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 * Reclaim all the outstanding zmbufs for a virtqueue.
100 drain_zmbuf_list(struct vhost_virtqueue *vq)
102 struct zcopy_mbuf *zmbuf, *next;
104 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
105 zmbuf != NULL; zmbuf = next) {
106 next = TAILQ_NEXT(zmbuf, next);
108 while (!mbuf_is_consumed(zmbuf->mbuf))
111 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
112 restore_mbuf(zmbuf->mbuf);
113 rte_pktmbuf_free(zmbuf->mbuf);
120 free_mem_region(struct virtio_net *dev)
123 struct rte_vhost_mem_region *reg;
124 struct vhost_virtqueue *vq;
126 if (!dev || !dev->mem)
129 if (dev->dequeue_zero_copy) {
130 for (i = 0; i < dev->nr_vring; i++) {
131 vq = dev->virtqueue[i];
133 drain_zmbuf_list(vq);
137 for (i = 0; i < dev->mem->nregions; i++) {
138 reg = &dev->mem->regions[i];
139 if (reg->host_user_addr) {
140 munmap(reg->mmap_addr, reg->mmap_size);
147 vhost_backend_cleanup(struct virtio_net *dev)
150 free_mem_region(dev);
155 free(dev->guest_pages);
156 dev->guest_pages = NULL;
159 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
163 if (dev->slave_req_fd >= 0) {
164 close(dev->slave_req_fd);
165 dev->slave_req_fd = -1;
168 if (dev->postcopy_ufd >= 0) {
169 close(dev->postcopy_ufd);
170 dev->postcopy_ufd = -1;
173 dev->postcopy_listening = 0;
177 * This function just returns success at the moment unless
178 * the device hasn't been initialised.
181 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
182 struct VhostUserMsg *msg __rte_unused,
183 int main_fd __rte_unused)
185 return RTE_VHOST_MSG_RESULT_OK;
189 vhost_user_reset_owner(struct virtio_net **pdev,
190 struct VhostUserMsg *msg __rte_unused,
191 int main_fd __rte_unused)
193 struct virtio_net *dev = *pdev;
194 vhost_destroy_device_notify(dev);
196 cleanup_device(dev, 0);
198 return RTE_VHOST_MSG_RESULT_OK;
202 * The features that we support are requested.
205 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
206 int main_fd __rte_unused)
208 struct virtio_net *dev = *pdev;
209 uint64_t features = 0;
211 rte_vhost_driver_get_features(dev->ifname, &features);
213 msg->payload.u64 = features;
214 msg->size = sizeof(msg->payload.u64);
217 return RTE_VHOST_MSG_RESULT_REPLY;
221 * The queue number that we support are requested.
224 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
225 int main_fd __rte_unused)
227 struct virtio_net *dev = *pdev;
228 uint32_t queue_num = 0;
230 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
232 msg->payload.u64 = (uint64_t)queue_num;
233 msg->size = sizeof(msg->payload.u64);
236 return RTE_VHOST_MSG_RESULT_REPLY;
240 * We receive the negotiated features supported by us and the virtio device.
243 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
244 int main_fd __rte_unused)
246 struct virtio_net *dev = *pdev;
247 uint64_t features = msg->payload.u64;
248 uint64_t vhost_features = 0;
249 struct rte_vdpa_device *vdpa_dev;
252 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
253 if (features & ~vhost_features) {
254 RTE_LOG(ERR, VHOST_CONFIG,
255 "(%d) received invalid negotiated features.\n",
257 return RTE_VHOST_MSG_RESULT_ERR;
260 if (dev->flags & VIRTIO_DEV_RUNNING) {
261 if (dev->features == features)
262 return RTE_VHOST_MSG_RESULT_OK;
265 * Error out if master tries to change features while device is
266 * in running state. The exception being VHOST_F_LOG_ALL, which
267 * is enabled when the live-migration starts.
269 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
270 RTE_LOG(ERR, VHOST_CONFIG,
271 "(%d) features changed while device is running.\n",
273 return RTE_VHOST_MSG_RESULT_ERR;
276 if (dev->notify_ops->features_changed)
277 dev->notify_ops->features_changed(dev->vid, features);
280 dev->features = features;
282 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
283 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
285 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
287 RTE_LOG(INFO, VHOST_CONFIG,
288 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
289 VHOST_LOG_DEBUG(VHOST_CONFIG,
290 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
292 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
293 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
295 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
296 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
298 * Remove all but first queue pair if MQ hasn't been
299 * negotiated. This is safe because the device is not
300 * running at this stage.
302 while (dev->nr_vring > 2) {
303 struct vhost_virtqueue *vq;
305 vq = dev->virtqueue[--dev->nr_vring];
309 dev->virtqueue[dev->nr_vring] = NULL;
315 did = dev->vdpa_dev_id;
316 vdpa_dev = rte_vdpa_get_device(did);
317 if (vdpa_dev && vdpa_dev->ops->set_features)
318 vdpa_dev->ops->set_features(dev->vid);
320 return RTE_VHOST_MSG_RESULT_OK;
324 * The virtio device sends us the size of the descriptor ring.
327 vhost_user_set_vring_num(struct virtio_net **pdev,
328 struct VhostUserMsg *msg,
329 int main_fd __rte_unused)
331 struct virtio_net *dev = *pdev;
332 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
334 vq->size = msg->payload.state.num;
336 /* VIRTIO 1.0, 2.4 Virtqueues says:
338 * Queue Size value is always a power of 2. The maximum Queue Size
341 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
342 RTE_LOG(ERR, VHOST_CONFIG,
343 "invalid virtqueue size %u\n", vq->size);
344 return RTE_VHOST_MSG_RESULT_ERR;
347 if (dev->dequeue_zero_copy) {
349 vq->last_zmbuf_idx = 0;
350 vq->zmbuf_size = vq->size;
351 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
352 sizeof(struct zcopy_mbuf), 0);
353 if (vq->zmbufs == NULL) {
354 RTE_LOG(WARNING, VHOST_CONFIG,
355 "failed to allocate mem for zero copy; "
356 "zero copy is force disabled\n");
357 dev->dequeue_zero_copy = 0;
359 TAILQ_INIT(&vq->zmbuf_list);
362 if (vq_is_packed(dev)) {
363 vq->shadow_used_packed = rte_malloc(NULL,
365 sizeof(struct vring_used_elem_packed),
366 RTE_CACHE_LINE_SIZE);
367 if (!vq->shadow_used_packed) {
368 RTE_LOG(ERR, VHOST_CONFIG,
369 "failed to allocate memory for shadow used ring.\n");
370 return RTE_VHOST_MSG_RESULT_ERR;
374 vq->shadow_used_split = rte_malloc(NULL,
375 vq->size * sizeof(struct vring_used_elem),
376 RTE_CACHE_LINE_SIZE);
377 if (!vq->shadow_used_split) {
378 RTE_LOG(ERR, VHOST_CONFIG,
379 "failed to allocate memory for shadow used ring.\n");
380 return RTE_VHOST_MSG_RESULT_ERR;
384 vq->batch_copy_elems = rte_malloc(NULL,
385 vq->size * sizeof(struct batch_copy_elem),
386 RTE_CACHE_LINE_SIZE);
387 if (!vq->batch_copy_elems) {
388 RTE_LOG(ERR, VHOST_CONFIG,
389 "failed to allocate memory for batching copy.\n");
390 return RTE_VHOST_MSG_RESULT_ERR;
393 return RTE_VHOST_MSG_RESULT_OK;
397 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
398 * same numa node as the memory of vring descriptor.
400 #ifdef RTE_LIBRTE_VHOST_NUMA
401 static struct virtio_net*
402 numa_realloc(struct virtio_net *dev, int index)
404 int oldnode, newnode;
405 struct virtio_net *old_dev;
406 struct vhost_virtqueue *old_vq, *vq;
407 struct zcopy_mbuf *new_zmbuf;
408 struct vring_used_elem *new_shadow_used_split;
409 struct vring_used_elem_packed *new_shadow_used_packed;
410 struct batch_copy_elem *new_batch_copy_elems;
414 vq = old_vq = dev->virtqueue[index];
416 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
417 MPOL_F_NODE | MPOL_F_ADDR);
419 /* check if we need to reallocate vq */
420 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
421 MPOL_F_NODE | MPOL_F_ADDR);
423 RTE_LOG(ERR, VHOST_CONFIG,
424 "Unable to get vq numa information.\n");
427 if (oldnode != newnode) {
428 RTE_LOG(INFO, VHOST_CONFIG,
429 "reallocate vq from %d to %d node\n", oldnode, newnode);
430 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
434 memcpy(vq, old_vq, sizeof(*vq));
435 TAILQ_INIT(&vq->zmbuf_list);
437 if (dev->dequeue_zero_copy) {
438 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
439 sizeof(struct zcopy_mbuf), 0, newnode);
441 rte_free(vq->zmbufs);
442 vq->zmbufs = new_zmbuf;
446 if (vq_is_packed(dev)) {
447 new_shadow_used_packed = rte_malloc_socket(NULL,
449 sizeof(struct vring_used_elem_packed),
452 if (new_shadow_used_packed) {
453 rte_free(vq->shadow_used_packed);
454 vq->shadow_used_packed = new_shadow_used_packed;
457 new_shadow_used_split = rte_malloc_socket(NULL,
459 sizeof(struct vring_used_elem),
462 if (new_shadow_used_split) {
463 rte_free(vq->shadow_used_split);
464 vq->shadow_used_split = new_shadow_used_split;
468 new_batch_copy_elems = rte_malloc_socket(NULL,
469 vq->size * sizeof(struct batch_copy_elem),
472 if (new_batch_copy_elems) {
473 rte_free(vq->batch_copy_elems);
474 vq->batch_copy_elems = new_batch_copy_elems;
480 /* check if we need to reallocate dev */
481 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
482 MPOL_F_NODE | MPOL_F_ADDR);
484 RTE_LOG(ERR, VHOST_CONFIG,
485 "Unable to get dev numa information.\n");
488 if (oldnode != newnode) {
489 RTE_LOG(INFO, VHOST_CONFIG,
490 "reallocate dev from %d to %d node\n",
492 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
498 memcpy(dev, old_dev, sizeof(*dev));
503 dev->virtqueue[index] = vq;
504 vhost_devices[dev->vid] = dev;
507 vhost_user_iotlb_init(dev, index);
512 static struct virtio_net*
513 numa_realloc(struct virtio_net *dev, int index __rte_unused)
519 /* Converts QEMU virtual address to Vhost virtual address. */
521 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
523 struct rte_vhost_mem_region *r;
526 if (unlikely(!dev || !dev->mem))
529 /* Find the region where the address lives. */
530 for (i = 0; i < dev->mem->nregions; i++) {
531 r = &dev->mem->regions[i];
533 if (qva >= r->guest_user_addr &&
534 qva < r->guest_user_addr + r->size) {
536 if (unlikely(*len > r->guest_user_addr + r->size - qva))
537 *len = r->guest_user_addr + r->size - qva;
539 return qva - r->guest_user_addr +
551 * Converts ring address to Vhost virtual address.
552 * If IOMMU is enabled, the ring address is a guest IO virtual address,
553 * else it is a QEMU virtual address.
556 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
557 uint64_t ra, uint64_t *size)
559 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
562 vva = vhost_user_iotlb_cache_find(vq, ra,
563 size, VHOST_ACCESS_RW);
565 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
570 return qva_to_vva(dev, ra, size);
573 static struct virtio_net *
574 translate_ring_addresses(struct virtio_net *dev, int vq_index)
576 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
577 struct vhost_vring_addr *addr = &vq->ring_addrs;
578 uint64_t len, expected_len;
580 if (vq_is_packed(dev)) {
581 len = sizeof(struct vring_packed_desc) * vq->size;
582 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
583 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
584 vq->log_guest_addr = 0;
585 if (vq->desc_packed == NULL ||
586 len != sizeof(struct vring_packed_desc) *
588 RTE_LOG(DEBUG, VHOST_CONFIG,
589 "(%d) failed to map desc_packed 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_packed_desc_event);
599 vq->driver_event = (struct vring_packed_desc_event *)
600 (uintptr_t)ring_addr_to_vva(dev,
601 vq, addr->avail_user_addr, &len);
602 if (vq->driver_event == NULL ||
603 len != sizeof(struct vring_packed_desc_event)) {
604 RTE_LOG(DEBUG, VHOST_CONFIG,
605 "(%d) failed to find driver area address.\n",
610 len = sizeof(struct vring_packed_desc_event);
611 vq->device_event = (struct vring_packed_desc_event *)
612 (uintptr_t)ring_addr_to_vva(dev,
613 vq, addr->used_user_addr, &len);
614 if (vq->device_event == NULL ||
615 len != sizeof(struct vring_packed_desc_event)) {
616 RTE_LOG(DEBUG, VHOST_CONFIG,
617 "(%d) failed to find device area address.\n",
625 /* The addresses are converted from QEMU virtual to Vhost virtual. */
626 if (vq->desc && vq->avail && vq->used)
629 len = sizeof(struct vring_desc) * vq->size;
630 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
631 vq, addr->desc_user_addr, &len);
632 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
633 RTE_LOG(DEBUG, VHOST_CONFIG,
634 "(%d) failed to map desc ring.\n",
639 dev = numa_realloc(dev, vq_index);
640 vq = dev->virtqueue[vq_index];
641 addr = &vq->ring_addrs;
643 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
644 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
645 len += sizeof(uint16_t);
647 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
648 vq, addr->avail_user_addr, &len);
649 if (vq->avail == 0 || len != expected_len) {
650 RTE_LOG(DEBUG, VHOST_CONFIG,
651 "(%d) failed to map avail ring.\n",
656 len = sizeof(struct vring_used) +
657 sizeof(struct vring_used_elem) * vq->size;
658 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
659 len += sizeof(uint16_t);
661 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
662 vq, addr->used_user_addr, &len);
663 if (vq->used == 0 || len != expected_len) {
664 RTE_LOG(DEBUG, VHOST_CONFIG,
665 "(%d) failed to map used ring.\n",
670 if (vq->last_used_idx != vq->used->idx) {
671 RTE_LOG(WARNING, VHOST_CONFIG,
672 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
673 "some packets maybe resent for Tx and dropped for Rx\n",
674 vq->last_used_idx, vq->used->idx);
675 vq->last_used_idx = vq->used->idx;
676 vq->last_avail_idx = vq->used->idx;
679 vq->log_guest_addr = addr->log_guest_addr;
681 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
683 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
684 dev->vid, vq->avail);
685 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
687 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
688 dev->vid, vq->log_guest_addr);
694 * The virtio device sends us the desc, used and avail ring addresses.
695 * This function then converts these to our address space.
698 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
699 int main_fd __rte_unused)
701 struct virtio_net *dev = *pdev;
702 struct vhost_virtqueue *vq;
703 struct vhost_vring_addr *addr = &msg->payload.addr;
705 if (dev->mem == NULL)
706 return RTE_VHOST_MSG_RESULT_ERR;
708 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
709 vq = dev->virtqueue[msg->payload.addr.index];
712 * Rings addresses should not be interpreted as long as the ring is not
713 * started and enabled
715 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
717 vring_invalidate(dev, vq);
719 if (vq->enabled && (dev->features &
720 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
721 dev = translate_ring_addresses(dev, msg->payload.addr.index);
723 return RTE_VHOST_MSG_RESULT_ERR;
728 return RTE_VHOST_MSG_RESULT_OK;
732 * The virtio device sends us the available ring last used index.
735 vhost_user_set_vring_base(struct virtio_net **pdev,
736 struct VhostUserMsg *msg,
737 int main_fd __rte_unused)
739 struct virtio_net *dev = *pdev;
740 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
741 uint64_t val = msg->payload.state.num;
743 if (vq_is_packed(dev)) {
745 * Bit[0:14]: avail index
746 * Bit[15]: avail wrap counter
748 vq->last_avail_idx = val & 0x7fff;
749 vq->avail_wrap_counter = !!(val & (0x1 << 15));
751 * Set used index to same value as available one, as
752 * their values should be the same since ring processing
753 * was stopped at get time.
755 vq->last_used_idx = vq->last_avail_idx;
756 vq->used_wrap_counter = vq->avail_wrap_counter;
758 vq->last_used_idx = msg->payload.state.num;
759 vq->last_avail_idx = msg->payload.state.num;
762 return RTE_VHOST_MSG_RESULT_OK;
766 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
767 uint64_t host_phys_addr, uint64_t size)
769 struct guest_page *page, *last_page;
770 struct guest_page *old_pages;
772 if (dev->nr_guest_pages == dev->max_guest_pages) {
773 dev->max_guest_pages *= 2;
774 old_pages = dev->guest_pages;
775 dev->guest_pages = realloc(dev->guest_pages,
776 dev->max_guest_pages * sizeof(*page));
777 if (!dev->guest_pages) {
778 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
784 if (dev->nr_guest_pages > 0) {
785 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
786 /* merge if the two pages are continuous */
787 if (host_phys_addr == last_page->host_phys_addr +
789 last_page->size += size;
794 page = &dev->guest_pages[dev->nr_guest_pages++];
795 page->guest_phys_addr = guest_phys_addr;
796 page->host_phys_addr = host_phys_addr;
803 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
806 uint64_t reg_size = reg->size;
807 uint64_t host_user_addr = reg->host_user_addr;
808 uint64_t guest_phys_addr = reg->guest_phys_addr;
809 uint64_t host_phys_addr;
812 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
813 size = page_size - (guest_phys_addr & (page_size - 1));
814 size = RTE_MIN(size, reg_size);
816 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
819 host_user_addr += size;
820 guest_phys_addr += size;
823 while (reg_size > 0) {
824 size = RTE_MIN(reg_size, page_size);
825 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
827 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
831 host_user_addr += size;
832 guest_phys_addr += size;
839 #ifdef RTE_LIBRTE_VHOST_DEBUG
840 /* TODO: enable it only in debug mode? */
842 dump_guest_pages(struct virtio_net *dev)
845 struct guest_page *page;
847 for (i = 0; i < dev->nr_guest_pages; i++) {
848 page = &dev->guest_pages[i];
850 RTE_LOG(INFO, VHOST_CONFIG,
851 "guest physical page region %u\n"
852 "\t guest_phys_addr: %" PRIx64 "\n"
853 "\t host_phys_addr : %" PRIx64 "\n"
854 "\t size : %" PRIx64 "\n",
856 page->guest_phys_addr,
857 page->host_phys_addr,
862 #define dump_guest_pages(dev)
866 vhost_memory_changed(struct VhostUserMemory *new,
867 struct rte_vhost_memory *old)
871 if (new->nregions != old->nregions)
874 for (i = 0; i < new->nregions; ++i) {
875 VhostUserMemoryRegion *new_r = &new->regions[i];
876 struct rte_vhost_mem_region *old_r = &old->regions[i];
878 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
880 if (new_r->memory_size != old_r->size)
882 if (new_r->userspace_addr != old_r->guest_user_addr)
890 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
893 struct virtio_net *dev = *pdev;
894 struct VhostUserMemory *memory = &msg->payload.memory;
895 struct rte_vhost_mem_region *reg;
898 uint64_t mmap_offset;
904 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
905 RTE_LOG(ERR, VHOST_CONFIG,
906 "too many memory regions (%u)\n", memory->nregions);
907 return RTE_VHOST_MSG_RESULT_ERR;
910 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
911 RTE_LOG(INFO, VHOST_CONFIG,
912 "(%d) memory regions not changed\n", dev->vid);
914 for (i = 0; i < memory->nregions; i++)
917 return RTE_VHOST_MSG_RESULT_OK;
921 free_mem_region(dev);
926 /* Flush IOTLB cache as previous HVAs are now invalid */
927 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
928 for (i = 0; i < dev->nr_vring; i++)
929 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
931 dev->nr_guest_pages = 0;
932 if (!dev->guest_pages) {
933 dev->max_guest_pages = 8;
934 dev->guest_pages = malloc(dev->max_guest_pages *
935 sizeof(struct guest_page));
936 if (dev->guest_pages == NULL) {
937 RTE_LOG(ERR, VHOST_CONFIG,
938 "(%d) failed to allocate memory "
939 "for dev->guest_pages\n",
941 return RTE_VHOST_MSG_RESULT_ERR;
945 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
946 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
947 if (dev->mem == NULL) {
948 RTE_LOG(ERR, VHOST_CONFIG,
949 "(%d) failed to allocate memory for dev->mem\n",
951 return RTE_VHOST_MSG_RESULT_ERR;
953 dev->mem->nregions = memory->nregions;
955 for (i = 0; i < memory->nregions; i++) {
957 reg = &dev->mem->regions[i];
959 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
960 reg->guest_user_addr = memory->regions[i].userspace_addr;
961 reg->size = memory->regions[i].memory_size;
964 mmap_offset = memory->regions[i].mmap_offset;
966 /* Check for memory_size + mmap_offset overflow */
967 if (mmap_offset >= -reg->size) {
968 RTE_LOG(ERR, VHOST_CONFIG,
969 "mmap_offset (%#"PRIx64") and memory_size "
970 "(%#"PRIx64") overflow\n",
971 mmap_offset, reg->size);
975 mmap_size = reg->size + mmap_offset;
977 /* mmap() without flag of MAP_ANONYMOUS, should be called
978 * with length argument aligned with hugepagesz at older
979 * longterm version Linux, like 2.6.32 and 3.2.72, or
980 * mmap() will fail with EINVAL.
982 * to avoid failure, make sure in caller to keep length
985 alignment = get_blk_size(fd);
986 if (alignment == (uint64_t)-1) {
987 RTE_LOG(ERR, VHOST_CONFIG,
988 "couldn't get hugepage size through fstat\n");
991 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
993 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
994 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
995 MAP_SHARED | populate, fd, 0);
997 if (mmap_addr == MAP_FAILED) {
998 RTE_LOG(ERR, VHOST_CONFIG,
999 "mmap region %u failed.\n", i);
1003 reg->mmap_addr = mmap_addr;
1004 reg->mmap_size = mmap_size;
1005 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1008 if (dev->dequeue_zero_copy)
1009 if (add_guest_pages(dev, reg, alignment) < 0) {
1010 RTE_LOG(ERR, VHOST_CONFIG,
1011 "adding guest pages to region %u failed.\n",
1016 RTE_LOG(INFO, VHOST_CONFIG,
1017 "guest memory region %u, size: 0x%" PRIx64 "\n"
1018 "\t guest physical addr: 0x%" PRIx64 "\n"
1019 "\t guest virtual addr: 0x%" PRIx64 "\n"
1020 "\t host virtual addr: 0x%" PRIx64 "\n"
1021 "\t mmap addr : 0x%" PRIx64 "\n"
1022 "\t mmap size : 0x%" PRIx64 "\n"
1023 "\t mmap align: 0x%" PRIx64 "\n"
1024 "\t mmap off : 0x%" PRIx64 "\n",
1026 reg->guest_phys_addr,
1027 reg->guest_user_addr,
1028 reg->host_user_addr,
1029 (uint64_t)(uintptr_t)mmap_addr,
1034 if (dev->postcopy_listening) {
1036 * We haven't a better way right now than sharing
1037 * DPDK's virtual address with Qemu, so that Qemu can
1038 * retrieve the region offset when handling userfaults.
1040 memory->regions[i].userspace_addr =
1041 reg->host_user_addr;
1044 if (dev->postcopy_listening) {
1045 /* Send the addresses back to qemu */
1047 send_vhost_reply(main_fd, msg);
1049 /* Wait for qemu to acknolwedge it's got the addresses
1050 * we've got to wait before we're allowed to generate faults.
1052 VhostUserMsg ack_msg;
1053 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1054 RTE_LOG(ERR, VHOST_CONFIG,
1055 "Failed to read qemu ack on postcopy set-mem-table\n");
1058 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1059 RTE_LOG(ERR, VHOST_CONFIG,
1060 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1061 ack_msg.request.master);
1065 /* Now userfault register and we can use the memory */
1066 for (i = 0; i < memory->nregions; i++) {
1067 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1068 reg = &dev->mem->regions[i];
1069 struct uffdio_register reg_struct;
1072 * Let's register all the mmap'ed area to ensure
1073 * alignment on page boundary.
1075 reg_struct.range.start =
1076 (uint64_t)(uintptr_t)reg->mmap_addr;
1077 reg_struct.range.len = reg->mmap_size;
1078 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1080 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1082 RTE_LOG(ERR, VHOST_CONFIG,
1083 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1084 i, dev->postcopy_ufd,
1088 RTE_LOG(INFO, VHOST_CONFIG,
1089 "\t userfaultfd registered for range : %llx - %llx\n",
1090 reg_struct.range.start,
1091 reg_struct.range.start +
1092 reg_struct.range.len - 1);
1099 for (i = 0; i < dev->nr_vring; i++) {
1100 struct vhost_virtqueue *vq = dev->virtqueue[i];
1102 if (vq->desc || vq->avail || vq->used) {
1104 * If the memory table got updated, the ring addresses
1105 * need to be translated again as virtual addresses have
1108 vring_invalidate(dev, vq);
1110 dev = translate_ring_addresses(dev, i);
1120 dump_guest_pages(dev);
1122 return RTE_VHOST_MSG_RESULT_OK;
1125 free_mem_region(dev);
1128 return RTE_VHOST_MSG_RESULT_ERR;
1132 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1139 if (vq_is_packed(dev))
1140 rings_ok = !!vq->desc_packed;
1142 rings_ok = vq->desc && vq->avail && vq->used;
1145 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1146 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1150 virtio_is_ready(struct virtio_net *dev)
1152 struct vhost_virtqueue *vq;
1155 if (dev->nr_vring == 0)
1158 for (i = 0; i < dev->nr_vring; i++) {
1159 vq = dev->virtqueue[i];
1161 if (!vq_is_ready(dev, vq))
1165 RTE_LOG(INFO, VHOST_CONFIG,
1166 "virtio is now ready for processing.\n");
1171 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1172 int main_fd __rte_unused)
1174 struct virtio_net *dev = *pdev;
1175 struct vhost_vring_file file;
1176 struct vhost_virtqueue *vq;
1178 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1179 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1180 file.fd = VIRTIO_INVALID_EVENTFD;
1182 file.fd = msg->fds[0];
1183 RTE_LOG(INFO, VHOST_CONFIG,
1184 "vring call idx:%d file:%d\n", file.index, file.fd);
1186 vq = dev->virtqueue[file.index];
1187 if (vq->callfd >= 0)
1190 vq->callfd = file.fd;
1192 return RTE_VHOST_MSG_RESULT_OK;
1195 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1196 struct VhostUserMsg *msg,
1197 int main_fd __rte_unused)
1199 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1201 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1203 return RTE_VHOST_MSG_RESULT_OK;
1207 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1208 int main_fd __rte_unused)
1210 struct virtio_net *dev = *pdev;
1211 struct vhost_vring_file file;
1212 struct vhost_virtqueue *vq;
1214 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1215 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1216 file.fd = VIRTIO_INVALID_EVENTFD;
1218 file.fd = msg->fds[0];
1219 RTE_LOG(INFO, VHOST_CONFIG,
1220 "vring kick idx:%d file:%d\n", file.index, file.fd);
1222 /* Interpret ring addresses only when ring is started. */
1223 dev = translate_ring_addresses(dev, file.index);
1225 return RTE_VHOST_MSG_RESULT_ERR;
1229 vq = dev->virtqueue[file.index];
1232 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1233 * the ring starts already enabled. Otherwise, it is enabled via
1234 * the SET_VRING_ENABLE message.
1236 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1238 if (dev->notify_ops->vring_state_changed)
1239 dev->notify_ops->vring_state_changed(
1240 dev->vid, file.index, 1);
1243 if (vq->kickfd >= 0)
1245 vq->kickfd = file.fd;
1247 return RTE_VHOST_MSG_RESULT_OK;
1251 free_zmbufs(struct vhost_virtqueue *vq)
1253 drain_zmbuf_list(vq);
1255 rte_free(vq->zmbufs);
1259 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1262 vhost_user_get_vring_base(struct virtio_net **pdev,
1263 struct VhostUserMsg *msg,
1264 int main_fd __rte_unused)
1266 struct virtio_net *dev = *pdev;
1267 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1270 /* We have to stop the queue (virtio) if it is running. */
1271 vhost_destroy_device_notify(dev);
1273 dev->flags &= ~VIRTIO_DEV_READY;
1274 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1276 /* Here we are safe to get the indexes */
1277 if (vq_is_packed(dev)) {
1279 * Bit[0:14]: avail index
1280 * Bit[15]: avail wrap counter
1282 val = vq->last_avail_idx & 0x7fff;
1283 val |= vq->avail_wrap_counter << 15;
1284 msg->payload.state.num = val;
1286 msg->payload.state.num = vq->last_avail_idx;
1289 RTE_LOG(INFO, VHOST_CONFIG,
1290 "vring base idx:%d file:%d\n", msg->payload.state.index,
1291 msg->payload.state.num);
1293 * Based on current qemu vhost-user implementation, this message is
1294 * sent and only sent in vhost_vring_stop.
1295 * TODO: cleanup the vring, it isn't usable since here.
1297 if (vq->kickfd >= 0)
1300 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1302 if (vq->callfd >= 0)
1305 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1307 vq->signalled_used_valid = false;
1309 if (dev->dequeue_zero_copy)
1311 if (vq_is_packed(dev)) {
1312 rte_free(vq->shadow_used_packed);
1313 vq->shadow_used_packed = NULL;
1315 rte_free(vq->shadow_used_split);
1316 vq->shadow_used_split = NULL;
1319 rte_free(vq->batch_copy_elems);
1320 vq->batch_copy_elems = NULL;
1322 msg->size = sizeof(msg->payload.state);
1325 return RTE_VHOST_MSG_RESULT_REPLY;
1329 * when virtio queues are ready to work, qemu will send us to
1330 * enable the virtio queue pair.
1333 vhost_user_set_vring_enable(struct virtio_net **pdev,
1334 struct VhostUserMsg *msg,
1335 int main_fd __rte_unused)
1337 struct virtio_net *dev = *pdev;
1338 int enable = (int)msg->payload.state.num;
1339 int index = (int)msg->payload.state.index;
1340 struct rte_vdpa_device *vdpa_dev;
1343 RTE_LOG(INFO, VHOST_CONFIG,
1344 "set queue enable: %d to qp idx: %d\n",
1347 did = dev->vdpa_dev_id;
1348 vdpa_dev = rte_vdpa_get_device(did);
1349 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1350 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1352 if (dev->notify_ops->vring_state_changed)
1353 dev->notify_ops->vring_state_changed(dev->vid,
1356 /* On disable, rings have to be stopped being processed. */
1357 if (!enable && dev->dequeue_zero_copy)
1358 drain_zmbuf_list(dev->virtqueue[index]);
1360 dev->virtqueue[index]->enabled = enable;
1362 return RTE_VHOST_MSG_RESULT_OK;
1366 vhost_user_get_protocol_features(struct virtio_net **pdev,
1367 struct VhostUserMsg *msg,
1368 int main_fd __rte_unused)
1370 struct virtio_net *dev = *pdev;
1371 uint64_t features, protocol_features;
1373 rte_vhost_driver_get_features(dev->ifname, &features);
1374 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1377 * REPLY_ACK protocol feature is only mandatory for now
1378 * for IOMMU feature. If IOMMU is explicitly disabled by the
1379 * application, disable also REPLY_ACK feature for older buggy
1380 * Qemu versions (from v2.7.0 to v2.9.0).
1382 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1383 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1385 msg->payload.u64 = protocol_features;
1386 msg->size = sizeof(msg->payload.u64);
1389 return RTE_VHOST_MSG_RESULT_REPLY;
1393 vhost_user_set_protocol_features(struct virtio_net **pdev,
1394 struct VhostUserMsg *msg,
1395 int main_fd __rte_unused)
1397 struct virtio_net *dev = *pdev;
1398 uint64_t protocol_features = msg->payload.u64;
1399 uint64_t slave_protocol_features = 0;
1401 rte_vhost_driver_get_protocol_features(dev->ifname,
1402 &slave_protocol_features);
1403 if (protocol_features & ~slave_protocol_features) {
1404 RTE_LOG(ERR, VHOST_CONFIG,
1405 "(%d) received invalid protocol features.\n",
1407 return RTE_VHOST_MSG_RESULT_ERR;
1410 dev->protocol_features = protocol_features;
1411 RTE_LOG(INFO, VHOST_CONFIG,
1412 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
1413 dev->protocol_features);
1415 return RTE_VHOST_MSG_RESULT_OK;
1419 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
1420 int main_fd __rte_unused)
1422 struct virtio_net *dev = *pdev;
1423 int fd = msg->fds[0];
1428 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1429 return RTE_VHOST_MSG_RESULT_ERR;
1432 if (msg->size != sizeof(VhostUserLog)) {
1433 RTE_LOG(ERR, VHOST_CONFIG,
1434 "invalid log base msg size: %"PRId32" != %d\n",
1435 msg->size, (int)sizeof(VhostUserLog));
1436 return RTE_VHOST_MSG_RESULT_ERR;
1439 size = msg->payload.log.mmap_size;
1440 off = msg->payload.log.mmap_offset;
1442 /* Don't allow mmap_offset to point outside the mmap region */
1444 RTE_LOG(ERR, VHOST_CONFIG,
1445 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1447 return RTE_VHOST_MSG_RESULT_ERR;
1450 RTE_LOG(INFO, VHOST_CONFIG,
1451 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1455 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1456 * fail when offset is not page size aligned.
1458 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1460 if (addr == MAP_FAILED) {
1461 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1462 return RTE_VHOST_MSG_RESULT_ERR;
1466 * Free previously mapped log memory on occasionally
1467 * multiple VHOST_USER_SET_LOG_BASE.
1469 if (dev->log_addr) {
1470 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1472 dev->log_addr = (uint64_t)(uintptr_t)addr;
1473 dev->log_base = dev->log_addr + off;
1474 dev->log_size = size;
1477 * The spec is not clear about it (yet), but QEMU doesn't expect
1478 * any payload in the reply.
1483 return RTE_VHOST_MSG_RESULT_REPLY;
1486 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1487 struct VhostUserMsg *msg,
1488 int main_fd __rte_unused)
1491 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1493 return RTE_VHOST_MSG_RESULT_OK;
1497 * An rarp packet is constructed and broadcasted to notify switches about
1498 * the new location of the migrated VM, so that packets from outside will
1499 * not be lost after migration.
1501 * However, we don't actually "send" a rarp packet here, instead, we set
1502 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1505 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
1506 int main_fd __rte_unused)
1508 struct virtio_net *dev = *pdev;
1509 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1510 struct rte_vdpa_device *vdpa_dev;
1513 RTE_LOG(DEBUG, VHOST_CONFIG,
1514 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1515 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1516 memcpy(dev->mac.addr_bytes, mac, 6);
1519 * Set the flag to inject a RARP broadcast packet at
1520 * rte_vhost_dequeue_burst().
1522 * rte_smp_wmb() is for making sure the mac is copied
1523 * before the flag is set.
1526 rte_atomic16_set(&dev->broadcast_rarp, 1);
1527 did = dev->vdpa_dev_id;
1528 vdpa_dev = rte_vdpa_get_device(did);
1529 if (vdpa_dev && vdpa_dev->ops->migration_done)
1530 vdpa_dev->ops->migration_done(dev->vid);
1532 return RTE_VHOST_MSG_RESULT_OK;
1536 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
1537 int main_fd __rte_unused)
1539 struct virtio_net *dev = *pdev;
1540 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1541 msg->payload.u64 > VIRTIO_MAX_MTU) {
1542 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1545 return RTE_VHOST_MSG_RESULT_ERR;
1548 dev->mtu = msg->payload.u64;
1550 return RTE_VHOST_MSG_RESULT_OK;
1554 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
1555 int main_fd __rte_unused)
1557 struct virtio_net *dev = *pdev;
1558 int fd = msg->fds[0];
1561 RTE_LOG(ERR, VHOST_CONFIG,
1562 "Invalid file descriptor for slave channel (%d)\n",
1564 return RTE_VHOST_MSG_RESULT_ERR;
1567 dev->slave_req_fd = fd;
1569 return RTE_VHOST_MSG_RESULT_OK;
1573 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1575 struct vhost_vring_addr *ra;
1576 uint64_t start, end;
1579 end = start + imsg->size;
1581 ra = &vq->ring_addrs;
1582 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1584 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1586 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1593 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1594 struct vhost_iotlb_msg *imsg)
1596 uint64_t istart, iend, vstart, vend;
1598 istart = imsg->iova;
1599 iend = istart + imsg->size - 1;
1601 vstart = (uintptr_t)vq->desc;
1602 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1603 if (vstart <= iend && istart <= vend)
1606 vstart = (uintptr_t)vq->avail;
1607 vend = vstart + sizeof(struct vring_avail);
1608 vend += sizeof(uint16_t) * vq->size - 1;
1609 if (vstart <= iend && istart <= vend)
1612 vstart = (uintptr_t)vq->used;
1613 vend = vstart + sizeof(struct vring_used);
1614 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1615 if (vstart <= iend && istart <= vend)
1622 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
1623 int main_fd __rte_unused)
1625 struct virtio_net *dev = *pdev;
1626 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1630 switch (imsg->type) {
1631 case VHOST_IOTLB_UPDATE:
1633 vva = qva_to_vva(dev, imsg->uaddr, &len);
1635 return RTE_VHOST_MSG_RESULT_ERR;
1637 for (i = 0; i < dev->nr_vring; i++) {
1638 struct vhost_virtqueue *vq = dev->virtqueue[i];
1640 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1643 if (is_vring_iotlb_update(vq, imsg))
1644 *pdev = dev = translate_ring_addresses(dev, i);
1647 case VHOST_IOTLB_INVALIDATE:
1648 for (i = 0; i < dev->nr_vring; i++) {
1649 struct vhost_virtqueue *vq = dev->virtqueue[i];
1651 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1654 if (is_vring_iotlb_invalidate(vq, imsg))
1655 vring_invalidate(dev, vq);
1659 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1661 return RTE_VHOST_MSG_RESULT_ERR;
1664 return RTE_VHOST_MSG_RESULT_OK;
1668 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
1669 struct VhostUserMsg *msg,
1670 int main_fd __rte_unused)
1672 struct virtio_net *dev = *pdev;
1673 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1674 struct uffdio_api api_struct;
1676 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
1678 if (dev->postcopy_ufd == -1) {
1679 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
1681 return RTE_VHOST_MSG_RESULT_ERR;
1683 api_struct.api = UFFD_API;
1684 api_struct.features = 0;
1685 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
1686 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
1688 close(dev->postcopy_ufd);
1689 dev->postcopy_ufd = -1;
1690 return RTE_VHOST_MSG_RESULT_ERR;
1692 msg->fds[0] = dev->postcopy_ufd;
1695 return RTE_VHOST_MSG_RESULT_REPLY;
1697 dev->postcopy_ufd = -1;
1700 return RTE_VHOST_MSG_RESULT_ERR;
1705 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
1706 struct VhostUserMsg *msg __rte_unused,
1707 int main_fd __rte_unused)
1709 struct virtio_net *dev = *pdev;
1711 if (dev->mem && dev->mem->nregions) {
1712 RTE_LOG(ERR, VHOST_CONFIG,
1713 "Regions already registered at postcopy-listen\n");
1714 return RTE_VHOST_MSG_RESULT_ERR;
1716 dev->postcopy_listening = 1;
1718 return RTE_VHOST_MSG_RESULT_OK;
1722 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
1723 int main_fd __rte_unused)
1725 struct virtio_net *dev = *pdev;
1727 dev->postcopy_listening = 0;
1728 if (dev->postcopy_ufd >= 0) {
1729 close(dev->postcopy_ufd);
1730 dev->postcopy_ufd = -1;
1733 msg->payload.u64 = 0;
1734 msg->size = sizeof(msg->payload.u64);
1737 return RTE_VHOST_MSG_RESULT_REPLY;
1740 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
1741 struct VhostUserMsg *msg,
1743 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
1744 [VHOST_USER_NONE] = NULL,
1745 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
1746 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
1747 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
1748 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
1749 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
1750 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
1751 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
1752 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
1753 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
1754 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
1755 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
1756 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
1757 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
1758 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
1759 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
1760 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
1761 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
1762 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
1763 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
1764 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
1765 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
1766 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
1767 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
1768 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
1769 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
1773 /* return bytes# of read on success or negative val on failure. */
1775 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1779 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1780 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
1785 if (msg->size > sizeof(msg->payload)) {
1786 RTE_LOG(ERR, VHOST_CONFIG,
1787 "invalid msg size: %d\n", msg->size);
1790 ret = read(sockfd, &msg->payload, msg->size);
1793 if (ret != (int)msg->size) {
1794 RTE_LOG(ERR, VHOST_CONFIG,
1795 "read control message failed\n");
1804 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1809 return send_fd_message(sockfd, (char *)msg,
1810 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
1814 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1819 msg->flags &= ~VHOST_USER_VERSION_MASK;
1820 msg->flags &= ~VHOST_USER_NEED_REPLY;
1821 msg->flags |= VHOST_USER_VERSION;
1822 msg->flags |= VHOST_USER_REPLY_MASK;
1824 return send_vhost_message(sockfd, msg);
1828 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
1832 if (msg->flags & VHOST_USER_NEED_REPLY)
1833 rte_spinlock_lock(&dev->slave_req_lock);
1835 ret = send_vhost_message(dev->slave_req_fd, msg);
1836 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1837 rte_spinlock_unlock(&dev->slave_req_lock);
1843 * Allocate a queue pair if it hasn't been allocated yet
1846 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1847 struct VhostUserMsg *msg)
1851 switch (msg->request.master) {
1852 case VHOST_USER_SET_VRING_KICK:
1853 case VHOST_USER_SET_VRING_CALL:
1854 case VHOST_USER_SET_VRING_ERR:
1855 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1857 case VHOST_USER_SET_VRING_NUM:
1858 case VHOST_USER_SET_VRING_BASE:
1859 case VHOST_USER_SET_VRING_ENABLE:
1860 vring_idx = msg->payload.state.index;
1862 case VHOST_USER_SET_VRING_ADDR:
1863 vring_idx = msg->payload.addr.index;
1869 if (vring_idx >= VHOST_MAX_VRING) {
1870 RTE_LOG(ERR, VHOST_CONFIG,
1871 "invalid vring index: %u\n", vring_idx);
1875 if (dev->virtqueue[vring_idx])
1878 return alloc_vring_queue(dev, vring_idx);
1882 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1885 unsigned int vq_num = 0;
1887 while (vq_num < dev->nr_vring) {
1888 struct vhost_virtqueue *vq = dev->virtqueue[i];
1891 rte_spinlock_lock(&vq->access_lock);
1899 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1902 unsigned int vq_num = 0;
1904 while (vq_num < dev->nr_vring) {
1905 struct vhost_virtqueue *vq = dev->virtqueue[i];
1908 rte_spinlock_unlock(&vq->access_lock);
1916 vhost_user_msg_handler(int vid, int fd)
1918 struct virtio_net *dev;
1919 struct VhostUserMsg msg;
1920 struct rte_vdpa_device *vdpa_dev;
1923 int unlock_required = 0;
1927 dev = get_device(vid);
1931 if (!dev->notify_ops) {
1932 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1933 if (!dev->notify_ops) {
1934 RTE_LOG(ERR, VHOST_CONFIG,
1935 "failed to get callback ops for driver %s\n",
1941 ret = read_vhost_message(fd, &msg);
1944 RTE_LOG(ERR, VHOST_CONFIG,
1945 "vhost read message failed\n");
1947 RTE_LOG(INFO, VHOST_CONFIG,
1948 "vhost peer closed\n");
1954 request = msg.request.master;
1955 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
1956 vhost_message_str[request]) {
1957 if (request != VHOST_USER_IOTLB_MSG)
1958 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1959 vhost_message_str[request]);
1961 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1962 vhost_message_str[request]);
1964 RTE_LOG(DEBUG, VHOST_CONFIG, "External request %d\n", request);
1967 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1969 RTE_LOG(ERR, VHOST_CONFIG,
1970 "failed to alloc queue\n");
1975 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1976 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1977 * and device is destroyed. destroy_device waits for queues to be
1978 * inactive, so it is safe. Otherwise taking the access_lock
1979 * would cause a dead lock.
1982 case VHOST_USER_SET_FEATURES:
1983 case VHOST_USER_SET_PROTOCOL_FEATURES:
1984 case VHOST_USER_SET_OWNER:
1985 case VHOST_USER_SET_MEM_TABLE:
1986 case VHOST_USER_SET_LOG_BASE:
1987 case VHOST_USER_SET_LOG_FD:
1988 case VHOST_USER_SET_VRING_NUM:
1989 case VHOST_USER_SET_VRING_ADDR:
1990 case VHOST_USER_SET_VRING_BASE:
1991 case VHOST_USER_SET_VRING_KICK:
1992 case VHOST_USER_SET_VRING_CALL:
1993 case VHOST_USER_SET_VRING_ERR:
1994 case VHOST_USER_SET_VRING_ENABLE:
1995 case VHOST_USER_SEND_RARP:
1996 case VHOST_USER_NET_SET_MTU:
1997 case VHOST_USER_SET_SLAVE_REQ_FD:
1998 vhost_user_lock_all_queue_pairs(dev);
1999 unlock_required = 1;
2007 if (dev->extern_ops.pre_msg_handle) {
2008 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2011 case RTE_VHOST_MSG_RESULT_REPLY:
2012 send_vhost_reply(fd, &msg);
2014 case RTE_VHOST_MSG_RESULT_ERR:
2015 case RTE_VHOST_MSG_RESULT_OK:
2017 goto skip_to_post_handle;
2018 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2024 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2025 if (!vhost_message_handlers[request])
2026 goto skip_to_post_handle;
2027 ret = vhost_message_handlers[request](&dev, &msg, fd);
2030 case RTE_VHOST_MSG_RESULT_ERR:
2031 RTE_LOG(ERR, VHOST_CONFIG,
2032 "Processing %s failed.\n",
2033 vhost_message_str[request]);
2036 case RTE_VHOST_MSG_RESULT_OK:
2037 RTE_LOG(DEBUG, VHOST_CONFIG,
2038 "Processing %s succeeded.\n",
2039 vhost_message_str[request]);
2042 case RTE_VHOST_MSG_RESULT_REPLY:
2043 RTE_LOG(DEBUG, VHOST_CONFIG,
2044 "Processing %s succeeded and needs reply.\n",
2045 vhost_message_str[request]);
2046 send_vhost_reply(fd, &msg);
2054 skip_to_post_handle:
2055 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2056 dev->extern_ops.post_msg_handle) {
2057 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2060 case RTE_VHOST_MSG_RESULT_REPLY:
2061 send_vhost_reply(fd, &msg);
2063 case RTE_VHOST_MSG_RESULT_ERR:
2064 case RTE_VHOST_MSG_RESULT_OK:
2066 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2072 if (unlock_required)
2073 vhost_user_unlock_all_queue_pairs(dev);
2075 /* If message was not handled at this stage, treat it as an error */
2077 RTE_LOG(ERR, VHOST_CONFIG,
2078 "vhost message (req: %d) was not handled.\n", request);
2079 ret = RTE_VHOST_MSG_RESULT_ERR;
2083 * If the request required a reply that was already sent,
2084 * this optional reply-ack won't be sent as the
2085 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2087 if (msg.flags & VHOST_USER_NEED_REPLY) {
2088 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2089 msg.size = sizeof(msg.payload.u64);
2091 send_vhost_reply(fd, &msg);
2092 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2093 RTE_LOG(ERR, VHOST_CONFIG,
2094 "vhost message handling failed.\n");
2098 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2099 dev->flags |= VIRTIO_DEV_READY;
2101 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2102 if (dev->dequeue_zero_copy) {
2103 RTE_LOG(INFO, VHOST_CONFIG,
2104 "dequeue zero copy is enabled\n");
2107 if (dev->notify_ops->new_device(dev->vid) == 0)
2108 dev->flags |= VIRTIO_DEV_RUNNING;
2112 did = dev->vdpa_dev_id;
2113 vdpa_dev = rte_vdpa_get_device(did);
2114 if (vdpa_dev && virtio_is_ready(dev) &&
2115 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2116 msg.request.master == VHOST_USER_SET_VRING_CALL) {
2117 if (vdpa_dev->ops->dev_conf)
2118 vdpa_dev->ops->dev_conf(dev->vid);
2119 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2125 static int process_slave_message_reply(struct virtio_net *dev,
2126 const struct VhostUserMsg *msg)
2128 struct VhostUserMsg msg_reply;
2131 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2134 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2139 if (msg_reply.request.slave != msg->request.slave) {
2140 RTE_LOG(ERR, VHOST_CONFIG,
2141 "Received unexpected msg type (%u), expected %u\n",
2142 msg_reply.request.slave, msg->request.slave);
2147 ret = msg_reply.payload.u64 ? -1 : 0;
2150 rte_spinlock_unlock(&dev->slave_req_lock);
2155 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2158 struct VhostUserMsg msg = {
2159 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2160 .flags = VHOST_USER_VERSION,
2161 .size = sizeof(msg.payload.iotlb),
2165 .type = VHOST_IOTLB_MISS,
2169 ret = send_vhost_message(dev->slave_req_fd, &msg);
2171 RTE_LOG(ERR, VHOST_CONFIG,
2172 "Failed to send IOTLB miss message (%d)\n",
2180 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2186 struct VhostUserMsg msg = {
2187 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2188 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2189 .size = sizeof(msg.payload.area),
2191 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2198 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2204 ret = send_vhost_slave_message(dev, &msg);
2206 RTE_LOG(ERR, VHOST_CONFIG,
2207 "Failed to set host notifier (%d)\n", ret);
2211 return process_slave_message_reply(dev, &msg);
2214 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2216 struct virtio_net *dev;
2217 struct rte_vdpa_device *vdpa_dev;
2218 int vfio_device_fd, did, ret = 0;
2219 uint64_t offset, size;
2222 dev = get_device(vid);
2226 did = dev->vdpa_dev_id;
2230 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2231 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2232 !(dev->protocol_features &
2233 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2234 !(dev->protocol_features &
2235 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2236 !(dev->protocol_features &
2237 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2240 vdpa_dev = rte_vdpa_get_device(did);
2244 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2245 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2247 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2248 if (vfio_device_fd < 0)
2252 for (i = 0; i < dev->nr_vring; i++) {
2253 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2259 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2260 vfio_device_fd, offset, size) < 0) {
2267 for (i = 0; i < dev->nr_vring; i++) {
2268 vhost_user_slave_set_vring_host_notifier(dev, i, -1,