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 /* VIRTIO 1.0, 2.4 Virtqueues says:
243 * Queue Size value is always a power of 2. The maximum Queue Size
246 if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
247 RTE_LOG(ERR, VHOST_CONFIG,
248 "invalid virtqueue size %u\n", vq->size);
252 if (dev->dequeue_zero_copy) {
254 vq->last_zmbuf_idx = 0;
255 vq->zmbuf_size = vq->size;
256 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
257 sizeof(struct zcopy_mbuf), 0);
258 if (vq->zmbufs == NULL) {
259 RTE_LOG(WARNING, VHOST_CONFIG,
260 "failed to allocate mem for zero copy; "
261 "zero copy is force disabled\n");
262 dev->dequeue_zero_copy = 0;
264 TAILQ_INIT(&vq->zmbuf_list);
267 vq->shadow_used_ring = rte_malloc(NULL,
268 vq->size * sizeof(struct vring_used_elem),
269 RTE_CACHE_LINE_SIZE);
270 if (!vq->shadow_used_ring) {
271 RTE_LOG(ERR, VHOST_CONFIG,
272 "failed to allocate memory for shadow used ring.\n");
276 vq->batch_copy_elems = rte_malloc(NULL,
277 vq->size * sizeof(struct batch_copy_elem),
278 RTE_CACHE_LINE_SIZE);
279 if (!vq->batch_copy_elems) {
280 RTE_LOG(ERR, VHOST_CONFIG,
281 "failed to allocate memory for batching copy.\n");
289 * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
290 * same numa node as the memory of vring descriptor.
292 #ifdef RTE_LIBRTE_VHOST_NUMA
293 static struct virtio_net*
294 numa_realloc(struct virtio_net *dev, int index)
296 int oldnode, newnode;
297 struct virtio_net *old_dev;
298 struct vhost_virtqueue *old_vq, *vq;
299 struct zcopy_mbuf *new_zmbuf;
300 struct vring_used_elem *new_shadow_used_ring;
301 struct batch_copy_elem *new_batch_copy_elems;
305 vq = old_vq = dev->virtqueue[index];
307 ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
308 MPOL_F_NODE | MPOL_F_ADDR);
310 /* check if we need to reallocate vq */
311 ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
312 MPOL_F_NODE | MPOL_F_ADDR);
314 RTE_LOG(ERR, VHOST_CONFIG,
315 "Unable to get vq numa information.\n");
318 if (oldnode != newnode) {
319 RTE_LOG(INFO, VHOST_CONFIG,
320 "reallocate vq from %d to %d node\n", oldnode, newnode);
321 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
325 memcpy(vq, old_vq, sizeof(*vq));
326 TAILQ_INIT(&vq->zmbuf_list);
328 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
329 sizeof(struct zcopy_mbuf), 0, newnode);
331 rte_free(vq->zmbufs);
332 vq->zmbufs = new_zmbuf;
335 new_shadow_used_ring = rte_malloc_socket(NULL,
336 vq->size * sizeof(struct vring_used_elem),
339 if (new_shadow_used_ring) {
340 rte_free(vq->shadow_used_ring);
341 vq->shadow_used_ring = new_shadow_used_ring;
344 new_batch_copy_elems = rte_malloc_socket(NULL,
345 vq->size * sizeof(struct batch_copy_elem),
348 if (new_batch_copy_elems) {
349 rte_free(vq->batch_copy_elems);
350 vq->batch_copy_elems = new_batch_copy_elems;
356 /* check if we need to reallocate dev */
357 ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
358 MPOL_F_NODE | MPOL_F_ADDR);
360 RTE_LOG(ERR, VHOST_CONFIG,
361 "Unable to get dev numa information.\n");
364 if (oldnode != newnode) {
365 RTE_LOG(INFO, VHOST_CONFIG,
366 "reallocate dev from %d to %d node\n",
368 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
374 memcpy(dev, old_dev, sizeof(*dev));
379 dev->virtqueue[index] = vq;
380 vhost_devices[dev->vid] = dev;
383 vhost_user_iotlb_init(dev, index);
388 static struct virtio_net*
389 numa_realloc(struct virtio_net *dev, int index __rte_unused)
395 /* Converts QEMU virtual address to Vhost virtual address. */
397 qva_to_vva(struct virtio_net *dev, uint64_t qva)
399 struct rte_vhost_mem_region *reg;
402 /* Find the region where the address lives. */
403 for (i = 0; i < dev->mem->nregions; i++) {
404 reg = &dev->mem->regions[i];
406 if (qva >= reg->guest_user_addr &&
407 qva < reg->guest_user_addr + reg->size) {
408 return qva - reg->guest_user_addr +
418 * Converts ring address to Vhost virtual address.
419 * If IOMMU is enabled, the ring address is a guest IO virtual address,
420 * else it is a QEMU virtual address.
423 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
424 uint64_t ra, uint64_t size)
426 if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
429 vva = vhost_user_iotlb_cache_find(vq, ra,
430 &size, VHOST_ACCESS_RW);
432 vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
437 return qva_to_vva(dev, ra);
440 static struct virtio_net *
441 translate_ring_addresses(struct virtio_net *dev, int vq_index)
443 struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
444 struct vhost_vring_addr *addr = &vq->ring_addrs;
446 /* The addresses are converted from QEMU virtual to Vhost virtual. */
447 if (vq->desc && vq->avail && vq->used)
450 vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
451 vq, addr->desc_user_addr, sizeof(struct vring_desc));
453 RTE_LOG(DEBUG, VHOST_CONFIG,
454 "(%d) failed to find desc ring address.\n",
459 dev = numa_realloc(dev, vq_index);
460 vq = dev->virtqueue[vq_index];
461 addr = &vq->ring_addrs;
463 vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
464 vq, addr->avail_user_addr, sizeof(struct vring_avail));
465 if (vq->avail == 0) {
466 RTE_LOG(DEBUG, VHOST_CONFIG,
467 "(%d) failed to find avail ring address.\n",
472 vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
473 vq, addr->used_user_addr, sizeof(struct vring_used));
475 RTE_LOG(DEBUG, VHOST_CONFIG,
476 "(%d) failed to find used ring address.\n",
481 if (vq->last_used_idx != vq->used->idx) {
482 RTE_LOG(WARNING, VHOST_CONFIG,
483 "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
484 "some packets maybe resent for Tx and dropped for Rx\n",
485 vq->last_used_idx, vq->used->idx);
486 vq->last_used_idx = vq->used->idx;
487 vq->last_avail_idx = vq->used->idx;
490 vq->log_guest_addr = addr->log_guest_addr;
492 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
494 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
495 dev->vid, vq->avail);
496 LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
498 LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
499 dev->vid, vq->log_guest_addr);
505 * The virtio device sends us the desc, used and avail ring addresses.
506 * This function then converts these to our address space.
509 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
511 struct vhost_virtqueue *vq;
512 struct vhost_vring_addr *addr = &msg->payload.addr;
513 struct virtio_net *dev = *pdev;
515 if (dev->mem == NULL)
518 /* addr->index refers to the queue index. The txq 1, rxq is 0. */
519 vq = dev->virtqueue[msg->payload.addr.index];
522 * Rings addresses should not be interpreted as long as the ring is not
523 * started and enabled
525 memcpy(&vq->ring_addrs, addr, sizeof(*addr));
527 vring_invalidate(dev, vq);
529 if (vq->enabled && (dev->features &
530 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
531 dev = translate_ring_addresses(dev, msg->payload.addr.index);
542 * The virtio device sends us the available ring last used index.
545 vhost_user_set_vring_base(struct virtio_net *dev,
548 dev->virtqueue[msg->payload.state.index]->last_used_idx =
549 msg->payload.state.num;
550 dev->virtqueue[msg->payload.state.index]->last_avail_idx =
551 msg->payload.state.num;
557 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
558 uint64_t host_phys_addr, uint64_t size)
560 struct guest_page *page, *last_page;
562 if (dev->nr_guest_pages == dev->max_guest_pages) {
563 dev->max_guest_pages *= 2;
564 dev->guest_pages = realloc(dev->guest_pages,
565 dev->max_guest_pages * sizeof(*page));
568 if (dev->nr_guest_pages > 0) {
569 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
570 /* merge if the two pages are continuous */
571 if (host_phys_addr == last_page->host_phys_addr +
573 last_page->size += size;
578 page = &dev->guest_pages[dev->nr_guest_pages++];
579 page->guest_phys_addr = guest_phys_addr;
580 page->host_phys_addr = host_phys_addr;
585 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
588 uint64_t reg_size = reg->size;
589 uint64_t host_user_addr = reg->host_user_addr;
590 uint64_t guest_phys_addr = reg->guest_phys_addr;
591 uint64_t host_phys_addr;
594 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
595 size = page_size - (guest_phys_addr & (page_size - 1));
596 size = RTE_MIN(size, reg_size);
598 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
599 host_user_addr += size;
600 guest_phys_addr += size;
603 while (reg_size > 0) {
604 size = RTE_MIN(reg_size, page_size);
605 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
607 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
609 host_user_addr += size;
610 guest_phys_addr += size;
615 #ifdef RTE_LIBRTE_VHOST_DEBUG
616 /* TODO: enable it only in debug mode? */
618 dump_guest_pages(struct virtio_net *dev)
621 struct guest_page *page;
623 for (i = 0; i < dev->nr_guest_pages; i++) {
624 page = &dev->guest_pages[i];
626 RTE_LOG(INFO, VHOST_CONFIG,
627 "guest physical page region %u\n"
628 "\t guest_phys_addr: %" PRIx64 "\n"
629 "\t host_phys_addr : %" PRIx64 "\n"
630 "\t size : %" PRIx64 "\n",
632 page->guest_phys_addr,
633 page->host_phys_addr,
638 #define dump_guest_pages(dev)
642 vhost_memory_changed(struct VhostUserMemory *new,
643 struct rte_vhost_memory *old)
647 if (new->nregions != old->nregions)
650 for (i = 0; i < new->nregions; ++i) {
651 VhostUserMemoryRegion *new_r = &new->regions[i];
652 struct rte_vhost_mem_region *old_r = &old->regions[i];
654 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
656 if (new_r->memory_size != old_r->size)
658 if (new_r->userspace_addr != old_r->guest_user_addr)
666 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
668 struct VhostUserMemory memory = pmsg->payload.memory;
669 struct rte_vhost_mem_region *reg;
672 uint64_t mmap_offset;
677 if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
678 RTE_LOG(ERR, VHOST_CONFIG,
679 "too many memory regions (%u)\n", memory.nregions);
683 if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
684 RTE_LOG(INFO, VHOST_CONFIG,
685 "(%d) memory regions not changed\n", dev->vid);
687 for (i = 0; i < memory.nregions; i++)
694 free_mem_region(dev);
699 dev->nr_guest_pages = 0;
700 if (!dev->guest_pages) {
701 dev->max_guest_pages = 8;
702 dev->guest_pages = malloc(dev->max_guest_pages *
703 sizeof(struct guest_page));
704 if (dev->guest_pages == NULL) {
705 RTE_LOG(ERR, VHOST_CONFIG,
706 "(%d) failed to allocate memory "
707 "for dev->guest_pages\n",
713 dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
714 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
715 if (dev->mem == NULL) {
716 RTE_LOG(ERR, VHOST_CONFIG,
717 "(%d) failed to allocate memory for dev->mem\n",
721 dev->mem->nregions = memory.nregions;
723 for (i = 0; i < memory.nregions; i++) {
725 reg = &dev->mem->regions[i];
727 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
728 reg->guest_user_addr = memory.regions[i].userspace_addr;
729 reg->size = memory.regions[i].memory_size;
732 mmap_offset = memory.regions[i].mmap_offset;
734 /* Check for memory_size + mmap_offset overflow */
735 if (mmap_offset >= -reg->size) {
736 RTE_LOG(ERR, VHOST_CONFIG,
737 "mmap_offset (%#"PRIx64") and memory_size "
738 "(%#"PRIx64") overflow\n",
739 mmap_offset, reg->size);
743 mmap_size = reg->size + mmap_offset;
745 /* mmap() without flag of MAP_ANONYMOUS, should be called
746 * with length argument aligned with hugepagesz at older
747 * longterm version Linux, like 2.6.32 and 3.2.72, or
748 * mmap() will fail with EINVAL.
750 * to avoid failure, make sure in caller to keep length
753 alignment = get_blk_size(fd);
754 if (alignment == (uint64_t)-1) {
755 RTE_LOG(ERR, VHOST_CONFIG,
756 "couldn't get hugepage size through fstat\n");
759 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
761 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
762 MAP_SHARED | MAP_POPULATE, fd, 0);
764 if (mmap_addr == MAP_FAILED) {
765 RTE_LOG(ERR, VHOST_CONFIG,
766 "mmap region %u failed.\n", i);
770 reg->mmap_addr = mmap_addr;
771 reg->mmap_size = mmap_size;
772 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
775 if (dev->dequeue_zero_copy)
776 add_guest_pages(dev, reg, alignment);
778 RTE_LOG(INFO, VHOST_CONFIG,
779 "guest memory region %u, size: 0x%" PRIx64 "\n"
780 "\t guest physical addr: 0x%" PRIx64 "\n"
781 "\t guest virtual addr: 0x%" PRIx64 "\n"
782 "\t host virtual addr: 0x%" PRIx64 "\n"
783 "\t mmap addr : 0x%" PRIx64 "\n"
784 "\t mmap size : 0x%" PRIx64 "\n"
785 "\t mmap align: 0x%" PRIx64 "\n"
786 "\t mmap off : 0x%" PRIx64 "\n",
788 reg->guest_phys_addr,
789 reg->guest_user_addr,
791 (uint64_t)(uintptr_t)mmap_addr,
797 dump_guest_pages(dev);
802 free_mem_region(dev);
809 vq_is_ready(struct vhost_virtqueue *vq)
811 return vq && vq->desc && vq->avail && vq->used &&
812 vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
813 vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
817 virtio_is_ready(struct virtio_net *dev)
819 struct vhost_virtqueue *vq;
822 if (dev->nr_vring == 0)
825 for (i = 0; i < dev->nr_vring; i++) {
826 vq = dev->virtqueue[i];
828 if (!vq_is_ready(vq))
832 RTE_LOG(INFO, VHOST_CONFIG,
833 "virtio is now ready for processing.\n");
838 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
840 struct vhost_vring_file file;
841 struct vhost_virtqueue *vq;
843 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
844 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
845 file.fd = VIRTIO_INVALID_EVENTFD;
847 file.fd = pmsg->fds[0];
848 RTE_LOG(INFO, VHOST_CONFIG,
849 "vring call idx:%d file:%d\n", file.index, file.fd);
851 vq = dev->virtqueue[file.index];
855 vq->callfd = file.fd;
859 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
861 struct vhost_vring_file file;
862 struct vhost_virtqueue *vq;
863 struct virtio_net *dev = *pdev;
865 file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
866 if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
867 file.fd = VIRTIO_INVALID_EVENTFD;
869 file.fd = pmsg->fds[0];
870 RTE_LOG(INFO, VHOST_CONFIG,
871 "vring kick idx:%d file:%d\n", file.index, file.fd);
873 /* Interpret ring addresses only when ring is started. */
874 dev = translate_ring_addresses(dev, file.index);
880 vq = dev->virtqueue[file.index];
883 * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
884 * the ring starts already enabled. Otherwise, it is enabled via
885 * the SET_VRING_ENABLE message.
887 if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
892 vq->kickfd = file.fd;
896 free_zmbufs(struct vhost_virtqueue *vq)
898 struct zcopy_mbuf *zmbuf, *next;
900 for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
901 zmbuf != NULL; zmbuf = next) {
902 next = TAILQ_NEXT(zmbuf, next);
904 rte_pktmbuf_free(zmbuf->mbuf);
905 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
908 rte_free(vq->zmbufs);
912 * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
915 vhost_user_get_vring_base(struct virtio_net *dev,
918 struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
920 /* We have to stop the queue (virtio) if it is running. */
921 if (dev->flags & VIRTIO_DEV_RUNNING) {
922 dev->flags &= ~VIRTIO_DEV_RUNNING;
923 dev->notify_ops->destroy_device(dev->vid);
926 dev->flags &= ~VIRTIO_DEV_READY;
928 /* Here we are safe to get the last used index */
929 msg->payload.state.num = vq->last_used_idx;
931 RTE_LOG(INFO, VHOST_CONFIG,
932 "vring base idx:%d file:%d\n", msg->payload.state.index,
933 msg->payload.state.num);
935 * Based on current qemu vhost-user implementation, this message is
936 * sent and only sent in vhost_vring_stop.
937 * TODO: cleanup the vring, it isn't usable since here.
942 vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
944 if (dev->dequeue_zero_copy)
946 rte_free(vq->shadow_used_ring);
947 vq->shadow_used_ring = NULL;
949 rte_free(vq->batch_copy_elems);
950 vq->batch_copy_elems = NULL;
956 * when virtio queues are ready to work, qemu will send us to
957 * enable the virtio queue pair.
960 vhost_user_set_vring_enable(struct virtio_net *dev,
963 int enable = (int)msg->payload.state.num;
965 RTE_LOG(INFO, VHOST_CONFIG,
966 "set queue enable: %d to qp idx: %d\n",
967 enable, msg->payload.state.index);
969 if (dev->notify_ops->vring_state_changed)
970 dev->notify_ops->vring_state_changed(dev->vid,
971 msg->payload.state.index, enable);
973 dev->virtqueue[msg->payload.state.index]->enabled = enable;
979 vhost_user_get_protocol_features(struct virtio_net *dev,
980 struct VhostUserMsg *msg)
982 uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
984 rte_vhost_driver_get_features(dev->ifname, &features);
987 * REPLY_ACK protocol feature is only mandatory for now
988 * for IOMMU feature. If IOMMU is explicitly disabled by the
989 * application, disable also REPLY_ACK feature for older buggy
990 * Qemu versions (from v2.7.0 to v2.9.0).
992 if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
993 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
995 msg->payload.u64 = protocol_features;
996 msg->size = sizeof(msg->payload.u64);
1000 vhost_user_set_protocol_features(struct virtio_net *dev,
1001 uint64_t protocol_features)
1003 if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1006 dev->protocol_features = protocol_features;
1010 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1012 int fd = msg->fds[0];
1017 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1021 if (msg->size != sizeof(VhostUserLog)) {
1022 RTE_LOG(ERR, VHOST_CONFIG,
1023 "invalid log base msg size: %"PRId32" != %d\n",
1024 msg->size, (int)sizeof(VhostUserLog));
1028 size = msg->payload.log.mmap_size;
1029 off = msg->payload.log.mmap_offset;
1031 /* Don't allow mmap_offset to point outside the mmap region */
1033 RTE_LOG(ERR, VHOST_CONFIG,
1034 "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1039 RTE_LOG(INFO, VHOST_CONFIG,
1040 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1044 * mmap from 0 to workaround a hugepage mmap bug: mmap will
1045 * fail when offset is not page size aligned.
1047 addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1049 if (addr == MAP_FAILED) {
1050 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1055 * Free previously mapped log memory on occasionally
1056 * multiple VHOST_USER_SET_LOG_BASE.
1058 if (dev->log_addr) {
1059 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1061 dev->log_addr = (uint64_t)(uintptr_t)addr;
1062 dev->log_base = dev->log_addr + off;
1063 dev->log_size = size;
1069 * An rarp packet is constructed and broadcasted to notify switches about
1070 * the new location of the migrated VM, so that packets from outside will
1071 * not be lost after migration.
1073 * However, we don't actually "send" a rarp packet here, instead, we set
1074 * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1077 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1079 uint8_t *mac = (uint8_t *)&msg->payload.u64;
1081 RTE_LOG(DEBUG, VHOST_CONFIG,
1082 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1083 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1084 memcpy(dev->mac.addr_bytes, mac, 6);
1087 * Set the flag to inject a RARP broadcast packet at
1088 * rte_vhost_dequeue_burst().
1090 * rte_smp_wmb() is for making sure the mac is copied
1091 * before the flag is set.
1094 rte_atomic16_set(&dev->broadcast_rarp, 1);
1100 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1102 if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1103 msg->payload.u64 > VIRTIO_MAX_MTU) {
1104 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1110 dev->mtu = msg->payload.u64;
1116 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1118 int fd = msg->fds[0];
1121 RTE_LOG(ERR, VHOST_CONFIG,
1122 "Invalid file descriptor for slave channel (%d)\n",
1127 dev->slave_req_fd = fd;
1133 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1135 struct vhost_vring_addr *ra;
1136 uint64_t start, end;
1139 end = start + imsg->size;
1141 ra = &vq->ring_addrs;
1142 if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1144 if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1146 if (ra->used_user_addr >= start && ra->used_user_addr < end)
1153 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1154 struct vhost_iotlb_msg *imsg)
1156 uint64_t istart, iend, vstart, vend;
1158 istart = imsg->iova;
1159 iend = istart + imsg->size - 1;
1161 vstart = (uintptr_t)vq->desc;
1162 vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1163 if (vstart <= iend && istart <= vend)
1166 vstart = (uintptr_t)vq->avail;
1167 vend = vstart + sizeof(struct vring_avail);
1168 vend += sizeof(uint16_t) * vq->size - 1;
1169 if (vstart <= iend && istart <= vend)
1172 vstart = (uintptr_t)vq->used;
1173 vend = vstart + sizeof(struct vring_used);
1174 vend += sizeof(struct vring_used_elem) * vq->size - 1;
1175 if (vstart <= iend && istart <= vend)
1182 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1184 struct virtio_net *dev = *pdev;
1185 struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1189 switch (imsg->type) {
1190 case VHOST_IOTLB_UPDATE:
1191 vva = qva_to_vva(dev, imsg->uaddr);
1195 for (i = 0; i < dev->nr_vring; i++) {
1196 struct vhost_virtqueue *vq = dev->virtqueue[i];
1198 vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1199 imsg->size, imsg->perm);
1201 if (is_vring_iotlb_update(vq, imsg))
1202 *pdev = dev = translate_ring_addresses(dev, i);
1205 case VHOST_IOTLB_INVALIDATE:
1206 for (i = 0; i < dev->nr_vring; i++) {
1207 struct vhost_virtqueue *vq = dev->virtqueue[i];
1209 vhost_user_iotlb_cache_remove(vq, imsg->iova,
1212 if (is_vring_iotlb_invalidate(vq, imsg))
1213 vring_invalidate(dev, vq);
1217 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1225 /* return bytes# of read on success or negative val on failure. */
1227 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1231 ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1232 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1236 if (msg && msg->size) {
1237 if (msg->size > sizeof(msg->payload)) {
1238 RTE_LOG(ERR, VHOST_CONFIG,
1239 "invalid msg size: %d\n", msg->size);
1242 ret = read(sockfd, &msg->payload, msg->size);
1245 if (ret != (int)msg->size) {
1246 RTE_LOG(ERR, VHOST_CONFIG,
1247 "read control message failed\n");
1256 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1261 return send_fd_message(sockfd, (char *)msg,
1262 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1266 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1271 msg->flags &= ~VHOST_USER_VERSION_MASK;
1272 msg->flags &= ~VHOST_USER_NEED_REPLY;
1273 msg->flags |= VHOST_USER_VERSION;
1274 msg->flags |= VHOST_USER_REPLY_MASK;
1276 return send_vhost_message(sockfd, msg);
1280 * Allocate a queue pair if it hasn't been allocated yet
1283 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1287 switch (msg->request.master) {
1288 case VHOST_USER_SET_VRING_KICK:
1289 case VHOST_USER_SET_VRING_CALL:
1290 case VHOST_USER_SET_VRING_ERR:
1291 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1293 case VHOST_USER_SET_VRING_NUM:
1294 case VHOST_USER_SET_VRING_BASE:
1295 case VHOST_USER_SET_VRING_ENABLE:
1296 vring_idx = msg->payload.state.index;
1298 case VHOST_USER_SET_VRING_ADDR:
1299 vring_idx = msg->payload.addr.index;
1305 if (vring_idx >= VHOST_MAX_VRING) {
1306 RTE_LOG(ERR, VHOST_CONFIG,
1307 "invalid vring index: %u\n", vring_idx);
1311 if (dev->virtqueue[vring_idx])
1314 return alloc_vring_queue(dev, vring_idx);
1318 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1321 unsigned int vq_num = 0;
1323 while (vq_num < dev->nr_vring) {
1324 struct vhost_virtqueue *vq = dev->virtqueue[i];
1327 rte_spinlock_lock(&vq->access_lock);
1335 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1338 unsigned int vq_num = 0;
1340 while (vq_num < dev->nr_vring) {
1341 struct vhost_virtqueue *vq = dev->virtqueue[i];
1344 rte_spinlock_unlock(&vq->access_lock);
1352 vhost_user_msg_handler(int vid, int fd)
1354 struct virtio_net *dev;
1355 struct VhostUserMsg msg;
1357 int unlock_required = 0;
1359 dev = get_device(vid);
1363 if (!dev->notify_ops) {
1364 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1365 if (!dev->notify_ops) {
1366 RTE_LOG(ERR, VHOST_CONFIG,
1367 "failed to get callback ops for driver %s\n",
1373 ret = read_vhost_message(fd, &msg);
1374 if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1376 RTE_LOG(ERR, VHOST_CONFIG,
1377 "vhost read message failed\n");
1379 RTE_LOG(INFO, VHOST_CONFIG,
1380 "vhost peer closed\n");
1382 RTE_LOG(ERR, VHOST_CONFIG,
1383 "vhost read incorrect message\n");
1389 if (msg.request.master != VHOST_USER_IOTLB_MSG)
1390 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1391 vhost_message_str[msg.request.master]);
1393 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1394 vhost_message_str[msg.request.master]);
1396 ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1398 RTE_LOG(ERR, VHOST_CONFIG,
1399 "failed to alloc queue\n");
1404 * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1405 * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1406 * and device is destroyed. destroy_device waits for queues to be
1407 * inactive, so it is safe. Otherwise taking the access_lock
1408 * would cause a dead lock.
1410 switch (msg.request.master) {
1411 case VHOST_USER_SET_FEATURES:
1412 case VHOST_USER_SET_PROTOCOL_FEATURES:
1413 case VHOST_USER_SET_OWNER:
1414 case VHOST_USER_SET_MEM_TABLE:
1415 case VHOST_USER_SET_LOG_BASE:
1416 case VHOST_USER_SET_LOG_FD:
1417 case VHOST_USER_SET_VRING_NUM:
1418 case VHOST_USER_SET_VRING_ADDR:
1419 case VHOST_USER_SET_VRING_BASE:
1420 case VHOST_USER_SET_VRING_KICK:
1421 case VHOST_USER_SET_VRING_CALL:
1422 case VHOST_USER_SET_VRING_ERR:
1423 case VHOST_USER_SET_VRING_ENABLE:
1424 case VHOST_USER_SEND_RARP:
1425 case VHOST_USER_NET_SET_MTU:
1426 case VHOST_USER_SET_SLAVE_REQ_FD:
1427 vhost_user_lock_all_queue_pairs(dev);
1428 unlock_required = 1;
1435 switch (msg.request.master) {
1436 case VHOST_USER_GET_FEATURES:
1437 msg.payload.u64 = vhost_user_get_features(dev);
1438 msg.size = sizeof(msg.payload.u64);
1439 send_vhost_reply(fd, &msg);
1441 case VHOST_USER_SET_FEATURES:
1442 ret = vhost_user_set_features(dev, msg.payload.u64);
1447 case VHOST_USER_GET_PROTOCOL_FEATURES:
1448 vhost_user_get_protocol_features(dev, &msg);
1449 send_vhost_reply(fd, &msg);
1451 case VHOST_USER_SET_PROTOCOL_FEATURES:
1452 vhost_user_set_protocol_features(dev, msg.payload.u64);
1455 case VHOST_USER_SET_OWNER:
1456 vhost_user_set_owner();
1458 case VHOST_USER_RESET_OWNER:
1459 vhost_user_reset_owner(dev);
1462 case VHOST_USER_SET_MEM_TABLE:
1463 ret = vhost_user_set_mem_table(dev, &msg);
1466 case VHOST_USER_SET_LOG_BASE:
1467 vhost_user_set_log_base(dev, &msg);
1469 /* it needs a reply */
1470 msg.size = sizeof(msg.payload.u64);
1471 send_vhost_reply(fd, &msg);
1473 case VHOST_USER_SET_LOG_FD:
1475 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1478 case VHOST_USER_SET_VRING_NUM:
1479 vhost_user_set_vring_num(dev, &msg);
1481 case VHOST_USER_SET_VRING_ADDR:
1482 vhost_user_set_vring_addr(&dev, &msg);
1484 case VHOST_USER_SET_VRING_BASE:
1485 vhost_user_set_vring_base(dev, &msg);
1488 case VHOST_USER_GET_VRING_BASE:
1489 vhost_user_get_vring_base(dev, &msg);
1490 msg.size = sizeof(msg.payload.state);
1491 send_vhost_reply(fd, &msg);
1494 case VHOST_USER_SET_VRING_KICK:
1495 vhost_user_set_vring_kick(&dev, &msg);
1497 case VHOST_USER_SET_VRING_CALL:
1498 vhost_user_set_vring_call(dev, &msg);
1501 case VHOST_USER_SET_VRING_ERR:
1502 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1504 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1507 case VHOST_USER_GET_QUEUE_NUM:
1508 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1509 msg.size = sizeof(msg.payload.u64);
1510 send_vhost_reply(fd, &msg);
1513 case VHOST_USER_SET_VRING_ENABLE:
1514 vhost_user_set_vring_enable(dev, &msg);
1516 case VHOST_USER_SEND_RARP:
1517 vhost_user_send_rarp(dev, &msg);
1520 case VHOST_USER_NET_SET_MTU:
1521 ret = vhost_user_net_set_mtu(dev, &msg);
1524 case VHOST_USER_SET_SLAVE_REQ_FD:
1525 ret = vhost_user_set_req_fd(dev, &msg);
1528 case VHOST_USER_IOTLB_MSG:
1529 ret = vhost_user_iotlb_msg(&dev, &msg);
1538 if (unlock_required)
1539 vhost_user_unlock_all_queue_pairs(dev);
1541 if (msg.flags & VHOST_USER_NEED_REPLY) {
1542 msg.payload.u64 = !!ret;
1543 msg.size = sizeof(msg.payload.u64);
1544 send_vhost_reply(fd, &msg);
1547 if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1548 dev->flags |= VIRTIO_DEV_READY;
1550 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1551 if (dev->dequeue_zero_copy) {
1552 RTE_LOG(INFO, VHOST_CONFIG,
1553 "dequeue zero copy is enabled\n");
1556 if (dev->notify_ops->new_device(dev->vid) == 0)
1557 dev->flags |= VIRTIO_DEV_RUNNING;
1565 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1568 struct VhostUserMsg msg = {
1569 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1570 .flags = VHOST_USER_VERSION,
1571 .size = sizeof(msg.payload.iotlb),
1575 .type = VHOST_IOTLB_MISS,
1579 ret = send_vhost_message(dev->slave_req_fd, &msg);
1581 RTE_LOG(ERR, VHOST_CONFIG,
1582 "Failed to send IOTLB miss message (%d)\n",