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         /*
 759          * Interpret ring addresses only when ring is started and enabled.
 760          * This is now if protocol features aren't supported.
 761          */
 762         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 763                 *pdev = dev = translate_ring_addresses(dev, file.index);
 764                 if (!dev)
 765                         return;
 766         }
 767
 768         vq = dev->virtqueue[file.index];
 769
 770         /*
 771          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 772          * the ring starts already enabled. Otherwise, it is enabled via
 773          * the SET_VRING_ENABLE message.
 774          */
 775         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 776                 vq->enabled = 1;
 777
 778         if (vq->kickfd >= 0)
 779                 close(vq->kickfd);
 780         vq->kickfd = file.fd;
 781 }
 782
 783 static void
 784 free_zmbufs(struct vhost_virtqueue *vq)
 785 {
 786         struct zcopy_mbuf *zmbuf, *next;
 787
 788         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 789              zmbuf != NULL; zmbuf = next) {
 790                 next = TAILQ_NEXT(zmbuf, next);
 791
 792                 rte_pktmbuf_free(zmbuf->mbuf);
 793                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 794         }
 795
 796         rte_free(vq->zmbufs);
 797 }
 798
 799 /*
 800  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 801  */
 802 static int
 803 vhost_user_get_vring_base(struct virtio_net *dev,
 804                           VhostUserMsg *msg)
 805 {
 806         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 807
 808         /* We have to stop the queue (virtio) if it is running. */
 809         if (dev->flags & VIRTIO_DEV_RUNNING) {
 810                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 811                 dev->notify_ops->destroy_device(dev->vid);
 812         }
 813
 814         dev->flags &= ~VIRTIO_DEV_READY;
 815
 816         /* Here we are safe to get the last used index */
 817         msg->payload.state.num = vq->last_used_idx;
 818
 819         RTE_LOG(INFO, VHOST_CONFIG,
 820                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 821                 msg->payload.state.num);
 822         /*
 823          * Based on current qemu vhost-user implementation, this message is
 824          * sent and only sent in vhost_vring_stop.
 825          * TODO: cleanup the vring, it isn't usable since here.
 826          */
 827         if (vq->kickfd >= 0)
 828                 close(vq->kickfd);
 829
 830         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 831
 832         if (dev->dequeue_zero_copy)
 833                 free_zmbufs(vq);
 834         rte_free(vq->shadow_used_ring);
 835         vq->shadow_used_ring = NULL;
 836
 837         rte_free(vq->batch_copy_elems);
 838         vq->batch_copy_elems = NULL;
 839
 840         return 0;
 841 }
 842
 843 /*
 844  * when virtio queues are ready to work, qemu will send us to
 845  * enable the virtio queue pair.
 846  */
 847 static int
 848 vhost_user_set_vring_enable(struct virtio_net **pdev,
 849                             VhostUserMsg *msg)
 850 {
 851         struct virtio_net *dev = *pdev;
 852         int enable = (int)msg->payload.state.num;
 853
 854         RTE_LOG(INFO, VHOST_CONFIG,
 855                 "set queue enable: %d to qp idx: %d\n",
 856                 enable, msg->payload.state.index);
 857
 858         /*
 859          * Interpret ring addresses only when ring is started and enabled.
 860          * This is now if protocol features are supported.
 861          */
 862         if (enable && (dev->features &
 863                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 864                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 865                 if (!dev)
 866                         return -1;
 867
 868                 *pdev = dev;
 869         }
 870
 871         if (dev->notify_ops->vring_state_changed)
 872                 dev->notify_ops->vring_state_changed(dev->vid,
 873                                 msg->payload.state.index, enable);
 874
 875         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 876
 877         return 0;
 878 }
 879
 880 static void
 881 vhost_user_set_protocol_features(struct virtio_net *dev,
 882                                  uint64_t protocol_features)
 883 {
 884         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 885                 return;
 886
 887         dev->protocol_features = protocol_features;
 888 }
 889
 890 static int
 891 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 892 {
 893         int fd = msg->fds[0];
 894         uint64_t size, off;
 895         void *addr;
 896
 897         if (fd < 0) {
 898                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 899                 return -1;
 900         }
 901
 902         if (msg->size != sizeof(VhostUserLog)) {
 903                 RTE_LOG(ERR, VHOST_CONFIG,
 904                         "invalid log base msg size: %"PRId32" != %d\n",
 905                         msg->size, (int)sizeof(VhostUserLog));
 906                 return -1;
 907         }
 908
 909         size = msg->payload.log.mmap_size;
 910         off  = msg->payload.log.mmap_offset;
 911         RTE_LOG(INFO, VHOST_CONFIG,
 912                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 913                 size, off);
 914
 915         /*
 916          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 917          * fail when offset is not page size aligned.
 918          */
 919         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 920         close(fd);
 921         if (addr == MAP_FAILED) {
 922                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 923                 return -1;
 924         }
 925
 926         /*
 927          * Free previously mapped log memory on occasionally
 928          * multiple VHOST_USER_SET_LOG_BASE.
 929          */
 930         if (dev->log_addr) {
 931                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 932         }
 933         dev->log_addr = (uint64_t)(uintptr_t)addr;
 934         dev->log_base = dev->log_addr + off;
 935         dev->log_size = size;
 936
 937         return 0;
 938 }
 939
 940 /*
 941  * An rarp packet is constructed and broadcasted to notify switches about
 942  * the new location of the migrated VM, so that packets from outside will
 943  * not be lost after migration.
 944  *
 945  * However, we don't actually "send" a rarp packet here, instead, we set
 946  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 947  */
 948 static int
 949 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 950 {
 951         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 952
 953         RTE_LOG(DEBUG, VHOST_CONFIG,
 954                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 955                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 956         memcpy(dev->mac.addr_bytes, mac, 6);
 957
 958         /*
 959          * Set the flag to inject a RARP broadcast packet at
 960          * rte_vhost_dequeue_burst().
 961          *
 962          * rte_smp_wmb() is for making sure the mac is copied
 963          * before the flag is set.
 964          */
 965         rte_smp_wmb();
 966         rte_atomic16_set(&dev->broadcast_rarp, 1);
 967
 968         return 0;
 969 }
 970
 971 static int
 972 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 973 {
 974         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 975                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 976                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 977                                 msg->payload.u64);
 978
 979                 return -1;
 980         }
 981
 982         dev->mtu = msg->payload.u64;
 983
 984         return 0;
 985 }
 986
 987 static int
 988 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 989 {
 990         int fd = msg->fds[0];
 991
 992         if (fd < 0) {
 993                 RTE_LOG(ERR, VHOST_CONFIG,
 994                                 "Invalid file descriptor for slave channel (%d)\n",
 995                                 fd);
 996                 return -1;
 997         }
 998
 999         dev->slave_req_fd = fd;
1000
1001         return 0;
1002 }
1003
1004 static int
1005 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1006 {
1007         struct vhost_vring_addr *ra;
1008         uint64_t start, end;
1009
1010         start = imsg->iova;
1011         end = start + imsg->size;
1012
1013         ra = &vq->ring_addrs;
1014         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1015                 return 1;
1016         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1017                 return 1;
1018         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1019                 return 1;
1020
1021         return 0;
1022 }
1023
1024 static int
1025 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1026                                 struct vhost_iotlb_msg *imsg)
1027 {
1028         uint64_t istart, iend, vstart, vend;
1029
1030         istart = imsg->iova;
1031         iend = istart + imsg->size - 1;
1032
1033         vstart = (uintptr_t)vq->desc;
1034         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1035         if (vstart <= iend && istart <= vend)
1036                 return 1;
1037
1038         vstart = (uintptr_t)vq->avail;
1039         vend = vstart + sizeof(struct vring_avail);
1040         vend += sizeof(uint16_t) * vq->size - 1;
1041         if (vstart <= iend && istart <= vend)
1042                 return 1;
1043
1044         vstart = (uintptr_t)vq->used;
1045         vend = vstart + sizeof(struct vring_used);
1046         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1047         if (vstart <= iend && istart <= vend)
1048                 return 1;
1049
1050         return 0;
1051 }
1052
1053 static int
1054 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1055 {
1056         struct virtio_net *dev = *pdev;
1057         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1058         uint16_t i;
1059         uint64_t vva;
1060
1061         switch (imsg->type) {
1062         case VHOST_IOTLB_UPDATE:
1063                 vva = qva_to_vva(dev, imsg->uaddr);
1064                 if (!vva)
1065                         return -1;
1066
1067                 for (i = 0; i < dev->nr_vring; i++) {
1068                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1069
1070                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1071                                         imsg->size, imsg->perm);
1072
1073                         if (is_vring_iotlb_update(vq, imsg))
1074                                 *pdev = dev = translate_ring_addresses(dev, i);
1075                 }
1076                 break;
1077         case VHOST_IOTLB_INVALIDATE:
1078                 for (i = 0; i < dev->nr_vring; i++) {
1079                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1080
1081                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1082                                         imsg->size);
1083
1084                         if (is_vring_iotlb_invalidate(vq, imsg))
1085                                 vring_invalidate(dev, vq);
1086                 }
1087                 break;
1088         default:
1089                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1090                                 imsg->type);
1091                 return -1;
1092         }
1093
1094         return 0;
1095 }
1096
1097 /* return bytes# of read on success or negative val on failure. */
1098 static int
1099 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1100 {
1101         int ret;
1102
1103         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1104                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1105         if (ret <= 0)
1106                 return ret;
1107
1108         if (msg && msg->size) {
1109                 if (msg->size > sizeof(msg->payload)) {
1110                         RTE_LOG(ERR, VHOST_CONFIG,
1111                                 "invalid msg size: %d\n", msg->size);
1112                         return -1;
1113                 }
1114                 ret = read(sockfd, &msg->payload, msg->size);
1115                 if (ret <= 0)
1116                         return ret;
1117                 if (ret != (int)msg->size) {
1118                         RTE_LOG(ERR, VHOST_CONFIG,
1119                                 "read control message failed\n");
1120                         return -1;
1121                 }
1122         }
1123
1124         return ret;
1125 }
1126
1127 static int
1128 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1129 {
1130         if (!msg)
1131                 return 0;
1132
1133         return send_fd_message(sockfd, (char *)msg,
1134                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1135 }
1136
1137 static int
1138 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1139 {
1140         if (!msg)
1141                 return 0;
1142
1143         msg->flags &= ~VHOST_USER_VERSION_MASK;
1144         msg->flags &= ~VHOST_USER_NEED_REPLY;
1145         msg->flags |= VHOST_USER_VERSION;
1146         msg->flags |= VHOST_USER_REPLY_MASK;
1147
1148         return send_vhost_message(sockfd, msg);
1149 }
1150
1151 /*
1152  * Allocate a queue pair if it hasn't been allocated yet
1153  */
1154 static int
1155 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1156 {
1157         uint16_t vring_idx;
1158
1159         switch (msg->request.master) {
1160         case VHOST_USER_SET_VRING_KICK:
1161         case VHOST_USER_SET_VRING_CALL:
1162         case VHOST_USER_SET_VRING_ERR:
1163                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1164                 break;
1165         case VHOST_USER_SET_VRING_NUM:
1166         case VHOST_USER_SET_VRING_BASE:
1167         case VHOST_USER_SET_VRING_ENABLE:
1168                 vring_idx = msg->payload.state.index;
1169                 break;
1170         case VHOST_USER_SET_VRING_ADDR:
1171                 vring_idx = msg->payload.addr.index;
1172                 break;
1173         default:
1174                 return 0;
1175         }
1176
1177         if (vring_idx >= VHOST_MAX_VRING) {
1178                 RTE_LOG(ERR, VHOST_CONFIG,
1179                         "invalid vring index: %u\n", vring_idx);
1180                 return -1;
1181         }
1182
1183         if (dev->virtqueue[vring_idx])
1184                 return 0;
1185
1186         return alloc_vring_queue(dev, vring_idx);
1187 }
1188
1189 int
1190 vhost_user_msg_handler(int vid, int fd)
1191 {
1192         struct virtio_net *dev;
1193         struct VhostUserMsg msg;
1194         int ret;
1195
1196         dev = get_device(vid);
1197         if (dev == NULL)
1198                 return -1;
1199
1200         if (!dev->notify_ops) {
1201                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1202                 if (!dev->notify_ops) {
1203                         RTE_LOG(ERR, VHOST_CONFIG,
1204                                 "failed to get callback ops for driver %s\n",
1205                                 dev->ifname);
1206                         return -1;
1207                 }
1208         }
1209
1210         ret = read_vhost_message(fd, &msg);
1211         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1212                 if (ret < 0)
1213                         RTE_LOG(ERR, VHOST_CONFIG,
1214                                 "vhost read message failed\n");
1215                 else if (ret == 0)
1216                         RTE_LOG(INFO, VHOST_CONFIG,
1217                                 "vhost peer closed\n");
1218                 else
1219                         RTE_LOG(ERR, VHOST_CONFIG,
1220                                 "vhost read incorrect message\n");
1221
1222                 return -1;
1223         }
1224
1225         ret = 0;
1226         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1227                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1228                         vhost_message_str[msg.request.master]);
1229         else
1230                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1231                         vhost_message_str[msg.request.master]);
1232
1233         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1234         if (ret < 0) {
1235                 RTE_LOG(ERR, VHOST_CONFIG,
1236                         "failed to alloc queue\n");
1237                 return -1;
1238         }
1239
1240         switch (msg.request.master) {
1241         case VHOST_USER_GET_FEATURES:
1242                 msg.payload.u64 = vhost_user_get_features(dev);
1243                 msg.size = sizeof(msg.payload.u64);
1244                 send_vhost_reply(fd, &msg);
1245                 break;
1246         case VHOST_USER_SET_FEATURES:
1247                 vhost_user_set_features(dev, msg.payload.u64);
1248                 break;
1249
1250         case VHOST_USER_GET_PROTOCOL_FEATURES:
1251                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1252                 msg.size = sizeof(msg.payload.u64);
1253                 send_vhost_reply(fd, &msg);
1254                 break;
1255         case VHOST_USER_SET_PROTOCOL_FEATURES:
1256                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1257                 break;
1258
1259         case VHOST_USER_SET_OWNER:
1260                 vhost_user_set_owner();
1261                 break;
1262         case VHOST_USER_RESET_OWNER:
1263                 vhost_user_reset_owner(dev);
1264                 break;
1265
1266         case VHOST_USER_SET_MEM_TABLE:
1267                 ret = vhost_user_set_mem_table(dev, &msg);
1268                 break;
1269
1270         case VHOST_USER_SET_LOG_BASE:
1271                 vhost_user_set_log_base(dev, &msg);
1272
1273                 /* it needs a reply */
1274                 msg.size = sizeof(msg.payload.u64);
1275                 send_vhost_reply(fd, &msg);
1276                 break;
1277         case VHOST_USER_SET_LOG_FD:
1278                 close(msg.fds[0]);
1279                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1280                 break;
1281
1282         case VHOST_USER_SET_VRING_NUM:
1283                 vhost_user_set_vring_num(dev, &msg);
1284                 break;
1285         case VHOST_USER_SET_VRING_ADDR:
1286                 vhost_user_set_vring_addr(&dev, &msg);
1287                 break;
1288         case VHOST_USER_SET_VRING_BASE:
1289                 vhost_user_set_vring_base(dev, &msg);
1290                 break;
1291
1292         case VHOST_USER_GET_VRING_BASE:
1293                 vhost_user_get_vring_base(dev, &msg);
1294                 msg.size = sizeof(msg.payload.state);
1295                 send_vhost_reply(fd, &msg);
1296                 break;
1297
1298         case VHOST_USER_SET_VRING_KICK:
1299                 vhost_user_set_vring_kick(&dev, &msg);
1300                 break;
1301         case VHOST_USER_SET_VRING_CALL:
1302                 vhost_user_set_vring_call(dev, &msg);
1303                 break;
1304
1305         case VHOST_USER_SET_VRING_ERR:
1306                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1307                         close(msg.fds[0]);
1308                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1309                 break;
1310
1311         case VHOST_USER_GET_QUEUE_NUM:
1312                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1313                 msg.size = sizeof(msg.payload.u64);
1314                 send_vhost_reply(fd, &msg);
1315                 break;
1316
1317         case VHOST_USER_SET_VRING_ENABLE:
1318                 vhost_user_set_vring_enable(&dev, &msg);
1319                 break;
1320         case VHOST_USER_SEND_RARP:
1321                 vhost_user_send_rarp(dev, &msg);
1322                 break;
1323
1324         case VHOST_USER_NET_SET_MTU:
1325                 ret = vhost_user_net_set_mtu(dev, &msg);
1326                 break;
1327
1328         case VHOST_USER_SET_SLAVE_REQ_FD:
1329                 ret = vhost_user_set_req_fd(dev, &msg);
1330                 break;
1331
1332         case VHOST_USER_IOTLB_MSG:
1333                 ret = vhost_user_iotlb_msg(&dev, &msg);
1334                 break;
1335
1336         default:
1337                 ret = -1;
1338                 break;
1339
1340         }
1341
1342         if (msg.flags & VHOST_USER_NEED_REPLY) {
1343                 msg.payload.u64 = !!ret;
1344                 msg.size = sizeof(msg.payload.u64);
1345                 send_vhost_reply(fd, &msg);
1346         }
1347
1348         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1349                 dev->flags |= VIRTIO_DEV_READY;
1350
1351                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1352                         if (dev->dequeue_zero_copy) {
1353                                 RTE_LOG(INFO, VHOST_CONFIG,
1354                                                 "dequeue zero copy is enabled\n");
1355                         }
1356
1357                         if (dev->notify_ops->new_device(dev->vid) == 0)
1358                                 dev->flags |= VIRTIO_DEV_RUNNING;
1359                 }
1360         }
1361
1362         return 0;
1363 }
1364
1365 int
1366 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1367 {
1368         int ret;
1369         struct VhostUserMsg msg = {
1370                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1371                 .flags = VHOST_USER_VERSION,
1372                 .size = sizeof(msg.payload.iotlb),
1373                 .payload.iotlb = {
1374                         .iova = iova,
1375                         .perm = perm,
1376                         .type = VHOST_IOTLB_MISS,
1377                 },
1378         };
1379
1380         ret = send_vhost_message(dev->slave_req_fd, &msg);
1381         if (ret < 0) {
1382                 RTE_LOG(ERR, VHOST_CONFIG,
1383                                 "Failed to send IOTLB miss message (%d)\n",
1384                                 ret);
1385                 return ret;
1386         }
1387
1388         return 0;
1389 }