1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2018 Intel Corporation
7 * The vhost-user protocol connection is an external interface, so it must be
8 * robust against invalid inputs.
10 * This is important because the vhost-user master is only one step removed
11 * from the guest. Malicious guests that have escaped will then launch further
12 * attacks from the vhost-user master.
14 * Even in deployments where guests are trusted, a bug in the vhost-user master
15 * can still cause invalid messages to be sent. Such messages must not
16 * compromise the stability of the DPDK application by causing crashes, memory
17 * corruption, or other problematic behavior.
19 * Do not assume received VhostUserMsg fields contain sensible values!
28 #include <sys/ioctl.h>
30 #include <sys/types.h>
32 #include <sys/syscall.h>
34 #ifdef RTE_LIBRTE_VHOST_NUMA
37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
38 #include <linux/userfaultfd.h>
40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
41 #include <linux/memfd.h>
42 #define MEMFD_SUPPORTED
45 #include <rte_common.h>
46 #include <rte_malloc.h>
51 #include "vhost_user.h"
53 #define VIRTIO_MIN_MTU 68
54 #define VIRTIO_MAX_MTU 65535
56 #define INFLIGHT_ALIGNMENT 64
57 #define INFLIGHT_VERSION 0x1
59 static const char *vhost_message_str[VHOST_USER_MAX] = {
60 [VHOST_USER_NONE] = "VHOST_USER_NONE",
61 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
62 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
63 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
64 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
65 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
66 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
67 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
68 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
69 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
70 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
71 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
72 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
73 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
74 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
75 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
76 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
77 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
78 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
79 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
80 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
81 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
82 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
83 [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
84 [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
85 [VHOST_USER_POSTCOPY_ADVISE] = "VHOST_USER_POSTCOPY_ADVISE",
86 [VHOST_USER_POSTCOPY_LISTEN] = "VHOST_USER_POSTCOPY_LISTEN",
87 [VHOST_USER_POSTCOPY_END] = "VHOST_USER_POSTCOPY_END",
88 [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
89 [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
92 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
93 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
96 close_msg_fds(struct VhostUserMsg *msg)
100 for (i = 0; i < msg->fd_num; i++)
105 * Ensure the expected number of FDs is received,
106 * close all FDs and return an error if this is not the case.
109 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
111 if (msg->fd_num == expected_fds)
114 VHOST_LOG_CONFIG(ERR,
115 " Expect %d FDs for request %s, received %d\n",
117 vhost_message_str[msg->request.master],
131 ret = fstat(fd, &stat);
132 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
136 * Reclaim all the outstanding zmbufs for a virtqueue.
139 drain_zmbuf_list(struct vhost_virtqueue *vq)
141 struct zcopy_mbuf *zmbuf, *next;
143 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
144 zmbuf != NULL; zmbuf = next) {
145 next = TAILQ_NEXT(zmbuf, next);
147 while (!mbuf_is_consumed(zmbuf->mbuf))
150 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
151 restore_mbuf(zmbuf->mbuf);
152 rte_pktmbuf_free(zmbuf->mbuf);
159 free_mem_region(struct virtio_net *dev)
162 struct rte_vhost_mem_region *reg;
163 struct vhost_virtqueue *vq;
165 if (!dev || !dev->mem)
168 if (dev->dequeue_zero_copy) {
169 for (i = 0; i < dev->nr_vring; i++) {
170 vq = dev->virtqueue[i];
172 drain_zmbuf_list(vq);
176 for (i = 0; i < dev->mem->nregions; i++) {
177 reg = &dev->mem->regions[i];
178 if (reg->host_user_addr) {
179 munmap(reg->mmap_addr, reg->mmap_size);
186 vhost_backend_cleanup(struct virtio_net *dev)
189 free_mem_region(dev);
194 free(dev->guest_pages);
195 dev->guest_pages = NULL;
198 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
202 if (dev->inflight_info) {
203 if (dev->inflight_info->addr) {
204 munmap(dev->inflight_info->addr,
205 dev->inflight_info->size);
206 dev->inflight_info->addr = NULL;
209 if (dev->inflight_info->fd > 0) {
210 close(dev->inflight_info->fd);
211 dev->inflight_info->fd = -1;
214 free(dev->inflight_info);
215 dev->inflight_info = NULL;
218 if (dev->slave_req_fd >= 0) {
219 close(dev->slave_req_fd);
220 dev->slave_req_fd = -1;
223 if (dev->postcopy_ufd >= 0) {
224 close(dev->postcopy_ufd);
225 dev->postcopy_ufd = -1;
228 dev->postcopy_listening = 0;
232 * This function just returns success at the moment unless
233 * the device hasn't been initialised.
236 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
237 struct VhostUserMsg *msg,
238 int main_fd __rte_unused)
240 if (validate_msg_fds(msg, 0) != 0)
241 return RTE_VHOST_MSG_RESULT_ERR;
243 return RTE_VHOST_MSG_RESULT_OK;
247 vhost_user_reset_owner(struct virtio_net **pdev,
248 struct VhostUserMsg *msg,
249 int main_fd __rte_unused)
251 struct virtio_net *dev = *pdev;
253 if (validate_msg_fds(msg, 0) != 0)
254 return RTE_VHOST_MSG_RESULT_ERR;
256 vhost_destroy_device_notify(dev);
258 cleanup_device(dev, 0);
260 return RTE_VHOST_MSG_RESULT_OK;
264 * The features that we support are requested.
267 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
268 int main_fd __rte_unused)
270 struct virtio_net *dev = *pdev;
271 uint64_t features = 0;
273 if (validate_msg_fds(msg, 0) != 0)
274 return RTE_VHOST_MSG_RESULT_ERR;
276 rte_vhost_driver_get_features(dev->ifname, &features);
278 msg->payload.u64 = features;
279 msg->size = sizeof(msg->payload.u64);
282 return RTE_VHOST_MSG_RESULT_REPLY;
286 * The queue number that we support are requested.
289 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
290 int main_fd __rte_unused)
292 struct virtio_net *dev = *pdev;
293 uint32_t queue_num = 0;
295 if (validate_msg_fds(msg, 0) != 0)
296 return RTE_VHOST_MSG_RESULT_ERR;
298 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
300 msg->payload.u64 = (uint64_t)queue_num;
301 msg->size = sizeof(msg->payload.u64);
304 return RTE_VHOST_MSG_RESULT_REPLY;
308 * We receive the negotiated features supported by us and the virtio device.
311 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
312 int main_fd __rte_unused)
314 struct virtio_net *dev = *pdev;
315 uint64_t features = msg->payload.u64;
316 uint64_t vhost_features = 0;
317 struct rte_vdpa_device *vdpa_dev;
320 if (validate_msg_fds(msg, 0) != 0)
321 return RTE_VHOST_MSG_RESULT_ERR;
323 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
324 if (features & ~vhost_features) {
325 VHOST_LOG_CONFIG(ERR,
326 "(%d) received invalid negotiated features.\n",
328 return RTE_VHOST_MSG_RESULT_ERR;
331 if (dev->flags & VIRTIO_DEV_RUNNING) {
332 if (dev->features == features)
333 return RTE_VHOST_MSG_RESULT_OK;
336 * Error out if master tries to change features while device is
337 * in running state. The exception being VHOST_F_LOG_ALL, which
338 * is enabled when the live-migration starts.
340 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
341 VHOST_LOG_CONFIG(ERR,
342 "(%d) features changed while device is running.\n",
344 return RTE_VHOST_MSG_RESULT_ERR;
347 if (dev->notify_ops->features_changed)
348 dev->notify_ops->features_changed(dev->vid, features);
351 dev->features = features;
353 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
354 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
356 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
358 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 did = dev->vdpa_dev_id;
388 vdpa_dev = rte_vdpa_get_device(did);
389 if (vdpa_dev && vdpa_dev->ops->set_features)
390 vdpa_dev->ops->set_features(dev->vid);
392 return RTE_VHOST_MSG_RESULT_OK;
396 * The virtio device sends us the size of the descriptor ring.
399 vhost_user_set_vring_num(struct virtio_net **pdev,
400 struct VhostUserMsg *msg,
401 int main_fd __rte_unused)
403 struct virtio_net *dev = *pdev;
404 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
406 if (validate_msg_fds(msg, 0) != 0)
407 return RTE_VHOST_MSG_RESULT_ERR;
409 vq->size = msg->payload.state.num;
411 /* VIRTIO 1.0, 2.4 Virtqueues says:
413 * Queue Size value is always a power of 2. The maximum Queue Size
416 * VIRTIO 1.1 2.7 Virtqueues says:
418 * Packed virtqueues support up to 2^15 entries each.
420 if (!vq_is_packed(dev)) {
421 if (vq->size & (vq->size - 1)) {
422 VHOST_LOG_CONFIG(ERR,
423 "invalid virtqueue size %u\n", vq->size);
424 return RTE_VHOST_MSG_RESULT_ERR;
428 if (vq->size > 32768) {
429 VHOST_LOG_CONFIG(ERR,
430 "invalid virtqueue size %u\n", vq->size);
431 return RTE_VHOST_MSG_RESULT_ERR;
434 if (dev->dequeue_zero_copy) {
436 vq->last_zmbuf_idx = 0;
437 vq->zmbuf_size = vq->size;
439 rte_free(vq->zmbufs);
440 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
441 sizeof(struct zcopy_mbuf), 0);
442 if (vq->zmbufs == NULL) {
443 VHOST_LOG_CONFIG(WARNING,
444 "failed to allocate mem for zero copy; "
445 "zero copy is force disabled\n");
446 dev->dequeue_zero_copy = 0;
448 TAILQ_INIT(&vq->zmbuf_list);
451 if (vq_is_packed(dev)) {
452 if (vq->shadow_used_packed)
453 rte_free(vq->shadow_used_packed);
454 vq->shadow_used_packed = rte_malloc(NULL,
456 sizeof(struct vring_used_elem_packed),
457 RTE_CACHE_LINE_SIZE);
458 if (!vq->shadow_used_packed) {
459 VHOST_LOG_CONFIG(ERR,
460 "failed to allocate memory for shadow used ring.\n");
461 return RTE_VHOST_MSG_RESULT_ERR;
465 if (vq->shadow_used_split)
466 rte_free(vq->shadow_used_split);
467 vq->shadow_used_split = rte_malloc(NULL,
468 vq->size * sizeof(struct vring_used_elem),
469 RTE_CACHE_LINE_SIZE);
470 if (!vq->shadow_used_split) {
471 VHOST_LOG_CONFIG(ERR,
472 "failed to allocate memory for shadow used ring.\n");
473 return RTE_VHOST_MSG_RESULT_ERR;
477 if (vq->batch_copy_elems)
478 rte_free(vq->batch_copy_elems);
479 vq->batch_copy_elems = rte_malloc(NULL,
480 vq->size * sizeof(struct batch_copy_elem),
481 RTE_CACHE_LINE_SIZE);
482 if (!vq->batch_copy_elems) {
483 VHOST_LOG_CONFIG(ERR,
484 "failed to allocate memory for batching copy.\n");
485 return RTE_VHOST_MSG_RESULT_ERR;
488 return RTE_VHOST_MSG_RESULT_OK;
492 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
493 * same numa node as the memory of vring descriptor.
495 #ifdef RTE_LIBRTE_VHOST_NUMA
496 static struct virtio_net*
497 numa_realloc(struct virtio_net *dev, int index)
499 int oldnode, newnode;
500 struct virtio_net *old_dev;
501 struct vhost_virtqueue *old_vq, *vq;
502 struct zcopy_mbuf *new_zmbuf;
503 struct vring_used_elem *new_shadow_used_split;
504 struct vring_used_elem_packed *new_shadow_used_packed;
505 struct batch_copy_elem *new_batch_copy_elems;
508 if (dev->flags & VIRTIO_DEV_RUNNING)
512 vq = old_vq = dev->virtqueue[index];
514 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
515 MPOL_F_NODE | MPOL_F_ADDR);
517 /* check if we need to reallocate vq */
518 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
519 MPOL_F_NODE | MPOL_F_ADDR);
521 VHOST_LOG_CONFIG(ERR,
522 "Unable to get vq numa information.\n");
525 if (oldnode != newnode) {
526 VHOST_LOG_CONFIG(INFO,
527 "reallocate vq from %d to %d node\n", oldnode, newnode);
528 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
532 memcpy(vq, old_vq, sizeof(*vq));
533 TAILQ_INIT(&vq->zmbuf_list);
535 if (dev->dequeue_zero_copy) {
536 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
537 sizeof(struct zcopy_mbuf), 0, newnode);
539 rte_free(vq->zmbufs);
540 vq->zmbufs = new_zmbuf;
544 if (vq_is_packed(dev)) {
545 new_shadow_used_packed = rte_malloc_socket(NULL,
547 sizeof(struct vring_used_elem_packed),
550 if (new_shadow_used_packed) {
551 rte_free(vq->shadow_used_packed);
552 vq->shadow_used_packed = new_shadow_used_packed;
555 new_shadow_used_split = rte_malloc_socket(NULL,
557 sizeof(struct vring_used_elem),
560 if (new_shadow_used_split) {
561 rte_free(vq->shadow_used_split);
562 vq->shadow_used_split = new_shadow_used_split;
566 new_batch_copy_elems = rte_malloc_socket(NULL,
567 vq->size * sizeof(struct batch_copy_elem),
570 if (new_batch_copy_elems) {
571 rte_free(vq->batch_copy_elems);
572 vq->batch_copy_elems = new_batch_copy_elems;
578 /* check if we need to reallocate dev */
579 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
580 MPOL_F_NODE | MPOL_F_ADDR);
582 VHOST_LOG_CONFIG(ERR,
583 "Unable to get dev numa information.\n");
586 if (oldnode != newnode) {
587 VHOST_LOG_CONFIG(INFO,
588 "reallocate dev from %d to %d node\n",
590 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
596 memcpy(dev, old_dev, sizeof(*dev));
601 dev->virtqueue[index] = vq;
602 vhost_devices[dev->vid] = dev;
605 vhost_user_iotlb_init(dev, index);
610 static struct virtio_net*
611 numa_realloc(struct virtio_net *dev, int index __rte_unused)
617 /* Converts QEMU virtual address to Vhost virtual address. */
619 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
621 struct rte_vhost_mem_region *r;
624 if (unlikely(!dev || !dev->mem))
627 /* Find the region where the address lives. */
628 for (i = 0; i < dev->mem->nregions; i++) {
629 r = &dev->mem->regions[i];
631 if (qva >= r->guest_user_addr &&
632 qva < r->guest_user_addr + r->size) {
634 if (unlikely(*len > r->guest_user_addr + r->size - qva))
635 *len = r->guest_user_addr + r->size - qva;
637 return qva - r->guest_user_addr +
649 * Converts ring address to Vhost virtual address.
650 * If IOMMU is enabled, the ring address is a guest IO virtual address,
651 * else it is a QEMU virtual address.
654 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
655 uint64_t ra, uint64_t *size)
657 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
659 uint64_t req_size = *size;
661 vva = vhost_user_iotlb_cache_find(vq, ra,
662 size, VHOST_ACCESS_RW);
663 if (req_size != *size)
664 vhost_user_iotlb_miss(dev, (ra + *size),
670 return qva_to_vva(dev, ra, size);
674 * Converts vring log address to GPA
675 * If IOMMU is enabled, the log address is IOVA
676 * If IOMMU not enabled, the log address is already GPA
679 translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
682 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
683 const uint64_t exp_size = sizeof(struct vring_used) +
684 sizeof(struct vring_used_elem) * vq->size;
686 uint64_t size = exp_size;
688 hva = vhost_iova_to_vva(dev, vq, log_addr,
689 &size, VHOST_ACCESS_RW);
690 if (size != exp_size)
693 gpa = hva_to_gpa(dev, hva, exp_size);
695 VHOST_LOG_CONFIG(ERR,
696 "VQ: Failed to find GPA for log_addr: 0x%" PRIx64 " hva: 0x%" PRIx64 "\n",
706 static struct virtio_net *
707 translate_ring_addresses(struct virtio_net *dev, int vq_index)
709 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
710 struct vhost_vring_addr *addr = &vq->ring_addrs;
711 uint64_t len, expected_len;
713 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
715 translate_log_addr(dev, vq, addr->log_guest_addr);
716 if (vq->log_guest_addr == 0) {
717 VHOST_LOG_CONFIG(DEBUG,
718 "(%d) failed to map log_guest_addr.\n",
724 if (vq_is_packed(dev)) {
725 len = sizeof(struct vring_packed_desc) * vq->size;
726 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
727 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
728 if (vq->desc_packed == NULL ||
729 len != sizeof(struct vring_packed_desc) *
731 VHOST_LOG_CONFIG(DEBUG,
732 "(%d) failed to map desc_packed ring.\n",
737 dev = numa_realloc(dev, vq_index);
738 vq = dev->virtqueue[vq_index];
739 addr = &vq->ring_addrs;
741 len = sizeof(struct vring_packed_desc_event);
742 vq->driver_event = (struct vring_packed_desc_event *)
743 (uintptr_t)ring_addr_to_vva(dev,
744 vq, addr->avail_user_addr, &len);
745 if (vq->driver_event == NULL ||
746 len != sizeof(struct vring_packed_desc_event)) {
747 VHOST_LOG_CONFIG(DEBUG,
748 "(%d) failed to find driver area address.\n",
753 len = sizeof(struct vring_packed_desc_event);
754 vq->device_event = (struct vring_packed_desc_event *)
755 (uintptr_t)ring_addr_to_vva(dev,
756 vq, addr->used_user_addr, &len);
757 if (vq->device_event == NULL ||
758 len != sizeof(struct vring_packed_desc_event)) {
759 VHOST_LOG_CONFIG(DEBUG,
760 "(%d) failed to find device area address.\n",
769 /* The addresses are converted from QEMU virtual to Vhost virtual. */
770 if (vq->desc && vq->avail && vq->used)
773 len = sizeof(struct vring_desc) * vq->size;
774 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
775 vq, addr->desc_user_addr, &len);
776 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
777 VHOST_LOG_CONFIG(DEBUG,
778 "(%d) failed to map desc ring.\n",
783 dev = numa_realloc(dev, vq_index);
784 vq = dev->virtqueue[vq_index];
785 addr = &vq->ring_addrs;
787 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
788 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
789 len += sizeof(uint16_t);
791 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
792 vq, addr->avail_user_addr, &len);
793 if (vq->avail == 0 || len != expected_len) {
794 VHOST_LOG_CONFIG(DEBUG,
795 "(%d) failed to map avail ring.\n",
800 len = sizeof(struct vring_used) +
801 sizeof(struct vring_used_elem) * vq->size;
802 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
803 len += sizeof(uint16_t);
805 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
806 vq, addr->used_user_addr, &len);
807 if (vq->used == 0 || len != expected_len) {
808 VHOST_LOG_CONFIG(DEBUG,
809 "(%d) failed to map used ring.\n",
814 if (vq->last_used_idx != vq->used->idx) {
815 VHOST_LOG_CONFIG(WARNING,
816 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
817 "some packets maybe resent for Tx and dropped for Rx\n",
818 vq->last_used_idx, vq->used->idx);
819 vq->last_used_idx = vq->used->idx;
820 vq->last_avail_idx = vq->used->idx;
825 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
827 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
828 dev->vid, vq->avail);
829 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
831 VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
832 dev->vid, vq->log_guest_addr);
838 * The virtio device sends us the desc, used and avail ring addresses.
839 * This function then converts these to our address space.
842 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
843 int main_fd __rte_unused)
845 struct virtio_net *dev = *pdev;
846 struct vhost_virtqueue *vq;
847 struct vhost_vring_addr *addr = &msg->payload.addr;
850 if (validate_msg_fds(msg, 0) != 0)
851 return RTE_VHOST_MSG_RESULT_ERR;
853 if (dev->mem == NULL)
854 return RTE_VHOST_MSG_RESULT_ERR;
856 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
857 vq = dev->virtqueue[msg->payload.addr.index];
859 access_ok = vq->access_ok;
862 * Rings addresses should not be interpreted as long as the ring is not
863 * started and enabled
865 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
867 vring_invalidate(dev, vq);
869 if ((vq->enabled && (dev->features &
870 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
872 dev = translate_ring_addresses(dev, msg->payload.addr.index);
874 return RTE_VHOST_MSG_RESULT_ERR;
879 return RTE_VHOST_MSG_RESULT_OK;
883 * The virtio device sends us the available ring last used index.
886 vhost_user_set_vring_base(struct virtio_net **pdev,
887 struct VhostUserMsg *msg,
888 int main_fd __rte_unused)
890 struct virtio_net *dev = *pdev;
891 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
892 uint64_t val = msg->payload.state.num;
894 if (validate_msg_fds(msg, 0) != 0)
895 return RTE_VHOST_MSG_RESULT_ERR;
897 if (vq_is_packed(dev)) {
899 * Bit[0:14]: avail index
900 * Bit[15]: avail wrap counter
902 vq->last_avail_idx = val & 0x7fff;
903 vq->avail_wrap_counter = !!(val & (0x1 << 15));
905 * Set used index to same value as available one, as
906 * their values should be the same since ring processing
907 * was stopped at get time.
909 vq->last_used_idx = vq->last_avail_idx;
910 vq->used_wrap_counter = vq->avail_wrap_counter;
912 vq->last_used_idx = msg->payload.state.num;
913 vq->last_avail_idx = msg->payload.state.num;
916 return RTE_VHOST_MSG_RESULT_OK;
920 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
921 uint64_t host_phys_addr, uint64_t size)
923 struct guest_page *page, *last_page;
924 struct guest_page *old_pages;
926 if (dev->nr_guest_pages == dev->max_guest_pages) {
927 dev->max_guest_pages *= 2;
928 old_pages = dev->guest_pages;
929 dev->guest_pages = realloc(dev->guest_pages,
930 dev->max_guest_pages * sizeof(*page));
931 if (!dev->guest_pages) {
932 VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
938 if (dev->nr_guest_pages > 0) {
939 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
940 /* merge if the two pages are continuous */
941 if (host_phys_addr == last_page->host_phys_addr +
943 last_page->size += size;
948 page = &dev->guest_pages[dev->nr_guest_pages++];
949 page->guest_phys_addr = guest_phys_addr;
950 page->host_phys_addr = host_phys_addr;
957 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
960 uint64_t reg_size = reg->size;
961 uint64_t host_user_addr = reg->host_user_addr;
962 uint64_t guest_phys_addr = reg->guest_phys_addr;
963 uint64_t host_phys_addr;
966 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
967 size = page_size - (guest_phys_addr & (page_size - 1));
968 size = RTE_MIN(size, reg_size);
970 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
973 host_user_addr += size;
974 guest_phys_addr += size;
977 while (reg_size > 0) {
978 size = RTE_MIN(reg_size, page_size);
979 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
981 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
985 host_user_addr += size;
986 guest_phys_addr += size;
993 #ifdef RTE_LIBRTE_VHOST_DEBUG
994 /* TODO: enable it only in debug mode? */
996 dump_guest_pages(struct virtio_net *dev)
999 struct guest_page *page;
1001 for (i = 0; i < dev->nr_guest_pages; i++) {
1002 page = &dev->guest_pages[i];
1004 VHOST_LOG_CONFIG(INFO,
1005 "guest physical page region %u\n"
1006 "\t guest_phys_addr: %" PRIx64 "\n"
1007 "\t host_phys_addr : %" PRIx64 "\n"
1008 "\t size : %" PRIx64 "\n",
1010 page->guest_phys_addr,
1011 page->host_phys_addr,
1016 #define dump_guest_pages(dev)
1020 vhost_memory_changed(struct VhostUserMemory *new,
1021 struct rte_vhost_memory *old)
1025 if (new->nregions != old->nregions)
1028 for (i = 0; i < new->nregions; ++i) {
1029 VhostUserMemoryRegion *new_r = &new->regions[i];
1030 struct rte_vhost_mem_region *old_r = &old->regions[i];
1032 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1034 if (new_r->memory_size != old_r->size)
1036 if (new_r->userspace_addr != old_r->guest_user_addr)
1044 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1047 struct virtio_net *dev = *pdev;
1048 struct VhostUserMemory *memory = &msg->payload.memory;
1049 struct rte_vhost_mem_region *reg;
1052 uint64_t mmap_offset;
1058 if (validate_msg_fds(msg, memory->nregions) != 0)
1059 return RTE_VHOST_MSG_RESULT_ERR;
1061 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1062 VHOST_LOG_CONFIG(ERR,
1063 "too many memory regions (%u)\n", memory->nregions);
1064 return RTE_VHOST_MSG_RESULT_ERR;
1067 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1068 VHOST_LOG_CONFIG(INFO,
1069 "(%d) memory regions not changed\n", dev->vid);
1073 return RTE_VHOST_MSG_RESULT_OK;
1077 free_mem_region(dev);
1082 /* Flush IOTLB cache as previous HVAs are now invalid */
1083 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1084 for (i = 0; i < dev->nr_vring; i++)
1085 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1087 dev->nr_guest_pages = 0;
1088 if (!dev->guest_pages) {
1089 dev->max_guest_pages = 8;
1090 dev->guest_pages = malloc(dev->max_guest_pages *
1091 sizeof(struct guest_page));
1092 if (dev->guest_pages == NULL) {
1093 VHOST_LOG_CONFIG(ERR,
1094 "(%d) failed to allocate memory "
1095 "for dev->guest_pages\n",
1097 return RTE_VHOST_MSG_RESULT_ERR;
1101 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1102 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1103 if (dev->mem == NULL) {
1104 VHOST_LOG_CONFIG(ERR,
1105 "(%d) failed to allocate memory for dev->mem\n",
1107 return RTE_VHOST_MSG_RESULT_ERR;
1109 dev->mem->nregions = memory->nregions;
1111 for (i = 0; i < memory->nregions; i++) {
1113 reg = &dev->mem->regions[i];
1115 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1116 reg->guest_user_addr = memory->regions[i].userspace_addr;
1117 reg->size = memory->regions[i].memory_size;
1120 mmap_offset = memory->regions[i].mmap_offset;
1122 /* Check for memory_size + mmap_offset overflow */
1123 if (mmap_offset >= -reg->size) {
1124 VHOST_LOG_CONFIG(ERR,
1125 "mmap_offset (%#"PRIx64") and memory_size "
1126 "(%#"PRIx64") overflow\n",
1127 mmap_offset, reg->size);
1131 mmap_size = reg->size + mmap_offset;
1133 /* mmap() without flag of MAP_ANONYMOUS, should be called
1134 * with length argument aligned with hugepagesz at older
1135 * longterm version Linux, like 2.6.32 and 3.2.72, or
1136 * mmap() will fail with EINVAL.
1138 * to avoid failure, make sure in caller to keep length
1141 alignment = get_blk_size(fd);
1142 if (alignment == (uint64_t)-1) {
1143 VHOST_LOG_CONFIG(ERR,
1144 "couldn't get hugepage size through fstat\n");
1147 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1148 if (mmap_size == 0) {
1150 * It could happen if initial mmap_size + alignment
1151 * overflows the sizeof uint64, which could happen if
1152 * either mmap_size or alignment value is wrong.
1154 * mmap() kernel implementation would return an error,
1155 * but better catch it before and provide useful info
1158 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1159 "or alignment (0x%" PRIx64 ") is invalid\n",
1160 reg->size + mmap_offset, alignment);
1164 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
1165 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1166 MAP_SHARED | populate, fd, 0);
1168 if (mmap_addr == MAP_FAILED) {
1169 VHOST_LOG_CONFIG(ERR,
1170 "mmap region %u failed.\n", i);
1174 reg->mmap_addr = mmap_addr;
1175 reg->mmap_size = mmap_size;
1176 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1179 if (dev->dequeue_zero_copy)
1180 if (add_guest_pages(dev, reg, alignment) < 0) {
1181 VHOST_LOG_CONFIG(ERR,
1182 "adding guest pages to region %u failed.\n",
1187 VHOST_LOG_CONFIG(INFO,
1188 "guest memory region %u, size: 0x%" PRIx64 "\n"
1189 "\t guest physical addr: 0x%" PRIx64 "\n"
1190 "\t guest virtual addr: 0x%" PRIx64 "\n"
1191 "\t host virtual addr: 0x%" PRIx64 "\n"
1192 "\t mmap addr : 0x%" PRIx64 "\n"
1193 "\t mmap size : 0x%" PRIx64 "\n"
1194 "\t mmap align: 0x%" PRIx64 "\n"
1195 "\t mmap off : 0x%" PRIx64 "\n",
1197 reg->guest_phys_addr,
1198 reg->guest_user_addr,
1199 reg->host_user_addr,
1200 (uint64_t)(uintptr_t)mmap_addr,
1205 if (dev->postcopy_listening) {
1207 * We haven't a better way right now than sharing
1208 * DPDK's virtual address with Qemu, so that Qemu can
1209 * retrieve the region offset when handling userfaults.
1211 memory->regions[i].userspace_addr =
1212 reg->host_user_addr;
1215 if (dev->postcopy_listening) {
1216 /* Send the addresses back to qemu */
1218 send_vhost_reply(main_fd, msg);
1220 /* Wait for qemu to acknolwedge it's got the addresses
1221 * we've got to wait before we're allowed to generate faults.
1223 VhostUserMsg ack_msg;
1224 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1225 VHOST_LOG_CONFIG(ERR,
1226 "Failed to read qemu ack on postcopy set-mem-table\n");
1230 if (validate_msg_fds(&ack_msg, 0) != 0)
1233 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1234 VHOST_LOG_CONFIG(ERR,
1235 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1236 ack_msg.request.master);
1240 /* Now userfault register and we can use the memory */
1241 for (i = 0; i < memory->nregions; i++) {
1242 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1243 reg = &dev->mem->regions[i];
1244 struct uffdio_register reg_struct;
1247 * Let's register all the mmap'ed area to ensure
1248 * alignment on page boundary.
1250 reg_struct.range.start =
1251 (uint64_t)(uintptr_t)reg->mmap_addr;
1252 reg_struct.range.len = reg->mmap_size;
1253 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1255 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1257 VHOST_LOG_CONFIG(ERR,
1258 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1259 i, dev->postcopy_ufd,
1263 VHOST_LOG_CONFIG(INFO,
1264 "\t userfaultfd registered for range : "
1265 "%" PRIx64 " - %" PRIx64 "\n",
1266 (uint64_t)reg_struct.range.start,
1267 (uint64_t)reg_struct.range.start +
1268 (uint64_t)reg_struct.range.len - 1);
1275 for (i = 0; i < dev->nr_vring; i++) {
1276 struct vhost_virtqueue *vq = dev->virtqueue[i];
1278 if (vq->desc || vq->avail || vq->used) {
1280 * If the memory table got updated, the ring addresses
1281 * need to be translated again as virtual addresses have
1284 vring_invalidate(dev, vq);
1286 dev = translate_ring_addresses(dev, i);
1296 dump_guest_pages(dev);
1298 return RTE_VHOST_MSG_RESULT_OK;
1301 free_mem_region(dev);
1304 return RTE_VHOST_MSG_RESULT_ERR;
1308 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1315 if (vq_is_packed(dev))
1316 rings_ok = vq->desc_packed && vq->driver_event &&
1319 rings_ok = vq->desc && vq->avail && vq->used;
1322 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1323 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1327 virtio_is_ready(struct virtio_net *dev)
1329 struct vhost_virtqueue *vq;
1332 if (dev->nr_vring == 0)
1335 for (i = 0; i < dev->nr_vring; i++) {
1336 vq = dev->virtqueue[i];
1338 if (!vq_is_ready(dev, vq))
1342 VHOST_LOG_CONFIG(INFO,
1343 "virtio is now ready for processing.\n");
1348 inflight_mem_alloc(const char *name, size_t size, int *fd)
1352 char fname[20] = "/tmp/memfd-XXXXXX";
1355 #ifdef MEMFD_SUPPORTED
1356 mfd = memfd_create(name, MFD_CLOEXEC);
1361 mfd = mkstemp(fname);
1363 VHOST_LOG_CONFIG(ERR,
1364 "failed to get inflight buffer fd\n");
1371 if (ftruncate(mfd, size) == -1) {
1372 VHOST_LOG_CONFIG(ERR,
1373 "failed to alloc inflight buffer\n");
1378 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1379 if (ptr == MAP_FAILED) {
1380 VHOST_LOG_CONFIG(ERR,
1381 "failed to mmap inflight buffer\n");
1391 get_pervq_shm_size_split(uint16_t queue_size)
1393 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1394 queue_size + sizeof(uint64_t) +
1395 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1399 get_pervq_shm_size_packed(uint16_t queue_size)
1401 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1402 * queue_size + sizeof(uint64_t) +
1403 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1404 INFLIGHT_ALIGNMENT);
1408 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1410 int main_fd __rte_unused)
1412 struct rte_vhost_inflight_info_packed *inflight_packed;
1413 uint64_t pervq_inflight_size, mmap_size;
1414 uint16_t num_queues, queue_size;
1415 struct virtio_net *dev = *pdev;
1419 if (msg->size != sizeof(msg->payload.inflight)) {
1420 VHOST_LOG_CONFIG(ERR,
1421 "invalid get_inflight_fd message size is %d\n",
1423 return RTE_VHOST_MSG_RESULT_ERR;
1426 if (dev->inflight_info == NULL) {
1427 dev->inflight_info = calloc(1,
1428 sizeof(struct inflight_mem_info));
1429 if (!dev->inflight_info) {
1430 VHOST_LOG_CONFIG(ERR,
1431 "failed to alloc dev inflight area\n");
1432 return RTE_VHOST_MSG_RESULT_ERR;
1436 num_queues = msg->payload.inflight.num_queues;
1437 queue_size = msg->payload.inflight.queue_size;
1439 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1440 msg->payload.inflight.num_queues);
1441 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1442 msg->payload.inflight.queue_size);
1444 if (vq_is_packed(dev))
1445 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1447 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1449 mmap_size = num_queues * pervq_inflight_size;
1450 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1452 VHOST_LOG_CONFIG(ERR,
1453 "failed to alloc vhost inflight area\n");
1454 msg->payload.inflight.mmap_size = 0;
1455 return RTE_VHOST_MSG_RESULT_ERR;
1457 memset(addr, 0, mmap_size);
1459 dev->inflight_info->addr = addr;
1460 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1461 dev->inflight_info->fd = msg->fds[0] = fd;
1462 msg->payload.inflight.mmap_offset = 0;
1465 if (vq_is_packed(dev)) {
1466 for (i = 0; i < num_queues; i++) {
1468 (struct rte_vhost_inflight_info_packed *)addr;
1469 inflight_packed->used_wrap_counter = 1;
1470 inflight_packed->old_used_wrap_counter = 1;
1471 for (j = 0; j < queue_size; j++)
1472 inflight_packed->desc[j].next = j + 1;
1473 addr = (void *)((char *)addr + pervq_inflight_size);
1477 VHOST_LOG_CONFIG(INFO,
1478 "send inflight mmap_size: %"PRIu64"\n",
1479 msg->payload.inflight.mmap_size);
1480 VHOST_LOG_CONFIG(INFO,
1481 "send inflight mmap_offset: %"PRIu64"\n",
1482 msg->payload.inflight.mmap_offset);
1483 VHOST_LOG_CONFIG(INFO,
1484 "send inflight fd: %d\n", msg->fds[0]);
1486 return RTE_VHOST_MSG_RESULT_REPLY;
1490 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1491 int main_fd __rte_unused)
1493 uint64_t mmap_size, mmap_offset;
1494 uint16_t num_queues, queue_size;
1495 struct virtio_net *dev = *pdev;
1496 uint32_t pervq_inflight_size;
1497 struct vhost_virtqueue *vq;
1502 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1503 VHOST_LOG_CONFIG(ERR,
1504 "invalid set_inflight_fd message size is %d,fd is %d\n",
1506 return RTE_VHOST_MSG_RESULT_ERR;
1509 mmap_size = msg->payload.inflight.mmap_size;
1510 mmap_offset = msg->payload.inflight.mmap_offset;
1511 num_queues = msg->payload.inflight.num_queues;
1512 queue_size = msg->payload.inflight.queue_size;
1514 if (vq_is_packed(dev))
1515 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1517 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1519 VHOST_LOG_CONFIG(INFO,
1520 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1521 VHOST_LOG_CONFIG(INFO,
1522 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1523 VHOST_LOG_CONFIG(INFO,
1524 "set_inflight_fd num_queues: %u\n", num_queues);
1525 VHOST_LOG_CONFIG(INFO,
1526 "set_inflight_fd queue_size: %u\n", queue_size);
1527 VHOST_LOG_CONFIG(INFO,
1528 "set_inflight_fd fd: %d\n", fd);
1529 VHOST_LOG_CONFIG(INFO,
1530 "set_inflight_fd pervq_inflight_size: %d\n",
1531 pervq_inflight_size);
1533 if (!dev->inflight_info) {
1534 dev->inflight_info = calloc(1,
1535 sizeof(struct inflight_mem_info));
1536 if (dev->inflight_info == NULL) {
1537 VHOST_LOG_CONFIG(ERR,
1538 "failed to alloc dev inflight area\n");
1539 return RTE_VHOST_MSG_RESULT_ERR;
1543 if (dev->inflight_info->addr)
1544 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1546 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1548 if (addr == MAP_FAILED) {
1549 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1550 return RTE_VHOST_MSG_RESULT_ERR;
1553 if (dev->inflight_info->fd)
1554 close(dev->inflight_info->fd);
1556 dev->inflight_info->fd = fd;
1557 dev->inflight_info->addr = addr;
1558 dev->inflight_info->size = mmap_size;
1560 for (i = 0; i < num_queues; i++) {
1561 vq = dev->virtqueue[i];
1562 if (vq_is_packed(dev)) {
1563 vq->inflight_packed = addr;
1564 vq->inflight_packed->desc_num = queue_size;
1566 vq->inflight_split = addr;
1567 vq->inflight_split->desc_num = queue_size;
1569 addr = (void *)((char *)addr + pervq_inflight_size);
1572 return RTE_VHOST_MSG_RESULT_OK;
1576 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1577 int main_fd __rte_unused)
1579 struct virtio_net *dev = *pdev;
1580 struct vhost_vring_file file;
1581 struct vhost_virtqueue *vq;
1584 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1585 if (validate_msg_fds(msg, expected_fds) != 0)
1586 return RTE_VHOST_MSG_RESULT_ERR;
1588 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1589 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1590 file.fd = VIRTIO_INVALID_EVENTFD;
1592 file.fd = msg->fds[0];
1593 VHOST_LOG_CONFIG(INFO,
1594 "vring call idx:%d file:%d\n", file.index, file.fd);
1596 vq = dev->virtqueue[file.index];
1597 if (vq->callfd >= 0)
1600 vq->callfd = file.fd;
1602 return RTE_VHOST_MSG_RESULT_OK;
1605 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1606 struct VhostUserMsg *msg,
1607 int main_fd __rte_unused)
1611 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1612 if (validate_msg_fds(msg, expected_fds) != 0)
1613 return RTE_VHOST_MSG_RESULT_ERR;
1615 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1617 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1619 return RTE_VHOST_MSG_RESULT_OK;
1623 resubmit_desc_compare(const void *a, const void *b)
1625 const struct rte_vhost_resubmit_desc *desc0 = a;
1626 const struct rte_vhost_resubmit_desc *desc1 = b;
1628 if (desc1->counter > desc0->counter)
1635 vhost_check_queue_inflights_split(struct virtio_net *dev,
1636 struct vhost_virtqueue *vq)
1639 uint16_t resubmit_num = 0, last_io, num;
1640 struct vring_used *used = vq->used;
1641 struct rte_vhost_resubmit_info *resubmit;
1642 struct rte_vhost_inflight_info_split *inflight_split;
1644 if (!(dev->protocol_features &
1645 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1646 return RTE_VHOST_MSG_RESULT_OK;
1648 if ((!vq->inflight_split))
1649 return RTE_VHOST_MSG_RESULT_ERR;
1651 if (!vq->inflight_split->version) {
1652 vq->inflight_split->version = INFLIGHT_VERSION;
1653 return RTE_VHOST_MSG_RESULT_OK;
1656 if (vq->resubmit_inflight)
1657 return RTE_VHOST_MSG_RESULT_OK;
1659 inflight_split = vq->inflight_split;
1660 vq->global_counter = 0;
1661 last_io = inflight_split->last_inflight_io;
1663 if (inflight_split->used_idx != used->idx) {
1664 inflight_split->desc[last_io].inflight = 0;
1666 inflight_split->used_idx = used->idx;
1669 for (i = 0; i < inflight_split->desc_num; i++) {
1670 if (inflight_split->desc[i].inflight == 1)
1674 vq->last_avail_idx += resubmit_num;
1677 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1679 VHOST_LOG_CONFIG(ERR,
1680 "failed to allocate memory for resubmit info.\n");
1681 return RTE_VHOST_MSG_RESULT_ERR;
1684 resubmit->resubmit_list = calloc(resubmit_num,
1685 sizeof(struct rte_vhost_resubmit_desc));
1686 if (!resubmit->resubmit_list) {
1687 VHOST_LOG_CONFIG(ERR,
1688 "failed to allocate memory for inflight desc.\n");
1690 return RTE_VHOST_MSG_RESULT_ERR;
1694 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1695 if (vq->inflight_split->desc[i].inflight == 1) {
1696 resubmit->resubmit_list[num].index = i;
1697 resubmit->resubmit_list[num].counter =
1698 inflight_split->desc[i].counter;
1702 resubmit->resubmit_num = num;
1704 if (resubmit->resubmit_num > 1)
1705 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1706 sizeof(struct rte_vhost_resubmit_desc),
1707 resubmit_desc_compare);
1709 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1710 vq->resubmit_inflight = resubmit;
1713 return RTE_VHOST_MSG_RESULT_OK;
1717 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1718 struct vhost_virtqueue *vq)
1721 uint16_t resubmit_num = 0, old_used_idx, num;
1722 struct rte_vhost_resubmit_info *resubmit;
1723 struct rte_vhost_inflight_info_packed *inflight_packed;
1725 if (!(dev->protocol_features &
1726 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1727 return RTE_VHOST_MSG_RESULT_OK;
1729 if ((!vq->inflight_packed))
1730 return RTE_VHOST_MSG_RESULT_ERR;
1732 if (!vq->inflight_packed->version) {
1733 vq->inflight_packed->version = INFLIGHT_VERSION;
1734 return RTE_VHOST_MSG_RESULT_OK;
1737 if (vq->resubmit_inflight)
1738 return RTE_VHOST_MSG_RESULT_OK;
1740 inflight_packed = vq->inflight_packed;
1741 vq->global_counter = 0;
1742 old_used_idx = inflight_packed->old_used_idx;
1744 if (inflight_packed->used_idx != old_used_idx) {
1745 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1746 inflight_packed->old_used_idx =
1747 inflight_packed->used_idx;
1748 inflight_packed->old_used_wrap_counter =
1749 inflight_packed->used_wrap_counter;
1750 inflight_packed->old_free_head =
1751 inflight_packed->free_head;
1753 inflight_packed->used_idx =
1754 inflight_packed->old_used_idx;
1755 inflight_packed->used_wrap_counter =
1756 inflight_packed->old_used_wrap_counter;
1757 inflight_packed->free_head =
1758 inflight_packed->old_free_head;
1762 for (i = 0; i < inflight_packed->desc_num; i++) {
1763 if (inflight_packed->desc[i].inflight == 1)
1768 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1769 if (resubmit == NULL) {
1770 VHOST_LOG_CONFIG(ERR,
1771 "failed to allocate memory for resubmit info.\n");
1772 return RTE_VHOST_MSG_RESULT_ERR;
1775 resubmit->resubmit_list = calloc(resubmit_num,
1776 sizeof(struct rte_vhost_resubmit_desc));
1777 if (resubmit->resubmit_list == NULL) {
1778 VHOST_LOG_CONFIG(ERR,
1779 "failed to allocate memory for resubmit desc.\n");
1781 return RTE_VHOST_MSG_RESULT_ERR;
1785 for (i = 0; i < inflight_packed->desc_num; i++) {
1786 if (vq->inflight_packed->desc[i].inflight == 1) {
1787 resubmit->resubmit_list[num].index = i;
1788 resubmit->resubmit_list[num].counter =
1789 inflight_packed->desc[i].counter;
1793 resubmit->resubmit_num = num;
1795 if (resubmit->resubmit_num > 1)
1796 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1797 sizeof(struct rte_vhost_resubmit_desc),
1798 resubmit_desc_compare);
1800 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1801 vq->resubmit_inflight = resubmit;
1804 return RTE_VHOST_MSG_RESULT_OK;
1808 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1809 int main_fd __rte_unused)
1811 struct virtio_net *dev = *pdev;
1812 struct vhost_vring_file file;
1813 struct vhost_virtqueue *vq;
1816 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1817 if (validate_msg_fds(msg, expected_fds) != 0)
1818 return RTE_VHOST_MSG_RESULT_ERR;
1820 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1821 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1822 file.fd = VIRTIO_INVALID_EVENTFD;
1824 file.fd = msg->fds[0];
1825 VHOST_LOG_CONFIG(INFO,
1826 "vring kick idx:%d file:%d\n", file.index, file.fd);
1828 /* Interpret ring addresses only when ring is started. */
1829 dev = translate_ring_addresses(dev, file.index);
1831 return RTE_VHOST_MSG_RESULT_ERR;
1835 vq = dev->virtqueue[file.index];
1838 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1839 * the ring starts already enabled. Otherwise, it is enabled via
1840 * the SET_VRING_ENABLE message.
1842 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1844 if (dev->notify_ops->vring_state_changed)
1845 dev->notify_ops->vring_state_changed(
1846 dev->vid, file.index, 1);
1849 if (vq->kickfd >= 0)
1851 vq->kickfd = file.fd;
1853 if (vq_is_packed(dev)) {
1854 if (vhost_check_queue_inflights_packed(dev, vq)) {
1855 VHOST_LOG_CONFIG(ERR,
1856 "failed to inflights for vq: %d\n", file.index);
1857 return RTE_VHOST_MSG_RESULT_ERR;
1860 if (vhost_check_queue_inflights_split(dev, vq)) {
1861 VHOST_LOG_CONFIG(ERR,
1862 "failed to inflights for vq: %d\n", file.index);
1863 return RTE_VHOST_MSG_RESULT_ERR;
1867 return RTE_VHOST_MSG_RESULT_OK;
1871 free_zmbufs(struct vhost_virtqueue *vq)
1873 drain_zmbuf_list(vq);
1875 rte_free(vq->zmbufs);
1879 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1882 vhost_user_get_vring_base(struct virtio_net **pdev,
1883 struct VhostUserMsg *msg,
1884 int main_fd __rte_unused)
1886 struct virtio_net *dev = *pdev;
1887 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1890 if (validate_msg_fds(msg, 0) != 0)
1891 return RTE_VHOST_MSG_RESULT_ERR;
1893 /* We have to stop the queue (virtio) if it is running. */
1894 vhost_destroy_device_notify(dev);
1896 dev->flags &= ~VIRTIO_DEV_READY;
1897 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1899 /* Here we are safe to get the indexes */
1900 if (vq_is_packed(dev)) {
1902 * Bit[0:14]: avail index
1903 * Bit[15]: avail wrap counter
1905 val = vq->last_avail_idx & 0x7fff;
1906 val |= vq->avail_wrap_counter << 15;
1907 msg->payload.state.num = val;
1909 msg->payload.state.num = vq->last_avail_idx;
1912 VHOST_LOG_CONFIG(INFO,
1913 "vring base idx:%d file:%d\n", msg->payload.state.index,
1914 msg->payload.state.num);
1916 * Based on current qemu vhost-user implementation, this message is
1917 * sent and only sent in vhost_vring_stop.
1918 * TODO: cleanup the vring, it isn't usable since here.
1920 if (vq->kickfd >= 0)
1923 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1925 if (vq->callfd >= 0)
1928 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1930 vq->signalled_used_valid = false;
1932 if (dev->dequeue_zero_copy)
1934 if (vq_is_packed(dev)) {
1935 rte_free(vq->shadow_used_packed);
1936 vq->shadow_used_packed = NULL;
1938 rte_free(vq->shadow_used_split);
1939 vq->shadow_used_split = NULL;
1942 rte_free(vq->batch_copy_elems);
1943 vq->batch_copy_elems = NULL;
1945 msg->size = sizeof(msg->payload.state);
1948 vring_invalidate(dev, vq);
1950 return RTE_VHOST_MSG_RESULT_REPLY;
1954 * when virtio queues are ready to work, qemu will send us to
1955 * enable the virtio queue pair.
1958 vhost_user_set_vring_enable(struct virtio_net **pdev,
1959 struct VhostUserMsg *msg,
1960 int main_fd __rte_unused)
1962 struct virtio_net *dev = *pdev;
1963 int enable = (int)msg->payload.state.num;
1964 int index = (int)msg->payload.state.index;
1965 struct rte_vdpa_device *vdpa_dev;
1968 if (validate_msg_fds(msg, 0) != 0)
1969 return RTE_VHOST_MSG_RESULT_ERR;
1971 VHOST_LOG_CONFIG(INFO,
1972 "set queue enable: %d to qp idx: %d\n",
1975 did = dev->vdpa_dev_id;
1976 vdpa_dev = rte_vdpa_get_device(did);
1977 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1978 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1980 if (dev->notify_ops->vring_state_changed)
1981 dev->notify_ops->vring_state_changed(dev->vid,
1984 /* On disable, rings have to be stopped being processed. */
1985 if (!enable && dev->dequeue_zero_copy)
1986 drain_zmbuf_list(dev->virtqueue[index]);
1988 dev->virtqueue[index]->enabled = enable;
1990 return RTE_VHOST_MSG_RESULT_OK;
1994 vhost_user_get_protocol_features(struct virtio_net **pdev,
1995 struct VhostUserMsg *msg,
1996 int main_fd __rte_unused)
1998 struct virtio_net *dev = *pdev;
1999 uint64_t features, protocol_features;
2001 if (validate_msg_fds(msg, 0) != 0)
2002 return RTE_VHOST_MSG_RESULT_ERR;
2004 rte_vhost_driver_get_features(dev->ifname, &features);
2005 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2008 * REPLY_ACK protocol feature is only mandatory for now
2009 * for IOMMU feature. If IOMMU is explicitly disabled by the
2010 * application, disable also REPLY_ACK feature for older buggy
2011 * Qemu versions (from v2.7.0 to v2.9.0).
2013 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
2014 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
2016 msg->payload.u64 = protocol_features;
2017 msg->size = sizeof(msg->payload.u64);
2020 return RTE_VHOST_MSG_RESULT_REPLY;
2024 vhost_user_set_protocol_features(struct virtio_net **pdev,
2025 struct VhostUserMsg *msg,
2026 int main_fd __rte_unused)
2028 struct virtio_net *dev = *pdev;
2029 uint64_t protocol_features = msg->payload.u64;
2030 uint64_t slave_protocol_features = 0;
2032 if (validate_msg_fds(msg, 0) != 0)
2033 return RTE_VHOST_MSG_RESULT_ERR;
2035 rte_vhost_driver_get_protocol_features(dev->ifname,
2036 &slave_protocol_features);
2037 if (protocol_features & ~slave_protocol_features) {
2038 VHOST_LOG_CONFIG(ERR,
2039 "(%d) received invalid protocol features.\n",
2041 return RTE_VHOST_MSG_RESULT_ERR;
2044 dev->protocol_features = protocol_features;
2045 VHOST_LOG_CONFIG(INFO,
2046 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2047 dev->protocol_features);
2049 return RTE_VHOST_MSG_RESULT_OK;
2053 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2054 int main_fd __rte_unused)
2056 struct virtio_net *dev = *pdev;
2057 int fd = msg->fds[0];
2061 if (validate_msg_fds(msg, 1) != 0)
2062 return RTE_VHOST_MSG_RESULT_ERR;
2065 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2066 return RTE_VHOST_MSG_RESULT_ERR;
2069 if (msg->size != sizeof(VhostUserLog)) {
2070 VHOST_LOG_CONFIG(ERR,
2071 "invalid log base msg size: %"PRId32" != %d\n",
2072 msg->size, (int)sizeof(VhostUserLog));
2073 return RTE_VHOST_MSG_RESULT_ERR;
2076 size = msg->payload.log.mmap_size;
2077 off = msg->payload.log.mmap_offset;
2079 /* Don't allow mmap_offset to point outside the mmap region */
2081 VHOST_LOG_CONFIG(ERR,
2082 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
2084 return RTE_VHOST_MSG_RESULT_ERR;
2087 VHOST_LOG_CONFIG(INFO,
2088 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2092 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2093 * fail when offset is not page size aligned.
2095 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2097 if (addr == MAP_FAILED) {
2098 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2099 return RTE_VHOST_MSG_RESULT_ERR;
2103 * Free previously mapped log memory on occasionally
2104 * multiple VHOST_USER_SET_LOG_BASE.
2106 if (dev->log_addr) {
2107 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2109 dev->log_addr = (uint64_t)(uintptr_t)addr;
2110 dev->log_base = dev->log_addr + off;
2111 dev->log_size = size;
2114 * The spec is not clear about it (yet), but QEMU doesn't expect
2115 * any payload in the reply.
2120 return RTE_VHOST_MSG_RESULT_REPLY;
2123 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2124 struct VhostUserMsg *msg,
2125 int main_fd __rte_unused)
2127 if (validate_msg_fds(msg, 1) != 0)
2128 return RTE_VHOST_MSG_RESULT_ERR;
2131 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2133 return RTE_VHOST_MSG_RESULT_OK;
2137 * An rarp packet is constructed and broadcasted to notify switches about
2138 * the new location of the migrated VM, so that packets from outside will
2139 * not be lost after migration.
2141 * However, we don't actually "send" a rarp packet here, instead, we set
2142 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2145 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2146 int main_fd __rte_unused)
2148 struct virtio_net *dev = *pdev;
2149 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2150 struct rte_vdpa_device *vdpa_dev;
2153 if (validate_msg_fds(msg, 0) != 0)
2154 return RTE_VHOST_MSG_RESULT_ERR;
2156 VHOST_LOG_CONFIG(DEBUG,
2157 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2158 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2159 memcpy(dev->mac.addr_bytes, mac, 6);
2162 * Set the flag to inject a RARP broadcast packet at
2163 * rte_vhost_dequeue_burst().
2165 * rte_smp_wmb() is for making sure the mac is copied
2166 * before the flag is set.
2169 rte_atomic16_set(&dev->broadcast_rarp, 1);
2170 did = dev->vdpa_dev_id;
2171 vdpa_dev = rte_vdpa_get_device(did);
2172 if (vdpa_dev && vdpa_dev->ops->migration_done)
2173 vdpa_dev->ops->migration_done(dev->vid);
2175 return RTE_VHOST_MSG_RESULT_OK;
2179 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2180 int main_fd __rte_unused)
2182 struct virtio_net *dev = *pdev;
2184 if (validate_msg_fds(msg, 0) != 0)
2185 return RTE_VHOST_MSG_RESULT_ERR;
2187 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2188 msg->payload.u64 > VIRTIO_MAX_MTU) {
2189 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2192 return RTE_VHOST_MSG_RESULT_ERR;
2195 dev->mtu = msg->payload.u64;
2197 return RTE_VHOST_MSG_RESULT_OK;
2201 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2202 int main_fd __rte_unused)
2204 struct virtio_net *dev = *pdev;
2205 int fd = msg->fds[0];
2207 if (validate_msg_fds(msg, 1) != 0)
2208 return RTE_VHOST_MSG_RESULT_ERR;
2211 VHOST_LOG_CONFIG(ERR,
2212 "Invalid file descriptor for slave channel (%d)\n",
2214 return RTE_VHOST_MSG_RESULT_ERR;
2217 if (dev->slave_req_fd >= 0)
2218 close(dev->slave_req_fd);
2220 dev->slave_req_fd = fd;
2222 return RTE_VHOST_MSG_RESULT_OK;
2226 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2228 struct vhost_vring_addr *ra;
2229 uint64_t start, end, len;
2232 end = start + imsg->size;
2234 ra = &vq->ring_addrs;
2235 len = sizeof(struct vring_desc) * vq->size;
2236 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2239 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2240 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2243 len = sizeof(struct vring_used) +
2244 sizeof(struct vring_used_elem) * vq->size;
2245 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2252 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2254 struct vhost_vring_addr *ra;
2255 uint64_t start, end, len;
2258 end = start + imsg->size;
2260 ra = &vq->ring_addrs;
2261 len = sizeof(struct vring_packed_desc) * vq->size;
2262 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2265 len = sizeof(struct vring_packed_desc_event);
2266 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2269 len = sizeof(struct vring_packed_desc_event);
2270 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2276 static int is_vring_iotlb(struct virtio_net *dev,
2277 struct vhost_virtqueue *vq,
2278 struct vhost_iotlb_msg *imsg)
2280 if (vq_is_packed(dev))
2281 return is_vring_iotlb_packed(vq, imsg);
2283 return is_vring_iotlb_split(vq, imsg);
2287 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2288 int main_fd __rte_unused)
2290 struct virtio_net *dev = *pdev;
2291 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2295 if (validate_msg_fds(msg, 0) != 0)
2296 return RTE_VHOST_MSG_RESULT_ERR;
2298 switch (imsg->type) {
2299 case VHOST_IOTLB_UPDATE:
2301 vva = qva_to_vva(dev, imsg->uaddr, &len);
2303 return RTE_VHOST_MSG_RESULT_ERR;
2305 for (i = 0; i < dev->nr_vring; i++) {
2306 struct vhost_virtqueue *vq = dev->virtqueue[i];
2308 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2311 if (is_vring_iotlb(dev, vq, imsg))
2312 *pdev = dev = translate_ring_addresses(dev, i);
2315 case VHOST_IOTLB_INVALIDATE:
2316 for (i = 0; i < dev->nr_vring; i++) {
2317 struct vhost_virtqueue *vq = dev->virtqueue[i];
2319 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2322 if (is_vring_iotlb(dev, vq, imsg))
2323 vring_invalidate(dev, vq);
2327 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2329 return RTE_VHOST_MSG_RESULT_ERR;
2332 return RTE_VHOST_MSG_RESULT_OK;
2336 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2337 struct VhostUserMsg *msg,
2338 int main_fd __rte_unused)
2340 struct virtio_net *dev = *pdev;
2341 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2342 struct uffdio_api api_struct;
2344 if (validate_msg_fds(msg, 0) != 0)
2345 return RTE_VHOST_MSG_RESULT_ERR;
2347 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2349 if (dev->postcopy_ufd == -1) {
2350 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2352 return RTE_VHOST_MSG_RESULT_ERR;
2354 api_struct.api = UFFD_API;
2355 api_struct.features = 0;
2356 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2357 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2359 close(dev->postcopy_ufd);
2360 dev->postcopy_ufd = -1;
2361 return RTE_VHOST_MSG_RESULT_ERR;
2363 msg->fds[0] = dev->postcopy_ufd;
2366 return RTE_VHOST_MSG_RESULT_REPLY;
2368 dev->postcopy_ufd = -1;
2371 return RTE_VHOST_MSG_RESULT_ERR;
2376 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2377 struct VhostUserMsg *msg __rte_unused,
2378 int main_fd __rte_unused)
2380 struct virtio_net *dev = *pdev;
2382 if (validate_msg_fds(msg, 0) != 0)
2383 return RTE_VHOST_MSG_RESULT_ERR;
2385 if (dev->mem && dev->mem->nregions) {
2386 VHOST_LOG_CONFIG(ERR,
2387 "Regions already registered at postcopy-listen\n");
2388 return RTE_VHOST_MSG_RESULT_ERR;
2390 dev->postcopy_listening = 1;
2392 return RTE_VHOST_MSG_RESULT_OK;
2396 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2397 int main_fd __rte_unused)
2399 struct virtio_net *dev = *pdev;
2401 if (validate_msg_fds(msg, 0) != 0)
2402 return RTE_VHOST_MSG_RESULT_ERR;
2404 dev->postcopy_listening = 0;
2405 if (dev->postcopy_ufd >= 0) {
2406 close(dev->postcopy_ufd);
2407 dev->postcopy_ufd = -1;
2410 msg->payload.u64 = 0;
2411 msg->size = sizeof(msg->payload.u64);
2414 return RTE_VHOST_MSG_RESULT_REPLY;
2417 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2418 struct VhostUserMsg *msg,
2420 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2421 [VHOST_USER_NONE] = NULL,
2422 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2423 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2424 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2425 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2426 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2427 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2428 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2429 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2430 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2431 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2432 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2433 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2434 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2435 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2436 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2437 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2438 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2439 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2440 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2441 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2442 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2443 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2444 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2445 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2446 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2447 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2448 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2451 /* return bytes# of read on success or negative val on failure. */
2453 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2457 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2458 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2463 if (msg->size > sizeof(msg->payload)) {
2464 VHOST_LOG_CONFIG(ERR,
2465 "invalid msg size: %d\n", msg->size);
2468 ret = read(sockfd, &msg->payload, msg->size);
2471 if (ret != (int)msg->size) {
2472 VHOST_LOG_CONFIG(ERR,
2473 "read control message failed\n");
2482 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2487 return send_fd_message(sockfd, (char *)msg,
2488 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2492 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2497 msg->flags &= ~VHOST_USER_VERSION_MASK;
2498 msg->flags &= ~VHOST_USER_NEED_REPLY;
2499 msg->flags |= VHOST_USER_VERSION;
2500 msg->flags |= VHOST_USER_REPLY_MASK;
2502 return send_vhost_message(sockfd, msg);
2506 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2510 if (msg->flags & VHOST_USER_NEED_REPLY)
2511 rte_spinlock_lock(&dev->slave_req_lock);
2513 ret = send_vhost_message(dev->slave_req_fd, msg);
2514 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2515 rte_spinlock_unlock(&dev->slave_req_lock);
2521 * Allocate a queue pair if it hasn't been allocated yet
2524 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2525 struct VhostUserMsg *msg)
2529 switch (msg->request.master) {
2530 case VHOST_USER_SET_VRING_KICK:
2531 case VHOST_USER_SET_VRING_CALL:
2532 case VHOST_USER_SET_VRING_ERR:
2533 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2535 case VHOST_USER_SET_VRING_NUM:
2536 case VHOST_USER_SET_VRING_BASE:
2537 case VHOST_USER_SET_VRING_ENABLE:
2538 vring_idx = msg->payload.state.index;
2540 case VHOST_USER_SET_VRING_ADDR:
2541 vring_idx = msg->payload.addr.index;
2547 if (vring_idx >= VHOST_MAX_VRING) {
2548 VHOST_LOG_CONFIG(ERR,
2549 "invalid vring index: %u\n", vring_idx);
2553 if (dev->virtqueue[vring_idx])
2556 return alloc_vring_queue(dev, vring_idx);
2560 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2563 unsigned int vq_num = 0;
2565 while (vq_num < dev->nr_vring) {
2566 struct vhost_virtqueue *vq = dev->virtqueue[i];
2569 rte_spinlock_lock(&vq->access_lock);
2577 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2580 unsigned int vq_num = 0;
2582 while (vq_num < dev->nr_vring) {
2583 struct vhost_virtqueue *vq = dev->virtqueue[i];
2586 rte_spinlock_unlock(&vq->access_lock);
2594 vhost_user_msg_handler(int vid, int fd)
2596 struct virtio_net *dev;
2597 struct VhostUserMsg msg;
2598 struct rte_vdpa_device *vdpa_dev;
2601 int unlock_required = 0;
2605 dev = get_device(vid);
2609 if (!dev->notify_ops) {
2610 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2611 if (!dev->notify_ops) {
2612 VHOST_LOG_CONFIG(ERR,
2613 "failed to get callback ops for driver %s\n",
2619 ret = read_vhost_message(fd, &msg);
2622 VHOST_LOG_CONFIG(ERR,
2623 "vhost read message failed\n");
2625 VHOST_LOG_CONFIG(INFO,
2626 "vhost peer closed\n");
2632 request = msg.request.master;
2633 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2634 vhost_message_str[request]) {
2635 if (request != VHOST_USER_IOTLB_MSG)
2636 VHOST_LOG_CONFIG(INFO, "read message %s\n",
2637 vhost_message_str[request]);
2639 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2640 vhost_message_str[request]);
2642 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2645 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2647 VHOST_LOG_CONFIG(ERR,
2648 "failed to alloc queue\n");
2653 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2654 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2655 * and device is destroyed. destroy_device waits for queues to be
2656 * inactive, so it is safe. Otherwise taking the access_lock
2657 * would cause a dead lock.
2660 case VHOST_USER_SET_FEATURES:
2661 case VHOST_USER_SET_PROTOCOL_FEATURES:
2662 case VHOST_USER_SET_OWNER:
2663 case VHOST_USER_SET_MEM_TABLE:
2664 case VHOST_USER_SET_LOG_BASE:
2665 case VHOST_USER_SET_LOG_FD:
2666 case VHOST_USER_SET_VRING_NUM:
2667 case VHOST_USER_SET_VRING_ADDR:
2668 case VHOST_USER_SET_VRING_BASE:
2669 case VHOST_USER_SET_VRING_KICK:
2670 case VHOST_USER_SET_VRING_CALL:
2671 case VHOST_USER_SET_VRING_ERR:
2672 case VHOST_USER_SET_VRING_ENABLE:
2673 case VHOST_USER_SEND_RARP:
2674 case VHOST_USER_NET_SET_MTU:
2675 case VHOST_USER_SET_SLAVE_REQ_FD:
2676 vhost_user_lock_all_queue_pairs(dev);
2677 unlock_required = 1;
2685 if (dev->extern_ops.pre_msg_handle) {
2686 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2689 case RTE_VHOST_MSG_RESULT_REPLY:
2690 send_vhost_reply(fd, &msg);
2692 case RTE_VHOST_MSG_RESULT_ERR:
2693 case RTE_VHOST_MSG_RESULT_OK:
2695 goto skip_to_post_handle;
2696 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2702 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2703 if (!vhost_message_handlers[request])
2704 goto skip_to_post_handle;
2705 ret = vhost_message_handlers[request](&dev, &msg, fd);
2708 case RTE_VHOST_MSG_RESULT_ERR:
2709 VHOST_LOG_CONFIG(ERR,
2710 "Processing %s failed.\n",
2711 vhost_message_str[request]);
2714 case RTE_VHOST_MSG_RESULT_OK:
2715 VHOST_LOG_CONFIG(DEBUG,
2716 "Processing %s succeeded.\n",
2717 vhost_message_str[request]);
2720 case RTE_VHOST_MSG_RESULT_REPLY:
2721 VHOST_LOG_CONFIG(DEBUG,
2722 "Processing %s succeeded and needs reply.\n",
2723 vhost_message_str[request]);
2724 send_vhost_reply(fd, &msg);
2732 skip_to_post_handle:
2733 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2734 dev->extern_ops.post_msg_handle) {
2735 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2738 case RTE_VHOST_MSG_RESULT_REPLY:
2739 send_vhost_reply(fd, &msg);
2741 case RTE_VHOST_MSG_RESULT_ERR:
2742 case RTE_VHOST_MSG_RESULT_OK:
2744 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2750 if (unlock_required)
2751 vhost_user_unlock_all_queue_pairs(dev);
2753 /* If message was not handled at this stage, treat it as an error */
2755 VHOST_LOG_CONFIG(ERR,
2756 "vhost message (req: %d) was not handled.\n", request);
2757 close_msg_fds(&msg);
2758 ret = RTE_VHOST_MSG_RESULT_ERR;
2762 * If the request required a reply that was already sent,
2763 * this optional reply-ack won't be sent as the
2764 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2766 if (msg.flags & VHOST_USER_NEED_REPLY) {
2767 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2768 msg.size = sizeof(msg.payload.u64);
2770 send_vhost_reply(fd, &msg);
2771 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2772 VHOST_LOG_CONFIG(ERR,
2773 "vhost message handling failed.\n");
2777 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2778 dev->flags |= VIRTIO_DEV_READY;
2780 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2781 if (dev->dequeue_zero_copy) {
2782 VHOST_LOG_CONFIG(INFO,
2783 "dequeue zero copy is enabled\n");
2786 if (dev->notify_ops->new_device(dev->vid) == 0)
2787 dev->flags |= VIRTIO_DEV_RUNNING;
2791 did = dev->vdpa_dev_id;
2792 vdpa_dev = rte_vdpa_get_device(did);
2793 if (vdpa_dev && virtio_is_ready(dev) &&
2794 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2795 msg.request.master == VHOST_USER_SET_VRING_CALL) {
2796 if (vdpa_dev->ops->dev_conf)
2797 vdpa_dev->ops->dev_conf(dev->vid);
2798 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2804 static int process_slave_message_reply(struct virtio_net *dev,
2805 const struct VhostUserMsg *msg)
2807 struct VhostUserMsg msg_reply;
2810 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2813 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2818 if (msg_reply.request.slave != msg->request.slave) {
2819 VHOST_LOG_CONFIG(ERR,
2820 "Received unexpected msg type (%u), expected %u\n",
2821 msg_reply.request.slave, msg->request.slave);
2826 ret = msg_reply.payload.u64 ? -1 : 0;
2829 rte_spinlock_unlock(&dev->slave_req_lock);
2834 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2837 struct VhostUserMsg msg = {
2838 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2839 .flags = VHOST_USER_VERSION,
2840 .size = sizeof(msg.payload.iotlb),
2844 .type = VHOST_IOTLB_MISS,
2848 ret = send_vhost_message(dev->slave_req_fd, &msg);
2850 VHOST_LOG_CONFIG(ERR,
2851 "Failed to send IOTLB miss message (%d)\n",
2860 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
2863 struct VhostUserMsg msg = {
2864 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
2865 .flags = VHOST_USER_VERSION,
2870 msg.flags |= VHOST_USER_NEED_REPLY;
2872 ret = send_vhost_slave_message(dev, &msg);
2874 VHOST_LOG_CONFIG(ERR,
2875 "Failed to send config change (%d)\n",
2880 return process_slave_message_reply(dev, &msg);
2884 rte_vhost_slave_config_change(int vid, bool need_reply)
2886 struct virtio_net *dev;
2888 dev = get_device(vid);
2892 return vhost_user_slave_config_change(dev, need_reply);
2895 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2901 struct VhostUserMsg msg = {
2902 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2903 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2904 .size = sizeof(msg.payload.area),
2906 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2913 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2919 ret = send_vhost_slave_message(dev, &msg);
2921 VHOST_LOG_CONFIG(ERR,
2922 "Failed to set host notifier (%d)\n", ret);
2926 return process_slave_message_reply(dev, &msg);
2929 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2931 struct virtio_net *dev;
2932 struct rte_vdpa_device *vdpa_dev;
2933 int vfio_device_fd, did, ret = 0;
2934 uint64_t offset, size;
2937 dev = get_device(vid);
2941 did = dev->vdpa_dev_id;
2945 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2946 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2947 !(dev->protocol_features &
2948 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2949 !(dev->protocol_features &
2950 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2951 !(dev->protocol_features &
2952 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2955 vdpa_dev = rte_vdpa_get_device(did);
2959 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2960 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2962 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2963 if (vfio_device_fd < 0)
2967 for (i = 0; i < dev->nr_vring; i++) {
2968 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2974 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2975 vfio_device_fd, offset, size) < 0) {
2982 for (i = 0; i < dev->nr_vring; i++) {
2983 vhost_user_slave_set_vring_host_notifier(dev, i, -1,