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 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
473 if (vhost_user_iotlb_init(dev, msg->payload.state.index))
474 return RTE_VHOST_MSG_RESULT_ERR;
476 return RTE_VHOST_MSG_RESULT_OK;
480 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
481 * same numa node as the memory of vring descriptor.
483 #ifdef RTE_LIBRTE_VHOST_NUMA
484 static struct virtio_net*
485 numa_realloc(struct virtio_net *dev, int index)
487 int oldnode, newnode;
488 struct virtio_net *old_dev;
489 struct vhost_virtqueue *old_vq, *vq;
490 struct vring_used_elem *new_shadow_used_split;
491 struct vring_used_elem_packed *new_shadow_used_packed;
492 struct batch_copy_elem *new_batch_copy_elems;
495 if (dev->flags & VIRTIO_DEV_RUNNING)
499 vq = old_vq = dev->virtqueue[index];
501 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
502 MPOL_F_NODE | MPOL_F_ADDR);
504 /* check if we need to reallocate vq */
505 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
506 MPOL_F_NODE | MPOL_F_ADDR);
508 VHOST_LOG_CONFIG(ERR,
509 "Unable to get vq numa information.\n");
512 if (oldnode != newnode) {
513 VHOST_LOG_CONFIG(INFO,
514 "reallocate vq from %d to %d node\n", oldnode, newnode);
515 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
519 memcpy(vq, old_vq, sizeof(*vq));
521 if (vq_is_packed(dev)) {
522 new_shadow_used_packed = rte_malloc_socket(NULL,
524 sizeof(struct vring_used_elem_packed),
527 if (new_shadow_used_packed) {
528 rte_free(vq->shadow_used_packed);
529 vq->shadow_used_packed = new_shadow_used_packed;
532 new_shadow_used_split = rte_malloc_socket(NULL,
534 sizeof(struct vring_used_elem),
537 if (new_shadow_used_split) {
538 rte_free(vq->shadow_used_split);
539 vq->shadow_used_split = new_shadow_used_split;
543 new_batch_copy_elems = rte_malloc_socket(NULL,
544 vq->size * sizeof(struct batch_copy_elem),
547 if (new_batch_copy_elems) {
548 rte_free(vq->batch_copy_elems);
549 vq->batch_copy_elems = new_batch_copy_elems;
555 /* check if we need to reallocate dev */
556 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
557 MPOL_F_NODE | MPOL_F_ADDR);
559 VHOST_LOG_CONFIG(ERR,
560 "Unable to get dev numa information.\n");
563 if (oldnode != newnode) {
564 VHOST_LOG_CONFIG(INFO,
565 "reallocate dev from %d to %d node\n",
567 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
573 memcpy(dev, old_dev, sizeof(*dev));
578 dev->virtqueue[index] = vq;
579 vhost_devices[dev->vid] = dev;
581 if (old_vq != vq && (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
582 vhost_user_iotlb_init(dev, index);
587 static struct virtio_net*
588 numa_realloc(struct virtio_net *dev, int index __rte_unused)
594 /* Converts QEMU virtual address to Vhost virtual address. */
596 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
598 struct rte_vhost_mem_region *r;
601 if (unlikely(!dev || !dev->mem))
604 /* Find the region where the address lives. */
605 for (i = 0; i < dev->mem->nregions; i++) {
606 r = &dev->mem->regions[i];
608 if (qva >= r->guest_user_addr &&
609 qva < r->guest_user_addr + r->size) {
611 if (unlikely(*len > r->guest_user_addr + r->size - qva))
612 *len = r->guest_user_addr + r->size - qva;
614 return qva - r->guest_user_addr +
626 * Converts ring address to Vhost virtual address.
627 * If IOMMU is enabled, the ring address is a guest IO virtual address,
628 * else it is a QEMU virtual address.
631 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
632 uint64_t ra, uint64_t *size)
634 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
637 vhost_user_iotlb_rd_lock(vq);
638 vva = vhost_iova_to_vva(dev, vq, ra,
639 size, VHOST_ACCESS_RW);
640 vhost_user_iotlb_rd_unlock(vq);
645 return qva_to_vva(dev, ra, size);
649 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
653 vhost_user_iotlb_rd_lock(vq);
654 log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
655 vhost_user_iotlb_rd_unlock(vq);
660 static struct virtio_net *
661 translate_ring_addresses(struct virtio_net *dev, int vq_index)
663 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
664 struct vhost_vring_addr *addr = &vq->ring_addrs;
665 uint64_t len, expected_len;
667 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
669 log_addr_to_gpa(dev, vq);
670 if (vq->log_guest_addr == 0) {
671 VHOST_LOG_CONFIG(DEBUG,
672 "(%d) failed to map log_guest_addr.\n",
678 if (vq_is_packed(dev)) {
679 len = sizeof(struct vring_packed_desc) * vq->size;
680 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
681 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
682 if (vq->desc_packed == NULL ||
683 len != sizeof(struct vring_packed_desc) *
685 VHOST_LOG_CONFIG(DEBUG,
686 "(%d) failed to map desc_packed ring.\n",
691 dev = numa_realloc(dev, vq_index);
692 vq = dev->virtqueue[vq_index];
693 addr = &vq->ring_addrs;
695 len = sizeof(struct vring_packed_desc_event);
696 vq->driver_event = (struct vring_packed_desc_event *)
697 (uintptr_t)ring_addr_to_vva(dev,
698 vq, addr->avail_user_addr, &len);
699 if (vq->driver_event == NULL ||
700 len != sizeof(struct vring_packed_desc_event)) {
701 VHOST_LOG_CONFIG(DEBUG,
702 "(%d) failed to find driver area address.\n",
707 len = sizeof(struct vring_packed_desc_event);
708 vq->device_event = (struct vring_packed_desc_event *)
709 (uintptr_t)ring_addr_to_vva(dev,
710 vq, addr->used_user_addr, &len);
711 if (vq->device_event == NULL ||
712 len != sizeof(struct vring_packed_desc_event)) {
713 VHOST_LOG_CONFIG(DEBUG,
714 "(%d) failed to find device area address.\n",
719 vq->access_ok = true;
723 /* The addresses are converted from QEMU virtual to Vhost virtual. */
724 if (vq->desc && vq->avail && vq->used)
727 len = sizeof(struct vring_desc) * vq->size;
728 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
729 vq, addr->desc_user_addr, &len);
730 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
731 VHOST_LOG_CONFIG(DEBUG,
732 "(%d) failed to map desc ring.\n",
737 dev = numa_realloc(dev, vq_index);
738 vq = dev->virtqueue[vq_index];
739 addr = &vq->ring_addrs;
741 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
742 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
743 len += sizeof(uint16_t);
745 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
746 vq, addr->avail_user_addr, &len);
747 if (vq->avail == 0 || len != expected_len) {
748 VHOST_LOG_CONFIG(DEBUG,
749 "(%d) failed to map avail ring.\n",
754 len = sizeof(struct vring_used) +
755 sizeof(struct vring_used_elem) * vq->size;
756 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
757 len += sizeof(uint16_t);
759 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
760 vq, addr->used_user_addr, &len);
761 if (vq->used == 0 || len != expected_len) {
762 VHOST_LOG_CONFIG(DEBUG,
763 "(%d) failed to map used ring.\n",
768 if (vq->last_used_idx != vq->used->idx) {
769 VHOST_LOG_CONFIG(WARNING,
770 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
771 "some packets maybe resent for Tx and dropped for Rx\n",
772 vq->last_used_idx, vq->used->idx);
773 vq->last_used_idx = vq->used->idx;
774 vq->last_avail_idx = vq->used->idx;
777 vq->access_ok = true;
779 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
781 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
782 dev->vid, vq->avail);
783 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
785 VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
786 dev->vid, vq->log_guest_addr);
792 * The virtio device sends us the desc, used and avail ring addresses.
793 * This function then converts these to our address space.
796 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
797 int main_fd __rte_unused)
799 struct virtio_net *dev = *pdev;
800 struct vhost_virtqueue *vq;
801 struct vhost_vring_addr *addr = &msg->payload.addr;
804 if (validate_msg_fds(msg, 0) != 0)
805 return RTE_VHOST_MSG_RESULT_ERR;
807 if (dev->mem == NULL)
808 return RTE_VHOST_MSG_RESULT_ERR;
810 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
811 vq = dev->virtqueue[msg->payload.addr.index];
813 access_ok = vq->access_ok;
816 * Rings addresses should not be interpreted as long as the ring is not
817 * started and enabled
819 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
821 vring_invalidate(dev, vq);
823 if ((vq->enabled && (dev->features &
824 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
826 dev = translate_ring_addresses(dev, msg->payload.addr.index);
828 return RTE_VHOST_MSG_RESULT_ERR;
833 return RTE_VHOST_MSG_RESULT_OK;
837 * The virtio device sends us the available ring last used index.
840 vhost_user_set_vring_base(struct virtio_net **pdev,
841 struct VhostUserMsg *msg,
842 int main_fd __rte_unused)
844 struct virtio_net *dev = *pdev;
845 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
846 uint64_t val = msg->payload.state.num;
848 if (validate_msg_fds(msg, 0) != 0)
849 return RTE_VHOST_MSG_RESULT_ERR;
851 if (vq_is_packed(dev)) {
853 * Bit[0:14]: avail index
854 * Bit[15]: avail wrap counter
856 vq->last_avail_idx = val & 0x7fff;
857 vq->avail_wrap_counter = !!(val & (0x1 << 15));
859 * Set used index to same value as available one, as
860 * their values should be the same since ring processing
861 * was stopped at get time.
863 vq->last_used_idx = vq->last_avail_idx;
864 vq->used_wrap_counter = vq->avail_wrap_counter;
866 vq->last_used_idx = msg->payload.state.num;
867 vq->last_avail_idx = msg->payload.state.num;
870 return RTE_VHOST_MSG_RESULT_OK;
874 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
875 uint64_t host_phys_addr, uint64_t size)
877 struct guest_page *page, *last_page;
878 struct guest_page *old_pages;
880 if (dev->nr_guest_pages == dev->max_guest_pages) {
881 dev->max_guest_pages *= 2;
882 old_pages = dev->guest_pages;
883 dev->guest_pages = rte_realloc(dev->guest_pages,
884 dev->max_guest_pages * sizeof(*page),
885 RTE_CACHE_LINE_SIZE);
886 if (dev->guest_pages == NULL) {
887 VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
893 if (dev->nr_guest_pages > 0) {
894 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
895 /* merge if the two pages are continuous */
896 if (host_phys_addr == last_page->host_phys_addr +
898 last_page->size += size;
903 page = &dev->guest_pages[dev->nr_guest_pages++];
904 page->guest_phys_addr = guest_phys_addr;
905 page->host_phys_addr = host_phys_addr;
912 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
915 uint64_t reg_size = reg->size;
916 uint64_t host_user_addr = reg->host_user_addr;
917 uint64_t guest_phys_addr = reg->guest_phys_addr;
918 uint64_t host_phys_addr;
921 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
922 size = page_size - (guest_phys_addr & (page_size - 1));
923 size = RTE_MIN(size, reg_size);
925 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
928 host_user_addr += size;
929 guest_phys_addr += size;
932 while (reg_size > 0) {
933 size = RTE_MIN(reg_size, page_size);
934 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
936 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
940 host_user_addr += size;
941 guest_phys_addr += size;
945 /* sort guest page array if over binary search threshold */
946 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
947 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
948 sizeof(struct guest_page), guest_page_addrcmp);
954 #ifdef RTE_LIBRTE_VHOST_DEBUG
955 /* TODO: enable it only in debug mode? */
957 dump_guest_pages(struct virtio_net *dev)
960 struct guest_page *page;
962 for (i = 0; i < dev->nr_guest_pages; i++) {
963 page = &dev->guest_pages[i];
965 VHOST_LOG_CONFIG(INFO,
966 "guest physical page region %u\n"
967 "\t guest_phys_addr: %" PRIx64 "\n"
968 "\t host_phys_addr : %" PRIx64 "\n"
969 "\t size : %" PRIx64 "\n",
971 page->guest_phys_addr,
972 page->host_phys_addr,
977 #define dump_guest_pages(dev)
981 vhost_memory_changed(struct VhostUserMemory *new,
982 struct rte_vhost_memory *old)
986 if (new->nregions != old->nregions)
989 for (i = 0; i < new->nregions; ++i) {
990 VhostUserMemoryRegion *new_r = &new->regions[i];
991 struct rte_vhost_mem_region *old_r = &old->regions[i];
993 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
995 if (new_r->memory_size != old_r->size)
997 if (new_r->userspace_addr != old_r->guest_user_addr)
1004 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1006 vhost_user_postcopy_region_register(struct virtio_net *dev,
1007 struct rte_vhost_mem_region *reg)
1009 struct uffdio_register reg_struct;
1012 * Let's register all the mmap'ed area to ensure
1013 * alignment on page boundary.
1015 reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1016 reg_struct.range.len = reg->mmap_size;
1017 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1019 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1021 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1022 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1023 (uint64_t)reg_struct.range.start,
1024 (uint64_t)reg_struct.range.start +
1025 (uint64_t)reg_struct.range.len - 1,
1031 VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1032 (uint64_t)reg_struct.range.start,
1033 (uint64_t)reg_struct.range.start +
1034 (uint64_t)reg_struct.range.len - 1);
1040 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1041 struct rte_vhost_mem_region *reg __rte_unused)
1048 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1049 struct VhostUserMsg *msg)
1051 struct VhostUserMemory *memory;
1052 struct rte_vhost_mem_region *reg;
1053 VhostUserMsg ack_msg;
1056 if (!dev->postcopy_listening)
1060 * We haven't a better way right now than sharing
1061 * DPDK's virtual address with Qemu, so that Qemu can
1062 * retrieve the region offset when handling userfaults.
1064 memory = &msg->payload.memory;
1065 for (i = 0; i < memory->nregions; i++) {
1066 reg = &dev->mem->regions[i];
1067 memory->regions[i].userspace_addr = reg->host_user_addr;
1070 /* Send the addresses back to qemu */
1072 send_vhost_reply(main_fd, msg);
1074 /* Wait for qemu to acknolwedge it's got the addresses
1075 * we've got to wait before we're allowed to generate faults.
1077 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1078 VHOST_LOG_CONFIG(ERR,
1079 "Failed to read qemu ack on postcopy set-mem-table\n");
1083 if (validate_msg_fds(&ack_msg, 0) != 0)
1086 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1087 VHOST_LOG_CONFIG(ERR,
1088 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1089 ack_msg.request.master);
1093 /* Now userfault register and we can use the memory */
1094 for (i = 0; i < memory->nregions; i++) {
1095 reg = &dev->mem->regions[i];
1096 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1104 vhost_user_mmap_region(struct virtio_net *dev,
1105 struct rte_vhost_mem_region *region,
1106 uint64_t mmap_offset)
1113 /* Check for memory_size + mmap_offset overflow */
1114 if (mmap_offset >= -region->size) {
1115 VHOST_LOG_CONFIG(ERR,
1116 "mmap_offset (%#"PRIx64") and memory_size "
1117 "(%#"PRIx64") overflow\n",
1118 mmap_offset, region->size);
1122 mmap_size = region->size + mmap_offset;
1124 /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1125 * argument aligned with hugepagesz at older longterm version Linux,
1126 * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1128 * To avoid failure, make sure in caller to keep length aligned.
1130 alignment = get_blk_size(region->fd);
1131 if (alignment == (uint64_t)-1) {
1132 VHOST_LOG_CONFIG(ERR,
1133 "couldn't get hugepage size through fstat\n");
1136 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1137 if (mmap_size == 0) {
1139 * It could happen if initial mmap_size + alignment overflows
1140 * the sizeof uint64, which could happen if either mmap_size or
1141 * alignment value is wrong.
1143 * mmap() kernel implementation would return an error, but
1144 * better catch it before and provide useful info in the logs.
1146 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1147 "or alignment (0x%" PRIx64 ") is invalid\n",
1148 region->size + mmap_offset, alignment);
1152 populate = dev->async_copy ? MAP_POPULATE : 0;
1153 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1154 MAP_SHARED | populate, region->fd, 0);
1156 if (mmap_addr == MAP_FAILED) {
1157 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1161 region->mmap_addr = mmap_addr;
1162 region->mmap_size = mmap_size;
1163 region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1165 if (dev->async_copy)
1166 if (add_guest_pages(dev, region, alignment) < 0) {
1167 VHOST_LOG_CONFIG(ERR,
1168 "adding guest pages to region failed.\n");
1172 VHOST_LOG_CONFIG(INFO,
1173 "guest memory region size: 0x%" PRIx64 "\n"
1174 "\t guest physical addr: 0x%" PRIx64 "\n"
1175 "\t guest virtual addr: 0x%" PRIx64 "\n"
1176 "\t host virtual addr: 0x%" PRIx64 "\n"
1177 "\t mmap addr : 0x%" PRIx64 "\n"
1178 "\t mmap size : 0x%" PRIx64 "\n"
1179 "\t mmap align: 0x%" PRIx64 "\n"
1180 "\t mmap off : 0x%" PRIx64 "\n",
1182 region->guest_phys_addr,
1183 region->guest_user_addr,
1184 region->host_user_addr,
1185 (uint64_t)(uintptr_t)mmap_addr,
1194 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1197 struct virtio_net *dev = *pdev;
1198 struct VhostUserMemory *memory = &msg->payload.memory;
1199 struct rte_vhost_mem_region *reg;
1201 uint64_t mmap_offset;
1204 if (validate_msg_fds(msg, memory->nregions) != 0)
1205 return RTE_VHOST_MSG_RESULT_ERR;
1207 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1208 VHOST_LOG_CONFIG(ERR,
1209 "too many memory regions (%u)\n", memory->nregions);
1213 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1214 VHOST_LOG_CONFIG(INFO,
1215 "(%d) memory regions not changed\n", dev->vid);
1219 return RTE_VHOST_MSG_RESULT_OK;
1223 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1224 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1226 if (vdpa_dev && vdpa_dev->ops->dev_close)
1227 vdpa_dev->ops->dev_close(dev->vid);
1228 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1230 free_mem_region(dev);
1235 /* Flush IOTLB cache as previous HVAs are now invalid */
1236 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1237 for (i = 0; i < dev->nr_vring; i++)
1238 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1240 dev->nr_guest_pages = 0;
1241 if (dev->guest_pages == NULL) {
1242 dev->max_guest_pages = 8;
1243 dev->guest_pages = rte_zmalloc(NULL,
1244 dev->max_guest_pages *
1245 sizeof(struct guest_page),
1246 RTE_CACHE_LINE_SIZE);
1247 if (dev->guest_pages == NULL) {
1248 VHOST_LOG_CONFIG(ERR,
1249 "(%d) failed to allocate memory "
1250 "for dev->guest_pages\n",
1256 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1257 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1258 if (dev->mem == NULL) {
1259 VHOST_LOG_CONFIG(ERR,
1260 "(%d) failed to allocate memory for dev->mem\n",
1262 goto free_guest_pages;
1265 for (i = 0; i < memory->nregions; i++) {
1266 reg = &dev->mem->regions[i];
1268 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1269 reg->guest_user_addr = memory->regions[i].userspace_addr;
1270 reg->size = memory->regions[i].memory_size;
1271 reg->fd = msg->fds[i];
1274 * Assign invalid file descriptor value to avoid double
1275 * closing on error path.
1279 mmap_offset = memory->regions[i].mmap_offset;
1281 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1282 VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1283 goto free_mem_table;
1286 dev->mem->nregions++;
1289 if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1290 goto free_mem_table;
1292 for (i = 0; i < dev->nr_vring; i++) {
1293 struct vhost_virtqueue *vq = dev->virtqueue[i];
1298 if (vq->desc || vq->avail || vq->used) {
1300 * If the memory table got updated, the ring addresses
1301 * need to be translated again as virtual addresses have
1304 vring_invalidate(dev, vq);
1306 dev = translate_ring_addresses(dev, i);
1309 goto free_mem_table;
1316 dump_guest_pages(dev);
1318 return RTE_VHOST_MSG_RESULT_OK;
1321 free_mem_region(dev);
1325 rte_free(dev->guest_pages);
1326 dev->guest_pages = NULL;
1329 return RTE_VHOST_MSG_RESULT_ERR;
1333 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1340 if (vq_is_packed(dev))
1341 rings_ok = vq->desc_packed && vq->driver_event &&
1344 rings_ok = vq->desc && vq->avail && vq->used;
1347 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1348 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1352 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1355 virtio_is_ready(struct virtio_net *dev)
1357 struct vhost_virtqueue *vq;
1358 uint32_t i, nr_vring = dev->nr_vring;
1360 if (dev->flags & VIRTIO_DEV_READY)
1366 if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1367 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1369 if (dev->nr_vring < nr_vring)
1373 for (i = 0; i < nr_vring; i++) {
1374 vq = dev->virtqueue[i];
1376 if (!vq_is_ready(dev, vq))
1380 /* If supported, ensure the frontend is really done with config */
1381 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1382 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1385 dev->flags |= VIRTIO_DEV_READY;
1387 if (!(dev->flags & VIRTIO_DEV_RUNNING))
1388 VHOST_LOG_CONFIG(INFO,
1389 "virtio is now ready for processing.\n");
1394 inflight_mem_alloc(const char *name, size_t size, int *fd)
1398 char fname[20] = "/tmp/memfd-XXXXXX";
1401 #ifdef MEMFD_SUPPORTED
1402 mfd = memfd_create(name, MFD_CLOEXEC);
1407 mfd = mkstemp(fname);
1409 VHOST_LOG_CONFIG(ERR,
1410 "failed to get inflight buffer fd\n");
1417 if (ftruncate(mfd, size) == -1) {
1418 VHOST_LOG_CONFIG(ERR,
1419 "failed to alloc inflight buffer\n");
1424 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1425 if (ptr == MAP_FAILED) {
1426 VHOST_LOG_CONFIG(ERR,
1427 "failed to mmap inflight buffer\n");
1437 get_pervq_shm_size_split(uint16_t queue_size)
1439 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1440 queue_size + sizeof(uint64_t) +
1441 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1445 get_pervq_shm_size_packed(uint16_t queue_size)
1447 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1448 * queue_size + sizeof(uint64_t) +
1449 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1450 INFLIGHT_ALIGNMENT);
1454 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1456 int main_fd __rte_unused)
1458 struct rte_vhost_inflight_info_packed *inflight_packed;
1459 uint64_t pervq_inflight_size, mmap_size;
1460 uint16_t num_queues, queue_size;
1461 struct virtio_net *dev = *pdev;
1465 if (msg->size != sizeof(msg->payload.inflight)) {
1466 VHOST_LOG_CONFIG(ERR,
1467 "invalid get_inflight_fd message size is %d\n",
1469 return RTE_VHOST_MSG_RESULT_ERR;
1472 if (dev->inflight_info == NULL) {
1473 dev->inflight_info = calloc(1,
1474 sizeof(struct inflight_mem_info));
1475 if (!dev->inflight_info) {
1476 VHOST_LOG_CONFIG(ERR,
1477 "failed to alloc dev inflight area\n");
1478 return RTE_VHOST_MSG_RESULT_ERR;
1480 dev->inflight_info->fd = -1;
1483 num_queues = msg->payload.inflight.num_queues;
1484 queue_size = msg->payload.inflight.queue_size;
1486 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1487 msg->payload.inflight.num_queues);
1488 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1489 msg->payload.inflight.queue_size);
1491 if (vq_is_packed(dev))
1492 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1494 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1496 mmap_size = num_queues * pervq_inflight_size;
1497 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1499 VHOST_LOG_CONFIG(ERR,
1500 "failed to alloc vhost inflight area\n");
1501 msg->payload.inflight.mmap_size = 0;
1502 return RTE_VHOST_MSG_RESULT_ERR;
1504 memset(addr, 0, mmap_size);
1506 if (dev->inflight_info->addr) {
1507 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1508 dev->inflight_info->addr = NULL;
1511 if (dev->inflight_info->fd >= 0) {
1512 close(dev->inflight_info->fd);
1513 dev->inflight_info->fd = -1;
1516 dev->inflight_info->addr = addr;
1517 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1518 dev->inflight_info->fd = msg->fds[0] = fd;
1519 msg->payload.inflight.mmap_offset = 0;
1522 if (vq_is_packed(dev)) {
1523 for (i = 0; i < num_queues; i++) {
1525 (struct rte_vhost_inflight_info_packed *)addr;
1526 inflight_packed->used_wrap_counter = 1;
1527 inflight_packed->old_used_wrap_counter = 1;
1528 for (j = 0; j < queue_size; j++)
1529 inflight_packed->desc[j].next = j + 1;
1530 addr = (void *)((char *)addr + pervq_inflight_size);
1534 VHOST_LOG_CONFIG(INFO,
1535 "send inflight mmap_size: %"PRIu64"\n",
1536 msg->payload.inflight.mmap_size);
1537 VHOST_LOG_CONFIG(INFO,
1538 "send inflight mmap_offset: %"PRIu64"\n",
1539 msg->payload.inflight.mmap_offset);
1540 VHOST_LOG_CONFIG(INFO,
1541 "send inflight fd: %d\n", msg->fds[0]);
1543 return RTE_VHOST_MSG_RESULT_REPLY;
1547 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1548 int main_fd __rte_unused)
1550 uint64_t mmap_size, mmap_offset;
1551 uint16_t num_queues, queue_size;
1552 struct virtio_net *dev = *pdev;
1553 uint32_t pervq_inflight_size;
1554 struct vhost_virtqueue *vq;
1559 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1560 VHOST_LOG_CONFIG(ERR,
1561 "invalid set_inflight_fd message size is %d,fd is %d\n",
1563 return RTE_VHOST_MSG_RESULT_ERR;
1566 mmap_size = msg->payload.inflight.mmap_size;
1567 mmap_offset = msg->payload.inflight.mmap_offset;
1568 num_queues = msg->payload.inflight.num_queues;
1569 queue_size = msg->payload.inflight.queue_size;
1571 if (vq_is_packed(dev))
1572 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1574 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1576 VHOST_LOG_CONFIG(INFO,
1577 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1578 VHOST_LOG_CONFIG(INFO,
1579 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1580 VHOST_LOG_CONFIG(INFO,
1581 "set_inflight_fd num_queues: %u\n", num_queues);
1582 VHOST_LOG_CONFIG(INFO,
1583 "set_inflight_fd queue_size: %u\n", queue_size);
1584 VHOST_LOG_CONFIG(INFO,
1585 "set_inflight_fd fd: %d\n", fd);
1586 VHOST_LOG_CONFIG(INFO,
1587 "set_inflight_fd pervq_inflight_size: %d\n",
1588 pervq_inflight_size);
1590 if (!dev->inflight_info) {
1591 dev->inflight_info = calloc(1,
1592 sizeof(struct inflight_mem_info));
1593 if (dev->inflight_info == NULL) {
1594 VHOST_LOG_CONFIG(ERR,
1595 "failed to alloc dev inflight area\n");
1596 return RTE_VHOST_MSG_RESULT_ERR;
1598 dev->inflight_info->fd = -1;
1601 if (dev->inflight_info->addr) {
1602 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1603 dev->inflight_info->addr = NULL;
1606 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1608 if (addr == MAP_FAILED) {
1609 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1610 return RTE_VHOST_MSG_RESULT_ERR;
1613 if (dev->inflight_info->fd >= 0) {
1614 close(dev->inflight_info->fd);
1615 dev->inflight_info->fd = -1;
1618 dev->inflight_info->fd = fd;
1619 dev->inflight_info->addr = addr;
1620 dev->inflight_info->size = mmap_size;
1622 for (i = 0; i < num_queues; i++) {
1623 vq = dev->virtqueue[i];
1627 if (vq_is_packed(dev)) {
1628 vq->inflight_packed = addr;
1629 vq->inflight_packed->desc_num = queue_size;
1631 vq->inflight_split = addr;
1632 vq->inflight_split->desc_num = queue_size;
1634 addr = (void *)((char *)addr + pervq_inflight_size);
1637 return RTE_VHOST_MSG_RESULT_OK;
1641 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1642 int main_fd __rte_unused)
1644 struct virtio_net *dev = *pdev;
1645 struct vhost_vring_file file;
1646 struct vhost_virtqueue *vq;
1649 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1650 if (validate_msg_fds(msg, expected_fds) != 0)
1651 return RTE_VHOST_MSG_RESULT_ERR;
1653 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1654 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1655 file.fd = VIRTIO_INVALID_EVENTFD;
1657 file.fd = msg->fds[0];
1658 VHOST_LOG_CONFIG(INFO,
1659 "vring call idx:%d file:%d\n", file.index, file.fd);
1661 vq = dev->virtqueue[file.index];
1665 vhost_user_notify_queue_state(dev, file.index, 0);
1668 if (vq->callfd >= 0)
1671 vq->callfd = file.fd;
1673 return RTE_VHOST_MSG_RESULT_OK;
1676 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1677 struct VhostUserMsg *msg,
1678 int main_fd __rte_unused)
1682 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1683 if (validate_msg_fds(msg, expected_fds) != 0)
1684 return RTE_VHOST_MSG_RESULT_ERR;
1686 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1688 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1690 return RTE_VHOST_MSG_RESULT_OK;
1694 resubmit_desc_compare(const void *a, const void *b)
1696 const struct rte_vhost_resubmit_desc *desc0 = a;
1697 const struct rte_vhost_resubmit_desc *desc1 = b;
1699 if (desc1->counter > desc0->counter)
1706 vhost_check_queue_inflights_split(struct virtio_net *dev,
1707 struct vhost_virtqueue *vq)
1710 uint16_t resubmit_num = 0, last_io, num;
1711 struct vring_used *used = vq->used;
1712 struct rte_vhost_resubmit_info *resubmit;
1713 struct rte_vhost_inflight_info_split *inflight_split;
1715 if (!(dev->protocol_features &
1716 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1717 return RTE_VHOST_MSG_RESULT_OK;
1719 /* The frontend may still not support the inflight feature
1720 * although we negotiate the protocol feature.
1722 if ((!vq->inflight_split))
1723 return RTE_VHOST_MSG_RESULT_OK;
1725 if (!vq->inflight_split->version) {
1726 vq->inflight_split->version = INFLIGHT_VERSION;
1727 return RTE_VHOST_MSG_RESULT_OK;
1730 if (vq->resubmit_inflight)
1731 return RTE_VHOST_MSG_RESULT_OK;
1733 inflight_split = vq->inflight_split;
1734 vq->global_counter = 0;
1735 last_io = inflight_split->last_inflight_io;
1737 if (inflight_split->used_idx != used->idx) {
1738 inflight_split->desc[last_io].inflight = 0;
1739 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1740 inflight_split->used_idx = used->idx;
1743 for (i = 0; i < inflight_split->desc_num; i++) {
1744 if (inflight_split->desc[i].inflight == 1)
1748 vq->last_avail_idx += resubmit_num;
1751 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1753 VHOST_LOG_CONFIG(ERR,
1754 "failed to allocate memory for resubmit info.\n");
1755 return RTE_VHOST_MSG_RESULT_ERR;
1758 resubmit->resubmit_list = calloc(resubmit_num,
1759 sizeof(struct rte_vhost_resubmit_desc));
1760 if (!resubmit->resubmit_list) {
1761 VHOST_LOG_CONFIG(ERR,
1762 "failed to allocate memory for inflight desc.\n");
1764 return RTE_VHOST_MSG_RESULT_ERR;
1768 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1769 if (vq->inflight_split->desc[i].inflight == 1) {
1770 resubmit->resubmit_list[num].index = i;
1771 resubmit->resubmit_list[num].counter =
1772 inflight_split->desc[i].counter;
1776 resubmit->resubmit_num = num;
1778 if (resubmit->resubmit_num > 1)
1779 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1780 sizeof(struct rte_vhost_resubmit_desc),
1781 resubmit_desc_compare);
1783 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1784 vq->resubmit_inflight = resubmit;
1787 return RTE_VHOST_MSG_RESULT_OK;
1791 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1792 struct vhost_virtqueue *vq)
1795 uint16_t resubmit_num = 0, old_used_idx, num;
1796 struct rte_vhost_resubmit_info *resubmit;
1797 struct rte_vhost_inflight_info_packed *inflight_packed;
1799 if (!(dev->protocol_features &
1800 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1801 return RTE_VHOST_MSG_RESULT_OK;
1803 /* The frontend may still not support the inflight feature
1804 * although we negotiate the protocol feature.
1806 if ((!vq->inflight_packed))
1807 return RTE_VHOST_MSG_RESULT_OK;
1809 if (!vq->inflight_packed->version) {
1810 vq->inflight_packed->version = INFLIGHT_VERSION;
1811 return RTE_VHOST_MSG_RESULT_OK;
1814 if (vq->resubmit_inflight)
1815 return RTE_VHOST_MSG_RESULT_OK;
1817 inflight_packed = vq->inflight_packed;
1818 vq->global_counter = 0;
1819 old_used_idx = inflight_packed->old_used_idx;
1821 if (inflight_packed->used_idx != old_used_idx) {
1822 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1823 inflight_packed->old_used_idx =
1824 inflight_packed->used_idx;
1825 inflight_packed->old_used_wrap_counter =
1826 inflight_packed->used_wrap_counter;
1827 inflight_packed->old_free_head =
1828 inflight_packed->free_head;
1830 inflight_packed->used_idx =
1831 inflight_packed->old_used_idx;
1832 inflight_packed->used_wrap_counter =
1833 inflight_packed->old_used_wrap_counter;
1834 inflight_packed->free_head =
1835 inflight_packed->old_free_head;
1839 for (i = 0; i < inflight_packed->desc_num; i++) {
1840 if (inflight_packed->desc[i].inflight == 1)
1845 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1846 if (resubmit == NULL) {
1847 VHOST_LOG_CONFIG(ERR,
1848 "failed to allocate memory for resubmit info.\n");
1849 return RTE_VHOST_MSG_RESULT_ERR;
1852 resubmit->resubmit_list = calloc(resubmit_num,
1853 sizeof(struct rte_vhost_resubmit_desc));
1854 if (resubmit->resubmit_list == NULL) {
1855 VHOST_LOG_CONFIG(ERR,
1856 "failed to allocate memory for resubmit desc.\n");
1858 return RTE_VHOST_MSG_RESULT_ERR;
1862 for (i = 0; i < inflight_packed->desc_num; i++) {
1863 if (vq->inflight_packed->desc[i].inflight == 1) {
1864 resubmit->resubmit_list[num].index = i;
1865 resubmit->resubmit_list[num].counter =
1866 inflight_packed->desc[i].counter;
1870 resubmit->resubmit_num = num;
1872 if (resubmit->resubmit_num > 1)
1873 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1874 sizeof(struct rte_vhost_resubmit_desc),
1875 resubmit_desc_compare);
1877 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1878 vq->resubmit_inflight = resubmit;
1881 return RTE_VHOST_MSG_RESULT_OK;
1885 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1886 int main_fd __rte_unused)
1888 struct virtio_net *dev = *pdev;
1889 struct vhost_vring_file file;
1890 struct vhost_virtqueue *vq;
1893 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1894 if (validate_msg_fds(msg, expected_fds) != 0)
1895 return RTE_VHOST_MSG_RESULT_ERR;
1897 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1898 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1899 file.fd = VIRTIO_INVALID_EVENTFD;
1901 file.fd = msg->fds[0];
1902 VHOST_LOG_CONFIG(INFO,
1903 "vring kick idx:%d file:%d\n", file.index, file.fd);
1905 /* Interpret ring addresses only when ring is started. */
1906 dev = translate_ring_addresses(dev, file.index);
1908 if (file.fd != VIRTIO_INVALID_EVENTFD)
1911 return RTE_VHOST_MSG_RESULT_ERR;
1916 vq = dev->virtqueue[file.index];
1919 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1920 * the ring starts already enabled. Otherwise, it is enabled via
1921 * the SET_VRING_ENABLE message.
1923 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1925 if (dev->notify_ops->vring_state_changed)
1926 dev->notify_ops->vring_state_changed(
1927 dev->vid, file.index, 1);
1932 vhost_user_notify_queue_state(dev, file.index, 0);
1935 if (vq->kickfd >= 0)
1937 vq->kickfd = file.fd;
1939 if (vq_is_packed(dev)) {
1940 if (vhost_check_queue_inflights_packed(dev, vq)) {
1941 VHOST_LOG_CONFIG(ERR,
1942 "failed to inflights for vq: %d\n", file.index);
1943 return RTE_VHOST_MSG_RESULT_ERR;
1946 if (vhost_check_queue_inflights_split(dev, vq)) {
1947 VHOST_LOG_CONFIG(ERR,
1948 "failed to inflights for vq: %d\n", file.index);
1949 return RTE_VHOST_MSG_RESULT_ERR;
1953 return RTE_VHOST_MSG_RESULT_OK;
1957 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1960 vhost_user_get_vring_base(struct virtio_net **pdev,
1961 struct VhostUserMsg *msg,
1962 int main_fd __rte_unused)
1964 struct virtio_net *dev = *pdev;
1965 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1968 if (validate_msg_fds(msg, 0) != 0)
1969 return RTE_VHOST_MSG_RESULT_ERR;
1971 /* We have to stop the queue (virtio) if it is running. */
1972 vhost_destroy_device_notify(dev);
1974 dev->flags &= ~VIRTIO_DEV_READY;
1975 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1977 /* Here we are safe to get the indexes */
1978 if (vq_is_packed(dev)) {
1980 * Bit[0:14]: avail index
1981 * Bit[15]: avail wrap counter
1983 val = vq->last_avail_idx & 0x7fff;
1984 val |= vq->avail_wrap_counter << 15;
1985 msg->payload.state.num = val;
1987 msg->payload.state.num = vq->last_avail_idx;
1990 VHOST_LOG_CONFIG(INFO,
1991 "vring base idx:%d file:%d\n", msg->payload.state.index,
1992 msg->payload.state.num);
1994 * Based on current qemu vhost-user implementation, this message is
1995 * sent and only sent in vhost_vring_stop.
1996 * TODO: cleanup the vring, it isn't usable since here.
1998 if (vq->kickfd >= 0)
2001 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2003 if (vq->callfd >= 0)
2006 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2008 vq->signalled_used_valid = false;
2010 if (vq_is_packed(dev)) {
2011 rte_free(vq->shadow_used_packed);
2012 vq->shadow_used_packed = NULL;
2014 rte_free(vq->shadow_used_split);
2015 vq->shadow_used_split = NULL;
2017 if (vq->async_pkts_info)
2018 rte_free(vq->async_pkts_info);
2019 if (vq->async_descs_split)
2020 rte_free(vq->async_descs_split);
2021 vq->async_pkts_info = NULL;
2022 vq->async_descs_split = NULL;
2025 rte_free(vq->batch_copy_elems);
2026 vq->batch_copy_elems = NULL;
2028 rte_free(vq->log_cache);
2029 vq->log_cache = NULL;
2031 msg->size = sizeof(msg->payload.state);
2034 vring_invalidate(dev, vq);
2036 return RTE_VHOST_MSG_RESULT_REPLY;
2040 * when virtio queues are ready to work, qemu will send us to
2041 * enable the virtio queue pair.
2044 vhost_user_set_vring_enable(struct virtio_net **pdev,
2045 struct VhostUserMsg *msg,
2046 int main_fd __rte_unused)
2048 struct virtio_net *dev = *pdev;
2049 bool enable = !!msg->payload.state.num;
2050 int index = (int)msg->payload.state.index;
2052 if (validate_msg_fds(msg, 0) != 0)
2053 return RTE_VHOST_MSG_RESULT_ERR;
2055 VHOST_LOG_CONFIG(INFO,
2056 "set queue enable: %d to qp idx: %d\n",
2059 if (enable && dev->virtqueue[index]->async_registered) {
2060 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2061 VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2062 "async inflight packets must be completed first\n");
2063 return RTE_VHOST_MSG_RESULT_ERR;
2067 dev->virtqueue[index]->enabled = enable;
2069 return RTE_VHOST_MSG_RESULT_OK;
2073 vhost_user_get_protocol_features(struct virtio_net **pdev,
2074 struct VhostUserMsg *msg,
2075 int main_fd __rte_unused)
2077 struct virtio_net *dev = *pdev;
2078 uint64_t features, protocol_features;
2080 if (validate_msg_fds(msg, 0) != 0)
2081 return RTE_VHOST_MSG_RESULT_ERR;
2083 rte_vhost_driver_get_features(dev->ifname, &features);
2084 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2086 msg->payload.u64 = protocol_features;
2087 msg->size = sizeof(msg->payload.u64);
2090 return RTE_VHOST_MSG_RESULT_REPLY;
2094 vhost_user_set_protocol_features(struct virtio_net **pdev,
2095 struct VhostUserMsg *msg,
2096 int main_fd __rte_unused)
2098 struct virtio_net *dev = *pdev;
2099 uint64_t protocol_features = msg->payload.u64;
2100 uint64_t slave_protocol_features = 0;
2102 if (validate_msg_fds(msg, 0) != 0)
2103 return RTE_VHOST_MSG_RESULT_ERR;
2105 rte_vhost_driver_get_protocol_features(dev->ifname,
2106 &slave_protocol_features);
2107 if (protocol_features & ~slave_protocol_features) {
2108 VHOST_LOG_CONFIG(ERR,
2109 "(%d) received invalid protocol features.\n",
2111 return RTE_VHOST_MSG_RESULT_ERR;
2114 dev->protocol_features = protocol_features;
2115 VHOST_LOG_CONFIG(INFO,
2116 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2117 dev->protocol_features);
2119 return RTE_VHOST_MSG_RESULT_OK;
2123 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2124 int main_fd __rte_unused)
2126 struct virtio_net *dev = *pdev;
2127 int fd = msg->fds[0];
2132 if (validate_msg_fds(msg, 1) != 0)
2133 return RTE_VHOST_MSG_RESULT_ERR;
2136 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2137 return RTE_VHOST_MSG_RESULT_ERR;
2140 if (msg->size != sizeof(VhostUserLog)) {
2141 VHOST_LOG_CONFIG(ERR,
2142 "invalid log base msg size: %"PRId32" != %d\n",
2143 msg->size, (int)sizeof(VhostUserLog));
2147 size = msg->payload.log.mmap_size;
2148 off = msg->payload.log.mmap_offset;
2150 /* Check for mmap size and offset overflow. */
2152 VHOST_LOG_CONFIG(ERR,
2153 "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2158 VHOST_LOG_CONFIG(INFO,
2159 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2163 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2164 * fail when offset is not page size aligned.
2166 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2168 if (addr == MAP_FAILED) {
2169 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2170 return RTE_VHOST_MSG_RESULT_ERR;
2174 * Free previously mapped log memory on occasionally
2175 * multiple VHOST_USER_SET_LOG_BASE.
2177 if (dev->log_addr) {
2178 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2180 dev->log_addr = (uint64_t)(uintptr_t)addr;
2181 dev->log_base = dev->log_addr + off;
2182 dev->log_size = size;
2184 for (i = 0; i < dev->nr_vring; i++) {
2185 struct vhost_virtqueue *vq = dev->virtqueue[i];
2187 rte_free(vq->log_cache);
2188 vq->log_cache = NULL;
2189 vq->log_cache_nb_elem = 0;
2190 vq->log_cache = rte_zmalloc("vq log cache",
2191 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2194 * If log cache alloc fail, don't fail migration, but no
2195 * caching will be done, which will impact performance
2198 VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2202 * The spec is not clear about it (yet), but QEMU doesn't expect
2203 * any payload in the reply.
2208 return RTE_VHOST_MSG_RESULT_REPLY;
2212 return RTE_VHOST_MSG_RESULT_ERR;
2215 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2216 struct VhostUserMsg *msg,
2217 int main_fd __rte_unused)
2219 if (validate_msg_fds(msg, 1) != 0)
2220 return RTE_VHOST_MSG_RESULT_ERR;
2223 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2225 return RTE_VHOST_MSG_RESULT_OK;
2229 * An rarp packet is constructed and broadcasted to notify switches about
2230 * the new location of the migrated VM, so that packets from outside will
2231 * not be lost after migration.
2233 * However, we don't actually "send" a rarp packet here, instead, we set
2234 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2237 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2238 int main_fd __rte_unused)
2240 struct virtio_net *dev = *pdev;
2241 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2242 struct rte_vdpa_device *vdpa_dev;
2244 if (validate_msg_fds(msg, 0) != 0)
2245 return RTE_VHOST_MSG_RESULT_ERR;
2247 VHOST_LOG_CONFIG(DEBUG,
2248 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2249 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2250 memcpy(dev->mac.addr_bytes, mac, 6);
2253 * Set the flag to inject a RARP broadcast packet at
2254 * rte_vhost_dequeue_burst().
2256 * __ATOMIC_RELEASE ordering is for making sure the mac is
2257 * copied before the flag is set.
2259 __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2260 vdpa_dev = dev->vdpa_dev;
2261 if (vdpa_dev && vdpa_dev->ops->migration_done)
2262 vdpa_dev->ops->migration_done(dev->vid);
2264 return RTE_VHOST_MSG_RESULT_OK;
2268 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2269 int main_fd __rte_unused)
2271 struct virtio_net *dev = *pdev;
2273 if (validate_msg_fds(msg, 0) != 0)
2274 return RTE_VHOST_MSG_RESULT_ERR;
2276 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2277 msg->payload.u64 > VIRTIO_MAX_MTU) {
2278 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2281 return RTE_VHOST_MSG_RESULT_ERR;
2284 dev->mtu = msg->payload.u64;
2286 return RTE_VHOST_MSG_RESULT_OK;
2290 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2291 int main_fd __rte_unused)
2293 struct virtio_net *dev = *pdev;
2294 int fd = msg->fds[0];
2296 if (validate_msg_fds(msg, 1) != 0)
2297 return RTE_VHOST_MSG_RESULT_ERR;
2300 VHOST_LOG_CONFIG(ERR,
2301 "Invalid file descriptor for slave channel (%d)\n",
2303 return RTE_VHOST_MSG_RESULT_ERR;
2306 if (dev->slave_req_fd >= 0)
2307 close(dev->slave_req_fd);
2309 dev->slave_req_fd = fd;
2311 return RTE_VHOST_MSG_RESULT_OK;
2315 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2317 struct vhost_vring_addr *ra;
2318 uint64_t start, end, len;
2321 end = start + imsg->size;
2323 ra = &vq->ring_addrs;
2324 len = sizeof(struct vring_desc) * vq->size;
2325 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2328 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2329 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2332 len = sizeof(struct vring_used) +
2333 sizeof(struct vring_used_elem) * vq->size;
2334 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2337 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2338 len = sizeof(uint64_t);
2339 if (ra->log_guest_addr < end &&
2340 (ra->log_guest_addr + len) > start)
2348 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2350 struct vhost_vring_addr *ra;
2351 uint64_t start, end, len;
2354 end = start + imsg->size;
2356 ra = &vq->ring_addrs;
2357 len = sizeof(struct vring_packed_desc) * vq->size;
2358 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2361 len = sizeof(struct vring_packed_desc_event);
2362 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2365 len = sizeof(struct vring_packed_desc_event);
2366 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2369 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2370 len = sizeof(uint64_t);
2371 if (ra->log_guest_addr < end &&
2372 (ra->log_guest_addr + len) > start)
2379 static int is_vring_iotlb(struct virtio_net *dev,
2380 struct vhost_virtqueue *vq,
2381 struct vhost_iotlb_msg *imsg)
2383 if (vq_is_packed(dev))
2384 return is_vring_iotlb_packed(vq, imsg);
2386 return is_vring_iotlb_split(vq, imsg);
2390 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2391 int main_fd __rte_unused)
2393 struct virtio_net *dev = *pdev;
2394 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2398 if (validate_msg_fds(msg, 0) != 0)
2399 return RTE_VHOST_MSG_RESULT_ERR;
2401 switch (imsg->type) {
2402 case VHOST_IOTLB_UPDATE:
2404 vva = qva_to_vva(dev, imsg->uaddr, &len);
2406 return RTE_VHOST_MSG_RESULT_ERR;
2408 for (i = 0; i < dev->nr_vring; i++) {
2409 struct vhost_virtqueue *vq = dev->virtqueue[i];
2414 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2417 if (is_vring_iotlb(dev, vq, imsg))
2418 *pdev = dev = translate_ring_addresses(dev, i);
2421 case VHOST_IOTLB_INVALIDATE:
2422 for (i = 0; i < dev->nr_vring; i++) {
2423 struct vhost_virtqueue *vq = dev->virtqueue[i];
2428 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2431 if (is_vring_iotlb(dev, vq, imsg))
2432 vring_invalidate(dev, vq);
2436 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2438 return RTE_VHOST_MSG_RESULT_ERR;
2441 return RTE_VHOST_MSG_RESULT_OK;
2445 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2446 struct VhostUserMsg *msg,
2447 int main_fd __rte_unused)
2449 struct virtio_net *dev = *pdev;
2450 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2451 struct uffdio_api api_struct;
2453 if (validate_msg_fds(msg, 0) != 0)
2454 return RTE_VHOST_MSG_RESULT_ERR;
2456 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2458 if (dev->postcopy_ufd == -1) {
2459 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2461 return RTE_VHOST_MSG_RESULT_ERR;
2463 api_struct.api = UFFD_API;
2464 api_struct.features = 0;
2465 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2466 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2468 close(dev->postcopy_ufd);
2469 dev->postcopy_ufd = -1;
2470 return RTE_VHOST_MSG_RESULT_ERR;
2472 msg->fds[0] = dev->postcopy_ufd;
2475 return RTE_VHOST_MSG_RESULT_REPLY;
2477 dev->postcopy_ufd = -1;
2480 return RTE_VHOST_MSG_RESULT_ERR;
2485 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2486 struct VhostUserMsg *msg __rte_unused,
2487 int main_fd __rte_unused)
2489 struct virtio_net *dev = *pdev;
2491 if (validate_msg_fds(msg, 0) != 0)
2492 return RTE_VHOST_MSG_RESULT_ERR;
2494 if (dev->mem && dev->mem->nregions) {
2495 VHOST_LOG_CONFIG(ERR,
2496 "Regions already registered at postcopy-listen\n");
2497 return RTE_VHOST_MSG_RESULT_ERR;
2499 dev->postcopy_listening = 1;
2501 return RTE_VHOST_MSG_RESULT_OK;
2505 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2506 int main_fd __rte_unused)
2508 struct virtio_net *dev = *pdev;
2510 if (validate_msg_fds(msg, 0) != 0)
2511 return RTE_VHOST_MSG_RESULT_ERR;
2513 dev->postcopy_listening = 0;
2514 if (dev->postcopy_ufd >= 0) {
2515 close(dev->postcopy_ufd);
2516 dev->postcopy_ufd = -1;
2519 msg->payload.u64 = 0;
2520 msg->size = sizeof(msg->payload.u64);
2523 return RTE_VHOST_MSG_RESULT_REPLY;
2527 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2528 int main_fd __rte_unused)
2530 struct virtio_net *dev = *pdev;
2532 if (validate_msg_fds(msg, 0) != 0)
2533 return RTE_VHOST_MSG_RESULT_ERR;
2535 msg->payload.u64 = dev->status;
2536 msg->size = sizeof(msg->payload.u64);
2539 return RTE_VHOST_MSG_RESULT_REPLY;
2543 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2544 int main_fd __rte_unused)
2546 struct virtio_net *dev = *pdev;
2548 if (validate_msg_fds(msg, 0) != 0)
2549 return RTE_VHOST_MSG_RESULT_ERR;
2551 /* As per Virtio specification, the device status is 8bits long */
2552 if (msg->payload.u64 > UINT8_MAX) {
2553 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2555 return RTE_VHOST_MSG_RESULT_ERR;
2558 dev->status = msg->payload.u64;
2560 if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2561 (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2562 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2564 * Clear the bit to let the driver know about the feature
2565 * negotiation failure
2567 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2570 VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2572 "\t-ACKNOWLEDGE: %u\n"
2574 "\t-FEATURES_OK: %u\n"
2575 "\t-DRIVER_OK: %u\n"
2576 "\t-DEVICE_NEED_RESET: %u\n"
2579 (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2580 !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2581 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2582 !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2583 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2584 !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2585 !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2587 return RTE_VHOST_MSG_RESULT_OK;
2590 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2591 struct VhostUserMsg *msg,
2593 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2594 [VHOST_USER_NONE] = NULL,
2595 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2596 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2597 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2598 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2599 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2600 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2601 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2602 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2603 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2604 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2605 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2606 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2607 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2608 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2609 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2610 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2611 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2612 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2613 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2614 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2615 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2616 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2617 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2618 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2619 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2620 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2621 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2622 [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2623 [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2626 /* return bytes# of read on success or negative val on failure. */
2628 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2632 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2633 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2636 } else if (ret != VHOST_USER_HDR_SIZE) {
2637 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2643 if (msg->size > sizeof(msg->payload)) {
2644 VHOST_LOG_CONFIG(ERR,
2645 "invalid msg size: %d\n", msg->size);
2648 ret = read(sockfd, &msg->payload, msg->size);
2651 if (ret != (int)msg->size) {
2652 VHOST_LOG_CONFIG(ERR,
2653 "read control message failed\n");
2662 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2667 return send_fd_message(sockfd, (char *)msg,
2668 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2672 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2677 msg->flags &= ~VHOST_USER_VERSION_MASK;
2678 msg->flags &= ~VHOST_USER_NEED_REPLY;
2679 msg->flags |= VHOST_USER_VERSION;
2680 msg->flags |= VHOST_USER_REPLY_MASK;
2682 return send_vhost_message(sockfd, msg);
2686 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2690 if (msg->flags & VHOST_USER_NEED_REPLY)
2691 rte_spinlock_lock(&dev->slave_req_lock);
2693 ret = send_vhost_message(dev->slave_req_fd, msg);
2694 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2695 rte_spinlock_unlock(&dev->slave_req_lock);
2701 * Allocate a queue pair if it hasn't been allocated yet
2704 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2705 struct VhostUserMsg *msg)
2709 switch (msg->request.master) {
2710 case VHOST_USER_SET_VRING_KICK:
2711 case VHOST_USER_SET_VRING_CALL:
2712 case VHOST_USER_SET_VRING_ERR:
2713 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2715 case VHOST_USER_SET_VRING_NUM:
2716 case VHOST_USER_SET_VRING_BASE:
2717 case VHOST_USER_SET_VRING_ENABLE:
2718 vring_idx = msg->payload.state.index;
2720 case VHOST_USER_SET_VRING_ADDR:
2721 vring_idx = msg->payload.addr.index;
2727 if (vring_idx >= VHOST_MAX_VRING) {
2728 VHOST_LOG_CONFIG(ERR,
2729 "invalid vring index: %u\n", vring_idx);
2733 if (dev->virtqueue[vring_idx])
2736 return alloc_vring_queue(dev, vring_idx);
2740 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2743 unsigned int vq_num = 0;
2745 while (vq_num < dev->nr_vring) {
2746 struct vhost_virtqueue *vq = dev->virtqueue[i];
2749 rte_spinlock_lock(&vq->access_lock);
2757 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2760 unsigned int vq_num = 0;
2762 while (vq_num < dev->nr_vring) {
2763 struct vhost_virtqueue *vq = dev->virtqueue[i];
2766 rte_spinlock_unlock(&vq->access_lock);
2774 vhost_user_msg_handler(int vid, int fd)
2776 struct virtio_net *dev;
2777 struct VhostUserMsg msg;
2778 struct rte_vdpa_device *vdpa_dev;
2780 int unlock_required = 0;
2785 dev = get_device(vid);
2789 if (!dev->notify_ops) {
2790 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2791 if (!dev->notify_ops) {
2792 VHOST_LOG_CONFIG(ERR,
2793 "failed to get callback ops for driver %s\n",
2799 ret = read_vhost_message(fd, &msg);
2802 VHOST_LOG_CONFIG(ERR,
2803 "vhost read message failed\n");
2805 VHOST_LOG_CONFIG(INFO,
2806 "vhost peer closed\n");
2812 request = msg.request.master;
2813 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2814 vhost_message_str[request]) {
2815 if (request != VHOST_USER_IOTLB_MSG)
2816 VHOST_LOG_CONFIG(INFO, "read message %s\n",
2817 vhost_message_str[request]);
2819 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2820 vhost_message_str[request]);
2822 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2825 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2827 VHOST_LOG_CONFIG(ERR,
2828 "failed to alloc queue\n");
2833 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2834 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2835 * and device is destroyed. destroy_device waits for queues to be
2836 * inactive, so it is safe. Otherwise taking the access_lock
2837 * would cause a dead lock.
2840 case VHOST_USER_SET_FEATURES:
2841 case VHOST_USER_SET_PROTOCOL_FEATURES:
2842 case VHOST_USER_SET_OWNER:
2843 case VHOST_USER_SET_MEM_TABLE:
2844 case VHOST_USER_SET_LOG_BASE:
2845 case VHOST_USER_SET_LOG_FD:
2846 case VHOST_USER_SET_VRING_NUM:
2847 case VHOST_USER_SET_VRING_ADDR:
2848 case VHOST_USER_SET_VRING_BASE:
2849 case VHOST_USER_SET_VRING_KICK:
2850 case VHOST_USER_SET_VRING_CALL:
2851 case VHOST_USER_SET_VRING_ERR:
2852 case VHOST_USER_SET_VRING_ENABLE:
2853 case VHOST_USER_SEND_RARP:
2854 case VHOST_USER_NET_SET_MTU:
2855 case VHOST_USER_SET_SLAVE_REQ_FD:
2856 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2857 vhost_user_lock_all_queue_pairs(dev);
2858 unlock_required = 1;
2867 if (dev->extern_ops.pre_msg_handle) {
2868 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2871 case RTE_VHOST_MSG_RESULT_REPLY:
2872 send_vhost_reply(fd, &msg);
2874 case RTE_VHOST_MSG_RESULT_ERR:
2875 case RTE_VHOST_MSG_RESULT_OK:
2877 goto skip_to_post_handle;
2878 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2884 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2885 if (!vhost_message_handlers[request])
2886 goto skip_to_post_handle;
2887 ret = vhost_message_handlers[request](&dev, &msg, fd);
2890 case RTE_VHOST_MSG_RESULT_ERR:
2891 VHOST_LOG_CONFIG(ERR,
2892 "Processing %s failed.\n",
2893 vhost_message_str[request]);
2896 case RTE_VHOST_MSG_RESULT_OK:
2897 VHOST_LOG_CONFIG(DEBUG,
2898 "Processing %s succeeded.\n",
2899 vhost_message_str[request]);
2902 case RTE_VHOST_MSG_RESULT_REPLY:
2903 VHOST_LOG_CONFIG(DEBUG,
2904 "Processing %s succeeded and needs reply.\n",
2905 vhost_message_str[request]);
2906 send_vhost_reply(fd, &msg);
2914 skip_to_post_handle:
2915 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2916 dev->extern_ops.post_msg_handle) {
2917 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2920 case RTE_VHOST_MSG_RESULT_REPLY:
2921 send_vhost_reply(fd, &msg);
2923 case RTE_VHOST_MSG_RESULT_ERR:
2924 case RTE_VHOST_MSG_RESULT_OK:
2926 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2932 if (unlock_required)
2933 vhost_user_unlock_all_queue_pairs(dev);
2935 /* If message was not handled at this stage, treat it as an error */
2937 VHOST_LOG_CONFIG(ERR,
2938 "vhost message (req: %d) was not handled.\n", request);
2939 close_msg_fds(&msg);
2940 ret = RTE_VHOST_MSG_RESULT_ERR;
2944 * If the request required a reply that was already sent,
2945 * this optional reply-ack won't be sent as the
2946 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2948 if (msg.flags & VHOST_USER_NEED_REPLY) {
2949 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2950 msg.size = sizeof(msg.payload.u64);
2952 send_vhost_reply(fd, &msg);
2953 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2954 VHOST_LOG_CONFIG(ERR,
2955 "vhost message handling failed.\n");
2959 for (i = 0; i < dev->nr_vring; i++) {
2960 struct vhost_virtqueue *vq = dev->virtqueue[i];
2961 bool cur_ready = vq_is_ready(dev, vq);
2963 if (cur_ready != (vq && vq->ready)) {
2964 vq->ready = cur_ready;
2965 vhost_user_notify_queue_state(dev, i, cur_ready);
2970 if (!virtio_is_ready(dev))
2974 * Virtio is now ready. If not done already, it is time
2975 * to notify the application it can process the rings and
2976 * configure the vDPA device if present.
2979 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2980 if (dev->notify_ops->new_device(dev->vid) == 0)
2981 dev->flags |= VIRTIO_DEV_RUNNING;
2984 vdpa_dev = dev->vdpa_dev;
2988 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2989 if (vdpa_dev->ops->dev_conf(dev->vid))
2990 VHOST_LOG_CONFIG(ERR,
2991 "Failed to configure vDPA device\n");
2993 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3000 static int process_slave_message_reply(struct virtio_net *dev,
3001 const struct VhostUserMsg *msg)
3003 struct VhostUserMsg msg_reply;
3006 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3009 ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3012 VHOST_LOG_CONFIG(ERR,
3013 "vhost read slave message reply failed\n");
3015 VHOST_LOG_CONFIG(INFO,
3016 "vhost peer closed\n");
3022 if (msg_reply.request.slave != msg->request.slave) {
3023 VHOST_LOG_CONFIG(ERR,
3024 "Received unexpected msg type (%u), expected %u\n",
3025 msg_reply.request.slave, msg->request.slave);
3030 ret = msg_reply.payload.u64 ? -1 : 0;
3033 rte_spinlock_unlock(&dev->slave_req_lock);
3038 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3041 struct VhostUserMsg msg = {
3042 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3043 .flags = VHOST_USER_VERSION,
3044 .size = sizeof(msg.payload.iotlb),
3048 .type = VHOST_IOTLB_MISS,
3052 ret = send_vhost_message(dev->slave_req_fd, &msg);
3054 VHOST_LOG_CONFIG(ERR,
3055 "Failed to send IOTLB miss message (%d)\n",
3064 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3067 struct VhostUserMsg msg = {
3068 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3069 .flags = VHOST_USER_VERSION,
3074 msg.flags |= VHOST_USER_NEED_REPLY;
3076 ret = send_vhost_slave_message(dev, &msg);
3078 VHOST_LOG_CONFIG(ERR,
3079 "Failed to send config change (%d)\n",
3084 return process_slave_message_reply(dev, &msg);
3088 rte_vhost_slave_config_change(int vid, bool need_reply)
3090 struct virtio_net *dev;
3092 dev = get_device(vid);
3096 return vhost_user_slave_config_change(dev, need_reply);
3099 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3105 struct VhostUserMsg msg = {
3106 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3107 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3108 .size = sizeof(msg.payload.area),
3110 .u64 = index & VHOST_USER_VRING_IDX_MASK,
3117 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3123 ret = send_vhost_slave_message(dev, &msg);
3125 VHOST_LOG_CONFIG(ERR,
3126 "Failed to set host notifier (%d)\n", ret);
3130 return process_slave_message_reply(dev, &msg);
3133 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3135 struct virtio_net *dev;
3136 struct rte_vdpa_device *vdpa_dev;
3137 int vfio_device_fd, ret = 0;
3138 uint64_t offset, size;
3139 unsigned int i, q_start, q_last;
3141 dev = get_device(vid);
3145 vdpa_dev = dev->vdpa_dev;
3146 if (vdpa_dev == NULL)
3149 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3150 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3151 !(dev->protocol_features &
3152 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3153 !(dev->protocol_features &
3154 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3155 !(dev->protocol_features &
3156 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3159 if (qid == RTE_VHOST_QUEUE_ALL) {
3161 q_last = dev->nr_vring - 1;
3163 if (qid >= dev->nr_vring)
3169 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3170 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3172 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3173 if (vfio_device_fd < 0)
3177 for (i = q_start; i <= q_last; i++) {
3178 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3184 if (vhost_user_slave_set_vring_host_notifier(dev, i,
3185 vfio_device_fd, offset, size) < 0) {
3192 for (i = q_start; i <= q_last; i++) {
3193 vhost_user_slave_set_vring_host_notifier(dev, i, -1,