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>
41 #include <rte_common.h>
42 #include <rte_malloc.h>
47 #include "vhost_user.h"
49 #define VIRTIO_MIN_MTU 68
50 #define VIRTIO_MAX_MTU 65535
52 static const char *vhost_message_str[VHOST_USER_MAX] = {
53 [VHOST_USER_NONE] = "VHOST_USER_NONE",
54 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
55 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
56 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
57 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
58 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
59 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
60 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
61 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
62 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
63 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
64 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
65 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
66 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
67 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
68 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
69 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
70 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
71 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
72 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
73 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
74 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
75 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
76 [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
77 [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
78 [VHOST_USER_POSTCOPY_ADVISE] = "VHOST_USER_POSTCOPY_ADVISE",
79 [VHOST_USER_POSTCOPY_LISTEN] = "VHOST_USER_POSTCOPY_LISTEN",
82 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
83 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
91 ret = fstat(fd, &stat);
92 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
96 free_mem_region(struct virtio_net *dev)
99 struct rte_vhost_mem_region *reg;
101 if (!dev || !dev->mem)
104 for (i = 0; i < dev->mem->nregions; i++) {
105 reg = &dev->mem->regions[i];
106 if (reg->host_user_addr) {
107 munmap(reg->mmap_addr, reg->mmap_size);
114 vhost_backend_cleanup(struct virtio_net *dev)
117 free_mem_region(dev);
122 free(dev->guest_pages);
123 dev->guest_pages = NULL;
126 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
130 if (dev->slave_req_fd >= 0) {
131 close(dev->slave_req_fd);
132 dev->slave_req_fd = -1;
135 if (dev->postcopy_ufd >= 0) {
136 close(dev->postcopy_ufd);
137 dev->postcopy_ufd = -1;
140 dev->postcopy_listening = 0;
144 * This function just returns success at the moment unless
145 * the device hasn't been initialised.
148 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
149 struct VhostUserMsg *msg __rte_unused,
150 int main_fd __rte_unused)
156 vhost_user_reset_owner(struct virtio_net **pdev,
157 struct VhostUserMsg *msg __rte_unused,
158 int main_fd __rte_unused)
160 struct virtio_net *dev = *pdev;
161 vhost_destroy_device_notify(dev);
163 cleanup_device(dev, 0);
169 * The features that we support are requested.
172 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
173 int main_fd __rte_unused)
175 struct virtio_net *dev = *pdev;
176 uint64_t features = 0;
178 rte_vhost_driver_get_features(dev->ifname, &features);
180 msg->payload.u64 = features;
181 msg->size = sizeof(msg->payload.u64);
184 return VH_RESULT_REPLY;
188 * The queue number that we support are requested.
191 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
192 int main_fd __rte_unused)
194 struct virtio_net *dev = *pdev;
195 uint32_t queue_num = 0;
197 rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
199 msg->payload.u64 = (uint64_t)queue_num;
200 msg->size = sizeof(msg->payload.u64);
203 return VH_RESULT_REPLY;
207 * We receive the negotiated features supported by us and the virtio device.
210 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
211 int main_fd __rte_unused)
213 struct virtio_net *dev = *pdev;
214 uint64_t features = msg->payload.u64;
215 uint64_t vhost_features = 0;
216 struct rte_vdpa_device *vdpa_dev;
219 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
220 if (features & ~vhost_features) {
221 RTE_LOG(ERR, VHOST_CONFIG,
222 "(%d) received invalid negotiated features.\n",
224 return VH_RESULT_ERR;
227 if (dev->flags & VIRTIO_DEV_RUNNING) {
228 if (dev->features == features)
232 * Error out if master tries to change features while device is
233 * in running state. The exception being VHOST_F_LOG_ALL, which
234 * is enabled when the live-migration starts.
236 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
237 RTE_LOG(ERR, VHOST_CONFIG,
238 "(%d) features changed while device is running.\n",
240 return VH_RESULT_ERR;
243 if (dev->notify_ops->features_changed)
244 dev->notify_ops->features_changed(dev->vid, features);
247 dev->features = features;
249 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
250 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
252 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
254 VHOST_LOG_DEBUG(VHOST_CONFIG,
255 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
257 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
258 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
260 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
261 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
263 * Remove all but first queue pair if MQ hasn't been
264 * negotiated. This is safe because the device is not
265 * running at this stage.
267 while (dev->nr_vring > 2) {
268 struct vhost_virtqueue *vq;
270 vq = dev->virtqueue[--dev->nr_vring];
274 dev->virtqueue[dev->nr_vring] = NULL;
280 did = dev->vdpa_dev_id;
281 vdpa_dev = rte_vdpa_get_device(did);
282 if (vdpa_dev && vdpa_dev->ops->set_features)
283 vdpa_dev->ops->set_features(dev->vid);
289 * The virtio device sends us the size of the descriptor ring.
292 vhost_user_set_vring_num(struct virtio_net **pdev,
293 struct VhostUserMsg *msg,
294 int main_fd __rte_unused)
296 struct virtio_net *dev = *pdev;
297 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
299 vq->size = msg->payload.state.num;
301 /* VIRTIO 1.0, 2.4 Virtqueues says:
303 * Queue Size value is always a power of 2. The maximum Queue Size
306 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
307 RTE_LOG(ERR, VHOST_CONFIG,
308 "invalid virtqueue size %u\n", vq->size);
309 return VH_RESULT_ERR;
312 if (dev->dequeue_zero_copy) {
314 vq->last_zmbuf_idx = 0;
315 vq->zmbuf_size = vq->size;
316 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
317 sizeof(struct zcopy_mbuf), 0);
318 if (vq->zmbufs == NULL) {
319 RTE_LOG(WARNING, VHOST_CONFIG,
320 "failed to allocate mem for zero copy; "
321 "zero copy is force disabled\n");
322 dev->dequeue_zero_copy = 0;
324 TAILQ_INIT(&vq->zmbuf_list);
327 if (vq_is_packed(dev)) {
328 vq->shadow_used_packed = rte_malloc(NULL,
330 sizeof(struct vring_used_elem_packed),
331 RTE_CACHE_LINE_SIZE);
332 if (!vq->shadow_used_packed) {
333 RTE_LOG(ERR, VHOST_CONFIG,
334 "failed to allocate memory for shadow used ring.\n");
335 return VH_RESULT_ERR;
339 vq->shadow_used_split = rte_malloc(NULL,
340 vq->size * sizeof(struct vring_used_elem),
341 RTE_CACHE_LINE_SIZE);
342 if (!vq->shadow_used_split) {
343 RTE_LOG(ERR, VHOST_CONFIG,
344 "failed to allocate memory for shadow used ring.\n");
345 return VH_RESULT_ERR;
349 vq->batch_copy_elems = rte_malloc(NULL,
350 vq->size * sizeof(struct batch_copy_elem),
351 RTE_CACHE_LINE_SIZE);
352 if (!vq->batch_copy_elems) {
353 RTE_LOG(ERR, VHOST_CONFIG,
354 "failed to allocate memory for batching copy.\n");
355 return VH_RESULT_ERR;
362 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
363 * same numa node as the memory of vring descriptor.
365 #ifdef RTE_LIBRTE_VHOST_NUMA
366 static struct virtio_net*
367 numa_realloc(struct virtio_net *dev, int index)
369 int oldnode, newnode;
370 struct virtio_net *old_dev;
371 struct vhost_virtqueue *old_vq, *vq;
372 struct zcopy_mbuf *new_zmbuf;
373 struct vring_used_elem *new_shadow_used_split;
374 struct vring_used_elem_packed *new_shadow_used_packed;
375 struct batch_copy_elem *new_batch_copy_elems;
379 vq = old_vq = dev->virtqueue[index];
381 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
382 MPOL_F_NODE | MPOL_F_ADDR);
384 /* check if we need to reallocate vq */
385 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
386 MPOL_F_NODE | MPOL_F_ADDR);
388 RTE_LOG(ERR, VHOST_CONFIG,
389 "Unable to get vq numa information.\n");
392 if (oldnode != newnode) {
393 RTE_LOG(INFO, VHOST_CONFIG,
394 "reallocate vq from %d to %d node\n", oldnode, newnode);
395 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
399 memcpy(vq, old_vq, sizeof(*vq));
400 TAILQ_INIT(&vq->zmbuf_list);
402 if (dev->dequeue_zero_copy) {
403 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
404 sizeof(struct zcopy_mbuf), 0, newnode);
406 rte_free(vq->zmbufs);
407 vq->zmbufs = new_zmbuf;
411 if (vq_is_packed(dev)) {
412 new_shadow_used_packed = rte_malloc_socket(NULL,
414 sizeof(struct vring_used_elem_packed),
417 if (new_shadow_used_packed) {
418 rte_free(vq->shadow_used_packed);
419 vq->shadow_used_packed = new_shadow_used_packed;
422 new_shadow_used_split = rte_malloc_socket(NULL,
424 sizeof(struct vring_used_elem),
427 if (new_shadow_used_split) {
428 rte_free(vq->shadow_used_split);
429 vq->shadow_used_split = new_shadow_used_split;
433 new_batch_copy_elems = rte_malloc_socket(NULL,
434 vq->size * sizeof(struct batch_copy_elem),
437 if (new_batch_copy_elems) {
438 rte_free(vq->batch_copy_elems);
439 vq->batch_copy_elems = new_batch_copy_elems;
445 /* check if we need to reallocate dev */
446 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
447 MPOL_F_NODE | MPOL_F_ADDR);
449 RTE_LOG(ERR, VHOST_CONFIG,
450 "Unable to get dev numa information.\n");
453 if (oldnode != newnode) {
454 RTE_LOG(INFO, VHOST_CONFIG,
455 "reallocate dev from %d to %d node\n",
457 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
463 memcpy(dev, old_dev, sizeof(*dev));
468 dev->virtqueue[index] = vq;
469 vhost_devices[dev->vid] = dev;
472 vhost_user_iotlb_init(dev, index);
477 static struct virtio_net*
478 numa_realloc(struct virtio_net *dev, int index __rte_unused)
484 /* Converts QEMU virtual address to Vhost virtual address. */
486 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
488 struct rte_vhost_mem_region *r;
491 /* Find the region where the address lives. */
492 for (i = 0; i < dev->mem->nregions; i++) {
493 r = &dev->mem->regions[i];
495 if (qva >= r->guest_user_addr &&
496 qva < r->guest_user_addr + r->size) {
498 if (unlikely(*len > r->guest_user_addr + r->size - qva))
499 *len = r->guest_user_addr + r->size - qva;
501 return qva - r->guest_user_addr +
512 * Converts ring address to Vhost virtual address.
513 * If IOMMU is enabled, the ring address is a guest IO virtual address,
514 * else it is a QEMU virtual address.
517 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
518 uint64_t ra, uint64_t *size)
520 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
523 vva = vhost_user_iotlb_cache_find(vq, ra,
524 size, VHOST_ACCESS_RW);
526 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
531 return qva_to_vva(dev, ra, size);
534 static struct virtio_net *
535 translate_ring_addresses(struct virtio_net *dev, int vq_index)
537 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
538 struct vhost_vring_addr *addr = &vq->ring_addrs;
541 if (vq_is_packed(dev)) {
542 len = sizeof(struct vring_packed_desc) * vq->size;
543 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
544 ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
545 vq->log_guest_addr = 0;
546 if (vq->desc_packed == NULL ||
547 len != sizeof(struct vring_packed_desc) *
549 RTE_LOG(DEBUG, VHOST_CONFIG,
550 "(%d) failed to map desc_packed ring.\n",
555 dev = numa_realloc(dev, vq_index);
556 vq = dev->virtqueue[vq_index];
557 addr = &vq->ring_addrs;
559 len = sizeof(struct vring_packed_desc_event);
560 vq->driver_event = (struct vring_packed_desc_event *)
561 (uintptr_t)ring_addr_to_vva(dev,
562 vq, addr->avail_user_addr, &len);
563 if (vq->driver_event == NULL ||
564 len != sizeof(struct vring_packed_desc_event)) {
565 RTE_LOG(DEBUG, VHOST_CONFIG,
566 "(%d) failed to find driver area address.\n",
571 len = sizeof(struct vring_packed_desc_event);
572 vq->device_event = (struct vring_packed_desc_event *)
573 (uintptr_t)ring_addr_to_vva(dev,
574 vq, addr->used_user_addr, &len);
575 if (vq->device_event == NULL ||
576 len != sizeof(struct vring_packed_desc_event)) {
577 RTE_LOG(DEBUG, VHOST_CONFIG,
578 "(%d) failed to find device area address.\n",
586 /* The addresses are converted from QEMU virtual to Vhost virtual. */
587 if (vq->desc && vq->avail && vq->used)
590 len = sizeof(struct vring_desc) * vq->size;
591 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
592 vq, addr->desc_user_addr, &len);
593 if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
594 RTE_LOG(DEBUG, VHOST_CONFIG,
595 "(%d) failed to map desc ring.\n",
600 dev = numa_realloc(dev, vq_index);
601 vq = dev->virtqueue[vq_index];
602 addr = &vq->ring_addrs;
604 len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
605 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
606 vq, addr->avail_user_addr, &len);
607 if (vq->avail == 0 ||
608 len != sizeof(struct vring_avail) +
609 sizeof(uint16_t) * vq->size) {
610 RTE_LOG(DEBUG, VHOST_CONFIG,
611 "(%d) failed to map avail ring.\n",
616 len = sizeof(struct vring_used) +
617 sizeof(struct vring_used_elem) * vq->size;
618 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
619 vq, addr->used_user_addr, &len);
620 if (vq->used == 0 || len != sizeof(struct vring_used) +
621 sizeof(struct vring_used_elem) * vq->size) {
622 RTE_LOG(DEBUG, VHOST_CONFIG,
623 "(%d) failed to map used ring.\n",
628 if (vq->last_used_idx != vq->used->idx) {
629 RTE_LOG(WARNING, VHOST_CONFIG,
630 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
631 "some packets maybe resent for Tx and dropped for Rx\n",
632 vq->last_used_idx, vq->used->idx);
633 vq->last_used_idx = vq->used->idx;
634 vq->last_avail_idx = vq->used->idx;
637 vq->log_guest_addr = addr->log_guest_addr;
639 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
641 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
642 dev->vid, vq->avail);
643 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
645 VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
646 dev->vid, vq->log_guest_addr);
652 * The virtio device sends us the desc, used and avail ring addresses.
653 * This function then converts these to our address space.
656 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
657 int main_fd __rte_unused)
659 struct virtio_net *dev = *pdev;
660 struct vhost_virtqueue *vq;
661 struct vhost_vring_addr *addr = &msg->payload.addr;
663 if (dev->mem == NULL)
664 return VH_RESULT_ERR;
666 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
667 vq = dev->virtqueue[msg->payload.addr.index];
670 * Rings addresses should not be interpreted as long as the ring is not
671 * started and enabled
673 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
675 vring_invalidate(dev, vq);
677 if (vq->enabled && (dev->features &
678 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
679 dev = translate_ring_addresses(dev, msg->payload.addr.index);
681 return VH_RESULT_ERR;
690 * The virtio device sends us the available ring last used index.
693 vhost_user_set_vring_base(struct virtio_net **pdev,
694 struct VhostUserMsg *msg,
695 int main_fd __rte_unused)
697 struct virtio_net *dev = *pdev;
698 dev->virtqueue[msg->payload.state.index]->last_used_idx =
699 msg->payload.state.num;
700 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
701 msg->payload.state.num;
707 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
708 uint64_t host_phys_addr, uint64_t size)
710 struct guest_page *page, *last_page;
712 if (dev->nr_guest_pages == dev->max_guest_pages) {
713 dev->max_guest_pages *= 2;
714 dev->guest_pages = realloc(dev->guest_pages,
715 dev->max_guest_pages * sizeof(*page));
716 if (!dev->guest_pages) {
717 RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
722 if (dev->nr_guest_pages > 0) {
723 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
724 /* merge if the two pages are continuous */
725 if (host_phys_addr == last_page->host_phys_addr +
727 last_page->size += size;
732 page = &dev->guest_pages[dev->nr_guest_pages++];
733 page->guest_phys_addr = guest_phys_addr;
734 page->host_phys_addr = host_phys_addr;
741 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
744 uint64_t reg_size = reg->size;
745 uint64_t host_user_addr = reg->host_user_addr;
746 uint64_t guest_phys_addr = reg->guest_phys_addr;
747 uint64_t host_phys_addr;
750 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
751 size = page_size - (guest_phys_addr & (page_size - 1));
752 size = RTE_MIN(size, reg_size);
754 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
757 host_user_addr += size;
758 guest_phys_addr += size;
761 while (reg_size > 0) {
762 size = RTE_MIN(reg_size, page_size);
763 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
765 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
769 host_user_addr += size;
770 guest_phys_addr += size;
777 #ifdef RTE_LIBRTE_VHOST_DEBUG
778 /* TODO: enable it only in debug mode? */
780 dump_guest_pages(struct virtio_net *dev)
783 struct guest_page *page;
785 for (i = 0; i < dev->nr_guest_pages; i++) {
786 page = &dev->guest_pages[i];
788 RTE_LOG(INFO, VHOST_CONFIG,
789 "guest physical page region %u\n"
790 "\t guest_phys_addr: %" PRIx64 "\n"
791 "\t host_phys_addr : %" PRIx64 "\n"
792 "\t size : %" PRIx64 "\n",
794 page->guest_phys_addr,
795 page->host_phys_addr,
800 #define dump_guest_pages(dev)
804 vhost_memory_changed(struct VhostUserMemory *new,
805 struct rte_vhost_memory *old)
809 if (new->nregions != old->nregions)
812 for (i = 0; i < new->nregions; ++i) {
813 VhostUserMemoryRegion *new_r = &new->regions[i];
814 struct rte_vhost_mem_region *old_r = &old->regions[i];
816 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
818 if (new_r->memory_size != old_r->size)
820 if (new_r->userspace_addr != old_r->guest_user_addr)
828 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
831 struct virtio_net *dev = *pdev;
832 struct VhostUserMemory *memory = &msg->payload.memory;
833 struct rte_vhost_mem_region *reg;
836 uint64_t mmap_offset;
842 if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
843 RTE_LOG(ERR, VHOST_CONFIG,
844 "too many memory regions (%u)\n", memory->nregions);
845 return VH_RESULT_ERR;
848 if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
849 RTE_LOG(INFO, VHOST_CONFIG,
850 "(%d) memory regions not changed\n", dev->vid);
852 for (i = 0; i < memory->nregions; i++)
859 free_mem_region(dev);
864 /* Flush IOTLB cache as previous HVAs are now invalid */
865 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
866 for (i = 0; i < dev->nr_vring; i++)
867 vhost_user_iotlb_flush_all(dev->virtqueue[i]);
869 dev->nr_guest_pages = 0;
870 if (!dev->guest_pages) {
871 dev->max_guest_pages = 8;
872 dev->guest_pages = malloc(dev->max_guest_pages *
873 sizeof(struct guest_page));
874 if (dev->guest_pages == NULL) {
875 RTE_LOG(ERR, VHOST_CONFIG,
876 "(%d) failed to allocate memory "
877 "for dev->guest_pages\n",
879 return VH_RESULT_ERR;
883 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
884 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
885 if (dev->mem == NULL) {
886 RTE_LOG(ERR, VHOST_CONFIG,
887 "(%d) failed to allocate memory for dev->mem\n",
889 return VH_RESULT_ERR;
891 dev->mem->nregions = memory->nregions;
893 for (i = 0; i < memory->nregions; i++) {
895 reg = &dev->mem->regions[i];
897 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
898 reg->guest_user_addr = memory->regions[i].userspace_addr;
899 reg->size = memory->regions[i].memory_size;
902 mmap_offset = memory->regions[i].mmap_offset;
904 /* Check for memory_size + mmap_offset overflow */
905 if (mmap_offset >= -reg->size) {
906 RTE_LOG(ERR, VHOST_CONFIG,
907 "mmap_offset (%#"PRIx64") and memory_size "
908 "(%#"PRIx64") overflow\n",
909 mmap_offset, reg->size);
913 mmap_size = reg->size + mmap_offset;
915 /* mmap() without flag of MAP_ANONYMOUS, should be called
916 * with length argument aligned with hugepagesz at older
917 * longterm version Linux, like 2.6.32 and 3.2.72, or
918 * mmap() will fail with EINVAL.
920 * to avoid failure, make sure in caller to keep length
923 alignment = get_blk_size(fd);
924 if (alignment == (uint64_t)-1) {
925 RTE_LOG(ERR, VHOST_CONFIG,
926 "couldn't get hugepage size through fstat\n");
929 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
931 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
932 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
933 MAP_SHARED | populate, fd, 0);
935 if (mmap_addr == MAP_FAILED) {
936 RTE_LOG(ERR, VHOST_CONFIG,
937 "mmap region %u failed.\n", i);
941 reg->mmap_addr = mmap_addr;
942 reg->mmap_size = mmap_size;
943 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
946 if (dev->dequeue_zero_copy)
947 if (add_guest_pages(dev, reg, alignment) < 0) {
948 RTE_LOG(ERR, VHOST_CONFIG,
949 "adding guest pages to region %u failed.\n",
954 RTE_LOG(INFO, VHOST_CONFIG,
955 "guest memory region %u, size: 0x%" PRIx64 "\n"
956 "\t guest physical addr: 0x%" PRIx64 "\n"
957 "\t guest virtual addr: 0x%" PRIx64 "\n"
958 "\t host virtual addr: 0x%" PRIx64 "\n"
959 "\t mmap addr : 0x%" PRIx64 "\n"
960 "\t mmap size : 0x%" PRIx64 "\n"
961 "\t mmap align: 0x%" PRIx64 "\n"
962 "\t mmap off : 0x%" PRIx64 "\n",
964 reg->guest_phys_addr,
965 reg->guest_user_addr,
967 (uint64_t)(uintptr_t)mmap_addr,
972 if (dev->postcopy_listening) {
974 * We haven't a better way right now than sharing
975 * DPDK's virtual address with Qemu, so that Qemu can
976 * retrieve the region offset when handling userfaults.
978 memory->regions[i].userspace_addr =
982 if (dev->postcopy_listening) {
983 /* Send the addresses back to qemu */
985 send_vhost_reply(main_fd, msg);
987 /* Wait for qemu to acknolwedge it's got the addresses
988 * we've got to wait before we're allowed to generate faults.
990 VhostUserMsg ack_msg;
991 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
992 RTE_LOG(ERR, VHOST_CONFIG,
993 "Failed to read qemu ack on postcopy set-mem-table\n");
996 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
997 RTE_LOG(ERR, VHOST_CONFIG,
998 "Bad qemu ack on postcopy set-mem-table (%d)\n",
999 ack_msg.request.master);
1003 /* Now userfault register and we can use the memory */
1004 for (i = 0; i < memory->nregions; i++) {
1005 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1006 reg = &dev->mem->regions[i];
1007 struct uffdio_register reg_struct;
1010 * Let's register all the mmap'ed area to ensure
1011 * alignment on page boundary.
1013 reg_struct.range.start =
1014 (uint64_t)(uintptr_t)reg->mmap_addr;
1015 reg_struct.range.len = reg->mmap_size;
1016 reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1018 if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1020 RTE_LOG(ERR, VHOST_CONFIG,
1021 "Failed to register ufd for region %d: (ufd = %d) %s\n",
1022 i, dev->postcopy_ufd,
1026 RTE_LOG(INFO, VHOST_CONFIG,
1027 "\t userfaultfd registered for range : %llx - %llx\n",
1028 reg_struct.range.start,
1029 reg_struct.range.start +
1030 reg_struct.range.len - 1);
1037 for (i = 0; i < dev->nr_vring; i++) {
1038 struct vhost_virtqueue *vq = dev->virtqueue[i];
1040 if (vq->desc || vq->avail || vq->used) {
1042 * If the memory table got updated, the ring addresses
1043 * need to be translated again as virtual addresses have
1046 vring_invalidate(dev, vq);
1048 dev = translate_ring_addresses(dev, i);
1058 dump_guest_pages(dev);
1060 return VH_RESULT_OK;
1063 free_mem_region(dev);
1066 return VH_RESULT_ERR;
1070 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1077 if (vq_is_packed(dev))
1078 rings_ok = !!vq->desc_packed;
1080 rings_ok = vq->desc && vq->avail && vq->used;
1083 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1084 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1088 virtio_is_ready(struct virtio_net *dev)
1090 struct vhost_virtqueue *vq;
1093 if (dev->nr_vring == 0)
1096 for (i = 0; i < dev->nr_vring; i++) {
1097 vq = dev->virtqueue[i];
1099 if (!vq_is_ready(dev, vq))
1103 RTE_LOG(INFO, VHOST_CONFIG,
1104 "virtio is now ready for processing.\n");
1109 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1110 int main_fd __rte_unused)
1112 struct virtio_net *dev = *pdev;
1113 struct vhost_vring_file file;
1114 struct vhost_virtqueue *vq;
1116 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1117 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1118 file.fd = VIRTIO_INVALID_EVENTFD;
1120 file.fd = msg->fds[0];
1121 RTE_LOG(INFO, VHOST_CONFIG,
1122 "vring call idx:%d file:%d\n", file.index, file.fd);
1124 vq = dev->virtqueue[file.index];
1125 if (vq->callfd >= 0)
1128 vq->callfd = file.fd;
1130 return VH_RESULT_OK;
1133 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1134 struct VhostUserMsg *msg,
1135 int main_fd __rte_unused)
1137 if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1139 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1141 return VH_RESULT_OK;
1145 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1146 int main_fd __rte_unused)
1148 struct virtio_net *dev = *pdev;
1149 struct vhost_vring_file file;
1150 struct vhost_virtqueue *vq;
1152 file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1153 if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1154 file.fd = VIRTIO_INVALID_EVENTFD;
1156 file.fd = msg->fds[0];
1157 RTE_LOG(INFO, VHOST_CONFIG,
1158 "vring kick idx:%d file:%d\n", file.index, file.fd);
1160 /* Interpret ring addresses only when ring is started. */
1161 dev = translate_ring_addresses(dev, file.index);
1163 return VH_RESULT_ERR;
1167 vq = dev->virtqueue[file.index];
1170 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1171 * the ring starts already enabled. Otherwise, it is enabled via
1172 * the SET_VRING_ENABLE message.
1174 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1177 if (vq->kickfd >= 0)
1179 vq->kickfd = file.fd;
1181 return VH_RESULT_OK;
1185 free_zmbufs(struct vhost_virtqueue *vq)
1187 struct zcopy_mbuf *zmbuf, *next;
1189 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1190 zmbuf != NULL; zmbuf = next) {
1191 next = TAILQ_NEXT(zmbuf, next);
1193 rte_pktmbuf_free(zmbuf->mbuf);
1194 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1197 rte_free(vq->zmbufs);
1201 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1204 vhost_user_get_vring_base(struct virtio_net **pdev,
1205 struct VhostUserMsg *msg,
1206 int main_fd __rte_unused)
1208 struct virtio_net *dev = *pdev;
1209 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1211 /* We have to stop the queue (virtio) if it is running. */
1212 vhost_destroy_device_notify(dev);
1214 dev->flags &= ~VIRTIO_DEV_READY;
1215 dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1217 /* Here we are safe to get the last avail index */
1218 msg->payload.state.num = vq->last_avail_idx;
1220 RTE_LOG(INFO, VHOST_CONFIG,
1221 "vring base idx:%d file:%d\n", msg->payload.state.index,
1222 msg->payload.state.num);
1224 * Based on current qemu vhost-user implementation, this message is
1225 * sent and only sent in vhost_vring_stop.
1226 * TODO: cleanup the vring, it isn't usable since here.
1228 if (vq->kickfd >= 0)
1231 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1233 if (vq->callfd >= 0)
1236 vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1238 if (dev->dequeue_zero_copy)
1240 if (vq_is_packed(dev)) {
1241 rte_free(vq->shadow_used_packed);
1242 vq->shadow_used_packed = NULL;
1244 rte_free(vq->shadow_used_split);
1245 vq->shadow_used_split = NULL;
1248 rte_free(vq->batch_copy_elems);
1249 vq->batch_copy_elems = NULL;
1251 msg->size = sizeof(msg->payload.state);
1254 return VH_RESULT_REPLY;
1258 * when virtio queues are ready to work, qemu will send us to
1259 * enable the virtio queue pair.
1262 vhost_user_set_vring_enable(struct virtio_net **pdev,
1263 struct VhostUserMsg *msg,
1264 int main_fd __rte_unused)
1266 struct virtio_net *dev = *pdev;
1267 int enable = (int)msg->payload.state.num;
1268 int index = (int)msg->payload.state.index;
1269 struct rte_vdpa_device *vdpa_dev;
1272 RTE_LOG(INFO, VHOST_CONFIG,
1273 "set queue enable: %d to qp idx: %d\n",
1276 did = dev->vdpa_dev_id;
1277 vdpa_dev = rte_vdpa_get_device(did);
1278 if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1279 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1281 if (dev->notify_ops->vring_state_changed)
1282 dev->notify_ops->vring_state_changed(dev->vid,
1285 dev->virtqueue[index]->enabled = enable;
1287 return VH_RESULT_OK;
1291 vhost_user_get_protocol_features(struct virtio_net **pdev,
1292 struct VhostUserMsg *msg,
1293 int main_fd __rte_unused)
1295 struct virtio_net *dev = *pdev;
1296 uint64_t features, protocol_features;
1298 rte_vhost_driver_get_features(dev->ifname, &features);
1299 rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1302 * REPLY_ACK protocol feature is only mandatory for now
1303 * for IOMMU feature. If IOMMU is explicitly disabled by the
1304 * application, disable also REPLY_ACK feature for older buggy
1305 * Qemu versions (from v2.7.0 to v2.9.0).
1307 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1308 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1310 msg->payload.u64 = protocol_features;
1311 msg->size = sizeof(msg->payload.u64);
1314 return VH_RESULT_REPLY;
1318 vhost_user_set_protocol_features(struct virtio_net **pdev,
1319 struct VhostUserMsg *msg,
1320 int main_fd __rte_unused)
1322 struct virtio_net *dev = *pdev;
1323 uint64_t protocol_features = msg->payload.u64;
1324 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES) {
1325 RTE_LOG(ERR, VHOST_CONFIG,
1326 "(%d) received invalid protocol features.\n",
1328 return VH_RESULT_ERR;
1331 dev->protocol_features = protocol_features;
1333 return VH_RESULT_OK;
1337 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
1338 int main_fd __rte_unused)
1340 struct virtio_net *dev = *pdev;
1341 int fd = msg->fds[0];
1346 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1347 return VH_RESULT_ERR;
1350 if (msg->size != sizeof(VhostUserLog)) {
1351 RTE_LOG(ERR, VHOST_CONFIG,
1352 "invalid log base msg size: %"PRId32" != %d\n",
1353 msg->size, (int)sizeof(VhostUserLog));
1354 return VH_RESULT_ERR;
1357 size = msg->payload.log.mmap_size;
1358 off = msg->payload.log.mmap_offset;
1360 /* Don't allow mmap_offset to point outside the mmap region */
1362 RTE_LOG(ERR, VHOST_CONFIG,
1363 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1365 return VH_RESULT_ERR;
1368 RTE_LOG(INFO, VHOST_CONFIG,
1369 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1373 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1374 * fail when offset is not page size aligned.
1376 addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1378 if (addr == MAP_FAILED) {
1379 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1380 return VH_RESULT_ERR;
1384 * Free previously mapped log memory on occasionally
1385 * multiple VHOST_USER_SET_LOG_BASE.
1387 if (dev->log_addr) {
1388 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1390 dev->log_addr = (uint64_t)(uintptr_t)addr;
1391 dev->log_base = dev->log_addr + off;
1392 dev->log_size = size;
1395 * The spec is not clear about it (yet), but QEMU doesn't expect
1396 * any payload in the reply.
1401 return VH_RESULT_REPLY;
1404 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
1405 struct VhostUserMsg *msg,
1406 int main_fd __rte_unused)
1409 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1411 return VH_RESULT_OK;
1415 * An rarp packet is constructed and broadcasted to notify switches about
1416 * the new location of the migrated VM, so that packets from outside will
1417 * not be lost after migration.
1419 * However, we don't actually "send" a rarp packet here, instead, we set
1420 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1423 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
1424 int main_fd __rte_unused)
1426 struct virtio_net *dev = *pdev;
1427 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1428 struct rte_vdpa_device *vdpa_dev;
1431 RTE_LOG(DEBUG, VHOST_CONFIG,
1432 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1433 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1434 memcpy(dev->mac.addr_bytes, mac, 6);
1437 * Set the flag to inject a RARP broadcast packet at
1438 * rte_vhost_dequeue_burst().
1440 * rte_smp_wmb() is for making sure the mac is copied
1441 * before the flag is set.
1444 rte_atomic16_set(&dev->broadcast_rarp, 1);
1445 did = dev->vdpa_dev_id;
1446 vdpa_dev = rte_vdpa_get_device(did);
1447 if (vdpa_dev && vdpa_dev->ops->migration_done)
1448 vdpa_dev->ops->migration_done(dev->vid);
1450 return VH_RESULT_OK;
1454 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
1455 int main_fd __rte_unused)
1457 struct virtio_net *dev = *pdev;
1458 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1459 msg->payload.u64 > VIRTIO_MAX_MTU) {
1460 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1463 return VH_RESULT_ERR;
1466 dev->mtu = msg->payload.u64;
1468 return VH_RESULT_OK;
1472 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
1473 int main_fd __rte_unused)
1475 struct virtio_net *dev = *pdev;
1476 int fd = msg->fds[0];
1479 RTE_LOG(ERR, VHOST_CONFIG,
1480 "Invalid file descriptor for slave channel (%d)\n",
1482 return VH_RESULT_ERR;
1485 dev->slave_req_fd = fd;
1487 return VH_RESULT_OK;
1491 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1493 struct vhost_vring_addr *ra;
1494 uint64_t start, end;
1497 end = start + imsg->size;
1499 ra = &vq->ring_addrs;
1500 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1502 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1504 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1511 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1512 struct vhost_iotlb_msg *imsg)
1514 uint64_t istart, iend, vstart, vend;
1516 istart = imsg->iova;
1517 iend = istart + imsg->size - 1;
1519 vstart = (uintptr_t)vq->desc;
1520 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1521 if (vstart <= iend && istart <= vend)
1524 vstart = (uintptr_t)vq->avail;
1525 vend = vstart + sizeof(struct vring_avail);
1526 vend += sizeof(uint16_t) * vq->size - 1;
1527 if (vstart <= iend && istart <= vend)
1530 vstart = (uintptr_t)vq->used;
1531 vend = vstart + sizeof(struct vring_used);
1532 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1533 if (vstart <= iend && istart <= vend)
1540 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
1541 int main_fd __rte_unused)
1543 struct virtio_net *dev = *pdev;
1544 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1548 switch (imsg->type) {
1549 case VHOST_IOTLB_UPDATE:
1551 vva = qva_to_vva(dev, imsg->uaddr, &len);
1553 return VH_RESULT_ERR;
1555 for (i = 0; i < dev->nr_vring; i++) {
1556 struct vhost_virtqueue *vq = dev->virtqueue[i];
1558 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1561 if (is_vring_iotlb_update(vq, imsg))
1562 *pdev = dev = translate_ring_addresses(dev, i);
1565 case VHOST_IOTLB_INVALIDATE:
1566 for (i = 0; i < dev->nr_vring; i++) {
1567 struct vhost_virtqueue *vq = dev->virtqueue[i];
1569 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1572 if (is_vring_iotlb_invalidate(vq, imsg))
1573 vring_invalidate(dev, vq);
1577 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1579 return VH_RESULT_ERR;
1582 return VH_RESULT_OK;
1586 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
1587 struct VhostUserMsg *msg,
1588 int main_fd __rte_unused)
1590 struct virtio_net *dev = *pdev;
1591 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1592 struct uffdio_api api_struct;
1594 dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
1596 if (dev->postcopy_ufd == -1) {
1597 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
1599 return VH_RESULT_ERR;
1601 api_struct.api = UFFD_API;
1602 api_struct.features = 0;
1603 if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
1604 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
1606 close(dev->postcopy_ufd);
1607 dev->postcopy_ufd = -1;
1608 return VH_RESULT_ERR;
1610 msg->fds[0] = dev->postcopy_ufd;
1613 return VH_RESULT_REPLY;
1615 dev->postcopy_ufd = -1;
1618 return VH_RESULT_ERR;
1623 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
1624 struct VhostUserMsg *msg __rte_unused,
1625 int main_fd __rte_unused)
1627 struct virtio_net *dev = *pdev;
1629 if (dev->mem && dev->mem->nregions) {
1630 RTE_LOG(ERR, VHOST_CONFIG,
1631 "Regions already registered at postcopy-listen\n");
1632 return VH_RESULT_ERR;
1634 dev->postcopy_listening = 1;
1636 return VH_RESULT_OK;
1639 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
1640 struct VhostUserMsg *msg,
1642 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
1643 [VHOST_USER_NONE] = NULL,
1644 [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
1645 [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
1646 [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
1647 [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
1648 [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
1649 [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
1650 [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
1651 [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
1652 [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
1653 [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
1654 [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
1655 [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
1656 [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
1657 [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
1658 [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
1659 [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
1660 [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
1661 [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
1662 [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
1663 [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
1664 [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
1665 [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
1666 [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
1667 [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
1671 /* return bytes# of read on success or negative val on failure. */
1673 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1677 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1678 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
1682 if (msg && msg->size) {
1683 if (msg->size > sizeof(msg->payload)) {
1684 RTE_LOG(ERR, VHOST_CONFIG,
1685 "invalid msg size: %d\n", msg->size);
1688 ret = read(sockfd, &msg->payload, msg->size);
1691 if (ret != (int)msg->size) {
1692 RTE_LOG(ERR, VHOST_CONFIG,
1693 "read control message failed\n");
1702 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1707 return send_fd_message(sockfd, (char *)msg,
1708 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
1712 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1717 msg->flags &= ~VHOST_USER_VERSION_MASK;
1718 msg->flags &= ~VHOST_USER_NEED_REPLY;
1719 msg->flags |= VHOST_USER_VERSION;
1720 msg->flags |= VHOST_USER_REPLY_MASK;
1722 return send_vhost_message(sockfd, msg);
1726 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
1730 if (msg->flags & VHOST_USER_NEED_REPLY)
1731 rte_spinlock_lock(&dev->slave_req_lock);
1733 ret = send_vhost_message(dev->slave_req_fd, msg);
1734 if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1735 rte_spinlock_unlock(&dev->slave_req_lock);
1741 * Allocate a queue pair if it hasn't been allocated yet
1744 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1745 struct VhostUserMsg *msg)
1749 switch (msg->request.master) {
1750 case VHOST_USER_SET_VRING_KICK:
1751 case VHOST_USER_SET_VRING_CALL:
1752 case VHOST_USER_SET_VRING_ERR:
1753 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1755 case VHOST_USER_SET_VRING_NUM:
1756 case VHOST_USER_SET_VRING_BASE:
1757 case VHOST_USER_SET_VRING_ENABLE:
1758 vring_idx = msg->payload.state.index;
1760 case VHOST_USER_SET_VRING_ADDR:
1761 vring_idx = msg->payload.addr.index;
1767 if (vring_idx >= VHOST_MAX_VRING) {
1768 RTE_LOG(ERR, VHOST_CONFIG,
1769 "invalid vring index: %u\n", vring_idx);
1773 if (dev->virtqueue[vring_idx])
1776 return alloc_vring_queue(dev, vring_idx);
1780 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1783 unsigned int vq_num = 0;
1785 while (vq_num < dev->nr_vring) {
1786 struct vhost_virtqueue *vq = dev->virtqueue[i];
1789 rte_spinlock_lock(&vq->access_lock);
1797 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1800 unsigned int vq_num = 0;
1802 while (vq_num < dev->nr_vring) {
1803 struct vhost_virtqueue *vq = dev->virtqueue[i];
1806 rte_spinlock_unlock(&vq->access_lock);
1814 vhost_user_msg_handler(int vid, int fd)
1816 struct virtio_net *dev;
1817 struct VhostUserMsg msg;
1818 struct rte_vdpa_device *vdpa_dev;
1821 int unlock_required = 0;
1822 uint32_t skip_master = 0;
1825 dev = get_device(vid);
1829 if (!dev->notify_ops) {
1830 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1831 if (!dev->notify_ops) {
1832 RTE_LOG(ERR, VHOST_CONFIG,
1833 "failed to get callback ops for driver %s\n",
1839 ret = read_vhost_message(fd, &msg);
1840 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1842 RTE_LOG(ERR, VHOST_CONFIG,
1843 "vhost read message failed\n");
1845 RTE_LOG(INFO, VHOST_CONFIG,
1846 "vhost peer closed\n");
1848 RTE_LOG(ERR, VHOST_CONFIG,
1849 "vhost read incorrect message\n");
1855 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1856 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1857 vhost_message_str[msg.request.master]);
1859 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1860 vhost_message_str[msg.request.master]);
1862 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1864 RTE_LOG(ERR, VHOST_CONFIG,
1865 "failed to alloc queue\n");
1870 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1871 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1872 * and device is destroyed. destroy_device waits for queues to be
1873 * inactive, so it is safe. Otherwise taking the access_lock
1874 * would cause a dead lock.
1876 switch (msg.request.master) {
1877 case VHOST_USER_SET_FEATURES:
1878 case VHOST_USER_SET_PROTOCOL_FEATURES:
1879 case VHOST_USER_SET_OWNER:
1880 case VHOST_USER_SET_MEM_TABLE:
1881 case VHOST_USER_SET_LOG_BASE:
1882 case VHOST_USER_SET_LOG_FD:
1883 case VHOST_USER_SET_VRING_NUM:
1884 case VHOST_USER_SET_VRING_ADDR:
1885 case VHOST_USER_SET_VRING_BASE:
1886 case VHOST_USER_SET_VRING_KICK:
1887 case VHOST_USER_SET_VRING_CALL:
1888 case VHOST_USER_SET_VRING_ERR:
1889 case VHOST_USER_SET_VRING_ENABLE:
1890 case VHOST_USER_SEND_RARP:
1891 case VHOST_USER_NET_SET_MTU:
1892 case VHOST_USER_SET_SLAVE_REQ_FD:
1893 vhost_user_lock_all_queue_pairs(dev);
1894 unlock_required = 1;
1901 if (dev->extern_ops.pre_msg_handle) {
1902 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1903 (void *)&msg, &skip_master);
1904 if (ret == VH_RESULT_ERR)
1906 else if (ret == VH_RESULT_REPLY)
1907 send_vhost_reply(fd, &msg);
1910 goto skip_to_post_handle;
1913 request = msg.request.master;
1914 if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
1915 if (!vhost_message_handlers[request])
1916 goto skip_to_post_handle;
1917 ret = vhost_message_handlers[request](&dev, &msg, fd);
1921 RTE_LOG(ERR, VHOST_CONFIG,
1922 "Processing %s failed.\n",
1923 vhost_message_str[request]);
1926 RTE_LOG(DEBUG, VHOST_CONFIG,
1927 "Processing %s succeeded.\n",
1928 vhost_message_str[request]);
1930 case VH_RESULT_REPLY:
1931 RTE_LOG(DEBUG, VHOST_CONFIG,
1932 "Processing %s succeeded and needs reply.\n",
1933 vhost_message_str[request]);
1934 send_vhost_reply(fd, &msg);
1938 RTE_LOG(ERR, VHOST_CONFIG,
1939 "Requested invalid message type %d.\n", request);
1940 ret = VH_RESULT_ERR;
1943 skip_to_post_handle:
1944 if (ret != VH_RESULT_ERR && dev->extern_ops.post_msg_handle) {
1945 ret = (*dev->extern_ops.post_msg_handle)(
1946 dev->vid, (void *)&msg);
1947 if (ret == VH_RESULT_ERR)
1949 else if (ret == VH_RESULT_REPLY)
1950 send_vhost_reply(fd, &msg);
1954 if (unlock_required)
1955 vhost_user_unlock_all_queue_pairs(dev);
1958 * If the request required a reply that was already sent,
1959 * this optional reply-ack won't be sent as the
1960 * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
1962 if (msg.flags & VHOST_USER_NEED_REPLY) {
1963 msg.payload.u64 = ret == VH_RESULT_ERR;
1964 msg.size = sizeof(msg.payload.u64);
1966 send_vhost_reply(fd, &msg);
1967 } else if (ret == VH_RESULT_ERR) {
1968 RTE_LOG(ERR, VHOST_CONFIG,
1969 "vhost message handling failed.\n");
1973 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1974 dev->flags |= VIRTIO_DEV_READY;
1976 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1977 if (dev->dequeue_zero_copy) {
1978 RTE_LOG(INFO, VHOST_CONFIG,
1979 "dequeue zero copy is enabled\n");
1982 if (dev->notify_ops->new_device(dev->vid) == 0)
1983 dev->flags |= VIRTIO_DEV_RUNNING;
1987 did = dev->vdpa_dev_id;
1988 vdpa_dev = rte_vdpa_get_device(did);
1989 if (vdpa_dev && virtio_is_ready(dev) &&
1990 !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1991 msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1992 if (vdpa_dev->ops->dev_conf)
1993 vdpa_dev->ops->dev_conf(dev->vid);
1994 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1995 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1996 RTE_LOG(INFO, VHOST_CONFIG,
1997 "(%d) software relay is used for vDPA, performance may be low.\n",
2005 static int process_slave_message_reply(struct virtio_net *dev,
2006 const struct VhostUserMsg *msg)
2008 struct VhostUserMsg msg_reply;
2011 if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2014 if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2019 if (msg_reply.request.slave != msg->request.slave) {
2020 RTE_LOG(ERR, VHOST_CONFIG,
2021 "Received unexpected msg type (%u), expected %u\n",
2022 msg_reply.request.slave, msg->request.slave);
2027 ret = msg_reply.payload.u64 ? -1 : 0;
2030 rte_spinlock_unlock(&dev->slave_req_lock);
2035 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2038 struct VhostUserMsg msg = {
2039 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2040 .flags = VHOST_USER_VERSION,
2041 .size = sizeof(msg.payload.iotlb),
2045 .type = VHOST_IOTLB_MISS,
2049 ret = send_vhost_message(dev->slave_req_fd, &msg);
2051 RTE_LOG(ERR, VHOST_CONFIG,
2052 "Failed to send IOTLB miss message (%d)\n",
2060 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2066 struct VhostUserMsg msg = {
2067 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2068 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2069 .size = sizeof(msg.payload.area),
2071 .u64 = index & VHOST_USER_VRING_IDX_MASK,
2078 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2084 ret = send_vhost_slave_message(dev, &msg);
2086 RTE_LOG(ERR, VHOST_CONFIG,
2087 "Failed to set host notifier (%d)\n", ret);
2091 return process_slave_message_reply(dev, &msg);
2094 int vhost_user_host_notifier_ctrl(int vid, bool enable)
2096 struct virtio_net *dev;
2097 struct rte_vdpa_device *vdpa_dev;
2098 int vfio_device_fd, did, ret = 0;
2099 uint64_t offset, size;
2102 dev = get_device(vid);
2106 did = dev->vdpa_dev_id;
2110 if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2111 !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2112 !(dev->protocol_features &
2113 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2114 !(dev->protocol_features &
2115 (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2116 !(dev->protocol_features &
2117 (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2120 vdpa_dev = rte_vdpa_get_device(did);
2124 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2125 RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2127 vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2128 if (vfio_device_fd < 0)
2132 for (i = 0; i < dev->nr_vring; i++) {
2133 if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2139 if (vhost_user_slave_set_vring_host_notifier(dev, i,
2140 vfio_device_fd, offset, size) < 0) {
2147 for (i = 0; i < dev->nr_vring; i++) {
2148 vhost_user_slave_set_vring_host_notifier(dev, i, -1,