lib/librte_vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <string.h>
   9 #include <unistd.h>
  10 #include <sys/mman.h>
  11 #include <sys/types.h>
  12 #include <sys/stat.h>
  13 #include <assert.h>
  14 #ifdef RTE_LIBRTE_VHOST_NUMA
  15 #include <numaif.h>
  16 #endif
  17
  18 #include <rte_common.h>
  19 #include <rte_malloc.h>
  20 #include <rte_log.h>
  21
  22 #include "iotlb.h"
  23 #include "vhost.h"
  24 #include "vhost_user.h"
  25
  26 #define VIRTIO_MIN_MTU 68
  27 #define VIRTIO_MAX_MTU 65535
  28
  29 static const char *vhost_message_str[VHOST_USER_MAX] = {
  30         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  31         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  32         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  33         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  34         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  35         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  36         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  37         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  38         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  39         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  40         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  41         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  42         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  43         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  44         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  45         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  46         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  47         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  48         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  49         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  50         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  51         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  52         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  53 };
  54
  55 static uint64_t
  56 get_blk_size(int fd)
  57 {
  58         struct stat stat;
  59         int ret;
  60
  61         ret = fstat(fd, &stat);
  62         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  63 }
  64
  65 static void
  66 free_mem_region(struct virtio_net *dev)
  67 {
  68         uint32_t i;
  69         struct rte_vhost_mem_region *reg;
  70
  71         if (!dev || !dev->mem)
  72                 return;
  73
  74         for (i = 0; i < dev->mem->nregions; i++) {
  75                 reg = &dev->mem->regions[i];
  76                 if (reg->host_user_addr) {
  77                         munmap(reg->mmap_addr, reg->mmap_size);
  78                         close(reg->fd);
  79                 }
  80         }
  81 }
  82
  83 void
  84 vhost_backend_cleanup(struct virtio_net *dev)
  85 {
  86         if (dev->mem) {
  87                 free_mem_region(dev);
  88                 rte_free(dev->mem);
  89                 dev->mem = NULL;
  90         }
  91
  92         free(dev->guest_pages);
  93         dev->guest_pages = NULL;
  94
  95         if (dev->log_addr) {
  96                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
  97                 dev->log_addr = 0;
  98         }
  99
 100         if (dev->slave_req_fd >= 0) {
 101                 close(dev->slave_req_fd);
 102                 dev->slave_req_fd = -1;
 103         }
 104 }
 105
 106 /*
 107  * This function just returns success at the moment unless
 108  * the device hasn't been initialised.
 109  */
 110 static int
 111 vhost_user_set_owner(void)
 112 {
 113         return 0;
 114 }
 115
 116 static int
 117 vhost_user_reset_owner(struct virtio_net *dev)
 118 {
 119         if (dev->flags & VIRTIO_DEV_RUNNING) {
 120                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 121                 dev->notify_ops->destroy_device(dev->vid);
 122         }
 123
 124         cleanup_device(dev, 0);
 125         reset_device(dev);
 126         return 0;
 127 }
 128
 129 /*
 130  * The features that we support are requested.
 131  */
 132 static uint64_t
 133 vhost_user_get_features(struct virtio_net *dev)
 134 {
 135         uint64_t features = 0;
 136
 137         rte_vhost_driver_get_features(dev->ifname, &features);
 138         return features;
 139 }
 140
 141 /*
 142  * We receive the negotiated features supported by us and the virtio device.
 143  */
 144 static int
 145 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 146 {
 147         uint64_t vhost_features = 0;
 148
 149         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 150         if (features & ~vhost_features) {
 151                 RTE_LOG(ERR, VHOST_CONFIG,
 152                         "(%d) received invalid negotiated features.\n",
 153                         dev->vid);
 154                 return -1;
 155         }
 156
 157         if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
 158                 if (dev->notify_ops->features_changed)
 159                         dev->notify_ops->features_changed(dev->vid, features);
 160         }
 161
 162         dev->features = features;
 163         if (dev->features &
 164                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 165                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 166         } else {
 167                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 168         }
 169         LOG_DEBUG(VHOST_CONFIG,
 170                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 171                 dev->vid,
 172                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 173                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 174
 175         return 0;
 176 }
 177
 178 /*
 179  * The virtio device sends us the size of the descriptor ring.
 180  */
 181 static int
 182 vhost_user_set_vring_num(struct virtio_net *dev,
 183                          VhostUserMsg *msg)
 184 {
 185         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 186
 187         vq->size = msg->payload.state.num;
 188
 189         if (dev->dequeue_zero_copy) {
 190                 vq->nr_zmbuf = 0;
 191                 vq->last_zmbuf_idx = 0;
 192                 vq->zmbuf_size = vq->size;
 193                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 194                                          sizeof(struct zcopy_mbuf), 0);
 195                 if (vq->zmbufs == NULL) {
 196                         RTE_LOG(WARNING, VHOST_CONFIG,
 197                                 "failed to allocate mem for zero copy; "
 198                                 "zero copy is force disabled\n");
 199                         dev->dequeue_zero_copy = 0;
 200                 }
 201         }
 202
 203         vq->shadow_used_ring = rte_malloc(NULL,
 204                                 vq->size * sizeof(struct vring_used_elem),
 205                                 RTE_CACHE_LINE_SIZE);
 206         if (!vq->shadow_used_ring) {
 207                 RTE_LOG(ERR, VHOST_CONFIG,
 208                         "failed to allocate memory for shadow used ring.\n");
 209                 return -1;
 210         }
 211
 212         vq->batch_copy_elems = rte_malloc(NULL,
 213                                 vq->size * sizeof(struct batch_copy_elem),
 214                                 RTE_CACHE_LINE_SIZE);
 215         if (!vq->batch_copy_elems) {
 216                 RTE_LOG(ERR, VHOST_CONFIG,
 217                         "failed to allocate memory for batching copy.\n");
 218                 return -1;
 219         }
 220
 221         return 0;
 222 }
 223
 224 /*
 225  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 226  * same numa node as the memory of vring descriptor.
 227  */
 228 #ifdef RTE_LIBRTE_VHOST_NUMA
 229 static struct virtio_net*
 230 numa_realloc(struct virtio_net *dev, int index)
 231 {
 232         int oldnode, newnode;
 233         struct virtio_net *old_dev;
 234         struct vhost_virtqueue *old_vq, *vq;
 235         int ret;
 236
 237         old_dev = dev;
 238         vq = old_vq = dev->virtqueue[index];
 239
 240         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 241                             MPOL_F_NODE | MPOL_F_ADDR);
 242
 243         /* check if we need to reallocate vq */
 244         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 245                              MPOL_F_NODE | MPOL_F_ADDR);
 246         if (ret) {
 247                 RTE_LOG(ERR, VHOST_CONFIG,
 248                         "Unable to get vq numa information.\n");
 249                 return dev;
 250         }
 251         if (oldnode != newnode) {
 252                 RTE_LOG(INFO, VHOST_CONFIG,
 253                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 254                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 255                 if (!vq)
 256                         return dev;
 257
 258                 memcpy(vq, old_vq, sizeof(*vq));
 259                 rte_free(old_vq);
 260         }
 261
 262         /* check if we need to reallocate dev */
 263         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 264                             MPOL_F_NODE | MPOL_F_ADDR);
 265         if (ret) {
 266                 RTE_LOG(ERR, VHOST_CONFIG,
 267                         "Unable to get dev numa information.\n");
 268                 goto out;
 269         }
 270         if (oldnode != newnode) {
 271                 RTE_LOG(INFO, VHOST_CONFIG,
 272                         "reallocate dev from %d to %d node\n",
 273                         oldnode, newnode);
 274                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 275                 if (!dev) {
 276                         dev = old_dev;
 277                         goto out;
 278                 }
 279
 280                 memcpy(dev, old_dev, sizeof(*dev));
 281                 rte_free(old_dev);
 282         }
 283
 284 out:
 285         dev->virtqueue[index] = vq;
 286         vhost_devices[dev->vid] = dev;
 287
 288         if (old_vq != vq)
 289                 vhost_user_iotlb_init(dev, index);
 290
 291         return dev;
 292 }
 293 #else
 294 static struct virtio_net*
 295 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 296 {
 297         return dev;
 298 }
 299 #endif
 300
 301 /* Converts QEMU virtual address to Vhost virtual address. */
 302 static uint64_t
 303 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 304 {
 305         struct rte_vhost_mem_region *reg;
 306         uint32_t i;
 307
 308         /* Find the region where the address lives. */
 309         for (i = 0; i < dev->mem->nregions; i++) {
 310                 reg = &dev->mem->regions[i];
 311
 312                 if (qva >= reg->guest_user_addr &&
 313                     qva <  reg->guest_user_addr + reg->size) {
 314                         return qva - reg->guest_user_addr +
 315                                reg->host_user_addr;
 316                 }
 317         }
 318
 319         return 0;
 320 }
 321
 322
 323 /*
 324  * Converts ring address to Vhost virtual address.
 325  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 326  * else it is a QEMU virtual address.
 327  */
 328 static uint64_t
 329 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 330                 uint64_t ra, uint64_t size)
 331 {
 332         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 333                 uint64_t vva;
 334
 335                 vva = vhost_user_iotlb_cache_find(vq, ra,
 336                                         &size, VHOST_ACCESS_RW);
 337                 if (!vva)
 338                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 339
 340                 return vva;
 341         }
 342
 343         return qva_to_vva(dev, ra);
 344 }
 345
 346 static struct virtio_net *
 347 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 348 {
 349         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 350         struct vhost_vring_addr *addr = &vq->ring_addrs;
 351
 352         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 353         if (vq->desc && vq->avail && vq->used)
 354                 return dev;
 355
 356         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 357                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 358         if (vq->desc == 0) {
 359                 RTE_LOG(DEBUG, VHOST_CONFIG,
 360                         "(%d) failed to find desc ring address.\n",
 361                         dev->vid);
 362                 return dev;
 363         }
 364
 365         dev = numa_realloc(dev, vq_index);
 366         vq = dev->virtqueue[vq_index];
 367         addr = &vq->ring_addrs;
 368
 369         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 370                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 371         if (vq->avail == 0) {
 372                 RTE_LOG(DEBUG, VHOST_CONFIG,
 373                         "(%d) failed to find avail ring address.\n",
 374                         dev->vid);
 375                 return dev;
 376         }
 377
 378         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 379                         vq, addr->used_user_addr, sizeof(struct vring_used));
 380         if (vq->used == 0) {
 381                 RTE_LOG(DEBUG, VHOST_CONFIG,
 382                         "(%d) failed to find used ring address.\n",
 383                         dev->vid);
 384                 return dev;
 385         }
 386
 387         if (vq->last_used_idx != vq->used->idx) {
 388                 RTE_LOG(WARNING, VHOST_CONFIG,
 389                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 390                         "some packets maybe resent for Tx and dropped for Rx\n",
 391                         vq->last_used_idx, vq->used->idx);
 392                 vq->last_used_idx  = vq->used->idx;
 393                 vq->last_avail_idx = vq->used->idx;
 394         }
 395
 396         vq->log_guest_addr = addr->log_guest_addr;
 397
 398         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 399                         dev->vid, vq->desc);
 400         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 401                         dev->vid, vq->avail);
 402         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 403                         dev->vid, vq->used);
 404         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 405                         dev->vid, vq->log_guest_addr);
 406
 407         return dev;
 408 }
 409
 410 /*
 411  * The virtio device sends us the desc, used and avail ring addresses.
 412  * This function then converts these to our address space.
 413  */
 414 static int
 415 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 416 {
 417         struct vhost_virtqueue *vq;
 418         struct vhost_vring_addr *addr = &msg->payload.addr;
 419         struct virtio_net *dev = *pdev;
 420
 421         if (dev->mem == NULL)
 422                 return -1;
 423
 424         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 425         vq = dev->virtqueue[msg->payload.addr.index];
 426
 427         /*
 428          * Rings addresses should not be interpreted as long as the ring is not
 429          * started and enabled
 430          */
 431         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 432
 433         vring_invalidate(dev, vq);
 434
 435         if (vq->enabled && (dev->features &
 436                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 437                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 438                 if (!dev)
 439                         return -1;
 440
 441                 *pdev = dev;
 442         }
 443
 444         return 0;
 445 }
 446
 447 /*
 448  * The virtio device sends us the available ring last used index.
 449  */
 450 static int
 451 vhost_user_set_vring_base(struct virtio_net *dev,
 452                           VhostUserMsg *msg)
 453 {
 454         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 455                         msg->payload.state.num;
 456         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 457                         msg->payload.state.num;
 458
 459         return 0;
 460 }
 461
 462 static void
 463 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 464                    uint64_t host_phys_addr, uint64_t size)
 465 {
 466         struct guest_page *page, *last_page;
 467
 468         if (dev->nr_guest_pages == dev->max_guest_pages) {
 469                 dev->max_guest_pages *= 2;
 470                 dev->guest_pages = realloc(dev->guest_pages,
 471                                         dev->max_guest_pages * sizeof(*page));
 472         }
 473
 474         if (dev->nr_guest_pages > 0) {
 475                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 476                 /* merge if the two pages are continuous */
 477                 if (host_phys_addr == last_page->host_phys_addr +
 478                                       last_page->size) {
 479                         last_page->size += size;
 480                         return;
 481                 }
 482         }
 483
 484         page = &dev->guest_pages[dev->nr_guest_pages++];
 485         page->guest_phys_addr = guest_phys_addr;
 486         page->host_phys_addr  = host_phys_addr;
 487         page->size = size;
 488 }
 489
 490 static void
 491 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 492                 uint64_t page_size)
 493 {
 494         uint64_t reg_size = reg->size;
 495         uint64_t host_user_addr  = reg->host_user_addr;
 496         uint64_t guest_phys_addr = reg->guest_phys_addr;
 497         uint64_t host_phys_addr;
 498         uint64_t size;
 499
 500         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 501         size = page_size - (guest_phys_addr & (page_size - 1));
 502         size = RTE_MIN(size, reg_size);
 503
 504         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 505         host_user_addr  += size;
 506         guest_phys_addr += size;
 507         reg_size -= size;
 508
 509         while (reg_size > 0) {
 510                 size = RTE_MIN(reg_size, page_size);
 511                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 512                                                   host_user_addr);
 513                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 514
 515                 host_user_addr  += size;
 516                 guest_phys_addr += size;
 517                 reg_size -= size;
 518         }
 519 }
 520
 521 #ifdef RTE_LIBRTE_VHOST_DEBUG
 522 /* TODO: enable it only in debug mode? */
 523 static void
 524 dump_guest_pages(struct virtio_net *dev)
 525 {
 526         uint32_t i;
 527         struct guest_page *page;
 528
 529         for (i = 0; i < dev->nr_guest_pages; i++) {
 530                 page = &dev->guest_pages[i];
 531
 532                 RTE_LOG(INFO, VHOST_CONFIG,
 533                         "guest physical page region %u\n"
 534                         "\t guest_phys_addr: %" PRIx64 "\n"
 535                         "\t host_phys_addr : %" PRIx64 "\n"
 536                         "\t size           : %" PRIx64 "\n",
 537                         i,
 538                         page->guest_phys_addr,
 539                         page->host_phys_addr,
 540                         page->size);
 541         }
 542 }
 543 #else
 544 #define dump_guest_pages(dev)
 545 #endif
 546
 547 static bool
 548 vhost_memory_changed(struct VhostUserMemory *new,
 549                      struct rte_vhost_memory *old)
 550 {
 551         uint32_t i;
 552
 553         if (new->nregions != old->nregions)
 554                 return true;
 555
 556         for (i = 0; i < new->nregions; ++i) {
 557                 VhostUserMemoryRegion *new_r = &new->regions[i];
 558                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 559
 560                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 561                         return true;
 562                 if (new_r->memory_size != old_r->size)
 563                         return true;
 564                 if (new_r->userspace_addr != old_r->guest_user_addr)
 565                         return true;
 566         }
 567
 568         return false;
 569 }
 570
 571 static int
 572 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 573 {
 574         struct VhostUserMemory memory = pmsg->payload.memory;
 575         struct rte_vhost_mem_region *reg;
 576         void *mmap_addr;
 577         uint64_t mmap_size;
 578         uint64_t mmap_offset;
 579         uint64_t alignment;
 580         uint32_t i;
 581         int fd;
 582
 583         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 584                 RTE_LOG(INFO, VHOST_CONFIG,
 585                         "(%d) memory regions not changed\n", dev->vid);
 586
 587                 for (i = 0; i < memory.nregions; i++)
 588                         close(pmsg->fds[i]);
 589
 590                 return 0;
 591         }
 592
 593         if (dev->mem) {
 594                 free_mem_region(dev);
 595                 rte_free(dev->mem);
 596                 dev->mem = NULL;
 597         }
 598
 599         dev->nr_guest_pages = 0;
 600         if (!dev->guest_pages) {
 601                 dev->max_guest_pages = 8;
 602                 dev->guest_pages = malloc(dev->max_guest_pages *
 603                                                 sizeof(struct guest_page));
 604                 if (dev->guest_pages == NULL) {
 605                         RTE_LOG(ERR, VHOST_CONFIG,
 606                                 "(%d) failed to allocate memory "
 607                                 "for dev->guest_pages\n",
 608                                 dev->vid);
 609                         return -1;
 610                 }
 611         }
 612
 613         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 614                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 615         if (dev->mem == NULL) {
 616                 RTE_LOG(ERR, VHOST_CONFIG,
 617                         "(%d) failed to allocate memory for dev->mem\n",
 618                         dev->vid);
 619                 return -1;
 620         }
 621         dev->mem->nregions = memory.nregions;
 622
 623         for (i = 0; i < memory.nregions; i++) {
 624                 fd  = pmsg->fds[i];
 625                 reg = &dev->mem->regions[i];
 626
 627                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 628                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 629                 reg->size            = memory.regions[i].memory_size;
 630                 reg->fd              = fd;
 631
 632                 mmap_offset = memory.regions[i].mmap_offset;
 633                 mmap_size   = reg->size + mmap_offset;
 634
 635                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 636                  * with length argument aligned with hugepagesz at older
 637                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 638                  * mmap() will fail with EINVAL.
 639                  *
 640                  * to avoid failure, make sure in caller to keep length
 641                  * aligned.
 642                  */
 643                 alignment = get_blk_size(fd);
 644                 if (alignment == (uint64_t)-1) {
 645                         RTE_LOG(ERR, VHOST_CONFIG,
 646                                 "couldn't get hugepage size through fstat\n");
 647                         goto err_mmap;
 648                 }
 649                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 650
 651                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 652                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 653
 654                 if (mmap_addr == MAP_FAILED) {
 655                         RTE_LOG(ERR, VHOST_CONFIG,
 656                                 "mmap region %u failed.\n", i);
 657                         goto err_mmap;
 658                 }
 659
 660                 reg->mmap_addr = mmap_addr;
 661                 reg->mmap_size = mmap_size;
 662                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 663                                       mmap_offset;
 664
 665                 if (dev->dequeue_zero_copy)
 666                         add_guest_pages(dev, reg, alignment);
 667
 668                 RTE_LOG(INFO, VHOST_CONFIG,
 669                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 670                         "\t guest physical addr: 0x%" PRIx64 "\n"
 671                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 672                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 673                         "\t mmap addr : 0x%" PRIx64 "\n"
 674                         "\t mmap size : 0x%" PRIx64 "\n"
 675                         "\t mmap align: 0x%" PRIx64 "\n"
 676                         "\t mmap off  : 0x%" PRIx64 "\n",
 677                         i, reg->size,
 678                         reg->guest_phys_addr,
 679                         reg->guest_user_addr,
 680                         reg->host_user_addr,
 681                         (uint64_t)(uintptr_t)mmap_addr,
 682                         mmap_size,
 683                         alignment,
 684                         mmap_offset);
 685         }
 686
 687         dump_guest_pages(dev);
 688
 689         return 0;
 690
 691 err_mmap:
 692         free_mem_region(dev);
 693         rte_free(dev->mem);
 694         dev->mem = NULL;
 695         return -1;
 696 }
 697
 698 static int
 699 vq_is_ready(struct vhost_virtqueue *vq)
 700 {
 701         return vq && vq->desc && vq->avail && vq->used &&
 702                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 703                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 704 }
 705
 706 static int
 707 virtio_is_ready(struct virtio_net *dev)
 708 {
 709         struct vhost_virtqueue *vq;
 710         uint32_t i;
 711
 712         if (dev->nr_vring == 0)
 713                 return 0;
 714
 715         for (i = 0; i < dev->nr_vring; i++) {
 716                 vq = dev->virtqueue[i];
 717
 718                 if (!vq_is_ready(vq))
 719                         return 0;
 720         }
 721
 722         RTE_LOG(INFO, VHOST_CONFIG,
 723                 "virtio is now ready for processing.\n");
 724         return 1;
 725 }
 726
 727 static void
 728 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 729 {
 730         struct vhost_vring_file file;
 731         struct vhost_virtqueue *vq;
 732
 733         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 734         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 735                 file.fd = VIRTIO_INVALID_EVENTFD;
 736         else
 737                 file.fd = pmsg->fds[0];
 738         RTE_LOG(INFO, VHOST_CONFIG,
 739                 "vring call idx:%d file:%d\n", file.index, file.fd);
 740
 741         vq = dev->virtqueue[file.index];
 742         if (vq->callfd >= 0)
 743                 close(vq->callfd);
 744
 745         vq->callfd = file.fd;
 746 }
 747
 748 static void
 749 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 750 {
 751         struct vhost_vring_file file;
 752         struct vhost_virtqueue *vq;
 753         struct virtio_net *dev = *pdev;
 754
 755         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 756         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 757                 file.fd = VIRTIO_INVALID_EVENTFD;
 758         else
 759                 file.fd = pmsg->fds[0];
 760         RTE_LOG(INFO, VHOST_CONFIG,
 761                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 762
 763         /* Interpret ring addresses only when ring is started. */
 764         dev = translate_ring_addresses(dev, file.index);
 765         if (!dev)
 766                 return;
 767
 768         *pdev = dev;
 769
 770         vq = dev->virtqueue[file.index];
 771
 772         /*
 773          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 774          * the ring starts already enabled. Otherwise, it is enabled via
 775          * the SET_VRING_ENABLE message.
 776          */
 777         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 778                 vq->enabled = 1;
 779
 780         if (vq->kickfd >= 0)
 781                 close(vq->kickfd);
 782         vq->kickfd = file.fd;
 783 }
 784
 785 static void
 786 free_zmbufs(struct vhost_virtqueue *vq)
 787 {
 788         struct zcopy_mbuf *zmbuf, *next;
 789
 790         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 791              zmbuf != NULL; zmbuf = next) {
 792                 next = TAILQ_NEXT(zmbuf, next);
 793
 794                 rte_pktmbuf_free(zmbuf->mbuf);
 795                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 796         }
 797
 798         rte_free(vq->zmbufs);
 799 }
 800
 801 /*
 802  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 803  */
 804 static int
 805 vhost_user_get_vring_base(struct virtio_net *dev,
 806                           VhostUserMsg *msg)
 807 {
 808         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 809
 810         /* We have to stop the queue (virtio) if it is running. */
 811         if (dev->flags & VIRTIO_DEV_RUNNING) {
 812                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 813                 dev->notify_ops->destroy_device(dev->vid);
 814         }
 815
 816         dev->flags &= ~VIRTIO_DEV_READY;
 817
 818         /* Here we are safe to get the last used index */
 819         msg->payload.state.num = vq->last_used_idx;
 820
 821         RTE_LOG(INFO, VHOST_CONFIG,
 822                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 823                 msg->payload.state.num);
 824         /*
 825          * Based on current qemu vhost-user implementation, this message is
 826          * sent and only sent in vhost_vring_stop.
 827          * TODO: cleanup the vring, it isn't usable since here.
 828          */
 829         if (vq->kickfd >= 0)
 830                 close(vq->kickfd);
 831
 832         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 833
 834         if (dev->dequeue_zero_copy)
 835                 free_zmbufs(vq);
 836         rte_free(vq->shadow_used_ring);
 837         vq->shadow_used_ring = NULL;
 838
 839         rte_free(vq->batch_copy_elems);
 840         vq->batch_copy_elems = NULL;
 841
 842         return 0;
 843 }
 844
 845 /*
 846  * when virtio queues are ready to work, qemu will send us to
 847  * enable the virtio queue pair.
 848  */
 849 static int
 850 vhost_user_set_vring_enable(struct virtio_net *dev,
 851                             VhostUserMsg *msg)
 852 {
 853         int enable = (int)msg->payload.state.num;
 854
 855         RTE_LOG(INFO, VHOST_CONFIG,
 856                 "set queue enable: %d to qp idx: %d\n",
 857                 enable, msg->payload.state.index);
 858
 859         if (dev->notify_ops->vring_state_changed)
 860                 dev->notify_ops->vring_state_changed(dev->vid,
 861                                 msg->payload.state.index, enable);
 862
 863         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 864
 865         return 0;
 866 }
 867
 868 static void
 869 vhost_user_get_protocol_features(struct virtio_net *dev,
 870                                  struct VhostUserMsg *msg)
 871 {
 872         uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
 873
 874         rte_vhost_driver_get_features(dev->ifname, &features);
 875
 876         /*
 877          * REPLY_ACK protocol feature is only mandatory for now
 878          * for IOMMU feature. If IOMMU is explicitly disabled by the
 879          * application, disable also REPLY_ACK feature for older buggy
 880          * Qemu versions (from v2.7.0 to v2.9.0).
 881          */
 882         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 883                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
 884
 885         msg->payload.u64 = protocol_features;
 886         msg->size = sizeof(msg->payload.u64);
 887 }
 888
 889 static void
 890 vhost_user_set_protocol_features(struct virtio_net *dev,
 891                                  uint64_t protocol_features)
 892 {
 893         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 894                 return;
 895
 896         dev->protocol_features = protocol_features;
 897 }
 898
 899 static int
 900 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 901 {
 902         int fd = msg->fds[0];
 903         uint64_t size, off;
 904         void *addr;
 905
 906         if (fd < 0) {
 907                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 908                 return -1;
 909         }
 910
 911         if (msg->size != sizeof(VhostUserLog)) {
 912                 RTE_LOG(ERR, VHOST_CONFIG,
 913                         "invalid log base msg size: %"PRId32" != %d\n",
 914                         msg->size, (int)sizeof(VhostUserLog));
 915                 return -1;
 916         }
 917
 918         size = msg->payload.log.mmap_size;
 919         off  = msg->payload.log.mmap_offset;
 920         RTE_LOG(INFO, VHOST_CONFIG,
 921                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 922                 size, off);
 923
 924         /*
 925          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 926          * fail when offset is not page size aligned.
 927          */
 928         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 929         close(fd);
 930         if (addr == MAP_FAILED) {
 931                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 932                 return -1;
 933         }
 934
 935         /*
 936          * Free previously mapped log memory on occasionally
 937          * multiple VHOST_USER_SET_LOG_BASE.
 938          */
 939         if (dev->log_addr) {
 940                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 941         }
 942         dev->log_addr = (uint64_t)(uintptr_t)addr;
 943         dev->log_base = dev->log_addr + off;
 944         dev->log_size = size;
 945
 946         return 0;
 947 }
 948
 949 /*
 950  * An rarp packet is constructed and broadcasted to notify switches about
 951  * the new location of the migrated VM, so that packets from outside will
 952  * not be lost after migration.
 953  *
 954  * However, we don't actually "send" a rarp packet here, instead, we set
 955  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 956  */
 957 static int
 958 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 959 {
 960         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 961
 962         RTE_LOG(DEBUG, VHOST_CONFIG,
 963                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 964                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 965         memcpy(dev->mac.addr_bytes, mac, 6);
 966
 967         /*
 968          * Set the flag to inject a RARP broadcast packet at
 969          * rte_vhost_dequeue_burst().
 970          *
 971          * rte_smp_wmb() is for making sure the mac is copied
 972          * before the flag is set.
 973          */
 974         rte_smp_wmb();
 975         rte_atomic16_set(&dev->broadcast_rarp, 1);
 976
 977         return 0;
 978 }
 979
 980 static int
 981 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 982 {
 983         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 984                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 985                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 986                                 msg->payload.u64);
 987
 988                 return -1;
 989         }
 990
 991         dev->mtu = msg->payload.u64;
 992
 993         return 0;
 994 }
 995
 996 static int
 997 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 998 {
 999         int fd = msg->fds[0];
1000
1001         if (fd < 0) {
1002                 RTE_LOG(ERR, VHOST_CONFIG,
1003                                 "Invalid file descriptor for slave channel (%d)\n",
1004                                 fd);
1005                 return -1;
1006         }
1007
1008         dev->slave_req_fd = fd;
1009
1010         return 0;
1011 }
1012
1013 static int
1014 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1015 {
1016         struct vhost_vring_addr *ra;
1017         uint64_t start, end;
1018
1019         start = imsg->iova;
1020         end = start + imsg->size;
1021
1022         ra = &vq->ring_addrs;
1023         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1024                 return 1;
1025         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1026                 return 1;
1027         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1028                 return 1;
1029
1030         return 0;
1031 }
1032
1033 static int
1034 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1035                                 struct vhost_iotlb_msg *imsg)
1036 {
1037         uint64_t istart, iend, vstart, vend;
1038
1039         istart = imsg->iova;
1040         iend = istart + imsg->size - 1;
1041
1042         vstart = (uintptr_t)vq->desc;
1043         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1044         if (vstart <= iend && istart <= vend)
1045                 return 1;
1046
1047         vstart = (uintptr_t)vq->avail;
1048         vend = vstart + sizeof(struct vring_avail);
1049         vend += sizeof(uint16_t) * vq->size - 1;
1050         if (vstart <= iend && istart <= vend)
1051                 return 1;
1052
1053         vstart = (uintptr_t)vq->used;
1054         vend = vstart + sizeof(struct vring_used);
1055         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1056         if (vstart <= iend && istart <= vend)
1057                 return 1;
1058
1059         return 0;
1060 }
1061
1062 static int
1063 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1064 {
1065         struct virtio_net *dev = *pdev;
1066         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1067         uint16_t i;
1068         uint64_t vva;
1069
1070         switch (imsg->type) {
1071         case VHOST_IOTLB_UPDATE:
1072                 vva = qva_to_vva(dev, imsg->uaddr);
1073                 if (!vva)
1074                         return -1;
1075
1076                 for (i = 0; i < dev->nr_vring; i++) {
1077                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1078
1079                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1080                                         imsg->size, imsg->perm);
1081
1082                         if (is_vring_iotlb_update(vq, imsg))
1083                                 *pdev = dev = translate_ring_addresses(dev, i);
1084                 }
1085                 break;
1086         case VHOST_IOTLB_INVALIDATE:
1087                 for (i = 0; i < dev->nr_vring; i++) {
1088                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1089
1090                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1091                                         imsg->size);
1092
1093                         if (is_vring_iotlb_invalidate(vq, imsg))
1094                                 vring_invalidate(dev, vq);
1095                 }
1096                 break;
1097         default:
1098                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1099                                 imsg->type);
1100                 return -1;
1101         }
1102
1103         return 0;
1104 }
1105
1106 /* return bytes# of read on success or negative val on failure. */
1107 static int
1108 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1109 {
1110         int ret;
1111
1112         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1113                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1114         if (ret <= 0)
1115                 return ret;
1116
1117         if (msg && msg->size) {
1118                 if (msg->size > sizeof(msg->payload)) {
1119                         RTE_LOG(ERR, VHOST_CONFIG,
1120                                 "invalid msg size: %d\n", msg->size);
1121                         return -1;
1122                 }
1123                 ret = read(sockfd, &msg->payload, msg->size);
1124                 if (ret <= 0)
1125                         return ret;
1126                 if (ret != (int)msg->size) {
1127                         RTE_LOG(ERR, VHOST_CONFIG,
1128                                 "read control message failed\n");
1129                         return -1;
1130                 }
1131         }
1132
1133         return ret;
1134 }
1135
1136 static int
1137 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1138 {
1139         if (!msg)
1140                 return 0;
1141
1142         return send_fd_message(sockfd, (char *)msg,
1143                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1144 }
1145
1146 static int
1147 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1148 {
1149         if (!msg)
1150                 return 0;
1151
1152         msg->flags &= ~VHOST_USER_VERSION_MASK;
1153         msg->flags &= ~VHOST_USER_NEED_REPLY;
1154         msg->flags |= VHOST_USER_VERSION;
1155         msg->flags |= VHOST_USER_REPLY_MASK;
1156
1157         return send_vhost_message(sockfd, msg);
1158 }
1159
1160 /*
1161  * Allocate a queue pair if it hasn't been allocated yet
1162  */
1163 static int
1164 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1165 {
1166         uint16_t vring_idx;
1167
1168         switch (msg->request.master) {
1169         case VHOST_USER_SET_VRING_KICK:
1170         case VHOST_USER_SET_VRING_CALL:
1171         case VHOST_USER_SET_VRING_ERR:
1172                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1173                 break;
1174         case VHOST_USER_SET_VRING_NUM:
1175         case VHOST_USER_SET_VRING_BASE:
1176         case VHOST_USER_SET_VRING_ENABLE:
1177                 vring_idx = msg->payload.state.index;
1178                 break;
1179         case VHOST_USER_SET_VRING_ADDR:
1180                 vring_idx = msg->payload.addr.index;
1181                 break;
1182         default:
1183                 return 0;
1184         }
1185
1186         if (vring_idx >= VHOST_MAX_VRING) {
1187                 RTE_LOG(ERR, VHOST_CONFIG,
1188                         "invalid vring index: %u\n", vring_idx);
1189                 return -1;
1190         }
1191
1192         if (dev->virtqueue[vring_idx])
1193                 return 0;
1194
1195         return alloc_vring_queue(dev, vring_idx);
1196 }
1197
1198 int
1199 vhost_user_msg_handler(int vid, int fd)
1200 {
1201         struct virtio_net *dev;
1202         struct VhostUserMsg msg;
1203         int ret;
1204
1205         dev = get_device(vid);
1206         if (dev == NULL)
1207                 return -1;
1208
1209         if (!dev->notify_ops) {
1210                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1211                 if (!dev->notify_ops) {
1212                         RTE_LOG(ERR, VHOST_CONFIG,
1213                                 "failed to get callback ops for driver %s\n",
1214                                 dev->ifname);
1215                         return -1;
1216                 }
1217         }
1218
1219         ret = read_vhost_message(fd, &msg);
1220         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1221                 if (ret < 0)
1222                         RTE_LOG(ERR, VHOST_CONFIG,
1223                                 "vhost read message failed\n");
1224                 else if (ret == 0)
1225                         RTE_LOG(INFO, VHOST_CONFIG,
1226                                 "vhost peer closed\n");
1227                 else
1228                         RTE_LOG(ERR, VHOST_CONFIG,
1229                                 "vhost read incorrect message\n");
1230
1231                 return -1;
1232         }
1233
1234         ret = 0;
1235         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1236                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1237                         vhost_message_str[msg.request.master]);
1238         else
1239                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1240                         vhost_message_str[msg.request.master]);
1241
1242         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1243         if (ret < 0) {
1244                 RTE_LOG(ERR, VHOST_CONFIG,
1245                         "failed to alloc queue\n");
1246                 return -1;
1247         }
1248
1249         switch (msg.request.master) {
1250         case VHOST_USER_GET_FEATURES:
1251                 msg.payload.u64 = vhost_user_get_features(dev);
1252                 msg.size = sizeof(msg.payload.u64);
1253                 send_vhost_reply(fd, &msg);
1254                 break;
1255         case VHOST_USER_SET_FEATURES:
1256                 vhost_user_set_features(dev, msg.payload.u64);
1257                 break;
1258
1259         case VHOST_USER_GET_PROTOCOL_FEATURES:
1260                 vhost_user_get_protocol_features(dev, &msg);
1261                 send_vhost_reply(fd, &msg);
1262                 break;
1263         case VHOST_USER_SET_PROTOCOL_FEATURES:
1264                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1265                 break;
1266
1267         case VHOST_USER_SET_OWNER:
1268                 vhost_user_set_owner();
1269                 break;
1270         case VHOST_USER_RESET_OWNER:
1271                 vhost_user_reset_owner(dev);
1272                 break;
1273
1274         case VHOST_USER_SET_MEM_TABLE:
1275                 ret = vhost_user_set_mem_table(dev, &msg);
1276                 break;
1277
1278         case VHOST_USER_SET_LOG_BASE:
1279                 vhost_user_set_log_base(dev, &msg);
1280
1281                 /* it needs a reply */
1282                 msg.size = sizeof(msg.payload.u64);
1283                 send_vhost_reply(fd, &msg);
1284                 break;
1285         case VHOST_USER_SET_LOG_FD:
1286                 close(msg.fds[0]);
1287                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1288                 break;
1289
1290         case VHOST_USER_SET_VRING_NUM:
1291                 vhost_user_set_vring_num(dev, &msg);
1292                 break;
1293         case VHOST_USER_SET_VRING_ADDR:
1294                 vhost_user_set_vring_addr(&dev, &msg);
1295                 break;
1296         case VHOST_USER_SET_VRING_BASE:
1297                 vhost_user_set_vring_base(dev, &msg);
1298                 break;
1299
1300         case VHOST_USER_GET_VRING_BASE:
1301                 vhost_user_get_vring_base(dev, &msg);
1302                 msg.size = sizeof(msg.payload.state);
1303                 send_vhost_reply(fd, &msg);
1304                 break;
1305
1306         case VHOST_USER_SET_VRING_KICK:
1307                 vhost_user_set_vring_kick(&dev, &msg);
1308                 break;
1309         case VHOST_USER_SET_VRING_CALL:
1310                 vhost_user_set_vring_call(dev, &msg);
1311                 break;
1312
1313         case VHOST_USER_SET_VRING_ERR:
1314                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1315                         close(msg.fds[0]);
1316                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1317                 break;
1318
1319         case VHOST_USER_GET_QUEUE_NUM:
1320                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1321                 msg.size = sizeof(msg.payload.u64);
1322                 send_vhost_reply(fd, &msg);
1323                 break;
1324
1325         case VHOST_USER_SET_VRING_ENABLE:
1326                 vhost_user_set_vring_enable(dev, &msg);
1327                 break;
1328         case VHOST_USER_SEND_RARP:
1329                 vhost_user_send_rarp(dev, &msg);
1330                 break;
1331
1332         case VHOST_USER_NET_SET_MTU:
1333                 ret = vhost_user_net_set_mtu(dev, &msg);
1334                 break;
1335
1336         case VHOST_USER_SET_SLAVE_REQ_FD:
1337                 ret = vhost_user_set_req_fd(dev, &msg);
1338                 break;
1339
1340         case VHOST_USER_IOTLB_MSG:
1341                 ret = vhost_user_iotlb_msg(&dev, &msg);
1342                 break;
1343
1344         default:
1345                 ret = -1;
1346                 break;
1347
1348         }
1349
1350         if (msg.flags & VHOST_USER_NEED_REPLY) {
1351                 msg.payload.u64 = !!ret;
1352                 msg.size = sizeof(msg.payload.u64);
1353                 send_vhost_reply(fd, &msg);
1354         }
1355
1356         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1357                 dev->flags |= VIRTIO_DEV_READY;
1358
1359                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1360                         if (dev->dequeue_zero_copy) {
1361                                 RTE_LOG(INFO, VHOST_CONFIG,
1362                                                 "dequeue zero copy is enabled\n");
1363                         }
1364
1365                         if (dev->notify_ops->new_device(dev->vid) == 0)
1366                                 dev->flags |= VIRTIO_DEV_RUNNING;
1367                 }
1368         }
1369
1370         return 0;
1371 }
1372
1373 int
1374 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1375 {
1376         int ret;
1377         struct VhostUserMsg msg = {
1378                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1379                 .flags = VHOST_USER_VERSION,
1380                 .size = sizeof(msg.payload.iotlb),
1381                 .payload.iotlb = {
1382                         .iova = iova,
1383                         .perm = perm,
1384                         .type = VHOST_IOTLB_MISS,
1385                 },
1386         };
1387
1388         ret = send_vhost_message(dev->slave_req_fd, &msg);
1389         if (ret < 0) {
1390                 RTE_LOG(ERR, VHOST_CONFIG,
1391                                 "Failed to send IOTLB miss message (%d)\n",
1392                                 ret);
1393                 return ret;
1394         }
1395
1396         return 0;
1397 }