VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2770 	[VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2771 	[VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2772 	[VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2773 	[VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2774 	[VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2775 	[VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2776 	[VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2777 	[VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2778 	[VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2779 	[VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2780 	[VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2781 	[VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2782 	[VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2783 	[VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2784 	[VHOST_USER_SET_STATUS] = vhost_user_set_status,
2785 	[VHOST_USER_GET_STATUS] = vhost_user_get_status,
2786 };
2787 
2788 /* return bytes# of read on success or negative val on failure. */
2789 static int
2790 read_vhost_message(struct virtio_net *dev, int sockfd, struct  vhu_msg_context *ctx)
2791 {
2792 	int ret;
2793 
2794 	ret = read_fd_message(dev->ifname, sockfd, (char *)&ctx->msg, VHOST_USER_HDR_SIZE,
2795 		ctx->fds, VHOST_MEMORY_MAX_NREGIONS, &ctx->fd_num);
2796 	if (ret <= 0) {
2797 		return ret;
2798 	} else if (ret != VHOST_USER_HDR_SIZE) {
2799 		VHOST_LOG_CONFIG(ERR, "(%s) Unexpected header size read\n", dev->ifname);
2800 		close_msg_fds(ctx);
2801 		return -1;
2802 	}
2803 
2804 	if (ctx->msg.size) {
2805 		if (ctx->msg.size > sizeof(ctx->msg.payload)) {
2806 			VHOST_LOG_CONFIG(ERR, "(%s) invalid msg size: %d\n",
2807 					dev->ifname, ctx->msg.size);
2808 			return -1;
2809 		}
2810 		ret = read(sockfd, &ctx->msg.payload, ctx->msg.size);
2811 		if (ret <= 0)
2812 			return ret;
2813 		if (ret != (int)ctx->msg.size) {
2814 			VHOST_LOG_CONFIG(ERR, "(%s) read control message failed\n", dev->ifname);
2815 			return -1;
2816 		}
2817 	}
2818 
2819 	return ret;
2820 }
2821 
2822 static int
2823 send_vhost_message(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx)
2824 {
2825 	if (!ctx)
2826 		return 0;
2827 
2828 	return send_fd_message(dev->ifname, sockfd, (char *)&ctx->msg,
2829 		VHOST_USER_HDR_SIZE + ctx->msg.size, ctx->fds, ctx->fd_num);
2830 }
2831 
2832 static int
2833 send_vhost_reply(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx)
2834 {
2835 	if (!ctx)
2836 		return 0;
2837 
2838 	ctx->msg.flags &= ~VHOST_USER_VERSION_MASK;
2839 	ctx->msg.flags &= ~VHOST_USER_NEED_REPLY;
2840 	ctx->msg.flags |= VHOST_USER_VERSION;
2841 	ctx->msg.flags |= VHOST_USER_REPLY_MASK;
2842 
2843 	return send_vhost_message(dev, sockfd, ctx);
2844 }
2845 
2846 static int
2847 send_vhost_slave_message(struct virtio_net *dev,
2848 		struct vhu_msg_context *ctx)
2849 {
2850 	int ret;
2851 
2852 	if (ctx->msg.flags & VHOST_USER_NEED_REPLY)
2853 		rte_spinlock_lock(&dev->slave_req_lock);
2854 
2855 	ret = send_vhost_message(dev, dev->slave_req_fd, ctx);
2856 	if (ret < 0 && (ctx->msg.flags & VHOST_USER_NEED_REPLY))
2857 		rte_spinlock_unlock(&dev->slave_req_lock);
2858 
2859 	return ret;
2860 }
2861 
2862 /*
2863  * Allocate a queue pair if it hasn't been allocated yet
2864  */
2865 static int
2866 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2867 			struct vhu_msg_context *ctx)
2868 {
2869 	uint32_t vring_idx;
2870 
2871 	switch (ctx->msg.request.master) {
2872 	case VHOST_USER_SET_VRING_KICK:
2873 	case VHOST_USER_SET_VRING_CALL:
2874 	case VHOST_USER_SET_VRING_ERR:
2875 		vring_idx = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
2876 		break;
2877 	case VHOST_USER_SET_VRING_NUM:
2878 	case VHOST_USER_SET_VRING_BASE:
2879 	case VHOST_USER_GET_VRING_BASE:
2880 	case VHOST_USER_SET_VRING_ENABLE:
2881 		vring_idx = ctx->msg.payload.state.index;
2882 		break;
2883 	case VHOST_USER_SET_VRING_ADDR:
2884 		vring_idx = ctx->msg.payload.addr.index;
2885 		break;
2886 	default:
2887 		return 0;
2888 	}
2889 
2890 	if (vring_idx >= VHOST_MAX_VRING) {
2891 		VHOST_LOG_CONFIG(ERR, "(%s) invalid vring index: %u\n", dev->ifname, vring_idx);
2892 		return -1;
2893 	}
2894 
2895 	if (dev->virtqueue[vring_idx])
2896 		return 0;
2897 
2898 	return alloc_vring_queue(dev, vring_idx);
2899 }
2900 
2901 static void
2902 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2903 {
2904 	unsigned int i = 0;
2905 	unsigned int vq_num = 0;
2906 
2907 	while (vq_num < dev->nr_vring) {
2908 		struct vhost_virtqueue *vq = dev->virtqueue[i];
2909 
2910 		if (vq) {
2911 			rte_spinlock_lock(&vq->access_lock);
2912 			vq_num++;
2913 		}
2914 		i++;
2915 	}
2916 }
2917 
2918 static void
2919 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2920 {
2921 	unsigned int i = 0;
2922 	unsigned int vq_num = 0;
2923 
2924 	while (vq_num < dev->nr_vring) {
2925 		struct vhost_virtqueue *vq = dev->virtqueue[i];
2926 
2927 		if (vq) {
2928 			rte_spinlock_unlock(&vq->access_lock);
2929 			vq_num++;
2930 		}
2931 		i++;
2932 	}
2933 }
2934 
2935 int
2936 vhost_user_msg_handler(int vid, int fd)
2937 {
2938 	struct virtio_net *dev;
2939 	struct vhu_msg_context ctx;
2940 	struct rte_vdpa_device *vdpa_dev;
2941 	int ret;
2942 	int unlock_required = 0;
2943 	bool handled;
2944 	int request;
2945 	uint32_t i;
2946 
2947 	dev = get_device(vid);
2948 	if (dev == NULL)
2949 		return -1;
2950 
2951 	if (!dev->notify_ops) {
2952 		dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2953 		if (!dev->notify_ops) {
2954 			VHOST_LOG_CONFIG(ERR, "(%s) failed to get callback ops for driver\n",
2955 				dev->ifname);
2956 			return -1;
2957 		}
2958 	}
2959 
2960 	ret = read_vhost_message(dev, fd, &ctx);
2961 	if (ret <= 0) {
2962 		if (ret < 0)
2963 			VHOST_LOG_CONFIG(ERR, "(%s) vhost read message failed\n", dev->ifname);
2964 		else
2965 			VHOST_LOG_CONFIG(INFO, "(%s) vhost peer closed\n", dev->ifname);
2966 
2967 		return -1;
2968 	}
2969 
2970 	ret = 0;
2971 	request = ctx.msg.request.master;
2972 	if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2973 			vhost_message_str[request]) {
2974 		if (request != VHOST_USER_IOTLB_MSG)
2975 			VHOST_LOG_CONFIG(INFO, "(%s) read message %s\n",
2976 				dev->ifname, vhost_message_str[request]);
2977 		else
2978 			VHOST_LOG_CONFIG(DEBUG, "(%s) read message %s\n",
2979 				dev->ifname, vhost_message_str[request]);
2980 	} else {
2981 		VHOST_LOG_CONFIG(DEBUG, "(%s) external request %d\n", dev->ifname, request);
2982 	}
2983 
2984 	ret = vhost_user_check_and_alloc_queue_pair(dev, &ctx);
2985 	if (ret < 0) {
2986 		VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc queue\n", dev->ifname);
2987 		return -1;
2988 	}
2989 
2990 	/*
2991 	 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2992 	 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2993 	 * and device is destroyed. destroy_device waits for queues to be
2994 	 * inactive, so it is safe. Otherwise taking the access_lock
2995 	 * would cause a dead lock.
2996 	 */
2997 	switch (request) {
2998 	case VHOST_USER_SET_FEATURES:
2999 	case VHOST_USER_SET_PROTOCOL_FEATURES:
3000 	case VHOST_USER_SET_OWNER:
3001 	case VHOST_USER_SET_MEM_TABLE:
3002 	case VHOST_USER_SET_LOG_BASE:
3003 	case VHOST_USER_SET_LOG_FD:
3004 	case VHOST_USER_SET_VRING_NUM:
3005 	case VHOST_USER_SET_VRING_ADDR:
3006 	case VHOST_USER_SET_VRING_BASE:
3007 	case VHOST_USER_SET_VRING_KICK:
3008 	case VHOST_USER_SET_VRING_CALL:
3009 	case VHOST_USER_SET_VRING_ERR:
3010 	case VHOST_USER_SET_VRING_ENABLE:
3011 	case VHOST_USER_SEND_RARP:
3012 	case VHOST_USER_NET_SET_MTU:
3013 	case VHOST_USER_SET_SLAVE_REQ_FD:
3014 		if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3015 			vhost_user_lock_all_queue_pairs(dev);
3016 			unlock_required = 1;
3017 		}
3018 		break;
3019 	default:
3020 		break;
3021 
3022 	}
3023 
3024 	handled = false;
3025 	if (dev->extern_ops.pre_msg_handle) {
3026 		ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
3027 				(void *)&ctx.msg);
3028 		switch (ret) {
3029 		case RTE_VHOST_MSG_RESULT_REPLY:
3030 			send_vhost_reply(dev, fd, &ctx);
3031 			/* Fall-through */
3032 		case RTE_VHOST_MSG_RESULT_ERR:
3033 		case RTE_VHOST_MSG_RESULT_OK:
3034 			handled = true;
3035 			goto skip_to_post_handle;
3036 		case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3037 		default:
3038 			break;
3039 		}
3040 	}
3041 
3042 	if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
3043 		if (!vhost_message_handlers[request])
3044 			goto skip_to_post_handle;
3045 		ret = vhost_message_handlers[request](&dev, &ctx, fd);
3046 
3047 		switch (ret) {
3048 		case RTE_VHOST_MSG_RESULT_ERR:
3049 			VHOST_LOG_CONFIG(ERR, "(%s) processing %s failed.\n",
3050 					dev->ifname, vhost_message_str[request]);
3051 			handled = true;
3052 			break;
3053 		case RTE_VHOST_MSG_RESULT_OK:
3054 			VHOST_LOG_CONFIG(DEBUG, "(%s) processing %s succeeded.\n",
3055 					dev->ifname, vhost_message_str[request]);
3056 			handled = true;
3057 			break;
3058 		case RTE_VHOST_MSG_RESULT_REPLY:
3059 			VHOST_LOG_CONFIG(DEBUG, "(%s) processing %s succeeded and needs reply.\n",
3060 					dev->ifname, vhost_message_str[request]);
3061 			send_vhost_reply(dev, fd, &ctx);
3062 			handled = true;
3063 			break;
3064 		default:
3065 			break;
3066 		}
3067 	}
3068 
3069 skip_to_post_handle:
3070 	if (ret != RTE_VHOST_MSG_RESULT_ERR &&
3071 			dev->extern_ops.post_msg_handle) {
3072 		ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
3073 				(void *)&ctx.msg);
3074 		switch (ret) {
3075 		case RTE_VHOST_MSG_RESULT_REPLY:
3076 			send_vhost_reply(dev, fd, &ctx);
3077 			/* Fall-through */
3078 		case RTE_VHOST_MSG_RESULT_ERR:
3079 		case RTE_VHOST_MSG_RESULT_OK:
3080 			handled = true;
3081 		case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3082 		default:
3083 			break;
3084 		}
3085 	}
3086 
3087 	/* If message was not handled at this stage, treat it as an error */
3088 	if (!handled) {
3089 		VHOST_LOG_CONFIG(ERR, "(%s) vhost message (req: %d) was not handled.\n",
3090 				dev->ifname, request);
3091 		close_msg_fds(&ctx);
3092 		ret = RTE_VHOST_MSG_RESULT_ERR;
3093 	}
3094 
3095 	/*
3096 	 * If the request required a reply that was already sent,
3097 	 * this optional reply-ack won't be sent as the
3098 	 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
3099 	 */
3100 	if (ctx.msg.flags & VHOST_USER_NEED_REPLY) {
3101 		ctx.msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
3102 		ctx.msg.size = sizeof(ctx.msg.payload.u64);
3103 		ctx.fd_num = 0;
3104 		send_vhost_reply(dev, fd, &ctx);
3105 	} else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
3106 		VHOST_LOG_CONFIG(ERR, "(%s) vhost message handling failed.\n", dev->ifname);
3107 		return -1;
3108 	}
3109 
3110 	for (i = 0; i < dev->nr_vring; i++) {
3111 		struct vhost_virtqueue *vq = dev->virtqueue[i];
3112 		bool cur_ready = vq_is_ready(dev, vq);
3113 
3114 		if (cur_ready != (vq && vq->ready)) {
3115 			vq->ready = cur_ready;
3116 			vhost_user_notify_queue_state(dev, i, cur_ready);
3117 		}
3118 	}
3119 
3120 	if (unlock_required)
3121 		vhost_user_unlock_all_queue_pairs(dev);
3122 
3123 	if (!virtio_is_ready(dev))
3124 		goto out;
3125 
3126 	/*
3127 	 * Virtio is now ready. If not done already, it is time
3128 	 * to notify the application it can process the rings and
3129 	 * configure the vDPA device if present.
3130 	 */
3131 
3132 	if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
3133 		if (dev->notify_ops->new_device(dev->vid) == 0)
3134 			dev->flags |= VIRTIO_DEV_RUNNING;
3135 	}
3136 
3137 	vdpa_dev = dev->vdpa_dev;
3138 	if (!vdpa_dev)
3139 		goto out;
3140 
3141 	if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3142 		if (vdpa_dev->ops->dev_conf(dev->vid))
3143 			VHOST_LOG_CONFIG(ERR, "(%s) failed to configure vDPA device\n",
3144 					dev->ifname);
3145 		else
3146 			dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3147 	}
3148 
3149 out:
3150 	return 0;
3151 }
3152 
3153 static int process_slave_message_reply(struct virtio_net *dev,
3154 				       const struct vhu_msg_context *ctx)
3155 {
3156 	struct vhu_msg_context msg_reply;
3157 	int ret;
3158 
3159 	if ((ctx->msg.flags & VHOST_USER_NEED_REPLY) == 0)
3160 		return 0;
3161 
3162 	ret = read_vhost_message(dev, dev->slave_req_fd, &msg_reply);
3163 	if (ret <= 0) {
3164 		if (ret < 0)
3165 			VHOST_LOG_CONFIG(ERR, "(%s) vhost read slave message reply failed\n",
3166 					dev->ifname);
3167 		else
3168 			VHOST_LOG_CONFIG(INFO, "(%s) vhost peer closed\n", dev->ifname);
3169 		ret = -1;
3170 		goto out;
3171 	}
3172 
3173 	ret = 0;
3174 	if (msg_reply.msg.request.slave != ctx->msg.request.slave) {
3175 		VHOST_LOG_CONFIG(ERR, "(%s) received unexpected msg type (%u), expected %u\n",
3176 				dev->ifname, msg_reply.msg.request.slave, ctx->msg.request.slave);
3177 		ret = -1;
3178 		goto out;
3179 	}
3180 
3181 	ret = msg_reply.msg.payload.u64 ? -1 : 0;
3182 
3183 out:
3184 	rte_spinlock_unlock(&dev->slave_req_lock);
3185 	return ret;
3186 }
3187 
3188 int
3189 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3190 {
3191 	int ret;
3192 	struct vhu_msg_context ctx = {
3193 		.msg = {
3194 			.request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3195 			.flags = VHOST_USER_VERSION,
3196 			.size = sizeof(ctx.msg.payload.iotlb),
3197 			.payload.iotlb = {
3198 				.iova = iova,
3199 				.perm = perm,
3200 				.type = VHOST_IOTLB_MISS,
3201 			},
3202 		},
3203 	};
3204 
3205 	ret = send_vhost_message(dev, dev->slave_req_fd, &ctx);
3206 	if (ret < 0) {
3207 		VHOST_LOG_CONFIG(ERR, "(%s) failed to send IOTLB miss message (%d)\n",
3208 				dev->ifname, ret);
3209 		return ret;
3210 	}
3211 
3212 	return 0;
3213 }
3214 
3215 static int
3216 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3217 {
3218 	int ret;
3219 	struct vhu_msg_context ctx = {
3220 		.msg = {
3221 			.request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3222 			.flags = VHOST_USER_VERSION,
3223 			.size = 0,
3224 		}
3225 	};
3226 
3227 	if (need_reply)
3228 		ctx.msg.flags |= VHOST_USER_NEED_REPLY;
3229 
3230 	ret = send_vhost_slave_message(dev, &ctx);
3231 	if (ret < 0) {
3232 		VHOST_LOG_CONFIG(ERR, "(%s) failed to send config change (%d)\n",
3233 				dev->ifname, ret);
3234 		return ret;
3235 	}
3236 
3237 	return process_slave_message_reply(dev, &ctx);
3238 }
3239 
3240 int
3241 rte_vhost_slave_config_change(int vid, bool need_reply)
3242 {
3243 	struct virtio_net *dev;
3244 
3245 	dev = get_device(vid);
3246 	if (!dev)
3247 		return -ENODEV;
3248 
3249 	return vhost_user_slave_config_change(dev, need_reply);
3250 }
3251 
3252 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3253 						    int index, int fd,
3254 						    uint64_t offset,
3255 						    uint64_t size)
3256 {
3257 	int ret;
3258 	struct vhu_msg_context ctx = {
3259 		.msg = {
3260 			.request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3261 			.flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3262 			.size = sizeof(ctx.msg.payload.area),
3263 			.payload.area = {
3264 				.u64 = index & VHOST_USER_VRING_IDX_MASK,
3265 				.size = size,
3266 				.offset = offset,
3267 			},
3268 		},
3269 	};
3270 
3271 	if (fd < 0)
3272 		ctx.msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3273 	else {
3274 		ctx.fds[0] = fd;
3275 		ctx.fd_num = 1;
3276 	}
3277 
3278 	ret = send_vhost_slave_message(dev, &ctx);
3279 	if (ret < 0) {
3280 		VHOST_LOG_CONFIG(ERR, "(%s) failed to set host notifier (%d)\n",
3281 				dev->ifname, ret);
3282 		return ret;
3283 	}
3284 
3285 	return process_slave_message_reply(dev, &ctx);
3286 }
3287 
3288 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3289 {
3290 	struct virtio_net *dev;
3291 	struct rte_vdpa_device *vdpa_dev;
3292 	int vfio_device_fd, ret = 0;
3293 	uint64_t offset, size;
3294 	unsigned int i, q_start, q_last;
3295 
3296 	dev = get_device(vid);
3297 	if (!dev)
3298 		return -ENODEV;
3299 
3300 	vdpa_dev = dev->vdpa_dev;
3301 	if (vdpa_dev == NULL)
3302 		return -ENODEV;
3303 
3304 	if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3305 	    !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3306 	    !(dev->protocol_features &
3307 			(1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3308 	    !(dev->protocol_features &
3309 			(1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3310 	    !(dev->protocol_features &
3311 			(1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3312 		return -ENOTSUP;
3313 
3314 	if (qid == RTE_VHOST_QUEUE_ALL) {
3315 		q_start = 0;
3316 		q_last = dev->nr_vring - 1;
3317 	} else {
3318 		if (qid >= dev->nr_vring)
3319 			return -EINVAL;
3320 		q_start = qid;
3321 		q_last = qid;
3322 	}
3323 
3324 	RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3325 	RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3326 
3327 	vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3328 	if (vfio_device_fd < 0)
3329 		return -ENOTSUP;
3330 
3331 	if (enable) {
3332 		for (i = q_start; i <= q_last; i++) {
3333 			if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3334 					&size) < 0) {
3335 				ret = -ENOTSUP;
3336 				goto disable;
3337 			}
3338 
3339 			if (vhost_user_slave_set_vring_host_notifier(dev, i,
3340 					vfio_device_fd, offset, size) < 0) {
3341 				ret = -EFAULT;
3342 				goto disable;
3343 			}
3344 		}
3345 	} else {
3346 disable:
3347 		for (i = q_start; i <= q_last; i++) {
3348 			vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3349 					0, 0);
3350 		}
3351 	}
3352 
3353 	return ret;
3354 }
3355