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 static struct virtio_net *
 376 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 377 {
 378         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 379         struct vhost_vring_addr *addr = &vq->ring_addrs;
 380
 381         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 382         if (vq->desc && vq->avail && vq->used)
 383                 return dev;
 384
 385         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 386                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 387         if (vq->desc == 0) {
 388                 RTE_LOG(DEBUG, VHOST_CONFIG,
 389                         "(%d) failed to find desc ring address.\n",
 390                         dev->vid);
 391                 return dev;
 392         }
 393
 394         dev = numa_realloc(dev, vq_index);
 395         vq = dev->virtqueue[vq_index];
 396         addr = &vq->ring_addrs;
 397
 398         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 399                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 400         if (vq->avail == 0) {
 401                 RTE_LOG(DEBUG, VHOST_CONFIG,
 402                         "(%d) failed to find avail ring address.\n",
 403                         dev->vid);
 404                 return dev;
 405         }
 406
 407         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 408                         vq, addr->used_user_addr, sizeof(struct vring_used));
 409         if (vq->used == 0) {
 410                 RTE_LOG(DEBUG, VHOST_CONFIG,
 411                         "(%d) failed to find used ring address.\n",
 412                         dev->vid);
 413                 return dev;
 414         }
 415
 416         if (vq->last_used_idx != vq->used->idx) {
 417                 RTE_LOG(WARNING, VHOST_CONFIG,
 418                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 419                         "some packets maybe resent for Tx and dropped for Rx\n",
 420                         vq->last_used_idx, vq->used->idx);
 421                 vq->last_used_idx  = vq->used->idx;
 422                 vq->last_avail_idx = vq->used->idx;
 423         }
 424
 425         vq->log_guest_addr = addr->log_guest_addr;
 426
 427         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 428                         dev->vid, vq->desc);
 429         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 430                         dev->vid, vq->avail);
 431         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 432                         dev->vid, vq->used);
 433         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 434                         dev->vid, vq->log_guest_addr);
 435
 436         return dev;
 437 }
 438
 439 /*
 440  * The virtio device sends us the desc, used and avail ring addresses.
 441  * This function then converts these to our address space.
 442  */
 443 static int
 444 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 445 {
 446         struct vhost_virtqueue *vq;
 447         struct vhost_vring_addr *addr = &msg->payload.addr;
 448         struct virtio_net *dev = *pdev;
 449
 450         if (dev->mem == NULL)
 451                 return -1;
 452
 453         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 454         vq = dev->virtqueue[msg->payload.addr.index];
 455
 456         /*
 457          * Rings addresses should not be interpreted as long as the ring is not
 458          * started and enabled
 459          */
 460         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 461
 462         vring_invalidate(dev, vq);
 463
 464         if (vq->enabled && (dev->features &
 465                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 466                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 467                 if (!dev)
 468                         return -1;
 469
 470                 *pdev = dev;
 471         }
 472
 473         return 0;
 474 }
 475
 476 /*
 477  * The virtio device sends us the available ring last used index.
 478  */
 479 static int
 480 vhost_user_set_vring_base(struct virtio_net *dev,
 481                           VhostUserMsg *msg)
 482 {
 483         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 484                         msg->payload.state.num;
 485         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 486                         msg->payload.state.num;
 487
 488         return 0;
 489 }
 490
 491 static void
 492 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 493                    uint64_t host_phys_addr, uint64_t size)
 494 {
 495         struct guest_page *page, *last_page;
 496
 497         if (dev->nr_guest_pages == dev->max_guest_pages) {
 498                 dev->max_guest_pages *= 2;
 499                 dev->guest_pages = realloc(dev->guest_pages,
 500                                         dev->max_guest_pages * sizeof(*page));
 501         }
 502
 503         if (dev->nr_guest_pages > 0) {
 504                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 505                 /* merge if the two pages are continuous */
 506                 if (host_phys_addr == last_page->host_phys_addr +
 507                                       last_page->size) {
 508                         last_page->size += size;
 509                         return;
 510                 }
 511         }
 512
 513         page = &dev->guest_pages[dev->nr_guest_pages++];
 514         page->guest_phys_addr = guest_phys_addr;
 515         page->host_phys_addr  = host_phys_addr;
 516         page->size = size;
 517 }
 518
 519 static void
 520 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 521                 uint64_t page_size)
 522 {
 523         uint64_t reg_size = reg->size;
 524         uint64_t host_user_addr  = reg->host_user_addr;
 525         uint64_t guest_phys_addr = reg->guest_phys_addr;
 526         uint64_t host_phys_addr;
 527         uint64_t size;
 528
 529         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 530         size = page_size - (guest_phys_addr & (page_size - 1));
 531         size = RTE_MIN(size, reg_size);
 532
 533         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 534         host_user_addr  += size;
 535         guest_phys_addr += size;
 536         reg_size -= size;
 537
 538         while (reg_size > 0) {
 539                 size = RTE_MIN(reg_size, page_size);
 540                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 541                                                   host_user_addr);
 542                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 543
 544                 host_user_addr  += size;
 545                 guest_phys_addr += size;
 546                 reg_size -= size;
 547         }
 548 }
 549
 550 #ifdef RTE_LIBRTE_VHOST_DEBUG
 551 /* TODO: enable it only in debug mode? */
 552 static void
 553 dump_guest_pages(struct virtio_net *dev)
 554 {
 555         uint32_t i;
 556         struct guest_page *page;
 557
 558         for (i = 0; i < dev->nr_guest_pages; i++) {
 559                 page = &dev->guest_pages[i];
 560
 561                 RTE_LOG(INFO, VHOST_CONFIG,
 562                         "guest physical page region %u\n"
 563                         "\t guest_phys_addr: %" PRIx64 "\n"
 564                         "\t host_phys_addr : %" PRIx64 "\n"
 565                         "\t size           : %" PRIx64 "\n",
 566                         i,
 567                         page->guest_phys_addr,
 568                         page->host_phys_addr,
 569                         page->size);
 570         }
 571 }
 572 #else
 573 #define dump_guest_pages(dev)
 574 #endif
 575
 576 static int
 577 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 578 {
 579         struct VhostUserMemory memory = pmsg->payload.memory;
 580         struct rte_vhost_mem_region *reg;
 581         void *mmap_addr;
 582         uint64_t mmap_size;
 583         uint64_t mmap_offset;
 584         uint64_t alignment;
 585         uint32_t i;
 586         int fd;
 587
 588         if (dev->mem) {
 589                 free_mem_region(dev);
 590                 rte_free(dev->mem);
 591                 dev->mem = NULL;
 592         }
 593
 594         dev->nr_guest_pages = 0;
 595         if (!dev->guest_pages) {
 596                 dev->max_guest_pages = 8;
 597                 dev->guest_pages = malloc(dev->max_guest_pages *
 598                                                 sizeof(struct guest_page));
 599                 if (dev->guest_pages == NULL) {
 600                         RTE_LOG(ERR, VHOST_CONFIG,
 601                                 "(%d) failed to allocate memory "
 602                                 "for dev->guest_pages\n",
 603                                 dev->vid);
 604                         return -1;
 605                 }
 606         }
 607
 608         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 609                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 610         if (dev->mem == NULL) {
 611                 RTE_LOG(ERR, VHOST_CONFIG,
 612                         "(%d) failed to allocate memory for dev->mem\n",
 613                         dev->vid);
 614                 return -1;
 615         }
 616         dev->mem->nregions = memory.nregions;
 617
 618         for (i = 0; i < memory.nregions; i++) {
 619                 fd  = pmsg->fds[i];
 620                 reg = &dev->mem->regions[i];
 621
 622                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 623                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 624                 reg->size            = memory.regions[i].memory_size;
 625                 reg->fd              = fd;
 626
 627                 mmap_offset = memory.regions[i].mmap_offset;
 628                 mmap_size   = reg->size + mmap_offset;
 629
 630                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 631                  * with length argument aligned with hugepagesz at older
 632                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 633                  * mmap() will fail with EINVAL.
 634                  *
 635                  * to avoid failure, make sure in caller to keep length
 636                  * aligned.
 637                  */
 638                 alignment = get_blk_size(fd);
 639                 if (alignment == (uint64_t)-1) {
 640                         RTE_LOG(ERR, VHOST_CONFIG,
 641                                 "couldn't get hugepage size through fstat\n");
 642                         goto err_mmap;
 643                 }
 644                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 645
 646                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 647                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 648
 649                 if (mmap_addr == MAP_FAILED) {
 650                         RTE_LOG(ERR, VHOST_CONFIG,
 651                                 "mmap region %u failed.\n", i);
 652                         goto err_mmap;
 653                 }
 654
 655                 reg->mmap_addr = mmap_addr;
 656                 reg->mmap_size = mmap_size;
 657                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 658                                       mmap_offset;
 659
 660                 if (dev->dequeue_zero_copy)
 661                         add_guest_pages(dev, reg, alignment);
 662
 663                 RTE_LOG(INFO, VHOST_CONFIG,
 664                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 665                         "\t guest physical addr: 0x%" PRIx64 "\n"
 666                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 667                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 668                         "\t mmap addr : 0x%" PRIx64 "\n"
 669                         "\t mmap size : 0x%" PRIx64 "\n"
 670                         "\t mmap align: 0x%" PRIx64 "\n"
 671                         "\t mmap off  : 0x%" PRIx64 "\n",
 672                         i, reg->size,
 673                         reg->guest_phys_addr,
 674                         reg->guest_user_addr,
 675                         reg->host_user_addr,
 676                         (uint64_t)(uintptr_t)mmap_addr,
 677                         mmap_size,
 678                         alignment,
 679                         mmap_offset);
 680         }
 681
 682         dump_guest_pages(dev);
 683
 684         return 0;
 685
 686 err_mmap:
 687         free_mem_region(dev);
 688         rte_free(dev->mem);
 689         dev->mem = NULL;
 690         return -1;
 691 }
 692
 693 static int
 694 vq_is_ready(struct vhost_virtqueue *vq)
 695 {
 696         return vq && vq->desc && vq->avail && vq->used &&
 697                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 698                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 699 }
 700
 701 static int
 702 virtio_is_ready(struct virtio_net *dev)
 703 {
 704         struct vhost_virtqueue *vq;
 705         uint32_t i;
 706
 707         if (dev->nr_vring == 0)
 708                 return 0;
 709
 710         for (i = 0; i < dev->nr_vring; i++) {
 711                 vq = dev->virtqueue[i];
 712
 713                 if (!vq_is_ready(vq))
 714                         return 0;
 715         }
 716
 717         RTE_LOG(INFO, VHOST_CONFIG,
 718                 "virtio is now ready for processing.\n");
 719         return 1;
 720 }
 721
 722 static void
 723 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 724 {
 725         struct vhost_vring_file file;
 726         struct vhost_virtqueue *vq;
 727
 728         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 729         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 730                 file.fd = VIRTIO_INVALID_EVENTFD;
 731         else
 732                 file.fd = pmsg->fds[0];
 733         RTE_LOG(INFO, VHOST_CONFIG,
 734                 "vring call idx:%d file:%d\n", file.index, file.fd);
 735
 736         vq = dev->virtqueue[file.index];
 737         if (vq->callfd >= 0)
 738                 close(vq->callfd);
 739
 740         vq->callfd = file.fd;
 741 }
 742
 743 static void
 744 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 745 {
 746         struct vhost_vring_file file;
 747         struct vhost_virtqueue *vq;
 748         struct virtio_net *dev = *pdev;
 749
 750         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 751         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 752                 file.fd = VIRTIO_INVALID_EVENTFD;
 753         else
 754                 file.fd = pmsg->fds[0];
 755         RTE_LOG(INFO, VHOST_CONFIG,
 756                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 757
 758         /* Interpret ring addresses only when ring is started. */
 759         dev = translate_ring_addresses(dev, file.index);
 760         if (!dev)
 761                 return;
 762
 763         *pdev = dev;
 764
 765         vq = dev->virtqueue[file.index];
 766
 767         /*
 768          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 769          * the ring starts already enabled. Otherwise, it is enabled via
 770          * the SET_VRING_ENABLE message.
 771          */
 772         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 773                 vq->enabled = 1;
 774
 775         if (vq->kickfd >= 0)
 776                 close(vq->kickfd);
 777         vq->kickfd = file.fd;
 778 }
 779
 780 static void
 781 free_zmbufs(struct vhost_virtqueue *vq)
 782 {
 783         struct zcopy_mbuf *zmbuf, *next;
 784
 785         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 786              zmbuf != NULL; zmbuf = next) {
 787                 next = TAILQ_NEXT(zmbuf, next);
 788
 789                 rte_pktmbuf_free(zmbuf->mbuf);
 790                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 791         }
 792
 793         rte_free(vq->zmbufs);
 794 }
 795
 796 /*
 797  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 798  */
 799 static int
 800 vhost_user_get_vring_base(struct virtio_net *dev,
 801                           VhostUserMsg *msg)
 802 {
 803         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 804
 805         /* We have to stop the queue (virtio) if it is running. */
 806         if (dev->flags & VIRTIO_DEV_RUNNING) {
 807                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 808                 dev->notify_ops->destroy_device(dev->vid);
 809         }
 810
 811         dev->flags &= ~VIRTIO_DEV_READY;
 812
 813         /* Here we are safe to get the last used index */
 814         msg->payload.state.num = vq->last_used_idx;
 815
 816         RTE_LOG(INFO, VHOST_CONFIG,
 817                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 818                 msg->payload.state.num);
 819         /*
 820          * Based on current qemu vhost-user implementation, this message is
 821          * sent and only sent in vhost_vring_stop.
 822          * TODO: cleanup the vring, it isn't usable since here.
 823          */
 824         if (vq->kickfd >= 0)
 825                 close(vq->kickfd);
 826
 827         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 828
 829         if (dev->dequeue_zero_copy)
 830                 free_zmbufs(vq);
 831         rte_free(vq->shadow_used_ring);
 832         vq->shadow_used_ring = NULL;
 833
 834         rte_free(vq->batch_copy_elems);
 835         vq->batch_copy_elems = NULL;
 836
 837         return 0;
 838 }
 839
 840 /*
 841  * when virtio queues are ready to work, qemu will send us to
 842  * enable the virtio queue pair.
 843  */
 844 static int
 845 vhost_user_set_vring_enable(struct virtio_net *dev,
 846                             VhostUserMsg *msg)
 847 {
 848         int enable = (int)msg->payload.state.num;
 849
 850         RTE_LOG(INFO, VHOST_CONFIG,
 851                 "set queue enable: %d to qp idx: %d\n",
 852                 enable, msg->payload.state.index);
 853
 854         if (dev->notify_ops->vring_state_changed)
 855                 dev->notify_ops->vring_state_changed(dev->vid,
 856                                 msg->payload.state.index, enable);
 857
 858         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 859
 860         return 0;
 861 }
 862
 863 static void
 864 vhost_user_set_protocol_features(struct virtio_net *dev,
 865                                  uint64_t protocol_features)
 866 {
 867         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 868                 return;
 869
 870         dev->protocol_features = protocol_features;
 871 }
 872
 873 static int
 874 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 875 {
 876         int fd = msg->fds[0];
 877         uint64_t size, off;
 878         void *addr;
 879
 880         if (fd < 0) {
 881                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 882                 return -1;
 883         }
 884
 885         if (msg->size != sizeof(VhostUserLog)) {
 886                 RTE_LOG(ERR, VHOST_CONFIG,
 887                         "invalid log base msg size: %"PRId32" != %d\n",
 888                         msg->size, (int)sizeof(VhostUserLog));
 889                 return -1;
 890         }
 891
 892         size = msg->payload.log.mmap_size;
 893         off  = msg->payload.log.mmap_offset;
 894         RTE_LOG(INFO, VHOST_CONFIG,
 895                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 896                 size, off);
 897
 898         /*
 899          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 900          * fail when offset is not page size aligned.
 901          */
 902         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 903         close(fd);
 904         if (addr == MAP_FAILED) {
 905                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 906                 return -1;
 907         }
 908
 909         /*
 910          * Free previously mapped log memory on occasionally
 911          * multiple VHOST_USER_SET_LOG_BASE.
 912          */
 913         if (dev->log_addr) {
 914                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 915         }
 916         dev->log_addr = (uint64_t)(uintptr_t)addr;
 917         dev->log_base = dev->log_addr + off;
 918         dev->log_size = size;
 919
 920         return 0;
 921 }
 922
 923 /*
 924  * An rarp packet is constructed and broadcasted to notify switches about
 925  * the new location of the migrated VM, so that packets from outside will
 926  * not be lost after migration.
 927  *
 928  * However, we don't actually "send" a rarp packet here, instead, we set
 929  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 930  */
 931 static int
 932 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 933 {
 934         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 935
 936         RTE_LOG(DEBUG, VHOST_CONFIG,
 937                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 938                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 939         memcpy(dev->mac.addr_bytes, mac, 6);
 940
 941         /*
 942          * Set the flag to inject a RARP broadcast packet at
 943          * rte_vhost_dequeue_burst().
 944          *
 945          * rte_smp_wmb() is for making sure the mac is copied
 946          * before the flag is set.
 947          */
 948         rte_smp_wmb();
 949         rte_atomic16_set(&dev->broadcast_rarp, 1);
 950
 951         return 0;
 952 }
 953
 954 static int
 955 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 956 {
 957         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 958                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 959                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 960                                 msg->payload.u64);
 961
 962                 return -1;
 963         }
 964
 965         dev->mtu = msg->payload.u64;
 966
 967         return 0;
 968 }
 969
 970 static int
 971 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 972 {
 973         int fd = msg->fds[0];
 974
 975         if (fd < 0) {
 976                 RTE_LOG(ERR, VHOST_CONFIG,
 977                                 "Invalid file descriptor for slave channel (%d)\n",
 978                                 fd);
 979                 return -1;
 980         }
 981
 982         dev->slave_req_fd = fd;
 983
 984         return 0;
 985 }
 986
 987 static int
 988 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
 989 {
 990         struct vhost_vring_addr *ra;
 991         uint64_t start, end;
 992
 993         start = imsg->iova;
 994         end = start + imsg->size;
 995
 996         ra = &vq->ring_addrs;
 997         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
 998                 return 1;
 999         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1000                 return 1;
1001         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1002                 return 1;
1003
1004         return 0;
1005 }
1006
1007 static int
1008 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1009                                 struct vhost_iotlb_msg *imsg)
1010 {
1011         uint64_t istart, iend, vstart, vend;
1012
1013         istart = imsg->iova;
1014         iend = istart + imsg->size - 1;
1015
1016         vstart = (uintptr_t)vq->desc;
1017         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1018         if (vstart <= iend && istart <= vend)
1019                 return 1;
1020
1021         vstart = (uintptr_t)vq->avail;
1022         vend = vstart + sizeof(struct vring_avail);
1023         vend += sizeof(uint16_t) * vq->size - 1;
1024         if (vstart <= iend && istart <= vend)
1025                 return 1;
1026
1027         vstart = (uintptr_t)vq->used;
1028         vend = vstart + sizeof(struct vring_used);
1029         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1030         if (vstart <= iend && istart <= vend)
1031                 return 1;
1032
1033         return 0;
1034 }
1035
1036 static int
1037 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1038 {
1039         struct virtio_net *dev = *pdev;
1040         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1041         uint16_t i;
1042         uint64_t vva;
1043
1044         switch (imsg->type) {
1045         case VHOST_IOTLB_UPDATE:
1046                 vva = qva_to_vva(dev, imsg->uaddr);
1047                 if (!vva)
1048                         return -1;
1049
1050                 for (i = 0; i < dev->nr_vring; i++) {
1051                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1052
1053                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1054                                         imsg->size, imsg->perm);
1055
1056                         if (is_vring_iotlb_update(vq, imsg))
1057                                 *pdev = dev = translate_ring_addresses(dev, i);
1058                 }
1059                 break;
1060         case VHOST_IOTLB_INVALIDATE:
1061                 for (i = 0; i < dev->nr_vring; i++) {
1062                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1063
1064                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1065                                         imsg->size);
1066
1067                         if (is_vring_iotlb_invalidate(vq, imsg))
1068                                 vring_invalidate(dev, vq);
1069                 }
1070                 break;
1071         default:
1072                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1073                                 imsg->type);
1074                 return -1;
1075         }
1076
1077         return 0;
1078 }
1079
1080 /* return bytes# of read on success or negative val on failure. */
1081 static int
1082 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1083 {
1084         int ret;
1085
1086         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1087                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1088         if (ret <= 0)
1089                 return ret;
1090
1091         if (msg && msg->size) {
1092                 if (msg->size > sizeof(msg->payload)) {
1093                         RTE_LOG(ERR, VHOST_CONFIG,
1094                                 "invalid msg size: %d\n", msg->size);
1095                         return -1;
1096                 }
1097                 ret = read(sockfd, &msg->payload, msg->size);
1098                 if (ret <= 0)
1099                         return ret;
1100                 if (ret != (int)msg->size) {
1101                         RTE_LOG(ERR, VHOST_CONFIG,
1102                                 "read control message failed\n");
1103                         return -1;
1104                 }
1105         }
1106
1107         return ret;
1108 }
1109
1110 static int
1111 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1112 {
1113         if (!msg)
1114                 return 0;
1115
1116         return send_fd_message(sockfd, (char *)msg,
1117                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1118 }
1119
1120 static int
1121 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1122 {
1123         if (!msg)
1124                 return 0;
1125
1126         msg->flags &= ~VHOST_USER_VERSION_MASK;
1127         msg->flags &= ~VHOST_USER_NEED_REPLY;
1128         msg->flags |= VHOST_USER_VERSION;
1129         msg->flags |= VHOST_USER_REPLY_MASK;
1130
1131         return send_vhost_message(sockfd, msg);
1132 }
1133
1134 /*
1135  * Allocate a queue pair if it hasn't been allocated yet
1136  */
1137 static int
1138 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1139 {
1140         uint16_t vring_idx;
1141
1142         switch (msg->request.master) {
1143         case VHOST_USER_SET_VRING_KICK:
1144         case VHOST_USER_SET_VRING_CALL:
1145         case VHOST_USER_SET_VRING_ERR:
1146                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1147                 break;
1148         case VHOST_USER_SET_VRING_NUM:
1149         case VHOST_USER_SET_VRING_BASE:
1150         case VHOST_USER_SET_VRING_ENABLE:
1151                 vring_idx = msg->payload.state.index;
1152                 break;
1153         case VHOST_USER_SET_VRING_ADDR:
1154                 vring_idx = msg->payload.addr.index;
1155                 break;
1156         default:
1157                 return 0;
1158         }
1159
1160         if (vring_idx >= VHOST_MAX_VRING) {
1161                 RTE_LOG(ERR, VHOST_CONFIG,
1162                         "invalid vring index: %u\n", vring_idx);
1163                 return -1;
1164         }
1165
1166         if (dev->virtqueue[vring_idx])
1167                 return 0;
1168
1169         return alloc_vring_queue(dev, vring_idx);
1170 }
1171
1172 int
1173 vhost_user_msg_handler(int vid, int fd)
1174 {
1175         struct virtio_net *dev;
1176         struct VhostUserMsg msg;
1177         int ret;
1178
1179         dev = get_device(vid);
1180         if (dev == NULL)
1181                 return -1;
1182
1183         if (!dev->notify_ops) {
1184                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1185                 if (!dev->notify_ops) {
1186                         RTE_LOG(ERR, VHOST_CONFIG,
1187                                 "failed to get callback ops for driver %s\n",
1188                                 dev->ifname);
1189                         return -1;
1190                 }
1191         }
1192
1193         ret = read_vhost_message(fd, &msg);
1194         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1195                 if (ret < 0)
1196                         RTE_LOG(ERR, VHOST_CONFIG,
1197                                 "vhost read message failed\n");
1198                 else if (ret == 0)
1199                         RTE_LOG(INFO, VHOST_CONFIG,
1200                                 "vhost peer closed\n");
1201                 else
1202                         RTE_LOG(ERR, VHOST_CONFIG,
1203                                 "vhost read incorrect message\n");
1204
1205                 return -1;
1206         }
1207
1208         ret = 0;
1209         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1210                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1211                         vhost_message_str[msg.request.master]);
1212         else
1213                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1214                         vhost_message_str[msg.request.master]);
1215
1216         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1217         if (ret < 0) {
1218                 RTE_LOG(ERR, VHOST_CONFIG,
1219                         "failed to alloc queue\n");
1220                 return -1;
1221         }
1222
1223         switch (msg.request.master) {
1224         case VHOST_USER_GET_FEATURES:
1225                 msg.payload.u64 = vhost_user_get_features(dev);
1226                 msg.size = sizeof(msg.payload.u64);
1227                 send_vhost_reply(fd, &msg);
1228                 break;
1229         case VHOST_USER_SET_FEATURES:
1230                 vhost_user_set_features(dev, msg.payload.u64);
1231                 break;
1232
1233         case VHOST_USER_GET_PROTOCOL_FEATURES:
1234                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1235                 msg.size = sizeof(msg.payload.u64);
1236                 send_vhost_reply(fd, &msg);
1237                 break;
1238         case VHOST_USER_SET_PROTOCOL_FEATURES:
1239                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1240                 break;
1241
1242         case VHOST_USER_SET_OWNER:
1243                 vhost_user_set_owner();
1244                 break;
1245         case VHOST_USER_RESET_OWNER:
1246                 vhost_user_reset_owner(dev);
1247                 break;
1248
1249         case VHOST_USER_SET_MEM_TABLE:
1250                 ret = vhost_user_set_mem_table(dev, &msg);
1251                 break;
1252
1253         case VHOST_USER_SET_LOG_BASE:
1254                 vhost_user_set_log_base(dev, &msg);
1255
1256                 /* it needs a reply */
1257                 msg.size = sizeof(msg.payload.u64);
1258                 send_vhost_reply(fd, &msg);
1259                 break;
1260         case VHOST_USER_SET_LOG_FD:
1261                 close(msg.fds[0]);
1262                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1263                 break;
1264
1265         case VHOST_USER_SET_VRING_NUM:
1266                 vhost_user_set_vring_num(dev, &msg);
1267                 break;
1268         case VHOST_USER_SET_VRING_ADDR:
1269                 vhost_user_set_vring_addr(&dev, &msg);
1270                 break;
1271         case VHOST_USER_SET_VRING_BASE:
1272                 vhost_user_set_vring_base(dev, &msg);
1273                 break;
1274
1275         case VHOST_USER_GET_VRING_BASE:
1276                 vhost_user_get_vring_base(dev, &msg);
1277                 msg.size = sizeof(msg.payload.state);
1278                 send_vhost_reply(fd, &msg);
1279                 break;
1280
1281         case VHOST_USER_SET_VRING_KICK:
1282                 vhost_user_set_vring_kick(&dev, &msg);
1283                 break;
1284         case VHOST_USER_SET_VRING_CALL:
1285                 vhost_user_set_vring_call(dev, &msg);
1286                 break;
1287
1288         case VHOST_USER_SET_VRING_ERR:
1289                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1290                         close(msg.fds[0]);
1291                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1292                 break;
1293
1294         case VHOST_USER_GET_QUEUE_NUM:
1295                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1296                 msg.size = sizeof(msg.payload.u64);
1297                 send_vhost_reply(fd, &msg);
1298                 break;
1299
1300         case VHOST_USER_SET_VRING_ENABLE:
1301                 vhost_user_set_vring_enable(dev, &msg);
1302                 break;
1303         case VHOST_USER_SEND_RARP:
1304                 vhost_user_send_rarp(dev, &msg);
1305                 break;
1306
1307         case VHOST_USER_NET_SET_MTU:
1308                 ret = vhost_user_net_set_mtu(dev, &msg);
1309                 break;
1310
1311         case VHOST_USER_SET_SLAVE_REQ_FD:
1312                 ret = vhost_user_set_req_fd(dev, &msg);
1313                 break;
1314
1315         case VHOST_USER_IOTLB_MSG:
1316                 ret = vhost_user_iotlb_msg(&dev, &msg);
1317                 break;
1318
1319         default:
1320                 ret = -1;
1321                 break;
1322
1323         }
1324
1325         if (msg.flags & VHOST_USER_NEED_REPLY) {
1326                 msg.payload.u64 = !!ret;
1327                 msg.size = sizeof(msg.payload.u64);
1328                 send_vhost_reply(fd, &msg);
1329         }
1330
1331         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1332                 dev->flags |= VIRTIO_DEV_READY;
1333
1334                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1335                         if (dev->dequeue_zero_copy) {
1336                                 RTE_LOG(INFO, VHOST_CONFIG,
1337                                                 "dequeue zero copy is enabled\n");
1338                         }
1339
1340                         if (dev->notify_ops->new_device(dev->vid) == 0)
1341                                 dev->flags |= VIRTIO_DEV_RUNNING;
1342                 }
1343         }
1344
1345         return 0;
1346 }
1347
1348 int
1349 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1350 {
1351         int ret;
1352         struct VhostUserMsg msg = {
1353                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1354                 .flags = VHOST_USER_VERSION,
1355                 .size = sizeof(msg.payload.iotlb),
1356                 .payload.iotlb = {
1357                         .iova = iova,
1358                         .perm = perm,
1359                         .type = VHOST_IOTLB_MISS,
1360                 },
1361         };
1362
1363         ret = send_vhost_message(dev->slave_req_fd, &msg);
1364         if (ret < 0) {
1365                 RTE_LOG(ERR, VHOST_CONFIG,
1366                                 "Failed to send IOTLB miss message (%d)\n",
1367                                 ret);
1368                 return ret;
1369         }
1370
1371         return 0;
1372 }