1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation
7 * The vhost-user protocol connection is an external interface, so it must be
8 * robust against invalid inputs.
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.
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.
19 * Do not assume received VhostUserMsg fields contain sensible values!
28 #include <sys/types.h>
31 #ifdef RTE_LIBRTE_VHOST_NUMA
35 #include <rte_common.h>
36 #include <rte_malloc.h>
41 #include "vhost_user.h"
43 #define VIRTIO_MIN_MTU 68
44 #define VIRTIO_MAX_MTU 65535
46 static const char *vhost_message_str[VHOST_USER_MAX] = {
47 [VHOST_USER_NONE] = "VHOST_USER_NONE",
48 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
49 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
50 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
51 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
52 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
53 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
54 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
55 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
56 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
57 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
58 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
59 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
60 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
61 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
62 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
63 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
64 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
65 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
66 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
67 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
68 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
69 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
70 [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
71 [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
80 ret = fstat(fd, &stat);
81 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
85 free_mem_region(struct virtio_net *dev)
88 struct rte_vhost_mem_region *reg;
90 if (!dev || !dev->mem)
93 for (i = 0; i < dev->mem->nregions; i++) {
94 reg = &dev->mem->regions[i];
95 if (reg->host_user_addr) {
96 munmap(reg->mmap_addr, reg->mmap_size);
103 vhost_backend_cleanup(struct virtio_net *dev)
106 free_mem_region(dev);
111 free(dev->guest_pages);
112 dev->guest_pages = NULL;
115 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
119 if (dev->slave_req_fd >= 0) {
120 close(dev->slave_req_fd);
121 dev->slave_req_fd = -1;
126 * This function just returns success at the moment unless
127 * the device hasn't been initialised.
130 vhost_user_set_owner(void)
136 vhost_user_reset_owner(struct virtio_net *dev)
138 struct rte_vdpa_device *vdpa_dev;
141 if (dev->flags & VIRTIO_DEV_RUNNING) {
142 did = dev->vdpa_dev_id;
143 vdpa_dev = rte_vdpa_get_device(did);
144 if (vdpa_dev && vdpa_dev->ops->dev_close)
145 vdpa_dev->ops->dev_close(dev->vid);
146 dev->flags &= ~VIRTIO_DEV_RUNNING;
147 dev->notify_ops->destroy_device(dev->vid);
150 cleanup_device(dev, 0);
156 * The features that we support are requested.
159 vhost_user_get_features(struct virtio_net *dev)
161 uint64_t features = 0;
163 rte_vhost_driver_get_features(dev->ifname, &features);
168 * The queue number that we support are requested.
171 vhost_user_get_queue_num(struct virtio_net *dev)
173 uint32_t queue_num = 0;
175 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
180 * We receive the negotiated features supported by us and the virtio device.
183 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
185 uint64_t vhost_features = 0;
186 struct rte_vdpa_device *vdpa_dev;
189 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
190 if (features & ~vhost_features) {
191 RTE_LOG(ERR, VHOST_CONFIG,
192 "(%d) received invalid negotiated features.\n",
197 if (dev->flags & VIRTIO_DEV_RUNNING) {
198 if (dev->features == features)
202 * Error out if master tries to change features while device is
203 * in running state. The exception being VHOST_F_LOG_ALL, which
204 * is enabled when the live-migration starts.
206 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
207 RTE_LOG(ERR, VHOST_CONFIG,
208 "(%d) features changed while device is running.\n",
213 if (dev->notify_ops->features_changed)
214 dev->notify_ops->features_changed(dev->vid, features);
217 dev->features = features;
219 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
220 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
222 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
224 VHOST_LOG_DEBUG(VHOST_CONFIG,
225 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
227 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
228 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
230 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
231 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
233 * Remove all but first queue pair if MQ hasn't been
234 * negotiated. This is safe because the device is not
235 * running at this stage.
237 while (dev->nr_vring > 2) {
238 struct vhost_virtqueue *vq;
240 vq = dev->virtqueue[--dev->nr_vring];
244 dev->virtqueue[dev->nr_vring] = NULL;
250 did = dev->vdpa_dev_id;
251 vdpa_dev = rte_vdpa_get_device(did);
252 if (vdpa_dev && vdpa_dev->ops->set_features)
253 vdpa_dev->ops->set_features(dev->vid);
259 * The virtio device sends us the size of the descriptor ring.
262 vhost_user_set_vring_num(struct virtio_net *dev,
265 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
267 vq->size = msg->payload.state.num;
269 /* VIRTIO 1.0, 2.4 Virtqueues says:
271 * Queue Size value is always a power of 2. The maximum Queue Size
274 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
275 RTE_LOG(ERR, VHOST_CONFIG,
276 "invalid virtqueue size %u\n", vq->size);
280 if (dev->dequeue_zero_copy) {
282 vq->last_zmbuf_idx = 0;
283 vq->zmbuf_size = vq->size;
284 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
285 sizeof(struct zcopy_mbuf), 0);
286 if (vq->zmbufs == NULL) {
287 RTE_LOG(WARNING, VHOST_CONFIG,
288 "failed to allocate mem for zero copy; "
289 "zero copy is force disabled\n");
290 dev->dequeue_zero_copy = 0;
292 TAILQ_INIT(&vq->zmbuf_list);
295 vq->shadow_used_ring = rte_malloc(NULL,
296 vq->size * sizeof(struct vring_used_elem),
297 RTE_CACHE_LINE_SIZE);
298 if (!vq->shadow_used_ring) {
299 RTE_LOG(ERR, VHOST_CONFIG,
300 "failed to allocate memory for shadow used ring.\n");
304 vq->batch_copy_elems = rte_malloc(NULL,
305 vq->size * sizeof(struct batch_copy_elem),
306 RTE_CACHE_LINE_SIZE);
307 if (!vq->batch_copy_elems) {
308 RTE_LOG(ERR, VHOST_CONFIG,
309 "failed to allocate memory for batching copy.\n");
317 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
318 * same numa node as the memory of vring descriptor.
320 #ifdef RTE_LIBRTE_VHOST_NUMA
321 static struct virtio_net*
322 numa_realloc(struct virtio_net *dev, int index)
324 int oldnode, newnode;
325 struct virtio_net *old_dev;
326 struct vhost_virtqueue *old_vq, *vq;
327 struct zcopy_mbuf *new_zmbuf;
328 struct vring_used_elem *new_shadow_used_ring;
329 struct batch_copy_elem *new_batch_copy_elems;
333 vq = old_vq = dev->virtqueue[index];
335 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
336 MPOL_F_NODE | MPOL_F_ADDR);
338 /* check if we need to reallocate vq */
339 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
340 MPOL_F_NODE | MPOL_F_ADDR);
342 RTE_LOG(ERR, VHOST_CONFIG,
343 "Unable to get vq numa information.\n");
346 if (oldnode != newnode) {
347 RTE_LOG(INFO, VHOST_CONFIG,
348 "reallocate vq from %d to %d node\n", oldnode, newnode);
349 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
353 memcpy(vq, old_vq, sizeof(*vq));
354 TAILQ_INIT(&vq->zmbuf_list);
356 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
357 sizeof(struct zcopy_mbuf), 0, newnode);
359 rte_free(vq->zmbufs);
360 vq->zmbufs = new_zmbuf;
363 new_shadow_used_ring = rte_malloc_socket(NULL,
364 vq->size * sizeof(struct vring_used_elem),
367 if (new_shadow_used_ring) {
368 rte_free(vq->shadow_used_ring);
369 vq->shadow_used_ring = new_shadow_used_ring;
372 new_batch_copy_elems = rte_malloc_socket(NULL,
373 vq->size * sizeof(struct batch_copy_elem),
376 if (new_batch_copy_elems) {
377 rte_free(vq->batch_copy_elems);
378 vq->batch_copy_elems = new_batch_copy_elems;
384 /* check if we need to reallocate dev */
385 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
386 MPOL_F_NODE | MPOL_F_ADDR);
388 RTE_LOG(ERR, VHOST_CONFIG,
389 "Unable to get dev numa information.\n");
392 if (oldnode != newnode) {
393 RTE_LOG(INFO, VHOST_CONFIG,
394 "reallocate dev from %d to %d node\n",
396 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
402 memcpy(dev, old_dev, sizeof(*dev));
407 dev->virtqueue[index] = vq;
408 vhost_devices[dev->vid] = dev;
411 vhost_user_iotlb_init(dev, index);
416 static struct virtio_net*
417 numa_realloc(struct virtio_net *dev, int index __rte_unused)
423 /* Converts QEMU virtual address to Vhost virtual address. */
425 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
427 struct rte_vhost_mem_region *r;
430 /* Find the region where the address lives. */
431 for (i = 0; i < dev->mem->nregions; i++) {
432 r = &dev->mem->regions[i];
434 if (qva >= r->guest_user_addr &&
435 qva < r->guest_user_addr + r->size) {
437 if (unlikely(*len > r->guest_user_addr + r->size - qva))
438 *len = r->guest_user_addr + r->size - qva;
440 return qva - r->guest_user_addr +
451 * Converts ring address to Vhost virtual address.
452 * If IOMMU is enabled, the ring address is a guest IO virtual address,
453 * else it is a QEMU virtual address.
456 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
457 uint64_t ra, uint64_t *size)
459 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
462 vva = vhost_user_iotlb_cache_find(vq, ra,
463 size, VHOST_ACCESS_RW);
465 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
470 return qva_to_vva(dev, ra, size);
473 static struct virtio_net *
474 translate_ring_addresses(struct virtio_net *dev, int vq_index)
476 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
477 struct vhost_vring_addr *addr = &vq->ring_addrs;
480 /* The addresses are converted from QEMU virtual to Vhost virtual. */
481 if (vq->desc && vq->avail && vq->used)
484 len = sizeof(struct vring_desc) * vq->size;
485 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
486 vq, addr->desc_user_addr, &len);
487 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
488 RTE_LOG(DEBUG, VHOST_CONFIG,
489 "(%d) failed to map desc ring.\n",
494 dev = numa_realloc(dev, vq_index);
495 vq = dev->virtqueue[vq_index];
496 addr = &vq->ring_addrs;
498 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
499 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
500 vq, addr->avail_user_addr, &len);
501 if (vq->avail == 0 ||
502 len != sizeof(struct vring_avail) +
503 sizeof(uint16_t) * vq->size) {
504 RTE_LOG(DEBUG, VHOST_CONFIG,
505 "(%d) failed to map avail ring.\n",
510 len = sizeof(struct vring_used) +
511 sizeof(struct vring_used_elem) * vq->size;
512 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
513 vq, addr->used_user_addr, &len);
514 if (vq->used == 0 || len != sizeof(struct vring_used) +
515 sizeof(struct vring_used_elem) * vq->size) {
516 RTE_LOG(DEBUG, VHOST_CONFIG,
517 "(%d) failed to map used ring.\n",
522 if (vq->last_used_idx != vq->used->idx) {
523 RTE_LOG(WARNING, VHOST_CONFIG,
524 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
525 "some packets maybe resent for Tx and dropped for Rx\n",
526 vq->last_used_idx, vq->used->idx);
527 vq->last_used_idx = vq->used->idx;
528 vq->last_avail_idx = vq->used->idx;
531 vq->log_guest_addr = addr->log_guest_addr;
533 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
535 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
536 dev->vid, vq->avail);
537 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
539 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
540 dev->vid, vq->log_guest_addr);
546 * The virtio device sends us the desc, used and avail ring addresses.
547 * This function then converts these to our address space.
550 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
552 struct vhost_virtqueue *vq;
553 struct vhost_vring_addr *addr = &msg->payload.addr;
554 struct virtio_net *dev = *pdev;
556 if (dev->mem == NULL)
559 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
560 vq = dev->virtqueue[msg->payload.addr.index];
563 * Rings addresses should not be interpreted as long as the ring is not
564 * started and enabled
566 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
568 vring_invalidate(dev, vq);
570 if (vq->enabled && (dev->features &
571 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
572 dev = translate_ring_addresses(dev, msg->payload.addr.index);
583 * The virtio device sends us the available ring last used index.
586 vhost_user_set_vring_base(struct virtio_net *dev,
589 dev->virtqueue[msg->payload.state.index]->last_used_idx =
590 msg->payload.state.num;
591 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
592 msg->payload.state.num;
598 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
599 uint64_t host_phys_addr, uint64_t size)
601 struct guest_page *page, *last_page;
603 if (dev->nr_guest_pages == dev->max_guest_pages) {
604 dev->max_guest_pages *= 2;
605 dev->guest_pages = realloc(dev->guest_pages,
606 dev->max_guest_pages * sizeof(*page));
607 if (!dev->guest_pages) {
608 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
613 if (dev->nr_guest_pages > 0) {
614 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
615 /* merge if the two pages are continuous */
616 if (host_phys_addr == last_page->host_phys_addr +
618 last_page->size += size;
623 page = &dev->guest_pages[dev->nr_guest_pages++];
624 page->guest_phys_addr = guest_phys_addr;
625 page->host_phys_addr = host_phys_addr;
632 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
635 uint64_t reg_size = reg->size;
636 uint64_t host_user_addr = reg->host_user_addr;
637 uint64_t guest_phys_addr = reg->guest_phys_addr;
638 uint64_t host_phys_addr;
641 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
642 size = page_size - (guest_phys_addr & (page_size - 1));
643 size = RTE_MIN(size, reg_size);
645 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
648 host_user_addr += size;
649 guest_phys_addr += size;
652 while (reg_size > 0) {
653 size = RTE_MIN(reg_size, page_size);
654 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
656 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
660 host_user_addr += size;
661 guest_phys_addr += size;
668 #ifdef RTE_LIBRTE_VHOST_DEBUG
669 /* TODO: enable it only in debug mode? */
671 dump_guest_pages(struct virtio_net *dev)
674 struct guest_page *page;
676 for (i = 0; i < dev->nr_guest_pages; i++) {
677 page = &dev->guest_pages[i];
679 RTE_LOG(INFO, VHOST_CONFIG,
680 "guest physical page region %u\n"
681 "\t guest_phys_addr: %" PRIx64 "\n"
682 "\t host_phys_addr : %" PRIx64 "\n"
683 "\t size : %" PRIx64 "\n",
685 page->guest_phys_addr,
686 page->host_phys_addr,
691 #define dump_guest_pages(dev)
695 vhost_memory_changed(struct VhostUserMemory *new,
696 struct rte_vhost_memory *old)
700 if (new->nregions != old->nregions)
703 for (i = 0; i < new->nregions; ++i) {
704 VhostUserMemoryRegion *new_r = &new->regions[i];
705 struct rte_vhost_mem_region *old_r = &old->regions[i];
707 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
709 if (new_r->memory_size != old_r->size)
711 if (new_r->userspace_addr != old_r->guest_user_addr)
719 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
721 struct virtio_net *dev = *pdev;
722 struct VhostUserMemory memory = pmsg->payload.memory;
723 struct rte_vhost_mem_region *reg;
726 uint64_t mmap_offset;
732 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
733 RTE_LOG(ERR, VHOST_CONFIG,
734 "too many memory regions (%u)\n", memory.nregions);
738 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
739 RTE_LOG(INFO, VHOST_CONFIG,
740 "(%d) memory regions not changed\n", dev->vid);
742 for (i = 0; i < memory.nregions; i++)
749 free_mem_region(dev);
754 dev->nr_guest_pages = 0;
755 if (!dev->guest_pages) {
756 dev->max_guest_pages = 8;
757 dev->guest_pages = malloc(dev->max_guest_pages *
758 sizeof(struct guest_page));
759 if (dev->guest_pages == NULL) {
760 RTE_LOG(ERR, VHOST_CONFIG,
761 "(%d) failed to allocate memory "
762 "for dev->guest_pages\n",
768 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
769 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
770 if (dev->mem == NULL) {
771 RTE_LOG(ERR, VHOST_CONFIG,
772 "(%d) failed to allocate memory for dev->mem\n",
776 dev->mem->nregions = memory.nregions;
778 for (i = 0; i < memory.nregions; i++) {
780 reg = &dev->mem->regions[i];
782 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
783 reg->guest_user_addr = memory.regions[i].userspace_addr;
784 reg->size = memory.regions[i].memory_size;
787 mmap_offset = memory.regions[i].mmap_offset;
789 /* Check for memory_size + mmap_offset overflow */
790 if (mmap_offset >= -reg->size) {
791 RTE_LOG(ERR, VHOST_CONFIG,
792 "mmap_offset (%#"PRIx64") and memory_size "
793 "(%#"PRIx64") overflow\n",
794 mmap_offset, reg->size);
798 mmap_size = reg->size + mmap_offset;
800 /* mmap() without flag of MAP_ANONYMOUS, should be called
801 * with length argument aligned with hugepagesz at older
802 * longterm version Linux, like 2.6.32 and 3.2.72, or
803 * mmap() will fail with EINVAL.
805 * to avoid failure, make sure in caller to keep length
808 alignment = get_blk_size(fd);
809 if (alignment == (uint64_t)-1) {
810 RTE_LOG(ERR, VHOST_CONFIG,
811 "couldn't get hugepage size through fstat\n");
814 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
816 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
817 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
818 MAP_SHARED | populate, fd, 0);
820 if (mmap_addr == MAP_FAILED) {
821 RTE_LOG(ERR, VHOST_CONFIG,
822 "mmap region %u failed.\n", i);
826 reg->mmap_addr = mmap_addr;
827 reg->mmap_size = mmap_size;
828 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
831 if (dev->dequeue_zero_copy)
832 if (add_guest_pages(dev, reg, alignment) < 0) {
833 RTE_LOG(ERR, VHOST_CONFIG,
834 "adding guest pages to region %u failed.\n",
839 RTE_LOG(INFO, VHOST_CONFIG,
840 "guest memory region %u, size: 0x%" PRIx64 "\n"
841 "\t guest physical addr: 0x%" PRIx64 "\n"
842 "\t guest virtual addr: 0x%" PRIx64 "\n"
843 "\t host virtual addr: 0x%" PRIx64 "\n"
844 "\t mmap addr : 0x%" PRIx64 "\n"
845 "\t mmap size : 0x%" PRIx64 "\n"
846 "\t mmap align: 0x%" PRIx64 "\n"
847 "\t mmap off : 0x%" PRIx64 "\n",
849 reg->guest_phys_addr,
850 reg->guest_user_addr,
852 (uint64_t)(uintptr_t)mmap_addr,
858 for (i = 0; i < dev->nr_vring; i++) {
859 struct vhost_virtqueue *vq = dev->virtqueue[i];
861 if (vq->desc || vq->avail || vq->used) {
863 * If the memory table got updated, the ring addresses
864 * need to be translated again as virtual addresses have
867 vring_invalidate(dev, vq);
869 dev = translate_ring_addresses(dev, i);
877 dump_guest_pages(dev);
882 free_mem_region(dev);
889 vq_is_ready(struct vhost_virtqueue *vq)
891 return vq && vq->desc && vq->avail && vq->used &&
892 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
893 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
897 virtio_is_ready(struct virtio_net *dev)
899 struct vhost_virtqueue *vq;
902 if (dev->nr_vring == 0)
905 for (i = 0; i < dev->nr_vring; i++) {
906 vq = dev->virtqueue[i];
908 if (!vq_is_ready(vq))
912 RTE_LOG(INFO, VHOST_CONFIG,
913 "virtio is now ready for processing.\n");
918 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
920 struct vhost_vring_file file;
921 struct vhost_virtqueue *vq;
923 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
924 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
925 file.fd = VIRTIO_INVALID_EVENTFD;
927 file.fd = pmsg->fds[0];
928 RTE_LOG(INFO, VHOST_CONFIG,
929 "vring call idx:%d file:%d\n", file.index, file.fd);
931 vq = dev->virtqueue[file.index];
935 vq->callfd = file.fd;
939 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
941 struct vhost_vring_file file;
942 struct vhost_virtqueue *vq;
943 struct virtio_net *dev = *pdev;
945 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
946 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
947 file.fd = VIRTIO_INVALID_EVENTFD;
949 file.fd = pmsg->fds[0];
950 RTE_LOG(INFO, VHOST_CONFIG,
951 "vring kick idx:%d file:%d\n", file.index, file.fd);
953 /* Interpret ring addresses only when ring is started. */
954 dev = translate_ring_addresses(dev, file.index);
960 vq = dev->virtqueue[file.index];
963 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
964 * the ring starts already enabled. Otherwise, it is enabled via
965 * the SET_VRING_ENABLE message.
967 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
972 vq->kickfd = file.fd;
976 free_zmbufs(struct vhost_virtqueue *vq)
978 struct zcopy_mbuf *zmbuf, *next;
980 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
981 zmbuf != NULL; zmbuf = next) {
982 next = TAILQ_NEXT(zmbuf, next);
984 rte_pktmbuf_free(zmbuf->mbuf);
985 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
988 rte_free(vq->zmbufs);
992 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
995 vhost_user_get_vring_base(struct virtio_net *dev,
998 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
999 struct rte_vdpa_device *vdpa_dev;
1002 /* We have to stop the queue (virtio) if it is running. */
1003 if (dev->flags & VIRTIO_DEV_RUNNING) {
1004 did = dev->vdpa_dev_id;
1005 vdpa_dev = rte_vdpa_get_device(did);
1006 if (vdpa_dev && vdpa_dev->ops->dev_close)
1007 vdpa_dev->ops->dev_close(dev->vid);
1008 dev->flags &= ~VIRTIO_DEV_RUNNING;
1009 dev->notify_ops->destroy_device(dev->vid);
1012 dev->flags &= ~VIRTIO_DEV_READY;
1013 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1015 /* Here we are safe to get the last avail index */
1016 msg->payload.state.num = vq->last_avail_idx;
1018 RTE_LOG(INFO, VHOST_CONFIG,
1019 "vring base idx:%d file:%d\n", msg->payload.state.index,
1020 msg->payload.state.num);
1022 * Based on current qemu vhost-user implementation, this message is
1023 * sent and only sent in vhost_vring_stop.
1024 * TODO: cleanup the vring, it isn't usable since here.
1026 if (vq->kickfd >= 0)
1029 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1031 if (vq->callfd >= 0)
1034 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1036 if (dev->dequeue_zero_copy)
1038 rte_free(vq->shadow_used_ring);
1039 vq->shadow_used_ring = NULL;
1041 rte_free(vq->batch_copy_elems);
1042 vq->batch_copy_elems = NULL;
1048 * when virtio queues are ready to work, qemu will send us to
1049 * enable the virtio queue pair.
1052 vhost_user_set_vring_enable(struct virtio_net *dev,
1055 int enable = (int)msg->payload.state.num;
1056 int index = (int)msg->payload.state.index;
1057 struct rte_vdpa_device *vdpa_dev;
1060 RTE_LOG(INFO, VHOST_CONFIG,
1061 "set queue enable: %d to qp idx: %d\n",
1064 did = dev->vdpa_dev_id;
1065 vdpa_dev = rte_vdpa_get_device(did);
1066 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1067 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1069 if (dev->notify_ops->vring_state_changed)
1070 dev->notify_ops->vring_state_changed(dev->vid,
1073 dev->virtqueue[index]->enabled = enable;
1079 vhost_user_get_protocol_features(struct virtio_net *dev,
1080 struct VhostUserMsg *msg)
1082 uint64_t features, protocol_features;
1084 rte_vhost_driver_get_features(dev->ifname, &features);
1085 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1088 * REPLY_ACK protocol feature is only mandatory for now
1089 * for IOMMU feature. If IOMMU is explicitly disabled by the
1090 * application, disable also REPLY_ACK feature for older buggy
1091 * Qemu versions (from v2.7.0 to v2.9.0).
1093 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1094 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1096 msg->payload.u64 = protocol_features;
1097 msg->size = sizeof(msg->payload.u64);
1101 vhost_user_set_protocol_features(struct virtio_net *dev,
1102 uint64_t protocol_features)
1104 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1107 dev->protocol_features = protocol_features;
1111 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1113 int fd = msg->fds[0];
1118 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1122 if (msg->size != sizeof(VhostUserLog)) {
1123 RTE_LOG(ERR, VHOST_CONFIG,
1124 "invalid log base msg size: %"PRId32" != %d\n",
1125 msg->size, (int)sizeof(VhostUserLog));
1129 size = msg->payload.log.mmap_size;
1130 off = msg->payload.log.mmap_offset;
1132 /* Don't allow mmap_offset to point outside the mmap region */
1134 RTE_LOG(ERR, VHOST_CONFIG,
1135 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1140 RTE_LOG(INFO, VHOST_CONFIG,
1141 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1145 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1146 * fail when offset is not page size aligned.
1148 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1150 if (addr == MAP_FAILED) {
1151 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1156 * Free previously mapped log memory on occasionally
1157 * multiple VHOST_USER_SET_LOG_BASE.
1159 if (dev->log_addr) {
1160 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1162 dev->log_addr = (uint64_t)(uintptr_t)addr;
1163 dev->log_base = dev->log_addr + off;
1164 dev->log_size = size;
1170 * An rarp packet is constructed and broadcasted to notify switches about
1171 * the new location of the migrated VM, so that packets from outside will
1172 * not be lost after migration.
1174 * However, we don't actually "send" a rarp packet here, instead, we set
1175 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1178 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1180 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1181 struct rte_vdpa_device *vdpa_dev;
1184 RTE_LOG(DEBUG, VHOST_CONFIG,
1185 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1186 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1187 memcpy(dev->mac.addr_bytes, mac, 6);
1190 * Set the flag to inject a RARP broadcast packet at
1191 * rte_vhost_dequeue_burst().
1193 * rte_smp_wmb() is for making sure the mac is copied
1194 * before the flag is set.
1197 rte_atomic16_set(&dev->broadcast_rarp, 1);
1198 did = dev->vdpa_dev_id;
1199 vdpa_dev = rte_vdpa_get_device(did);
1200 if (vdpa_dev && vdpa_dev->ops->migration_done)
1201 vdpa_dev->ops->migration_done(dev->vid);
1207 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1209 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1210 msg->payload.u64 > VIRTIO_MAX_MTU) {
1211 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1217 dev->mtu = msg->payload.u64;
1223 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1225 int fd = msg->fds[0];
1228 RTE_LOG(ERR, VHOST_CONFIG,
1229 "Invalid file descriptor for slave channel (%d)\n",
1234 dev->slave_req_fd = fd;
1240 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1242 struct vhost_vring_addr *ra;
1243 uint64_t start, end;
1246 end = start + imsg->size;
1248 ra = &vq->ring_addrs;
1249 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1251 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1253 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1260 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1261 struct vhost_iotlb_msg *imsg)
1263 uint64_t istart, iend, vstart, vend;
1265 istart = imsg->iova;
1266 iend = istart + imsg->size - 1;
1268 vstart = (uintptr_t)vq->desc;
1269 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1270 if (vstart <= iend && istart <= vend)
1273 vstart = (uintptr_t)vq->avail;
1274 vend = vstart + sizeof(struct vring_avail);
1275 vend += sizeof(uint16_t) * vq->size - 1;
1276 if (vstart <= iend && istart <= vend)
1279 vstart = (uintptr_t)vq->used;
1280 vend = vstart + sizeof(struct vring_used);
1281 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1282 if (vstart <= iend && istart <= vend)
1289 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1291 struct virtio_net *dev = *pdev;
1292 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1296 switch (imsg->type) {
1297 case VHOST_IOTLB_UPDATE:
1299 vva = qva_to_vva(dev, imsg->uaddr, &len);
1303 for (i = 0; i < dev->nr_vring; i++) {
1304 struct vhost_virtqueue *vq = dev->virtqueue[i];
1306 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1309 if (is_vring_iotlb_update(vq, imsg))
1310 *pdev = dev = translate_ring_addresses(dev, i);
1313 case VHOST_IOTLB_INVALIDATE:
1314 for (i = 0; i < dev->nr_vring; i++) {
1315 struct vhost_virtqueue *vq = dev->virtqueue[i];
1317 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1320 if (is_vring_iotlb_invalidate(vq, imsg))
1321 vring_invalidate(dev, vq);
1325 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1333 /* return bytes# of read on success or negative val on failure. */
1335 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1339 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1340 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1344 if (msg && msg->size) {
1345 if (msg->size > sizeof(msg->payload)) {
1346 RTE_LOG(ERR, VHOST_CONFIG,
1347 "invalid msg size: %d\n", msg->size);
1350 ret = read(sockfd, &msg->payload, msg->size);
1353 if (ret != (int)msg->size) {
1354 RTE_LOG(ERR, VHOST_CONFIG,
1355 "read control message failed\n");
1364 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1369 return send_fd_message(sockfd, (char *)msg,
1370 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1374 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1379 msg->flags &= ~VHOST_USER_VERSION_MASK;
1380 msg->flags &= ~VHOST_USER_NEED_REPLY;
1381 msg->flags |= VHOST_USER_VERSION;
1382 msg->flags |= VHOST_USER_REPLY_MASK;
1384 return send_vhost_message(sockfd, msg, NULL, 0);
1388 * Allocate a queue pair if it hasn't been allocated yet
1391 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1395 switch (msg->request.master) {
1396 case VHOST_USER_SET_VRING_KICK:
1397 case VHOST_USER_SET_VRING_CALL:
1398 case VHOST_USER_SET_VRING_ERR:
1399 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1401 case VHOST_USER_SET_VRING_NUM:
1402 case VHOST_USER_SET_VRING_BASE:
1403 case VHOST_USER_SET_VRING_ENABLE:
1404 vring_idx = msg->payload.state.index;
1406 case VHOST_USER_SET_VRING_ADDR:
1407 vring_idx = msg->payload.addr.index;
1413 if (vring_idx >= VHOST_MAX_VRING) {
1414 RTE_LOG(ERR, VHOST_CONFIG,
1415 "invalid vring index: %u\n", vring_idx);
1419 if (dev->virtqueue[vring_idx])
1422 return alloc_vring_queue(dev, vring_idx);
1426 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1429 unsigned int vq_num = 0;
1431 while (vq_num < dev->nr_vring) {
1432 struct vhost_virtqueue *vq = dev->virtqueue[i];
1435 rte_spinlock_lock(&vq->access_lock);
1443 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1446 unsigned int vq_num = 0;
1448 while (vq_num < dev->nr_vring) {
1449 struct vhost_virtqueue *vq = dev->virtqueue[i];
1452 rte_spinlock_unlock(&vq->access_lock);
1460 vhost_user_msg_handler(int vid, int fd)
1462 struct virtio_net *dev;
1463 struct VhostUserMsg msg;
1464 struct rte_vdpa_device *vdpa_dev;
1467 int unlock_required = 0;
1468 uint32_t skip_master = 0;
1470 dev = get_device(vid);
1474 if (!dev->notify_ops) {
1475 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1476 if (!dev->notify_ops) {
1477 RTE_LOG(ERR, VHOST_CONFIG,
1478 "failed to get callback ops for driver %s\n",
1484 ret = read_vhost_message(fd, &msg);
1485 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1487 RTE_LOG(ERR, VHOST_CONFIG,
1488 "vhost read message failed\n");
1490 RTE_LOG(INFO, VHOST_CONFIG,
1491 "vhost peer closed\n");
1493 RTE_LOG(ERR, VHOST_CONFIG,
1494 "vhost read incorrect message\n");
1500 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1501 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1502 vhost_message_str[msg.request.master]);
1504 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1505 vhost_message_str[msg.request.master]);
1507 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1509 RTE_LOG(ERR, VHOST_CONFIG,
1510 "failed to alloc queue\n");
1515 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1516 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1517 * and device is destroyed. destroy_device waits for queues to be
1518 * inactive, so it is safe. Otherwise taking the access_lock
1519 * would cause a dead lock.
1521 switch (msg.request.master) {
1522 case VHOST_USER_SET_FEATURES:
1523 case VHOST_USER_SET_PROTOCOL_FEATURES:
1524 case VHOST_USER_SET_OWNER:
1525 case VHOST_USER_SET_MEM_TABLE:
1526 case VHOST_USER_SET_LOG_BASE:
1527 case VHOST_USER_SET_LOG_FD:
1528 case VHOST_USER_SET_VRING_NUM:
1529 case VHOST_USER_SET_VRING_ADDR:
1530 case VHOST_USER_SET_VRING_BASE:
1531 case VHOST_USER_SET_VRING_KICK:
1532 case VHOST_USER_SET_VRING_CALL:
1533 case VHOST_USER_SET_VRING_ERR:
1534 case VHOST_USER_SET_VRING_ENABLE:
1535 case VHOST_USER_SEND_RARP:
1536 case VHOST_USER_NET_SET_MTU:
1537 case VHOST_USER_SET_SLAVE_REQ_FD:
1538 vhost_user_lock_all_queue_pairs(dev);
1539 unlock_required = 1;
1546 if (dev->extern_ops.pre_msg_handle) {
1547 uint32_t need_reply;
1549 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1550 (void *)&msg, &need_reply, &skip_master);
1555 send_vhost_reply(fd, &msg);
1558 goto skip_to_post_handle;
1561 switch (msg.request.master) {
1562 case VHOST_USER_GET_FEATURES:
1563 msg.payload.u64 = vhost_user_get_features(dev);
1564 msg.size = sizeof(msg.payload.u64);
1565 send_vhost_reply(fd, &msg);
1567 case VHOST_USER_SET_FEATURES:
1568 ret = vhost_user_set_features(dev, msg.payload.u64);
1573 case VHOST_USER_GET_PROTOCOL_FEATURES:
1574 vhost_user_get_protocol_features(dev, &msg);
1575 send_vhost_reply(fd, &msg);
1577 case VHOST_USER_SET_PROTOCOL_FEATURES:
1578 vhost_user_set_protocol_features(dev, msg.payload.u64);
1581 case VHOST_USER_SET_OWNER:
1582 vhost_user_set_owner();
1584 case VHOST_USER_RESET_OWNER:
1585 vhost_user_reset_owner(dev);
1588 case VHOST_USER_SET_MEM_TABLE:
1589 ret = vhost_user_set_mem_table(&dev, &msg);
1592 case VHOST_USER_SET_LOG_BASE:
1593 vhost_user_set_log_base(dev, &msg);
1595 /* it needs a reply */
1596 msg.size = sizeof(msg.payload.u64);
1597 send_vhost_reply(fd, &msg);
1599 case VHOST_USER_SET_LOG_FD:
1601 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1604 case VHOST_USER_SET_VRING_NUM:
1605 vhost_user_set_vring_num(dev, &msg);
1607 case VHOST_USER_SET_VRING_ADDR:
1608 vhost_user_set_vring_addr(&dev, &msg);
1610 case VHOST_USER_SET_VRING_BASE:
1611 vhost_user_set_vring_base(dev, &msg);
1614 case VHOST_USER_GET_VRING_BASE:
1615 vhost_user_get_vring_base(dev, &msg);
1616 msg.size = sizeof(msg.payload.state);
1617 send_vhost_reply(fd, &msg);
1620 case VHOST_USER_SET_VRING_KICK:
1621 vhost_user_set_vring_kick(&dev, &msg);
1623 case VHOST_USER_SET_VRING_CALL:
1624 vhost_user_set_vring_call(dev, &msg);
1627 case VHOST_USER_SET_VRING_ERR:
1628 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1630 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1633 case VHOST_USER_GET_QUEUE_NUM:
1634 msg.payload.u64 = (uint64_t)vhost_user_get_queue_num(dev);
1635 msg.size = sizeof(msg.payload.u64);
1636 send_vhost_reply(fd, &msg);
1639 case VHOST_USER_SET_VRING_ENABLE:
1640 vhost_user_set_vring_enable(dev, &msg);
1642 case VHOST_USER_SEND_RARP:
1643 vhost_user_send_rarp(dev, &msg);
1646 case VHOST_USER_NET_SET_MTU:
1647 ret = vhost_user_net_set_mtu(dev, &msg);
1650 case VHOST_USER_SET_SLAVE_REQ_FD:
1651 ret = vhost_user_set_req_fd(dev, &msg);
1654 case VHOST_USER_IOTLB_MSG:
1655 ret = vhost_user_iotlb_msg(&dev, &msg);
1663 skip_to_post_handle:
1664 if (dev->extern_ops.post_msg_handle) {
1665 uint32_t need_reply;
1667 ret = (*dev->extern_ops.post_msg_handle)(
1668 dev->vid, (void *)&msg, &need_reply);
1673 send_vhost_reply(fd, &msg);
1677 if (unlock_required)
1678 vhost_user_unlock_all_queue_pairs(dev);
1680 if (msg.flags & VHOST_USER_NEED_REPLY) {
1681 msg.payload.u64 = !!ret;
1682 msg.size = sizeof(msg.payload.u64);
1683 send_vhost_reply(fd, &msg);
1686 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1687 dev->flags |= VIRTIO_DEV_READY;
1689 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1690 if (dev->dequeue_zero_copy) {
1691 RTE_LOG(INFO, VHOST_CONFIG,
1692 "dequeue zero copy is enabled\n");
1695 if (dev->notify_ops->new_device(dev->vid) == 0)
1696 dev->flags |= VIRTIO_DEV_RUNNING;
1700 did = dev->vdpa_dev_id;
1701 vdpa_dev = rte_vdpa_get_device(did);
1702 if (vdpa_dev && virtio_is_ready(dev) &&
1703 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1704 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1705 if (vdpa_dev->ops->dev_conf)
1706 vdpa_dev->ops->dev_conf(dev->vid);
1707 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1714 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1717 struct VhostUserMsg msg = {
1718 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1719 .flags = VHOST_USER_VERSION,
1720 .size = sizeof(msg.payload.iotlb),
1724 .type = VHOST_IOTLB_MISS,
1728 ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1730 RTE_LOG(ERR, VHOST_CONFIG,
1731 "Failed to send IOTLB miss message (%d)\n",