lib/librte_vhost/vhost_user.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <string.h>
  38 #include <unistd.h>
  39 #include <sys/mman.h>
  40 #include <sys/types.h>
  41 #include <sys/stat.h>
  42 #include <assert.h>
  43 #ifdef RTE_LIBRTE_VHOST_NUMA
  44 #include <numaif.h>
  45 #endif
  46
  47 #include <rte_common.h>
  48 #include <rte_malloc.h>
  49 #include <rte_log.h>
  50
  51 #include "iotlb.h"
  52 #include "vhost.h"
  53 #include "vhost_user.h"
  54
  55 #define VIRTIO_MIN_MTU 68
  56 #define VIRTIO_MAX_MTU 65535
  57
  58 static const char *vhost_message_str[VHOST_USER_MAX] = {
  59         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  60         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  61         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  62         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  63         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  64         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  65         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  66         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  67         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  68         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  69         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  70         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  71         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  72         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  73         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  74         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  75         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  76         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  77         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  78         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  79         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  80         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  81         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  82 };
  83
  84 static uint64_t
  85 get_blk_size(int fd)
  86 {
  87         struct stat stat;
  88         int ret;
  89
  90         ret = fstat(fd, &stat);
  91         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  92 }
  93
  94 static void
  95 free_mem_region(struct virtio_net *dev)
  96 {
  97         uint32_t i;
  98         struct rte_vhost_mem_region *reg;
  99
 100         if (!dev || !dev->mem)
 101                 return;
 102
 103         for (i = 0; i < dev->mem->nregions; i++) {
 104                 reg = &dev->mem->regions[i];
 105                 if (reg->host_user_addr) {
 106                         munmap(reg->mmap_addr, reg->mmap_size);
 107                         close(reg->fd);
 108                 }
 109         }
 110 }
 111
 112 void
 113 vhost_backend_cleanup(struct virtio_net *dev)
 114 {
 115         if (dev->mem) {
 116                 free_mem_region(dev);
 117                 rte_free(dev->mem);
 118                 dev->mem = NULL;
 119         }
 120
 121         free(dev->guest_pages);
 122         dev->guest_pages = NULL;
 123
 124         if (dev->log_addr) {
 125                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 126                 dev->log_addr = 0;
 127         }
 128
 129         if (dev->slave_req_fd >= 0) {
 130                 close(dev->slave_req_fd);
 131                 dev->slave_req_fd = -1;
 132         }
 133 }
 134
 135 /*
 136  * This function just returns success at the moment unless
 137  * the device hasn't been initialised.
 138  */
 139 static int
 140 vhost_user_set_owner(void)
 141 {
 142         return 0;
 143 }
 144
 145 static int
 146 vhost_user_reset_owner(struct virtio_net *dev)
 147 {
 148         if (dev->flags & VIRTIO_DEV_RUNNING) {
 149                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 150                 dev->notify_ops->destroy_device(dev->vid);
 151         }
 152
 153         cleanup_device(dev, 0);
 154         reset_device(dev);
 155         return 0;
 156 }
 157
 158 /*
 159  * The features that we support are requested.
 160  */
 161 static uint64_t
 162 vhost_user_get_features(struct virtio_net *dev)
 163 {
 164         uint64_t features = 0;
 165
 166         rte_vhost_driver_get_features(dev->ifname, &features);
 167         return features;
 168 }
 169
 170 /*
 171  * We receive the negotiated features supported by us and the virtio device.
 172  */
 173 static int
 174 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 175 {
 176         uint64_t vhost_features = 0;
 177
 178         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 179         if (features & ~vhost_features) {
 180                 RTE_LOG(ERR, VHOST_CONFIG,
 181                         "(%d) received invalid negotiated features.\n",
 182                         dev->vid);
 183                 return -1;
 184         }
 185
 186         if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
 187                 if (dev->notify_ops->features_changed)
 188                         dev->notify_ops->features_changed(dev->vid, features);
 189         }
 190
 191         dev->features = features;
 192         if (dev->features &
 193                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 194                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 195         } else {
 196                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 197         }
 198         LOG_DEBUG(VHOST_CONFIG,
 199                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 200                 dev->vid,
 201                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 202                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 203
 204         return 0;
 205 }
 206
 207 /*
 208  * The virtio device sends us the size of the descriptor ring.
 209  */
 210 static int
 211 vhost_user_set_vring_num(struct virtio_net *dev,
 212                          VhostUserMsg *msg)
 213 {
 214         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 215
 216         vq->size = msg->payload.state.num;
 217
 218         if (dev->dequeue_zero_copy) {
 219                 vq->nr_zmbuf = 0;
 220                 vq->last_zmbuf_idx = 0;
 221                 vq->zmbuf_size = vq->size;
 222                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 223                                          sizeof(struct zcopy_mbuf), 0);
 224                 if (vq->zmbufs == NULL) {
 225                         RTE_LOG(WARNING, VHOST_CONFIG,
 226                                 "failed to allocate mem for zero copy; "
 227                                 "zero copy is force disabled\n");
 228                         dev->dequeue_zero_copy = 0;
 229                 }
 230         }
 231
 232         vq->shadow_used_ring = rte_malloc(NULL,
 233                                 vq->size * sizeof(struct vring_used_elem),
 234                                 RTE_CACHE_LINE_SIZE);
 235         if (!vq->shadow_used_ring) {
 236                 RTE_LOG(ERR, VHOST_CONFIG,
 237                         "failed to allocate memory for shadow used ring.\n");
 238                 return -1;
 239         }
 240
 241         vq->batch_copy_elems = rte_malloc(NULL,
 242                                 vq->size * sizeof(struct batch_copy_elem),
 243                                 RTE_CACHE_LINE_SIZE);
 244         if (!vq->batch_copy_elems) {
 245                 RTE_LOG(ERR, VHOST_CONFIG,
 246                         "failed to allocate memory for batching copy.\n");
 247                 return -1;
 248         }
 249
 250         return 0;
 251 }
 252
 253 /*
 254  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 255  * same numa node as the memory of vring descriptor.
 256  */
 257 #ifdef RTE_LIBRTE_VHOST_NUMA
 258 static struct virtio_net*
 259 numa_realloc(struct virtio_net *dev, int index)
 260 {
 261         int oldnode, newnode;
 262         struct virtio_net *old_dev;
 263         struct vhost_virtqueue *old_vq, *vq;
 264         int ret;
 265
 266         old_dev = dev;
 267         vq = old_vq = dev->virtqueue[index];
 268
 269         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 270                             MPOL_F_NODE | MPOL_F_ADDR);
 271
 272         /* check if we need to reallocate vq */
 273         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 274                              MPOL_F_NODE | MPOL_F_ADDR);
 275         if (ret) {
 276                 RTE_LOG(ERR, VHOST_CONFIG,
 277                         "Unable to get vq numa information.\n");
 278                 return dev;
 279         }
 280         if (oldnode != newnode) {
 281                 RTE_LOG(INFO, VHOST_CONFIG,
 282                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 283                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 284                 if (!vq)
 285                         return dev;
 286
 287                 memcpy(vq, old_vq, sizeof(*vq));
 288                 rte_free(old_vq);
 289         }
 290
 291         /* check if we need to reallocate dev */
 292         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 293                             MPOL_F_NODE | MPOL_F_ADDR);
 294         if (ret) {
 295                 RTE_LOG(ERR, VHOST_CONFIG,
 296                         "Unable to get dev numa information.\n");
 297                 goto out;
 298         }
 299         if (oldnode != newnode) {
 300                 RTE_LOG(INFO, VHOST_CONFIG,
 301                         "reallocate dev from %d to %d node\n",
 302                         oldnode, newnode);
 303                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 304                 if (!dev) {
 305                         dev = old_dev;
 306                         goto out;
 307                 }
 308
 309                 memcpy(dev, old_dev, sizeof(*dev));
 310                 rte_free(old_dev);
 311         }
 312
 313 out:
 314         dev->virtqueue[index] = vq;
 315         vhost_devices[dev->vid] = dev;
 316
 317         return dev;
 318 }
 319 #else
 320 static struct virtio_net*
 321 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 322 {
 323         return dev;
 324 }
 325 #endif
 326
 327 /*
 328  * Converts QEMU virtual address to Vhost virtual address. This function is
 329  * used to convert the ring addresses to our address space.
 330  */
 331 static uint64_t
 332 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 333 {
 334         struct rte_vhost_mem_region *reg;
 335         uint32_t i;
 336
 337         /* Find the region where the address lives. */
 338         for (i = 0; i < dev->mem->nregions; i++) {
 339                 reg = &dev->mem->regions[i];
 340
 341                 if (qva >= reg->guest_user_addr &&
 342                     qva <  reg->guest_user_addr + reg->size) {
 343                         return qva - reg->guest_user_addr +
 344                                reg->host_user_addr;
 345                 }
 346         }
 347
 348         return 0;
 349 }
 350
 351 /*
 352  * The virtio device sends us the desc, used and avail ring addresses.
 353  * This function then converts these to our address space.
 354  */
 355 static int
 356 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
 357 {
 358         struct vhost_virtqueue *vq;
 359         struct vhost_vring_addr *addr = &msg->payload.addr;
 360
 361         if (dev->mem == NULL)
 362                 return -1;
 363
 364         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 365         vq = dev->virtqueue[msg->payload.addr.index];
 366
 367         /*
 368          * Rings addresses should not be interpreted as long as the ring is not
 369          * started and enabled
 370          */
 371         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 372
 373         return 0;
 374 }
 375
 376 static struct virtio_net *
 377 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 378 {
 379         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 380         struct vhost_vring_addr *addr = &vq->ring_addrs;
 381
 382         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 383         vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
 384                         addr->desc_user_addr);
 385         if (vq->desc == 0) {
 386                 RTE_LOG(ERR, VHOST_CONFIG,
 387                         "(%d) failed to find desc ring address.\n",
 388                         dev->vid);
 389                 return NULL;
 390         }
 391
 392         dev = numa_realloc(dev, vq_index);
 393         vq = dev->virtqueue[vq_index];
 394
 395         vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
 396                         addr->avail_user_addr);
 397         if (vq->avail == 0) {
 398                 RTE_LOG(ERR, VHOST_CONFIG,
 399                         "(%d) failed to find avail ring address.\n",
 400                         dev->vid);
 401                 return NULL;
 402         }
 403
 404         vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
 405                         addr->used_user_addr);
 406         if (vq->used == 0) {
 407                 RTE_LOG(ERR, VHOST_CONFIG,
 408                         "(%d) failed to find used ring address.\n",
 409                         dev->vid);
 410                 return NULL;
 411         }
 412
 413         if (vq->last_used_idx != vq->used->idx) {
 414                 RTE_LOG(WARNING, VHOST_CONFIG,
 415                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 416                         "some packets maybe resent for Tx and dropped for Rx\n",
 417                         vq->last_used_idx, vq->used->idx);
 418                 vq->last_used_idx  = vq->used->idx;
 419                 vq->last_avail_idx = vq->used->idx;
 420         }
 421
 422         vq->log_guest_addr = addr->log_guest_addr;
 423
 424         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 425                         dev->vid, vq->desc);
 426         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 427                         dev->vid, vq->avail);
 428         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 429                         dev->vid, vq->used);
 430         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 431                         dev->vid, vq->log_guest_addr);
 432
 433         return dev;
 434 }
 435
 436 /*
 437  * The virtio device sends us the available ring last used index.
 438  */
 439 static int
 440 vhost_user_set_vring_base(struct virtio_net *dev,
 441                           VhostUserMsg *msg)
 442 {
 443         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 444                         msg->payload.state.num;
 445         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 446                         msg->payload.state.num;
 447
 448         return 0;
 449 }
 450
 451 static void
 452 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 453                    uint64_t host_phys_addr, uint64_t size)
 454 {
 455         struct guest_page *page, *last_page;
 456
 457         if (dev->nr_guest_pages == dev->max_guest_pages) {
 458                 dev->max_guest_pages *= 2;
 459                 dev->guest_pages = realloc(dev->guest_pages,
 460                                         dev->max_guest_pages * sizeof(*page));
 461         }
 462
 463         if (dev->nr_guest_pages > 0) {
 464                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 465                 /* merge if the two pages are continuous */
 466                 if (host_phys_addr == last_page->host_phys_addr +
 467                                       last_page->size) {
 468                         last_page->size += size;
 469                         return;
 470                 }
 471         }
 472
 473         page = &dev->guest_pages[dev->nr_guest_pages++];
 474         page->guest_phys_addr = guest_phys_addr;
 475         page->host_phys_addr  = host_phys_addr;
 476         page->size = size;
 477 }
 478
 479 static void
 480 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 481                 uint64_t page_size)
 482 {
 483         uint64_t reg_size = reg->size;
 484         uint64_t host_user_addr  = reg->host_user_addr;
 485         uint64_t guest_phys_addr = reg->guest_phys_addr;
 486         uint64_t host_phys_addr;
 487         uint64_t size;
 488
 489         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 490         size = page_size - (guest_phys_addr & (page_size - 1));
 491         size = RTE_MIN(size, reg_size);
 492
 493         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 494         host_user_addr  += size;
 495         guest_phys_addr += size;
 496         reg_size -= size;
 497
 498         while (reg_size > 0) {
 499                 size = RTE_MIN(reg_size, page_size);
 500                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 501                                                   host_user_addr);
 502                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 503
 504                 host_user_addr  += size;
 505                 guest_phys_addr += size;
 506                 reg_size -= size;
 507         }
 508 }
 509
 510 #ifdef RTE_LIBRTE_VHOST_DEBUG
 511 /* TODO: enable it only in debug mode? */
 512 static void
 513 dump_guest_pages(struct virtio_net *dev)
 514 {
 515         uint32_t i;
 516         struct guest_page *page;
 517
 518         for (i = 0; i < dev->nr_guest_pages; i++) {
 519                 page = &dev->guest_pages[i];
 520
 521                 RTE_LOG(INFO, VHOST_CONFIG,
 522                         "guest physical page region %u\n"
 523                         "\t guest_phys_addr: %" PRIx64 "\n"
 524                         "\t host_phys_addr : %" PRIx64 "\n"
 525                         "\t size           : %" PRIx64 "\n",
 526                         i,
 527                         page->guest_phys_addr,
 528                         page->host_phys_addr,
 529                         page->size);
 530         }
 531 }
 532 #else
 533 #define dump_guest_pages(dev)
 534 #endif
 535
 536 static int
 537 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 538 {
 539         struct VhostUserMemory memory = pmsg->payload.memory;
 540         struct rte_vhost_mem_region *reg;
 541         void *mmap_addr;
 542         uint64_t mmap_size;
 543         uint64_t mmap_offset;
 544         uint64_t alignment;
 545         uint32_t i;
 546         int fd;
 547
 548         if (dev->mem) {
 549                 free_mem_region(dev);
 550                 rte_free(dev->mem);
 551                 dev->mem = NULL;
 552         }
 553
 554         dev->nr_guest_pages = 0;
 555         if (!dev->guest_pages) {
 556                 dev->max_guest_pages = 8;
 557                 dev->guest_pages = malloc(dev->max_guest_pages *
 558                                                 sizeof(struct guest_page));
 559                 if (dev->guest_pages == NULL) {
 560                         RTE_LOG(ERR, VHOST_CONFIG,
 561                                 "(%d) failed to allocate memory "
 562                                 "for dev->guest_pages\n",
 563                                 dev->vid);
 564                         return -1;
 565                 }
 566         }
 567
 568         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 569                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 570         if (dev->mem == NULL) {
 571                 RTE_LOG(ERR, VHOST_CONFIG,
 572                         "(%d) failed to allocate memory for dev->mem\n",
 573                         dev->vid);
 574                 return -1;
 575         }
 576         dev->mem->nregions = memory.nregions;
 577
 578         for (i = 0; i < memory.nregions; i++) {
 579                 fd  = pmsg->fds[i];
 580                 reg = &dev->mem->regions[i];
 581
 582                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 583                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 584                 reg->size            = memory.regions[i].memory_size;
 585                 reg->fd              = fd;
 586
 587                 mmap_offset = memory.regions[i].mmap_offset;
 588                 mmap_size   = reg->size + mmap_offset;
 589
 590                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 591                  * with length argument aligned with hugepagesz at older
 592                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 593                  * mmap() will fail with EINVAL.
 594                  *
 595                  * to avoid failure, make sure in caller to keep length
 596                  * aligned.
 597                  */
 598                 alignment = get_blk_size(fd);
 599                 if (alignment == (uint64_t)-1) {
 600                         RTE_LOG(ERR, VHOST_CONFIG,
 601                                 "couldn't get hugepage size through fstat\n");
 602                         goto err_mmap;
 603                 }
 604                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 605
 606                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 607                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 608
 609                 if (mmap_addr == MAP_FAILED) {
 610                         RTE_LOG(ERR, VHOST_CONFIG,
 611                                 "mmap region %u failed.\n", i);
 612                         goto err_mmap;
 613                 }
 614
 615                 reg->mmap_addr = mmap_addr;
 616                 reg->mmap_size = mmap_size;
 617                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 618                                       mmap_offset;
 619
 620                 if (dev->dequeue_zero_copy)
 621                         add_guest_pages(dev, reg, alignment);
 622
 623                 RTE_LOG(INFO, VHOST_CONFIG,
 624                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 625                         "\t guest physical addr: 0x%" PRIx64 "\n"
 626                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 627                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 628                         "\t mmap addr : 0x%" PRIx64 "\n"
 629                         "\t mmap size : 0x%" PRIx64 "\n"
 630                         "\t mmap align: 0x%" PRIx64 "\n"
 631                         "\t mmap off  : 0x%" PRIx64 "\n",
 632                         i, reg->size,
 633                         reg->guest_phys_addr,
 634                         reg->guest_user_addr,
 635                         reg->host_user_addr,
 636                         (uint64_t)(uintptr_t)mmap_addr,
 637                         mmap_size,
 638                         alignment,
 639                         mmap_offset);
 640         }
 641
 642         dump_guest_pages(dev);
 643
 644         return 0;
 645
 646 err_mmap:
 647         free_mem_region(dev);
 648         rte_free(dev->mem);
 649         dev->mem = NULL;
 650         return -1;
 651 }
 652
 653 static int
 654 vq_is_ready(struct vhost_virtqueue *vq)
 655 {
 656         return vq && vq->desc   &&
 657                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 658                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 659 }
 660
 661 static int
 662 virtio_is_ready(struct virtio_net *dev)
 663 {
 664         struct vhost_virtqueue *vq;
 665         uint32_t i;
 666
 667         if (dev->nr_vring == 0)
 668                 return 0;
 669
 670         for (i = 0; i < dev->nr_vring; i++) {
 671                 vq = dev->virtqueue[i];
 672
 673                 if (!vq_is_ready(vq))
 674                         return 0;
 675         }
 676
 677         RTE_LOG(INFO, VHOST_CONFIG,
 678                 "virtio is now ready for processing.\n");
 679         return 1;
 680 }
 681
 682 static void
 683 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 684 {
 685         struct vhost_vring_file file;
 686         struct vhost_virtqueue *vq;
 687
 688         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 689         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 690                 file.fd = VIRTIO_INVALID_EVENTFD;
 691         else
 692                 file.fd = pmsg->fds[0];
 693         RTE_LOG(INFO, VHOST_CONFIG,
 694                 "vring call idx:%d file:%d\n", file.index, file.fd);
 695
 696         vq = dev->virtqueue[file.index];
 697         if (vq->callfd >= 0)
 698                 close(vq->callfd);
 699
 700         vq->callfd = file.fd;
 701 }
 702
 703 static void
 704 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 705 {
 706         struct vhost_vring_file file;
 707         struct vhost_virtqueue *vq;
 708         struct virtio_net *dev = *pdev;
 709
 710         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 711         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 712                 file.fd = VIRTIO_INVALID_EVENTFD;
 713         else
 714                 file.fd = pmsg->fds[0];
 715         RTE_LOG(INFO, VHOST_CONFIG,
 716                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 717
 718         /*
 719          * Interpret ring addresses only when ring is started and enabled.
 720          * This is now if protocol features aren't supported.
 721          */
 722         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 723                 *pdev = dev = translate_ring_addresses(dev, file.index);
 724                 if (!dev)
 725                         return;
 726         }
 727
 728         vq = dev->virtqueue[file.index];
 729
 730         /*
 731          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 732          * the ring starts already enabled. Otherwise, it is enabled via
 733          * the SET_VRING_ENABLE message.
 734          */
 735         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 736                 vq->enabled = 1;
 737
 738         if (vq->kickfd >= 0)
 739                 close(vq->kickfd);
 740         vq->kickfd = file.fd;
 741 }
 742
 743 static void
 744 free_zmbufs(struct vhost_virtqueue *vq)
 745 {
 746         struct zcopy_mbuf *zmbuf, *next;
 747
 748         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 749              zmbuf != NULL; zmbuf = next) {
 750                 next = TAILQ_NEXT(zmbuf, next);
 751
 752                 rte_pktmbuf_free(zmbuf->mbuf);
 753                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 754         }
 755
 756         rte_free(vq->zmbufs);
 757 }
 758
 759 /*
 760  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 761  */
 762 static int
 763 vhost_user_get_vring_base(struct virtio_net *dev,
 764                           VhostUserMsg *msg)
 765 {
 766         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 767
 768         /* We have to stop the queue (virtio) if it is running. */
 769         if (dev->flags & VIRTIO_DEV_RUNNING) {
 770                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 771                 dev->notify_ops->destroy_device(dev->vid);
 772         }
 773
 774         dev->flags &= ~VIRTIO_DEV_READY;
 775
 776         /* Here we are safe to get the last used index */
 777         msg->payload.state.num = vq->last_used_idx;
 778
 779         RTE_LOG(INFO, VHOST_CONFIG,
 780                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 781                 msg->payload.state.num);
 782         /*
 783          * Based on current qemu vhost-user implementation, this message is
 784          * sent and only sent in vhost_vring_stop.
 785          * TODO: cleanup the vring, it isn't usable since here.
 786          */
 787         if (vq->kickfd >= 0)
 788                 close(vq->kickfd);
 789
 790         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 791
 792         if (dev->dequeue_zero_copy)
 793                 free_zmbufs(vq);
 794         rte_free(vq->shadow_used_ring);
 795         vq->shadow_used_ring = NULL;
 796
 797         rte_free(vq->batch_copy_elems);
 798         vq->batch_copy_elems = NULL;
 799
 800         return 0;
 801 }
 802
 803 /*
 804  * when virtio queues are ready to work, qemu will send us to
 805  * enable the virtio queue pair.
 806  */
 807 static int
 808 vhost_user_set_vring_enable(struct virtio_net **pdev,
 809                             VhostUserMsg *msg)
 810 {
 811         struct virtio_net *dev = *pdev;
 812         int enable = (int)msg->payload.state.num;
 813
 814         RTE_LOG(INFO, VHOST_CONFIG,
 815                 "set queue enable: %d to qp idx: %d\n",
 816                 enable, msg->payload.state.index);
 817
 818         /*
 819          * Interpret ring addresses only when ring is started and enabled.
 820          * This is now if protocol features are supported.
 821          */
 822         if (enable && (dev->features &
 823                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 824                 dev = translate_ring_addresses(dev, msg->payload.state.index);
 825                 if (!dev)
 826                         return -1;
 827
 828                 *pdev = dev;
 829         }
 830
 831         if (dev->notify_ops->vring_state_changed)
 832                 dev->notify_ops->vring_state_changed(dev->vid,
 833                                 msg->payload.state.index, enable);
 834
 835         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 836
 837         return 0;
 838 }
 839
 840 static void
 841 vhost_user_set_protocol_features(struct virtio_net *dev,
 842                                  uint64_t protocol_features)
 843 {
 844         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 845                 return;
 846
 847         dev->protocol_features = protocol_features;
 848 }
 849
 850 static int
 851 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 852 {
 853         int fd = msg->fds[0];
 854         uint64_t size, off;
 855         void *addr;
 856
 857         if (fd < 0) {
 858                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 859                 return -1;
 860         }
 861
 862         if (msg->size != sizeof(VhostUserLog)) {
 863                 RTE_LOG(ERR, VHOST_CONFIG,
 864                         "invalid log base msg size: %"PRId32" != %d\n",
 865                         msg->size, (int)sizeof(VhostUserLog));
 866                 return -1;
 867         }
 868
 869         size = msg->payload.log.mmap_size;
 870         off  = msg->payload.log.mmap_offset;
 871         RTE_LOG(INFO, VHOST_CONFIG,
 872                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 873                 size, off);
 874
 875         /*
 876          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 877          * fail when offset is not page size aligned.
 878          */
 879         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 880         close(fd);
 881         if (addr == MAP_FAILED) {
 882                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 883                 return -1;
 884         }
 885
 886         /*
 887          * Free previously mapped log memory on occasionally
 888          * multiple VHOST_USER_SET_LOG_BASE.
 889          */
 890         if (dev->log_addr) {
 891                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 892         }
 893         dev->log_addr = (uint64_t)(uintptr_t)addr;
 894         dev->log_base = dev->log_addr + off;
 895         dev->log_size = size;
 896
 897         return 0;
 898 }
 899
 900 /*
 901  * An rarp packet is constructed and broadcasted to notify switches about
 902  * the new location of the migrated VM, so that packets from outside will
 903  * not be lost after migration.
 904  *
 905  * However, we don't actually "send" a rarp packet here, instead, we set
 906  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 907  */
 908 static int
 909 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 910 {
 911         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 912
 913         RTE_LOG(DEBUG, VHOST_CONFIG,
 914                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 915                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 916         memcpy(dev->mac.addr_bytes, mac, 6);
 917
 918         /*
 919          * Set the flag to inject a RARP broadcast packet at
 920          * rte_vhost_dequeue_burst().
 921          *
 922          * rte_smp_wmb() is for making sure the mac is copied
 923          * before the flag is set.
 924          */
 925         rte_smp_wmb();
 926         rte_atomic16_set(&dev->broadcast_rarp, 1);
 927
 928         return 0;
 929 }
 930
 931 static int
 932 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 933 {
 934         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 935                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 936                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 937                                 msg->payload.u64);
 938
 939                 return -1;
 940         }
 941
 942         dev->mtu = msg->payload.u64;
 943
 944         return 0;
 945 }
 946
 947 static int
 948 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 949 {
 950         int fd = msg->fds[0];
 951
 952         if (fd < 0) {
 953                 RTE_LOG(ERR, VHOST_CONFIG,
 954                                 "Invalid file descriptor for slave channel (%d)\n",
 955                                 fd);
 956                 return -1;
 957         }
 958
 959         dev->slave_req_fd = fd;
 960
 961         return 0;
 962 }
 963
 964 static int
 965 vhost_user_iotlb_msg(struct virtio_net *dev, struct VhostUserMsg *msg)
 966 {
 967         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
 968         uint16_t i;
 969         uint64_t vva;
 970
 971         switch (imsg->type) {
 972         case VHOST_IOTLB_UPDATE:
 973                 vva = qva_to_vva(dev, imsg->uaddr);
 974                 if (!vva)
 975                         return -1;
 976
 977                 for (i = 0; i < dev->nr_vring; i++) {
 978                         struct vhost_virtqueue *vq = dev->virtqueue[i];
 979
 980                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
 981                                         imsg->size, imsg->perm);
 982                 }
 983                 break;
 984         case VHOST_IOTLB_INVALIDATE:
 985                 for (i = 0; i < dev->nr_vring; i++) {
 986                         struct vhost_virtqueue *vq = dev->virtqueue[i];
 987
 988                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
 989                                         imsg->size);
 990                 }
 991                 break;
 992         default:
 993                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
 994                                 imsg->type);
 995                 return -1;
 996         }
 997
 998         return 0;
 999 }
1000
1001 /* return bytes# of read on success or negative val on failure. */
1002 static int
1003 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1004 {
1005         int ret;
1006
1007         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1008                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1009         if (ret <= 0)
1010                 return ret;
1011
1012         if (msg && msg->size) {
1013                 if (msg->size > sizeof(msg->payload)) {
1014                         RTE_LOG(ERR, VHOST_CONFIG,
1015                                 "invalid msg size: %d\n", msg->size);
1016                         return -1;
1017                 }
1018                 ret = read(sockfd, &msg->payload, msg->size);
1019                 if (ret <= 0)
1020                         return ret;
1021                 if (ret != (int)msg->size) {
1022                         RTE_LOG(ERR, VHOST_CONFIG,
1023                                 "read control message failed\n");
1024                         return -1;
1025                 }
1026         }
1027
1028         return ret;
1029 }
1030
1031 static int
1032 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1033 {
1034         if (!msg)
1035                 return 0;
1036
1037         return send_fd_message(sockfd, (char *)msg,
1038                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1039 }
1040
1041 static int
1042 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1043 {
1044         if (!msg)
1045                 return 0;
1046
1047         msg->flags &= ~VHOST_USER_VERSION_MASK;
1048         msg->flags &= ~VHOST_USER_NEED_REPLY;
1049         msg->flags |= VHOST_USER_VERSION;
1050         msg->flags |= VHOST_USER_REPLY_MASK;
1051
1052         return send_vhost_message(sockfd, msg);
1053 }
1054
1055 /*
1056  * Allocate a queue pair if it hasn't been allocated yet
1057  */
1058 static int
1059 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1060 {
1061         uint16_t vring_idx;
1062
1063         switch (msg->request) {
1064         case VHOST_USER_SET_VRING_KICK:
1065         case VHOST_USER_SET_VRING_CALL:
1066         case VHOST_USER_SET_VRING_ERR:
1067                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1068                 break;
1069         case VHOST_USER_SET_VRING_NUM:
1070         case VHOST_USER_SET_VRING_BASE:
1071         case VHOST_USER_SET_VRING_ENABLE:
1072                 vring_idx = msg->payload.state.index;
1073                 break;
1074         case VHOST_USER_SET_VRING_ADDR:
1075                 vring_idx = msg->payload.addr.index;
1076                 break;
1077         default:
1078                 return 0;
1079         }
1080
1081         if (vring_idx >= VHOST_MAX_VRING) {
1082                 RTE_LOG(ERR, VHOST_CONFIG,
1083                         "invalid vring index: %u\n", vring_idx);
1084                 return -1;
1085         }
1086
1087         if (dev->virtqueue[vring_idx])
1088                 return 0;
1089
1090         return alloc_vring_queue(dev, vring_idx);
1091 }
1092
1093 int
1094 vhost_user_msg_handler(int vid, int fd)
1095 {
1096         struct virtio_net *dev;
1097         struct VhostUserMsg msg;
1098         int ret;
1099
1100         dev = get_device(vid);
1101         if (dev == NULL)
1102                 return -1;
1103
1104         if (!dev->notify_ops) {
1105                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1106                 if (!dev->notify_ops) {
1107                         RTE_LOG(ERR, VHOST_CONFIG,
1108                                 "failed to get callback ops for driver %s\n",
1109                                 dev->ifname);
1110                         return -1;
1111                 }
1112         }
1113
1114         ret = read_vhost_message(fd, &msg);
1115         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1116                 if (ret < 0)
1117                         RTE_LOG(ERR, VHOST_CONFIG,
1118                                 "vhost read message failed\n");
1119                 else if (ret == 0)
1120                         RTE_LOG(INFO, VHOST_CONFIG,
1121                                 "vhost peer closed\n");
1122                 else
1123                         RTE_LOG(ERR, VHOST_CONFIG,
1124                                 "vhost read incorrect message\n");
1125
1126                 return -1;
1127         }
1128
1129         ret = 0;
1130         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1131                 vhost_message_str[msg.request]);
1132
1133         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1134         if (ret < 0) {
1135                 RTE_LOG(ERR, VHOST_CONFIG,
1136                         "failed to alloc queue\n");
1137                 return -1;
1138         }
1139
1140         switch (msg.request) {
1141         case VHOST_USER_GET_FEATURES:
1142                 msg.payload.u64 = vhost_user_get_features(dev);
1143                 msg.size = sizeof(msg.payload.u64);
1144                 send_vhost_reply(fd, &msg);
1145                 break;
1146         case VHOST_USER_SET_FEATURES:
1147                 vhost_user_set_features(dev, msg.payload.u64);
1148                 break;
1149
1150         case VHOST_USER_GET_PROTOCOL_FEATURES:
1151                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1152                 msg.size = sizeof(msg.payload.u64);
1153                 send_vhost_reply(fd, &msg);
1154                 break;
1155         case VHOST_USER_SET_PROTOCOL_FEATURES:
1156                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1157                 break;
1158
1159         case VHOST_USER_SET_OWNER:
1160                 vhost_user_set_owner();
1161                 break;
1162         case VHOST_USER_RESET_OWNER:
1163                 vhost_user_reset_owner(dev);
1164                 break;
1165
1166         case VHOST_USER_SET_MEM_TABLE:
1167                 ret = vhost_user_set_mem_table(dev, &msg);
1168                 break;
1169
1170         case VHOST_USER_SET_LOG_BASE:
1171                 vhost_user_set_log_base(dev, &msg);
1172
1173                 /* it needs a reply */
1174                 msg.size = sizeof(msg.payload.u64);
1175                 send_vhost_reply(fd, &msg);
1176                 break;
1177         case VHOST_USER_SET_LOG_FD:
1178                 close(msg.fds[0]);
1179                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1180                 break;
1181
1182         case VHOST_USER_SET_VRING_NUM:
1183                 vhost_user_set_vring_num(dev, &msg);
1184                 break;
1185         case VHOST_USER_SET_VRING_ADDR:
1186                 vhost_user_set_vring_addr(dev, &msg);
1187                 break;
1188         case VHOST_USER_SET_VRING_BASE:
1189                 vhost_user_set_vring_base(dev, &msg);
1190                 break;
1191
1192         case VHOST_USER_GET_VRING_BASE:
1193                 vhost_user_get_vring_base(dev, &msg);
1194                 msg.size = sizeof(msg.payload.state);
1195                 send_vhost_reply(fd, &msg);
1196                 break;
1197
1198         case VHOST_USER_SET_VRING_KICK:
1199                 vhost_user_set_vring_kick(&dev, &msg);
1200                 break;
1201         case VHOST_USER_SET_VRING_CALL:
1202                 vhost_user_set_vring_call(dev, &msg);
1203                 break;
1204
1205         case VHOST_USER_SET_VRING_ERR:
1206                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1207                         close(msg.fds[0]);
1208                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1209                 break;
1210
1211         case VHOST_USER_GET_QUEUE_NUM:
1212                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1213                 msg.size = sizeof(msg.payload.u64);
1214                 send_vhost_reply(fd, &msg);
1215                 break;
1216
1217         case VHOST_USER_SET_VRING_ENABLE:
1218                 vhost_user_set_vring_enable(&dev, &msg);
1219                 break;
1220         case VHOST_USER_SEND_RARP:
1221                 vhost_user_send_rarp(dev, &msg);
1222                 break;
1223
1224         case VHOST_USER_NET_SET_MTU:
1225                 ret = vhost_user_net_set_mtu(dev, &msg);
1226                 break;
1227
1228         case VHOST_USER_SET_SLAVE_REQ_FD:
1229                 ret = vhost_user_set_req_fd(dev, &msg);
1230                 break;
1231
1232         case VHOST_USER_IOTLB_MSG:
1233                 ret = vhost_user_iotlb_msg(dev, &msg);
1234                 break;
1235
1236         default:
1237                 ret = -1;
1238                 break;
1239
1240         }
1241
1242         /*
1243          * The virtio_net struct might have been reallocated on a different
1244          * NUMA node, so dev pointer might no more be valid.
1245          */
1246         dev = get_device(vid);
1247
1248         if (msg.flags & VHOST_USER_NEED_REPLY) {
1249                 msg.payload.u64 = !!ret;
1250                 msg.size = sizeof(msg.payload.u64);
1251                 send_vhost_reply(fd, &msg);
1252         }
1253
1254         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1255                 dev->flags |= VIRTIO_DEV_READY;
1256
1257                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1258                         if (dev->dequeue_zero_copy) {
1259                                 RTE_LOG(INFO, VHOST_CONFIG,
1260                                                 "dequeue zero copy is enabled\n");
1261                         }
1262
1263                         if (dev->notify_ops->new_device(dev->vid) == 0)
1264                                 dev->flags |= VIRTIO_DEV_RUNNING;
1265                 }
1266         }
1267
1268         return 0;
1269 }
1270
1271 int
1272 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1273 {
1274         int ret;
1275         struct VhostUserMsg msg = {
1276                 .request = (enum VhostUserRequest)VHOST_USER_SLAVE_IOTLB_MSG,
1277                 .flags = VHOST_USER_VERSION,
1278                 .size = sizeof(msg.payload.iotlb),
1279                 .payload.iotlb = {
1280                         .iova = iova,
1281                         .perm = perm,
1282                         .type = VHOST_IOTLB_MISS,
1283                 },
1284         };
1285
1286         ret = send_vhost_message(dev->slave_req_fd, &msg);
1287         if (ret < 0) {
1288                 RTE_LOG(ERR, VHOST_CONFIG,
1289                                 "Failed to send IOTLB miss message (%d)\n",
1290                                 ret);
1291                 return ret;
1292         }
1293
1294         return 0;
1295 }