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",
93 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
94 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
97 close_msg_fds(struct VhostUserMsg *msg)
101 for (i = 0; i < msg->fd_num; i++)
106 * Ensure the expected number of FDs is received,
107 * close all FDs and return an error if this is not the case.
110 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
112 if (msg->fd_num == expected_fds)
115 VHOST_LOG_CONFIG(ERR,
116 " Expect %d FDs for request %s, received %d\n",
118 vhost_message_str[msg->request.master],
132 ret = fstat(fd, &stat);
133 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
137 * Reclaim all the outstanding zmbufs for a virtqueue.
140 drain_zmbuf_list(struct vhost_virtqueue *vq)
142 struct zcopy_mbuf *zmbuf, *next;
144 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
145 zmbuf != NULL; zmbuf = next) {
146 next = TAILQ_NEXT(zmbuf, next);
148 while (!mbuf_is_consumed(zmbuf->mbuf))
151 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
152 restore_mbuf(zmbuf->mbuf);
153 rte_pktmbuf_free(zmbuf->mbuf);
160 free_mem_region(struct virtio_net *dev)
163 struct rte_vhost_mem_region *reg;
164 struct vhost_virtqueue *vq;
166 if (!dev || !dev->mem)
169 if (dev->dequeue_zero_copy) {
170 for (i = 0; i < dev->nr_vring; i++) {
171 vq = dev->virtqueue[i];
173 drain_zmbuf_list(vq);
177 for (i = 0; i < dev->mem->nregions; i++) {
178 reg = &dev->mem->regions[i];
179 if (reg->host_user_addr) {
180 munmap(reg->mmap_addr, reg->mmap_size);
187 vhost_backend_cleanup(struct virtio_net *dev)
190 free_mem_region(dev);
195 rte_free(dev->guest_pages);
196 dev->guest_pages = NULL;
199 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
203 if (dev->inflight_info) {
204 if (dev->inflight_info->addr) {
205 munmap(dev->inflight_info->addr,
206 dev->inflight_info->size);
207 dev->inflight_info->addr = NULL;
210 if (dev->inflight_info->fd >= 0) {
211 close(dev->inflight_info->fd);
212 dev->inflight_info->fd = -1;
215 free(dev->inflight_info);
216 dev->inflight_info = NULL;
219 if (dev->slave_req_fd >= 0) {
220 close(dev->slave_req_fd);
221 dev->slave_req_fd = -1;
224 if (dev->postcopy_ufd >= 0) {
225 close(dev->postcopy_ufd);
226 dev->postcopy_ufd = -1;
229 dev->postcopy_listening = 0;
233 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
236 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
238 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
239 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
241 if (dev->notify_ops->vring_state_changed)
242 dev->notify_ops->vring_state_changed(dev->vid,
247 * This function just returns success at the moment unless
248 * the device hasn't been initialised.
251 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
252 struct VhostUserMsg *msg,
253 int main_fd __rte_unused)
255 if (validate_msg_fds(msg, 0) != 0)
256 return RTE_VHOST_MSG_RESULT_ERR;
258 return RTE_VHOST_MSG_RESULT_OK;
262 vhost_user_reset_owner(struct virtio_net **pdev,
263 struct VhostUserMsg *msg,
264 int main_fd __rte_unused)
266 struct virtio_net *dev = *pdev;
268 if (validate_msg_fds(msg, 0) != 0)
269 return RTE_VHOST_MSG_RESULT_ERR;
271 vhost_destroy_device_notify(dev);
273 cleanup_device(dev, 0);
275 return RTE_VHOST_MSG_RESULT_OK;
279 * The features that we support are requested.
282 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
283 int main_fd __rte_unused)
285 struct virtio_net *dev = *pdev;
286 uint64_t features = 0;
288 if (validate_msg_fds(msg, 0) != 0)
289 return RTE_VHOST_MSG_RESULT_ERR;
291 rte_vhost_driver_get_features(dev->ifname, &features);
293 msg->payload.u64 = features;
294 msg->size = sizeof(msg->payload.u64);
297 return RTE_VHOST_MSG_RESULT_REPLY;
301 * The queue number that we support are requested.
304 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
305 int main_fd __rte_unused)
307 struct virtio_net *dev = *pdev;
308 uint32_t queue_num = 0;
310 if (validate_msg_fds(msg, 0) != 0)
311 return RTE_VHOST_MSG_RESULT_ERR;
313 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
315 msg->payload.u64 = (uint64_t)queue_num;
316 msg->size = sizeof(msg->payload.u64);
319 return RTE_VHOST_MSG_RESULT_REPLY;
323 * We receive the negotiated features supported by us and the virtio device.
326 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
327 int main_fd __rte_unused)
329 struct virtio_net *dev = *pdev;
330 uint64_t features = msg->payload.u64;
331 uint64_t vhost_features = 0;
332 struct rte_vdpa_device *vdpa_dev;
334 if (validate_msg_fds(msg, 0) != 0)
335 return RTE_VHOST_MSG_RESULT_ERR;
337 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
338 if (features & ~vhost_features) {
339 VHOST_LOG_CONFIG(ERR,
340 "(%d) received invalid negotiated features.\n",
342 return RTE_VHOST_MSG_RESULT_ERR;
345 if (dev->flags & VIRTIO_DEV_RUNNING) {
346 if (dev->features == features)
347 return RTE_VHOST_MSG_RESULT_OK;
350 * Error out if master tries to change features while device is
351 * in running state. The exception being VHOST_F_LOG_ALL, which
352 * is enabled when the live-migration starts.
354 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
355 VHOST_LOG_CONFIG(ERR,
356 "(%d) features changed while device is running.\n",
358 return RTE_VHOST_MSG_RESULT_ERR;
361 if (dev->notify_ops->features_changed)
362 dev->notify_ops->features_changed(dev->vid, features);
365 dev->features = features;
367 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
368 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
370 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
372 VHOST_LOG_CONFIG(INFO,
373 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
374 VHOST_LOG_CONFIG(DEBUG,
375 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
377 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
378 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
380 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
381 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
383 * Remove all but first queue pair if MQ hasn't been
384 * negotiated. This is safe because the device is not
385 * running at this stage.
387 while (dev->nr_vring > 2) {
388 struct vhost_virtqueue *vq;
390 vq = dev->virtqueue[--dev->nr_vring];
394 dev->virtqueue[dev->nr_vring] = NULL;
396 cleanup_vq_inflight(dev, vq);
401 vdpa_dev = dev->vdpa_dev;
403 vdpa_dev->ops->set_features(dev->vid);
405 return RTE_VHOST_MSG_RESULT_OK;
409 * The virtio device sends us the size of the descriptor ring.
412 vhost_user_set_vring_num(struct virtio_net **pdev,
413 struct VhostUserMsg *msg,
414 int main_fd __rte_unused)
416 struct virtio_net *dev = *pdev;
417 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
419 if (validate_msg_fds(msg, 0) != 0)
420 return RTE_VHOST_MSG_RESULT_ERR;
422 vq->size = msg->payload.state.num;
424 /* VIRTIO 1.0, 2.4 Virtqueues says:
426 * Queue Size value is always a power of 2. The maximum Queue Size
429 * VIRTIO 1.1 2.7 Virtqueues says:
431 * Packed virtqueues support up to 2^15 entries each.
433 if (!vq_is_packed(dev)) {
434 if (vq->size & (vq->size - 1)) {
435 VHOST_LOG_CONFIG(ERR,
436 "invalid virtqueue size %u\n", vq->size);
437 return RTE_VHOST_MSG_RESULT_ERR;
441 if (vq->size > 32768) {
442 VHOST_LOG_CONFIG(ERR,
443 "invalid virtqueue size %u\n", vq->size);
444 return RTE_VHOST_MSG_RESULT_ERR;
447 if (dev->dequeue_zero_copy) {
449 vq->last_zmbuf_idx = 0;
450 vq->zmbuf_size = vq->size;
452 rte_free(vq->zmbufs);
453 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
454 sizeof(struct zcopy_mbuf), 0);
455 if (vq->zmbufs == NULL) {
456 VHOST_LOG_CONFIG(WARNING,
457 "failed to allocate mem for zero copy; "
458 "zero copy is force disabled\n");
459 dev->dequeue_zero_copy = 0;
461 TAILQ_INIT(&vq->zmbuf_list);
464 if (vq_is_packed(dev)) {
465 if (vq->shadow_used_packed)
466 rte_free(vq->shadow_used_packed);
467 vq->shadow_used_packed = rte_malloc(NULL,
469 sizeof(struct vring_used_elem_packed),
470 RTE_CACHE_LINE_SIZE);
471 if (!vq->shadow_used_packed) {
472 VHOST_LOG_CONFIG(ERR,
473 "failed to allocate memory for shadow used ring.\n");
474 return RTE_VHOST_MSG_RESULT_ERR;
478 if (vq->shadow_used_split)
479 rte_free(vq->shadow_used_split);
481 vq->shadow_used_split = rte_malloc(NULL,
482 vq->size * sizeof(struct vring_used_elem),
483 RTE_CACHE_LINE_SIZE);
485 if (!vq->shadow_used_split) {
486 VHOST_LOG_CONFIG(ERR,
487 "failed to allocate memory for vq internal data.\n");
488 return RTE_VHOST_MSG_RESULT_ERR;
492 if (vq->batch_copy_elems)
493 rte_free(vq->batch_copy_elems);
494 vq->batch_copy_elems = rte_malloc(NULL,
495 vq->size * sizeof(struct batch_copy_elem),
496 RTE_CACHE_LINE_SIZE);
497 if (!vq->batch_copy_elems) {
498 VHOST_LOG_CONFIG(ERR,
499 "failed to allocate memory for batching copy.\n");
500 return RTE_VHOST_MSG_RESULT_ERR;
503 return RTE_VHOST_MSG_RESULT_OK;
507 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
508 * same numa node as the memory of vring descriptor.
510 #ifdef RTE_LIBRTE_VHOST_NUMA
511 static struct virtio_net*
512 numa_realloc(struct virtio_net *dev, int index)
514 int oldnode, newnode;
515 struct virtio_net *old_dev;
516 struct vhost_virtqueue *old_vq, *vq;
517 struct zcopy_mbuf *new_zmbuf;
518 struct vring_used_elem *new_shadow_used_split;
519 struct vring_used_elem_packed *new_shadow_used_packed;
520 struct batch_copy_elem *new_batch_copy_elems;
523 if (dev->flags & VIRTIO_DEV_RUNNING)
527 vq = old_vq = dev->virtqueue[index];
529 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
530 MPOL_F_NODE | MPOL_F_ADDR);
532 /* check if we need to reallocate vq */
533 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
534 MPOL_F_NODE | MPOL_F_ADDR);
536 VHOST_LOG_CONFIG(ERR,
537 "Unable to get vq numa information.\n");
540 if (oldnode != newnode) {
541 VHOST_LOG_CONFIG(INFO,
542 "reallocate vq from %d to %d node\n", oldnode, newnode);
543 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
547 memcpy(vq, old_vq, sizeof(*vq));
548 TAILQ_INIT(&vq->zmbuf_list);
550 if (dev->dequeue_zero_copy) {
551 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
552 sizeof(struct zcopy_mbuf), 0, newnode);
554 rte_free(vq->zmbufs);
555 vq->zmbufs = new_zmbuf;
559 if (vq_is_packed(dev)) {
560 new_shadow_used_packed = rte_malloc_socket(NULL,
562 sizeof(struct vring_used_elem_packed),
565 if (new_shadow_used_packed) {
566 rte_free(vq->shadow_used_packed);
567 vq->shadow_used_packed = new_shadow_used_packed;
570 new_shadow_used_split = rte_malloc_socket(NULL,
572 sizeof(struct vring_used_elem),
575 if (new_shadow_used_split) {
576 rte_free(vq->shadow_used_split);
577 vq->shadow_used_split = new_shadow_used_split;
581 new_batch_copy_elems = rte_malloc_socket(NULL,
582 vq->size * sizeof(struct batch_copy_elem),
585 if (new_batch_copy_elems) {
586 rte_free(vq->batch_copy_elems);
587 vq->batch_copy_elems = new_batch_copy_elems;
593 /* check if we need to reallocate dev */
594 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
595 MPOL_F_NODE | MPOL_F_ADDR);
597 VHOST_LOG_CONFIG(ERR,
598 "Unable to get dev numa information.\n");
601 if (oldnode != newnode) {
602 VHOST_LOG_CONFIG(INFO,
603 "reallocate dev from %d to %d node\n",
605 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
611 memcpy(dev, old_dev, sizeof(*dev));
616 dev->virtqueue[index] = vq;
617 vhost_devices[dev->vid] = dev;
620 vhost_user_iotlb_init(dev, index);
625 static struct virtio_net*
626 numa_realloc(struct virtio_net *dev, int index __rte_unused)
632 /* Converts QEMU virtual address to Vhost virtual address. */
634 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
636 struct rte_vhost_mem_region *r;
639 if (unlikely(!dev || !dev->mem))
642 /* Find the region where the address lives. */
643 for (i = 0; i < dev->mem->nregions; i++) {
644 r = &dev->mem->regions[i];
646 if (qva >= r->guest_user_addr &&
647 qva < r->guest_user_addr + r->size) {
649 if (unlikely(*len > r->guest_user_addr + r->size - qva))
650 *len = r->guest_user_addr + r->size - qva;
652 return qva - r->guest_user_addr +
664 * Converts ring address to Vhost virtual address.
665 * If IOMMU is enabled, the ring address is a guest IO virtual address,
666 * else it is a QEMU virtual address.
669 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
670 uint64_t ra, uint64_t *size)
672 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
675 vhost_user_iotlb_rd_lock(vq);
676 vva = vhost_iova_to_vva(dev, vq, ra,
677 size, VHOST_ACCESS_RW);
678 vhost_user_iotlb_rd_unlock(vq);
683 return qva_to_vva(dev, ra, size);
687 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
691 vhost_user_iotlb_rd_lock(vq);
692 log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
693 vhost_user_iotlb_rd_unlock(vq);
698 static struct virtio_net *
699 translate_ring_addresses(struct virtio_net *dev, int vq_index)
701 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
702 struct vhost_vring_addr *addr = &vq->ring_addrs;
703 uint64_t len, expected_len;
705 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
707 log_addr_to_gpa(dev, vq);
708 if (vq->log_guest_addr == 0) {
709 VHOST_LOG_CONFIG(DEBUG,
710 "(%d) failed to map log_guest_addr.\n",
716 if (vq_is_packed(dev)) {
717 len = sizeof(struct vring_packed_desc) * vq->size;
718 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
719 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
720 if (vq->desc_packed == NULL ||
721 len != sizeof(struct vring_packed_desc) *
723 VHOST_LOG_CONFIG(DEBUG,
724 "(%d) failed to map desc_packed ring.\n",
729 dev = numa_realloc(dev, vq_index);
730 vq = dev->virtqueue[vq_index];
731 addr = &vq->ring_addrs;
733 len = sizeof(struct vring_packed_desc_event);
734 vq->driver_event = (struct vring_packed_desc_event *)
735 (uintptr_t)ring_addr_to_vva(dev,
736 vq, addr->avail_user_addr, &len);
737 if (vq->driver_event == NULL ||
738 len != sizeof(struct vring_packed_desc_event)) {
739 VHOST_LOG_CONFIG(DEBUG,
740 "(%d) failed to find driver area address.\n",
745 len = sizeof(struct vring_packed_desc_event);
746 vq->device_event = (struct vring_packed_desc_event *)
747 (uintptr_t)ring_addr_to_vva(dev,
748 vq, addr->used_user_addr, &len);
749 if (vq->device_event == NULL ||
750 len != sizeof(struct vring_packed_desc_event)) {
751 VHOST_LOG_CONFIG(DEBUG,
752 "(%d) failed to find device area address.\n",
761 /* The addresses are converted from QEMU virtual to Vhost virtual. */
762 if (vq->desc && vq->avail && vq->used)
765 len = sizeof(struct vring_desc) * vq->size;
766 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
767 vq, addr->desc_user_addr, &len);
768 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
769 VHOST_LOG_CONFIG(DEBUG,
770 "(%d) failed to map desc ring.\n",
775 dev = numa_realloc(dev, vq_index);
776 vq = dev->virtqueue[vq_index];
777 addr = &vq->ring_addrs;
779 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
780 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
781 len += sizeof(uint16_t);
783 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
784 vq, addr->avail_user_addr, &len);
785 if (vq->avail == 0 || len != expected_len) {
786 VHOST_LOG_CONFIG(DEBUG,
787 "(%d) failed to map avail ring.\n",
792 len = sizeof(struct vring_used) +
793 sizeof(struct vring_used_elem) * vq->size;
794 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
795 len += sizeof(uint16_t);
797 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
798 vq, addr->used_user_addr, &len);
799 if (vq->used == 0 || len != expected_len) {
800 VHOST_LOG_CONFIG(DEBUG,
801 "(%d) failed to map used ring.\n",
806 if (vq->last_used_idx != vq->used->idx) {
807 VHOST_LOG_CONFIG(WARNING,
808 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
809 "some packets maybe resent for Tx and dropped for Rx\n",
810 vq->last_used_idx, vq->used->idx);
811 vq->last_used_idx = vq->used->idx;
812 vq->last_avail_idx = vq->used->idx;
817 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
819 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
820 dev->vid, vq->avail);
821 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
823 VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
824 dev->vid, vq->log_guest_addr);
830 * The virtio device sends us the desc, used and avail ring addresses.
831 * This function then converts these to our address space.
834 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
835 int main_fd __rte_unused)
837 struct virtio_net *dev = *pdev;
838 struct vhost_virtqueue *vq;
839 struct vhost_vring_addr *addr = &msg->payload.addr;
842 if (validate_msg_fds(msg, 0) != 0)
843 return RTE_VHOST_MSG_RESULT_ERR;
845 if (dev->mem == NULL)
846 return RTE_VHOST_MSG_RESULT_ERR;
848 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
849 vq = dev->virtqueue[msg->payload.addr.index];
851 access_ok = vq->access_ok;
854 * Rings addresses should not be interpreted as long as the ring is not
855 * started and enabled
857 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
859 vring_invalidate(dev, vq);
861 if ((vq->enabled && (dev->features &
862 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
864 dev = translate_ring_addresses(dev, msg->payload.addr.index);
866 return RTE_VHOST_MSG_RESULT_ERR;
871 return RTE_VHOST_MSG_RESULT_OK;
875 * The virtio device sends us the available ring last used index.
878 vhost_user_set_vring_base(struct virtio_net **pdev,
879 struct VhostUserMsg *msg,
880 int main_fd __rte_unused)
882 struct virtio_net *dev = *pdev;
883 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
884 uint64_t val = msg->payload.state.num;
886 if (validate_msg_fds(msg, 0) != 0)
887 return RTE_VHOST_MSG_RESULT_ERR;
889 if (vq_is_packed(dev)) {
891 * Bit[0:14]: avail index
892 * Bit[15]: avail wrap counter
894 vq->last_avail_idx = val & 0x7fff;
895 vq->avail_wrap_counter = !!(val & (0x1 << 15));
897 * Set used index to same value as available one, as
898 * their values should be the same since ring processing
899 * was stopped at get time.
901 vq->last_used_idx = vq->last_avail_idx;
902 vq->used_wrap_counter = vq->avail_wrap_counter;
904 vq->last_used_idx = msg->payload.state.num;
905 vq->last_avail_idx = msg->payload.state.num;
908 return RTE_VHOST_MSG_RESULT_OK;
912 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
913 uint64_t host_phys_addr, uint64_t size)
915 struct guest_page *page, *last_page;
916 struct guest_page *old_pages;
918 if (dev->nr_guest_pages == dev->max_guest_pages) {
919 dev->max_guest_pages *= 2;
920 old_pages = dev->guest_pages;
921 dev->guest_pages = rte_realloc(dev->guest_pages,
922 dev->max_guest_pages * sizeof(*page),
923 RTE_CACHE_LINE_SIZE);
924 if (dev->guest_pages == NULL) {
925 VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
931 if (dev->nr_guest_pages > 0) {
932 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
933 /* merge if the two pages are continuous */
934 if (host_phys_addr == last_page->host_phys_addr +
936 last_page->size += size;
941 page = &dev->guest_pages[dev->nr_guest_pages++];
942 page->guest_phys_addr = guest_phys_addr;
943 page->host_phys_addr = host_phys_addr;
950 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
953 uint64_t reg_size = reg->size;
954 uint64_t host_user_addr = reg->host_user_addr;
955 uint64_t guest_phys_addr = reg->guest_phys_addr;
956 uint64_t host_phys_addr;
959 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
960 size = page_size - (guest_phys_addr & (page_size - 1));
961 size = RTE_MIN(size, reg_size);
963 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
966 host_user_addr += size;
967 guest_phys_addr += size;
970 while (reg_size > 0) {
971 size = RTE_MIN(reg_size, page_size);
972 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
974 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
978 host_user_addr += size;
979 guest_phys_addr += size;
983 /* sort guest page array if over binary search threshold */
984 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
985 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
986 sizeof(struct guest_page), guest_page_addrcmp);
992 #ifdef RTE_LIBRTE_VHOST_DEBUG
993 /* TODO: enable it only in debug mode? */
995 dump_guest_pages(struct virtio_net *dev)
998 struct guest_page *page;
1000 for (i = 0; i < dev->nr_guest_pages; i++) {
1001 page = &dev->guest_pages[i];
1003 VHOST_LOG_CONFIG(INFO,
1004 "guest physical page region %u\n"
1005 "\t guest_phys_addr: %" PRIx64 "\n"
1006 "\t host_phys_addr : %" PRIx64 "\n"
1007 "\t size : %" PRIx64 "\n",
1009 page->guest_phys_addr,
1010 page->host_phys_addr,
1015 #define dump_guest_pages(dev)
1019 vhost_memory_changed(struct VhostUserMemory *new,
1020 struct rte_vhost_memory *old)
1024 if (new->nregions != old->nregions)
1027 for (i = 0; i < new->nregions; ++i) {
1028 VhostUserMemoryRegion *new_r = &new->regions[i];
1029 struct rte_vhost_mem_region *old_r = &old->regions[i];
1031 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1033 if (new_r->memory_size != old_r->size)
1035 if (new_r->userspace_addr != old_r->guest_user_addr)
1043 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1046 struct virtio_net *dev = *pdev;
1047 struct VhostUserMemory *memory = &msg->payload.memory;
1048 struct rte_vhost_mem_region *reg;
1051 uint64_t mmap_offset;
1057 if (validate_msg_fds(msg, memory->nregions) != 0)
1058 return RTE_VHOST_MSG_RESULT_ERR;
1060 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1061 VHOST_LOG_CONFIG(ERR,
1062 "too many memory regions (%u)\n", memory->nregions);
1063 return RTE_VHOST_MSG_RESULT_ERR;
1066 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1067 VHOST_LOG_CONFIG(INFO,
1068 "(%d) memory regions not changed\n", dev->vid);
1072 return RTE_VHOST_MSG_RESULT_OK;
1076 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1077 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1079 if (vdpa_dev && vdpa_dev->ops->dev_close)
1080 vdpa_dev->ops->dev_close(dev->vid);
1081 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1083 free_mem_region(dev);
1088 /* Flush IOTLB cache as previous HVAs are now invalid */
1089 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1090 for (i = 0; i < dev->nr_vring; i++)
1091 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1093 dev->nr_guest_pages = 0;
1094 if (dev->guest_pages == NULL) {
1095 dev->max_guest_pages = 8;
1096 dev->guest_pages = rte_zmalloc(NULL,
1097 dev->max_guest_pages *
1098 sizeof(struct guest_page),
1099 RTE_CACHE_LINE_SIZE);
1100 if (dev->guest_pages == NULL) {
1101 VHOST_LOG_CONFIG(ERR,
1102 "(%d) failed to allocate memory "
1103 "for dev->guest_pages\n",
1105 return RTE_VHOST_MSG_RESULT_ERR;
1109 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1110 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1111 if (dev->mem == NULL) {
1112 VHOST_LOG_CONFIG(ERR,
1113 "(%d) failed to allocate memory for dev->mem\n",
1115 return RTE_VHOST_MSG_RESULT_ERR;
1117 dev->mem->nregions = memory->nregions;
1119 for (i = 0; i < memory->nregions; i++) {
1121 reg = &dev->mem->regions[i];
1123 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1124 reg->guest_user_addr = memory->regions[i].userspace_addr;
1125 reg->size = memory->regions[i].memory_size;
1128 mmap_offset = memory->regions[i].mmap_offset;
1130 /* Check for memory_size + mmap_offset overflow */
1131 if (mmap_offset >= -reg->size) {
1132 VHOST_LOG_CONFIG(ERR,
1133 "mmap_offset (%#"PRIx64") and memory_size "
1134 "(%#"PRIx64") overflow\n",
1135 mmap_offset, reg->size);
1139 mmap_size = reg->size + mmap_offset;
1141 /* mmap() without flag of MAP_ANONYMOUS, should be called
1142 * with length argument aligned with hugepagesz at older
1143 * longterm version Linux, like 2.6.32 and 3.2.72, or
1144 * mmap() will fail with EINVAL.
1146 * to avoid failure, make sure in caller to keep length
1149 alignment = get_blk_size(fd);
1150 if (alignment == (uint64_t)-1) {
1151 VHOST_LOG_CONFIG(ERR,
1152 "couldn't get hugepage size through fstat\n");
1155 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1156 if (mmap_size == 0) {
1158 * It could happen if initial mmap_size + alignment
1159 * overflows the sizeof uint64, which could happen if
1160 * either mmap_size or alignment value is wrong.
1162 * mmap() kernel implementation would return an error,
1163 * but better catch it before and provide useful info
1166 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1167 "or alignment (0x%" PRIx64 ") is invalid\n",
1168 reg->size + mmap_offset, alignment);
1172 populate = (dev->dequeue_zero_copy || dev->async_copy) ?
1174 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1175 MAP_SHARED | populate, fd, 0);
1177 if (mmap_addr == MAP_FAILED) {
1178 VHOST_LOG_CONFIG(ERR,
1179 "mmap region %u failed.\n", i);
1183 reg->mmap_addr = mmap_addr;
1184 reg->mmap_size = mmap_size;
1185 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1188 if (dev->dequeue_zero_copy || dev->async_copy)
1189 if (add_guest_pages(dev, reg, alignment) < 0) {
1190 VHOST_LOG_CONFIG(ERR,
1191 "adding guest pages to region %u failed.\n",
1196 VHOST_LOG_CONFIG(INFO,
1197 "guest memory region %u, size: 0x%" PRIx64 "\n"
1198 "\t guest physical addr: 0x%" PRIx64 "\n"
1199 "\t guest virtual addr: 0x%" PRIx64 "\n"
1200 "\t host virtual addr: 0x%" PRIx64 "\n"
1201 "\t mmap addr : 0x%" PRIx64 "\n"
1202 "\t mmap size : 0x%" PRIx64 "\n"
1203 "\t mmap align: 0x%" PRIx64 "\n"
1204 "\t mmap off : 0x%" PRIx64 "\n",
1206 reg->guest_phys_addr,
1207 reg->guest_user_addr,
1208 reg->host_user_addr,
1209 (uint64_t)(uintptr_t)mmap_addr,
1214 if (dev->postcopy_listening) {
1216 * We haven't a better way right now than sharing
1217 * DPDK's virtual address with Qemu, so that Qemu can
1218 * retrieve the region offset when handling userfaults.
1220 memory->regions[i].userspace_addr =
1221 reg->host_user_addr;
1224 if (dev->postcopy_listening) {
1225 /* Send the addresses back to qemu */
1227 send_vhost_reply(main_fd, msg);
1229 /* Wait for qemu to acknolwedge it's got the addresses
1230 * we've got to wait before we're allowed to generate faults.
1232 VhostUserMsg ack_msg;
1233 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1234 VHOST_LOG_CONFIG(ERR,
1235 "Failed to read qemu ack on postcopy set-mem-table\n");
1239 if (validate_msg_fds(&ack_msg, 0) != 0)
1242 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1243 VHOST_LOG_CONFIG(ERR,
1244 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1245 ack_msg.request.master);
1249 /* Now userfault register and we can use the memory */
1250 for (i = 0; i < memory->nregions; i++) {
1251 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1252 reg = &dev->mem->regions[i];
1253 struct uffdio_register reg_struct;
1256 * Let's register all the mmap'ed area to ensure
1257 * alignment on page boundary.
1259 reg_struct.range.start =
1260 (uint64_t)(uintptr_t)reg->mmap_addr;
1261 reg_struct.range.len = reg->mmap_size;
1262 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1264 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1266 VHOST_LOG_CONFIG(ERR,
1267 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1268 i, dev->postcopy_ufd,
1272 VHOST_LOG_CONFIG(INFO,
1273 "\t userfaultfd registered for range : "
1274 "%" PRIx64 " - %" PRIx64 "\n",
1275 (uint64_t)reg_struct.range.start,
1276 (uint64_t)reg_struct.range.start +
1277 (uint64_t)reg_struct.range.len - 1);
1284 for (i = 0; i < dev->nr_vring; i++) {
1285 struct vhost_virtqueue *vq = dev->virtqueue[i];
1287 if (vq->desc || vq->avail || vq->used) {
1289 * If the memory table got updated, the ring addresses
1290 * need to be translated again as virtual addresses have
1293 vring_invalidate(dev, vq);
1295 dev = translate_ring_addresses(dev, i);
1305 dump_guest_pages(dev);
1307 return RTE_VHOST_MSG_RESULT_OK;
1310 free_mem_region(dev);
1313 return RTE_VHOST_MSG_RESULT_ERR;
1317 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1324 if (vq_is_packed(dev))
1325 rings_ok = vq->desc_packed && vq->driver_event &&
1328 rings_ok = vq->desc && vq->avail && vq->used;
1331 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1332 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1336 #define VIRTIO_DEV_NUM_VQS_TO_BE_READY 2u
1339 virtio_is_ready(struct virtio_net *dev)
1341 struct vhost_virtqueue *vq;
1344 if (dev->flags & VIRTIO_DEV_READY)
1347 if (dev->nr_vring < VIRTIO_DEV_NUM_VQS_TO_BE_READY)
1350 for (i = 0; i < VIRTIO_DEV_NUM_VQS_TO_BE_READY; i++) {
1351 vq = dev->virtqueue[i];
1353 if (!vq_is_ready(dev, vq))
1357 /* If supported, ensure the frontend is really done with config */
1358 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1359 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1362 dev->flags |= VIRTIO_DEV_READY;
1364 if (!(dev->flags & VIRTIO_DEV_RUNNING))
1365 VHOST_LOG_CONFIG(INFO,
1366 "virtio is now ready for processing.\n");
1371 inflight_mem_alloc(const char *name, size_t size, int *fd)
1375 char fname[20] = "/tmp/memfd-XXXXXX";
1378 #ifdef MEMFD_SUPPORTED
1379 mfd = memfd_create(name, MFD_CLOEXEC);
1384 mfd = mkstemp(fname);
1386 VHOST_LOG_CONFIG(ERR,
1387 "failed to get inflight buffer fd\n");
1394 if (ftruncate(mfd, size) == -1) {
1395 VHOST_LOG_CONFIG(ERR,
1396 "failed to alloc inflight buffer\n");
1401 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1402 if (ptr == MAP_FAILED) {
1403 VHOST_LOG_CONFIG(ERR,
1404 "failed to mmap inflight buffer\n");
1414 get_pervq_shm_size_split(uint16_t queue_size)
1416 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1417 queue_size + sizeof(uint64_t) +
1418 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1422 get_pervq_shm_size_packed(uint16_t queue_size)
1424 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1425 * queue_size + sizeof(uint64_t) +
1426 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1427 INFLIGHT_ALIGNMENT);
1431 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1433 int main_fd __rte_unused)
1435 struct rte_vhost_inflight_info_packed *inflight_packed;
1436 uint64_t pervq_inflight_size, mmap_size;
1437 uint16_t num_queues, queue_size;
1438 struct virtio_net *dev = *pdev;
1442 if (msg->size != sizeof(msg->payload.inflight)) {
1443 VHOST_LOG_CONFIG(ERR,
1444 "invalid get_inflight_fd message size is %d\n",
1446 return RTE_VHOST_MSG_RESULT_ERR;
1449 if (dev->inflight_info == NULL) {
1450 dev->inflight_info = calloc(1,
1451 sizeof(struct inflight_mem_info));
1452 if (!dev->inflight_info) {
1453 VHOST_LOG_CONFIG(ERR,
1454 "failed to alloc dev inflight area\n");
1455 return RTE_VHOST_MSG_RESULT_ERR;
1457 dev->inflight_info->fd = -1;
1460 num_queues = msg->payload.inflight.num_queues;
1461 queue_size = msg->payload.inflight.queue_size;
1463 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1464 msg->payload.inflight.num_queues);
1465 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1466 msg->payload.inflight.queue_size);
1468 if (vq_is_packed(dev))
1469 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1471 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1473 mmap_size = num_queues * pervq_inflight_size;
1474 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1476 VHOST_LOG_CONFIG(ERR,
1477 "failed to alloc vhost inflight area\n");
1478 msg->payload.inflight.mmap_size = 0;
1479 return RTE_VHOST_MSG_RESULT_ERR;
1481 memset(addr, 0, mmap_size);
1483 if (dev->inflight_info->addr) {
1484 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1485 dev->inflight_info->addr = NULL;
1488 if (dev->inflight_info->fd >= 0) {
1489 close(dev->inflight_info->fd);
1490 dev->inflight_info->fd = -1;
1493 dev->inflight_info->addr = addr;
1494 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1495 dev->inflight_info->fd = msg->fds[0] = fd;
1496 msg->payload.inflight.mmap_offset = 0;
1499 if (vq_is_packed(dev)) {
1500 for (i = 0; i < num_queues; i++) {
1502 (struct rte_vhost_inflight_info_packed *)addr;
1503 inflight_packed->used_wrap_counter = 1;
1504 inflight_packed->old_used_wrap_counter = 1;
1505 for (j = 0; j < queue_size; j++)
1506 inflight_packed->desc[j].next = j + 1;
1507 addr = (void *)((char *)addr + pervq_inflight_size);
1511 VHOST_LOG_CONFIG(INFO,
1512 "send inflight mmap_size: %"PRIu64"\n",
1513 msg->payload.inflight.mmap_size);
1514 VHOST_LOG_CONFIG(INFO,
1515 "send inflight mmap_offset: %"PRIu64"\n",
1516 msg->payload.inflight.mmap_offset);
1517 VHOST_LOG_CONFIG(INFO,
1518 "send inflight fd: %d\n", msg->fds[0]);
1520 return RTE_VHOST_MSG_RESULT_REPLY;
1524 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1525 int main_fd __rte_unused)
1527 uint64_t mmap_size, mmap_offset;
1528 uint16_t num_queues, queue_size;
1529 struct virtio_net *dev = *pdev;
1530 uint32_t pervq_inflight_size;
1531 struct vhost_virtqueue *vq;
1536 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1537 VHOST_LOG_CONFIG(ERR,
1538 "invalid set_inflight_fd message size is %d,fd is %d\n",
1540 return RTE_VHOST_MSG_RESULT_ERR;
1543 mmap_size = msg->payload.inflight.mmap_size;
1544 mmap_offset = msg->payload.inflight.mmap_offset;
1545 num_queues = msg->payload.inflight.num_queues;
1546 queue_size = msg->payload.inflight.queue_size;
1548 if (vq_is_packed(dev))
1549 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1551 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1553 VHOST_LOG_CONFIG(INFO,
1554 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1555 VHOST_LOG_CONFIG(INFO,
1556 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1557 VHOST_LOG_CONFIG(INFO,
1558 "set_inflight_fd num_queues: %u\n", num_queues);
1559 VHOST_LOG_CONFIG(INFO,
1560 "set_inflight_fd queue_size: %u\n", queue_size);
1561 VHOST_LOG_CONFIG(INFO,
1562 "set_inflight_fd fd: %d\n", fd);
1563 VHOST_LOG_CONFIG(INFO,
1564 "set_inflight_fd pervq_inflight_size: %d\n",
1565 pervq_inflight_size);
1567 if (!dev->inflight_info) {
1568 dev->inflight_info = calloc(1,
1569 sizeof(struct inflight_mem_info));
1570 if (dev->inflight_info == NULL) {
1571 VHOST_LOG_CONFIG(ERR,
1572 "failed to alloc dev inflight area\n");
1573 return RTE_VHOST_MSG_RESULT_ERR;
1575 dev->inflight_info->fd = -1;
1578 if (dev->inflight_info->addr) {
1579 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1580 dev->inflight_info->addr = NULL;
1583 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1585 if (addr == MAP_FAILED) {
1586 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1587 return RTE_VHOST_MSG_RESULT_ERR;
1590 if (dev->inflight_info->fd >= 0) {
1591 close(dev->inflight_info->fd);
1592 dev->inflight_info->fd = -1;
1595 dev->inflight_info->fd = fd;
1596 dev->inflight_info->addr = addr;
1597 dev->inflight_info->size = mmap_size;
1599 for (i = 0; i < num_queues; i++) {
1600 vq = dev->virtqueue[i];
1601 if (vq_is_packed(dev)) {
1602 vq->inflight_packed = addr;
1603 vq->inflight_packed->desc_num = queue_size;
1605 vq->inflight_split = addr;
1606 vq->inflight_split->desc_num = queue_size;
1608 addr = (void *)((char *)addr + pervq_inflight_size);
1611 return RTE_VHOST_MSG_RESULT_OK;
1615 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1616 int main_fd __rte_unused)
1618 struct virtio_net *dev = *pdev;
1619 struct vhost_vring_file file;
1620 struct vhost_virtqueue *vq;
1623 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1624 if (validate_msg_fds(msg, expected_fds) != 0)
1625 return RTE_VHOST_MSG_RESULT_ERR;
1627 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1628 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1629 file.fd = VIRTIO_INVALID_EVENTFD;
1631 file.fd = msg->fds[0];
1632 VHOST_LOG_CONFIG(INFO,
1633 "vring call idx:%d file:%d\n", file.index, file.fd);
1635 vq = dev->virtqueue[file.index];
1638 vhost_user_notify_queue_state(dev, file.index, 0);
1642 if (vq->callfd >= 0)
1645 vq->callfd = file.fd;
1647 return RTE_VHOST_MSG_RESULT_OK;
1650 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1651 struct VhostUserMsg *msg,
1652 int main_fd __rte_unused)
1656 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1657 if (validate_msg_fds(msg, expected_fds) != 0)
1658 return RTE_VHOST_MSG_RESULT_ERR;
1660 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1662 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1664 return RTE_VHOST_MSG_RESULT_OK;
1668 resubmit_desc_compare(const void *a, const void *b)
1670 const struct rte_vhost_resubmit_desc *desc0 = a;
1671 const struct rte_vhost_resubmit_desc *desc1 = b;
1673 if (desc1->counter > desc0->counter)
1680 vhost_check_queue_inflights_split(struct virtio_net *dev,
1681 struct vhost_virtqueue *vq)
1684 uint16_t resubmit_num = 0, last_io, num;
1685 struct vring_used *used = vq->used;
1686 struct rte_vhost_resubmit_info *resubmit;
1687 struct rte_vhost_inflight_info_split *inflight_split;
1689 if (!(dev->protocol_features &
1690 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1691 return RTE_VHOST_MSG_RESULT_OK;
1693 /* The frontend may still not support the inflight feature
1694 * although we negotiate the protocol feature.
1696 if ((!vq->inflight_split))
1697 return RTE_VHOST_MSG_RESULT_OK;
1699 if (!vq->inflight_split->version) {
1700 vq->inflight_split->version = INFLIGHT_VERSION;
1701 return RTE_VHOST_MSG_RESULT_OK;
1704 if (vq->resubmit_inflight)
1705 return RTE_VHOST_MSG_RESULT_OK;
1707 inflight_split = vq->inflight_split;
1708 vq->global_counter = 0;
1709 last_io = inflight_split->last_inflight_io;
1711 if (inflight_split->used_idx != used->idx) {
1712 inflight_split->desc[last_io].inflight = 0;
1714 inflight_split->used_idx = used->idx;
1717 for (i = 0; i < inflight_split->desc_num; i++) {
1718 if (inflight_split->desc[i].inflight == 1)
1722 vq->last_avail_idx += resubmit_num;
1725 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1727 VHOST_LOG_CONFIG(ERR,
1728 "failed to allocate memory for resubmit info.\n");
1729 return RTE_VHOST_MSG_RESULT_ERR;
1732 resubmit->resubmit_list = calloc(resubmit_num,
1733 sizeof(struct rte_vhost_resubmit_desc));
1734 if (!resubmit->resubmit_list) {
1735 VHOST_LOG_CONFIG(ERR,
1736 "failed to allocate memory for inflight desc.\n");
1738 return RTE_VHOST_MSG_RESULT_ERR;
1742 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1743 if (vq->inflight_split->desc[i].inflight == 1) {
1744 resubmit->resubmit_list[num].index = i;
1745 resubmit->resubmit_list[num].counter =
1746 inflight_split->desc[i].counter;
1750 resubmit->resubmit_num = num;
1752 if (resubmit->resubmit_num > 1)
1753 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1754 sizeof(struct rte_vhost_resubmit_desc),
1755 resubmit_desc_compare);
1757 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1758 vq->resubmit_inflight = resubmit;
1761 return RTE_VHOST_MSG_RESULT_OK;
1765 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1766 struct vhost_virtqueue *vq)
1769 uint16_t resubmit_num = 0, old_used_idx, num;
1770 struct rte_vhost_resubmit_info *resubmit;
1771 struct rte_vhost_inflight_info_packed *inflight_packed;
1773 if (!(dev->protocol_features &
1774 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1775 return RTE_VHOST_MSG_RESULT_OK;
1777 /* The frontend may still not support the inflight feature
1778 * although we negotiate the protocol feature.
1780 if ((!vq->inflight_packed))
1781 return RTE_VHOST_MSG_RESULT_OK;
1783 if (!vq->inflight_packed->version) {
1784 vq->inflight_packed->version = INFLIGHT_VERSION;
1785 return RTE_VHOST_MSG_RESULT_OK;
1788 if (vq->resubmit_inflight)
1789 return RTE_VHOST_MSG_RESULT_OK;
1791 inflight_packed = vq->inflight_packed;
1792 vq->global_counter = 0;
1793 old_used_idx = inflight_packed->old_used_idx;
1795 if (inflight_packed->used_idx != old_used_idx) {
1796 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1797 inflight_packed->old_used_idx =
1798 inflight_packed->used_idx;
1799 inflight_packed->old_used_wrap_counter =
1800 inflight_packed->used_wrap_counter;
1801 inflight_packed->old_free_head =
1802 inflight_packed->free_head;
1804 inflight_packed->used_idx =
1805 inflight_packed->old_used_idx;
1806 inflight_packed->used_wrap_counter =
1807 inflight_packed->old_used_wrap_counter;
1808 inflight_packed->free_head =
1809 inflight_packed->old_free_head;
1813 for (i = 0; i < inflight_packed->desc_num; i++) {
1814 if (inflight_packed->desc[i].inflight == 1)
1819 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1820 if (resubmit == NULL) {
1821 VHOST_LOG_CONFIG(ERR,
1822 "failed to allocate memory for resubmit info.\n");
1823 return RTE_VHOST_MSG_RESULT_ERR;
1826 resubmit->resubmit_list = calloc(resubmit_num,
1827 sizeof(struct rte_vhost_resubmit_desc));
1828 if (resubmit->resubmit_list == NULL) {
1829 VHOST_LOG_CONFIG(ERR,
1830 "failed to allocate memory for resubmit desc.\n");
1832 return RTE_VHOST_MSG_RESULT_ERR;
1836 for (i = 0; i < inflight_packed->desc_num; i++) {
1837 if (vq->inflight_packed->desc[i].inflight == 1) {
1838 resubmit->resubmit_list[num].index = i;
1839 resubmit->resubmit_list[num].counter =
1840 inflight_packed->desc[i].counter;
1844 resubmit->resubmit_num = num;
1846 if (resubmit->resubmit_num > 1)
1847 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1848 sizeof(struct rte_vhost_resubmit_desc),
1849 resubmit_desc_compare);
1851 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1852 vq->resubmit_inflight = resubmit;
1855 return RTE_VHOST_MSG_RESULT_OK;
1859 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1860 int main_fd __rte_unused)
1862 struct virtio_net *dev = *pdev;
1863 struct vhost_vring_file file;
1864 struct vhost_virtqueue *vq;
1867 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1868 if (validate_msg_fds(msg, expected_fds) != 0)
1869 return RTE_VHOST_MSG_RESULT_ERR;
1871 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1872 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1873 file.fd = VIRTIO_INVALID_EVENTFD;
1875 file.fd = msg->fds[0];
1876 VHOST_LOG_CONFIG(INFO,
1877 "vring kick idx:%d file:%d\n", file.index, file.fd);
1879 /* Interpret ring addresses only when ring is started. */
1880 dev = translate_ring_addresses(dev, file.index);
1882 return RTE_VHOST_MSG_RESULT_ERR;
1886 vq = dev->virtqueue[file.index];
1889 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1890 * the ring starts already enabled. Otherwise, it is enabled via
1891 * the SET_VRING_ENABLE message.
1893 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1895 if (dev->notify_ops->vring_state_changed)
1896 dev->notify_ops->vring_state_changed(
1897 dev->vid, file.index, 1);
1901 vhost_user_notify_queue_state(dev, file.index, 0);
1905 if (vq->kickfd >= 0)
1907 vq->kickfd = file.fd;
1909 if (vq_is_packed(dev)) {
1910 if (vhost_check_queue_inflights_packed(dev, vq)) {
1911 VHOST_LOG_CONFIG(ERR,
1912 "failed to inflights for vq: %d\n", file.index);
1913 return RTE_VHOST_MSG_RESULT_ERR;
1916 if (vhost_check_queue_inflights_split(dev, vq)) {
1917 VHOST_LOG_CONFIG(ERR,
1918 "failed to inflights for vq: %d\n", file.index);
1919 return RTE_VHOST_MSG_RESULT_ERR;
1923 return RTE_VHOST_MSG_RESULT_OK;
1927 free_zmbufs(struct vhost_virtqueue *vq)
1929 drain_zmbuf_list(vq);
1931 rte_free(vq->zmbufs);
1935 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1938 vhost_user_get_vring_base(struct virtio_net **pdev,
1939 struct VhostUserMsg *msg,
1940 int main_fd __rte_unused)
1942 struct virtio_net *dev = *pdev;
1943 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1946 if (validate_msg_fds(msg, 0) != 0)
1947 return RTE_VHOST_MSG_RESULT_ERR;
1949 /* We have to stop the queue (virtio) if it is running. */
1950 vhost_destroy_device_notify(dev);
1952 dev->flags &= ~VIRTIO_DEV_READY;
1953 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1955 /* Here we are safe to get the indexes */
1956 if (vq_is_packed(dev)) {
1958 * Bit[0:14]: avail index
1959 * Bit[15]: avail wrap counter
1961 val = vq->last_avail_idx & 0x7fff;
1962 val |= vq->avail_wrap_counter << 15;
1963 msg->payload.state.num = val;
1965 msg->payload.state.num = vq->last_avail_idx;
1968 VHOST_LOG_CONFIG(INFO,
1969 "vring base idx:%d file:%d\n", msg->payload.state.index,
1970 msg->payload.state.num);
1972 * Based on current qemu vhost-user implementation, this message is
1973 * sent and only sent in vhost_vring_stop.
1974 * TODO: cleanup the vring, it isn't usable since here.
1976 if (vq->kickfd >= 0)
1979 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1981 if (vq->callfd >= 0)
1984 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1986 vq->signalled_used_valid = false;
1988 if (dev->dequeue_zero_copy)
1990 if (vq_is_packed(dev)) {
1991 rte_free(vq->shadow_used_packed);
1992 vq->shadow_used_packed = NULL;
1994 rte_free(vq->shadow_used_split);
1995 vq->shadow_used_split = NULL;
1996 if (vq->async_pkts_pending)
1997 rte_free(vq->async_pkts_pending);
1998 if (vq->async_pending_info)
1999 rte_free(vq->async_pending_info);
2000 vq->async_pkts_pending = NULL;
2001 vq->async_pending_info = NULL;
2004 rte_free(vq->batch_copy_elems);
2005 vq->batch_copy_elems = NULL;
2007 msg->size = sizeof(msg->payload.state);
2010 vring_invalidate(dev, vq);
2012 return RTE_VHOST_MSG_RESULT_REPLY;
2016 * when virtio queues are ready to work, qemu will send us to
2017 * enable the virtio queue pair.
2020 vhost_user_set_vring_enable(struct virtio_net **pdev,
2021 struct VhostUserMsg *msg,
2022 int main_fd __rte_unused)
2024 struct virtio_net *dev = *pdev;
2025 int enable = (int)msg->payload.state.num;
2026 int index = (int)msg->payload.state.index;
2028 if (validate_msg_fds(msg, 0) != 0)
2029 return RTE_VHOST_MSG_RESULT_ERR;
2031 VHOST_LOG_CONFIG(INFO,
2032 "set queue enable: %d to qp idx: %d\n",
2035 if (!enable && dev->virtqueue[index]->async_registered) {
2036 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2037 VHOST_LOG_CONFIG(ERR, "failed to disable vring. "
2038 "async inflight packets must be completed first\n");
2039 return RTE_VHOST_MSG_RESULT_ERR;
2043 /* On disable, rings have to be stopped being processed. */
2044 if (!enable && dev->dequeue_zero_copy)
2045 drain_zmbuf_list(dev->virtqueue[index]);
2047 dev->virtqueue[index]->enabled = enable;
2049 return RTE_VHOST_MSG_RESULT_OK;
2053 vhost_user_get_protocol_features(struct virtio_net **pdev,
2054 struct VhostUserMsg *msg,
2055 int main_fd __rte_unused)
2057 struct virtio_net *dev = *pdev;
2058 uint64_t features, protocol_features;
2060 if (validate_msg_fds(msg, 0) != 0)
2061 return RTE_VHOST_MSG_RESULT_ERR;
2063 rte_vhost_driver_get_features(dev->ifname, &features);
2064 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2066 msg->payload.u64 = protocol_features;
2067 msg->size = sizeof(msg->payload.u64);
2070 return RTE_VHOST_MSG_RESULT_REPLY;
2074 vhost_user_set_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 protocol_features = msg->payload.u64;
2080 uint64_t slave_protocol_features = 0;
2082 if (validate_msg_fds(msg, 0) != 0)
2083 return RTE_VHOST_MSG_RESULT_ERR;
2085 rte_vhost_driver_get_protocol_features(dev->ifname,
2086 &slave_protocol_features);
2087 if (protocol_features & ~slave_protocol_features) {
2088 VHOST_LOG_CONFIG(ERR,
2089 "(%d) received invalid protocol features.\n",
2091 return RTE_VHOST_MSG_RESULT_ERR;
2094 dev->protocol_features = protocol_features;
2095 VHOST_LOG_CONFIG(INFO,
2096 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2097 dev->protocol_features);
2099 return RTE_VHOST_MSG_RESULT_OK;
2103 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2104 int main_fd __rte_unused)
2106 struct virtio_net *dev = *pdev;
2107 int fd = msg->fds[0];
2111 if (validate_msg_fds(msg, 1) != 0)
2112 return RTE_VHOST_MSG_RESULT_ERR;
2115 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2116 return RTE_VHOST_MSG_RESULT_ERR;
2119 if (msg->size != sizeof(VhostUserLog)) {
2120 VHOST_LOG_CONFIG(ERR,
2121 "invalid log base msg size: %"PRId32" != %d\n",
2122 msg->size, (int)sizeof(VhostUserLog));
2123 return RTE_VHOST_MSG_RESULT_ERR;
2126 size = msg->payload.log.mmap_size;
2127 off = msg->payload.log.mmap_offset;
2129 /* Check for mmap size and offset overflow. */
2131 VHOST_LOG_CONFIG(ERR,
2132 "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2134 return RTE_VHOST_MSG_RESULT_ERR;
2137 VHOST_LOG_CONFIG(INFO,
2138 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2142 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2143 * fail when offset is not page size aligned.
2145 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2147 if (addr == MAP_FAILED) {
2148 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2149 return RTE_VHOST_MSG_RESULT_ERR;
2153 * Free previously mapped log memory on occasionally
2154 * multiple VHOST_USER_SET_LOG_BASE.
2156 if (dev->log_addr) {
2157 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2159 dev->log_addr = (uint64_t)(uintptr_t)addr;
2160 dev->log_base = dev->log_addr + off;
2161 dev->log_size = size;
2164 * The spec is not clear about it (yet), but QEMU doesn't expect
2165 * any payload in the reply.
2170 return RTE_VHOST_MSG_RESULT_REPLY;
2173 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2174 struct VhostUserMsg *msg,
2175 int main_fd __rte_unused)
2177 if (validate_msg_fds(msg, 1) != 0)
2178 return RTE_VHOST_MSG_RESULT_ERR;
2181 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2183 return RTE_VHOST_MSG_RESULT_OK;
2187 * An rarp packet is constructed and broadcasted to notify switches about
2188 * the new location of the migrated VM, so that packets from outside will
2189 * not be lost after migration.
2191 * However, we don't actually "send" a rarp packet here, instead, we set
2192 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2195 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2196 int main_fd __rte_unused)
2198 struct virtio_net *dev = *pdev;
2199 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2200 struct rte_vdpa_device *vdpa_dev;
2202 if (validate_msg_fds(msg, 0) != 0)
2203 return RTE_VHOST_MSG_RESULT_ERR;
2205 VHOST_LOG_CONFIG(DEBUG,
2206 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2207 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2208 memcpy(dev->mac.addr_bytes, mac, 6);
2211 * Set the flag to inject a RARP broadcast packet at
2212 * rte_vhost_dequeue_burst().
2214 * __ATOMIC_RELEASE ordering is for making sure the mac is
2215 * copied before the flag is set.
2217 __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2218 vdpa_dev = dev->vdpa_dev;
2219 if (vdpa_dev && vdpa_dev->ops->migration_done)
2220 vdpa_dev->ops->migration_done(dev->vid);
2222 return RTE_VHOST_MSG_RESULT_OK;
2226 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2227 int main_fd __rte_unused)
2229 struct virtio_net *dev = *pdev;
2231 if (validate_msg_fds(msg, 0) != 0)
2232 return RTE_VHOST_MSG_RESULT_ERR;
2234 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2235 msg->payload.u64 > VIRTIO_MAX_MTU) {
2236 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2239 return RTE_VHOST_MSG_RESULT_ERR;
2242 dev->mtu = msg->payload.u64;
2244 return RTE_VHOST_MSG_RESULT_OK;
2248 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2249 int main_fd __rte_unused)
2251 struct virtio_net *dev = *pdev;
2252 int fd = msg->fds[0];
2254 if (validate_msg_fds(msg, 1) != 0)
2255 return RTE_VHOST_MSG_RESULT_ERR;
2258 VHOST_LOG_CONFIG(ERR,
2259 "Invalid file descriptor for slave channel (%d)\n",
2261 return RTE_VHOST_MSG_RESULT_ERR;
2264 if (dev->slave_req_fd >= 0)
2265 close(dev->slave_req_fd);
2267 dev->slave_req_fd = fd;
2269 return RTE_VHOST_MSG_RESULT_OK;
2273 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2275 struct vhost_vring_addr *ra;
2276 uint64_t start, end, len;
2279 end = start + imsg->size;
2281 ra = &vq->ring_addrs;
2282 len = sizeof(struct vring_desc) * vq->size;
2283 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2286 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2287 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2290 len = sizeof(struct vring_used) +
2291 sizeof(struct vring_used_elem) * vq->size;
2292 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2295 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2296 len = sizeof(uint64_t);
2297 if (ra->log_guest_addr < end &&
2298 (ra->log_guest_addr + len) > start)
2306 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2308 struct vhost_vring_addr *ra;
2309 uint64_t start, end, len;
2312 end = start + imsg->size;
2314 ra = &vq->ring_addrs;
2315 len = sizeof(struct vring_packed_desc) * vq->size;
2316 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2319 len = sizeof(struct vring_packed_desc_event);
2320 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2323 len = sizeof(struct vring_packed_desc_event);
2324 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2327 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2328 len = sizeof(uint64_t);
2329 if (ra->log_guest_addr < end &&
2330 (ra->log_guest_addr + len) > start)
2337 static int is_vring_iotlb(struct virtio_net *dev,
2338 struct vhost_virtqueue *vq,
2339 struct vhost_iotlb_msg *imsg)
2341 if (vq_is_packed(dev))
2342 return is_vring_iotlb_packed(vq, imsg);
2344 return is_vring_iotlb_split(vq, imsg);
2348 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2349 int main_fd __rte_unused)
2351 struct virtio_net *dev = *pdev;
2352 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2356 if (validate_msg_fds(msg, 0) != 0)
2357 return RTE_VHOST_MSG_RESULT_ERR;
2359 switch (imsg->type) {
2360 case VHOST_IOTLB_UPDATE:
2362 vva = qva_to_vva(dev, imsg->uaddr, &len);
2364 return RTE_VHOST_MSG_RESULT_ERR;
2366 for (i = 0; i < dev->nr_vring; i++) {
2367 struct vhost_virtqueue *vq = dev->virtqueue[i];
2369 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2372 if (is_vring_iotlb(dev, vq, imsg))
2373 *pdev = dev = translate_ring_addresses(dev, i);
2376 case VHOST_IOTLB_INVALIDATE:
2377 for (i = 0; i < dev->nr_vring; i++) {
2378 struct vhost_virtqueue *vq = dev->virtqueue[i];
2380 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2383 if (is_vring_iotlb(dev, vq, imsg))
2384 vring_invalidate(dev, vq);
2388 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2390 return RTE_VHOST_MSG_RESULT_ERR;
2393 return RTE_VHOST_MSG_RESULT_OK;
2397 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2398 struct VhostUserMsg *msg,
2399 int main_fd __rte_unused)
2401 struct virtio_net *dev = *pdev;
2402 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2403 struct uffdio_api api_struct;
2405 if (validate_msg_fds(msg, 0) != 0)
2406 return RTE_VHOST_MSG_RESULT_ERR;
2408 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2410 if (dev->postcopy_ufd == -1) {
2411 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2413 return RTE_VHOST_MSG_RESULT_ERR;
2415 api_struct.api = UFFD_API;
2416 api_struct.features = 0;
2417 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2418 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2420 close(dev->postcopy_ufd);
2421 dev->postcopy_ufd = -1;
2422 return RTE_VHOST_MSG_RESULT_ERR;
2424 msg->fds[0] = dev->postcopy_ufd;
2427 return RTE_VHOST_MSG_RESULT_REPLY;
2429 dev->postcopy_ufd = -1;
2432 return RTE_VHOST_MSG_RESULT_ERR;
2437 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2438 struct VhostUserMsg *msg __rte_unused,
2439 int main_fd __rte_unused)
2441 struct virtio_net *dev = *pdev;
2443 if (validate_msg_fds(msg, 0) != 0)
2444 return RTE_VHOST_MSG_RESULT_ERR;
2446 if (dev->mem && dev->mem->nregions) {
2447 VHOST_LOG_CONFIG(ERR,
2448 "Regions already registered at postcopy-listen\n");
2449 return RTE_VHOST_MSG_RESULT_ERR;
2451 dev->postcopy_listening = 1;
2453 return RTE_VHOST_MSG_RESULT_OK;
2457 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2458 int main_fd __rte_unused)
2460 struct virtio_net *dev = *pdev;
2462 if (validate_msg_fds(msg, 0) != 0)
2463 return RTE_VHOST_MSG_RESULT_ERR;
2465 dev->postcopy_listening = 0;
2466 if (dev->postcopy_ufd >= 0) {
2467 close(dev->postcopy_ufd);
2468 dev->postcopy_ufd = -1;
2471 msg->payload.u64 = 0;
2472 msg->size = sizeof(msg->payload.u64);
2475 return RTE_VHOST_MSG_RESULT_REPLY;
2479 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2480 int main_fd __rte_unused)
2482 struct virtio_net *dev = *pdev;
2484 if (validate_msg_fds(msg, 0) != 0)
2485 return RTE_VHOST_MSG_RESULT_ERR;
2487 /* As per Virtio specification, the device status is 8bits long */
2488 if (msg->payload.u64 > UINT8_MAX) {
2489 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2491 return RTE_VHOST_MSG_RESULT_ERR;
2494 dev->status = msg->payload.u64;
2496 VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2497 "\t-ACKNOWLEDGE: %u\n"
2499 "\t-FEATURES_OK: %u\n"
2500 "\t-DRIVER_OK: %u\n"
2501 "\t-DEVICE_NEED_RESET: %u\n"
2504 !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2505 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2506 !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2507 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2508 !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2509 !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2511 return RTE_VHOST_MSG_RESULT_OK;
2514 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2515 struct VhostUserMsg *msg,
2517 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2518 [VHOST_USER_NONE] = NULL,
2519 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2520 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2521 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2522 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2523 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2524 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2525 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2526 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2527 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2528 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2529 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2530 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2531 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2532 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2533 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2534 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2535 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2536 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2537 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2538 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2539 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2540 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2541 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2542 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2543 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2544 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2545 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2546 [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2549 /* return bytes# of read on success or negative val on failure. */
2551 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2555 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2556 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2559 } else if (ret != VHOST_USER_HDR_SIZE) {
2560 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2566 if (msg->size > sizeof(msg->payload)) {
2567 VHOST_LOG_CONFIG(ERR,
2568 "invalid msg size: %d\n", msg->size);
2571 ret = read(sockfd, &msg->payload, msg->size);
2574 if (ret != (int)msg->size) {
2575 VHOST_LOG_CONFIG(ERR,
2576 "read control message failed\n");
2585 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2590 return send_fd_message(sockfd, (char *)msg,
2591 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2595 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2600 msg->flags &= ~VHOST_USER_VERSION_MASK;
2601 msg->flags &= ~VHOST_USER_NEED_REPLY;
2602 msg->flags |= VHOST_USER_VERSION;
2603 msg->flags |= VHOST_USER_REPLY_MASK;
2605 return send_vhost_message(sockfd, msg);
2609 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2613 if (msg->flags & VHOST_USER_NEED_REPLY)
2614 rte_spinlock_lock(&dev->slave_req_lock);
2616 ret = send_vhost_message(dev->slave_req_fd, msg);
2617 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2618 rte_spinlock_unlock(&dev->slave_req_lock);
2624 * Allocate a queue pair if it hasn't been allocated yet
2627 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2628 struct VhostUserMsg *msg)
2632 switch (msg->request.master) {
2633 case VHOST_USER_SET_VRING_KICK:
2634 case VHOST_USER_SET_VRING_CALL:
2635 case VHOST_USER_SET_VRING_ERR:
2636 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2638 case VHOST_USER_SET_VRING_NUM:
2639 case VHOST_USER_SET_VRING_BASE:
2640 case VHOST_USER_SET_VRING_ENABLE:
2641 vring_idx = msg->payload.state.index;
2643 case VHOST_USER_SET_VRING_ADDR:
2644 vring_idx = msg->payload.addr.index;
2650 if (vring_idx >= VHOST_MAX_VRING) {
2651 VHOST_LOG_CONFIG(ERR,
2652 "invalid vring index: %u\n", vring_idx);
2656 if (dev->virtqueue[vring_idx])
2659 return alloc_vring_queue(dev, vring_idx);
2663 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2666 unsigned int vq_num = 0;
2668 while (vq_num < dev->nr_vring) {
2669 struct vhost_virtqueue *vq = dev->virtqueue[i];
2672 rte_spinlock_lock(&vq->access_lock);
2680 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2683 unsigned int vq_num = 0;
2685 while (vq_num < dev->nr_vring) {
2686 struct vhost_virtqueue *vq = dev->virtqueue[i];
2689 rte_spinlock_unlock(&vq->access_lock);
2697 vhost_user_msg_handler(int vid, int fd)
2699 struct virtio_net *dev;
2700 struct VhostUserMsg msg;
2701 struct rte_vdpa_device *vdpa_dev;
2703 int unlock_required = 0;
2708 dev = get_device(vid);
2712 if (!dev->notify_ops) {
2713 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2714 if (!dev->notify_ops) {
2715 VHOST_LOG_CONFIG(ERR,
2716 "failed to get callback ops for driver %s\n",
2722 ret = read_vhost_message(fd, &msg);
2725 VHOST_LOG_CONFIG(ERR,
2726 "vhost read message failed\n");
2728 VHOST_LOG_CONFIG(INFO,
2729 "vhost peer closed\n");
2735 request = msg.request.master;
2736 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2737 vhost_message_str[request]) {
2738 if (request != VHOST_USER_IOTLB_MSG)
2739 VHOST_LOG_CONFIG(INFO, "read message %s\n",
2740 vhost_message_str[request]);
2742 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2743 vhost_message_str[request]);
2745 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2748 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2750 VHOST_LOG_CONFIG(ERR,
2751 "failed to alloc queue\n");
2756 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2757 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2758 * and device is destroyed. destroy_device waits for queues to be
2759 * inactive, so it is safe. Otherwise taking the access_lock
2760 * would cause a dead lock.
2763 case VHOST_USER_SET_FEATURES:
2764 case VHOST_USER_SET_PROTOCOL_FEATURES:
2765 case VHOST_USER_SET_OWNER:
2766 case VHOST_USER_SET_MEM_TABLE:
2767 case VHOST_USER_SET_LOG_BASE:
2768 case VHOST_USER_SET_LOG_FD:
2769 case VHOST_USER_SET_VRING_NUM:
2770 case VHOST_USER_SET_VRING_ADDR:
2771 case VHOST_USER_SET_VRING_BASE:
2772 case VHOST_USER_SET_VRING_KICK:
2773 case VHOST_USER_SET_VRING_CALL:
2774 case VHOST_USER_SET_VRING_ERR:
2775 case VHOST_USER_SET_VRING_ENABLE:
2776 case VHOST_USER_SEND_RARP:
2777 case VHOST_USER_NET_SET_MTU:
2778 case VHOST_USER_SET_SLAVE_REQ_FD:
2779 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2780 vhost_user_lock_all_queue_pairs(dev);
2781 unlock_required = 1;
2790 if (dev->extern_ops.pre_msg_handle) {
2791 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2794 case RTE_VHOST_MSG_RESULT_REPLY:
2795 send_vhost_reply(fd, &msg);
2797 case RTE_VHOST_MSG_RESULT_ERR:
2798 case RTE_VHOST_MSG_RESULT_OK:
2800 goto skip_to_post_handle;
2801 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2807 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2808 if (!vhost_message_handlers[request])
2809 goto skip_to_post_handle;
2810 ret = vhost_message_handlers[request](&dev, &msg, fd);
2813 case RTE_VHOST_MSG_RESULT_ERR:
2814 VHOST_LOG_CONFIG(ERR,
2815 "Processing %s failed.\n",
2816 vhost_message_str[request]);
2819 case RTE_VHOST_MSG_RESULT_OK:
2820 VHOST_LOG_CONFIG(DEBUG,
2821 "Processing %s succeeded.\n",
2822 vhost_message_str[request]);
2825 case RTE_VHOST_MSG_RESULT_REPLY:
2826 VHOST_LOG_CONFIG(DEBUG,
2827 "Processing %s succeeded and needs reply.\n",
2828 vhost_message_str[request]);
2829 send_vhost_reply(fd, &msg);
2837 skip_to_post_handle:
2838 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2839 dev->extern_ops.post_msg_handle) {
2840 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2843 case RTE_VHOST_MSG_RESULT_REPLY:
2844 send_vhost_reply(fd, &msg);
2846 case RTE_VHOST_MSG_RESULT_ERR:
2847 case RTE_VHOST_MSG_RESULT_OK:
2849 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2855 if (unlock_required)
2856 vhost_user_unlock_all_queue_pairs(dev);
2858 /* If message was not handled at this stage, treat it as an error */
2860 VHOST_LOG_CONFIG(ERR,
2861 "vhost message (req: %d) was not handled.\n", request);
2862 close_msg_fds(&msg);
2863 ret = RTE_VHOST_MSG_RESULT_ERR;
2867 * If the request required a reply that was already sent,
2868 * this optional reply-ack won't be sent as the
2869 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2871 if (msg.flags & VHOST_USER_NEED_REPLY) {
2872 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2873 msg.size = sizeof(msg.payload.u64);
2875 send_vhost_reply(fd, &msg);
2876 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2877 VHOST_LOG_CONFIG(ERR,
2878 "vhost message handling failed.\n");
2882 for (i = 0; i < dev->nr_vring; i++) {
2883 struct vhost_virtqueue *vq = dev->virtqueue[i];
2884 bool cur_ready = vq_is_ready(dev, vq);
2886 if (cur_ready != (vq && vq->ready)) {
2887 vhost_user_notify_queue_state(dev, i, cur_ready);
2888 vq->ready = cur_ready;
2893 if (!virtio_is_ready(dev))
2897 * Virtio is now ready. If not done already, it is time
2898 * to notify the application it can process the rings and
2899 * configure the vDPA device if present.
2902 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2903 if (dev->notify_ops->new_device(dev->vid) == 0)
2904 dev->flags |= VIRTIO_DEV_RUNNING;
2907 vdpa_dev = dev->vdpa_dev;
2911 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2912 if (vdpa_dev->ops->dev_conf(dev->vid))
2913 VHOST_LOG_CONFIG(ERR,
2914 "Failed to configure vDPA device\n");
2916 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2923 static int process_slave_message_reply(struct virtio_net *dev,
2924 const struct VhostUserMsg *msg)
2926 struct VhostUserMsg msg_reply;
2929 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2932 ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
2935 VHOST_LOG_CONFIG(ERR,
2936 "vhost read slave message reply failed\n");
2938 VHOST_LOG_CONFIG(INFO,
2939 "vhost peer closed\n");
2945 if (msg_reply.request.slave != msg->request.slave) {
2946 VHOST_LOG_CONFIG(ERR,
2947 "Received unexpected msg type (%u), expected %u\n",
2948 msg_reply.request.slave, msg->request.slave);
2953 ret = msg_reply.payload.u64 ? -1 : 0;
2956 rte_spinlock_unlock(&dev->slave_req_lock);
2961 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2964 struct VhostUserMsg msg = {
2965 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2966 .flags = VHOST_USER_VERSION,
2967 .size = sizeof(msg.payload.iotlb),
2971 .type = VHOST_IOTLB_MISS,
2975 ret = send_vhost_message(dev->slave_req_fd, &msg);
2977 VHOST_LOG_CONFIG(ERR,
2978 "Failed to send IOTLB miss message (%d)\n",
2987 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
2990 struct VhostUserMsg msg = {
2991 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
2992 .flags = VHOST_USER_VERSION,
2997 msg.flags |= VHOST_USER_NEED_REPLY;
2999 ret = send_vhost_slave_message(dev, &msg);
3001 VHOST_LOG_CONFIG(ERR,
3002 "Failed to send config change (%d)\n",
3007 return process_slave_message_reply(dev, &msg);
3011 rte_vhost_slave_config_change(int vid, bool need_reply)
3013 struct virtio_net *dev;
3015 dev = get_device(vid);
3019 return vhost_user_slave_config_change(dev, need_reply);
3022 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3028 struct VhostUserMsg msg = {
3029 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3030 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3031 .size = sizeof(msg.payload.area),
3033 .u64 = index & VHOST_USER_VRING_IDX_MASK,
3040 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3046 ret = send_vhost_slave_message(dev, &msg);
3048 VHOST_LOG_CONFIG(ERR,
3049 "Failed to set host notifier (%d)\n", ret);
3053 return process_slave_message_reply(dev, &msg);
3056 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3058 struct virtio_net *dev;
3059 struct rte_vdpa_device *vdpa_dev;
3060 int vfio_device_fd, ret = 0;
3061 uint64_t offset, size;
3062 unsigned int i, q_start, q_last;
3064 dev = get_device(vid);
3068 vdpa_dev = dev->vdpa_dev;
3069 if (vdpa_dev == NULL)
3072 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3073 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3074 !(dev->protocol_features &
3075 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3076 !(dev->protocol_features &
3077 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3078 !(dev->protocol_features &
3079 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3082 if (qid == RTE_VHOST_QUEUE_ALL) {
3084 q_last = dev->nr_vring - 1;
3086 if (qid >= dev->nr_vring)
3092 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3093 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3095 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3096 if (vfio_device_fd < 0)
3100 for (i = q_start; i <= q_last; i++) {
3101 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3107 if (vhost_user_slave_set_vring_host_notifier(dev, i,
3108 vfio_device_fd, offset, size) < 0) {
3115 for (i = q_start; i <= q_last; i++) {
3116 vhost_user_slave_set_vring_host_notifier(dev, i, -1,