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;
413 if (dev->flags & VIRTIO_DEV_RUNNING)
417 vq = old_vq = dev->virtqueue[index];
419 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
420 MPOL_F_NODE | MPOL_F_ADDR);
422 /* check if we need to reallocate vq */
423 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
424 MPOL_F_NODE | MPOL_F_ADDR);
426 RTE_LOG(ERR, VHOST_CONFIG,
427 "Unable to get vq numa information.\n");
430 if (oldnode != newnode) {
431 RTE_LOG(INFO, VHOST_CONFIG,
432 "reallocate vq from %d to %d node\n", oldnode, newnode);
433 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
437 memcpy(vq, old_vq, sizeof(*vq));
438 TAILQ_INIT(&vq->zmbuf_list);
440 if (dev->dequeue_zero_copy) {
441 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
442 sizeof(struct zcopy_mbuf), 0, newnode);
444 rte_free(vq->zmbufs);
445 vq->zmbufs = new_zmbuf;
449 if (vq_is_packed(dev)) {
450 new_shadow_used_packed = rte_malloc_socket(NULL,
452 sizeof(struct vring_used_elem_packed),
455 if (new_shadow_used_packed) {
456 rte_free(vq->shadow_used_packed);
457 vq->shadow_used_packed = new_shadow_used_packed;
460 new_shadow_used_split = rte_malloc_socket(NULL,
462 sizeof(struct vring_used_elem),
465 if (new_shadow_used_split) {
466 rte_free(vq->shadow_used_split);
467 vq->shadow_used_split = new_shadow_used_split;
471 new_batch_copy_elems = rte_malloc_socket(NULL,
472 vq->size * sizeof(struct batch_copy_elem),
475 if (new_batch_copy_elems) {
476 rte_free(vq->batch_copy_elems);
477 vq->batch_copy_elems = new_batch_copy_elems;
483 /* check if we need to reallocate dev */
484 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
485 MPOL_F_NODE | MPOL_F_ADDR);
487 RTE_LOG(ERR, VHOST_CONFIG,
488 "Unable to get dev numa information.\n");
491 if (oldnode != newnode) {
492 RTE_LOG(INFO, VHOST_CONFIG,
493 "reallocate dev from %d to %d node\n",
495 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
501 memcpy(dev, old_dev, sizeof(*dev));
506 dev->virtqueue[index] = vq;
507 vhost_devices[dev->vid] = dev;
510 vhost_user_iotlb_init(dev, index);
515 static struct virtio_net*
516 numa_realloc(struct virtio_net *dev, int index __rte_unused)
522 /* Converts QEMU virtual address to Vhost virtual address. */
524 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
526 struct rte_vhost_mem_region *r;
529 if (unlikely(!dev || !dev->mem))
532 /* Find the region where the address lives. */
533 for (i = 0; i < dev->mem->nregions; i++) {
534 r = &dev->mem->regions[i];
536 if (qva >= r->guest_user_addr &&
537 qva < r->guest_user_addr + r->size) {
539 if (unlikely(*len > r->guest_user_addr + r->size - qva))
540 *len = r->guest_user_addr + r->size - qva;
542 return qva - r->guest_user_addr +
554 * Converts ring address to Vhost virtual address.
555 * If IOMMU is enabled, the ring address is a guest IO virtual address,
556 * else it is a QEMU virtual address.
559 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
560 uint64_t ra, uint64_t *size)
562 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
565 vva = vhost_user_iotlb_cache_find(vq, ra,
566 size, VHOST_ACCESS_RW);
568 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
573 return qva_to_vva(dev, ra, size);
576 static struct virtio_net *
577 translate_ring_addresses(struct virtio_net *dev, int vq_index)
579 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
580 struct vhost_vring_addr *addr = &vq->ring_addrs;
581 uint64_t len, expected_len;
583 if (vq_is_packed(dev)) {
584 len = sizeof(struct vring_packed_desc) * vq->size;
585 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
586 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
587 vq->log_guest_addr = 0;
588 if (vq->desc_packed == NULL ||
589 len != sizeof(struct vring_packed_desc) *
591 RTE_LOG(DEBUG, VHOST_CONFIG,
592 "(%d) failed to map desc_packed ring.\n",
597 dev = numa_realloc(dev, vq_index);
598 vq = dev->virtqueue[vq_index];
599 addr = &vq->ring_addrs;
601 len = sizeof(struct vring_packed_desc_event);
602 vq->driver_event = (struct vring_packed_desc_event *)
603 (uintptr_t)ring_addr_to_vva(dev,
604 vq, addr->avail_user_addr, &len);
605 if (vq->driver_event == NULL ||
606 len != sizeof(struct vring_packed_desc_event)) {
607 RTE_LOG(DEBUG, VHOST_CONFIG,
608 "(%d) failed to find driver area address.\n",
613 len = sizeof(struct vring_packed_desc_event);
614 vq->device_event = (struct vring_packed_desc_event *)
615 (uintptr_t)ring_addr_to_vva(dev,
616 vq, addr->used_user_addr, &len);
617 if (vq->device_event == NULL ||
618 len != sizeof(struct vring_packed_desc_event)) {
619 RTE_LOG(DEBUG, VHOST_CONFIG,
620 "(%d) failed to find device area address.\n",
628 /* The addresses are converted from QEMU virtual to Vhost virtual. */
629 if (vq->desc && vq->avail && vq->used)
632 len = sizeof(struct vring_desc) * vq->size;
633 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
634 vq, addr->desc_user_addr, &len);
635 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
636 RTE_LOG(DEBUG, VHOST_CONFIG,
637 "(%d) failed to map desc ring.\n",
642 dev = numa_realloc(dev, vq_index);
643 vq = dev->virtqueue[vq_index];
644 addr = &vq->ring_addrs;
646 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
647 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
648 len += sizeof(uint16_t);
650 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
651 vq, addr->avail_user_addr, &len);
652 if (vq->avail == 0 || len != expected_len) {
653 RTE_LOG(DEBUG, VHOST_CONFIG,
654 "(%d) failed to map avail ring.\n",
659 len = sizeof(struct vring_used) +
660 sizeof(struct vring_used_elem) * vq->size;
661 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
662 len += sizeof(uint16_t);
664 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
665 vq, addr->used_user_addr, &len);
666 if (vq->used == 0 || len != expected_len) {
667 RTE_LOG(DEBUG, VHOST_CONFIG,
668 "(%d) failed to map used ring.\n",
673 if (vq->last_used_idx != vq->used->idx) {
674 RTE_LOG(WARNING, VHOST_CONFIG,
675 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
676 "some packets maybe resent for Tx and dropped for Rx\n",
677 vq->last_used_idx, vq->used->idx);
678 vq->last_used_idx = vq->used->idx;
679 vq->last_avail_idx = vq->used->idx;
682 vq->log_guest_addr = addr->log_guest_addr;
684 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
686 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
687 dev->vid, vq->avail);
688 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
690 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
691 dev->vid, vq->log_guest_addr);
697 * The virtio device sends us the desc, used and avail ring addresses.
698 * This function then converts these to our address space.
701 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
702 int main_fd __rte_unused)
704 struct virtio_net *dev = *pdev;
705 struct vhost_virtqueue *vq;
706 struct vhost_vring_addr *addr = &msg->payload.addr;
708 if (dev->mem == NULL)
709 return RTE_VHOST_MSG_RESULT_ERR;
711 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
712 vq = dev->virtqueue[msg->payload.addr.index];
715 * Rings addresses should not be interpreted as long as the ring is not
716 * started and enabled
718 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
720 vring_invalidate(dev, vq);
722 if (vq->enabled && (dev->features &
723 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
724 dev = translate_ring_addresses(dev, msg->payload.addr.index);
726 return RTE_VHOST_MSG_RESULT_ERR;
731 return RTE_VHOST_MSG_RESULT_OK;
735 * The virtio device sends us the available ring last used index.
738 vhost_user_set_vring_base(struct virtio_net **pdev,
739 struct VhostUserMsg *msg,
740 int main_fd __rte_unused)
742 struct virtio_net *dev = *pdev;
743 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
744 uint64_t val = msg->payload.state.num;
746 if (vq_is_packed(dev)) {
748 * Bit[0:14]: avail index
749 * Bit[15]: avail wrap counter
751 vq->last_avail_idx = val & 0x7fff;
752 vq->avail_wrap_counter = !!(val & (0x1 << 15));
754 * Set used index to same value as available one, as
755 * their values should be the same since ring processing
756 * was stopped at get time.
758 vq->last_used_idx = vq->last_avail_idx;
759 vq->used_wrap_counter = vq->avail_wrap_counter;
761 vq->last_used_idx = msg->payload.state.num;
762 vq->last_avail_idx = msg->payload.state.num;
765 return RTE_VHOST_MSG_RESULT_OK;
769 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
770 uint64_t host_phys_addr, uint64_t size)
772 struct guest_page *page, *last_page;
773 struct guest_page *old_pages;
775 if (dev->nr_guest_pages == dev->max_guest_pages) {
776 dev->max_guest_pages *= 2;
777 old_pages = dev->guest_pages;
778 dev->guest_pages = realloc(dev->guest_pages,
779 dev->max_guest_pages * sizeof(*page));
780 if (!dev->guest_pages) {
781 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
787 if (dev->nr_guest_pages > 0) {
788 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
789 /* merge if the two pages are continuous */
790 if (host_phys_addr == last_page->host_phys_addr +
792 last_page->size += size;
797 page = &dev->guest_pages[dev->nr_guest_pages++];
798 page->guest_phys_addr = guest_phys_addr;
799 page->host_phys_addr = host_phys_addr;
806 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
809 uint64_t reg_size = reg->size;
810 uint64_t host_user_addr = reg->host_user_addr;
811 uint64_t guest_phys_addr = reg->guest_phys_addr;
812 uint64_t host_phys_addr;
815 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
816 size = page_size - (guest_phys_addr & (page_size - 1));
817 size = RTE_MIN(size, reg_size);
819 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
822 host_user_addr += size;
823 guest_phys_addr += size;
826 while (reg_size > 0) {
827 size = RTE_MIN(reg_size, page_size);
828 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
830 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
834 host_user_addr += size;
835 guest_phys_addr += size;
842 #ifdef RTE_LIBRTE_VHOST_DEBUG
843 /* TODO: enable it only in debug mode? */
845 dump_guest_pages(struct virtio_net *dev)
848 struct guest_page *page;
850 for (i = 0; i < dev->nr_guest_pages; i++) {
851 page = &dev->guest_pages[i];
853 RTE_LOG(INFO, VHOST_CONFIG,
854 "guest physical page region %u\n"
855 "\t guest_phys_addr: %" PRIx64 "\n"
856 "\t host_phys_addr : %" PRIx64 "\n"
857 "\t size : %" PRIx64 "\n",
859 page->guest_phys_addr,
860 page->host_phys_addr,
865 #define dump_guest_pages(dev)
869 vhost_memory_changed(struct VhostUserMemory *new,
870 struct rte_vhost_memory *old)
874 if (new->nregions != old->nregions)
877 for (i = 0; i < new->nregions; ++i) {
878 VhostUserMemoryRegion *new_r = &new->regions[i];
879 struct rte_vhost_mem_region *old_r = &old->regions[i];
881 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
883 if (new_r->memory_size != old_r->size)
885 if (new_r->userspace_addr != old_r->guest_user_addr)
893 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
896 struct virtio_net *dev = *pdev;
897 struct VhostUserMemory *memory = &msg->payload.memory;
898 struct rte_vhost_mem_region *reg;
901 uint64_t mmap_offset;
907 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
908 RTE_LOG(ERR, VHOST_CONFIG,
909 "too many memory regions (%u)\n", memory->nregions);
910 return RTE_VHOST_MSG_RESULT_ERR;
913 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
914 RTE_LOG(INFO, VHOST_CONFIG,
915 "(%d) memory regions not changed\n", dev->vid);
917 for (i = 0; i < memory->nregions; i++)
920 return RTE_VHOST_MSG_RESULT_OK;
924 free_mem_region(dev);
929 /* Flush IOTLB cache as previous HVAs are now invalid */
930 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
931 for (i = 0; i < dev->nr_vring; i++)
932 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
934 dev->nr_guest_pages = 0;
935 if (!dev->guest_pages) {
936 dev->max_guest_pages = 8;
937 dev->guest_pages = malloc(dev->max_guest_pages *
938 sizeof(struct guest_page));
939 if (dev->guest_pages == NULL) {
940 RTE_LOG(ERR, VHOST_CONFIG,
941 "(%d) failed to allocate memory "
942 "for dev->guest_pages\n",
944 return RTE_VHOST_MSG_RESULT_ERR;
948 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
949 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
950 if (dev->mem == NULL) {
951 RTE_LOG(ERR, VHOST_CONFIG,
952 "(%d) failed to allocate memory for dev->mem\n",
954 return RTE_VHOST_MSG_RESULT_ERR;
956 dev->mem->nregions = memory->nregions;
958 for (i = 0; i < memory->nregions; i++) {
960 reg = &dev->mem->regions[i];
962 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
963 reg->guest_user_addr = memory->regions[i].userspace_addr;
964 reg->size = memory->regions[i].memory_size;
967 mmap_offset = memory->regions[i].mmap_offset;
969 /* Check for memory_size + mmap_offset overflow */
970 if (mmap_offset >= -reg->size) {
971 RTE_LOG(ERR, VHOST_CONFIG,
972 "mmap_offset (%#"PRIx64") and memory_size "
973 "(%#"PRIx64") overflow\n",
974 mmap_offset, reg->size);
978 mmap_size = reg->size + mmap_offset;
980 /* mmap() without flag of MAP_ANONYMOUS, should be called
981 * with length argument aligned with hugepagesz at older
982 * longterm version Linux, like 2.6.32 and 3.2.72, or
983 * mmap() will fail with EINVAL.
985 * to avoid failure, make sure in caller to keep length
988 alignment = get_blk_size(fd);
989 if (alignment == (uint64_t)-1) {
990 RTE_LOG(ERR, VHOST_CONFIG,
991 "couldn't get hugepage size through fstat\n");
994 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
996 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
997 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
998 MAP_SHARED | populate, fd, 0);
1000 if (mmap_addr == MAP_FAILED) {
1001 RTE_LOG(ERR, VHOST_CONFIG,
1002 "mmap region %u failed.\n", i);
1006 reg->mmap_addr = mmap_addr;
1007 reg->mmap_size = mmap_size;
1008 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1011 if (dev->dequeue_zero_copy)
1012 if (add_guest_pages(dev, reg, alignment) < 0) {
1013 RTE_LOG(ERR, VHOST_CONFIG,
1014 "adding guest pages to region %u failed.\n",
1019 RTE_LOG(INFO, VHOST_CONFIG,
1020 "guest memory region %u, size: 0x%" PRIx64 "\n"
1021 "\t guest physical addr: 0x%" PRIx64 "\n"
1022 "\t guest virtual addr: 0x%" PRIx64 "\n"
1023 "\t host virtual addr: 0x%" PRIx64 "\n"
1024 "\t mmap addr : 0x%" PRIx64 "\n"
1025 "\t mmap size : 0x%" PRIx64 "\n"
1026 "\t mmap align: 0x%" PRIx64 "\n"
1027 "\t mmap off : 0x%" PRIx64 "\n",
1029 reg->guest_phys_addr,
1030 reg->guest_user_addr,
1031 reg->host_user_addr,
1032 (uint64_t)(uintptr_t)mmap_addr,
1037 if (dev->postcopy_listening) {
1039 * We haven't a better way right now than sharing
1040 * DPDK's virtual address with Qemu, so that Qemu can
1041 * retrieve the region offset when handling userfaults.
1043 memory->regions[i].userspace_addr =
1044 reg->host_user_addr;
1047 if (dev->postcopy_listening) {
1048 /* Send the addresses back to qemu */
1050 send_vhost_reply(main_fd, msg);
1052 /* Wait for qemu to acknolwedge it's got the addresses
1053 * we've got to wait before we're allowed to generate faults.
1055 VhostUserMsg ack_msg;
1056 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1057 RTE_LOG(ERR, VHOST_CONFIG,
1058 "Failed to read qemu ack on postcopy set-mem-table\n");
1061 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1062 RTE_LOG(ERR, VHOST_CONFIG,
1063 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1064 ack_msg.request.master);
1068 /* Now userfault register and we can use the memory */
1069 for (i = 0; i < memory->nregions; i++) {
1070 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1071 reg = &dev->mem->regions[i];
1072 struct uffdio_register reg_struct;
1075 * Let's register all the mmap'ed area to ensure
1076 * alignment on page boundary.
1078 reg_struct.range.start =
1079 (uint64_t)(uintptr_t)reg->mmap_addr;
1080 reg_struct.range.len = reg->mmap_size;
1081 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1083 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1085 RTE_LOG(ERR, VHOST_CONFIG,
1086 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1087 i, dev->postcopy_ufd,
1091 RTE_LOG(INFO, VHOST_CONFIG,
1092 "\t userfaultfd registered for range : %llx - %llx\n",
1093 reg_struct.range.start,
1094 reg_struct.range.start +
1095 reg_struct.range.len - 1);
1102 for (i = 0; i < dev->nr_vring; i++) {
1103 struct vhost_virtqueue *vq = dev->virtqueue[i];
1105 if (vq->desc || vq->avail || vq->used) {
1107 * If the memory table got updated, the ring addresses
1108 * need to be translated again as virtual addresses have
1111 vring_invalidate(dev, vq);
1113 dev = translate_ring_addresses(dev, i);
1123 dump_guest_pages(dev);
1125 return RTE_VHOST_MSG_RESULT_OK;
1128 free_mem_region(dev);
1131 return RTE_VHOST_MSG_RESULT_ERR;
1135 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1142 if (vq_is_packed(dev))
1143 rings_ok = !!vq->desc_packed;
1145 rings_ok = vq->desc && vq->avail && vq->used;
1148 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1149 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1153 virtio_is_ready(struct virtio_net *dev)
1155 struct vhost_virtqueue *vq;
1158 if (dev->nr_vring == 0)
1161 for (i = 0; i < dev->nr_vring; i++) {
1162 vq = dev->virtqueue[i];
1164 if (!vq_is_ready(dev, vq))
1168 RTE_LOG(INFO, VHOST_CONFIG,
1169 "virtio is now ready for processing.\n");
1174 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1175 int main_fd __rte_unused)
1177 struct virtio_net *dev = *pdev;
1178 struct vhost_vring_file file;
1179 struct vhost_virtqueue *vq;
1181 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1182 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1183 file.fd = VIRTIO_INVALID_EVENTFD;
1185 file.fd = msg->fds[0];
1186 RTE_LOG(INFO, VHOST_CONFIG,
1187 "vring call idx:%d file:%d\n", file.index, file.fd);
1189 vq = dev->virtqueue[file.index];
1190 if (vq->callfd >= 0)
1193 vq->callfd = file.fd;
1195 return RTE_VHOST_MSG_RESULT_OK;
1198 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1199 struct VhostUserMsg *msg,
1200 int main_fd __rte_unused)
1202 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1204 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1206 return RTE_VHOST_MSG_RESULT_OK;
1210 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1211 int main_fd __rte_unused)
1213 struct virtio_net *dev = *pdev;
1214 struct vhost_vring_file file;
1215 struct vhost_virtqueue *vq;
1217 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1218 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1219 file.fd = VIRTIO_INVALID_EVENTFD;
1221 file.fd = msg->fds[0];
1222 RTE_LOG(INFO, VHOST_CONFIG,
1223 "vring kick idx:%d file:%d\n", file.index, file.fd);
1225 /* Interpret ring addresses only when ring is started. */
1226 dev = translate_ring_addresses(dev, file.index);
1228 return RTE_VHOST_MSG_RESULT_ERR;
1232 vq = dev->virtqueue[file.index];
1235 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1236 * the ring starts already enabled. Otherwise, it is enabled via
1237 * the SET_VRING_ENABLE message.
1239 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1241 if (dev->notify_ops->vring_state_changed)
1242 dev->notify_ops->vring_state_changed(
1243 dev->vid, file.index, 1);
1246 if (vq->kickfd >= 0)
1248 vq->kickfd = file.fd;
1250 return RTE_VHOST_MSG_RESULT_OK;
1254 free_zmbufs(struct vhost_virtqueue *vq)
1256 drain_zmbuf_list(vq);
1258 rte_free(vq->zmbufs);
1262 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1265 vhost_user_get_vring_base(struct virtio_net **pdev,
1266 struct VhostUserMsg *msg,
1267 int main_fd __rte_unused)
1269 struct virtio_net *dev = *pdev;
1270 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1273 /* We have to stop the queue (virtio) if it is running. */
1274 vhost_destroy_device_notify(dev);
1276 dev->flags &= ~VIRTIO_DEV_READY;
1277 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1279 /* Here we are safe to get the indexes */
1280 if (vq_is_packed(dev)) {
1282 * Bit[0:14]: avail index
1283 * Bit[15]: avail wrap counter
1285 val = vq->last_avail_idx & 0x7fff;
1286 val |= vq->avail_wrap_counter << 15;
1287 msg->payload.state.num = val;
1289 msg->payload.state.num = vq->last_avail_idx;
1292 RTE_LOG(INFO, VHOST_CONFIG,
1293 "vring base idx:%d file:%d\n", msg->payload.state.index,
1294 msg->payload.state.num);
1296 * Based on current qemu vhost-user implementation, this message is
1297 * sent and only sent in vhost_vring_stop.
1298 * TODO: cleanup the vring, it isn't usable since here.
1300 if (vq->kickfd >= 0)
1303 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1305 if (vq->callfd >= 0)
1308 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1310 vq->signalled_used_valid = false;
1312 if (dev->dequeue_zero_copy)
1314 if (vq_is_packed(dev)) {
1315 rte_free(vq->shadow_used_packed);
1316 vq->shadow_used_packed = NULL;
1318 rte_free(vq->shadow_used_split);
1319 vq->shadow_used_split = NULL;
1322 rte_free(vq->batch_copy_elems);
1323 vq->batch_copy_elems = NULL;
1325 msg->size = sizeof(msg->payload.state);
1328 return RTE_VHOST_MSG_RESULT_REPLY;
1332 * when virtio queues are ready to work, qemu will send us to
1333 * enable the virtio queue pair.
1336 vhost_user_set_vring_enable(struct virtio_net **pdev,
1337 struct VhostUserMsg *msg,
1338 int main_fd __rte_unused)
1340 struct virtio_net *dev = *pdev;
1341 int enable = (int)msg->payload.state.num;
1342 int index = (int)msg->payload.state.index;
1343 struct rte_vdpa_device *vdpa_dev;
1346 RTE_LOG(INFO, VHOST_CONFIG,
1347 "set queue enable: %d to qp idx: %d\n",
1350 did = dev->vdpa_dev_id;
1351 vdpa_dev = rte_vdpa_get_device(did);
1352 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1353 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1355 if (dev->notify_ops->vring_state_changed)
1356 dev->notify_ops->vring_state_changed(dev->vid,
1359 /* On disable, rings have to be stopped being processed. */
1360 if (!enable && dev->dequeue_zero_copy)
1361 drain_zmbuf_list(dev->virtqueue[index]);
1363 dev->virtqueue[index]->enabled = enable;
1365 return RTE_VHOST_MSG_RESULT_OK;
1369 vhost_user_get_protocol_features(struct virtio_net **pdev,
1370 struct VhostUserMsg *msg,
1371 int main_fd __rte_unused)
1373 struct virtio_net *dev = *pdev;
1374 uint64_t features, protocol_features;
1376 rte_vhost_driver_get_features(dev->ifname, &features);
1377 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1380 * REPLY_ACK protocol feature is only mandatory for now
1381 * for IOMMU feature. If IOMMU is explicitly disabled by the
1382 * application, disable also REPLY_ACK feature for older buggy
1383 * Qemu versions (from v2.7.0 to v2.9.0).
1385 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1386 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1388 msg->payload.u64 = protocol_features;
1389 msg->size = sizeof(msg->payload.u64);
1392 return RTE_VHOST_MSG_RESULT_REPLY;
1396 vhost_user_set_protocol_features(struct virtio_net **pdev,
1397 struct VhostUserMsg *msg,
1398 int main_fd __rte_unused)
1400 struct virtio_net *dev = *pdev;
1401 uint64_t protocol_features = msg->payload.u64;
1402 uint64_t slave_protocol_features = 0;
1404 rte_vhost_driver_get_protocol_features(dev->ifname,
1405 &slave_protocol_features);
1406 if (protocol_features & ~slave_protocol_features) {
1407 RTE_LOG(ERR, VHOST_CONFIG,
1408 "(%d) received invalid protocol features.\n",
1410 return RTE_VHOST_MSG_RESULT_ERR;
1413 dev->protocol_features = protocol_features;
1414 RTE_LOG(INFO, VHOST_CONFIG,
1415 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
1416 dev->protocol_features);
1418 return RTE_VHOST_MSG_RESULT_OK;
1422 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
1423 int main_fd __rte_unused)
1425 struct virtio_net *dev = *pdev;
1426 int fd = msg->fds[0];
1431 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1432 return RTE_VHOST_MSG_RESULT_ERR;
1435 if (msg->size != sizeof(VhostUserLog)) {
1436 RTE_LOG(ERR, VHOST_CONFIG,
1437 "invalid log base msg size: %"PRId32" != %d\n",
1438 msg->size, (int)sizeof(VhostUserLog));
1439 return RTE_VHOST_MSG_RESULT_ERR;
1442 size = msg->payload.log.mmap_size;
1443 off = msg->payload.log.mmap_offset;
1445 /* Don't allow mmap_offset to point outside the mmap region */
1447 RTE_LOG(ERR, VHOST_CONFIG,
1448 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1450 return RTE_VHOST_MSG_RESULT_ERR;
1453 RTE_LOG(INFO, VHOST_CONFIG,
1454 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1458 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1459 * fail when offset is not page size aligned.
1461 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1463 if (addr == MAP_FAILED) {
1464 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1465 return RTE_VHOST_MSG_RESULT_ERR;
1469 * Free previously mapped log memory on occasionally
1470 * multiple VHOST_USER_SET_LOG_BASE.
1472 if (dev->log_addr) {
1473 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1475 dev->log_addr = (uint64_t)(uintptr_t)addr;
1476 dev->log_base = dev->log_addr + off;
1477 dev->log_size = size;
1480 * The spec is not clear about it (yet), but QEMU doesn't expect
1481 * any payload in the reply.
1486 return RTE_VHOST_MSG_RESULT_REPLY;
1489 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1490 struct VhostUserMsg *msg,
1491 int main_fd __rte_unused)
1494 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1496 return RTE_VHOST_MSG_RESULT_OK;
1500 * An rarp packet is constructed and broadcasted to notify switches about
1501 * the new location of the migrated VM, so that packets from outside will
1502 * not be lost after migration.
1504 * However, we don't actually "send" a rarp packet here, instead, we set
1505 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1508 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
1509 int main_fd __rte_unused)
1511 struct virtio_net *dev = *pdev;
1512 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1513 struct rte_vdpa_device *vdpa_dev;
1516 RTE_LOG(DEBUG, VHOST_CONFIG,
1517 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1518 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1519 memcpy(dev->mac.addr_bytes, mac, 6);
1522 * Set the flag to inject a RARP broadcast packet at
1523 * rte_vhost_dequeue_burst().
1525 * rte_smp_wmb() is for making sure the mac is copied
1526 * before the flag is set.
1529 rte_atomic16_set(&dev->broadcast_rarp, 1);
1530 did = dev->vdpa_dev_id;
1531 vdpa_dev = rte_vdpa_get_device(did);
1532 if (vdpa_dev && vdpa_dev->ops->migration_done)
1533 vdpa_dev->ops->migration_done(dev->vid);
1535 return RTE_VHOST_MSG_RESULT_OK;
1539 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
1540 int main_fd __rte_unused)
1542 struct virtio_net *dev = *pdev;
1543 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1544 msg->payload.u64 > VIRTIO_MAX_MTU) {
1545 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1548 return RTE_VHOST_MSG_RESULT_ERR;
1551 dev->mtu = msg->payload.u64;
1553 return RTE_VHOST_MSG_RESULT_OK;
1557 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
1558 int main_fd __rte_unused)
1560 struct virtio_net *dev = *pdev;
1561 int fd = msg->fds[0];
1564 RTE_LOG(ERR, VHOST_CONFIG,
1565 "Invalid file descriptor for slave channel (%d)\n",
1567 return RTE_VHOST_MSG_RESULT_ERR;
1570 if (dev->slave_req_fd >= 0)
1571 close(dev->slave_req_fd);
1573 dev->slave_req_fd = fd;
1575 return RTE_VHOST_MSG_RESULT_OK;
1579 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1581 struct vhost_vring_addr *ra;
1582 uint64_t start, end;
1585 end = start + imsg->size;
1587 ra = &vq->ring_addrs;
1588 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1590 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1592 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1599 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1600 struct vhost_iotlb_msg *imsg)
1602 uint64_t istart, iend, vstart, vend;
1604 istart = imsg->iova;
1605 iend = istart + imsg->size - 1;
1607 vstart = (uintptr_t)vq->desc;
1608 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1609 if (vstart <= iend && istart <= vend)
1612 vstart = (uintptr_t)vq->avail;
1613 vend = vstart + sizeof(struct vring_avail);
1614 vend += sizeof(uint16_t) * vq->size - 1;
1615 if (vstart <= iend && istart <= vend)
1618 vstart = (uintptr_t)vq->used;
1619 vend = vstart + sizeof(struct vring_used);
1620 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1621 if (vstart <= iend && istart <= vend)
1628 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
1629 int main_fd __rte_unused)
1631 struct virtio_net *dev = *pdev;
1632 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1636 switch (imsg->type) {
1637 case VHOST_IOTLB_UPDATE:
1639 vva = qva_to_vva(dev, imsg->uaddr, &len);
1641 return RTE_VHOST_MSG_RESULT_ERR;
1643 for (i = 0; i < dev->nr_vring; i++) {
1644 struct vhost_virtqueue *vq = dev->virtqueue[i];
1646 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1649 if (is_vring_iotlb_update(vq, imsg))
1650 *pdev = dev = translate_ring_addresses(dev, i);
1653 case VHOST_IOTLB_INVALIDATE:
1654 for (i = 0; i < dev->nr_vring; i++) {
1655 struct vhost_virtqueue *vq = dev->virtqueue[i];
1657 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1660 if (is_vring_iotlb_invalidate(vq, imsg))
1661 vring_invalidate(dev, vq);
1665 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1667 return RTE_VHOST_MSG_RESULT_ERR;
1670 return RTE_VHOST_MSG_RESULT_OK;
1674 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
1675 struct VhostUserMsg *msg,
1676 int main_fd __rte_unused)
1678 struct virtio_net *dev = *pdev;
1679 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1680 struct uffdio_api api_struct;
1682 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
1684 if (dev->postcopy_ufd == -1) {
1685 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
1687 return RTE_VHOST_MSG_RESULT_ERR;
1689 api_struct.api = UFFD_API;
1690 api_struct.features = 0;
1691 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
1692 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
1694 close(dev->postcopy_ufd);
1695 dev->postcopy_ufd = -1;
1696 return RTE_VHOST_MSG_RESULT_ERR;
1698 msg->fds[0] = dev->postcopy_ufd;
1701 return RTE_VHOST_MSG_RESULT_REPLY;
1703 dev->postcopy_ufd = -1;
1706 return RTE_VHOST_MSG_RESULT_ERR;
1711 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
1712 struct VhostUserMsg *msg __rte_unused,
1713 int main_fd __rte_unused)
1715 struct virtio_net *dev = *pdev;
1717 if (dev->mem && dev->mem->nregions) {
1718 RTE_LOG(ERR, VHOST_CONFIG,
1719 "Regions already registered at postcopy-listen\n");
1720 return RTE_VHOST_MSG_RESULT_ERR;
1722 dev->postcopy_listening = 1;
1724 return RTE_VHOST_MSG_RESULT_OK;
1728 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
1729 int main_fd __rte_unused)
1731 struct virtio_net *dev = *pdev;
1733 dev->postcopy_listening = 0;
1734 if (dev->postcopy_ufd >= 0) {
1735 close(dev->postcopy_ufd);
1736 dev->postcopy_ufd = -1;
1739 msg->payload.u64 = 0;
1740 msg->size = sizeof(msg->payload.u64);
1743 return RTE_VHOST_MSG_RESULT_REPLY;
1746 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
1747 struct VhostUserMsg *msg,
1749 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
1750 [VHOST_USER_NONE] = NULL,
1751 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
1752 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
1753 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
1754 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
1755 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
1756 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
1757 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
1758 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
1759 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
1760 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
1761 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
1762 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
1763 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
1764 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
1765 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
1766 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
1767 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
1768 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
1769 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
1770 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
1771 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
1772 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
1773 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
1774 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
1775 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
1779 /* return bytes# of read on success or negative val on failure. */
1781 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1785 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1786 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
1791 if (msg->size > sizeof(msg->payload)) {
1792 RTE_LOG(ERR, VHOST_CONFIG,
1793 "invalid msg size: %d\n", msg->size);
1796 ret = read(sockfd, &msg->payload, msg->size);
1799 if (ret != (int)msg->size) {
1800 RTE_LOG(ERR, VHOST_CONFIG,
1801 "read control message failed\n");
1810 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1815 return send_fd_message(sockfd, (char *)msg,
1816 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
1820 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1825 msg->flags &= ~VHOST_USER_VERSION_MASK;
1826 msg->flags &= ~VHOST_USER_NEED_REPLY;
1827 msg->flags |= VHOST_USER_VERSION;
1828 msg->flags |= VHOST_USER_REPLY_MASK;
1830 return send_vhost_message(sockfd, msg);
1834 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
1838 if (msg->flags & VHOST_USER_NEED_REPLY)
1839 rte_spinlock_lock(&dev->slave_req_lock);
1841 ret = send_vhost_message(dev->slave_req_fd, msg);
1842 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1843 rte_spinlock_unlock(&dev->slave_req_lock);
1849 * Allocate a queue pair if it hasn't been allocated yet
1852 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1853 struct VhostUserMsg *msg)
1857 switch (msg->request.master) {
1858 case VHOST_USER_SET_VRING_KICK:
1859 case VHOST_USER_SET_VRING_CALL:
1860 case VHOST_USER_SET_VRING_ERR:
1861 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1863 case VHOST_USER_SET_VRING_NUM:
1864 case VHOST_USER_SET_VRING_BASE:
1865 case VHOST_USER_SET_VRING_ENABLE:
1866 vring_idx = msg->payload.state.index;
1868 case VHOST_USER_SET_VRING_ADDR:
1869 vring_idx = msg->payload.addr.index;
1875 if (vring_idx >= VHOST_MAX_VRING) {
1876 RTE_LOG(ERR, VHOST_CONFIG,
1877 "invalid vring index: %u\n", vring_idx);
1881 if (dev->virtqueue[vring_idx])
1884 return alloc_vring_queue(dev, vring_idx);
1888 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1891 unsigned int vq_num = 0;
1893 while (vq_num < dev->nr_vring) {
1894 struct vhost_virtqueue *vq = dev->virtqueue[i];
1897 rte_spinlock_lock(&vq->access_lock);
1905 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1908 unsigned int vq_num = 0;
1910 while (vq_num < dev->nr_vring) {
1911 struct vhost_virtqueue *vq = dev->virtqueue[i];
1914 rte_spinlock_unlock(&vq->access_lock);
1922 vhost_user_msg_handler(int vid, int fd)
1924 struct virtio_net *dev;
1925 struct VhostUserMsg msg;
1926 struct rte_vdpa_device *vdpa_dev;
1929 int unlock_required = 0;
1933 dev = get_device(vid);
1937 if (!dev->notify_ops) {
1938 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1939 if (!dev->notify_ops) {
1940 RTE_LOG(ERR, VHOST_CONFIG,
1941 "failed to get callback ops for driver %s\n",
1947 ret = read_vhost_message(fd, &msg);
1950 RTE_LOG(ERR, VHOST_CONFIG,
1951 "vhost read message failed\n");
1953 RTE_LOG(INFO, VHOST_CONFIG,
1954 "vhost peer closed\n");
1960 request = msg.request.master;
1961 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
1962 vhost_message_str[request]) {
1963 if (request != VHOST_USER_IOTLB_MSG)
1964 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1965 vhost_message_str[request]);
1967 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1968 vhost_message_str[request]);
1970 RTE_LOG(DEBUG, VHOST_CONFIG, "External request %d\n", request);
1973 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1975 RTE_LOG(ERR, VHOST_CONFIG,
1976 "failed to alloc queue\n");
1981 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1982 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1983 * and device is destroyed. destroy_device waits for queues to be
1984 * inactive, so it is safe. Otherwise taking the access_lock
1985 * would cause a dead lock.
1988 case VHOST_USER_SET_FEATURES:
1989 case VHOST_USER_SET_PROTOCOL_FEATURES:
1990 case VHOST_USER_SET_OWNER:
1991 case VHOST_USER_SET_MEM_TABLE:
1992 case VHOST_USER_SET_LOG_BASE:
1993 case VHOST_USER_SET_LOG_FD:
1994 case VHOST_USER_SET_VRING_NUM:
1995 case VHOST_USER_SET_VRING_ADDR:
1996 case VHOST_USER_SET_VRING_BASE:
1997 case VHOST_USER_SET_VRING_KICK:
1998 case VHOST_USER_SET_VRING_CALL:
1999 case VHOST_USER_SET_VRING_ERR:
2000 case VHOST_USER_SET_VRING_ENABLE:
2001 case VHOST_USER_SEND_RARP:
2002 case VHOST_USER_NET_SET_MTU:
2003 case VHOST_USER_SET_SLAVE_REQ_FD:
2004 vhost_user_lock_all_queue_pairs(dev);
2005 unlock_required = 1;
2013 if (dev->extern_ops.pre_msg_handle) {
2014 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2017 case RTE_VHOST_MSG_RESULT_REPLY:
2018 send_vhost_reply(fd, &msg);
2020 case RTE_VHOST_MSG_RESULT_ERR:
2021 case RTE_VHOST_MSG_RESULT_OK:
2023 goto skip_to_post_handle;
2024 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2030 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2031 if (!vhost_message_handlers[request])
2032 goto skip_to_post_handle;
2033 ret = vhost_message_handlers[request](&dev, &msg, fd);
2036 case RTE_VHOST_MSG_RESULT_ERR:
2037 RTE_LOG(ERR, VHOST_CONFIG,
2038 "Processing %s failed.\n",
2039 vhost_message_str[request]);
2042 case RTE_VHOST_MSG_RESULT_OK:
2043 RTE_LOG(DEBUG, VHOST_CONFIG,
2044 "Processing %s succeeded.\n",
2045 vhost_message_str[request]);
2048 case RTE_VHOST_MSG_RESULT_REPLY:
2049 RTE_LOG(DEBUG, VHOST_CONFIG,
2050 "Processing %s succeeded and needs reply.\n",
2051 vhost_message_str[request]);
2052 send_vhost_reply(fd, &msg);
2060 skip_to_post_handle:
2061 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2062 dev->extern_ops.post_msg_handle) {
2063 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2066 case RTE_VHOST_MSG_RESULT_REPLY:
2067 send_vhost_reply(fd, &msg);
2069 case RTE_VHOST_MSG_RESULT_ERR:
2070 case RTE_VHOST_MSG_RESULT_OK:
2072 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2078 if (unlock_required)
2079 vhost_user_unlock_all_queue_pairs(dev);
2081 /* If message was not handled at this stage, treat it as an error */
2083 RTE_LOG(ERR, VHOST_CONFIG,
2084 "vhost message (req: %d) was not handled.\n", request);
2085 ret = RTE_VHOST_MSG_RESULT_ERR;
2089 * If the request required a reply that was already sent,
2090 * this optional reply-ack won't be sent as the
2091 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2093 if (msg.flags & VHOST_USER_NEED_REPLY) {
2094 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2095 msg.size = sizeof(msg.payload.u64);
2097 send_vhost_reply(fd, &msg);
2098 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2099 RTE_LOG(ERR, VHOST_CONFIG,
2100 "vhost message handling failed.\n");
2104 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2105 dev->flags |= VIRTIO_DEV_READY;
2107 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2108 if (dev->dequeue_zero_copy) {
2109 RTE_LOG(INFO, VHOST_CONFIG,
2110 "dequeue zero copy is enabled\n");
2113 if (dev->notify_ops->new_device(dev->vid) == 0)
2114 dev->flags |= VIRTIO_DEV_RUNNING;
2118 did = dev->vdpa_dev_id;
2119 vdpa_dev = rte_vdpa_get_device(did);
2120 if (vdpa_dev && virtio_is_ready(dev) &&
2121 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2122 msg.request.master == VHOST_USER_SET_VRING_CALL) {
2123 if (vdpa_dev->ops->dev_conf)
2124 vdpa_dev->ops->dev_conf(dev->vid);
2125 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2131 static int process_slave_message_reply(struct virtio_net *dev,
2132 const struct VhostUserMsg *msg)
2134 struct VhostUserMsg msg_reply;
2137 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2140 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2145 if (msg_reply.request.slave != msg->request.slave) {
2146 RTE_LOG(ERR, VHOST_CONFIG,
2147 "Received unexpected msg type (%u), expected %u\n",
2148 msg_reply.request.slave, msg->request.slave);
2153 ret = msg_reply.payload.u64 ? -1 : 0;
2156 rte_spinlock_unlock(&dev->slave_req_lock);
2161 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2164 struct VhostUserMsg msg = {
2165 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2166 .flags = VHOST_USER_VERSION,
2167 .size = sizeof(msg.payload.iotlb),
2171 .type = VHOST_IOTLB_MISS,
2175 ret = send_vhost_message(dev->slave_req_fd, &msg);
2177 RTE_LOG(ERR, VHOST_CONFIG,
2178 "Failed to send IOTLB miss message (%d)\n",
2186 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2192 struct VhostUserMsg msg = {
2193 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2194 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2195 .size = sizeof(msg.payload.area),
2197 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2204 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2210 ret = send_vhost_slave_message(dev, &msg);
2212 RTE_LOG(ERR, VHOST_CONFIG,
2213 "Failed to set host notifier (%d)\n", ret);
2217 return process_slave_message_reply(dev, &msg);
2220 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2222 struct virtio_net *dev;
2223 struct rte_vdpa_device *vdpa_dev;
2224 int vfio_device_fd, did, ret = 0;
2225 uint64_t offset, size;
2228 dev = get_device(vid);
2232 did = dev->vdpa_dev_id;
2236 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2237 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2238 !(dev->protocol_features &
2239 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2240 !(dev->protocol_features &
2241 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2242 !(dev->protocol_features &
2243 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2246 vdpa_dev = rte_vdpa_get_device(did);
2250 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2251 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2253 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2254 if (vfio_device_fd < 0)
2258 for (i = 0; i < dev->nr_vring; i++) {
2259 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2265 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2266 vfio_device_fd, offset, size) < 0) {
2273 for (i = 0; i < dev->nr_vring; i++) {
2274 vhost_user_slave_set_vring_host_notifier(dev, i, -1,