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/ioctl.h>
30 #include <sys/types.h>
32 #include <sys/syscall.h>
34 #ifdef RTE_LIBRTE_VHOST_NUMA
37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
38 #include <linux/userfaultfd.h>
41 #include <rte_common.h>
42 #include <rte_malloc.h>
47 #include "vhost_user.h"
49 #define VIRTIO_MIN_MTU 68
50 #define VIRTIO_MAX_MTU 65535
52 static const char *vhost_message_str[VHOST_USER_MAX] = {
53 [VHOST_USER_NONE] = "VHOST_USER_NONE",
54 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
55 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
56 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
57 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
58 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
59 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
60 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
61 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
62 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
63 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
64 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
65 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
66 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
67 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
68 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
69 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
70 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
71 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
72 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
73 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
74 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
75 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
76 [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
77 [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
78 [VHOST_USER_POSTCOPY_ADVISE] = "VHOST_USER_POSTCOPY_ADVISE",
87 ret = fstat(fd, &stat);
88 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
92 free_mem_region(struct virtio_net *dev)
95 struct rte_vhost_mem_region *reg;
97 if (!dev || !dev->mem)
100 for (i = 0; i < dev->mem->nregions; i++) {
101 reg = &dev->mem->regions[i];
102 if (reg->host_user_addr) {
103 munmap(reg->mmap_addr, reg->mmap_size);
110 vhost_backend_cleanup(struct virtio_net *dev)
113 free_mem_region(dev);
118 free(dev->guest_pages);
119 dev->guest_pages = NULL;
122 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
126 if (dev->slave_req_fd >= 0) {
127 close(dev->slave_req_fd);
128 dev->slave_req_fd = -1;
131 if (dev->postcopy_ufd >= 0) {
132 close(dev->postcopy_ufd);
133 dev->postcopy_ufd = -1;
138 * This function just returns success at the moment unless
139 * the device hasn't been initialised.
142 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
143 struct VhostUserMsg *msg __rte_unused,
144 int main_fd __rte_unused)
150 vhost_user_reset_owner(struct virtio_net **pdev,
151 struct VhostUserMsg *msg __rte_unused,
152 int main_fd __rte_unused)
154 struct virtio_net *dev = *pdev;
155 vhost_destroy_device_notify(dev);
157 cleanup_device(dev, 0);
163 * The features that we support are requested.
166 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
167 int main_fd __rte_unused)
169 struct virtio_net *dev = *pdev;
170 uint64_t features = 0;
172 rte_vhost_driver_get_features(dev->ifname, &features);
174 msg->payload.u64 = features;
175 msg->size = sizeof(msg->payload.u64);
178 return VH_RESULT_REPLY;
182 * The queue number that we support are requested.
185 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
186 int main_fd __rte_unused)
188 struct virtio_net *dev = *pdev;
189 uint32_t queue_num = 0;
191 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
193 msg->payload.u64 = (uint64_t)queue_num;
194 msg->size = sizeof(msg->payload.u64);
197 return VH_RESULT_REPLY;
201 * We receive the negotiated features supported by us and the virtio device.
204 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
205 int main_fd __rte_unused)
207 struct virtio_net *dev = *pdev;
208 uint64_t features = msg->payload.u64;
209 uint64_t vhost_features = 0;
210 struct rte_vdpa_device *vdpa_dev;
213 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
214 if (features & ~vhost_features) {
215 RTE_LOG(ERR, VHOST_CONFIG,
216 "(%d) received invalid negotiated features.\n",
218 return VH_RESULT_ERR;
221 if (dev->flags & VIRTIO_DEV_RUNNING) {
222 if (dev->features == features)
226 * Error out if master tries to change features while device is
227 * in running state. The exception being VHOST_F_LOG_ALL, which
228 * is enabled when the live-migration starts.
230 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
231 RTE_LOG(ERR, VHOST_CONFIG,
232 "(%d) features changed while device is running.\n",
234 return VH_RESULT_ERR;
237 if (dev->notify_ops->features_changed)
238 dev->notify_ops->features_changed(dev->vid, features);
241 dev->features = features;
243 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
244 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
246 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
248 VHOST_LOG_DEBUG(VHOST_CONFIG,
249 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
251 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
252 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
254 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
255 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
257 * Remove all but first queue pair if MQ hasn't been
258 * negotiated. This is safe because the device is not
259 * running at this stage.
261 while (dev->nr_vring > 2) {
262 struct vhost_virtqueue *vq;
264 vq = dev->virtqueue[--dev->nr_vring];
268 dev->virtqueue[dev->nr_vring] = NULL;
274 did = dev->vdpa_dev_id;
275 vdpa_dev = rte_vdpa_get_device(did);
276 if (vdpa_dev && vdpa_dev->ops->set_features)
277 vdpa_dev->ops->set_features(dev->vid);
283 * The virtio device sends us the size of the descriptor ring.
286 vhost_user_set_vring_num(struct virtio_net **pdev,
287 struct VhostUserMsg *msg,
288 int main_fd __rte_unused)
290 struct virtio_net *dev = *pdev;
291 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
293 vq->size = msg->payload.state.num;
295 /* VIRTIO 1.0, 2.4 Virtqueues says:
297 * Queue Size value is always a power of 2. The maximum Queue Size
300 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
301 RTE_LOG(ERR, VHOST_CONFIG,
302 "invalid virtqueue size %u\n", vq->size);
303 return VH_RESULT_ERR;
306 if (dev->dequeue_zero_copy) {
308 vq->last_zmbuf_idx = 0;
309 vq->zmbuf_size = vq->size;
310 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
311 sizeof(struct zcopy_mbuf), 0);
312 if (vq->zmbufs == NULL) {
313 RTE_LOG(WARNING, VHOST_CONFIG,
314 "failed to allocate mem for zero copy; "
315 "zero copy is force disabled\n");
316 dev->dequeue_zero_copy = 0;
318 TAILQ_INIT(&vq->zmbuf_list);
321 if (vq_is_packed(dev)) {
322 vq->shadow_used_packed = rte_malloc(NULL,
324 sizeof(struct vring_used_elem_packed),
325 RTE_CACHE_LINE_SIZE);
326 if (!vq->shadow_used_packed) {
327 RTE_LOG(ERR, VHOST_CONFIG,
328 "failed to allocate memory for shadow used ring.\n");
329 return VH_RESULT_ERR;
333 vq->shadow_used_split = rte_malloc(NULL,
334 vq->size * sizeof(struct vring_used_elem),
335 RTE_CACHE_LINE_SIZE);
336 if (!vq->shadow_used_split) {
337 RTE_LOG(ERR, VHOST_CONFIG,
338 "failed to allocate memory for shadow used ring.\n");
339 return VH_RESULT_ERR;
343 vq->batch_copy_elems = rte_malloc(NULL,
344 vq->size * sizeof(struct batch_copy_elem),
345 RTE_CACHE_LINE_SIZE);
346 if (!vq->batch_copy_elems) {
347 RTE_LOG(ERR, VHOST_CONFIG,
348 "failed to allocate memory for batching copy.\n");
349 return VH_RESULT_ERR;
356 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
357 * same numa node as the memory of vring descriptor.
359 #ifdef RTE_LIBRTE_VHOST_NUMA
360 static struct virtio_net*
361 numa_realloc(struct virtio_net *dev, int index)
363 int oldnode, newnode;
364 struct virtio_net *old_dev;
365 struct vhost_virtqueue *old_vq, *vq;
366 struct zcopy_mbuf *new_zmbuf;
367 struct vring_used_elem *new_shadow_used_split;
368 struct vring_used_elem_packed *new_shadow_used_packed;
369 struct batch_copy_elem *new_batch_copy_elems;
373 vq = old_vq = dev->virtqueue[index];
375 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
376 MPOL_F_NODE | MPOL_F_ADDR);
378 /* check if we need to reallocate vq */
379 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
380 MPOL_F_NODE | MPOL_F_ADDR);
382 RTE_LOG(ERR, VHOST_CONFIG,
383 "Unable to get vq numa information.\n");
386 if (oldnode != newnode) {
387 RTE_LOG(INFO, VHOST_CONFIG,
388 "reallocate vq from %d to %d node\n", oldnode, newnode);
389 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
393 memcpy(vq, old_vq, sizeof(*vq));
394 TAILQ_INIT(&vq->zmbuf_list);
396 if (dev->dequeue_zero_copy) {
397 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
398 sizeof(struct zcopy_mbuf), 0, newnode);
400 rte_free(vq->zmbufs);
401 vq->zmbufs = new_zmbuf;
405 if (vq_is_packed(dev)) {
406 new_shadow_used_packed = rte_malloc_socket(NULL,
408 sizeof(struct vring_used_elem_packed),
411 if (new_shadow_used_packed) {
412 rte_free(vq->shadow_used_packed);
413 vq->shadow_used_packed = new_shadow_used_packed;
416 new_shadow_used_split = rte_malloc_socket(NULL,
418 sizeof(struct vring_used_elem),
421 if (new_shadow_used_split) {
422 rte_free(vq->shadow_used_split);
423 vq->shadow_used_split = new_shadow_used_split;
427 new_batch_copy_elems = rte_malloc_socket(NULL,
428 vq->size * sizeof(struct batch_copy_elem),
431 if (new_batch_copy_elems) {
432 rte_free(vq->batch_copy_elems);
433 vq->batch_copy_elems = new_batch_copy_elems;
439 /* check if we need to reallocate dev */
440 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
441 MPOL_F_NODE | MPOL_F_ADDR);
443 RTE_LOG(ERR, VHOST_CONFIG,
444 "Unable to get dev numa information.\n");
447 if (oldnode != newnode) {
448 RTE_LOG(INFO, VHOST_CONFIG,
449 "reallocate dev from %d to %d node\n",
451 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
457 memcpy(dev, old_dev, sizeof(*dev));
462 dev->virtqueue[index] = vq;
463 vhost_devices[dev->vid] = dev;
466 vhost_user_iotlb_init(dev, index);
471 static struct virtio_net*
472 numa_realloc(struct virtio_net *dev, int index __rte_unused)
478 /* Converts QEMU virtual address to Vhost virtual address. */
480 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
482 struct rte_vhost_mem_region *r;
485 /* Find the region where the address lives. */
486 for (i = 0; i < dev->mem->nregions; i++) {
487 r = &dev->mem->regions[i];
489 if (qva >= r->guest_user_addr &&
490 qva < r->guest_user_addr + r->size) {
492 if (unlikely(*len > r->guest_user_addr + r->size - qva))
493 *len = r->guest_user_addr + r->size - qva;
495 return qva - r->guest_user_addr +
506 * Converts ring address to Vhost virtual address.
507 * If IOMMU is enabled, the ring address is a guest IO virtual address,
508 * else it is a QEMU virtual address.
511 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
512 uint64_t ra, uint64_t *size)
514 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
517 vva = vhost_user_iotlb_cache_find(vq, ra,
518 size, VHOST_ACCESS_RW);
520 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
525 return qva_to_vva(dev, ra, size);
528 static struct virtio_net *
529 translate_ring_addresses(struct virtio_net *dev, int vq_index)
531 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
532 struct vhost_vring_addr *addr = &vq->ring_addrs;
535 if (vq_is_packed(dev)) {
536 len = sizeof(struct vring_packed_desc) * vq->size;
537 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
538 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
539 vq->log_guest_addr = 0;
540 if (vq->desc_packed == NULL ||
541 len != sizeof(struct vring_packed_desc) *
543 RTE_LOG(DEBUG, VHOST_CONFIG,
544 "(%d) failed to map desc_packed ring.\n",
549 dev = numa_realloc(dev, vq_index);
550 vq = dev->virtqueue[vq_index];
551 addr = &vq->ring_addrs;
553 len = sizeof(struct vring_packed_desc_event);
554 vq->driver_event = (struct vring_packed_desc_event *)
555 (uintptr_t)ring_addr_to_vva(dev,
556 vq, addr->avail_user_addr, &len);
557 if (vq->driver_event == NULL ||
558 len != sizeof(struct vring_packed_desc_event)) {
559 RTE_LOG(DEBUG, VHOST_CONFIG,
560 "(%d) failed to find driver area address.\n",
565 len = sizeof(struct vring_packed_desc_event);
566 vq->device_event = (struct vring_packed_desc_event *)
567 (uintptr_t)ring_addr_to_vva(dev,
568 vq, addr->used_user_addr, &len);
569 if (vq->device_event == NULL ||
570 len != sizeof(struct vring_packed_desc_event)) {
571 RTE_LOG(DEBUG, VHOST_CONFIG,
572 "(%d) failed to find device area address.\n",
580 /* The addresses are converted from QEMU virtual to Vhost virtual. */
581 if (vq->desc && vq->avail && vq->used)
584 len = sizeof(struct vring_desc) * vq->size;
585 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
586 vq, addr->desc_user_addr, &len);
587 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
588 RTE_LOG(DEBUG, VHOST_CONFIG,
589 "(%d) failed to map desc ring.\n",
594 dev = numa_realloc(dev, vq_index);
595 vq = dev->virtqueue[vq_index];
596 addr = &vq->ring_addrs;
598 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
599 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
600 vq, addr->avail_user_addr, &len);
601 if (vq->avail == 0 ||
602 len != sizeof(struct vring_avail) +
603 sizeof(uint16_t) * vq->size) {
604 RTE_LOG(DEBUG, VHOST_CONFIG,
605 "(%d) failed to map avail ring.\n",
610 len = sizeof(struct vring_used) +
611 sizeof(struct vring_used_elem) * vq->size;
612 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
613 vq, addr->used_user_addr, &len);
614 if (vq->used == 0 || len != sizeof(struct vring_used) +
615 sizeof(struct vring_used_elem) * vq->size) {
616 RTE_LOG(DEBUG, VHOST_CONFIG,
617 "(%d) failed to map used ring.\n",
622 if (vq->last_used_idx != vq->used->idx) {
623 RTE_LOG(WARNING, VHOST_CONFIG,
624 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
625 "some packets maybe resent for Tx and dropped for Rx\n",
626 vq->last_used_idx, vq->used->idx);
627 vq->last_used_idx = vq->used->idx;
628 vq->last_avail_idx = vq->used->idx;
631 vq->log_guest_addr = addr->log_guest_addr;
633 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
635 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
636 dev->vid, vq->avail);
637 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
639 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
640 dev->vid, vq->log_guest_addr);
646 * The virtio device sends us the desc, used and avail ring addresses.
647 * This function then converts these to our address space.
650 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
651 int main_fd __rte_unused)
653 struct virtio_net *dev = *pdev;
654 struct vhost_virtqueue *vq;
655 struct vhost_vring_addr *addr = &msg->payload.addr;
657 if (dev->mem == NULL)
658 return VH_RESULT_ERR;
660 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
661 vq = dev->virtqueue[msg->payload.addr.index];
664 * Rings addresses should not be interpreted as long as the ring is not
665 * started and enabled
667 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
669 vring_invalidate(dev, vq);
671 if (vq->enabled && (dev->features &
672 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
673 dev = translate_ring_addresses(dev, msg->payload.addr.index);
675 return VH_RESULT_ERR;
684 * The virtio device sends us the available ring last used index.
687 vhost_user_set_vring_base(struct virtio_net **pdev,
688 struct VhostUserMsg *msg,
689 int main_fd __rte_unused)
691 struct virtio_net *dev = *pdev;
692 dev->virtqueue[msg->payload.state.index]->last_used_idx =
693 msg->payload.state.num;
694 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
695 msg->payload.state.num;
701 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
702 uint64_t host_phys_addr, uint64_t size)
704 struct guest_page *page, *last_page;
706 if (dev->nr_guest_pages == dev->max_guest_pages) {
707 dev->max_guest_pages *= 2;
708 dev->guest_pages = realloc(dev->guest_pages,
709 dev->max_guest_pages * sizeof(*page));
710 if (!dev->guest_pages) {
711 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
716 if (dev->nr_guest_pages > 0) {
717 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
718 /* merge if the two pages are continuous */
719 if (host_phys_addr == last_page->host_phys_addr +
721 last_page->size += size;
726 page = &dev->guest_pages[dev->nr_guest_pages++];
727 page->guest_phys_addr = guest_phys_addr;
728 page->host_phys_addr = host_phys_addr;
735 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
738 uint64_t reg_size = reg->size;
739 uint64_t host_user_addr = reg->host_user_addr;
740 uint64_t guest_phys_addr = reg->guest_phys_addr;
741 uint64_t host_phys_addr;
744 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
745 size = page_size - (guest_phys_addr & (page_size - 1));
746 size = RTE_MIN(size, reg_size);
748 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
751 host_user_addr += size;
752 guest_phys_addr += size;
755 while (reg_size > 0) {
756 size = RTE_MIN(reg_size, page_size);
757 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
759 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
763 host_user_addr += size;
764 guest_phys_addr += size;
771 #ifdef RTE_LIBRTE_VHOST_DEBUG
772 /* TODO: enable it only in debug mode? */
774 dump_guest_pages(struct virtio_net *dev)
777 struct guest_page *page;
779 for (i = 0; i < dev->nr_guest_pages; i++) {
780 page = &dev->guest_pages[i];
782 RTE_LOG(INFO, VHOST_CONFIG,
783 "guest physical page region %u\n"
784 "\t guest_phys_addr: %" PRIx64 "\n"
785 "\t host_phys_addr : %" PRIx64 "\n"
786 "\t size : %" PRIx64 "\n",
788 page->guest_phys_addr,
789 page->host_phys_addr,
794 #define dump_guest_pages(dev)
798 vhost_memory_changed(struct VhostUserMemory *new,
799 struct rte_vhost_memory *old)
803 if (new->nregions != old->nregions)
806 for (i = 0; i < new->nregions; ++i) {
807 VhostUserMemoryRegion *new_r = &new->regions[i];
808 struct rte_vhost_mem_region *old_r = &old->regions[i];
810 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
812 if (new_r->memory_size != old_r->size)
814 if (new_r->userspace_addr != old_r->guest_user_addr)
822 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
823 int main_fd __rte_unused)
825 struct virtio_net *dev = *pdev;
826 struct VhostUserMemory memory = msg->payload.memory;
827 struct rte_vhost_mem_region *reg;
830 uint64_t mmap_offset;
836 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
837 RTE_LOG(ERR, VHOST_CONFIG,
838 "too many memory regions (%u)\n", memory.nregions);
839 return VH_RESULT_ERR;
842 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
843 RTE_LOG(INFO, VHOST_CONFIG,
844 "(%d) memory regions not changed\n", dev->vid);
846 for (i = 0; i < memory.nregions; i++)
853 free_mem_region(dev);
858 /* Flush IOTLB cache as previous HVAs are now invalid */
859 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
860 for (i = 0; i < dev->nr_vring; i++)
861 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
863 dev->nr_guest_pages = 0;
864 if (!dev->guest_pages) {
865 dev->max_guest_pages = 8;
866 dev->guest_pages = malloc(dev->max_guest_pages *
867 sizeof(struct guest_page));
868 if (dev->guest_pages == NULL) {
869 RTE_LOG(ERR, VHOST_CONFIG,
870 "(%d) failed to allocate memory "
871 "for dev->guest_pages\n",
873 return VH_RESULT_ERR;
877 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
878 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
879 if (dev->mem == NULL) {
880 RTE_LOG(ERR, VHOST_CONFIG,
881 "(%d) failed to allocate memory for dev->mem\n",
883 return VH_RESULT_ERR;
885 dev->mem->nregions = memory.nregions;
887 for (i = 0; i < memory.nregions; i++) {
889 reg = &dev->mem->regions[i];
891 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
892 reg->guest_user_addr = memory.regions[i].userspace_addr;
893 reg->size = memory.regions[i].memory_size;
896 mmap_offset = memory.regions[i].mmap_offset;
898 /* Check for memory_size + mmap_offset overflow */
899 if (mmap_offset >= -reg->size) {
900 RTE_LOG(ERR, VHOST_CONFIG,
901 "mmap_offset (%#"PRIx64") and memory_size "
902 "(%#"PRIx64") overflow\n",
903 mmap_offset, reg->size);
907 mmap_size = reg->size + mmap_offset;
909 /* mmap() without flag of MAP_ANONYMOUS, should be called
910 * with length argument aligned with hugepagesz at older
911 * longterm version Linux, like 2.6.32 and 3.2.72, or
912 * mmap() will fail with EINVAL.
914 * to avoid failure, make sure in caller to keep length
917 alignment = get_blk_size(fd);
918 if (alignment == (uint64_t)-1) {
919 RTE_LOG(ERR, VHOST_CONFIG,
920 "couldn't get hugepage size through fstat\n");
923 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
925 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
926 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
927 MAP_SHARED | populate, fd, 0);
929 if (mmap_addr == MAP_FAILED) {
930 RTE_LOG(ERR, VHOST_CONFIG,
931 "mmap region %u failed.\n", i);
935 reg->mmap_addr = mmap_addr;
936 reg->mmap_size = mmap_size;
937 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
940 if (dev->dequeue_zero_copy)
941 if (add_guest_pages(dev, reg, alignment) < 0) {
942 RTE_LOG(ERR, VHOST_CONFIG,
943 "adding guest pages to region %u failed.\n",
948 RTE_LOG(INFO, VHOST_CONFIG,
949 "guest memory region %u, size: 0x%" PRIx64 "\n"
950 "\t guest physical addr: 0x%" PRIx64 "\n"
951 "\t guest virtual addr: 0x%" PRIx64 "\n"
952 "\t host virtual addr: 0x%" PRIx64 "\n"
953 "\t mmap addr : 0x%" PRIx64 "\n"
954 "\t mmap size : 0x%" PRIx64 "\n"
955 "\t mmap align: 0x%" PRIx64 "\n"
956 "\t mmap off : 0x%" PRIx64 "\n",
958 reg->guest_phys_addr,
959 reg->guest_user_addr,
961 (uint64_t)(uintptr_t)mmap_addr,
967 for (i = 0; i < dev->nr_vring; i++) {
968 struct vhost_virtqueue *vq = dev->virtqueue[i];
970 if (vq->desc || vq->avail || vq->used) {
972 * If the memory table got updated, the ring addresses
973 * need to be translated again as virtual addresses have
976 vring_invalidate(dev, vq);
978 dev = translate_ring_addresses(dev, i);
988 dump_guest_pages(dev);
993 free_mem_region(dev);
996 return VH_RESULT_ERR;
1000 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1007 if (vq_is_packed(dev))
1008 rings_ok = !!vq->desc_packed;
1010 rings_ok = vq->desc && vq->avail && vq->used;
1013 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1014 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1018 virtio_is_ready(struct virtio_net *dev)
1020 struct vhost_virtqueue *vq;
1023 if (dev->nr_vring == 0)
1026 for (i = 0; i < dev->nr_vring; i++) {
1027 vq = dev->virtqueue[i];
1029 if (!vq_is_ready(dev, vq))
1033 RTE_LOG(INFO, VHOST_CONFIG,
1034 "virtio is now ready for processing.\n");
1039 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1040 int main_fd __rte_unused)
1042 struct virtio_net *dev = *pdev;
1043 struct vhost_vring_file file;
1044 struct vhost_virtqueue *vq;
1046 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1047 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1048 file.fd = VIRTIO_INVALID_EVENTFD;
1050 file.fd = msg->fds[0];
1051 RTE_LOG(INFO, VHOST_CONFIG,
1052 "vring call idx:%d file:%d\n", file.index, file.fd);
1054 vq = dev->virtqueue[file.index];
1055 if (vq->callfd >= 0)
1058 vq->callfd = file.fd;
1060 return VH_RESULT_OK;
1063 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1064 struct VhostUserMsg *msg,
1065 int main_fd __rte_unused)
1067 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1069 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1071 return VH_RESULT_OK;
1075 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1076 int main_fd __rte_unused)
1078 struct virtio_net *dev = *pdev;
1079 struct vhost_vring_file file;
1080 struct vhost_virtqueue *vq;
1082 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1083 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1084 file.fd = VIRTIO_INVALID_EVENTFD;
1086 file.fd = msg->fds[0];
1087 RTE_LOG(INFO, VHOST_CONFIG,
1088 "vring kick idx:%d file:%d\n", file.index, file.fd);
1090 /* Interpret ring addresses only when ring is started. */
1091 dev = translate_ring_addresses(dev, file.index);
1093 return VH_RESULT_ERR;
1097 vq = dev->virtqueue[file.index];
1100 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1101 * the ring starts already enabled. Otherwise, it is enabled via
1102 * the SET_VRING_ENABLE message.
1104 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1107 if (vq->kickfd >= 0)
1109 vq->kickfd = file.fd;
1111 return VH_RESULT_OK;
1115 free_zmbufs(struct vhost_virtqueue *vq)
1117 struct zcopy_mbuf *zmbuf, *next;
1119 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1120 zmbuf != NULL; zmbuf = next) {
1121 next = TAILQ_NEXT(zmbuf, next);
1123 rte_pktmbuf_free(zmbuf->mbuf);
1124 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1127 rte_free(vq->zmbufs);
1131 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1134 vhost_user_get_vring_base(struct virtio_net **pdev,
1135 struct VhostUserMsg *msg,
1136 int main_fd __rte_unused)
1138 struct virtio_net *dev = *pdev;
1139 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1141 /* We have to stop the queue (virtio) if it is running. */
1142 vhost_destroy_device_notify(dev);
1144 dev->flags &= ~VIRTIO_DEV_READY;
1145 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1147 /* Here we are safe to get the last avail index */
1148 msg->payload.state.num = vq->last_avail_idx;
1150 RTE_LOG(INFO, VHOST_CONFIG,
1151 "vring base idx:%d file:%d\n", msg->payload.state.index,
1152 msg->payload.state.num);
1154 * Based on current qemu vhost-user implementation, this message is
1155 * sent and only sent in vhost_vring_stop.
1156 * TODO: cleanup the vring, it isn't usable since here.
1158 if (vq->kickfd >= 0)
1161 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1163 if (vq->callfd >= 0)
1166 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1168 if (dev->dequeue_zero_copy)
1170 if (vq_is_packed(dev)) {
1171 rte_free(vq->shadow_used_packed);
1172 vq->shadow_used_packed = NULL;
1174 rte_free(vq->shadow_used_split);
1175 vq->shadow_used_split = NULL;
1178 rte_free(vq->batch_copy_elems);
1179 vq->batch_copy_elems = NULL;
1181 msg->size = sizeof(msg->payload.state);
1184 return VH_RESULT_REPLY;
1188 * when virtio queues are ready to work, qemu will send us to
1189 * enable the virtio queue pair.
1192 vhost_user_set_vring_enable(struct virtio_net **pdev,
1193 struct VhostUserMsg *msg,
1194 int main_fd __rte_unused)
1196 struct virtio_net *dev = *pdev;
1197 int enable = (int)msg->payload.state.num;
1198 int index = (int)msg->payload.state.index;
1199 struct rte_vdpa_device *vdpa_dev;
1202 RTE_LOG(INFO, VHOST_CONFIG,
1203 "set queue enable: %d to qp idx: %d\n",
1206 did = dev->vdpa_dev_id;
1207 vdpa_dev = rte_vdpa_get_device(did);
1208 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1209 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1211 if (dev->notify_ops->vring_state_changed)
1212 dev->notify_ops->vring_state_changed(dev->vid,
1215 dev->virtqueue[index]->enabled = enable;
1217 return VH_RESULT_OK;
1221 vhost_user_get_protocol_features(struct virtio_net **pdev,
1222 struct VhostUserMsg *msg,
1223 int main_fd __rte_unused)
1225 struct virtio_net *dev = *pdev;
1226 uint64_t features, protocol_features;
1228 rte_vhost_driver_get_features(dev->ifname, &features);
1229 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1232 * REPLY_ACK protocol feature is only mandatory for now
1233 * for IOMMU feature. If IOMMU is explicitly disabled by the
1234 * application, disable also REPLY_ACK feature for older buggy
1235 * Qemu versions (from v2.7.0 to v2.9.0).
1237 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1238 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1240 msg->payload.u64 = protocol_features;
1241 msg->size = sizeof(msg->payload.u64);
1244 return VH_RESULT_REPLY;
1248 vhost_user_set_protocol_features(struct virtio_net **pdev,
1249 struct VhostUserMsg *msg,
1250 int main_fd __rte_unused)
1252 struct virtio_net *dev = *pdev;
1253 uint64_t protocol_features = msg->payload.u64;
1254 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES) {
1255 RTE_LOG(ERR, VHOST_CONFIG,
1256 "(%d) received invalid protocol features.\n",
1258 return VH_RESULT_ERR;
1261 dev->protocol_features = protocol_features;
1263 return VH_RESULT_OK;
1267 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
1268 int main_fd __rte_unused)
1270 struct virtio_net *dev = *pdev;
1271 int fd = msg->fds[0];
1276 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1277 return VH_RESULT_ERR;
1280 if (msg->size != sizeof(VhostUserLog)) {
1281 RTE_LOG(ERR, VHOST_CONFIG,
1282 "invalid log base msg size: %"PRId32" != %d\n",
1283 msg->size, (int)sizeof(VhostUserLog));
1284 return VH_RESULT_ERR;
1287 size = msg->payload.log.mmap_size;
1288 off = msg->payload.log.mmap_offset;
1290 /* Don't allow mmap_offset to point outside the mmap region */
1292 RTE_LOG(ERR, VHOST_CONFIG,
1293 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1295 return VH_RESULT_ERR;
1298 RTE_LOG(INFO, VHOST_CONFIG,
1299 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1303 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1304 * fail when offset is not page size aligned.
1306 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1308 if (addr == MAP_FAILED) {
1309 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1310 return VH_RESULT_ERR;
1314 * Free previously mapped log memory on occasionally
1315 * multiple VHOST_USER_SET_LOG_BASE.
1317 if (dev->log_addr) {
1318 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1320 dev->log_addr = (uint64_t)(uintptr_t)addr;
1321 dev->log_base = dev->log_addr + off;
1322 dev->log_size = size;
1325 * The spec is not clear about it (yet), but QEMU doesn't expect
1326 * any payload in the reply.
1331 return VH_RESULT_REPLY;
1334 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1335 struct VhostUserMsg *msg,
1336 int main_fd __rte_unused)
1339 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1341 return VH_RESULT_OK;
1345 * An rarp packet is constructed and broadcasted to notify switches about
1346 * the new location of the migrated VM, so that packets from outside will
1347 * not be lost after migration.
1349 * However, we don't actually "send" a rarp packet here, instead, we set
1350 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1353 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
1354 int main_fd __rte_unused)
1356 struct virtio_net *dev = *pdev;
1357 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1358 struct rte_vdpa_device *vdpa_dev;
1361 RTE_LOG(DEBUG, VHOST_CONFIG,
1362 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1363 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1364 memcpy(dev->mac.addr_bytes, mac, 6);
1367 * Set the flag to inject a RARP broadcast packet at
1368 * rte_vhost_dequeue_burst().
1370 * rte_smp_wmb() is for making sure the mac is copied
1371 * before the flag is set.
1374 rte_atomic16_set(&dev->broadcast_rarp, 1);
1375 did = dev->vdpa_dev_id;
1376 vdpa_dev = rte_vdpa_get_device(did);
1377 if (vdpa_dev && vdpa_dev->ops->migration_done)
1378 vdpa_dev->ops->migration_done(dev->vid);
1380 return VH_RESULT_OK;
1384 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
1385 int main_fd __rte_unused)
1387 struct virtio_net *dev = *pdev;
1388 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1389 msg->payload.u64 > VIRTIO_MAX_MTU) {
1390 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1393 return VH_RESULT_ERR;
1396 dev->mtu = msg->payload.u64;
1398 return VH_RESULT_OK;
1402 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
1403 int main_fd __rte_unused)
1405 struct virtio_net *dev = *pdev;
1406 int fd = msg->fds[0];
1409 RTE_LOG(ERR, VHOST_CONFIG,
1410 "Invalid file descriptor for slave channel (%d)\n",
1412 return VH_RESULT_ERR;
1415 dev->slave_req_fd = fd;
1417 return VH_RESULT_OK;
1421 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1423 struct vhost_vring_addr *ra;
1424 uint64_t start, end;
1427 end = start + imsg->size;
1429 ra = &vq->ring_addrs;
1430 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1432 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1434 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1441 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1442 struct vhost_iotlb_msg *imsg)
1444 uint64_t istart, iend, vstart, vend;
1446 istart = imsg->iova;
1447 iend = istart + imsg->size - 1;
1449 vstart = (uintptr_t)vq->desc;
1450 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1451 if (vstart <= iend && istart <= vend)
1454 vstart = (uintptr_t)vq->avail;
1455 vend = vstart + sizeof(struct vring_avail);
1456 vend += sizeof(uint16_t) * vq->size - 1;
1457 if (vstart <= iend && istart <= vend)
1460 vstart = (uintptr_t)vq->used;
1461 vend = vstart + sizeof(struct vring_used);
1462 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1463 if (vstart <= iend && istart <= vend)
1470 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
1471 int main_fd __rte_unused)
1473 struct virtio_net *dev = *pdev;
1474 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1478 switch (imsg->type) {
1479 case VHOST_IOTLB_UPDATE:
1481 vva = qva_to_vva(dev, imsg->uaddr, &len);
1483 return VH_RESULT_ERR;
1485 for (i = 0; i < dev->nr_vring; i++) {
1486 struct vhost_virtqueue *vq = dev->virtqueue[i];
1488 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1491 if (is_vring_iotlb_update(vq, imsg))
1492 *pdev = dev = translate_ring_addresses(dev, i);
1495 case VHOST_IOTLB_INVALIDATE:
1496 for (i = 0; i < dev->nr_vring; i++) {
1497 struct vhost_virtqueue *vq = dev->virtqueue[i];
1499 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1502 if (is_vring_iotlb_invalidate(vq, imsg))
1503 vring_invalidate(dev, vq);
1507 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1509 return VH_RESULT_ERR;
1512 return VH_RESULT_OK;
1516 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
1517 struct VhostUserMsg *msg,
1518 int main_fd __rte_unused)
1520 struct virtio_net *dev = *pdev;
1521 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1522 struct uffdio_api api_struct;
1524 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
1526 if (dev->postcopy_ufd == -1) {
1527 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
1529 return VH_RESULT_ERR;
1531 api_struct.api = UFFD_API;
1532 api_struct.features = 0;
1533 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
1534 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
1536 close(dev->postcopy_ufd);
1537 dev->postcopy_ufd = -1;
1538 return VH_RESULT_ERR;
1540 msg->fds[0] = dev->postcopy_ufd;
1543 return VH_RESULT_REPLY;
1545 dev->postcopy_ufd = -1;
1548 return VH_RESULT_ERR;
1552 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
1553 struct VhostUserMsg *msg,
1555 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
1556 [VHOST_USER_NONE] = NULL,
1557 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
1558 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
1559 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
1560 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
1561 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
1562 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
1563 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
1564 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
1565 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
1566 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
1567 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
1568 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
1569 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
1570 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
1571 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
1572 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
1573 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
1574 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
1575 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
1576 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
1577 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
1578 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
1579 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
1583 /* return bytes# of read on success or negative val on failure. */
1585 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1589 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1590 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
1594 if (msg && msg->size) {
1595 if (msg->size > sizeof(msg->payload)) {
1596 RTE_LOG(ERR, VHOST_CONFIG,
1597 "invalid msg size: %d\n", msg->size);
1600 ret = read(sockfd, &msg->payload, msg->size);
1603 if (ret != (int)msg->size) {
1604 RTE_LOG(ERR, VHOST_CONFIG,
1605 "read control message failed\n");
1614 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1619 return send_fd_message(sockfd, (char *)msg,
1620 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
1624 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1629 msg->flags &= ~VHOST_USER_VERSION_MASK;
1630 msg->flags &= ~VHOST_USER_NEED_REPLY;
1631 msg->flags |= VHOST_USER_VERSION;
1632 msg->flags |= VHOST_USER_REPLY_MASK;
1634 return send_vhost_message(sockfd, msg);
1638 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
1642 if (msg->flags & VHOST_USER_NEED_REPLY)
1643 rte_spinlock_lock(&dev->slave_req_lock);
1645 ret = send_vhost_message(dev->slave_req_fd, msg);
1646 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1647 rte_spinlock_unlock(&dev->slave_req_lock);
1653 * Allocate a queue pair if it hasn't been allocated yet
1656 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1657 struct VhostUserMsg *msg)
1661 switch (msg->request.master) {
1662 case VHOST_USER_SET_VRING_KICK:
1663 case VHOST_USER_SET_VRING_CALL:
1664 case VHOST_USER_SET_VRING_ERR:
1665 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1667 case VHOST_USER_SET_VRING_NUM:
1668 case VHOST_USER_SET_VRING_BASE:
1669 case VHOST_USER_SET_VRING_ENABLE:
1670 vring_idx = msg->payload.state.index;
1672 case VHOST_USER_SET_VRING_ADDR:
1673 vring_idx = msg->payload.addr.index;
1679 if (vring_idx >= VHOST_MAX_VRING) {
1680 RTE_LOG(ERR, VHOST_CONFIG,
1681 "invalid vring index: %u\n", vring_idx);
1685 if (dev->virtqueue[vring_idx])
1688 return alloc_vring_queue(dev, vring_idx);
1692 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1695 unsigned int vq_num = 0;
1697 while (vq_num < dev->nr_vring) {
1698 struct vhost_virtqueue *vq = dev->virtqueue[i];
1701 rte_spinlock_lock(&vq->access_lock);
1709 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1712 unsigned int vq_num = 0;
1714 while (vq_num < dev->nr_vring) {
1715 struct vhost_virtqueue *vq = dev->virtqueue[i];
1718 rte_spinlock_unlock(&vq->access_lock);
1726 vhost_user_msg_handler(int vid, int fd)
1728 struct virtio_net *dev;
1729 struct VhostUserMsg msg;
1730 struct rte_vdpa_device *vdpa_dev;
1733 int unlock_required = 0;
1734 uint32_t skip_master = 0;
1737 dev = get_device(vid);
1741 if (!dev->notify_ops) {
1742 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1743 if (!dev->notify_ops) {
1744 RTE_LOG(ERR, VHOST_CONFIG,
1745 "failed to get callback ops for driver %s\n",
1751 ret = read_vhost_message(fd, &msg);
1752 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1754 RTE_LOG(ERR, VHOST_CONFIG,
1755 "vhost read message failed\n");
1757 RTE_LOG(INFO, VHOST_CONFIG,
1758 "vhost peer closed\n");
1760 RTE_LOG(ERR, VHOST_CONFIG,
1761 "vhost read incorrect message\n");
1767 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1768 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1769 vhost_message_str[msg.request.master]);
1771 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1772 vhost_message_str[msg.request.master]);
1774 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1776 RTE_LOG(ERR, VHOST_CONFIG,
1777 "failed to alloc queue\n");
1782 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1783 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1784 * and device is destroyed. destroy_device waits for queues to be
1785 * inactive, so it is safe. Otherwise taking the access_lock
1786 * would cause a dead lock.
1788 switch (msg.request.master) {
1789 case VHOST_USER_SET_FEATURES:
1790 case VHOST_USER_SET_PROTOCOL_FEATURES:
1791 case VHOST_USER_SET_OWNER:
1792 case VHOST_USER_SET_MEM_TABLE:
1793 case VHOST_USER_SET_LOG_BASE:
1794 case VHOST_USER_SET_LOG_FD:
1795 case VHOST_USER_SET_VRING_NUM:
1796 case VHOST_USER_SET_VRING_ADDR:
1797 case VHOST_USER_SET_VRING_BASE:
1798 case VHOST_USER_SET_VRING_KICK:
1799 case VHOST_USER_SET_VRING_CALL:
1800 case VHOST_USER_SET_VRING_ERR:
1801 case VHOST_USER_SET_VRING_ENABLE:
1802 case VHOST_USER_SEND_RARP:
1803 case VHOST_USER_NET_SET_MTU:
1804 case VHOST_USER_SET_SLAVE_REQ_FD:
1805 vhost_user_lock_all_queue_pairs(dev);
1806 unlock_required = 1;
1813 if (dev->extern_ops.pre_msg_handle) {
1814 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1815 (void *)&msg, &skip_master);
1816 if (ret == VH_RESULT_ERR)
1818 else if (ret == VH_RESULT_REPLY)
1819 send_vhost_reply(fd, &msg);
1822 goto skip_to_post_handle;
1825 request = msg.request.master;
1826 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
1827 if (!vhost_message_handlers[request])
1828 goto skip_to_post_handle;
1829 ret = vhost_message_handlers[request](&dev, &msg, fd);
1833 RTE_LOG(ERR, VHOST_CONFIG,
1834 "Processing %s failed.\n",
1835 vhost_message_str[request]);
1838 RTE_LOG(DEBUG, VHOST_CONFIG,
1839 "Processing %s succeeded.\n",
1840 vhost_message_str[request]);
1842 case VH_RESULT_REPLY:
1843 RTE_LOG(DEBUG, VHOST_CONFIG,
1844 "Processing %s succeeded and needs reply.\n",
1845 vhost_message_str[request]);
1846 send_vhost_reply(fd, &msg);
1850 RTE_LOG(ERR, VHOST_CONFIG,
1851 "Requested invalid message type %d.\n", request);
1852 ret = VH_RESULT_ERR;
1855 skip_to_post_handle:
1856 if (ret != VH_RESULT_ERR && dev->extern_ops.post_msg_handle) {
1857 ret = (*dev->extern_ops.post_msg_handle)(
1858 dev->vid, (void *)&msg);
1859 if (ret == VH_RESULT_ERR)
1861 else if (ret == VH_RESULT_REPLY)
1862 send_vhost_reply(fd, &msg);
1866 if (unlock_required)
1867 vhost_user_unlock_all_queue_pairs(dev);
1870 * If the request required a reply that was already sent,
1871 * this optional reply-ack won't be sent as the
1872 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
1874 if (msg.flags & VHOST_USER_NEED_REPLY) {
1875 msg.payload.u64 = ret == VH_RESULT_ERR;
1876 msg.size = sizeof(msg.payload.u64);
1878 send_vhost_reply(fd, &msg);
1879 } else if (ret == VH_RESULT_ERR) {
1880 RTE_LOG(ERR, VHOST_CONFIG,
1881 "vhost message handling failed.\n");
1885 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1886 dev->flags |= VIRTIO_DEV_READY;
1888 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1889 if (dev->dequeue_zero_copy) {
1890 RTE_LOG(INFO, VHOST_CONFIG,
1891 "dequeue zero copy is enabled\n");
1894 if (dev->notify_ops->new_device(dev->vid) == 0)
1895 dev->flags |= VIRTIO_DEV_RUNNING;
1899 did = dev->vdpa_dev_id;
1900 vdpa_dev = rte_vdpa_get_device(did);
1901 if (vdpa_dev && virtio_is_ready(dev) &&
1902 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1903 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1904 if (vdpa_dev->ops->dev_conf)
1905 vdpa_dev->ops->dev_conf(dev->vid);
1906 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1907 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1908 RTE_LOG(INFO, VHOST_CONFIG,
1909 "(%d) software relay is used for vDPA, performance may be low.\n",
1917 static int process_slave_message_reply(struct virtio_net *dev,
1918 const struct VhostUserMsg *msg)
1920 struct VhostUserMsg msg_reply;
1923 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1926 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1931 if (msg_reply.request.slave != msg->request.slave) {
1932 RTE_LOG(ERR, VHOST_CONFIG,
1933 "Received unexpected msg type (%u), expected %u\n",
1934 msg_reply.request.slave, msg->request.slave);
1939 ret = msg_reply.payload.u64 ? -1 : 0;
1942 rte_spinlock_unlock(&dev->slave_req_lock);
1947 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1950 struct VhostUserMsg msg = {
1951 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1952 .flags = VHOST_USER_VERSION,
1953 .size = sizeof(msg.payload.iotlb),
1957 .type = VHOST_IOTLB_MISS,
1961 ret = send_vhost_message(dev->slave_req_fd, &msg);
1963 RTE_LOG(ERR, VHOST_CONFIG,
1964 "Failed to send IOTLB miss message (%d)\n",
1972 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1978 struct VhostUserMsg msg = {
1979 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1980 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1981 .size = sizeof(msg.payload.area),
1983 .u64 = index & VHOST_USER_VRING_IDX_MASK,
1990 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1996 ret = send_vhost_slave_message(dev, &msg);
1998 RTE_LOG(ERR, VHOST_CONFIG,
1999 "Failed to set host notifier (%d)\n", ret);
2003 return process_slave_message_reply(dev, &msg);
2006 int vhost_user_host_notifier_ctrl(int vid, bool enable)
2008 struct virtio_net *dev;
2009 struct rte_vdpa_device *vdpa_dev;
2010 int vfio_device_fd, did, ret = 0;
2011 uint64_t offset, size;
2014 dev = get_device(vid);
2018 did = dev->vdpa_dev_id;
2022 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2023 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2024 !(dev->protocol_features &
2025 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2026 !(dev->protocol_features &
2027 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2028 !(dev->protocol_features &
2029 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2032 vdpa_dev = rte_vdpa_get_device(did);
2036 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2037 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2039 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2040 if (vfio_device_fd < 0)
2044 for (i = 0; i < dev->nr_vring; i++) {
2045 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2051 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2052 vfio_device_fd, offset, size) < 0) {
2059 for (i = 0; i < dev->nr_vring; i++) {
2060 vhost_user_slave_set_vring_host_notifier(dev, i, -1,