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         if (!(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 191                 /*
 192                  * Remove all but first queue pair if MQ hasn't been
 193                  * negotiated. This is safe because the device is not
 194                  * running at this stage.
 195                  */
 196                 while (dev->nr_vring > 2) {
 197                         struct vhost_virtqueue *vq;
 198
 199                         vq = dev->virtqueue[--dev->nr_vring];
 200                         if (!vq)
 201                                 continue;
 202
 203                         dev->virtqueue[dev->nr_vring] = NULL;
 204                         cleanup_vq(vq, 1);
 205                         free_vq(vq);
 206                 }
 207         }
 208
 209         return 0;
 210 }
 211
 212 /*
 213  * The virtio device sends us the size of the descriptor ring.
 214  */
 215 static int
 216 vhost_user_set_vring_num(struct virtio_net *dev,
 217                          VhostUserMsg *msg)
 218 {
 219         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 220
 221         vq->size = msg->payload.state.num;
 222
 223         if (dev->dequeue_zero_copy) {
 224                 vq->nr_zmbuf = 0;
 225                 vq->last_zmbuf_idx = 0;
 226                 vq->zmbuf_size = vq->size;
 227                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 228                                          sizeof(struct zcopy_mbuf), 0);
 229                 if (vq->zmbufs == NULL) {
 230                         RTE_LOG(WARNING, VHOST_CONFIG,
 231                                 "failed to allocate mem for zero copy; "
 232                                 "zero copy is force disabled\n");
 233                         dev->dequeue_zero_copy = 0;
 234                 }
 235                 TAILQ_INIT(&vq->zmbuf_list);
 236         }
 237
 238         vq->shadow_used_ring = rte_malloc(NULL,
 239                                 vq->size * sizeof(struct vring_used_elem),
 240                                 RTE_CACHE_LINE_SIZE);
 241         if (!vq->shadow_used_ring) {
 242                 RTE_LOG(ERR, VHOST_CONFIG,
 243                         "failed to allocate memory for shadow used ring.\n");
 244                 return -1;
 245         }
 246
 247         vq->batch_copy_elems = rte_malloc(NULL,
 248                                 vq->size * sizeof(struct batch_copy_elem),
 249                                 RTE_CACHE_LINE_SIZE);
 250         if (!vq->batch_copy_elems) {
 251                 RTE_LOG(ERR, VHOST_CONFIG,
 252                         "failed to allocate memory for batching copy.\n");
 253                 return -1;
 254         }
 255
 256         return 0;
 257 }
 258
 259 /*
 260  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 261  * same numa node as the memory of vring descriptor.
 262  */
 263 #ifdef RTE_LIBRTE_VHOST_NUMA
 264 static struct virtio_net*
 265 numa_realloc(struct virtio_net *dev, int index)
 266 {
 267         int oldnode, newnode;
 268         struct virtio_net *old_dev;
 269         struct vhost_virtqueue *old_vq, *vq;
 270         struct zcopy_mbuf *new_zmbuf;
 271         struct vring_used_elem *new_shadow_used_ring;
 272         struct batch_copy_elem *new_batch_copy_elems;
 273         int ret;
 274
 275         old_dev = dev;
 276         vq = old_vq = dev->virtqueue[index];
 277
 278         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 279                             MPOL_F_NODE | MPOL_F_ADDR);
 280
 281         /* check if we need to reallocate vq */
 282         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 283                              MPOL_F_NODE | MPOL_F_ADDR);
 284         if (ret) {
 285                 RTE_LOG(ERR, VHOST_CONFIG,
 286                         "Unable to get vq numa information.\n");
 287                 return dev;
 288         }
 289         if (oldnode != newnode) {
 290                 RTE_LOG(INFO, VHOST_CONFIG,
 291                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 292                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 293                 if (!vq)
 294                         return dev;
 295
 296                 memcpy(vq, old_vq, sizeof(*vq));
 297                 TAILQ_INIT(&vq->zmbuf_list);
 298
 299                 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 300                         sizeof(struct zcopy_mbuf), 0, newnode);
 301                 if (new_zmbuf) {
 302                         rte_free(vq->zmbufs);
 303                         vq->zmbufs = new_zmbuf;
 304                 }
 305
 306                 new_shadow_used_ring = rte_malloc_socket(NULL,
 307                         vq->size * sizeof(struct vring_used_elem),
 308                         RTE_CACHE_LINE_SIZE,
 309                         newnode);
 310                 if (new_shadow_used_ring) {
 311                         rte_free(vq->shadow_used_ring);
 312                         vq->shadow_used_ring = new_shadow_used_ring;
 313                 }
 314
 315                 new_batch_copy_elems = rte_malloc_socket(NULL,
 316                         vq->size * sizeof(struct batch_copy_elem),
 317                         RTE_CACHE_LINE_SIZE,
 318                         newnode);
 319                 if (new_batch_copy_elems) {
 320                         rte_free(vq->batch_copy_elems);
 321                         vq->batch_copy_elems = new_batch_copy_elems;
 322                 }
 323
 324                 rte_free(old_vq);
 325         }
 326
 327         /* check if we need to reallocate dev */
 328         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 329                             MPOL_F_NODE | MPOL_F_ADDR);
 330         if (ret) {
 331                 RTE_LOG(ERR, VHOST_CONFIG,
 332                         "Unable to get dev numa information.\n");
 333                 goto out;
 334         }
 335         if (oldnode != newnode) {
 336                 RTE_LOG(INFO, VHOST_CONFIG,
 337                         "reallocate dev from %d to %d node\n",
 338                         oldnode, newnode);
 339                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 340                 if (!dev) {
 341                         dev = old_dev;
 342                         goto out;
 343                 }
 344
 345                 memcpy(dev, old_dev, sizeof(*dev));
 346                 rte_free(old_dev);
 347         }
 348
 349 out:
 350         dev->virtqueue[index] = vq;
 351         vhost_devices[dev->vid] = dev;
 352
 353         if (old_vq != vq)
 354                 vhost_user_iotlb_init(dev, index);
 355
 356         return dev;
 357 }
 358 #else
 359 static struct virtio_net*
 360 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 361 {
 362         return dev;
 363 }
 364 #endif
 365
 366 /* Converts QEMU virtual address to Vhost virtual address. */
 367 static uint64_t
 368 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 369 {
 370         struct rte_vhost_mem_region *reg;
 371         uint32_t i;
 372
 373         /* Find the region where the address lives. */
 374         for (i = 0; i < dev->mem->nregions; i++) {
 375                 reg = &dev->mem->regions[i];
 376
 377                 if (qva >= reg->guest_user_addr &&
 378                     qva <  reg->guest_user_addr + reg->size) {
 379                         return qva - reg->guest_user_addr +
 380                                reg->host_user_addr;
 381                 }
 382         }
 383
 384         return 0;
 385 }
 386
 387
 388 /*
 389  * Converts ring address to Vhost virtual address.
 390  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 391  * else it is a QEMU virtual address.
 392  */
 393 static uint64_t
 394 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 395                 uint64_t ra, uint64_t size)
 396 {
 397         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 398                 uint64_t vva;
 399
 400                 vva = vhost_user_iotlb_cache_find(vq, ra,
 401                                         &size, VHOST_ACCESS_RW);
 402                 if (!vva)
 403                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 404
 405                 return vva;
 406         }
 407
 408         return qva_to_vva(dev, ra);
 409 }
 410
 411 static struct virtio_net *
 412 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 413 {
 414         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 415         struct vhost_vring_addr *addr = &vq->ring_addrs;
 416
 417         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 418         if (vq->desc && vq->avail && vq->used)
 419                 return dev;
 420
 421         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 422                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 423         if (vq->desc == 0) {
 424                 RTE_LOG(DEBUG, VHOST_CONFIG,
 425                         "(%d) failed to find desc ring address.\n",
 426                         dev->vid);
 427                 return dev;
 428         }
 429
 430         dev = numa_realloc(dev, vq_index);
 431         vq = dev->virtqueue[vq_index];
 432         addr = &vq->ring_addrs;
 433
 434         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 435                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 436         if (vq->avail == 0) {
 437                 RTE_LOG(DEBUG, VHOST_CONFIG,
 438                         "(%d) failed to find avail ring address.\n",
 439                         dev->vid);
 440                 return dev;
 441         }
 442
 443         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 444                         vq, addr->used_user_addr, sizeof(struct vring_used));
 445         if (vq->used == 0) {
 446                 RTE_LOG(DEBUG, VHOST_CONFIG,
 447                         "(%d) failed to find used ring address.\n",
 448                         dev->vid);
 449                 return dev;
 450         }
 451
 452         if (vq->last_used_idx != vq->used->idx) {
 453                 RTE_LOG(WARNING, VHOST_CONFIG,
 454                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 455                         "some packets maybe resent for Tx and dropped for Rx\n",
 456                         vq->last_used_idx, vq->used->idx);
 457                 vq->last_used_idx  = vq->used->idx;
 458                 vq->last_avail_idx = vq->used->idx;
 459         }
 460
 461         vq->log_guest_addr = addr->log_guest_addr;
 462
 463         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 464                         dev->vid, vq->desc);
 465         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 466                         dev->vid, vq->avail);
 467         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 468                         dev->vid, vq->used);
 469         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 470                         dev->vid, vq->log_guest_addr);
 471
 472         return dev;
 473 }
 474
 475 /*
 476  * The virtio device sends us the desc, used and avail ring addresses.
 477  * This function then converts these to our address space.
 478  */
 479 static int
 480 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 481 {
 482         struct vhost_virtqueue *vq;
 483         struct vhost_vring_addr *addr = &msg->payload.addr;
 484         struct virtio_net *dev = *pdev;
 485
 486         if (dev->mem == NULL)
 487                 return -1;
 488
 489         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 490         vq = dev->virtqueue[msg->payload.addr.index];
 491
 492         /*
 493          * Rings addresses should not be interpreted as long as the ring is not
 494          * started and enabled
 495          */
 496         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 497
 498         vring_invalidate(dev, vq);
 499
 500         if (vq->enabled && (dev->features &
 501                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 502                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 503                 if (!dev)
 504                         return -1;
 505
 506                 *pdev = dev;
 507         }
 508
 509         return 0;
 510 }
 511
 512 /*
 513  * The virtio device sends us the available ring last used index.
 514  */
 515 static int
 516 vhost_user_set_vring_base(struct virtio_net *dev,
 517                           VhostUserMsg *msg)
 518 {
 519         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 520                         msg->payload.state.num;
 521         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 522                         msg->payload.state.num;
 523
 524         return 0;
 525 }
 526
 527 static void
 528 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 529                    uint64_t host_phys_addr, uint64_t size)
 530 {
 531         struct guest_page *page, *last_page;
 532
 533         if (dev->nr_guest_pages == dev->max_guest_pages) {
 534                 dev->max_guest_pages *= 2;
 535                 dev->guest_pages = realloc(dev->guest_pages,
 536                                         dev->max_guest_pages * sizeof(*page));
 537         }
 538
 539         if (dev->nr_guest_pages > 0) {
 540                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 541                 /* merge if the two pages are continuous */
 542                 if (host_phys_addr == last_page->host_phys_addr +
 543                                       last_page->size) {
 544                         last_page->size += size;
 545                         return;
 546                 }
 547         }
 548
 549         page = &dev->guest_pages[dev->nr_guest_pages++];
 550         page->guest_phys_addr = guest_phys_addr;
 551         page->host_phys_addr  = host_phys_addr;
 552         page->size = size;
 553 }
 554
 555 static void
 556 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 557                 uint64_t page_size)
 558 {
 559         uint64_t reg_size = reg->size;
 560         uint64_t host_user_addr  = reg->host_user_addr;
 561         uint64_t guest_phys_addr = reg->guest_phys_addr;
 562         uint64_t host_phys_addr;
 563         uint64_t size;
 564
 565         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 566         size = page_size - (guest_phys_addr & (page_size - 1));
 567         size = RTE_MIN(size, reg_size);
 568
 569         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 570         host_user_addr  += size;
 571         guest_phys_addr += size;
 572         reg_size -= size;
 573
 574         while (reg_size > 0) {
 575                 size = RTE_MIN(reg_size, page_size);
 576                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 577                                                   host_user_addr);
 578                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 579
 580                 host_user_addr  += size;
 581                 guest_phys_addr += size;
 582                 reg_size -= size;
 583         }
 584 }
 585
 586 #ifdef RTE_LIBRTE_VHOST_DEBUG
 587 /* TODO: enable it only in debug mode? */
 588 static void
 589 dump_guest_pages(struct virtio_net *dev)
 590 {
 591         uint32_t i;
 592         struct guest_page *page;
 593
 594         for (i = 0; i < dev->nr_guest_pages; i++) {
 595                 page = &dev->guest_pages[i];
 596
 597                 RTE_LOG(INFO, VHOST_CONFIG,
 598                         "guest physical page region %u\n"
 599                         "\t guest_phys_addr: %" PRIx64 "\n"
 600                         "\t host_phys_addr : %" PRIx64 "\n"
 601                         "\t size           : %" PRIx64 "\n",
 602                         i,
 603                         page->guest_phys_addr,
 604                         page->host_phys_addr,
 605                         page->size);
 606         }
 607 }
 608 #else
 609 #define dump_guest_pages(dev)
 610 #endif
 611
 612 static bool
 613 vhost_memory_changed(struct VhostUserMemory *new,
 614                      struct rte_vhost_memory *old)
 615 {
 616         uint32_t i;
 617
 618         if (new->nregions != old->nregions)
 619                 return true;
 620
 621         for (i = 0; i < new->nregions; ++i) {
 622                 VhostUserMemoryRegion *new_r = &new->regions[i];
 623                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 624
 625                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 626                         return true;
 627                 if (new_r->memory_size != old_r->size)
 628                         return true;
 629                 if (new_r->userspace_addr != old_r->guest_user_addr)
 630                         return true;
 631         }
 632
 633         return false;
 634 }
 635
 636 static int
 637 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 638 {
 639         struct VhostUserMemory memory = pmsg->payload.memory;
 640         struct rte_vhost_mem_region *reg;
 641         void *mmap_addr;
 642         uint64_t mmap_size;
 643         uint64_t mmap_offset;
 644         uint64_t alignment;
 645         uint32_t i;
 646         int fd;
 647
 648         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 649                 RTE_LOG(INFO, VHOST_CONFIG,
 650                         "(%d) memory regions not changed\n", dev->vid);
 651
 652                 for (i = 0; i < memory.nregions; i++)
 653                         close(pmsg->fds[i]);
 654
 655                 return 0;
 656         }
 657
 658         if (dev->mem) {
 659                 free_mem_region(dev);
 660                 rte_free(dev->mem);
 661                 dev->mem = NULL;
 662         }
 663
 664         dev->nr_guest_pages = 0;
 665         if (!dev->guest_pages) {
 666                 dev->max_guest_pages = 8;
 667                 dev->guest_pages = malloc(dev->max_guest_pages *
 668                                                 sizeof(struct guest_page));
 669                 if (dev->guest_pages == NULL) {
 670                         RTE_LOG(ERR, VHOST_CONFIG,
 671                                 "(%d) failed to allocate memory "
 672                                 "for dev->guest_pages\n",
 673                                 dev->vid);
 674                         return -1;
 675                 }
 676         }
 677
 678         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 679                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 680         if (dev->mem == NULL) {
 681                 RTE_LOG(ERR, VHOST_CONFIG,
 682                         "(%d) failed to allocate memory for dev->mem\n",
 683                         dev->vid);
 684                 return -1;
 685         }
 686         dev->mem->nregions = memory.nregions;
 687
 688         for (i = 0; i < memory.nregions; i++) {
 689                 fd  = pmsg->fds[i];
 690                 reg = &dev->mem->regions[i];
 691
 692                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 693                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 694                 reg->size            = memory.regions[i].memory_size;
 695                 reg->fd              = fd;
 696
 697                 mmap_offset = memory.regions[i].mmap_offset;
 698                 mmap_size   = reg->size + mmap_offset;
 699
 700                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 701                  * with length argument aligned with hugepagesz at older
 702                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 703                  * mmap() will fail with EINVAL.
 704                  *
 705                  * to avoid failure, make sure in caller to keep length
 706                  * aligned.
 707                  */
 708                 alignment = get_blk_size(fd);
 709                 if (alignment == (uint64_t)-1) {
 710                         RTE_LOG(ERR, VHOST_CONFIG,
 711                                 "couldn't get hugepage size through fstat\n");
 712                         goto err_mmap;
 713                 }
 714                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 715
 716                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 717                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 718
 719                 if (mmap_addr == MAP_FAILED) {
 720                         RTE_LOG(ERR, VHOST_CONFIG,
 721                                 "mmap region %u failed.\n", i);
 722                         goto err_mmap;
 723                 }
 724
 725                 reg->mmap_addr = mmap_addr;
 726                 reg->mmap_size = mmap_size;
 727                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 728                                       mmap_offset;
 729
 730                 if (dev->dequeue_zero_copy)
 731                         add_guest_pages(dev, reg, alignment);
 732
 733                 RTE_LOG(INFO, VHOST_CONFIG,
 734                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 735                         "\t guest physical addr: 0x%" PRIx64 "\n"
 736                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 737                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 738                         "\t mmap addr : 0x%" PRIx64 "\n"
 739                         "\t mmap size : 0x%" PRIx64 "\n"
 740                         "\t mmap align: 0x%" PRIx64 "\n"
 741                         "\t mmap off  : 0x%" PRIx64 "\n",
 742                         i, reg->size,
 743                         reg->guest_phys_addr,
 744                         reg->guest_user_addr,
 745                         reg->host_user_addr,
 746                         (uint64_t)(uintptr_t)mmap_addr,
 747                         mmap_size,
 748                         alignment,
 749                         mmap_offset);
 750         }
 751
 752         dump_guest_pages(dev);
 753
 754         return 0;
 755
 756 err_mmap:
 757         free_mem_region(dev);
 758         rte_free(dev->mem);
 759         dev->mem = NULL;
 760         return -1;
 761 }
 762
 763 static int
 764 vq_is_ready(struct vhost_virtqueue *vq)
 765 {
 766         return vq && vq->desc && vq->avail && vq->used &&
 767                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 768                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 769 }
 770
 771 static int
 772 virtio_is_ready(struct virtio_net *dev)
 773 {
 774         struct vhost_virtqueue *vq;
 775         uint32_t i;
 776
 777         if (dev->nr_vring == 0)
 778                 return 0;
 779
 780         for (i = 0; i < dev->nr_vring; i++) {
 781                 vq = dev->virtqueue[i];
 782
 783                 if (!vq_is_ready(vq))
 784                         return 0;
 785         }
 786
 787         RTE_LOG(INFO, VHOST_CONFIG,
 788                 "virtio is now ready for processing.\n");
 789         return 1;
 790 }
 791
 792 static void
 793 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 794 {
 795         struct vhost_vring_file file;
 796         struct vhost_virtqueue *vq;
 797
 798         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 799         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 800                 file.fd = VIRTIO_INVALID_EVENTFD;
 801         else
 802                 file.fd = pmsg->fds[0];
 803         RTE_LOG(INFO, VHOST_CONFIG,
 804                 "vring call idx:%d file:%d\n", file.index, file.fd);
 805
 806         vq = dev->virtqueue[file.index];
 807         if (vq->callfd >= 0)
 808                 close(vq->callfd);
 809
 810         vq->callfd = file.fd;
 811 }
 812
 813 static void
 814 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 815 {
 816         struct vhost_vring_file file;
 817         struct vhost_virtqueue *vq;
 818         struct virtio_net *dev = *pdev;
 819
 820         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 821         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 822                 file.fd = VIRTIO_INVALID_EVENTFD;
 823         else
 824                 file.fd = pmsg->fds[0];
 825         RTE_LOG(INFO, VHOST_CONFIG,
 826                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 827
 828         /* Interpret ring addresses only when ring is started. */
 829         dev = translate_ring_addresses(dev, file.index);
 830         if (!dev)
 831                 return;
 832
 833         *pdev = dev;
 834
 835         vq = dev->virtqueue[file.index];
 836
 837         /*
 838          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 839          * the ring starts already enabled. Otherwise, it is enabled via
 840          * the SET_VRING_ENABLE message.
 841          */
 842         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 843                 vq->enabled = 1;
 844
 845         if (vq->kickfd >= 0)
 846                 close(vq->kickfd);
 847         vq->kickfd = file.fd;
 848 }
 849
 850 static void
 851 free_zmbufs(struct vhost_virtqueue *vq)
 852 {
 853         struct zcopy_mbuf *zmbuf, *next;
 854
 855         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 856              zmbuf != NULL; zmbuf = next) {
 857                 next = TAILQ_NEXT(zmbuf, next);
 858
 859                 rte_pktmbuf_free(zmbuf->mbuf);
 860                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 861         }
 862
 863         rte_free(vq->zmbufs);
 864 }
 865
 866 /*
 867  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 868  */
 869 static int
 870 vhost_user_get_vring_base(struct virtio_net *dev,
 871                           VhostUserMsg *msg)
 872 {
 873         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 874
 875         /* We have to stop the queue (virtio) if it is running. */
 876         if (dev->flags & VIRTIO_DEV_RUNNING) {
 877                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 878                 dev->notify_ops->destroy_device(dev->vid);
 879         }
 880
 881         dev->flags &= ~VIRTIO_DEV_READY;
 882
 883         /* Here we are safe to get the last used index */
 884         msg->payload.state.num = vq->last_used_idx;
 885
 886         RTE_LOG(INFO, VHOST_CONFIG,
 887                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 888                 msg->payload.state.num);
 889         /*
 890          * Based on current qemu vhost-user implementation, this message is
 891          * sent and only sent in vhost_vring_stop.
 892          * TODO: cleanup the vring, it isn't usable since here.
 893          */
 894         if (vq->kickfd >= 0)
 895                 close(vq->kickfd);
 896
 897         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 898
 899         if (dev->dequeue_zero_copy)
 900                 free_zmbufs(vq);
 901         rte_free(vq->shadow_used_ring);
 902         vq->shadow_used_ring = NULL;
 903
 904         rte_free(vq->batch_copy_elems);
 905         vq->batch_copy_elems = NULL;
 906
 907         return 0;
 908 }
 909
 910 /*
 911  * when virtio queues are ready to work, qemu will send us to
 912  * enable the virtio queue pair.
 913  */
 914 static int
 915 vhost_user_set_vring_enable(struct virtio_net *dev,
 916                             VhostUserMsg *msg)
 917 {
 918         int enable = (int)msg->payload.state.num;
 919
 920         RTE_LOG(INFO, VHOST_CONFIG,
 921                 "set queue enable: %d to qp idx: %d\n",
 922                 enable, msg->payload.state.index);
 923
 924         if (dev->notify_ops->vring_state_changed)
 925                 dev->notify_ops->vring_state_changed(dev->vid,
 926                                 msg->payload.state.index, enable);
 927
 928         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 929
 930         return 0;
 931 }
 932
 933 static void
 934 vhost_user_get_protocol_features(struct virtio_net *dev,
 935                                  struct VhostUserMsg *msg)
 936 {
 937         uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
 938
 939         rte_vhost_driver_get_features(dev->ifname, &features);
 940
 941         /*
 942          * REPLY_ACK protocol feature is only mandatory for now
 943          * for IOMMU feature. If IOMMU is explicitly disabled by the
 944          * application, disable also REPLY_ACK feature for older buggy
 945          * Qemu versions (from v2.7.0 to v2.9.0).
 946          */
 947         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
 948                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
 949
 950         msg->payload.u64 = protocol_features;
 951         msg->size = sizeof(msg->payload.u64);
 952 }
 953
 954 static void
 955 vhost_user_set_protocol_features(struct virtio_net *dev,
 956                                  uint64_t protocol_features)
 957 {
 958         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 959                 return;
 960
 961         dev->protocol_features = protocol_features;
 962 }
 963
 964 static int
 965 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 966 {
 967         int fd = msg->fds[0];
 968         uint64_t size, off;
 969         void *addr;
 970
 971         if (fd < 0) {
 972                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 973                 return -1;
 974         }
 975
 976         if (msg->size != sizeof(VhostUserLog)) {
 977                 RTE_LOG(ERR, VHOST_CONFIG,
 978                         "invalid log base msg size: %"PRId32" != %d\n",
 979                         msg->size, (int)sizeof(VhostUserLog));
 980                 return -1;
 981         }
 982
 983         size = msg->payload.log.mmap_size;
 984         off  = msg->payload.log.mmap_offset;
 985         RTE_LOG(INFO, VHOST_CONFIG,
 986                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 987                 size, off);
 988
 989         /*
 990          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 991          * fail when offset is not page size aligned.
 992          */
 993         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 994         close(fd);
 995         if (addr == MAP_FAILED) {
 996                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 997                 return -1;
 998         }
 999
1000         /*
1001          * Free previously mapped log memory on occasionally
1002          * multiple VHOST_USER_SET_LOG_BASE.
1003          */
1004         if (dev->log_addr) {
1005                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1006         }
1007         dev->log_addr = (uint64_t)(uintptr_t)addr;
1008         dev->log_base = dev->log_addr + off;
1009         dev->log_size = size;
1010
1011         return 0;
1012 }
1013
1014 /*
1015  * An rarp packet is constructed and broadcasted to notify switches about
1016  * the new location of the migrated VM, so that packets from outside will
1017  * not be lost after migration.
1018  *
1019  * However, we don't actually "send" a rarp packet here, instead, we set
1020  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1021  */
1022 static int
1023 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1024 {
1025         uint8_t *mac = (uint8_t *)&msg->payload.u64;
1026
1027         RTE_LOG(DEBUG, VHOST_CONFIG,
1028                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1029                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1030         memcpy(dev->mac.addr_bytes, mac, 6);
1031
1032         /*
1033          * Set the flag to inject a RARP broadcast packet at
1034          * rte_vhost_dequeue_burst().
1035          *
1036          * rte_smp_wmb() is for making sure the mac is copied
1037          * before the flag is set.
1038          */
1039         rte_smp_wmb();
1040         rte_atomic16_set(&dev->broadcast_rarp, 1);
1041
1042         return 0;
1043 }
1044
1045 static int
1046 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1047 {
1048         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1049                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1050                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1051                                 msg->payload.u64);
1052
1053                 return -1;
1054         }
1055
1056         dev->mtu = msg->payload.u64;
1057
1058         return 0;
1059 }
1060
1061 static int
1062 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1063 {
1064         int fd = msg->fds[0];
1065
1066         if (fd < 0) {
1067                 RTE_LOG(ERR, VHOST_CONFIG,
1068                                 "Invalid file descriptor for slave channel (%d)\n",
1069                                 fd);
1070                 return -1;
1071         }
1072
1073         dev->slave_req_fd = fd;
1074
1075         return 0;
1076 }
1077
1078 static int
1079 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1080 {
1081         struct vhost_vring_addr *ra;
1082         uint64_t start, end;
1083
1084         start = imsg->iova;
1085         end = start + imsg->size;
1086
1087         ra = &vq->ring_addrs;
1088         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1089                 return 1;
1090         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1091                 return 1;
1092         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1093                 return 1;
1094
1095         return 0;
1096 }
1097
1098 static int
1099 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1100                                 struct vhost_iotlb_msg *imsg)
1101 {
1102         uint64_t istart, iend, vstart, vend;
1103
1104         istart = imsg->iova;
1105         iend = istart + imsg->size - 1;
1106
1107         vstart = (uintptr_t)vq->desc;
1108         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1109         if (vstart <= iend && istart <= vend)
1110                 return 1;
1111
1112         vstart = (uintptr_t)vq->avail;
1113         vend = vstart + sizeof(struct vring_avail);
1114         vend += sizeof(uint16_t) * vq->size - 1;
1115         if (vstart <= iend && istart <= vend)
1116                 return 1;
1117
1118         vstart = (uintptr_t)vq->used;
1119         vend = vstart + sizeof(struct vring_used);
1120         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1121         if (vstart <= iend && istart <= vend)
1122                 return 1;
1123
1124         return 0;
1125 }
1126
1127 static int
1128 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1129 {
1130         struct virtio_net *dev = *pdev;
1131         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1132         uint16_t i;
1133         uint64_t vva;
1134
1135         switch (imsg->type) {
1136         case VHOST_IOTLB_UPDATE:
1137                 vva = qva_to_vva(dev, imsg->uaddr);
1138                 if (!vva)
1139                         return -1;
1140
1141                 for (i = 0; i < dev->nr_vring; i++) {
1142                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1143
1144                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1145                                         imsg->size, imsg->perm);
1146
1147                         if (is_vring_iotlb_update(vq, imsg))
1148                                 *pdev = dev = translate_ring_addresses(dev, i);
1149                 }
1150                 break;
1151         case VHOST_IOTLB_INVALIDATE:
1152                 for (i = 0; i < dev->nr_vring; i++) {
1153                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1154
1155                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1156                                         imsg->size);
1157
1158                         if (is_vring_iotlb_invalidate(vq, imsg))
1159                                 vring_invalidate(dev, vq);
1160                 }
1161                 break;
1162         default:
1163                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1164                                 imsg->type);
1165                 return -1;
1166         }
1167
1168         return 0;
1169 }
1170
1171 /* return bytes# of read on success or negative val on failure. */
1172 static int
1173 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1174 {
1175         int ret;
1176
1177         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1178                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1179         if (ret <= 0)
1180                 return ret;
1181
1182         if (msg && msg->size) {
1183                 if (msg->size > sizeof(msg->payload)) {
1184                         RTE_LOG(ERR, VHOST_CONFIG,
1185                                 "invalid msg size: %d\n", msg->size);
1186                         return -1;
1187                 }
1188                 ret = read(sockfd, &msg->payload, msg->size);
1189                 if (ret <= 0)
1190                         return ret;
1191                 if (ret != (int)msg->size) {
1192                         RTE_LOG(ERR, VHOST_CONFIG,
1193                                 "read control message failed\n");
1194                         return -1;
1195                 }
1196         }
1197
1198         return ret;
1199 }
1200
1201 static int
1202 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1203 {
1204         if (!msg)
1205                 return 0;
1206
1207         return send_fd_message(sockfd, (char *)msg,
1208                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1209 }
1210
1211 static int
1212 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1213 {
1214         if (!msg)
1215                 return 0;
1216
1217         msg->flags &= ~VHOST_USER_VERSION_MASK;
1218         msg->flags &= ~VHOST_USER_NEED_REPLY;
1219         msg->flags |= VHOST_USER_VERSION;
1220         msg->flags |= VHOST_USER_REPLY_MASK;
1221
1222         return send_vhost_message(sockfd, msg);
1223 }
1224
1225 /*
1226  * Allocate a queue pair if it hasn't been allocated yet
1227  */
1228 static int
1229 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1230 {
1231         uint16_t vring_idx;
1232
1233         switch (msg->request.master) {
1234         case VHOST_USER_SET_VRING_KICK:
1235         case VHOST_USER_SET_VRING_CALL:
1236         case VHOST_USER_SET_VRING_ERR:
1237                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1238                 break;
1239         case VHOST_USER_SET_VRING_NUM:
1240         case VHOST_USER_SET_VRING_BASE:
1241         case VHOST_USER_SET_VRING_ENABLE:
1242                 vring_idx = msg->payload.state.index;
1243                 break;
1244         case VHOST_USER_SET_VRING_ADDR:
1245                 vring_idx = msg->payload.addr.index;
1246                 break;
1247         default:
1248                 return 0;
1249         }
1250
1251         if (vring_idx >= VHOST_MAX_VRING) {
1252                 RTE_LOG(ERR, VHOST_CONFIG,
1253                         "invalid vring index: %u\n", vring_idx);
1254                 return -1;
1255         }
1256
1257         if (dev->virtqueue[vring_idx])
1258                 return 0;
1259
1260         return alloc_vring_queue(dev, vring_idx);
1261 }
1262
1263 int
1264 vhost_user_msg_handler(int vid, int fd)
1265 {
1266         struct virtio_net *dev;
1267         struct VhostUserMsg msg;
1268         int ret;
1269
1270         dev = get_device(vid);
1271         if (dev == NULL)
1272                 return -1;
1273
1274         if (!dev->notify_ops) {
1275                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1276                 if (!dev->notify_ops) {
1277                         RTE_LOG(ERR, VHOST_CONFIG,
1278                                 "failed to get callback ops for driver %s\n",
1279                                 dev->ifname);
1280                         return -1;
1281                 }
1282         }
1283
1284         ret = read_vhost_message(fd, &msg);
1285         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1286                 if (ret < 0)
1287                         RTE_LOG(ERR, VHOST_CONFIG,
1288                                 "vhost read message failed\n");
1289                 else if (ret == 0)
1290                         RTE_LOG(INFO, VHOST_CONFIG,
1291                                 "vhost peer closed\n");
1292                 else
1293                         RTE_LOG(ERR, VHOST_CONFIG,
1294                                 "vhost read incorrect message\n");
1295
1296                 return -1;
1297         }
1298
1299         ret = 0;
1300         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1301                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1302                         vhost_message_str[msg.request.master]);
1303         else
1304                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1305                         vhost_message_str[msg.request.master]);
1306
1307         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1308         if (ret < 0) {
1309                 RTE_LOG(ERR, VHOST_CONFIG,
1310                         "failed to alloc queue\n");
1311                 return -1;
1312         }
1313
1314         switch (msg.request.master) {
1315         case VHOST_USER_GET_FEATURES:
1316                 msg.payload.u64 = vhost_user_get_features(dev);
1317                 msg.size = sizeof(msg.payload.u64);
1318                 send_vhost_reply(fd, &msg);
1319                 break;
1320         case VHOST_USER_SET_FEATURES:
1321                 ret = vhost_user_set_features(dev, msg.payload.u64);
1322                 if (ret)
1323                         return -1;
1324                 break;
1325
1326         case VHOST_USER_GET_PROTOCOL_FEATURES:
1327                 vhost_user_get_protocol_features(dev, &msg);
1328                 send_vhost_reply(fd, &msg);
1329                 break;
1330         case VHOST_USER_SET_PROTOCOL_FEATURES:
1331                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1332                 break;
1333
1334         case VHOST_USER_SET_OWNER:
1335                 vhost_user_set_owner();
1336                 break;
1337         case VHOST_USER_RESET_OWNER:
1338                 vhost_user_reset_owner(dev);
1339                 break;
1340
1341         case VHOST_USER_SET_MEM_TABLE:
1342                 ret = vhost_user_set_mem_table(dev, &msg);
1343                 break;
1344
1345         case VHOST_USER_SET_LOG_BASE:
1346                 vhost_user_set_log_base(dev, &msg);
1347
1348                 /* it needs a reply */
1349                 msg.size = sizeof(msg.payload.u64);
1350                 send_vhost_reply(fd, &msg);
1351                 break;
1352         case VHOST_USER_SET_LOG_FD:
1353                 close(msg.fds[0]);
1354                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1355                 break;
1356
1357         case VHOST_USER_SET_VRING_NUM:
1358                 vhost_user_set_vring_num(dev, &msg);
1359                 break;
1360         case VHOST_USER_SET_VRING_ADDR:
1361                 vhost_user_set_vring_addr(&dev, &msg);
1362                 break;
1363         case VHOST_USER_SET_VRING_BASE:
1364                 vhost_user_set_vring_base(dev, &msg);
1365                 break;
1366
1367         case VHOST_USER_GET_VRING_BASE:
1368                 vhost_user_get_vring_base(dev, &msg);
1369                 msg.size = sizeof(msg.payload.state);
1370                 send_vhost_reply(fd, &msg);
1371                 break;
1372
1373         case VHOST_USER_SET_VRING_KICK:
1374                 vhost_user_set_vring_kick(&dev, &msg);
1375                 break;
1376         case VHOST_USER_SET_VRING_CALL:
1377                 vhost_user_set_vring_call(dev, &msg);
1378                 break;
1379
1380         case VHOST_USER_SET_VRING_ERR:
1381                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1382                         close(msg.fds[0]);
1383                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1384                 break;
1385
1386         case VHOST_USER_GET_QUEUE_NUM:
1387                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1388                 msg.size = sizeof(msg.payload.u64);
1389                 send_vhost_reply(fd, &msg);
1390                 break;
1391
1392         case VHOST_USER_SET_VRING_ENABLE:
1393                 vhost_user_set_vring_enable(dev, &msg);
1394                 break;
1395         case VHOST_USER_SEND_RARP:
1396                 vhost_user_send_rarp(dev, &msg);
1397                 break;
1398
1399         case VHOST_USER_NET_SET_MTU:
1400                 ret = vhost_user_net_set_mtu(dev, &msg);
1401                 break;
1402
1403         case VHOST_USER_SET_SLAVE_REQ_FD:
1404                 ret = vhost_user_set_req_fd(dev, &msg);
1405                 break;
1406
1407         case VHOST_USER_IOTLB_MSG:
1408                 ret = vhost_user_iotlb_msg(&dev, &msg);
1409                 break;
1410
1411         default:
1412                 ret = -1;
1413                 break;
1414
1415         }
1416
1417         if (msg.flags & VHOST_USER_NEED_REPLY) {
1418                 msg.payload.u64 = !!ret;
1419                 msg.size = sizeof(msg.payload.u64);
1420                 send_vhost_reply(fd, &msg);
1421         }
1422
1423         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1424                 dev->flags |= VIRTIO_DEV_READY;
1425
1426                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1427                         if (dev->dequeue_zero_copy) {
1428                                 RTE_LOG(INFO, VHOST_CONFIG,
1429                                                 "dequeue zero copy is enabled\n");
1430                         }
1431
1432                         if (dev->notify_ops->new_device(dev->vid) == 0)
1433                                 dev->flags |= VIRTIO_DEV_RUNNING;
1434                 }
1435         }
1436
1437         return 0;
1438 }
1439
1440 int
1441 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1442 {
1443         int ret;
1444         struct VhostUserMsg msg = {
1445                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1446                 .flags = VHOST_USER_VERSION,
1447                 .size = sizeof(msg.payload.iotlb),
1448                 .payload.iotlb = {
1449                         .iova = iova,
1450                         .perm = perm,
1451                         .type = VHOST_IOTLB_MISS,
1452                 },
1453         };
1454
1455         ret = send_vhost_message(dev->slave_req_fd, &msg);
1456         if (ret < 0) {
1457                 RTE_LOG(ERR, VHOST_CONFIG,
1458                                 "Failed to send IOTLB miss message (%d)\n",
1459                                 ret);
1460                 return ret;
1461         }
1462
1463         return 0;
1464 }