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(struct virtio_net **pdev __rte_unused,
131 struct VhostUserMsg *msg __rte_unused)
137 vhost_user_reset_owner(struct virtio_net **pdev,
138 struct VhostUserMsg *msg __rte_unused)
140 struct virtio_net *dev = *pdev;
141 vhost_destroy_device_notify(dev);
143 cleanup_device(dev, 0);
149 * The features that we support are requested.
152 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg)
154 struct virtio_net *dev = *pdev;
155 uint64_t features = 0;
157 rte_vhost_driver_get_features(dev->ifname, &features);
159 msg->payload.u64 = features;
160 msg->size = sizeof(msg->payload.u64);
162 return VH_RESULT_REPLY;
166 * The queue number that we support are requested.
169 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg)
171 struct virtio_net *dev = *pdev;
172 uint32_t queue_num = 0;
174 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
176 msg->payload.u64 = (uint64_t)queue_num;
177 msg->size = sizeof(msg->payload.u64);
179 return VH_RESULT_REPLY;
183 * We receive the negotiated features supported by us and the virtio device.
186 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg)
188 struct virtio_net *dev = *pdev;
189 uint64_t features = msg->payload.u64;
190 uint64_t vhost_features = 0;
191 struct rte_vdpa_device *vdpa_dev;
194 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
195 if (features & ~vhost_features) {
196 RTE_LOG(ERR, VHOST_CONFIG,
197 "(%d) received invalid negotiated features.\n",
199 return VH_RESULT_ERR;
202 if (dev->flags & VIRTIO_DEV_RUNNING) {
203 if (dev->features == features)
207 * Error out if master tries to change features while device is
208 * in running state. The exception being VHOST_F_LOG_ALL, which
209 * is enabled when the live-migration starts.
211 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
212 RTE_LOG(ERR, VHOST_CONFIG,
213 "(%d) features changed while device is running.\n",
215 return VH_RESULT_ERR;
218 if (dev->notify_ops->features_changed)
219 dev->notify_ops->features_changed(dev->vid, features);
222 dev->features = features;
224 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
225 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
227 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
229 VHOST_LOG_DEBUG(VHOST_CONFIG,
230 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
232 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
233 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
235 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
236 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
238 * Remove all but first queue pair if MQ hasn't been
239 * negotiated. This is safe because the device is not
240 * running at this stage.
242 while (dev->nr_vring > 2) {
243 struct vhost_virtqueue *vq;
245 vq = dev->virtqueue[--dev->nr_vring];
249 dev->virtqueue[dev->nr_vring] = NULL;
255 did = dev->vdpa_dev_id;
256 vdpa_dev = rte_vdpa_get_device(did);
257 if (vdpa_dev && vdpa_dev->ops->set_features)
258 vdpa_dev->ops->set_features(dev->vid);
264 * The virtio device sends us the size of the descriptor ring.
267 vhost_user_set_vring_num(struct virtio_net **pdev,
268 struct VhostUserMsg *msg)
270 struct virtio_net *dev = *pdev;
271 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
273 vq->size = msg->payload.state.num;
275 /* VIRTIO 1.0, 2.4 Virtqueues says:
277 * Queue Size value is always a power of 2. The maximum Queue Size
280 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
281 RTE_LOG(ERR, VHOST_CONFIG,
282 "invalid virtqueue size %u\n", vq->size);
283 return VH_RESULT_ERR;
286 if (dev->dequeue_zero_copy) {
288 vq->last_zmbuf_idx = 0;
289 vq->zmbuf_size = vq->size;
290 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
291 sizeof(struct zcopy_mbuf), 0);
292 if (vq->zmbufs == NULL) {
293 RTE_LOG(WARNING, VHOST_CONFIG,
294 "failed to allocate mem for zero copy; "
295 "zero copy is force disabled\n");
296 dev->dequeue_zero_copy = 0;
298 TAILQ_INIT(&vq->zmbuf_list);
301 if (vq_is_packed(dev)) {
302 vq->shadow_used_packed = rte_malloc(NULL,
304 sizeof(struct vring_used_elem_packed),
305 RTE_CACHE_LINE_SIZE);
306 if (!vq->shadow_used_packed) {
307 RTE_LOG(ERR, VHOST_CONFIG,
308 "failed to allocate memory for shadow used ring.\n");
309 return VH_RESULT_ERR;
313 vq->shadow_used_split = rte_malloc(NULL,
314 vq->size * sizeof(struct vring_used_elem),
315 RTE_CACHE_LINE_SIZE);
316 if (!vq->shadow_used_split) {
317 RTE_LOG(ERR, VHOST_CONFIG,
318 "failed to allocate memory for shadow used ring.\n");
319 return VH_RESULT_ERR;
323 vq->batch_copy_elems = rte_malloc(NULL,
324 vq->size * sizeof(struct batch_copy_elem),
325 RTE_CACHE_LINE_SIZE);
326 if (!vq->batch_copy_elems) {
327 RTE_LOG(ERR, VHOST_CONFIG,
328 "failed to allocate memory for batching copy.\n");
329 return VH_RESULT_ERR;
336 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
337 * same numa node as the memory of vring descriptor.
339 #ifdef RTE_LIBRTE_VHOST_NUMA
340 static struct virtio_net*
341 numa_realloc(struct virtio_net *dev, int index)
343 int oldnode, newnode;
344 struct virtio_net *old_dev;
345 struct vhost_virtqueue *old_vq, *vq;
346 struct zcopy_mbuf *new_zmbuf;
347 struct vring_used_elem *new_shadow_used_split;
348 struct vring_used_elem_packed *new_shadow_used_packed;
349 struct batch_copy_elem *new_batch_copy_elems;
353 vq = old_vq = dev->virtqueue[index];
355 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
356 MPOL_F_NODE | MPOL_F_ADDR);
358 /* check if we need to reallocate vq */
359 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
360 MPOL_F_NODE | MPOL_F_ADDR);
362 RTE_LOG(ERR, VHOST_CONFIG,
363 "Unable to get vq numa information.\n");
366 if (oldnode != newnode) {
367 RTE_LOG(INFO, VHOST_CONFIG,
368 "reallocate vq from %d to %d node\n", oldnode, newnode);
369 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
373 memcpy(vq, old_vq, sizeof(*vq));
374 TAILQ_INIT(&vq->zmbuf_list);
376 if (dev->dequeue_zero_copy) {
377 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
378 sizeof(struct zcopy_mbuf), 0, newnode);
380 rte_free(vq->zmbufs);
381 vq->zmbufs = new_zmbuf;
385 if (vq_is_packed(dev)) {
386 new_shadow_used_packed = rte_malloc_socket(NULL,
388 sizeof(struct vring_used_elem_packed),
391 if (new_shadow_used_packed) {
392 rte_free(vq->shadow_used_packed);
393 vq->shadow_used_packed = new_shadow_used_packed;
396 new_shadow_used_split = rte_malloc_socket(NULL,
398 sizeof(struct vring_used_elem),
401 if (new_shadow_used_split) {
402 rte_free(vq->shadow_used_split);
403 vq->shadow_used_split = new_shadow_used_split;
407 new_batch_copy_elems = rte_malloc_socket(NULL,
408 vq->size * sizeof(struct batch_copy_elem),
411 if (new_batch_copy_elems) {
412 rte_free(vq->batch_copy_elems);
413 vq->batch_copy_elems = new_batch_copy_elems;
419 /* check if we need to reallocate dev */
420 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
421 MPOL_F_NODE | MPOL_F_ADDR);
423 RTE_LOG(ERR, VHOST_CONFIG,
424 "Unable to get dev numa information.\n");
427 if (oldnode != newnode) {
428 RTE_LOG(INFO, VHOST_CONFIG,
429 "reallocate dev from %d to %d node\n",
431 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
437 memcpy(dev, old_dev, sizeof(*dev));
442 dev->virtqueue[index] = vq;
443 vhost_devices[dev->vid] = dev;
446 vhost_user_iotlb_init(dev, index);
451 static struct virtio_net*
452 numa_realloc(struct virtio_net *dev, int index __rte_unused)
458 /* Converts QEMU virtual address to Vhost virtual address. */
460 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
462 struct rte_vhost_mem_region *r;
465 /* Find the region where the address lives. */
466 for (i = 0; i < dev->mem->nregions; i++) {
467 r = &dev->mem->regions[i];
469 if (qva >= r->guest_user_addr &&
470 qva < r->guest_user_addr + r->size) {
472 if (unlikely(*len > r->guest_user_addr + r->size - qva))
473 *len = r->guest_user_addr + r->size - qva;
475 return qva - r->guest_user_addr +
486 * Converts ring address to Vhost virtual address.
487 * If IOMMU is enabled, the ring address is a guest IO virtual address,
488 * else it is a QEMU virtual address.
491 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
492 uint64_t ra, uint64_t *size)
494 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
497 vva = vhost_user_iotlb_cache_find(vq, ra,
498 size, VHOST_ACCESS_RW);
500 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
505 return qva_to_vva(dev, ra, size);
508 static struct virtio_net *
509 translate_ring_addresses(struct virtio_net *dev, int vq_index)
511 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
512 struct vhost_vring_addr *addr = &vq->ring_addrs;
515 if (vq_is_packed(dev)) {
516 len = sizeof(struct vring_packed_desc) * vq->size;
517 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
518 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
519 vq->log_guest_addr = 0;
520 if (vq->desc_packed == NULL ||
521 len != sizeof(struct vring_packed_desc) *
523 RTE_LOG(DEBUG, VHOST_CONFIG,
524 "(%d) failed to map desc_packed ring.\n",
529 dev = numa_realloc(dev, vq_index);
530 vq = dev->virtqueue[vq_index];
531 addr = &vq->ring_addrs;
533 len = sizeof(struct vring_packed_desc_event);
534 vq->driver_event = (struct vring_packed_desc_event *)
535 (uintptr_t)ring_addr_to_vva(dev,
536 vq, addr->avail_user_addr, &len);
537 if (vq->driver_event == NULL ||
538 len != sizeof(struct vring_packed_desc_event)) {
539 RTE_LOG(DEBUG, VHOST_CONFIG,
540 "(%d) failed to find driver area address.\n",
545 len = sizeof(struct vring_packed_desc_event);
546 vq->device_event = (struct vring_packed_desc_event *)
547 (uintptr_t)ring_addr_to_vva(dev,
548 vq, addr->used_user_addr, &len);
549 if (vq->device_event == NULL ||
550 len != sizeof(struct vring_packed_desc_event)) {
551 RTE_LOG(DEBUG, VHOST_CONFIG,
552 "(%d) failed to find device area address.\n",
560 /* The addresses are converted from QEMU virtual to Vhost virtual. */
561 if (vq->desc && vq->avail && vq->used)
564 len = sizeof(struct vring_desc) * vq->size;
565 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
566 vq, addr->desc_user_addr, &len);
567 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
568 RTE_LOG(DEBUG, VHOST_CONFIG,
569 "(%d) failed to map desc ring.\n",
574 dev = numa_realloc(dev, vq_index);
575 vq = dev->virtqueue[vq_index];
576 addr = &vq->ring_addrs;
578 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
579 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
580 vq, addr->avail_user_addr, &len);
581 if (vq->avail == 0 ||
582 len != sizeof(struct vring_avail) +
583 sizeof(uint16_t) * vq->size) {
584 RTE_LOG(DEBUG, VHOST_CONFIG,
585 "(%d) failed to map avail ring.\n",
590 len = sizeof(struct vring_used) +
591 sizeof(struct vring_used_elem) * vq->size;
592 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
593 vq, addr->used_user_addr, &len);
594 if (vq->used == 0 || len != sizeof(struct vring_used) +
595 sizeof(struct vring_used_elem) * vq->size) {
596 RTE_LOG(DEBUG, VHOST_CONFIG,
597 "(%d) failed to map used ring.\n",
602 if (vq->last_used_idx != vq->used->idx) {
603 RTE_LOG(WARNING, VHOST_CONFIG,
604 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
605 "some packets maybe resent for Tx and dropped for Rx\n",
606 vq->last_used_idx, vq->used->idx);
607 vq->last_used_idx = vq->used->idx;
608 vq->last_avail_idx = vq->used->idx;
611 vq->log_guest_addr = addr->log_guest_addr;
613 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
615 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
616 dev->vid, vq->avail);
617 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
619 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
620 dev->vid, vq->log_guest_addr);
626 * The virtio device sends us the desc, used and avail ring addresses.
627 * This function then converts these to our address space.
630 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg)
632 struct virtio_net *dev = *pdev;
633 struct vhost_virtqueue *vq;
634 struct vhost_vring_addr *addr = &msg->payload.addr;
636 if (dev->mem == NULL)
637 return VH_RESULT_ERR;
639 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
640 vq = dev->virtqueue[msg->payload.addr.index];
643 * Rings addresses should not be interpreted as long as the ring is not
644 * started and enabled
646 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
648 vring_invalidate(dev, vq);
650 if (vq->enabled && (dev->features &
651 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
652 dev = translate_ring_addresses(dev, msg->payload.addr.index);
654 return VH_RESULT_ERR;
663 * The virtio device sends us the available ring last used index.
666 vhost_user_set_vring_base(struct virtio_net **pdev,
667 struct VhostUserMsg *msg)
669 struct virtio_net *dev = *pdev;
670 dev->virtqueue[msg->payload.state.index]->last_used_idx =
671 msg->payload.state.num;
672 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
673 msg->payload.state.num;
679 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
680 uint64_t host_phys_addr, uint64_t size)
682 struct guest_page *page, *last_page;
684 if (dev->nr_guest_pages == dev->max_guest_pages) {
685 dev->max_guest_pages *= 2;
686 dev->guest_pages = realloc(dev->guest_pages,
687 dev->max_guest_pages * sizeof(*page));
688 if (!dev->guest_pages) {
689 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
694 if (dev->nr_guest_pages > 0) {
695 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
696 /* merge if the two pages are continuous */
697 if (host_phys_addr == last_page->host_phys_addr +
699 last_page->size += size;
704 page = &dev->guest_pages[dev->nr_guest_pages++];
705 page->guest_phys_addr = guest_phys_addr;
706 page->host_phys_addr = host_phys_addr;
713 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
716 uint64_t reg_size = reg->size;
717 uint64_t host_user_addr = reg->host_user_addr;
718 uint64_t guest_phys_addr = reg->guest_phys_addr;
719 uint64_t host_phys_addr;
722 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
723 size = page_size - (guest_phys_addr & (page_size - 1));
724 size = RTE_MIN(size, reg_size);
726 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
729 host_user_addr += size;
730 guest_phys_addr += size;
733 while (reg_size > 0) {
734 size = RTE_MIN(reg_size, page_size);
735 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
737 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
741 host_user_addr += size;
742 guest_phys_addr += size;
749 #ifdef RTE_LIBRTE_VHOST_DEBUG
750 /* TODO: enable it only in debug mode? */
752 dump_guest_pages(struct virtio_net *dev)
755 struct guest_page *page;
757 for (i = 0; i < dev->nr_guest_pages; i++) {
758 page = &dev->guest_pages[i];
760 RTE_LOG(INFO, VHOST_CONFIG,
761 "guest physical page region %u\n"
762 "\t guest_phys_addr: %" PRIx64 "\n"
763 "\t host_phys_addr : %" PRIx64 "\n"
764 "\t size : %" PRIx64 "\n",
766 page->guest_phys_addr,
767 page->host_phys_addr,
772 #define dump_guest_pages(dev)
776 vhost_memory_changed(struct VhostUserMemory *new,
777 struct rte_vhost_memory *old)
781 if (new->nregions != old->nregions)
784 for (i = 0; i < new->nregions; ++i) {
785 VhostUserMemoryRegion *new_r = &new->regions[i];
786 struct rte_vhost_mem_region *old_r = &old->regions[i];
788 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
790 if (new_r->memory_size != old_r->size)
792 if (new_r->userspace_addr != old_r->guest_user_addr)
800 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg)
802 struct virtio_net *dev = *pdev;
803 struct VhostUserMemory memory = msg->payload.memory;
804 struct rte_vhost_mem_region *reg;
807 uint64_t mmap_offset;
813 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
814 RTE_LOG(ERR, VHOST_CONFIG,
815 "too many memory regions (%u)\n", memory.nregions);
816 return VH_RESULT_ERR;
819 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
820 RTE_LOG(INFO, VHOST_CONFIG,
821 "(%d) memory regions not changed\n", dev->vid);
823 for (i = 0; i < memory.nregions; i++)
830 free_mem_region(dev);
835 /* Flush IOTLB cache as previous HVAs are now invalid */
836 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
837 for (i = 0; i < dev->nr_vring; i++)
838 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
840 dev->nr_guest_pages = 0;
841 if (!dev->guest_pages) {
842 dev->max_guest_pages = 8;
843 dev->guest_pages = malloc(dev->max_guest_pages *
844 sizeof(struct guest_page));
845 if (dev->guest_pages == NULL) {
846 RTE_LOG(ERR, VHOST_CONFIG,
847 "(%d) failed to allocate memory "
848 "for dev->guest_pages\n",
850 return VH_RESULT_ERR;
854 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
855 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
856 if (dev->mem == NULL) {
857 RTE_LOG(ERR, VHOST_CONFIG,
858 "(%d) failed to allocate memory for dev->mem\n",
860 return VH_RESULT_ERR;
862 dev->mem->nregions = memory.nregions;
864 for (i = 0; i < memory.nregions; i++) {
866 reg = &dev->mem->regions[i];
868 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
869 reg->guest_user_addr = memory.regions[i].userspace_addr;
870 reg->size = memory.regions[i].memory_size;
873 mmap_offset = memory.regions[i].mmap_offset;
875 /* Check for memory_size + mmap_offset overflow */
876 if (mmap_offset >= -reg->size) {
877 RTE_LOG(ERR, VHOST_CONFIG,
878 "mmap_offset (%#"PRIx64") and memory_size "
879 "(%#"PRIx64") overflow\n",
880 mmap_offset, reg->size);
884 mmap_size = reg->size + mmap_offset;
886 /* mmap() without flag of MAP_ANONYMOUS, should be called
887 * with length argument aligned with hugepagesz at older
888 * longterm version Linux, like 2.6.32 and 3.2.72, or
889 * mmap() will fail with EINVAL.
891 * to avoid failure, make sure in caller to keep length
894 alignment = get_blk_size(fd);
895 if (alignment == (uint64_t)-1) {
896 RTE_LOG(ERR, VHOST_CONFIG,
897 "couldn't get hugepage size through fstat\n");
900 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
902 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
903 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
904 MAP_SHARED | populate, fd, 0);
906 if (mmap_addr == MAP_FAILED) {
907 RTE_LOG(ERR, VHOST_CONFIG,
908 "mmap region %u failed.\n", i);
912 reg->mmap_addr = mmap_addr;
913 reg->mmap_size = mmap_size;
914 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
917 if (dev->dequeue_zero_copy)
918 if (add_guest_pages(dev, reg, alignment) < 0) {
919 RTE_LOG(ERR, VHOST_CONFIG,
920 "adding guest pages to region %u failed.\n",
925 RTE_LOG(INFO, VHOST_CONFIG,
926 "guest memory region %u, size: 0x%" PRIx64 "\n"
927 "\t guest physical addr: 0x%" PRIx64 "\n"
928 "\t guest virtual addr: 0x%" PRIx64 "\n"
929 "\t host virtual addr: 0x%" PRIx64 "\n"
930 "\t mmap addr : 0x%" PRIx64 "\n"
931 "\t mmap size : 0x%" PRIx64 "\n"
932 "\t mmap align: 0x%" PRIx64 "\n"
933 "\t mmap off : 0x%" PRIx64 "\n",
935 reg->guest_phys_addr,
936 reg->guest_user_addr,
938 (uint64_t)(uintptr_t)mmap_addr,
944 for (i = 0; i < dev->nr_vring; i++) {
945 struct vhost_virtqueue *vq = dev->virtqueue[i];
947 if (vq->desc || vq->avail || vq->used) {
949 * If the memory table got updated, the ring addresses
950 * need to be translated again as virtual addresses have
953 vring_invalidate(dev, vq);
955 dev = translate_ring_addresses(dev, i);
957 return VH_RESULT_ERR;
963 dump_guest_pages(dev);
968 free_mem_region(dev);
971 return VH_RESULT_ERR;
975 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
982 if (vq_is_packed(dev))
983 rings_ok = !!vq->desc_packed;
985 rings_ok = vq->desc && vq->avail && vq->used;
988 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
989 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
993 virtio_is_ready(struct virtio_net *dev)
995 struct vhost_virtqueue *vq;
998 if (dev->nr_vring == 0)
1001 for (i = 0; i < dev->nr_vring; i++) {
1002 vq = dev->virtqueue[i];
1004 if (!vq_is_ready(dev, vq))
1008 RTE_LOG(INFO, VHOST_CONFIG,
1009 "virtio is now ready for processing.\n");
1014 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg)
1016 struct virtio_net *dev = *pdev;
1017 struct vhost_vring_file file;
1018 struct vhost_virtqueue *vq;
1020 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1021 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1022 file.fd = VIRTIO_INVALID_EVENTFD;
1024 file.fd = msg->fds[0];
1025 RTE_LOG(INFO, VHOST_CONFIG,
1026 "vring call idx:%d file:%d\n", file.index, file.fd);
1028 vq = dev->virtqueue[file.index];
1029 if (vq->callfd >= 0)
1032 vq->callfd = file.fd;
1034 return VH_RESULT_OK;
1037 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1038 struct VhostUserMsg *msg)
1040 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1042 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1044 return VH_RESULT_OK;
1048 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg)
1050 struct virtio_net *dev = *pdev;
1051 struct vhost_vring_file file;
1052 struct vhost_virtqueue *vq;
1054 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1055 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1056 file.fd = VIRTIO_INVALID_EVENTFD;
1058 file.fd = msg->fds[0];
1059 RTE_LOG(INFO, VHOST_CONFIG,
1060 "vring kick idx:%d file:%d\n", file.index, file.fd);
1062 /* Interpret ring addresses only when ring is started. */
1063 dev = translate_ring_addresses(dev, file.index);
1065 return VH_RESULT_ERR;
1069 vq = dev->virtqueue[file.index];
1072 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1073 * the ring starts already enabled. Otherwise, it is enabled via
1074 * the SET_VRING_ENABLE message.
1076 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1079 if (vq->kickfd >= 0)
1081 vq->kickfd = file.fd;
1083 return VH_RESULT_OK;
1087 free_zmbufs(struct vhost_virtqueue *vq)
1089 struct zcopy_mbuf *zmbuf, *next;
1091 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1092 zmbuf != NULL; zmbuf = next) {
1093 next = TAILQ_NEXT(zmbuf, next);
1095 rte_pktmbuf_free(zmbuf->mbuf);
1096 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1099 rte_free(vq->zmbufs);
1103 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1106 vhost_user_get_vring_base(struct virtio_net **pdev,
1107 struct VhostUserMsg *msg)
1109 struct virtio_net *dev = *pdev;
1110 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1112 /* We have to stop the queue (virtio) if it is running. */
1113 vhost_destroy_device_notify(dev);
1115 dev->flags &= ~VIRTIO_DEV_READY;
1116 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1118 /* Here we are safe to get the last avail index */
1119 msg->payload.state.num = vq->last_avail_idx;
1121 RTE_LOG(INFO, VHOST_CONFIG,
1122 "vring base idx:%d file:%d\n", msg->payload.state.index,
1123 msg->payload.state.num);
1125 * Based on current qemu vhost-user implementation, this message is
1126 * sent and only sent in vhost_vring_stop.
1127 * TODO: cleanup the vring, it isn't usable since here.
1129 if (vq->kickfd >= 0)
1132 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1134 if (vq->callfd >= 0)
1137 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1139 if (dev->dequeue_zero_copy)
1141 if (vq_is_packed(dev)) {
1142 rte_free(vq->shadow_used_packed);
1143 vq->shadow_used_packed = NULL;
1145 rte_free(vq->shadow_used_split);
1146 vq->shadow_used_split = NULL;
1149 rte_free(vq->batch_copy_elems);
1150 vq->batch_copy_elems = NULL;
1152 msg->size = sizeof(msg->payload.state);
1154 return VH_RESULT_REPLY;
1158 * when virtio queues are ready to work, qemu will send us to
1159 * enable the virtio queue pair.
1162 vhost_user_set_vring_enable(struct virtio_net **pdev,
1163 struct VhostUserMsg *msg)
1165 struct virtio_net *dev = *pdev;
1166 int enable = (int)msg->payload.state.num;
1167 int index = (int)msg->payload.state.index;
1168 struct rte_vdpa_device *vdpa_dev;
1171 RTE_LOG(INFO, VHOST_CONFIG,
1172 "set queue enable: %d to qp idx: %d\n",
1175 did = dev->vdpa_dev_id;
1176 vdpa_dev = rte_vdpa_get_device(did);
1177 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1178 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1180 if (dev->notify_ops->vring_state_changed)
1181 dev->notify_ops->vring_state_changed(dev->vid,
1184 dev->virtqueue[index]->enabled = enable;
1186 return VH_RESULT_OK;
1190 vhost_user_get_protocol_features(struct virtio_net **pdev,
1191 struct VhostUserMsg *msg)
1193 struct virtio_net *dev = *pdev;
1194 uint64_t features, protocol_features;
1196 rte_vhost_driver_get_features(dev->ifname, &features);
1197 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1200 * REPLY_ACK protocol feature is only mandatory for now
1201 * for IOMMU feature. If IOMMU is explicitly disabled by the
1202 * application, disable also REPLY_ACK feature for older buggy
1203 * Qemu versions (from v2.7.0 to v2.9.0).
1205 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1206 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1208 msg->payload.u64 = protocol_features;
1209 msg->size = sizeof(msg->payload.u64);
1211 return VH_RESULT_REPLY;
1215 vhost_user_set_protocol_features(struct virtio_net **pdev,
1216 struct VhostUserMsg *msg)
1218 struct virtio_net *dev = *pdev;
1219 uint64_t protocol_features = msg->payload.u64;
1220 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES) {
1221 RTE_LOG(ERR, VHOST_CONFIG,
1222 "(%d) received invalid protocol features.\n",
1224 return VH_RESULT_ERR;
1227 dev->protocol_features = protocol_features;
1229 return VH_RESULT_OK;
1233 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg)
1235 struct virtio_net *dev = *pdev;
1236 int fd = msg->fds[0];
1241 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1242 return VH_RESULT_ERR;
1245 if (msg->size != sizeof(VhostUserLog)) {
1246 RTE_LOG(ERR, VHOST_CONFIG,
1247 "invalid log base msg size: %"PRId32" != %d\n",
1248 msg->size, (int)sizeof(VhostUserLog));
1249 return VH_RESULT_ERR;
1252 size = msg->payload.log.mmap_size;
1253 off = msg->payload.log.mmap_offset;
1255 /* Don't allow mmap_offset to point outside the mmap region */
1257 RTE_LOG(ERR, VHOST_CONFIG,
1258 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1260 return VH_RESULT_ERR;
1263 RTE_LOG(INFO, VHOST_CONFIG,
1264 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1268 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1269 * fail when offset is not page size aligned.
1271 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1273 if (addr == MAP_FAILED) {
1274 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1275 return VH_RESULT_ERR;
1279 * Free previously mapped log memory on occasionally
1280 * multiple VHOST_USER_SET_LOG_BASE.
1282 if (dev->log_addr) {
1283 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1285 dev->log_addr = (uint64_t)(uintptr_t)addr;
1286 dev->log_base = dev->log_addr + off;
1287 dev->log_size = size;
1290 * The spec is not clear about it (yet), but QEMU doesn't expect
1291 * any payload in the reply.
1295 return VH_RESULT_REPLY;
1298 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1299 struct VhostUserMsg *msg)
1302 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1304 return VH_RESULT_OK;
1308 * An rarp packet is constructed and broadcasted to notify switches about
1309 * the new location of the migrated VM, so that packets from outside will
1310 * not be lost after migration.
1312 * However, we don't actually "send" a rarp packet here, instead, we set
1313 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1316 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg)
1318 struct virtio_net *dev = *pdev;
1319 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1320 struct rte_vdpa_device *vdpa_dev;
1323 RTE_LOG(DEBUG, VHOST_CONFIG,
1324 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1325 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1326 memcpy(dev->mac.addr_bytes, mac, 6);
1329 * Set the flag to inject a RARP broadcast packet at
1330 * rte_vhost_dequeue_burst().
1332 * rte_smp_wmb() is for making sure the mac is copied
1333 * before the flag is set.
1336 rte_atomic16_set(&dev->broadcast_rarp, 1);
1337 did = dev->vdpa_dev_id;
1338 vdpa_dev = rte_vdpa_get_device(did);
1339 if (vdpa_dev && vdpa_dev->ops->migration_done)
1340 vdpa_dev->ops->migration_done(dev->vid);
1342 return VH_RESULT_OK;
1346 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg)
1348 struct virtio_net *dev = *pdev;
1349 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1350 msg->payload.u64 > VIRTIO_MAX_MTU) {
1351 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1354 return VH_RESULT_ERR;
1357 dev->mtu = msg->payload.u64;
1359 return VH_RESULT_OK;
1363 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg)
1365 struct virtio_net *dev = *pdev;
1366 int fd = msg->fds[0];
1369 RTE_LOG(ERR, VHOST_CONFIG,
1370 "Invalid file descriptor for slave channel (%d)\n",
1372 return VH_RESULT_ERR;
1375 dev->slave_req_fd = fd;
1377 return VH_RESULT_OK;
1381 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1383 struct vhost_vring_addr *ra;
1384 uint64_t start, end;
1387 end = start + imsg->size;
1389 ra = &vq->ring_addrs;
1390 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1392 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1394 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1401 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1402 struct vhost_iotlb_msg *imsg)
1404 uint64_t istart, iend, vstart, vend;
1406 istart = imsg->iova;
1407 iend = istart + imsg->size - 1;
1409 vstart = (uintptr_t)vq->desc;
1410 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1411 if (vstart <= iend && istart <= vend)
1414 vstart = (uintptr_t)vq->avail;
1415 vend = vstart + sizeof(struct vring_avail);
1416 vend += sizeof(uint16_t) * vq->size - 1;
1417 if (vstart <= iend && istart <= vend)
1420 vstart = (uintptr_t)vq->used;
1421 vend = vstart + sizeof(struct vring_used);
1422 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1423 if (vstart <= iend && istart <= vend)
1430 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1432 struct virtio_net *dev = *pdev;
1433 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1437 switch (imsg->type) {
1438 case VHOST_IOTLB_UPDATE:
1440 vva = qva_to_vva(dev, imsg->uaddr, &len);
1442 return VH_RESULT_ERR;
1444 for (i = 0; i < dev->nr_vring; i++) {
1445 struct vhost_virtqueue *vq = dev->virtqueue[i];
1447 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1450 if (is_vring_iotlb_update(vq, imsg))
1451 *pdev = dev = translate_ring_addresses(dev, i);
1454 case VHOST_IOTLB_INVALIDATE:
1455 for (i = 0; i < dev->nr_vring; i++) {
1456 struct vhost_virtqueue *vq = dev->virtqueue[i];
1458 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1461 if (is_vring_iotlb_invalidate(vq, imsg))
1462 vring_invalidate(dev, vq);
1466 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1468 return VH_RESULT_ERR;
1471 return VH_RESULT_OK;
1474 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
1475 struct VhostUserMsg *msg);
1476 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
1477 [VHOST_USER_NONE] = NULL,
1478 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
1479 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
1480 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
1481 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
1482 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
1483 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
1484 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
1485 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
1486 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
1487 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
1488 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
1489 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
1490 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
1491 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
1492 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
1493 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
1494 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
1495 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
1496 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
1497 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
1498 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
1499 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
1503 /* return bytes# of read on success or negative val on failure. */
1505 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1509 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1510 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1514 if (msg && msg->size) {
1515 if (msg->size > sizeof(msg->payload)) {
1516 RTE_LOG(ERR, VHOST_CONFIG,
1517 "invalid msg size: %d\n", msg->size);
1520 ret = read(sockfd, &msg->payload, msg->size);
1523 if (ret != (int)msg->size) {
1524 RTE_LOG(ERR, VHOST_CONFIG,
1525 "read control message failed\n");
1534 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1539 return send_fd_message(sockfd, (char *)msg,
1540 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1544 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1549 msg->flags &= ~VHOST_USER_VERSION_MASK;
1550 msg->flags &= ~VHOST_USER_NEED_REPLY;
1551 msg->flags |= VHOST_USER_VERSION;
1552 msg->flags |= VHOST_USER_REPLY_MASK;
1554 return send_vhost_message(sockfd, msg, NULL, 0);
1558 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1559 int *fds, int fd_num)
1563 if (msg->flags & VHOST_USER_NEED_REPLY)
1564 rte_spinlock_lock(&dev->slave_req_lock);
1566 ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1567 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1568 rte_spinlock_unlock(&dev->slave_req_lock);
1574 * Allocate a queue pair if it hasn't been allocated yet
1577 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1578 struct VhostUserMsg *msg)
1582 switch (msg->request.master) {
1583 case VHOST_USER_SET_VRING_KICK:
1584 case VHOST_USER_SET_VRING_CALL:
1585 case VHOST_USER_SET_VRING_ERR:
1586 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1588 case VHOST_USER_SET_VRING_NUM:
1589 case VHOST_USER_SET_VRING_BASE:
1590 case VHOST_USER_SET_VRING_ENABLE:
1591 vring_idx = msg->payload.state.index;
1593 case VHOST_USER_SET_VRING_ADDR:
1594 vring_idx = msg->payload.addr.index;
1600 if (vring_idx >= VHOST_MAX_VRING) {
1601 RTE_LOG(ERR, VHOST_CONFIG,
1602 "invalid vring index: %u\n", vring_idx);
1606 if (dev->virtqueue[vring_idx])
1609 return alloc_vring_queue(dev, vring_idx);
1613 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1616 unsigned int vq_num = 0;
1618 while (vq_num < dev->nr_vring) {
1619 struct vhost_virtqueue *vq = dev->virtqueue[i];
1622 rte_spinlock_lock(&vq->access_lock);
1630 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1633 unsigned int vq_num = 0;
1635 while (vq_num < dev->nr_vring) {
1636 struct vhost_virtqueue *vq = dev->virtqueue[i];
1639 rte_spinlock_unlock(&vq->access_lock);
1647 vhost_user_msg_handler(int vid, int fd)
1649 struct virtio_net *dev;
1650 struct VhostUserMsg msg;
1651 struct rte_vdpa_device *vdpa_dev;
1654 int unlock_required = 0;
1655 uint32_t skip_master = 0;
1658 dev = get_device(vid);
1662 if (!dev->notify_ops) {
1663 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1664 if (!dev->notify_ops) {
1665 RTE_LOG(ERR, VHOST_CONFIG,
1666 "failed to get callback ops for driver %s\n",
1672 ret = read_vhost_message(fd, &msg);
1673 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1675 RTE_LOG(ERR, VHOST_CONFIG,
1676 "vhost read message failed\n");
1678 RTE_LOG(INFO, VHOST_CONFIG,
1679 "vhost peer closed\n");
1681 RTE_LOG(ERR, VHOST_CONFIG,
1682 "vhost read incorrect message\n");
1688 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1689 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1690 vhost_message_str[msg.request.master]);
1692 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1693 vhost_message_str[msg.request.master]);
1695 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1697 RTE_LOG(ERR, VHOST_CONFIG,
1698 "failed to alloc queue\n");
1703 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1704 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1705 * and device is destroyed. destroy_device waits for queues to be
1706 * inactive, so it is safe. Otherwise taking the access_lock
1707 * would cause a dead lock.
1709 switch (msg.request.master) {
1710 case VHOST_USER_SET_FEATURES:
1711 case VHOST_USER_SET_PROTOCOL_FEATURES:
1712 case VHOST_USER_SET_OWNER:
1713 case VHOST_USER_SET_MEM_TABLE:
1714 case VHOST_USER_SET_LOG_BASE:
1715 case VHOST_USER_SET_LOG_FD:
1716 case VHOST_USER_SET_VRING_NUM:
1717 case VHOST_USER_SET_VRING_ADDR:
1718 case VHOST_USER_SET_VRING_BASE:
1719 case VHOST_USER_SET_VRING_KICK:
1720 case VHOST_USER_SET_VRING_CALL:
1721 case VHOST_USER_SET_VRING_ERR:
1722 case VHOST_USER_SET_VRING_ENABLE:
1723 case VHOST_USER_SEND_RARP:
1724 case VHOST_USER_NET_SET_MTU:
1725 case VHOST_USER_SET_SLAVE_REQ_FD:
1726 vhost_user_lock_all_queue_pairs(dev);
1727 unlock_required = 1;
1734 if (dev->extern_ops.pre_msg_handle) {
1735 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1736 (void *)&msg, &skip_master);
1737 if (ret == VH_RESULT_ERR)
1739 else if (ret == VH_RESULT_REPLY)
1740 send_vhost_reply(fd, &msg);
1743 goto skip_to_post_handle;
1746 request = msg.request.master;
1747 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
1748 if (!vhost_message_handlers[request])
1749 goto skip_to_post_handle;
1750 ret = vhost_message_handlers[request](&dev, &msg);
1754 RTE_LOG(ERR, VHOST_CONFIG,
1755 "Processing %s failed.\n",
1756 vhost_message_str[request]);
1759 RTE_LOG(DEBUG, VHOST_CONFIG,
1760 "Processing %s succeeded.\n",
1761 vhost_message_str[request]);
1763 case VH_RESULT_REPLY:
1764 RTE_LOG(DEBUG, VHOST_CONFIG,
1765 "Processing %s succeeded and needs reply.\n",
1766 vhost_message_str[request]);
1767 send_vhost_reply(fd, &msg);
1771 RTE_LOG(ERR, VHOST_CONFIG,
1772 "Requested invalid message type %d.\n", request);
1773 ret = VH_RESULT_ERR;
1776 skip_to_post_handle:
1777 if (ret != VH_RESULT_ERR && dev->extern_ops.post_msg_handle) {
1778 ret = (*dev->extern_ops.post_msg_handle)(
1779 dev->vid, (void *)&msg);
1780 if (ret == VH_RESULT_ERR)
1782 else if (ret == VH_RESULT_REPLY)
1783 send_vhost_reply(fd, &msg);
1787 if (unlock_required)
1788 vhost_user_unlock_all_queue_pairs(dev);
1791 * If the request required a reply that was already sent,
1792 * this optional reply-ack won't be sent as the
1793 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
1795 if (msg.flags & VHOST_USER_NEED_REPLY) {
1796 msg.payload.u64 = ret == VH_RESULT_ERR;
1797 msg.size = sizeof(msg.payload.u64);
1798 send_vhost_reply(fd, &msg);
1799 } else if (ret == VH_RESULT_ERR) {
1800 RTE_LOG(ERR, VHOST_CONFIG,
1801 "vhost message handling failed.\n");
1805 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1806 dev->flags |= VIRTIO_DEV_READY;
1808 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1809 if (dev->dequeue_zero_copy) {
1810 RTE_LOG(INFO, VHOST_CONFIG,
1811 "dequeue zero copy is enabled\n");
1814 if (dev->notify_ops->new_device(dev->vid) == 0)
1815 dev->flags |= VIRTIO_DEV_RUNNING;
1819 did = dev->vdpa_dev_id;
1820 vdpa_dev = rte_vdpa_get_device(did);
1821 if (vdpa_dev && virtio_is_ready(dev) &&
1822 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1823 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1824 if (vdpa_dev->ops->dev_conf)
1825 vdpa_dev->ops->dev_conf(dev->vid);
1826 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1827 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1828 RTE_LOG(INFO, VHOST_CONFIG,
1829 "(%d) software relay is used for vDPA, performance may be low.\n",
1837 static int process_slave_message_reply(struct virtio_net *dev,
1838 const struct VhostUserMsg *msg)
1840 struct VhostUserMsg msg_reply;
1843 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1846 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1851 if (msg_reply.request.slave != msg->request.slave) {
1852 RTE_LOG(ERR, VHOST_CONFIG,
1853 "Received unexpected msg type (%u), expected %u\n",
1854 msg_reply.request.slave, msg->request.slave);
1859 ret = msg_reply.payload.u64 ? -1 : 0;
1862 rte_spinlock_unlock(&dev->slave_req_lock);
1867 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1870 struct VhostUserMsg msg = {
1871 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1872 .flags = VHOST_USER_VERSION,
1873 .size = sizeof(msg.payload.iotlb),
1877 .type = VHOST_IOTLB_MISS,
1881 ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1883 RTE_LOG(ERR, VHOST_CONFIG,
1884 "Failed to send IOTLB miss message (%d)\n",
1892 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1900 struct VhostUserMsg msg = {
1901 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1902 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1903 .size = sizeof(msg.payload.area),
1905 .u64 = index & VHOST_USER_VRING_IDX_MASK,
1912 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1918 ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1920 RTE_LOG(ERR, VHOST_CONFIG,
1921 "Failed to set host notifier (%d)\n", ret);
1925 return process_slave_message_reply(dev, &msg);
1928 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1930 struct virtio_net *dev;
1931 struct rte_vdpa_device *vdpa_dev;
1932 int vfio_device_fd, did, ret = 0;
1933 uint64_t offset, size;
1936 dev = get_device(vid);
1940 did = dev->vdpa_dev_id;
1944 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1945 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1946 !(dev->protocol_features &
1947 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1948 !(dev->protocol_features &
1949 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1950 !(dev->protocol_features &
1951 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1954 vdpa_dev = rte_vdpa_get_device(did);
1958 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1959 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1961 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1962 if (vfio_device_fd < 0)
1966 for (i = 0; i < dev->nr_vring; i++) {
1967 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1973 if (vhost_user_slave_set_vring_host_notifier(dev, i,
1974 vfio_device_fd, offset, size) < 0) {
1981 for (i = 0; i < dev->nr_vring; i++) {
1982 vhost_user_slave_set_vring_host_notifier(dev, i, -1,