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 if (vq_is_packed(dev)) {
471 len = sizeof(struct vring_packed_desc) * vq->size;
472 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
473 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
474 vq->log_guest_addr = 0;
475 if (vq->desc_packed == NULL ||
476 len != sizeof(struct vring_packed_desc) *
478 RTE_LOG(DEBUG, VHOST_CONFIG,
479 "(%d) failed to map desc_packed ring.\n",
484 dev = numa_realloc(dev, vq_index);
485 vq = dev->virtqueue[vq_index];
486 addr = &vq->ring_addrs;
491 /* The addresses are converted from QEMU virtual to Vhost virtual. */
492 if (vq->desc && vq->avail && vq->used)
495 len = sizeof(struct vring_desc) * vq->size;
496 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
497 vq, addr->desc_user_addr, &len);
498 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
499 RTE_LOG(DEBUG, VHOST_CONFIG,
500 "(%d) failed to map desc ring.\n",
505 dev = numa_realloc(dev, vq_index);
506 vq = dev->virtqueue[vq_index];
507 addr = &vq->ring_addrs;
509 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
510 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
511 vq, addr->avail_user_addr, &len);
512 if (vq->avail == 0 ||
513 len != sizeof(struct vring_avail) +
514 sizeof(uint16_t) * vq->size) {
515 RTE_LOG(DEBUG, VHOST_CONFIG,
516 "(%d) failed to map avail ring.\n",
521 len = sizeof(struct vring_used) +
522 sizeof(struct vring_used_elem) * vq->size;
523 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
524 vq, addr->used_user_addr, &len);
525 if (vq->used == 0 || len != sizeof(struct vring_used) +
526 sizeof(struct vring_used_elem) * vq->size) {
527 RTE_LOG(DEBUG, VHOST_CONFIG,
528 "(%d) failed to map used ring.\n",
533 if (vq->last_used_idx != vq->used->idx) {
534 RTE_LOG(WARNING, VHOST_CONFIG,
535 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
536 "some packets maybe resent for Tx and dropped for Rx\n",
537 vq->last_used_idx, vq->used->idx);
538 vq->last_used_idx = vq->used->idx;
539 vq->last_avail_idx = vq->used->idx;
542 vq->log_guest_addr = addr->log_guest_addr;
544 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
546 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
547 dev->vid, vq->avail);
548 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
550 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
551 dev->vid, vq->log_guest_addr);
557 * The virtio device sends us the desc, used and avail ring addresses.
558 * This function then converts these to our address space.
561 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
563 struct vhost_virtqueue *vq;
564 struct vhost_vring_addr *addr = &msg->payload.addr;
565 struct virtio_net *dev = *pdev;
567 if (dev->mem == NULL)
570 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
571 vq = dev->virtqueue[msg->payload.addr.index];
574 * Rings addresses should not be interpreted as long as the ring is not
575 * started and enabled
577 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
579 vring_invalidate(dev, vq);
581 if (vq->enabled && (dev->features &
582 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
583 dev = translate_ring_addresses(dev, msg->payload.addr.index);
594 * The virtio device sends us the available ring last used index.
597 vhost_user_set_vring_base(struct virtio_net *dev,
600 dev->virtqueue[msg->payload.state.index]->last_used_idx =
601 msg->payload.state.num;
602 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
603 msg->payload.state.num;
609 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
610 uint64_t host_phys_addr, uint64_t size)
612 struct guest_page *page, *last_page;
614 if (dev->nr_guest_pages == dev->max_guest_pages) {
615 dev->max_guest_pages *= 2;
616 dev->guest_pages = realloc(dev->guest_pages,
617 dev->max_guest_pages * sizeof(*page));
618 if (!dev->guest_pages) {
619 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
624 if (dev->nr_guest_pages > 0) {
625 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
626 /* merge if the two pages are continuous */
627 if (host_phys_addr == last_page->host_phys_addr +
629 last_page->size += size;
634 page = &dev->guest_pages[dev->nr_guest_pages++];
635 page->guest_phys_addr = guest_phys_addr;
636 page->host_phys_addr = host_phys_addr;
643 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
646 uint64_t reg_size = reg->size;
647 uint64_t host_user_addr = reg->host_user_addr;
648 uint64_t guest_phys_addr = reg->guest_phys_addr;
649 uint64_t host_phys_addr;
652 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
653 size = page_size - (guest_phys_addr & (page_size - 1));
654 size = RTE_MIN(size, reg_size);
656 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
659 host_user_addr += size;
660 guest_phys_addr += size;
663 while (reg_size > 0) {
664 size = RTE_MIN(reg_size, page_size);
665 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
667 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
671 host_user_addr += size;
672 guest_phys_addr += size;
679 #ifdef RTE_LIBRTE_VHOST_DEBUG
680 /* TODO: enable it only in debug mode? */
682 dump_guest_pages(struct virtio_net *dev)
685 struct guest_page *page;
687 for (i = 0; i < dev->nr_guest_pages; i++) {
688 page = &dev->guest_pages[i];
690 RTE_LOG(INFO, VHOST_CONFIG,
691 "guest physical page region %u\n"
692 "\t guest_phys_addr: %" PRIx64 "\n"
693 "\t host_phys_addr : %" PRIx64 "\n"
694 "\t size : %" PRIx64 "\n",
696 page->guest_phys_addr,
697 page->host_phys_addr,
702 #define dump_guest_pages(dev)
706 vhost_memory_changed(struct VhostUserMemory *new,
707 struct rte_vhost_memory *old)
711 if (new->nregions != old->nregions)
714 for (i = 0; i < new->nregions; ++i) {
715 VhostUserMemoryRegion *new_r = &new->regions[i];
716 struct rte_vhost_mem_region *old_r = &old->regions[i];
718 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
720 if (new_r->memory_size != old_r->size)
722 if (new_r->userspace_addr != old_r->guest_user_addr)
730 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
732 struct virtio_net *dev = *pdev;
733 struct VhostUserMemory memory = pmsg->payload.memory;
734 struct rte_vhost_mem_region *reg;
737 uint64_t mmap_offset;
743 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
744 RTE_LOG(ERR, VHOST_CONFIG,
745 "too many memory regions (%u)\n", memory.nregions);
749 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
750 RTE_LOG(INFO, VHOST_CONFIG,
751 "(%d) memory regions not changed\n", dev->vid);
753 for (i = 0; i < memory.nregions; i++)
760 free_mem_region(dev);
765 dev->nr_guest_pages = 0;
766 if (!dev->guest_pages) {
767 dev->max_guest_pages = 8;
768 dev->guest_pages = malloc(dev->max_guest_pages *
769 sizeof(struct guest_page));
770 if (dev->guest_pages == NULL) {
771 RTE_LOG(ERR, VHOST_CONFIG,
772 "(%d) failed to allocate memory "
773 "for dev->guest_pages\n",
779 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
780 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
781 if (dev->mem == NULL) {
782 RTE_LOG(ERR, VHOST_CONFIG,
783 "(%d) failed to allocate memory for dev->mem\n",
787 dev->mem->nregions = memory.nregions;
789 for (i = 0; i < memory.nregions; i++) {
791 reg = &dev->mem->regions[i];
793 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
794 reg->guest_user_addr = memory.regions[i].userspace_addr;
795 reg->size = memory.regions[i].memory_size;
798 mmap_offset = memory.regions[i].mmap_offset;
800 /* Check for memory_size + mmap_offset overflow */
801 if (mmap_offset >= -reg->size) {
802 RTE_LOG(ERR, VHOST_CONFIG,
803 "mmap_offset (%#"PRIx64") and memory_size "
804 "(%#"PRIx64") overflow\n",
805 mmap_offset, reg->size);
809 mmap_size = reg->size + mmap_offset;
811 /* mmap() without flag of MAP_ANONYMOUS, should be called
812 * with length argument aligned with hugepagesz at older
813 * longterm version Linux, like 2.6.32 and 3.2.72, or
814 * mmap() will fail with EINVAL.
816 * to avoid failure, make sure in caller to keep length
819 alignment = get_blk_size(fd);
820 if (alignment == (uint64_t)-1) {
821 RTE_LOG(ERR, VHOST_CONFIG,
822 "couldn't get hugepage size through fstat\n");
825 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
827 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
828 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
829 MAP_SHARED | populate, fd, 0);
831 if (mmap_addr == MAP_FAILED) {
832 RTE_LOG(ERR, VHOST_CONFIG,
833 "mmap region %u failed.\n", i);
837 reg->mmap_addr = mmap_addr;
838 reg->mmap_size = mmap_size;
839 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
842 if (dev->dequeue_zero_copy)
843 if (add_guest_pages(dev, reg, alignment) < 0) {
844 RTE_LOG(ERR, VHOST_CONFIG,
845 "adding guest pages to region %u failed.\n",
850 RTE_LOG(INFO, VHOST_CONFIG,
851 "guest memory region %u, size: 0x%" PRIx64 "\n"
852 "\t guest physical addr: 0x%" PRIx64 "\n"
853 "\t guest virtual addr: 0x%" PRIx64 "\n"
854 "\t host virtual addr: 0x%" PRIx64 "\n"
855 "\t mmap addr : 0x%" PRIx64 "\n"
856 "\t mmap size : 0x%" PRIx64 "\n"
857 "\t mmap align: 0x%" PRIx64 "\n"
858 "\t mmap off : 0x%" PRIx64 "\n",
860 reg->guest_phys_addr,
861 reg->guest_user_addr,
863 (uint64_t)(uintptr_t)mmap_addr,
869 for (i = 0; i < dev->nr_vring; i++) {
870 struct vhost_virtqueue *vq = dev->virtqueue[i];
872 if (vq->desc || vq->avail || vq->used) {
874 * If the memory table got updated, the ring addresses
875 * need to be translated again as virtual addresses have
878 vring_invalidate(dev, vq);
880 dev = translate_ring_addresses(dev, i);
888 dump_guest_pages(dev);
893 free_mem_region(dev);
900 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
907 if (vq_is_packed(dev))
908 rings_ok = !!vq->desc_packed;
910 rings_ok = vq->desc && vq->avail && vq->used;
913 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
914 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
918 virtio_is_ready(struct virtio_net *dev)
920 struct vhost_virtqueue *vq;
923 if (dev->nr_vring == 0)
926 for (i = 0; i < dev->nr_vring; i++) {
927 vq = dev->virtqueue[i];
929 if (!vq_is_ready(dev, vq))
933 RTE_LOG(INFO, VHOST_CONFIG,
934 "virtio is now ready for processing.\n");
939 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
941 struct vhost_vring_file file;
942 struct vhost_virtqueue *vq;
944 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
945 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
946 file.fd = VIRTIO_INVALID_EVENTFD;
948 file.fd = pmsg->fds[0];
949 RTE_LOG(INFO, VHOST_CONFIG,
950 "vring call idx:%d file:%d\n", file.index, file.fd);
952 vq = dev->virtqueue[file.index];
956 vq->callfd = file.fd;
960 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
962 struct vhost_vring_file file;
963 struct vhost_virtqueue *vq;
964 struct virtio_net *dev = *pdev;
966 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
967 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
968 file.fd = VIRTIO_INVALID_EVENTFD;
970 file.fd = pmsg->fds[0];
971 RTE_LOG(INFO, VHOST_CONFIG,
972 "vring kick idx:%d file:%d\n", file.index, file.fd);
974 /* Interpret ring addresses only when ring is started. */
975 dev = translate_ring_addresses(dev, file.index);
981 vq = dev->virtqueue[file.index];
984 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
985 * the ring starts already enabled. Otherwise, it is enabled via
986 * the SET_VRING_ENABLE message.
988 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
993 vq->kickfd = file.fd;
997 free_zmbufs(struct vhost_virtqueue *vq)
999 struct zcopy_mbuf *zmbuf, *next;
1001 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1002 zmbuf != NULL; zmbuf = next) {
1003 next = TAILQ_NEXT(zmbuf, next);
1005 rte_pktmbuf_free(zmbuf->mbuf);
1006 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1009 rte_free(vq->zmbufs);
1013 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1016 vhost_user_get_vring_base(struct virtio_net *dev,
1019 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1021 /* We have to stop the queue (virtio) if it is running. */
1022 vhost_destroy_device_notify(dev);
1024 dev->flags &= ~VIRTIO_DEV_READY;
1025 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1027 /* Here we are safe to get the last avail index */
1028 msg->payload.state.num = vq->last_avail_idx;
1030 RTE_LOG(INFO, VHOST_CONFIG,
1031 "vring base idx:%d file:%d\n", msg->payload.state.index,
1032 msg->payload.state.num);
1034 * Based on current qemu vhost-user implementation, this message is
1035 * sent and only sent in vhost_vring_stop.
1036 * TODO: cleanup the vring, it isn't usable since here.
1038 if (vq->kickfd >= 0)
1041 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1043 if (vq->callfd >= 0)
1046 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1048 if (dev->dequeue_zero_copy)
1050 rte_free(vq->shadow_used_ring);
1051 vq->shadow_used_ring = NULL;
1053 rte_free(vq->batch_copy_elems);
1054 vq->batch_copy_elems = NULL;
1060 * when virtio queues are ready to work, qemu will send us to
1061 * enable the virtio queue pair.
1064 vhost_user_set_vring_enable(struct virtio_net *dev,
1067 int enable = (int)msg->payload.state.num;
1068 int index = (int)msg->payload.state.index;
1069 struct rte_vdpa_device *vdpa_dev;
1072 RTE_LOG(INFO, VHOST_CONFIG,
1073 "set queue enable: %d to qp idx: %d\n",
1076 did = dev->vdpa_dev_id;
1077 vdpa_dev = rte_vdpa_get_device(did);
1078 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1079 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1081 if (dev->notify_ops->vring_state_changed)
1082 dev->notify_ops->vring_state_changed(dev->vid,
1085 dev->virtqueue[index]->enabled = enable;
1091 vhost_user_get_protocol_features(struct virtio_net *dev,
1092 struct VhostUserMsg *msg)
1094 uint64_t features, protocol_features;
1096 rte_vhost_driver_get_features(dev->ifname, &features);
1097 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1100 * REPLY_ACK protocol feature is only mandatory for now
1101 * for IOMMU feature. If IOMMU is explicitly disabled by the
1102 * application, disable also REPLY_ACK feature for older buggy
1103 * Qemu versions (from v2.7.0 to v2.9.0).
1105 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1106 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1108 msg->payload.u64 = protocol_features;
1109 msg->size = sizeof(msg->payload.u64);
1113 vhost_user_set_protocol_features(struct virtio_net *dev,
1114 uint64_t protocol_features)
1116 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1119 dev->protocol_features = protocol_features;
1123 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1125 int fd = msg->fds[0];
1130 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1134 if (msg->size != sizeof(VhostUserLog)) {
1135 RTE_LOG(ERR, VHOST_CONFIG,
1136 "invalid log base msg size: %"PRId32" != %d\n",
1137 msg->size, (int)sizeof(VhostUserLog));
1141 size = msg->payload.log.mmap_size;
1142 off = msg->payload.log.mmap_offset;
1144 /* Don't allow mmap_offset to point outside the mmap region */
1146 RTE_LOG(ERR, VHOST_CONFIG,
1147 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1152 RTE_LOG(INFO, VHOST_CONFIG,
1153 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1157 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1158 * fail when offset is not page size aligned.
1160 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1162 if (addr == MAP_FAILED) {
1163 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1168 * Free previously mapped log memory on occasionally
1169 * multiple VHOST_USER_SET_LOG_BASE.
1171 if (dev->log_addr) {
1172 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1174 dev->log_addr = (uint64_t)(uintptr_t)addr;
1175 dev->log_base = dev->log_addr + off;
1176 dev->log_size = size;
1182 * An rarp packet is constructed and broadcasted to notify switches about
1183 * the new location of the migrated VM, so that packets from outside will
1184 * not be lost after migration.
1186 * However, we don't actually "send" a rarp packet here, instead, we set
1187 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1190 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1192 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1193 struct rte_vdpa_device *vdpa_dev;
1196 RTE_LOG(DEBUG, VHOST_CONFIG,
1197 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1198 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1199 memcpy(dev->mac.addr_bytes, mac, 6);
1202 * Set the flag to inject a RARP broadcast packet at
1203 * rte_vhost_dequeue_burst().
1205 * rte_smp_wmb() is for making sure the mac is copied
1206 * before the flag is set.
1209 rte_atomic16_set(&dev->broadcast_rarp, 1);
1210 did = dev->vdpa_dev_id;
1211 vdpa_dev = rte_vdpa_get_device(did);
1212 if (vdpa_dev && vdpa_dev->ops->migration_done)
1213 vdpa_dev->ops->migration_done(dev->vid);
1219 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1221 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1222 msg->payload.u64 > VIRTIO_MAX_MTU) {
1223 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1229 dev->mtu = msg->payload.u64;
1235 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1237 int fd = msg->fds[0];
1240 RTE_LOG(ERR, VHOST_CONFIG,
1241 "Invalid file descriptor for slave channel (%d)\n",
1246 dev->slave_req_fd = fd;
1252 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1254 struct vhost_vring_addr *ra;
1255 uint64_t start, end;
1258 end = start + imsg->size;
1260 ra = &vq->ring_addrs;
1261 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1263 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1265 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1272 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1273 struct vhost_iotlb_msg *imsg)
1275 uint64_t istart, iend, vstart, vend;
1277 istart = imsg->iova;
1278 iend = istart + imsg->size - 1;
1280 vstart = (uintptr_t)vq->desc;
1281 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1282 if (vstart <= iend && istart <= vend)
1285 vstart = (uintptr_t)vq->avail;
1286 vend = vstart + sizeof(struct vring_avail);
1287 vend += sizeof(uint16_t) * vq->size - 1;
1288 if (vstart <= iend && istart <= vend)
1291 vstart = (uintptr_t)vq->used;
1292 vend = vstart + sizeof(struct vring_used);
1293 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1294 if (vstart <= iend && istart <= vend)
1301 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1303 struct virtio_net *dev = *pdev;
1304 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1308 switch (imsg->type) {
1309 case VHOST_IOTLB_UPDATE:
1311 vva = qva_to_vva(dev, imsg->uaddr, &len);
1315 for (i = 0; i < dev->nr_vring; i++) {
1316 struct vhost_virtqueue *vq = dev->virtqueue[i];
1318 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1321 if (is_vring_iotlb_update(vq, imsg))
1322 *pdev = dev = translate_ring_addresses(dev, i);
1325 case VHOST_IOTLB_INVALIDATE:
1326 for (i = 0; i < dev->nr_vring; i++) {
1327 struct vhost_virtqueue *vq = dev->virtqueue[i];
1329 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1332 if (is_vring_iotlb_invalidate(vq, imsg))
1333 vring_invalidate(dev, vq);
1337 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1345 /* return bytes# of read on success or negative val on failure. */
1347 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1351 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1352 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1356 if (msg && msg->size) {
1357 if (msg->size > sizeof(msg->payload)) {
1358 RTE_LOG(ERR, VHOST_CONFIG,
1359 "invalid msg size: %d\n", msg->size);
1362 ret = read(sockfd, &msg->payload, msg->size);
1365 if (ret != (int)msg->size) {
1366 RTE_LOG(ERR, VHOST_CONFIG,
1367 "read control message failed\n");
1376 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1381 return send_fd_message(sockfd, (char *)msg,
1382 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1386 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1391 msg->flags &= ~VHOST_USER_VERSION_MASK;
1392 msg->flags &= ~VHOST_USER_NEED_REPLY;
1393 msg->flags |= VHOST_USER_VERSION;
1394 msg->flags |= VHOST_USER_REPLY_MASK;
1396 return send_vhost_message(sockfd, msg, NULL, 0);
1400 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1401 int *fds, int fd_num)
1405 if (msg->flags & VHOST_USER_NEED_REPLY)
1406 rte_spinlock_lock(&dev->slave_req_lock);
1408 ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1409 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1410 rte_spinlock_unlock(&dev->slave_req_lock);
1416 * Allocate a queue pair if it hasn't been allocated yet
1419 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1423 switch (msg->request.master) {
1424 case VHOST_USER_SET_VRING_KICK:
1425 case VHOST_USER_SET_VRING_CALL:
1426 case VHOST_USER_SET_VRING_ERR:
1427 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1429 case VHOST_USER_SET_VRING_NUM:
1430 case VHOST_USER_SET_VRING_BASE:
1431 case VHOST_USER_SET_VRING_ENABLE:
1432 vring_idx = msg->payload.state.index;
1434 case VHOST_USER_SET_VRING_ADDR:
1435 vring_idx = msg->payload.addr.index;
1441 if (vring_idx >= VHOST_MAX_VRING) {
1442 RTE_LOG(ERR, VHOST_CONFIG,
1443 "invalid vring index: %u\n", vring_idx);
1447 if (dev->virtqueue[vring_idx])
1450 return alloc_vring_queue(dev, vring_idx);
1454 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1457 unsigned int vq_num = 0;
1459 while (vq_num < dev->nr_vring) {
1460 struct vhost_virtqueue *vq = dev->virtqueue[i];
1463 rte_spinlock_lock(&vq->access_lock);
1471 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1474 unsigned int vq_num = 0;
1476 while (vq_num < dev->nr_vring) {
1477 struct vhost_virtqueue *vq = dev->virtqueue[i];
1480 rte_spinlock_unlock(&vq->access_lock);
1488 vhost_user_msg_handler(int vid, int fd)
1490 struct virtio_net *dev;
1491 struct VhostUserMsg msg;
1492 struct rte_vdpa_device *vdpa_dev;
1495 int unlock_required = 0;
1496 uint32_t skip_master = 0;
1498 dev = get_device(vid);
1502 if (!dev->notify_ops) {
1503 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1504 if (!dev->notify_ops) {
1505 RTE_LOG(ERR, VHOST_CONFIG,
1506 "failed to get callback ops for driver %s\n",
1512 ret = read_vhost_message(fd, &msg);
1513 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1515 RTE_LOG(ERR, VHOST_CONFIG,
1516 "vhost read message failed\n");
1518 RTE_LOG(INFO, VHOST_CONFIG,
1519 "vhost peer closed\n");
1521 RTE_LOG(ERR, VHOST_CONFIG,
1522 "vhost read incorrect message\n");
1528 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1529 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1530 vhost_message_str[msg.request.master]);
1532 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1533 vhost_message_str[msg.request.master]);
1535 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1537 RTE_LOG(ERR, VHOST_CONFIG,
1538 "failed to alloc queue\n");
1543 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1544 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1545 * and device is destroyed. destroy_device waits for queues to be
1546 * inactive, so it is safe. Otherwise taking the access_lock
1547 * would cause a dead lock.
1549 switch (msg.request.master) {
1550 case VHOST_USER_SET_FEATURES:
1551 case VHOST_USER_SET_PROTOCOL_FEATURES:
1552 case VHOST_USER_SET_OWNER:
1553 case VHOST_USER_SET_MEM_TABLE:
1554 case VHOST_USER_SET_LOG_BASE:
1555 case VHOST_USER_SET_LOG_FD:
1556 case VHOST_USER_SET_VRING_NUM:
1557 case VHOST_USER_SET_VRING_ADDR:
1558 case VHOST_USER_SET_VRING_BASE:
1559 case VHOST_USER_SET_VRING_KICK:
1560 case VHOST_USER_SET_VRING_CALL:
1561 case VHOST_USER_SET_VRING_ERR:
1562 case VHOST_USER_SET_VRING_ENABLE:
1563 case VHOST_USER_SEND_RARP:
1564 case VHOST_USER_NET_SET_MTU:
1565 case VHOST_USER_SET_SLAVE_REQ_FD:
1566 vhost_user_lock_all_queue_pairs(dev);
1567 unlock_required = 1;
1574 if (dev->extern_ops.pre_msg_handle) {
1575 uint32_t need_reply;
1577 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1578 (void *)&msg, &need_reply, &skip_master);
1583 send_vhost_reply(fd, &msg);
1586 goto skip_to_post_handle;
1589 switch (msg.request.master) {
1590 case VHOST_USER_GET_FEATURES:
1591 msg.payload.u64 = vhost_user_get_features(dev);
1592 msg.size = sizeof(msg.payload.u64);
1593 send_vhost_reply(fd, &msg);
1595 case VHOST_USER_SET_FEATURES:
1596 ret = vhost_user_set_features(dev, msg.payload.u64);
1601 case VHOST_USER_GET_PROTOCOL_FEATURES:
1602 vhost_user_get_protocol_features(dev, &msg);
1603 send_vhost_reply(fd, &msg);
1605 case VHOST_USER_SET_PROTOCOL_FEATURES:
1606 vhost_user_set_protocol_features(dev, msg.payload.u64);
1609 case VHOST_USER_SET_OWNER:
1610 vhost_user_set_owner();
1612 case VHOST_USER_RESET_OWNER:
1613 vhost_user_reset_owner(dev);
1616 case VHOST_USER_SET_MEM_TABLE:
1617 ret = vhost_user_set_mem_table(&dev, &msg);
1620 case VHOST_USER_SET_LOG_BASE:
1621 vhost_user_set_log_base(dev, &msg);
1623 /* it needs a reply */
1624 msg.size = sizeof(msg.payload.u64);
1625 send_vhost_reply(fd, &msg);
1627 case VHOST_USER_SET_LOG_FD:
1629 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1632 case VHOST_USER_SET_VRING_NUM:
1633 vhost_user_set_vring_num(dev, &msg);
1635 case VHOST_USER_SET_VRING_ADDR:
1636 vhost_user_set_vring_addr(&dev, &msg);
1638 case VHOST_USER_SET_VRING_BASE:
1639 vhost_user_set_vring_base(dev, &msg);
1642 case VHOST_USER_GET_VRING_BASE:
1643 vhost_user_get_vring_base(dev, &msg);
1644 msg.size = sizeof(msg.payload.state);
1645 send_vhost_reply(fd, &msg);
1648 case VHOST_USER_SET_VRING_KICK:
1649 vhost_user_set_vring_kick(&dev, &msg);
1651 case VHOST_USER_SET_VRING_CALL:
1652 vhost_user_set_vring_call(dev, &msg);
1655 case VHOST_USER_SET_VRING_ERR:
1656 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1658 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1661 case VHOST_USER_GET_QUEUE_NUM:
1662 msg.payload.u64 = (uint64_t)vhost_user_get_queue_num(dev);
1663 msg.size = sizeof(msg.payload.u64);
1664 send_vhost_reply(fd, &msg);
1667 case VHOST_USER_SET_VRING_ENABLE:
1668 vhost_user_set_vring_enable(dev, &msg);
1670 case VHOST_USER_SEND_RARP:
1671 vhost_user_send_rarp(dev, &msg);
1674 case VHOST_USER_NET_SET_MTU:
1675 ret = vhost_user_net_set_mtu(dev, &msg);
1678 case VHOST_USER_SET_SLAVE_REQ_FD:
1679 ret = vhost_user_set_req_fd(dev, &msg);
1682 case VHOST_USER_IOTLB_MSG:
1683 ret = vhost_user_iotlb_msg(&dev, &msg);
1691 skip_to_post_handle:
1692 if (dev->extern_ops.post_msg_handle) {
1693 uint32_t need_reply;
1695 ret = (*dev->extern_ops.post_msg_handle)(
1696 dev->vid, (void *)&msg, &need_reply);
1701 send_vhost_reply(fd, &msg);
1705 if (unlock_required)
1706 vhost_user_unlock_all_queue_pairs(dev);
1708 if (msg.flags & VHOST_USER_NEED_REPLY) {
1709 msg.payload.u64 = !!ret;
1710 msg.size = sizeof(msg.payload.u64);
1711 send_vhost_reply(fd, &msg);
1714 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1715 dev->flags |= VIRTIO_DEV_READY;
1717 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1718 if (dev->dequeue_zero_copy) {
1719 RTE_LOG(INFO, VHOST_CONFIG,
1720 "dequeue zero copy is enabled\n");
1723 if (dev->notify_ops->new_device(dev->vid) == 0)
1724 dev->flags |= VIRTIO_DEV_RUNNING;
1728 did = dev->vdpa_dev_id;
1729 vdpa_dev = rte_vdpa_get_device(did);
1730 if (vdpa_dev && virtio_is_ready(dev) &&
1731 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1732 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1733 if (vdpa_dev->ops->dev_conf)
1734 vdpa_dev->ops->dev_conf(dev->vid);
1735 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1736 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1737 RTE_LOG(INFO, VHOST_CONFIG,
1738 "(%d) software relay is used for vDPA, performance may be low.\n",
1746 static int process_slave_message_reply(struct virtio_net *dev,
1747 const VhostUserMsg *msg)
1749 VhostUserMsg msg_reply;
1752 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1755 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1760 if (msg_reply.request.slave != msg->request.slave) {
1761 RTE_LOG(ERR, VHOST_CONFIG,
1762 "Received unexpected msg type (%u), expected %u\n",
1763 msg_reply.request.slave, msg->request.slave);
1768 ret = msg_reply.payload.u64 ? -1 : 0;
1771 rte_spinlock_unlock(&dev->slave_req_lock);
1776 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1779 struct VhostUserMsg msg = {
1780 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1781 .flags = VHOST_USER_VERSION,
1782 .size = sizeof(msg.payload.iotlb),
1786 .type = VHOST_IOTLB_MISS,
1790 ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1792 RTE_LOG(ERR, VHOST_CONFIG,
1793 "Failed to send IOTLB miss message (%d)\n",
1801 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1809 struct VhostUserMsg msg = {
1810 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1811 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1812 .size = sizeof(msg.payload.area),
1814 .u64 = index & VHOST_USER_VRING_IDX_MASK,
1821 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1827 ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1829 RTE_LOG(ERR, VHOST_CONFIG,
1830 "Failed to set host notifier (%d)\n", ret);
1834 return process_slave_message_reply(dev, &msg);
1837 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1839 struct virtio_net *dev;
1840 struct rte_vdpa_device *vdpa_dev;
1841 int vfio_device_fd, did, ret = 0;
1842 uint64_t offset, size;
1845 dev = get_device(vid);
1849 did = dev->vdpa_dev_id;
1853 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1854 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1855 !(dev->protocol_features &
1856 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1857 !(dev->protocol_features &
1858 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1859 !(dev->protocol_features &
1860 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1863 vdpa_dev = rte_vdpa_get_device(did);
1867 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1868 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1870 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1871 if (vfio_device_fd < 0)
1875 for (i = 0; i < dev->nr_vring; i++) {
1876 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1882 if (vhost_user_slave_set_vring_host_notifier(dev, i,
1883 vfio_device_fd, offset, size) < 0) {
1890 for (i = 0; i < dev->nr_vring; i++) {
1891 vhost_user_slave_set_vring_host_notifier(dev, i, -1,