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);
101 ret = fstat(fd, &stat);
102 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
106 * Reclaim all the outstanding zmbufs for a virtqueue.
109 drain_zmbuf_list(struct vhost_virtqueue *vq)
111 struct zcopy_mbuf *zmbuf, *next;
113 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
114 zmbuf != NULL; zmbuf = next) {
115 next = TAILQ_NEXT(zmbuf, next);
117 while (!mbuf_is_consumed(zmbuf->mbuf))
120 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
121 restore_mbuf(zmbuf->mbuf);
122 rte_pktmbuf_free(zmbuf->mbuf);
129 free_mem_region(struct virtio_net *dev)
132 struct rte_vhost_mem_region *reg;
133 struct vhost_virtqueue *vq;
135 if (!dev || !dev->mem)
138 if (dev->dequeue_zero_copy) {
139 for (i = 0; i < dev->nr_vring; i++) {
140 vq = dev->virtqueue[i];
142 drain_zmbuf_list(vq);
146 for (i = 0; i < dev->mem->nregions; i++) {
147 reg = &dev->mem->regions[i];
148 if (reg->host_user_addr) {
149 munmap(reg->mmap_addr, reg->mmap_size);
156 vhost_backend_cleanup(struct virtio_net *dev)
159 free_mem_region(dev);
164 free(dev->guest_pages);
165 dev->guest_pages = NULL;
168 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
172 if (dev->inflight_info) {
173 if (dev->inflight_info->addr) {
174 munmap(dev->inflight_info->addr,
175 dev->inflight_info->size);
176 dev->inflight_info->addr = NULL;
179 if (dev->inflight_info->fd > 0) {
180 close(dev->inflight_info->fd);
181 dev->inflight_info->fd = -1;
184 free(dev->inflight_info);
185 dev->inflight_info = NULL;
188 if (dev->slave_req_fd >= 0) {
189 close(dev->slave_req_fd);
190 dev->slave_req_fd = -1;
193 if (dev->postcopy_ufd >= 0) {
194 close(dev->postcopy_ufd);
195 dev->postcopy_ufd = -1;
198 dev->postcopy_listening = 0;
202 * This function just returns success at the moment unless
203 * the device hasn't been initialised.
206 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
207 struct VhostUserMsg *msg __rte_unused,
208 int main_fd __rte_unused)
210 return RTE_VHOST_MSG_RESULT_OK;
214 vhost_user_reset_owner(struct virtio_net **pdev,
215 struct VhostUserMsg *msg __rte_unused,
216 int main_fd __rte_unused)
218 struct virtio_net *dev = *pdev;
219 vhost_destroy_device_notify(dev);
221 cleanup_device(dev, 0);
223 return RTE_VHOST_MSG_RESULT_OK;
227 * The features that we support are requested.
230 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
231 int main_fd __rte_unused)
233 struct virtio_net *dev = *pdev;
234 uint64_t features = 0;
236 rte_vhost_driver_get_features(dev->ifname, &features);
238 msg->payload.u64 = features;
239 msg->size = sizeof(msg->payload.u64);
242 return RTE_VHOST_MSG_RESULT_REPLY;
246 * The queue number that we support are requested.
249 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
250 int main_fd __rte_unused)
252 struct virtio_net *dev = *pdev;
253 uint32_t queue_num = 0;
255 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
257 msg->payload.u64 = (uint64_t)queue_num;
258 msg->size = sizeof(msg->payload.u64);
261 return RTE_VHOST_MSG_RESULT_REPLY;
265 * We receive the negotiated features supported by us and the virtio device.
268 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
269 int main_fd __rte_unused)
271 struct virtio_net *dev = *pdev;
272 uint64_t features = msg->payload.u64;
273 uint64_t vhost_features = 0;
274 struct rte_vdpa_device *vdpa_dev;
277 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
278 if (features & ~vhost_features) {
279 RTE_LOG(ERR, VHOST_CONFIG,
280 "(%d) received invalid negotiated features.\n",
282 return RTE_VHOST_MSG_RESULT_ERR;
285 if (dev->flags & VIRTIO_DEV_RUNNING) {
286 if (dev->features == features)
287 return RTE_VHOST_MSG_RESULT_OK;
290 * Error out if master tries to change features while device is
291 * in running state. The exception being VHOST_F_LOG_ALL, which
292 * is enabled when the live-migration starts.
294 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
295 RTE_LOG(ERR, VHOST_CONFIG,
296 "(%d) features changed while device is running.\n",
298 return RTE_VHOST_MSG_RESULT_ERR;
301 if (dev->notify_ops->features_changed)
302 dev->notify_ops->features_changed(dev->vid, features);
305 dev->features = features;
307 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
308 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
310 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
312 RTE_LOG(INFO, VHOST_CONFIG,
313 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
314 VHOST_LOG_DEBUG(VHOST_CONFIG,
315 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
317 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
318 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
320 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
321 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
323 * Remove all but first queue pair if MQ hasn't been
324 * negotiated. This is safe because the device is not
325 * running at this stage.
327 while (dev->nr_vring > 2) {
328 struct vhost_virtqueue *vq;
330 vq = dev->virtqueue[--dev->nr_vring];
334 dev->virtqueue[dev->nr_vring] = NULL;
336 cleanup_vq_inflight(dev, vq);
341 did = dev->vdpa_dev_id;
342 vdpa_dev = rte_vdpa_get_device(did);
343 if (vdpa_dev && vdpa_dev->ops->set_features)
344 vdpa_dev->ops->set_features(dev->vid);
346 return RTE_VHOST_MSG_RESULT_OK;
350 * The virtio device sends us the size of the descriptor ring.
353 vhost_user_set_vring_num(struct virtio_net **pdev,
354 struct VhostUserMsg *msg,
355 int main_fd __rte_unused)
357 struct virtio_net *dev = *pdev;
358 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
360 vq->size = msg->payload.state.num;
362 /* VIRTIO 1.0, 2.4 Virtqueues says:
364 * Queue Size value is always a power of 2. The maximum Queue Size
367 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
368 RTE_LOG(ERR, VHOST_CONFIG,
369 "invalid virtqueue size %u\n", vq->size);
370 return RTE_VHOST_MSG_RESULT_ERR;
373 if (dev->dequeue_zero_copy) {
375 vq->last_zmbuf_idx = 0;
376 vq->zmbuf_size = vq->size;
377 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
378 sizeof(struct zcopy_mbuf), 0);
379 if (vq->zmbufs == NULL) {
380 RTE_LOG(WARNING, VHOST_CONFIG,
381 "failed to allocate mem for zero copy; "
382 "zero copy is force disabled\n");
383 dev->dequeue_zero_copy = 0;
385 TAILQ_INIT(&vq->zmbuf_list);
388 if (vq_is_packed(dev)) {
389 vq->shadow_used_packed = rte_malloc(NULL,
391 sizeof(struct vring_used_elem_packed),
392 RTE_CACHE_LINE_SIZE);
393 if (!vq->shadow_used_packed) {
394 RTE_LOG(ERR, VHOST_CONFIG,
395 "failed to allocate memory for shadow used ring.\n");
396 return RTE_VHOST_MSG_RESULT_ERR;
400 vq->shadow_used_split = rte_malloc(NULL,
401 vq->size * sizeof(struct vring_used_elem),
402 RTE_CACHE_LINE_SIZE);
403 if (!vq->shadow_used_split) {
404 RTE_LOG(ERR, VHOST_CONFIG,
405 "failed to allocate memory for shadow used ring.\n");
406 return RTE_VHOST_MSG_RESULT_ERR;
410 vq->batch_copy_elems = rte_malloc(NULL,
411 vq->size * sizeof(struct batch_copy_elem),
412 RTE_CACHE_LINE_SIZE);
413 if (!vq->batch_copy_elems) {
414 RTE_LOG(ERR, VHOST_CONFIG,
415 "failed to allocate memory for batching copy.\n");
416 return RTE_VHOST_MSG_RESULT_ERR;
419 return RTE_VHOST_MSG_RESULT_OK;
423 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
424 * same numa node as the memory of vring descriptor.
426 #ifdef RTE_LIBRTE_VHOST_NUMA
427 static struct virtio_net*
428 numa_realloc(struct virtio_net *dev, int index)
430 int oldnode, newnode;
431 struct virtio_net *old_dev;
432 struct vhost_virtqueue *old_vq, *vq;
433 struct zcopy_mbuf *new_zmbuf;
434 struct vring_used_elem *new_shadow_used_split;
435 struct vring_used_elem_packed *new_shadow_used_packed;
436 struct batch_copy_elem *new_batch_copy_elems;
439 if (dev->flags & VIRTIO_DEV_RUNNING)
443 vq = old_vq = dev->virtqueue[index];
445 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
446 MPOL_F_NODE | MPOL_F_ADDR);
448 /* check if we need to reallocate vq */
449 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
450 MPOL_F_NODE | MPOL_F_ADDR);
452 RTE_LOG(ERR, VHOST_CONFIG,
453 "Unable to get vq numa information.\n");
456 if (oldnode != newnode) {
457 RTE_LOG(INFO, VHOST_CONFIG,
458 "reallocate vq from %d to %d node\n", oldnode, newnode);
459 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
463 memcpy(vq, old_vq, sizeof(*vq));
464 TAILQ_INIT(&vq->zmbuf_list);
466 if (dev->dequeue_zero_copy) {
467 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
468 sizeof(struct zcopy_mbuf), 0, newnode);
470 rte_free(vq->zmbufs);
471 vq->zmbufs = new_zmbuf;
475 if (vq_is_packed(dev)) {
476 new_shadow_used_packed = rte_malloc_socket(NULL,
478 sizeof(struct vring_used_elem_packed),
481 if (new_shadow_used_packed) {
482 rte_free(vq->shadow_used_packed);
483 vq->shadow_used_packed = new_shadow_used_packed;
486 new_shadow_used_split = rte_malloc_socket(NULL,
488 sizeof(struct vring_used_elem),
491 if (new_shadow_used_split) {
492 rte_free(vq->shadow_used_split);
493 vq->shadow_used_split = new_shadow_used_split;
497 new_batch_copy_elems = rte_malloc_socket(NULL,
498 vq->size * sizeof(struct batch_copy_elem),
501 if (new_batch_copy_elems) {
502 rte_free(vq->batch_copy_elems);
503 vq->batch_copy_elems = new_batch_copy_elems;
509 /* check if we need to reallocate dev */
510 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
511 MPOL_F_NODE | MPOL_F_ADDR);
513 RTE_LOG(ERR, VHOST_CONFIG,
514 "Unable to get dev numa information.\n");
517 if (oldnode != newnode) {
518 RTE_LOG(INFO, VHOST_CONFIG,
519 "reallocate dev from %d to %d node\n",
521 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
527 memcpy(dev, old_dev, sizeof(*dev));
532 dev->virtqueue[index] = vq;
533 vhost_devices[dev->vid] = dev;
536 vhost_user_iotlb_init(dev, index);
541 static struct virtio_net*
542 numa_realloc(struct virtio_net *dev, int index __rte_unused)
548 /* Converts QEMU virtual address to Vhost virtual address. */
550 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
552 struct rte_vhost_mem_region *r;
555 if (unlikely(!dev || !dev->mem))
558 /* Find the region where the address lives. */
559 for (i = 0; i < dev->mem->nregions; i++) {
560 r = &dev->mem->regions[i];
562 if (qva >= r->guest_user_addr &&
563 qva < r->guest_user_addr + r->size) {
565 if (unlikely(*len > r->guest_user_addr + r->size - qva))
566 *len = r->guest_user_addr + r->size - qva;
568 return qva - r->guest_user_addr +
580 * Converts ring address to Vhost virtual address.
581 * If IOMMU is enabled, the ring address is a guest IO virtual address,
582 * else it is a QEMU virtual address.
585 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
586 uint64_t ra, uint64_t *size)
588 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
590 uint64_t req_size = *size;
592 vva = vhost_user_iotlb_cache_find(vq, ra,
593 size, VHOST_ACCESS_RW);
594 if (req_size != *size)
595 vhost_user_iotlb_miss(dev, (ra + *size),
601 return qva_to_vva(dev, ra, size);
605 * Converts vring log address to GPA
606 * If IOMMU is enabled, the log address is IOVA
607 * If IOMMU not enabled, the log address is already GPA
610 translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
613 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
614 const uint64_t exp_size = sizeof(struct vring_used) +
615 sizeof(struct vring_used_elem) * vq->size;
617 uint64_t size = exp_size;
619 hva = vhost_iova_to_vva(dev, vq, log_addr,
620 &size, VHOST_ACCESS_RW);
621 if (size != exp_size)
624 gpa = hva_to_gpa(dev, hva, exp_size);
626 RTE_LOG(ERR, VHOST_CONFIG,
627 "VQ: Failed to find GPA for log_addr: 0x%" PRIx64 " hva: 0x%" PRIx64 "\n",
637 static struct virtio_net *
638 translate_ring_addresses(struct virtio_net *dev, int vq_index)
640 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
641 struct vhost_vring_addr *addr = &vq->ring_addrs;
642 uint64_t len, expected_len;
644 if (vq_is_packed(dev)) {
645 len = sizeof(struct vring_packed_desc) * vq->size;
646 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
647 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
648 vq->log_guest_addr = 0;
649 if (vq->desc_packed == NULL ||
650 len != sizeof(struct vring_packed_desc) *
652 RTE_LOG(DEBUG, VHOST_CONFIG,
653 "(%d) failed to map desc_packed ring.\n",
658 dev = numa_realloc(dev, vq_index);
659 vq = dev->virtqueue[vq_index];
660 addr = &vq->ring_addrs;
662 len = sizeof(struct vring_packed_desc_event);
663 vq->driver_event = (struct vring_packed_desc_event *)
664 (uintptr_t)ring_addr_to_vva(dev,
665 vq, addr->avail_user_addr, &len);
666 if (vq->driver_event == NULL ||
667 len != sizeof(struct vring_packed_desc_event)) {
668 RTE_LOG(DEBUG, VHOST_CONFIG,
669 "(%d) failed to find driver area address.\n",
674 len = sizeof(struct vring_packed_desc_event);
675 vq->device_event = (struct vring_packed_desc_event *)
676 (uintptr_t)ring_addr_to_vva(dev,
677 vq, addr->used_user_addr, &len);
678 if (vq->device_event == NULL ||
679 len != sizeof(struct vring_packed_desc_event)) {
680 RTE_LOG(DEBUG, VHOST_CONFIG,
681 "(%d) failed to find device area address.\n",
690 /* The addresses are converted from QEMU virtual to Vhost virtual. */
691 if (vq->desc && vq->avail && vq->used)
694 len = sizeof(struct vring_desc) * vq->size;
695 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
696 vq, addr->desc_user_addr, &len);
697 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
698 RTE_LOG(DEBUG, VHOST_CONFIG,
699 "(%d) failed to map desc ring.\n",
704 dev = numa_realloc(dev, vq_index);
705 vq = dev->virtqueue[vq_index];
706 addr = &vq->ring_addrs;
708 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
709 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
710 len += sizeof(uint16_t);
712 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
713 vq, addr->avail_user_addr, &len);
714 if (vq->avail == 0 || len != expected_len) {
715 RTE_LOG(DEBUG, VHOST_CONFIG,
716 "(%d) failed to map avail ring.\n",
721 len = sizeof(struct vring_used) +
722 sizeof(struct vring_used_elem) * vq->size;
723 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
724 len += sizeof(uint16_t);
726 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
727 vq, addr->used_user_addr, &len);
728 if (vq->used == 0 || len != expected_len) {
729 RTE_LOG(DEBUG, VHOST_CONFIG,
730 "(%d) failed to map used ring.\n",
735 if (vq->last_used_idx != vq->used->idx) {
736 RTE_LOG(WARNING, VHOST_CONFIG,
737 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
738 "some packets maybe resent for Tx and dropped for Rx\n",
739 vq->last_used_idx, vq->used->idx);
740 vq->last_used_idx = vq->used->idx;
741 vq->last_avail_idx = vq->used->idx;
745 translate_log_addr(dev, vq, addr->log_guest_addr);
746 if (vq->log_guest_addr == 0) {
747 RTE_LOG(DEBUG, VHOST_CONFIG,
748 "(%d) failed to map log_guest_addr .\n",
754 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
756 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
757 dev->vid, vq->avail);
758 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
760 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
761 dev->vid, vq->log_guest_addr);
767 * The virtio device sends us the desc, used and avail ring addresses.
768 * This function then converts these to our address space.
771 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
772 int main_fd __rte_unused)
774 struct virtio_net *dev = *pdev;
775 struct vhost_virtqueue *vq;
776 struct vhost_vring_addr *addr = &msg->payload.addr;
779 if (dev->mem == NULL)
780 return RTE_VHOST_MSG_RESULT_ERR;
782 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
783 vq = dev->virtqueue[msg->payload.addr.index];
785 access_ok = vq->access_ok;
788 * Rings addresses should not be interpreted as long as the ring is not
789 * started and enabled
791 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
793 vring_invalidate(dev, vq);
795 if ((vq->enabled && (dev->features &
796 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
798 dev = translate_ring_addresses(dev, msg->payload.addr.index);
800 return RTE_VHOST_MSG_RESULT_ERR;
805 return RTE_VHOST_MSG_RESULT_OK;
809 * The virtio device sends us the available ring last used index.
812 vhost_user_set_vring_base(struct virtio_net **pdev,
813 struct VhostUserMsg *msg,
814 int main_fd __rte_unused)
816 struct virtio_net *dev = *pdev;
817 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
818 uint64_t val = msg->payload.state.num;
820 if (vq_is_packed(dev)) {
822 * Bit[0:14]: avail index
823 * Bit[15]: avail wrap counter
825 vq->last_avail_idx = val & 0x7fff;
826 vq->avail_wrap_counter = !!(val & (0x1 << 15));
828 * Set used index to same value as available one, as
829 * their values should be the same since ring processing
830 * was stopped at get time.
832 vq->last_used_idx = vq->last_avail_idx;
833 vq->used_wrap_counter = vq->avail_wrap_counter;
835 vq->last_used_idx = msg->payload.state.num;
836 vq->last_avail_idx = msg->payload.state.num;
839 return RTE_VHOST_MSG_RESULT_OK;
843 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
844 uint64_t host_phys_addr, uint64_t size)
846 struct guest_page *page, *last_page;
847 struct guest_page *old_pages;
849 if (dev->nr_guest_pages == dev->max_guest_pages) {
850 dev->max_guest_pages *= 2;
851 old_pages = dev->guest_pages;
852 dev->guest_pages = realloc(dev->guest_pages,
853 dev->max_guest_pages * sizeof(*page));
854 if (!dev->guest_pages) {
855 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
861 if (dev->nr_guest_pages > 0) {
862 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
863 /* merge if the two pages are continuous */
864 if (host_phys_addr == last_page->host_phys_addr +
866 last_page->size += size;
871 page = &dev->guest_pages[dev->nr_guest_pages++];
872 page->guest_phys_addr = guest_phys_addr;
873 page->host_phys_addr = host_phys_addr;
880 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
883 uint64_t reg_size = reg->size;
884 uint64_t host_user_addr = reg->host_user_addr;
885 uint64_t guest_phys_addr = reg->guest_phys_addr;
886 uint64_t host_phys_addr;
889 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
890 size = page_size - (guest_phys_addr & (page_size - 1));
891 size = RTE_MIN(size, reg_size);
893 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
896 host_user_addr += size;
897 guest_phys_addr += size;
900 while (reg_size > 0) {
901 size = RTE_MIN(reg_size, page_size);
902 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
904 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
908 host_user_addr += size;
909 guest_phys_addr += size;
916 #ifdef RTE_LIBRTE_VHOST_DEBUG
917 /* TODO: enable it only in debug mode? */
919 dump_guest_pages(struct virtio_net *dev)
922 struct guest_page *page;
924 for (i = 0; i < dev->nr_guest_pages; i++) {
925 page = &dev->guest_pages[i];
927 RTE_LOG(INFO, VHOST_CONFIG,
928 "guest physical page region %u\n"
929 "\t guest_phys_addr: %" PRIx64 "\n"
930 "\t host_phys_addr : %" PRIx64 "\n"
931 "\t size : %" PRIx64 "\n",
933 page->guest_phys_addr,
934 page->host_phys_addr,
939 #define dump_guest_pages(dev)
943 vhost_memory_changed(struct VhostUserMemory *new,
944 struct rte_vhost_memory *old)
948 if (new->nregions != old->nregions)
951 for (i = 0; i < new->nregions; ++i) {
952 VhostUserMemoryRegion *new_r = &new->regions[i];
953 struct rte_vhost_mem_region *old_r = &old->regions[i];
955 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
957 if (new_r->memory_size != old_r->size)
959 if (new_r->userspace_addr != old_r->guest_user_addr)
967 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
970 struct virtio_net *dev = *pdev;
971 struct VhostUserMemory *memory = &msg->payload.memory;
972 struct rte_vhost_mem_region *reg;
975 uint64_t mmap_offset;
981 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
982 RTE_LOG(ERR, VHOST_CONFIG,
983 "too many memory regions (%u)\n", memory->nregions);
984 return RTE_VHOST_MSG_RESULT_ERR;
987 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
988 RTE_LOG(INFO, VHOST_CONFIG,
989 "(%d) memory regions not changed\n", dev->vid);
991 for (i = 0; i < memory->nregions; i++)
994 return RTE_VHOST_MSG_RESULT_OK;
998 free_mem_region(dev);
1003 /* Flush IOTLB cache as previous HVAs are now invalid */
1004 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1005 for (i = 0; i < dev->nr_vring; i++)
1006 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1008 dev->nr_guest_pages = 0;
1009 if (!dev->guest_pages) {
1010 dev->max_guest_pages = 8;
1011 dev->guest_pages = malloc(dev->max_guest_pages *
1012 sizeof(struct guest_page));
1013 if (dev->guest_pages == NULL) {
1014 RTE_LOG(ERR, VHOST_CONFIG,
1015 "(%d) failed to allocate memory "
1016 "for dev->guest_pages\n",
1018 return RTE_VHOST_MSG_RESULT_ERR;
1022 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1023 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1024 if (dev->mem == NULL) {
1025 RTE_LOG(ERR, VHOST_CONFIG,
1026 "(%d) failed to allocate memory for dev->mem\n",
1028 return RTE_VHOST_MSG_RESULT_ERR;
1030 dev->mem->nregions = memory->nregions;
1032 for (i = 0; i < memory->nregions; i++) {
1034 reg = &dev->mem->regions[i];
1036 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1037 reg->guest_user_addr = memory->regions[i].userspace_addr;
1038 reg->size = memory->regions[i].memory_size;
1041 mmap_offset = memory->regions[i].mmap_offset;
1043 /* Check for memory_size + mmap_offset overflow */
1044 if (mmap_offset >= -reg->size) {
1045 RTE_LOG(ERR, VHOST_CONFIG,
1046 "mmap_offset (%#"PRIx64") and memory_size "
1047 "(%#"PRIx64") overflow\n",
1048 mmap_offset, reg->size);
1052 mmap_size = reg->size + mmap_offset;
1054 /* mmap() without flag of MAP_ANONYMOUS, should be called
1055 * with length argument aligned with hugepagesz at older
1056 * longterm version Linux, like 2.6.32 and 3.2.72, or
1057 * mmap() will fail with EINVAL.
1059 * to avoid failure, make sure in caller to keep length
1062 alignment = get_blk_size(fd);
1063 if (alignment == (uint64_t)-1) {
1064 RTE_LOG(ERR, VHOST_CONFIG,
1065 "couldn't get hugepage size through fstat\n");
1068 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1070 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
1071 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1072 MAP_SHARED | populate, fd, 0);
1074 if (mmap_addr == MAP_FAILED) {
1075 RTE_LOG(ERR, VHOST_CONFIG,
1076 "mmap region %u failed.\n", i);
1080 reg->mmap_addr = mmap_addr;
1081 reg->mmap_size = mmap_size;
1082 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1085 if (dev->dequeue_zero_copy)
1086 if (add_guest_pages(dev, reg, alignment) < 0) {
1087 RTE_LOG(ERR, VHOST_CONFIG,
1088 "adding guest pages to region %u failed.\n",
1093 RTE_LOG(INFO, VHOST_CONFIG,
1094 "guest memory region %u, size: 0x%" PRIx64 "\n"
1095 "\t guest physical addr: 0x%" PRIx64 "\n"
1096 "\t guest virtual addr: 0x%" PRIx64 "\n"
1097 "\t host virtual addr: 0x%" PRIx64 "\n"
1098 "\t mmap addr : 0x%" PRIx64 "\n"
1099 "\t mmap size : 0x%" PRIx64 "\n"
1100 "\t mmap align: 0x%" PRIx64 "\n"
1101 "\t mmap off : 0x%" PRIx64 "\n",
1103 reg->guest_phys_addr,
1104 reg->guest_user_addr,
1105 reg->host_user_addr,
1106 (uint64_t)(uintptr_t)mmap_addr,
1111 if (dev->postcopy_listening) {
1113 * We haven't a better way right now than sharing
1114 * DPDK's virtual address with Qemu, so that Qemu can
1115 * retrieve the region offset when handling userfaults.
1117 memory->regions[i].userspace_addr =
1118 reg->host_user_addr;
1121 if (dev->postcopy_listening) {
1122 /* Send the addresses back to qemu */
1124 send_vhost_reply(main_fd, msg);
1126 /* Wait for qemu to acknolwedge it's got the addresses
1127 * we've got to wait before we're allowed to generate faults.
1129 VhostUserMsg ack_msg;
1130 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1131 RTE_LOG(ERR, VHOST_CONFIG,
1132 "Failed to read qemu ack on postcopy set-mem-table\n");
1135 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1136 RTE_LOG(ERR, VHOST_CONFIG,
1137 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1138 ack_msg.request.master);
1142 /* Now userfault register and we can use the memory */
1143 for (i = 0; i < memory->nregions; i++) {
1144 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1145 reg = &dev->mem->regions[i];
1146 struct uffdio_register reg_struct;
1149 * Let's register all the mmap'ed area to ensure
1150 * alignment on page boundary.
1152 reg_struct.range.start =
1153 (uint64_t)(uintptr_t)reg->mmap_addr;
1154 reg_struct.range.len = reg->mmap_size;
1155 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1157 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1159 RTE_LOG(ERR, VHOST_CONFIG,
1160 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1161 i, dev->postcopy_ufd,
1165 RTE_LOG(INFO, VHOST_CONFIG,
1166 "\t userfaultfd registered for range : "
1167 "%" PRIx64 " - %" PRIx64 "\n",
1168 (uint64_t)reg_struct.range.start,
1169 (uint64_t)reg_struct.range.start +
1170 (uint64_t)reg_struct.range.len - 1);
1177 for (i = 0; i < dev->nr_vring; i++) {
1178 struct vhost_virtqueue *vq = dev->virtqueue[i];
1180 if (vq->desc || vq->avail || vq->used) {
1182 * If the memory table got updated, the ring addresses
1183 * need to be translated again as virtual addresses have
1186 vring_invalidate(dev, vq);
1188 dev = translate_ring_addresses(dev, i);
1198 dump_guest_pages(dev);
1200 return RTE_VHOST_MSG_RESULT_OK;
1203 free_mem_region(dev);
1206 return RTE_VHOST_MSG_RESULT_ERR;
1210 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1217 if (vq_is_packed(dev))
1218 rings_ok = !!vq->desc_packed;
1220 rings_ok = vq->desc && vq->avail && vq->used;
1223 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1224 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1228 virtio_is_ready(struct virtio_net *dev)
1230 struct vhost_virtqueue *vq;
1233 if (dev->nr_vring == 0)
1236 for (i = 0; i < dev->nr_vring; i++) {
1237 vq = dev->virtqueue[i];
1239 if (!vq_is_ready(dev, vq))
1243 RTE_LOG(INFO, VHOST_CONFIG,
1244 "virtio is now ready for processing.\n");
1249 inflight_mem_alloc(const char *name, size_t size, int *fd)
1253 char fname[20] = "/tmp/memfd-XXXXXX";
1256 #ifdef MEMFD_SUPPORTED
1257 mfd = memfd_create(name, MFD_CLOEXEC);
1262 mfd = mkstemp(fname);
1264 RTE_LOG(ERR, VHOST_CONFIG,
1265 "failed to get inflight buffer fd\n");
1272 if (ftruncate(mfd, size) == -1) {
1273 RTE_LOG(ERR, VHOST_CONFIG,
1274 "failed to alloc inflight buffer\n");
1279 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1280 if (ptr == MAP_FAILED) {
1281 RTE_LOG(ERR, VHOST_CONFIG,
1282 "failed to mmap inflight buffer\n");
1292 get_pervq_shm_size_split(uint16_t queue_size)
1294 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1295 queue_size + sizeof(uint64_t) +
1296 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1300 get_pervq_shm_size_packed(uint16_t queue_size)
1302 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1303 * queue_size + sizeof(uint64_t) +
1304 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1305 INFLIGHT_ALIGNMENT);
1309 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1311 int main_fd __rte_unused)
1313 struct rte_vhost_inflight_info_packed *inflight_packed;
1314 uint64_t pervq_inflight_size, mmap_size;
1315 uint16_t num_queues, queue_size;
1316 struct virtio_net *dev = *pdev;
1320 if (msg->size != sizeof(msg->payload.inflight)) {
1321 RTE_LOG(ERR, VHOST_CONFIG,
1322 "invalid get_inflight_fd message size is %d\n",
1324 return RTE_VHOST_MSG_RESULT_ERR;
1327 if (dev->inflight_info == NULL) {
1328 dev->inflight_info = calloc(1,
1329 sizeof(struct inflight_mem_info));
1330 if (!dev->inflight_info) {
1331 RTE_LOG(ERR, VHOST_CONFIG,
1332 "failed to alloc dev inflight area\n");
1333 return RTE_VHOST_MSG_RESULT_ERR;
1337 num_queues = msg->payload.inflight.num_queues;
1338 queue_size = msg->payload.inflight.queue_size;
1340 RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
1341 msg->payload.inflight.num_queues);
1342 RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
1343 msg->payload.inflight.queue_size);
1345 if (vq_is_packed(dev))
1346 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1348 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1350 mmap_size = num_queues * pervq_inflight_size;
1351 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1353 RTE_LOG(ERR, VHOST_CONFIG,
1354 "failed to alloc vhost inflight area\n");
1355 msg->payload.inflight.mmap_size = 0;
1356 return RTE_VHOST_MSG_RESULT_ERR;
1358 memset(addr, 0, mmap_size);
1360 dev->inflight_info->addr = addr;
1361 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1362 dev->inflight_info->fd = msg->fds[0] = fd;
1363 msg->payload.inflight.mmap_offset = 0;
1366 if (vq_is_packed(dev)) {
1367 for (i = 0; i < num_queues; i++) {
1369 (struct rte_vhost_inflight_info_packed *)addr;
1370 inflight_packed->used_wrap_counter = 1;
1371 inflight_packed->old_used_wrap_counter = 1;
1372 for (j = 0; j < queue_size; j++)
1373 inflight_packed->desc[j].next = j + 1;
1374 addr = (void *)((char *)addr + pervq_inflight_size);
1378 RTE_LOG(INFO, VHOST_CONFIG,
1379 "send inflight mmap_size: %"PRIu64"\n",
1380 msg->payload.inflight.mmap_size);
1381 RTE_LOG(INFO, VHOST_CONFIG,
1382 "send inflight mmap_offset: %"PRIu64"\n",
1383 msg->payload.inflight.mmap_offset);
1384 RTE_LOG(INFO, VHOST_CONFIG,
1385 "send inflight fd: %d\n", msg->fds[0]);
1387 return RTE_VHOST_MSG_RESULT_REPLY;
1391 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1392 int main_fd __rte_unused)
1394 uint64_t mmap_size, mmap_offset;
1395 uint16_t num_queues, queue_size;
1396 struct virtio_net *dev = *pdev;
1397 uint32_t pervq_inflight_size;
1398 struct vhost_virtqueue *vq;
1403 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1404 RTE_LOG(ERR, VHOST_CONFIG,
1405 "invalid set_inflight_fd message size is %d,fd is %d\n",
1407 return RTE_VHOST_MSG_RESULT_ERR;
1410 mmap_size = msg->payload.inflight.mmap_size;
1411 mmap_offset = msg->payload.inflight.mmap_offset;
1412 num_queues = msg->payload.inflight.num_queues;
1413 queue_size = msg->payload.inflight.queue_size;
1415 if (vq_is_packed(dev))
1416 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1418 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1420 RTE_LOG(INFO, VHOST_CONFIG,
1421 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1422 RTE_LOG(INFO, VHOST_CONFIG,
1423 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1424 RTE_LOG(INFO, VHOST_CONFIG,
1425 "set_inflight_fd num_queues: %u\n", num_queues);
1426 RTE_LOG(INFO, VHOST_CONFIG,
1427 "set_inflight_fd queue_size: %u\n", queue_size);
1428 RTE_LOG(INFO, VHOST_CONFIG,
1429 "set_inflight_fd fd: %d\n", fd);
1430 RTE_LOG(INFO, VHOST_CONFIG,
1431 "set_inflight_fd pervq_inflight_size: %d\n",
1432 pervq_inflight_size);
1434 if (!dev->inflight_info) {
1435 dev->inflight_info = calloc(1,
1436 sizeof(struct inflight_mem_info));
1437 if (dev->inflight_info == NULL) {
1438 RTE_LOG(ERR, VHOST_CONFIG,
1439 "failed to alloc dev inflight area\n");
1440 return RTE_VHOST_MSG_RESULT_ERR;
1444 if (dev->inflight_info->addr)
1445 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1447 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1449 if (addr == MAP_FAILED) {
1450 RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
1451 return RTE_VHOST_MSG_RESULT_ERR;
1454 if (dev->inflight_info->fd)
1455 close(dev->inflight_info->fd);
1457 dev->inflight_info->fd = fd;
1458 dev->inflight_info->addr = addr;
1459 dev->inflight_info->size = mmap_size;
1461 for (i = 0; i < num_queues; i++) {
1462 vq = dev->virtqueue[i];
1463 if (vq_is_packed(dev)) {
1464 vq->inflight_packed = addr;
1465 vq->inflight_packed->desc_num = queue_size;
1467 vq->inflight_split = addr;
1468 vq->inflight_split->desc_num = queue_size;
1470 addr = (void *)((char *)addr + pervq_inflight_size);
1473 return RTE_VHOST_MSG_RESULT_OK;
1477 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1478 int main_fd __rte_unused)
1480 struct virtio_net *dev = *pdev;
1481 struct vhost_vring_file file;
1482 struct vhost_virtqueue *vq;
1484 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1485 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1486 file.fd = VIRTIO_INVALID_EVENTFD;
1488 file.fd = msg->fds[0];
1489 RTE_LOG(INFO, VHOST_CONFIG,
1490 "vring call idx:%d file:%d\n", file.index, file.fd);
1492 vq = dev->virtqueue[file.index];
1493 if (vq->callfd >= 0)
1496 vq->callfd = file.fd;
1498 return RTE_VHOST_MSG_RESULT_OK;
1501 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1502 struct VhostUserMsg *msg,
1503 int main_fd __rte_unused)
1505 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1507 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1509 return RTE_VHOST_MSG_RESULT_OK;
1513 resubmit_desc_compare(const void *a, const void *b)
1515 const struct rte_vhost_resubmit_desc *desc0 = a;
1516 const struct rte_vhost_resubmit_desc *desc1 = b;
1518 if (desc1->counter > desc0->counter)
1525 vhost_check_queue_inflights_split(struct virtio_net *dev,
1526 struct vhost_virtqueue *vq)
1529 uint16_t resubmit_num = 0, last_io, num;
1530 struct vring_used *used = vq->used;
1531 struct rte_vhost_resubmit_info *resubmit;
1532 struct rte_vhost_inflight_info_split *inflight_split;
1534 if (!(dev->protocol_features &
1535 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1536 return RTE_VHOST_MSG_RESULT_OK;
1538 if ((!vq->inflight_split))
1539 return RTE_VHOST_MSG_RESULT_ERR;
1541 if (!vq->inflight_split->version) {
1542 vq->inflight_split->version = INFLIGHT_VERSION;
1543 return RTE_VHOST_MSG_RESULT_OK;
1546 if (vq->resubmit_inflight)
1547 return RTE_VHOST_MSG_RESULT_OK;
1549 inflight_split = vq->inflight_split;
1550 vq->global_counter = 0;
1551 last_io = inflight_split->last_inflight_io;
1553 if (inflight_split->used_idx != used->idx) {
1554 inflight_split->desc[last_io].inflight = 0;
1556 inflight_split->used_idx = used->idx;
1559 for (i = 0; i < inflight_split->desc_num; i++) {
1560 if (inflight_split->desc[i].inflight == 1)
1564 vq->last_avail_idx += resubmit_num;
1567 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1569 RTE_LOG(ERR, VHOST_CONFIG,
1570 "failed to allocate memory for resubmit info.\n");
1571 return RTE_VHOST_MSG_RESULT_ERR;
1574 resubmit->resubmit_list = calloc(resubmit_num,
1575 sizeof(struct rte_vhost_resubmit_desc));
1576 if (!resubmit->resubmit_list) {
1577 RTE_LOG(ERR, VHOST_CONFIG,
1578 "failed to allocate memory for inflight desc.\n");
1580 return RTE_VHOST_MSG_RESULT_ERR;
1584 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1585 if (vq->inflight_split->desc[i].inflight == 1) {
1586 resubmit->resubmit_list[num].index = i;
1587 resubmit->resubmit_list[num].counter =
1588 inflight_split->desc[i].counter;
1592 resubmit->resubmit_num = num;
1594 if (resubmit->resubmit_num > 1)
1595 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1596 sizeof(struct rte_vhost_resubmit_desc),
1597 resubmit_desc_compare);
1599 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1600 vq->resubmit_inflight = resubmit;
1603 return RTE_VHOST_MSG_RESULT_OK;
1607 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1608 struct vhost_virtqueue *vq)
1611 uint16_t resubmit_num = 0, old_used_idx, num;
1612 struct rte_vhost_resubmit_info *resubmit;
1613 struct rte_vhost_inflight_info_packed *inflight_packed;
1615 if (!(dev->protocol_features &
1616 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1617 return RTE_VHOST_MSG_RESULT_OK;
1619 if ((!vq->inflight_packed))
1620 return RTE_VHOST_MSG_RESULT_ERR;
1622 if (!vq->inflight_packed->version) {
1623 vq->inflight_packed->version = INFLIGHT_VERSION;
1624 return RTE_VHOST_MSG_RESULT_OK;
1627 if (vq->resubmit_inflight)
1628 return RTE_VHOST_MSG_RESULT_OK;
1630 inflight_packed = vq->inflight_packed;
1631 vq->global_counter = 0;
1632 old_used_idx = inflight_packed->old_used_idx;
1634 if (inflight_packed->used_idx != old_used_idx) {
1635 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1636 inflight_packed->old_used_idx =
1637 inflight_packed->used_idx;
1638 inflight_packed->old_used_wrap_counter =
1639 inflight_packed->used_wrap_counter;
1640 inflight_packed->old_free_head =
1641 inflight_packed->free_head;
1643 inflight_packed->used_idx =
1644 inflight_packed->old_used_idx;
1645 inflight_packed->used_wrap_counter =
1646 inflight_packed->old_used_wrap_counter;
1647 inflight_packed->free_head =
1648 inflight_packed->old_free_head;
1652 for (i = 0; i < inflight_packed->desc_num; i++) {
1653 if (inflight_packed->desc[i].inflight == 1)
1658 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1659 if (resubmit == NULL) {
1660 RTE_LOG(ERR, VHOST_CONFIG,
1661 "failed to allocate memory for resubmit info.\n");
1662 return RTE_VHOST_MSG_RESULT_ERR;
1665 resubmit->resubmit_list = calloc(resubmit_num,
1666 sizeof(struct rte_vhost_resubmit_desc));
1667 if (resubmit->resubmit_list == NULL) {
1668 RTE_LOG(ERR, VHOST_CONFIG,
1669 "failed to allocate memory for resubmit desc.\n");
1671 return RTE_VHOST_MSG_RESULT_ERR;
1675 for (i = 0; i < inflight_packed->desc_num; i++) {
1676 if (vq->inflight_packed->desc[i].inflight == 1) {
1677 resubmit->resubmit_list[num].index = i;
1678 resubmit->resubmit_list[num].counter =
1679 inflight_packed->desc[i].counter;
1683 resubmit->resubmit_num = num;
1685 if (resubmit->resubmit_num > 1)
1686 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1687 sizeof(struct rte_vhost_resubmit_desc),
1688 resubmit_desc_compare);
1690 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1691 vq->resubmit_inflight = resubmit;
1694 return RTE_VHOST_MSG_RESULT_OK;
1698 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1699 int main_fd __rte_unused)
1701 struct virtio_net *dev = *pdev;
1702 struct vhost_vring_file file;
1703 struct vhost_virtqueue *vq;
1705 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1706 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1707 file.fd = VIRTIO_INVALID_EVENTFD;
1709 file.fd = msg->fds[0];
1710 RTE_LOG(INFO, VHOST_CONFIG,
1711 "vring kick idx:%d file:%d\n", file.index, file.fd);
1713 /* Interpret ring addresses only when ring is started. */
1714 dev = translate_ring_addresses(dev, file.index);
1716 return RTE_VHOST_MSG_RESULT_ERR;
1720 vq = dev->virtqueue[file.index];
1723 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1724 * the ring starts already enabled. Otherwise, it is enabled via
1725 * the SET_VRING_ENABLE message.
1727 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1729 if (dev->notify_ops->vring_state_changed)
1730 dev->notify_ops->vring_state_changed(
1731 dev->vid, file.index, 1);
1734 if (vq->kickfd >= 0)
1736 vq->kickfd = file.fd;
1738 if (vq_is_packed(dev)) {
1739 if (vhost_check_queue_inflights_packed(dev, vq)) {
1740 RTE_LOG(ERR, VHOST_CONFIG,
1741 "failed to inflights for vq: %d\n", file.index);
1742 return RTE_VHOST_MSG_RESULT_ERR;
1745 if (vhost_check_queue_inflights_split(dev, vq)) {
1746 RTE_LOG(ERR, VHOST_CONFIG,
1747 "failed to inflights for vq: %d\n", file.index);
1748 return RTE_VHOST_MSG_RESULT_ERR;
1752 return RTE_VHOST_MSG_RESULT_OK;
1756 free_zmbufs(struct vhost_virtqueue *vq)
1758 drain_zmbuf_list(vq);
1760 rte_free(vq->zmbufs);
1764 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1767 vhost_user_get_vring_base(struct virtio_net **pdev,
1768 struct VhostUserMsg *msg,
1769 int main_fd __rte_unused)
1771 struct virtio_net *dev = *pdev;
1772 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1775 /* We have to stop the queue (virtio) if it is running. */
1776 vhost_destroy_device_notify(dev);
1778 dev->flags &= ~VIRTIO_DEV_READY;
1779 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1781 /* Here we are safe to get the indexes */
1782 if (vq_is_packed(dev)) {
1784 * Bit[0:14]: avail index
1785 * Bit[15]: avail wrap counter
1787 val = vq->last_avail_idx & 0x7fff;
1788 val |= vq->avail_wrap_counter << 15;
1789 msg->payload.state.num = val;
1791 msg->payload.state.num = vq->last_avail_idx;
1794 RTE_LOG(INFO, VHOST_CONFIG,
1795 "vring base idx:%d file:%d\n", msg->payload.state.index,
1796 msg->payload.state.num);
1798 * Based on current qemu vhost-user implementation, this message is
1799 * sent and only sent in vhost_vring_stop.
1800 * TODO: cleanup the vring, it isn't usable since here.
1802 if (vq->kickfd >= 0)
1805 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1807 if (vq->callfd >= 0)
1810 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1812 vq->signalled_used_valid = false;
1814 if (dev->dequeue_zero_copy)
1816 if (vq_is_packed(dev)) {
1817 rte_free(vq->shadow_used_packed);
1818 vq->shadow_used_packed = NULL;
1820 rte_free(vq->shadow_used_split);
1821 vq->shadow_used_split = NULL;
1824 rte_free(vq->batch_copy_elems);
1825 vq->batch_copy_elems = NULL;
1827 msg->size = sizeof(msg->payload.state);
1830 vring_invalidate(dev, vq);
1832 return RTE_VHOST_MSG_RESULT_REPLY;
1836 * when virtio queues are ready to work, qemu will send us to
1837 * enable the virtio queue pair.
1840 vhost_user_set_vring_enable(struct virtio_net **pdev,
1841 struct VhostUserMsg *msg,
1842 int main_fd __rte_unused)
1844 struct virtio_net *dev = *pdev;
1845 int enable = (int)msg->payload.state.num;
1846 int index = (int)msg->payload.state.index;
1847 struct rte_vdpa_device *vdpa_dev;
1850 RTE_LOG(INFO, VHOST_CONFIG,
1851 "set queue enable: %d to qp idx: %d\n",
1854 did = dev->vdpa_dev_id;
1855 vdpa_dev = rte_vdpa_get_device(did);
1856 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1857 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1859 if (dev->notify_ops->vring_state_changed)
1860 dev->notify_ops->vring_state_changed(dev->vid,
1863 /* On disable, rings have to be stopped being processed. */
1864 if (!enable && dev->dequeue_zero_copy)
1865 drain_zmbuf_list(dev->virtqueue[index]);
1867 dev->virtqueue[index]->enabled = enable;
1869 return RTE_VHOST_MSG_RESULT_OK;
1873 vhost_user_get_protocol_features(struct virtio_net **pdev,
1874 struct VhostUserMsg *msg,
1875 int main_fd __rte_unused)
1877 struct virtio_net *dev = *pdev;
1878 uint64_t features, protocol_features;
1880 rte_vhost_driver_get_features(dev->ifname, &features);
1881 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1884 * REPLY_ACK protocol feature is only mandatory for now
1885 * for IOMMU feature. If IOMMU is explicitly disabled by the
1886 * application, disable also REPLY_ACK feature for older buggy
1887 * Qemu versions (from v2.7.0 to v2.9.0).
1889 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1890 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1892 msg->payload.u64 = protocol_features;
1893 msg->size = sizeof(msg->payload.u64);
1896 return RTE_VHOST_MSG_RESULT_REPLY;
1900 vhost_user_set_protocol_features(struct virtio_net **pdev,
1901 struct VhostUserMsg *msg,
1902 int main_fd __rte_unused)
1904 struct virtio_net *dev = *pdev;
1905 uint64_t protocol_features = msg->payload.u64;
1906 uint64_t slave_protocol_features = 0;
1908 rte_vhost_driver_get_protocol_features(dev->ifname,
1909 &slave_protocol_features);
1910 if (protocol_features & ~slave_protocol_features) {
1911 RTE_LOG(ERR, VHOST_CONFIG,
1912 "(%d) received invalid protocol features.\n",
1914 return RTE_VHOST_MSG_RESULT_ERR;
1917 dev->protocol_features = protocol_features;
1918 RTE_LOG(INFO, VHOST_CONFIG,
1919 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
1920 dev->protocol_features);
1922 return RTE_VHOST_MSG_RESULT_OK;
1926 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
1927 int main_fd __rte_unused)
1929 struct virtio_net *dev = *pdev;
1930 int fd = msg->fds[0];
1935 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1936 return RTE_VHOST_MSG_RESULT_ERR;
1939 if (msg->size != sizeof(VhostUserLog)) {
1940 RTE_LOG(ERR, VHOST_CONFIG,
1941 "invalid log base msg size: %"PRId32" != %d\n",
1942 msg->size, (int)sizeof(VhostUserLog));
1943 return RTE_VHOST_MSG_RESULT_ERR;
1946 size = msg->payload.log.mmap_size;
1947 off = msg->payload.log.mmap_offset;
1949 /* Don't allow mmap_offset to point outside the mmap region */
1951 RTE_LOG(ERR, VHOST_CONFIG,
1952 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1954 return RTE_VHOST_MSG_RESULT_ERR;
1957 RTE_LOG(INFO, VHOST_CONFIG,
1958 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1962 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1963 * fail when offset is not page size aligned.
1965 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1967 if (addr == MAP_FAILED) {
1968 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1969 return RTE_VHOST_MSG_RESULT_ERR;
1973 * Free previously mapped log memory on occasionally
1974 * multiple VHOST_USER_SET_LOG_BASE.
1976 if (dev->log_addr) {
1977 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1979 dev->log_addr = (uint64_t)(uintptr_t)addr;
1980 dev->log_base = dev->log_addr + off;
1981 dev->log_size = size;
1984 * The spec is not clear about it (yet), but QEMU doesn't expect
1985 * any payload in the reply.
1990 return RTE_VHOST_MSG_RESULT_REPLY;
1993 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1994 struct VhostUserMsg *msg,
1995 int main_fd __rte_unused)
1998 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
2000 return RTE_VHOST_MSG_RESULT_OK;
2004 * An rarp packet is constructed and broadcasted to notify switches about
2005 * the new location of the migrated VM, so that packets from outside will
2006 * not be lost after migration.
2008 * However, we don't actually "send" a rarp packet here, instead, we set
2009 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2012 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2013 int main_fd __rte_unused)
2015 struct virtio_net *dev = *pdev;
2016 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2017 struct rte_vdpa_device *vdpa_dev;
2020 RTE_LOG(DEBUG, VHOST_CONFIG,
2021 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2022 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2023 memcpy(dev->mac.addr_bytes, mac, 6);
2026 * Set the flag to inject a RARP broadcast packet at
2027 * rte_vhost_dequeue_burst().
2029 * rte_smp_wmb() is for making sure the mac is copied
2030 * before the flag is set.
2033 rte_atomic16_set(&dev->broadcast_rarp, 1);
2034 did = dev->vdpa_dev_id;
2035 vdpa_dev = rte_vdpa_get_device(did);
2036 if (vdpa_dev && vdpa_dev->ops->migration_done)
2037 vdpa_dev->ops->migration_done(dev->vid);
2039 return RTE_VHOST_MSG_RESULT_OK;
2043 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2044 int main_fd __rte_unused)
2046 struct virtio_net *dev = *pdev;
2047 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2048 msg->payload.u64 > VIRTIO_MAX_MTU) {
2049 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
2052 return RTE_VHOST_MSG_RESULT_ERR;
2055 dev->mtu = msg->payload.u64;
2057 return RTE_VHOST_MSG_RESULT_OK;
2061 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2062 int main_fd __rte_unused)
2064 struct virtio_net *dev = *pdev;
2065 int fd = msg->fds[0];
2068 RTE_LOG(ERR, VHOST_CONFIG,
2069 "Invalid file descriptor for slave channel (%d)\n",
2071 return RTE_VHOST_MSG_RESULT_ERR;
2074 if (dev->slave_req_fd >= 0)
2075 close(dev->slave_req_fd);
2077 dev->slave_req_fd = fd;
2079 return RTE_VHOST_MSG_RESULT_OK;
2083 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2085 struct vhost_vring_addr *ra;
2086 uint64_t start, end, len;
2089 end = start + imsg->size;
2091 ra = &vq->ring_addrs;
2092 len = sizeof(struct vring_desc) * vq->size;
2093 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2096 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2097 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2100 len = sizeof(struct vring_used) +
2101 sizeof(struct vring_used_elem) * vq->size;
2102 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2109 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2111 struct vhost_vring_addr *ra;
2112 uint64_t start, end, len;
2115 end = start + imsg->size;
2117 ra = &vq->ring_addrs;
2118 len = sizeof(struct vring_packed_desc) * vq->size;
2119 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2122 len = sizeof(struct vring_packed_desc_event);
2123 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2126 len = sizeof(struct vring_packed_desc_event);
2127 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2133 static int is_vring_iotlb(struct virtio_net *dev,
2134 struct vhost_virtqueue *vq,
2135 struct vhost_iotlb_msg *imsg)
2137 if (vq_is_packed(dev))
2138 return is_vring_iotlb_packed(vq, imsg);
2140 return is_vring_iotlb_split(vq, imsg);
2144 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2145 int main_fd __rte_unused)
2147 struct virtio_net *dev = *pdev;
2148 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2152 switch (imsg->type) {
2153 case VHOST_IOTLB_UPDATE:
2155 vva = qva_to_vva(dev, imsg->uaddr, &len);
2157 return RTE_VHOST_MSG_RESULT_ERR;
2159 for (i = 0; i < dev->nr_vring; i++) {
2160 struct vhost_virtqueue *vq = dev->virtqueue[i];
2162 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2165 if (is_vring_iotlb(dev, vq, imsg))
2166 *pdev = dev = translate_ring_addresses(dev, i);
2169 case VHOST_IOTLB_INVALIDATE:
2170 for (i = 0; i < dev->nr_vring; i++) {
2171 struct vhost_virtqueue *vq = dev->virtqueue[i];
2173 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2176 if (is_vring_iotlb(dev, vq, imsg))
2177 vring_invalidate(dev, vq);
2181 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
2183 return RTE_VHOST_MSG_RESULT_ERR;
2186 return RTE_VHOST_MSG_RESULT_OK;
2190 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2191 struct VhostUserMsg *msg,
2192 int main_fd __rte_unused)
2194 struct virtio_net *dev = *pdev;
2195 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2196 struct uffdio_api api_struct;
2198 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2200 if (dev->postcopy_ufd == -1) {
2201 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
2203 return RTE_VHOST_MSG_RESULT_ERR;
2205 api_struct.api = UFFD_API;
2206 api_struct.features = 0;
2207 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2208 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
2210 close(dev->postcopy_ufd);
2211 dev->postcopy_ufd = -1;
2212 return RTE_VHOST_MSG_RESULT_ERR;
2214 msg->fds[0] = dev->postcopy_ufd;
2217 return RTE_VHOST_MSG_RESULT_REPLY;
2219 dev->postcopy_ufd = -1;
2222 return RTE_VHOST_MSG_RESULT_ERR;
2227 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2228 struct VhostUserMsg *msg __rte_unused,
2229 int main_fd __rte_unused)
2231 struct virtio_net *dev = *pdev;
2233 if (dev->mem && dev->mem->nregions) {
2234 RTE_LOG(ERR, VHOST_CONFIG,
2235 "Regions already registered at postcopy-listen\n");
2236 return RTE_VHOST_MSG_RESULT_ERR;
2238 dev->postcopy_listening = 1;
2240 return RTE_VHOST_MSG_RESULT_OK;
2244 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2245 int main_fd __rte_unused)
2247 struct virtio_net *dev = *pdev;
2249 dev->postcopy_listening = 0;
2250 if (dev->postcopy_ufd >= 0) {
2251 close(dev->postcopy_ufd);
2252 dev->postcopy_ufd = -1;
2255 msg->payload.u64 = 0;
2256 msg->size = sizeof(msg->payload.u64);
2259 return RTE_VHOST_MSG_RESULT_REPLY;
2262 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2263 struct VhostUserMsg *msg,
2265 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2266 [VHOST_USER_NONE] = NULL,
2267 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2268 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2269 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2270 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2271 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2272 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2273 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2274 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2275 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2276 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2277 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2278 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2279 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2280 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2281 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2282 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2283 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2284 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2285 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2286 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2287 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2288 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2289 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2290 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2291 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2292 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2293 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2296 /* return bytes# of read on success or negative val on failure. */
2298 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2302 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2303 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2308 if (msg->size > sizeof(msg->payload)) {
2309 RTE_LOG(ERR, VHOST_CONFIG,
2310 "invalid msg size: %d\n", msg->size);
2313 ret = read(sockfd, &msg->payload, msg->size);
2316 if (ret != (int)msg->size) {
2317 RTE_LOG(ERR, VHOST_CONFIG,
2318 "read control message failed\n");
2327 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2332 return send_fd_message(sockfd, (char *)msg,
2333 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2337 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2342 msg->flags &= ~VHOST_USER_VERSION_MASK;
2343 msg->flags &= ~VHOST_USER_NEED_REPLY;
2344 msg->flags |= VHOST_USER_VERSION;
2345 msg->flags |= VHOST_USER_REPLY_MASK;
2347 return send_vhost_message(sockfd, msg);
2351 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2355 if (msg->flags & VHOST_USER_NEED_REPLY)
2356 rte_spinlock_lock(&dev->slave_req_lock);
2358 ret = send_vhost_message(dev->slave_req_fd, msg);
2359 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2360 rte_spinlock_unlock(&dev->slave_req_lock);
2366 * Allocate a queue pair if it hasn't been allocated yet
2369 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2370 struct VhostUserMsg *msg)
2374 switch (msg->request.master) {
2375 case VHOST_USER_SET_VRING_KICK:
2376 case VHOST_USER_SET_VRING_CALL:
2377 case VHOST_USER_SET_VRING_ERR:
2378 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2380 case VHOST_USER_SET_VRING_NUM:
2381 case VHOST_USER_SET_VRING_BASE:
2382 case VHOST_USER_SET_VRING_ENABLE:
2383 vring_idx = msg->payload.state.index;
2385 case VHOST_USER_SET_VRING_ADDR:
2386 vring_idx = msg->payload.addr.index;
2392 if (vring_idx >= VHOST_MAX_VRING) {
2393 RTE_LOG(ERR, VHOST_CONFIG,
2394 "invalid vring index: %u\n", vring_idx);
2398 if (dev->virtqueue[vring_idx])
2401 return alloc_vring_queue(dev, vring_idx);
2405 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2408 unsigned int vq_num = 0;
2410 while (vq_num < dev->nr_vring) {
2411 struct vhost_virtqueue *vq = dev->virtqueue[i];
2414 rte_spinlock_lock(&vq->access_lock);
2422 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2425 unsigned int vq_num = 0;
2427 while (vq_num < dev->nr_vring) {
2428 struct vhost_virtqueue *vq = dev->virtqueue[i];
2431 rte_spinlock_unlock(&vq->access_lock);
2439 vhost_user_msg_handler(int vid, int fd)
2441 struct virtio_net *dev;
2442 struct VhostUserMsg msg;
2443 struct rte_vdpa_device *vdpa_dev;
2446 int unlock_required = 0;
2450 dev = get_device(vid);
2454 if (!dev->notify_ops) {
2455 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2456 if (!dev->notify_ops) {
2457 RTE_LOG(ERR, VHOST_CONFIG,
2458 "failed to get callback ops for driver %s\n",
2464 ret = read_vhost_message(fd, &msg);
2467 RTE_LOG(ERR, VHOST_CONFIG,
2468 "vhost read message failed\n");
2470 RTE_LOG(INFO, VHOST_CONFIG,
2471 "vhost peer closed\n");
2477 request = msg.request.master;
2478 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2479 vhost_message_str[request]) {
2480 if (request != VHOST_USER_IOTLB_MSG)
2481 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
2482 vhost_message_str[request]);
2484 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
2485 vhost_message_str[request]);
2487 RTE_LOG(DEBUG, VHOST_CONFIG, "External request %d\n", request);
2490 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2492 RTE_LOG(ERR, VHOST_CONFIG,
2493 "failed to alloc queue\n");
2498 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2499 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2500 * and device is destroyed. destroy_device waits for queues to be
2501 * inactive, so it is safe. Otherwise taking the access_lock
2502 * would cause a dead lock.
2505 case VHOST_USER_SET_FEATURES:
2506 case VHOST_USER_SET_PROTOCOL_FEATURES:
2507 case VHOST_USER_SET_OWNER:
2508 case VHOST_USER_SET_MEM_TABLE:
2509 case VHOST_USER_SET_LOG_BASE:
2510 case VHOST_USER_SET_LOG_FD:
2511 case VHOST_USER_SET_VRING_NUM:
2512 case VHOST_USER_SET_VRING_ADDR:
2513 case VHOST_USER_SET_VRING_BASE:
2514 case VHOST_USER_SET_VRING_KICK:
2515 case VHOST_USER_SET_VRING_CALL:
2516 case VHOST_USER_SET_VRING_ERR:
2517 case VHOST_USER_SET_VRING_ENABLE:
2518 case VHOST_USER_SEND_RARP:
2519 case VHOST_USER_NET_SET_MTU:
2520 case VHOST_USER_SET_SLAVE_REQ_FD:
2521 vhost_user_lock_all_queue_pairs(dev);
2522 unlock_required = 1;
2530 if (dev->extern_ops.pre_msg_handle) {
2531 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2534 case RTE_VHOST_MSG_RESULT_REPLY:
2535 send_vhost_reply(fd, &msg);
2537 case RTE_VHOST_MSG_RESULT_ERR:
2538 case RTE_VHOST_MSG_RESULT_OK:
2540 goto skip_to_post_handle;
2541 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2547 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2548 if (!vhost_message_handlers[request])
2549 goto skip_to_post_handle;
2550 ret = vhost_message_handlers[request](&dev, &msg, fd);
2553 case RTE_VHOST_MSG_RESULT_ERR:
2554 RTE_LOG(ERR, VHOST_CONFIG,
2555 "Processing %s failed.\n",
2556 vhost_message_str[request]);
2559 case RTE_VHOST_MSG_RESULT_OK:
2560 RTE_LOG(DEBUG, VHOST_CONFIG,
2561 "Processing %s succeeded.\n",
2562 vhost_message_str[request]);
2565 case RTE_VHOST_MSG_RESULT_REPLY:
2566 RTE_LOG(DEBUG, VHOST_CONFIG,
2567 "Processing %s succeeded and needs reply.\n",
2568 vhost_message_str[request]);
2569 send_vhost_reply(fd, &msg);
2577 skip_to_post_handle:
2578 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2579 dev->extern_ops.post_msg_handle) {
2580 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2583 case RTE_VHOST_MSG_RESULT_REPLY:
2584 send_vhost_reply(fd, &msg);
2586 case RTE_VHOST_MSG_RESULT_ERR:
2587 case RTE_VHOST_MSG_RESULT_OK:
2589 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2595 if (unlock_required)
2596 vhost_user_unlock_all_queue_pairs(dev);
2598 /* If message was not handled at this stage, treat it as an error */
2600 RTE_LOG(ERR, VHOST_CONFIG,
2601 "vhost message (req: %d) was not handled.\n", request);
2602 ret = RTE_VHOST_MSG_RESULT_ERR;
2606 * If the request required a reply that was already sent,
2607 * this optional reply-ack won't be sent as the
2608 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2610 if (msg.flags & VHOST_USER_NEED_REPLY) {
2611 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2612 msg.size = sizeof(msg.payload.u64);
2614 send_vhost_reply(fd, &msg);
2615 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2616 RTE_LOG(ERR, VHOST_CONFIG,
2617 "vhost message handling failed.\n");
2621 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2622 dev->flags |= VIRTIO_DEV_READY;
2624 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2625 if (dev->dequeue_zero_copy) {
2626 RTE_LOG(INFO, VHOST_CONFIG,
2627 "dequeue zero copy is enabled\n");
2630 if (dev->notify_ops->new_device(dev->vid) == 0)
2631 dev->flags |= VIRTIO_DEV_RUNNING;
2635 did = dev->vdpa_dev_id;
2636 vdpa_dev = rte_vdpa_get_device(did);
2637 if (vdpa_dev && virtio_is_ready(dev) &&
2638 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2639 msg.request.master == VHOST_USER_SET_VRING_CALL) {
2640 if (vdpa_dev->ops->dev_conf)
2641 vdpa_dev->ops->dev_conf(dev->vid);
2642 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2648 static int process_slave_message_reply(struct virtio_net *dev,
2649 const struct VhostUserMsg *msg)
2651 struct VhostUserMsg msg_reply;
2654 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2657 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2662 if (msg_reply.request.slave != msg->request.slave) {
2663 RTE_LOG(ERR, VHOST_CONFIG,
2664 "Received unexpected msg type (%u), expected %u\n",
2665 msg_reply.request.slave, msg->request.slave);
2670 ret = msg_reply.payload.u64 ? -1 : 0;
2673 rte_spinlock_unlock(&dev->slave_req_lock);
2678 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2681 struct VhostUserMsg msg = {
2682 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2683 .flags = VHOST_USER_VERSION,
2684 .size = sizeof(msg.payload.iotlb),
2688 .type = VHOST_IOTLB_MISS,
2692 ret = send_vhost_message(dev->slave_req_fd, &msg);
2694 RTE_LOG(ERR, VHOST_CONFIG,
2695 "Failed to send IOTLB miss message (%d)\n",
2703 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2709 struct VhostUserMsg msg = {
2710 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2711 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2712 .size = sizeof(msg.payload.area),
2714 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2721 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2727 ret = send_vhost_slave_message(dev, &msg);
2729 RTE_LOG(ERR, VHOST_CONFIG,
2730 "Failed to set host notifier (%d)\n", ret);
2734 return process_slave_message_reply(dev, &msg);
2737 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2739 struct virtio_net *dev;
2740 struct rte_vdpa_device *vdpa_dev;
2741 int vfio_device_fd, did, ret = 0;
2742 uint64_t offset, size;
2745 dev = get_device(vid);
2749 did = dev->vdpa_dev_id;
2753 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2754 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2755 !(dev->protocol_features &
2756 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2757 !(dev->protocol_features &
2758 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2759 !(dev->protocol_features &
2760 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2763 vdpa_dev = rte_vdpa_get_device(did);
2767 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2768 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2770 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2771 if (vfio_device_fd < 0)
2775 for (i = 0; i < dev->nr_vring; i++) {
2776 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2782 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2783 vfio_device_fd, offset, size) < 0) {
2790 for (i = 0; i < dev->nr_vring; i++) {
2791 vhost_user_slave_set_vring_host_notifier(dev, i, -1,