lib/librte_vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <string.h>
   9 #include <unistd.h>
  10 #include <sys/mman.h>
  11 #include <sys/types.h>
  12 #include <sys/stat.h>
  13 #include <assert.h>
  14 #ifdef RTE_LIBRTE_VHOST_NUMA
  15 #include <numaif.h>
  16 #endif
  17
  18 #include <rte_common.h>
  19 #include <rte_malloc.h>
  20 #include <rte_log.h>
  21
  22 #include "iotlb.h"
  23 #include "vhost.h"
  24 #include "vhost_user.h"
  25
  26 #define VIRTIO_MIN_MTU 68
  27 #define VIRTIO_MAX_MTU 65535
  28
  29 static const char *vhost_message_str[VHOST_USER_MAX] = {
  30         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  31         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  32         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  33         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  34         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  35         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  36         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  37         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  38         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  39         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  40         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  41         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  42         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  43         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  44         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  45         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  46         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  47         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  48         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  49         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  50         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  51         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  52         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  53 };
  54
  55 static uint64_t
  56 get_blk_size(int fd)
  57 {
  58         struct stat stat;
  59         int ret;
  60
  61         ret = fstat(fd, &stat);
  62         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  63 }
  64
  65 static void
  66 free_mem_region(struct virtio_net *dev)
  67 {
  68         uint32_t i;
  69         struct rte_vhost_mem_region *reg;
  70
  71         if (!dev || !dev->mem)
  72                 return;
  73
  74         for (i = 0; i < dev->mem->nregions; i++) {
  75                 reg = &dev->mem->regions[i];
  76                 if (reg->host_user_addr) {
  77                         munmap(reg->mmap_addr, reg->mmap_size);
  78                         close(reg->fd);
  79                 }
  80         }
  81 }
  82
  83 void
  84 vhost_backend_cleanup(struct virtio_net *dev)
  85 {
  86         if (dev->mem) {
  87                 free_mem_region(dev);
  88                 rte_free(dev->mem);
  89                 dev->mem = NULL;
  90         }
  91
  92         free(dev->guest_pages);
  93         dev->guest_pages = NULL;
  94
  95         if (dev->log_addr) {
  96                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
  97                 dev->log_addr = 0;
  98         }
  99
 100         if (dev->slave_req_fd >= 0) {
 101                 close(dev->slave_req_fd);
 102                 dev->slave_req_fd = -1;
 103         }
 104 }
 105
 106 /*
 107  * This function just returns success at the moment unless
 108  * the device hasn't been initialised.
 109  */
 110 static int
 111 vhost_user_set_owner(void)
 112 {
 113         return 0;
 114 }
 115
 116 static int
 117 vhost_user_reset_owner(struct virtio_net *dev)
 118 {
 119         if (dev->flags & VIRTIO_DEV_RUNNING) {
 120                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 121                 dev->notify_ops->destroy_device(dev->vid);
 122         }
 123
 124         cleanup_device(dev, 0);
 125         reset_device(dev);
 126         return 0;
 127 }
 128
 129 /*
 130  * The features that we support are requested.
 131  */
 132 static uint64_t
 133 vhost_user_get_features(struct virtio_net *dev)
 134 {
 135         uint64_t features = 0;
 136
 137         rte_vhost_driver_get_features(dev->ifname, &features);
 138         return features;
 139 }
 140
 141 /*
 142  * We receive the negotiated features supported by us and the virtio device.
 143  */
 144 static int
 145 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 146 {
 147         uint64_t vhost_features = 0;
 148
 149         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 150         if (features & ~vhost_features) {
 151                 RTE_LOG(ERR, VHOST_CONFIG,
 152                         "(%d) received invalid negotiated features.\n",
 153                         dev->vid);
 154                 return -1;
 155         }
 156
 157         if (dev->flags & VIRTIO_DEV_RUNNING) {
 158                 if (dev->features == features)
 159                         return 0;
 160
 161                 /*
 162                  * Error out if master tries to change features while device is
 163                  * in running state. The exception being VHOST_F_LOG_ALL, which
 164                  * is enabled when the live-migration starts.
 165                  */
 166                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 167                         RTE_LOG(ERR, VHOST_CONFIG,
 168                                 "(%d) features changed while device is running.\n",
 169                                 dev->vid);
 170                         return -1;
 171                 }
 172
 173                 if (dev->notify_ops->features_changed)
 174                         dev->notify_ops->features_changed(dev->vid, features);
 175         }
 176
 177         dev->features = features;
 178         if (dev->features &
 179                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 180                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 181         } else {
 182                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 183         }
 184         LOG_DEBUG(VHOST_CONFIG,
 185                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 186                 dev->vid,
 187                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 188                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 189
 190         return 0;
 191 }
 192
 193 /*
 194  * The virtio device sends us the size of the descriptor ring.
 195  */
 196 static int
 197 vhost_user_set_vring_num(struct virtio_net *dev,
 198                          VhostUserMsg *msg)
 199 {
 200         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 201
 202         vq->size = msg->payload.state.num;
 203
 204         if (dev->dequeue_zero_copy) {
 205                 vq->nr_zmbuf = 0;
 206                 vq->last_zmbuf_idx = 0;
 207                 vq->zmbuf_size = vq->size;
 208                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 209                                          sizeof(struct zcopy_mbuf), 0);
 210                 if (vq->zmbufs == NULL) {
 211                         RTE_LOG(WARNING, VHOST_CONFIG,
 212                                 "failed to allocate mem for zero copy; "
 213                                 "zero copy is force disabled\n");
 214                         dev->dequeue_zero_copy = 0;
 215                 }
 216         }
 217
 218         vq->shadow_used_ring = rte_malloc(NULL,
 219                                 vq->size * sizeof(struct vring_used_elem),
 220                                 RTE_CACHE_LINE_SIZE);
 221         if (!vq->shadow_used_ring) {
 222                 RTE_LOG(ERR, VHOST_CONFIG,
 223                         "failed to allocate memory for shadow used ring.\n");
 224                 return -1;
 225         }
 226
 227         vq->batch_copy_elems = rte_malloc(NULL,
 228                                 vq->size * sizeof(struct batch_copy_elem),
 229                                 RTE_CACHE_LINE_SIZE);
 230         if (!vq->batch_copy_elems) {
 231                 RTE_LOG(ERR, VHOST_CONFIG,
 232                         "failed to allocate memory for batching copy.\n");
 233                 return -1;
 234         }
 235
 236         return 0;
 237 }
 238
 239 /*
 240  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 241  * same numa node as the memory of vring descriptor.
 242  */
 243 #ifdef RTE_LIBRTE_VHOST_NUMA
 244 static struct virtio_net*
 245 numa_realloc(struct virtio_net *dev, int index)
 246 {
 247         int oldnode, newnode;
 248         struct virtio_net *old_dev;
 249         struct vhost_virtqueue *old_vq, *vq;
 250         int ret;
 251
 252         old_dev = dev;
 253         vq = old_vq = dev->virtqueue[index];
 254
 255         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 256                             MPOL_F_NODE | MPOL_F_ADDR);
 257
 258         /* check if we need to reallocate vq */
 259         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 260                              MPOL_F_NODE | MPOL_F_ADDR);
 261         if (ret) {
 262                 RTE_LOG(ERR, VHOST_CONFIG,
 263                         "Unable to get vq numa information.\n");
 264                 return dev;
 265         }
 266         if (oldnode != newnode) {
 267                 RTE_LOG(INFO, VHOST_CONFIG,
 268                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 269                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 270                 if (!vq)
 271                         return dev;
 272
 273                 memcpy(vq, old_vq, sizeof(*vq));
 274                 rte_free(old_vq);
 275         }
 276
 277         /* check if we need to reallocate dev */
 278         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 279                             MPOL_F_NODE | MPOL_F_ADDR);
 280         if (ret) {
 281                 RTE_LOG(ERR, VHOST_CONFIG,
 282                         "Unable to get dev numa information.\n");
 283                 goto out;
 284         }
 285         if (oldnode != newnode) {
 286                 RTE_LOG(INFO, VHOST_CONFIG,
 287                         "reallocate dev from %d to %d node\n",
 288                         oldnode, newnode);
 289                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 290                 if (!dev) {
 291                         dev = old_dev;
 292                         goto out;
 293                 }
 294
 295                 memcpy(dev, old_dev, sizeof(*dev));
 296                 rte_free(old_dev);
 297         }
 298
 299 out:
 300         dev->virtqueue[index] = vq;
 301         vhost_devices[dev->vid] = dev;
 302
 303         if (old_vq != vq)
 304                 vhost_user_iotlb_init(dev, index);
 305
 306         return dev;
 307 }
 308 #else
 309 static struct virtio_net*
 310 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 311 {
 312         return dev;
 313 }
 314 #endif
 315
 316 /* Converts QEMU virtual address to Vhost virtual address. */
 317 static uint64_t
 318 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 319 {
 320         struct rte_vhost_mem_region *reg;
 321         uint32_t i;
 322
 323         /* Find the region where the address lives. */
 324         for (i = 0; i < dev->mem->nregions; i++) {
 325                 reg = &dev->mem->regions[i];
 326
 327                 if (qva >= reg->guest_user_addr &&
 328                     qva <  reg->guest_user_addr + reg->size) {
 329                         return qva - reg->guest_user_addr +
 330                                reg->host_user_addr;
 331                 }
 332         }
 333
 334         return 0;
 335 }
 336
 337
 338 /*
 339  * Converts ring address to Vhost virtual address.
 340  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 341  * else it is a QEMU virtual address.
 342  */
 343 static uint64_t
 344 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 345                 uint64_t ra, uint64_t size)
 346 {
 347         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 348                 uint64_t vva;
 349
 350                 vva = vhost_user_iotlb_cache_find(vq, ra,
 351                                         &size, VHOST_ACCESS_RW);
 352                 if (!vva)
 353                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 354
 355                 return vva;
 356         }
 357
 358         return qva_to_vva(dev, ra);
 359 }
 360
 361 static struct virtio_net *
 362 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 363 {
 364         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 365         struct vhost_vring_addr *addr = &vq->ring_addrs;
 366
 367         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 368         if (vq->desc && vq->avail && vq->used)
 369                 return dev;
 370
 371         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 372                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 373         if (vq->desc == 0) {
 374                 RTE_LOG(DEBUG, VHOST_CONFIG,
 375                         "(%d) failed to find desc ring address.\n",
 376                         dev->vid);
 377                 return dev;
 378         }
 379
 380         dev = numa_realloc(dev, vq_index);
 381         vq = dev->virtqueue[vq_index];
 382         addr = &vq->ring_addrs;
 383
 384         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 385                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 386         if (vq->avail == 0) {
 387                 RTE_LOG(DEBUG, VHOST_CONFIG,
 388                         "(%d) failed to find avail ring address.\n",
 389                         dev->vid);
 390                 return dev;
 391         }
 392
 393         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 394                         vq, addr->used_user_addr, sizeof(struct vring_used));
 395         if (vq->used == 0) {
 396                 RTE_LOG(DEBUG, VHOST_CONFIG,
 397                         "(%d) failed to find used ring address.\n",
 398                         dev->vid);
 399                 return dev;
 400         }
 401
 402         if (vq->last_used_idx != vq->used->idx) {
 403                 RTE_LOG(WARNING, VHOST_CONFIG,
 404                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 405                         "some packets maybe resent for Tx and dropped for Rx\n",
 406                         vq->last_used_idx, vq->used->idx);
 407                 vq->last_used_idx  = vq->used->idx;
 408                 vq->last_avail_idx = vq->used->idx;
 409         }
 410
 411         vq->log_guest_addr = addr->log_guest_addr;
 412
 413         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 414                         dev->vid, vq->desc);
 415         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 416                         dev->vid, vq->avail);
 417         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 418                         dev->vid, vq->used);
 419         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 420                         dev->vid, vq->log_guest_addr);
 421
 422         return dev;
 423 }
 424
 425 /*
 426  * The virtio device sends us the desc, used and avail ring addresses.
 427  * This function then converts these to our address space.
 428  */
 429 static int
 430 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 431 {
 432         struct vhost_virtqueue *vq;
 433         struct vhost_vring_addr *addr = &msg->payload.addr;
 434         struct virtio_net *dev = *pdev;
 435
 436         if (dev->mem == NULL)
 437                 return -1;
 438
 439         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 440         vq = dev->virtqueue[msg->payload.addr.index];
 441
 442         /*
 443          * Rings addresses should not be interpreted as long as the ring is not
 444          * started and enabled
 445          */
 446         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 447
 448         vring_invalidate(dev, vq);
 449
 450         if (vq->enabled && (dev->features &
 451                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 452                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 453                 if (!dev)
 454                         return -1;
 455
 456                 *pdev = dev;
 457         }
 458
 459         return 0;
 460 }
 461
 462 /*
 463  * The virtio device sends us the available ring last used index.
 464  */
 465 static int
 466 vhost_user_set_vring_base(struct virtio_net *dev,
 467                           VhostUserMsg *msg)
 468 {
 469         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 470                         msg->payload.state.num;
 471         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 472                         msg->payload.state.num;
 473
 474         return 0;
 475 }
 476
 477 static void
 478 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 479                    uint64_t host_phys_addr, uint64_t size)
 480 {
 481         struct guest_page *page, *last_page;
 482
 483         if (dev->nr_guest_pages == dev->max_guest_pages) {
 484                 dev->max_guest_pages *= 2;
 485                 dev->guest_pages = realloc(dev->guest_pages,
 486                                         dev->max_guest_pages * sizeof(*page));
 487         }
 488
 489         if (dev->nr_guest_pages > 0) {
 490                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 491                 /* merge if the two pages are continuous */
 492                 if (host_phys_addr == last_page->host_phys_addr +
 493                                       last_page->size) {
 494                         last_page->size += size;
 495                         return;
 496                 }
 497         }
 498
 499         page = &dev->guest_pages[dev->nr_guest_pages++];
 500         page->guest_phys_addr = guest_phys_addr;
 501         page->host_phys_addr  = host_phys_addr;
 502         page->size = size;
 503 }
 504
 505 static void
 506 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 507                 uint64_t page_size)
 508 {
 509         uint64_t reg_size = reg->size;
 510         uint64_t host_user_addr  = reg->host_user_addr;
 511         uint64_t guest_phys_addr = reg->guest_phys_addr;
 512         uint64_t host_phys_addr;
 513         uint64_t size;
 514
 515         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 516         size = page_size - (guest_phys_addr & (page_size - 1));
 517         size = RTE_MIN(size, reg_size);
 518
 519         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 520         host_user_addr  += size;
 521         guest_phys_addr += size;
 522         reg_size -= size;
 523
 524         while (reg_size > 0) {
 525                 size = RTE_MIN(reg_size, page_size);
 526                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 527                                                   host_user_addr);
 528                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 529
 530                 host_user_addr  += size;
 531                 guest_phys_addr += size;
 532                 reg_size -= size;
 533         }
 534 }
 535
 536 #ifdef RTE_LIBRTE_VHOST_DEBUG
 537 /* TODO: enable it only in debug mode? */
 538 static void
 539 dump_guest_pages(struct virtio_net *dev)
 540 {
 541         uint32_t i;
 542         struct guest_page *page;
 543
 544         for (i = 0; i < dev->nr_guest_pages; i++) {
 545                 page = &dev->guest_pages[i];
 546
 547                 RTE_LOG(INFO, VHOST_CONFIG,
 548                         "guest physical page region %u\n"
 549                         "\t guest_phys_addr: %" PRIx64 "\n"
 550                         "\t host_phys_addr : %" PRIx64 "\n"
 551                         "\t size           : %" PRIx64 "\n",
 552                         i,
 553                         page->guest_phys_addr,
 554                         page->host_phys_addr,
 555                         page->size);
 556         }
 557 }
 558 #else
 559 #define dump_guest_pages(dev)
 560 #endif
 561
 562 static bool
 563 vhost_memory_changed(struct VhostUserMemory *new,
 564                      struct rte_vhost_memory *old)
 565 {
 566         uint32_t i;
 567
 568         if (new->nregions != old->nregions)
 569                 return true;
 570
 571         for (i = 0; i < new->nregions; ++i) {
 572                 VhostUserMemoryRegion *new_r = &new->regions[i];
 573                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 574
 575                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 576                         return true;
 577                 if (new_r->memory_size != old_r->size)
 578                         return true;
 579                 if (new_r->userspace_addr != old_r->guest_user_addr)
 580                         return true;
 581         }
 582
 583         return false;
 584 }
 585
 586 static int
 587 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 588 {
 589         struct VhostUserMemory memory = pmsg->payload.memory;
 590         struct rte_vhost_mem_region *reg;
 591         void *mmap_addr;
 592         uint64_t mmap_size;
 593         uint64_t mmap_offset;
 594         uint64_t alignment;
 595         uint32_t i;
 596         int fd;
 597
 598         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 599                 RTE_LOG(INFO, VHOST_CONFIG,
 600                         "(%d) memory regions not changed\n", dev->vid);
 601
 602                 for (i = 0; i < memory.nregions; i++)
 603                         close(pmsg->fds[i]);
 604
 605                 return 0;
 606         }
 607
 608         if (dev->mem) {
 609                 free_mem_region(dev);
 610                 rte_free(dev->mem);
 611                 dev->mem = NULL;
 612         }
 613
 614         dev->nr_guest_pages = 0;
 615         if (!dev->guest_pages) {
 616                 dev->max_guest_pages = 8;
 617                 dev->guest_pages = malloc(dev->max_guest_pages *
 618                                                 sizeof(struct guest_page));
 619                 if (dev->guest_pages == NULL) {
 620                         RTE_LOG(ERR, VHOST_CONFIG,
 621                                 "(%d) failed to allocate memory "
 622                                 "for dev->guest_pages\n",
 623                                 dev->vid);
 624                         return -1;
 625                 }
 626         }
 627
 628         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 629                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 630         if (dev->mem == NULL) {
 631                 RTE_LOG(ERR, VHOST_CONFIG,
 632                         "(%d) failed to allocate memory for dev->mem\n",
 633                         dev->vid);
 634                 return -1;
 635         }
 636         dev->mem->nregions = memory.nregions;
 637
 638         for (i = 0; i < memory.nregions; i++) {
 639                 fd  = pmsg->fds[i];
 640                 reg = &dev->mem->regions[i];
 641
 642                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 643                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 644                 reg->size            = memory.regions[i].memory_size;
 645                 reg->fd              = fd;
 646
 647                 mmap_offset = memory.regions[i].mmap_offset;
 648                 mmap_size   = reg->size + mmap_offset;
 649
 650                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 651                  * with length argument aligned with hugepagesz at older
 652                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 653                  * mmap() will fail with EINVAL.
 654                  *
 655                  * to avoid failure, make sure in caller to keep length
 656                  * aligned.
 657                  */
 658                 alignment = get_blk_size(fd);
 659                 if (alignment == (uint64_t)-1) {
 660                         RTE_LOG(ERR, VHOST_CONFIG,
 661                                 "couldn't get hugepage size through fstat\n");
 662                         goto err_mmap;
 663                 }
 664                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 665
 666                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 667                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 668
 669                 if (mmap_addr == MAP_FAILED) {
 670                         RTE_LOG(ERR, VHOST_CONFIG,
 671                                 "mmap region %u failed.\n", i);
 672                         goto err_mmap;
 673                 }
 674
 675                 reg->mmap_addr = mmap_addr;
 676                 reg->mmap_size = mmap_size;
 677                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 678                                       mmap_offset;
 679
 680                 if (dev->dequeue_zero_copy)
 681                         add_guest_pages(dev, reg, alignment);
 682
 683                 RTE_LOG(INFO, VHOST_CONFIG,
 684                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 685                         "\t guest physical addr: 0x%" PRIx64 "\n"
 686                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 687                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 688                         "\t mmap addr : 0x%" PRIx64 "\n"
 689                         "\t mmap size : 0x%" PRIx64 "\n"
 690                         "\t mmap align: 0x%" PRIx64 "\n"
 691                         "\t mmap off  : 0x%" PRIx64 "\n",
 692                         i, reg->size,
 693                         reg->guest_phys_addr,
 694                         reg->guest_user_addr,
 695                         reg->host_user_addr,
 696                         (uint64_t)(uintptr_t)mmap_addr,
 697                         mmap_size,
 698                         alignment,
 699                         mmap_offset);
 700         }
 701
 702         dump_guest_pages(dev);
 703
 704         return 0;
 705
 706 err_mmap:
 707         free_mem_region(dev);
 708         rte_free(dev->mem);
 709         dev->mem = NULL;
 710         return -1;
 711 }
 712
 713 static int
 714 vq_is_ready(struct vhost_virtqueue *vq)
 715 {
 716         return vq && vq->desc && vq->avail && vq->used &&
 717                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 718                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 719 }
 720
 721 static int
 722 virtio_is_ready(struct virtio_net *dev)
 723 {
 724         struct vhost_virtqueue *vq;
 725         uint32_t i;
 726
 727         if (dev->nr_vring == 0)
 728                 return 0;
 729
 730         for (i = 0; i < dev->nr_vring; i++) {
 731                 vq = dev->virtqueue[i];
 732
 733                 if (!vq_is_ready(vq))
 734                         return 0;
 735         }
 736
 737         RTE_LOG(INFO, VHOST_CONFIG,
 738                 "virtio is now ready for processing.\n");
 739         return 1;
 740 }
 741
 742 static void
 743 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 744 {
 745         struct vhost_vring_file file;
 746         struct vhost_virtqueue *vq;
 747
 748         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 749         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 750                 file.fd = VIRTIO_INVALID_EVENTFD;
 751         else
 752                 file.fd = pmsg->fds[0];
 753         RTE_LOG(INFO, VHOST_CONFIG,
 754                 "vring call idx:%d file:%d\n", file.index, file.fd);
 755
 756         vq = dev->virtqueue[file.index];
 757         if (vq->callfd >= 0)
 758                 close(vq->callfd);
 759
 760         vq->callfd = file.fd;
 761 }
 762
 763 static void
 764 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 765 {
 766         struct vhost_vring_file file;
 767         struct vhost_virtqueue *vq;
 768         struct virtio_net *dev = *pdev;
 769
 770         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 771         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 772                 file.fd = VIRTIO_INVALID_EVENTFD;
 773         else
 774                 file.fd = pmsg->fds[0];
 775         RTE_LOG(INFO, VHOST_CONFIG,
 776                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 777
 778         /* Interpret ring addresses only when ring is started. */
 779         dev = translate_ring_addresses(dev, file.index);
 780         if (!dev)
 781                 return;
 782
 783         *pdev = dev;
 784
 785         vq = dev->virtqueue[file.index];
 786
 787         /*
 788          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 789          * the ring starts already enabled. Otherwise, it is enabled via
 790          * the SET_VRING_ENABLE message.
 791          */
 792         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 793                 vq->enabled = 1;
 794
 795         if (vq->kickfd >= 0)
 796                 close(vq->kickfd);
 797         vq->kickfd = file.fd;
 798 }
 799
 800 static void
 801 free_zmbufs(struct vhost_virtqueue *vq)
 802 {
 803         struct zcopy_mbuf *zmbuf, *next;
 804
 805         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 806              zmbuf != NULL; zmbuf = next) {
 807                 next = TAILQ_NEXT(zmbuf, next);
 808
 809                 rte_pktmbuf_free(zmbuf->mbuf);
 810                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 811         }
 812
 813         rte_free(vq->zmbufs);
 814 }
 815
 816 /*
 817  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 818  */
 819 static int
 820 vhost_user_get_vring_base(struct virtio_net *dev,
 821                           VhostUserMsg *msg)
 822 {
 823         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 824
 825         /* We have to stop the queue (virtio) if it is running. */
 826         if (dev->flags & VIRTIO_DEV_RUNNING) {
 827                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 828                 dev->notify_ops->destroy_device(dev->vid);
 829         }
 830
 831         dev->flags &= ~VIRTIO_DEV_READY;
 832
 833         /* Here we are safe to get the last used index */
 834         msg->payload.state.num = vq->last_used_idx;
 835
 836         RTE_LOG(INFO, VHOST_CONFIG,
 837                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 838                 msg->payload.state.num);
 839         /*
 840          * Based on current qemu vhost-user implementation, this message is
 841          * sent and only sent in vhost_vring_stop.
 842          * TODO: cleanup the vring, it isn't usable since here.
 843          */
 844         if (vq->kickfd >= 0)
 845                 close(vq->kickfd);
 846
 847         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 848
 849         if (dev->dequeue_zero_copy)
 850                 free_zmbufs(vq);
 851         rte_free(vq->shadow_used_ring);
 852         vq->shadow_used_ring = NULL;
 853
 854         rte_free(vq->batch_copy_elems);
 855         vq->batch_copy_elems = NULL;
 856
 857         return 0;
 858 }
 859
 860 /*
 861  * when virtio queues are ready to work, qemu will send us to
 862  * enable the virtio queue pair.
 863  */
 864 static int
 865 vhost_user_set_vring_enable(struct virtio_net *dev,
 866                             VhostUserMsg *msg)
 867 {
 868         int enable = (int)msg->payload.state.num;
 869
 870         RTE_LOG(INFO, VHOST_CONFIG,
 871                 "set queue enable: %d to qp idx: %d\n",
 872                 enable, msg->payload.state.index);
 873
 874         if (dev->notify_ops->vring_state_changed)
 875                 dev->notify_ops->vring_state_changed(dev->vid,
 876                                 msg->payload.state.index, enable);
 877
 878         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 879
 880         return 0;
 881 }
 882
 883 static void
 884 vhost_user_get_protocol_features(struct virtio_net *dev,
 885                                  struct VhostUserMsg *msg)
 886 {
 887         uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
 888
 889         rte_vhost_driver_get_features(dev->ifname, &features);
 890
 891         /*
 892          * REPLY_ACK protocol feature is only mandatory for now
 893          * for IOMMU feature. If IOMMU is explicitly disabled by the
 894          * application, disable also REPLY_ACK feature for older buggy
 895          * Qemu versions (from v2.7.0 to v2.9.0).
 896          */
 897         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 898                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
 899
 900         msg->payload.u64 = protocol_features;
 901         msg->size = sizeof(msg->payload.u64);
 902 }
 903
 904 static void
 905 vhost_user_set_protocol_features(struct virtio_net *dev,
 906                                  uint64_t protocol_features)
 907 {
 908         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 909                 return;
 910
 911         dev->protocol_features = protocol_features;
 912 }
 913
 914 static int
 915 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 916 {
 917         int fd = msg->fds[0];
 918         uint64_t size, off;
 919         void *addr;
 920
 921         if (fd < 0) {
 922                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 923                 return -1;
 924         }
 925
 926         if (msg->size != sizeof(VhostUserLog)) {
 927                 RTE_LOG(ERR, VHOST_CONFIG,
 928                         "invalid log base msg size: %"PRId32" != %d\n",
 929                         msg->size, (int)sizeof(VhostUserLog));
 930                 return -1;
 931         }
 932
 933         size = msg->payload.log.mmap_size;
 934         off  = msg->payload.log.mmap_offset;
 935         RTE_LOG(INFO, VHOST_CONFIG,
 936                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 937                 size, off);
 938
 939         /*
 940          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 941          * fail when offset is not page size aligned.
 942          */
 943         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 944         close(fd);
 945         if (addr == MAP_FAILED) {
 946                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 947                 return -1;
 948         }
 949
 950         /*
 951          * Free previously mapped log memory on occasionally
 952          * multiple VHOST_USER_SET_LOG_BASE.
 953          */
 954         if (dev->log_addr) {
 955                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 956         }
 957         dev->log_addr = (uint64_t)(uintptr_t)addr;
 958         dev->log_base = dev->log_addr + off;
 959         dev->log_size = size;
 960
 961         return 0;
 962 }
 963
 964 /*
 965  * An rarp packet is constructed and broadcasted to notify switches about
 966  * the new location of the migrated VM, so that packets from outside will
 967  * not be lost after migration.
 968  *
 969  * However, we don't actually "send" a rarp packet here, instead, we set
 970  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 971  */
 972 static int
 973 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 974 {
 975         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 976
 977         RTE_LOG(DEBUG, VHOST_CONFIG,
 978                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 979                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 980         memcpy(dev->mac.addr_bytes, mac, 6);
 981
 982         /*
 983          * Set the flag to inject a RARP broadcast packet at
 984          * rte_vhost_dequeue_burst().
 985          *
 986          * rte_smp_wmb() is for making sure the mac is copied
 987          * before the flag is set.
 988          */
 989         rte_smp_wmb();
 990         rte_atomic16_set(&dev->broadcast_rarp, 1);
 991
 992         return 0;
 993 }
 994
 995 static int
 996 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 997 {
 998         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 999                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1000                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1001                                 msg->payload.u64);
1002
1003                 return -1;
1004         }
1005
1006         dev->mtu = msg->payload.u64;
1007
1008         return 0;
1009 }
1010
1011 static int
1012 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1013 {
1014         int fd = msg->fds[0];
1015
1016         if (fd < 0) {
1017                 RTE_LOG(ERR, VHOST_CONFIG,
1018                                 "Invalid file descriptor for slave channel (%d)\n",
1019                                 fd);
1020                 return -1;
1021         }
1022
1023         dev->slave_req_fd = fd;
1024
1025         return 0;
1026 }
1027
1028 static int
1029 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1030 {
1031         struct vhost_vring_addr *ra;
1032         uint64_t start, end;
1033
1034         start = imsg->iova;
1035         end = start + imsg->size;
1036
1037         ra = &vq->ring_addrs;
1038         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1039                 return 1;
1040         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1041                 return 1;
1042         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1043                 return 1;
1044
1045         return 0;
1046 }
1047
1048 static int
1049 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1050                                 struct vhost_iotlb_msg *imsg)
1051 {
1052         uint64_t istart, iend, vstart, vend;
1053
1054         istart = imsg->iova;
1055         iend = istart + imsg->size - 1;
1056
1057         vstart = (uintptr_t)vq->desc;
1058         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1059         if (vstart <= iend && istart <= vend)
1060                 return 1;
1061
1062         vstart = (uintptr_t)vq->avail;
1063         vend = vstart + sizeof(struct vring_avail);
1064         vend += sizeof(uint16_t) * vq->size - 1;
1065         if (vstart <= iend && istart <= vend)
1066                 return 1;
1067
1068         vstart = (uintptr_t)vq->used;
1069         vend = vstart + sizeof(struct vring_used);
1070         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1071         if (vstart <= iend && istart <= vend)
1072                 return 1;
1073
1074         return 0;
1075 }
1076
1077 static int
1078 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1079 {
1080         struct virtio_net *dev = *pdev;
1081         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1082         uint16_t i;
1083         uint64_t vva;
1084
1085         switch (imsg->type) {
1086         case VHOST_IOTLB_UPDATE:
1087                 vva = qva_to_vva(dev, imsg->uaddr);
1088                 if (!vva)
1089                         return -1;
1090
1091                 for (i = 0; i < dev->nr_vring; i++) {
1092                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1093
1094                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1095                                         imsg->size, imsg->perm);
1096
1097                         if (is_vring_iotlb_update(vq, imsg))
1098                                 *pdev = dev = translate_ring_addresses(dev, i);
1099                 }
1100                 break;
1101         case VHOST_IOTLB_INVALIDATE:
1102                 for (i = 0; i < dev->nr_vring; i++) {
1103                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1104
1105                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1106                                         imsg->size);
1107
1108                         if (is_vring_iotlb_invalidate(vq, imsg))
1109                                 vring_invalidate(dev, vq);
1110                 }
1111                 break;
1112         default:
1113                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1114                                 imsg->type);
1115                 return -1;
1116         }
1117
1118         return 0;
1119 }
1120
1121 /* return bytes# of read on success or negative val on failure. */
1122 static int
1123 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1124 {
1125         int ret;
1126
1127         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1128                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1129         if (ret <= 0)
1130                 return ret;
1131
1132         if (msg && msg->size) {
1133                 if (msg->size > sizeof(msg->payload)) {
1134                         RTE_LOG(ERR, VHOST_CONFIG,
1135                                 "invalid msg size: %d\n", msg->size);
1136                         return -1;
1137                 }
1138                 ret = read(sockfd, &msg->payload, msg->size);
1139                 if (ret <= 0)
1140                         return ret;
1141                 if (ret != (int)msg->size) {
1142                         RTE_LOG(ERR, VHOST_CONFIG,
1143                                 "read control message failed\n");
1144                         return -1;
1145                 }
1146         }
1147
1148         return ret;
1149 }
1150
1151 static int
1152 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1153 {
1154         if (!msg)
1155                 return 0;
1156
1157         return send_fd_message(sockfd, (char *)msg,
1158                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1159 }
1160
1161 static int
1162 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1163 {
1164         if (!msg)
1165                 return 0;
1166
1167         msg->flags &= ~VHOST_USER_VERSION_MASK;
1168         msg->flags &= ~VHOST_USER_NEED_REPLY;
1169         msg->flags |= VHOST_USER_VERSION;
1170         msg->flags |= VHOST_USER_REPLY_MASK;
1171
1172         return send_vhost_message(sockfd, msg);
1173 }
1174
1175 /*
1176  * Allocate a queue pair if it hasn't been allocated yet
1177  */
1178 static int
1179 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1180 {
1181         uint16_t vring_idx;
1182
1183         switch (msg->request.master) {
1184         case VHOST_USER_SET_VRING_KICK:
1185         case VHOST_USER_SET_VRING_CALL:
1186         case VHOST_USER_SET_VRING_ERR:
1187                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1188                 break;
1189         case VHOST_USER_SET_VRING_NUM:
1190         case VHOST_USER_SET_VRING_BASE:
1191         case VHOST_USER_SET_VRING_ENABLE:
1192                 vring_idx = msg->payload.state.index;
1193                 break;
1194         case VHOST_USER_SET_VRING_ADDR:
1195                 vring_idx = msg->payload.addr.index;
1196                 break;
1197         default:
1198                 return 0;
1199         }
1200
1201         if (vring_idx >= VHOST_MAX_VRING) {
1202                 RTE_LOG(ERR, VHOST_CONFIG,
1203                         "invalid vring index: %u\n", vring_idx);
1204                 return -1;
1205         }
1206
1207         if (dev->virtqueue[vring_idx])
1208                 return 0;
1209
1210         return alloc_vring_queue(dev, vring_idx);
1211 }
1212
1213 int
1214 vhost_user_msg_handler(int vid, int fd)
1215 {
1216         struct virtio_net *dev;
1217         struct VhostUserMsg msg;
1218         int ret;
1219
1220         dev = get_device(vid);
1221         if (dev == NULL)
1222                 return -1;
1223
1224         if (!dev->notify_ops) {
1225                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1226                 if (!dev->notify_ops) {
1227                         RTE_LOG(ERR, VHOST_CONFIG,
1228                                 "failed to get callback ops for driver %s\n",
1229                                 dev->ifname);
1230                         return -1;
1231                 }
1232         }
1233
1234         ret = read_vhost_message(fd, &msg);
1235         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1236                 if (ret < 0)
1237                         RTE_LOG(ERR, VHOST_CONFIG,
1238                                 "vhost read message failed\n");
1239                 else if (ret == 0)
1240                         RTE_LOG(INFO, VHOST_CONFIG,
1241                                 "vhost peer closed\n");
1242                 else
1243                         RTE_LOG(ERR, VHOST_CONFIG,
1244                                 "vhost read incorrect message\n");
1245
1246                 return -1;
1247         }
1248
1249         ret = 0;
1250         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1251                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1252                         vhost_message_str[msg.request.master]);
1253         else
1254                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1255                         vhost_message_str[msg.request.master]);
1256
1257         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1258         if (ret < 0) {
1259                 RTE_LOG(ERR, VHOST_CONFIG,
1260                         "failed to alloc queue\n");
1261                 return -1;
1262         }
1263
1264         switch (msg.request.master) {
1265         case VHOST_USER_GET_FEATURES:
1266                 msg.payload.u64 = vhost_user_get_features(dev);
1267                 msg.size = sizeof(msg.payload.u64);
1268                 send_vhost_reply(fd, &msg);
1269                 break;
1270         case VHOST_USER_SET_FEATURES:
1271                 ret = vhost_user_set_features(dev, msg.payload.u64);
1272                 if (ret)
1273                         return -1;
1274                 break;
1275
1276         case VHOST_USER_GET_PROTOCOL_FEATURES:
1277                 vhost_user_get_protocol_features(dev, &msg);
1278                 send_vhost_reply(fd, &msg);
1279                 break;
1280         case VHOST_USER_SET_PROTOCOL_FEATURES:
1281                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1282                 break;
1283
1284         case VHOST_USER_SET_OWNER:
1285                 vhost_user_set_owner();
1286                 break;
1287         case VHOST_USER_RESET_OWNER:
1288                 vhost_user_reset_owner(dev);
1289                 break;
1290
1291         case VHOST_USER_SET_MEM_TABLE:
1292                 ret = vhost_user_set_mem_table(dev, &msg);
1293                 break;
1294
1295         case VHOST_USER_SET_LOG_BASE:
1296                 vhost_user_set_log_base(dev, &msg);
1297
1298                 /* it needs a reply */
1299                 msg.size = sizeof(msg.payload.u64);
1300                 send_vhost_reply(fd, &msg);
1301                 break;
1302         case VHOST_USER_SET_LOG_FD:
1303                 close(msg.fds[0]);
1304                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1305                 break;
1306
1307         case VHOST_USER_SET_VRING_NUM:
1308                 vhost_user_set_vring_num(dev, &msg);
1309                 break;
1310         case VHOST_USER_SET_VRING_ADDR:
1311                 vhost_user_set_vring_addr(&dev, &msg);
1312                 break;
1313         case VHOST_USER_SET_VRING_BASE:
1314                 vhost_user_set_vring_base(dev, &msg);
1315                 break;
1316
1317         case VHOST_USER_GET_VRING_BASE:
1318                 vhost_user_get_vring_base(dev, &msg);
1319                 msg.size = sizeof(msg.payload.state);
1320                 send_vhost_reply(fd, &msg);
1321                 break;
1322
1323         case VHOST_USER_SET_VRING_KICK:
1324                 vhost_user_set_vring_kick(&dev, &msg);
1325                 break;
1326         case VHOST_USER_SET_VRING_CALL:
1327                 vhost_user_set_vring_call(dev, &msg);
1328                 break;
1329
1330         case VHOST_USER_SET_VRING_ERR:
1331                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1332                         close(msg.fds[0]);
1333                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1334                 break;
1335
1336         case VHOST_USER_GET_QUEUE_NUM:
1337                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1338                 msg.size = sizeof(msg.payload.u64);
1339                 send_vhost_reply(fd, &msg);
1340                 break;
1341
1342         case VHOST_USER_SET_VRING_ENABLE:
1343                 vhost_user_set_vring_enable(dev, &msg);
1344                 break;
1345         case VHOST_USER_SEND_RARP:
1346                 vhost_user_send_rarp(dev, &msg);
1347                 break;
1348
1349         case VHOST_USER_NET_SET_MTU:
1350                 ret = vhost_user_net_set_mtu(dev, &msg);
1351                 break;
1352
1353         case VHOST_USER_SET_SLAVE_REQ_FD:
1354                 ret = vhost_user_set_req_fd(dev, &msg);
1355                 break;
1356
1357         case VHOST_USER_IOTLB_MSG:
1358                 ret = vhost_user_iotlb_msg(&dev, &msg);
1359                 break;
1360
1361         default:
1362                 ret = -1;
1363                 break;
1364
1365         }
1366
1367         if (msg.flags & VHOST_USER_NEED_REPLY) {
1368                 msg.payload.u64 = !!ret;
1369                 msg.size = sizeof(msg.payload.u64);
1370                 send_vhost_reply(fd, &msg);
1371         }
1372
1373         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1374                 dev->flags |= VIRTIO_DEV_READY;
1375
1376                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1377                         if (dev->dequeue_zero_copy) {
1378                                 RTE_LOG(INFO, VHOST_CONFIG,
1379                                                 "dequeue zero copy is enabled\n");
1380                         }
1381
1382                         if (dev->notify_ops->new_device(dev->vid) == 0)
1383                                 dev->flags |= VIRTIO_DEV_RUNNING;
1384                 }
1385         }
1386
1387         return 0;
1388 }
1389
1390 int
1391 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1392 {
1393         int ret;
1394         struct VhostUserMsg msg = {
1395                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1396                 .flags = VHOST_USER_VERSION,
1397                 .size = sizeof(msg.payload.iotlb),
1398                 .payload.iotlb = {
1399                         .iova = iova,
1400                         .perm = perm,
1401                         .type = VHOST_IOTLB_MISS,
1402                 },
1403         };
1404
1405         ret = send_vhost_message(dev->slave_req_fd, &msg);
1406         if (ret < 0) {
1407                 RTE_LOG(ERR, VHOST_CONFIG,
1408                                 "Failed to send IOTLB miss message (%d)\n",
1409                                 ret);
1410                 return ret;
1411         }
1412
1413         return 0;
1414 }