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);
1149 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
1150 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1151 MAP_SHARED | populate, fd, 0);
1153 if (mmap_addr == MAP_FAILED) {
1154 VHOST_LOG_CONFIG(ERR,
1155 "mmap region %u failed.\n", i);
1159 reg->mmap_addr = mmap_addr;
1160 reg->mmap_size = mmap_size;
1161 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1164 if (dev->dequeue_zero_copy)
1165 if (add_guest_pages(dev, reg, alignment) < 0) {
1166 VHOST_LOG_CONFIG(ERR,
1167 "adding guest pages to region %u failed.\n",
1172 VHOST_LOG_CONFIG(INFO,
1173 "guest memory region %u, size: 0x%" PRIx64 "\n"
1174 "\t guest physical addr: 0x%" PRIx64 "\n"
1175 "\t guest virtual addr: 0x%" PRIx64 "\n"
1176 "\t host virtual addr: 0x%" PRIx64 "\n"
1177 "\t mmap addr : 0x%" PRIx64 "\n"
1178 "\t mmap size : 0x%" PRIx64 "\n"
1179 "\t mmap align: 0x%" PRIx64 "\n"
1180 "\t mmap off : 0x%" PRIx64 "\n",
1182 reg->guest_phys_addr,
1183 reg->guest_user_addr,
1184 reg->host_user_addr,
1185 (uint64_t)(uintptr_t)mmap_addr,
1190 if (dev->postcopy_listening) {
1192 * We haven't a better way right now than sharing
1193 * DPDK's virtual address with Qemu, so that Qemu can
1194 * retrieve the region offset when handling userfaults.
1196 memory->regions[i].userspace_addr =
1197 reg->host_user_addr;
1200 if (dev->postcopy_listening) {
1201 /* Send the addresses back to qemu */
1203 send_vhost_reply(main_fd, msg);
1205 /* Wait for qemu to acknolwedge it's got the addresses
1206 * we've got to wait before we're allowed to generate faults.
1208 VhostUserMsg ack_msg;
1209 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1210 VHOST_LOG_CONFIG(ERR,
1211 "Failed to read qemu ack on postcopy set-mem-table\n");
1215 if (validate_msg_fds(&ack_msg, 0) != 0)
1218 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1219 VHOST_LOG_CONFIG(ERR,
1220 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1221 ack_msg.request.master);
1225 /* Now userfault register and we can use the memory */
1226 for (i = 0; i < memory->nregions; i++) {
1227 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1228 reg = &dev->mem->regions[i];
1229 struct uffdio_register reg_struct;
1232 * Let's register all the mmap'ed area to ensure
1233 * alignment on page boundary.
1235 reg_struct.range.start =
1236 (uint64_t)(uintptr_t)reg->mmap_addr;
1237 reg_struct.range.len = reg->mmap_size;
1238 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1240 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1242 VHOST_LOG_CONFIG(ERR,
1243 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1244 i, dev->postcopy_ufd,
1248 VHOST_LOG_CONFIG(INFO,
1249 "\t userfaultfd registered for range : "
1250 "%" PRIx64 " - %" PRIx64 "\n",
1251 (uint64_t)reg_struct.range.start,
1252 (uint64_t)reg_struct.range.start +
1253 (uint64_t)reg_struct.range.len - 1);
1260 for (i = 0; i < dev->nr_vring; i++) {
1261 struct vhost_virtqueue *vq = dev->virtqueue[i];
1263 if (vq->desc || vq->avail || vq->used) {
1265 * If the memory table got updated, the ring addresses
1266 * need to be translated again as virtual addresses have
1269 vring_invalidate(dev, vq);
1271 dev = translate_ring_addresses(dev, i);
1281 dump_guest_pages(dev);
1283 return RTE_VHOST_MSG_RESULT_OK;
1286 free_mem_region(dev);
1289 return RTE_VHOST_MSG_RESULT_ERR;
1293 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1300 if (vq_is_packed(dev))
1301 rings_ok = vq->desc_packed && vq->driver_event &&
1304 rings_ok = vq->desc && vq->avail && vq->used;
1307 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1308 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1312 virtio_is_ready(struct virtio_net *dev)
1314 struct vhost_virtqueue *vq;
1317 if (dev->nr_vring == 0)
1320 for (i = 0; i < dev->nr_vring; i++) {
1321 vq = dev->virtqueue[i];
1323 if (!vq_is_ready(dev, vq))
1327 VHOST_LOG_CONFIG(INFO,
1328 "virtio is now ready for processing.\n");
1333 inflight_mem_alloc(const char *name, size_t size, int *fd)
1337 char fname[20] = "/tmp/memfd-XXXXXX";
1340 #ifdef MEMFD_SUPPORTED
1341 mfd = memfd_create(name, MFD_CLOEXEC);
1346 mfd = mkstemp(fname);
1348 VHOST_LOG_CONFIG(ERR,
1349 "failed to get inflight buffer fd\n");
1356 if (ftruncate(mfd, size) == -1) {
1357 VHOST_LOG_CONFIG(ERR,
1358 "failed to alloc inflight buffer\n");
1363 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1364 if (ptr == MAP_FAILED) {
1365 VHOST_LOG_CONFIG(ERR,
1366 "failed to mmap inflight buffer\n");
1376 get_pervq_shm_size_split(uint16_t queue_size)
1378 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1379 queue_size + sizeof(uint64_t) +
1380 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1384 get_pervq_shm_size_packed(uint16_t queue_size)
1386 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1387 * queue_size + sizeof(uint64_t) +
1388 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1389 INFLIGHT_ALIGNMENT);
1393 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1395 int main_fd __rte_unused)
1397 struct rte_vhost_inflight_info_packed *inflight_packed;
1398 uint64_t pervq_inflight_size, mmap_size;
1399 uint16_t num_queues, queue_size;
1400 struct virtio_net *dev = *pdev;
1404 if (msg->size != sizeof(msg->payload.inflight)) {
1405 VHOST_LOG_CONFIG(ERR,
1406 "invalid get_inflight_fd message size is %d\n",
1408 return RTE_VHOST_MSG_RESULT_ERR;
1411 if (dev->inflight_info == NULL) {
1412 dev->inflight_info = calloc(1,
1413 sizeof(struct inflight_mem_info));
1414 if (!dev->inflight_info) {
1415 VHOST_LOG_CONFIG(ERR,
1416 "failed to alloc dev inflight area\n");
1417 return RTE_VHOST_MSG_RESULT_ERR;
1421 num_queues = msg->payload.inflight.num_queues;
1422 queue_size = msg->payload.inflight.queue_size;
1424 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1425 msg->payload.inflight.num_queues);
1426 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1427 msg->payload.inflight.queue_size);
1429 if (vq_is_packed(dev))
1430 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1432 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1434 mmap_size = num_queues * pervq_inflight_size;
1435 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1437 VHOST_LOG_CONFIG(ERR,
1438 "failed to alloc vhost inflight area\n");
1439 msg->payload.inflight.mmap_size = 0;
1440 return RTE_VHOST_MSG_RESULT_ERR;
1442 memset(addr, 0, mmap_size);
1444 dev->inflight_info->addr = addr;
1445 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1446 dev->inflight_info->fd = msg->fds[0] = fd;
1447 msg->payload.inflight.mmap_offset = 0;
1450 if (vq_is_packed(dev)) {
1451 for (i = 0; i < num_queues; i++) {
1453 (struct rte_vhost_inflight_info_packed *)addr;
1454 inflight_packed->used_wrap_counter = 1;
1455 inflight_packed->old_used_wrap_counter = 1;
1456 for (j = 0; j < queue_size; j++)
1457 inflight_packed->desc[j].next = j + 1;
1458 addr = (void *)((char *)addr + pervq_inflight_size);
1462 VHOST_LOG_CONFIG(INFO,
1463 "send inflight mmap_size: %"PRIu64"\n",
1464 msg->payload.inflight.mmap_size);
1465 VHOST_LOG_CONFIG(INFO,
1466 "send inflight mmap_offset: %"PRIu64"\n",
1467 msg->payload.inflight.mmap_offset);
1468 VHOST_LOG_CONFIG(INFO,
1469 "send inflight fd: %d\n", msg->fds[0]);
1471 return RTE_VHOST_MSG_RESULT_REPLY;
1475 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1476 int main_fd __rte_unused)
1478 uint64_t mmap_size, mmap_offset;
1479 uint16_t num_queues, queue_size;
1480 struct virtio_net *dev = *pdev;
1481 uint32_t pervq_inflight_size;
1482 struct vhost_virtqueue *vq;
1487 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1488 VHOST_LOG_CONFIG(ERR,
1489 "invalid set_inflight_fd message size is %d,fd is %d\n",
1491 return RTE_VHOST_MSG_RESULT_ERR;
1494 mmap_size = msg->payload.inflight.mmap_size;
1495 mmap_offset = msg->payload.inflight.mmap_offset;
1496 num_queues = msg->payload.inflight.num_queues;
1497 queue_size = msg->payload.inflight.queue_size;
1499 if (vq_is_packed(dev))
1500 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1502 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1504 VHOST_LOG_CONFIG(INFO,
1505 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1506 VHOST_LOG_CONFIG(INFO,
1507 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1508 VHOST_LOG_CONFIG(INFO,
1509 "set_inflight_fd num_queues: %u\n", num_queues);
1510 VHOST_LOG_CONFIG(INFO,
1511 "set_inflight_fd queue_size: %u\n", queue_size);
1512 VHOST_LOG_CONFIG(INFO,
1513 "set_inflight_fd fd: %d\n", fd);
1514 VHOST_LOG_CONFIG(INFO,
1515 "set_inflight_fd pervq_inflight_size: %d\n",
1516 pervq_inflight_size);
1518 if (!dev->inflight_info) {
1519 dev->inflight_info = calloc(1,
1520 sizeof(struct inflight_mem_info));
1521 if (dev->inflight_info == NULL) {
1522 VHOST_LOG_CONFIG(ERR,
1523 "failed to alloc dev inflight area\n");
1524 return RTE_VHOST_MSG_RESULT_ERR;
1528 if (dev->inflight_info->addr)
1529 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1531 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1533 if (addr == MAP_FAILED) {
1534 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1535 return RTE_VHOST_MSG_RESULT_ERR;
1538 if (dev->inflight_info->fd)
1539 close(dev->inflight_info->fd);
1541 dev->inflight_info->fd = fd;
1542 dev->inflight_info->addr = addr;
1543 dev->inflight_info->size = mmap_size;
1545 for (i = 0; i < num_queues; i++) {
1546 vq = dev->virtqueue[i];
1547 if (vq_is_packed(dev)) {
1548 vq->inflight_packed = addr;
1549 vq->inflight_packed->desc_num = queue_size;
1551 vq->inflight_split = addr;
1552 vq->inflight_split->desc_num = queue_size;
1554 addr = (void *)((char *)addr + pervq_inflight_size);
1557 return RTE_VHOST_MSG_RESULT_OK;
1561 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1562 int main_fd __rte_unused)
1564 struct virtio_net *dev = *pdev;
1565 struct vhost_vring_file file;
1566 struct vhost_virtqueue *vq;
1569 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1570 if (validate_msg_fds(msg, expected_fds) != 0)
1571 return RTE_VHOST_MSG_RESULT_ERR;
1573 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1574 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1575 file.fd = VIRTIO_INVALID_EVENTFD;
1577 file.fd = msg->fds[0];
1578 VHOST_LOG_CONFIG(INFO,
1579 "vring call idx:%d file:%d\n", file.index, file.fd);
1581 vq = dev->virtqueue[file.index];
1582 if (vq->callfd >= 0)
1585 vq->callfd = file.fd;
1587 return RTE_VHOST_MSG_RESULT_OK;
1590 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1591 struct VhostUserMsg *msg,
1592 int main_fd __rte_unused)
1596 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1597 if (validate_msg_fds(msg, expected_fds) != 0)
1598 return RTE_VHOST_MSG_RESULT_ERR;
1600 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1602 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1604 return RTE_VHOST_MSG_RESULT_OK;
1608 resubmit_desc_compare(const void *a, const void *b)
1610 const struct rte_vhost_resubmit_desc *desc0 = a;
1611 const struct rte_vhost_resubmit_desc *desc1 = b;
1613 if (desc1->counter > desc0->counter)
1620 vhost_check_queue_inflights_split(struct virtio_net *dev,
1621 struct vhost_virtqueue *vq)
1624 uint16_t resubmit_num = 0, last_io, num;
1625 struct vring_used *used = vq->used;
1626 struct rte_vhost_resubmit_info *resubmit;
1627 struct rte_vhost_inflight_info_split *inflight_split;
1629 if (!(dev->protocol_features &
1630 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1631 return RTE_VHOST_MSG_RESULT_OK;
1633 if ((!vq->inflight_split))
1634 return RTE_VHOST_MSG_RESULT_ERR;
1636 if (!vq->inflight_split->version) {
1637 vq->inflight_split->version = INFLIGHT_VERSION;
1638 return RTE_VHOST_MSG_RESULT_OK;
1641 if (vq->resubmit_inflight)
1642 return RTE_VHOST_MSG_RESULT_OK;
1644 inflight_split = vq->inflight_split;
1645 vq->global_counter = 0;
1646 last_io = inflight_split->last_inflight_io;
1648 if (inflight_split->used_idx != used->idx) {
1649 inflight_split->desc[last_io].inflight = 0;
1651 inflight_split->used_idx = used->idx;
1654 for (i = 0; i < inflight_split->desc_num; i++) {
1655 if (inflight_split->desc[i].inflight == 1)
1659 vq->last_avail_idx += resubmit_num;
1662 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1664 VHOST_LOG_CONFIG(ERR,
1665 "failed to allocate memory for resubmit info.\n");
1666 return RTE_VHOST_MSG_RESULT_ERR;
1669 resubmit->resubmit_list = calloc(resubmit_num,
1670 sizeof(struct rte_vhost_resubmit_desc));
1671 if (!resubmit->resubmit_list) {
1672 VHOST_LOG_CONFIG(ERR,
1673 "failed to allocate memory for inflight desc.\n");
1675 return RTE_VHOST_MSG_RESULT_ERR;
1679 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1680 if (vq->inflight_split->desc[i].inflight == 1) {
1681 resubmit->resubmit_list[num].index = i;
1682 resubmit->resubmit_list[num].counter =
1683 inflight_split->desc[i].counter;
1687 resubmit->resubmit_num = num;
1689 if (resubmit->resubmit_num > 1)
1690 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1691 sizeof(struct rte_vhost_resubmit_desc),
1692 resubmit_desc_compare);
1694 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1695 vq->resubmit_inflight = resubmit;
1698 return RTE_VHOST_MSG_RESULT_OK;
1702 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1703 struct vhost_virtqueue *vq)
1706 uint16_t resubmit_num = 0, old_used_idx, num;
1707 struct rte_vhost_resubmit_info *resubmit;
1708 struct rte_vhost_inflight_info_packed *inflight_packed;
1710 if (!(dev->protocol_features &
1711 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1712 return RTE_VHOST_MSG_RESULT_OK;
1714 if ((!vq->inflight_packed))
1715 return RTE_VHOST_MSG_RESULT_ERR;
1717 if (!vq->inflight_packed->version) {
1718 vq->inflight_packed->version = INFLIGHT_VERSION;
1719 return RTE_VHOST_MSG_RESULT_OK;
1722 if (vq->resubmit_inflight)
1723 return RTE_VHOST_MSG_RESULT_OK;
1725 inflight_packed = vq->inflight_packed;
1726 vq->global_counter = 0;
1727 old_used_idx = inflight_packed->old_used_idx;
1729 if (inflight_packed->used_idx != old_used_idx) {
1730 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1731 inflight_packed->old_used_idx =
1732 inflight_packed->used_idx;
1733 inflight_packed->old_used_wrap_counter =
1734 inflight_packed->used_wrap_counter;
1735 inflight_packed->old_free_head =
1736 inflight_packed->free_head;
1738 inflight_packed->used_idx =
1739 inflight_packed->old_used_idx;
1740 inflight_packed->used_wrap_counter =
1741 inflight_packed->old_used_wrap_counter;
1742 inflight_packed->free_head =
1743 inflight_packed->old_free_head;
1747 for (i = 0; i < inflight_packed->desc_num; i++) {
1748 if (inflight_packed->desc[i].inflight == 1)
1753 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1754 if (resubmit == NULL) {
1755 VHOST_LOG_CONFIG(ERR,
1756 "failed to allocate memory for resubmit info.\n");
1757 return RTE_VHOST_MSG_RESULT_ERR;
1760 resubmit->resubmit_list = calloc(resubmit_num,
1761 sizeof(struct rte_vhost_resubmit_desc));
1762 if (resubmit->resubmit_list == NULL) {
1763 VHOST_LOG_CONFIG(ERR,
1764 "failed to allocate memory for resubmit desc.\n");
1766 return RTE_VHOST_MSG_RESULT_ERR;
1770 for (i = 0; i < inflight_packed->desc_num; i++) {
1771 if (vq->inflight_packed->desc[i].inflight == 1) {
1772 resubmit->resubmit_list[num].index = i;
1773 resubmit->resubmit_list[num].counter =
1774 inflight_packed->desc[i].counter;
1778 resubmit->resubmit_num = num;
1780 if (resubmit->resubmit_num > 1)
1781 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1782 sizeof(struct rte_vhost_resubmit_desc),
1783 resubmit_desc_compare);
1785 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1786 vq->resubmit_inflight = resubmit;
1789 return RTE_VHOST_MSG_RESULT_OK;
1793 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1794 int main_fd __rte_unused)
1796 struct virtio_net *dev = *pdev;
1797 struct vhost_vring_file file;
1798 struct vhost_virtqueue *vq;
1801 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1802 if (validate_msg_fds(msg, expected_fds) != 0)
1803 return RTE_VHOST_MSG_RESULT_ERR;
1805 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1806 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1807 file.fd = VIRTIO_INVALID_EVENTFD;
1809 file.fd = msg->fds[0];
1810 VHOST_LOG_CONFIG(INFO,
1811 "vring kick idx:%d file:%d\n", file.index, file.fd);
1813 /* Interpret ring addresses only when ring is started. */
1814 dev = translate_ring_addresses(dev, file.index);
1816 return RTE_VHOST_MSG_RESULT_ERR;
1820 vq = dev->virtqueue[file.index];
1823 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1824 * the ring starts already enabled. Otherwise, it is enabled via
1825 * the SET_VRING_ENABLE message.
1827 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1829 if (dev->notify_ops->vring_state_changed)
1830 dev->notify_ops->vring_state_changed(
1831 dev->vid, file.index, 1);
1834 if (vq->kickfd >= 0)
1836 vq->kickfd = file.fd;
1838 if (vq_is_packed(dev)) {
1839 if (vhost_check_queue_inflights_packed(dev, vq)) {
1840 VHOST_LOG_CONFIG(ERR,
1841 "failed to inflights for vq: %d\n", file.index);
1842 return RTE_VHOST_MSG_RESULT_ERR;
1845 if (vhost_check_queue_inflights_split(dev, vq)) {
1846 VHOST_LOG_CONFIG(ERR,
1847 "failed to inflights for vq: %d\n", file.index);
1848 return RTE_VHOST_MSG_RESULT_ERR;
1852 return RTE_VHOST_MSG_RESULT_OK;
1856 free_zmbufs(struct vhost_virtqueue *vq)
1858 drain_zmbuf_list(vq);
1860 rte_free(vq->zmbufs);
1864 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1867 vhost_user_get_vring_base(struct virtio_net **pdev,
1868 struct VhostUserMsg *msg,
1869 int main_fd __rte_unused)
1871 struct virtio_net *dev = *pdev;
1872 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1875 if (validate_msg_fds(msg, 0) != 0)
1876 return RTE_VHOST_MSG_RESULT_ERR;
1878 /* We have to stop the queue (virtio) if it is running. */
1879 vhost_destroy_device_notify(dev);
1881 dev->flags &= ~VIRTIO_DEV_READY;
1882 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1884 /* Here we are safe to get the indexes */
1885 if (vq_is_packed(dev)) {
1887 * Bit[0:14]: avail index
1888 * Bit[15]: avail wrap counter
1890 val = vq->last_avail_idx & 0x7fff;
1891 val |= vq->avail_wrap_counter << 15;
1892 msg->payload.state.num = val;
1894 msg->payload.state.num = vq->last_avail_idx;
1897 VHOST_LOG_CONFIG(INFO,
1898 "vring base idx:%d file:%d\n", msg->payload.state.index,
1899 msg->payload.state.num);
1901 * Based on current qemu vhost-user implementation, this message is
1902 * sent and only sent in vhost_vring_stop.
1903 * TODO: cleanup the vring, it isn't usable since here.
1905 if (vq->kickfd >= 0)
1908 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1910 if (vq->callfd >= 0)
1913 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1915 vq->signalled_used_valid = false;
1917 if (dev->dequeue_zero_copy)
1919 if (vq_is_packed(dev)) {
1920 rte_free(vq->shadow_used_packed);
1921 vq->shadow_used_packed = NULL;
1923 rte_free(vq->shadow_used_split);
1924 vq->shadow_used_split = NULL;
1927 rte_free(vq->batch_copy_elems);
1928 vq->batch_copy_elems = NULL;
1930 msg->size = sizeof(msg->payload.state);
1933 vring_invalidate(dev, vq);
1935 return RTE_VHOST_MSG_RESULT_REPLY;
1939 * when virtio queues are ready to work, qemu will send us to
1940 * enable the virtio queue pair.
1943 vhost_user_set_vring_enable(struct virtio_net **pdev,
1944 struct VhostUserMsg *msg,
1945 int main_fd __rte_unused)
1947 struct virtio_net *dev = *pdev;
1948 int enable = (int)msg->payload.state.num;
1949 int index = (int)msg->payload.state.index;
1950 struct rte_vdpa_device *vdpa_dev;
1953 if (validate_msg_fds(msg, 0) != 0)
1954 return RTE_VHOST_MSG_RESULT_ERR;
1956 VHOST_LOG_CONFIG(INFO,
1957 "set queue enable: %d to qp idx: %d\n",
1960 did = dev->vdpa_dev_id;
1961 vdpa_dev = rte_vdpa_get_device(did);
1962 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1963 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1965 if (dev->notify_ops->vring_state_changed)
1966 dev->notify_ops->vring_state_changed(dev->vid,
1969 /* On disable, rings have to be stopped being processed. */
1970 if (!enable && dev->dequeue_zero_copy)
1971 drain_zmbuf_list(dev->virtqueue[index]);
1973 dev->virtqueue[index]->enabled = enable;
1975 return RTE_VHOST_MSG_RESULT_OK;
1979 vhost_user_get_protocol_features(struct virtio_net **pdev,
1980 struct VhostUserMsg *msg,
1981 int main_fd __rte_unused)
1983 struct virtio_net *dev = *pdev;
1984 uint64_t features, protocol_features;
1986 if (validate_msg_fds(msg, 0) != 0)
1987 return RTE_VHOST_MSG_RESULT_ERR;
1989 rte_vhost_driver_get_features(dev->ifname, &features);
1990 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1993 * REPLY_ACK protocol feature is only mandatory for now
1994 * for IOMMU feature. If IOMMU is explicitly disabled by the
1995 * application, disable also REPLY_ACK feature for older buggy
1996 * Qemu versions (from v2.7.0 to v2.9.0).
1998 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1999 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
2001 msg->payload.u64 = protocol_features;
2002 msg->size = sizeof(msg->payload.u64);
2005 return RTE_VHOST_MSG_RESULT_REPLY;
2009 vhost_user_set_protocol_features(struct virtio_net **pdev,
2010 struct VhostUserMsg *msg,
2011 int main_fd __rte_unused)
2013 struct virtio_net *dev = *pdev;
2014 uint64_t protocol_features = msg->payload.u64;
2015 uint64_t slave_protocol_features = 0;
2017 if (validate_msg_fds(msg, 0) != 0)
2018 return RTE_VHOST_MSG_RESULT_ERR;
2020 rte_vhost_driver_get_protocol_features(dev->ifname,
2021 &slave_protocol_features);
2022 if (protocol_features & ~slave_protocol_features) {
2023 VHOST_LOG_CONFIG(ERR,
2024 "(%d) received invalid protocol features.\n",
2026 return RTE_VHOST_MSG_RESULT_ERR;
2029 dev->protocol_features = protocol_features;
2030 VHOST_LOG_CONFIG(INFO,
2031 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2032 dev->protocol_features);
2034 return RTE_VHOST_MSG_RESULT_OK;
2038 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2039 int main_fd __rte_unused)
2041 struct virtio_net *dev = *pdev;
2042 int fd = msg->fds[0];
2046 if (validate_msg_fds(msg, 1) != 0)
2047 return RTE_VHOST_MSG_RESULT_ERR;
2050 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2051 return RTE_VHOST_MSG_RESULT_ERR;
2054 if (msg->size != sizeof(VhostUserLog)) {
2055 VHOST_LOG_CONFIG(ERR,
2056 "invalid log base msg size: %"PRId32" != %d\n",
2057 msg->size, (int)sizeof(VhostUserLog));
2058 return RTE_VHOST_MSG_RESULT_ERR;
2061 size = msg->payload.log.mmap_size;
2062 off = msg->payload.log.mmap_offset;
2064 /* Don't allow mmap_offset to point outside the mmap region */
2066 VHOST_LOG_CONFIG(ERR,
2067 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
2069 return RTE_VHOST_MSG_RESULT_ERR;
2072 VHOST_LOG_CONFIG(INFO,
2073 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2077 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2078 * fail when offset is not page size aligned.
2080 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2082 if (addr == MAP_FAILED) {
2083 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2084 return RTE_VHOST_MSG_RESULT_ERR;
2088 * Free previously mapped log memory on occasionally
2089 * multiple VHOST_USER_SET_LOG_BASE.
2091 if (dev->log_addr) {
2092 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2094 dev->log_addr = (uint64_t)(uintptr_t)addr;
2095 dev->log_base = dev->log_addr + off;
2096 dev->log_size = size;
2099 * The spec is not clear about it (yet), but QEMU doesn't expect
2100 * any payload in the reply.
2105 return RTE_VHOST_MSG_RESULT_REPLY;
2108 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2109 struct VhostUserMsg *msg,
2110 int main_fd __rte_unused)
2112 if (validate_msg_fds(msg, 1) != 0)
2113 return RTE_VHOST_MSG_RESULT_ERR;
2116 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2118 return RTE_VHOST_MSG_RESULT_OK;
2122 * An rarp packet is constructed and broadcasted to notify switches about
2123 * the new location of the migrated VM, so that packets from outside will
2124 * not be lost after migration.
2126 * However, we don't actually "send" a rarp packet here, instead, we set
2127 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2130 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2131 int main_fd __rte_unused)
2133 struct virtio_net *dev = *pdev;
2134 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2135 struct rte_vdpa_device *vdpa_dev;
2138 if (validate_msg_fds(msg, 0) != 0)
2139 return RTE_VHOST_MSG_RESULT_ERR;
2141 VHOST_LOG_CONFIG(DEBUG,
2142 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2143 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2144 memcpy(dev->mac.addr_bytes, mac, 6);
2147 * Set the flag to inject a RARP broadcast packet at
2148 * rte_vhost_dequeue_burst().
2150 * rte_smp_wmb() is for making sure the mac is copied
2151 * before the flag is set.
2154 rte_atomic16_set(&dev->broadcast_rarp, 1);
2155 did = dev->vdpa_dev_id;
2156 vdpa_dev = rte_vdpa_get_device(did);
2157 if (vdpa_dev && vdpa_dev->ops->migration_done)
2158 vdpa_dev->ops->migration_done(dev->vid);
2160 return RTE_VHOST_MSG_RESULT_OK;
2164 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2165 int main_fd __rte_unused)
2167 struct virtio_net *dev = *pdev;
2169 if (validate_msg_fds(msg, 0) != 0)
2170 return RTE_VHOST_MSG_RESULT_ERR;
2172 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2173 msg->payload.u64 > VIRTIO_MAX_MTU) {
2174 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2177 return RTE_VHOST_MSG_RESULT_ERR;
2180 dev->mtu = msg->payload.u64;
2182 return RTE_VHOST_MSG_RESULT_OK;
2186 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2187 int main_fd __rte_unused)
2189 struct virtio_net *dev = *pdev;
2190 int fd = msg->fds[0];
2192 if (validate_msg_fds(msg, 1) != 0)
2193 return RTE_VHOST_MSG_RESULT_ERR;
2196 VHOST_LOG_CONFIG(ERR,
2197 "Invalid file descriptor for slave channel (%d)\n",
2199 return RTE_VHOST_MSG_RESULT_ERR;
2202 if (dev->slave_req_fd >= 0)
2203 close(dev->slave_req_fd);
2205 dev->slave_req_fd = fd;
2207 return RTE_VHOST_MSG_RESULT_OK;
2211 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2213 struct vhost_vring_addr *ra;
2214 uint64_t start, end, len;
2217 end = start + imsg->size;
2219 ra = &vq->ring_addrs;
2220 len = sizeof(struct vring_desc) * vq->size;
2221 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2224 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2225 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2228 len = sizeof(struct vring_used) +
2229 sizeof(struct vring_used_elem) * vq->size;
2230 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2237 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2239 struct vhost_vring_addr *ra;
2240 uint64_t start, end, len;
2243 end = start + imsg->size;
2245 ra = &vq->ring_addrs;
2246 len = sizeof(struct vring_packed_desc) * vq->size;
2247 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2250 len = sizeof(struct vring_packed_desc_event);
2251 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2254 len = sizeof(struct vring_packed_desc_event);
2255 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2261 static int is_vring_iotlb(struct virtio_net *dev,
2262 struct vhost_virtqueue *vq,
2263 struct vhost_iotlb_msg *imsg)
2265 if (vq_is_packed(dev))
2266 return is_vring_iotlb_packed(vq, imsg);
2268 return is_vring_iotlb_split(vq, imsg);
2272 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2273 int main_fd __rte_unused)
2275 struct virtio_net *dev = *pdev;
2276 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2280 if (validate_msg_fds(msg, 0) != 0)
2281 return RTE_VHOST_MSG_RESULT_ERR;
2283 switch (imsg->type) {
2284 case VHOST_IOTLB_UPDATE:
2286 vva = qva_to_vva(dev, imsg->uaddr, &len);
2288 return RTE_VHOST_MSG_RESULT_ERR;
2290 for (i = 0; i < dev->nr_vring; i++) {
2291 struct vhost_virtqueue *vq = dev->virtqueue[i];
2293 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2296 if (is_vring_iotlb(dev, vq, imsg))
2297 *pdev = dev = translate_ring_addresses(dev, i);
2300 case VHOST_IOTLB_INVALIDATE:
2301 for (i = 0; i < dev->nr_vring; i++) {
2302 struct vhost_virtqueue *vq = dev->virtqueue[i];
2304 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2307 if (is_vring_iotlb(dev, vq, imsg))
2308 vring_invalidate(dev, vq);
2312 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2314 return RTE_VHOST_MSG_RESULT_ERR;
2317 return RTE_VHOST_MSG_RESULT_OK;
2321 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2322 struct VhostUserMsg *msg,
2323 int main_fd __rte_unused)
2325 struct virtio_net *dev = *pdev;
2326 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2327 struct uffdio_api api_struct;
2329 if (validate_msg_fds(msg, 0) != 0)
2330 return RTE_VHOST_MSG_RESULT_ERR;
2332 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2334 if (dev->postcopy_ufd == -1) {
2335 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2337 return RTE_VHOST_MSG_RESULT_ERR;
2339 api_struct.api = UFFD_API;
2340 api_struct.features = 0;
2341 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2342 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2344 close(dev->postcopy_ufd);
2345 dev->postcopy_ufd = -1;
2346 return RTE_VHOST_MSG_RESULT_ERR;
2348 msg->fds[0] = dev->postcopy_ufd;
2351 return RTE_VHOST_MSG_RESULT_REPLY;
2353 dev->postcopy_ufd = -1;
2356 return RTE_VHOST_MSG_RESULT_ERR;
2361 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2362 struct VhostUserMsg *msg __rte_unused,
2363 int main_fd __rte_unused)
2365 struct virtio_net *dev = *pdev;
2367 if (validate_msg_fds(msg, 0) != 0)
2368 return RTE_VHOST_MSG_RESULT_ERR;
2370 if (dev->mem && dev->mem->nregions) {
2371 VHOST_LOG_CONFIG(ERR,
2372 "Regions already registered at postcopy-listen\n");
2373 return RTE_VHOST_MSG_RESULT_ERR;
2375 dev->postcopy_listening = 1;
2377 return RTE_VHOST_MSG_RESULT_OK;
2381 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2382 int main_fd __rte_unused)
2384 struct virtio_net *dev = *pdev;
2386 if (validate_msg_fds(msg, 0) != 0)
2387 return RTE_VHOST_MSG_RESULT_ERR;
2389 dev->postcopy_listening = 0;
2390 if (dev->postcopy_ufd >= 0) {
2391 close(dev->postcopy_ufd);
2392 dev->postcopy_ufd = -1;
2395 msg->payload.u64 = 0;
2396 msg->size = sizeof(msg->payload.u64);
2399 return RTE_VHOST_MSG_RESULT_REPLY;
2402 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2403 struct VhostUserMsg *msg,
2405 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2406 [VHOST_USER_NONE] = NULL,
2407 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2408 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2409 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2410 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2411 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2412 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2413 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2414 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2415 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2416 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2417 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2418 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2419 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2420 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2421 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2422 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2423 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2424 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2425 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2426 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2427 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2428 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2429 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2430 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2431 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2432 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2433 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2436 /* return bytes# of read on success or negative val on failure. */
2438 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2442 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2443 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2448 if (msg->size > sizeof(msg->payload)) {
2449 VHOST_LOG_CONFIG(ERR,
2450 "invalid msg size: %d\n", msg->size);
2453 ret = read(sockfd, &msg->payload, msg->size);
2456 if (ret != (int)msg->size) {
2457 VHOST_LOG_CONFIG(ERR,
2458 "read control message failed\n");
2467 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2472 return send_fd_message(sockfd, (char *)msg,
2473 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2477 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2482 msg->flags &= ~VHOST_USER_VERSION_MASK;
2483 msg->flags &= ~VHOST_USER_NEED_REPLY;
2484 msg->flags |= VHOST_USER_VERSION;
2485 msg->flags |= VHOST_USER_REPLY_MASK;
2487 return send_vhost_message(sockfd, msg);
2491 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2495 if (msg->flags & VHOST_USER_NEED_REPLY)
2496 rte_spinlock_lock(&dev->slave_req_lock);
2498 ret = send_vhost_message(dev->slave_req_fd, msg);
2499 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2500 rte_spinlock_unlock(&dev->slave_req_lock);
2506 * Allocate a queue pair if it hasn't been allocated yet
2509 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2510 struct VhostUserMsg *msg)
2514 switch (msg->request.master) {
2515 case VHOST_USER_SET_VRING_KICK:
2516 case VHOST_USER_SET_VRING_CALL:
2517 case VHOST_USER_SET_VRING_ERR:
2518 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2520 case VHOST_USER_SET_VRING_NUM:
2521 case VHOST_USER_SET_VRING_BASE:
2522 case VHOST_USER_SET_VRING_ENABLE:
2523 vring_idx = msg->payload.state.index;
2525 case VHOST_USER_SET_VRING_ADDR:
2526 vring_idx = msg->payload.addr.index;
2532 if (vring_idx >= VHOST_MAX_VRING) {
2533 VHOST_LOG_CONFIG(ERR,
2534 "invalid vring index: %u\n", vring_idx);
2538 if (dev->virtqueue[vring_idx])
2541 return alloc_vring_queue(dev, vring_idx);
2545 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2548 unsigned int vq_num = 0;
2550 while (vq_num < dev->nr_vring) {
2551 struct vhost_virtqueue *vq = dev->virtqueue[i];
2554 rte_spinlock_lock(&vq->access_lock);
2562 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2565 unsigned int vq_num = 0;
2567 while (vq_num < dev->nr_vring) {
2568 struct vhost_virtqueue *vq = dev->virtqueue[i];
2571 rte_spinlock_unlock(&vq->access_lock);
2579 vhost_user_msg_handler(int vid, int fd)
2581 struct virtio_net *dev;
2582 struct VhostUserMsg msg;
2583 struct rte_vdpa_device *vdpa_dev;
2586 int unlock_required = 0;
2590 dev = get_device(vid);
2594 if (!dev->notify_ops) {
2595 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2596 if (!dev->notify_ops) {
2597 VHOST_LOG_CONFIG(ERR,
2598 "failed to get callback ops for driver %s\n",
2604 ret = read_vhost_message(fd, &msg);
2607 VHOST_LOG_CONFIG(ERR,
2608 "vhost read message failed\n");
2610 VHOST_LOG_CONFIG(INFO,
2611 "vhost peer closed\n");
2617 request = msg.request.master;
2618 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2619 vhost_message_str[request]) {
2620 if (request != VHOST_USER_IOTLB_MSG)
2621 VHOST_LOG_CONFIG(INFO, "read message %s\n",
2622 vhost_message_str[request]);
2624 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2625 vhost_message_str[request]);
2627 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2630 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2632 VHOST_LOG_CONFIG(ERR,
2633 "failed to alloc queue\n");
2638 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2639 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2640 * and device is destroyed. destroy_device waits for queues to be
2641 * inactive, so it is safe. Otherwise taking the access_lock
2642 * would cause a dead lock.
2645 case VHOST_USER_SET_FEATURES:
2646 case VHOST_USER_SET_PROTOCOL_FEATURES:
2647 case VHOST_USER_SET_OWNER:
2648 case VHOST_USER_SET_MEM_TABLE:
2649 case VHOST_USER_SET_LOG_BASE:
2650 case VHOST_USER_SET_LOG_FD:
2651 case VHOST_USER_SET_VRING_NUM:
2652 case VHOST_USER_SET_VRING_ADDR:
2653 case VHOST_USER_SET_VRING_BASE:
2654 case VHOST_USER_SET_VRING_KICK:
2655 case VHOST_USER_SET_VRING_CALL:
2656 case VHOST_USER_SET_VRING_ERR:
2657 case VHOST_USER_SET_VRING_ENABLE:
2658 case VHOST_USER_SEND_RARP:
2659 case VHOST_USER_NET_SET_MTU:
2660 case VHOST_USER_SET_SLAVE_REQ_FD:
2661 vhost_user_lock_all_queue_pairs(dev);
2662 unlock_required = 1;
2670 if (dev->extern_ops.pre_msg_handle) {
2671 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2674 case RTE_VHOST_MSG_RESULT_REPLY:
2675 send_vhost_reply(fd, &msg);
2677 case RTE_VHOST_MSG_RESULT_ERR:
2678 case RTE_VHOST_MSG_RESULT_OK:
2680 goto skip_to_post_handle;
2681 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2687 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2688 if (!vhost_message_handlers[request])
2689 goto skip_to_post_handle;
2690 ret = vhost_message_handlers[request](&dev, &msg, fd);
2693 case RTE_VHOST_MSG_RESULT_ERR:
2694 VHOST_LOG_CONFIG(ERR,
2695 "Processing %s failed.\n",
2696 vhost_message_str[request]);
2699 case RTE_VHOST_MSG_RESULT_OK:
2700 VHOST_LOG_CONFIG(DEBUG,
2701 "Processing %s succeeded.\n",
2702 vhost_message_str[request]);
2705 case RTE_VHOST_MSG_RESULT_REPLY:
2706 VHOST_LOG_CONFIG(DEBUG,
2707 "Processing %s succeeded and needs reply.\n",
2708 vhost_message_str[request]);
2709 send_vhost_reply(fd, &msg);
2717 skip_to_post_handle:
2718 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2719 dev->extern_ops.post_msg_handle) {
2720 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2723 case RTE_VHOST_MSG_RESULT_REPLY:
2724 send_vhost_reply(fd, &msg);
2726 case RTE_VHOST_MSG_RESULT_ERR:
2727 case RTE_VHOST_MSG_RESULT_OK:
2729 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2735 if (unlock_required)
2736 vhost_user_unlock_all_queue_pairs(dev);
2738 /* If message was not handled at this stage, treat it as an error */
2740 VHOST_LOG_CONFIG(ERR,
2741 "vhost message (req: %d) was not handled.\n", request);
2742 close_msg_fds(&msg);
2743 ret = RTE_VHOST_MSG_RESULT_ERR;
2747 * If the request required a reply that was already sent,
2748 * this optional reply-ack won't be sent as the
2749 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2751 if (msg.flags & VHOST_USER_NEED_REPLY) {
2752 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2753 msg.size = sizeof(msg.payload.u64);
2755 send_vhost_reply(fd, &msg);
2756 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2757 VHOST_LOG_CONFIG(ERR,
2758 "vhost message handling failed.\n");
2762 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2763 dev->flags |= VIRTIO_DEV_READY;
2765 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2766 if (dev->dequeue_zero_copy) {
2767 VHOST_LOG_CONFIG(INFO,
2768 "dequeue zero copy is enabled\n");
2771 if (dev->notify_ops->new_device(dev->vid) == 0)
2772 dev->flags |= VIRTIO_DEV_RUNNING;
2776 did = dev->vdpa_dev_id;
2777 vdpa_dev = rte_vdpa_get_device(did);
2778 if (vdpa_dev && virtio_is_ready(dev) &&
2779 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2780 msg.request.master == VHOST_USER_SET_VRING_CALL) {
2781 if (vdpa_dev->ops->dev_conf)
2782 vdpa_dev->ops->dev_conf(dev->vid);
2783 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2789 static int process_slave_message_reply(struct virtio_net *dev,
2790 const struct VhostUserMsg *msg)
2792 struct VhostUserMsg msg_reply;
2795 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2798 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2803 if (msg_reply.request.slave != msg->request.slave) {
2804 VHOST_LOG_CONFIG(ERR,
2805 "Received unexpected msg type (%u), expected %u\n",
2806 msg_reply.request.slave, msg->request.slave);
2811 ret = msg_reply.payload.u64 ? -1 : 0;
2814 rte_spinlock_unlock(&dev->slave_req_lock);
2819 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2822 struct VhostUserMsg msg = {
2823 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2824 .flags = VHOST_USER_VERSION,
2825 .size = sizeof(msg.payload.iotlb),
2829 .type = VHOST_IOTLB_MISS,
2833 ret = send_vhost_message(dev->slave_req_fd, &msg);
2835 VHOST_LOG_CONFIG(ERR,
2836 "Failed to send IOTLB miss message (%d)\n",
2845 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
2848 struct VhostUserMsg msg = {
2849 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
2850 .flags = VHOST_USER_VERSION,
2855 msg.flags |= VHOST_USER_NEED_REPLY;
2857 ret = send_vhost_slave_message(dev, &msg);
2859 VHOST_LOG_CONFIG(ERR,
2860 "Failed to send config change (%d)\n",
2865 return process_slave_message_reply(dev, &msg);
2869 rte_vhost_slave_config_change(int vid, bool need_reply)
2871 struct virtio_net *dev;
2873 dev = get_device(vid);
2877 return vhost_user_slave_config_change(dev, need_reply);
2880 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2886 struct VhostUserMsg msg = {
2887 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2888 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2889 .size = sizeof(msg.payload.area),
2891 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2898 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2904 ret = send_vhost_slave_message(dev, &msg);
2906 VHOST_LOG_CONFIG(ERR,
2907 "Failed to set host notifier (%d)\n", ret);
2911 return process_slave_message_reply(dev, &msg);
2914 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2916 struct virtio_net *dev;
2917 struct rte_vdpa_device *vdpa_dev;
2918 int vfio_device_fd, did, ret = 0;
2919 uint64_t offset, size;
2922 dev = get_device(vid);
2926 did = dev->vdpa_dev_id;
2930 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2931 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2932 !(dev->protocol_features &
2933 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2934 !(dev->protocol_features &
2935 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2936 !(dev->protocol_features &
2937 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2940 vdpa_dev = rte_vdpa_get_device(did);
2944 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2945 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2947 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2948 if (vfio_device_fd < 0)
2952 for (i = 0; i < dev->nr_vring; i++) {
2953 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2959 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2960 vfio_device_fd, offset, size) < 0) {
2967 for (i = 0; i < dev->nr_vring; i++) {
2968 vhost_user_slave_set_vring_host_notifier(dev, i, -1,