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 vhost_destroy_device_notify(dev);
140 cleanup_device(dev, 0);
146 * The features that we support are requested.
149 vhost_user_get_features(struct virtio_net *dev)
151 uint64_t features = 0;
153 rte_vhost_driver_get_features(dev->ifname, &features);
158 * The queue number that we support are requested.
161 vhost_user_get_queue_num(struct virtio_net *dev)
163 uint32_t queue_num = 0;
165 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
170 * We receive the negotiated features supported by us and the virtio device.
173 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
175 uint64_t vhost_features = 0;
176 struct rte_vdpa_device *vdpa_dev;
179 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
180 if (features & ~vhost_features) {
181 RTE_LOG(ERR, VHOST_CONFIG,
182 "(%d) received invalid negotiated features.\n",
187 if (dev->flags & VIRTIO_DEV_RUNNING) {
188 if (dev->features == features)
192 * Error out if master tries to change features while device is
193 * in running state. The exception being VHOST_F_LOG_ALL, which
194 * is enabled when the live-migration starts.
196 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
197 RTE_LOG(ERR, VHOST_CONFIG,
198 "(%d) features changed while device is running.\n",
203 if (dev->notify_ops->features_changed)
204 dev->notify_ops->features_changed(dev->vid, features);
207 dev->features = features;
209 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
210 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
212 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
214 VHOST_LOG_DEBUG(VHOST_CONFIG,
215 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
217 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
218 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
220 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
221 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
223 * Remove all but first queue pair if MQ hasn't been
224 * negotiated. This is safe because the device is not
225 * running at this stage.
227 while (dev->nr_vring > 2) {
228 struct vhost_virtqueue *vq;
230 vq = dev->virtqueue[--dev->nr_vring];
234 dev->virtqueue[dev->nr_vring] = NULL;
240 did = dev->vdpa_dev_id;
241 vdpa_dev = rte_vdpa_get_device(did);
242 if (vdpa_dev && vdpa_dev->ops->set_features)
243 vdpa_dev->ops->set_features(dev->vid);
249 * The virtio device sends us the size of the descriptor ring.
252 vhost_user_set_vring_num(struct virtio_net *dev,
255 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
257 vq->size = msg->payload.state.num;
259 /* VIRTIO 1.0, 2.4 Virtqueues says:
261 * Queue Size value is always a power of 2. The maximum Queue Size
264 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
265 RTE_LOG(ERR, VHOST_CONFIG,
266 "invalid virtqueue size %u\n", vq->size);
270 if (dev->dequeue_zero_copy) {
272 vq->last_zmbuf_idx = 0;
273 vq->zmbuf_size = vq->size;
274 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
275 sizeof(struct zcopy_mbuf), 0);
276 if (vq->zmbufs == NULL) {
277 RTE_LOG(WARNING, VHOST_CONFIG,
278 "failed to allocate mem for zero copy; "
279 "zero copy is force disabled\n");
280 dev->dequeue_zero_copy = 0;
282 TAILQ_INIT(&vq->zmbuf_list);
285 vq->shadow_used_ring = rte_malloc(NULL,
286 vq->size * sizeof(struct vring_used_elem),
287 RTE_CACHE_LINE_SIZE);
288 if (!vq->shadow_used_ring) {
289 RTE_LOG(ERR, VHOST_CONFIG,
290 "failed to allocate memory for shadow used ring.\n");
294 vq->batch_copy_elems = rte_malloc(NULL,
295 vq->size * sizeof(struct batch_copy_elem),
296 RTE_CACHE_LINE_SIZE);
297 if (!vq->batch_copy_elems) {
298 RTE_LOG(ERR, VHOST_CONFIG,
299 "failed to allocate memory for batching copy.\n");
307 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
308 * same numa node as the memory of vring descriptor.
310 #ifdef RTE_LIBRTE_VHOST_NUMA
311 static struct virtio_net*
312 numa_realloc(struct virtio_net *dev, int index)
314 int oldnode, newnode;
315 struct virtio_net *old_dev;
316 struct vhost_virtqueue *old_vq, *vq;
317 struct zcopy_mbuf *new_zmbuf;
318 struct vring_used_elem *new_shadow_used_ring;
319 struct batch_copy_elem *new_batch_copy_elems;
323 vq = old_vq = dev->virtqueue[index];
325 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
326 MPOL_F_NODE | MPOL_F_ADDR);
328 /* check if we need to reallocate vq */
329 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
330 MPOL_F_NODE | MPOL_F_ADDR);
332 RTE_LOG(ERR, VHOST_CONFIG,
333 "Unable to get vq numa information.\n");
336 if (oldnode != newnode) {
337 RTE_LOG(INFO, VHOST_CONFIG,
338 "reallocate vq from %d to %d node\n", oldnode, newnode);
339 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
343 memcpy(vq, old_vq, sizeof(*vq));
344 TAILQ_INIT(&vq->zmbuf_list);
346 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
347 sizeof(struct zcopy_mbuf), 0, newnode);
349 rte_free(vq->zmbufs);
350 vq->zmbufs = new_zmbuf;
353 new_shadow_used_ring = rte_malloc_socket(NULL,
354 vq->size * sizeof(struct vring_used_elem),
357 if (new_shadow_used_ring) {
358 rte_free(vq->shadow_used_ring);
359 vq->shadow_used_ring = new_shadow_used_ring;
362 new_batch_copy_elems = rte_malloc_socket(NULL,
363 vq->size * sizeof(struct batch_copy_elem),
366 if (new_batch_copy_elems) {
367 rte_free(vq->batch_copy_elems);
368 vq->batch_copy_elems = new_batch_copy_elems;
374 /* check if we need to reallocate dev */
375 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
376 MPOL_F_NODE | MPOL_F_ADDR);
378 RTE_LOG(ERR, VHOST_CONFIG,
379 "Unable to get dev numa information.\n");
382 if (oldnode != newnode) {
383 RTE_LOG(INFO, VHOST_CONFIG,
384 "reallocate dev from %d to %d node\n",
386 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
392 memcpy(dev, old_dev, sizeof(*dev));
397 dev->virtqueue[index] = vq;
398 vhost_devices[dev->vid] = dev;
401 vhost_user_iotlb_init(dev, index);
406 static struct virtio_net*
407 numa_realloc(struct virtio_net *dev, int index __rte_unused)
413 /* Converts QEMU virtual address to Vhost virtual address. */
415 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
417 struct rte_vhost_mem_region *r;
420 /* Find the region where the address lives. */
421 for (i = 0; i < dev->mem->nregions; i++) {
422 r = &dev->mem->regions[i];
424 if (qva >= r->guest_user_addr &&
425 qva < r->guest_user_addr + r->size) {
427 if (unlikely(*len > r->guest_user_addr + r->size - qva))
428 *len = r->guest_user_addr + r->size - qva;
430 return qva - r->guest_user_addr +
441 * Converts ring address to Vhost virtual address.
442 * If IOMMU is enabled, the ring address is a guest IO virtual address,
443 * else it is a QEMU virtual address.
446 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
447 uint64_t ra, uint64_t *size)
449 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
452 vva = vhost_user_iotlb_cache_find(vq, ra,
453 size, VHOST_ACCESS_RW);
455 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
460 return qva_to_vva(dev, ra, size);
463 static struct virtio_net *
464 translate_ring_addresses(struct virtio_net *dev, int vq_index)
466 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
467 struct vhost_vring_addr *addr = &vq->ring_addrs;
470 /* The addresses are converted from QEMU virtual to Vhost virtual. */
471 if (vq->desc && vq->avail && vq->used)
474 len = sizeof(struct vring_desc) * vq->size;
475 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
476 vq, addr->desc_user_addr, &len);
477 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
478 RTE_LOG(DEBUG, VHOST_CONFIG,
479 "(%d) failed to map desc ring.\n",
484 dev = numa_realloc(dev, vq_index);
485 vq = dev->virtqueue[vq_index];
486 addr = &vq->ring_addrs;
488 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
489 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
490 vq, addr->avail_user_addr, &len);
491 if (vq->avail == 0 ||
492 len != sizeof(struct vring_avail) +
493 sizeof(uint16_t) * vq->size) {
494 RTE_LOG(DEBUG, VHOST_CONFIG,
495 "(%d) failed to map avail ring.\n",
500 len = sizeof(struct vring_used) +
501 sizeof(struct vring_used_elem) * vq->size;
502 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
503 vq, addr->used_user_addr, &len);
504 if (vq->used == 0 || len != sizeof(struct vring_used) +
505 sizeof(struct vring_used_elem) * vq->size) {
506 RTE_LOG(DEBUG, VHOST_CONFIG,
507 "(%d) failed to map used ring.\n",
512 if (vq->last_used_idx != vq->used->idx) {
513 RTE_LOG(WARNING, VHOST_CONFIG,
514 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
515 "some packets maybe resent for Tx and dropped for Rx\n",
516 vq->last_used_idx, vq->used->idx);
517 vq->last_used_idx = vq->used->idx;
518 vq->last_avail_idx = vq->used->idx;
521 vq->log_guest_addr = addr->log_guest_addr;
523 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
525 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
526 dev->vid, vq->avail);
527 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
529 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
530 dev->vid, vq->log_guest_addr);
536 * The virtio device sends us the desc, used and avail ring addresses.
537 * This function then converts these to our address space.
540 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
542 struct vhost_virtqueue *vq;
543 struct vhost_vring_addr *addr = &msg->payload.addr;
544 struct virtio_net *dev = *pdev;
546 if (dev->mem == NULL)
549 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
550 vq = dev->virtqueue[msg->payload.addr.index];
553 * Rings addresses should not be interpreted as long as the ring is not
554 * started and enabled
556 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
558 vring_invalidate(dev, vq);
560 if (vq->enabled && (dev->features &
561 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
562 dev = translate_ring_addresses(dev, msg->payload.addr.index);
573 * The virtio device sends us the available ring last used index.
576 vhost_user_set_vring_base(struct virtio_net *dev,
579 dev->virtqueue[msg->payload.state.index]->last_used_idx =
580 msg->payload.state.num;
581 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
582 msg->payload.state.num;
588 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
589 uint64_t host_phys_addr, uint64_t size)
591 struct guest_page *page, *last_page;
593 if (dev->nr_guest_pages == dev->max_guest_pages) {
594 dev->max_guest_pages *= 2;
595 dev->guest_pages = realloc(dev->guest_pages,
596 dev->max_guest_pages * sizeof(*page));
597 if (!dev->guest_pages) {
598 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
603 if (dev->nr_guest_pages > 0) {
604 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
605 /* merge if the two pages are continuous */
606 if (host_phys_addr == last_page->host_phys_addr +
608 last_page->size += size;
613 page = &dev->guest_pages[dev->nr_guest_pages++];
614 page->guest_phys_addr = guest_phys_addr;
615 page->host_phys_addr = host_phys_addr;
622 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
625 uint64_t reg_size = reg->size;
626 uint64_t host_user_addr = reg->host_user_addr;
627 uint64_t guest_phys_addr = reg->guest_phys_addr;
628 uint64_t host_phys_addr;
631 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
632 size = page_size - (guest_phys_addr & (page_size - 1));
633 size = RTE_MIN(size, reg_size);
635 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
638 host_user_addr += size;
639 guest_phys_addr += size;
642 while (reg_size > 0) {
643 size = RTE_MIN(reg_size, page_size);
644 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
646 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
650 host_user_addr += size;
651 guest_phys_addr += size;
658 #ifdef RTE_LIBRTE_VHOST_DEBUG
659 /* TODO: enable it only in debug mode? */
661 dump_guest_pages(struct virtio_net *dev)
664 struct guest_page *page;
666 for (i = 0; i < dev->nr_guest_pages; i++) {
667 page = &dev->guest_pages[i];
669 RTE_LOG(INFO, VHOST_CONFIG,
670 "guest physical page region %u\n"
671 "\t guest_phys_addr: %" PRIx64 "\n"
672 "\t host_phys_addr : %" PRIx64 "\n"
673 "\t size : %" PRIx64 "\n",
675 page->guest_phys_addr,
676 page->host_phys_addr,
681 #define dump_guest_pages(dev)
685 vhost_memory_changed(struct VhostUserMemory *new,
686 struct rte_vhost_memory *old)
690 if (new->nregions != old->nregions)
693 for (i = 0; i < new->nregions; ++i) {
694 VhostUserMemoryRegion *new_r = &new->regions[i];
695 struct rte_vhost_mem_region *old_r = &old->regions[i];
697 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
699 if (new_r->memory_size != old_r->size)
701 if (new_r->userspace_addr != old_r->guest_user_addr)
709 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
711 struct virtio_net *dev = *pdev;
712 struct VhostUserMemory memory = pmsg->payload.memory;
713 struct rte_vhost_mem_region *reg;
716 uint64_t mmap_offset;
722 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
723 RTE_LOG(ERR, VHOST_CONFIG,
724 "too many memory regions (%u)\n", memory.nregions);
728 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
729 RTE_LOG(INFO, VHOST_CONFIG,
730 "(%d) memory regions not changed\n", dev->vid);
732 for (i = 0; i < memory.nregions; i++)
739 free_mem_region(dev);
744 dev->nr_guest_pages = 0;
745 if (!dev->guest_pages) {
746 dev->max_guest_pages = 8;
747 dev->guest_pages = malloc(dev->max_guest_pages *
748 sizeof(struct guest_page));
749 if (dev->guest_pages == NULL) {
750 RTE_LOG(ERR, VHOST_CONFIG,
751 "(%d) failed to allocate memory "
752 "for dev->guest_pages\n",
758 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
759 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
760 if (dev->mem == NULL) {
761 RTE_LOG(ERR, VHOST_CONFIG,
762 "(%d) failed to allocate memory for dev->mem\n",
766 dev->mem->nregions = memory.nregions;
768 for (i = 0; i < memory.nregions; i++) {
770 reg = &dev->mem->regions[i];
772 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
773 reg->guest_user_addr = memory.regions[i].userspace_addr;
774 reg->size = memory.regions[i].memory_size;
777 mmap_offset = memory.regions[i].mmap_offset;
779 /* Check for memory_size + mmap_offset overflow */
780 if (mmap_offset >= -reg->size) {
781 RTE_LOG(ERR, VHOST_CONFIG,
782 "mmap_offset (%#"PRIx64") and memory_size "
783 "(%#"PRIx64") overflow\n",
784 mmap_offset, reg->size);
788 mmap_size = reg->size + mmap_offset;
790 /* mmap() without flag of MAP_ANONYMOUS, should be called
791 * with length argument aligned with hugepagesz at older
792 * longterm version Linux, like 2.6.32 and 3.2.72, or
793 * mmap() will fail with EINVAL.
795 * to avoid failure, make sure in caller to keep length
798 alignment = get_blk_size(fd);
799 if (alignment == (uint64_t)-1) {
800 RTE_LOG(ERR, VHOST_CONFIG,
801 "couldn't get hugepage size through fstat\n");
804 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
806 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
807 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
808 MAP_SHARED | populate, fd, 0);
810 if (mmap_addr == MAP_FAILED) {
811 RTE_LOG(ERR, VHOST_CONFIG,
812 "mmap region %u failed.\n", i);
816 reg->mmap_addr = mmap_addr;
817 reg->mmap_size = mmap_size;
818 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
821 if (dev->dequeue_zero_copy)
822 if (add_guest_pages(dev, reg, alignment) < 0) {
823 RTE_LOG(ERR, VHOST_CONFIG,
824 "adding guest pages to region %u failed.\n",
829 RTE_LOG(INFO, VHOST_CONFIG,
830 "guest memory region %u, size: 0x%" PRIx64 "\n"
831 "\t guest physical addr: 0x%" PRIx64 "\n"
832 "\t guest virtual addr: 0x%" PRIx64 "\n"
833 "\t host virtual addr: 0x%" PRIx64 "\n"
834 "\t mmap addr : 0x%" PRIx64 "\n"
835 "\t mmap size : 0x%" PRIx64 "\n"
836 "\t mmap align: 0x%" PRIx64 "\n"
837 "\t mmap off : 0x%" PRIx64 "\n",
839 reg->guest_phys_addr,
840 reg->guest_user_addr,
842 (uint64_t)(uintptr_t)mmap_addr,
848 for (i = 0; i < dev->nr_vring; i++) {
849 struct vhost_virtqueue *vq = dev->virtqueue[i];
851 if (vq->desc || vq->avail || vq->used) {
853 * If the memory table got updated, the ring addresses
854 * need to be translated again as virtual addresses have
857 vring_invalidate(dev, vq);
859 dev = translate_ring_addresses(dev, i);
867 dump_guest_pages(dev);
872 free_mem_region(dev);
879 vq_is_ready(struct vhost_virtqueue *vq)
881 return vq && vq->desc && vq->avail && vq->used &&
882 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
883 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
887 virtio_is_ready(struct virtio_net *dev)
889 struct vhost_virtqueue *vq;
892 if (dev->nr_vring == 0)
895 for (i = 0; i < dev->nr_vring; i++) {
896 vq = dev->virtqueue[i];
898 if (!vq_is_ready(vq))
902 RTE_LOG(INFO, VHOST_CONFIG,
903 "virtio is now ready for processing.\n");
908 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
910 struct vhost_vring_file file;
911 struct vhost_virtqueue *vq;
913 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
914 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
915 file.fd = VIRTIO_INVALID_EVENTFD;
917 file.fd = pmsg->fds[0];
918 RTE_LOG(INFO, VHOST_CONFIG,
919 "vring call idx:%d file:%d\n", file.index, file.fd);
921 vq = dev->virtqueue[file.index];
925 vq->callfd = file.fd;
929 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
931 struct vhost_vring_file file;
932 struct vhost_virtqueue *vq;
933 struct virtio_net *dev = *pdev;
935 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
936 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
937 file.fd = VIRTIO_INVALID_EVENTFD;
939 file.fd = pmsg->fds[0];
940 RTE_LOG(INFO, VHOST_CONFIG,
941 "vring kick idx:%d file:%d\n", file.index, file.fd);
943 /* Interpret ring addresses only when ring is started. */
944 dev = translate_ring_addresses(dev, file.index);
950 vq = dev->virtqueue[file.index];
953 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
954 * the ring starts already enabled. Otherwise, it is enabled via
955 * the SET_VRING_ENABLE message.
957 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
962 vq->kickfd = file.fd;
966 free_zmbufs(struct vhost_virtqueue *vq)
968 struct zcopy_mbuf *zmbuf, *next;
970 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
971 zmbuf != NULL; zmbuf = next) {
972 next = TAILQ_NEXT(zmbuf, next);
974 rte_pktmbuf_free(zmbuf->mbuf);
975 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
978 rte_free(vq->zmbufs);
982 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
985 vhost_user_get_vring_base(struct virtio_net *dev,
988 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
990 /* We have to stop the queue (virtio) if it is running. */
991 vhost_destroy_device_notify(dev);
993 dev->flags &= ~VIRTIO_DEV_READY;
994 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
996 /* Here we are safe to get the last avail index */
997 msg->payload.state.num = vq->last_avail_idx;
999 RTE_LOG(INFO, VHOST_CONFIG,
1000 "vring base idx:%d file:%d\n", msg->payload.state.index,
1001 msg->payload.state.num);
1003 * Based on current qemu vhost-user implementation, this message is
1004 * sent and only sent in vhost_vring_stop.
1005 * TODO: cleanup the vring, it isn't usable since here.
1007 if (vq->kickfd >= 0)
1010 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1012 if (vq->callfd >= 0)
1015 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1017 if (dev->dequeue_zero_copy)
1019 rte_free(vq->shadow_used_ring);
1020 vq->shadow_used_ring = NULL;
1022 rte_free(vq->batch_copy_elems);
1023 vq->batch_copy_elems = NULL;
1029 * when virtio queues are ready to work, qemu will send us to
1030 * enable the virtio queue pair.
1033 vhost_user_set_vring_enable(struct virtio_net *dev,
1036 int enable = (int)msg->payload.state.num;
1037 int index = (int)msg->payload.state.index;
1038 struct rte_vdpa_device *vdpa_dev;
1041 RTE_LOG(INFO, VHOST_CONFIG,
1042 "set queue enable: %d to qp idx: %d\n",
1045 did = dev->vdpa_dev_id;
1046 vdpa_dev = rte_vdpa_get_device(did);
1047 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1048 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1050 if (dev->notify_ops->vring_state_changed)
1051 dev->notify_ops->vring_state_changed(dev->vid,
1054 dev->virtqueue[index]->enabled = enable;
1060 vhost_user_get_protocol_features(struct virtio_net *dev,
1061 struct VhostUserMsg *msg)
1063 uint64_t features, protocol_features;
1065 rte_vhost_driver_get_features(dev->ifname, &features);
1066 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1069 * REPLY_ACK protocol feature is only mandatory for now
1070 * for IOMMU feature. If IOMMU is explicitly disabled by the
1071 * application, disable also REPLY_ACK feature for older buggy
1072 * Qemu versions (from v2.7.0 to v2.9.0).
1074 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1075 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1077 msg->payload.u64 = protocol_features;
1078 msg->size = sizeof(msg->payload.u64);
1082 vhost_user_set_protocol_features(struct virtio_net *dev,
1083 uint64_t protocol_features)
1085 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1088 dev->protocol_features = protocol_features;
1092 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1094 int fd = msg->fds[0];
1099 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1103 if (msg->size != sizeof(VhostUserLog)) {
1104 RTE_LOG(ERR, VHOST_CONFIG,
1105 "invalid log base msg size: %"PRId32" != %d\n",
1106 msg->size, (int)sizeof(VhostUserLog));
1110 size = msg->payload.log.mmap_size;
1111 off = msg->payload.log.mmap_offset;
1113 /* Don't allow mmap_offset to point outside the mmap region */
1115 RTE_LOG(ERR, VHOST_CONFIG,
1116 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1121 RTE_LOG(INFO, VHOST_CONFIG,
1122 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1126 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1127 * fail when offset is not page size aligned.
1129 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1131 if (addr == MAP_FAILED) {
1132 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1137 * Free previously mapped log memory on occasionally
1138 * multiple VHOST_USER_SET_LOG_BASE.
1140 if (dev->log_addr) {
1141 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1143 dev->log_addr = (uint64_t)(uintptr_t)addr;
1144 dev->log_base = dev->log_addr + off;
1145 dev->log_size = size;
1151 * An rarp packet is constructed and broadcasted to notify switches about
1152 * the new location of the migrated VM, so that packets from outside will
1153 * not be lost after migration.
1155 * However, we don't actually "send" a rarp packet here, instead, we set
1156 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1159 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1161 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1162 struct rte_vdpa_device *vdpa_dev;
1165 RTE_LOG(DEBUG, VHOST_CONFIG,
1166 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1167 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1168 memcpy(dev->mac.addr_bytes, mac, 6);
1171 * Set the flag to inject a RARP broadcast packet at
1172 * rte_vhost_dequeue_burst().
1174 * rte_smp_wmb() is for making sure the mac is copied
1175 * before the flag is set.
1178 rte_atomic16_set(&dev->broadcast_rarp, 1);
1179 did = dev->vdpa_dev_id;
1180 vdpa_dev = rte_vdpa_get_device(did);
1181 if (vdpa_dev && vdpa_dev->ops->migration_done)
1182 vdpa_dev->ops->migration_done(dev->vid);
1188 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1190 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1191 msg->payload.u64 > VIRTIO_MAX_MTU) {
1192 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1198 dev->mtu = msg->payload.u64;
1204 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1206 int fd = msg->fds[0];
1209 RTE_LOG(ERR, VHOST_CONFIG,
1210 "Invalid file descriptor for slave channel (%d)\n",
1215 dev->slave_req_fd = fd;
1221 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1223 struct vhost_vring_addr *ra;
1224 uint64_t start, end;
1227 end = start + imsg->size;
1229 ra = &vq->ring_addrs;
1230 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1232 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1234 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1241 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1242 struct vhost_iotlb_msg *imsg)
1244 uint64_t istart, iend, vstart, vend;
1246 istart = imsg->iova;
1247 iend = istart + imsg->size - 1;
1249 vstart = (uintptr_t)vq->desc;
1250 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1251 if (vstart <= iend && istart <= vend)
1254 vstart = (uintptr_t)vq->avail;
1255 vend = vstart + sizeof(struct vring_avail);
1256 vend += sizeof(uint16_t) * vq->size - 1;
1257 if (vstart <= iend && istart <= vend)
1260 vstart = (uintptr_t)vq->used;
1261 vend = vstart + sizeof(struct vring_used);
1262 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1263 if (vstart <= iend && istart <= vend)
1270 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1272 struct virtio_net *dev = *pdev;
1273 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1277 switch (imsg->type) {
1278 case VHOST_IOTLB_UPDATE:
1280 vva = qva_to_vva(dev, imsg->uaddr, &len);
1284 for (i = 0; i < dev->nr_vring; i++) {
1285 struct vhost_virtqueue *vq = dev->virtqueue[i];
1287 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1290 if (is_vring_iotlb_update(vq, imsg))
1291 *pdev = dev = translate_ring_addresses(dev, i);
1294 case VHOST_IOTLB_INVALIDATE:
1295 for (i = 0; i < dev->nr_vring; i++) {
1296 struct vhost_virtqueue *vq = dev->virtqueue[i];
1298 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1301 if (is_vring_iotlb_invalidate(vq, imsg))
1302 vring_invalidate(dev, vq);
1306 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1314 /* return bytes# of read on success or negative val on failure. */
1316 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1320 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1321 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1325 if (msg && msg->size) {
1326 if (msg->size > sizeof(msg->payload)) {
1327 RTE_LOG(ERR, VHOST_CONFIG,
1328 "invalid msg size: %d\n", msg->size);
1331 ret = read(sockfd, &msg->payload, msg->size);
1334 if (ret != (int)msg->size) {
1335 RTE_LOG(ERR, VHOST_CONFIG,
1336 "read control message failed\n");
1345 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1350 return send_fd_message(sockfd, (char *)msg,
1351 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1355 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1360 msg->flags &= ~VHOST_USER_VERSION_MASK;
1361 msg->flags &= ~VHOST_USER_NEED_REPLY;
1362 msg->flags |= VHOST_USER_VERSION;
1363 msg->flags |= VHOST_USER_REPLY_MASK;
1365 return send_vhost_message(sockfd, msg, NULL, 0);
1369 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1370 int *fds, int fd_num)
1374 if (msg->flags & VHOST_USER_NEED_REPLY)
1375 rte_spinlock_lock(&dev->slave_req_lock);
1377 ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1378 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1379 rte_spinlock_unlock(&dev->slave_req_lock);
1385 * Allocate a queue pair if it hasn't been allocated yet
1388 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1392 switch (msg->request.master) {
1393 case VHOST_USER_SET_VRING_KICK:
1394 case VHOST_USER_SET_VRING_CALL:
1395 case VHOST_USER_SET_VRING_ERR:
1396 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1398 case VHOST_USER_SET_VRING_NUM:
1399 case VHOST_USER_SET_VRING_BASE:
1400 case VHOST_USER_SET_VRING_ENABLE:
1401 vring_idx = msg->payload.state.index;
1403 case VHOST_USER_SET_VRING_ADDR:
1404 vring_idx = msg->payload.addr.index;
1410 if (vring_idx >= VHOST_MAX_VRING) {
1411 RTE_LOG(ERR, VHOST_CONFIG,
1412 "invalid vring index: %u\n", vring_idx);
1416 if (dev->virtqueue[vring_idx])
1419 return alloc_vring_queue(dev, vring_idx);
1423 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1426 unsigned int vq_num = 0;
1428 while (vq_num < dev->nr_vring) {
1429 struct vhost_virtqueue *vq = dev->virtqueue[i];
1432 rte_spinlock_lock(&vq->access_lock);
1440 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1443 unsigned int vq_num = 0;
1445 while (vq_num < dev->nr_vring) {
1446 struct vhost_virtqueue *vq = dev->virtqueue[i];
1449 rte_spinlock_unlock(&vq->access_lock);
1457 vhost_user_msg_handler(int vid, int fd)
1459 struct virtio_net *dev;
1460 struct VhostUserMsg msg;
1461 struct rte_vdpa_device *vdpa_dev;
1464 int unlock_required = 0;
1465 uint32_t skip_master = 0;
1467 dev = get_device(vid);
1471 if (!dev->notify_ops) {
1472 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1473 if (!dev->notify_ops) {
1474 RTE_LOG(ERR, VHOST_CONFIG,
1475 "failed to get callback ops for driver %s\n",
1481 ret = read_vhost_message(fd, &msg);
1482 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1484 RTE_LOG(ERR, VHOST_CONFIG,
1485 "vhost read message failed\n");
1487 RTE_LOG(INFO, VHOST_CONFIG,
1488 "vhost peer closed\n");
1490 RTE_LOG(ERR, VHOST_CONFIG,
1491 "vhost read incorrect message\n");
1497 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1498 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1499 vhost_message_str[msg.request.master]);
1501 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1502 vhost_message_str[msg.request.master]);
1504 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1506 RTE_LOG(ERR, VHOST_CONFIG,
1507 "failed to alloc queue\n");
1512 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1513 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1514 * and device is destroyed. destroy_device waits for queues to be
1515 * inactive, so it is safe. Otherwise taking the access_lock
1516 * would cause a dead lock.
1518 switch (msg.request.master) {
1519 case VHOST_USER_SET_FEATURES:
1520 case VHOST_USER_SET_PROTOCOL_FEATURES:
1521 case VHOST_USER_SET_OWNER:
1522 case VHOST_USER_SET_MEM_TABLE:
1523 case VHOST_USER_SET_LOG_BASE:
1524 case VHOST_USER_SET_LOG_FD:
1525 case VHOST_USER_SET_VRING_NUM:
1526 case VHOST_USER_SET_VRING_ADDR:
1527 case VHOST_USER_SET_VRING_BASE:
1528 case VHOST_USER_SET_VRING_KICK:
1529 case VHOST_USER_SET_VRING_CALL:
1530 case VHOST_USER_SET_VRING_ERR:
1531 case VHOST_USER_SET_VRING_ENABLE:
1532 case VHOST_USER_SEND_RARP:
1533 case VHOST_USER_NET_SET_MTU:
1534 case VHOST_USER_SET_SLAVE_REQ_FD:
1535 vhost_user_lock_all_queue_pairs(dev);
1536 unlock_required = 1;
1543 if (dev->extern_ops.pre_msg_handle) {
1544 uint32_t need_reply;
1546 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1547 (void *)&msg, &need_reply, &skip_master);
1552 send_vhost_reply(fd, &msg);
1555 goto skip_to_post_handle;
1558 switch (msg.request.master) {
1559 case VHOST_USER_GET_FEATURES:
1560 msg.payload.u64 = vhost_user_get_features(dev);
1561 msg.size = sizeof(msg.payload.u64);
1562 send_vhost_reply(fd, &msg);
1564 case VHOST_USER_SET_FEATURES:
1565 ret = vhost_user_set_features(dev, msg.payload.u64);
1570 case VHOST_USER_GET_PROTOCOL_FEATURES:
1571 vhost_user_get_protocol_features(dev, &msg);
1572 send_vhost_reply(fd, &msg);
1574 case VHOST_USER_SET_PROTOCOL_FEATURES:
1575 vhost_user_set_protocol_features(dev, msg.payload.u64);
1578 case VHOST_USER_SET_OWNER:
1579 vhost_user_set_owner();
1581 case VHOST_USER_RESET_OWNER:
1582 vhost_user_reset_owner(dev);
1585 case VHOST_USER_SET_MEM_TABLE:
1586 ret = vhost_user_set_mem_table(&dev, &msg);
1589 case VHOST_USER_SET_LOG_BASE:
1590 vhost_user_set_log_base(dev, &msg);
1592 /* it needs a reply */
1593 msg.size = sizeof(msg.payload.u64);
1594 send_vhost_reply(fd, &msg);
1596 case VHOST_USER_SET_LOG_FD:
1598 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1601 case VHOST_USER_SET_VRING_NUM:
1602 vhost_user_set_vring_num(dev, &msg);
1604 case VHOST_USER_SET_VRING_ADDR:
1605 vhost_user_set_vring_addr(&dev, &msg);
1607 case VHOST_USER_SET_VRING_BASE:
1608 vhost_user_set_vring_base(dev, &msg);
1611 case VHOST_USER_GET_VRING_BASE:
1612 vhost_user_get_vring_base(dev, &msg);
1613 msg.size = sizeof(msg.payload.state);
1614 send_vhost_reply(fd, &msg);
1617 case VHOST_USER_SET_VRING_KICK:
1618 vhost_user_set_vring_kick(&dev, &msg);
1620 case VHOST_USER_SET_VRING_CALL:
1621 vhost_user_set_vring_call(dev, &msg);
1624 case VHOST_USER_SET_VRING_ERR:
1625 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1627 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1630 case VHOST_USER_GET_QUEUE_NUM:
1631 msg.payload.u64 = (uint64_t)vhost_user_get_queue_num(dev);
1632 msg.size = sizeof(msg.payload.u64);
1633 send_vhost_reply(fd, &msg);
1636 case VHOST_USER_SET_VRING_ENABLE:
1637 vhost_user_set_vring_enable(dev, &msg);
1639 case VHOST_USER_SEND_RARP:
1640 vhost_user_send_rarp(dev, &msg);
1643 case VHOST_USER_NET_SET_MTU:
1644 ret = vhost_user_net_set_mtu(dev, &msg);
1647 case VHOST_USER_SET_SLAVE_REQ_FD:
1648 ret = vhost_user_set_req_fd(dev, &msg);
1651 case VHOST_USER_IOTLB_MSG:
1652 ret = vhost_user_iotlb_msg(&dev, &msg);
1660 skip_to_post_handle:
1661 if (dev->extern_ops.post_msg_handle) {
1662 uint32_t need_reply;
1664 ret = (*dev->extern_ops.post_msg_handle)(
1665 dev->vid, (void *)&msg, &need_reply);
1670 send_vhost_reply(fd, &msg);
1674 if (unlock_required)
1675 vhost_user_unlock_all_queue_pairs(dev);
1677 if (msg.flags & VHOST_USER_NEED_REPLY) {
1678 msg.payload.u64 = !!ret;
1679 msg.size = sizeof(msg.payload.u64);
1680 send_vhost_reply(fd, &msg);
1683 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1684 dev->flags |= VIRTIO_DEV_READY;
1686 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1687 if (dev->dequeue_zero_copy) {
1688 RTE_LOG(INFO, VHOST_CONFIG,
1689 "dequeue zero copy is enabled\n");
1692 if (dev->notify_ops->new_device(dev->vid) == 0)
1693 dev->flags |= VIRTIO_DEV_RUNNING;
1697 did = dev->vdpa_dev_id;
1698 vdpa_dev = rte_vdpa_get_device(did);
1699 if (vdpa_dev && virtio_is_ready(dev) &&
1700 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1701 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1702 if (vdpa_dev->ops->dev_conf)
1703 vdpa_dev->ops->dev_conf(dev->vid);
1704 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1705 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1706 RTE_LOG(INFO, VHOST_CONFIG,
1707 "(%d) software relay is used for vDPA, performance may be low.\n",
1715 static int process_slave_message_reply(struct virtio_net *dev,
1716 const VhostUserMsg *msg)
1718 VhostUserMsg msg_reply;
1721 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1724 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1729 if (msg_reply.request.slave != msg->request.slave) {
1730 RTE_LOG(ERR, VHOST_CONFIG,
1731 "Received unexpected msg type (%u), expected %u\n",
1732 msg_reply.request.slave, msg->request.slave);
1737 ret = msg_reply.payload.u64 ? -1 : 0;
1740 rte_spinlock_unlock(&dev->slave_req_lock);
1745 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1748 struct VhostUserMsg msg = {
1749 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1750 .flags = VHOST_USER_VERSION,
1751 .size = sizeof(msg.payload.iotlb),
1755 .type = VHOST_IOTLB_MISS,
1759 ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1761 RTE_LOG(ERR, VHOST_CONFIG,
1762 "Failed to send IOTLB miss message (%d)\n",
1770 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1778 struct VhostUserMsg msg = {
1779 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1780 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1781 .size = sizeof(msg.payload.area),
1783 .u64 = index & VHOST_USER_VRING_IDX_MASK,
1790 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1796 ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1798 RTE_LOG(ERR, VHOST_CONFIG,
1799 "Failed to set host notifier (%d)\n", ret);
1803 return process_slave_message_reply(dev, &msg);
1806 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1808 struct virtio_net *dev;
1809 struct rte_vdpa_device *vdpa_dev;
1810 int vfio_device_fd, did, ret = 0;
1811 uint64_t offset, size;
1814 dev = get_device(vid);
1818 did = dev->vdpa_dev_id;
1822 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1823 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1824 !(dev->protocol_features &
1825 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1826 !(dev->protocol_features &
1827 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1828 !(dev->protocol_features &
1829 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1832 vdpa_dev = rte_vdpa_get_device(did);
1836 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1837 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1839 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1840 if (vfio_device_fd < 0)
1844 for (i = 0; i < dev->nr_vring; i++) {
1845 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1851 if (vhost_user_slave_set_vring_host_notifier(dev, i,
1852 vfio_device_fd, offset, size) < 0) {
1859 for (i = 0; i < dev->nr_vring; i++) {
1860 vhost_user_slave_set_vring_host_notifier(dev, i, -1,