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 if (msg->payload.state.num > 32768) {
410 VHOST_LOG_CONFIG(ERR, "invalid virtqueue size %u\n", msg->payload.state.num);
411 return RTE_VHOST_MSG_RESULT_ERR;
414 vq->size = msg->payload.state.num;
416 /* VIRTIO 1.0, 2.4 Virtqueues says:
418 * Queue Size value is always a power of 2. The maximum Queue Size
421 * VIRTIO 1.1 2.7 Virtqueues says:
423 * Packed virtqueues support up to 2^15 entries each.
425 if (!vq_is_packed(dev)) {
426 if (vq->size & (vq->size - 1)) {
427 VHOST_LOG_CONFIG(ERR,
428 "invalid virtqueue size %u\n", vq->size);
429 return RTE_VHOST_MSG_RESULT_ERR;
433 if (vq_is_packed(dev)) {
434 if (vq->shadow_used_packed)
435 rte_free(vq->shadow_used_packed);
436 vq->shadow_used_packed = rte_malloc(NULL,
438 sizeof(struct vring_used_elem_packed),
439 RTE_CACHE_LINE_SIZE);
440 if (!vq->shadow_used_packed) {
441 VHOST_LOG_CONFIG(ERR,
442 "failed to allocate memory for shadow used ring.\n");
443 return RTE_VHOST_MSG_RESULT_ERR;
447 if (vq->shadow_used_split)
448 rte_free(vq->shadow_used_split);
450 vq->shadow_used_split = rte_malloc(NULL,
451 vq->size * sizeof(struct vring_used_elem),
452 RTE_CACHE_LINE_SIZE);
454 if (!vq->shadow_used_split) {
455 VHOST_LOG_CONFIG(ERR,
456 "failed to allocate memory for vq internal data.\n");
457 return RTE_VHOST_MSG_RESULT_ERR;
461 if (vq->batch_copy_elems)
462 rte_free(vq->batch_copy_elems);
463 vq->batch_copy_elems = rte_malloc(NULL,
464 vq->size * sizeof(struct batch_copy_elem),
465 RTE_CACHE_LINE_SIZE);
466 if (!vq->batch_copy_elems) {
467 VHOST_LOG_CONFIG(ERR,
468 "failed to allocate memory for batching copy.\n");
469 return RTE_VHOST_MSG_RESULT_ERR;
472 return RTE_VHOST_MSG_RESULT_OK;
476 * Reallocate virtio_dev, vhost_virtqueue and related data structures to
477 * make them on the same numa node as the memory of vring descriptor.
479 #ifdef RTE_LIBRTE_VHOST_NUMA
480 static struct virtio_net*
481 numa_realloc(struct virtio_net *dev, int index)
483 int oldnode, newnode;
484 struct virtio_net *old_dev;
485 struct vhost_virtqueue *old_vq, *vq;
486 struct vring_used_elem *new_shadow_used_split;
487 struct vring_used_elem_packed *new_shadow_used_packed;
488 struct batch_copy_elem *new_batch_copy_elems;
491 if (dev->flags & VIRTIO_DEV_RUNNING)
495 vq = old_vq = dev->virtqueue[index];
497 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
498 MPOL_F_NODE | MPOL_F_ADDR);
500 /* check if we need to reallocate vq */
501 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
502 MPOL_F_NODE | MPOL_F_ADDR);
504 VHOST_LOG_CONFIG(ERR,
505 "Unable to get vq numa information.\n");
508 if (oldnode != newnode) {
509 VHOST_LOG_CONFIG(INFO,
510 "reallocate vq from %d to %d node\n", oldnode, newnode);
511 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
515 memcpy(vq, old_vq, sizeof(*vq));
517 if (vq_is_packed(dev)) {
518 new_shadow_used_packed = rte_malloc_socket(NULL,
520 sizeof(struct vring_used_elem_packed),
523 if (new_shadow_used_packed) {
524 rte_free(vq->shadow_used_packed);
525 vq->shadow_used_packed = new_shadow_used_packed;
528 new_shadow_used_split = rte_malloc_socket(NULL,
530 sizeof(struct vring_used_elem),
533 if (new_shadow_used_split) {
534 rte_free(vq->shadow_used_split);
535 vq->shadow_used_split = new_shadow_used_split;
539 new_batch_copy_elems = rte_malloc_socket(NULL,
540 vq->size * sizeof(struct batch_copy_elem),
543 if (new_batch_copy_elems) {
544 rte_free(vq->batch_copy_elems);
545 vq->batch_copy_elems = new_batch_copy_elems;
551 /* check if we need to reallocate dev */
552 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
553 MPOL_F_NODE | MPOL_F_ADDR);
555 VHOST_LOG_CONFIG(ERR,
556 "Unable to get dev numa information.\n");
559 if (oldnode != newnode) {
560 struct rte_vhost_memory *old_mem;
563 VHOST_LOG_CONFIG(INFO,
564 "reallocate dev from %d to %d node\n",
566 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
572 memcpy(dev, old_dev, sizeof(*dev));
575 mem_size = sizeof(struct rte_vhost_memory) +
576 sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
578 dev->mem = rte_malloc_socket(NULL, mem_size, 0, newnode);
584 memcpy(dev->mem, old_mem, mem_size);
589 dev->virtqueue[index] = vq;
590 vhost_devices[dev->vid] = dev;
593 vhost_user_iotlb_init(dev, index);
598 static struct virtio_net*
599 numa_realloc(struct virtio_net *dev, int index __rte_unused)
605 /* Converts QEMU virtual address to Vhost virtual address. */
607 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
609 struct rte_vhost_mem_region *r;
612 if (unlikely(!dev || !dev->mem))
615 /* Find the region where the address lives. */
616 for (i = 0; i < dev->mem->nregions; i++) {
617 r = &dev->mem->regions[i];
619 if (qva >= r->guest_user_addr &&
620 qva < r->guest_user_addr + r->size) {
622 if (unlikely(*len > r->guest_user_addr + r->size - qva))
623 *len = r->guest_user_addr + r->size - qva;
625 return qva - r->guest_user_addr +
637 * Converts ring address to Vhost virtual address.
638 * If IOMMU is enabled, the ring address is a guest IO virtual address,
639 * else it is a QEMU virtual address.
642 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
643 uint64_t ra, uint64_t *size)
645 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
648 vhost_user_iotlb_rd_lock(vq);
649 vva = vhost_iova_to_vva(dev, vq, ra,
650 size, VHOST_ACCESS_RW);
651 vhost_user_iotlb_rd_unlock(vq);
656 return qva_to_vva(dev, ra, size);
660 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
664 vhost_user_iotlb_rd_lock(vq);
665 log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
666 vhost_user_iotlb_rd_unlock(vq);
671 static struct virtio_net *
672 translate_ring_addresses(struct virtio_net *dev, int vq_index)
674 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
675 struct vhost_vring_addr *addr = &vq->ring_addrs;
676 uint64_t len, expected_len;
678 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
680 log_addr_to_gpa(dev, vq);
681 if (vq->log_guest_addr == 0) {
682 VHOST_LOG_CONFIG(DEBUG,
683 "(%d) failed to map log_guest_addr.\n",
689 if (vq_is_packed(dev)) {
690 len = sizeof(struct vring_packed_desc) * vq->size;
691 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
692 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
693 if (vq->desc_packed == NULL ||
694 len != sizeof(struct vring_packed_desc) *
696 VHOST_LOG_CONFIG(DEBUG,
697 "(%d) failed to map desc_packed ring.\n",
702 dev = numa_realloc(dev, vq_index);
703 vq = dev->virtqueue[vq_index];
704 addr = &vq->ring_addrs;
706 len = sizeof(struct vring_packed_desc_event);
707 vq->driver_event = (struct vring_packed_desc_event *)
708 (uintptr_t)ring_addr_to_vva(dev,
709 vq, addr->avail_user_addr, &len);
710 if (vq->driver_event == NULL ||
711 len != sizeof(struct vring_packed_desc_event)) {
712 VHOST_LOG_CONFIG(DEBUG,
713 "(%d) failed to find driver area address.\n",
718 len = sizeof(struct vring_packed_desc_event);
719 vq->device_event = (struct vring_packed_desc_event *)
720 (uintptr_t)ring_addr_to_vva(dev,
721 vq, addr->used_user_addr, &len);
722 if (vq->device_event == NULL ||
723 len != sizeof(struct vring_packed_desc_event)) {
724 VHOST_LOG_CONFIG(DEBUG,
725 "(%d) failed to find device area address.\n",
730 vq->access_ok = true;
734 /* The addresses are converted from QEMU virtual to Vhost virtual. */
735 if (vq->desc && vq->avail && vq->used)
738 len = sizeof(struct vring_desc) * vq->size;
739 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
740 vq, addr->desc_user_addr, &len);
741 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
742 VHOST_LOG_CONFIG(DEBUG,
743 "(%d) failed to map desc ring.\n",
748 dev = numa_realloc(dev, vq_index);
749 vq = dev->virtqueue[vq_index];
750 addr = &vq->ring_addrs;
752 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
753 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
754 len += sizeof(uint16_t);
756 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
757 vq, addr->avail_user_addr, &len);
758 if (vq->avail == 0 || len != expected_len) {
759 VHOST_LOG_CONFIG(DEBUG,
760 "(%d) failed to map avail ring.\n",
765 len = sizeof(struct vring_used) +
766 sizeof(struct vring_used_elem) * vq->size;
767 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
768 len += sizeof(uint16_t);
770 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
771 vq, addr->used_user_addr, &len);
772 if (vq->used == 0 || len != expected_len) {
773 VHOST_LOG_CONFIG(DEBUG,
774 "(%d) failed to map used ring.\n",
779 if (vq->last_used_idx != vq->used->idx) {
780 VHOST_LOG_CONFIG(WARNING,
781 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
782 "some packets maybe resent for Tx and dropped for Rx\n",
783 vq->last_used_idx, vq->used->idx);
784 vq->last_used_idx = vq->used->idx;
785 vq->last_avail_idx = vq->used->idx;
788 vq->access_ok = true;
790 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
792 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
793 dev->vid, vq->avail);
794 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
796 VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
797 dev->vid, vq->log_guest_addr);
803 * The virtio device sends us the desc, used and avail ring addresses.
804 * This function then converts these to our address space.
807 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
808 int main_fd __rte_unused)
810 struct virtio_net *dev = *pdev;
811 struct vhost_virtqueue *vq;
812 struct vhost_vring_addr *addr = &msg->payload.addr;
815 if (validate_msg_fds(msg, 0) != 0)
816 return RTE_VHOST_MSG_RESULT_ERR;
818 if (dev->mem == NULL)
819 return RTE_VHOST_MSG_RESULT_ERR;
821 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
822 vq = dev->virtqueue[msg->payload.addr.index];
824 access_ok = vq->access_ok;
827 * Rings addresses should not be interpreted as long as the ring is not
828 * started and enabled
830 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
832 vring_invalidate(dev, vq);
834 if ((vq->enabled && (dev->features &
835 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
837 dev = translate_ring_addresses(dev, msg->payload.addr.index);
839 return RTE_VHOST_MSG_RESULT_ERR;
844 return RTE_VHOST_MSG_RESULT_OK;
848 * The virtio device sends us the available ring last used index.
851 vhost_user_set_vring_base(struct virtio_net **pdev,
852 struct VhostUserMsg *msg,
853 int main_fd __rte_unused)
855 struct virtio_net *dev = *pdev;
856 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
857 uint64_t val = msg->payload.state.num;
859 if (validate_msg_fds(msg, 0) != 0)
860 return RTE_VHOST_MSG_RESULT_ERR;
862 if (vq_is_packed(dev)) {
864 * Bit[0:14]: avail index
865 * Bit[15]: avail wrap counter
867 vq->last_avail_idx = val & 0x7fff;
868 vq->avail_wrap_counter = !!(val & (0x1 << 15));
870 * Set used index to same value as available one, as
871 * their values should be the same since ring processing
872 * was stopped at get time.
874 vq->last_used_idx = vq->last_avail_idx;
875 vq->used_wrap_counter = vq->avail_wrap_counter;
877 vq->last_used_idx = msg->payload.state.num;
878 vq->last_avail_idx = msg->payload.state.num;
881 return RTE_VHOST_MSG_RESULT_OK;
885 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
886 uint64_t host_phys_addr, uint64_t size)
888 struct guest_page *page, *last_page;
889 struct guest_page *old_pages;
891 if (dev->nr_guest_pages == dev->max_guest_pages) {
892 dev->max_guest_pages *= 2;
893 old_pages = dev->guest_pages;
894 dev->guest_pages = rte_realloc(dev->guest_pages,
895 dev->max_guest_pages * sizeof(*page),
896 RTE_CACHE_LINE_SIZE);
897 if (dev->guest_pages == NULL) {
898 VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
904 if (dev->nr_guest_pages > 0) {
905 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
906 /* merge if the two pages are continuous */
907 if (host_phys_addr == last_page->host_phys_addr +
909 last_page->size += size;
914 page = &dev->guest_pages[dev->nr_guest_pages++];
915 page->guest_phys_addr = guest_phys_addr;
916 page->host_phys_addr = host_phys_addr;
923 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
926 uint64_t reg_size = reg->size;
927 uint64_t host_user_addr = reg->host_user_addr;
928 uint64_t guest_phys_addr = reg->guest_phys_addr;
929 uint64_t host_phys_addr;
932 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
933 size = page_size - (guest_phys_addr & (page_size - 1));
934 size = RTE_MIN(size, reg_size);
936 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
939 host_user_addr += size;
940 guest_phys_addr += size;
943 while (reg_size > 0) {
944 size = RTE_MIN(reg_size, page_size);
945 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
947 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
951 host_user_addr += size;
952 guest_phys_addr += size;
956 /* sort guest page array if over binary search threshold */
957 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
958 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
959 sizeof(struct guest_page), guest_page_addrcmp);
965 #ifdef RTE_LIBRTE_VHOST_DEBUG
966 /* TODO: enable it only in debug mode? */
968 dump_guest_pages(struct virtio_net *dev)
971 struct guest_page *page;
973 for (i = 0; i < dev->nr_guest_pages; i++) {
974 page = &dev->guest_pages[i];
976 VHOST_LOG_CONFIG(INFO,
977 "guest physical page region %u\n"
978 "\t guest_phys_addr: %" PRIx64 "\n"
979 "\t host_phys_addr : %" PRIx64 "\n"
980 "\t size : %" PRIx64 "\n",
982 page->guest_phys_addr,
983 page->host_phys_addr,
988 #define dump_guest_pages(dev)
992 vhost_memory_changed(struct VhostUserMemory *new,
993 struct rte_vhost_memory *old)
997 if (new->nregions != old->nregions)
1000 for (i = 0; i < new->nregions; ++i) {
1001 VhostUserMemoryRegion *new_r = &new->regions[i];
1002 struct rte_vhost_mem_region *old_r = &old->regions[i];
1004 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1006 if (new_r->memory_size != old_r->size)
1008 if (new_r->userspace_addr != old_r->guest_user_addr)
1015 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1017 vhost_user_postcopy_region_register(struct virtio_net *dev,
1018 struct rte_vhost_mem_region *reg)
1020 struct uffdio_register reg_struct;
1023 * Let's register all the mmap'ed area to ensure
1024 * alignment on page boundary.
1026 reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1027 reg_struct.range.len = reg->mmap_size;
1028 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1030 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1032 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1033 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1034 (uint64_t)reg_struct.range.start,
1035 (uint64_t)reg_struct.range.start +
1036 (uint64_t)reg_struct.range.len - 1,
1042 VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1043 (uint64_t)reg_struct.range.start,
1044 (uint64_t)reg_struct.range.start +
1045 (uint64_t)reg_struct.range.len - 1);
1051 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1052 struct rte_vhost_mem_region *reg __rte_unused)
1059 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1060 struct VhostUserMsg *msg)
1062 struct VhostUserMemory *memory;
1063 struct rte_vhost_mem_region *reg;
1064 VhostUserMsg ack_msg;
1067 if (!dev->postcopy_listening)
1071 * We haven't a better way right now than sharing
1072 * DPDK's virtual address with Qemu, so that Qemu can
1073 * retrieve the region offset when handling userfaults.
1075 memory = &msg->payload.memory;
1076 for (i = 0; i < memory->nregions; i++) {
1077 reg = &dev->mem->regions[i];
1078 memory->regions[i].userspace_addr = reg->host_user_addr;
1081 /* Send the addresses back to qemu */
1083 send_vhost_reply(main_fd, msg);
1085 /* Wait for qemu to acknolwedge it's got the addresses
1086 * we've got to wait before we're allowed to generate faults.
1088 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1089 VHOST_LOG_CONFIG(ERR,
1090 "Failed to read qemu ack on postcopy set-mem-table\n");
1094 if (validate_msg_fds(&ack_msg, 0) != 0)
1097 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1098 VHOST_LOG_CONFIG(ERR,
1099 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1100 ack_msg.request.master);
1104 /* Now userfault register and we can use the memory */
1105 for (i = 0; i < memory->nregions; i++) {
1106 reg = &dev->mem->regions[i];
1107 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1115 vhost_user_mmap_region(struct virtio_net *dev,
1116 struct rte_vhost_mem_region *region,
1117 uint64_t mmap_offset)
1124 /* Check for memory_size + mmap_offset overflow */
1125 if (mmap_offset >= -region->size) {
1126 VHOST_LOG_CONFIG(ERR,
1127 "mmap_offset (%#"PRIx64") and memory_size "
1128 "(%#"PRIx64") overflow\n",
1129 mmap_offset, region->size);
1133 mmap_size = region->size + mmap_offset;
1135 /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1136 * argument aligned with hugepagesz at older longterm version Linux,
1137 * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1139 * To avoid failure, make sure in caller to keep length aligned.
1141 alignment = get_blk_size(region->fd);
1142 if (alignment == (uint64_t)-1) {
1143 VHOST_LOG_CONFIG(ERR,
1144 "couldn't get hugepage size through fstat\n");
1147 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1148 if (mmap_size == 0) {
1150 * It could happen if initial mmap_size + alignment overflows
1151 * the sizeof uint64, which could happen if either mmap_size or
1152 * alignment value is wrong.
1154 * mmap() kernel implementation would return an error, but
1155 * better catch it before and provide useful info in the logs.
1157 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1158 "or alignment (0x%" PRIx64 ") is invalid\n",
1159 region->size + mmap_offset, alignment);
1163 populate = dev->async_copy ? MAP_POPULATE : 0;
1164 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1165 MAP_SHARED | populate, region->fd, 0);
1167 if (mmap_addr == MAP_FAILED) {
1168 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1172 region->mmap_addr = mmap_addr;
1173 region->mmap_size = mmap_size;
1174 region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1176 if (dev->async_copy)
1177 if (add_guest_pages(dev, region, alignment) < 0) {
1178 VHOST_LOG_CONFIG(ERR,
1179 "adding guest pages to region failed.\n");
1183 VHOST_LOG_CONFIG(INFO,
1184 "guest memory region size: 0x%" PRIx64 "\n"
1185 "\t guest physical addr: 0x%" PRIx64 "\n"
1186 "\t guest virtual addr: 0x%" PRIx64 "\n"
1187 "\t host virtual addr: 0x%" PRIx64 "\n"
1188 "\t mmap addr : 0x%" PRIx64 "\n"
1189 "\t mmap size : 0x%" PRIx64 "\n"
1190 "\t mmap align: 0x%" PRIx64 "\n"
1191 "\t mmap off : 0x%" PRIx64 "\n",
1193 region->guest_phys_addr,
1194 region->guest_user_addr,
1195 region->host_user_addr,
1196 (uint64_t)(uintptr_t)mmap_addr,
1205 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1208 struct virtio_net *dev = *pdev;
1209 struct VhostUserMemory *memory = &msg->payload.memory;
1210 struct rte_vhost_mem_region *reg;
1212 uint64_t mmap_offset;
1215 if (validate_msg_fds(msg, memory->nregions) != 0)
1216 return RTE_VHOST_MSG_RESULT_ERR;
1218 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1219 VHOST_LOG_CONFIG(ERR,
1220 "too many memory regions (%u)\n", memory->nregions);
1224 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1225 VHOST_LOG_CONFIG(INFO,
1226 "(%d) memory regions not changed\n", dev->vid);
1230 return RTE_VHOST_MSG_RESULT_OK;
1234 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1235 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1237 if (vdpa_dev && vdpa_dev->ops->dev_close)
1238 vdpa_dev->ops->dev_close(dev->vid);
1239 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1241 free_mem_region(dev);
1246 /* Flush IOTLB cache as previous HVAs are now invalid */
1247 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1248 for (i = 0; i < dev->nr_vring; i++)
1249 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1251 dev->nr_guest_pages = 0;
1252 if (dev->guest_pages == NULL) {
1253 dev->max_guest_pages = 8;
1254 dev->guest_pages = rte_zmalloc(NULL,
1255 dev->max_guest_pages *
1256 sizeof(struct guest_page),
1257 RTE_CACHE_LINE_SIZE);
1258 if (dev->guest_pages == NULL) {
1259 VHOST_LOG_CONFIG(ERR,
1260 "(%d) failed to allocate memory "
1261 "for dev->guest_pages\n",
1267 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1268 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1269 if (dev->mem == NULL) {
1270 VHOST_LOG_CONFIG(ERR,
1271 "(%d) failed to allocate memory for dev->mem\n",
1273 goto free_guest_pages;
1276 for (i = 0; i < memory->nregions; i++) {
1277 reg = &dev->mem->regions[i];
1279 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1280 reg->guest_user_addr = memory->regions[i].userspace_addr;
1281 reg->size = memory->regions[i].memory_size;
1282 reg->fd = msg->fds[i];
1285 * Assign invalid file descriptor value to avoid double
1286 * closing on error path.
1290 mmap_offset = memory->regions[i].mmap_offset;
1292 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1293 VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1294 goto free_mem_table;
1297 dev->mem->nregions++;
1300 if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1301 goto free_mem_table;
1303 for (i = 0; i < dev->nr_vring; i++) {
1304 struct vhost_virtqueue *vq = dev->virtqueue[i];
1309 if (vq->desc || vq->avail || vq->used) {
1311 * If the memory table got updated, the ring addresses
1312 * need to be translated again as virtual addresses have
1315 vring_invalidate(dev, vq);
1317 dev = translate_ring_addresses(dev, i);
1320 goto free_mem_table;
1327 dump_guest_pages(dev);
1329 return RTE_VHOST_MSG_RESULT_OK;
1332 free_mem_region(dev);
1336 rte_free(dev->guest_pages);
1337 dev->guest_pages = NULL;
1340 return RTE_VHOST_MSG_RESULT_ERR;
1344 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1351 if (vq_is_packed(dev))
1352 rings_ok = vq->desc_packed && vq->driver_event &&
1355 rings_ok = vq->desc && vq->avail && vq->used;
1358 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1359 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1363 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1366 virtio_is_ready(struct virtio_net *dev)
1368 struct vhost_virtqueue *vq;
1369 uint32_t i, nr_vring = dev->nr_vring;
1371 if (dev->flags & VIRTIO_DEV_READY)
1377 if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1378 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1380 if (dev->nr_vring < nr_vring)
1384 for (i = 0; i < nr_vring; i++) {
1385 vq = dev->virtqueue[i];
1387 if (!vq_is_ready(dev, vq))
1391 /* If supported, ensure the frontend is really done with config */
1392 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1393 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1396 dev->flags |= VIRTIO_DEV_READY;
1398 if (!(dev->flags & VIRTIO_DEV_RUNNING))
1399 VHOST_LOG_CONFIG(INFO,
1400 "virtio is now ready for processing.\n");
1405 inflight_mem_alloc(const char *name, size_t size, int *fd)
1409 char fname[20] = "/tmp/memfd-XXXXXX";
1412 #ifdef MEMFD_SUPPORTED
1413 mfd = memfd_create(name, MFD_CLOEXEC);
1418 mfd = mkstemp(fname);
1420 VHOST_LOG_CONFIG(ERR,
1421 "failed to get inflight buffer fd\n");
1428 if (ftruncate(mfd, size) == -1) {
1429 VHOST_LOG_CONFIG(ERR,
1430 "failed to alloc inflight buffer\n");
1435 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1436 if (ptr == MAP_FAILED) {
1437 VHOST_LOG_CONFIG(ERR,
1438 "failed to mmap inflight buffer\n");
1448 get_pervq_shm_size_split(uint16_t queue_size)
1450 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1451 queue_size + sizeof(uint64_t) +
1452 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1456 get_pervq_shm_size_packed(uint16_t queue_size)
1458 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1459 * queue_size + sizeof(uint64_t) +
1460 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1461 INFLIGHT_ALIGNMENT);
1465 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1467 int main_fd __rte_unused)
1469 struct rte_vhost_inflight_info_packed *inflight_packed;
1470 uint64_t pervq_inflight_size, mmap_size;
1471 uint16_t num_queues, queue_size;
1472 struct virtio_net *dev = *pdev;
1476 if (msg->size != sizeof(msg->payload.inflight)) {
1477 VHOST_LOG_CONFIG(ERR,
1478 "invalid get_inflight_fd message size is %d\n",
1480 return RTE_VHOST_MSG_RESULT_ERR;
1483 if (dev->inflight_info == NULL) {
1484 dev->inflight_info = calloc(1,
1485 sizeof(struct inflight_mem_info));
1486 if (!dev->inflight_info) {
1487 VHOST_LOG_CONFIG(ERR,
1488 "failed to alloc dev inflight area\n");
1489 return RTE_VHOST_MSG_RESULT_ERR;
1491 dev->inflight_info->fd = -1;
1494 num_queues = msg->payload.inflight.num_queues;
1495 queue_size = msg->payload.inflight.queue_size;
1497 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1498 msg->payload.inflight.num_queues);
1499 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1500 msg->payload.inflight.queue_size);
1502 if (vq_is_packed(dev))
1503 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1505 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1507 mmap_size = num_queues * pervq_inflight_size;
1508 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1510 VHOST_LOG_CONFIG(ERR,
1511 "failed to alloc vhost inflight area\n");
1512 msg->payload.inflight.mmap_size = 0;
1513 return RTE_VHOST_MSG_RESULT_ERR;
1515 memset(addr, 0, mmap_size);
1517 if (dev->inflight_info->addr) {
1518 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1519 dev->inflight_info->addr = NULL;
1522 if (dev->inflight_info->fd >= 0) {
1523 close(dev->inflight_info->fd);
1524 dev->inflight_info->fd = -1;
1527 dev->inflight_info->addr = addr;
1528 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1529 dev->inflight_info->fd = msg->fds[0] = fd;
1530 msg->payload.inflight.mmap_offset = 0;
1533 if (vq_is_packed(dev)) {
1534 for (i = 0; i < num_queues; i++) {
1536 (struct rte_vhost_inflight_info_packed *)addr;
1537 inflight_packed->used_wrap_counter = 1;
1538 inflight_packed->old_used_wrap_counter = 1;
1539 for (j = 0; j < queue_size; j++)
1540 inflight_packed->desc[j].next = j + 1;
1541 addr = (void *)((char *)addr + pervq_inflight_size);
1545 VHOST_LOG_CONFIG(INFO,
1546 "send inflight mmap_size: %"PRIu64"\n",
1547 msg->payload.inflight.mmap_size);
1548 VHOST_LOG_CONFIG(INFO,
1549 "send inflight mmap_offset: %"PRIu64"\n",
1550 msg->payload.inflight.mmap_offset);
1551 VHOST_LOG_CONFIG(INFO,
1552 "send inflight fd: %d\n", msg->fds[0]);
1554 return RTE_VHOST_MSG_RESULT_REPLY;
1558 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1559 int main_fd __rte_unused)
1561 uint64_t mmap_size, mmap_offset;
1562 uint16_t num_queues, queue_size;
1563 struct virtio_net *dev = *pdev;
1564 uint32_t pervq_inflight_size;
1565 struct vhost_virtqueue *vq;
1570 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1571 VHOST_LOG_CONFIG(ERR,
1572 "invalid set_inflight_fd message size is %d,fd is %d\n",
1574 return RTE_VHOST_MSG_RESULT_ERR;
1577 mmap_size = msg->payload.inflight.mmap_size;
1578 mmap_offset = msg->payload.inflight.mmap_offset;
1579 num_queues = msg->payload.inflight.num_queues;
1580 queue_size = msg->payload.inflight.queue_size;
1582 if (vq_is_packed(dev))
1583 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1585 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1587 VHOST_LOG_CONFIG(INFO,
1588 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1589 VHOST_LOG_CONFIG(INFO,
1590 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1591 VHOST_LOG_CONFIG(INFO,
1592 "set_inflight_fd num_queues: %u\n", num_queues);
1593 VHOST_LOG_CONFIG(INFO,
1594 "set_inflight_fd queue_size: %u\n", queue_size);
1595 VHOST_LOG_CONFIG(INFO,
1596 "set_inflight_fd fd: %d\n", fd);
1597 VHOST_LOG_CONFIG(INFO,
1598 "set_inflight_fd pervq_inflight_size: %d\n",
1599 pervq_inflight_size);
1601 if (!dev->inflight_info) {
1602 dev->inflight_info = calloc(1,
1603 sizeof(struct inflight_mem_info));
1604 if (dev->inflight_info == NULL) {
1605 VHOST_LOG_CONFIG(ERR,
1606 "failed to alloc dev inflight area\n");
1607 return RTE_VHOST_MSG_RESULT_ERR;
1609 dev->inflight_info->fd = -1;
1612 if (dev->inflight_info->addr) {
1613 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1614 dev->inflight_info->addr = NULL;
1617 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1619 if (addr == MAP_FAILED) {
1620 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1621 return RTE_VHOST_MSG_RESULT_ERR;
1624 if (dev->inflight_info->fd >= 0) {
1625 close(dev->inflight_info->fd);
1626 dev->inflight_info->fd = -1;
1629 dev->inflight_info->fd = fd;
1630 dev->inflight_info->addr = addr;
1631 dev->inflight_info->size = mmap_size;
1633 for (i = 0; i < num_queues; i++) {
1634 vq = dev->virtqueue[i];
1638 if (vq_is_packed(dev)) {
1639 vq->inflight_packed = addr;
1640 vq->inflight_packed->desc_num = queue_size;
1642 vq->inflight_split = addr;
1643 vq->inflight_split->desc_num = queue_size;
1645 addr = (void *)((char *)addr + pervq_inflight_size);
1648 return RTE_VHOST_MSG_RESULT_OK;
1652 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1653 int main_fd __rte_unused)
1655 struct virtio_net *dev = *pdev;
1656 struct vhost_vring_file file;
1657 struct vhost_virtqueue *vq;
1660 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1661 if (validate_msg_fds(msg, expected_fds) != 0)
1662 return RTE_VHOST_MSG_RESULT_ERR;
1664 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1665 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1666 file.fd = VIRTIO_INVALID_EVENTFD;
1668 file.fd = msg->fds[0];
1669 VHOST_LOG_CONFIG(INFO,
1670 "vring call idx:%d file:%d\n", file.index, file.fd);
1672 vq = dev->virtqueue[file.index];
1676 vhost_user_notify_queue_state(dev, file.index, 0);
1679 if (vq->callfd >= 0)
1682 vq->callfd = file.fd;
1684 return RTE_VHOST_MSG_RESULT_OK;
1687 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1688 struct VhostUserMsg *msg,
1689 int main_fd __rte_unused)
1693 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1694 if (validate_msg_fds(msg, expected_fds) != 0)
1695 return RTE_VHOST_MSG_RESULT_ERR;
1697 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1699 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1701 return RTE_VHOST_MSG_RESULT_OK;
1705 resubmit_desc_compare(const void *a, const void *b)
1707 const struct rte_vhost_resubmit_desc *desc0 = a;
1708 const struct rte_vhost_resubmit_desc *desc1 = b;
1710 if (desc1->counter > desc0->counter)
1717 vhost_check_queue_inflights_split(struct virtio_net *dev,
1718 struct vhost_virtqueue *vq)
1721 uint16_t resubmit_num = 0, last_io, num;
1722 struct vring_used *used = vq->used;
1723 struct rte_vhost_resubmit_info *resubmit;
1724 struct rte_vhost_inflight_info_split *inflight_split;
1726 if (!(dev->protocol_features &
1727 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1728 return RTE_VHOST_MSG_RESULT_OK;
1730 /* The frontend may still not support the inflight feature
1731 * although we negotiate the protocol feature.
1733 if ((!vq->inflight_split))
1734 return RTE_VHOST_MSG_RESULT_OK;
1736 if (!vq->inflight_split->version) {
1737 vq->inflight_split->version = INFLIGHT_VERSION;
1738 return RTE_VHOST_MSG_RESULT_OK;
1741 if (vq->resubmit_inflight)
1742 return RTE_VHOST_MSG_RESULT_OK;
1744 inflight_split = vq->inflight_split;
1745 vq->global_counter = 0;
1746 last_io = inflight_split->last_inflight_io;
1748 if (inflight_split->used_idx != used->idx) {
1749 inflight_split->desc[last_io].inflight = 0;
1750 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1751 inflight_split->used_idx = used->idx;
1754 for (i = 0; i < inflight_split->desc_num; i++) {
1755 if (inflight_split->desc[i].inflight == 1)
1759 vq->last_avail_idx += resubmit_num;
1762 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1764 VHOST_LOG_CONFIG(ERR,
1765 "failed to allocate memory for resubmit info.\n");
1766 return RTE_VHOST_MSG_RESULT_ERR;
1769 resubmit->resubmit_list = calloc(resubmit_num,
1770 sizeof(struct rte_vhost_resubmit_desc));
1771 if (!resubmit->resubmit_list) {
1772 VHOST_LOG_CONFIG(ERR,
1773 "failed to allocate memory for inflight desc.\n");
1775 return RTE_VHOST_MSG_RESULT_ERR;
1779 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1780 if (vq->inflight_split->desc[i].inflight == 1) {
1781 resubmit->resubmit_list[num].index = i;
1782 resubmit->resubmit_list[num].counter =
1783 inflight_split->desc[i].counter;
1787 resubmit->resubmit_num = num;
1789 if (resubmit->resubmit_num > 1)
1790 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1791 sizeof(struct rte_vhost_resubmit_desc),
1792 resubmit_desc_compare);
1794 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1795 vq->resubmit_inflight = resubmit;
1798 return RTE_VHOST_MSG_RESULT_OK;
1802 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1803 struct vhost_virtqueue *vq)
1806 uint16_t resubmit_num = 0, old_used_idx, num;
1807 struct rte_vhost_resubmit_info *resubmit;
1808 struct rte_vhost_inflight_info_packed *inflight_packed;
1810 if (!(dev->protocol_features &
1811 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1812 return RTE_VHOST_MSG_RESULT_OK;
1814 /* The frontend may still not support the inflight feature
1815 * although we negotiate the protocol feature.
1817 if ((!vq->inflight_packed))
1818 return RTE_VHOST_MSG_RESULT_OK;
1820 if (!vq->inflight_packed->version) {
1821 vq->inflight_packed->version = INFLIGHT_VERSION;
1822 return RTE_VHOST_MSG_RESULT_OK;
1825 if (vq->resubmit_inflight)
1826 return RTE_VHOST_MSG_RESULT_OK;
1828 inflight_packed = vq->inflight_packed;
1829 vq->global_counter = 0;
1830 old_used_idx = inflight_packed->old_used_idx;
1832 if (inflight_packed->used_idx != old_used_idx) {
1833 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1834 inflight_packed->old_used_idx =
1835 inflight_packed->used_idx;
1836 inflight_packed->old_used_wrap_counter =
1837 inflight_packed->used_wrap_counter;
1838 inflight_packed->old_free_head =
1839 inflight_packed->free_head;
1841 inflight_packed->used_idx =
1842 inflight_packed->old_used_idx;
1843 inflight_packed->used_wrap_counter =
1844 inflight_packed->old_used_wrap_counter;
1845 inflight_packed->free_head =
1846 inflight_packed->old_free_head;
1850 for (i = 0; i < inflight_packed->desc_num; i++) {
1851 if (inflight_packed->desc[i].inflight == 1)
1856 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1857 if (resubmit == NULL) {
1858 VHOST_LOG_CONFIG(ERR,
1859 "failed to allocate memory for resubmit info.\n");
1860 return RTE_VHOST_MSG_RESULT_ERR;
1863 resubmit->resubmit_list = calloc(resubmit_num,
1864 sizeof(struct rte_vhost_resubmit_desc));
1865 if (resubmit->resubmit_list == NULL) {
1866 VHOST_LOG_CONFIG(ERR,
1867 "failed to allocate memory for resubmit desc.\n");
1869 return RTE_VHOST_MSG_RESULT_ERR;
1873 for (i = 0; i < inflight_packed->desc_num; i++) {
1874 if (vq->inflight_packed->desc[i].inflight == 1) {
1875 resubmit->resubmit_list[num].index = i;
1876 resubmit->resubmit_list[num].counter =
1877 inflight_packed->desc[i].counter;
1881 resubmit->resubmit_num = num;
1883 if (resubmit->resubmit_num > 1)
1884 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1885 sizeof(struct rte_vhost_resubmit_desc),
1886 resubmit_desc_compare);
1888 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1889 vq->resubmit_inflight = resubmit;
1892 return RTE_VHOST_MSG_RESULT_OK;
1896 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1897 int main_fd __rte_unused)
1899 struct virtio_net *dev = *pdev;
1900 struct vhost_vring_file file;
1901 struct vhost_virtqueue *vq;
1904 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1905 if (validate_msg_fds(msg, expected_fds) != 0)
1906 return RTE_VHOST_MSG_RESULT_ERR;
1908 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1909 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1910 file.fd = VIRTIO_INVALID_EVENTFD;
1912 file.fd = msg->fds[0];
1913 VHOST_LOG_CONFIG(INFO,
1914 "vring kick idx:%d file:%d\n", file.index, file.fd);
1916 /* Interpret ring addresses only when ring is started. */
1917 dev = translate_ring_addresses(dev, file.index);
1919 if (file.fd != VIRTIO_INVALID_EVENTFD)
1922 return RTE_VHOST_MSG_RESULT_ERR;
1927 vq = dev->virtqueue[file.index];
1930 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1931 * the ring starts already enabled. Otherwise, it is enabled via
1932 * the SET_VRING_ENABLE message.
1934 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1940 vhost_user_notify_queue_state(dev, file.index, 0);
1943 if (vq->kickfd >= 0)
1945 vq->kickfd = file.fd;
1947 if (vq_is_packed(dev)) {
1948 if (vhost_check_queue_inflights_packed(dev, vq)) {
1949 VHOST_LOG_CONFIG(ERR,
1950 "failed to inflights for vq: %d\n", file.index);
1951 return RTE_VHOST_MSG_RESULT_ERR;
1954 if (vhost_check_queue_inflights_split(dev, vq)) {
1955 VHOST_LOG_CONFIG(ERR,
1956 "failed to inflights for vq: %d\n", file.index);
1957 return RTE_VHOST_MSG_RESULT_ERR;
1961 return RTE_VHOST_MSG_RESULT_OK;
1965 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1968 vhost_user_get_vring_base(struct virtio_net **pdev,
1969 struct VhostUserMsg *msg,
1970 int main_fd __rte_unused)
1972 struct virtio_net *dev = *pdev;
1973 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1976 if (validate_msg_fds(msg, 0) != 0)
1977 return RTE_VHOST_MSG_RESULT_ERR;
1979 /* We have to stop the queue (virtio) if it is running. */
1980 vhost_destroy_device_notify(dev);
1982 dev->flags &= ~VIRTIO_DEV_READY;
1983 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1985 /* Here we are safe to get the indexes */
1986 if (vq_is_packed(dev)) {
1988 * Bit[0:14]: avail index
1989 * Bit[15]: avail wrap counter
1991 val = vq->last_avail_idx & 0x7fff;
1992 val |= vq->avail_wrap_counter << 15;
1993 msg->payload.state.num = val;
1995 msg->payload.state.num = vq->last_avail_idx;
1998 VHOST_LOG_CONFIG(INFO,
1999 "vring base idx:%d file:%d\n", msg->payload.state.index,
2000 msg->payload.state.num);
2002 * Based on current qemu vhost-user implementation, this message is
2003 * sent and only sent in vhost_vring_stop.
2004 * TODO: cleanup the vring, it isn't usable since here.
2006 if (vq->kickfd >= 0)
2009 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2011 if (vq->callfd >= 0)
2014 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2016 vq->signalled_used_valid = false;
2018 if (vq_is_packed(dev)) {
2019 rte_free(vq->shadow_used_packed);
2020 vq->shadow_used_packed = NULL;
2022 rte_free(vq->shadow_used_split);
2023 vq->shadow_used_split = NULL;
2026 rte_free(vq->batch_copy_elems);
2027 vq->batch_copy_elems = NULL;
2029 rte_free(vq->log_cache);
2030 vq->log_cache = NULL;
2032 msg->size = sizeof(msg->payload.state);
2035 vring_invalidate(dev, vq);
2037 return RTE_VHOST_MSG_RESULT_REPLY;
2041 * when virtio queues are ready to work, qemu will send us to
2042 * enable the virtio queue pair.
2045 vhost_user_set_vring_enable(struct virtio_net **pdev,
2046 struct VhostUserMsg *msg,
2047 int main_fd __rte_unused)
2049 struct virtio_net *dev = *pdev;
2050 bool enable = !!msg->payload.state.num;
2051 int index = (int)msg->payload.state.index;
2053 if (validate_msg_fds(msg, 0) != 0)
2054 return RTE_VHOST_MSG_RESULT_ERR;
2056 VHOST_LOG_CONFIG(INFO,
2057 "set queue enable: %d to qp idx: %d\n",
2060 if (enable && dev->virtqueue[index]->async_registered) {
2061 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2062 VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2063 "async inflight packets must be completed first\n");
2064 return RTE_VHOST_MSG_RESULT_ERR;
2068 dev->virtqueue[index]->enabled = enable;
2070 return RTE_VHOST_MSG_RESULT_OK;
2074 vhost_user_get_protocol_features(struct virtio_net **pdev,
2075 struct VhostUserMsg *msg,
2076 int main_fd __rte_unused)
2078 struct virtio_net *dev = *pdev;
2079 uint64_t features, protocol_features;
2081 if (validate_msg_fds(msg, 0) != 0)
2082 return RTE_VHOST_MSG_RESULT_ERR;
2084 rte_vhost_driver_get_features(dev->ifname, &features);
2085 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2087 msg->payload.u64 = protocol_features;
2088 msg->size = sizeof(msg->payload.u64);
2091 return RTE_VHOST_MSG_RESULT_REPLY;
2095 vhost_user_set_protocol_features(struct virtio_net **pdev,
2096 struct VhostUserMsg *msg,
2097 int main_fd __rte_unused)
2099 struct virtio_net *dev = *pdev;
2100 uint64_t protocol_features = msg->payload.u64;
2101 uint64_t slave_protocol_features = 0;
2103 if (validate_msg_fds(msg, 0) != 0)
2104 return RTE_VHOST_MSG_RESULT_ERR;
2106 rte_vhost_driver_get_protocol_features(dev->ifname,
2107 &slave_protocol_features);
2108 if (protocol_features & ~slave_protocol_features) {
2109 VHOST_LOG_CONFIG(ERR,
2110 "(%d) received invalid protocol features.\n",
2112 return RTE_VHOST_MSG_RESULT_ERR;
2115 dev->protocol_features = protocol_features;
2116 VHOST_LOG_CONFIG(INFO,
2117 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2118 dev->protocol_features);
2120 return RTE_VHOST_MSG_RESULT_OK;
2124 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2125 int main_fd __rte_unused)
2127 struct virtio_net *dev = *pdev;
2128 int fd = msg->fds[0];
2133 if (validate_msg_fds(msg, 1) != 0)
2134 return RTE_VHOST_MSG_RESULT_ERR;
2137 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2138 return RTE_VHOST_MSG_RESULT_ERR;
2141 if (msg->size != sizeof(VhostUserLog)) {
2142 VHOST_LOG_CONFIG(ERR,
2143 "invalid log base msg size: %"PRId32" != %d\n",
2144 msg->size, (int)sizeof(VhostUserLog));
2148 size = msg->payload.log.mmap_size;
2149 off = msg->payload.log.mmap_offset;
2151 /* Check for mmap size and offset overflow. */
2153 VHOST_LOG_CONFIG(ERR,
2154 "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2159 VHOST_LOG_CONFIG(INFO,
2160 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2164 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2165 * fail when offset is not page size aligned.
2167 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2169 if (addr == MAP_FAILED) {
2170 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2171 return RTE_VHOST_MSG_RESULT_ERR;
2175 * Free previously mapped log memory on occasionally
2176 * multiple VHOST_USER_SET_LOG_BASE.
2178 if (dev->log_addr) {
2179 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2181 dev->log_addr = (uint64_t)(uintptr_t)addr;
2182 dev->log_base = dev->log_addr + off;
2183 dev->log_size = size;
2185 for (i = 0; i < dev->nr_vring; i++) {
2186 struct vhost_virtqueue *vq = dev->virtqueue[i];
2188 rte_free(vq->log_cache);
2189 vq->log_cache = NULL;
2190 vq->log_cache_nb_elem = 0;
2191 vq->log_cache = rte_zmalloc("vq log cache",
2192 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2195 * If log cache alloc fail, don't fail migration, but no
2196 * caching will be done, which will impact performance
2199 VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2203 * The spec is not clear about it (yet), but QEMU doesn't expect
2204 * any payload in the reply.
2209 return RTE_VHOST_MSG_RESULT_REPLY;
2213 return RTE_VHOST_MSG_RESULT_ERR;
2216 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2217 struct VhostUserMsg *msg,
2218 int main_fd __rte_unused)
2220 if (validate_msg_fds(msg, 1) != 0)
2221 return RTE_VHOST_MSG_RESULT_ERR;
2224 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2226 return RTE_VHOST_MSG_RESULT_OK;
2230 * An rarp packet is constructed and broadcasted to notify switches about
2231 * the new location of the migrated VM, so that packets from outside will
2232 * not be lost after migration.
2234 * However, we don't actually "send" a rarp packet here, instead, we set
2235 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2238 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2239 int main_fd __rte_unused)
2241 struct virtio_net *dev = *pdev;
2242 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2243 struct rte_vdpa_device *vdpa_dev;
2245 if (validate_msg_fds(msg, 0) != 0)
2246 return RTE_VHOST_MSG_RESULT_ERR;
2248 VHOST_LOG_CONFIG(DEBUG,
2249 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2250 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2251 memcpy(dev->mac.addr_bytes, mac, 6);
2254 * Set the flag to inject a RARP broadcast packet at
2255 * rte_vhost_dequeue_burst().
2257 * __ATOMIC_RELEASE ordering is for making sure the mac is
2258 * copied before the flag is set.
2260 __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2261 vdpa_dev = dev->vdpa_dev;
2262 if (vdpa_dev && vdpa_dev->ops->migration_done)
2263 vdpa_dev->ops->migration_done(dev->vid);
2265 return RTE_VHOST_MSG_RESULT_OK;
2269 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2270 int main_fd __rte_unused)
2272 struct virtio_net *dev = *pdev;
2274 if (validate_msg_fds(msg, 0) != 0)
2275 return RTE_VHOST_MSG_RESULT_ERR;
2277 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2278 msg->payload.u64 > VIRTIO_MAX_MTU) {
2279 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2282 return RTE_VHOST_MSG_RESULT_ERR;
2285 dev->mtu = msg->payload.u64;
2287 return RTE_VHOST_MSG_RESULT_OK;
2291 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2292 int main_fd __rte_unused)
2294 struct virtio_net *dev = *pdev;
2295 int fd = msg->fds[0];
2297 if (validate_msg_fds(msg, 1) != 0)
2298 return RTE_VHOST_MSG_RESULT_ERR;
2301 VHOST_LOG_CONFIG(ERR,
2302 "Invalid file descriptor for slave channel (%d)\n",
2304 return RTE_VHOST_MSG_RESULT_ERR;
2307 if (dev->slave_req_fd >= 0)
2308 close(dev->slave_req_fd);
2310 dev->slave_req_fd = fd;
2312 return RTE_VHOST_MSG_RESULT_OK;
2316 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2318 struct vhost_vring_addr *ra;
2319 uint64_t start, end, len;
2322 end = start + imsg->size;
2324 ra = &vq->ring_addrs;
2325 len = sizeof(struct vring_desc) * vq->size;
2326 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2329 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2330 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2333 len = sizeof(struct vring_used) +
2334 sizeof(struct vring_used_elem) * vq->size;
2335 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2338 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2339 len = sizeof(uint64_t);
2340 if (ra->log_guest_addr < end &&
2341 (ra->log_guest_addr + len) > start)
2349 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2351 struct vhost_vring_addr *ra;
2352 uint64_t start, end, len;
2355 end = start + imsg->size;
2357 ra = &vq->ring_addrs;
2358 len = sizeof(struct vring_packed_desc) * vq->size;
2359 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2362 len = sizeof(struct vring_packed_desc_event);
2363 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2366 len = sizeof(struct vring_packed_desc_event);
2367 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2370 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2371 len = sizeof(uint64_t);
2372 if (ra->log_guest_addr < end &&
2373 (ra->log_guest_addr + len) > start)
2380 static int is_vring_iotlb(struct virtio_net *dev,
2381 struct vhost_virtqueue *vq,
2382 struct vhost_iotlb_msg *imsg)
2384 if (vq_is_packed(dev))
2385 return is_vring_iotlb_packed(vq, imsg);
2387 return is_vring_iotlb_split(vq, imsg);
2391 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2392 int main_fd __rte_unused)
2394 struct virtio_net *dev = *pdev;
2395 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2399 if (validate_msg_fds(msg, 0) != 0)
2400 return RTE_VHOST_MSG_RESULT_ERR;
2402 switch (imsg->type) {
2403 case VHOST_IOTLB_UPDATE:
2405 vva = qva_to_vva(dev, imsg->uaddr, &len);
2407 return RTE_VHOST_MSG_RESULT_ERR;
2409 for (i = 0; i < dev->nr_vring; i++) {
2410 struct vhost_virtqueue *vq = dev->virtqueue[i];
2415 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2418 if (is_vring_iotlb(dev, vq, imsg))
2419 *pdev = dev = translate_ring_addresses(dev, i);
2422 case VHOST_IOTLB_INVALIDATE:
2423 for (i = 0; i < dev->nr_vring; i++) {
2424 struct vhost_virtqueue *vq = dev->virtqueue[i];
2429 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2432 if (is_vring_iotlb(dev, vq, imsg))
2433 vring_invalidate(dev, vq);
2437 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2439 return RTE_VHOST_MSG_RESULT_ERR;
2442 return RTE_VHOST_MSG_RESULT_OK;
2446 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2447 struct VhostUserMsg *msg,
2448 int main_fd __rte_unused)
2450 struct virtio_net *dev = *pdev;
2451 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2452 struct uffdio_api api_struct;
2454 if (validate_msg_fds(msg, 0) != 0)
2455 return RTE_VHOST_MSG_RESULT_ERR;
2457 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2459 if (dev->postcopy_ufd == -1) {
2460 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2462 return RTE_VHOST_MSG_RESULT_ERR;
2464 api_struct.api = UFFD_API;
2465 api_struct.features = 0;
2466 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2467 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2469 close(dev->postcopy_ufd);
2470 dev->postcopy_ufd = -1;
2471 return RTE_VHOST_MSG_RESULT_ERR;
2473 msg->fds[0] = dev->postcopy_ufd;
2476 return RTE_VHOST_MSG_RESULT_REPLY;
2478 dev->postcopy_ufd = -1;
2481 return RTE_VHOST_MSG_RESULT_ERR;
2486 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2487 struct VhostUserMsg *msg __rte_unused,
2488 int main_fd __rte_unused)
2490 struct virtio_net *dev = *pdev;
2492 if (validate_msg_fds(msg, 0) != 0)
2493 return RTE_VHOST_MSG_RESULT_ERR;
2495 if (dev->mem && dev->mem->nregions) {
2496 VHOST_LOG_CONFIG(ERR,
2497 "Regions already registered at postcopy-listen\n");
2498 return RTE_VHOST_MSG_RESULT_ERR;
2500 dev->postcopy_listening = 1;
2502 return RTE_VHOST_MSG_RESULT_OK;
2506 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2507 int main_fd __rte_unused)
2509 struct virtio_net *dev = *pdev;
2511 if (validate_msg_fds(msg, 0) != 0)
2512 return RTE_VHOST_MSG_RESULT_ERR;
2514 dev->postcopy_listening = 0;
2515 if (dev->postcopy_ufd >= 0) {
2516 close(dev->postcopy_ufd);
2517 dev->postcopy_ufd = -1;
2520 msg->payload.u64 = 0;
2521 msg->size = sizeof(msg->payload.u64);
2524 return RTE_VHOST_MSG_RESULT_REPLY;
2528 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2529 int main_fd __rte_unused)
2531 struct virtio_net *dev = *pdev;
2533 if (validate_msg_fds(msg, 0) != 0)
2534 return RTE_VHOST_MSG_RESULT_ERR;
2536 msg->payload.u64 = dev->status;
2537 msg->size = sizeof(msg->payload.u64);
2540 return RTE_VHOST_MSG_RESULT_REPLY;
2544 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2545 int main_fd __rte_unused)
2547 struct virtio_net *dev = *pdev;
2549 if (validate_msg_fds(msg, 0) != 0)
2550 return RTE_VHOST_MSG_RESULT_ERR;
2552 /* As per Virtio specification, the device status is 8bits long */
2553 if (msg->payload.u64 > UINT8_MAX) {
2554 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2556 return RTE_VHOST_MSG_RESULT_ERR;
2559 dev->status = msg->payload.u64;
2561 if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2562 (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2563 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2565 * Clear the bit to let the driver know about the feature
2566 * negotiation failure
2568 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2571 VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2573 "\t-ACKNOWLEDGE: %u\n"
2575 "\t-FEATURES_OK: %u\n"
2576 "\t-DRIVER_OK: %u\n"
2577 "\t-DEVICE_NEED_RESET: %u\n"
2580 (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2581 !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2582 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2583 !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2584 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2585 !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2586 !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2588 return RTE_VHOST_MSG_RESULT_OK;
2591 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2592 struct VhostUserMsg *msg,
2594 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2595 [VHOST_USER_NONE] = NULL,
2596 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2597 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2598 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2599 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2600 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2601 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2602 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2603 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2604 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2605 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2606 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2607 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2608 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2609 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2610 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2611 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2612 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2613 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2614 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2615 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2616 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2617 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2618 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2619 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2620 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2621 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2622 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2623 [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2624 [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2627 /* return bytes# of read on success or negative val on failure. */
2629 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2633 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2634 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2637 } else if (ret != VHOST_USER_HDR_SIZE) {
2638 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2644 if (msg->size > sizeof(msg->payload)) {
2645 VHOST_LOG_CONFIG(ERR,
2646 "invalid msg size: %d\n", msg->size);
2649 ret = read(sockfd, &msg->payload, msg->size);
2652 if (ret != (int)msg->size) {
2653 VHOST_LOG_CONFIG(ERR,
2654 "read control message failed\n");
2663 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2668 return send_fd_message(sockfd, (char *)msg,
2669 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2673 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2678 msg->flags &= ~VHOST_USER_VERSION_MASK;
2679 msg->flags &= ~VHOST_USER_NEED_REPLY;
2680 msg->flags |= VHOST_USER_VERSION;
2681 msg->flags |= VHOST_USER_REPLY_MASK;
2683 return send_vhost_message(sockfd, msg);
2687 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2691 if (msg->flags & VHOST_USER_NEED_REPLY)
2692 rte_spinlock_lock(&dev->slave_req_lock);
2694 ret = send_vhost_message(dev->slave_req_fd, msg);
2695 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2696 rte_spinlock_unlock(&dev->slave_req_lock);
2702 * Allocate a queue pair if it hasn't been allocated yet
2705 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2706 struct VhostUserMsg *msg)
2710 switch (msg->request.master) {
2711 case VHOST_USER_SET_VRING_KICK:
2712 case VHOST_USER_SET_VRING_CALL:
2713 case VHOST_USER_SET_VRING_ERR:
2714 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2716 case VHOST_USER_SET_VRING_NUM:
2717 case VHOST_USER_SET_VRING_BASE:
2718 case VHOST_USER_SET_VRING_ENABLE:
2719 vring_idx = msg->payload.state.index;
2721 case VHOST_USER_SET_VRING_ADDR:
2722 vring_idx = msg->payload.addr.index;
2728 if (vring_idx >= VHOST_MAX_VRING) {
2729 VHOST_LOG_CONFIG(ERR,
2730 "invalid vring index: %u\n", vring_idx);
2734 if (dev->virtqueue[vring_idx])
2737 return alloc_vring_queue(dev, vring_idx);
2741 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2744 unsigned int vq_num = 0;
2746 while (vq_num < dev->nr_vring) {
2747 struct vhost_virtqueue *vq = dev->virtqueue[i];
2750 rte_spinlock_lock(&vq->access_lock);
2758 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2761 unsigned int vq_num = 0;
2763 while (vq_num < dev->nr_vring) {
2764 struct vhost_virtqueue *vq = dev->virtqueue[i];
2767 rte_spinlock_unlock(&vq->access_lock);
2775 vhost_user_msg_handler(int vid, int fd)
2777 struct virtio_net *dev;
2778 struct VhostUserMsg msg;
2779 struct rte_vdpa_device *vdpa_dev;
2781 int unlock_required = 0;
2786 dev = get_device(vid);
2790 if (!dev->notify_ops) {
2791 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2792 if (!dev->notify_ops) {
2793 VHOST_LOG_CONFIG(ERR,
2794 "failed to get callback ops for driver %s\n",
2800 ret = read_vhost_message(fd, &msg);
2803 VHOST_LOG_CONFIG(ERR,
2804 "vhost read message failed\n");
2806 VHOST_LOG_CONFIG(INFO,
2807 "vhost peer closed\n");
2813 request = msg.request.master;
2814 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2815 vhost_message_str[request]) {
2816 if (request != VHOST_USER_IOTLB_MSG)
2817 VHOST_LOG_CONFIG(INFO, "read message %s\n",
2818 vhost_message_str[request]);
2820 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2821 vhost_message_str[request]);
2823 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2826 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2828 VHOST_LOG_CONFIG(ERR,
2829 "failed to alloc queue\n");
2834 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2835 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2836 * and device is destroyed. destroy_device waits for queues to be
2837 * inactive, so it is safe. Otherwise taking the access_lock
2838 * would cause a dead lock.
2841 case VHOST_USER_SET_FEATURES:
2842 case VHOST_USER_SET_PROTOCOL_FEATURES:
2843 case VHOST_USER_SET_OWNER:
2844 case VHOST_USER_SET_MEM_TABLE:
2845 case VHOST_USER_SET_LOG_BASE:
2846 case VHOST_USER_SET_LOG_FD:
2847 case VHOST_USER_SET_VRING_NUM:
2848 case VHOST_USER_SET_VRING_ADDR:
2849 case VHOST_USER_SET_VRING_BASE:
2850 case VHOST_USER_SET_VRING_KICK:
2851 case VHOST_USER_SET_VRING_CALL:
2852 case VHOST_USER_SET_VRING_ERR:
2853 case VHOST_USER_SET_VRING_ENABLE:
2854 case VHOST_USER_SEND_RARP:
2855 case VHOST_USER_NET_SET_MTU:
2856 case VHOST_USER_SET_SLAVE_REQ_FD:
2857 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2858 vhost_user_lock_all_queue_pairs(dev);
2859 unlock_required = 1;
2868 if (dev->extern_ops.pre_msg_handle) {
2869 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2872 case RTE_VHOST_MSG_RESULT_REPLY:
2873 send_vhost_reply(fd, &msg);
2875 case RTE_VHOST_MSG_RESULT_ERR:
2876 case RTE_VHOST_MSG_RESULT_OK:
2878 goto skip_to_post_handle;
2879 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2885 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2886 if (!vhost_message_handlers[request])
2887 goto skip_to_post_handle;
2888 ret = vhost_message_handlers[request](&dev, &msg, fd);
2891 case RTE_VHOST_MSG_RESULT_ERR:
2892 VHOST_LOG_CONFIG(ERR,
2893 "Processing %s failed.\n",
2894 vhost_message_str[request]);
2897 case RTE_VHOST_MSG_RESULT_OK:
2898 VHOST_LOG_CONFIG(DEBUG,
2899 "Processing %s succeeded.\n",
2900 vhost_message_str[request]);
2903 case RTE_VHOST_MSG_RESULT_REPLY:
2904 VHOST_LOG_CONFIG(DEBUG,
2905 "Processing %s succeeded and needs reply.\n",
2906 vhost_message_str[request]);
2907 send_vhost_reply(fd, &msg);
2915 skip_to_post_handle:
2916 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2917 dev->extern_ops.post_msg_handle) {
2918 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2921 case RTE_VHOST_MSG_RESULT_REPLY:
2922 send_vhost_reply(fd, &msg);
2924 case RTE_VHOST_MSG_RESULT_ERR:
2925 case RTE_VHOST_MSG_RESULT_OK:
2927 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2933 if (unlock_required)
2934 vhost_user_unlock_all_queue_pairs(dev);
2936 /* If message was not handled at this stage, treat it as an error */
2938 VHOST_LOG_CONFIG(ERR,
2939 "vhost message (req: %d) was not handled.\n", request);
2940 close_msg_fds(&msg);
2941 ret = RTE_VHOST_MSG_RESULT_ERR;
2945 * If the request required a reply that was already sent,
2946 * this optional reply-ack won't be sent as the
2947 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2949 if (msg.flags & VHOST_USER_NEED_REPLY) {
2950 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2951 msg.size = sizeof(msg.payload.u64);
2953 send_vhost_reply(fd, &msg);
2954 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2955 VHOST_LOG_CONFIG(ERR,
2956 "vhost message handling failed.\n");
2960 for (i = 0; i < dev->nr_vring; i++) {
2961 struct vhost_virtqueue *vq = dev->virtqueue[i];
2962 bool cur_ready = vq_is_ready(dev, vq);
2964 if (cur_ready != (vq && vq->ready)) {
2965 vq->ready = cur_ready;
2966 vhost_user_notify_queue_state(dev, i, cur_ready);
2971 if (!virtio_is_ready(dev))
2975 * Virtio is now ready. If not done already, it is time
2976 * to notify the application it can process the rings and
2977 * configure the vDPA device if present.
2980 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2981 if (dev->notify_ops->new_device(dev->vid) == 0)
2982 dev->flags |= VIRTIO_DEV_RUNNING;
2985 vdpa_dev = dev->vdpa_dev;
2989 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2990 if (vdpa_dev->ops->dev_conf(dev->vid))
2991 VHOST_LOG_CONFIG(ERR,
2992 "Failed to configure vDPA device\n");
2994 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3001 static int process_slave_message_reply(struct virtio_net *dev,
3002 const struct VhostUserMsg *msg)
3004 struct VhostUserMsg msg_reply;
3007 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3010 ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3013 VHOST_LOG_CONFIG(ERR,
3014 "vhost read slave message reply failed\n");
3016 VHOST_LOG_CONFIG(INFO,
3017 "vhost peer closed\n");
3023 if (msg_reply.request.slave != msg->request.slave) {
3024 VHOST_LOG_CONFIG(ERR,
3025 "Received unexpected msg type (%u), expected %u\n",
3026 msg_reply.request.slave, msg->request.slave);
3031 ret = msg_reply.payload.u64 ? -1 : 0;
3034 rte_spinlock_unlock(&dev->slave_req_lock);
3039 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3042 struct VhostUserMsg msg = {
3043 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3044 .flags = VHOST_USER_VERSION,
3045 .size = sizeof(msg.payload.iotlb),
3049 .type = VHOST_IOTLB_MISS,
3053 ret = send_vhost_message(dev->slave_req_fd, &msg);
3055 VHOST_LOG_CONFIG(ERR,
3056 "Failed to send IOTLB miss message (%d)\n",
3065 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3068 struct VhostUserMsg msg = {
3069 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3070 .flags = VHOST_USER_VERSION,
3075 msg.flags |= VHOST_USER_NEED_REPLY;
3077 ret = send_vhost_slave_message(dev, &msg);
3079 VHOST_LOG_CONFIG(ERR,
3080 "Failed to send config change (%d)\n",
3085 return process_slave_message_reply(dev, &msg);
3089 rte_vhost_slave_config_change(int vid, bool need_reply)
3091 struct virtio_net *dev;
3093 dev = get_device(vid);
3097 return vhost_user_slave_config_change(dev, need_reply);
3100 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3106 struct VhostUserMsg msg = {
3107 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3108 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3109 .size = sizeof(msg.payload.area),
3111 .u64 = index & VHOST_USER_VRING_IDX_MASK,
3118 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3124 ret = send_vhost_slave_message(dev, &msg);
3126 VHOST_LOG_CONFIG(ERR,
3127 "Failed to set host notifier (%d)\n", ret);
3131 return process_slave_message_reply(dev, &msg);
3134 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3136 struct virtio_net *dev;
3137 struct rte_vdpa_device *vdpa_dev;
3138 int vfio_device_fd, ret = 0;
3139 uint64_t offset, size;
3140 unsigned int i, q_start, q_last;
3142 dev = get_device(vid);
3146 vdpa_dev = dev->vdpa_dev;
3147 if (vdpa_dev == NULL)
3150 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3151 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3152 !(dev->protocol_features &
3153 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3154 !(dev->protocol_features &
3155 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3156 !(dev->protocol_features &
3157 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3160 if (qid == RTE_VHOST_QUEUE_ALL) {
3162 q_last = dev->nr_vring - 1;
3164 if (qid >= dev->nr_vring)
3170 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3171 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3173 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3174 if (vfio_device_fd < 0)
3178 for (i = q_start; i <= q_last; i++) {
3179 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3185 if (vhost_user_slave_set_vring_host_notifier(dev, i,
3186 vfio_device_fd, offset, size) < 0) {
3193 for (i = q_start; i <= q_last; i++) {
3194 vhost_user_slave_set_vring_host_notifier(dev, i, -1,