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 if (vq_is_packed(dev)) {
286 vq->shadow_used_packed = rte_malloc(NULL,
288 sizeof(struct vring_used_elem_packed),
289 RTE_CACHE_LINE_SIZE);
290 if (!vq->shadow_used_packed) {
291 RTE_LOG(ERR, VHOST_CONFIG,
292 "failed to allocate memory for shadow used ring.\n");
297 vq->shadow_used_split = rte_malloc(NULL,
298 vq->size * sizeof(struct vring_used_elem),
299 RTE_CACHE_LINE_SIZE);
300 if (!vq->shadow_used_split) {
301 RTE_LOG(ERR, VHOST_CONFIG,
302 "failed to allocate memory for shadow used ring.\n");
307 vq->batch_copy_elems = rte_malloc(NULL,
308 vq->size * sizeof(struct batch_copy_elem),
309 RTE_CACHE_LINE_SIZE);
310 if (!vq->batch_copy_elems) {
311 RTE_LOG(ERR, VHOST_CONFIG,
312 "failed to allocate memory for batching copy.\n");
320 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
321 * same numa node as the memory of vring descriptor.
323 #ifdef RTE_LIBRTE_VHOST_NUMA
324 static struct virtio_net*
325 numa_realloc(struct virtio_net *dev, int index)
327 int oldnode, newnode;
328 struct virtio_net *old_dev;
329 struct vhost_virtqueue *old_vq, *vq;
330 struct zcopy_mbuf *new_zmbuf;
331 struct vring_used_elem *new_shadow_used_split;
332 struct vring_used_elem_packed *new_shadow_used_packed;
333 struct batch_copy_elem *new_batch_copy_elems;
337 vq = old_vq = dev->virtqueue[index];
339 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
340 MPOL_F_NODE | MPOL_F_ADDR);
342 /* check if we need to reallocate vq */
343 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
344 MPOL_F_NODE | MPOL_F_ADDR);
346 RTE_LOG(ERR, VHOST_CONFIG,
347 "Unable to get vq numa information.\n");
350 if (oldnode != newnode) {
351 RTE_LOG(INFO, VHOST_CONFIG,
352 "reallocate vq from %d to %d node\n", oldnode, newnode);
353 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
357 memcpy(vq, old_vq, sizeof(*vq));
358 TAILQ_INIT(&vq->zmbuf_list);
360 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
361 sizeof(struct zcopy_mbuf), 0, newnode);
363 rte_free(vq->zmbufs);
364 vq->zmbufs = new_zmbuf;
367 if (vq_is_packed(dev)) {
368 new_shadow_used_packed = rte_malloc_socket(NULL,
370 sizeof(struct vring_used_elem_packed),
373 if (new_shadow_used_packed) {
374 rte_free(vq->shadow_used_packed);
375 vq->shadow_used_packed = new_shadow_used_packed;
378 new_shadow_used_split = rte_malloc_socket(NULL,
380 sizeof(struct vring_used_elem),
383 if (new_shadow_used_split) {
384 rte_free(vq->shadow_used_split);
385 vq->shadow_used_split = new_shadow_used_split;
389 new_batch_copy_elems = rte_malloc_socket(NULL,
390 vq->size * sizeof(struct batch_copy_elem),
393 if (new_batch_copy_elems) {
394 rte_free(vq->batch_copy_elems);
395 vq->batch_copy_elems = new_batch_copy_elems;
401 /* check if we need to reallocate dev */
402 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
403 MPOL_F_NODE | MPOL_F_ADDR);
405 RTE_LOG(ERR, VHOST_CONFIG,
406 "Unable to get dev numa information.\n");
409 if (oldnode != newnode) {
410 RTE_LOG(INFO, VHOST_CONFIG,
411 "reallocate dev from %d to %d node\n",
413 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
419 memcpy(dev, old_dev, sizeof(*dev));
424 dev->virtqueue[index] = vq;
425 vhost_devices[dev->vid] = dev;
428 vhost_user_iotlb_init(dev, index);
433 static struct virtio_net*
434 numa_realloc(struct virtio_net *dev, int index __rte_unused)
440 /* Converts QEMU virtual address to Vhost virtual address. */
442 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
444 struct rte_vhost_mem_region *r;
447 /* Find the region where the address lives. */
448 for (i = 0; i < dev->mem->nregions; i++) {
449 r = &dev->mem->regions[i];
451 if (qva >= r->guest_user_addr &&
452 qva < r->guest_user_addr + r->size) {
454 if (unlikely(*len > r->guest_user_addr + r->size - qva))
455 *len = r->guest_user_addr + r->size - qva;
457 return qva - r->guest_user_addr +
468 * Converts ring address to Vhost virtual address.
469 * If IOMMU is enabled, the ring address is a guest IO virtual address,
470 * else it is a QEMU virtual address.
473 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
474 uint64_t ra, uint64_t *size)
476 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
479 vva = vhost_user_iotlb_cache_find(vq, ra,
480 size, VHOST_ACCESS_RW);
482 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
487 return qva_to_vva(dev, ra, size);
490 static struct virtio_net *
491 translate_ring_addresses(struct virtio_net *dev, int vq_index)
493 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
494 struct vhost_vring_addr *addr = &vq->ring_addrs;
497 if (vq_is_packed(dev)) {
498 len = sizeof(struct vring_packed_desc) * vq->size;
499 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
500 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
501 vq->log_guest_addr = 0;
502 if (vq->desc_packed == NULL ||
503 len != sizeof(struct vring_packed_desc) *
505 RTE_LOG(DEBUG, VHOST_CONFIG,
506 "(%d) failed to map desc_packed ring.\n",
511 dev = numa_realloc(dev, vq_index);
512 vq = dev->virtqueue[vq_index];
513 addr = &vq->ring_addrs;
515 len = sizeof(struct vring_packed_desc_event);
516 vq->driver_event = (struct vring_packed_desc_event *)
517 (uintptr_t)ring_addr_to_vva(dev,
518 vq, addr->avail_user_addr, &len);
519 if (vq->driver_event == NULL ||
520 len != sizeof(struct vring_packed_desc_event)) {
521 RTE_LOG(DEBUG, VHOST_CONFIG,
522 "(%d) failed to find driver area address.\n",
527 len = sizeof(struct vring_packed_desc_event);
528 vq->device_event = (struct vring_packed_desc_event *)
529 (uintptr_t)ring_addr_to_vva(dev,
530 vq, addr->used_user_addr, &len);
531 if (vq->device_event == NULL ||
532 len != sizeof(struct vring_packed_desc_event)) {
533 RTE_LOG(DEBUG, VHOST_CONFIG,
534 "(%d) failed to find device area address.\n",
542 /* The addresses are converted from QEMU virtual to Vhost virtual. */
543 if (vq->desc && vq->avail && vq->used)
546 len = sizeof(struct vring_desc) * vq->size;
547 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
548 vq, addr->desc_user_addr, &len);
549 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
550 RTE_LOG(DEBUG, VHOST_CONFIG,
551 "(%d) failed to map desc ring.\n",
556 dev = numa_realloc(dev, vq_index);
557 vq = dev->virtqueue[vq_index];
558 addr = &vq->ring_addrs;
560 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
561 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
562 vq, addr->avail_user_addr, &len);
563 if (vq->avail == 0 ||
564 len != sizeof(struct vring_avail) +
565 sizeof(uint16_t) * vq->size) {
566 RTE_LOG(DEBUG, VHOST_CONFIG,
567 "(%d) failed to map avail ring.\n",
572 len = sizeof(struct vring_used) +
573 sizeof(struct vring_used_elem) * vq->size;
574 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
575 vq, addr->used_user_addr, &len);
576 if (vq->used == 0 || len != sizeof(struct vring_used) +
577 sizeof(struct vring_used_elem) * vq->size) {
578 RTE_LOG(DEBUG, VHOST_CONFIG,
579 "(%d) failed to map used ring.\n",
584 if (vq->last_used_idx != vq->used->idx) {
585 RTE_LOG(WARNING, VHOST_CONFIG,
586 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
587 "some packets maybe resent for Tx and dropped for Rx\n",
588 vq->last_used_idx, vq->used->idx);
589 vq->last_used_idx = vq->used->idx;
590 vq->last_avail_idx = vq->used->idx;
593 vq->log_guest_addr = addr->log_guest_addr;
595 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
597 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
598 dev->vid, vq->avail);
599 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
601 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
602 dev->vid, vq->log_guest_addr);
608 * The virtio device sends us the desc, used and avail ring addresses.
609 * This function then converts these to our address space.
612 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
614 struct vhost_virtqueue *vq;
615 struct vhost_vring_addr *addr = &msg->payload.addr;
616 struct virtio_net *dev = *pdev;
618 if (dev->mem == NULL)
621 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
622 vq = dev->virtqueue[msg->payload.addr.index];
625 * Rings addresses should not be interpreted as long as the ring is not
626 * started and enabled
628 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
630 vring_invalidate(dev, vq);
632 if (vq->enabled && (dev->features &
633 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
634 dev = translate_ring_addresses(dev, msg->payload.addr.index);
645 * The virtio device sends us the available ring last used index.
648 vhost_user_set_vring_base(struct virtio_net *dev,
651 dev->virtqueue[msg->payload.state.index]->last_used_idx =
652 msg->payload.state.num;
653 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
654 msg->payload.state.num;
660 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
661 uint64_t host_phys_addr, uint64_t size)
663 struct guest_page *page, *last_page;
665 if (dev->nr_guest_pages == dev->max_guest_pages) {
666 dev->max_guest_pages *= 2;
667 dev->guest_pages = realloc(dev->guest_pages,
668 dev->max_guest_pages * sizeof(*page));
669 if (!dev->guest_pages) {
670 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
675 if (dev->nr_guest_pages > 0) {
676 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
677 /* merge if the two pages are continuous */
678 if (host_phys_addr == last_page->host_phys_addr +
680 last_page->size += size;
685 page = &dev->guest_pages[dev->nr_guest_pages++];
686 page->guest_phys_addr = guest_phys_addr;
687 page->host_phys_addr = host_phys_addr;
694 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
697 uint64_t reg_size = reg->size;
698 uint64_t host_user_addr = reg->host_user_addr;
699 uint64_t guest_phys_addr = reg->guest_phys_addr;
700 uint64_t host_phys_addr;
703 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
704 size = page_size - (guest_phys_addr & (page_size - 1));
705 size = RTE_MIN(size, reg_size);
707 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
710 host_user_addr += size;
711 guest_phys_addr += size;
714 while (reg_size > 0) {
715 size = RTE_MIN(reg_size, page_size);
716 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
718 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
722 host_user_addr += size;
723 guest_phys_addr += size;
730 #ifdef RTE_LIBRTE_VHOST_DEBUG
731 /* TODO: enable it only in debug mode? */
733 dump_guest_pages(struct virtio_net *dev)
736 struct guest_page *page;
738 for (i = 0; i < dev->nr_guest_pages; i++) {
739 page = &dev->guest_pages[i];
741 RTE_LOG(INFO, VHOST_CONFIG,
742 "guest physical page region %u\n"
743 "\t guest_phys_addr: %" PRIx64 "\n"
744 "\t host_phys_addr : %" PRIx64 "\n"
745 "\t size : %" PRIx64 "\n",
747 page->guest_phys_addr,
748 page->host_phys_addr,
753 #define dump_guest_pages(dev)
757 vhost_memory_changed(struct VhostUserMemory *new,
758 struct rte_vhost_memory *old)
762 if (new->nregions != old->nregions)
765 for (i = 0; i < new->nregions; ++i) {
766 VhostUserMemoryRegion *new_r = &new->regions[i];
767 struct rte_vhost_mem_region *old_r = &old->regions[i];
769 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
771 if (new_r->memory_size != old_r->size)
773 if (new_r->userspace_addr != old_r->guest_user_addr)
781 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
783 struct virtio_net *dev = *pdev;
784 struct VhostUserMemory memory = pmsg->payload.memory;
785 struct rte_vhost_mem_region *reg;
788 uint64_t mmap_offset;
794 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
795 RTE_LOG(ERR, VHOST_CONFIG,
796 "too many memory regions (%u)\n", memory.nregions);
800 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
801 RTE_LOG(INFO, VHOST_CONFIG,
802 "(%d) memory regions not changed\n", dev->vid);
804 for (i = 0; i < memory.nregions; i++)
811 free_mem_region(dev);
816 /* Flush IOTLB cache as previous HVAs are now invalid */
817 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
818 for (i = 0; i < dev->nr_vring; i++)
819 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
821 dev->nr_guest_pages = 0;
822 if (!dev->guest_pages) {
823 dev->max_guest_pages = 8;
824 dev->guest_pages = malloc(dev->max_guest_pages *
825 sizeof(struct guest_page));
826 if (dev->guest_pages == NULL) {
827 RTE_LOG(ERR, VHOST_CONFIG,
828 "(%d) failed to allocate memory "
829 "for dev->guest_pages\n",
835 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
836 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
837 if (dev->mem == NULL) {
838 RTE_LOG(ERR, VHOST_CONFIG,
839 "(%d) failed to allocate memory for dev->mem\n",
843 dev->mem->nregions = memory.nregions;
845 for (i = 0; i < memory.nregions; i++) {
847 reg = &dev->mem->regions[i];
849 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
850 reg->guest_user_addr = memory.regions[i].userspace_addr;
851 reg->size = memory.regions[i].memory_size;
854 mmap_offset = memory.regions[i].mmap_offset;
856 /* Check for memory_size + mmap_offset overflow */
857 if (mmap_offset >= -reg->size) {
858 RTE_LOG(ERR, VHOST_CONFIG,
859 "mmap_offset (%#"PRIx64") and memory_size "
860 "(%#"PRIx64") overflow\n",
861 mmap_offset, reg->size);
865 mmap_size = reg->size + mmap_offset;
867 /* mmap() without flag of MAP_ANONYMOUS, should be called
868 * with length argument aligned with hugepagesz at older
869 * longterm version Linux, like 2.6.32 and 3.2.72, or
870 * mmap() will fail with EINVAL.
872 * to avoid failure, make sure in caller to keep length
875 alignment = get_blk_size(fd);
876 if (alignment == (uint64_t)-1) {
877 RTE_LOG(ERR, VHOST_CONFIG,
878 "couldn't get hugepage size through fstat\n");
881 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
883 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
884 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
885 MAP_SHARED | populate, fd, 0);
887 if (mmap_addr == MAP_FAILED) {
888 RTE_LOG(ERR, VHOST_CONFIG,
889 "mmap region %u failed.\n", i);
893 reg->mmap_addr = mmap_addr;
894 reg->mmap_size = mmap_size;
895 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
898 if (dev->dequeue_zero_copy)
899 if (add_guest_pages(dev, reg, alignment) < 0) {
900 RTE_LOG(ERR, VHOST_CONFIG,
901 "adding guest pages to region %u failed.\n",
906 RTE_LOG(INFO, VHOST_CONFIG,
907 "guest memory region %u, size: 0x%" PRIx64 "\n"
908 "\t guest physical addr: 0x%" PRIx64 "\n"
909 "\t guest virtual addr: 0x%" PRIx64 "\n"
910 "\t host virtual addr: 0x%" PRIx64 "\n"
911 "\t mmap addr : 0x%" PRIx64 "\n"
912 "\t mmap size : 0x%" PRIx64 "\n"
913 "\t mmap align: 0x%" PRIx64 "\n"
914 "\t mmap off : 0x%" PRIx64 "\n",
916 reg->guest_phys_addr,
917 reg->guest_user_addr,
919 (uint64_t)(uintptr_t)mmap_addr,
925 for (i = 0; i < dev->nr_vring; i++) {
926 struct vhost_virtqueue *vq = dev->virtqueue[i];
928 if (vq->desc || vq->avail || vq->used) {
930 * If the memory table got updated, the ring addresses
931 * need to be translated again as virtual addresses have
934 vring_invalidate(dev, vq);
936 dev = translate_ring_addresses(dev, i);
944 dump_guest_pages(dev);
949 free_mem_region(dev);
956 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
963 if (vq_is_packed(dev))
964 rings_ok = !!vq->desc_packed;
966 rings_ok = vq->desc && vq->avail && vq->used;
969 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
970 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
974 virtio_is_ready(struct virtio_net *dev)
976 struct vhost_virtqueue *vq;
979 if (dev->nr_vring == 0)
982 for (i = 0; i < dev->nr_vring; i++) {
983 vq = dev->virtqueue[i];
985 if (!vq_is_ready(dev, vq))
989 RTE_LOG(INFO, VHOST_CONFIG,
990 "virtio is now ready for processing.\n");
995 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
997 struct vhost_vring_file file;
998 struct vhost_virtqueue *vq;
1000 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1001 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1002 file.fd = VIRTIO_INVALID_EVENTFD;
1004 file.fd = pmsg->fds[0];
1005 RTE_LOG(INFO, VHOST_CONFIG,
1006 "vring call idx:%d file:%d\n", file.index, file.fd);
1008 vq = dev->virtqueue[file.index];
1009 if (vq->callfd >= 0)
1012 vq->callfd = file.fd;
1016 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
1018 struct vhost_vring_file file;
1019 struct vhost_virtqueue *vq;
1020 struct virtio_net *dev = *pdev;
1022 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1023 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1024 file.fd = VIRTIO_INVALID_EVENTFD;
1026 file.fd = pmsg->fds[0];
1027 RTE_LOG(INFO, VHOST_CONFIG,
1028 "vring kick idx:%d file:%d\n", file.index, file.fd);
1030 /* Interpret ring addresses only when ring is started. */
1031 dev = translate_ring_addresses(dev, file.index);
1037 vq = dev->virtqueue[file.index];
1040 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1041 * the ring starts already enabled. Otherwise, it is enabled via
1042 * the SET_VRING_ENABLE message.
1044 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1047 if (vq->kickfd >= 0)
1049 vq->kickfd = file.fd;
1053 free_zmbufs(struct vhost_virtqueue *vq)
1055 struct zcopy_mbuf *zmbuf, *next;
1057 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1058 zmbuf != NULL; zmbuf = next) {
1059 next = TAILQ_NEXT(zmbuf, next);
1061 rte_pktmbuf_free(zmbuf->mbuf);
1062 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1065 rte_free(vq->zmbufs);
1069 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1072 vhost_user_get_vring_base(struct virtio_net *dev,
1075 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1077 /* We have to stop the queue (virtio) if it is running. */
1078 vhost_destroy_device_notify(dev);
1080 dev->flags &= ~VIRTIO_DEV_READY;
1081 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1083 /* Here we are safe to get the last avail index */
1084 msg->payload.state.num = vq->last_avail_idx;
1086 RTE_LOG(INFO, VHOST_CONFIG,
1087 "vring base idx:%d file:%d\n", msg->payload.state.index,
1088 msg->payload.state.num);
1090 * Based on current qemu vhost-user implementation, this message is
1091 * sent and only sent in vhost_vring_stop.
1092 * TODO: cleanup the vring, it isn't usable since here.
1094 if (vq->kickfd >= 0)
1097 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1099 if (vq->callfd >= 0)
1102 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1104 if (dev->dequeue_zero_copy)
1106 if (vq_is_packed(dev)) {
1107 rte_free(vq->shadow_used_packed);
1108 vq->shadow_used_packed = NULL;
1110 rte_free(vq->shadow_used_split);
1111 vq->shadow_used_split = NULL;
1114 rte_free(vq->batch_copy_elems);
1115 vq->batch_copy_elems = NULL;
1121 * when virtio queues are ready to work, qemu will send us to
1122 * enable the virtio queue pair.
1125 vhost_user_set_vring_enable(struct virtio_net *dev,
1128 int enable = (int)msg->payload.state.num;
1129 int index = (int)msg->payload.state.index;
1130 struct rte_vdpa_device *vdpa_dev;
1133 RTE_LOG(INFO, VHOST_CONFIG,
1134 "set queue enable: %d to qp idx: %d\n",
1137 did = dev->vdpa_dev_id;
1138 vdpa_dev = rte_vdpa_get_device(did);
1139 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1140 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1142 if (dev->notify_ops->vring_state_changed)
1143 dev->notify_ops->vring_state_changed(dev->vid,
1146 dev->virtqueue[index]->enabled = enable;
1152 vhost_user_get_protocol_features(struct virtio_net *dev,
1153 struct VhostUserMsg *msg)
1155 uint64_t features, protocol_features;
1157 rte_vhost_driver_get_features(dev->ifname, &features);
1158 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1161 * REPLY_ACK protocol feature is only mandatory for now
1162 * for IOMMU feature. If IOMMU is explicitly disabled by the
1163 * application, disable also REPLY_ACK feature for older buggy
1164 * Qemu versions (from v2.7.0 to v2.9.0).
1166 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1167 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1169 msg->payload.u64 = protocol_features;
1170 msg->size = sizeof(msg->payload.u64);
1174 vhost_user_set_protocol_features(struct virtio_net *dev,
1175 uint64_t protocol_features)
1177 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1180 dev->protocol_features = protocol_features;
1184 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1186 int fd = msg->fds[0];
1191 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1195 if (msg->size != sizeof(VhostUserLog)) {
1196 RTE_LOG(ERR, VHOST_CONFIG,
1197 "invalid log base msg size: %"PRId32" != %d\n",
1198 msg->size, (int)sizeof(VhostUserLog));
1202 size = msg->payload.log.mmap_size;
1203 off = msg->payload.log.mmap_offset;
1205 /* Don't allow mmap_offset to point outside the mmap region */
1207 RTE_LOG(ERR, VHOST_CONFIG,
1208 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1213 RTE_LOG(INFO, VHOST_CONFIG,
1214 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1218 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1219 * fail when offset is not page size aligned.
1221 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1223 if (addr == MAP_FAILED) {
1224 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1229 * Free previously mapped log memory on occasionally
1230 * multiple VHOST_USER_SET_LOG_BASE.
1232 if (dev->log_addr) {
1233 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1235 dev->log_addr = (uint64_t)(uintptr_t)addr;
1236 dev->log_base = dev->log_addr + off;
1237 dev->log_size = size;
1243 * An rarp packet is constructed and broadcasted to notify switches about
1244 * the new location of the migrated VM, so that packets from outside will
1245 * not be lost after migration.
1247 * However, we don't actually "send" a rarp packet here, instead, we set
1248 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1251 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1253 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1254 struct rte_vdpa_device *vdpa_dev;
1257 RTE_LOG(DEBUG, VHOST_CONFIG,
1258 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1259 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1260 memcpy(dev->mac.addr_bytes, mac, 6);
1263 * Set the flag to inject a RARP broadcast packet at
1264 * rte_vhost_dequeue_burst().
1266 * rte_smp_wmb() is for making sure the mac is copied
1267 * before the flag is set.
1270 rte_atomic16_set(&dev->broadcast_rarp, 1);
1271 did = dev->vdpa_dev_id;
1272 vdpa_dev = rte_vdpa_get_device(did);
1273 if (vdpa_dev && vdpa_dev->ops->migration_done)
1274 vdpa_dev->ops->migration_done(dev->vid);
1280 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1282 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1283 msg->payload.u64 > VIRTIO_MAX_MTU) {
1284 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1290 dev->mtu = msg->payload.u64;
1296 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1298 int fd = msg->fds[0];
1301 RTE_LOG(ERR, VHOST_CONFIG,
1302 "Invalid file descriptor for slave channel (%d)\n",
1307 dev->slave_req_fd = fd;
1313 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1315 struct vhost_vring_addr *ra;
1316 uint64_t start, end;
1319 end = start + imsg->size;
1321 ra = &vq->ring_addrs;
1322 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1324 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1326 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1333 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1334 struct vhost_iotlb_msg *imsg)
1336 uint64_t istart, iend, vstart, vend;
1338 istart = imsg->iova;
1339 iend = istart + imsg->size - 1;
1341 vstart = (uintptr_t)vq->desc;
1342 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1343 if (vstart <= iend && istart <= vend)
1346 vstart = (uintptr_t)vq->avail;
1347 vend = vstart + sizeof(struct vring_avail);
1348 vend += sizeof(uint16_t) * vq->size - 1;
1349 if (vstart <= iend && istart <= vend)
1352 vstart = (uintptr_t)vq->used;
1353 vend = vstart + sizeof(struct vring_used);
1354 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1355 if (vstart <= iend && istart <= vend)
1362 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1364 struct virtio_net *dev = *pdev;
1365 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1369 switch (imsg->type) {
1370 case VHOST_IOTLB_UPDATE:
1372 vva = qva_to_vva(dev, imsg->uaddr, &len);
1376 for (i = 0; i < dev->nr_vring; i++) {
1377 struct vhost_virtqueue *vq = dev->virtqueue[i];
1379 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1382 if (is_vring_iotlb_update(vq, imsg))
1383 *pdev = dev = translate_ring_addresses(dev, i);
1386 case VHOST_IOTLB_INVALIDATE:
1387 for (i = 0; i < dev->nr_vring; i++) {
1388 struct vhost_virtqueue *vq = dev->virtqueue[i];
1390 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1393 if (is_vring_iotlb_invalidate(vq, imsg))
1394 vring_invalidate(dev, vq);
1398 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1406 /* return bytes# of read on success or negative val on failure. */
1408 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1412 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1413 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1417 if (msg && msg->size) {
1418 if (msg->size > sizeof(msg->payload)) {
1419 RTE_LOG(ERR, VHOST_CONFIG,
1420 "invalid msg size: %d\n", msg->size);
1423 ret = read(sockfd, &msg->payload, msg->size);
1426 if (ret != (int)msg->size) {
1427 RTE_LOG(ERR, VHOST_CONFIG,
1428 "read control message failed\n");
1437 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1442 return send_fd_message(sockfd, (char *)msg,
1443 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1447 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1452 msg->flags &= ~VHOST_USER_VERSION_MASK;
1453 msg->flags &= ~VHOST_USER_NEED_REPLY;
1454 msg->flags |= VHOST_USER_VERSION;
1455 msg->flags |= VHOST_USER_REPLY_MASK;
1457 return send_vhost_message(sockfd, msg, NULL, 0);
1461 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1462 int *fds, int fd_num)
1466 if (msg->flags & VHOST_USER_NEED_REPLY)
1467 rte_spinlock_lock(&dev->slave_req_lock);
1469 ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1470 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1471 rte_spinlock_unlock(&dev->slave_req_lock);
1477 * Allocate a queue pair if it hasn't been allocated yet
1480 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1484 switch (msg->request.master) {
1485 case VHOST_USER_SET_VRING_KICK:
1486 case VHOST_USER_SET_VRING_CALL:
1487 case VHOST_USER_SET_VRING_ERR:
1488 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1490 case VHOST_USER_SET_VRING_NUM:
1491 case VHOST_USER_SET_VRING_BASE:
1492 case VHOST_USER_SET_VRING_ENABLE:
1493 vring_idx = msg->payload.state.index;
1495 case VHOST_USER_SET_VRING_ADDR:
1496 vring_idx = msg->payload.addr.index;
1502 if (vring_idx >= VHOST_MAX_VRING) {
1503 RTE_LOG(ERR, VHOST_CONFIG,
1504 "invalid vring index: %u\n", vring_idx);
1508 if (dev->virtqueue[vring_idx])
1511 return alloc_vring_queue(dev, vring_idx);
1515 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1518 unsigned int vq_num = 0;
1520 while (vq_num < dev->nr_vring) {
1521 struct vhost_virtqueue *vq = dev->virtqueue[i];
1524 rte_spinlock_lock(&vq->access_lock);
1532 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1535 unsigned int vq_num = 0;
1537 while (vq_num < dev->nr_vring) {
1538 struct vhost_virtqueue *vq = dev->virtqueue[i];
1541 rte_spinlock_unlock(&vq->access_lock);
1549 vhost_user_msg_handler(int vid, int fd)
1551 struct virtio_net *dev;
1552 struct VhostUserMsg msg;
1553 struct rte_vdpa_device *vdpa_dev;
1556 int unlock_required = 0;
1557 uint32_t skip_master = 0;
1559 dev = get_device(vid);
1563 if (!dev->notify_ops) {
1564 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1565 if (!dev->notify_ops) {
1566 RTE_LOG(ERR, VHOST_CONFIG,
1567 "failed to get callback ops for driver %s\n",
1573 ret = read_vhost_message(fd, &msg);
1574 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1576 RTE_LOG(ERR, VHOST_CONFIG,
1577 "vhost read message failed\n");
1579 RTE_LOG(INFO, VHOST_CONFIG,
1580 "vhost peer closed\n");
1582 RTE_LOG(ERR, VHOST_CONFIG,
1583 "vhost read incorrect message\n");
1589 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1590 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1591 vhost_message_str[msg.request.master]);
1593 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1594 vhost_message_str[msg.request.master]);
1596 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1598 RTE_LOG(ERR, VHOST_CONFIG,
1599 "failed to alloc queue\n");
1604 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1605 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1606 * and device is destroyed. destroy_device waits for queues to be
1607 * inactive, so it is safe. Otherwise taking the access_lock
1608 * would cause a dead lock.
1610 switch (msg.request.master) {
1611 case VHOST_USER_SET_FEATURES:
1612 case VHOST_USER_SET_PROTOCOL_FEATURES:
1613 case VHOST_USER_SET_OWNER:
1614 case VHOST_USER_SET_MEM_TABLE:
1615 case VHOST_USER_SET_LOG_BASE:
1616 case VHOST_USER_SET_LOG_FD:
1617 case VHOST_USER_SET_VRING_NUM:
1618 case VHOST_USER_SET_VRING_ADDR:
1619 case VHOST_USER_SET_VRING_BASE:
1620 case VHOST_USER_SET_VRING_KICK:
1621 case VHOST_USER_SET_VRING_CALL:
1622 case VHOST_USER_SET_VRING_ERR:
1623 case VHOST_USER_SET_VRING_ENABLE:
1624 case VHOST_USER_SEND_RARP:
1625 case VHOST_USER_NET_SET_MTU:
1626 case VHOST_USER_SET_SLAVE_REQ_FD:
1627 vhost_user_lock_all_queue_pairs(dev);
1628 unlock_required = 1;
1635 if (dev->extern_ops.pre_msg_handle) {
1636 uint32_t need_reply;
1638 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1639 (void *)&msg, &need_reply, &skip_master);
1644 send_vhost_reply(fd, &msg);
1647 goto skip_to_post_handle;
1650 switch (msg.request.master) {
1651 case VHOST_USER_GET_FEATURES:
1652 msg.payload.u64 = vhost_user_get_features(dev);
1653 msg.size = sizeof(msg.payload.u64);
1654 send_vhost_reply(fd, &msg);
1656 case VHOST_USER_SET_FEATURES:
1657 ret = vhost_user_set_features(dev, msg.payload.u64);
1662 case VHOST_USER_GET_PROTOCOL_FEATURES:
1663 vhost_user_get_protocol_features(dev, &msg);
1664 send_vhost_reply(fd, &msg);
1666 case VHOST_USER_SET_PROTOCOL_FEATURES:
1667 vhost_user_set_protocol_features(dev, msg.payload.u64);
1670 case VHOST_USER_SET_OWNER:
1671 vhost_user_set_owner();
1673 case VHOST_USER_RESET_OWNER:
1674 vhost_user_reset_owner(dev);
1677 case VHOST_USER_SET_MEM_TABLE:
1678 ret = vhost_user_set_mem_table(&dev, &msg);
1681 case VHOST_USER_SET_LOG_BASE:
1682 vhost_user_set_log_base(dev, &msg);
1684 /* it needs a reply */
1685 msg.size = sizeof(msg.payload.u64);
1686 send_vhost_reply(fd, &msg);
1688 case VHOST_USER_SET_LOG_FD:
1690 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1693 case VHOST_USER_SET_VRING_NUM:
1694 vhost_user_set_vring_num(dev, &msg);
1696 case VHOST_USER_SET_VRING_ADDR:
1697 vhost_user_set_vring_addr(&dev, &msg);
1699 case VHOST_USER_SET_VRING_BASE:
1700 vhost_user_set_vring_base(dev, &msg);
1703 case VHOST_USER_GET_VRING_BASE:
1704 vhost_user_get_vring_base(dev, &msg);
1705 msg.size = sizeof(msg.payload.state);
1706 send_vhost_reply(fd, &msg);
1709 case VHOST_USER_SET_VRING_KICK:
1710 vhost_user_set_vring_kick(&dev, &msg);
1712 case VHOST_USER_SET_VRING_CALL:
1713 vhost_user_set_vring_call(dev, &msg);
1716 case VHOST_USER_SET_VRING_ERR:
1717 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1719 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1722 case VHOST_USER_GET_QUEUE_NUM:
1723 msg.payload.u64 = (uint64_t)vhost_user_get_queue_num(dev);
1724 msg.size = sizeof(msg.payload.u64);
1725 send_vhost_reply(fd, &msg);
1728 case VHOST_USER_SET_VRING_ENABLE:
1729 vhost_user_set_vring_enable(dev, &msg);
1731 case VHOST_USER_SEND_RARP:
1732 vhost_user_send_rarp(dev, &msg);
1735 case VHOST_USER_NET_SET_MTU:
1736 ret = vhost_user_net_set_mtu(dev, &msg);
1739 case VHOST_USER_SET_SLAVE_REQ_FD:
1740 ret = vhost_user_set_req_fd(dev, &msg);
1743 case VHOST_USER_IOTLB_MSG:
1744 ret = vhost_user_iotlb_msg(&dev, &msg);
1752 skip_to_post_handle:
1753 if (dev->extern_ops.post_msg_handle) {
1754 uint32_t need_reply;
1756 ret = (*dev->extern_ops.post_msg_handle)(
1757 dev->vid, (void *)&msg, &need_reply);
1762 send_vhost_reply(fd, &msg);
1766 if (unlock_required)
1767 vhost_user_unlock_all_queue_pairs(dev);
1769 if (msg.flags & VHOST_USER_NEED_REPLY) {
1770 msg.payload.u64 = !!ret;
1771 msg.size = sizeof(msg.payload.u64);
1772 send_vhost_reply(fd, &msg);
1775 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1776 dev->flags |= VIRTIO_DEV_READY;
1778 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1779 if (dev->dequeue_zero_copy) {
1780 RTE_LOG(INFO, VHOST_CONFIG,
1781 "dequeue zero copy is enabled\n");
1784 if (dev->notify_ops->new_device(dev->vid) == 0)
1785 dev->flags |= VIRTIO_DEV_RUNNING;
1789 did = dev->vdpa_dev_id;
1790 vdpa_dev = rte_vdpa_get_device(did);
1791 if (vdpa_dev && virtio_is_ready(dev) &&
1792 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1793 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1794 if (vdpa_dev->ops->dev_conf)
1795 vdpa_dev->ops->dev_conf(dev->vid);
1796 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1797 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1798 RTE_LOG(INFO, VHOST_CONFIG,
1799 "(%d) software relay is used for vDPA, performance may be low.\n",
1807 static int process_slave_message_reply(struct virtio_net *dev,
1808 const VhostUserMsg *msg)
1810 VhostUserMsg msg_reply;
1813 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1816 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1821 if (msg_reply.request.slave != msg->request.slave) {
1822 RTE_LOG(ERR, VHOST_CONFIG,
1823 "Received unexpected msg type (%u), expected %u\n",
1824 msg_reply.request.slave, msg->request.slave);
1829 ret = msg_reply.payload.u64 ? -1 : 0;
1832 rte_spinlock_unlock(&dev->slave_req_lock);
1837 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1840 struct VhostUserMsg msg = {
1841 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1842 .flags = VHOST_USER_VERSION,
1843 .size = sizeof(msg.payload.iotlb),
1847 .type = VHOST_IOTLB_MISS,
1851 ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1853 RTE_LOG(ERR, VHOST_CONFIG,
1854 "Failed to send IOTLB miss message (%d)\n",
1862 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1870 struct VhostUserMsg msg = {
1871 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1872 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1873 .size = sizeof(msg.payload.area),
1875 .u64 = index & VHOST_USER_VRING_IDX_MASK,
1882 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1888 ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1890 RTE_LOG(ERR, VHOST_CONFIG,
1891 "Failed to set host notifier (%d)\n", ret);
1895 return process_slave_message_reply(dev, &msg);
1898 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1900 struct virtio_net *dev;
1901 struct rte_vdpa_device *vdpa_dev;
1902 int vfio_device_fd, did, ret = 0;
1903 uint64_t offset, size;
1906 dev = get_device(vid);
1910 did = dev->vdpa_dev_id;
1914 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1915 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1916 !(dev->protocol_features &
1917 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1918 !(dev->protocol_features &
1919 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1920 !(dev->protocol_features &
1921 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1924 vdpa_dev = rte_vdpa_get_device(did);
1928 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1929 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1931 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1932 if (vfio_device_fd < 0)
1936 for (i = 0; i < dev->nr_vring; i++) {
1937 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1943 if (vhost_user_slave_set_vring_host_notifier(dev, i,
1944 vfio_device_fd, offset, size) < 0) {
1951 for (i = 0; i < dev->nr_vring; i++) {
1952 vhost_user_slave_set_vring_host_notifier(dev, i, -1,