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>
40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
41 #include <linux/memfd.h>
42 #define MEMFD_SUPPORTED
45 #include <rte_common.h>
46 #include <rte_malloc.h>
51 #include "vhost_user.h"
53 #define VIRTIO_MIN_MTU 68
54 #define VIRTIO_MAX_MTU 65535
56 #define INFLIGHT_ALIGNMENT 64
57 #define INFLIGHT_VERSION 0x1
59 static const char *vhost_message_str[VHOST_USER_MAX] = {
60 [VHOST_USER_NONE] = "VHOST_USER_NONE",
61 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
62 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
63 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
64 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
65 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
66 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
67 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
68 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
69 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
70 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
71 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
72 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
73 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
74 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
75 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
76 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
77 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
78 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
79 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
80 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
81 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
82 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
83 [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
84 [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
85 [VHOST_USER_POSTCOPY_ADVISE] = "VHOST_USER_POSTCOPY_ADVISE",
86 [VHOST_USER_POSTCOPY_LISTEN] = "VHOST_USER_POSTCOPY_LISTEN",
87 [VHOST_USER_POSTCOPY_END] = "VHOST_USER_POSTCOPY_END",
88 [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
89 [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
92 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
93 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
96 close_msg_fds(struct VhostUserMsg *msg)
100 for (i = 0; i < msg->fd_num; i++)
105 * Ensure the expected number of FDs is received,
106 * close all FDs and return an error if this is not the case.
109 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
111 if (msg->fd_num == expected_fds)
114 RTE_LOG(ERR, VHOST_CONFIG,
115 " Expect %d FDs for request %s, received %d\n",
117 vhost_message_str[msg->request.master],
131 ret = fstat(fd, &stat);
132 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
136 * Reclaim all the outstanding zmbufs for a virtqueue.
139 drain_zmbuf_list(struct vhost_virtqueue *vq)
141 struct zcopy_mbuf *zmbuf, *next;
143 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
144 zmbuf != NULL; zmbuf = next) {
145 next = TAILQ_NEXT(zmbuf, next);
147 while (!mbuf_is_consumed(zmbuf->mbuf))
150 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
151 restore_mbuf(zmbuf->mbuf);
152 rte_pktmbuf_free(zmbuf->mbuf);
159 free_mem_region(struct virtio_net *dev)
162 struct rte_vhost_mem_region *reg;
163 struct vhost_virtqueue *vq;
165 if (!dev || !dev->mem)
168 if (dev->dequeue_zero_copy) {
169 for (i = 0; i < dev->nr_vring; i++) {
170 vq = dev->virtqueue[i];
172 drain_zmbuf_list(vq);
176 for (i = 0; i < dev->mem->nregions; i++) {
177 reg = &dev->mem->regions[i];
178 if (reg->host_user_addr) {
179 munmap(reg->mmap_addr, reg->mmap_size);
186 vhost_backend_cleanup(struct virtio_net *dev)
189 free_mem_region(dev);
194 free(dev->guest_pages);
195 dev->guest_pages = NULL;
198 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
202 if (dev->inflight_info) {
203 if (dev->inflight_info->addr) {
204 munmap(dev->inflight_info->addr,
205 dev->inflight_info->size);
206 dev->inflight_info->addr = NULL;
209 if (dev->inflight_info->fd > 0) {
210 close(dev->inflight_info->fd);
211 dev->inflight_info->fd = -1;
214 free(dev->inflight_info);
215 dev->inflight_info = NULL;
218 if (dev->slave_req_fd >= 0) {
219 close(dev->slave_req_fd);
220 dev->slave_req_fd = -1;
223 if (dev->postcopy_ufd >= 0) {
224 close(dev->postcopy_ufd);
225 dev->postcopy_ufd = -1;
228 dev->postcopy_listening = 0;
232 * This function just returns success at the moment unless
233 * the device hasn't been initialised.
236 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
237 struct VhostUserMsg *msg,
238 int main_fd __rte_unused)
240 if (validate_msg_fds(msg, 0) != 0)
241 return RTE_VHOST_MSG_RESULT_ERR;
243 return RTE_VHOST_MSG_RESULT_OK;
247 vhost_user_reset_owner(struct virtio_net **pdev,
248 struct VhostUserMsg *msg,
249 int main_fd __rte_unused)
251 struct virtio_net *dev = *pdev;
253 if (validate_msg_fds(msg, 0) != 0)
254 return RTE_VHOST_MSG_RESULT_ERR;
256 vhost_destroy_device_notify(dev);
258 cleanup_device(dev, 0);
260 return RTE_VHOST_MSG_RESULT_OK;
264 * The features that we support are requested.
267 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
268 int main_fd __rte_unused)
270 struct virtio_net *dev = *pdev;
271 uint64_t features = 0;
273 if (validate_msg_fds(msg, 0) != 0)
274 return RTE_VHOST_MSG_RESULT_ERR;
276 rte_vhost_driver_get_features(dev->ifname, &features);
278 msg->payload.u64 = features;
279 msg->size = sizeof(msg->payload.u64);
282 return RTE_VHOST_MSG_RESULT_REPLY;
286 * The queue number that we support are requested.
289 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
290 int main_fd __rte_unused)
292 struct virtio_net *dev = *pdev;
293 uint32_t queue_num = 0;
295 if (validate_msg_fds(msg, 0) != 0)
296 return RTE_VHOST_MSG_RESULT_ERR;
298 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
300 msg->payload.u64 = (uint64_t)queue_num;
301 msg->size = sizeof(msg->payload.u64);
304 return RTE_VHOST_MSG_RESULT_REPLY;
308 * We receive the negotiated features supported by us and the virtio device.
311 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
312 int main_fd __rte_unused)
314 struct virtio_net *dev = *pdev;
315 uint64_t features = msg->payload.u64;
316 uint64_t vhost_features = 0;
317 struct rte_vdpa_device *vdpa_dev;
320 if (validate_msg_fds(msg, 0) != 0)
321 return RTE_VHOST_MSG_RESULT_ERR;
323 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
324 if (features & ~vhost_features) {
325 RTE_LOG(ERR, VHOST_CONFIG,
326 "(%d) received invalid negotiated features.\n",
328 return RTE_VHOST_MSG_RESULT_ERR;
331 if (dev->flags & VIRTIO_DEV_RUNNING) {
332 if (dev->features == features)
333 return RTE_VHOST_MSG_RESULT_OK;
336 * Error out if master tries to change features while device is
337 * in running state. The exception being VHOST_F_LOG_ALL, which
338 * is enabled when the live-migration starts.
340 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
341 RTE_LOG(ERR, VHOST_CONFIG,
342 "(%d) features changed while device is running.\n",
344 return RTE_VHOST_MSG_RESULT_ERR;
347 if (dev->notify_ops->features_changed)
348 dev->notify_ops->features_changed(dev->vid, features);
351 dev->features = features;
353 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
354 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
356 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
358 RTE_LOG(INFO, VHOST_CONFIG,
359 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
360 VHOST_LOG_DEBUG(VHOST_CONFIG,
361 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
363 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
364 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
366 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
367 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
369 * Remove all but first queue pair if MQ hasn't been
370 * negotiated. This is safe because the device is not
371 * running at this stage.
373 while (dev->nr_vring > 2) {
374 struct vhost_virtqueue *vq;
376 vq = dev->virtqueue[--dev->nr_vring];
380 dev->virtqueue[dev->nr_vring] = NULL;
382 cleanup_vq_inflight(dev, vq);
387 did = dev->vdpa_dev_id;
388 vdpa_dev = rte_vdpa_get_device(did);
389 if (vdpa_dev && vdpa_dev->ops->set_features)
390 vdpa_dev->ops->set_features(dev->vid);
392 return RTE_VHOST_MSG_RESULT_OK;
396 * The virtio device sends us the size of the descriptor ring.
399 vhost_user_set_vring_num(struct virtio_net **pdev,
400 struct VhostUserMsg *msg,
401 int main_fd __rte_unused)
403 struct virtio_net *dev = *pdev;
404 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
406 if (validate_msg_fds(msg, 0) != 0)
407 return RTE_VHOST_MSG_RESULT_ERR;
409 vq->size = msg->payload.state.num;
411 /* VIRTIO 1.0, 2.4 Virtqueues says:
413 * Queue Size value is always a power of 2. The maximum Queue Size
416 * VIRTIO 1.1 2.7 Virtqueues says:
418 * Packed virtqueues support up to 2^15 entries each.
420 if (!vq_is_packed(dev)) {
421 if (vq->size & (vq->size - 1)) {
422 RTE_LOG(ERR, VHOST_CONFIG,
423 "invalid virtqueue size %u\n", vq->size);
424 return RTE_VHOST_MSG_RESULT_ERR;
428 if (vq->size > 32768) {
429 RTE_LOG(ERR, VHOST_CONFIG,
430 "invalid virtqueue size %u\n", vq->size);
431 return RTE_VHOST_MSG_RESULT_ERR;
434 if (dev->dequeue_zero_copy) {
436 vq->last_zmbuf_idx = 0;
437 vq->zmbuf_size = vq->size;
439 rte_free(vq->zmbufs);
440 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
441 sizeof(struct zcopy_mbuf), 0);
442 if (vq->zmbufs == NULL) {
443 RTE_LOG(WARNING, VHOST_CONFIG,
444 "failed to allocate mem for zero copy; "
445 "zero copy is force disabled\n");
446 dev->dequeue_zero_copy = 0;
448 TAILQ_INIT(&vq->zmbuf_list);
451 if (vq_is_packed(dev)) {
452 if (vq->shadow_used_packed)
453 rte_free(vq->shadow_used_packed);
454 vq->shadow_used_packed = rte_malloc(NULL,
456 sizeof(struct vring_used_elem_packed),
457 RTE_CACHE_LINE_SIZE);
458 if (!vq->shadow_used_packed) {
459 RTE_LOG(ERR, VHOST_CONFIG,
460 "failed to allocate memory for shadow used ring.\n");
461 return RTE_VHOST_MSG_RESULT_ERR;
465 if (vq->shadow_used_split)
466 rte_free(vq->shadow_used_split);
467 vq->shadow_used_split = rte_malloc(NULL,
468 vq->size * sizeof(struct vring_used_elem),
469 RTE_CACHE_LINE_SIZE);
470 if (!vq->shadow_used_split) {
471 RTE_LOG(ERR, VHOST_CONFIG,
472 "failed to allocate memory for shadow used ring.\n");
473 return RTE_VHOST_MSG_RESULT_ERR;
477 if (vq->batch_copy_elems)
478 rte_free(vq->batch_copy_elems);
479 vq->batch_copy_elems = rte_malloc(NULL,
480 vq->size * sizeof(struct batch_copy_elem),
481 RTE_CACHE_LINE_SIZE);
482 if (!vq->batch_copy_elems) {
483 RTE_LOG(ERR, VHOST_CONFIG,
484 "failed to allocate memory for batching copy.\n");
485 return RTE_VHOST_MSG_RESULT_ERR;
488 return RTE_VHOST_MSG_RESULT_OK;
492 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
493 * same numa node as the memory of vring descriptor.
495 #ifdef RTE_LIBRTE_VHOST_NUMA
496 static struct virtio_net*
497 numa_realloc(struct virtio_net *dev, int index)
499 int oldnode, newnode;
500 struct virtio_net *old_dev;
501 struct vhost_virtqueue *old_vq, *vq;
502 struct zcopy_mbuf *new_zmbuf;
503 struct vring_used_elem *new_shadow_used_split;
504 struct vring_used_elem_packed *new_shadow_used_packed;
505 struct batch_copy_elem *new_batch_copy_elems;
508 if (dev->flags & VIRTIO_DEV_RUNNING)
512 vq = old_vq = dev->virtqueue[index];
514 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
515 MPOL_F_NODE | MPOL_F_ADDR);
517 /* check if we need to reallocate vq */
518 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
519 MPOL_F_NODE | MPOL_F_ADDR);
521 RTE_LOG(ERR, VHOST_CONFIG,
522 "Unable to get vq numa information.\n");
525 if (oldnode != newnode) {
526 RTE_LOG(INFO, VHOST_CONFIG,
527 "reallocate vq from %d to %d node\n", oldnode, newnode);
528 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
532 memcpy(vq, old_vq, sizeof(*vq));
533 TAILQ_INIT(&vq->zmbuf_list);
535 if (dev->dequeue_zero_copy) {
536 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
537 sizeof(struct zcopy_mbuf), 0, newnode);
539 rte_free(vq->zmbufs);
540 vq->zmbufs = new_zmbuf;
544 if (vq_is_packed(dev)) {
545 new_shadow_used_packed = rte_malloc_socket(NULL,
547 sizeof(struct vring_used_elem_packed),
550 if (new_shadow_used_packed) {
551 rte_free(vq->shadow_used_packed);
552 vq->shadow_used_packed = new_shadow_used_packed;
555 new_shadow_used_split = rte_malloc_socket(NULL,
557 sizeof(struct vring_used_elem),
560 if (new_shadow_used_split) {
561 rte_free(vq->shadow_used_split);
562 vq->shadow_used_split = new_shadow_used_split;
566 new_batch_copy_elems = rte_malloc_socket(NULL,
567 vq->size * sizeof(struct batch_copy_elem),
570 if (new_batch_copy_elems) {
571 rte_free(vq->batch_copy_elems);
572 vq->batch_copy_elems = new_batch_copy_elems;
578 /* check if we need to reallocate dev */
579 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
580 MPOL_F_NODE | MPOL_F_ADDR);
582 RTE_LOG(ERR, VHOST_CONFIG,
583 "Unable to get dev numa information.\n");
586 if (oldnode != newnode) {
587 RTE_LOG(INFO, VHOST_CONFIG,
588 "reallocate dev from %d to %d node\n",
590 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
596 memcpy(dev, old_dev, sizeof(*dev));
601 dev->virtqueue[index] = vq;
602 vhost_devices[dev->vid] = dev;
605 vhost_user_iotlb_init(dev, index);
610 static struct virtio_net*
611 numa_realloc(struct virtio_net *dev, int index __rte_unused)
617 /* Converts QEMU virtual address to Vhost virtual address. */
619 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
621 struct rte_vhost_mem_region *r;
624 if (unlikely(!dev || !dev->mem))
627 /* Find the region where the address lives. */
628 for (i = 0; i < dev->mem->nregions; i++) {
629 r = &dev->mem->regions[i];
631 if (qva >= r->guest_user_addr &&
632 qva < r->guest_user_addr + r->size) {
634 if (unlikely(*len > r->guest_user_addr + r->size - qva))
635 *len = r->guest_user_addr + r->size - qva;
637 return qva - r->guest_user_addr +
649 * Converts ring address to Vhost virtual address.
650 * If IOMMU is enabled, the ring address is a guest IO virtual address,
651 * else it is a QEMU virtual address.
654 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
655 uint64_t ra, uint64_t *size)
657 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
659 uint64_t req_size = *size;
661 vva = vhost_user_iotlb_cache_find(vq, ra,
662 size, VHOST_ACCESS_RW);
663 if (req_size != *size)
664 vhost_user_iotlb_miss(dev, (ra + *size),
670 return qva_to_vva(dev, ra, size);
674 * Converts vring log address to GPA
675 * If IOMMU is enabled, the log address is IOVA
676 * If IOMMU not enabled, the log address is already GPA
679 translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
682 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
683 const uint64_t exp_size = sizeof(struct vring_used) +
684 sizeof(struct vring_used_elem) * vq->size;
686 uint64_t size = exp_size;
688 hva = vhost_iova_to_vva(dev, vq, log_addr,
689 &size, VHOST_ACCESS_RW);
690 if (size != exp_size)
693 gpa = hva_to_gpa(dev, hva, exp_size);
695 RTE_LOG(ERR, VHOST_CONFIG,
696 "VQ: Failed to find GPA for log_addr: 0x%" PRIx64 " hva: 0x%" PRIx64 "\n",
706 static struct virtio_net *
707 translate_ring_addresses(struct virtio_net *dev, int vq_index)
709 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
710 struct vhost_vring_addr *addr = &vq->ring_addrs;
711 uint64_t len, expected_len;
713 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
715 translate_log_addr(dev, vq, addr->log_guest_addr);
716 if (vq->log_guest_addr == 0) {
717 RTE_LOG(DEBUG, VHOST_CONFIG,
718 "(%d) failed to map log_guest_addr.\n",
724 if (vq_is_packed(dev)) {
725 len = sizeof(struct vring_packed_desc) * vq->size;
726 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
727 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
728 if (vq->desc_packed == NULL ||
729 len != sizeof(struct vring_packed_desc) *
731 RTE_LOG(DEBUG, VHOST_CONFIG,
732 "(%d) failed to map desc_packed ring.\n",
737 dev = numa_realloc(dev, vq_index);
738 vq = dev->virtqueue[vq_index];
739 addr = &vq->ring_addrs;
741 len = sizeof(struct vring_packed_desc_event);
742 vq->driver_event = (struct vring_packed_desc_event *)
743 (uintptr_t)ring_addr_to_vva(dev,
744 vq, addr->avail_user_addr, &len);
745 if (vq->driver_event == NULL ||
746 len != sizeof(struct vring_packed_desc_event)) {
747 RTE_LOG(DEBUG, VHOST_CONFIG,
748 "(%d) failed to find driver area address.\n",
753 len = sizeof(struct vring_packed_desc_event);
754 vq->device_event = (struct vring_packed_desc_event *)
755 (uintptr_t)ring_addr_to_vva(dev,
756 vq, addr->used_user_addr, &len);
757 if (vq->device_event == NULL ||
758 len != sizeof(struct vring_packed_desc_event)) {
759 RTE_LOG(DEBUG, VHOST_CONFIG,
760 "(%d) failed to find device area address.\n",
769 /* The addresses are converted from QEMU virtual to Vhost virtual. */
770 if (vq->desc && vq->avail && vq->used)
773 len = sizeof(struct vring_desc) * vq->size;
774 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
775 vq, addr->desc_user_addr, &len);
776 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
777 RTE_LOG(DEBUG, VHOST_CONFIG,
778 "(%d) failed to map desc ring.\n",
783 dev = numa_realloc(dev, vq_index);
784 vq = dev->virtqueue[vq_index];
785 addr = &vq->ring_addrs;
787 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
788 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
789 len += sizeof(uint16_t);
791 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
792 vq, addr->avail_user_addr, &len);
793 if (vq->avail == 0 || len != expected_len) {
794 RTE_LOG(DEBUG, VHOST_CONFIG,
795 "(%d) failed to map avail ring.\n",
800 len = sizeof(struct vring_used) +
801 sizeof(struct vring_used_elem) * vq->size;
802 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
803 len += sizeof(uint16_t);
805 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
806 vq, addr->used_user_addr, &len);
807 if (vq->used == 0 || len != expected_len) {
808 RTE_LOG(DEBUG, VHOST_CONFIG,
809 "(%d) failed to map used ring.\n",
814 if (vq->last_used_idx != vq->used->idx) {
815 RTE_LOG(WARNING, VHOST_CONFIG,
816 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
817 "some packets maybe resent for Tx and dropped for Rx\n",
818 vq->last_used_idx, vq->used->idx);
819 vq->last_used_idx = vq->used->idx;
820 vq->last_avail_idx = vq->used->idx;
825 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
827 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
828 dev->vid, vq->avail);
829 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
831 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
832 dev->vid, vq->log_guest_addr);
838 * The virtio device sends us the desc, used and avail ring addresses.
839 * This function then converts these to our address space.
842 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
843 int main_fd __rte_unused)
845 struct virtio_net *dev = *pdev;
846 struct vhost_virtqueue *vq;
847 struct vhost_vring_addr *addr = &msg->payload.addr;
850 if (validate_msg_fds(msg, 0) != 0)
851 return RTE_VHOST_MSG_RESULT_ERR;
853 if (dev->mem == NULL)
854 return RTE_VHOST_MSG_RESULT_ERR;
856 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
857 vq = dev->virtqueue[msg->payload.addr.index];
859 access_ok = vq->access_ok;
862 * Rings addresses should not be interpreted as long as the ring is not
863 * started and enabled
865 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
867 vring_invalidate(dev, vq);
869 if ((vq->enabled && (dev->features &
870 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
872 dev = translate_ring_addresses(dev, msg->payload.addr.index);
874 return RTE_VHOST_MSG_RESULT_ERR;
879 return RTE_VHOST_MSG_RESULT_OK;
883 * The virtio device sends us the available ring last used index.
886 vhost_user_set_vring_base(struct virtio_net **pdev,
887 struct VhostUserMsg *msg,
888 int main_fd __rte_unused)
890 struct virtio_net *dev = *pdev;
891 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
892 uint64_t val = msg->payload.state.num;
894 if (validate_msg_fds(msg, 0) != 0)
895 return RTE_VHOST_MSG_RESULT_ERR;
897 if (vq_is_packed(dev)) {
899 * Bit[0:14]: avail index
900 * Bit[15]: avail wrap counter
902 vq->last_avail_idx = val & 0x7fff;
903 vq->avail_wrap_counter = !!(val & (0x1 << 15));
905 * Set used index to same value as available one, as
906 * their values should be the same since ring processing
907 * was stopped at get time.
909 vq->last_used_idx = vq->last_avail_idx;
910 vq->used_wrap_counter = vq->avail_wrap_counter;
912 vq->last_used_idx = msg->payload.state.num;
913 vq->last_avail_idx = msg->payload.state.num;
916 return RTE_VHOST_MSG_RESULT_OK;
920 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
921 uint64_t host_phys_addr, uint64_t size)
923 struct guest_page *page, *last_page;
924 struct guest_page *old_pages;
926 if (dev->nr_guest_pages == dev->max_guest_pages) {
927 dev->max_guest_pages *= 2;
928 old_pages = dev->guest_pages;
929 dev->guest_pages = realloc(dev->guest_pages,
930 dev->max_guest_pages * sizeof(*page));
931 if (!dev->guest_pages) {
932 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
938 if (dev->nr_guest_pages > 0) {
939 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
940 /* merge if the two pages are continuous */
941 if (host_phys_addr == last_page->host_phys_addr +
943 last_page->size += size;
948 page = &dev->guest_pages[dev->nr_guest_pages++];
949 page->guest_phys_addr = guest_phys_addr;
950 page->host_phys_addr = host_phys_addr;
957 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
960 uint64_t reg_size = reg->size;
961 uint64_t host_user_addr = reg->host_user_addr;
962 uint64_t guest_phys_addr = reg->guest_phys_addr;
963 uint64_t host_phys_addr;
966 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
967 size = page_size - (guest_phys_addr & (page_size - 1));
968 size = RTE_MIN(size, reg_size);
970 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
973 host_user_addr += size;
974 guest_phys_addr += size;
977 while (reg_size > 0) {
978 size = RTE_MIN(reg_size, page_size);
979 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
981 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
985 host_user_addr += size;
986 guest_phys_addr += size;
993 #ifdef RTE_LIBRTE_VHOST_DEBUG
994 /* TODO: enable it only in debug mode? */
996 dump_guest_pages(struct virtio_net *dev)
999 struct guest_page *page;
1001 for (i = 0; i < dev->nr_guest_pages; i++) {
1002 page = &dev->guest_pages[i];
1004 RTE_LOG(INFO, VHOST_CONFIG,
1005 "guest physical page region %u\n"
1006 "\t guest_phys_addr: %" PRIx64 "\n"
1007 "\t host_phys_addr : %" PRIx64 "\n"
1008 "\t size : %" PRIx64 "\n",
1010 page->guest_phys_addr,
1011 page->host_phys_addr,
1016 #define dump_guest_pages(dev)
1020 vhost_memory_changed(struct VhostUserMemory *new,
1021 struct rte_vhost_memory *old)
1025 if (new->nregions != old->nregions)
1028 for (i = 0; i < new->nregions; ++i) {
1029 VhostUserMemoryRegion *new_r = &new->regions[i];
1030 struct rte_vhost_mem_region *old_r = &old->regions[i];
1032 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1034 if (new_r->memory_size != old_r->size)
1036 if (new_r->userspace_addr != old_r->guest_user_addr)
1044 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1047 struct virtio_net *dev = *pdev;
1048 struct VhostUserMemory *memory = &msg->payload.memory;
1049 struct rte_vhost_mem_region *reg;
1052 uint64_t mmap_offset;
1058 if (validate_msg_fds(msg, memory->nregions) != 0)
1059 return RTE_VHOST_MSG_RESULT_ERR;
1061 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1062 RTE_LOG(ERR, VHOST_CONFIG,
1063 "too many memory regions (%u)\n", memory->nregions);
1064 return RTE_VHOST_MSG_RESULT_ERR;
1067 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1068 RTE_LOG(INFO, VHOST_CONFIG,
1069 "(%d) memory regions not changed\n", dev->vid);
1073 return RTE_VHOST_MSG_RESULT_OK;
1077 free_mem_region(dev);
1082 /* Flush IOTLB cache as previous HVAs are now invalid */
1083 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1084 for (i = 0; i < dev->nr_vring; i++)
1085 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1087 dev->nr_guest_pages = 0;
1088 if (!dev->guest_pages) {
1089 dev->max_guest_pages = 8;
1090 dev->guest_pages = malloc(dev->max_guest_pages *
1091 sizeof(struct guest_page));
1092 if (dev->guest_pages == NULL) {
1093 RTE_LOG(ERR, VHOST_CONFIG,
1094 "(%d) failed to allocate memory "
1095 "for dev->guest_pages\n",
1097 return RTE_VHOST_MSG_RESULT_ERR;
1101 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1102 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1103 if (dev->mem == NULL) {
1104 RTE_LOG(ERR, VHOST_CONFIG,
1105 "(%d) failed to allocate memory for dev->mem\n",
1107 return RTE_VHOST_MSG_RESULT_ERR;
1109 dev->mem->nregions = memory->nregions;
1111 for (i = 0; i < memory->nregions; i++) {
1113 reg = &dev->mem->regions[i];
1115 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1116 reg->guest_user_addr = memory->regions[i].userspace_addr;
1117 reg->size = memory->regions[i].memory_size;
1120 mmap_offset = memory->regions[i].mmap_offset;
1122 /* Check for memory_size + mmap_offset overflow */
1123 if (mmap_offset >= -reg->size) {
1124 RTE_LOG(ERR, VHOST_CONFIG,
1125 "mmap_offset (%#"PRIx64") and memory_size "
1126 "(%#"PRIx64") overflow\n",
1127 mmap_offset, reg->size);
1131 mmap_size = reg->size + mmap_offset;
1133 /* mmap() without flag of MAP_ANONYMOUS, should be called
1134 * with length argument aligned with hugepagesz at older
1135 * longterm version Linux, like 2.6.32 and 3.2.72, or
1136 * mmap() will fail with EINVAL.
1138 * to avoid failure, make sure in caller to keep length
1141 alignment = get_blk_size(fd);
1142 if (alignment == (uint64_t)-1) {
1143 RTE_LOG(ERR, VHOST_CONFIG,
1144 "couldn't get hugepage size through fstat\n");
1147 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1149 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
1150 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1151 MAP_SHARED | populate, fd, 0);
1153 if (mmap_addr == MAP_FAILED) {
1154 RTE_LOG(ERR, VHOST_CONFIG,
1155 "mmap region %u failed.\n", i);
1159 reg->mmap_addr = mmap_addr;
1160 reg->mmap_size = mmap_size;
1161 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1164 if (dev->dequeue_zero_copy)
1165 if (add_guest_pages(dev, reg, alignment) < 0) {
1166 RTE_LOG(ERR, VHOST_CONFIG,
1167 "adding guest pages to region %u failed.\n",
1172 RTE_LOG(INFO, VHOST_CONFIG,
1173 "guest memory region %u, size: 0x%" PRIx64 "\n"
1174 "\t guest physical addr: 0x%" PRIx64 "\n"
1175 "\t guest virtual addr: 0x%" PRIx64 "\n"
1176 "\t host virtual addr: 0x%" PRIx64 "\n"
1177 "\t mmap addr : 0x%" PRIx64 "\n"
1178 "\t mmap size : 0x%" PRIx64 "\n"
1179 "\t mmap align: 0x%" PRIx64 "\n"
1180 "\t mmap off : 0x%" PRIx64 "\n",
1182 reg->guest_phys_addr,
1183 reg->guest_user_addr,
1184 reg->host_user_addr,
1185 (uint64_t)(uintptr_t)mmap_addr,
1190 if (dev->postcopy_listening) {
1192 * We haven't a better way right now than sharing
1193 * DPDK's virtual address with Qemu, so that Qemu can
1194 * retrieve the region offset when handling userfaults.
1196 memory->regions[i].userspace_addr =
1197 reg->host_user_addr;
1200 if (dev->postcopy_listening) {
1201 /* Send the addresses back to qemu */
1203 send_vhost_reply(main_fd, msg);
1205 /* Wait for qemu to acknolwedge it's got the addresses
1206 * we've got to wait before we're allowed to generate faults.
1208 VhostUserMsg ack_msg;
1209 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1210 RTE_LOG(ERR, VHOST_CONFIG,
1211 "Failed to read qemu ack on postcopy set-mem-table\n");
1215 if (validate_msg_fds(&ack_msg, 0) != 0)
1218 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1219 RTE_LOG(ERR, VHOST_CONFIG,
1220 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1221 ack_msg.request.master);
1225 /* Now userfault register and we can use the memory */
1226 for (i = 0; i < memory->nregions; i++) {
1227 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1228 reg = &dev->mem->regions[i];
1229 struct uffdio_register reg_struct;
1232 * Let's register all the mmap'ed area to ensure
1233 * alignment on page boundary.
1235 reg_struct.range.start =
1236 (uint64_t)(uintptr_t)reg->mmap_addr;
1237 reg_struct.range.len = reg->mmap_size;
1238 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1240 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1242 RTE_LOG(ERR, VHOST_CONFIG,
1243 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1244 i, dev->postcopy_ufd,
1248 RTE_LOG(INFO, VHOST_CONFIG,
1249 "\t userfaultfd registered for range : "
1250 "%" PRIx64 " - %" PRIx64 "\n",
1251 (uint64_t)reg_struct.range.start,
1252 (uint64_t)reg_struct.range.start +
1253 (uint64_t)reg_struct.range.len - 1);
1260 for (i = 0; i < dev->nr_vring; i++) {
1261 struct vhost_virtqueue *vq = dev->virtqueue[i];
1263 if (vq->desc || vq->avail || vq->used) {
1265 * If the memory table got updated, the ring addresses
1266 * need to be translated again as virtual addresses have
1269 vring_invalidate(dev, vq);
1271 dev = translate_ring_addresses(dev, i);
1281 dump_guest_pages(dev);
1283 return RTE_VHOST_MSG_RESULT_OK;
1286 free_mem_region(dev);
1289 return RTE_VHOST_MSG_RESULT_ERR;
1293 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1300 if (vq_is_packed(dev))
1301 rings_ok = !!vq->desc_packed;
1303 rings_ok = vq->desc && vq->avail && vq->used;
1306 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1307 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1311 virtio_is_ready(struct virtio_net *dev)
1313 struct vhost_virtqueue *vq;
1316 if (dev->nr_vring == 0)
1319 for (i = 0; i < dev->nr_vring; i++) {
1320 vq = dev->virtqueue[i];
1322 if (!vq_is_ready(dev, vq))
1326 RTE_LOG(INFO, VHOST_CONFIG,
1327 "virtio is now ready for processing.\n");
1332 inflight_mem_alloc(const char *name, size_t size, int *fd)
1336 char fname[20] = "/tmp/memfd-XXXXXX";
1339 #ifdef MEMFD_SUPPORTED
1340 mfd = memfd_create(name, MFD_CLOEXEC);
1345 mfd = mkstemp(fname);
1347 RTE_LOG(ERR, VHOST_CONFIG,
1348 "failed to get inflight buffer fd\n");
1355 if (ftruncate(mfd, size) == -1) {
1356 RTE_LOG(ERR, VHOST_CONFIG,
1357 "failed to alloc inflight buffer\n");
1362 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1363 if (ptr == MAP_FAILED) {
1364 RTE_LOG(ERR, VHOST_CONFIG,
1365 "failed to mmap inflight buffer\n");
1375 get_pervq_shm_size_split(uint16_t queue_size)
1377 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1378 queue_size + sizeof(uint64_t) +
1379 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1383 get_pervq_shm_size_packed(uint16_t queue_size)
1385 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1386 * queue_size + sizeof(uint64_t) +
1387 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1388 INFLIGHT_ALIGNMENT);
1392 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1394 int main_fd __rte_unused)
1396 struct rte_vhost_inflight_info_packed *inflight_packed;
1397 uint64_t pervq_inflight_size, mmap_size;
1398 uint16_t num_queues, queue_size;
1399 struct virtio_net *dev = *pdev;
1403 if (msg->size != sizeof(msg->payload.inflight)) {
1404 RTE_LOG(ERR, VHOST_CONFIG,
1405 "invalid get_inflight_fd message size is %d\n",
1407 return RTE_VHOST_MSG_RESULT_ERR;
1410 if (dev->inflight_info == NULL) {
1411 dev->inflight_info = calloc(1,
1412 sizeof(struct inflight_mem_info));
1413 if (!dev->inflight_info) {
1414 RTE_LOG(ERR, VHOST_CONFIG,
1415 "failed to alloc dev inflight area\n");
1416 return RTE_VHOST_MSG_RESULT_ERR;
1420 num_queues = msg->payload.inflight.num_queues;
1421 queue_size = msg->payload.inflight.queue_size;
1423 RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
1424 msg->payload.inflight.num_queues);
1425 RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
1426 msg->payload.inflight.queue_size);
1428 if (vq_is_packed(dev))
1429 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1431 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1433 mmap_size = num_queues * pervq_inflight_size;
1434 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1436 RTE_LOG(ERR, VHOST_CONFIG,
1437 "failed to alloc vhost inflight area\n");
1438 msg->payload.inflight.mmap_size = 0;
1439 return RTE_VHOST_MSG_RESULT_ERR;
1441 memset(addr, 0, mmap_size);
1443 dev->inflight_info->addr = addr;
1444 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1445 dev->inflight_info->fd = msg->fds[0] = fd;
1446 msg->payload.inflight.mmap_offset = 0;
1449 if (vq_is_packed(dev)) {
1450 for (i = 0; i < num_queues; i++) {
1452 (struct rte_vhost_inflight_info_packed *)addr;
1453 inflight_packed->used_wrap_counter = 1;
1454 inflight_packed->old_used_wrap_counter = 1;
1455 for (j = 0; j < queue_size; j++)
1456 inflight_packed->desc[j].next = j + 1;
1457 addr = (void *)((char *)addr + pervq_inflight_size);
1461 RTE_LOG(INFO, VHOST_CONFIG,
1462 "send inflight mmap_size: %"PRIu64"\n",
1463 msg->payload.inflight.mmap_size);
1464 RTE_LOG(INFO, VHOST_CONFIG,
1465 "send inflight mmap_offset: %"PRIu64"\n",
1466 msg->payload.inflight.mmap_offset);
1467 RTE_LOG(INFO, VHOST_CONFIG,
1468 "send inflight fd: %d\n", msg->fds[0]);
1470 return RTE_VHOST_MSG_RESULT_REPLY;
1474 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1475 int main_fd __rte_unused)
1477 uint64_t mmap_size, mmap_offset;
1478 uint16_t num_queues, queue_size;
1479 struct virtio_net *dev = *pdev;
1480 uint32_t pervq_inflight_size;
1481 struct vhost_virtqueue *vq;
1486 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1487 RTE_LOG(ERR, VHOST_CONFIG,
1488 "invalid set_inflight_fd message size is %d,fd is %d\n",
1490 return RTE_VHOST_MSG_RESULT_ERR;
1493 mmap_size = msg->payload.inflight.mmap_size;
1494 mmap_offset = msg->payload.inflight.mmap_offset;
1495 num_queues = msg->payload.inflight.num_queues;
1496 queue_size = msg->payload.inflight.queue_size;
1498 if (vq_is_packed(dev))
1499 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1501 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1503 RTE_LOG(INFO, VHOST_CONFIG,
1504 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1505 RTE_LOG(INFO, VHOST_CONFIG,
1506 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1507 RTE_LOG(INFO, VHOST_CONFIG,
1508 "set_inflight_fd num_queues: %u\n", num_queues);
1509 RTE_LOG(INFO, VHOST_CONFIG,
1510 "set_inflight_fd queue_size: %u\n", queue_size);
1511 RTE_LOG(INFO, VHOST_CONFIG,
1512 "set_inflight_fd fd: %d\n", fd);
1513 RTE_LOG(INFO, VHOST_CONFIG,
1514 "set_inflight_fd pervq_inflight_size: %d\n",
1515 pervq_inflight_size);
1517 if (!dev->inflight_info) {
1518 dev->inflight_info = calloc(1,
1519 sizeof(struct inflight_mem_info));
1520 if (dev->inflight_info == NULL) {
1521 RTE_LOG(ERR, VHOST_CONFIG,
1522 "failed to alloc dev inflight area\n");
1523 return RTE_VHOST_MSG_RESULT_ERR;
1527 if (dev->inflight_info->addr)
1528 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1530 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1532 if (addr == MAP_FAILED) {
1533 RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
1534 return RTE_VHOST_MSG_RESULT_ERR;
1537 if (dev->inflight_info->fd)
1538 close(dev->inflight_info->fd);
1540 dev->inflight_info->fd = fd;
1541 dev->inflight_info->addr = addr;
1542 dev->inflight_info->size = mmap_size;
1544 for (i = 0; i < num_queues; i++) {
1545 vq = dev->virtqueue[i];
1546 if (vq_is_packed(dev)) {
1547 vq->inflight_packed = addr;
1548 vq->inflight_packed->desc_num = queue_size;
1550 vq->inflight_split = addr;
1551 vq->inflight_split->desc_num = queue_size;
1553 addr = (void *)((char *)addr + pervq_inflight_size);
1556 return RTE_VHOST_MSG_RESULT_OK;
1560 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1561 int main_fd __rte_unused)
1563 struct virtio_net *dev = *pdev;
1564 struct vhost_vring_file file;
1565 struct vhost_virtqueue *vq;
1567 if (validate_msg_fds(msg, 1) != 0)
1568 return RTE_VHOST_MSG_RESULT_ERR;
1570 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1571 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1572 file.fd = VIRTIO_INVALID_EVENTFD;
1574 file.fd = msg->fds[0];
1575 RTE_LOG(INFO, VHOST_CONFIG,
1576 "vring call idx:%d file:%d\n", file.index, file.fd);
1578 vq = dev->virtqueue[file.index];
1579 if (vq->callfd >= 0)
1582 vq->callfd = file.fd;
1584 return RTE_VHOST_MSG_RESULT_OK;
1587 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1588 struct VhostUserMsg *msg,
1589 int main_fd __rte_unused)
1591 if (validate_msg_fds(msg, 1) != 0)
1592 return RTE_VHOST_MSG_RESULT_ERR;
1594 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1596 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1598 return RTE_VHOST_MSG_RESULT_OK;
1602 resubmit_desc_compare(const void *a, const void *b)
1604 const struct rte_vhost_resubmit_desc *desc0 = a;
1605 const struct rte_vhost_resubmit_desc *desc1 = b;
1607 if (desc1->counter > desc0->counter)
1614 vhost_check_queue_inflights_split(struct virtio_net *dev,
1615 struct vhost_virtqueue *vq)
1618 uint16_t resubmit_num = 0, last_io, num;
1619 struct vring_used *used = vq->used;
1620 struct rte_vhost_resubmit_info *resubmit;
1621 struct rte_vhost_inflight_info_split *inflight_split;
1623 if (!(dev->protocol_features &
1624 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1625 return RTE_VHOST_MSG_RESULT_OK;
1627 if ((!vq->inflight_split))
1628 return RTE_VHOST_MSG_RESULT_ERR;
1630 if (!vq->inflight_split->version) {
1631 vq->inflight_split->version = INFLIGHT_VERSION;
1632 return RTE_VHOST_MSG_RESULT_OK;
1635 if (vq->resubmit_inflight)
1636 return RTE_VHOST_MSG_RESULT_OK;
1638 inflight_split = vq->inflight_split;
1639 vq->global_counter = 0;
1640 last_io = inflight_split->last_inflight_io;
1642 if (inflight_split->used_idx != used->idx) {
1643 inflight_split->desc[last_io].inflight = 0;
1645 inflight_split->used_idx = used->idx;
1648 for (i = 0; i < inflight_split->desc_num; i++) {
1649 if (inflight_split->desc[i].inflight == 1)
1653 vq->last_avail_idx += resubmit_num;
1656 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1658 RTE_LOG(ERR, VHOST_CONFIG,
1659 "failed to allocate memory for resubmit info.\n");
1660 return RTE_VHOST_MSG_RESULT_ERR;
1663 resubmit->resubmit_list = calloc(resubmit_num,
1664 sizeof(struct rte_vhost_resubmit_desc));
1665 if (!resubmit->resubmit_list) {
1666 RTE_LOG(ERR, VHOST_CONFIG,
1667 "failed to allocate memory for inflight desc.\n");
1669 return RTE_VHOST_MSG_RESULT_ERR;
1673 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1674 if (vq->inflight_split->desc[i].inflight == 1) {
1675 resubmit->resubmit_list[num].index = i;
1676 resubmit->resubmit_list[num].counter =
1677 inflight_split->desc[i].counter;
1681 resubmit->resubmit_num = num;
1683 if (resubmit->resubmit_num > 1)
1684 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1685 sizeof(struct rte_vhost_resubmit_desc),
1686 resubmit_desc_compare);
1688 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1689 vq->resubmit_inflight = resubmit;
1692 return RTE_VHOST_MSG_RESULT_OK;
1696 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1697 struct vhost_virtqueue *vq)
1700 uint16_t resubmit_num = 0, old_used_idx, num;
1701 struct rte_vhost_resubmit_info *resubmit;
1702 struct rte_vhost_inflight_info_packed *inflight_packed;
1704 if (!(dev->protocol_features &
1705 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1706 return RTE_VHOST_MSG_RESULT_OK;
1708 if ((!vq->inflight_packed))
1709 return RTE_VHOST_MSG_RESULT_ERR;
1711 if (!vq->inflight_packed->version) {
1712 vq->inflight_packed->version = INFLIGHT_VERSION;
1713 return RTE_VHOST_MSG_RESULT_OK;
1716 if (vq->resubmit_inflight)
1717 return RTE_VHOST_MSG_RESULT_OK;
1719 inflight_packed = vq->inflight_packed;
1720 vq->global_counter = 0;
1721 old_used_idx = inflight_packed->old_used_idx;
1723 if (inflight_packed->used_idx != old_used_idx) {
1724 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1725 inflight_packed->old_used_idx =
1726 inflight_packed->used_idx;
1727 inflight_packed->old_used_wrap_counter =
1728 inflight_packed->used_wrap_counter;
1729 inflight_packed->old_free_head =
1730 inflight_packed->free_head;
1732 inflight_packed->used_idx =
1733 inflight_packed->old_used_idx;
1734 inflight_packed->used_wrap_counter =
1735 inflight_packed->old_used_wrap_counter;
1736 inflight_packed->free_head =
1737 inflight_packed->old_free_head;
1741 for (i = 0; i < inflight_packed->desc_num; i++) {
1742 if (inflight_packed->desc[i].inflight == 1)
1747 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1748 if (resubmit == NULL) {
1749 RTE_LOG(ERR, VHOST_CONFIG,
1750 "failed to allocate memory for resubmit info.\n");
1751 return RTE_VHOST_MSG_RESULT_ERR;
1754 resubmit->resubmit_list = calloc(resubmit_num,
1755 sizeof(struct rte_vhost_resubmit_desc));
1756 if (resubmit->resubmit_list == NULL) {
1757 RTE_LOG(ERR, VHOST_CONFIG,
1758 "failed to allocate memory for resubmit desc.\n");
1760 return RTE_VHOST_MSG_RESULT_ERR;
1764 for (i = 0; i < inflight_packed->desc_num; i++) {
1765 if (vq->inflight_packed->desc[i].inflight == 1) {
1766 resubmit->resubmit_list[num].index = i;
1767 resubmit->resubmit_list[num].counter =
1768 inflight_packed->desc[i].counter;
1772 resubmit->resubmit_num = num;
1774 if (resubmit->resubmit_num > 1)
1775 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1776 sizeof(struct rte_vhost_resubmit_desc),
1777 resubmit_desc_compare);
1779 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1780 vq->resubmit_inflight = resubmit;
1783 return RTE_VHOST_MSG_RESULT_OK;
1787 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1788 int main_fd __rte_unused)
1790 struct virtio_net *dev = *pdev;
1791 struct vhost_vring_file file;
1792 struct vhost_virtqueue *vq;
1794 if (validate_msg_fds(msg, 1) != 0)
1795 return RTE_VHOST_MSG_RESULT_ERR;
1797 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1798 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1799 file.fd = VIRTIO_INVALID_EVENTFD;
1801 file.fd = msg->fds[0];
1802 RTE_LOG(INFO, VHOST_CONFIG,
1803 "vring kick idx:%d file:%d\n", file.index, file.fd);
1805 /* Interpret ring addresses only when ring is started. */
1806 dev = translate_ring_addresses(dev, file.index);
1808 return RTE_VHOST_MSG_RESULT_ERR;
1812 vq = dev->virtqueue[file.index];
1815 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1816 * the ring starts already enabled. Otherwise, it is enabled via
1817 * the SET_VRING_ENABLE message.
1819 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1821 if (dev->notify_ops->vring_state_changed)
1822 dev->notify_ops->vring_state_changed(
1823 dev->vid, file.index, 1);
1826 if (vq->kickfd >= 0)
1828 vq->kickfd = file.fd;
1830 if (vq_is_packed(dev)) {
1831 if (vhost_check_queue_inflights_packed(dev, vq)) {
1832 RTE_LOG(ERR, VHOST_CONFIG,
1833 "failed to inflights for vq: %d\n", file.index);
1834 return RTE_VHOST_MSG_RESULT_ERR;
1837 if (vhost_check_queue_inflights_split(dev, vq)) {
1838 RTE_LOG(ERR, VHOST_CONFIG,
1839 "failed to inflights for vq: %d\n", file.index);
1840 return RTE_VHOST_MSG_RESULT_ERR;
1844 return RTE_VHOST_MSG_RESULT_OK;
1848 free_zmbufs(struct vhost_virtqueue *vq)
1850 drain_zmbuf_list(vq);
1852 rte_free(vq->zmbufs);
1856 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1859 vhost_user_get_vring_base(struct virtio_net **pdev,
1860 struct VhostUserMsg *msg,
1861 int main_fd __rte_unused)
1863 struct virtio_net *dev = *pdev;
1864 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1867 if (validate_msg_fds(msg, 0) != 0)
1868 return RTE_VHOST_MSG_RESULT_ERR;
1870 /* We have to stop the queue (virtio) if it is running. */
1871 vhost_destroy_device_notify(dev);
1873 dev->flags &= ~VIRTIO_DEV_READY;
1874 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1876 /* Here we are safe to get the indexes */
1877 if (vq_is_packed(dev)) {
1879 * Bit[0:14]: avail index
1880 * Bit[15]: avail wrap counter
1882 val = vq->last_avail_idx & 0x7fff;
1883 val |= vq->avail_wrap_counter << 15;
1884 msg->payload.state.num = val;
1886 msg->payload.state.num = vq->last_avail_idx;
1889 RTE_LOG(INFO, VHOST_CONFIG,
1890 "vring base idx:%d file:%d\n", msg->payload.state.index,
1891 msg->payload.state.num);
1893 * Based on current qemu vhost-user implementation, this message is
1894 * sent and only sent in vhost_vring_stop.
1895 * TODO: cleanup the vring, it isn't usable since here.
1897 if (vq->kickfd >= 0)
1900 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1902 if (vq->callfd >= 0)
1905 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1907 vq->signalled_used_valid = false;
1909 if (dev->dequeue_zero_copy)
1911 if (vq_is_packed(dev)) {
1912 rte_free(vq->shadow_used_packed);
1913 vq->shadow_used_packed = NULL;
1915 rte_free(vq->shadow_used_split);
1916 vq->shadow_used_split = NULL;
1919 rte_free(vq->batch_copy_elems);
1920 vq->batch_copy_elems = NULL;
1922 msg->size = sizeof(msg->payload.state);
1925 vring_invalidate(dev, vq);
1927 return RTE_VHOST_MSG_RESULT_REPLY;
1931 * when virtio queues are ready to work, qemu will send us to
1932 * enable the virtio queue pair.
1935 vhost_user_set_vring_enable(struct virtio_net **pdev,
1936 struct VhostUserMsg *msg,
1937 int main_fd __rte_unused)
1939 struct virtio_net *dev = *pdev;
1940 int enable = (int)msg->payload.state.num;
1941 int index = (int)msg->payload.state.index;
1942 struct rte_vdpa_device *vdpa_dev;
1945 if (validate_msg_fds(msg, 0) != 0)
1946 return RTE_VHOST_MSG_RESULT_ERR;
1948 RTE_LOG(INFO, VHOST_CONFIG,
1949 "set queue enable: %d to qp idx: %d\n",
1952 did = dev->vdpa_dev_id;
1953 vdpa_dev = rte_vdpa_get_device(did);
1954 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1955 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1957 if (dev->notify_ops->vring_state_changed)
1958 dev->notify_ops->vring_state_changed(dev->vid,
1961 /* On disable, rings have to be stopped being processed. */
1962 if (!enable && dev->dequeue_zero_copy)
1963 drain_zmbuf_list(dev->virtqueue[index]);
1965 dev->virtqueue[index]->enabled = enable;
1967 return RTE_VHOST_MSG_RESULT_OK;
1971 vhost_user_get_protocol_features(struct virtio_net **pdev,
1972 struct VhostUserMsg *msg,
1973 int main_fd __rte_unused)
1975 struct virtio_net *dev = *pdev;
1976 uint64_t features, protocol_features;
1978 if (validate_msg_fds(msg, 0) != 0)
1979 return RTE_VHOST_MSG_RESULT_ERR;
1981 rte_vhost_driver_get_features(dev->ifname, &features);
1982 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1985 * REPLY_ACK protocol feature is only mandatory for now
1986 * for IOMMU feature. If IOMMU is explicitly disabled by the
1987 * application, disable also REPLY_ACK feature for older buggy
1988 * Qemu versions (from v2.7.0 to v2.9.0).
1990 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1991 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1993 msg->payload.u64 = protocol_features;
1994 msg->size = sizeof(msg->payload.u64);
1997 return RTE_VHOST_MSG_RESULT_REPLY;
2001 vhost_user_set_protocol_features(struct virtio_net **pdev,
2002 struct VhostUserMsg *msg,
2003 int main_fd __rte_unused)
2005 struct virtio_net *dev = *pdev;
2006 uint64_t protocol_features = msg->payload.u64;
2007 uint64_t slave_protocol_features = 0;
2009 if (validate_msg_fds(msg, 0) != 0)
2010 return RTE_VHOST_MSG_RESULT_ERR;
2012 rte_vhost_driver_get_protocol_features(dev->ifname,
2013 &slave_protocol_features);
2014 if (protocol_features & ~slave_protocol_features) {
2015 RTE_LOG(ERR, VHOST_CONFIG,
2016 "(%d) received invalid protocol features.\n",
2018 return RTE_VHOST_MSG_RESULT_ERR;
2021 dev->protocol_features = protocol_features;
2022 RTE_LOG(INFO, VHOST_CONFIG,
2023 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2024 dev->protocol_features);
2026 return RTE_VHOST_MSG_RESULT_OK;
2030 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2031 int main_fd __rte_unused)
2033 struct virtio_net *dev = *pdev;
2034 int fd = msg->fds[0];
2038 if (validate_msg_fds(msg, 1) != 0)
2039 return RTE_VHOST_MSG_RESULT_ERR;
2042 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
2043 return RTE_VHOST_MSG_RESULT_ERR;
2046 if (msg->size != sizeof(VhostUserLog)) {
2047 RTE_LOG(ERR, VHOST_CONFIG,
2048 "invalid log base msg size: %"PRId32" != %d\n",
2049 msg->size, (int)sizeof(VhostUserLog));
2050 return RTE_VHOST_MSG_RESULT_ERR;
2053 size = msg->payload.log.mmap_size;
2054 off = msg->payload.log.mmap_offset;
2056 /* Don't allow mmap_offset to point outside the mmap region */
2058 RTE_LOG(ERR, VHOST_CONFIG,
2059 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
2061 return RTE_VHOST_MSG_RESULT_ERR;
2064 RTE_LOG(INFO, VHOST_CONFIG,
2065 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2069 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2070 * fail when offset is not page size aligned.
2072 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2074 if (addr == MAP_FAILED) {
2075 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
2076 return RTE_VHOST_MSG_RESULT_ERR;
2080 * Free previously mapped log memory on occasionally
2081 * multiple VHOST_USER_SET_LOG_BASE.
2083 if (dev->log_addr) {
2084 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2086 dev->log_addr = (uint64_t)(uintptr_t)addr;
2087 dev->log_base = dev->log_addr + off;
2088 dev->log_size = size;
2091 * The spec is not clear about it (yet), but QEMU doesn't expect
2092 * any payload in the reply.
2097 return RTE_VHOST_MSG_RESULT_REPLY;
2100 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2101 struct VhostUserMsg *msg,
2102 int main_fd __rte_unused)
2104 if (validate_msg_fds(msg, 1) != 0)
2105 return RTE_VHOST_MSG_RESULT_ERR;
2108 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
2110 return RTE_VHOST_MSG_RESULT_OK;
2114 * An rarp packet is constructed and broadcasted to notify switches about
2115 * the new location of the migrated VM, so that packets from outside will
2116 * not be lost after migration.
2118 * However, we don't actually "send" a rarp packet here, instead, we set
2119 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2122 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2123 int main_fd __rte_unused)
2125 struct virtio_net *dev = *pdev;
2126 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2127 struct rte_vdpa_device *vdpa_dev;
2130 if (validate_msg_fds(msg, 0) != 0)
2131 return RTE_VHOST_MSG_RESULT_ERR;
2133 RTE_LOG(DEBUG, VHOST_CONFIG,
2134 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2135 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2136 memcpy(dev->mac.addr_bytes, mac, 6);
2139 * Set the flag to inject a RARP broadcast packet at
2140 * rte_vhost_dequeue_burst().
2142 * rte_smp_wmb() is for making sure the mac is copied
2143 * before the flag is set.
2146 rte_atomic16_set(&dev->broadcast_rarp, 1);
2147 did = dev->vdpa_dev_id;
2148 vdpa_dev = rte_vdpa_get_device(did);
2149 if (vdpa_dev && vdpa_dev->ops->migration_done)
2150 vdpa_dev->ops->migration_done(dev->vid);
2152 return RTE_VHOST_MSG_RESULT_OK;
2156 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2157 int main_fd __rte_unused)
2159 struct virtio_net *dev = *pdev;
2161 if (validate_msg_fds(msg, 0) != 0)
2162 return RTE_VHOST_MSG_RESULT_ERR;
2164 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2165 msg->payload.u64 > VIRTIO_MAX_MTU) {
2166 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
2169 return RTE_VHOST_MSG_RESULT_ERR;
2172 dev->mtu = msg->payload.u64;
2174 return RTE_VHOST_MSG_RESULT_OK;
2178 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2179 int main_fd __rte_unused)
2181 struct virtio_net *dev = *pdev;
2182 int fd = msg->fds[0];
2184 if (validate_msg_fds(msg, 1) != 0)
2185 return RTE_VHOST_MSG_RESULT_ERR;
2188 RTE_LOG(ERR, VHOST_CONFIG,
2189 "Invalid file descriptor for slave channel (%d)\n",
2191 return RTE_VHOST_MSG_RESULT_ERR;
2194 if (dev->slave_req_fd >= 0)
2195 close(dev->slave_req_fd);
2197 dev->slave_req_fd = fd;
2199 return RTE_VHOST_MSG_RESULT_OK;
2203 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2205 struct vhost_vring_addr *ra;
2206 uint64_t start, end, len;
2209 end = start + imsg->size;
2211 ra = &vq->ring_addrs;
2212 len = sizeof(struct vring_desc) * vq->size;
2213 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2216 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2217 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2220 len = sizeof(struct vring_used) +
2221 sizeof(struct vring_used_elem) * vq->size;
2222 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2229 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2231 struct vhost_vring_addr *ra;
2232 uint64_t start, end, len;
2235 end = start + imsg->size;
2237 ra = &vq->ring_addrs;
2238 len = sizeof(struct vring_packed_desc) * vq->size;
2239 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2242 len = sizeof(struct vring_packed_desc_event);
2243 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2246 len = sizeof(struct vring_packed_desc_event);
2247 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2253 static int is_vring_iotlb(struct virtio_net *dev,
2254 struct vhost_virtqueue *vq,
2255 struct vhost_iotlb_msg *imsg)
2257 if (vq_is_packed(dev))
2258 return is_vring_iotlb_packed(vq, imsg);
2260 return is_vring_iotlb_split(vq, imsg);
2264 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2265 int main_fd __rte_unused)
2267 struct virtio_net *dev = *pdev;
2268 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2272 if (validate_msg_fds(msg, 0) != 0)
2273 return RTE_VHOST_MSG_RESULT_ERR;
2275 switch (imsg->type) {
2276 case VHOST_IOTLB_UPDATE:
2278 vva = qva_to_vva(dev, imsg->uaddr, &len);
2280 return RTE_VHOST_MSG_RESULT_ERR;
2282 for (i = 0; i < dev->nr_vring; i++) {
2283 struct vhost_virtqueue *vq = dev->virtqueue[i];
2285 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2288 if (is_vring_iotlb(dev, vq, imsg))
2289 *pdev = dev = translate_ring_addresses(dev, i);
2292 case VHOST_IOTLB_INVALIDATE:
2293 for (i = 0; i < dev->nr_vring; i++) {
2294 struct vhost_virtqueue *vq = dev->virtqueue[i];
2296 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2299 if (is_vring_iotlb(dev, vq, imsg))
2300 vring_invalidate(dev, vq);
2304 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
2306 return RTE_VHOST_MSG_RESULT_ERR;
2309 return RTE_VHOST_MSG_RESULT_OK;
2313 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2314 struct VhostUserMsg *msg,
2315 int main_fd __rte_unused)
2317 struct virtio_net *dev = *pdev;
2318 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2319 struct uffdio_api api_struct;
2321 if (validate_msg_fds(msg, 0) != 0)
2322 return RTE_VHOST_MSG_RESULT_ERR;
2324 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2326 if (dev->postcopy_ufd == -1) {
2327 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
2329 return RTE_VHOST_MSG_RESULT_ERR;
2331 api_struct.api = UFFD_API;
2332 api_struct.features = 0;
2333 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2334 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
2336 close(dev->postcopy_ufd);
2337 dev->postcopy_ufd = -1;
2338 return RTE_VHOST_MSG_RESULT_ERR;
2340 msg->fds[0] = dev->postcopy_ufd;
2343 return RTE_VHOST_MSG_RESULT_REPLY;
2345 dev->postcopy_ufd = -1;
2348 return RTE_VHOST_MSG_RESULT_ERR;
2353 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2354 struct VhostUserMsg *msg __rte_unused,
2355 int main_fd __rte_unused)
2357 struct virtio_net *dev = *pdev;
2359 if (validate_msg_fds(msg, 0) != 0)
2360 return RTE_VHOST_MSG_RESULT_ERR;
2362 if (dev->mem && dev->mem->nregions) {
2363 RTE_LOG(ERR, VHOST_CONFIG,
2364 "Regions already registered at postcopy-listen\n");
2365 return RTE_VHOST_MSG_RESULT_ERR;
2367 dev->postcopy_listening = 1;
2369 return RTE_VHOST_MSG_RESULT_OK;
2373 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2374 int main_fd __rte_unused)
2376 struct virtio_net *dev = *pdev;
2378 if (validate_msg_fds(msg, 0) != 0)
2379 return RTE_VHOST_MSG_RESULT_ERR;
2381 dev->postcopy_listening = 0;
2382 if (dev->postcopy_ufd >= 0) {
2383 close(dev->postcopy_ufd);
2384 dev->postcopy_ufd = -1;
2387 msg->payload.u64 = 0;
2388 msg->size = sizeof(msg->payload.u64);
2391 return RTE_VHOST_MSG_RESULT_REPLY;
2394 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2395 struct VhostUserMsg *msg,
2397 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2398 [VHOST_USER_NONE] = NULL,
2399 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2400 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2401 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2402 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2403 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2404 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2405 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2406 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2407 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2408 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2409 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2410 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2411 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2412 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2413 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2414 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2415 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2416 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2417 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2418 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2419 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2420 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2421 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2422 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2423 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2424 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2425 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2428 /* return bytes# of read on success or negative val on failure. */
2430 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2434 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2435 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2440 if (msg->size > sizeof(msg->payload)) {
2441 RTE_LOG(ERR, VHOST_CONFIG,
2442 "invalid msg size: %d\n", msg->size);
2445 ret = read(sockfd, &msg->payload, msg->size);
2448 if (ret != (int)msg->size) {
2449 RTE_LOG(ERR, VHOST_CONFIG,
2450 "read control message failed\n");
2459 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2464 return send_fd_message(sockfd, (char *)msg,
2465 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2469 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2474 msg->flags &= ~VHOST_USER_VERSION_MASK;
2475 msg->flags &= ~VHOST_USER_NEED_REPLY;
2476 msg->flags |= VHOST_USER_VERSION;
2477 msg->flags |= VHOST_USER_REPLY_MASK;
2479 return send_vhost_message(sockfd, msg);
2483 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2487 if (msg->flags & VHOST_USER_NEED_REPLY)
2488 rte_spinlock_lock(&dev->slave_req_lock);
2490 ret = send_vhost_message(dev->slave_req_fd, msg);
2491 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2492 rte_spinlock_unlock(&dev->slave_req_lock);
2498 * Allocate a queue pair if it hasn't been allocated yet
2501 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2502 struct VhostUserMsg *msg)
2506 switch (msg->request.master) {
2507 case VHOST_USER_SET_VRING_KICK:
2508 case VHOST_USER_SET_VRING_CALL:
2509 case VHOST_USER_SET_VRING_ERR:
2510 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2512 case VHOST_USER_SET_VRING_NUM:
2513 case VHOST_USER_SET_VRING_BASE:
2514 case VHOST_USER_SET_VRING_ENABLE:
2515 vring_idx = msg->payload.state.index;
2517 case VHOST_USER_SET_VRING_ADDR:
2518 vring_idx = msg->payload.addr.index;
2524 if (vring_idx >= VHOST_MAX_VRING) {
2525 RTE_LOG(ERR, VHOST_CONFIG,
2526 "invalid vring index: %u\n", vring_idx);
2530 if (dev->virtqueue[vring_idx])
2533 return alloc_vring_queue(dev, vring_idx);
2537 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2540 unsigned int vq_num = 0;
2542 while (vq_num < dev->nr_vring) {
2543 struct vhost_virtqueue *vq = dev->virtqueue[i];
2546 rte_spinlock_lock(&vq->access_lock);
2554 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2557 unsigned int vq_num = 0;
2559 while (vq_num < dev->nr_vring) {
2560 struct vhost_virtqueue *vq = dev->virtqueue[i];
2563 rte_spinlock_unlock(&vq->access_lock);
2571 vhost_user_msg_handler(int vid, int fd)
2573 struct virtio_net *dev;
2574 struct VhostUserMsg msg;
2575 struct rte_vdpa_device *vdpa_dev;
2578 int unlock_required = 0;
2582 dev = get_device(vid);
2586 if (!dev->notify_ops) {
2587 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2588 if (!dev->notify_ops) {
2589 RTE_LOG(ERR, VHOST_CONFIG,
2590 "failed to get callback ops for driver %s\n",
2596 ret = read_vhost_message(fd, &msg);
2599 RTE_LOG(ERR, VHOST_CONFIG,
2600 "vhost read message failed\n");
2602 RTE_LOG(INFO, VHOST_CONFIG,
2603 "vhost peer closed\n");
2609 request = msg.request.master;
2610 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2611 vhost_message_str[request]) {
2612 if (request != VHOST_USER_IOTLB_MSG)
2613 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
2614 vhost_message_str[request]);
2616 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
2617 vhost_message_str[request]);
2619 RTE_LOG(DEBUG, VHOST_CONFIG, "External request %d\n", request);
2622 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2624 RTE_LOG(ERR, VHOST_CONFIG,
2625 "failed to alloc queue\n");
2630 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2631 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2632 * and device is destroyed. destroy_device waits for queues to be
2633 * inactive, so it is safe. Otherwise taking the access_lock
2634 * would cause a dead lock.
2637 case VHOST_USER_SET_FEATURES:
2638 case VHOST_USER_SET_PROTOCOL_FEATURES:
2639 case VHOST_USER_SET_OWNER:
2640 case VHOST_USER_SET_MEM_TABLE:
2641 case VHOST_USER_SET_LOG_BASE:
2642 case VHOST_USER_SET_LOG_FD:
2643 case VHOST_USER_SET_VRING_NUM:
2644 case VHOST_USER_SET_VRING_ADDR:
2645 case VHOST_USER_SET_VRING_BASE:
2646 case VHOST_USER_SET_VRING_KICK:
2647 case VHOST_USER_SET_VRING_CALL:
2648 case VHOST_USER_SET_VRING_ERR:
2649 case VHOST_USER_SET_VRING_ENABLE:
2650 case VHOST_USER_SEND_RARP:
2651 case VHOST_USER_NET_SET_MTU:
2652 case VHOST_USER_SET_SLAVE_REQ_FD:
2653 vhost_user_lock_all_queue_pairs(dev);
2654 unlock_required = 1;
2662 if (dev->extern_ops.pre_msg_handle) {
2663 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2666 case RTE_VHOST_MSG_RESULT_REPLY:
2667 send_vhost_reply(fd, &msg);
2669 case RTE_VHOST_MSG_RESULT_ERR:
2670 case RTE_VHOST_MSG_RESULT_OK:
2672 goto skip_to_post_handle;
2673 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2679 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2680 if (!vhost_message_handlers[request])
2681 goto skip_to_post_handle;
2682 ret = vhost_message_handlers[request](&dev, &msg, fd);
2685 case RTE_VHOST_MSG_RESULT_ERR:
2686 RTE_LOG(ERR, VHOST_CONFIG,
2687 "Processing %s failed.\n",
2688 vhost_message_str[request]);
2691 case RTE_VHOST_MSG_RESULT_OK:
2692 RTE_LOG(DEBUG, VHOST_CONFIG,
2693 "Processing %s succeeded.\n",
2694 vhost_message_str[request]);
2697 case RTE_VHOST_MSG_RESULT_REPLY:
2698 RTE_LOG(DEBUG, VHOST_CONFIG,
2699 "Processing %s succeeded and needs reply.\n",
2700 vhost_message_str[request]);
2701 send_vhost_reply(fd, &msg);
2709 skip_to_post_handle:
2710 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2711 dev->extern_ops.post_msg_handle) {
2712 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2715 case RTE_VHOST_MSG_RESULT_REPLY:
2716 send_vhost_reply(fd, &msg);
2718 case RTE_VHOST_MSG_RESULT_ERR:
2719 case RTE_VHOST_MSG_RESULT_OK:
2721 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2727 if (unlock_required)
2728 vhost_user_unlock_all_queue_pairs(dev);
2730 /* If message was not handled at this stage, treat it as an error */
2732 RTE_LOG(ERR, VHOST_CONFIG,
2733 "vhost message (req: %d) was not handled.\n", request);
2734 close_msg_fds(&msg);
2735 ret = RTE_VHOST_MSG_RESULT_ERR;
2739 * If the request required a reply that was already sent,
2740 * this optional reply-ack won't be sent as the
2741 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2743 if (msg.flags & VHOST_USER_NEED_REPLY) {
2744 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2745 msg.size = sizeof(msg.payload.u64);
2747 send_vhost_reply(fd, &msg);
2748 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2749 RTE_LOG(ERR, VHOST_CONFIG,
2750 "vhost message handling failed.\n");
2754 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2755 dev->flags |= VIRTIO_DEV_READY;
2757 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2758 if (dev->dequeue_zero_copy) {
2759 RTE_LOG(INFO, VHOST_CONFIG,
2760 "dequeue zero copy is enabled\n");
2763 if (dev->notify_ops->new_device(dev->vid) == 0)
2764 dev->flags |= VIRTIO_DEV_RUNNING;
2768 did = dev->vdpa_dev_id;
2769 vdpa_dev = rte_vdpa_get_device(did);
2770 if (vdpa_dev && virtio_is_ready(dev) &&
2771 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2772 msg.request.master == VHOST_USER_SET_VRING_CALL) {
2773 if (vdpa_dev->ops->dev_conf)
2774 vdpa_dev->ops->dev_conf(dev->vid);
2775 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2781 static int process_slave_message_reply(struct virtio_net *dev,
2782 const struct VhostUserMsg *msg)
2784 struct VhostUserMsg msg_reply;
2787 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2790 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2795 if (msg_reply.request.slave != msg->request.slave) {
2796 RTE_LOG(ERR, VHOST_CONFIG,
2797 "Received unexpected msg type (%u), expected %u\n",
2798 msg_reply.request.slave, msg->request.slave);
2803 ret = msg_reply.payload.u64 ? -1 : 0;
2806 rte_spinlock_unlock(&dev->slave_req_lock);
2811 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2814 struct VhostUserMsg msg = {
2815 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2816 .flags = VHOST_USER_VERSION,
2817 .size = sizeof(msg.payload.iotlb),
2821 .type = VHOST_IOTLB_MISS,
2825 ret = send_vhost_message(dev->slave_req_fd, &msg);
2827 RTE_LOG(ERR, VHOST_CONFIG,
2828 "Failed to send IOTLB miss message (%d)\n",
2836 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2842 struct VhostUserMsg msg = {
2843 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2844 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2845 .size = sizeof(msg.payload.area),
2847 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2854 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2860 ret = send_vhost_slave_message(dev, &msg);
2862 RTE_LOG(ERR, VHOST_CONFIG,
2863 "Failed to set host notifier (%d)\n", ret);
2867 return process_slave_message_reply(dev, &msg);
2870 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2872 struct virtio_net *dev;
2873 struct rte_vdpa_device *vdpa_dev;
2874 int vfio_device_fd, did, ret = 0;
2875 uint64_t offset, size;
2878 dev = get_device(vid);
2882 did = dev->vdpa_dev_id;
2886 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2887 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2888 !(dev->protocol_features &
2889 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2890 !(dev->protocol_features &
2891 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2892 !(dev->protocol_features &
2893 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2896 vdpa_dev = rte_vdpa_get_device(did);
2900 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2901 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2903 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2904 if (vfio_device_fd < 0)
2908 for (i = 0; i < dev->nr_vring; i++) {
2909 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2915 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2916 vfio_device_fd, offset, size) < 0) {
2923 for (i = 0; i < dev->nr_vring; i++) {
2924 vhost_user_slave_set_vring_host_notifier(dev, i, -1,