1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2016 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/types.h>
31 #ifdef RTE_LIBRTE_VHOST_NUMA
35 #include <rte_common.h>
36 #include <rte_malloc.h>
41 #include "vhost_user.h"
43 #define VIRTIO_MIN_MTU 68
44 #define VIRTIO_MAX_MTU 65535
46 static const char *vhost_message_str[VHOST_USER_MAX] = {
47 [VHOST_USER_NONE] = "VHOST_USER_NONE",
48 [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
49 [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
50 [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
51 [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
52 [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
53 [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
54 [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
55 [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
56 [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
57 [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
58 [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
59 [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
60 [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
61 [VHOST_USER_SET_VRING_ERR] = "VHOST_USER_SET_VRING_ERR",
62 [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
63 [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
64 [VHOST_USER_GET_QUEUE_NUM] = "VHOST_USER_GET_QUEUE_NUM",
65 [VHOST_USER_SET_VRING_ENABLE] = "VHOST_USER_SET_VRING_ENABLE",
66 [VHOST_USER_SEND_RARP] = "VHOST_USER_SEND_RARP",
67 [VHOST_USER_NET_SET_MTU] = "VHOST_USER_NET_SET_MTU",
68 [VHOST_USER_SET_SLAVE_REQ_FD] = "VHOST_USER_SET_SLAVE_REQ_FD",
69 [VHOST_USER_IOTLB_MSG] = "VHOST_USER_IOTLB_MSG",
78 ret = fstat(fd, &stat);
79 return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
83 free_mem_region(struct virtio_net *dev)
86 struct rte_vhost_mem_region *reg;
88 if (!dev || !dev->mem)
91 for (i = 0; i < dev->mem->nregions; i++) {
92 reg = &dev->mem->regions[i];
93 if (reg->host_user_addr) {
94 munmap(reg->mmap_addr, reg->mmap_size);
101 vhost_backend_cleanup(struct virtio_net *dev)
104 free_mem_region(dev);
109 free(dev->guest_pages);
110 dev->guest_pages = NULL;
113 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
117 if (dev->slave_req_fd >= 0) {
118 close(dev->slave_req_fd);
119 dev->slave_req_fd = -1;
124 * This function just returns success at the moment unless
125 * the device hasn't been initialised.
128 vhost_user_set_owner(void)
134 vhost_user_reset_owner(struct virtio_net *dev)
136 if (dev->flags & VIRTIO_DEV_RUNNING) {
137 dev->flags &= ~VIRTIO_DEV_RUNNING;
138 dev->notify_ops->destroy_device(dev->vid);
141 cleanup_device(dev, 0);
147 * The features that we support are requested.
150 vhost_user_get_features(struct virtio_net *dev)
152 uint64_t features = 0;
154 rte_vhost_driver_get_features(dev->ifname, &features);
159 * We receive the negotiated features supported by us and the virtio device.
162 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
164 uint64_t vhost_features = 0;
166 rte_vhost_driver_get_features(dev->ifname, &vhost_features);
167 if (features & ~vhost_features) {
168 RTE_LOG(ERR, VHOST_CONFIG,
169 "(%d) received invalid negotiated features.\n",
174 if (dev->flags & VIRTIO_DEV_RUNNING) {
175 if (dev->features == features)
179 * Error out if master tries to change features while device is
180 * in running state. The exception being VHOST_F_LOG_ALL, which
181 * is enabled when the live-migration starts.
183 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
184 RTE_LOG(ERR, VHOST_CONFIG,
185 "(%d) features changed while device is running.\n",
190 if (dev->notify_ops->features_changed)
191 dev->notify_ops->features_changed(dev->vid, features);
194 dev->features = features;
196 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
197 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
199 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
201 LOG_DEBUG(VHOST_CONFIG,
202 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
204 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
205 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
207 if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
208 !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
210 * Remove all but first queue pair if MQ hasn't been
211 * negotiated. This is safe because the device is not
212 * running at this stage.
214 while (dev->nr_vring > 2) {
215 struct vhost_virtqueue *vq;
217 vq = dev->virtqueue[--dev->nr_vring];
221 dev->virtqueue[dev->nr_vring] = NULL;
231 * The virtio device sends us the size of the descriptor ring.
234 vhost_user_set_vring_num(struct virtio_net *dev,
237 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
239 vq->size = msg->payload.state.num;
241 if (dev->dequeue_zero_copy) {
243 vq->last_zmbuf_idx = 0;
244 vq->zmbuf_size = vq->size;
245 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
246 sizeof(struct zcopy_mbuf), 0);
247 if (vq->zmbufs == NULL) {
248 RTE_LOG(WARNING, VHOST_CONFIG,
249 "failed to allocate mem for zero copy; "
250 "zero copy is force disabled\n");
251 dev->dequeue_zero_copy = 0;
253 TAILQ_INIT(&vq->zmbuf_list);
256 vq->shadow_used_ring = rte_malloc(NULL,
257 vq->size * sizeof(struct vring_used_elem),
258 RTE_CACHE_LINE_SIZE);
259 if (!vq->shadow_used_ring) {
260 RTE_LOG(ERR, VHOST_CONFIG,
261 "failed to allocate memory for shadow used ring.\n");
265 vq->batch_copy_elems = rte_malloc(NULL,
266 vq->size * sizeof(struct batch_copy_elem),
267 RTE_CACHE_LINE_SIZE);
268 if (!vq->batch_copy_elems) {
269 RTE_LOG(ERR, VHOST_CONFIG,
270 "failed to allocate memory for batching copy.\n");
278 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
279 * same numa node as the memory of vring descriptor.
281 #ifdef RTE_LIBRTE_VHOST_NUMA
282 static struct virtio_net*
283 numa_realloc(struct virtio_net *dev, int index)
285 int oldnode, newnode;
286 struct virtio_net *old_dev;
287 struct vhost_virtqueue *old_vq, *vq;
288 struct zcopy_mbuf *new_zmbuf;
289 struct vring_used_elem *new_shadow_used_ring;
290 struct batch_copy_elem *new_batch_copy_elems;
294 vq = old_vq = dev->virtqueue[index];
296 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
297 MPOL_F_NODE | MPOL_F_ADDR);
299 /* check if we need to reallocate vq */
300 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
301 MPOL_F_NODE | MPOL_F_ADDR);
303 RTE_LOG(ERR, VHOST_CONFIG,
304 "Unable to get vq numa information.\n");
307 if (oldnode != newnode) {
308 RTE_LOG(INFO, VHOST_CONFIG,
309 "reallocate vq from %d to %d node\n", oldnode, newnode);
310 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
314 memcpy(vq, old_vq, sizeof(*vq));
315 TAILQ_INIT(&vq->zmbuf_list);
317 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
318 sizeof(struct zcopy_mbuf), 0, newnode);
320 rte_free(vq->zmbufs);
321 vq->zmbufs = new_zmbuf;
324 new_shadow_used_ring = rte_malloc_socket(NULL,
325 vq->size * sizeof(struct vring_used_elem),
328 if (new_shadow_used_ring) {
329 rte_free(vq->shadow_used_ring);
330 vq->shadow_used_ring = new_shadow_used_ring;
333 new_batch_copy_elems = rte_malloc_socket(NULL,
334 vq->size * sizeof(struct batch_copy_elem),
337 if (new_batch_copy_elems) {
338 rte_free(vq->batch_copy_elems);
339 vq->batch_copy_elems = new_batch_copy_elems;
345 /* check if we need to reallocate dev */
346 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
347 MPOL_F_NODE | MPOL_F_ADDR);
349 RTE_LOG(ERR, VHOST_CONFIG,
350 "Unable to get dev numa information.\n");
353 if (oldnode != newnode) {
354 RTE_LOG(INFO, VHOST_CONFIG,
355 "reallocate dev from %d to %d node\n",
357 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
363 memcpy(dev, old_dev, sizeof(*dev));
368 dev->virtqueue[index] = vq;
369 vhost_devices[dev->vid] = dev;
372 vhost_user_iotlb_init(dev, index);
377 static struct virtio_net*
378 numa_realloc(struct virtio_net *dev, int index __rte_unused)
384 /* Converts QEMU virtual address to Vhost virtual address. */
386 qva_to_vva(struct virtio_net *dev, uint64_t qva)
388 struct rte_vhost_mem_region *reg;
391 /* Find the region where the address lives. */
392 for (i = 0; i < dev->mem->nregions; i++) {
393 reg = &dev->mem->regions[i];
395 if (qva >= reg->guest_user_addr &&
396 qva < reg->guest_user_addr + reg->size) {
397 return qva - reg->guest_user_addr +
407 * Converts ring address to Vhost virtual address.
408 * If IOMMU is enabled, the ring address is a guest IO virtual address,
409 * else it is a QEMU virtual address.
412 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
413 uint64_t ra, uint64_t size)
415 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
418 vva = vhost_user_iotlb_cache_find(vq, ra,
419 &size, VHOST_ACCESS_RW);
421 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
426 return qva_to_vva(dev, ra);
429 static struct virtio_net *
430 translate_ring_addresses(struct virtio_net *dev, int vq_index)
432 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
433 struct vhost_vring_addr *addr = &vq->ring_addrs;
435 /* The addresses are converted from QEMU virtual to Vhost virtual. */
436 if (vq->desc && vq->avail && vq->used)
439 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
440 vq, addr->desc_user_addr, sizeof(struct vring_desc));
442 RTE_LOG(DEBUG, VHOST_CONFIG,
443 "(%d) failed to find desc ring address.\n",
448 dev = numa_realloc(dev, vq_index);
449 vq = dev->virtqueue[vq_index];
450 addr = &vq->ring_addrs;
452 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
453 vq, addr->avail_user_addr, sizeof(struct vring_avail));
454 if (vq->avail == 0) {
455 RTE_LOG(DEBUG, VHOST_CONFIG,
456 "(%d) failed to find avail ring address.\n",
461 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
462 vq, addr->used_user_addr, sizeof(struct vring_used));
464 RTE_LOG(DEBUG, VHOST_CONFIG,
465 "(%d) failed to find used ring address.\n",
470 if (vq->last_used_idx != vq->used->idx) {
471 RTE_LOG(WARNING, VHOST_CONFIG,
472 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
473 "some packets maybe resent for Tx and dropped for Rx\n",
474 vq->last_used_idx, vq->used->idx);
475 vq->last_used_idx = vq->used->idx;
476 vq->last_avail_idx = vq->used->idx;
479 vq->log_guest_addr = addr->log_guest_addr;
481 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
483 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
484 dev->vid, vq->avail);
485 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
487 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
488 dev->vid, vq->log_guest_addr);
494 * The virtio device sends us the desc, used and avail ring addresses.
495 * This function then converts these to our address space.
498 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
500 struct vhost_virtqueue *vq;
501 struct vhost_vring_addr *addr = &msg->payload.addr;
502 struct virtio_net *dev = *pdev;
504 if (dev->mem == NULL)
507 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
508 vq = dev->virtqueue[msg->payload.addr.index];
511 * Rings addresses should not be interpreted as long as the ring is not
512 * started and enabled
514 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
516 vring_invalidate(dev, vq);
518 if (vq->enabled && (dev->features &
519 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
520 dev = translate_ring_addresses(dev, msg->payload.addr.index);
531 * The virtio device sends us the available ring last used index.
534 vhost_user_set_vring_base(struct virtio_net *dev,
537 dev->virtqueue[msg->payload.state.index]->last_used_idx =
538 msg->payload.state.num;
539 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
540 msg->payload.state.num;
546 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
547 uint64_t host_phys_addr, uint64_t size)
549 struct guest_page *page, *last_page;
551 if (dev->nr_guest_pages == dev->max_guest_pages) {
552 dev->max_guest_pages *= 2;
553 dev->guest_pages = realloc(dev->guest_pages,
554 dev->max_guest_pages * sizeof(*page));
557 if (dev->nr_guest_pages > 0) {
558 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
559 /* merge if the two pages are continuous */
560 if (host_phys_addr == last_page->host_phys_addr +
562 last_page->size += size;
567 page = &dev->guest_pages[dev->nr_guest_pages++];
568 page->guest_phys_addr = guest_phys_addr;
569 page->host_phys_addr = host_phys_addr;
574 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
577 uint64_t reg_size = reg->size;
578 uint64_t host_user_addr = reg->host_user_addr;
579 uint64_t guest_phys_addr = reg->guest_phys_addr;
580 uint64_t host_phys_addr;
583 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
584 size = page_size - (guest_phys_addr & (page_size - 1));
585 size = RTE_MIN(size, reg_size);
587 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
588 host_user_addr += size;
589 guest_phys_addr += size;
592 while (reg_size > 0) {
593 size = RTE_MIN(reg_size, page_size);
594 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
596 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
598 host_user_addr += size;
599 guest_phys_addr += size;
604 #ifdef RTE_LIBRTE_VHOST_DEBUG
605 /* TODO: enable it only in debug mode? */
607 dump_guest_pages(struct virtio_net *dev)
610 struct guest_page *page;
612 for (i = 0; i < dev->nr_guest_pages; i++) {
613 page = &dev->guest_pages[i];
615 RTE_LOG(INFO, VHOST_CONFIG,
616 "guest physical page region %u\n"
617 "\t guest_phys_addr: %" PRIx64 "\n"
618 "\t host_phys_addr : %" PRIx64 "\n"
619 "\t size : %" PRIx64 "\n",
621 page->guest_phys_addr,
622 page->host_phys_addr,
627 #define dump_guest_pages(dev)
631 vhost_memory_changed(struct VhostUserMemory *new,
632 struct rte_vhost_memory *old)
636 if (new->nregions != old->nregions)
639 for (i = 0; i < new->nregions; ++i) {
640 VhostUserMemoryRegion *new_r = &new->regions[i];
641 struct rte_vhost_mem_region *old_r = &old->regions[i];
643 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
645 if (new_r->memory_size != old_r->size)
647 if (new_r->userspace_addr != old_r->guest_user_addr)
655 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
657 struct VhostUserMemory memory = pmsg->payload.memory;
658 struct rte_vhost_mem_region *reg;
661 uint64_t mmap_offset;
666 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
667 RTE_LOG(ERR, VHOST_CONFIG,
668 "too many memory regions (%u)\n", memory.nregions);
672 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
673 RTE_LOG(INFO, VHOST_CONFIG,
674 "(%d) memory regions not changed\n", dev->vid);
676 for (i = 0; i < memory.nregions; i++)
683 free_mem_region(dev);
688 dev->nr_guest_pages = 0;
689 if (!dev->guest_pages) {
690 dev->max_guest_pages = 8;
691 dev->guest_pages = malloc(dev->max_guest_pages *
692 sizeof(struct guest_page));
693 if (dev->guest_pages == NULL) {
694 RTE_LOG(ERR, VHOST_CONFIG,
695 "(%d) failed to allocate memory "
696 "for dev->guest_pages\n",
702 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
703 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
704 if (dev->mem == NULL) {
705 RTE_LOG(ERR, VHOST_CONFIG,
706 "(%d) failed to allocate memory for dev->mem\n",
710 dev->mem->nregions = memory.nregions;
712 for (i = 0; i < memory.nregions; i++) {
714 reg = &dev->mem->regions[i];
716 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
717 reg->guest_user_addr = memory.regions[i].userspace_addr;
718 reg->size = memory.regions[i].memory_size;
721 mmap_offset = memory.regions[i].mmap_offset;
722 mmap_size = reg->size + mmap_offset;
724 /* mmap() without flag of MAP_ANONYMOUS, should be called
725 * with length argument aligned with hugepagesz at older
726 * longterm version Linux, like 2.6.32 and 3.2.72, or
727 * mmap() will fail with EINVAL.
729 * to avoid failure, make sure in caller to keep length
732 alignment = get_blk_size(fd);
733 if (alignment == (uint64_t)-1) {
734 RTE_LOG(ERR, VHOST_CONFIG,
735 "couldn't get hugepage size through fstat\n");
738 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
740 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
741 MAP_SHARED | MAP_POPULATE, fd, 0);
743 if (mmap_addr == MAP_FAILED) {
744 RTE_LOG(ERR, VHOST_CONFIG,
745 "mmap region %u failed.\n", i);
749 reg->mmap_addr = mmap_addr;
750 reg->mmap_size = mmap_size;
751 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
754 if (dev->dequeue_zero_copy)
755 add_guest_pages(dev, reg, alignment);
757 RTE_LOG(INFO, VHOST_CONFIG,
758 "guest memory region %u, size: 0x%" PRIx64 "\n"
759 "\t guest physical addr: 0x%" PRIx64 "\n"
760 "\t guest virtual addr: 0x%" PRIx64 "\n"
761 "\t host virtual addr: 0x%" PRIx64 "\n"
762 "\t mmap addr : 0x%" PRIx64 "\n"
763 "\t mmap size : 0x%" PRIx64 "\n"
764 "\t mmap align: 0x%" PRIx64 "\n"
765 "\t mmap off : 0x%" PRIx64 "\n",
767 reg->guest_phys_addr,
768 reg->guest_user_addr,
770 (uint64_t)(uintptr_t)mmap_addr,
776 dump_guest_pages(dev);
781 free_mem_region(dev);
788 vq_is_ready(struct vhost_virtqueue *vq)
790 return vq && vq->desc && vq->avail && vq->used &&
791 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
792 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
796 virtio_is_ready(struct virtio_net *dev)
798 struct vhost_virtqueue *vq;
801 if (dev->nr_vring == 0)
804 for (i = 0; i < dev->nr_vring; i++) {
805 vq = dev->virtqueue[i];
807 if (!vq_is_ready(vq))
811 RTE_LOG(INFO, VHOST_CONFIG,
812 "virtio is now ready for processing.\n");
817 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
819 struct vhost_vring_file file;
820 struct vhost_virtqueue *vq;
822 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
823 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
824 file.fd = VIRTIO_INVALID_EVENTFD;
826 file.fd = pmsg->fds[0];
827 RTE_LOG(INFO, VHOST_CONFIG,
828 "vring call idx:%d file:%d\n", file.index, file.fd);
830 vq = dev->virtqueue[file.index];
834 vq->callfd = file.fd;
838 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
840 struct vhost_vring_file file;
841 struct vhost_virtqueue *vq;
842 struct virtio_net *dev = *pdev;
844 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
845 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
846 file.fd = VIRTIO_INVALID_EVENTFD;
848 file.fd = pmsg->fds[0];
849 RTE_LOG(INFO, VHOST_CONFIG,
850 "vring kick idx:%d file:%d\n", file.index, file.fd);
852 /* Interpret ring addresses only when ring is started. */
853 dev = translate_ring_addresses(dev, file.index);
859 vq = dev->virtqueue[file.index];
862 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
863 * the ring starts already enabled. Otherwise, it is enabled via
864 * the SET_VRING_ENABLE message.
866 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
871 vq->kickfd = file.fd;
875 free_zmbufs(struct vhost_virtqueue *vq)
877 struct zcopy_mbuf *zmbuf, *next;
879 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
880 zmbuf != NULL; zmbuf = next) {
881 next = TAILQ_NEXT(zmbuf, next);
883 rte_pktmbuf_free(zmbuf->mbuf);
884 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
887 rte_free(vq->zmbufs);
891 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
894 vhost_user_get_vring_base(struct virtio_net *dev,
897 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
899 /* We have to stop the queue (virtio) if it is running. */
900 if (dev->flags & VIRTIO_DEV_RUNNING) {
901 dev->flags &= ~VIRTIO_DEV_RUNNING;
902 dev->notify_ops->destroy_device(dev->vid);
905 dev->flags &= ~VIRTIO_DEV_READY;
907 /* Here we are safe to get the last used index */
908 msg->payload.state.num = vq->last_used_idx;
910 RTE_LOG(INFO, VHOST_CONFIG,
911 "vring base idx:%d file:%d\n", msg->payload.state.index,
912 msg->payload.state.num);
914 * Based on current qemu vhost-user implementation, this message is
915 * sent and only sent in vhost_vring_stop.
916 * TODO: cleanup the vring, it isn't usable since here.
921 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
923 if (dev->dequeue_zero_copy)
925 rte_free(vq->shadow_used_ring);
926 vq->shadow_used_ring = NULL;
928 rte_free(vq->batch_copy_elems);
929 vq->batch_copy_elems = NULL;
935 * when virtio queues are ready to work, qemu will send us to
936 * enable the virtio queue pair.
939 vhost_user_set_vring_enable(struct virtio_net *dev,
942 int enable = (int)msg->payload.state.num;
944 RTE_LOG(INFO, VHOST_CONFIG,
945 "set queue enable: %d to qp idx: %d\n",
946 enable, msg->payload.state.index);
948 if (dev->notify_ops->vring_state_changed)
949 dev->notify_ops->vring_state_changed(dev->vid,
950 msg->payload.state.index, enable);
952 dev->virtqueue[msg->payload.state.index]->enabled = enable;
958 vhost_user_get_protocol_features(struct virtio_net *dev,
959 struct VhostUserMsg *msg)
961 uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
963 rte_vhost_driver_get_features(dev->ifname, &features);
966 * REPLY_ACK protocol feature is only mandatory for now
967 * for IOMMU feature. If IOMMU is explicitly disabled by the
968 * application, disable also REPLY_ACK feature for older buggy
969 * Qemu versions (from v2.7.0 to v2.9.0).
971 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
972 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
974 msg->payload.u64 = protocol_features;
975 msg->size = sizeof(msg->payload.u64);
979 vhost_user_set_protocol_features(struct virtio_net *dev,
980 uint64_t protocol_features)
982 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
985 dev->protocol_features = protocol_features;
989 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
991 int fd = msg->fds[0];
996 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1000 if (msg->size != sizeof(VhostUserLog)) {
1001 RTE_LOG(ERR, VHOST_CONFIG,
1002 "invalid log base msg size: %"PRId32" != %d\n",
1003 msg->size, (int)sizeof(VhostUserLog));
1007 size = msg->payload.log.mmap_size;
1008 off = msg->payload.log.mmap_offset;
1010 /* Don't allow mmap_offset to point outside the mmap region */
1012 RTE_LOG(ERR, VHOST_CONFIG,
1013 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1018 RTE_LOG(INFO, VHOST_CONFIG,
1019 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1023 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1024 * fail when offset is not page size aligned.
1026 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1028 if (addr == MAP_FAILED) {
1029 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1034 * Free previously mapped log memory on occasionally
1035 * multiple VHOST_USER_SET_LOG_BASE.
1037 if (dev->log_addr) {
1038 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1040 dev->log_addr = (uint64_t)(uintptr_t)addr;
1041 dev->log_base = dev->log_addr + off;
1042 dev->log_size = size;
1048 * An rarp packet is constructed and broadcasted to notify switches about
1049 * the new location of the migrated VM, so that packets from outside will
1050 * not be lost after migration.
1052 * However, we don't actually "send" a rarp packet here, instead, we set
1053 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1056 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1058 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1060 RTE_LOG(DEBUG, VHOST_CONFIG,
1061 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1062 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1063 memcpy(dev->mac.addr_bytes, mac, 6);
1066 * Set the flag to inject a RARP broadcast packet at
1067 * rte_vhost_dequeue_burst().
1069 * rte_smp_wmb() is for making sure the mac is copied
1070 * before the flag is set.
1073 rte_atomic16_set(&dev->broadcast_rarp, 1);
1079 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1081 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1082 msg->payload.u64 > VIRTIO_MAX_MTU) {
1083 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1089 dev->mtu = msg->payload.u64;
1095 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1097 int fd = msg->fds[0];
1100 RTE_LOG(ERR, VHOST_CONFIG,
1101 "Invalid file descriptor for slave channel (%d)\n",
1106 dev->slave_req_fd = fd;
1112 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1114 struct vhost_vring_addr *ra;
1115 uint64_t start, end;
1118 end = start + imsg->size;
1120 ra = &vq->ring_addrs;
1121 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1123 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1125 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1132 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1133 struct vhost_iotlb_msg *imsg)
1135 uint64_t istart, iend, vstart, vend;
1137 istart = imsg->iova;
1138 iend = istart + imsg->size - 1;
1140 vstart = (uintptr_t)vq->desc;
1141 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1142 if (vstart <= iend && istart <= vend)
1145 vstart = (uintptr_t)vq->avail;
1146 vend = vstart + sizeof(struct vring_avail);
1147 vend += sizeof(uint16_t) * vq->size - 1;
1148 if (vstart <= iend && istart <= vend)
1151 vstart = (uintptr_t)vq->used;
1152 vend = vstart + sizeof(struct vring_used);
1153 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1154 if (vstart <= iend && istart <= vend)
1161 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1163 struct virtio_net *dev = *pdev;
1164 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1168 switch (imsg->type) {
1169 case VHOST_IOTLB_UPDATE:
1170 vva = qva_to_vva(dev, imsg->uaddr);
1174 for (i = 0; i < dev->nr_vring; i++) {
1175 struct vhost_virtqueue *vq = dev->virtqueue[i];
1177 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1178 imsg->size, imsg->perm);
1180 if (is_vring_iotlb_update(vq, imsg))
1181 *pdev = dev = translate_ring_addresses(dev, i);
1184 case VHOST_IOTLB_INVALIDATE:
1185 for (i = 0; i < dev->nr_vring; i++) {
1186 struct vhost_virtqueue *vq = dev->virtqueue[i];
1188 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1191 if (is_vring_iotlb_invalidate(vq, imsg))
1192 vring_invalidate(dev, vq);
1196 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1204 /* return bytes# of read on success or negative val on failure. */
1206 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1210 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1211 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1215 if (msg && msg->size) {
1216 if (msg->size > sizeof(msg->payload)) {
1217 RTE_LOG(ERR, VHOST_CONFIG,
1218 "invalid msg size: %d\n", msg->size);
1221 ret = read(sockfd, &msg->payload, msg->size);
1224 if (ret != (int)msg->size) {
1225 RTE_LOG(ERR, VHOST_CONFIG,
1226 "read control message failed\n");
1235 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1240 return send_fd_message(sockfd, (char *)msg,
1241 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1245 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1250 msg->flags &= ~VHOST_USER_VERSION_MASK;
1251 msg->flags &= ~VHOST_USER_NEED_REPLY;
1252 msg->flags |= VHOST_USER_VERSION;
1253 msg->flags |= VHOST_USER_REPLY_MASK;
1255 return send_vhost_message(sockfd, msg);
1259 * Allocate a queue pair if it hasn't been allocated yet
1262 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1266 switch (msg->request.master) {
1267 case VHOST_USER_SET_VRING_KICK:
1268 case VHOST_USER_SET_VRING_CALL:
1269 case VHOST_USER_SET_VRING_ERR:
1270 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1272 case VHOST_USER_SET_VRING_NUM:
1273 case VHOST_USER_SET_VRING_BASE:
1274 case VHOST_USER_SET_VRING_ENABLE:
1275 vring_idx = msg->payload.state.index;
1277 case VHOST_USER_SET_VRING_ADDR:
1278 vring_idx = msg->payload.addr.index;
1284 if (vring_idx >= VHOST_MAX_VRING) {
1285 RTE_LOG(ERR, VHOST_CONFIG,
1286 "invalid vring index: %u\n", vring_idx);
1290 if (dev->virtqueue[vring_idx])
1293 return alloc_vring_queue(dev, vring_idx);
1297 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1300 unsigned int vq_num = 0;
1302 while (vq_num < dev->nr_vring) {
1303 struct vhost_virtqueue *vq = dev->virtqueue[i];
1306 rte_spinlock_lock(&vq->access_lock);
1314 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1317 unsigned int vq_num = 0;
1319 while (vq_num < dev->nr_vring) {
1320 struct vhost_virtqueue *vq = dev->virtqueue[i];
1323 rte_spinlock_unlock(&vq->access_lock);
1331 vhost_user_msg_handler(int vid, int fd)
1333 struct virtio_net *dev;
1334 struct VhostUserMsg msg;
1336 int unlock_required = 0;
1338 dev = get_device(vid);
1342 if (!dev->notify_ops) {
1343 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1344 if (!dev->notify_ops) {
1345 RTE_LOG(ERR, VHOST_CONFIG,
1346 "failed to get callback ops for driver %s\n",
1352 ret = read_vhost_message(fd, &msg);
1353 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1355 RTE_LOG(ERR, VHOST_CONFIG,
1356 "vhost read message failed\n");
1358 RTE_LOG(INFO, VHOST_CONFIG,
1359 "vhost peer closed\n");
1361 RTE_LOG(ERR, VHOST_CONFIG,
1362 "vhost read incorrect message\n");
1368 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1369 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1370 vhost_message_str[msg.request.master]);
1372 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1373 vhost_message_str[msg.request.master]);
1375 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1377 RTE_LOG(ERR, VHOST_CONFIG,
1378 "failed to alloc queue\n");
1383 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1384 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1385 * and device is destroyed. destroy_device waits for queues to be
1386 * inactive, so it is safe. Otherwise taking the access_lock
1387 * would cause a dead lock.
1389 switch (msg.request.master) {
1390 case VHOST_USER_SET_FEATURES:
1391 case VHOST_USER_SET_PROTOCOL_FEATURES:
1392 case VHOST_USER_SET_OWNER:
1393 case VHOST_USER_SET_MEM_TABLE:
1394 case VHOST_USER_SET_LOG_BASE:
1395 case VHOST_USER_SET_LOG_FD:
1396 case VHOST_USER_SET_VRING_NUM:
1397 case VHOST_USER_SET_VRING_ADDR:
1398 case VHOST_USER_SET_VRING_BASE:
1399 case VHOST_USER_SET_VRING_KICK:
1400 case VHOST_USER_SET_VRING_CALL:
1401 case VHOST_USER_SET_VRING_ERR:
1402 case VHOST_USER_SET_VRING_ENABLE:
1403 case VHOST_USER_SEND_RARP:
1404 case VHOST_USER_NET_SET_MTU:
1405 case VHOST_USER_SET_SLAVE_REQ_FD:
1406 vhost_user_lock_all_queue_pairs(dev);
1407 unlock_required = 1;
1414 switch (msg.request.master) {
1415 case VHOST_USER_GET_FEATURES:
1416 msg.payload.u64 = vhost_user_get_features(dev);
1417 msg.size = sizeof(msg.payload.u64);
1418 send_vhost_reply(fd, &msg);
1420 case VHOST_USER_SET_FEATURES:
1421 ret = vhost_user_set_features(dev, msg.payload.u64);
1426 case VHOST_USER_GET_PROTOCOL_FEATURES:
1427 vhost_user_get_protocol_features(dev, &msg);
1428 send_vhost_reply(fd, &msg);
1430 case VHOST_USER_SET_PROTOCOL_FEATURES:
1431 vhost_user_set_protocol_features(dev, msg.payload.u64);
1434 case VHOST_USER_SET_OWNER:
1435 vhost_user_set_owner();
1437 case VHOST_USER_RESET_OWNER:
1438 vhost_user_reset_owner(dev);
1441 case VHOST_USER_SET_MEM_TABLE:
1442 ret = vhost_user_set_mem_table(dev, &msg);
1445 case VHOST_USER_SET_LOG_BASE:
1446 vhost_user_set_log_base(dev, &msg);
1448 /* it needs a reply */
1449 msg.size = sizeof(msg.payload.u64);
1450 send_vhost_reply(fd, &msg);
1452 case VHOST_USER_SET_LOG_FD:
1454 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1457 case VHOST_USER_SET_VRING_NUM:
1458 vhost_user_set_vring_num(dev, &msg);
1460 case VHOST_USER_SET_VRING_ADDR:
1461 vhost_user_set_vring_addr(&dev, &msg);
1463 case VHOST_USER_SET_VRING_BASE:
1464 vhost_user_set_vring_base(dev, &msg);
1467 case VHOST_USER_GET_VRING_BASE:
1468 vhost_user_get_vring_base(dev, &msg);
1469 msg.size = sizeof(msg.payload.state);
1470 send_vhost_reply(fd, &msg);
1473 case VHOST_USER_SET_VRING_KICK:
1474 vhost_user_set_vring_kick(&dev, &msg);
1476 case VHOST_USER_SET_VRING_CALL:
1477 vhost_user_set_vring_call(dev, &msg);
1480 case VHOST_USER_SET_VRING_ERR:
1481 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1483 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1486 case VHOST_USER_GET_QUEUE_NUM:
1487 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1488 msg.size = sizeof(msg.payload.u64);
1489 send_vhost_reply(fd, &msg);
1492 case VHOST_USER_SET_VRING_ENABLE:
1493 vhost_user_set_vring_enable(dev, &msg);
1495 case VHOST_USER_SEND_RARP:
1496 vhost_user_send_rarp(dev, &msg);
1499 case VHOST_USER_NET_SET_MTU:
1500 ret = vhost_user_net_set_mtu(dev, &msg);
1503 case VHOST_USER_SET_SLAVE_REQ_FD:
1504 ret = vhost_user_set_req_fd(dev, &msg);
1507 case VHOST_USER_IOTLB_MSG:
1508 ret = vhost_user_iotlb_msg(&dev, &msg);
1517 if (unlock_required)
1518 vhost_user_unlock_all_queue_pairs(dev);
1520 if (msg.flags & VHOST_USER_NEED_REPLY) {
1521 msg.payload.u64 = !!ret;
1522 msg.size = sizeof(msg.payload.u64);
1523 send_vhost_reply(fd, &msg);
1526 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1527 dev->flags |= VIRTIO_DEV_READY;
1529 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1530 if (dev->dequeue_zero_copy) {
1531 RTE_LOG(INFO, VHOST_CONFIG,
1532 "dequeue zero copy is enabled\n");
1535 if (dev->notify_ops->new_device(dev->vid) == 0)
1536 dev->flags |= VIRTIO_DEV_RUNNING;
1544 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1547 struct VhostUserMsg msg = {
1548 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1549 .flags = VHOST_USER_VERSION,
1550 .size = sizeof(msg.payload.iotlb),
1554 .type = VHOST_IOTLB_MISS,
1558 ret = send_vhost_message(dev->slave_req_fd, &msg);
1560 RTE_LOG(ERR, VHOST_CONFIG,
1561 "Failed to send IOTLB miss message (%d)\n",