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",
74 /* The possible results of a message handling function */
76 /* Message handling failed */
78 /* Message handling successful */
80 /* Message handling successful and reply prepared */
90 ret = fstat(fd, &stat);
91 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
95 free_mem_region(struct virtio_net *dev)
98 struct rte_vhost_mem_region *reg;
100 if (!dev || !dev->mem)
103 for (i = 0; i < dev->mem->nregions; i++) {
104 reg = &dev->mem->regions[i];
105 if (reg->host_user_addr) {
106 munmap(reg->mmap_addr, reg->mmap_size);
113 vhost_backend_cleanup(struct virtio_net *dev)
116 free_mem_region(dev);
121 free(dev->guest_pages);
122 dev->guest_pages = NULL;
125 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
129 if (dev->slave_req_fd >= 0) {
130 close(dev->slave_req_fd);
131 dev->slave_req_fd = -1;
136 * This function just returns success at the moment unless
137 * the device hasn't been initialised.
140 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
141 struct VhostUserMsg *msg __rte_unused)
147 vhost_user_reset_owner(struct virtio_net **pdev,
148 struct VhostUserMsg *msg __rte_unused)
150 struct virtio_net *dev = *pdev;
151 vhost_destroy_device_notify(dev);
153 cleanup_device(dev, 0);
159 * The features that we support are requested.
162 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg)
164 struct virtio_net *dev = *pdev;
165 uint64_t features = 0;
167 rte_vhost_driver_get_features(dev->ifname, &features);
169 msg->payload.u64 = features;
170 msg->size = sizeof(msg->payload.u64);
172 return VH_RESULT_REPLY;
176 * The queue number that we support are requested.
179 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg)
181 struct virtio_net *dev = *pdev;
182 uint32_t queue_num = 0;
184 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
186 msg->payload.u64 = (uint64_t)queue_num;
187 msg->size = sizeof(msg->payload.u64);
189 return VH_RESULT_REPLY;
193 * We receive the negotiated features supported by us and the virtio device.
196 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg)
198 struct virtio_net *dev = *pdev;
199 uint64_t features = msg->payload.u64;
200 uint64_t vhost_features = 0;
201 struct rte_vdpa_device *vdpa_dev;
204 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
205 if (features & ~vhost_features) {
206 RTE_LOG(ERR, VHOST_CONFIG,
207 "(%d) received invalid negotiated features.\n",
209 return VH_RESULT_ERR;
212 if (dev->flags & VIRTIO_DEV_RUNNING) {
213 if (dev->features == features)
217 * Error out if master tries to change features while device is
218 * in running state. The exception being VHOST_F_LOG_ALL, which
219 * is enabled when the live-migration starts.
221 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
222 RTE_LOG(ERR, VHOST_CONFIG,
223 "(%d) features changed while device is running.\n",
225 return VH_RESULT_ERR;
228 if (dev->notify_ops->features_changed)
229 dev->notify_ops->features_changed(dev->vid, features);
232 dev->features = features;
234 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
235 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
237 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
239 VHOST_LOG_DEBUG(VHOST_CONFIG,
240 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
242 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
243 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
245 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
246 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
248 * Remove all but first queue pair if MQ hasn't been
249 * negotiated. This is safe because the device is not
250 * running at this stage.
252 while (dev->nr_vring > 2) {
253 struct vhost_virtqueue *vq;
255 vq = dev->virtqueue[--dev->nr_vring];
259 dev->virtqueue[dev->nr_vring] = NULL;
265 did = dev->vdpa_dev_id;
266 vdpa_dev = rte_vdpa_get_device(did);
267 if (vdpa_dev && vdpa_dev->ops->set_features)
268 vdpa_dev->ops->set_features(dev->vid);
274 * The virtio device sends us the size of the descriptor ring.
277 vhost_user_set_vring_num(struct virtio_net **pdev,
278 struct VhostUserMsg *msg)
280 struct virtio_net *dev = *pdev;
281 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
283 vq->size = msg->payload.state.num;
285 /* VIRTIO 1.0, 2.4 Virtqueues says:
287 * Queue Size value is always a power of 2. The maximum Queue Size
290 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
291 RTE_LOG(ERR, VHOST_CONFIG,
292 "invalid virtqueue size %u\n", vq->size);
293 return VH_RESULT_ERR;
296 if (dev->dequeue_zero_copy) {
298 vq->last_zmbuf_idx = 0;
299 vq->zmbuf_size = vq->size;
300 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
301 sizeof(struct zcopy_mbuf), 0);
302 if (vq->zmbufs == NULL) {
303 RTE_LOG(WARNING, VHOST_CONFIG,
304 "failed to allocate mem for zero copy; "
305 "zero copy is force disabled\n");
306 dev->dequeue_zero_copy = 0;
308 TAILQ_INIT(&vq->zmbuf_list);
311 if (vq_is_packed(dev)) {
312 vq->shadow_used_packed = rte_malloc(NULL,
314 sizeof(struct vring_used_elem_packed),
315 RTE_CACHE_LINE_SIZE);
316 if (!vq->shadow_used_packed) {
317 RTE_LOG(ERR, VHOST_CONFIG,
318 "failed to allocate memory for shadow used ring.\n");
319 return VH_RESULT_ERR;
323 vq->shadow_used_split = rte_malloc(NULL,
324 vq->size * sizeof(struct vring_used_elem),
325 RTE_CACHE_LINE_SIZE);
326 if (!vq->shadow_used_split) {
327 RTE_LOG(ERR, VHOST_CONFIG,
328 "failed to allocate memory for shadow used ring.\n");
329 return VH_RESULT_ERR;
333 vq->batch_copy_elems = rte_malloc(NULL,
334 vq->size * sizeof(struct batch_copy_elem),
335 RTE_CACHE_LINE_SIZE);
336 if (!vq->batch_copy_elems) {
337 RTE_LOG(ERR, VHOST_CONFIG,
338 "failed to allocate memory for batching copy.\n");
339 return VH_RESULT_ERR;
346 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
347 * same numa node as the memory of vring descriptor.
349 #ifdef RTE_LIBRTE_VHOST_NUMA
350 static struct virtio_net*
351 numa_realloc(struct virtio_net *dev, int index)
353 int oldnode, newnode;
354 struct virtio_net *old_dev;
355 struct vhost_virtqueue *old_vq, *vq;
356 struct zcopy_mbuf *new_zmbuf;
357 struct vring_used_elem *new_shadow_used_split;
358 struct vring_used_elem_packed *new_shadow_used_packed;
359 struct batch_copy_elem *new_batch_copy_elems;
363 vq = old_vq = dev->virtqueue[index];
365 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
366 MPOL_F_NODE | MPOL_F_ADDR);
368 /* check if we need to reallocate vq */
369 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
370 MPOL_F_NODE | MPOL_F_ADDR);
372 RTE_LOG(ERR, VHOST_CONFIG,
373 "Unable to get vq numa information.\n");
376 if (oldnode != newnode) {
377 RTE_LOG(INFO, VHOST_CONFIG,
378 "reallocate vq from %d to %d node\n", oldnode, newnode);
379 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
383 memcpy(vq, old_vq, sizeof(*vq));
384 TAILQ_INIT(&vq->zmbuf_list);
386 if (dev->dequeue_zero_copy) {
387 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
388 sizeof(struct zcopy_mbuf), 0, newnode);
390 rte_free(vq->zmbufs);
391 vq->zmbufs = new_zmbuf;
395 if (vq_is_packed(dev)) {
396 new_shadow_used_packed = rte_malloc_socket(NULL,
398 sizeof(struct vring_used_elem_packed),
401 if (new_shadow_used_packed) {
402 rte_free(vq->shadow_used_packed);
403 vq->shadow_used_packed = new_shadow_used_packed;
406 new_shadow_used_split = rte_malloc_socket(NULL,
408 sizeof(struct vring_used_elem),
411 if (new_shadow_used_split) {
412 rte_free(vq->shadow_used_split);
413 vq->shadow_used_split = new_shadow_used_split;
417 new_batch_copy_elems = rte_malloc_socket(NULL,
418 vq->size * sizeof(struct batch_copy_elem),
421 if (new_batch_copy_elems) {
422 rte_free(vq->batch_copy_elems);
423 vq->batch_copy_elems = new_batch_copy_elems;
429 /* check if we need to reallocate dev */
430 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
431 MPOL_F_NODE | MPOL_F_ADDR);
433 RTE_LOG(ERR, VHOST_CONFIG,
434 "Unable to get dev numa information.\n");
437 if (oldnode != newnode) {
438 RTE_LOG(INFO, VHOST_CONFIG,
439 "reallocate dev from %d to %d node\n",
441 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
447 memcpy(dev, old_dev, sizeof(*dev));
452 dev->virtqueue[index] = vq;
453 vhost_devices[dev->vid] = dev;
456 vhost_user_iotlb_init(dev, index);
461 static struct virtio_net*
462 numa_realloc(struct virtio_net *dev, int index __rte_unused)
468 /* Converts QEMU virtual address to Vhost virtual address. */
470 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
472 struct rte_vhost_mem_region *r;
475 /* Find the region where the address lives. */
476 for (i = 0; i < dev->mem->nregions; i++) {
477 r = &dev->mem->regions[i];
479 if (qva >= r->guest_user_addr &&
480 qva < r->guest_user_addr + r->size) {
482 if (unlikely(*len > r->guest_user_addr + r->size - qva))
483 *len = r->guest_user_addr + r->size - qva;
485 return qva - r->guest_user_addr +
496 * Converts ring address to Vhost virtual address.
497 * If IOMMU is enabled, the ring address is a guest IO virtual address,
498 * else it is a QEMU virtual address.
501 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
502 uint64_t ra, uint64_t *size)
504 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
507 vva = vhost_user_iotlb_cache_find(vq, ra,
508 size, VHOST_ACCESS_RW);
510 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
515 return qva_to_vva(dev, ra, size);
518 static struct virtio_net *
519 translate_ring_addresses(struct virtio_net *dev, int vq_index)
521 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
522 struct vhost_vring_addr *addr = &vq->ring_addrs;
525 if (vq_is_packed(dev)) {
526 len = sizeof(struct vring_packed_desc) * vq->size;
527 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
528 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
529 vq->log_guest_addr = 0;
530 if (vq->desc_packed == NULL ||
531 len != sizeof(struct vring_packed_desc) *
533 RTE_LOG(DEBUG, VHOST_CONFIG,
534 "(%d) failed to map desc_packed ring.\n",
539 dev = numa_realloc(dev, vq_index);
540 vq = dev->virtqueue[vq_index];
541 addr = &vq->ring_addrs;
543 len = sizeof(struct vring_packed_desc_event);
544 vq->driver_event = (struct vring_packed_desc_event *)
545 (uintptr_t)ring_addr_to_vva(dev,
546 vq, addr->avail_user_addr, &len);
547 if (vq->driver_event == NULL ||
548 len != sizeof(struct vring_packed_desc_event)) {
549 RTE_LOG(DEBUG, VHOST_CONFIG,
550 "(%d) failed to find driver area address.\n",
555 len = sizeof(struct vring_packed_desc_event);
556 vq->device_event = (struct vring_packed_desc_event *)
557 (uintptr_t)ring_addr_to_vva(dev,
558 vq, addr->used_user_addr, &len);
559 if (vq->device_event == NULL ||
560 len != sizeof(struct vring_packed_desc_event)) {
561 RTE_LOG(DEBUG, VHOST_CONFIG,
562 "(%d) failed to find device area address.\n",
570 /* The addresses are converted from QEMU virtual to Vhost virtual. */
571 if (vq->desc && vq->avail && vq->used)
574 len = sizeof(struct vring_desc) * vq->size;
575 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
576 vq, addr->desc_user_addr, &len);
577 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
578 RTE_LOG(DEBUG, VHOST_CONFIG,
579 "(%d) failed to map desc ring.\n",
584 dev = numa_realloc(dev, vq_index);
585 vq = dev->virtqueue[vq_index];
586 addr = &vq->ring_addrs;
588 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
589 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
590 vq, addr->avail_user_addr, &len);
591 if (vq->avail == 0 ||
592 len != sizeof(struct vring_avail) +
593 sizeof(uint16_t) * vq->size) {
594 RTE_LOG(DEBUG, VHOST_CONFIG,
595 "(%d) failed to map avail ring.\n",
600 len = sizeof(struct vring_used) +
601 sizeof(struct vring_used_elem) * vq->size;
602 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
603 vq, addr->used_user_addr, &len);
604 if (vq->used == 0 || len != sizeof(struct vring_used) +
605 sizeof(struct vring_used_elem) * vq->size) {
606 RTE_LOG(DEBUG, VHOST_CONFIG,
607 "(%d) failed to map used ring.\n",
612 if (vq->last_used_idx != vq->used->idx) {
613 RTE_LOG(WARNING, VHOST_CONFIG,
614 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
615 "some packets maybe resent for Tx and dropped for Rx\n",
616 vq->last_used_idx, vq->used->idx);
617 vq->last_used_idx = vq->used->idx;
618 vq->last_avail_idx = vq->used->idx;
621 vq->log_guest_addr = addr->log_guest_addr;
623 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
625 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
626 dev->vid, vq->avail);
627 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
629 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
630 dev->vid, vq->log_guest_addr);
636 * The virtio device sends us the desc, used and avail ring addresses.
637 * This function then converts these to our address space.
640 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg)
642 struct virtio_net *dev = *pdev;
643 struct vhost_virtqueue *vq;
644 struct vhost_vring_addr *addr = &msg->payload.addr;
646 if (dev->mem == NULL)
647 return VH_RESULT_ERR;
649 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
650 vq = dev->virtqueue[msg->payload.addr.index];
653 * Rings addresses should not be interpreted as long as the ring is not
654 * started and enabled
656 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
658 vring_invalidate(dev, vq);
660 if (vq->enabled && (dev->features &
661 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
662 dev = translate_ring_addresses(dev, msg->payload.addr.index);
664 return VH_RESULT_ERR;
673 * The virtio device sends us the available ring last used index.
676 vhost_user_set_vring_base(struct virtio_net **pdev,
677 struct VhostUserMsg *msg)
679 struct virtio_net *dev = *pdev;
680 dev->virtqueue[msg->payload.state.index]->last_used_idx =
681 msg->payload.state.num;
682 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
683 msg->payload.state.num;
689 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
690 uint64_t host_phys_addr, uint64_t size)
692 struct guest_page *page, *last_page;
694 if (dev->nr_guest_pages == dev->max_guest_pages) {
695 dev->max_guest_pages *= 2;
696 dev->guest_pages = realloc(dev->guest_pages,
697 dev->max_guest_pages * sizeof(*page));
698 if (!dev->guest_pages) {
699 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
704 if (dev->nr_guest_pages > 0) {
705 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
706 /* merge if the two pages are continuous */
707 if (host_phys_addr == last_page->host_phys_addr +
709 last_page->size += size;
714 page = &dev->guest_pages[dev->nr_guest_pages++];
715 page->guest_phys_addr = guest_phys_addr;
716 page->host_phys_addr = host_phys_addr;
723 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
726 uint64_t reg_size = reg->size;
727 uint64_t host_user_addr = reg->host_user_addr;
728 uint64_t guest_phys_addr = reg->guest_phys_addr;
729 uint64_t host_phys_addr;
732 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
733 size = page_size - (guest_phys_addr & (page_size - 1));
734 size = RTE_MIN(size, reg_size);
736 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
739 host_user_addr += size;
740 guest_phys_addr += size;
743 while (reg_size > 0) {
744 size = RTE_MIN(reg_size, page_size);
745 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
747 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
751 host_user_addr += size;
752 guest_phys_addr += size;
759 #ifdef RTE_LIBRTE_VHOST_DEBUG
760 /* TODO: enable it only in debug mode? */
762 dump_guest_pages(struct virtio_net *dev)
765 struct guest_page *page;
767 for (i = 0; i < dev->nr_guest_pages; i++) {
768 page = &dev->guest_pages[i];
770 RTE_LOG(INFO, VHOST_CONFIG,
771 "guest physical page region %u\n"
772 "\t guest_phys_addr: %" PRIx64 "\n"
773 "\t host_phys_addr : %" PRIx64 "\n"
774 "\t size : %" PRIx64 "\n",
776 page->guest_phys_addr,
777 page->host_phys_addr,
782 #define dump_guest_pages(dev)
786 vhost_memory_changed(struct VhostUserMemory *new,
787 struct rte_vhost_memory *old)
791 if (new->nregions != old->nregions)
794 for (i = 0; i < new->nregions; ++i) {
795 VhostUserMemoryRegion *new_r = &new->regions[i];
796 struct rte_vhost_mem_region *old_r = &old->regions[i];
798 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
800 if (new_r->memory_size != old_r->size)
802 if (new_r->userspace_addr != old_r->guest_user_addr)
810 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg)
812 struct virtio_net *dev = *pdev;
813 struct VhostUserMemory memory = msg->payload.memory;
814 struct rte_vhost_mem_region *reg;
817 uint64_t mmap_offset;
823 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
824 RTE_LOG(ERR, VHOST_CONFIG,
825 "too many memory regions (%u)\n", memory.nregions);
826 return VH_RESULT_ERR;
829 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
830 RTE_LOG(INFO, VHOST_CONFIG,
831 "(%d) memory regions not changed\n", dev->vid);
833 for (i = 0; i < memory.nregions; i++)
840 free_mem_region(dev);
845 /* Flush IOTLB cache as previous HVAs are now invalid */
846 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
847 for (i = 0; i < dev->nr_vring; i++)
848 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
850 dev->nr_guest_pages = 0;
851 if (!dev->guest_pages) {
852 dev->max_guest_pages = 8;
853 dev->guest_pages = malloc(dev->max_guest_pages *
854 sizeof(struct guest_page));
855 if (dev->guest_pages == NULL) {
856 RTE_LOG(ERR, VHOST_CONFIG,
857 "(%d) failed to allocate memory "
858 "for dev->guest_pages\n",
860 return VH_RESULT_ERR;
864 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
865 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
866 if (dev->mem == NULL) {
867 RTE_LOG(ERR, VHOST_CONFIG,
868 "(%d) failed to allocate memory for dev->mem\n",
870 return VH_RESULT_ERR;
872 dev->mem->nregions = memory.nregions;
874 for (i = 0; i < memory.nregions; i++) {
876 reg = &dev->mem->regions[i];
878 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
879 reg->guest_user_addr = memory.regions[i].userspace_addr;
880 reg->size = memory.regions[i].memory_size;
883 mmap_offset = memory.regions[i].mmap_offset;
885 /* Check for memory_size + mmap_offset overflow */
886 if (mmap_offset >= -reg->size) {
887 RTE_LOG(ERR, VHOST_CONFIG,
888 "mmap_offset (%#"PRIx64") and memory_size "
889 "(%#"PRIx64") overflow\n",
890 mmap_offset, reg->size);
894 mmap_size = reg->size + mmap_offset;
896 /* mmap() without flag of MAP_ANONYMOUS, should be called
897 * with length argument aligned with hugepagesz at older
898 * longterm version Linux, like 2.6.32 and 3.2.72, or
899 * mmap() will fail with EINVAL.
901 * to avoid failure, make sure in caller to keep length
904 alignment = get_blk_size(fd);
905 if (alignment == (uint64_t)-1) {
906 RTE_LOG(ERR, VHOST_CONFIG,
907 "couldn't get hugepage size through fstat\n");
910 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
912 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
913 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
914 MAP_SHARED | populate, fd, 0);
916 if (mmap_addr == MAP_FAILED) {
917 RTE_LOG(ERR, VHOST_CONFIG,
918 "mmap region %u failed.\n", i);
922 reg->mmap_addr = mmap_addr;
923 reg->mmap_size = mmap_size;
924 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
927 if (dev->dequeue_zero_copy)
928 if (add_guest_pages(dev, reg, alignment) < 0) {
929 RTE_LOG(ERR, VHOST_CONFIG,
930 "adding guest pages to region %u failed.\n",
935 RTE_LOG(INFO, VHOST_CONFIG,
936 "guest memory region %u, size: 0x%" PRIx64 "\n"
937 "\t guest physical addr: 0x%" PRIx64 "\n"
938 "\t guest virtual addr: 0x%" PRIx64 "\n"
939 "\t host virtual addr: 0x%" PRIx64 "\n"
940 "\t mmap addr : 0x%" PRIx64 "\n"
941 "\t mmap size : 0x%" PRIx64 "\n"
942 "\t mmap align: 0x%" PRIx64 "\n"
943 "\t mmap off : 0x%" PRIx64 "\n",
945 reg->guest_phys_addr,
946 reg->guest_user_addr,
948 (uint64_t)(uintptr_t)mmap_addr,
954 for (i = 0; i < dev->nr_vring; i++) {
955 struct vhost_virtqueue *vq = dev->virtqueue[i];
957 if (vq->desc || vq->avail || vq->used) {
959 * If the memory table got updated, the ring addresses
960 * need to be translated again as virtual addresses have
963 vring_invalidate(dev, vq);
965 dev = translate_ring_addresses(dev, i);
967 return VH_RESULT_ERR;
973 dump_guest_pages(dev);
978 free_mem_region(dev);
981 return VH_RESULT_ERR;
985 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
992 if (vq_is_packed(dev))
993 rings_ok = !!vq->desc_packed;
995 rings_ok = vq->desc && vq->avail && vq->used;
998 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
999 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1003 virtio_is_ready(struct virtio_net *dev)
1005 struct vhost_virtqueue *vq;
1008 if (dev->nr_vring == 0)
1011 for (i = 0; i < dev->nr_vring; i++) {
1012 vq = dev->virtqueue[i];
1014 if (!vq_is_ready(dev, vq))
1018 RTE_LOG(INFO, VHOST_CONFIG,
1019 "virtio is now ready for processing.\n");
1024 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg)
1026 struct virtio_net *dev = *pdev;
1027 struct vhost_vring_file file;
1028 struct vhost_virtqueue *vq;
1030 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1031 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1032 file.fd = VIRTIO_INVALID_EVENTFD;
1034 file.fd = msg->fds[0];
1035 RTE_LOG(INFO, VHOST_CONFIG,
1036 "vring call idx:%d file:%d\n", file.index, file.fd);
1038 vq = dev->virtqueue[file.index];
1039 if (vq->callfd >= 0)
1042 vq->callfd = file.fd;
1044 return VH_RESULT_OK;
1047 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1048 struct VhostUserMsg *msg)
1050 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1052 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1054 return VH_RESULT_OK;
1058 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg)
1060 struct virtio_net *dev = *pdev;
1061 struct vhost_vring_file file;
1062 struct vhost_virtqueue *vq;
1064 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1065 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1066 file.fd = VIRTIO_INVALID_EVENTFD;
1068 file.fd = msg->fds[0];
1069 RTE_LOG(INFO, VHOST_CONFIG,
1070 "vring kick idx:%d file:%d\n", file.index, file.fd);
1072 /* Interpret ring addresses only when ring is started. */
1073 dev = translate_ring_addresses(dev, file.index);
1075 return VH_RESULT_ERR;
1079 vq = dev->virtqueue[file.index];
1082 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1083 * the ring starts already enabled. Otherwise, it is enabled via
1084 * the SET_VRING_ENABLE message.
1086 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1089 if (vq->kickfd >= 0)
1091 vq->kickfd = file.fd;
1093 return VH_RESULT_OK;
1097 free_zmbufs(struct vhost_virtqueue *vq)
1099 struct zcopy_mbuf *zmbuf, *next;
1101 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1102 zmbuf != NULL; zmbuf = next) {
1103 next = TAILQ_NEXT(zmbuf, next);
1105 rte_pktmbuf_free(zmbuf->mbuf);
1106 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1109 rte_free(vq->zmbufs);
1113 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1116 vhost_user_get_vring_base(struct virtio_net **pdev,
1117 struct VhostUserMsg *msg)
1119 struct virtio_net *dev = *pdev;
1120 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1122 /* We have to stop the queue (virtio) if it is running. */
1123 vhost_destroy_device_notify(dev);
1125 dev->flags &= ~VIRTIO_DEV_READY;
1126 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1128 /* Here we are safe to get the last avail index */
1129 msg->payload.state.num = vq->last_avail_idx;
1131 RTE_LOG(INFO, VHOST_CONFIG,
1132 "vring base idx:%d file:%d\n", msg->payload.state.index,
1133 msg->payload.state.num);
1135 * Based on current qemu vhost-user implementation, this message is
1136 * sent and only sent in vhost_vring_stop.
1137 * TODO: cleanup the vring, it isn't usable since here.
1139 if (vq->kickfd >= 0)
1142 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1144 if (vq->callfd >= 0)
1147 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1149 if (dev->dequeue_zero_copy)
1151 if (vq_is_packed(dev)) {
1152 rte_free(vq->shadow_used_packed);
1153 vq->shadow_used_packed = NULL;
1155 rte_free(vq->shadow_used_split);
1156 vq->shadow_used_split = NULL;
1159 rte_free(vq->batch_copy_elems);
1160 vq->batch_copy_elems = NULL;
1162 msg->size = sizeof(msg->payload.state);
1164 return VH_RESULT_OK;
1168 * when virtio queues are ready to work, qemu will send us to
1169 * enable the virtio queue pair.
1172 vhost_user_set_vring_enable(struct virtio_net **pdev,
1173 struct VhostUserMsg *msg)
1175 struct virtio_net *dev = *pdev;
1176 int enable = (int)msg->payload.state.num;
1177 int index = (int)msg->payload.state.index;
1178 struct rte_vdpa_device *vdpa_dev;
1181 RTE_LOG(INFO, VHOST_CONFIG,
1182 "set queue enable: %d to qp idx: %d\n",
1185 did = dev->vdpa_dev_id;
1186 vdpa_dev = rte_vdpa_get_device(did);
1187 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1188 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1190 if (dev->notify_ops->vring_state_changed)
1191 dev->notify_ops->vring_state_changed(dev->vid,
1194 dev->virtqueue[index]->enabled = enable;
1196 return VH_RESULT_OK;
1200 vhost_user_get_protocol_features(struct virtio_net **pdev,
1201 struct VhostUserMsg *msg)
1203 struct virtio_net *dev = *pdev;
1204 uint64_t features, protocol_features;
1206 rte_vhost_driver_get_features(dev->ifname, &features);
1207 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1210 * REPLY_ACK protocol feature is only mandatory for now
1211 * for IOMMU feature. If IOMMU is explicitly disabled by the
1212 * application, disable also REPLY_ACK feature for older buggy
1213 * Qemu versions (from v2.7.0 to v2.9.0).
1215 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1216 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1218 msg->payload.u64 = protocol_features;
1219 msg->size = sizeof(msg->payload.u64);
1221 return VH_RESULT_OK;
1225 vhost_user_set_protocol_features(struct virtio_net **pdev,
1226 struct VhostUserMsg *msg)
1228 struct virtio_net *dev = *pdev;
1229 uint64_t protocol_features = msg->payload.u64;
1230 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES) {
1231 RTE_LOG(ERR, VHOST_CONFIG,
1232 "(%d) received invalid protocol features.\n",
1234 return VH_RESULT_ERR;
1237 dev->protocol_features = protocol_features;
1239 return VH_RESULT_OK;
1243 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg)
1245 struct virtio_net *dev = *pdev;
1246 int fd = msg->fds[0];
1251 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1252 return VH_RESULT_ERR;
1255 if (msg->size != sizeof(VhostUserLog)) {
1256 RTE_LOG(ERR, VHOST_CONFIG,
1257 "invalid log base msg size: %"PRId32" != %d\n",
1258 msg->size, (int)sizeof(VhostUserLog));
1259 return VH_RESULT_ERR;
1262 size = msg->payload.log.mmap_size;
1263 off = msg->payload.log.mmap_offset;
1265 /* Don't allow mmap_offset to point outside the mmap region */
1267 RTE_LOG(ERR, VHOST_CONFIG,
1268 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1270 return VH_RESULT_ERR;
1273 RTE_LOG(INFO, VHOST_CONFIG,
1274 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1278 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1279 * fail when offset is not page size aligned.
1281 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1283 if (addr == MAP_FAILED) {
1284 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1285 return VH_RESULT_ERR;
1289 * Free previously mapped log memory on occasionally
1290 * multiple VHOST_USER_SET_LOG_BASE.
1292 if (dev->log_addr) {
1293 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1295 dev->log_addr = (uint64_t)(uintptr_t)addr;
1296 dev->log_base = dev->log_addr + off;
1297 dev->log_size = size;
1299 msg->size = sizeof(msg->payload.u64);
1301 return VH_RESULT_OK;
1304 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1305 struct VhostUserMsg *msg)
1308 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1310 return VH_RESULT_OK;
1314 * An rarp packet is constructed and broadcasted to notify switches about
1315 * the new location of the migrated VM, so that packets from outside will
1316 * not be lost after migration.
1318 * However, we don't actually "send" a rarp packet here, instead, we set
1319 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1322 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg)
1324 struct virtio_net *dev = *pdev;
1325 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1326 struct rte_vdpa_device *vdpa_dev;
1329 RTE_LOG(DEBUG, VHOST_CONFIG,
1330 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1331 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1332 memcpy(dev->mac.addr_bytes, mac, 6);
1335 * Set the flag to inject a RARP broadcast packet at
1336 * rte_vhost_dequeue_burst().
1338 * rte_smp_wmb() is for making sure the mac is copied
1339 * before the flag is set.
1342 rte_atomic16_set(&dev->broadcast_rarp, 1);
1343 did = dev->vdpa_dev_id;
1344 vdpa_dev = rte_vdpa_get_device(did);
1345 if (vdpa_dev && vdpa_dev->ops->migration_done)
1346 vdpa_dev->ops->migration_done(dev->vid);
1348 return VH_RESULT_OK;
1352 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg)
1354 struct virtio_net *dev = *pdev;
1355 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1356 msg->payload.u64 > VIRTIO_MAX_MTU) {
1357 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1360 return VH_RESULT_ERR;
1363 dev->mtu = msg->payload.u64;
1365 return VH_RESULT_OK;
1369 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg)
1371 struct virtio_net *dev = *pdev;
1372 int fd = msg->fds[0];
1375 RTE_LOG(ERR, VHOST_CONFIG,
1376 "Invalid file descriptor for slave channel (%d)\n",
1378 return VH_RESULT_ERR;
1381 dev->slave_req_fd = fd;
1383 return VH_RESULT_OK;
1387 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1389 struct vhost_vring_addr *ra;
1390 uint64_t start, end;
1393 end = start + imsg->size;
1395 ra = &vq->ring_addrs;
1396 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1398 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1400 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1407 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1408 struct vhost_iotlb_msg *imsg)
1410 uint64_t istart, iend, vstart, vend;
1412 istart = imsg->iova;
1413 iend = istart + imsg->size - 1;
1415 vstart = (uintptr_t)vq->desc;
1416 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1417 if (vstart <= iend && istart <= vend)
1420 vstart = (uintptr_t)vq->avail;
1421 vend = vstart + sizeof(struct vring_avail);
1422 vend += sizeof(uint16_t) * vq->size - 1;
1423 if (vstart <= iend && istart <= vend)
1426 vstart = (uintptr_t)vq->used;
1427 vend = vstart + sizeof(struct vring_used);
1428 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1429 if (vstart <= iend && istart <= vend)
1436 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1438 struct virtio_net *dev = *pdev;
1439 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1443 switch (imsg->type) {
1444 case VHOST_IOTLB_UPDATE:
1446 vva = qva_to_vva(dev, imsg->uaddr, &len);
1448 return VH_RESULT_ERR;
1450 for (i = 0; i < dev->nr_vring; i++) {
1451 struct vhost_virtqueue *vq = dev->virtqueue[i];
1453 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1456 if (is_vring_iotlb_update(vq, imsg))
1457 *pdev = dev = translate_ring_addresses(dev, i);
1460 case VHOST_IOTLB_INVALIDATE:
1461 for (i = 0; i < dev->nr_vring; i++) {
1462 struct vhost_virtqueue *vq = dev->virtqueue[i];
1464 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1467 if (is_vring_iotlb_invalidate(vq, imsg))
1468 vring_invalidate(dev, vq);
1472 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1474 return VH_RESULT_ERR;
1477 return VH_RESULT_OK;
1480 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
1481 struct VhostUserMsg *msg);
1482 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
1483 [VHOST_USER_NONE] = NULL,
1484 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
1485 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
1486 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
1487 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
1488 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
1489 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
1490 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
1491 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
1492 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
1493 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
1494 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
1495 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
1496 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
1497 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
1498 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
1499 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
1500 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
1501 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
1502 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
1503 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
1504 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
1505 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
1509 /* return bytes# of read on success or negative val on failure. */
1511 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1515 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1516 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1520 if (msg && msg->size) {
1521 if (msg->size > sizeof(msg->payload)) {
1522 RTE_LOG(ERR, VHOST_CONFIG,
1523 "invalid msg size: %d\n", msg->size);
1526 ret = read(sockfd, &msg->payload, msg->size);
1529 if (ret != (int)msg->size) {
1530 RTE_LOG(ERR, VHOST_CONFIG,
1531 "read control message failed\n");
1540 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1545 return send_fd_message(sockfd, (char *)msg,
1546 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1550 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1555 msg->flags &= ~VHOST_USER_VERSION_MASK;
1556 msg->flags &= ~VHOST_USER_NEED_REPLY;
1557 msg->flags |= VHOST_USER_VERSION;
1558 msg->flags |= VHOST_USER_REPLY_MASK;
1560 return send_vhost_message(sockfd, msg, NULL, 0);
1564 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1565 int *fds, int fd_num)
1569 if (msg->flags & VHOST_USER_NEED_REPLY)
1570 rte_spinlock_lock(&dev->slave_req_lock);
1572 ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1573 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1574 rte_spinlock_unlock(&dev->slave_req_lock);
1580 * Allocate a queue pair if it hasn't been allocated yet
1583 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1584 struct VhostUserMsg *msg)
1588 switch (msg->request.master) {
1589 case VHOST_USER_SET_VRING_KICK:
1590 case VHOST_USER_SET_VRING_CALL:
1591 case VHOST_USER_SET_VRING_ERR:
1592 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1594 case VHOST_USER_SET_VRING_NUM:
1595 case VHOST_USER_SET_VRING_BASE:
1596 case VHOST_USER_SET_VRING_ENABLE:
1597 vring_idx = msg->payload.state.index;
1599 case VHOST_USER_SET_VRING_ADDR:
1600 vring_idx = msg->payload.addr.index;
1606 if (vring_idx >= VHOST_MAX_VRING) {
1607 RTE_LOG(ERR, VHOST_CONFIG,
1608 "invalid vring index: %u\n", vring_idx);
1612 if (dev->virtqueue[vring_idx])
1615 return alloc_vring_queue(dev, vring_idx);
1619 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1622 unsigned int vq_num = 0;
1624 while (vq_num < dev->nr_vring) {
1625 struct vhost_virtqueue *vq = dev->virtqueue[i];
1628 rte_spinlock_lock(&vq->access_lock);
1636 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1639 unsigned int vq_num = 0;
1641 while (vq_num < dev->nr_vring) {
1642 struct vhost_virtqueue *vq = dev->virtqueue[i];
1645 rte_spinlock_unlock(&vq->access_lock);
1653 vhost_user_msg_handler(int vid, int fd)
1655 struct virtio_net *dev;
1656 struct VhostUserMsg msg;
1657 struct rte_vdpa_device *vdpa_dev;
1660 int unlock_required = 0;
1661 uint32_t skip_master = 0;
1664 dev = get_device(vid);
1668 if (!dev->notify_ops) {
1669 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1670 if (!dev->notify_ops) {
1671 RTE_LOG(ERR, VHOST_CONFIG,
1672 "failed to get callback ops for driver %s\n",
1678 ret = read_vhost_message(fd, &msg);
1679 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1681 RTE_LOG(ERR, VHOST_CONFIG,
1682 "vhost read message failed\n");
1684 RTE_LOG(INFO, VHOST_CONFIG,
1685 "vhost peer closed\n");
1687 RTE_LOG(ERR, VHOST_CONFIG,
1688 "vhost read incorrect message\n");
1694 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1695 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1696 vhost_message_str[msg.request.master]);
1698 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1699 vhost_message_str[msg.request.master]);
1701 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1703 RTE_LOG(ERR, VHOST_CONFIG,
1704 "failed to alloc queue\n");
1709 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1710 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1711 * and device is destroyed. destroy_device waits for queues to be
1712 * inactive, so it is safe. Otherwise taking the access_lock
1713 * would cause a dead lock.
1715 switch (msg.request.master) {
1716 case VHOST_USER_SET_FEATURES:
1717 case VHOST_USER_SET_PROTOCOL_FEATURES:
1718 case VHOST_USER_SET_OWNER:
1719 case VHOST_USER_SET_MEM_TABLE:
1720 case VHOST_USER_SET_LOG_BASE:
1721 case VHOST_USER_SET_LOG_FD:
1722 case VHOST_USER_SET_VRING_NUM:
1723 case VHOST_USER_SET_VRING_ADDR:
1724 case VHOST_USER_SET_VRING_BASE:
1725 case VHOST_USER_SET_VRING_KICK:
1726 case VHOST_USER_SET_VRING_CALL:
1727 case VHOST_USER_SET_VRING_ERR:
1728 case VHOST_USER_SET_VRING_ENABLE:
1729 case VHOST_USER_SEND_RARP:
1730 case VHOST_USER_NET_SET_MTU:
1731 case VHOST_USER_SET_SLAVE_REQ_FD:
1732 vhost_user_lock_all_queue_pairs(dev);
1733 unlock_required = 1;
1740 if (dev->extern_ops.pre_msg_handle) {
1741 uint32_t need_reply;
1743 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1744 (void *)&msg, &need_reply, &skip_master);
1749 send_vhost_reply(fd, &msg);
1752 goto skip_to_post_handle;
1755 request = msg.request.master;
1756 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
1757 if (!vhost_message_handlers[request])
1758 goto skip_to_post_handle;
1759 ret = vhost_message_handlers[request](&dev, &msg);
1763 RTE_LOG(ERR, VHOST_CONFIG,
1764 "Processing %s failed.\n",
1765 vhost_message_str[request]);
1768 RTE_LOG(DEBUG, VHOST_CONFIG,
1769 "Processing %s succeeded.\n",
1770 vhost_message_str[request]);
1772 case VH_RESULT_REPLY:
1773 RTE_LOG(DEBUG, VHOST_CONFIG,
1774 "Processing %s succeeded and needs reply.\n",
1775 vhost_message_str[request]);
1776 send_vhost_reply(fd, &msg);
1780 RTE_LOG(ERR, VHOST_CONFIG,
1781 "Requested invalid message type %d.\n", request);
1782 ret = VH_RESULT_ERR;
1785 skip_to_post_handle:
1786 if (!ret && dev->extern_ops.post_msg_handle) {
1787 uint32_t need_reply;
1789 ret = (*dev->extern_ops.post_msg_handle)(
1790 dev->vid, (void *)&msg, &need_reply);
1795 send_vhost_reply(fd, &msg);
1799 if (unlock_required)
1800 vhost_user_unlock_all_queue_pairs(dev);
1802 if (msg.flags & VHOST_USER_NEED_REPLY) {
1803 msg.payload.u64 = !!ret;
1804 msg.size = sizeof(msg.payload.u64);
1805 send_vhost_reply(fd, &msg);
1807 RTE_LOG(ERR, VHOST_CONFIG,
1808 "vhost message handling failed.\n");
1812 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1813 dev->flags |= VIRTIO_DEV_READY;
1815 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1816 if (dev->dequeue_zero_copy) {
1817 RTE_LOG(INFO, VHOST_CONFIG,
1818 "dequeue zero copy is enabled\n");
1821 if (dev->notify_ops->new_device(dev->vid) == 0)
1822 dev->flags |= VIRTIO_DEV_RUNNING;
1826 did = dev->vdpa_dev_id;
1827 vdpa_dev = rte_vdpa_get_device(did);
1828 if (vdpa_dev && virtio_is_ready(dev) &&
1829 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1830 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1831 if (vdpa_dev->ops->dev_conf)
1832 vdpa_dev->ops->dev_conf(dev->vid);
1833 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1834 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1835 RTE_LOG(INFO, VHOST_CONFIG,
1836 "(%d) software relay is used for vDPA, performance may be low.\n",
1844 static int process_slave_message_reply(struct virtio_net *dev,
1845 const struct VhostUserMsg *msg)
1847 struct VhostUserMsg msg_reply;
1850 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1853 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1858 if (msg_reply.request.slave != msg->request.slave) {
1859 RTE_LOG(ERR, VHOST_CONFIG,
1860 "Received unexpected msg type (%u), expected %u\n",
1861 msg_reply.request.slave, msg->request.slave);
1866 ret = msg_reply.payload.u64 ? -1 : 0;
1869 rte_spinlock_unlock(&dev->slave_req_lock);
1874 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1877 struct VhostUserMsg msg = {
1878 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1879 .flags = VHOST_USER_VERSION,
1880 .size = sizeof(msg.payload.iotlb),
1884 .type = VHOST_IOTLB_MISS,
1888 ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1890 RTE_LOG(ERR, VHOST_CONFIG,
1891 "Failed to send IOTLB miss message (%d)\n",
1899 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1907 struct VhostUserMsg msg = {
1908 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1909 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1910 .size = sizeof(msg.payload.area),
1912 .u64 = index & VHOST_USER_VRING_IDX_MASK,
1919 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1925 ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1927 RTE_LOG(ERR, VHOST_CONFIG,
1928 "Failed to set host notifier (%d)\n", ret);
1932 return process_slave_message_reply(dev, &msg);
1935 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1937 struct virtio_net *dev;
1938 struct rte_vdpa_device *vdpa_dev;
1939 int vfio_device_fd, did, ret = 0;
1940 uint64_t offset, size;
1943 dev = get_device(vid);
1947 did = dev->vdpa_dev_id;
1951 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1952 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1953 !(dev->protocol_features &
1954 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1955 !(dev->protocol_features &
1956 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1957 !(dev->protocol_features &
1958 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1961 vdpa_dev = rte_vdpa_get_device(did);
1965 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1966 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1968 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1969 if (vfio_device_fd < 0)
1973 for (i = 0; i < dev->nr_vring; i++) {
1974 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1980 if (vhost_user_slave_set_vring_host_notifier(dev, i,
1981 vfio_device_fd, offset, size) < 0) {
1988 for (i = 0; i < dev->nr_vring; i++) {
1989 vhost_user_slave_set_vring_host_notifier(dev, i, -1,