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 int
 548 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 549 {
 550         struct VhostUserMemory memory = pmsg->payload.memory;
 551         struct rte_vhost_mem_region *reg;
 552         void *mmap_addr;
 553         uint64_t mmap_size;
 554         uint64_t mmap_offset;
 555         uint64_t alignment;
 556         uint32_t i;
 557         int fd;
 558
 559         if (dev->mem) {
 560                 free_mem_region(dev);
 561                 rte_free(dev->mem);
 562                 dev->mem = NULL;
 563         }
 564
 565         dev->nr_guest_pages = 0;
 566         if (!dev->guest_pages) {
 567                 dev->max_guest_pages = 8;
 568                 dev->guest_pages = malloc(dev->max_guest_pages *
 569                                                 sizeof(struct guest_page));
 570                 if (dev->guest_pages == NULL) {
 571                         RTE_LOG(ERR, VHOST_CONFIG,
 572                                 "(%d) failed to allocate memory "
 573                                 "for dev->guest_pages\n",
 574                                 dev->vid);
 575                         return -1;
 576                 }
 577         }
 578
 579         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 580                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 581         if (dev->mem == NULL) {
 582                 RTE_LOG(ERR, VHOST_CONFIG,
 583                         "(%d) failed to allocate memory for dev->mem\n",
 584                         dev->vid);
 585                 return -1;
 586         }
 587         dev->mem->nregions = memory.nregions;
 588
 589         for (i = 0; i < memory.nregions; i++) {
 590                 fd  = pmsg->fds[i];
 591                 reg = &dev->mem->regions[i];
 592
 593                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 594                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 595                 reg->size            = memory.regions[i].memory_size;
 596                 reg->fd              = fd;
 597
 598                 mmap_offset = memory.regions[i].mmap_offset;
 599                 mmap_size   = reg->size + mmap_offset;
 600
 601                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 602                  * with length argument aligned with hugepagesz at older
 603                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 604                  * mmap() will fail with EINVAL.
 605                  *
 606                  * to avoid failure, make sure in caller to keep length
 607                  * aligned.
 608                  */
 609                 alignment = get_blk_size(fd);
 610                 if (alignment == (uint64_t)-1) {
 611                         RTE_LOG(ERR, VHOST_CONFIG,
 612                                 "couldn't get hugepage size through fstat\n");
 613                         goto err_mmap;
 614                 }
 615                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 616
 617                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 618                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 619
 620                 if (mmap_addr == MAP_FAILED) {
 621                         RTE_LOG(ERR, VHOST_CONFIG,
 622                                 "mmap region %u failed.\n", i);
 623                         goto err_mmap;
 624                 }
 625
 626                 reg->mmap_addr = mmap_addr;
 627                 reg->mmap_size = mmap_size;
 628                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 629                                       mmap_offset;
 630
 631                 if (dev->dequeue_zero_copy)
 632                         add_guest_pages(dev, reg, alignment);
 633
 634                 RTE_LOG(INFO, VHOST_CONFIG,
 635                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 636                         "\t guest physical addr: 0x%" PRIx64 "\n"
 637                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 638                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 639                         "\t mmap addr : 0x%" PRIx64 "\n"
 640                         "\t mmap size : 0x%" PRIx64 "\n"
 641                         "\t mmap align: 0x%" PRIx64 "\n"
 642                         "\t mmap off  : 0x%" PRIx64 "\n",
 643                         i, reg->size,
 644                         reg->guest_phys_addr,
 645                         reg->guest_user_addr,
 646                         reg->host_user_addr,
 647                         (uint64_t)(uintptr_t)mmap_addr,
 648                         mmap_size,
 649                         alignment,
 650                         mmap_offset);
 651         }
 652
 653         dump_guest_pages(dev);
 654
 655         return 0;
 656
 657 err_mmap:
 658         free_mem_region(dev);
 659         rte_free(dev->mem);
 660         dev->mem = NULL;
 661         return -1;
 662 }
 663
 664 static int
 665 vq_is_ready(struct vhost_virtqueue *vq)
 666 {
 667         return vq && vq->desc && vq->avail && vq->used &&
 668                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 669                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 670 }
 671
 672 static int
 673 virtio_is_ready(struct virtio_net *dev)
 674 {
 675         struct vhost_virtqueue *vq;
 676         uint32_t i;
 677
 678         if (dev->nr_vring == 0)
 679                 return 0;
 680
 681         for (i = 0; i < dev->nr_vring; i++) {
 682                 vq = dev->virtqueue[i];
 683
 684                 if (!vq_is_ready(vq))
 685                         return 0;
 686         }
 687
 688         RTE_LOG(INFO, VHOST_CONFIG,
 689                 "virtio is now ready for processing.\n");
 690         return 1;
 691 }
 692
 693 static void
 694 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 695 {
 696         struct vhost_vring_file file;
 697         struct vhost_virtqueue *vq;
 698
 699         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 700         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 701                 file.fd = VIRTIO_INVALID_EVENTFD;
 702         else
 703                 file.fd = pmsg->fds[0];
 704         RTE_LOG(INFO, VHOST_CONFIG,
 705                 "vring call idx:%d file:%d\n", file.index, file.fd);
 706
 707         vq = dev->virtqueue[file.index];
 708         if (vq->callfd >= 0)
 709                 close(vq->callfd);
 710
 711         vq->callfd = file.fd;
 712 }
 713
 714 static void
 715 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 716 {
 717         struct vhost_vring_file file;
 718         struct vhost_virtqueue *vq;
 719         struct virtio_net *dev = *pdev;
 720
 721         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 722         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 723                 file.fd = VIRTIO_INVALID_EVENTFD;
 724         else
 725                 file.fd = pmsg->fds[0];
 726         RTE_LOG(INFO, VHOST_CONFIG,
 727                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 728
 729         /* Interpret ring addresses only when ring is started. */
 730         dev = translate_ring_addresses(dev, file.index);
 731         if (!dev)
 732                 return;
 733
 734         *pdev = dev;
 735
 736         vq = dev->virtqueue[file.index];
 737
 738         /*
 739          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 740          * the ring starts already enabled. Otherwise, it is enabled via
 741          * the SET_VRING_ENABLE message.
 742          */
 743         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 744                 vq->enabled = 1;
 745
 746         if (vq->kickfd >= 0)
 747                 close(vq->kickfd);
 748         vq->kickfd = file.fd;
 749 }
 750
 751 static void
 752 free_zmbufs(struct vhost_virtqueue *vq)
 753 {
 754         struct zcopy_mbuf *zmbuf, *next;
 755
 756         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 757              zmbuf != NULL; zmbuf = next) {
 758                 next = TAILQ_NEXT(zmbuf, next);
 759
 760                 rte_pktmbuf_free(zmbuf->mbuf);
 761                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 762         }
 763
 764         rte_free(vq->zmbufs);
 765 }
 766
 767 /*
 768  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 769  */
 770 static int
 771 vhost_user_get_vring_base(struct virtio_net *dev,
 772                           VhostUserMsg *msg)
 773 {
 774         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 775
 776         /* We have to stop the queue (virtio) if it is running. */
 777         if (dev->flags & VIRTIO_DEV_RUNNING) {
 778                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 779                 dev->notify_ops->destroy_device(dev->vid);
 780         }
 781
 782         dev->flags &= ~VIRTIO_DEV_READY;
 783
 784         /* Here we are safe to get the last used index */
 785         msg->payload.state.num = vq->last_used_idx;
 786
 787         RTE_LOG(INFO, VHOST_CONFIG,
 788                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 789                 msg->payload.state.num);
 790         /*
 791          * Based on current qemu vhost-user implementation, this message is
 792          * sent and only sent in vhost_vring_stop.
 793          * TODO: cleanup the vring, it isn't usable since here.
 794          */
 795         if (vq->kickfd >= 0)
 796                 close(vq->kickfd);
 797
 798         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 799
 800         if (dev->dequeue_zero_copy)
 801                 free_zmbufs(vq);
 802         rte_free(vq->shadow_used_ring);
 803         vq->shadow_used_ring = NULL;
 804
 805         rte_free(vq->batch_copy_elems);
 806         vq->batch_copy_elems = NULL;
 807
 808         return 0;
 809 }
 810
 811 /*
 812  * when virtio queues are ready to work, qemu will send us to
 813  * enable the virtio queue pair.
 814  */
 815 static int
 816 vhost_user_set_vring_enable(struct virtio_net *dev,
 817                             VhostUserMsg *msg)
 818 {
 819         int enable = (int)msg->payload.state.num;
 820
 821         RTE_LOG(INFO, VHOST_CONFIG,
 822                 "set queue enable: %d to qp idx: %d\n",
 823                 enable, msg->payload.state.index);
 824
 825         if (dev->notify_ops->vring_state_changed)
 826                 dev->notify_ops->vring_state_changed(dev->vid,
 827                                 msg->payload.state.index, enable);
 828
 829         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 830
 831         return 0;
 832 }
 833
 834 static void
 835 vhost_user_get_protocol_features(struct virtio_net *dev,
 836                                  struct VhostUserMsg *msg)
 837 {
 838         uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
 839
 840         rte_vhost_driver_get_features(dev->ifname, &features);
 841
 842         /*
 843          * REPLY_ACK protocol feature is only mandatory for now
 844          * for IOMMU feature. If IOMMU is explicitly disabled by the
 845          * application, disable also REPLY_ACK feature for older buggy
 846          * Qemu versions (from v2.7.0 to v2.9.0).
 847          */
 848         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 849                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
 850
 851         msg->payload.u64 = protocol_features;
 852         msg->size = sizeof(msg->payload.u64);
 853 }
 854
 855 static void
 856 vhost_user_set_protocol_features(struct virtio_net *dev,
 857                                  uint64_t protocol_features)
 858 {
 859         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 860                 return;
 861
 862         dev->protocol_features = protocol_features;
 863 }
 864
 865 static int
 866 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 867 {
 868         int fd = msg->fds[0];
 869         uint64_t size, off;
 870         void *addr;
 871
 872         if (fd < 0) {
 873                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 874                 return -1;
 875         }
 876
 877         if (msg->size != sizeof(VhostUserLog)) {
 878                 RTE_LOG(ERR, VHOST_CONFIG,
 879                         "invalid log base msg size: %"PRId32" != %d\n",
 880                         msg->size, (int)sizeof(VhostUserLog));
 881                 return -1;
 882         }
 883
 884         size = msg->payload.log.mmap_size;
 885         off  = msg->payload.log.mmap_offset;
 886         RTE_LOG(INFO, VHOST_CONFIG,
 887                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 888                 size, off);
 889
 890         /*
 891          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 892          * fail when offset is not page size aligned.
 893          */
 894         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 895         close(fd);
 896         if (addr == MAP_FAILED) {
 897                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 898                 return -1;
 899         }
 900
 901         /*
 902          * Free previously mapped log memory on occasionally
 903          * multiple VHOST_USER_SET_LOG_BASE.
 904          */
 905         if (dev->log_addr) {
 906                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 907         }
 908         dev->log_addr = (uint64_t)(uintptr_t)addr;
 909         dev->log_base = dev->log_addr + off;
 910         dev->log_size = size;
 911
 912         return 0;
 913 }
 914
 915 /*
 916  * An rarp packet is constructed and broadcasted to notify switches about
 917  * the new location of the migrated VM, so that packets from outside will
 918  * not be lost after migration.
 919  *
 920  * However, we don't actually "send" a rarp packet here, instead, we set
 921  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 922  */
 923 static int
 924 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 925 {
 926         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 927
 928         RTE_LOG(DEBUG, VHOST_CONFIG,
 929                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 930                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 931         memcpy(dev->mac.addr_bytes, mac, 6);
 932
 933         /*
 934          * Set the flag to inject a RARP broadcast packet at
 935          * rte_vhost_dequeue_burst().
 936          *
 937          * rte_smp_wmb() is for making sure the mac is copied
 938          * before the flag is set.
 939          */
 940         rte_smp_wmb();
 941         rte_atomic16_set(&dev->broadcast_rarp, 1);
 942
 943         return 0;
 944 }
 945
 946 static int
 947 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 948 {
 949         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 950                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 951                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 952                                 msg->payload.u64);
 953
 954                 return -1;
 955         }
 956
 957         dev->mtu = msg->payload.u64;
 958
 959         return 0;
 960 }
 961
 962 static int
 963 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 964 {
 965         int fd = msg->fds[0];
 966
 967         if (fd < 0) {
 968                 RTE_LOG(ERR, VHOST_CONFIG,
 969                                 "Invalid file descriptor for slave channel (%d)\n",
 970                                 fd);
 971                 return -1;
 972         }
 973
 974         dev->slave_req_fd = fd;
 975
 976         return 0;
 977 }
 978
 979 static int
 980 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
 981 {
 982         struct vhost_vring_addr *ra;
 983         uint64_t start, end;
 984
 985         start = imsg->iova;
 986         end = start + imsg->size;
 987
 988         ra = &vq->ring_addrs;
 989         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
 990                 return 1;
 991         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
 992                 return 1;
 993         if (ra->used_user_addr >= start && ra->used_user_addr < end)
 994                 return 1;
 995
 996         return 0;
 997 }
 998
 999 static int
1000 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1001                                 struct vhost_iotlb_msg *imsg)
1002 {
1003         uint64_t istart, iend, vstart, vend;
1004
1005         istart = imsg->iova;
1006         iend = istart + imsg->size - 1;
1007
1008         vstart = (uintptr_t)vq->desc;
1009         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1010         if (vstart <= iend && istart <= vend)
1011                 return 1;
1012
1013         vstart = (uintptr_t)vq->avail;
1014         vend = vstart + sizeof(struct vring_avail);
1015         vend += sizeof(uint16_t) * vq->size - 1;
1016         if (vstart <= iend && istart <= vend)
1017                 return 1;
1018
1019         vstart = (uintptr_t)vq->used;
1020         vend = vstart + sizeof(struct vring_used);
1021         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1022         if (vstart <= iend && istart <= vend)
1023                 return 1;
1024
1025         return 0;
1026 }
1027
1028 static int
1029 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1030 {
1031         struct virtio_net *dev = *pdev;
1032         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1033         uint16_t i;
1034         uint64_t vva;
1035
1036         switch (imsg->type) {
1037         case VHOST_IOTLB_UPDATE:
1038                 vva = qva_to_vva(dev, imsg->uaddr);
1039                 if (!vva)
1040                         return -1;
1041
1042                 for (i = 0; i < dev->nr_vring; i++) {
1043                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1044
1045                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1046                                         imsg->size, imsg->perm);
1047
1048                         if (is_vring_iotlb_update(vq, imsg))
1049                                 *pdev = dev = translate_ring_addresses(dev, i);
1050                 }
1051                 break;
1052         case VHOST_IOTLB_INVALIDATE:
1053                 for (i = 0; i < dev->nr_vring; i++) {
1054                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1055
1056                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1057                                         imsg->size);
1058
1059                         if (is_vring_iotlb_invalidate(vq, imsg))
1060                                 vring_invalidate(dev, vq);
1061                 }
1062                 break;
1063         default:
1064                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1065                                 imsg->type);
1066                 return -1;
1067         }
1068
1069         return 0;
1070 }
1071
1072 /* return bytes# of read on success or negative val on failure. */
1073 static int
1074 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1075 {
1076         int ret;
1077
1078         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1079                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1080         if (ret <= 0)
1081                 return ret;
1082
1083         if (msg && msg->size) {
1084                 if (msg->size > sizeof(msg->payload)) {
1085                         RTE_LOG(ERR, VHOST_CONFIG,
1086                                 "invalid msg size: %d\n", msg->size);
1087                         return -1;
1088                 }
1089                 ret = read(sockfd, &msg->payload, msg->size);
1090                 if (ret <= 0)
1091                         return ret;
1092                 if (ret != (int)msg->size) {
1093                         RTE_LOG(ERR, VHOST_CONFIG,
1094                                 "read control message failed\n");
1095                         return -1;
1096                 }
1097         }
1098
1099         return ret;
1100 }
1101
1102 static int
1103 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1104 {
1105         if (!msg)
1106                 return 0;
1107
1108         return send_fd_message(sockfd, (char *)msg,
1109                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1110 }
1111
1112 static int
1113 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1114 {
1115         if (!msg)
1116                 return 0;
1117
1118         msg->flags &= ~VHOST_USER_VERSION_MASK;
1119         msg->flags &= ~VHOST_USER_NEED_REPLY;
1120         msg->flags |= VHOST_USER_VERSION;
1121         msg->flags |= VHOST_USER_REPLY_MASK;
1122
1123         return send_vhost_message(sockfd, msg);
1124 }
1125
1126 /*
1127  * Allocate a queue pair if it hasn't been allocated yet
1128  */
1129 static int
1130 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1131 {
1132         uint16_t vring_idx;
1133
1134         switch (msg->request.master) {
1135         case VHOST_USER_SET_VRING_KICK:
1136         case VHOST_USER_SET_VRING_CALL:
1137         case VHOST_USER_SET_VRING_ERR:
1138                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1139                 break;
1140         case VHOST_USER_SET_VRING_NUM:
1141         case VHOST_USER_SET_VRING_BASE:
1142         case VHOST_USER_SET_VRING_ENABLE:
1143                 vring_idx = msg->payload.state.index;
1144                 break;
1145         case VHOST_USER_SET_VRING_ADDR:
1146                 vring_idx = msg->payload.addr.index;
1147                 break;
1148         default:
1149                 return 0;
1150         }
1151
1152         if (vring_idx >= VHOST_MAX_VRING) {
1153                 RTE_LOG(ERR, VHOST_CONFIG,
1154                         "invalid vring index: %u\n", vring_idx);
1155                 return -1;
1156         }
1157
1158         if (dev->virtqueue[vring_idx])
1159                 return 0;
1160
1161         return alloc_vring_queue(dev, vring_idx);
1162 }
1163
1164 int
1165 vhost_user_msg_handler(int vid, int fd)
1166 {
1167         struct virtio_net *dev;
1168         struct VhostUserMsg msg;
1169         int ret;
1170
1171         dev = get_device(vid);
1172         if (dev == NULL)
1173                 return -1;
1174
1175         if (!dev->notify_ops) {
1176                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1177                 if (!dev->notify_ops) {
1178                         RTE_LOG(ERR, VHOST_CONFIG,
1179                                 "failed to get callback ops for driver %s\n",
1180                                 dev->ifname);
1181                         return -1;
1182                 }
1183         }
1184
1185         ret = read_vhost_message(fd, &msg);
1186         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1187                 if (ret < 0)
1188                         RTE_LOG(ERR, VHOST_CONFIG,
1189                                 "vhost read message failed\n");
1190                 else if (ret == 0)
1191                         RTE_LOG(INFO, VHOST_CONFIG,
1192                                 "vhost peer closed\n");
1193                 else
1194                         RTE_LOG(ERR, VHOST_CONFIG,
1195                                 "vhost read incorrect message\n");
1196
1197                 return -1;
1198         }
1199
1200         ret = 0;
1201         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1202                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1203                         vhost_message_str[msg.request.master]);
1204         else
1205                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1206                         vhost_message_str[msg.request.master]);
1207
1208         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1209         if (ret < 0) {
1210                 RTE_LOG(ERR, VHOST_CONFIG,
1211                         "failed to alloc queue\n");
1212                 return -1;
1213         }
1214
1215         switch (msg.request.master) {
1216         case VHOST_USER_GET_FEATURES:
1217                 msg.payload.u64 = vhost_user_get_features(dev);
1218                 msg.size = sizeof(msg.payload.u64);
1219                 send_vhost_reply(fd, &msg);
1220                 break;
1221         case VHOST_USER_SET_FEATURES:
1222                 vhost_user_set_features(dev, msg.payload.u64);
1223                 break;
1224
1225         case VHOST_USER_GET_PROTOCOL_FEATURES:
1226                 vhost_user_get_protocol_features(dev, &msg);
1227                 send_vhost_reply(fd, &msg);
1228                 break;
1229         case VHOST_USER_SET_PROTOCOL_FEATURES:
1230                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1231                 break;
1232
1233         case VHOST_USER_SET_OWNER:
1234                 vhost_user_set_owner();
1235                 break;
1236         case VHOST_USER_RESET_OWNER:
1237                 vhost_user_reset_owner(dev);
1238                 break;
1239
1240         case VHOST_USER_SET_MEM_TABLE:
1241                 ret = vhost_user_set_mem_table(dev, &msg);
1242                 break;
1243
1244         case VHOST_USER_SET_LOG_BASE:
1245                 vhost_user_set_log_base(dev, &msg);
1246
1247                 /* it needs a reply */
1248                 msg.size = sizeof(msg.payload.u64);
1249                 send_vhost_reply(fd, &msg);
1250                 break;
1251         case VHOST_USER_SET_LOG_FD:
1252                 close(msg.fds[0]);
1253                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1254                 break;
1255
1256         case VHOST_USER_SET_VRING_NUM:
1257                 vhost_user_set_vring_num(dev, &msg);
1258                 break;
1259         case VHOST_USER_SET_VRING_ADDR:
1260                 vhost_user_set_vring_addr(&dev, &msg);
1261                 break;
1262         case VHOST_USER_SET_VRING_BASE:
1263                 vhost_user_set_vring_base(dev, &msg);
1264                 break;
1265
1266         case VHOST_USER_GET_VRING_BASE:
1267                 vhost_user_get_vring_base(dev, &msg);
1268                 msg.size = sizeof(msg.payload.state);
1269                 send_vhost_reply(fd, &msg);
1270                 break;
1271
1272         case VHOST_USER_SET_VRING_KICK:
1273                 vhost_user_set_vring_kick(&dev, &msg);
1274                 break;
1275         case VHOST_USER_SET_VRING_CALL:
1276                 vhost_user_set_vring_call(dev, &msg);
1277                 break;
1278
1279         case VHOST_USER_SET_VRING_ERR:
1280                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1281                         close(msg.fds[0]);
1282                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1283                 break;
1284
1285         case VHOST_USER_GET_QUEUE_NUM:
1286                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1287                 msg.size = sizeof(msg.payload.u64);
1288                 send_vhost_reply(fd, &msg);
1289                 break;
1290
1291         case VHOST_USER_SET_VRING_ENABLE:
1292                 vhost_user_set_vring_enable(dev, &msg);
1293                 break;
1294         case VHOST_USER_SEND_RARP:
1295                 vhost_user_send_rarp(dev, &msg);
1296                 break;
1297
1298         case VHOST_USER_NET_SET_MTU:
1299                 ret = vhost_user_net_set_mtu(dev, &msg);
1300                 break;
1301
1302         case VHOST_USER_SET_SLAVE_REQ_FD:
1303                 ret = vhost_user_set_req_fd(dev, &msg);
1304                 break;
1305
1306         case VHOST_USER_IOTLB_MSG:
1307                 ret = vhost_user_iotlb_msg(&dev, &msg);
1308                 break;
1309
1310         default:
1311                 ret = -1;
1312                 break;
1313
1314         }
1315
1316         if (msg.flags & VHOST_USER_NEED_REPLY) {
1317                 msg.payload.u64 = !!ret;
1318                 msg.size = sizeof(msg.payload.u64);
1319                 send_vhost_reply(fd, &msg);
1320         }
1321
1322         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1323                 dev->flags |= VIRTIO_DEV_READY;
1324
1325                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1326                         if (dev->dequeue_zero_copy) {
1327                                 RTE_LOG(INFO, VHOST_CONFIG,
1328                                                 "dequeue zero copy is enabled\n");
1329                         }
1330
1331                         if (dev->notify_ops->new_device(dev->vid) == 0)
1332                                 dev->flags |= VIRTIO_DEV_RUNNING;
1333                 }
1334         }
1335
1336         return 0;
1337 }
1338
1339 int
1340 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1341 {
1342         int ret;
1343         struct VhostUserMsg msg = {
1344                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1345                 .flags = VHOST_USER_VERSION,
1346                 .size = sizeof(msg.payload.iotlb),
1347                 .payload.iotlb = {
1348                         .iova = iova,
1349                         .perm = perm,
1350                         .type = VHOST_IOTLB_MISS,
1351                 },
1352         };
1353
1354         ret = send_vhost_message(dev->slave_req_fd, &msg);
1355         if (ret < 0) {
1356                 RTE_LOG(ERR, VHOST_CONFIG,
1357                                 "Failed to send IOTLB miss message (%d)\n",
1358                                 ret);
1359                 return ret;
1360         }
1361
1362         return 0;
1363 }