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",
90 [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
91 [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
98 close_msg_fds(struct VhostUserMsg *msg)
102 for (i = 0; i < msg->fd_num; i++) {
103 int fd = msg->fds[i];
114 * Ensure the expected number of FDs is received,
115 * close all FDs and return an error if this is not the case.
118 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
120 if (msg->fd_num == expected_fds)
123 VHOST_LOG_CONFIG(ERR,
124 " Expect %d FDs for request %s, received %d\n",
126 vhost_message_str[msg->request.master],
140 ret = fstat(fd, &stat);
141 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
145 free_mem_region(struct virtio_net *dev)
148 struct rte_vhost_mem_region *reg;
150 if (!dev || !dev->mem)
153 for (i = 0; i < dev->mem->nregions; i++) {
154 reg = &dev->mem->regions[i];
155 if (reg->host_user_addr) {
156 munmap(reg->mmap_addr, reg->mmap_size);
163 vhost_backend_cleanup(struct virtio_net *dev)
166 free_mem_region(dev);
171 rte_free(dev->guest_pages);
172 dev->guest_pages = NULL;
175 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
179 if (dev->inflight_info) {
180 if (dev->inflight_info->addr) {
181 munmap(dev->inflight_info->addr,
182 dev->inflight_info->size);
183 dev->inflight_info->addr = NULL;
186 if (dev->inflight_info->fd >= 0) {
187 close(dev->inflight_info->fd);
188 dev->inflight_info->fd = -1;
191 free(dev->inflight_info);
192 dev->inflight_info = NULL;
195 if (dev->slave_req_fd >= 0) {
196 close(dev->slave_req_fd);
197 dev->slave_req_fd = -1;
200 if (dev->postcopy_ufd >= 0) {
201 close(dev->postcopy_ufd);
202 dev->postcopy_ufd = -1;
205 dev->postcopy_listening = 0;
209 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
212 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
213 struct vhost_virtqueue *vq = dev->virtqueue[index];
215 /* Configure guest notifications on enable */
216 if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
217 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
219 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
220 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
222 if (dev->notify_ops->vring_state_changed)
223 dev->notify_ops->vring_state_changed(dev->vid,
228 * This function just returns success at the moment unless
229 * the device hasn't been initialised.
232 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
233 struct VhostUserMsg *msg,
234 int main_fd __rte_unused)
236 if (validate_msg_fds(msg, 0) != 0)
237 return RTE_VHOST_MSG_RESULT_ERR;
239 return RTE_VHOST_MSG_RESULT_OK;
243 vhost_user_reset_owner(struct virtio_net **pdev,
244 struct VhostUserMsg *msg,
245 int main_fd __rte_unused)
247 struct virtio_net *dev = *pdev;
249 if (validate_msg_fds(msg, 0) != 0)
250 return RTE_VHOST_MSG_RESULT_ERR;
252 vhost_destroy_device_notify(dev);
254 cleanup_device(dev, 0);
256 return RTE_VHOST_MSG_RESULT_OK;
260 * The features that we support are requested.
263 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
264 int main_fd __rte_unused)
266 struct virtio_net *dev = *pdev;
267 uint64_t features = 0;
269 if (validate_msg_fds(msg, 0) != 0)
270 return RTE_VHOST_MSG_RESULT_ERR;
272 rte_vhost_driver_get_features(dev->ifname, &features);
274 msg->payload.u64 = features;
275 msg->size = sizeof(msg->payload.u64);
278 return RTE_VHOST_MSG_RESULT_REPLY;
282 * The queue number that we support are requested.
285 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
286 int main_fd __rte_unused)
288 struct virtio_net *dev = *pdev;
289 uint32_t queue_num = 0;
291 if (validate_msg_fds(msg, 0) != 0)
292 return RTE_VHOST_MSG_RESULT_ERR;
294 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
296 msg->payload.u64 = (uint64_t)queue_num;
297 msg->size = sizeof(msg->payload.u64);
300 return RTE_VHOST_MSG_RESULT_REPLY;
304 * We receive the negotiated features supported by us and the virtio device.
307 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
308 int main_fd __rte_unused)
310 struct virtio_net *dev = *pdev;
311 uint64_t features = msg->payload.u64;
312 uint64_t vhost_features = 0;
313 struct rte_vdpa_device *vdpa_dev;
315 if (validate_msg_fds(msg, 0) != 0)
316 return RTE_VHOST_MSG_RESULT_ERR;
318 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
319 if (features & ~vhost_features) {
320 VHOST_LOG_CONFIG(ERR,
321 "(%d) received invalid negotiated features.\n",
323 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
324 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
326 return RTE_VHOST_MSG_RESULT_ERR;
329 if (dev->flags & VIRTIO_DEV_RUNNING) {
330 if (dev->features == features)
331 return RTE_VHOST_MSG_RESULT_OK;
334 * Error out if master tries to change features while device is
335 * in running state. The exception being VHOST_F_LOG_ALL, which
336 * is enabled when the live-migration starts.
338 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
339 VHOST_LOG_CONFIG(ERR,
340 "(%d) features changed while device is running.\n",
342 return RTE_VHOST_MSG_RESULT_ERR;
345 if (dev->notify_ops->features_changed)
346 dev->notify_ops->features_changed(dev->vid, features);
349 dev->features = features;
351 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
352 (1ULL << VIRTIO_F_VERSION_1) |
353 (1ULL << VIRTIO_F_RING_PACKED))) {
354 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
356 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
358 VHOST_LOG_CONFIG(INFO,
359 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
360 VHOST_LOG_CONFIG(DEBUG,
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 vdpa_dev = dev->vdpa_dev;
389 vdpa_dev->ops->set_features(dev->vid);
391 dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
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 VHOST_LOG_CONFIG(ERR,
423 "invalid virtqueue size %u\n", vq->size);
424 return RTE_VHOST_MSG_RESULT_ERR;
428 if (vq->size > 32768) {
429 VHOST_LOG_CONFIG(ERR,
430 "invalid virtqueue size %u\n", vq->size);
431 return RTE_VHOST_MSG_RESULT_ERR;
434 if (vq_is_packed(dev)) {
435 if (vq->shadow_used_packed)
436 rte_free(vq->shadow_used_packed);
437 vq->shadow_used_packed = rte_malloc(NULL,
439 sizeof(struct vring_used_elem_packed),
440 RTE_CACHE_LINE_SIZE);
441 if (!vq->shadow_used_packed) {
442 VHOST_LOG_CONFIG(ERR,
443 "failed to allocate memory for shadow used ring.\n");
444 return RTE_VHOST_MSG_RESULT_ERR;
448 if (vq->shadow_used_split)
449 rte_free(vq->shadow_used_split);
451 vq->shadow_used_split = rte_malloc(NULL,
452 vq->size * sizeof(struct vring_used_elem),
453 RTE_CACHE_LINE_SIZE);
455 if (!vq->shadow_used_split) {
456 VHOST_LOG_CONFIG(ERR,
457 "failed to allocate memory for vq internal data.\n");
458 return RTE_VHOST_MSG_RESULT_ERR;
462 if (vq->batch_copy_elems)
463 rte_free(vq->batch_copy_elems);
464 vq->batch_copy_elems = rte_malloc(NULL,
465 vq->size * sizeof(struct batch_copy_elem),
466 RTE_CACHE_LINE_SIZE);
467 if (!vq->batch_copy_elems) {
468 VHOST_LOG_CONFIG(ERR,
469 "failed to allocate memory for batching copy.\n");
470 return RTE_VHOST_MSG_RESULT_ERR;
473 return RTE_VHOST_MSG_RESULT_OK;
477 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
478 * same numa node as the memory of vring descriptor.
480 #ifdef RTE_LIBRTE_VHOST_NUMA
481 static struct virtio_net*
482 numa_realloc(struct virtio_net *dev, int index)
484 int oldnode, newnode;
485 struct virtio_net *old_dev;
486 struct vhost_virtqueue *old_vq, *vq;
487 struct vring_used_elem *new_shadow_used_split;
488 struct vring_used_elem_packed *new_shadow_used_packed;
489 struct batch_copy_elem *new_batch_copy_elems;
492 if (dev->flags & VIRTIO_DEV_RUNNING)
496 vq = old_vq = dev->virtqueue[index];
498 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
499 MPOL_F_NODE | MPOL_F_ADDR);
501 /* check if we need to reallocate vq */
502 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
503 MPOL_F_NODE | MPOL_F_ADDR);
505 VHOST_LOG_CONFIG(ERR,
506 "Unable to get vq numa information.\n");
509 if (oldnode != newnode) {
510 VHOST_LOG_CONFIG(INFO,
511 "reallocate vq from %d to %d node\n", oldnode, newnode);
512 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
516 memcpy(vq, old_vq, sizeof(*vq));
518 if (vq_is_packed(dev)) {
519 new_shadow_used_packed = rte_malloc_socket(NULL,
521 sizeof(struct vring_used_elem_packed),
524 if (new_shadow_used_packed) {
525 rte_free(vq->shadow_used_packed);
526 vq->shadow_used_packed = new_shadow_used_packed;
529 new_shadow_used_split = rte_malloc_socket(NULL,
531 sizeof(struct vring_used_elem),
534 if (new_shadow_used_split) {
535 rte_free(vq->shadow_used_split);
536 vq->shadow_used_split = new_shadow_used_split;
540 new_batch_copy_elems = rte_malloc_socket(NULL,
541 vq->size * sizeof(struct batch_copy_elem),
544 if (new_batch_copy_elems) {
545 rte_free(vq->batch_copy_elems);
546 vq->batch_copy_elems = new_batch_copy_elems;
552 /* check if we need to reallocate dev */
553 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
554 MPOL_F_NODE | MPOL_F_ADDR);
556 VHOST_LOG_CONFIG(ERR,
557 "Unable to get dev numa information.\n");
560 if (oldnode != newnode) {
561 VHOST_LOG_CONFIG(INFO,
562 "reallocate dev from %d to %d node\n",
564 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
570 memcpy(dev, old_dev, sizeof(*dev));
575 dev->virtqueue[index] = vq;
576 vhost_devices[dev->vid] = dev;
579 vhost_user_iotlb_init(dev, index);
584 static struct virtio_net*
585 numa_realloc(struct virtio_net *dev, int index __rte_unused)
591 /* Converts QEMU virtual address to Vhost virtual address. */
593 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
595 struct rte_vhost_mem_region *r;
598 if (unlikely(!dev || !dev->mem))
601 /* Find the region where the address lives. */
602 for (i = 0; i < dev->mem->nregions; i++) {
603 r = &dev->mem->regions[i];
605 if (qva >= r->guest_user_addr &&
606 qva < r->guest_user_addr + r->size) {
608 if (unlikely(*len > r->guest_user_addr + r->size - qva))
609 *len = r->guest_user_addr + r->size - qva;
611 return qva - r->guest_user_addr +
623 * Converts ring address to Vhost virtual address.
624 * If IOMMU is enabled, the ring address is a guest IO virtual address,
625 * else it is a QEMU virtual address.
628 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
629 uint64_t ra, uint64_t *size)
631 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
634 vhost_user_iotlb_rd_lock(vq);
635 vva = vhost_iova_to_vva(dev, vq, ra,
636 size, VHOST_ACCESS_RW);
637 vhost_user_iotlb_rd_unlock(vq);
642 return qva_to_vva(dev, ra, size);
646 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
650 vhost_user_iotlb_rd_lock(vq);
651 log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
652 vhost_user_iotlb_rd_unlock(vq);
657 static struct virtio_net *
658 translate_ring_addresses(struct virtio_net *dev, int vq_index)
660 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
661 struct vhost_vring_addr *addr = &vq->ring_addrs;
662 uint64_t len, expected_len;
664 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
666 log_addr_to_gpa(dev, vq);
667 if (vq->log_guest_addr == 0) {
668 VHOST_LOG_CONFIG(DEBUG,
669 "(%d) failed to map log_guest_addr.\n",
675 if (vq_is_packed(dev)) {
676 len = sizeof(struct vring_packed_desc) * vq->size;
677 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
678 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
679 if (vq->desc_packed == NULL ||
680 len != sizeof(struct vring_packed_desc) *
682 VHOST_LOG_CONFIG(DEBUG,
683 "(%d) failed to map desc_packed ring.\n",
688 dev = numa_realloc(dev, vq_index);
689 vq = dev->virtqueue[vq_index];
690 addr = &vq->ring_addrs;
692 len = sizeof(struct vring_packed_desc_event);
693 vq->driver_event = (struct vring_packed_desc_event *)
694 (uintptr_t)ring_addr_to_vva(dev,
695 vq, addr->avail_user_addr, &len);
696 if (vq->driver_event == NULL ||
697 len != sizeof(struct vring_packed_desc_event)) {
698 VHOST_LOG_CONFIG(DEBUG,
699 "(%d) failed to find driver area address.\n",
704 len = sizeof(struct vring_packed_desc_event);
705 vq->device_event = (struct vring_packed_desc_event *)
706 (uintptr_t)ring_addr_to_vva(dev,
707 vq, addr->used_user_addr, &len);
708 if (vq->device_event == NULL ||
709 len != sizeof(struct vring_packed_desc_event)) {
710 VHOST_LOG_CONFIG(DEBUG,
711 "(%d) failed to find device area address.\n",
720 /* The addresses are converted from QEMU virtual to Vhost virtual. */
721 if (vq->desc && vq->avail && vq->used)
724 len = sizeof(struct vring_desc) * vq->size;
725 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
726 vq, addr->desc_user_addr, &len);
727 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
728 VHOST_LOG_CONFIG(DEBUG,
729 "(%d) failed to map desc ring.\n",
734 dev = numa_realloc(dev, vq_index);
735 vq = dev->virtqueue[vq_index];
736 addr = &vq->ring_addrs;
738 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
739 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
740 len += sizeof(uint16_t);
742 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
743 vq, addr->avail_user_addr, &len);
744 if (vq->avail == 0 || len != expected_len) {
745 VHOST_LOG_CONFIG(DEBUG,
746 "(%d) failed to map avail ring.\n",
751 len = sizeof(struct vring_used) +
752 sizeof(struct vring_used_elem) * vq->size;
753 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
754 len += sizeof(uint16_t);
756 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
757 vq, addr->used_user_addr, &len);
758 if (vq->used == 0 || len != expected_len) {
759 VHOST_LOG_CONFIG(DEBUG,
760 "(%d) failed to map used ring.\n",
765 if (vq->last_used_idx != vq->used->idx) {
766 VHOST_LOG_CONFIG(WARNING,
767 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
768 "some packets maybe resent for Tx and dropped for Rx\n",
769 vq->last_used_idx, vq->used->idx);
770 vq->last_used_idx = vq->used->idx;
771 vq->last_avail_idx = vq->used->idx;
776 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
778 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
779 dev->vid, vq->avail);
780 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
782 VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
783 dev->vid, vq->log_guest_addr);
789 * The virtio device sends us the desc, used and avail ring addresses.
790 * This function then converts these to our address space.
793 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
794 int main_fd __rte_unused)
796 struct virtio_net *dev = *pdev;
797 struct vhost_virtqueue *vq;
798 struct vhost_vring_addr *addr = &msg->payload.addr;
801 if (validate_msg_fds(msg, 0) != 0)
802 return RTE_VHOST_MSG_RESULT_ERR;
804 if (dev->mem == NULL)
805 return RTE_VHOST_MSG_RESULT_ERR;
807 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
808 vq = dev->virtqueue[msg->payload.addr.index];
810 access_ok = vq->access_ok;
813 * Rings addresses should not be interpreted as long as the ring is not
814 * started and enabled
816 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
818 vring_invalidate(dev, vq);
820 if ((vq->enabled && (dev->features &
821 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
823 dev = translate_ring_addresses(dev, msg->payload.addr.index);
825 return RTE_VHOST_MSG_RESULT_ERR;
830 return RTE_VHOST_MSG_RESULT_OK;
834 * The virtio device sends us the available ring last used index.
837 vhost_user_set_vring_base(struct virtio_net **pdev,
838 struct VhostUserMsg *msg,
839 int main_fd __rte_unused)
841 struct virtio_net *dev = *pdev;
842 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
843 uint64_t val = msg->payload.state.num;
845 if (validate_msg_fds(msg, 0) != 0)
846 return RTE_VHOST_MSG_RESULT_ERR;
848 if (vq_is_packed(dev)) {
850 * Bit[0:14]: avail index
851 * Bit[15]: avail wrap counter
853 vq->last_avail_idx = val & 0x7fff;
854 vq->avail_wrap_counter = !!(val & (0x1 << 15));
856 * Set used index to same value as available one, as
857 * their values should be the same since ring processing
858 * was stopped at get time.
860 vq->last_used_idx = vq->last_avail_idx;
861 vq->used_wrap_counter = vq->avail_wrap_counter;
863 vq->last_used_idx = msg->payload.state.num;
864 vq->last_avail_idx = msg->payload.state.num;
867 return RTE_VHOST_MSG_RESULT_OK;
871 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
872 uint64_t host_phys_addr, uint64_t size)
874 struct guest_page *page, *last_page;
875 struct guest_page *old_pages;
877 if (dev->nr_guest_pages == dev->max_guest_pages) {
878 dev->max_guest_pages *= 2;
879 old_pages = dev->guest_pages;
880 dev->guest_pages = rte_realloc(dev->guest_pages,
881 dev->max_guest_pages * sizeof(*page),
882 RTE_CACHE_LINE_SIZE);
883 if (dev->guest_pages == NULL) {
884 VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
890 if (dev->nr_guest_pages > 0) {
891 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
892 /* merge if the two pages are continuous */
893 if (host_phys_addr == last_page->host_phys_addr +
895 last_page->size += size;
900 page = &dev->guest_pages[dev->nr_guest_pages++];
901 page->guest_phys_addr = guest_phys_addr;
902 page->host_phys_addr = host_phys_addr;
909 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
912 uint64_t reg_size = reg->size;
913 uint64_t host_user_addr = reg->host_user_addr;
914 uint64_t guest_phys_addr = reg->guest_phys_addr;
915 uint64_t host_phys_addr;
918 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
919 size = page_size - (guest_phys_addr & (page_size - 1));
920 size = RTE_MIN(size, reg_size);
922 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
925 host_user_addr += size;
926 guest_phys_addr += size;
929 while (reg_size > 0) {
930 size = RTE_MIN(reg_size, page_size);
931 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
933 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
937 host_user_addr += size;
938 guest_phys_addr += size;
942 /* sort guest page array if over binary search threshold */
943 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
944 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
945 sizeof(struct guest_page), guest_page_addrcmp);
951 #ifdef RTE_LIBRTE_VHOST_DEBUG
952 /* TODO: enable it only in debug mode? */
954 dump_guest_pages(struct virtio_net *dev)
957 struct guest_page *page;
959 for (i = 0; i < dev->nr_guest_pages; i++) {
960 page = &dev->guest_pages[i];
962 VHOST_LOG_CONFIG(INFO,
963 "guest physical page region %u\n"
964 "\t guest_phys_addr: %" PRIx64 "\n"
965 "\t host_phys_addr : %" PRIx64 "\n"
966 "\t size : %" PRIx64 "\n",
968 page->guest_phys_addr,
969 page->host_phys_addr,
974 #define dump_guest_pages(dev)
978 vhost_memory_changed(struct VhostUserMemory *new,
979 struct rte_vhost_memory *old)
983 if (new->nregions != old->nregions)
986 for (i = 0; i < new->nregions; ++i) {
987 VhostUserMemoryRegion *new_r = &new->regions[i];
988 struct rte_vhost_mem_region *old_r = &old->regions[i];
990 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
992 if (new_r->memory_size != old_r->size)
994 if (new_r->userspace_addr != old_r->guest_user_addr)
1002 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1005 struct virtio_net *dev = *pdev;
1006 struct VhostUserMemory *memory = &msg->payload.memory;
1007 struct rte_vhost_mem_region *reg;
1010 uint64_t mmap_offset;
1015 if (validate_msg_fds(msg, memory->nregions) != 0)
1016 return RTE_VHOST_MSG_RESULT_ERR;
1018 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1019 VHOST_LOG_CONFIG(ERR,
1020 "too many memory regions (%u)\n", memory->nregions);
1024 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1025 VHOST_LOG_CONFIG(INFO,
1026 "(%d) memory regions not changed\n", dev->vid);
1030 return RTE_VHOST_MSG_RESULT_OK;
1034 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1035 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1037 if (vdpa_dev && vdpa_dev->ops->dev_close)
1038 vdpa_dev->ops->dev_close(dev->vid);
1039 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1041 free_mem_region(dev);
1046 /* Flush IOTLB cache as previous HVAs are now invalid */
1047 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1048 for (i = 0; i < dev->nr_vring; i++)
1049 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1051 dev->nr_guest_pages = 0;
1052 if (dev->guest_pages == NULL) {
1053 dev->max_guest_pages = 8;
1054 dev->guest_pages = rte_zmalloc(NULL,
1055 dev->max_guest_pages *
1056 sizeof(struct guest_page),
1057 RTE_CACHE_LINE_SIZE);
1058 if (dev->guest_pages == NULL) {
1059 VHOST_LOG_CONFIG(ERR,
1060 "(%d) failed to allocate memory "
1061 "for dev->guest_pages\n",
1067 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1068 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1069 if (dev->mem == NULL) {
1070 VHOST_LOG_CONFIG(ERR,
1071 "(%d) failed to allocate memory for dev->mem\n",
1073 goto free_guest_pages;
1075 dev->mem->nregions = memory->nregions;
1077 for (i = 0; i < memory->nregions; i++) {
1078 reg = &dev->mem->regions[i];
1080 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1081 reg->guest_user_addr = memory->regions[i].userspace_addr;
1082 reg->size = memory->regions[i].memory_size;
1083 reg->fd = msg->fds[i];
1086 * Assign invalid file descriptor value to avoid double
1087 * closing on error path.
1091 mmap_offset = memory->regions[i].mmap_offset;
1093 /* Check for memory_size + mmap_offset overflow */
1094 if (mmap_offset >= -reg->size) {
1095 VHOST_LOG_CONFIG(ERR,
1096 "mmap_offset (%#"PRIx64") and memory_size "
1097 "(%#"PRIx64") overflow\n",
1098 mmap_offset, reg->size);
1099 goto free_mem_table;
1102 mmap_size = reg->size + mmap_offset;
1104 /* mmap() without flag of MAP_ANONYMOUS, should be called
1105 * with length argument aligned with hugepagesz at older
1106 * longterm version Linux, like 2.6.32 and 3.2.72, or
1107 * mmap() will fail with EINVAL.
1109 * to avoid failure, make sure in caller to keep length
1112 alignment = get_blk_size(reg->fd);
1113 if (alignment == (uint64_t)-1) {
1114 VHOST_LOG_CONFIG(ERR,
1115 "couldn't get hugepage size through fstat\n");
1116 goto free_mem_table;
1118 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1119 if (mmap_size == 0) {
1121 * It could happen if initial mmap_size + alignment
1122 * overflows the sizeof uint64, which could happen if
1123 * either mmap_size or alignment value is wrong.
1125 * mmap() kernel implementation would return an error,
1126 * but better catch it before and provide useful info
1129 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1130 "or alignment (0x%" PRIx64 ") is invalid\n",
1131 reg->size + mmap_offset, alignment);
1132 goto free_mem_table;
1135 populate = dev->async_copy ? MAP_POPULATE : 0;
1136 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1137 MAP_SHARED | populate, reg->fd, 0);
1139 if (mmap_addr == MAP_FAILED) {
1140 VHOST_LOG_CONFIG(ERR,
1141 "mmap region %u failed.\n", i);
1142 goto free_mem_table;
1145 reg->mmap_addr = mmap_addr;
1146 reg->mmap_size = mmap_size;
1147 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1150 if (dev->async_copy)
1151 if (add_guest_pages(dev, reg, alignment) < 0) {
1152 VHOST_LOG_CONFIG(ERR,
1153 "adding guest pages to region %u failed.\n",
1155 goto free_mem_table;
1158 VHOST_LOG_CONFIG(INFO,
1159 "guest memory region %u, size: 0x%" PRIx64 "\n"
1160 "\t guest physical addr: 0x%" PRIx64 "\n"
1161 "\t guest virtual addr: 0x%" PRIx64 "\n"
1162 "\t host virtual addr: 0x%" PRIx64 "\n"
1163 "\t mmap addr : 0x%" PRIx64 "\n"
1164 "\t mmap size : 0x%" PRIx64 "\n"
1165 "\t mmap align: 0x%" PRIx64 "\n"
1166 "\t mmap off : 0x%" PRIx64 "\n",
1168 reg->guest_phys_addr,
1169 reg->guest_user_addr,
1170 reg->host_user_addr,
1171 (uint64_t)(uintptr_t)mmap_addr,
1176 if (dev->postcopy_listening) {
1178 * We haven't a better way right now than sharing
1179 * DPDK's virtual address with Qemu, so that Qemu can
1180 * retrieve the region offset when handling userfaults.
1182 memory->regions[i].userspace_addr =
1183 reg->host_user_addr;
1186 if (dev->postcopy_listening) {
1187 /* Send the addresses back to qemu */
1189 send_vhost_reply(main_fd, msg);
1191 /* Wait for qemu to acknolwedge it's got the addresses
1192 * we've got to wait before we're allowed to generate faults.
1194 VhostUserMsg ack_msg;
1195 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1196 VHOST_LOG_CONFIG(ERR,
1197 "Failed to read qemu ack on postcopy set-mem-table\n");
1198 goto free_mem_table;
1201 if (validate_msg_fds(&ack_msg, 0) != 0)
1202 goto free_mem_table;
1204 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1205 VHOST_LOG_CONFIG(ERR,
1206 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1207 ack_msg.request.master);
1208 goto free_mem_table;
1211 /* Now userfault register and we can use the memory */
1212 for (i = 0; i < memory->nregions; i++) {
1213 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1214 reg = &dev->mem->regions[i];
1215 struct uffdio_register reg_struct;
1218 * Let's register all the mmap'ed area to ensure
1219 * alignment on page boundary.
1221 reg_struct.range.start =
1222 (uint64_t)(uintptr_t)reg->mmap_addr;
1223 reg_struct.range.len = reg->mmap_size;
1224 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1226 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1228 VHOST_LOG_CONFIG(ERR,
1229 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1230 i, dev->postcopy_ufd,
1232 goto free_mem_table;
1234 VHOST_LOG_CONFIG(INFO,
1235 "\t userfaultfd registered for range : "
1236 "%" PRIx64 " - %" PRIx64 "\n",
1237 (uint64_t)reg_struct.range.start,
1238 (uint64_t)reg_struct.range.start +
1239 (uint64_t)reg_struct.range.len - 1);
1241 goto free_mem_table;
1246 for (i = 0; i < dev->nr_vring; i++) {
1247 struct vhost_virtqueue *vq = dev->virtqueue[i];
1252 if (vq->desc || vq->avail || vq->used) {
1254 * If the memory table got updated, the ring addresses
1255 * need to be translated again as virtual addresses have
1258 vring_invalidate(dev, vq);
1260 dev = translate_ring_addresses(dev, i);
1263 goto free_mem_table;
1270 dump_guest_pages(dev);
1272 return RTE_VHOST_MSG_RESULT_OK;
1275 free_mem_region(dev);
1279 rte_free(dev->guest_pages);
1280 dev->guest_pages = NULL;
1283 return RTE_VHOST_MSG_RESULT_ERR;
1287 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1294 if (vq_is_packed(dev))
1295 rings_ok = vq->desc_packed && vq->driver_event &&
1298 rings_ok = vq->desc && vq->avail && vq->used;
1301 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1302 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1306 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1309 virtio_is_ready(struct virtio_net *dev)
1311 struct vhost_virtqueue *vq;
1312 uint32_t i, nr_vring = dev->nr_vring;
1314 if (dev->flags & VIRTIO_DEV_READY)
1320 if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1321 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1323 if (dev->nr_vring < nr_vring)
1327 for (i = 0; i < nr_vring; i++) {
1328 vq = dev->virtqueue[i];
1330 if (!vq_is_ready(dev, vq))
1334 /* If supported, ensure the frontend is really done with config */
1335 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1336 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1339 dev->flags |= VIRTIO_DEV_READY;
1341 if (!(dev->flags & VIRTIO_DEV_RUNNING))
1342 VHOST_LOG_CONFIG(INFO,
1343 "virtio is now ready for processing.\n");
1348 inflight_mem_alloc(const char *name, size_t size, int *fd)
1352 char fname[20] = "/tmp/memfd-XXXXXX";
1355 #ifdef MEMFD_SUPPORTED
1356 mfd = memfd_create(name, MFD_CLOEXEC);
1361 mfd = mkstemp(fname);
1363 VHOST_LOG_CONFIG(ERR,
1364 "failed to get inflight buffer fd\n");
1371 if (ftruncate(mfd, size) == -1) {
1372 VHOST_LOG_CONFIG(ERR,
1373 "failed to alloc inflight buffer\n");
1378 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1379 if (ptr == MAP_FAILED) {
1380 VHOST_LOG_CONFIG(ERR,
1381 "failed to mmap inflight buffer\n");
1391 get_pervq_shm_size_split(uint16_t queue_size)
1393 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1394 queue_size + sizeof(uint64_t) +
1395 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1399 get_pervq_shm_size_packed(uint16_t queue_size)
1401 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1402 * queue_size + sizeof(uint64_t) +
1403 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1404 INFLIGHT_ALIGNMENT);
1408 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1410 int main_fd __rte_unused)
1412 struct rte_vhost_inflight_info_packed *inflight_packed;
1413 uint64_t pervq_inflight_size, mmap_size;
1414 uint16_t num_queues, queue_size;
1415 struct virtio_net *dev = *pdev;
1419 if (msg->size != sizeof(msg->payload.inflight)) {
1420 VHOST_LOG_CONFIG(ERR,
1421 "invalid get_inflight_fd message size is %d\n",
1423 return RTE_VHOST_MSG_RESULT_ERR;
1426 if (dev->inflight_info == NULL) {
1427 dev->inflight_info = calloc(1,
1428 sizeof(struct inflight_mem_info));
1429 if (!dev->inflight_info) {
1430 VHOST_LOG_CONFIG(ERR,
1431 "failed to alloc dev inflight area\n");
1432 return RTE_VHOST_MSG_RESULT_ERR;
1434 dev->inflight_info->fd = -1;
1437 num_queues = msg->payload.inflight.num_queues;
1438 queue_size = msg->payload.inflight.queue_size;
1440 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1441 msg->payload.inflight.num_queues);
1442 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1443 msg->payload.inflight.queue_size);
1445 if (vq_is_packed(dev))
1446 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1448 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1450 mmap_size = num_queues * pervq_inflight_size;
1451 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1453 VHOST_LOG_CONFIG(ERR,
1454 "failed to alloc vhost inflight area\n");
1455 msg->payload.inflight.mmap_size = 0;
1456 return RTE_VHOST_MSG_RESULT_ERR;
1458 memset(addr, 0, mmap_size);
1460 if (dev->inflight_info->addr) {
1461 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1462 dev->inflight_info->addr = NULL;
1465 if (dev->inflight_info->fd >= 0) {
1466 close(dev->inflight_info->fd);
1467 dev->inflight_info->fd = -1;
1470 dev->inflight_info->addr = addr;
1471 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1472 dev->inflight_info->fd = msg->fds[0] = fd;
1473 msg->payload.inflight.mmap_offset = 0;
1476 if (vq_is_packed(dev)) {
1477 for (i = 0; i < num_queues; i++) {
1479 (struct rte_vhost_inflight_info_packed *)addr;
1480 inflight_packed->used_wrap_counter = 1;
1481 inflight_packed->old_used_wrap_counter = 1;
1482 for (j = 0; j < queue_size; j++)
1483 inflight_packed->desc[j].next = j + 1;
1484 addr = (void *)((char *)addr + pervq_inflight_size);
1488 VHOST_LOG_CONFIG(INFO,
1489 "send inflight mmap_size: %"PRIu64"\n",
1490 msg->payload.inflight.mmap_size);
1491 VHOST_LOG_CONFIG(INFO,
1492 "send inflight mmap_offset: %"PRIu64"\n",
1493 msg->payload.inflight.mmap_offset);
1494 VHOST_LOG_CONFIG(INFO,
1495 "send inflight fd: %d\n", msg->fds[0]);
1497 return RTE_VHOST_MSG_RESULT_REPLY;
1501 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1502 int main_fd __rte_unused)
1504 uint64_t mmap_size, mmap_offset;
1505 uint16_t num_queues, queue_size;
1506 struct virtio_net *dev = *pdev;
1507 uint32_t pervq_inflight_size;
1508 struct vhost_virtqueue *vq;
1513 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1514 VHOST_LOG_CONFIG(ERR,
1515 "invalid set_inflight_fd message size is %d,fd is %d\n",
1517 return RTE_VHOST_MSG_RESULT_ERR;
1520 mmap_size = msg->payload.inflight.mmap_size;
1521 mmap_offset = msg->payload.inflight.mmap_offset;
1522 num_queues = msg->payload.inflight.num_queues;
1523 queue_size = msg->payload.inflight.queue_size;
1525 if (vq_is_packed(dev))
1526 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1528 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1530 VHOST_LOG_CONFIG(INFO,
1531 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1532 VHOST_LOG_CONFIG(INFO,
1533 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1534 VHOST_LOG_CONFIG(INFO,
1535 "set_inflight_fd num_queues: %u\n", num_queues);
1536 VHOST_LOG_CONFIG(INFO,
1537 "set_inflight_fd queue_size: %u\n", queue_size);
1538 VHOST_LOG_CONFIG(INFO,
1539 "set_inflight_fd fd: %d\n", fd);
1540 VHOST_LOG_CONFIG(INFO,
1541 "set_inflight_fd pervq_inflight_size: %d\n",
1542 pervq_inflight_size);
1544 if (!dev->inflight_info) {
1545 dev->inflight_info = calloc(1,
1546 sizeof(struct inflight_mem_info));
1547 if (dev->inflight_info == NULL) {
1548 VHOST_LOG_CONFIG(ERR,
1549 "failed to alloc dev inflight area\n");
1550 return RTE_VHOST_MSG_RESULT_ERR;
1552 dev->inflight_info->fd = -1;
1555 if (dev->inflight_info->addr) {
1556 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1557 dev->inflight_info->addr = NULL;
1560 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1562 if (addr == MAP_FAILED) {
1563 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1564 return RTE_VHOST_MSG_RESULT_ERR;
1567 if (dev->inflight_info->fd >= 0) {
1568 close(dev->inflight_info->fd);
1569 dev->inflight_info->fd = -1;
1572 dev->inflight_info->fd = fd;
1573 dev->inflight_info->addr = addr;
1574 dev->inflight_info->size = mmap_size;
1576 for (i = 0; i < num_queues; i++) {
1577 vq = dev->virtqueue[i];
1581 if (vq_is_packed(dev)) {
1582 vq->inflight_packed = addr;
1583 vq->inflight_packed->desc_num = queue_size;
1585 vq->inflight_split = addr;
1586 vq->inflight_split->desc_num = queue_size;
1588 addr = (void *)((char *)addr + pervq_inflight_size);
1591 return RTE_VHOST_MSG_RESULT_OK;
1595 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1596 int main_fd __rte_unused)
1598 struct virtio_net *dev = *pdev;
1599 struct vhost_vring_file file;
1600 struct vhost_virtqueue *vq;
1603 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1604 if (validate_msg_fds(msg, expected_fds) != 0)
1605 return RTE_VHOST_MSG_RESULT_ERR;
1607 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1608 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1609 file.fd = VIRTIO_INVALID_EVENTFD;
1611 file.fd = msg->fds[0];
1612 VHOST_LOG_CONFIG(INFO,
1613 "vring call idx:%d file:%d\n", file.index, file.fd);
1615 vq = dev->virtqueue[file.index];
1619 vhost_user_notify_queue_state(dev, file.index, 0);
1622 if (vq->callfd >= 0)
1625 vq->callfd = file.fd;
1627 return RTE_VHOST_MSG_RESULT_OK;
1630 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1631 struct VhostUserMsg *msg,
1632 int main_fd __rte_unused)
1636 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1637 if (validate_msg_fds(msg, expected_fds) != 0)
1638 return RTE_VHOST_MSG_RESULT_ERR;
1640 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1642 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1644 return RTE_VHOST_MSG_RESULT_OK;
1648 resubmit_desc_compare(const void *a, const void *b)
1650 const struct rte_vhost_resubmit_desc *desc0 = a;
1651 const struct rte_vhost_resubmit_desc *desc1 = b;
1653 if (desc1->counter > desc0->counter)
1660 vhost_check_queue_inflights_split(struct virtio_net *dev,
1661 struct vhost_virtqueue *vq)
1664 uint16_t resubmit_num = 0, last_io, num;
1665 struct vring_used *used = vq->used;
1666 struct rte_vhost_resubmit_info *resubmit;
1667 struct rte_vhost_inflight_info_split *inflight_split;
1669 if (!(dev->protocol_features &
1670 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1671 return RTE_VHOST_MSG_RESULT_OK;
1673 /* The frontend may still not support the inflight feature
1674 * although we negotiate the protocol feature.
1676 if ((!vq->inflight_split))
1677 return RTE_VHOST_MSG_RESULT_OK;
1679 if (!vq->inflight_split->version) {
1680 vq->inflight_split->version = INFLIGHT_VERSION;
1681 return RTE_VHOST_MSG_RESULT_OK;
1684 if (vq->resubmit_inflight)
1685 return RTE_VHOST_MSG_RESULT_OK;
1687 inflight_split = vq->inflight_split;
1688 vq->global_counter = 0;
1689 last_io = inflight_split->last_inflight_io;
1691 if (inflight_split->used_idx != used->idx) {
1692 inflight_split->desc[last_io].inflight = 0;
1694 inflight_split->used_idx = used->idx;
1697 for (i = 0; i < inflight_split->desc_num; i++) {
1698 if (inflight_split->desc[i].inflight == 1)
1702 vq->last_avail_idx += resubmit_num;
1705 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1707 VHOST_LOG_CONFIG(ERR,
1708 "failed to allocate memory for resubmit info.\n");
1709 return RTE_VHOST_MSG_RESULT_ERR;
1712 resubmit->resubmit_list = calloc(resubmit_num,
1713 sizeof(struct rte_vhost_resubmit_desc));
1714 if (!resubmit->resubmit_list) {
1715 VHOST_LOG_CONFIG(ERR,
1716 "failed to allocate memory for inflight desc.\n");
1718 return RTE_VHOST_MSG_RESULT_ERR;
1722 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1723 if (vq->inflight_split->desc[i].inflight == 1) {
1724 resubmit->resubmit_list[num].index = i;
1725 resubmit->resubmit_list[num].counter =
1726 inflight_split->desc[i].counter;
1730 resubmit->resubmit_num = num;
1732 if (resubmit->resubmit_num > 1)
1733 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1734 sizeof(struct rte_vhost_resubmit_desc),
1735 resubmit_desc_compare);
1737 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1738 vq->resubmit_inflight = resubmit;
1741 return RTE_VHOST_MSG_RESULT_OK;
1745 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1746 struct vhost_virtqueue *vq)
1749 uint16_t resubmit_num = 0, old_used_idx, num;
1750 struct rte_vhost_resubmit_info *resubmit;
1751 struct rte_vhost_inflight_info_packed *inflight_packed;
1753 if (!(dev->protocol_features &
1754 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1755 return RTE_VHOST_MSG_RESULT_OK;
1757 /* The frontend may still not support the inflight feature
1758 * although we negotiate the protocol feature.
1760 if ((!vq->inflight_packed))
1761 return RTE_VHOST_MSG_RESULT_OK;
1763 if (!vq->inflight_packed->version) {
1764 vq->inflight_packed->version = INFLIGHT_VERSION;
1765 return RTE_VHOST_MSG_RESULT_OK;
1768 if (vq->resubmit_inflight)
1769 return RTE_VHOST_MSG_RESULT_OK;
1771 inflight_packed = vq->inflight_packed;
1772 vq->global_counter = 0;
1773 old_used_idx = inflight_packed->old_used_idx;
1775 if (inflight_packed->used_idx != old_used_idx) {
1776 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1777 inflight_packed->old_used_idx =
1778 inflight_packed->used_idx;
1779 inflight_packed->old_used_wrap_counter =
1780 inflight_packed->used_wrap_counter;
1781 inflight_packed->old_free_head =
1782 inflight_packed->free_head;
1784 inflight_packed->used_idx =
1785 inflight_packed->old_used_idx;
1786 inflight_packed->used_wrap_counter =
1787 inflight_packed->old_used_wrap_counter;
1788 inflight_packed->free_head =
1789 inflight_packed->old_free_head;
1793 for (i = 0; i < inflight_packed->desc_num; i++) {
1794 if (inflight_packed->desc[i].inflight == 1)
1799 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1800 if (resubmit == NULL) {
1801 VHOST_LOG_CONFIG(ERR,
1802 "failed to allocate memory for resubmit info.\n");
1803 return RTE_VHOST_MSG_RESULT_ERR;
1806 resubmit->resubmit_list = calloc(resubmit_num,
1807 sizeof(struct rte_vhost_resubmit_desc));
1808 if (resubmit->resubmit_list == NULL) {
1809 VHOST_LOG_CONFIG(ERR,
1810 "failed to allocate memory for resubmit desc.\n");
1812 return RTE_VHOST_MSG_RESULT_ERR;
1816 for (i = 0; i < inflight_packed->desc_num; i++) {
1817 if (vq->inflight_packed->desc[i].inflight == 1) {
1818 resubmit->resubmit_list[num].index = i;
1819 resubmit->resubmit_list[num].counter =
1820 inflight_packed->desc[i].counter;
1824 resubmit->resubmit_num = num;
1826 if (resubmit->resubmit_num > 1)
1827 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1828 sizeof(struct rte_vhost_resubmit_desc),
1829 resubmit_desc_compare);
1831 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1832 vq->resubmit_inflight = resubmit;
1835 return RTE_VHOST_MSG_RESULT_OK;
1839 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1840 int main_fd __rte_unused)
1842 struct virtio_net *dev = *pdev;
1843 struct vhost_vring_file file;
1844 struct vhost_virtqueue *vq;
1847 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1848 if (validate_msg_fds(msg, expected_fds) != 0)
1849 return RTE_VHOST_MSG_RESULT_ERR;
1851 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1852 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1853 file.fd = VIRTIO_INVALID_EVENTFD;
1855 file.fd = msg->fds[0];
1856 VHOST_LOG_CONFIG(INFO,
1857 "vring kick idx:%d file:%d\n", file.index, file.fd);
1859 /* Interpret ring addresses only when ring is started. */
1860 dev = translate_ring_addresses(dev, file.index);
1862 if (file.fd != VIRTIO_INVALID_EVENTFD)
1865 return RTE_VHOST_MSG_RESULT_ERR;
1870 vq = dev->virtqueue[file.index];
1873 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1874 * the ring starts already enabled. Otherwise, it is enabled via
1875 * the SET_VRING_ENABLE message.
1877 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1879 if (dev->notify_ops->vring_state_changed)
1880 dev->notify_ops->vring_state_changed(
1881 dev->vid, file.index, 1);
1886 vhost_user_notify_queue_state(dev, file.index, 0);
1889 if (vq->kickfd >= 0)
1891 vq->kickfd = file.fd;
1893 if (vq_is_packed(dev)) {
1894 if (vhost_check_queue_inflights_packed(dev, vq)) {
1895 VHOST_LOG_CONFIG(ERR,
1896 "failed to inflights for vq: %d\n", file.index);
1897 return RTE_VHOST_MSG_RESULT_ERR;
1900 if (vhost_check_queue_inflights_split(dev, vq)) {
1901 VHOST_LOG_CONFIG(ERR,
1902 "failed to inflights for vq: %d\n", file.index);
1903 return RTE_VHOST_MSG_RESULT_ERR;
1907 return RTE_VHOST_MSG_RESULT_OK;
1911 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1914 vhost_user_get_vring_base(struct virtio_net **pdev,
1915 struct VhostUserMsg *msg,
1916 int main_fd __rte_unused)
1918 struct virtio_net *dev = *pdev;
1919 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1922 if (validate_msg_fds(msg, 0) != 0)
1923 return RTE_VHOST_MSG_RESULT_ERR;
1925 /* We have to stop the queue (virtio) if it is running. */
1926 vhost_destroy_device_notify(dev);
1928 dev->flags &= ~VIRTIO_DEV_READY;
1929 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1931 /* Here we are safe to get the indexes */
1932 if (vq_is_packed(dev)) {
1934 * Bit[0:14]: avail index
1935 * Bit[15]: avail wrap counter
1937 val = vq->last_avail_idx & 0x7fff;
1938 val |= vq->avail_wrap_counter << 15;
1939 msg->payload.state.num = val;
1941 msg->payload.state.num = vq->last_avail_idx;
1944 VHOST_LOG_CONFIG(INFO,
1945 "vring base idx:%d file:%d\n", msg->payload.state.index,
1946 msg->payload.state.num);
1948 * Based on current qemu vhost-user implementation, this message is
1949 * sent and only sent in vhost_vring_stop.
1950 * TODO: cleanup the vring, it isn't usable since here.
1952 if (vq->kickfd >= 0)
1955 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1957 if (vq->callfd >= 0)
1960 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1962 vq->signalled_used_valid = false;
1964 if (vq_is_packed(dev)) {
1965 rte_free(vq->shadow_used_packed);
1966 vq->shadow_used_packed = NULL;
1968 rte_free(vq->shadow_used_split);
1969 vq->shadow_used_split = NULL;
1970 if (vq->async_pkts_pending)
1971 rte_free(vq->async_pkts_pending);
1972 if (vq->async_pkts_info)
1973 rte_free(vq->async_pkts_info);
1974 vq->async_pkts_pending = NULL;
1975 vq->async_pkts_info = NULL;
1978 rte_free(vq->batch_copy_elems);
1979 vq->batch_copy_elems = NULL;
1981 msg->size = sizeof(msg->payload.state);
1984 vring_invalidate(dev, vq);
1986 return RTE_VHOST_MSG_RESULT_REPLY;
1990 * when virtio queues are ready to work, qemu will send us to
1991 * enable the virtio queue pair.
1994 vhost_user_set_vring_enable(struct virtio_net **pdev,
1995 struct VhostUserMsg *msg,
1996 int main_fd __rte_unused)
1998 struct virtio_net *dev = *pdev;
1999 int enable = (int)msg->payload.state.num;
2000 int index = (int)msg->payload.state.index;
2002 if (validate_msg_fds(msg, 0) != 0)
2003 return RTE_VHOST_MSG_RESULT_ERR;
2005 VHOST_LOG_CONFIG(INFO,
2006 "set queue enable: %d to qp idx: %d\n",
2009 if (enable && dev->virtqueue[index]->async_registered) {
2010 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2011 VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2012 "async inflight packets must be completed first\n");
2013 return RTE_VHOST_MSG_RESULT_ERR;
2017 dev->virtqueue[index]->enabled = enable;
2019 return RTE_VHOST_MSG_RESULT_OK;
2023 vhost_user_get_protocol_features(struct virtio_net **pdev,
2024 struct VhostUserMsg *msg,
2025 int main_fd __rte_unused)
2027 struct virtio_net *dev = *pdev;
2028 uint64_t features, protocol_features;
2030 if (validate_msg_fds(msg, 0) != 0)
2031 return RTE_VHOST_MSG_RESULT_ERR;
2033 rte_vhost_driver_get_features(dev->ifname, &features);
2034 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2036 msg->payload.u64 = protocol_features;
2037 msg->size = sizeof(msg->payload.u64);
2040 return RTE_VHOST_MSG_RESULT_REPLY;
2044 vhost_user_set_protocol_features(struct virtio_net **pdev,
2045 struct VhostUserMsg *msg,
2046 int main_fd __rte_unused)
2048 struct virtio_net *dev = *pdev;
2049 uint64_t protocol_features = msg->payload.u64;
2050 uint64_t slave_protocol_features = 0;
2052 if (validate_msg_fds(msg, 0) != 0)
2053 return RTE_VHOST_MSG_RESULT_ERR;
2055 rte_vhost_driver_get_protocol_features(dev->ifname,
2056 &slave_protocol_features);
2057 if (protocol_features & ~slave_protocol_features) {
2058 VHOST_LOG_CONFIG(ERR,
2059 "(%d) received invalid protocol features.\n",
2061 return RTE_VHOST_MSG_RESULT_ERR;
2064 dev->protocol_features = protocol_features;
2065 VHOST_LOG_CONFIG(INFO,
2066 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2067 dev->protocol_features);
2069 return RTE_VHOST_MSG_RESULT_OK;
2073 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2074 int main_fd __rte_unused)
2076 struct virtio_net *dev = *pdev;
2077 int fd = msg->fds[0];
2081 if (validate_msg_fds(msg, 1) != 0)
2082 return RTE_VHOST_MSG_RESULT_ERR;
2085 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2086 return RTE_VHOST_MSG_RESULT_ERR;
2089 if (msg->size != sizeof(VhostUserLog)) {
2090 VHOST_LOG_CONFIG(ERR,
2091 "invalid log base msg size: %"PRId32" != %d\n",
2092 msg->size, (int)sizeof(VhostUserLog));
2096 size = msg->payload.log.mmap_size;
2097 off = msg->payload.log.mmap_offset;
2099 /* Check for mmap size and offset overflow. */
2101 VHOST_LOG_CONFIG(ERR,
2102 "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2107 VHOST_LOG_CONFIG(INFO,
2108 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2112 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2113 * fail when offset is not page size aligned.
2115 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2117 if (addr == MAP_FAILED) {
2118 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2119 return RTE_VHOST_MSG_RESULT_ERR;
2123 * Free previously mapped log memory on occasionally
2124 * multiple VHOST_USER_SET_LOG_BASE.
2126 if (dev->log_addr) {
2127 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2129 dev->log_addr = (uint64_t)(uintptr_t)addr;
2130 dev->log_base = dev->log_addr + off;
2131 dev->log_size = size;
2134 * The spec is not clear about it (yet), but QEMU doesn't expect
2135 * any payload in the reply.
2140 return RTE_VHOST_MSG_RESULT_REPLY;
2144 return RTE_VHOST_MSG_RESULT_ERR;
2147 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2148 struct VhostUserMsg *msg,
2149 int main_fd __rte_unused)
2151 if (validate_msg_fds(msg, 1) != 0)
2152 return RTE_VHOST_MSG_RESULT_ERR;
2155 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2157 return RTE_VHOST_MSG_RESULT_OK;
2161 * An rarp packet is constructed and broadcasted to notify switches about
2162 * the new location of the migrated VM, so that packets from outside will
2163 * not be lost after migration.
2165 * However, we don't actually "send" a rarp packet here, instead, we set
2166 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2169 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2170 int main_fd __rte_unused)
2172 struct virtio_net *dev = *pdev;
2173 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2174 struct rte_vdpa_device *vdpa_dev;
2176 if (validate_msg_fds(msg, 0) != 0)
2177 return RTE_VHOST_MSG_RESULT_ERR;
2179 VHOST_LOG_CONFIG(DEBUG,
2180 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2181 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2182 memcpy(dev->mac.addr_bytes, mac, 6);
2185 * Set the flag to inject a RARP broadcast packet at
2186 * rte_vhost_dequeue_burst().
2188 * __ATOMIC_RELEASE ordering is for making sure the mac is
2189 * copied before the flag is set.
2191 __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2192 vdpa_dev = dev->vdpa_dev;
2193 if (vdpa_dev && vdpa_dev->ops->migration_done)
2194 vdpa_dev->ops->migration_done(dev->vid);
2196 return RTE_VHOST_MSG_RESULT_OK;
2200 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2201 int main_fd __rte_unused)
2203 struct virtio_net *dev = *pdev;
2205 if (validate_msg_fds(msg, 0) != 0)
2206 return RTE_VHOST_MSG_RESULT_ERR;
2208 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2209 msg->payload.u64 > VIRTIO_MAX_MTU) {
2210 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2213 return RTE_VHOST_MSG_RESULT_ERR;
2216 dev->mtu = msg->payload.u64;
2218 return RTE_VHOST_MSG_RESULT_OK;
2222 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2223 int main_fd __rte_unused)
2225 struct virtio_net *dev = *pdev;
2226 int fd = msg->fds[0];
2228 if (validate_msg_fds(msg, 1) != 0)
2229 return RTE_VHOST_MSG_RESULT_ERR;
2232 VHOST_LOG_CONFIG(ERR,
2233 "Invalid file descriptor for slave channel (%d)\n",
2235 return RTE_VHOST_MSG_RESULT_ERR;
2238 if (dev->slave_req_fd >= 0)
2239 close(dev->slave_req_fd);
2241 dev->slave_req_fd = fd;
2243 return RTE_VHOST_MSG_RESULT_OK;
2247 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2249 struct vhost_vring_addr *ra;
2250 uint64_t start, end, len;
2253 end = start + imsg->size;
2255 ra = &vq->ring_addrs;
2256 len = sizeof(struct vring_desc) * vq->size;
2257 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2260 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2261 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2264 len = sizeof(struct vring_used) +
2265 sizeof(struct vring_used_elem) * vq->size;
2266 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2269 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2270 len = sizeof(uint64_t);
2271 if (ra->log_guest_addr < end &&
2272 (ra->log_guest_addr + len) > start)
2280 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2282 struct vhost_vring_addr *ra;
2283 uint64_t start, end, len;
2286 end = start + imsg->size;
2288 ra = &vq->ring_addrs;
2289 len = sizeof(struct vring_packed_desc) * vq->size;
2290 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2293 len = sizeof(struct vring_packed_desc_event);
2294 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2297 len = sizeof(struct vring_packed_desc_event);
2298 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2301 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2302 len = sizeof(uint64_t);
2303 if (ra->log_guest_addr < end &&
2304 (ra->log_guest_addr + len) > start)
2311 static int is_vring_iotlb(struct virtio_net *dev,
2312 struct vhost_virtqueue *vq,
2313 struct vhost_iotlb_msg *imsg)
2315 if (vq_is_packed(dev))
2316 return is_vring_iotlb_packed(vq, imsg);
2318 return is_vring_iotlb_split(vq, imsg);
2322 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2323 int main_fd __rte_unused)
2325 struct virtio_net *dev = *pdev;
2326 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2330 if (validate_msg_fds(msg, 0) != 0)
2331 return RTE_VHOST_MSG_RESULT_ERR;
2333 switch (imsg->type) {
2334 case VHOST_IOTLB_UPDATE:
2336 vva = qva_to_vva(dev, imsg->uaddr, &len);
2338 return RTE_VHOST_MSG_RESULT_ERR;
2340 for (i = 0; i < dev->nr_vring; i++) {
2341 struct vhost_virtqueue *vq = dev->virtqueue[i];
2346 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2349 if (is_vring_iotlb(dev, vq, imsg))
2350 *pdev = dev = translate_ring_addresses(dev, i);
2353 case VHOST_IOTLB_INVALIDATE:
2354 for (i = 0; i < dev->nr_vring; i++) {
2355 struct vhost_virtqueue *vq = dev->virtqueue[i];
2360 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2363 if (is_vring_iotlb(dev, vq, imsg))
2364 vring_invalidate(dev, vq);
2368 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2370 return RTE_VHOST_MSG_RESULT_ERR;
2373 return RTE_VHOST_MSG_RESULT_OK;
2377 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2378 struct VhostUserMsg *msg,
2379 int main_fd __rte_unused)
2381 struct virtio_net *dev = *pdev;
2382 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2383 struct uffdio_api api_struct;
2385 if (validate_msg_fds(msg, 0) != 0)
2386 return RTE_VHOST_MSG_RESULT_ERR;
2388 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2390 if (dev->postcopy_ufd == -1) {
2391 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2393 return RTE_VHOST_MSG_RESULT_ERR;
2395 api_struct.api = UFFD_API;
2396 api_struct.features = 0;
2397 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2398 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2400 close(dev->postcopy_ufd);
2401 dev->postcopy_ufd = -1;
2402 return RTE_VHOST_MSG_RESULT_ERR;
2404 msg->fds[0] = dev->postcopy_ufd;
2407 return RTE_VHOST_MSG_RESULT_REPLY;
2409 dev->postcopy_ufd = -1;
2412 return RTE_VHOST_MSG_RESULT_ERR;
2417 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2418 struct VhostUserMsg *msg __rte_unused,
2419 int main_fd __rte_unused)
2421 struct virtio_net *dev = *pdev;
2423 if (validate_msg_fds(msg, 0) != 0)
2424 return RTE_VHOST_MSG_RESULT_ERR;
2426 if (dev->mem && dev->mem->nregions) {
2427 VHOST_LOG_CONFIG(ERR,
2428 "Regions already registered at postcopy-listen\n");
2429 return RTE_VHOST_MSG_RESULT_ERR;
2431 dev->postcopy_listening = 1;
2433 return RTE_VHOST_MSG_RESULT_OK;
2437 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2438 int main_fd __rte_unused)
2440 struct virtio_net *dev = *pdev;
2442 if (validate_msg_fds(msg, 0) != 0)
2443 return RTE_VHOST_MSG_RESULT_ERR;
2445 dev->postcopy_listening = 0;
2446 if (dev->postcopy_ufd >= 0) {
2447 close(dev->postcopy_ufd);
2448 dev->postcopy_ufd = -1;
2451 msg->payload.u64 = 0;
2452 msg->size = sizeof(msg->payload.u64);
2455 return RTE_VHOST_MSG_RESULT_REPLY;
2459 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2460 int main_fd __rte_unused)
2462 struct virtio_net *dev = *pdev;
2464 if (validate_msg_fds(msg, 0) != 0)
2465 return RTE_VHOST_MSG_RESULT_ERR;
2467 msg->payload.u64 = dev->status;
2468 msg->size = sizeof(msg->payload.u64);
2471 return RTE_VHOST_MSG_RESULT_REPLY;
2475 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2476 int main_fd __rte_unused)
2478 struct virtio_net *dev = *pdev;
2480 if (validate_msg_fds(msg, 0) != 0)
2481 return RTE_VHOST_MSG_RESULT_ERR;
2483 /* As per Virtio specification, the device status is 8bits long */
2484 if (msg->payload.u64 > UINT8_MAX) {
2485 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2487 return RTE_VHOST_MSG_RESULT_ERR;
2490 dev->status = msg->payload.u64;
2492 if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2493 (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2494 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2496 * Clear the bit to let the driver know about the feature
2497 * negotiation failure
2499 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2502 VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2504 "\t-ACKNOWLEDGE: %u\n"
2506 "\t-FEATURES_OK: %u\n"
2507 "\t-DRIVER_OK: %u\n"
2508 "\t-DEVICE_NEED_RESET: %u\n"
2511 (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2512 !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2513 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2514 !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2515 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2516 !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2517 !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2519 return RTE_VHOST_MSG_RESULT_OK;
2522 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2523 struct VhostUserMsg *msg,
2525 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2526 [VHOST_USER_NONE] = NULL,
2527 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2528 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2529 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2530 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2531 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2532 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2533 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2534 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2535 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2536 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2537 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2538 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2539 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2540 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2541 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2542 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2543 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2544 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2545 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2546 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2547 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2548 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2549 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2550 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2551 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2552 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2553 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2554 [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2555 [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2558 /* return bytes# of read on success or negative val on failure. */
2560 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2564 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2565 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2568 } else if (ret != VHOST_USER_HDR_SIZE) {
2569 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2575 if (msg->size > sizeof(msg->payload)) {
2576 VHOST_LOG_CONFIG(ERR,
2577 "invalid msg size: %d\n", msg->size);
2580 ret = read(sockfd, &msg->payload, msg->size);
2583 if (ret != (int)msg->size) {
2584 VHOST_LOG_CONFIG(ERR,
2585 "read control message failed\n");
2594 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2599 return send_fd_message(sockfd, (char *)msg,
2600 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2604 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2609 msg->flags &= ~VHOST_USER_VERSION_MASK;
2610 msg->flags &= ~VHOST_USER_NEED_REPLY;
2611 msg->flags |= VHOST_USER_VERSION;
2612 msg->flags |= VHOST_USER_REPLY_MASK;
2614 return send_vhost_message(sockfd, msg);
2618 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2622 if (msg->flags & VHOST_USER_NEED_REPLY)
2623 rte_spinlock_lock(&dev->slave_req_lock);
2625 ret = send_vhost_message(dev->slave_req_fd, msg);
2626 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2627 rte_spinlock_unlock(&dev->slave_req_lock);
2633 * Allocate a queue pair if it hasn't been allocated yet
2636 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2637 struct VhostUserMsg *msg)
2641 switch (msg->request.master) {
2642 case VHOST_USER_SET_VRING_KICK:
2643 case VHOST_USER_SET_VRING_CALL:
2644 case VHOST_USER_SET_VRING_ERR:
2645 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2647 case VHOST_USER_SET_VRING_NUM:
2648 case VHOST_USER_SET_VRING_BASE:
2649 case VHOST_USER_SET_VRING_ENABLE:
2650 vring_idx = msg->payload.state.index;
2652 case VHOST_USER_SET_VRING_ADDR:
2653 vring_idx = msg->payload.addr.index;
2659 if (vring_idx >= VHOST_MAX_VRING) {
2660 VHOST_LOG_CONFIG(ERR,
2661 "invalid vring index: %u\n", vring_idx);
2665 if (dev->virtqueue[vring_idx])
2668 return alloc_vring_queue(dev, vring_idx);
2672 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2675 unsigned int vq_num = 0;
2677 while (vq_num < dev->nr_vring) {
2678 struct vhost_virtqueue *vq = dev->virtqueue[i];
2681 rte_spinlock_lock(&vq->access_lock);
2689 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2692 unsigned int vq_num = 0;
2694 while (vq_num < dev->nr_vring) {
2695 struct vhost_virtqueue *vq = dev->virtqueue[i];
2698 rte_spinlock_unlock(&vq->access_lock);
2706 vhost_user_msg_handler(int vid, int fd)
2708 struct virtio_net *dev;
2709 struct VhostUserMsg msg;
2710 struct rte_vdpa_device *vdpa_dev;
2712 int unlock_required = 0;
2717 dev = get_device(vid);
2721 if (!dev->notify_ops) {
2722 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2723 if (!dev->notify_ops) {
2724 VHOST_LOG_CONFIG(ERR,
2725 "failed to get callback ops for driver %s\n",
2731 ret = read_vhost_message(fd, &msg);
2734 VHOST_LOG_CONFIG(ERR,
2735 "vhost read message failed\n");
2737 VHOST_LOG_CONFIG(INFO,
2738 "vhost peer closed\n");
2744 request = msg.request.master;
2745 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2746 vhost_message_str[request]) {
2747 if (request != VHOST_USER_IOTLB_MSG)
2748 VHOST_LOG_CONFIG(INFO, "read message %s\n",
2749 vhost_message_str[request]);
2751 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2752 vhost_message_str[request]);
2754 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2757 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2759 VHOST_LOG_CONFIG(ERR,
2760 "failed to alloc queue\n");
2765 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2766 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2767 * and device is destroyed. destroy_device waits for queues to be
2768 * inactive, so it is safe. Otherwise taking the access_lock
2769 * would cause a dead lock.
2772 case VHOST_USER_SET_FEATURES:
2773 case VHOST_USER_SET_PROTOCOL_FEATURES:
2774 case VHOST_USER_SET_OWNER:
2775 case VHOST_USER_SET_MEM_TABLE:
2776 case VHOST_USER_SET_LOG_BASE:
2777 case VHOST_USER_SET_LOG_FD:
2778 case VHOST_USER_SET_VRING_NUM:
2779 case VHOST_USER_SET_VRING_ADDR:
2780 case VHOST_USER_SET_VRING_BASE:
2781 case VHOST_USER_SET_VRING_KICK:
2782 case VHOST_USER_SET_VRING_CALL:
2783 case VHOST_USER_SET_VRING_ERR:
2784 case VHOST_USER_SET_VRING_ENABLE:
2785 case VHOST_USER_SEND_RARP:
2786 case VHOST_USER_NET_SET_MTU:
2787 case VHOST_USER_SET_SLAVE_REQ_FD:
2788 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2789 vhost_user_lock_all_queue_pairs(dev);
2790 unlock_required = 1;
2799 if (dev->extern_ops.pre_msg_handle) {
2800 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2803 case RTE_VHOST_MSG_RESULT_REPLY:
2804 send_vhost_reply(fd, &msg);
2806 case RTE_VHOST_MSG_RESULT_ERR:
2807 case RTE_VHOST_MSG_RESULT_OK:
2809 goto skip_to_post_handle;
2810 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2816 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2817 if (!vhost_message_handlers[request])
2818 goto skip_to_post_handle;
2819 ret = vhost_message_handlers[request](&dev, &msg, fd);
2822 case RTE_VHOST_MSG_RESULT_ERR:
2823 VHOST_LOG_CONFIG(ERR,
2824 "Processing %s failed.\n",
2825 vhost_message_str[request]);
2828 case RTE_VHOST_MSG_RESULT_OK:
2829 VHOST_LOG_CONFIG(DEBUG,
2830 "Processing %s succeeded.\n",
2831 vhost_message_str[request]);
2834 case RTE_VHOST_MSG_RESULT_REPLY:
2835 VHOST_LOG_CONFIG(DEBUG,
2836 "Processing %s succeeded and needs reply.\n",
2837 vhost_message_str[request]);
2838 send_vhost_reply(fd, &msg);
2846 skip_to_post_handle:
2847 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2848 dev->extern_ops.post_msg_handle) {
2849 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2852 case RTE_VHOST_MSG_RESULT_REPLY:
2853 send_vhost_reply(fd, &msg);
2855 case RTE_VHOST_MSG_RESULT_ERR:
2856 case RTE_VHOST_MSG_RESULT_OK:
2858 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2864 if (unlock_required)
2865 vhost_user_unlock_all_queue_pairs(dev);
2867 /* If message was not handled at this stage, treat it as an error */
2869 VHOST_LOG_CONFIG(ERR,
2870 "vhost message (req: %d) was not handled.\n", request);
2871 close_msg_fds(&msg);
2872 ret = RTE_VHOST_MSG_RESULT_ERR;
2876 * If the request required a reply that was already sent,
2877 * this optional reply-ack won't be sent as the
2878 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2880 if (msg.flags & VHOST_USER_NEED_REPLY) {
2881 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2882 msg.size = sizeof(msg.payload.u64);
2884 send_vhost_reply(fd, &msg);
2885 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2886 VHOST_LOG_CONFIG(ERR,
2887 "vhost message handling failed.\n");
2891 for (i = 0; i < dev->nr_vring; i++) {
2892 struct vhost_virtqueue *vq = dev->virtqueue[i];
2893 bool cur_ready = vq_is_ready(dev, vq);
2895 if (cur_ready != (vq && vq->ready)) {
2896 vq->ready = cur_ready;
2897 vhost_user_notify_queue_state(dev, i, cur_ready);
2902 if (!virtio_is_ready(dev))
2906 * Virtio is now ready. If not done already, it is time
2907 * to notify the application it can process the rings and
2908 * configure the vDPA device if present.
2911 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2912 if (dev->notify_ops->new_device(dev->vid) == 0)
2913 dev->flags |= VIRTIO_DEV_RUNNING;
2916 vdpa_dev = dev->vdpa_dev;
2920 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2921 if (vdpa_dev->ops->dev_conf(dev->vid))
2922 VHOST_LOG_CONFIG(ERR,
2923 "Failed to configure vDPA device\n");
2925 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2932 static int process_slave_message_reply(struct virtio_net *dev,
2933 const struct VhostUserMsg *msg)
2935 struct VhostUserMsg msg_reply;
2938 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2941 ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
2944 VHOST_LOG_CONFIG(ERR,
2945 "vhost read slave message reply failed\n");
2947 VHOST_LOG_CONFIG(INFO,
2948 "vhost peer closed\n");
2954 if (msg_reply.request.slave != msg->request.slave) {
2955 VHOST_LOG_CONFIG(ERR,
2956 "Received unexpected msg type (%u), expected %u\n",
2957 msg_reply.request.slave, msg->request.slave);
2962 ret = msg_reply.payload.u64 ? -1 : 0;
2965 rte_spinlock_unlock(&dev->slave_req_lock);
2970 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2973 struct VhostUserMsg msg = {
2974 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2975 .flags = VHOST_USER_VERSION,
2976 .size = sizeof(msg.payload.iotlb),
2980 .type = VHOST_IOTLB_MISS,
2984 ret = send_vhost_message(dev->slave_req_fd, &msg);
2986 VHOST_LOG_CONFIG(ERR,
2987 "Failed to send IOTLB miss message (%d)\n",
2996 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
2999 struct VhostUserMsg msg = {
3000 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3001 .flags = VHOST_USER_VERSION,
3006 msg.flags |= VHOST_USER_NEED_REPLY;
3008 ret = send_vhost_slave_message(dev, &msg);
3010 VHOST_LOG_CONFIG(ERR,
3011 "Failed to send config change (%d)\n",
3016 return process_slave_message_reply(dev, &msg);
3020 rte_vhost_slave_config_change(int vid, bool need_reply)
3022 struct virtio_net *dev;
3024 dev = get_device(vid);
3028 return vhost_user_slave_config_change(dev, need_reply);
3031 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3037 struct VhostUserMsg msg = {
3038 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3039 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3040 .size = sizeof(msg.payload.area),
3042 .u64 = index & VHOST_USER_VRING_IDX_MASK,
3049 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3055 ret = send_vhost_slave_message(dev, &msg);
3057 VHOST_LOG_CONFIG(ERR,
3058 "Failed to set host notifier (%d)\n", ret);
3062 return process_slave_message_reply(dev, &msg);
3065 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3067 struct virtio_net *dev;
3068 struct rte_vdpa_device *vdpa_dev;
3069 int vfio_device_fd, ret = 0;
3070 uint64_t offset, size;
3071 unsigned int i, q_start, q_last;
3073 dev = get_device(vid);
3077 vdpa_dev = dev->vdpa_dev;
3078 if (vdpa_dev == NULL)
3081 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3082 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3083 !(dev->protocol_features &
3084 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3085 !(dev->protocol_features &
3086 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3087 !(dev->protocol_features &
3088 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3091 if (qid == RTE_VHOST_QUEUE_ALL) {
3093 q_last = dev->nr_vring - 1;
3095 if (qid >= dev->nr_vring)
3101 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3102 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3104 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3105 if (vfio_device_fd < 0)
3109 for (i = q_start; i <= q_last; i++) {
3110 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3116 if (vhost_user_slave_set_vring_host_notifier(dev, i,
3117 vfio_device_fd, offset, size) < 0) {
3124 for (i = q_start; i <= q_last; i++) {
3125 vhost_user_slave_set_vring_host_notifier(dev, i, -1,