lib/librte_vhost/vhost_user.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <string.h>
  38 #include <unistd.h>
  39 #include <sys/mman.h>
  40 #include <sys/types.h>
  41 #include <sys/stat.h>
  42 #include <assert.h>
  43 #ifdef RTE_LIBRTE_VHOST_NUMA
  44 #include <numaif.h>
  45 #endif
  46
  47 #include <rte_common.h>
  48 #include <rte_malloc.h>
  49 #include <rte_log.h>
  50
  51 #include "iotlb.h"
  52 #include "vhost.h"
  53 #include "vhost_user.h"
  54
  55 #define VIRTIO_MIN_MTU 68
  56 #define VIRTIO_MAX_MTU 65535
  57
  58 static const char *vhost_message_str[VHOST_USER_MAX] = {
  59         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  60         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  61         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  62         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  63         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  64         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  65         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  66         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  67         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  68         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  69         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  70         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  71         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  72         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  73         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  74         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  75         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  76         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  77         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  78         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  79         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  80         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  81         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  82 };
  83
  84 static uint64_t
  85 get_blk_size(int fd)
  86 {
  87         struct stat stat;
  88         int ret;
  89
  90         ret = fstat(fd, &stat);
  91         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  92 }
  93
  94 static void
  95 free_mem_region(struct virtio_net *dev)
  96 {
  97         uint32_t i;
  98         struct rte_vhost_mem_region *reg;
  99
 100         if (!dev || !dev->mem)
 101                 return;
 102
 103         for (i = 0; i < dev->mem->nregions; i++) {
 104                 reg = &dev->mem->regions[i];
 105                 if (reg->host_user_addr) {
 106                         munmap(reg->mmap_addr, reg->mmap_size);
 107                         close(reg->fd);
 108                 }
 109         }
 110 }
 111
 112 void
 113 vhost_backend_cleanup(struct virtio_net *dev)
 114 {
 115         if (dev->mem) {
 116                 free_mem_region(dev);
 117                 rte_free(dev->mem);
 118                 dev->mem = NULL;
 119         }
 120
 121         free(dev->guest_pages);
 122         dev->guest_pages = NULL;
 123
 124         if (dev->log_addr) {
 125                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 126                 dev->log_addr = 0;
 127         }
 128
 129         if (dev->slave_req_fd >= 0) {
 130                 close(dev->slave_req_fd);
 131                 dev->slave_req_fd = -1;
 132         }
 133 }
 134
 135 /*
 136  * This function just returns success at the moment unless
 137  * the device hasn't been initialised.
 138  */
 139 static int
 140 vhost_user_set_owner(void)
 141 {
 142         return 0;
 143 }
 144
 145 static int
 146 vhost_user_reset_owner(struct virtio_net *dev)
 147 {
 148         if (dev->flags & VIRTIO_DEV_RUNNING) {
 149                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 150                 dev->notify_ops->destroy_device(dev->vid);
 151         }
 152
 153         cleanup_device(dev, 0);
 154         reset_device(dev);
 155         return 0;
 156 }
 157
 158 /*
 159  * The features that we support are requested.
 160  */
 161 static uint64_t
 162 vhost_user_get_features(struct virtio_net *dev)
 163 {
 164         uint64_t features = 0;
 165
 166         rte_vhost_driver_get_features(dev->ifname, &features);
 167         return features;
 168 }
 169
 170 /*
 171  * We receive the negotiated features supported by us and the virtio device.
 172  */
 173 static int
 174 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 175 {
 176         uint64_t vhost_features = 0;
 177
 178         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 179         if (features & ~vhost_features) {
 180                 RTE_LOG(ERR, VHOST_CONFIG,
 181                         "(%d) received invalid negotiated features.\n",
 182                         dev->vid);
 183                 return -1;
 184         }
 185
 186         if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
 187                 if (dev->notify_ops->features_changed)
 188                         dev->notify_ops->features_changed(dev->vid, features);
 189         }
 190
 191         dev->features = features;
 192         if (dev->features &
 193                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 194                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 195         } else {
 196                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 197         }
 198         LOG_DEBUG(VHOST_CONFIG,
 199                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 200                 dev->vid,
 201                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 202                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 203
 204         return 0;
 205 }
 206
 207 /*
 208  * The virtio device sends us the size of the descriptor ring.
 209  */
 210 static int
 211 vhost_user_set_vring_num(struct virtio_net *dev,
 212                          VhostUserMsg *msg)
 213 {
 214         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 215
 216         vq->size = msg->payload.state.num;
 217
 218         if (dev->dequeue_zero_copy) {
 219                 vq->nr_zmbuf = 0;
 220                 vq->last_zmbuf_idx = 0;
 221                 vq->zmbuf_size = vq->size;
 222                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 223                                          sizeof(struct zcopy_mbuf), 0);
 224                 if (vq->zmbufs == NULL) {
 225                         RTE_LOG(WARNING, VHOST_CONFIG,
 226                                 "failed to allocate mem for zero copy; "
 227                                 "zero copy is force disabled\n");
 228                         dev->dequeue_zero_copy = 0;
 229                 }
 230         }
 231
 232         vq->shadow_used_ring = rte_malloc(NULL,
 233                                 vq->size * sizeof(struct vring_used_elem),
 234                                 RTE_CACHE_LINE_SIZE);
 235         if (!vq->shadow_used_ring) {
 236                 RTE_LOG(ERR, VHOST_CONFIG,
 237                         "failed to allocate memory for shadow used ring.\n");
 238                 return -1;
 239         }
 240
 241         vq->batch_copy_elems = rte_malloc(NULL,
 242                                 vq->size * sizeof(struct batch_copy_elem),
 243                                 RTE_CACHE_LINE_SIZE);
 244         if (!vq->batch_copy_elems) {
 245                 RTE_LOG(ERR, VHOST_CONFIG,
 246                         "failed to allocate memory for batching copy.\n");
 247                 return -1;
 248         }
 249
 250         return 0;
 251 }
 252
 253 /*
 254  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 255  * same numa node as the memory of vring descriptor.
 256  */
 257 #ifdef RTE_LIBRTE_VHOST_NUMA
 258 static struct virtio_net*
 259 numa_realloc(struct virtio_net *dev, int index)
 260 {
 261         int oldnode, newnode;
 262         struct virtio_net *old_dev;
 263         struct vhost_virtqueue *old_vq, *vq;
 264         int ret;
 265
 266         old_dev = dev;
 267         vq = old_vq = dev->virtqueue[index];
 268
 269         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 270                             MPOL_F_NODE | MPOL_F_ADDR);
 271
 272         /* check if we need to reallocate vq */
 273         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 274                              MPOL_F_NODE | MPOL_F_ADDR);
 275         if (ret) {
 276                 RTE_LOG(ERR, VHOST_CONFIG,
 277                         "Unable to get vq numa information.\n");
 278                 return dev;
 279         }
 280         if (oldnode != newnode) {
 281                 RTE_LOG(INFO, VHOST_CONFIG,
 282                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 283                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 284                 if (!vq)
 285                         return dev;
 286
 287                 memcpy(vq, old_vq, sizeof(*vq));
 288                 rte_free(old_vq);
 289         }
 290
 291         /* check if we need to reallocate dev */
 292         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 293                             MPOL_F_NODE | MPOL_F_ADDR);
 294         if (ret) {
 295                 RTE_LOG(ERR, VHOST_CONFIG,
 296                         "Unable to get dev numa information.\n");
 297                 goto out;
 298         }
 299         if (oldnode != newnode) {
 300                 RTE_LOG(INFO, VHOST_CONFIG,
 301                         "reallocate dev from %d to %d node\n",
 302                         oldnode, newnode);
 303                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 304                 if (!dev) {
 305                         dev = old_dev;
 306                         goto out;
 307                 }
 308
 309                 memcpy(dev, old_dev, sizeof(*dev));
 310                 rte_free(old_dev);
 311         }
 312
 313 out:
 314         dev->virtqueue[index] = vq;
 315         vhost_devices[dev->vid] = dev;
 316
 317         return dev;
 318 }
 319 #else
 320 static struct virtio_net*
 321 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 322 {
 323         return dev;
 324 }
 325 #endif
 326
 327 /* Converts QEMU virtual address to Vhost virtual address. */
 328 static uint64_t
 329 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 330 {
 331         struct rte_vhost_mem_region *reg;
 332         uint32_t i;
 333
 334         /* Find the region where the address lives. */
 335         for (i = 0; i < dev->mem->nregions; i++) {
 336                 reg = &dev->mem->regions[i];
 337
 338                 if (qva >= reg->guest_user_addr &&
 339                     qva <  reg->guest_user_addr + reg->size) {
 340                         return qva - reg->guest_user_addr +
 341                                reg->host_user_addr;
 342                 }
 343         }
 344
 345         return 0;
 346 }
 347
 348
 349 /*
 350  * Converts ring address to Vhost virtual address.
 351  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 352  * else it is a QEMU virtual address.
 353  */
 354 static uint64_t
 355 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 356                 uint64_t ra, uint64_t size)
 357 {
 358         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 359                 uint64_t vva;
 360
 361                 vva = vhost_user_iotlb_cache_find(vq, ra,
 362                                         &size, VHOST_ACCESS_RW);
 363                 if (!vva)
 364                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 365
 366                 return vva;
 367         }
 368
 369         return qva_to_vva(dev, ra);
 370 }
 371
 372 /*
 373  * The virtio device sends us the desc, used and avail ring addresses.
 374  * This function then converts these to our address space.
 375  */
 376 static int
 377 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
 378 {
 379         struct vhost_virtqueue *vq;
 380         struct vhost_vring_addr *addr = &msg->payload.addr;
 381
 382         if (dev->mem == NULL)
 383                 return -1;
 384
 385         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 386         vq = dev->virtqueue[msg->payload.addr.index];
 387
 388         /*
 389          * Rings addresses should not be interpreted as long as the ring is not
 390          * started and enabled
 391          */
 392         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 393
 394         return 0;
 395 }
 396
 397 static struct virtio_net *
 398 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 399 {
 400         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 401         struct vhost_vring_addr *addr = &vq->ring_addrs;
 402
 403         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 404         if (vq->desc && vq->avail && vq->used)
 405                 return dev;
 406
 407         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 408                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 409         if (vq->desc == 0) {
 410                 RTE_LOG(ERR, VHOST_CONFIG,
 411                         "(%d) failed to find desc ring address.\n",
 412                         dev->vid);
 413                 return NULL;
 414         }
 415
 416         dev = numa_realloc(dev, vq_index);
 417         vq = dev->virtqueue[vq_index];
 418
 419         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 420                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 421         if (vq->avail == 0) {
 422                 RTE_LOG(ERR, VHOST_CONFIG,
 423                         "(%d) failed to find avail ring address.\n",
 424                         dev->vid);
 425                 return NULL;
 426         }
 427
 428         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 429                         vq, addr->used_user_addr, sizeof(struct vring_used));
 430         if (vq->used == 0) {
 431                 RTE_LOG(ERR, VHOST_CONFIG,
 432                         "(%d) failed to find used ring address.\n",
 433                         dev->vid);
 434                 return NULL;
 435         }
 436
 437         if (vq->last_used_idx != vq->used->idx) {
 438                 RTE_LOG(WARNING, VHOST_CONFIG,
 439                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 440                         "some packets maybe resent for Tx and dropped for Rx\n",
 441                         vq->last_used_idx, vq->used->idx);
 442                 vq->last_used_idx  = vq->used->idx;
 443                 vq->last_avail_idx = vq->used->idx;
 444         }
 445
 446         vq->log_guest_addr = addr->log_guest_addr;
 447
 448         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 449                         dev->vid, vq->desc);
 450         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 451                         dev->vid, vq->avail);
 452         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 453                         dev->vid, vq->used);
 454         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 455                         dev->vid, vq->log_guest_addr);
 456
 457         return dev;
 458 }
 459
 460 /*
 461  * The virtio device sends us the available ring last used index.
 462  */
 463 static int
 464 vhost_user_set_vring_base(struct virtio_net *dev,
 465                           VhostUserMsg *msg)
 466 {
 467         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 468                         msg->payload.state.num;
 469         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 470                         msg->payload.state.num;
 471
 472         return 0;
 473 }
 474
 475 static void
 476 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 477                    uint64_t host_phys_addr, uint64_t size)
 478 {
 479         struct guest_page *page, *last_page;
 480
 481         if (dev->nr_guest_pages == dev->max_guest_pages) {
 482                 dev->max_guest_pages *= 2;
 483                 dev->guest_pages = realloc(dev->guest_pages,
 484                                         dev->max_guest_pages * sizeof(*page));
 485         }
 486
 487         if (dev->nr_guest_pages > 0) {
 488                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 489                 /* merge if the two pages are continuous */
 490                 if (host_phys_addr == last_page->host_phys_addr +
 491                                       last_page->size) {
 492                         last_page->size += size;
 493                         return;
 494                 }
 495         }
 496
 497         page = &dev->guest_pages[dev->nr_guest_pages++];
 498         page->guest_phys_addr = guest_phys_addr;
 499         page->host_phys_addr  = host_phys_addr;
 500         page->size = size;
 501 }
 502
 503 static void
 504 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 505                 uint64_t page_size)
 506 {
 507         uint64_t reg_size = reg->size;
 508         uint64_t host_user_addr  = reg->host_user_addr;
 509         uint64_t guest_phys_addr = reg->guest_phys_addr;
 510         uint64_t host_phys_addr;
 511         uint64_t size;
 512
 513         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 514         size = page_size - (guest_phys_addr & (page_size - 1));
 515         size = RTE_MIN(size, reg_size);
 516
 517         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 518         host_user_addr  += size;
 519         guest_phys_addr += size;
 520         reg_size -= size;
 521
 522         while (reg_size > 0) {
 523                 size = RTE_MIN(reg_size, page_size);
 524                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 525                                                   host_user_addr);
 526                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 527
 528                 host_user_addr  += size;
 529                 guest_phys_addr += size;
 530                 reg_size -= size;
 531         }
 532 }
 533
 534 #ifdef RTE_LIBRTE_VHOST_DEBUG
 535 /* TODO: enable it only in debug mode? */
 536 static void
 537 dump_guest_pages(struct virtio_net *dev)
 538 {
 539         uint32_t i;
 540         struct guest_page *page;
 541
 542         for (i = 0; i < dev->nr_guest_pages; i++) {
 543                 page = &dev->guest_pages[i];
 544
 545                 RTE_LOG(INFO, VHOST_CONFIG,
 546                         "guest physical page region %u\n"
 547                         "\t guest_phys_addr: %" PRIx64 "\n"
 548                         "\t host_phys_addr : %" PRIx64 "\n"
 549                         "\t size           : %" PRIx64 "\n",
 550                         i,
 551                         page->guest_phys_addr,
 552                         page->host_phys_addr,
 553                         page->size);
 554         }
 555 }
 556 #else
 557 #define dump_guest_pages(dev)
 558 #endif
 559
 560 static int
 561 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 562 {
 563         struct VhostUserMemory memory = pmsg->payload.memory;
 564         struct rte_vhost_mem_region *reg;
 565         void *mmap_addr;
 566         uint64_t mmap_size;
 567         uint64_t mmap_offset;
 568         uint64_t alignment;
 569         uint32_t i;
 570         int fd;
 571
 572         if (dev->mem) {
 573                 free_mem_region(dev);
 574                 rte_free(dev->mem);
 575                 dev->mem = NULL;
 576         }
 577
 578         dev->nr_guest_pages = 0;
 579         if (!dev->guest_pages) {
 580                 dev->max_guest_pages = 8;
 581                 dev->guest_pages = malloc(dev->max_guest_pages *
 582                                                 sizeof(struct guest_page));
 583                 if (dev->guest_pages == NULL) {
 584                         RTE_LOG(ERR, VHOST_CONFIG,
 585                                 "(%d) failed to allocate memory "
 586                                 "for dev->guest_pages\n",
 587                                 dev->vid);
 588                         return -1;
 589                 }
 590         }
 591
 592         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 593                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 594         if (dev->mem == NULL) {
 595                 RTE_LOG(ERR, VHOST_CONFIG,
 596                         "(%d) failed to allocate memory for dev->mem\n",
 597                         dev->vid);
 598                 return -1;
 599         }
 600         dev->mem->nregions = memory.nregions;
 601
 602         for (i = 0; i < memory.nregions; i++) {
 603                 fd  = pmsg->fds[i];
 604                 reg = &dev->mem->regions[i];
 605
 606                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 607                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 608                 reg->size            = memory.regions[i].memory_size;
 609                 reg->fd              = fd;
 610
 611                 mmap_offset = memory.regions[i].mmap_offset;
 612                 mmap_size   = reg->size + mmap_offset;
 613
 614                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 615                  * with length argument aligned with hugepagesz at older
 616                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 617                  * mmap() will fail with EINVAL.
 618                  *
 619                  * to avoid failure, make sure in caller to keep length
 620                  * aligned.
 621                  */
 622                 alignment = get_blk_size(fd);
 623                 if (alignment == (uint64_t)-1) {
 624                         RTE_LOG(ERR, VHOST_CONFIG,
 625                                 "couldn't get hugepage size through fstat\n");
 626                         goto err_mmap;
 627                 }
 628                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 629
 630                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 631                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 632
 633                 if (mmap_addr == MAP_FAILED) {
 634                         RTE_LOG(ERR, VHOST_CONFIG,
 635                                 "mmap region %u failed.\n", i);
 636                         goto err_mmap;
 637                 }
 638
 639                 reg->mmap_addr = mmap_addr;
 640                 reg->mmap_size = mmap_size;
 641                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 642                                       mmap_offset;
 643
 644                 if (dev->dequeue_zero_copy)
 645                         add_guest_pages(dev, reg, alignment);
 646
 647                 RTE_LOG(INFO, VHOST_CONFIG,
 648                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 649                         "\t guest physical addr: 0x%" PRIx64 "\n"
 650                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 651                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 652                         "\t mmap addr : 0x%" PRIx64 "\n"
 653                         "\t mmap size : 0x%" PRIx64 "\n"
 654                         "\t mmap align: 0x%" PRIx64 "\n"
 655                         "\t mmap off  : 0x%" PRIx64 "\n",
 656                         i, reg->size,
 657                         reg->guest_phys_addr,
 658                         reg->guest_user_addr,
 659                         reg->host_user_addr,
 660                         (uint64_t)(uintptr_t)mmap_addr,
 661                         mmap_size,
 662                         alignment,
 663                         mmap_offset);
 664         }
 665
 666         dump_guest_pages(dev);
 667
 668         return 0;
 669
 670 err_mmap:
 671         free_mem_region(dev);
 672         rte_free(dev->mem);
 673         dev->mem = NULL;
 674         return -1;
 675 }
 676
 677 static int
 678 vq_is_ready(struct vhost_virtqueue *vq)
 679 {
 680         return vq && vq->desc   &&
 681                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 682                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 683 }
 684
 685 static int
 686 virtio_is_ready(struct virtio_net *dev)
 687 {
 688         struct vhost_virtqueue *vq;
 689         uint32_t i;
 690
 691         if (dev->nr_vring == 0)
 692                 return 0;
 693
 694         for (i = 0; i < dev->nr_vring; i++) {
 695                 vq = dev->virtqueue[i];
 696
 697                 if (!vq_is_ready(vq))
 698                         return 0;
 699         }
 700
 701         RTE_LOG(INFO, VHOST_CONFIG,
 702                 "virtio is now ready for processing.\n");
 703         return 1;
 704 }
 705
 706 static void
 707 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 708 {
 709         struct vhost_vring_file file;
 710         struct vhost_virtqueue *vq;
 711
 712         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 713         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 714                 file.fd = VIRTIO_INVALID_EVENTFD;
 715         else
 716                 file.fd = pmsg->fds[0];
 717         RTE_LOG(INFO, VHOST_CONFIG,
 718                 "vring call idx:%d file:%d\n", file.index, file.fd);
 719
 720         vq = dev->virtqueue[file.index];
 721         if (vq->callfd >= 0)
 722                 close(vq->callfd);
 723
 724         vq->callfd = file.fd;
 725 }
 726
 727 static void
 728 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 729 {
 730         struct vhost_vring_file file;
 731         struct vhost_virtqueue *vq;
 732         struct virtio_net *dev = *pdev;
 733
 734         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 735         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 736                 file.fd = VIRTIO_INVALID_EVENTFD;
 737         else
 738                 file.fd = pmsg->fds[0];
 739         RTE_LOG(INFO, VHOST_CONFIG,
 740                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 741
 742         /*
 743          * Interpret ring addresses only when ring is started and enabled.
 744          * This is now if protocol features aren't supported.
 745          */
 746         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 747                 *pdev = dev = translate_ring_addresses(dev, file.index);
 748                 if (!dev)
 749                         return;
 750         }
 751
 752         vq = dev->virtqueue[file.index];
 753
 754         /*
 755          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 756          * the ring starts already enabled. Otherwise, it is enabled via
 757          * the SET_VRING_ENABLE message.
 758          */
 759         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 760                 vq->enabled = 1;
 761
 762         if (vq->kickfd >= 0)
 763                 close(vq->kickfd);
 764         vq->kickfd = file.fd;
 765 }
 766
 767 static void
 768 free_zmbufs(struct vhost_virtqueue *vq)
 769 {
 770         struct zcopy_mbuf *zmbuf, *next;
 771
 772         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 773              zmbuf != NULL; zmbuf = next) {
 774                 next = TAILQ_NEXT(zmbuf, next);
 775
 776                 rte_pktmbuf_free(zmbuf->mbuf);
 777                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 778         }
 779
 780         rte_free(vq->zmbufs);
 781 }
 782
 783 /*
 784  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 785  */
 786 static int
 787 vhost_user_get_vring_base(struct virtio_net *dev,
 788                           VhostUserMsg *msg)
 789 {
 790         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 791
 792         /* We have to stop the queue (virtio) if it is running. */
 793         if (dev->flags & VIRTIO_DEV_RUNNING) {
 794                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 795                 dev->notify_ops->destroy_device(dev->vid);
 796         }
 797
 798         dev->flags &= ~VIRTIO_DEV_READY;
 799
 800         /* Here we are safe to get the last used index */
 801         msg->payload.state.num = vq->last_used_idx;
 802
 803         RTE_LOG(INFO, VHOST_CONFIG,
 804                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 805                 msg->payload.state.num);
 806         /*
 807          * Based on current qemu vhost-user implementation, this message is
 808          * sent and only sent in vhost_vring_stop.
 809          * TODO: cleanup the vring, it isn't usable since here.
 810          */
 811         if (vq->kickfd >= 0)
 812                 close(vq->kickfd);
 813
 814         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 815
 816         if (dev->dequeue_zero_copy)
 817                 free_zmbufs(vq);
 818         rte_free(vq->shadow_used_ring);
 819         vq->shadow_used_ring = NULL;
 820
 821         rte_free(vq->batch_copy_elems);
 822         vq->batch_copy_elems = NULL;
 823
 824         return 0;
 825 }
 826
 827 /*
 828  * when virtio queues are ready to work, qemu will send us to
 829  * enable the virtio queue pair.
 830  */
 831 static int
 832 vhost_user_set_vring_enable(struct virtio_net **pdev,
 833                             VhostUserMsg *msg)
 834 {
 835         struct virtio_net *dev = *pdev;
 836         int enable = (int)msg->payload.state.num;
 837
 838         RTE_LOG(INFO, VHOST_CONFIG,
 839                 "set queue enable: %d to qp idx: %d\n",
 840                 enable, msg->payload.state.index);
 841
 842         /*
 843          * Interpret ring addresses only when ring is started and enabled.
 844          * This is now if protocol features are supported.
 845          */
 846         if (enable && (dev->features &
 847                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 848                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 849                 if (!dev)
 850                         return -1;
 851
 852                 *pdev = dev;
 853         }
 854
 855         if (dev->notify_ops->vring_state_changed)
 856                 dev->notify_ops->vring_state_changed(dev->vid,
 857                                 msg->payload.state.index, enable);
 858
 859         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 860
 861         return 0;
 862 }
 863
 864 static void
 865 vhost_user_set_protocol_features(struct virtio_net *dev,
 866                                  uint64_t protocol_features)
 867 {
 868         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 869                 return;
 870
 871         dev->protocol_features = protocol_features;
 872 }
 873
 874 static int
 875 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 876 {
 877         int fd = msg->fds[0];
 878         uint64_t size, off;
 879         void *addr;
 880
 881         if (fd < 0) {
 882                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 883                 return -1;
 884         }
 885
 886         if (msg->size != sizeof(VhostUserLog)) {
 887                 RTE_LOG(ERR, VHOST_CONFIG,
 888                         "invalid log base msg size: %"PRId32" != %d\n",
 889                         msg->size, (int)sizeof(VhostUserLog));
 890                 return -1;
 891         }
 892
 893         size = msg->payload.log.mmap_size;
 894         off  = msg->payload.log.mmap_offset;
 895         RTE_LOG(INFO, VHOST_CONFIG,
 896                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 897                 size, off);
 898
 899         /*
 900          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 901          * fail when offset is not page size aligned.
 902          */
 903         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 904         close(fd);
 905         if (addr == MAP_FAILED) {
 906                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 907                 return -1;
 908         }
 909
 910         /*
 911          * Free previously mapped log memory on occasionally
 912          * multiple VHOST_USER_SET_LOG_BASE.
 913          */
 914         if (dev->log_addr) {
 915                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 916         }
 917         dev->log_addr = (uint64_t)(uintptr_t)addr;
 918         dev->log_base = dev->log_addr + off;
 919         dev->log_size = size;
 920
 921         return 0;
 922 }
 923
 924 /*
 925  * An rarp packet is constructed and broadcasted to notify switches about
 926  * the new location of the migrated VM, so that packets from outside will
 927  * not be lost after migration.
 928  *
 929  * However, we don't actually "send" a rarp packet here, instead, we set
 930  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 931  */
 932 static int
 933 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 934 {
 935         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 936
 937         RTE_LOG(DEBUG, VHOST_CONFIG,
 938                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 939                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 940         memcpy(dev->mac.addr_bytes, mac, 6);
 941
 942         /*
 943          * Set the flag to inject a RARP broadcast packet at
 944          * rte_vhost_dequeue_burst().
 945          *
 946          * rte_smp_wmb() is for making sure the mac is copied
 947          * before the flag is set.
 948          */
 949         rte_smp_wmb();
 950         rte_atomic16_set(&dev->broadcast_rarp, 1);
 951
 952         return 0;
 953 }
 954
 955 static int
 956 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 957 {
 958         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 959                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 960                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 961                                 msg->payload.u64);
 962
 963                 return -1;
 964         }
 965
 966         dev->mtu = msg->payload.u64;
 967
 968         return 0;
 969 }
 970
 971 static int
 972 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 973 {
 974         int fd = msg->fds[0];
 975
 976         if (fd < 0) {
 977                 RTE_LOG(ERR, VHOST_CONFIG,
 978                                 "Invalid file descriptor for slave channel (%d)\n",
 979                                 fd);
 980                 return -1;
 981         }
 982
 983         dev->slave_req_fd = fd;
 984
 985         return 0;
 986 }
 987
 988 static int
 989 vhost_user_iotlb_msg(struct virtio_net *dev, struct VhostUserMsg *msg)
 990 {
 991         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
 992         uint16_t i;
 993         uint64_t vva;
 994
 995         switch (imsg->type) {
 996         case VHOST_IOTLB_UPDATE:
 997                 vva = qva_to_vva(dev, imsg->uaddr);
 998                 if (!vva)
 999                         return -1;
1000
1001                 for (i = 0; i < dev->nr_vring; i++) {
1002                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1003
1004                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1005                                         imsg->size, imsg->perm);
1006                 }
1007                 break;
1008         case VHOST_IOTLB_INVALIDATE:
1009                 for (i = 0; i < dev->nr_vring; i++) {
1010                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1011
1012                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1013                                         imsg->size);
1014                 }
1015                 break;
1016         default:
1017                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1018                                 imsg->type);
1019                 return -1;
1020         }
1021
1022         return 0;
1023 }
1024
1025 /* return bytes# of read on success or negative val on failure. */
1026 static int
1027 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1028 {
1029         int ret;
1030
1031         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1032                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1033         if (ret <= 0)
1034                 return ret;
1035
1036         if (msg && msg->size) {
1037                 if (msg->size > sizeof(msg->payload)) {
1038                         RTE_LOG(ERR, VHOST_CONFIG,
1039                                 "invalid msg size: %d\n", msg->size);
1040                         return -1;
1041                 }
1042                 ret = read(sockfd, &msg->payload, msg->size);
1043                 if (ret <= 0)
1044                         return ret;
1045                 if (ret != (int)msg->size) {
1046                         RTE_LOG(ERR, VHOST_CONFIG,
1047                                 "read control message failed\n");
1048                         return -1;
1049                 }
1050         }
1051
1052         return ret;
1053 }
1054
1055 static int
1056 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1057 {
1058         if (!msg)
1059                 return 0;
1060
1061         return send_fd_message(sockfd, (char *)msg,
1062                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1063 }
1064
1065 static int
1066 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1067 {
1068         if (!msg)
1069                 return 0;
1070
1071         msg->flags &= ~VHOST_USER_VERSION_MASK;
1072         msg->flags &= ~VHOST_USER_NEED_REPLY;
1073         msg->flags |= VHOST_USER_VERSION;
1074         msg->flags |= VHOST_USER_REPLY_MASK;
1075
1076         return send_vhost_message(sockfd, msg);
1077 }
1078
1079 /*
1080  * Allocate a queue pair if it hasn't been allocated yet
1081  */
1082 static int
1083 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1084 {
1085         uint16_t vring_idx;
1086
1087         switch (msg->request) {
1088         case VHOST_USER_SET_VRING_KICK:
1089         case VHOST_USER_SET_VRING_CALL:
1090         case VHOST_USER_SET_VRING_ERR:
1091                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1092                 break;
1093         case VHOST_USER_SET_VRING_NUM:
1094         case VHOST_USER_SET_VRING_BASE:
1095         case VHOST_USER_SET_VRING_ENABLE:
1096                 vring_idx = msg->payload.state.index;
1097                 break;
1098         case VHOST_USER_SET_VRING_ADDR:
1099                 vring_idx = msg->payload.addr.index;
1100                 break;
1101         default:
1102                 return 0;
1103         }
1104
1105         if (vring_idx >= VHOST_MAX_VRING) {
1106                 RTE_LOG(ERR, VHOST_CONFIG,
1107                         "invalid vring index: %u\n", vring_idx);
1108                 return -1;
1109         }
1110
1111         if (dev->virtqueue[vring_idx])
1112                 return 0;
1113
1114         return alloc_vring_queue(dev, vring_idx);
1115 }
1116
1117 int
1118 vhost_user_msg_handler(int vid, int fd)
1119 {
1120         struct virtio_net *dev;
1121         struct VhostUserMsg msg;
1122         int ret;
1123
1124         dev = get_device(vid);
1125         if (dev == NULL)
1126                 return -1;
1127
1128         if (!dev->notify_ops) {
1129                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1130                 if (!dev->notify_ops) {
1131                         RTE_LOG(ERR, VHOST_CONFIG,
1132                                 "failed to get callback ops for driver %s\n",
1133                                 dev->ifname);
1134                         return -1;
1135                 }
1136         }
1137
1138         ret = read_vhost_message(fd, &msg);
1139         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1140                 if (ret < 0)
1141                         RTE_LOG(ERR, VHOST_CONFIG,
1142                                 "vhost read message failed\n");
1143                 else if (ret == 0)
1144                         RTE_LOG(INFO, VHOST_CONFIG,
1145                                 "vhost peer closed\n");
1146                 else
1147                         RTE_LOG(ERR, VHOST_CONFIG,
1148                                 "vhost read incorrect message\n");
1149
1150                 return -1;
1151         }
1152
1153         ret = 0;
1154         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1155                 vhost_message_str[msg.request]);
1156
1157         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1158         if (ret < 0) {
1159                 RTE_LOG(ERR, VHOST_CONFIG,
1160                         "failed to alloc queue\n");
1161                 return -1;
1162         }
1163
1164         switch (msg.request) {
1165         case VHOST_USER_GET_FEATURES:
1166                 msg.payload.u64 = vhost_user_get_features(dev);
1167                 msg.size = sizeof(msg.payload.u64);
1168                 send_vhost_reply(fd, &msg);
1169                 break;
1170         case VHOST_USER_SET_FEATURES:
1171                 vhost_user_set_features(dev, msg.payload.u64);
1172                 break;
1173
1174         case VHOST_USER_GET_PROTOCOL_FEATURES:
1175                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1176                 msg.size = sizeof(msg.payload.u64);
1177                 send_vhost_reply(fd, &msg);
1178                 break;
1179         case VHOST_USER_SET_PROTOCOL_FEATURES:
1180                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1181                 break;
1182
1183         case VHOST_USER_SET_OWNER:
1184                 vhost_user_set_owner();
1185                 break;
1186         case VHOST_USER_RESET_OWNER:
1187                 vhost_user_reset_owner(dev);
1188                 break;
1189
1190         case VHOST_USER_SET_MEM_TABLE:
1191                 ret = vhost_user_set_mem_table(dev, &msg);
1192                 break;
1193
1194         case VHOST_USER_SET_LOG_BASE:
1195                 vhost_user_set_log_base(dev, &msg);
1196
1197                 /* it needs a reply */
1198                 msg.size = sizeof(msg.payload.u64);
1199                 send_vhost_reply(fd, &msg);
1200                 break;
1201         case VHOST_USER_SET_LOG_FD:
1202                 close(msg.fds[0]);
1203                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1204                 break;
1205
1206         case VHOST_USER_SET_VRING_NUM:
1207                 vhost_user_set_vring_num(dev, &msg);
1208                 break;
1209         case VHOST_USER_SET_VRING_ADDR:
1210                 vhost_user_set_vring_addr(dev, &msg);
1211                 break;
1212         case VHOST_USER_SET_VRING_BASE:
1213                 vhost_user_set_vring_base(dev, &msg);
1214                 break;
1215
1216         case VHOST_USER_GET_VRING_BASE:
1217                 vhost_user_get_vring_base(dev, &msg);
1218                 msg.size = sizeof(msg.payload.state);
1219                 send_vhost_reply(fd, &msg);
1220                 break;
1221
1222         case VHOST_USER_SET_VRING_KICK:
1223                 vhost_user_set_vring_kick(&dev, &msg);
1224                 break;
1225         case VHOST_USER_SET_VRING_CALL:
1226                 vhost_user_set_vring_call(dev, &msg);
1227                 break;
1228
1229         case VHOST_USER_SET_VRING_ERR:
1230                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1231                         close(msg.fds[0]);
1232                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1233                 break;
1234
1235         case VHOST_USER_GET_QUEUE_NUM:
1236                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1237                 msg.size = sizeof(msg.payload.u64);
1238                 send_vhost_reply(fd, &msg);
1239                 break;
1240
1241         case VHOST_USER_SET_VRING_ENABLE:
1242                 vhost_user_set_vring_enable(&dev, &msg);
1243                 break;
1244         case VHOST_USER_SEND_RARP:
1245                 vhost_user_send_rarp(dev, &msg);
1246                 break;
1247
1248         case VHOST_USER_NET_SET_MTU:
1249                 ret = vhost_user_net_set_mtu(dev, &msg);
1250                 break;
1251
1252         case VHOST_USER_SET_SLAVE_REQ_FD:
1253                 ret = vhost_user_set_req_fd(dev, &msg);
1254                 break;
1255
1256         case VHOST_USER_IOTLB_MSG:
1257                 ret = vhost_user_iotlb_msg(dev, &msg);
1258                 break;
1259
1260         default:
1261                 ret = -1;
1262                 break;
1263
1264         }
1265
1266         /*
1267          * The virtio_net struct might have been reallocated on a different
1268          * NUMA node, so dev pointer might no more be valid.
1269          */
1270         dev = get_device(vid);
1271
1272         if (msg.flags & VHOST_USER_NEED_REPLY) {
1273                 msg.payload.u64 = !!ret;
1274                 msg.size = sizeof(msg.payload.u64);
1275                 send_vhost_reply(fd, &msg);
1276         }
1277
1278         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1279                 dev->flags |= VIRTIO_DEV_READY;
1280
1281                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1282                         if (dev->dequeue_zero_copy) {
1283                                 RTE_LOG(INFO, VHOST_CONFIG,
1284                                                 "dequeue zero copy is enabled\n");
1285                         }
1286
1287                         if (dev->notify_ops->new_device(dev->vid) == 0)
1288                                 dev->flags |= VIRTIO_DEV_RUNNING;
1289                 }
1290         }
1291
1292         return 0;
1293 }
1294
1295 int
1296 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1297 {
1298         int ret;
1299         struct VhostUserMsg msg = {
1300                 .request = (enum VhostUserRequest)VHOST_USER_SLAVE_IOTLB_MSG,
1301                 .flags = VHOST_USER_VERSION,
1302                 .size = sizeof(msg.payload.iotlb),
1303                 .payload.iotlb = {
1304                         .iova = iova,
1305                         .perm = perm,
1306                         .type = VHOST_IOTLB_MISS,
1307                 },
1308         };
1309
1310         ret = send_vhost_message(dev->slave_req_fd, &msg);
1311         if (ret < 0) {
1312                 RTE_LOG(ERR, VHOST_CONFIG,
1313                                 "Failed to send IOTLB miss message (%d)\n",
1314                                 ret);
1315                 return ret;
1316         }
1317
1318         return 0;
1319 }