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",
79 [VHOST_USER_POSTCOPY_LISTEN] = "VHOST_USER_POSTCOPY_LISTEN",
80 [VHOST_USER_POSTCOPY_END] = "VHOST_USER_POSTCOPY_END",
83 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
84 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
92 ret = fstat(fd, &stat);
93 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
97 free_mem_region(struct virtio_net *dev)
100 struct rte_vhost_mem_region *reg;
102 if (!dev || !dev->mem)
105 for (i = 0; i < dev->mem->nregions; i++) {
106 reg = &dev->mem->regions[i];
107 if (reg->host_user_addr) {
108 munmap(reg->mmap_addr, reg->mmap_size);
115 vhost_backend_cleanup(struct virtio_net *dev)
118 free_mem_region(dev);
123 free(dev->guest_pages);
124 dev->guest_pages = NULL;
127 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
131 if (dev->slave_req_fd >= 0) {
132 close(dev->slave_req_fd);
133 dev->slave_req_fd = -1;
136 if (dev->postcopy_ufd >= 0) {
137 close(dev->postcopy_ufd);
138 dev->postcopy_ufd = -1;
141 dev->postcopy_listening = 0;
145 * This function just returns success at the moment unless
146 * the device hasn't been initialised.
149 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
150 struct VhostUserMsg *msg __rte_unused,
151 int main_fd __rte_unused)
157 vhost_user_reset_owner(struct virtio_net **pdev,
158 struct VhostUserMsg *msg __rte_unused,
159 int main_fd __rte_unused)
161 struct virtio_net *dev = *pdev;
162 vhost_destroy_device_notify(dev);
164 cleanup_device(dev, 0);
170 * The features that we support are requested.
173 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
174 int main_fd __rte_unused)
176 struct virtio_net *dev = *pdev;
177 uint64_t features = 0;
179 rte_vhost_driver_get_features(dev->ifname, &features);
181 msg->payload.u64 = features;
182 msg->size = sizeof(msg->payload.u64);
185 return VH_RESULT_REPLY;
189 * The queue number that we support are requested.
192 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
193 int main_fd __rte_unused)
195 struct virtio_net *dev = *pdev;
196 uint32_t queue_num = 0;
198 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
200 msg->payload.u64 = (uint64_t)queue_num;
201 msg->size = sizeof(msg->payload.u64);
204 return VH_RESULT_REPLY;
208 * We receive the negotiated features supported by us and the virtio device.
211 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
212 int main_fd __rte_unused)
214 struct virtio_net *dev = *pdev;
215 uint64_t features = msg->payload.u64;
216 uint64_t vhost_features = 0;
217 struct rte_vdpa_device *vdpa_dev;
220 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
221 if (features & ~vhost_features) {
222 RTE_LOG(ERR, VHOST_CONFIG,
223 "(%d) received invalid negotiated features.\n",
225 return VH_RESULT_ERR;
228 if (dev->flags & VIRTIO_DEV_RUNNING) {
229 if (dev->features == features)
233 * Error out if master tries to change features while device is
234 * in running state. The exception being VHOST_F_LOG_ALL, which
235 * is enabled when the live-migration starts.
237 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
238 RTE_LOG(ERR, VHOST_CONFIG,
239 "(%d) features changed while device is running.\n",
241 return VH_RESULT_ERR;
244 if (dev->notify_ops->features_changed)
245 dev->notify_ops->features_changed(dev->vid, features);
248 dev->features = features;
250 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
251 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
253 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
255 VHOST_LOG_DEBUG(VHOST_CONFIG,
256 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
258 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
259 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
261 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
262 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
264 * Remove all but first queue pair if MQ hasn't been
265 * negotiated. This is safe because the device is not
266 * running at this stage.
268 while (dev->nr_vring > 2) {
269 struct vhost_virtqueue *vq;
271 vq = dev->virtqueue[--dev->nr_vring];
275 dev->virtqueue[dev->nr_vring] = NULL;
281 did = dev->vdpa_dev_id;
282 vdpa_dev = rte_vdpa_get_device(did);
283 if (vdpa_dev && vdpa_dev->ops->set_features)
284 vdpa_dev->ops->set_features(dev->vid);
290 * The virtio device sends us the size of the descriptor ring.
293 vhost_user_set_vring_num(struct virtio_net **pdev,
294 struct VhostUserMsg *msg,
295 int main_fd __rte_unused)
297 struct virtio_net *dev = *pdev;
298 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
300 vq->size = msg->payload.state.num;
302 /* VIRTIO 1.0, 2.4 Virtqueues says:
304 * Queue Size value is always a power of 2. The maximum Queue Size
307 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
308 RTE_LOG(ERR, VHOST_CONFIG,
309 "invalid virtqueue size %u\n", vq->size);
310 return VH_RESULT_ERR;
313 if (dev->dequeue_zero_copy) {
315 vq->last_zmbuf_idx = 0;
316 vq->zmbuf_size = vq->size;
317 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
318 sizeof(struct zcopy_mbuf), 0);
319 if (vq->zmbufs == NULL) {
320 RTE_LOG(WARNING, VHOST_CONFIG,
321 "failed to allocate mem for zero copy; "
322 "zero copy is force disabled\n");
323 dev->dequeue_zero_copy = 0;
325 TAILQ_INIT(&vq->zmbuf_list);
328 if (vq_is_packed(dev)) {
329 vq->shadow_used_packed = rte_malloc(NULL,
331 sizeof(struct vring_used_elem_packed),
332 RTE_CACHE_LINE_SIZE);
333 if (!vq->shadow_used_packed) {
334 RTE_LOG(ERR, VHOST_CONFIG,
335 "failed to allocate memory for shadow used ring.\n");
336 return VH_RESULT_ERR;
340 vq->shadow_used_split = rte_malloc(NULL,
341 vq->size * sizeof(struct vring_used_elem),
342 RTE_CACHE_LINE_SIZE);
343 if (!vq->shadow_used_split) {
344 RTE_LOG(ERR, VHOST_CONFIG,
345 "failed to allocate memory for shadow used ring.\n");
346 return VH_RESULT_ERR;
350 vq->batch_copy_elems = rte_malloc(NULL,
351 vq->size * sizeof(struct batch_copy_elem),
352 RTE_CACHE_LINE_SIZE);
353 if (!vq->batch_copy_elems) {
354 RTE_LOG(ERR, VHOST_CONFIG,
355 "failed to allocate memory for batching copy.\n");
356 return VH_RESULT_ERR;
363 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
364 * same numa node as the memory of vring descriptor.
366 #ifdef RTE_LIBRTE_VHOST_NUMA
367 static struct virtio_net*
368 numa_realloc(struct virtio_net *dev, int index)
370 int oldnode, newnode;
371 struct virtio_net *old_dev;
372 struct vhost_virtqueue *old_vq, *vq;
373 struct zcopy_mbuf *new_zmbuf;
374 struct vring_used_elem *new_shadow_used_split;
375 struct vring_used_elem_packed *new_shadow_used_packed;
376 struct batch_copy_elem *new_batch_copy_elems;
380 vq = old_vq = dev->virtqueue[index];
382 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
383 MPOL_F_NODE | MPOL_F_ADDR);
385 /* check if we need to reallocate vq */
386 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
387 MPOL_F_NODE | MPOL_F_ADDR);
389 RTE_LOG(ERR, VHOST_CONFIG,
390 "Unable to get vq numa information.\n");
393 if (oldnode != newnode) {
394 RTE_LOG(INFO, VHOST_CONFIG,
395 "reallocate vq from %d to %d node\n", oldnode, newnode);
396 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
400 memcpy(vq, old_vq, sizeof(*vq));
401 TAILQ_INIT(&vq->zmbuf_list);
403 if (dev->dequeue_zero_copy) {
404 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
405 sizeof(struct zcopy_mbuf), 0, newnode);
407 rte_free(vq->zmbufs);
408 vq->zmbufs = new_zmbuf;
412 if (vq_is_packed(dev)) {
413 new_shadow_used_packed = rte_malloc_socket(NULL,
415 sizeof(struct vring_used_elem_packed),
418 if (new_shadow_used_packed) {
419 rte_free(vq->shadow_used_packed);
420 vq->shadow_used_packed = new_shadow_used_packed;
423 new_shadow_used_split = rte_malloc_socket(NULL,
425 sizeof(struct vring_used_elem),
428 if (new_shadow_used_split) {
429 rte_free(vq->shadow_used_split);
430 vq->shadow_used_split = new_shadow_used_split;
434 new_batch_copy_elems = rte_malloc_socket(NULL,
435 vq->size * sizeof(struct batch_copy_elem),
438 if (new_batch_copy_elems) {
439 rte_free(vq->batch_copy_elems);
440 vq->batch_copy_elems = new_batch_copy_elems;
446 /* check if we need to reallocate dev */
447 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
448 MPOL_F_NODE | MPOL_F_ADDR);
450 RTE_LOG(ERR, VHOST_CONFIG,
451 "Unable to get dev numa information.\n");
454 if (oldnode != newnode) {
455 RTE_LOG(INFO, VHOST_CONFIG,
456 "reallocate dev from %d to %d node\n",
458 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
464 memcpy(dev, old_dev, sizeof(*dev));
469 dev->virtqueue[index] = vq;
470 vhost_devices[dev->vid] = dev;
473 vhost_user_iotlb_init(dev, index);
478 static struct virtio_net*
479 numa_realloc(struct virtio_net *dev, int index __rte_unused)
485 /* Converts QEMU virtual address to Vhost virtual address. */
487 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
489 struct rte_vhost_mem_region *r;
492 if (unlikely(!dev || !dev->mem))
495 /* Find the region where the address lives. */
496 for (i = 0; i < dev->mem->nregions; i++) {
497 r = &dev->mem->regions[i];
499 if (qva >= r->guest_user_addr &&
500 qva < r->guest_user_addr + r->size) {
502 if (unlikely(*len > r->guest_user_addr + r->size - qva))
503 *len = r->guest_user_addr + r->size - qva;
505 return qva - r->guest_user_addr +
517 * Converts ring address to Vhost virtual address.
518 * If IOMMU is enabled, the ring address is a guest IO virtual address,
519 * else it is a QEMU virtual address.
522 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
523 uint64_t ra, uint64_t *size)
525 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
528 vva = vhost_user_iotlb_cache_find(vq, ra,
529 size, VHOST_ACCESS_RW);
531 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
536 return qva_to_vva(dev, ra, size);
539 static struct virtio_net *
540 translate_ring_addresses(struct virtio_net *dev, int vq_index)
542 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
543 struct vhost_vring_addr *addr = &vq->ring_addrs;
544 uint64_t len, expected_len;
546 if (vq_is_packed(dev)) {
547 len = sizeof(struct vring_packed_desc) * vq->size;
548 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
549 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
550 vq->log_guest_addr = 0;
551 if (vq->desc_packed == NULL ||
552 len != sizeof(struct vring_packed_desc) *
554 RTE_LOG(DEBUG, VHOST_CONFIG,
555 "(%d) failed to map desc_packed ring.\n",
560 dev = numa_realloc(dev, vq_index);
561 vq = dev->virtqueue[vq_index];
562 addr = &vq->ring_addrs;
564 len = sizeof(struct vring_packed_desc_event);
565 vq->driver_event = (struct vring_packed_desc_event *)
566 (uintptr_t)ring_addr_to_vva(dev,
567 vq, addr->avail_user_addr, &len);
568 if (vq->driver_event == NULL ||
569 len != sizeof(struct vring_packed_desc_event)) {
570 RTE_LOG(DEBUG, VHOST_CONFIG,
571 "(%d) failed to find driver area address.\n",
576 len = sizeof(struct vring_packed_desc_event);
577 vq->device_event = (struct vring_packed_desc_event *)
578 (uintptr_t)ring_addr_to_vva(dev,
579 vq, addr->used_user_addr, &len);
580 if (vq->device_event == NULL ||
581 len != sizeof(struct vring_packed_desc_event)) {
582 RTE_LOG(DEBUG, VHOST_CONFIG,
583 "(%d) failed to find device area address.\n",
591 /* The addresses are converted from QEMU virtual to Vhost virtual. */
592 if (vq->desc && vq->avail && vq->used)
595 len = sizeof(struct vring_desc) * vq->size;
596 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
597 vq, addr->desc_user_addr, &len);
598 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
599 RTE_LOG(DEBUG, VHOST_CONFIG,
600 "(%d) failed to map desc ring.\n",
605 dev = numa_realloc(dev, vq_index);
606 vq = dev->virtqueue[vq_index];
607 addr = &vq->ring_addrs;
609 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
610 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
611 len += sizeof(uint16_t);
613 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
614 vq, addr->avail_user_addr, &len);
615 if (vq->avail == 0 || len != expected_len) {
616 RTE_LOG(DEBUG, VHOST_CONFIG,
617 "(%d) failed to map avail ring.\n",
622 len = sizeof(struct vring_used) +
623 sizeof(struct vring_used_elem) * vq->size;
624 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
625 len += sizeof(uint16_t);
627 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
628 vq, addr->used_user_addr, &len);
629 if (vq->used == 0 || len != expected_len) {
630 RTE_LOG(DEBUG, VHOST_CONFIG,
631 "(%d) failed to map used ring.\n",
636 if (vq->last_used_idx != vq->used->idx) {
637 RTE_LOG(WARNING, VHOST_CONFIG,
638 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
639 "some packets maybe resent for Tx and dropped for Rx\n",
640 vq->last_used_idx, vq->used->idx);
641 vq->last_used_idx = vq->used->idx;
642 vq->last_avail_idx = vq->used->idx;
645 vq->log_guest_addr = addr->log_guest_addr;
647 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
649 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
650 dev->vid, vq->avail);
651 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
653 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
654 dev->vid, vq->log_guest_addr);
660 * The virtio device sends us the desc, used and avail ring addresses.
661 * This function then converts these to our address space.
664 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
665 int main_fd __rte_unused)
667 struct virtio_net *dev = *pdev;
668 struct vhost_virtqueue *vq;
669 struct vhost_vring_addr *addr = &msg->payload.addr;
671 if (dev->mem == NULL)
672 return VH_RESULT_ERR;
674 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
675 vq = dev->virtqueue[msg->payload.addr.index];
678 * Rings addresses should not be interpreted as long as the ring is not
679 * started and enabled
681 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
683 vring_invalidate(dev, vq);
685 if (vq->enabled && (dev->features &
686 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
687 dev = translate_ring_addresses(dev, msg->payload.addr.index);
689 return VH_RESULT_ERR;
698 * The virtio device sends us the available ring last used index.
701 vhost_user_set_vring_base(struct virtio_net **pdev,
702 struct VhostUserMsg *msg,
703 int main_fd __rte_unused)
705 struct virtio_net *dev = *pdev;
706 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
707 uint64_t val = msg->payload.state.num;
709 if (vq_is_packed(dev)) {
711 * Bit[0:14]: avail index
712 * Bit[15]: avail wrap counter
714 vq->last_avail_idx = val & 0x7fff;
715 vq->avail_wrap_counter = !!(val & (0x1 << 15));
717 * Set used index to same value as available one, as
718 * their values should be the same since ring processing
719 * was stopped at get time.
721 vq->last_used_idx = vq->last_avail_idx;
722 vq->used_wrap_counter = vq->avail_wrap_counter;
724 vq->last_used_idx = msg->payload.state.num;
725 vq->last_avail_idx = msg->payload.state.num;
732 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
733 uint64_t host_phys_addr, uint64_t size)
735 struct guest_page *page, *last_page;
736 struct guest_page *old_pages;
738 if (dev->nr_guest_pages == dev->max_guest_pages) {
739 dev->max_guest_pages *= 2;
740 old_pages = dev->guest_pages;
741 dev->guest_pages = realloc(dev->guest_pages,
742 dev->max_guest_pages * sizeof(*page));
743 if (!dev->guest_pages) {
744 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
750 if (dev->nr_guest_pages > 0) {
751 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
752 /* merge if the two pages are continuous */
753 if (host_phys_addr == last_page->host_phys_addr +
755 last_page->size += size;
760 page = &dev->guest_pages[dev->nr_guest_pages++];
761 page->guest_phys_addr = guest_phys_addr;
762 page->host_phys_addr = host_phys_addr;
769 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
772 uint64_t reg_size = reg->size;
773 uint64_t host_user_addr = reg->host_user_addr;
774 uint64_t guest_phys_addr = reg->guest_phys_addr;
775 uint64_t host_phys_addr;
778 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
779 size = page_size - (guest_phys_addr & (page_size - 1));
780 size = RTE_MIN(size, reg_size);
782 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
785 host_user_addr += size;
786 guest_phys_addr += size;
789 while (reg_size > 0) {
790 size = RTE_MIN(reg_size, page_size);
791 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
793 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
797 host_user_addr += size;
798 guest_phys_addr += size;
805 #ifdef RTE_LIBRTE_VHOST_DEBUG
806 /* TODO: enable it only in debug mode? */
808 dump_guest_pages(struct virtio_net *dev)
811 struct guest_page *page;
813 for (i = 0; i < dev->nr_guest_pages; i++) {
814 page = &dev->guest_pages[i];
816 RTE_LOG(INFO, VHOST_CONFIG,
817 "guest physical page region %u\n"
818 "\t guest_phys_addr: %" PRIx64 "\n"
819 "\t host_phys_addr : %" PRIx64 "\n"
820 "\t size : %" PRIx64 "\n",
822 page->guest_phys_addr,
823 page->host_phys_addr,
828 #define dump_guest_pages(dev)
832 vhost_memory_changed(struct VhostUserMemory *new,
833 struct rte_vhost_memory *old)
837 if (new->nregions != old->nregions)
840 for (i = 0; i < new->nregions; ++i) {
841 VhostUserMemoryRegion *new_r = &new->regions[i];
842 struct rte_vhost_mem_region *old_r = &old->regions[i];
844 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
846 if (new_r->memory_size != old_r->size)
848 if (new_r->userspace_addr != old_r->guest_user_addr)
856 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
859 struct virtio_net *dev = *pdev;
860 struct VhostUserMemory *memory = &msg->payload.memory;
861 struct rte_vhost_mem_region *reg;
864 uint64_t mmap_offset;
870 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
871 RTE_LOG(ERR, VHOST_CONFIG,
872 "too many memory regions (%u)\n", memory->nregions);
873 return VH_RESULT_ERR;
876 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
877 RTE_LOG(INFO, VHOST_CONFIG,
878 "(%d) memory regions not changed\n", dev->vid);
880 for (i = 0; i < memory->nregions; i++)
887 free_mem_region(dev);
892 /* Flush IOTLB cache as previous HVAs are now invalid */
893 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
894 for (i = 0; i < dev->nr_vring; i++)
895 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
897 dev->nr_guest_pages = 0;
898 if (!dev->guest_pages) {
899 dev->max_guest_pages = 8;
900 dev->guest_pages = malloc(dev->max_guest_pages *
901 sizeof(struct guest_page));
902 if (dev->guest_pages == NULL) {
903 RTE_LOG(ERR, VHOST_CONFIG,
904 "(%d) failed to allocate memory "
905 "for dev->guest_pages\n",
907 return VH_RESULT_ERR;
911 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
912 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
913 if (dev->mem == NULL) {
914 RTE_LOG(ERR, VHOST_CONFIG,
915 "(%d) failed to allocate memory for dev->mem\n",
917 return VH_RESULT_ERR;
919 dev->mem->nregions = memory->nregions;
921 for (i = 0; i < memory->nregions; i++) {
923 reg = &dev->mem->regions[i];
925 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
926 reg->guest_user_addr = memory->regions[i].userspace_addr;
927 reg->size = memory->regions[i].memory_size;
930 mmap_offset = memory->regions[i].mmap_offset;
932 /* Check for memory_size + mmap_offset overflow */
933 if (mmap_offset >= -reg->size) {
934 RTE_LOG(ERR, VHOST_CONFIG,
935 "mmap_offset (%#"PRIx64") and memory_size "
936 "(%#"PRIx64") overflow\n",
937 mmap_offset, reg->size);
941 mmap_size = reg->size + mmap_offset;
943 /* mmap() without flag of MAP_ANONYMOUS, should be called
944 * with length argument aligned with hugepagesz at older
945 * longterm version Linux, like 2.6.32 and 3.2.72, or
946 * mmap() will fail with EINVAL.
948 * to avoid failure, make sure in caller to keep length
951 alignment = get_blk_size(fd);
952 if (alignment == (uint64_t)-1) {
953 RTE_LOG(ERR, VHOST_CONFIG,
954 "couldn't get hugepage size through fstat\n");
957 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
959 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
960 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
961 MAP_SHARED | populate, fd, 0);
963 if (mmap_addr == MAP_FAILED) {
964 RTE_LOG(ERR, VHOST_CONFIG,
965 "mmap region %u failed.\n", i);
969 reg->mmap_addr = mmap_addr;
970 reg->mmap_size = mmap_size;
971 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
974 if (dev->dequeue_zero_copy)
975 if (add_guest_pages(dev, reg, alignment) < 0) {
976 RTE_LOG(ERR, VHOST_CONFIG,
977 "adding guest pages to region %u failed.\n",
982 RTE_LOG(INFO, VHOST_CONFIG,
983 "guest memory region %u, size: 0x%" PRIx64 "\n"
984 "\t guest physical addr: 0x%" PRIx64 "\n"
985 "\t guest virtual addr: 0x%" PRIx64 "\n"
986 "\t host virtual addr: 0x%" PRIx64 "\n"
987 "\t mmap addr : 0x%" PRIx64 "\n"
988 "\t mmap size : 0x%" PRIx64 "\n"
989 "\t mmap align: 0x%" PRIx64 "\n"
990 "\t mmap off : 0x%" PRIx64 "\n",
992 reg->guest_phys_addr,
993 reg->guest_user_addr,
995 (uint64_t)(uintptr_t)mmap_addr,
1000 if (dev->postcopy_listening) {
1002 * We haven't a better way right now than sharing
1003 * DPDK's virtual address with Qemu, so that Qemu can
1004 * retrieve the region offset when handling userfaults.
1006 memory->regions[i].userspace_addr =
1007 reg->host_user_addr;
1010 if (dev->postcopy_listening) {
1011 /* Send the addresses back to qemu */
1013 send_vhost_reply(main_fd, msg);
1015 /* Wait for qemu to acknolwedge it's got the addresses
1016 * we've got to wait before we're allowed to generate faults.
1018 VhostUserMsg ack_msg;
1019 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1020 RTE_LOG(ERR, VHOST_CONFIG,
1021 "Failed to read qemu ack on postcopy set-mem-table\n");
1024 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1025 RTE_LOG(ERR, VHOST_CONFIG,
1026 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1027 ack_msg.request.master);
1031 /* Now userfault register and we can use the memory */
1032 for (i = 0; i < memory->nregions; i++) {
1033 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1034 reg = &dev->mem->regions[i];
1035 struct uffdio_register reg_struct;
1038 * Let's register all the mmap'ed area to ensure
1039 * alignment on page boundary.
1041 reg_struct.range.start =
1042 (uint64_t)(uintptr_t)reg->mmap_addr;
1043 reg_struct.range.len = reg->mmap_size;
1044 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1046 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1048 RTE_LOG(ERR, VHOST_CONFIG,
1049 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1050 i, dev->postcopy_ufd,
1054 RTE_LOG(INFO, VHOST_CONFIG,
1055 "\t userfaultfd registered for range : %llx - %llx\n",
1056 reg_struct.range.start,
1057 reg_struct.range.start +
1058 reg_struct.range.len - 1);
1065 for (i = 0; i < dev->nr_vring; i++) {
1066 struct vhost_virtqueue *vq = dev->virtqueue[i];
1068 if (vq->desc || vq->avail || vq->used) {
1070 * If the memory table got updated, the ring addresses
1071 * need to be translated again as virtual addresses have
1074 vring_invalidate(dev, vq);
1076 dev = translate_ring_addresses(dev, i);
1086 dump_guest_pages(dev);
1088 return VH_RESULT_OK;
1091 free_mem_region(dev);
1094 return VH_RESULT_ERR;
1098 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1105 if (vq_is_packed(dev))
1106 rings_ok = !!vq->desc_packed;
1108 rings_ok = vq->desc && vq->avail && vq->used;
1111 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1112 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1116 virtio_is_ready(struct virtio_net *dev)
1118 struct vhost_virtqueue *vq;
1121 if (dev->nr_vring == 0)
1124 for (i = 0; i < dev->nr_vring; i++) {
1125 vq = dev->virtqueue[i];
1127 if (!vq_is_ready(dev, vq))
1131 RTE_LOG(INFO, VHOST_CONFIG,
1132 "virtio is now ready for processing.\n");
1137 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1138 int main_fd __rte_unused)
1140 struct virtio_net *dev = *pdev;
1141 struct vhost_vring_file file;
1142 struct vhost_virtqueue *vq;
1144 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1145 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1146 file.fd = VIRTIO_INVALID_EVENTFD;
1148 file.fd = msg->fds[0];
1149 RTE_LOG(INFO, VHOST_CONFIG,
1150 "vring call idx:%d file:%d\n", file.index, file.fd);
1152 vq = dev->virtqueue[file.index];
1153 if (vq->callfd >= 0)
1156 vq->callfd = file.fd;
1158 return VH_RESULT_OK;
1161 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1162 struct VhostUserMsg *msg,
1163 int main_fd __rte_unused)
1165 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1167 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1169 return VH_RESULT_OK;
1173 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1174 int main_fd __rte_unused)
1176 struct virtio_net *dev = *pdev;
1177 struct vhost_vring_file file;
1178 struct vhost_virtqueue *vq;
1180 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1181 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1182 file.fd = VIRTIO_INVALID_EVENTFD;
1184 file.fd = msg->fds[0];
1185 RTE_LOG(INFO, VHOST_CONFIG,
1186 "vring kick idx:%d file:%d\n", file.index, file.fd);
1188 /* Interpret ring addresses only when ring is started. */
1189 dev = translate_ring_addresses(dev, file.index);
1191 return VH_RESULT_ERR;
1195 vq = dev->virtqueue[file.index];
1198 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1199 * the ring starts already enabled. Otherwise, it is enabled via
1200 * the SET_VRING_ENABLE message.
1202 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1205 if (vq->kickfd >= 0)
1207 vq->kickfd = file.fd;
1209 return VH_RESULT_OK;
1213 free_zmbufs(struct vhost_virtqueue *vq)
1215 struct zcopy_mbuf *zmbuf, *next;
1217 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1218 zmbuf != NULL; zmbuf = next) {
1219 next = TAILQ_NEXT(zmbuf, next);
1221 rte_pktmbuf_free(zmbuf->mbuf);
1222 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1225 rte_free(vq->zmbufs);
1229 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1232 vhost_user_get_vring_base(struct virtio_net **pdev,
1233 struct VhostUserMsg *msg,
1234 int main_fd __rte_unused)
1236 struct virtio_net *dev = *pdev;
1237 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1240 /* We have to stop the queue (virtio) if it is running. */
1241 vhost_destroy_device_notify(dev);
1243 dev->flags &= ~VIRTIO_DEV_READY;
1244 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1246 /* Here we are safe to get the indexes */
1247 if (vq_is_packed(dev)) {
1249 * Bit[0:14]: avail index
1250 * Bit[15]: avail wrap counter
1252 val = vq->last_avail_idx & 0x7fff;
1253 val |= vq->avail_wrap_counter << 15;
1254 msg->payload.state.num = val;
1256 msg->payload.state.num = vq->last_avail_idx;
1259 RTE_LOG(INFO, VHOST_CONFIG,
1260 "vring base idx:%d file:%d\n", msg->payload.state.index,
1261 msg->payload.state.num);
1263 * Based on current qemu vhost-user implementation, this message is
1264 * sent and only sent in vhost_vring_stop.
1265 * TODO: cleanup the vring, it isn't usable since here.
1267 if (vq->kickfd >= 0)
1270 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1272 if (vq->callfd >= 0)
1275 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1277 if (dev->dequeue_zero_copy)
1279 if (vq_is_packed(dev)) {
1280 rte_free(vq->shadow_used_packed);
1281 vq->shadow_used_packed = NULL;
1283 rte_free(vq->shadow_used_split);
1284 vq->shadow_used_split = NULL;
1287 rte_free(vq->batch_copy_elems);
1288 vq->batch_copy_elems = NULL;
1290 msg->size = sizeof(msg->payload.state);
1293 return VH_RESULT_REPLY;
1297 * when virtio queues are ready to work, qemu will send us to
1298 * enable the virtio queue pair.
1301 vhost_user_set_vring_enable(struct virtio_net **pdev,
1302 struct VhostUserMsg *msg,
1303 int main_fd __rte_unused)
1305 struct virtio_net *dev = *pdev;
1306 int enable = (int)msg->payload.state.num;
1307 int index = (int)msg->payload.state.index;
1308 struct rte_vdpa_device *vdpa_dev;
1311 RTE_LOG(INFO, VHOST_CONFIG,
1312 "set queue enable: %d to qp idx: %d\n",
1315 did = dev->vdpa_dev_id;
1316 vdpa_dev = rte_vdpa_get_device(did);
1317 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1318 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1320 if (dev->notify_ops->vring_state_changed)
1321 dev->notify_ops->vring_state_changed(dev->vid,
1324 dev->virtqueue[index]->enabled = enable;
1326 return VH_RESULT_OK;
1330 vhost_user_get_protocol_features(struct virtio_net **pdev,
1331 struct VhostUserMsg *msg,
1332 int main_fd __rte_unused)
1334 struct virtio_net *dev = *pdev;
1335 uint64_t features, protocol_features;
1337 rte_vhost_driver_get_features(dev->ifname, &features);
1338 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1341 * REPLY_ACK protocol feature is only mandatory for now
1342 * for IOMMU feature. If IOMMU is explicitly disabled by the
1343 * application, disable also REPLY_ACK feature for older buggy
1344 * Qemu versions (from v2.7.0 to v2.9.0).
1346 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1347 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1349 msg->payload.u64 = protocol_features;
1350 msg->size = sizeof(msg->payload.u64);
1353 return VH_RESULT_REPLY;
1357 vhost_user_set_protocol_features(struct virtio_net **pdev,
1358 struct VhostUserMsg *msg,
1359 int main_fd __rte_unused)
1361 struct virtio_net *dev = *pdev;
1362 uint64_t protocol_features = msg->payload.u64;
1363 uint64_t slave_protocol_features = 0;
1365 rte_vhost_driver_get_protocol_features(dev->ifname,
1366 &slave_protocol_features);
1367 if (protocol_features & ~slave_protocol_features) {
1368 RTE_LOG(ERR, VHOST_CONFIG,
1369 "(%d) received invalid protocol features.\n",
1371 return VH_RESULT_ERR;
1374 dev->protocol_features = protocol_features;
1376 return VH_RESULT_OK;
1380 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
1381 int main_fd __rte_unused)
1383 struct virtio_net *dev = *pdev;
1384 int fd = msg->fds[0];
1389 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1390 return VH_RESULT_ERR;
1393 if (msg->size != sizeof(VhostUserLog)) {
1394 RTE_LOG(ERR, VHOST_CONFIG,
1395 "invalid log base msg size: %"PRId32" != %d\n",
1396 msg->size, (int)sizeof(VhostUserLog));
1397 return VH_RESULT_ERR;
1400 size = msg->payload.log.mmap_size;
1401 off = msg->payload.log.mmap_offset;
1403 /* Don't allow mmap_offset to point outside the mmap region */
1405 RTE_LOG(ERR, VHOST_CONFIG,
1406 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1408 return VH_RESULT_ERR;
1411 RTE_LOG(INFO, VHOST_CONFIG,
1412 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1416 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1417 * fail when offset is not page size aligned.
1419 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1421 if (addr == MAP_FAILED) {
1422 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1423 return VH_RESULT_ERR;
1427 * Free previously mapped log memory on occasionally
1428 * multiple VHOST_USER_SET_LOG_BASE.
1430 if (dev->log_addr) {
1431 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1433 dev->log_addr = (uint64_t)(uintptr_t)addr;
1434 dev->log_base = dev->log_addr + off;
1435 dev->log_size = size;
1438 * The spec is not clear about it (yet), but QEMU doesn't expect
1439 * any payload in the reply.
1444 return VH_RESULT_REPLY;
1447 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1448 struct VhostUserMsg *msg,
1449 int main_fd __rte_unused)
1452 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1454 return VH_RESULT_OK;
1458 * An rarp packet is constructed and broadcasted to notify switches about
1459 * the new location of the migrated VM, so that packets from outside will
1460 * not be lost after migration.
1462 * However, we don't actually "send" a rarp packet here, instead, we set
1463 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1466 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
1467 int main_fd __rte_unused)
1469 struct virtio_net *dev = *pdev;
1470 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1471 struct rte_vdpa_device *vdpa_dev;
1474 RTE_LOG(DEBUG, VHOST_CONFIG,
1475 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1476 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1477 memcpy(dev->mac.addr_bytes, mac, 6);
1480 * Set the flag to inject a RARP broadcast packet at
1481 * rte_vhost_dequeue_burst().
1483 * rte_smp_wmb() is for making sure the mac is copied
1484 * before the flag is set.
1487 rte_atomic16_set(&dev->broadcast_rarp, 1);
1488 did = dev->vdpa_dev_id;
1489 vdpa_dev = rte_vdpa_get_device(did);
1490 if (vdpa_dev && vdpa_dev->ops->migration_done)
1491 vdpa_dev->ops->migration_done(dev->vid);
1493 return VH_RESULT_OK;
1497 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
1498 int main_fd __rte_unused)
1500 struct virtio_net *dev = *pdev;
1501 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1502 msg->payload.u64 > VIRTIO_MAX_MTU) {
1503 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1506 return VH_RESULT_ERR;
1509 dev->mtu = msg->payload.u64;
1511 return VH_RESULT_OK;
1515 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
1516 int main_fd __rte_unused)
1518 struct virtio_net *dev = *pdev;
1519 int fd = msg->fds[0];
1522 RTE_LOG(ERR, VHOST_CONFIG,
1523 "Invalid file descriptor for slave channel (%d)\n",
1525 return VH_RESULT_ERR;
1528 dev->slave_req_fd = fd;
1530 return VH_RESULT_OK;
1534 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1536 struct vhost_vring_addr *ra;
1537 uint64_t start, end;
1540 end = start + imsg->size;
1542 ra = &vq->ring_addrs;
1543 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1545 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1547 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1554 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1555 struct vhost_iotlb_msg *imsg)
1557 uint64_t istart, iend, vstart, vend;
1559 istart = imsg->iova;
1560 iend = istart + imsg->size - 1;
1562 vstart = (uintptr_t)vq->desc;
1563 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1564 if (vstart <= iend && istart <= vend)
1567 vstart = (uintptr_t)vq->avail;
1568 vend = vstart + sizeof(struct vring_avail);
1569 vend += sizeof(uint16_t) * vq->size - 1;
1570 if (vstart <= iend && istart <= vend)
1573 vstart = (uintptr_t)vq->used;
1574 vend = vstart + sizeof(struct vring_used);
1575 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1576 if (vstart <= iend && istart <= vend)
1583 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
1584 int main_fd __rte_unused)
1586 struct virtio_net *dev = *pdev;
1587 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1591 switch (imsg->type) {
1592 case VHOST_IOTLB_UPDATE:
1594 vva = qva_to_vva(dev, imsg->uaddr, &len);
1596 return VH_RESULT_ERR;
1598 for (i = 0; i < dev->nr_vring; i++) {
1599 struct vhost_virtqueue *vq = dev->virtqueue[i];
1601 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1604 if (is_vring_iotlb_update(vq, imsg))
1605 *pdev = dev = translate_ring_addresses(dev, i);
1608 case VHOST_IOTLB_INVALIDATE:
1609 for (i = 0; i < dev->nr_vring; i++) {
1610 struct vhost_virtqueue *vq = dev->virtqueue[i];
1612 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1615 if (is_vring_iotlb_invalidate(vq, imsg))
1616 vring_invalidate(dev, vq);
1620 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1622 return VH_RESULT_ERR;
1625 return VH_RESULT_OK;
1629 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
1630 struct VhostUserMsg *msg,
1631 int main_fd __rte_unused)
1633 struct virtio_net *dev = *pdev;
1634 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1635 struct uffdio_api api_struct;
1637 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
1639 if (dev->postcopy_ufd == -1) {
1640 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
1642 return VH_RESULT_ERR;
1644 api_struct.api = UFFD_API;
1645 api_struct.features = 0;
1646 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
1647 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
1649 close(dev->postcopy_ufd);
1650 dev->postcopy_ufd = -1;
1651 return VH_RESULT_ERR;
1653 msg->fds[0] = dev->postcopy_ufd;
1656 return VH_RESULT_REPLY;
1658 dev->postcopy_ufd = -1;
1661 return VH_RESULT_ERR;
1666 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
1667 struct VhostUserMsg *msg __rte_unused,
1668 int main_fd __rte_unused)
1670 struct virtio_net *dev = *pdev;
1672 if (dev->mem && dev->mem->nregions) {
1673 RTE_LOG(ERR, VHOST_CONFIG,
1674 "Regions already registered at postcopy-listen\n");
1675 return VH_RESULT_ERR;
1677 dev->postcopy_listening = 1;
1679 return VH_RESULT_OK;
1683 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
1684 int main_fd __rte_unused)
1686 struct virtio_net *dev = *pdev;
1688 dev->postcopy_listening = 0;
1689 if (dev->postcopy_ufd >= 0) {
1690 close(dev->postcopy_ufd);
1691 dev->postcopy_ufd = -1;
1694 msg->payload.u64 = 0;
1695 msg->size = sizeof(msg->payload.u64);
1698 return VH_RESULT_REPLY;
1701 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
1702 struct VhostUserMsg *msg,
1704 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
1705 [VHOST_USER_NONE] = NULL,
1706 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
1707 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
1708 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
1709 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
1710 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
1711 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
1712 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
1713 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
1714 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
1715 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
1716 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
1717 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
1718 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
1719 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
1720 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
1721 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
1722 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
1723 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
1724 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
1725 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
1726 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
1727 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
1728 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
1729 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
1730 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
1734 /* return bytes# of read on success or negative val on failure. */
1736 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1740 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1741 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
1746 if (msg->size > sizeof(msg->payload)) {
1747 RTE_LOG(ERR, VHOST_CONFIG,
1748 "invalid msg size: %d\n", msg->size);
1751 ret = read(sockfd, &msg->payload, msg->size);
1754 if (ret != (int)msg->size) {
1755 RTE_LOG(ERR, VHOST_CONFIG,
1756 "read control message failed\n");
1765 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1770 return send_fd_message(sockfd, (char *)msg,
1771 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
1775 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1780 msg->flags &= ~VHOST_USER_VERSION_MASK;
1781 msg->flags &= ~VHOST_USER_NEED_REPLY;
1782 msg->flags |= VHOST_USER_VERSION;
1783 msg->flags |= VHOST_USER_REPLY_MASK;
1785 return send_vhost_message(sockfd, msg);
1789 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
1793 if (msg->flags & VHOST_USER_NEED_REPLY)
1794 rte_spinlock_lock(&dev->slave_req_lock);
1796 ret = send_vhost_message(dev->slave_req_fd, msg);
1797 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1798 rte_spinlock_unlock(&dev->slave_req_lock);
1804 * Allocate a queue pair if it hasn't been allocated yet
1807 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1808 struct VhostUserMsg *msg)
1812 switch (msg->request.master) {
1813 case VHOST_USER_SET_VRING_KICK:
1814 case VHOST_USER_SET_VRING_CALL:
1815 case VHOST_USER_SET_VRING_ERR:
1816 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1818 case VHOST_USER_SET_VRING_NUM:
1819 case VHOST_USER_SET_VRING_BASE:
1820 case VHOST_USER_SET_VRING_ENABLE:
1821 vring_idx = msg->payload.state.index;
1823 case VHOST_USER_SET_VRING_ADDR:
1824 vring_idx = msg->payload.addr.index;
1830 if (vring_idx >= VHOST_MAX_VRING) {
1831 RTE_LOG(ERR, VHOST_CONFIG,
1832 "invalid vring index: %u\n", vring_idx);
1836 if (dev->virtqueue[vring_idx])
1839 return alloc_vring_queue(dev, vring_idx);
1843 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1846 unsigned int vq_num = 0;
1848 while (vq_num < dev->nr_vring) {
1849 struct vhost_virtqueue *vq = dev->virtqueue[i];
1852 rte_spinlock_lock(&vq->access_lock);
1860 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1863 unsigned int vq_num = 0;
1865 while (vq_num < dev->nr_vring) {
1866 struct vhost_virtqueue *vq = dev->virtqueue[i];
1869 rte_spinlock_unlock(&vq->access_lock);
1877 vhost_user_msg_handler(int vid, int fd)
1879 struct virtio_net *dev;
1880 struct VhostUserMsg msg;
1881 struct rte_vdpa_device *vdpa_dev;
1884 int unlock_required = 0;
1885 uint32_t skip_master = 0;
1888 dev = get_device(vid);
1892 if (!dev->notify_ops) {
1893 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1894 if (!dev->notify_ops) {
1895 RTE_LOG(ERR, VHOST_CONFIG,
1896 "failed to get callback ops for driver %s\n",
1902 ret = read_vhost_message(fd, &msg);
1903 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1905 RTE_LOG(ERR, VHOST_CONFIG,
1906 "vhost read message failed\n");
1908 RTE_LOG(INFO, VHOST_CONFIG,
1909 "vhost peer closed\n");
1911 RTE_LOG(ERR, VHOST_CONFIG,
1912 "vhost read incorrect message\n");
1918 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1919 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1920 vhost_message_str[msg.request.master]);
1922 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1923 vhost_message_str[msg.request.master]);
1925 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1927 RTE_LOG(ERR, VHOST_CONFIG,
1928 "failed to alloc queue\n");
1933 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1934 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1935 * and device is destroyed. destroy_device waits for queues to be
1936 * inactive, so it is safe. Otherwise taking the access_lock
1937 * would cause a dead lock.
1939 switch (msg.request.master) {
1940 case VHOST_USER_SET_FEATURES:
1941 case VHOST_USER_SET_PROTOCOL_FEATURES:
1942 case VHOST_USER_SET_OWNER:
1943 case VHOST_USER_SET_MEM_TABLE:
1944 case VHOST_USER_SET_LOG_BASE:
1945 case VHOST_USER_SET_LOG_FD:
1946 case VHOST_USER_SET_VRING_NUM:
1947 case VHOST_USER_SET_VRING_ADDR:
1948 case VHOST_USER_SET_VRING_BASE:
1949 case VHOST_USER_SET_VRING_KICK:
1950 case VHOST_USER_SET_VRING_CALL:
1951 case VHOST_USER_SET_VRING_ERR:
1952 case VHOST_USER_SET_VRING_ENABLE:
1953 case VHOST_USER_SEND_RARP:
1954 case VHOST_USER_NET_SET_MTU:
1955 case VHOST_USER_SET_SLAVE_REQ_FD:
1956 vhost_user_lock_all_queue_pairs(dev);
1957 unlock_required = 1;
1964 if (dev->extern_ops.pre_msg_handle) {
1965 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1966 (void *)&msg, &skip_master);
1967 if (ret == VH_RESULT_ERR)
1969 else if (ret == VH_RESULT_REPLY)
1970 send_vhost_reply(fd, &msg);
1973 goto skip_to_post_handle;
1976 request = msg.request.master;
1977 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
1978 if (!vhost_message_handlers[request])
1979 goto skip_to_post_handle;
1980 ret = vhost_message_handlers[request](&dev, &msg, fd);
1984 RTE_LOG(ERR, VHOST_CONFIG,
1985 "Processing %s failed.\n",
1986 vhost_message_str[request]);
1989 RTE_LOG(DEBUG, VHOST_CONFIG,
1990 "Processing %s succeeded.\n",
1991 vhost_message_str[request]);
1993 case VH_RESULT_REPLY:
1994 RTE_LOG(DEBUG, VHOST_CONFIG,
1995 "Processing %s succeeded and needs reply.\n",
1996 vhost_message_str[request]);
1997 send_vhost_reply(fd, &msg);
2001 RTE_LOG(ERR, VHOST_CONFIG,
2002 "Requested invalid message type %d.\n", request);
2003 ret = VH_RESULT_ERR;
2006 skip_to_post_handle:
2007 if (ret != VH_RESULT_ERR && dev->extern_ops.post_msg_handle) {
2008 ret = (*dev->extern_ops.post_msg_handle)(
2009 dev->vid, (void *)&msg);
2010 if (ret == VH_RESULT_ERR)
2012 else if (ret == VH_RESULT_REPLY)
2013 send_vhost_reply(fd, &msg);
2017 if (unlock_required)
2018 vhost_user_unlock_all_queue_pairs(dev);
2021 * If the request required a reply that was already sent,
2022 * this optional reply-ack won't be sent as the
2023 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2025 if (msg.flags & VHOST_USER_NEED_REPLY) {
2026 msg.payload.u64 = ret == VH_RESULT_ERR;
2027 msg.size = sizeof(msg.payload.u64);
2029 send_vhost_reply(fd, &msg);
2030 } else if (ret == VH_RESULT_ERR) {
2031 RTE_LOG(ERR, VHOST_CONFIG,
2032 "vhost message handling failed.\n");
2036 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2037 dev->flags |= VIRTIO_DEV_READY;
2039 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2040 if (dev->dequeue_zero_copy) {
2041 RTE_LOG(INFO, VHOST_CONFIG,
2042 "dequeue zero copy is enabled\n");
2045 if (dev->notify_ops->new_device(dev->vid) == 0)
2046 dev->flags |= VIRTIO_DEV_RUNNING;
2050 did = dev->vdpa_dev_id;
2051 vdpa_dev = rte_vdpa_get_device(did);
2052 if (vdpa_dev && virtio_is_ready(dev) &&
2053 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2054 msg.request.master == VHOST_USER_SET_VRING_CALL) {
2055 if (vdpa_dev->ops->dev_conf)
2056 vdpa_dev->ops->dev_conf(dev->vid);
2057 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2063 static int process_slave_message_reply(struct virtio_net *dev,
2064 const struct VhostUserMsg *msg)
2066 struct VhostUserMsg msg_reply;
2069 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2072 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2077 if (msg_reply.request.slave != msg->request.slave) {
2078 RTE_LOG(ERR, VHOST_CONFIG,
2079 "Received unexpected msg type (%u), expected %u\n",
2080 msg_reply.request.slave, msg->request.slave);
2085 ret = msg_reply.payload.u64 ? -1 : 0;
2088 rte_spinlock_unlock(&dev->slave_req_lock);
2093 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2096 struct VhostUserMsg msg = {
2097 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2098 .flags = VHOST_USER_VERSION,
2099 .size = sizeof(msg.payload.iotlb),
2103 .type = VHOST_IOTLB_MISS,
2107 ret = send_vhost_message(dev->slave_req_fd, &msg);
2109 RTE_LOG(ERR, VHOST_CONFIG,
2110 "Failed to send IOTLB miss message (%d)\n",
2118 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2124 struct VhostUserMsg msg = {
2125 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2126 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2127 .size = sizeof(msg.payload.area),
2129 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2136 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2142 ret = send_vhost_slave_message(dev, &msg);
2144 RTE_LOG(ERR, VHOST_CONFIG,
2145 "Failed to set host notifier (%d)\n", ret);
2149 return process_slave_message_reply(dev, &msg);
2152 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2154 struct virtio_net *dev;
2155 struct rte_vdpa_device *vdpa_dev;
2156 int vfio_device_fd, did, ret = 0;
2157 uint64_t offset, size;
2160 dev = get_device(vid);
2164 did = dev->vdpa_dev_id;
2168 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2169 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2170 !(dev->protocol_features &
2171 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2172 !(dev->protocol_features &
2173 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2174 !(dev->protocol_features &
2175 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2178 vdpa_dev = rte_vdpa_get_device(did);
2182 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2183 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2185 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2186 if (vfio_device_fd < 0)
2190 for (i = 0; i < dev->nr_vring; i++) {
2191 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2197 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2198 vfio_device_fd, offset, size) < 0) {
2205 for (i = 0; i < dev->nr_vring; i++) {
2206 vhost_user_slave_set_vring_host_notifier(dev, i, -1,