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         if (old_vq != vq)
 318                 vhost_user_iotlb_init(dev, index);
 319
 320         return dev;
 321 }
 322 #else
 323 static struct virtio_net*
 324 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 325 {
 326         return dev;
 327 }
 328 #endif
 329
 330 /* Converts QEMU virtual address to Vhost virtual address. */
 331 static uint64_t
 332 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 333 {
 334         struct rte_vhost_mem_region *reg;
 335         uint32_t i;
 336
 337         /* Find the region where the address lives. */
 338         for (i = 0; i < dev->mem->nregions; i++) {
 339                 reg = &dev->mem->regions[i];
 340
 341                 if (qva >= reg->guest_user_addr &&
 342                     qva <  reg->guest_user_addr + reg->size) {
 343                         return qva - reg->guest_user_addr +
 344                                reg->host_user_addr;
 345                 }
 346         }
 347
 348         return 0;
 349 }
 350
 351
 352 /*
 353  * Converts ring address to Vhost virtual address.
 354  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 355  * else it is a QEMU virtual address.
 356  */
 357 static uint64_t
 358 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 359                 uint64_t ra, uint64_t size)
 360 {
 361         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 362                 uint64_t vva;
 363
 364                 vva = vhost_user_iotlb_cache_find(vq, ra,
 365                                         &size, VHOST_ACCESS_RW);
 366                 if (!vva)
 367                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 368
 369                 return vva;
 370         }
 371
 372         return qva_to_vva(dev, ra);
 373 }
 374
 375 /*
 376  * The virtio device sends us the desc, used and avail ring addresses.
 377  * This function then converts these to our address space.
 378  */
 379 static int
 380 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
 381 {
 382         struct vhost_virtqueue *vq;
 383         struct vhost_vring_addr *addr = &msg->payload.addr;
 384
 385         if (dev->mem == NULL)
 386                 return -1;
 387
 388         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 389         vq = dev->virtqueue[msg->payload.addr.index];
 390
 391         /*
 392          * Rings addresses should not be interpreted as long as the ring is not
 393          * started and enabled
 394          */
 395         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 396
 397         vring_invalidate(dev, vq);
 398
 399         return 0;
 400 }
 401
 402 static struct virtio_net *
 403 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 404 {
 405         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 406         struct vhost_vring_addr *addr = &vq->ring_addrs;
 407
 408         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 409         if (vq->desc && vq->avail && vq->used)
 410                 return dev;
 411
 412         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 413                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 414         if (vq->desc == 0) {
 415                 RTE_LOG(DEBUG, VHOST_CONFIG,
 416                         "(%d) failed to find desc ring address.\n",
 417                         dev->vid);
 418                 return dev;
 419         }
 420
 421         dev = numa_realloc(dev, vq_index);
 422         vq = dev->virtqueue[vq_index];
 423
 424         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 425                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 426         if (vq->avail == 0) {
 427                 RTE_LOG(DEBUG, VHOST_CONFIG,
 428                         "(%d) failed to find avail ring address.\n",
 429                         dev->vid);
 430                 return dev;
 431         }
 432
 433         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 434                         vq, addr->used_user_addr, sizeof(struct vring_used));
 435         if (vq->used == 0) {
 436                 RTE_LOG(DEBUG, VHOST_CONFIG,
 437                         "(%d) failed to find used ring address.\n",
 438                         dev->vid);
 439                 return dev;
 440         }
 441
 442         if (vq->last_used_idx != vq->used->idx) {
 443                 RTE_LOG(WARNING, VHOST_CONFIG,
 444                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 445                         "some packets maybe resent for Tx and dropped for Rx\n",
 446                         vq->last_used_idx, vq->used->idx);
 447                 vq->last_used_idx  = vq->used->idx;
 448                 vq->last_avail_idx = vq->used->idx;
 449         }
 450
 451         vq->log_guest_addr = addr->log_guest_addr;
 452
 453         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 454                         dev->vid, vq->desc);
 455         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 456                         dev->vid, vq->avail);
 457         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 458                         dev->vid, vq->used);
 459         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 460                         dev->vid, vq->log_guest_addr);
 461
 462         return dev;
 463 }
 464
 465 /*
 466  * The virtio device sends us the available ring last used index.
 467  */
 468 static int
 469 vhost_user_set_vring_base(struct virtio_net *dev,
 470                           VhostUserMsg *msg)
 471 {
 472         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 473                         msg->payload.state.num;
 474         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 475                         msg->payload.state.num;
 476
 477         return 0;
 478 }
 479
 480 static void
 481 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 482                    uint64_t host_phys_addr, uint64_t size)
 483 {
 484         struct guest_page *page, *last_page;
 485
 486         if (dev->nr_guest_pages == dev->max_guest_pages) {
 487                 dev->max_guest_pages *= 2;
 488                 dev->guest_pages = realloc(dev->guest_pages,
 489                                         dev->max_guest_pages * sizeof(*page));
 490         }
 491
 492         if (dev->nr_guest_pages > 0) {
 493                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 494                 /* merge if the two pages are continuous */
 495                 if (host_phys_addr == last_page->host_phys_addr +
 496                                       last_page->size) {
 497                         last_page->size += size;
 498                         return;
 499                 }
 500         }
 501
 502         page = &dev->guest_pages[dev->nr_guest_pages++];
 503         page->guest_phys_addr = guest_phys_addr;
 504         page->host_phys_addr  = host_phys_addr;
 505         page->size = size;
 506 }
 507
 508 static void
 509 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 510                 uint64_t page_size)
 511 {
 512         uint64_t reg_size = reg->size;
 513         uint64_t host_user_addr  = reg->host_user_addr;
 514         uint64_t guest_phys_addr = reg->guest_phys_addr;
 515         uint64_t host_phys_addr;
 516         uint64_t size;
 517
 518         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 519         size = page_size - (guest_phys_addr & (page_size - 1));
 520         size = RTE_MIN(size, reg_size);
 521
 522         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 523         host_user_addr  += size;
 524         guest_phys_addr += size;
 525         reg_size -= size;
 526
 527         while (reg_size > 0) {
 528                 size = RTE_MIN(reg_size, page_size);
 529                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 530                                                   host_user_addr);
 531                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 532
 533                 host_user_addr  += size;
 534                 guest_phys_addr += size;
 535                 reg_size -= size;
 536         }
 537 }
 538
 539 #ifdef RTE_LIBRTE_VHOST_DEBUG
 540 /* TODO: enable it only in debug mode? */
 541 static void
 542 dump_guest_pages(struct virtio_net *dev)
 543 {
 544         uint32_t i;
 545         struct guest_page *page;
 546
 547         for (i = 0; i < dev->nr_guest_pages; i++) {
 548                 page = &dev->guest_pages[i];
 549
 550                 RTE_LOG(INFO, VHOST_CONFIG,
 551                         "guest physical page region %u\n"
 552                         "\t guest_phys_addr: %" PRIx64 "\n"
 553                         "\t host_phys_addr : %" PRIx64 "\n"
 554                         "\t size           : %" PRIx64 "\n",
 555                         i,
 556                         page->guest_phys_addr,
 557                         page->host_phys_addr,
 558                         page->size);
 559         }
 560 }
 561 #else
 562 #define dump_guest_pages(dev)
 563 #endif
 564
 565 static int
 566 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 567 {
 568         struct VhostUserMemory memory = pmsg->payload.memory;
 569         struct rte_vhost_mem_region *reg;
 570         void *mmap_addr;
 571         uint64_t mmap_size;
 572         uint64_t mmap_offset;
 573         uint64_t alignment;
 574         uint32_t i;
 575         int fd;
 576
 577         if (dev->mem) {
 578                 free_mem_region(dev);
 579                 rte_free(dev->mem);
 580                 dev->mem = NULL;
 581         }
 582
 583         dev->nr_guest_pages = 0;
 584         if (!dev->guest_pages) {
 585                 dev->max_guest_pages = 8;
 586                 dev->guest_pages = malloc(dev->max_guest_pages *
 587                                                 sizeof(struct guest_page));
 588                 if (dev->guest_pages == NULL) {
 589                         RTE_LOG(ERR, VHOST_CONFIG,
 590                                 "(%d) failed to allocate memory "
 591                                 "for dev->guest_pages\n",
 592                                 dev->vid);
 593                         return -1;
 594                 }
 595         }
 596
 597         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 598                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 599         if (dev->mem == NULL) {
 600                 RTE_LOG(ERR, VHOST_CONFIG,
 601                         "(%d) failed to allocate memory for dev->mem\n",
 602                         dev->vid);
 603                 return -1;
 604         }
 605         dev->mem->nregions = memory.nregions;
 606
 607         for (i = 0; i < memory.nregions; i++) {
 608                 fd  = pmsg->fds[i];
 609                 reg = &dev->mem->regions[i];
 610
 611                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 612                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 613                 reg->size            = memory.regions[i].memory_size;
 614                 reg->fd              = fd;
 615
 616                 mmap_offset = memory.regions[i].mmap_offset;
 617                 mmap_size   = reg->size + mmap_offset;
 618
 619                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 620                  * with length argument aligned with hugepagesz at older
 621                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 622                  * mmap() will fail with EINVAL.
 623                  *
 624                  * to avoid failure, make sure in caller to keep length
 625                  * aligned.
 626                  */
 627                 alignment = get_blk_size(fd);
 628                 if (alignment == (uint64_t)-1) {
 629                         RTE_LOG(ERR, VHOST_CONFIG,
 630                                 "couldn't get hugepage size through fstat\n");
 631                         goto err_mmap;
 632                 }
 633                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 634
 635                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 636                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 637
 638                 if (mmap_addr == MAP_FAILED) {
 639                         RTE_LOG(ERR, VHOST_CONFIG,
 640                                 "mmap region %u failed.\n", i);
 641                         goto err_mmap;
 642                 }
 643
 644                 reg->mmap_addr = mmap_addr;
 645                 reg->mmap_size = mmap_size;
 646                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 647                                       mmap_offset;
 648
 649                 if (dev->dequeue_zero_copy)
 650                         add_guest_pages(dev, reg, alignment);
 651
 652                 RTE_LOG(INFO, VHOST_CONFIG,
 653                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 654                         "\t guest physical addr: 0x%" PRIx64 "\n"
 655                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 656                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 657                         "\t mmap addr : 0x%" PRIx64 "\n"
 658                         "\t mmap size : 0x%" PRIx64 "\n"
 659                         "\t mmap align: 0x%" PRIx64 "\n"
 660                         "\t mmap off  : 0x%" PRIx64 "\n",
 661                         i, reg->size,
 662                         reg->guest_phys_addr,
 663                         reg->guest_user_addr,
 664                         reg->host_user_addr,
 665                         (uint64_t)(uintptr_t)mmap_addr,
 666                         mmap_size,
 667                         alignment,
 668                         mmap_offset);
 669         }
 670
 671         dump_guest_pages(dev);
 672
 673         return 0;
 674
 675 err_mmap:
 676         free_mem_region(dev);
 677         rte_free(dev->mem);
 678         dev->mem = NULL;
 679         return -1;
 680 }
 681
 682 static int
 683 vq_is_ready(struct vhost_virtqueue *vq)
 684 {
 685         return vq && vq->desc && vq->avail && vq->used &&
 686                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 687                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 688 }
 689
 690 static int
 691 virtio_is_ready(struct virtio_net *dev)
 692 {
 693         struct vhost_virtqueue *vq;
 694         uint32_t i;
 695
 696         if (dev->nr_vring == 0)
 697                 return 0;
 698
 699         for (i = 0; i < dev->nr_vring; i++) {
 700                 vq = dev->virtqueue[i];
 701
 702                 if (!vq_is_ready(vq))
 703                         return 0;
 704         }
 705
 706         RTE_LOG(INFO, VHOST_CONFIG,
 707                 "virtio is now ready for processing.\n");
 708         return 1;
 709 }
 710
 711 static void
 712 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 713 {
 714         struct vhost_vring_file file;
 715         struct vhost_virtqueue *vq;
 716
 717         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 718         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 719                 file.fd = VIRTIO_INVALID_EVENTFD;
 720         else
 721                 file.fd = pmsg->fds[0];
 722         RTE_LOG(INFO, VHOST_CONFIG,
 723                 "vring call idx:%d file:%d\n", file.index, file.fd);
 724
 725         vq = dev->virtqueue[file.index];
 726         if (vq->callfd >= 0)
 727                 close(vq->callfd);
 728
 729         vq->callfd = file.fd;
 730 }
 731
 732 static void
 733 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 734 {
 735         struct vhost_vring_file file;
 736         struct vhost_virtqueue *vq;
 737         struct virtio_net *dev = *pdev;
 738
 739         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 740         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 741                 file.fd = VIRTIO_INVALID_EVENTFD;
 742         else
 743                 file.fd = pmsg->fds[0];
 744         RTE_LOG(INFO, VHOST_CONFIG,
 745                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 746
 747         /*
 748          * Interpret ring addresses only when ring is started and enabled.
 749          * This is now if protocol features aren't supported.
 750          */
 751         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 752                 *pdev = dev = translate_ring_addresses(dev, file.index);
 753                 if (!dev)
 754                         return;
 755         }
 756
 757         vq = dev->virtqueue[file.index];
 758
 759         /*
 760          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 761          * the ring starts already enabled. Otherwise, it is enabled via
 762          * the SET_VRING_ENABLE message.
 763          */
 764         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 765                 vq->enabled = 1;
 766
 767         if (vq->kickfd >= 0)
 768                 close(vq->kickfd);
 769         vq->kickfd = file.fd;
 770 }
 771
 772 static void
 773 free_zmbufs(struct vhost_virtqueue *vq)
 774 {
 775         struct zcopy_mbuf *zmbuf, *next;
 776
 777         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 778              zmbuf != NULL; zmbuf = next) {
 779                 next = TAILQ_NEXT(zmbuf, next);
 780
 781                 rte_pktmbuf_free(zmbuf->mbuf);
 782                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 783         }
 784
 785         rte_free(vq->zmbufs);
 786 }
 787
 788 /*
 789  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 790  */
 791 static int
 792 vhost_user_get_vring_base(struct virtio_net *dev,
 793                           VhostUserMsg *msg)
 794 {
 795         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 796
 797         /* We have to stop the queue (virtio) if it is running. */
 798         if (dev->flags & VIRTIO_DEV_RUNNING) {
 799                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 800                 dev->notify_ops->destroy_device(dev->vid);
 801         }
 802
 803         dev->flags &= ~VIRTIO_DEV_READY;
 804
 805         /* Here we are safe to get the last used index */
 806         msg->payload.state.num = vq->last_used_idx;
 807
 808         RTE_LOG(INFO, VHOST_CONFIG,
 809                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 810                 msg->payload.state.num);
 811         /*
 812          * Based on current qemu vhost-user implementation, this message is
 813          * sent and only sent in vhost_vring_stop.
 814          * TODO: cleanup the vring, it isn't usable since here.
 815          */
 816         if (vq->kickfd >= 0)
 817                 close(vq->kickfd);
 818
 819         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 820
 821         if (dev->dequeue_zero_copy)
 822                 free_zmbufs(vq);
 823         rte_free(vq->shadow_used_ring);
 824         vq->shadow_used_ring = NULL;
 825
 826         rte_free(vq->batch_copy_elems);
 827         vq->batch_copy_elems = NULL;
 828
 829         return 0;
 830 }
 831
 832 /*
 833  * when virtio queues are ready to work, qemu will send us to
 834  * enable the virtio queue pair.
 835  */
 836 static int
 837 vhost_user_set_vring_enable(struct virtio_net **pdev,
 838                             VhostUserMsg *msg)
 839 {
 840         struct virtio_net *dev = *pdev;
 841         int enable = (int)msg->payload.state.num;
 842
 843         RTE_LOG(INFO, VHOST_CONFIG,
 844                 "set queue enable: %d to qp idx: %d\n",
 845                 enable, msg->payload.state.index);
 846
 847         /*
 848          * Interpret ring addresses only when ring is started and enabled.
 849          * This is now if protocol features are supported.
 850          */
 851         if (enable && (dev->features &
 852                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 853                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 854                 if (!dev)
 855                         return -1;
 856
 857                 *pdev = dev;
 858         }
 859
 860         if (dev->notify_ops->vring_state_changed)
 861                 dev->notify_ops->vring_state_changed(dev->vid,
 862                                 msg->payload.state.index, enable);
 863
 864         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 865
 866         return 0;
 867 }
 868
 869 static void
 870 vhost_user_set_protocol_features(struct virtio_net *dev,
 871                                  uint64_t protocol_features)
 872 {
 873         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 874                 return;
 875
 876         dev->protocol_features = protocol_features;
 877 }
 878
 879 static int
 880 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 881 {
 882         int fd = msg->fds[0];
 883         uint64_t size, off;
 884         void *addr;
 885
 886         if (fd < 0) {
 887                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 888                 return -1;
 889         }
 890
 891         if (msg->size != sizeof(VhostUserLog)) {
 892                 RTE_LOG(ERR, VHOST_CONFIG,
 893                         "invalid log base msg size: %"PRId32" != %d\n",
 894                         msg->size, (int)sizeof(VhostUserLog));
 895                 return -1;
 896         }
 897
 898         size = msg->payload.log.mmap_size;
 899         off  = msg->payload.log.mmap_offset;
 900         RTE_LOG(INFO, VHOST_CONFIG,
 901                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 902                 size, off);
 903
 904         /*
 905          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 906          * fail when offset is not page size aligned.
 907          */
 908         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 909         close(fd);
 910         if (addr == MAP_FAILED) {
 911                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 912                 return -1;
 913         }
 914
 915         /*
 916          * Free previously mapped log memory on occasionally
 917          * multiple VHOST_USER_SET_LOG_BASE.
 918          */
 919         if (dev->log_addr) {
 920                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 921         }
 922         dev->log_addr = (uint64_t)(uintptr_t)addr;
 923         dev->log_base = dev->log_addr + off;
 924         dev->log_size = size;
 925
 926         return 0;
 927 }
 928
 929 /*
 930  * An rarp packet is constructed and broadcasted to notify switches about
 931  * the new location of the migrated VM, so that packets from outside will
 932  * not be lost after migration.
 933  *
 934  * However, we don't actually "send" a rarp packet here, instead, we set
 935  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 936  */
 937 static int
 938 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 939 {
 940         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 941
 942         RTE_LOG(DEBUG, VHOST_CONFIG,
 943                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 944                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 945         memcpy(dev->mac.addr_bytes, mac, 6);
 946
 947         /*
 948          * Set the flag to inject a RARP broadcast packet at
 949          * rte_vhost_dequeue_burst().
 950          *
 951          * rte_smp_wmb() is for making sure the mac is copied
 952          * before the flag is set.
 953          */
 954         rte_smp_wmb();
 955         rte_atomic16_set(&dev->broadcast_rarp, 1);
 956
 957         return 0;
 958 }
 959
 960 static int
 961 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 962 {
 963         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 964                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 965                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 966                                 msg->payload.u64);
 967
 968                 return -1;
 969         }
 970
 971         dev->mtu = msg->payload.u64;
 972
 973         return 0;
 974 }
 975
 976 static int
 977 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 978 {
 979         int fd = msg->fds[0];
 980
 981         if (fd < 0) {
 982                 RTE_LOG(ERR, VHOST_CONFIG,
 983                                 "Invalid file descriptor for slave channel (%d)\n",
 984                                 fd);
 985                 return -1;
 986         }
 987
 988         dev->slave_req_fd = fd;
 989
 990         return 0;
 991 }
 992
 993 static int
 994 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
 995 {
 996         struct vhost_vring_addr *ra;
 997         uint64_t start, end;
 998
 999         start = imsg->iova;
1000         end = start + imsg->size;
1001
1002         ra = &vq->ring_addrs;
1003         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1004                 return 1;
1005         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1006                 return 1;
1007         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1008                 return 1;
1009
1010         return 0;
1011 }
1012
1013 static int
1014 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1015                                 struct vhost_iotlb_msg *imsg)
1016 {
1017         uint64_t istart, iend, vstart, vend;
1018
1019         istart = imsg->iova;
1020         iend = istart + imsg->size - 1;
1021
1022         vstart = (uintptr_t)vq->desc;
1023         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1024         if (vstart <= iend && istart <= vend)
1025                 return 1;
1026
1027         vstart = (uintptr_t)vq->avail;
1028         vend = vstart + sizeof(struct vring_avail);
1029         vend += sizeof(uint16_t) * vq->size - 1;
1030         if (vstart <= iend && istart <= vend)
1031                 return 1;
1032
1033         vstart = (uintptr_t)vq->used;
1034         vend = vstart + sizeof(struct vring_used);
1035         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1036         if (vstart <= iend && istart <= vend)
1037                 return 1;
1038
1039         return 0;
1040 }
1041
1042 static int
1043 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1044 {
1045         struct virtio_net *dev = *pdev;
1046         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1047         uint16_t i;
1048         uint64_t vva;
1049
1050         switch (imsg->type) {
1051         case VHOST_IOTLB_UPDATE:
1052                 vva = qva_to_vva(dev, imsg->uaddr);
1053                 if (!vva)
1054                         return -1;
1055
1056                 for (i = 0; i < dev->nr_vring; i++) {
1057                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1058
1059                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1060                                         imsg->size, imsg->perm);
1061
1062                         if (is_vring_iotlb_update(vq, imsg))
1063                                 *pdev = dev = translate_ring_addresses(dev, i);
1064                 }
1065                 break;
1066         case VHOST_IOTLB_INVALIDATE:
1067                 for (i = 0; i < dev->nr_vring; i++) {
1068                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1069
1070                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1071                                         imsg->size);
1072
1073                         if (is_vring_iotlb_invalidate(vq, imsg))
1074                                 vring_invalidate(dev, vq);
1075                 }
1076                 break;
1077         default:
1078                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1079                                 imsg->type);
1080                 return -1;
1081         }
1082
1083         return 0;
1084 }
1085
1086 /* return bytes# of read on success or negative val on failure. */
1087 static int
1088 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1089 {
1090         int ret;
1091
1092         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1093                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1094         if (ret <= 0)
1095                 return ret;
1096
1097         if (msg && msg->size) {
1098                 if (msg->size > sizeof(msg->payload)) {
1099                         RTE_LOG(ERR, VHOST_CONFIG,
1100                                 "invalid msg size: %d\n", msg->size);
1101                         return -1;
1102                 }
1103                 ret = read(sockfd, &msg->payload, msg->size);
1104                 if (ret <= 0)
1105                         return ret;
1106                 if (ret != (int)msg->size) {
1107                         RTE_LOG(ERR, VHOST_CONFIG,
1108                                 "read control message failed\n");
1109                         return -1;
1110                 }
1111         }
1112
1113         return ret;
1114 }
1115
1116 static int
1117 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1118 {
1119         if (!msg)
1120                 return 0;
1121
1122         return send_fd_message(sockfd, (char *)msg,
1123                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1124 }
1125
1126 static int
1127 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1128 {
1129         if (!msg)
1130                 return 0;
1131
1132         msg->flags &= ~VHOST_USER_VERSION_MASK;
1133         msg->flags &= ~VHOST_USER_NEED_REPLY;
1134         msg->flags |= VHOST_USER_VERSION;
1135         msg->flags |= VHOST_USER_REPLY_MASK;
1136
1137         return send_vhost_message(sockfd, msg);
1138 }
1139
1140 /*
1141  * Allocate a queue pair if it hasn't been allocated yet
1142  */
1143 static int
1144 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1145 {
1146         uint16_t vring_idx;
1147
1148         switch (msg->request.master) {
1149         case VHOST_USER_SET_VRING_KICK:
1150         case VHOST_USER_SET_VRING_CALL:
1151         case VHOST_USER_SET_VRING_ERR:
1152                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1153                 break;
1154         case VHOST_USER_SET_VRING_NUM:
1155         case VHOST_USER_SET_VRING_BASE:
1156         case VHOST_USER_SET_VRING_ENABLE:
1157                 vring_idx = msg->payload.state.index;
1158                 break;
1159         case VHOST_USER_SET_VRING_ADDR:
1160                 vring_idx = msg->payload.addr.index;
1161                 break;
1162         default:
1163                 return 0;
1164         }
1165
1166         if (vring_idx >= VHOST_MAX_VRING) {
1167                 RTE_LOG(ERR, VHOST_CONFIG,
1168                         "invalid vring index: %u\n", vring_idx);
1169                 return -1;
1170         }
1171
1172         if (dev->virtqueue[vring_idx])
1173                 return 0;
1174
1175         return alloc_vring_queue(dev, vring_idx);
1176 }
1177
1178 int
1179 vhost_user_msg_handler(int vid, int fd)
1180 {
1181         struct virtio_net *dev;
1182         struct VhostUserMsg msg;
1183         int ret;
1184
1185         dev = get_device(vid);
1186         if (dev == NULL)
1187                 return -1;
1188
1189         if (!dev->notify_ops) {
1190                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1191                 if (!dev->notify_ops) {
1192                         RTE_LOG(ERR, VHOST_CONFIG,
1193                                 "failed to get callback ops for driver %s\n",
1194                                 dev->ifname);
1195                         return -1;
1196                 }
1197         }
1198
1199         ret = read_vhost_message(fd, &msg);
1200         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1201                 if (ret < 0)
1202                         RTE_LOG(ERR, VHOST_CONFIG,
1203                                 "vhost read message failed\n");
1204                 else if (ret == 0)
1205                         RTE_LOG(INFO, VHOST_CONFIG,
1206                                 "vhost peer closed\n");
1207                 else
1208                         RTE_LOG(ERR, VHOST_CONFIG,
1209                                 "vhost read incorrect message\n");
1210
1211                 return -1;
1212         }
1213
1214         ret = 0;
1215         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1216                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1217                         vhost_message_str[msg.request.master]);
1218         else
1219                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1220                         vhost_message_str[msg.request.master]);
1221
1222         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1223         if (ret < 0) {
1224                 RTE_LOG(ERR, VHOST_CONFIG,
1225                         "failed to alloc queue\n");
1226                 return -1;
1227         }
1228
1229         switch (msg.request.master) {
1230         case VHOST_USER_GET_FEATURES:
1231                 msg.payload.u64 = vhost_user_get_features(dev);
1232                 msg.size = sizeof(msg.payload.u64);
1233                 send_vhost_reply(fd, &msg);
1234                 break;
1235         case VHOST_USER_SET_FEATURES:
1236                 vhost_user_set_features(dev, msg.payload.u64);
1237                 break;
1238
1239         case VHOST_USER_GET_PROTOCOL_FEATURES:
1240                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1241                 msg.size = sizeof(msg.payload.u64);
1242                 send_vhost_reply(fd, &msg);
1243                 break;
1244         case VHOST_USER_SET_PROTOCOL_FEATURES:
1245                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1246                 break;
1247
1248         case VHOST_USER_SET_OWNER:
1249                 vhost_user_set_owner();
1250                 break;
1251         case VHOST_USER_RESET_OWNER:
1252                 vhost_user_reset_owner(dev);
1253                 break;
1254
1255         case VHOST_USER_SET_MEM_TABLE:
1256                 ret = vhost_user_set_mem_table(dev, &msg);
1257                 break;
1258
1259         case VHOST_USER_SET_LOG_BASE:
1260                 vhost_user_set_log_base(dev, &msg);
1261
1262                 /* it needs a reply */
1263                 msg.size = sizeof(msg.payload.u64);
1264                 send_vhost_reply(fd, &msg);
1265                 break;
1266         case VHOST_USER_SET_LOG_FD:
1267                 close(msg.fds[0]);
1268                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1269                 break;
1270
1271         case VHOST_USER_SET_VRING_NUM:
1272                 vhost_user_set_vring_num(dev, &msg);
1273                 break;
1274         case VHOST_USER_SET_VRING_ADDR:
1275                 vhost_user_set_vring_addr(dev, &msg);
1276                 break;
1277         case VHOST_USER_SET_VRING_BASE:
1278                 vhost_user_set_vring_base(dev, &msg);
1279                 break;
1280
1281         case VHOST_USER_GET_VRING_BASE:
1282                 vhost_user_get_vring_base(dev, &msg);
1283                 msg.size = sizeof(msg.payload.state);
1284                 send_vhost_reply(fd, &msg);
1285                 break;
1286
1287         case VHOST_USER_SET_VRING_KICK:
1288                 vhost_user_set_vring_kick(&dev, &msg);
1289                 break;
1290         case VHOST_USER_SET_VRING_CALL:
1291                 vhost_user_set_vring_call(dev, &msg);
1292                 break;
1293
1294         case VHOST_USER_SET_VRING_ERR:
1295                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1296                         close(msg.fds[0]);
1297                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1298                 break;
1299
1300         case VHOST_USER_GET_QUEUE_NUM:
1301                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1302                 msg.size = sizeof(msg.payload.u64);
1303                 send_vhost_reply(fd, &msg);
1304                 break;
1305
1306         case VHOST_USER_SET_VRING_ENABLE:
1307                 vhost_user_set_vring_enable(&dev, &msg);
1308                 break;
1309         case VHOST_USER_SEND_RARP:
1310                 vhost_user_send_rarp(dev, &msg);
1311                 break;
1312
1313         case VHOST_USER_NET_SET_MTU:
1314                 ret = vhost_user_net_set_mtu(dev, &msg);
1315                 break;
1316
1317         case VHOST_USER_SET_SLAVE_REQ_FD:
1318                 ret = vhost_user_set_req_fd(dev, &msg);
1319                 break;
1320
1321         case VHOST_USER_IOTLB_MSG:
1322                 ret = vhost_user_iotlb_msg(&dev, &msg);
1323                 break;
1324
1325         default:
1326                 ret = -1;
1327                 break;
1328
1329         }
1330
1331         if (msg.flags & VHOST_USER_NEED_REPLY) {
1332                 msg.payload.u64 = !!ret;
1333                 msg.size = sizeof(msg.payload.u64);
1334                 send_vhost_reply(fd, &msg);
1335         }
1336
1337         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1338                 dev->flags |= VIRTIO_DEV_READY;
1339
1340                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1341                         if (dev->dequeue_zero_copy) {
1342                                 RTE_LOG(INFO, VHOST_CONFIG,
1343                                                 "dequeue zero copy is enabled\n");
1344                         }
1345
1346                         if (dev->notify_ops->new_device(dev->vid) == 0)
1347                                 dev->flags |= VIRTIO_DEV_RUNNING;
1348                 }
1349         }
1350
1351         return 0;
1352 }
1353
1354 int
1355 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1356 {
1357         int ret;
1358         struct VhostUserMsg msg = {
1359                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1360                 .flags = VHOST_USER_VERSION,
1361                 .size = sizeof(msg.payload.iotlb),
1362                 .payload.iotlb = {
1363                         .iova = iova,
1364                         .perm = perm,
1365                         .type = VHOST_IOTLB_MISS,
1366                 },
1367         };
1368
1369         ret = send_vhost_message(dev->slave_req_fd, &msg);
1370         if (ret < 0) {
1371                 RTE_LOG(ERR, VHOST_CONFIG,
1372                                 "Failed to send IOTLB miss message (%d)\n",
1373                                 ret);
1374                 return ret;
1375         }
1376
1377         return 0;
1378 }