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>
49 #include <rte_errno.h>
53 #include "vhost_user.h"
55 #define VIRTIO_MIN_MTU 68
56 #define VIRTIO_MAX_MTU 65535
58 #define INFLIGHT_ALIGNMENT 64
59 #define INFLIGHT_VERSION 0x1
61 static const char *vhost_message_str[VHOST_USER_MAX] = {
62 [VHOST_USER_NONE] = "VHOST_USER_NONE",
63 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
64 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
65 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
66 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
67 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
68 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
69 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
70 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
71 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
72 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
73 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
74 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
75 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
76 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
77 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
78 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
79 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
80 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
81 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
82 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
83 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
84 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
85 [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
86 [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
87 [VHOST_USER_POSTCOPY_ADVISE] = "VHOST_USER_POSTCOPY_ADVISE",
88 [VHOST_USER_POSTCOPY_LISTEN] = "VHOST_USER_POSTCOPY_LISTEN",
89 [VHOST_USER_POSTCOPY_END] = "VHOST_USER_POSTCOPY_END",
90 [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
91 [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
92 [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
93 [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
96 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
97 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
100 close_msg_fds(struct VhostUserMsg *msg)
104 for (i = 0; i < msg->fd_num; i++) {
105 int fd = msg->fds[i];
116 * Ensure the expected number of FDs is received,
117 * close all FDs and return an error if this is not the case.
120 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
122 if (msg->fd_num == expected_fds)
125 VHOST_LOG_CONFIG(ERR,
126 " Expect %d FDs for request %s, received %d\n",
128 vhost_message_str[msg->request.master],
142 ret = fstat(fd, &stat);
143 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
147 async_dma_map(struct rte_vhost_mem_region *region, bool *dma_map_success, bool do_map)
152 host_iova = rte_mem_virt2iova((void *)(uintptr_t)region->host_user_addr);
154 /* Add mapped region into the default container of DPDK. */
155 ret = rte_vfio_container_dma_map(RTE_VFIO_DEFAULT_CONTAINER_FD,
156 region->host_user_addr,
159 *dma_map_success = ret == 0;
163 * DMA device may bind with kernel driver, in this case,
164 * we don't need to program IOMMU manually. However, if no
165 * device is bound with vfio/uio in DPDK, and vfio kernel
166 * module is loaded, the API will still be called and return
167 * with ENODEV/ENOSUP.
169 * DPDK vfio only returns ENODEV/ENOSUP in very similar
170 * situations(vfio either unsupported, or supported
171 * but no devices found). Either way, no mappings could be
172 * performed. We treat it as normal case in async path.
174 if (rte_errno == ENODEV || rte_errno == ENOTSUP)
177 VHOST_LOG_CONFIG(ERR, "DMA engine map failed\n");
183 /* No need to do vfio unmap if the map failed. */
184 if (!*dma_map_success)
187 /* Remove mapped region from the default container of DPDK. */
188 ret = rte_vfio_container_dma_unmap(RTE_VFIO_DEFAULT_CONTAINER_FD,
189 region->host_user_addr,
193 VHOST_LOG_CONFIG(ERR, "DMA engine unmap failed\n");
196 /* Clear the flag once the unmap succeeds. */
197 *dma_map_success = 0;
204 free_mem_region(struct virtio_net *dev)
207 struct rte_vhost_mem_region *reg;
209 if (!dev || !dev->mem)
212 for (i = 0; i < dev->mem->nregions; i++) {
213 reg = &dev->mem->regions[i];
214 if (reg->host_user_addr) {
215 if (dev->async_copy && rte_vfio_is_enabled("vfio"))
216 async_dma_map(reg, &dev->async_map_status[i], false);
218 munmap(reg->mmap_addr, reg->mmap_size);
225 vhost_backend_cleanup(struct virtio_net *dev)
228 free_mem_region(dev);
232 if (dev->async_map_status) {
233 rte_free(dev->async_map_status);
234 dev->async_map_status = NULL;
238 rte_free(dev->guest_pages);
239 dev->guest_pages = NULL;
242 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
246 if (dev->inflight_info) {
247 if (dev->inflight_info->addr) {
248 munmap(dev->inflight_info->addr,
249 dev->inflight_info->size);
250 dev->inflight_info->addr = NULL;
253 if (dev->inflight_info->fd >= 0) {
254 close(dev->inflight_info->fd);
255 dev->inflight_info->fd = -1;
258 rte_free(dev->inflight_info);
259 dev->inflight_info = NULL;
262 if (dev->slave_req_fd >= 0) {
263 close(dev->slave_req_fd);
264 dev->slave_req_fd = -1;
267 if (dev->postcopy_ufd >= 0) {
268 close(dev->postcopy_ufd);
269 dev->postcopy_ufd = -1;
272 dev->postcopy_listening = 0;
276 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
279 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
280 struct vhost_virtqueue *vq = dev->virtqueue[index];
282 /* Configure guest notifications on enable */
283 if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
284 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
286 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
287 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
289 if (dev->notify_ops->vring_state_changed)
290 dev->notify_ops->vring_state_changed(dev->vid,
295 * This function just returns success at the moment unless
296 * the device hasn't been initialised.
299 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
300 struct VhostUserMsg *msg,
301 int main_fd __rte_unused)
303 if (validate_msg_fds(msg, 0) != 0)
304 return RTE_VHOST_MSG_RESULT_ERR;
306 return RTE_VHOST_MSG_RESULT_OK;
310 vhost_user_reset_owner(struct virtio_net **pdev,
311 struct VhostUserMsg *msg,
312 int main_fd __rte_unused)
314 struct virtio_net *dev = *pdev;
316 if (validate_msg_fds(msg, 0) != 0)
317 return RTE_VHOST_MSG_RESULT_ERR;
319 vhost_destroy_device_notify(dev);
321 cleanup_device(dev, 0);
323 return RTE_VHOST_MSG_RESULT_OK;
327 * The features that we support are requested.
330 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
331 int main_fd __rte_unused)
333 struct virtio_net *dev = *pdev;
334 uint64_t features = 0;
336 if (validate_msg_fds(msg, 0) != 0)
337 return RTE_VHOST_MSG_RESULT_ERR;
339 rte_vhost_driver_get_features(dev->ifname, &features);
341 msg->payload.u64 = features;
342 msg->size = sizeof(msg->payload.u64);
345 return RTE_VHOST_MSG_RESULT_REPLY;
349 * The queue number that we support are requested.
352 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
353 int main_fd __rte_unused)
355 struct virtio_net *dev = *pdev;
356 uint32_t queue_num = 0;
358 if (validate_msg_fds(msg, 0) != 0)
359 return RTE_VHOST_MSG_RESULT_ERR;
361 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
363 msg->payload.u64 = (uint64_t)queue_num;
364 msg->size = sizeof(msg->payload.u64);
367 return RTE_VHOST_MSG_RESULT_REPLY;
371 * We receive the negotiated features supported by us and the virtio device.
374 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
375 int main_fd __rte_unused)
377 struct virtio_net *dev = *pdev;
378 uint64_t features = msg->payload.u64;
379 uint64_t vhost_features = 0;
380 struct rte_vdpa_device *vdpa_dev;
382 if (validate_msg_fds(msg, 0) != 0)
383 return RTE_VHOST_MSG_RESULT_ERR;
385 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
386 if (features & ~vhost_features) {
387 VHOST_LOG_CONFIG(ERR,
388 "(%d) received invalid negotiated features.\n",
390 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
391 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
393 return RTE_VHOST_MSG_RESULT_ERR;
396 if (dev->flags & VIRTIO_DEV_RUNNING) {
397 if (dev->features == features)
398 return RTE_VHOST_MSG_RESULT_OK;
401 * Error out if master tries to change features while device is
402 * in running state. The exception being VHOST_F_LOG_ALL, which
403 * is enabled when the live-migration starts.
405 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
406 VHOST_LOG_CONFIG(ERR,
407 "(%d) features changed while device is running.\n",
409 return RTE_VHOST_MSG_RESULT_ERR;
412 if (dev->notify_ops->features_changed)
413 dev->notify_ops->features_changed(dev->vid, features);
416 dev->features = features;
418 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
419 (1ULL << VIRTIO_F_VERSION_1) |
420 (1ULL << VIRTIO_F_RING_PACKED))) {
421 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
423 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
425 VHOST_LOG_CONFIG(INFO,
426 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
427 VHOST_LOG_CONFIG(DEBUG,
428 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
430 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
431 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
433 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
434 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
436 * Remove all but first queue pair if MQ hasn't been
437 * negotiated. This is safe because the device is not
438 * running at this stage.
440 while (dev->nr_vring > 2) {
441 struct vhost_virtqueue *vq;
443 vq = dev->virtqueue[--dev->nr_vring];
447 dev->virtqueue[dev->nr_vring] = NULL;
449 cleanup_vq_inflight(dev, vq);
454 vdpa_dev = dev->vdpa_dev;
456 vdpa_dev->ops->set_features(dev->vid);
458 dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
459 return RTE_VHOST_MSG_RESULT_OK;
463 * The virtio device sends us the size of the descriptor ring.
466 vhost_user_set_vring_num(struct virtio_net **pdev,
467 struct VhostUserMsg *msg,
468 int main_fd __rte_unused)
470 struct virtio_net *dev = *pdev;
471 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
473 if (validate_msg_fds(msg, 0) != 0)
474 return RTE_VHOST_MSG_RESULT_ERR;
476 if (msg->payload.state.num > 32768) {
477 VHOST_LOG_CONFIG(ERR, "invalid virtqueue size %u\n", msg->payload.state.num);
478 return RTE_VHOST_MSG_RESULT_ERR;
481 vq->size = msg->payload.state.num;
483 /* VIRTIO 1.0, 2.4 Virtqueues says:
485 * Queue Size value is always a power of 2. The maximum Queue Size
488 * VIRTIO 1.1 2.7 Virtqueues says:
490 * Packed virtqueues support up to 2^15 entries each.
492 if (!vq_is_packed(dev)) {
493 if (vq->size & (vq->size - 1)) {
494 VHOST_LOG_CONFIG(ERR,
495 "invalid virtqueue size %u\n", vq->size);
496 return RTE_VHOST_MSG_RESULT_ERR;
500 if (vq_is_packed(dev)) {
501 if (vq->shadow_used_packed)
502 rte_free(vq->shadow_used_packed);
503 vq->shadow_used_packed = rte_malloc_socket(NULL,
505 sizeof(struct vring_used_elem_packed),
506 RTE_CACHE_LINE_SIZE, vq->numa_node);
507 if (!vq->shadow_used_packed) {
508 VHOST_LOG_CONFIG(ERR,
509 "failed to allocate memory for shadow used ring.\n");
510 return RTE_VHOST_MSG_RESULT_ERR;
514 if (vq->shadow_used_split)
515 rte_free(vq->shadow_used_split);
517 vq->shadow_used_split = rte_malloc_socket(NULL,
518 vq->size * sizeof(struct vring_used_elem),
519 RTE_CACHE_LINE_SIZE, vq->numa_node);
521 if (!vq->shadow_used_split) {
522 VHOST_LOG_CONFIG(ERR,
523 "failed to allocate memory for vq internal data.\n");
524 return RTE_VHOST_MSG_RESULT_ERR;
528 if (vq->batch_copy_elems)
529 rte_free(vq->batch_copy_elems);
530 vq->batch_copy_elems = rte_malloc_socket(NULL,
531 vq->size * sizeof(struct batch_copy_elem),
532 RTE_CACHE_LINE_SIZE, vq->numa_node);
533 if (!vq->batch_copy_elems) {
534 VHOST_LOG_CONFIG(ERR,
535 "failed to allocate memory for batching copy.\n");
536 return RTE_VHOST_MSG_RESULT_ERR;
539 return RTE_VHOST_MSG_RESULT_OK;
543 * Reallocate virtio_dev, vhost_virtqueue and related data structures to
544 * make them on the same numa node as the memory of vring descriptor.
546 #ifdef RTE_LIBRTE_VHOST_NUMA
547 static struct virtio_net*
548 numa_realloc(struct virtio_net *dev, int index)
551 struct virtio_net *old_dev;
552 struct vhost_virtqueue *vq;
553 struct batch_copy_elem *bce;
554 struct guest_page *gp;
555 struct rte_vhost_memory *mem;
560 vq = dev->virtqueue[index];
563 * If VQ is ready, it is too late to reallocate, it certainly already
564 * happened anyway on VHOST_USER_SET_VRING_ADRR.
569 ret = get_mempolicy(&node, NULL, 0, vq->desc, MPOL_F_NODE | MPOL_F_ADDR);
571 VHOST_LOG_CONFIG(ERR, "Unable to get virtqueue %d numa information.\n", index);
575 if (node == vq->numa_node)
576 goto out_dev_realloc;
578 vq = rte_realloc_socket(vq, sizeof(*vq), 0, node);
580 VHOST_LOG_CONFIG(ERR, "Failed to realloc virtqueue %d on node %d\n",
585 if (vq != dev->virtqueue[index]) {
586 VHOST_LOG_CONFIG(INFO, "reallocated virtqueue on node %d\n", node);
587 dev->virtqueue[index] = vq;
588 vhost_user_iotlb_init(dev, index);
591 if (vq_is_packed(dev)) {
592 struct vring_used_elem_packed *sup;
594 sup = rte_realloc_socket(vq->shadow_used_packed, vq->size * sizeof(*sup),
595 RTE_CACHE_LINE_SIZE, node);
597 VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow packed on node %d\n", node);
600 vq->shadow_used_packed = sup;
602 struct vring_used_elem *sus;
604 sus = rte_realloc_socket(vq->shadow_used_split, vq->size * sizeof(*sus),
605 RTE_CACHE_LINE_SIZE, node);
607 VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow split on node %d\n", node);
610 vq->shadow_used_split = sus;
613 bce = rte_realloc_socket(vq->batch_copy_elems, vq->size * sizeof(*bce),
614 RTE_CACHE_LINE_SIZE, node);
616 VHOST_LOG_CONFIG(ERR, "Failed to realloc batch copy elem on node %d\n", node);
619 vq->batch_copy_elems = bce;
622 struct log_cache_entry *lc;
624 lc = rte_realloc_socket(vq->log_cache, sizeof(*lc) * VHOST_LOG_CACHE_NR, 0, node);
626 VHOST_LOG_CONFIG(ERR, "Failed to realloc log cache on node %d\n", node);
632 if (vq->resubmit_inflight) {
633 struct rte_vhost_resubmit_info *ri;
635 ri = rte_realloc_socket(vq->resubmit_inflight, sizeof(*ri), 0, node);
637 VHOST_LOG_CONFIG(ERR, "Failed to realloc resubmit inflight on node %d\n",
641 vq->resubmit_inflight = ri;
643 if (ri->resubmit_list) {
644 struct rte_vhost_resubmit_desc *rd;
646 rd = rte_realloc_socket(ri->resubmit_list, sizeof(*rd) * ri->resubmit_num,
649 VHOST_LOG_CONFIG(ERR, "Failed to realloc resubmit list on node %d\n",
653 ri->resubmit_list = rd;
657 vq->numa_node = node;
661 if (dev->flags & VIRTIO_DEV_RUNNING)
664 ret = get_mempolicy(&dev_node, NULL, 0, dev, MPOL_F_NODE | MPOL_F_ADDR);
666 VHOST_LOG_CONFIG(ERR, "Unable to get Virtio dev %d numa information.\n", dev->vid);
670 if (dev_node == node)
673 dev = rte_realloc_socket(old_dev, sizeof(*dev), 0, node);
675 VHOST_LOG_CONFIG(ERR, "Failed to realloc dev on node %d\n", node);
679 VHOST_LOG_CONFIG(INFO, "reallocated device on node %d\n", node);
680 vhost_devices[dev->vid] = dev;
682 mem_size = sizeof(struct rte_vhost_memory) +
683 sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
684 mem = rte_realloc_socket(dev->mem, mem_size, 0, node);
686 VHOST_LOG_CONFIG(ERR, "Failed to realloc mem table on node %d\n", node);
691 if (dev->async_copy && rte_vfio_is_enabled("vfio")) {
692 if (dev->async_map_status == NULL) {
693 dev->async_map_status = rte_zmalloc_socket("async-dma-map-status",
694 sizeof(bool) * dev->mem->nregions, 0, node);
695 if (!dev->async_map_status) {
696 VHOST_LOG_CONFIG(ERR,
697 "(%d) failed to realloc dma mapping status on node\n",
704 gp = rte_realloc_socket(dev->guest_pages, dev->max_guest_pages * sizeof(*gp),
705 RTE_CACHE_LINE_SIZE, node);
707 VHOST_LOG_CONFIG(ERR, "Failed to realloc guest pages on node %d\n", node);
710 dev->guest_pages = gp;
715 static struct virtio_net*
716 numa_realloc(struct virtio_net *dev, int index __rte_unused)
722 /* Converts QEMU virtual address to Vhost virtual address. */
724 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
726 struct rte_vhost_mem_region *r;
729 if (unlikely(!dev || !dev->mem))
732 /* Find the region where the address lives. */
733 for (i = 0; i < dev->mem->nregions; i++) {
734 r = &dev->mem->regions[i];
736 if (qva >= r->guest_user_addr &&
737 qva < r->guest_user_addr + r->size) {
739 if (unlikely(*len > r->guest_user_addr + r->size - qva))
740 *len = r->guest_user_addr + r->size - qva;
742 return qva - r->guest_user_addr +
754 * Converts ring address to Vhost virtual address.
755 * If IOMMU is enabled, the ring address is a guest IO virtual address,
756 * else it is a QEMU virtual address.
759 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
760 uint64_t ra, uint64_t *size)
762 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
765 vhost_user_iotlb_rd_lock(vq);
766 vva = vhost_iova_to_vva(dev, vq, ra,
767 size, VHOST_ACCESS_RW);
768 vhost_user_iotlb_rd_unlock(vq);
773 return qva_to_vva(dev, ra, size);
777 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
781 vhost_user_iotlb_rd_lock(vq);
782 log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
783 vhost_user_iotlb_rd_unlock(vq);
788 static struct virtio_net *
789 translate_ring_addresses(struct virtio_net *dev, int vq_index)
791 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
792 struct vhost_vring_addr *addr = &vq->ring_addrs;
793 uint64_t len, expected_len;
795 if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
797 log_addr_to_gpa(dev, vq);
798 if (vq->log_guest_addr == 0) {
799 VHOST_LOG_CONFIG(DEBUG,
800 "(%d) failed to map log_guest_addr.\n",
806 if (vq_is_packed(dev)) {
807 len = sizeof(struct vring_packed_desc) * vq->size;
808 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
809 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
810 if (vq->desc_packed == NULL ||
811 len != sizeof(struct vring_packed_desc) *
813 VHOST_LOG_CONFIG(DEBUG,
814 "(%d) failed to map desc_packed ring.\n",
819 dev = numa_realloc(dev, vq_index);
820 vq = dev->virtqueue[vq_index];
821 addr = &vq->ring_addrs;
823 len = sizeof(struct vring_packed_desc_event);
824 vq->driver_event = (struct vring_packed_desc_event *)
825 (uintptr_t)ring_addr_to_vva(dev,
826 vq, addr->avail_user_addr, &len);
827 if (vq->driver_event == NULL ||
828 len != sizeof(struct vring_packed_desc_event)) {
829 VHOST_LOG_CONFIG(DEBUG,
830 "(%d) failed to find driver area address.\n",
835 len = sizeof(struct vring_packed_desc_event);
836 vq->device_event = (struct vring_packed_desc_event *)
837 (uintptr_t)ring_addr_to_vva(dev,
838 vq, addr->used_user_addr, &len);
839 if (vq->device_event == NULL ||
840 len != sizeof(struct vring_packed_desc_event)) {
841 VHOST_LOG_CONFIG(DEBUG,
842 "(%d) failed to find device area address.\n",
847 vq->access_ok = true;
851 /* The addresses are converted from QEMU virtual to Vhost virtual. */
852 if (vq->desc && vq->avail && vq->used)
855 len = sizeof(struct vring_desc) * vq->size;
856 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
857 vq, addr->desc_user_addr, &len);
858 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
859 VHOST_LOG_CONFIG(DEBUG,
860 "(%d) failed to map desc ring.\n",
865 dev = numa_realloc(dev, vq_index);
866 vq = dev->virtqueue[vq_index];
867 addr = &vq->ring_addrs;
869 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
870 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
871 len += sizeof(uint16_t);
873 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
874 vq, addr->avail_user_addr, &len);
875 if (vq->avail == 0 || len != expected_len) {
876 VHOST_LOG_CONFIG(DEBUG,
877 "(%d) failed to map avail ring.\n",
882 len = sizeof(struct vring_used) +
883 sizeof(struct vring_used_elem) * vq->size;
884 if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
885 len += sizeof(uint16_t);
887 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
888 vq, addr->used_user_addr, &len);
889 if (vq->used == 0 || len != expected_len) {
890 VHOST_LOG_CONFIG(DEBUG,
891 "(%d) failed to map used ring.\n",
896 if (vq->last_used_idx != vq->used->idx) {
897 VHOST_LOG_CONFIG(WARNING,
898 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
899 "some packets maybe resent for Tx and dropped for Rx\n",
900 vq->last_used_idx, vq->used->idx);
901 vq->last_used_idx = vq->used->idx;
902 vq->last_avail_idx = vq->used->idx;
905 vq->access_ok = true;
907 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
909 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
910 dev->vid, vq->avail);
911 VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
913 VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
914 dev->vid, vq->log_guest_addr);
920 * The virtio device sends us the desc, used and avail ring addresses.
921 * This function then converts these to our address space.
924 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
925 int main_fd __rte_unused)
927 struct virtio_net *dev = *pdev;
928 struct vhost_virtqueue *vq;
929 struct vhost_vring_addr *addr = &msg->payload.addr;
932 if (validate_msg_fds(msg, 0) != 0)
933 return RTE_VHOST_MSG_RESULT_ERR;
935 if (dev->mem == NULL)
936 return RTE_VHOST_MSG_RESULT_ERR;
938 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
939 vq = dev->virtqueue[msg->payload.addr.index];
941 access_ok = vq->access_ok;
944 * Rings addresses should not be interpreted as long as the ring is not
945 * started and enabled
947 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
949 vring_invalidate(dev, vq);
951 if ((vq->enabled && (dev->features &
952 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
954 dev = translate_ring_addresses(dev, msg->payload.addr.index);
956 return RTE_VHOST_MSG_RESULT_ERR;
961 return RTE_VHOST_MSG_RESULT_OK;
965 * The virtio device sends us the available ring last used index.
968 vhost_user_set_vring_base(struct virtio_net **pdev,
969 struct VhostUserMsg *msg,
970 int main_fd __rte_unused)
972 struct virtio_net *dev = *pdev;
973 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
974 uint64_t val = msg->payload.state.num;
976 if (validate_msg_fds(msg, 0) != 0)
977 return RTE_VHOST_MSG_RESULT_ERR;
979 if (vq_is_packed(dev)) {
981 * Bit[0:14]: avail index
982 * Bit[15]: avail wrap counter
984 vq->last_avail_idx = val & 0x7fff;
985 vq->avail_wrap_counter = !!(val & (0x1 << 15));
987 * Set used index to same value as available one, as
988 * their values should be the same since ring processing
989 * was stopped at get time.
991 vq->last_used_idx = vq->last_avail_idx;
992 vq->used_wrap_counter = vq->avail_wrap_counter;
994 vq->last_used_idx = msg->payload.state.num;
995 vq->last_avail_idx = msg->payload.state.num;
998 return RTE_VHOST_MSG_RESULT_OK;
1002 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
1003 uint64_t host_phys_addr, uint64_t size)
1005 struct guest_page *page, *last_page;
1006 struct guest_page *old_pages;
1008 if (dev->nr_guest_pages == dev->max_guest_pages) {
1009 dev->max_guest_pages *= 2;
1010 old_pages = dev->guest_pages;
1011 dev->guest_pages = rte_realloc(dev->guest_pages,
1012 dev->max_guest_pages * sizeof(*page),
1013 RTE_CACHE_LINE_SIZE);
1014 if (dev->guest_pages == NULL) {
1015 VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
1016 rte_free(old_pages);
1021 if (dev->nr_guest_pages > 0) {
1022 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
1023 /* merge if the two pages are continuous */
1024 if (host_phys_addr == last_page->host_phys_addr +
1026 last_page->size += size;
1031 page = &dev->guest_pages[dev->nr_guest_pages++];
1032 page->guest_phys_addr = guest_phys_addr;
1033 page->host_phys_addr = host_phys_addr;
1040 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
1043 uint64_t reg_size = reg->size;
1044 uint64_t host_user_addr = reg->host_user_addr;
1045 uint64_t guest_phys_addr = reg->guest_phys_addr;
1046 uint64_t host_phys_addr;
1049 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
1050 size = page_size - (guest_phys_addr & (page_size - 1));
1051 size = RTE_MIN(size, reg_size);
1053 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
1056 host_user_addr += size;
1057 guest_phys_addr += size;
1060 while (reg_size > 0) {
1061 size = RTE_MIN(reg_size, page_size);
1062 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
1064 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
1068 host_user_addr += size;
1069 guest_phys_addr += size;
1073 /* sort guest page array if over binary search threshold */
1074 if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
1075 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
1076 sizeof(struct guest_page), guest_page_addrcmp);
1082 #ifdef RTE_LIBRTE_VHOST_DEBUG
1083 /* TODO: enable it only in debug mode? */
1085 dump_guest_pages(struct virtio_net *dev)
1088 struct guest_page *page;
1090 for (i = 0; i < dev->nr_guest_pages; i++) {
1091 page = &dev->guest_pages[i];
1093 VHOST_LOG_CONFIG(INFO,
1094 "guest physical page region %u\n"
1095 "\t guest_phys_addr: %" PRIx64 "\n"
1096 "\t host_phys_addr : %" PRIx64 "\n"
1097 "\t size : %" PRIx64 "\n",
1099 page->guest_phys_addr,
1100 page->host_phys_addr,
1105 #define dump_guest_pages(dev)
1109 vhost_memory_changed(struct VhostUserMemory *new,
1110 struct rte_vhost_memory *old)
1114 if (new->nregions != old->nregions)
1117 for (i = 0; i < new->nregions; ++i) {
1118 VhostUserMemoryRegion *new_r = &new->regions[i];
1119 struct rte_vhost_mem_region *old_r = &old->regions[i];
1121 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1123 if (new_r->memory_size != old_r->size)
1125 if (new_r->userspace_addr != old_r->guest_user_addr)
1132 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1134 vhost_user_postcopy_region_register(struct virtio_net *dev,
1135 struct rte_vhost_mem_region *reg)
1137 struct uffdio_register reg_struct;
1140 * Let's register all the mmap'ed area to ensure
1141 * alignment on page boundary.
1143 reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1144 reg_struct.range.len = reg->mmap_size;
1145 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1147 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1149 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1150 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1151 (uint64_t)reg_struct.range.start,
1152 (uint64_t)reg_struct.range.start +
1153 (uint64_t)reg_struct.range.len - 1,
1159 VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1160 (uint64_t)reg_struct.range.start,
1161 (uint64_t)reg_struct.range.start +
1162 (uint64_t)reg_struct.range.len - 1);
1168 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1169 struct rte_vhost_mem_region *reg __rte_unused)
1176 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1177 struct VhostUserMsg *msg)
1179 struct VhostUserMemory *memory;
1180 struct rte_vhost_mem_region *reg;
1181 VhostUserMsg ack_msg;
1184 if (!dev->postcopy_listening)
1188 * We haven't a better way right now than sharing
1189 * DPDK's virtual address with Qemu, so that Qemu can
1190 * retrieve the region offset when handling userfaults.
1192 memory = &msg->payload.memory;
1193 for (i = 0; i < memory->nregions; i++) {
1194 reg = &dev->mem->regions[i];
1195 memory->regions[i].userspace_addr = reg->host_user_addr;
1198 /* Send the addresses back to qemu */
1200 send_vhost_reply(main_fd, msg);
1202 /* Wait for qemu to acknolwedge it's got the addresses
1203 * we've got to wait before we're allowed to generate faults.
1205 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1206 VHOST_LOG_CONFIG(ERR,
1207 "Failed to read qemu ack on postcopy set-mem-table\n");
1211 if (validate_msg_fds(&ack_msg, 0) != 0)
1214 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1215 VHOST_LOG_CONFIG(ERR,
1216 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1217 ack_msg.request.master);
1221 /* Now userfault register and we can use the memory */
1222 for (i = 0; i < memory->nregions; i++) {
1223 reg = &dev->mem->regions[i];
1224 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1232 vhost_user_mmap_region(struct virtio_net *dev,
1233 struct rte_vhost_mem_region *region,
1234 uint32_t region_index,
1235 uint64_t mmap_offset)
1243 /* Check for memory_size + mmap_offset overflow */
1244 if (mmap_offset >= -region->size) {
1245 VHOST_LOG_CONFIG(ERR,
1246 "mmap_offset (%#"PRIx64") and memory_size "
1247 "(%#"PRIx64") overflow\n",
1248 mmap_offset, region->size);
1252 mmap_size = region->size + mmap_offset;
1254 /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1255 * argument aligned with hugepagesz at older longterm version Linux,
1256 * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1258 * To avoid failure, make sure in caller to keep length aligned.
1260 alignment = get_blk_size(region->fd);
1261 if (alignment == (uint64_t)-1) {
1262 VHOST_LOG_CONFIG(ERR,
1263 "couldn't get hugepage size through fstat\n");
1266 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1267 if (mmap_size == 0) {
1269 * It could happen if initial mmap_size + alignment overflows
1270 * the sizeof uint64, which could happen if either mmap_size or
1271 * alignment value is wrong.
1273 * mmap() kernel implementation would return an error, but
1274 * better catch it before and provide useful info in the logs.
1276 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1277 "or alignment (0x%" PRIx64 ") is invalid\n",
1278 region->size + mmap_offset, alignment);
1282 populate = dev->async_copy ? MAP_POPULATE : 0;
1283 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1284 MAP_SHARED | populate, region->fd, 0);
1286 if (mmap_addr == MAP_FAILED) {
1287 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1291 region->mmap_addr = mmap_addr;
1292 region->mmap_size = mmap_size;
1293 region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1295 if (dev->async_copy) {
1296 if (add_guest_pages(dev, region, alignment) < 0) {
1297 VHOST_LOG_CONFIG(ERR,
1298 "adding guest pages to region failed.\n");
1302 if (rte_vfio_is_enabled("vfio")) {
1303 ret = async_dma_map(region, &dev->async_map_status[region_index], true);
1305 VHOST_LOG_CONFIG(ERR, "Configure IOMMU for DMA "
1312 VHOST_LOG_CONFIG(INFO,
1313 "guest memory region size: 0x%" PRIx64 "\n"
1314 "\t guest physical addr: 0x%" PRIx64 "\n"
1315 "\t guest virtual addr: 0x%" PRIx64 "\n"
1316 "\t host virtual addr: 0x%" PRIx64 "\n"
1317 "\t mmap addr : 0x%" PRIx64 "\n"
1318 "\t mmap size : 0x%" PRIx64 "\n"
1319 "\t mmap align: 0x%" PRIx64 "\n"
1320 "\t mmap off : 0x%" PRIx64 "\n",
1322 region->guest_phys_addr,
1323 region->guest_user_addr,
1324 region->host_user_addr,
1325 (uint64_t)(uintptr_t)mmap_addr,
1334 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1337 struct virtio_net *dev = *pdev;
1338 struct VhostUserMemory *memory = &msg->payload.memory;
1339 struct rte_vhost_mem_region *reg;
1340 int numa_node = SOCKET_ID_ANY;
1341 uint64_t mmap_offset;
1343 bool async_notify = false;
1345 if (validate_msg_fds(msg, memory->nregions) != 0)
1346 return RTE_VHOST_MSG_RESULT_ERR;
1348 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1349 VHOST_LOG_CONFIG(ERR,
1350 "too many memory regions (%u)\n", memory->nregions);
1354 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1355 VHOST_LOG_CONFIG(INFO,
1356 "(%d) memory regions not changed\n", dev->vid);
1360 return RTE_VHOST_MSG_RESULT_OK;
1364 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1365 struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1367 if (vdpa_dev && vdpa_dev->ops->dev_close)
1368 vdpa_dev->ops->dev_close(dev->vid);
1369 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1372 /* notify the vhost application to stop DMA transfers */
1373 if (dev->async_copy && dev->notify_ops->vring_state_changed) {
1374 for (i = 0; i < dev->nr_vring; i++) {
1375 dev->notify_ops->vring_state_changed(dev->vid,
1378 async_notify = true;
1381 free_mem_region(dev);
1385 if (dev->async_map_status) {
1386 rte_free(dev->async_map_status);
1387 dev->async_map_status = NULL;
1391 /* Flush IOTLB cache as previous HVAs are now invalid */
1392 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1393 for (i = 0; i < dev->nr_vring; i++)
1394 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1397 * If VQ 0 has already been allocated, try to allocate on the same
1398 * NUMA node. It can be reallocated later in numa_realloc().
1400 if (dev->nr_vring > 0)
1401 numa_node = dev->virtqueue[0]->numa_node;
1403 dev->nr_guest_pages = 0;
1404 if (dev->guest_pages == NULL) {
1405 dev->max_guest_pages = 8;
1406 dev->guest_pages = rte_zmalloc_socket(NULL,
1407 dev->max_guest_pages *
1408 sizeof(struct guest_page),
1409 RTE_CACHE_LINE_SIZE,
1411 if (dev->guest_pages == NULL) {
1412 VHOST_LOG_CONFIG(ERR,
1413 "(%d) failed to allocate memory "
1414 "for dev->guest_pages\n",
1420 dev->mem = rte_zmalloc_socket("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1421 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0, numa_node);
1422 if (dev->mem == NULL) {
1423 VHOST_LOG_CONFIG(ERR,
1424 "(%d) failed to allocate memory for dev->mem\n",
1426 goto free_guest_pages;
1429 if (dev->async_copy) {
1430 dev->async_map_status = rte_zmalloc_socket("async-dma-map-status",
1431 sizeof(bool) * memory->nregions, 0, numa_node);
1432 if (!dev->async_map_status) {
1433 VHOST_LOG_CONFIG(ERR,
1434 "(%d) failed to allocate memory for dma mapping status\n",
1436 goto free_mem_table;
1440 for (i = 0; i < memory->nregions; i++) {
1441 reg = &dev->mem->regions[i];
1443 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1444 reg->guest_user_addr = memory->regions[i].userspace_addr;
1445 reg->size = memory->regions[i].memory_size;
1446 reg->fd = msg->fds[i];
1449 * Assign invalid file descriptor value to avoid double
1450 * closing on error path.
1454 mmap_offset = memory->regions[i].mmap_offset;
1456 if (vhost_user_mmap_region(dev, reg, i, mmap_offset) < 0) {
1457 VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1458 goto free_mem_table;
1461 dev->mem->nregions++;
1464 if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1465 goto free_mem_table;
1467 for (i = 0; i < dev->nr_vring; i++) {
1468 struct vhost_virtqueue *vq = dev->virtqueue[i];
1473 if (vq->desc || vq->avail || vq->used) {
1475 * If the memory table got updated, the ring addresses
1476 * need to be translated again as virtual addresses have
1479 vring_invalidate(dev, vq);
1481 dev = translate_ring_addresses(dev, i);
1484 goto free_mem_table;
1491 dump_guest_pages(dev);
1494 for (i = 0; i < dev->nr_vring; i++)
1495 dev->notify_ops->vring_state_changed(dev->vid, i, 1);
1498 return RTE_VHOST_MSG_RESULT_OK;
1501 free_mem_region(dev);
1504 if (dev->async_map_status) {
1505 rte_free(dev->async_map_status);
1506 dev->async_map_status = NULL;
1509 rte_free(dev->guest_pages);
1510 dev->guest_pages = NULL;
1513 return RTE_VHOST_MSG_RESULT_ERR;
1517 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1524 if (vq_is_packed(dev))
1525 rings_ok = vq->desc_packed && vq->driver_event &&
1528 rings_ok = vq->desc && vq->avail && vq->used;
1531 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1532 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1536 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1539 virtio_is_ready(struct virtio_net *dev)
1541 struct vhost_virtqueue *vq;
1542 uint32_t i, nr_vring = dev->nr_vring;
1544 if (dev->flags & VIRTIO_DEV_READY)
1550 if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1551 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1553 if (dev->nr_vring < nr_vring)
1557 for (i = 0; i < nr_vring; i++) {
1558 vq = dev->virtqueue[i];
1560 if (!vq_is_ready(dev, vq))
1564 /* If supported, ensure the frontend is really done with config */
1565 if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1566 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1569 dev->flags |= VIRTIO_DEV_READY;
1571 if (!(dev->flags & VIRTIO_DEV_RUNNING))
1572 VHOST_LOG_CONFIG(INFO,
1573 "virtio is now ready for processing.\n");
1578 inflight_mem_alloc(const char *name, size_t size, int *fd)
1582 char fname[20] = "/tmp/memfd-XXXXXX";
1585 #ifdef MEMFD_SUPPORTED
1586 mfd = memfd_create(name, MFD_CLOEXEC);
1591 mfd = mkstemp(fname);
1593 VHOST_LOG_CONFIG(ERR,
1594 "failed to get inflight buffer fd\n");
1601 if (ftruncate(mfd, size) == -1) {
1602 VHOST_LOG_CONFIG(ERR,
1603 "failed to alloc inflight buffer\n");
1608 ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1609 if (ptr == MAP_FAILED) {
1610 VHOST_LOG_CONFIG(ERR,
1611 "failed to mmap inflight buffer\n");
1621 get_pervq_shm_size_split(uint16_t queue_size)
1623 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1624 queue_size + sizeof(uint64_t) +
1625 sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1629 get_pervq_shm_size_packed(uint16_t queue_size)
1631 return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1632 * queue_size + sizeof(uint64_t) +
1633 sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1634 INFLIGHT_ALIGNMENT);
1638 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1640 int main_fd __rte_unused)
1642 struct rte_vhost_inflight_info_packed *inflight_packed;
1643 uint64_t pervq_inflight_size, mmap_size;
1644 uint16_t num_queues, queue_size;
1645 struct virtio_net *dev = *pdev;
1647 int numa_node = SOCKET_ID_ANY;
1650 if (msg->size != sizeof(msg->payload.inflight)) {
1651 VHOST_LOG_CONFIG(ERR,
1652 "invalid get_inflight_fd message size is %d\n",
1654 return RTE_VHOST_MSG_RESULT_ERR;
1658 * If VQ 0 has already been allocated, try to allocate on the same
1659 * NUMA node. It can be reallocated later in numa_realloc().
1661 if (dev->nr_vring > 0)
1662 numa_node = dev->virtqueue[0]->numa_node;
1664 if (dev->inflight_info == NULL) {
1665 dev->inflight_info = rte_zmalloc_socket("inflight_info",
1666 sizeof(struct inflight_mem_info), 0, numa_node);
1667 if (!dev->inflight_info) {
1668 VHOST_LOG_CONFIG(ERR,
1669 "failed to alloc dev inflight area\n");
1670 return RTE_VHOST_MSG_RESULT_ERR;
1672 dev->inflight_info->fd = -1;
1675 num_queues = msg->payload.inflight.num_queues;
1676 queue_size = msg->payload.inflight.queue_size;
1678 VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1679 msg->payload.inflight.num_queues);
1680 VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1681 msg->payload.inflight.queue_size);
1683 if (vq_is_packed(dev))
1684 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1686 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1688 mmap_size = num_queues * pervq_inflight_size;
1689 addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1691 VHOST_LOG_CONFIG(ERR,
1692 "failed to alloc vhost inflight area\n");
1693 msg->payload.inflight.mmap_size = 0;
1694 return RTE_VHOST_MSG_RESULT_ERR;
1696 memset(addr, 0, mmap_size);
1698 if (dev->inflight_info->addr) {
1699 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1700 dev->inflight_info->addr = NULL;
1703 if (dev->inflight_info->fd >= 0) {
1704 close(dev->inflight_info->fd);
1705 dev->inflight_info->fd = -1;
1708 dev->inflight_info->addr = addr;
1709 dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1710 dev->inflight_info->fd = msg->fds[0] = fd;
1711 msg->payload.inflight.mmap_offset = 0;
1714 if (vq_is_packed(dev)) {
1715 for (i = 0; i < num_queues; i++) {
1717 (struct rte_vhost_inflight_info_packed *)addr;
1718 inflight_packed->used_wrap_counter = 1;
1719 inflight_packed->old_used_wrap_counter = 1;
1720 for (j = 0; j < queue_size; j++)
1721 inflight_packed->desc[j].next = j + 1;
1722 addr = (void *)((char *)addr + pervq_inflight_size);
1726 VHOST_LOG_CONFIG(INFO,
1727 "send inflight mmap_size: %"PRIu64"\n",
1728 msg->payload.inflight.mmap_size);
1729 VHOST_LOG_CONFIG(INFO,
1730 "send inflight mmap_offset: %"PRIu64"\n",
1731 msg->payload.inflight.mmap_offset);
1732 VHOST_LOG_CONFIG(INFO,
1733 "send inflight fd: %d\n", msg->fds[0]);
1735 return RTE_VHOST_MSG_RESULT_REPLY;
1739 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1740 int main_fd __rte_unused)
1742 uint64_t mmap_size, mmap_offset;
1743 uint16_t num_queues, queue_size;
1744 struct virtio_net *dev = *pdev;
1745 uint32_t pervq_inflight_size;
1746 struct vhost_virtqueue *vq;
1749 int numa_node = SOCKET_ID_ANY;
1752 if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1753 VHOST_LOG_CONFIG(ERR,
1754 "invalid set_inflight_fd message size is %d,fd is %d\n",
1756 return RTE_VHOST_MSG_RESULT_ERR;
1759 mmap_size = msg->payload.inflight.mmap_size;
1760 mmap_offset = msg->payload.inflight.mmap_offset;
1761 num_queues = msg->payload.inflight.num_queues;
1762 queue_size = msg->payload.inflight.queue_size;
1764 if (vq_is_packed(dev))
1765 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1767 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1769 VHOST_LOG_CONFIG(INFO,
1770 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1771 VHOST_LOG_CONFIG(INFO,
1772 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1773 VHOST_LOG_CONFIG(INFO,
1774 "set_inflight_fd num_queues: %u\n", num_queues);
1775 VHOST_LOG_CONFIG(INFO,
1776 "set_inflight_fd queue_size: %u\n", queue_size);
1777 VHOST_LOG_CONFIG(INFO,
1778 "set_inflight_fd fd: %d\n", fd);
1779 VHOST_LOG_CONFIG(INFO,
1780 "set_inflight_fd pervq_inflight_size: %d\n",
1781 pervq_inflight_size);
1784 * If VQ 0 has already been allocated, try to allocate on the same
1785 * NUMA node. It can be reallocated later in numa_realloc().
1787 if (dev->nr_vring > 0)
1788 numa_node = dev->virtqueue[0]->numa_node;
1790 if (!dev->inflight_info) {
1791 dev->inflight_info = rte_zmalloc_socket("inflight_info",
1792 sizeof(struct inflight_mem_info), 0, numa_node);
1793 if (dev->inflight_info == NULL) {
1794 VHOST_LOG_CONFIG(ERR,
1795 "failed to alloc dev inflight area\n");
1796 return RTE_VHOST_MSG_RESULT_ERR;
1798 dev->inflight_info->fd = -1;
1801 if (dev->inflight_info->addr) {
1802 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1803 dev->inflight_info->addr = NULL;
1806 addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1808 if (addr == MAP_FAILED) {
1809 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1810 return RTE_VHOST_MSG_RESULT_ERR;
1813 if (dev->inflight_info->fd >= 0) {
1814 close(dev->inflight_info->fd);
1815 dev->inflight_info->fd = -1;
1818 dev->inflight_info->fd = fd;
1819 dev->inflight_info->addr = addr;
1820 dev->inflight_info->size = mmap_size;
1822 for (i = 0; i < num_queues; i++) {
1823 vq = dev->virtqueue[i];
1827 if (vq_is_packed(dev)) {
1828 vq->inflight_packed = addr;
1829 vq->inflight_packed->desc_num = queue_size;
1831 vq->inflight_split = addr;
1832 vq->inflight_split->desc_num = queue_size;
1834 addr = (void *)((char *)addr + pervq_inflight_size);
1837 return RTE_VHOST_MSG_RESULT_OK;
1841 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1842 int main_fd __rte_unused)
1844 struct virtio_net *dev = *pdev;
1845 struct vhost_vring_file file;
1846 struct vhost_virtqueue *vq;
1849 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1850 if (validate_msg_fds(msg, expected_fds) != 0)
1851 return RTE_VHOST_MSG_RESULT_ERR;
1853 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1854 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1855 file.fd = VIRTIO_INVALID_EVENTFD;
1857 file.fd = msg->fds[0];
1858 VHOST_LOG_CONFIG(INFO,
1859 "vring call idx:%d file:%d\n", file.index, file.fd);
1861 vq = dev->virtqueue[file.index];
1865 vhost_user_notify_queue_state(dev, file.index, 0);
1868 if (vq->callfd >= 0)
1871 vq->callfd = file.fd;
1873 return RTE_VHOST_MSG_RESULT_OK;
1876 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1877 struct VhostUserMsg *msg,
1878 int main_fd __rte_unused)
1882 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1883 if (validate_msg_fds(msg, expected_fds) != 0)
1884 return RTE_VHOST_MSG_RESULT_ERR;
1886 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1888 VHOST_LOG_CONFIG(INFO, "not implemented\n");
1890 return RTE_VHOST_MSG_RESULT_OK;
1894 resubmit_desc_compare(const void *a, const void *b)
1896 const struct rte_vhost_resubmit_desc *desc0 = a;
1897 const struct rte_vhost_resubmit_desc *desc1 = b;
1899 if (desc1->counter > desc0->counter)
1906 vhost_check_queue_inflights_split(struct virtio_net *dev,
1907 struct vhost_virtqueue *vq)
1910 uint16_t resubmit_num = 0, last_io, num;
1911 struct vring_used *used = vq->used;
1912 struct rte_vhost_resubmit_info *resubmit;
1913 struct rte_vhost_inflight_info_split *inflight_split;
1915 if (!(dev->protocol_features &
1916 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1917 return RTE_VHOST_MSG_RESULT_OK;
1919 /* The frontend may still not support the inflight feature
1920 * although we negotiate the protocol feature.
1922 if ((!vq->inflight_split))
1923 return RTE_VHOST_MSG_RESULT_OK;
1925 if (!vq->inflight_split->version) {
1926 vq->inflight_split->version = INFLIGHT_VERSION;
1927 return RTE_VHOST_MSG_RESULT_OK;
1930 if (vq->resubmit_inflight)
1931 return RTE_VHOST_MSG_RESULT_OK;
1933 inflight_split = vq->inflight_split;
1934 vq->global_counter = 0;
1935 last_io = inflight_split->last_inflight_io;
1937 if (inflight_split->used_idx != used->idx) {
1938 inflight_split->desc[last_io].inflight = 0;
1939 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1940 inflight_split->used_idx = used->idx;
1943 for (i = 0; i < inflight_split->desc_num; i++) {
1944 if (inflight_split->desc[i].inflight == 1)
1948 vq->last_avail_idx += resubmit_num;
1951 resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1954 VHOST_LOG_CONFIG(ERR,
1955 "failed to allocate memory for resubmit info.\n");
1956 return RTE_VHOST_MSG_RESULT_ERR;
1959 resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
1960 resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
1962 if (!resubmit->resubmit_list) {
1963 VHOST_LOG_CONFIG(ERR,
1964 "failed to allocate memory for inflight desc.\n");
1966 return RTE_VHOST_MSG_RESULT_ERR;
1970 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1971 if (vq->inflight_split->desc[i].inflight == 1) {
1972 resubmit->resubmit_list[num].index = i;
1973 resubmit->resubmit_list[num].counter =
1974 inflight_split->desc[i].counter;
1978 resubmit->resubmit_num = num;
1980 if (resubmit->resubmit_num > 1)
1981 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1982 sizeof(struct rte_vhost_resubmit_desc),
1983 resubmit_desc_compare);
1985 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1986 vq->resubmit_inflight = resubmit;
1989 return RTE_VHOST_MSG_RESULT_OK;
1993 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1994 struct vhost_virtqueue *vq)
1997 uint16_t resubmit_num = 0, old_used_idx, num;
1998 struct rte_vhost_resubmit_info *resubmit;
1999 struct rte_vhost_inflight_info_packed *inflight_packed;
2001 if (!(dev->protocol_features &
2002 (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
2003 return RTE_VHOST_MSG_RESULT_OK;
2005 /* The frontend may still not support the inflight feature
2006 * although we negotiate the protocol feature.
2008 if ((!vq->inflight_packed))
2009 return RTE_VHOST_MSG_RESULT_OK;
2011 if (!vq->inflight_packed->version) {
2012 vq->inflight_packed->version = INFLIGHT_VERSION;
2013 return RTE_VHOST_MSG_RESULT_OK;
2016 if (vq->resubmit_inflight)
2017 return RTE_VHOST_MSG_RESULT_OK;
2019 inflight_packed = vq->inflight_packed;
2020 vq->global_counter = 0;
2021 old_used_idx = inflight_packed->old_used_idx;
2023 if (inflight_packed->used_idx != old_used_idx) {
2024 if (inflight_packed->desc[old_used_idx].inflight == 0) {
2025 inflight_packed->old_used_idx =
2026 inflight_packed->used_idx;
2027 inflight_packed->old_used_wrap_counter =
2028 inflight_packed->used_wrap_counter;
2029 inflight_packed->old_free_head =
2030 inflight_packed->free_head;
2032 inflight_packed->used_idx =
2033 inflight_packed->old_used_idx;
2034 inflight_packed->used_wrap_counter =
2035 inflight_packed->old_used_wrap_counter;
2036 inflight_packed->free_head =
2037 inflight_packed->old_free_head;
2041 for (i = 0; i < inflight_packed->desc_num; i++) {
2042 if (inflight_packed->desc[i].inflight == 1)
2047 resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
2049 if (resubmit == NULL) {
2050 VHOST_LOG_CONFIG(ERR,
2051 "failed to allocate memory for resubmit info.\n");
2052 return RTE_VHOST_MSG_RESULT_ERR;
2055 resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
2056 resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
2058 if (resubmit->resubmit_list == NULL) {
2059 VHOST_LOG_CONFIG(ERR,
2060 "failed to allocate memory for resubmit desc.\n");
2062 return RTE_VHOST_MSG_RESULT_ERR;
2066 for (i = 0; i < inflight_packed->desc_num; i++) {
2067 if (vq->inflight_packed->desc[i].inflight == 1) {
2068 resubmit->resubmit_list[num].index = i;
2069 resubmit->resubmit_list[num].counter =
2070 inflight_packed->desc[i].counter;
2074 resubmit->resubmit_num = num;
2076 if (resubmit->resubmit_num > 1)
2077 qsort(resubmit->resubmit_list, resubmit->resubmit_num,
2078 sizeof(struct rte_vhost_resubmit_desc),
2079 resubmit_desc_compare);
2081 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
2082 vq->resubmit_inflight = resubmit;
2085 return RTE_VHOST_MSG_RESULT_OK;
2089 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
2090 int main_fd __rte_unused)
2092 struct virtio_net *dev = *pdev;
2093 struct vhost_vring_file file;
2094 struct vhost_virtqueue *vq;
2097 expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
2098 if (validate_msg_fds(msg, expected_fds) != 0)
2099 return RTE_VHOST_MSG_RESULT_ERR;
2101 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2102 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
2103 file.fd = VIRTIO_INVALID_EVENTFD;
2105 file.fd = msg->fds[0];
2106 VHOST_LOG_CONFIG(INFO,
2107 "vring kick idx:%d file:%d\n", file.index, file.fd);
2109 /* Interpret ring addresses only when ring is started. */
2110 dev = translate_ring_addresses(dev, file.index);
2112 if (file.fd != VIRTIO_INVALID_EVENTFD)
2115 return RTE_VHOST_MSG_RESULT_ERR;
2120 vq = dev->virtqueue[file.index];
2123 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
2124 * the ring starts already enabled. Otherwise, it is enabled via
2125 * the SET_VRING_ENABLE message.
2127 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
2133 vhost_user_notify_queue_state(dev, file.index, 0);
2136 if (vq->kickfd >= 0)
2138 vq->kickfd = file.fd;
2140 if (vq_is_packed(dev)) {
2141 if (vhost_check_queue_inflights_packed(dev, vq)) {
2142 VHOST_LOG_CONFIG(ERR,
2143 "failed to inflights for vq: %d\n", file.index);
2144 return RTE_VHOST_MSG_RESULT_ERR;
2147 if (vhost_check_queue_inflights_split(dev, vq)) {
2148 VHOST_LOG_CONFIG(ERR,
2149 "failed to inflights for vq: %d\n", file.index);
2150 return RTE_VHOST_MSG_RESULT_ERR;
2154 return RTE_VHOST_MSG_RESULT_OK;
2158 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
2161 vhost_user_get_vring_base(struct virtio_net **pdev,
2162 struct VhostUserMsg *msg,
2163 int main_fd __rte_unused)
2165 struct virtio_net *dev = *pdev;
2166 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
2169 if (validate_msg_fds(msg, 0) != 0)
2170 return RTE_VHOST_MSG_RESULT_ERR;
2172 /* We have to stop the queue (virtio) if it is running. */
2173 vhost_destroy_device_notify(dev);
2175 dev->flags &= ~VIRTIO_DEV_READY;
2176 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
2178 /* Here we are safe to get the indexes */
2179 if (vq_is_packed(dev)) {
2181 * Bit[0:14]: avail index
2182 * Bit[15]: avail wrap counter
2184 val = vq->last_avail_idx & 0x7fff;
2185 val |= vq->avail_wrap_counter << 15;
2186 msg->payload.state.num = val;
2188 msg->payload.state.num = vq->last_avail_idx;
2191 VHOST_LOG_CONFIG(INFO,
2192 "vring base idx:%d file:%d\n", msg->payload.state.index,
2193 msg->payload.state.num);
2195 * Based on current qemu vhost-user implementation, this message is
2196 * sent and only sent in vhost_vring_stop.
2197 * TODO: cleanup the vring, it isn't usable since here.
2199 if (vq->kickfd >= 0)
2202 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2204 if (vq->callfd >= 0)
2207 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2209 vq->signalled_used_valid = false;
2211 if (vq_is_packed(dev)) {
2212 rte_free(vq->shadow_used_packed);
2213 vq->shadow_used_packed = NULL;
2215 rte_free(vq->shadow_used_split);
2216 vq->shadow_used_split = NULL;
2219 rte_free(vq->batch_copy_elems);
2220 vq->batch_copy_elems = NULL;
2222 rte_free(vq->log_cache);
2223 vq->log_cache = NULL;
2225 msg->size = sizeof(msg->payload.state);
2228 vhost_user_iotlb_flush_all(vq);
2230 vring_invalidate(dev, vq);
2232 return RTE_VHOST_MSG_RESULT_REPLY;
2236 * when virtio queues are ready to work, qemu will send us to
2237 * enable the virtio queue pair.
2240 vhost_user_set_vring_enable(struct virtio_net **pdev,
2241 struct VhostUserMsg *msg,
2242 int main_fd __rte_unused)
2244 struct virtio_net *dev = *pdev;
2245 bool enable = !!msg->payload.state.num;
2246 int index = (int)msg->payload.state.index;
2248 if (validate_msg_fds(msg, 0) != 0)
2249 return RTE_VHOST_MSG_RESULT_ERR;
2251 VHOST_LOG_CONFIG(INFO,
2252 "set queue enable: %d to qp idx: %d\n",
2255 if (enable && dev->virtqueue[index]->async_registered) {
2256 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2257 VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2258 "async inflight packets must be completed first\n");
2259 return RTE_VHOST_MSG_RESULT_ERR;
2263 dev->virtqueue[index]->enabled = enable;
2265 return RTE_VHOST_MSG_RESULT_OK;
2269 vhost_user_get_protocol_features(struct virtio_net **pdev,
2270 struct VhostUserMsg *msg,
2271 int main_fd __rte_unused)
2273 struct virtio_net *dev = *pdev;
2274 uint64_t features, protocol_features;
2276 if (validate_msg_fds(msg, 0) != 0)
2277 return RTE_VHOST_MSG_RESULT_ERR;
2279 rte_vhost_driver_get_features(dev->ifname, &features);
2280 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2282 msg->payload.u64 = protocol_features;
2283 msg->size = sizeof(msg->payload.u64);
2286 return RTE_VHOST_MSG_RESULT_REPLY;
2290 vhost_user_set_protocol_features(struct virtio_net **pdev,
2291 struct VhostUserMsg *msg,
2292 int main_fd __rte_unused)
2294 struct virtio_net *dev = *pdev;
2295 uint64_t protocol_features = msg->payload.u64;
2296 uint64_t slave_protocol_features = 0;
2298 if (validate_msg_fds(msg, 0) != 0)
2299 return RTE_VHOST_MSG_RESULT_ERR;
2301 rte_vhost_driver_get_protocol_features(dev->ifname,
2302 &slave_protocol_features);
2303 if (protocol_features & ~slave_protocol_features) {
2304 VHOST_LOG_CONFIG(ERR,
2305 "(%d) received invalid protocol features.\n",
2307 return RTE_VHOST_MSG_RESULT_ERR;
2310 dev->protocol_features = protocol_features;
2311 VHOST_LOG_CONFIG(INFO,
2312 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2313 dev->protocol_features);
2315 return RTE_VHOST_MSG_RESULT_OK;
2319 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2320 int main_fd __rte_unused)
2322 struct virtio_net *dev = *pdev;
2323 int fd = msg->fds[0];
2328 if (validate_msg_fds(msg, 1) != 0)
2329 return RTE_VHOST_MSG_RESULT_ERR;
2332 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2333 return RTE_VHOST_MSG_RESULT_ERR;
2336 if (msg->size != sizeof(VhostUserLog)) {
2337 VHOST_LOG_CONFIG(ERR,
2338 "invalid log base msg size: %"PRId32" != %d\n",
2339 msg->size, (int)sizeof(VhostUserLog));
2343 size = msg->payload.log.mmap_size;
2344 off = msg->payload.log.mmap_offset;
2346 /* Check for mmap size and offset overflow. */
2348 VHOST_LOG_CONFIG(ERR,
2349 "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2354 VHOST_LOG_CONFIG(INFO,
2355 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2359 * mmap from 0 to workaround a hugepage mmap bug: mmap will
2360 * fail when offset is not page size aligned.
2362 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2364 if (addr == MAP_FAILED) {
2365 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2366 return RTE_VHOST_MSG_RESULT_ERR;
2370 * Free previously mapped log memory on occasionally
2371 * multiple VHOST_USER_SET_LOG_BASE.
2373 if (dev->log_addr) {
2374 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2376 dev->log_addr = (uint64_t)(uintptr_t)addr;
2377 dev->log_base = dev->log_addr + off;
2378 dev->log_size = size;
2380 for (i = 0; i < dev->nr_vring; i++) {
2381 struct vhost_virtqueue *vq = dev->virtqueue[i];
2383 rte_free(vq->log_cache);
2384 vq->log_cache = NULL;
2385 vq->log_cache_nb_elem = 0;
2386 vq->log_cache = rte_malloc_socket("vq log cache",
2387 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2390 * If log cache alloc fail, don't fail migration, but no
2391 * caching will be done, which will impact performance
2394 VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2398 * The spec is not clear about it (yet), but QEMU doesn't expect
2399 * any payload in the reply.
2404 return RTE_VHOST_MSG_RESULT_REPLY;
2408 return RTE_VHOST_MSG_RESULT_ERR;
2411 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2412 struct VhostUserMsg *msg,
2413 int main_fd __rte_unused)
2415 if (validate_msg_fds(msg, 1) != 0)
2416 return RTE_VHOST_MSG_RESULT_ERR;
2419 VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2421 return RTE_VHOST_MSG_RESULT_OK;
2425 * An rarp packet is constructed and broadcasted to notify switches about
2426 * the new location of the migrated VM, so that packets from outside will
2427 * not be lost after migration.
2429 * However, we don't actually "send" a rarp packet here, instead, we set
2430 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2433 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2434 int main_fd __rte_unused)
2436 struct virtio_net *dev = *pdev;
2437 uint8_t *mac = (uint8_t *)&msg->payload.u64;
2438 struct rte_vdpa_device *vdpa_dev;
2440 if (validate_msg_fds(msg, 0) != 0)
2441 return RTE_VHOST_MSG_RESULT_ERR;
2443 VHOST_LOG_CONFIG(DEBUG,
2444 ":: mac: " RTE_ETHER_ADDR_PRT_FMT "\n",
2445 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2446 memcpy(dev->mac.addr_bytes, mac, 6);
2449 * Set the flag to inject a RARP broadcast packet at
2450 * rte_vhost_dequeue_burst().
2452 * __ATOMIC_RELEASE ordering is for making sure the mac is
2453 * copied before the flag is set.
2455 __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2456 vdpa_dev = dev->vdpa_dev;
2457 if (vdpa_dev && vdpa_dev->ops->migration_done)
2458 vdpa_dev->ops->migration_done(dev->vid);
2460 return RTE_VHOST_MSG_RESULT_OK;
2464 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2465 int main_fd __rte_unused)
2467 struct virtio_net *dev = *pdev;
2469 if (validate_msg_fds(msg, 0) != 0)
2470 return RTE_VHOST_MSG_RESULT_ERR;
2472 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2473 msg->payload.u64 > VIRTIO_MAX_MTU) {
2474 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2477 return RTE_VHOST_MSG_RESULT_ERR;
2480 dev->mtu = msg->payload.u64;
2482 return RTE_VHOST_MSG_RESULT_OK;
2486 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2487 int main_fd __rte_unused)
2489 struct virtio_net *dev = *pdev;
2490 int fd = msg->fds[0];
2492 if (validate_msg_fds(msg, 1) != 0)
2493 return RTE_VHOST_MSG_RESULT_ERR;
2496 VHOST_LOG_CONFIG(ERR,
2497 "Invalid file descriptor for slave channel (%d)\n",
2499 return RTE_VHOST_MSG_RESULT_ERR;
2502 if (dev->slave_req_fd >= 0)
2503 close(dev->slave_req_fd);
2505 dev->slave_req_fd = fd;
2507 return RTE_VHOST_MSG_RESULT_OK;
2511 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2513 struct vhost_vring_addr *ra;
2514 uint64_t start, end, len;
2517 end = start + imsg->size;
2519 ra = &vq->ring_addrs;
2520 len = sizeof(struct vring_desc) * vq->size;
2521 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2524 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2525 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2528 len = sizeof(struct vring_used) +
2529 sizeof(struct vring_used_elem) * vq->size;
2530 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2533 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2534 len = sizeof(uint64_t);
2535 if (ra->log_guest_addr < end &&
2536 (ra->log_guest_addr + len) > start)
2544 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2546 struct vhost_vring_addr *ra;
2547 uint64_t start, end, len;
2550 end = start + imsg->size;
2552 ra = &vq->ring_addrs;
2553 len = sizeof(struct vring_packed_desc) * vq->size;
2554 if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2557 len = sizeof(struct vring_packed_desc_event);
2558 if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2561 len = sizeof(struct vring_packed_desc_event);
2562 if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2565 if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2566 len = sizeof(uint64_t);
2567 if (ra->log_guest_addr < end &&
2568 (ra->log_guest_addr + len) > start)
2575 static int is_vring_iotlb(struct virtio_net *dev,
2576 struct vhost_virtqueue *vq,
2577 struct vhost_iotlb_msg *imsg)
2579 if (vq_is_packed(dev))
2580 return is_vring_iotlb_packed(vq, imsg);
2582 return is_vring_iotlb_split(vq, imsg);
2586 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2587 int main_fd __rte_unused)
2589 struct virtio_net *dev = *pdev;
2590 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2594 if (validate_msg_fds(msg, 0) != 0)
2595 return RTE_VHOST_MSG_RESULT_ERR;
2597 switch (imsg->type) {
2598 case VHOST_IOTLB_UPDATE:
2600 vva = qva_to_vva(dev, imsg->uaddr, &len);
2602 return RTE_VHOST_MSG_RESULT_ERR;
2604 for (i = 0; i < dev->nr_vring; i++) {
2605 struct vhost_virtqueue *vq = dev->virtqueue[i];
2610 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2613 if (is_vring_iotlb(dev, vq, imsg))
2614 *pdev = dev = translate_ring_addresses(dev, i);
2617 case VHOST_IOTLB_INVALIDATE:
2618 for (i = 0; i < dev->nr_vring; i++) {
2619 struct vhost_virtqueue *vq = dev->virtqueue[i];
2624 vhost_user_iotlb_cache_remove(vq, imsg->iova,
2627 if (is_vring_iotlb(dev, vq, imsg))
2628 vring_invalidate(dev, vq);
2632 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2634 return RTE_VHOST_MSG_RESULT_ERR;
2637 return RTE_VHOST_MSG_RESULT_OK;
2641 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2642 struct VhostUserMsg *msg,
2643 int main_fd __rte_unused)
2645 struct virtio_net *dev = *pdev;
2646 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2647 struct uffdio_api api_struct;
2649 if (validate_msg_fds(msg, 0) != 0)
2650 return RTE_VHOST_MSG_RESULT_ERR;
2652 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2654 if (dev->postcopy_ufd == -1) {
2655 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2657 return RTE_VHOST_MSG_RESULT_ERR;
2659 api_struct.api = UFFD_API;
2660 api_struct.features = 0;
2661 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2662 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2664 close(dev->postcopy_ufd);
2665 dev->postcopy_ufd = -1;
2666 return RTE_VHOST_MSG_RESULT_ERR;
2668 msg->fds[0] = dev->postcopy_ufd;
2671 return RTE_VHOST_MSG_RESULT_REPLY;
2673 dev->postcopy_ufd = -1;
2676 return RTE_VHOST_MSG_RESULT_ERR;
2681 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2682 struct VhostUserMsg *msg __rte_unused,
2683 int main_fd __rte_unused)
2685 struct virtio_net *dev = *pdev;
2687 if (validate_msg_fds(msg, 0) != 0)
2688 return RTE_VHOST_MSG_RESULT_ERR;
2690 if (dev->mem && dev->mem->nregions) {
2691 VHOST_LOG_CONFIG(ERR,
2692 "Regions already registered at postcopy-listen\n");
2693 return RTE_VHOST_MSG_RESULT_ERR;
2695 dev->postcopy_listening = 1;
2697 return RTE_VHOST_MSG_RESULT_OK;
2701 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2702 int main_fd __rte_unused)
2704 struct virtio_net *dev = *pdev;
2706 if (validate_msg_fds(msg, 0) != 0)
2707 return RTE_VHOST_MSG_RESULT_ERR;
2709 dev->postcopy_listening = 0;
2710 if (dev->postcopy_ufd >= 0) {
2711 close(dev->postcopy_ufd);
2712 dev->postcopy_ufd = -1;
2715 msg->payload.u64 = 0;
2716 msg->size = sizeof(msg->payload.u64);
2719 return RTE_VHOST_MSG_RESULT_REPLY;
2723 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2724 int main_fd __rte_unused)
2726 struct virtio_net *dev = *pdev;
2728 if (validate_msg_fds(msg, 0) != 0)
2729 return RTE_VHOST_MSG_RESULT_ERR;
2731 msg->payload.u64 = dev->status;
2732 msg->size = sizeof(msg->payload.u64);
2735 return RTE_VHOST_MSG_RESULT_REPLY;
2739 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2740 int main_fd __rte_unused)
2742 struct virtio_net *dev = *pdev;
2744 if (validate_msg_fds(msg, 0) != 0)
2745 return RTE_VHOST_MSG_RESULT_ERR;
2747 /* As per Virtio specification, the device status is 8bits long */
2748 if (msg->payload.u64 > UINT8_MAX) {
2749 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2751 return RTE_VHOST_MSG_RESULT_ERR;
2754 dev->status = msg->payload.u64;
2756 if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2757 (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2758 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2760 * Clear the bit to let the driver know about the feature
2761 * negotiation failure
2763 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2766 VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2768 "\t-ACKNOWLEDGE: %u\n"
2770 "\t-FEATURES_OK: %u\n"
2771 "\t-DRIVER_OK: %u\n"
2772 "\t-DEVICE_NEED_RESET: %u\n"
2775 (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2776 !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2777 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2778 !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2779 !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2780 !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2781 !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2783 return RTE_VHOST_MSG_RESULT_OK;
2786 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2787 struct VhostUserMsg *msg,
2789 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2790 [VHOST_USER_NONE] = NULL,
2791 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2792 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2793 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2794 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2795 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2796 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2797 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2798 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2799 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2800 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2801 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2802 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2803 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2804 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2805 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2806 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2807 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2808 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2809 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2810 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2811 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2812 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2813 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2814 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2815 [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2816 [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2817 [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2818 [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2819 [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2822 /* return bytes# of read on success or negative val on failure. */
2824 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2828 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2829 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2832 } else if (ret != VHOST_USER_HDR_SIZE) {
2833 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2839 if (msg->size > sizeof(msg->payload)) {
2840 VHOST_LOG_CONFIG(ERR,
2841 "invalid msg size: %d\n", msg->size);
2844 ret = read(sockfd, &msg->payload, msg->size);
2847 if (ret != (int)msg->size) {
2848 VHOST_LOG_CONFIG(ERR,
2849 "read control message failed\n");
2858 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2863 return send_fd_message(sockfd, (char *)msg,
2864 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2868 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2873 msg->flags &= ~VHOST_USER_VERSION_MASK;
2874 msg->flags &= ~VHOST_USER_NEED_REPLY;
2875 msg->flags |= VHOST_USER_VERSION;
2876 msg->flags |= VHOST_USER_REPLY_MASK;
2878 return send_vhost_message(sockfd, msg);
2882 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2886 if (msg->flags & VHOST_USER_NEED_REPLY)
2887 rte_spinlock_lock(&dev->slave_req_lock);
2889 ret = send_vhost_message(dev->slave_req_fd, msg);
2890 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2891 rte_spinlock_unlock(&dev->slave_req_lock);
2897 * Allocate a queue pair if it hasn't been allocated yet
2900 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2901 struct VhostUserMsg *msg)
2905 switch (msg->request.master) {
2906 case VHOST_USER_SET_VRING_KICK:
2907 case VHOST_USER_SET_VRING_CALL:
2908 case VHOST_USER_SET_VRING_ERR:
2909 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2911 case VHOST_USER_SET_VRING_NUM:
2912 case VHOST_USER_SET_VRING_BASE:
2913 case VHOST_USER_GET_VRING_BASE:
2914 case VHOST_USER_SET_VRING_ENABLE:
2915 vring_idx = msg->payload.state.index;
2917 case VHOST_USER_SET_VRING_ADDR:
2918 vring_idx = msg->payload.addr.index;
2924 if (vring_idx >= VHOST_MAX_VRING) {
2925 VHOST_LOG_CONFIG(ERR,
2926 "invalid vring index: %u\n", vring_idx);
2930 if (dev->virtqueue[vring_idx])
2933 return alloc_vring_queue(dev, vring_idx);
2937 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2940 unsigned int vq_num = 0;
2942 while (vq_num < dev->nr_vring) {
2943 struct vhost_virtqueue *vq = dev->virtqueue[i];
2946 rte_spinlock_lock(&vq->access_lock);
2954 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2957 unsigned int vq_num = 0;
2959 while (vq_num < dev->nr_vring) {
2960 struct vhost_virtqueue *vq = dev->virtqueue[i];
2963 rte_spinlock_unlock(&vq->access_lock);
2971 vhost_user_msg_handler(int vid, int fd)
2973 struct virtio_net *dev;
2974 struct VhostUserMsg msg;
2975 struct rte_vdpa_device *vdpa_dev;
2977 int unlock_required = 0;
2982 dev = get_device(vid);
2986 if (!dev->notify_ops) {
2987 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2988 if (!dev->notify_ops) {
2989 VHOST_LOG_CONFIG(ERR,
2990 "failed to get callback ops for driver %s\n",
2996 ret = read_vhost_message(fd, &msg);
2999 VHOST_LOG_CONFIG(ERR,
3000 "vhost read message failed\n");
3002 VHOST_LOG_CONFIG(INFO,
3003 "vhost peer closed\n");
3009 request = msg.request.master;
3010 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
3011 vhost_message_str[request]) {
3012 if (request != VHOST_USER_IOTLB_MSG)
3013 VHOST_LOG_CONFIG(INFO, "read message %s\n",
3014 vhost_message_str[request]);
3016 VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
3017 vhost_message_str[request]);
3019 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
3022 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
3024 VHOST_LOG_CONFIG(ERR,
3025 "failed to alloc queue\n");
3030 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
3031 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
3032 * and device is destroyed. destroy_device waits for queues to be
3033 * inactive, so it is safe. Otherwise taking the access_lock
3034 * would cause a dead lock.
3037 case VHOST_USER_SET_FEATURES:
3038 case VHOST_USER_SET_PROTOCOL_FEATURES:
3039 case VHOST_USER_SET_OWNER:
3040 case VHOST_USER_SET_MEM_TABLE:
3041 case VHOST_USER_SET_LOG_BASE:
3042 case VHOST_USER_SET_LOG_FD:
3043 case VHOST_USER_SET_VRING_NUM:
3044 case VHOST_USER_SET_VRING_ADDR:
3045 case VHOST_USER_SET_VRING_BASE:
3046 case VHOST_USER_SET_VRING_KICK:
3047 case VHOST_USER_SET_VRING_CALL:
3048 case VHOST_USER_SET_VRING_ERR:
3049 case VHOST_USER_SET_VRING_ENABLE:
3050 case VHOST_USER_SEND_RARP:
3051 case VHOST_USER_NET_SET_MTU:
3052 case VHOST_USER_SET_SLAVE_REQ_FD:
3053 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3054 vhost_user_lock_all_queue_pairs(dev);
3055 unlock_required = 1;
3064 if (dev->extern_ops.pre_msg_handle) {
3065 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
3068 case RTE_VHOST_MSG_RESULT_REPLY:
3069 send_vhost_reply(fd, &msg);
3071 case RTE_VHOST_MSG_RESULT_ERR:
3072 case RTE_VHOST_MSG_RESULT_OK:
3074 goto skip_to_post_handle;
3075 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3081 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
3082 if (!vhost_message_handlers[request])
3083 goto skip_to_post_handle;
3084 ret = vhost_message_handlers[request](&dev, &msg, fd);
3087 case RTE_VHOST_MSG_RESULT_ERR:
3088 VHOST_LOG_CONFIG(ERR,
3089 "Processing %s failed.\n",
3090 vhost_message_str[request]);
3093 case RTE_VHOST_MSG_RESULT_OK:
3094 VHOST_LOG_CONFIG(DEBUG,
3095 "Processing %s succeeded.\n",
3096 vhost_message_str[request]);
3099 case RTE_VHOST_MSG_RESULT_REPLY:
3100 VHOST_LOG_CONFIG(DEBUG,
3101 "Processing %s succeeded and needs reply.\n",
3102 vhost_message_str[request]);
3103 send_vhost_reply(fd, &msg);
3111 skip_to_post_handle:
3112 if (ret != RTE_VHOST_MSG_RESULT_ERR &&
3113 dev->extern_ops.post_msg_handle) {
3114 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
3117 case RTE_VHOST_MSG_RESULT_REPLY:
3118 send_vhost_reply(fd, &msg);
3120 case RTE_VHOST_MSG_RESULT_ERR:
3121 case RTE_VHOST_MSG_RESULT_OK:
3123 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3129 /* If message was not handled at this stage, treat it as an error */
3131 VHOST_LOG_CONFIG(ERR,
3132 "vhost message (req: %d) was not handled.\n", request);
3133 close_msg_fds(&msg);
3134 ret = RTE_VHOST_MSG_RESULT_ERR;
3138 * If the request required a reply that was already sent,
3139 * this optional reply-ack won't be sent as the
3140 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
3142 if (msg.flags & VHOST_USER_NEED_REPLY) {
3143 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
3144 msg.size = sizeof(msg.payload.u64);
3146 send_vhost_reply(fd, &msg);
3147 } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
3148 VHOST_LOG_CONFIG(ERR,
3149 "vhost message handling failed.\n");
3153 for (i = 0; i < dev->nr_vring; i++) {
3154 struct vhost_virtqueue *vq = dev->virtqueue[i];
3155 bool cur_ready = vq_is_ready(dev, vq);
3157 if (cur_ready != (vq && vq->ready)) {
3158 vq->ready = cur_ready;
3159 vhost_user_notify_queue_state(dev, i, cur_ready);
3163 if (unlock_required)
3164 vhost_user_unlock_all_queue_pairs(dev);
3166 if (!virtio_is_ready(dev))
3170 * Virtio is now ready. If not done already, it is time
3171 * to notify the application it can process the rings and
3172 * configure the vDPA device if present.
3175 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
3176 if (dev->notify_ops->new_device(dev->vid) == 0)
3177 dev->flags |= VIRTIO_DEV_RUNNING;
3180 vdpa_dev = dev->vdpa_dev;
3184 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3185 if (vdpa_dev->ops->dev_conf(dev->vid))
3186 VHOST_LOG_CONFIG(ERR,
3187 "Failed to configure vDPA device\n");
3189 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3196 static int process_slave_message_reply(struct virtio_net *dev,
3197 const struct VhostUserMsg *msg)
3199 struct VhostUserMsg msg_reply;
3202 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3205 ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3208 VHOST_LOG_CONFIG(ERR,
3209 "vhost read slave message reply failed\n");
3211 VHOST_LOG_CONFIG(INFO,
3212 "vhost peer closed\n");
3218 if (msg_reply.request.slave != msg->request.slave) {
3219 VHOST_LOG_CONFIG(ERR,
3220 "Received unexpected msg type (%u), expected %u\n",
3221 msg_reply.request.slave, msg->request.slave);
3226 ret = msg_reply.payload.u64 ? -1 : 0;
3229 rte_spinlock_unlock(&dev->slave_req_lock);
3234 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3237 struct VhostUserMsg msg = {
3238 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3239 .flags = VHOST_USER_VERSION,
3240 .size = sizeof(msg.payload.iotlb),
3244 .type = VHOST_IOTLB_MISS,
3248 ret = send_vhost_message(dev->slave_req_fd, &msg);
3250 VHOST_LOG_CONFIG(ERR,
3251 "Failed to send IOTLB miss message (%d)\n",
3260 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3263 struct VhostUserMsg msg = {
3264 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3265 .flags = VHOST_USER_VERSION,
3270 msg.flags |= VHOST_USER_NEED_REPLY;
3272 ret = send_vhost_slave_message(dev, &msg);
3274 VHOST_LOG_CONFIG(ERR,
3275 "Failed to send config change (%d)\n",
3280 return process_slave_message_reply(dev, &msg);
3284 rte_vhost_slave_config_change(int vid, bool need_reply)
3286 struct virtio_net *dev;
3288 dev = get_device(vid);
3292 return vhost_user_slave_config_change(dev, need_reply);
3295 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3301 struct VhostUserMsg msg = {
3302 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3303 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3304 .size = sizeof(msg.payload.area),
3306 .u64 = index & VHOST_USER_VRING_IDX_MASK,
3313 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3319 ret = send_vhost_slave_message(dev, &msg);
3321 VHOST_LOG_CONFIG(ERR,
3322 "Failed to set host notifier (%d)\n", ret);
3326 return process_slave_message_reply(dev, &msg);
3329 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3331 struct virtio_net *dev;
3332 struct rte_vdpa_device *vdpa_dev;
3333 int vfio_device_fd, ret = 0;
3334 uint64_t offset, size;
3335 unsigned int i, q_start, q_last;
3337 dev = get_device(vid);
3341 vdpa_dev = dev->vdpa_dev;
3342 if (vdpa_dev == NULL)
3345 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3346 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3347 !(dev->protocol_features &
3348 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3349 !(dev->protocol_features &
3350 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3351 !(dev->protocol_features &
3352 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3355 if (qid == RTE_VHOST_QUEUE_ALL) {
3357 q_last = dev->nr_vring - 1;
3359 if (qid >= dev->nr_vring)
3365 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3366 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3368 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3369 if (vfio_device_fd < 0)
3373 for (i = q_start; i <= q_last; i++) {
3374 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3380 if (vhost_user_slave_set_vring_host_notifier(dev, i,
3381 vfio_device_fd, offset, size) < 0) {
3388 for (i = q_start; i <= q_last; i++) {
3389 vhost_user_slave_set_vring_host_notifier(dev, i, -1,