lib/librte_vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 /* Security model
   6  * --------------
   7  * The vhost-user protocol connection is an external interface, so it must be
   8  * robust against invalid inputs.
   9  *
  10  * This is important because the vhost-user master is only one step removed
  11  * from the guest.  Malicious guests that have escaped will then launch further
  12  * attacks from the vhost-user master.
  13  *
  14  * Even in deployments where guests are trusted, a bug in the vhost-user master
  15  * can still cause invalid messages to be sent.  Such messages must not
  16  * compromise the stability of the DPDK application by causing crashes, memory
  17  * corruption, or other problematic behavior.
  18  *
  19  * Do not assume received VhostUserMsg fields contain sensible values!
  20  */
  21
  22 #include <stdint.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <sys/mman.h>
  28 #include <sys/types.h>
  29 #include <sys/stat.h>
  30 #include <assert.h>
  31 #ifdef RTE_LIBRTE_VHOST_NUMA
  32 #include <numaif.h>
  33 #endif
  34
  35 #include <rte_common.h>
  36 #include <rte_malloc.h>
  37 #include <rte_log.h>
  38
  39 #include "iotlb.h"
  40 #include "vhost.h"
  41 #include "vhost_user.h"
  42
  43 #define VIRTIO_MIN_MTU 68
  44 #define VIRTIO_MAX_MTU 65535
  45
  46 static const char *vhost_message_str[VHOST_USER_MAX] = {
  47         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  48         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  49         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  50         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  51         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  52         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  53         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  54         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  55         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  56         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  57         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  58         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  59         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  60         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  61         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  62         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  63         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  64         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  65         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  66         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  67         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  68         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  69         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  70         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  71         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  72 };
  73
  74 static uint64_t
  75 get_blk_size(int fd)
  76 {
  77         struct stat stat;
  78         int ret;
  79
  80         ret = fstat(fd, &stat);
  81         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  82 }
  83
  84 static void
  85 free_mem_region(struct virtio_net *dev)
  86 {
  87         uint32_t i;
  88         struct rte_vhost_mem_region *reg;
  89
  90         if (!dev || !dev->mem)
  91                 return;
  92
  93         for (i = 0; i < dev->mem->nregions; i++) {
  94                 reg = &dev->mem->regions[i];
  95                 if (reg->host_user_addr) {
  96                         munmap(reg->mmap_addr, reg->mmap_size);
  97                         close(reg->fd);
  98                 }
  99         }
 100 }
 101
 102 void
 103 vhost_backend_cleanup(struct virtio_net *dev)
 104 {
 105         if (dev->mem) {
 106                 free_mem_region(dev);
 107                 rte_free(dev->mem);
 108                 dev->mem = NULL;
 109         }
 110
 111         free(dev->guest_pages);
 112         dev->guest_pages = NULL;
 113
 114         if (dev->log_addr) {
 115                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 116                 dev->log_addr = 0;
 117         }
 118
 119         if (dev->slave_req_fd >= 0) {
 120                 close(dev->slave_req_fd);
 121                 dev->slave_req_fd = -1;
 122         }
 123 }
 124
 125 /*
 126  * This function just returns success at the moment unless
 127  * the device hasn't been initialised.
 128  */
 129 static int
 130 vhost_user_set_owner(void)
 131 {
 132         return 0;
 133 }
 134
 135 static int
 136 vhost_user_reset_owner(struct virtio_net *dev)
 137 {
 138         vhost_destroy_device_notify(dev);
 139
 140         cleanup_device(dev, 0);
 141         reset_device(dev);
 142         return 0;
 143 }
 144
 145 /*
 146  * The features that we support are requested.
 147  */
 148 static uint64_t
 149 vhost_user_get_features(struct virtio_net *dev)
 150 {
 151         uint64_t features = 0;
 152
 153         rte_vhost_driver_get_features(dev->ifname, &features);
 154         return features;
 155 }
 156
 157 /*
 158  * The queue number that we support are requested.
 159  */
 160 static uint32_t
 161 vhost_user_get_queue_num(struct virtio_net *dev)
 162 {
 163         uint32_t queue_num = 0;
 164
 165         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 166         return queue_num;
 167 }
 168
 169 /*
 170  * We receive the negotiated features supported by us and the virtio device.
 171  */
 172 static int
 173 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 174 {
 175         uint64_t vhost_features = 0;
 176         struct rte_vdpa_device *vdpa_dev;
 177         int did = -1;
 178
 179         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 180         if (features & ~vhost_features) {
 181                 RTE_LOG(ERR, VHOST_CONFIG,
 182                         "(%d) received invalid negotiated features.\n",
 183                         dev->vid);
 184                 return -1;
 185         }
 186
 187         if (dev->flags & VIRTIO_DEV_RUNNING) {
 188                 if (dev->features == features)
 189                         return 0;
 190
 191                 /*
 192                  * Error out if master tries to change features while device is
 193                  * in running state. The exception being VHOST_F_LOG_ALL, which
 194                  * is enabled when the live-migration starts.
 195                  */
 196                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 197                         RTE_LOG(ERR, VHOST_CONFIG,
 198                                 "(%d) features changed while device is running.\n",
 199                                 dev->vid);
 200                         return -1;
 201                 }
 202
 203                 if (dev->notify_ops->features_changed)
 204                         dev->notify_ops->features_changed(dev->vid, features);
 205         }
 206
 207         dev->features = features;
 208         if (dev->features &
 209                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 210                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 211         } else {
 212                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 213         }
 214         VHOST_LOG_DEBUG(VHOST_CONFIG,
 215                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 216                 dev->vid,
 217                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 218                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 219
 220         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 221             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 222                 /*
 223                  * Remove all but first queue pair if MQ hasn't been
 224                  * negotiated. This is safe because the device is not
 225                  * running at this stage.
 226                  */
 227                 while (dev->nr_vring > 2) {
 228                         struct vhost_virtqueue *vq;
 229
 230                         vq = dev->virtqueue[--dev->nr_vring];
 231                         if (!vq)
 232                                 continue;
 233
 234                         dev->virtqueue[dev->nr_vring] = NULL;
 235                         cleanup_vq(vq, 1);
 236                         free_vq(vq);
 237                 }
 238         }
 239
 240         did = dev->vdpa_dev_id;
 241         vdpa_dev = rte_vdpa_get_device(did);
 242         if (vdpa_dev && vdpa_dev->ops->set_features)
 243                 vdpa_dev->ops->set_features(dev->vid);
 244
 245         return 0;
 246 }
 247
 248 /*
 249  * The virtio device sends us the size of the descriptor ring.
 250  */
 251 static int
 252 vhost_user_set_vring_num(struct virtio_net *dev,
 253                          VhostUserMsg *msg)
 254 {
 255         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 256
 257         vq->size = msg->payload.state.num;
 258
 259         /* VIRTIO 1.0, 2.4 Virtqueues says:
 260          *
 261          *   Queue Size value is always a power of 2. The maximum Queue Size
 262          *   value is 32768.
 263          */
 264         if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
 265                 RTE_LOG(ERR, VHOST_CONFIG,
 266                         "invalid virtqueue size %u\n", vq->size);
 267                 return -1;
 268         }
 269
 270         if (dev->dequeue_zero_copy) {
 271                 vq->nr_zmbuf = 0;
 272                 vq->last_zmbuf_idx = 0;
 273                 vq->zmbuf_size = vq->size;
 274                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 275                                          sizeof(struct zcopy_mbuf), 0);
 276                 if (vq->zmbufs == NULL) {
 277                         RTE_LOG(WARNING, VHOST_CONFIG,
 278                                 "failed to allocate mem for zero copy; "
 279                                 "zero copy is force disabled\n");
 280                         dev->dequeue_zero_copy = 0;
 281                 }
 282                 TAILQ_INIT(&vq->zmbuf_list);
 283         }
 284
 285         vq->shadow_used_ring = rte_malloc(NULL,
 286                                 vq->size * sizeof(struct vring_used_elem),
 287                                 RTE_CACHE_LINE_SIZE);
 288         if (!vq->shadow_used_ring) {
 289                 RTE_LOG(ERR, VHOST_CONFIG,
 290                         "failed to allocate memory for shadow used ring.\n");
 291                 return -1;
 292         }
 293
 294         vq->batch_copy_elems = rte_malloc(NULL,
 295                                 vq->size * sizeof(struct batch_copy_elem),
 296                                 RTE_CACHE_LINE_SIZE);
 297         if (!vq->batch_copy_elems) {
 298                 RTE_LOG(ERR, VHOST_CONFIG,
 299                         "failed to allocate memory for batching copy.\n");
 300                 return -1;
 301         }
 302
 303         return 0;
 304 }
 305
 306 /*
 307  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 308  * same numa node as the memory of vring descriptor.
 309  */
 310 #ifdef RTE_LIBRTE_VHOST_NUMA
 311 static struct virtio_net*
 312 numa_realloc(struct virtio_net *dev, int index)
 313 {
 314         int oldnode, newnode;
 315         struct virtio_net *old_dev;
 316         struct vhost_virtqueue *old_vq, *vq;
 317         struct zcopy_mbuf *new_zmbuf;
 318         struct vring_used_elem *new_shadow_used_ring;
 319         struct batch_copy_elem *new_batch_copy_elems;
 320         int ret;
 321
 322         old_dev = dev;
 323         vq = old_vq = dev->virtqueue[index];
 324
 325         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 326                             MPOL_F_NODE | MPOL_F_ADDR);
 327
 328         /* check if we need to reallocate vq */
 329         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 330                              MPOL_F_NODE | MPOL_F_ADDR);
 331         if (ret) {
 332                 RTE_LOG(ERR, VHOST_CONFIG,
 333                         "Unable to get vq numa information.\n");
 334                 return dev;
 335         }
 336         if (oldnode != newnode) {
 337                 RTE_LOG(INFO, VHOST_CONFIG,
 338                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 339                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 340                 if (!vq)
 341                         return dev;
 342
 343                 memcpy(vq, old_vq, sizeof(*vq));
 344                 TAILQ_INIT(&vq->zmbuf_list);
 345
 346                 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 347                         sizeof(struct zcopy_mbuf), 0, newnode);
 348                 if (new_zmbuf) {
 349                         rte_free(vq->zmbufs);
 350                         vq->zmbufs = new_zmbuf;
 351                 }
 352
 353                 new_shadow_used_ring = rte_malloc_socket(NULL,
 354                         vq->size * sizeof(struct vring_used_elem),
 355                         RTE_CACHE_LINE_SIZE,
 356                         newnode);
 357                 if (new_shadow_used_ring) {
 358                         rte_free(vq->shadow_used_ring);
 359                         vq->shadow_used_ring = new_shadow_used_ring;
 360                 }
 361
 362                 new_batch_copy_elems = rte_malloc_socket(NULL,
 363                         vq->size * sizeof(struct batch_copy_elem),
 364                         RTE_CACHE_LINE_SIZE,
 365                         newnode);
 366                 if (new_batch_copy_elems) {
 367                         rte_free(vq->batch_copy_elems);
 368                         vq->batch_copy_elems = new_batch_copy_elems;
 369                 }
 370
 371                 rte_free(old_vq);
 372         }
 373
 374         /* check if we need to reallocate dev */
 375         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 376                             MPOL_F_NODE | MPOL_F_ADDR);
 377         if (ret) {
 378                 RTE_LOG(ERR, VHOST_CONFIG,
 379                         "Unable to get dev numa information.\n");
 380                 goto out;
 381         }
 382         if (oldnode != newnode) {
 383                 RTE_LOG(INFO, VHOST_CONFIG,
 384                         "reallocate dev from %d to %d node\n",
 385                         oldnode, newnode);
 386                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 387                 if (!dev) {
 388                         dev = old_dev;
 389                         goto out;
 390                 }
 391
 392                 memcpy(dev, old_dev, sizeof(*dev));
 393                 rte_free(old_dev);
 394         }
 395
 396 out:
 397         dev->virtqueue[index] = vq;
 398         vhost_devices[dev->vid] = dev;
 399
 400         if (old_vq != vq)
 401                 vhost_user_iotlb_init(dev, index);
 402
 403         return dev;
 404 }
 405 #else
 406 static struct virtio_net*
 407 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 408 {
 409         return dev;
 410 }
 411 #endif
 412
 413 /* Converts QEMU virtual address to Vhost virtual address. */
 414 static uint64_t
 415 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 416 {
 417         struct rte_vhost_mem_region *r;
 418         uint32_t i;
 419
 420         /* Find the region where the address lives. */
 421         for (i = 0; i < dev->mem->nregions; i++) {
 422                 r = &dev->mem->regions[i];
 423
 424                 if (qva >= r->guest_user_addr &&
 425                     qva <  r->guest_user_addr + r->size) {
 426
 427                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 428                                 *len = r->guest_user_addr + r->size - qva;
 429
 430                         return qva - r->guest_user_addr +
 431                                r->host_user_addr;
 432                 }
 433         }
 434         *len = 0;
 435
 436         return 0;
 437 }
 438
 439
 440 /*
 441  * Converts ring address to Vhost virtual address.
 442  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 443  * else it is a QEMU virtual address.
 444  */
 445 static uint64_t
 446 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 447                 uint64_t ra, uint64_t *size)
 448 {
 449         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 450                 uint64_t vva;
 451
 452                 vva = vhost_user_iotlb_cache_find(vq, ra,
 453                                         size, VHOST_ACCESS_RW);
 454                 if (!vva)
 455                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 456
 457                 return vva;
 458         }
 459
 460         return qva_to_vva(dev, ra, size);
 461 }
 462
 463 static struct virtio_net *
 464 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 465 {
 466         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 467         struct vhost_vring_addr *addr = &vq->ring_addrs;
 468         uint64_t len;
 469
 470         if (vq_is_packed(dev)) {
 471                 len = sizeof(struct vring_packed_desc) * vq->size;
 472                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 473                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 474                 vq->log_guest_addr = 0;
 475                 if (vq->desc_packed == NULL ||
 476                                 len != sizeof(struct vring_packed_desc) *
 477                                 vq->size) {
 478                         RTE_LOG(DEBUG, VHOST_CONFIG,
 479                                 "(%d) failed to map desc_packed ring.\n",
 480                                 dev->vid);
 481                         return dev;
 482                 }
 483
 484                 dev = numa_realloc(dev, vq_index);
 485                 vq = dev->virtqueue[vq_index];
 486                 addr = &vq->ring_addrs;
 487
 488                 return dev;
 489         }
 490
 491         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 492         if (vq->desc && vq->avail && vq->used)
 493                 return dev;
 494
 495         len = sizeof(struct vring_desc) * vq->size;
 496         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 497                         vq, addr->desc_user_addr, &len);
 498         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 499                 RTE_LOG(DEBUG, VHOST_CONFIG,
 500                         "(%d) failed to map desc ring.\n",
 501                         dev->vid);
 502                 return dev;
 503         }
 504
 505         dev = numa_realloc(dev, vq_index);
 506         vq = dev->virtqueue[vq_index];
 507         addr = &vq->ring_addrs;
 508
 509         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 510         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 511                         vq, addr->avail_user_addr, &len);
 512         if (vq->avail == 0 ||
 513                         len != sizeof(struct vring_avail) +
 514                         sizeof(uint16_t) * vq->size) {
 515                 RTE_LOG(DEBUG, VHOST_CONFIG,
 516                         "(%d) failed to map avail ring.\n",
 517                         dev->vid);
 518                 return dev;
 519         }
 520
 521         len = sizeof(struct vring_used) +
 522                 sizeof(struct vring_used_elem) * vq->size;
 523         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 524                         vq, addr->used_user_addr, &len);
 525         if (vq->used == 0 || len != sizeof(struct vring_used) +
 526                         sizeof(struct vring_used_elem) * vq->size) {
 527                 RTE_LOG(DEBUG, VHOST_CONFIG,
 528                         "(%d) failed to map used ring.\n",
 529                         dev->vid);
 530                 return dev;
 531         }
 532
 533         if (vq->last_used_idx != vq->used->idx) {
 534                 RTE_LOG(WARNING, VHOST_CONFIG,
 535                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 536                         "some packets maybe resent for Tx and dropped for Rx\n",
 537                         vq->last_used_idx, vq->used->idx);
 538                 vq->last_used_idx  = vq->used->idx;
 539                 vq->last_avail_idx = vq->used->idx;
 540         }
 541
 542         vq->log_guest_addr = addr->log_guest_addr;
 543
 544         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 545                         dev->vid, vq->desc);
 546         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 547                         dev->vid, vq->avail);
 548         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 549                         dev->vid, vq->used);
 550         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 551                         dev->vid, vq->log_guest_addr);
 552
 553         return dev;
 554 }
 555
 556 /*
 557  * The virtio device sends us the desc, used and avail ring addresses.
 558  * This function then converts these to our address space.
 559  */
 560 static int
 561 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 562 {
 563         struct vhost_virtqueue *vq;
 564         struct vhost_vring_addr *addr = &msg->payload.addr;
 565         struct virtio_net *dev = *pdev;
 566
 567         if (dev->mem == NULL)
 568                 return -1;
 569
 570         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 571         vq = dev->virtqueue[msg->payload.addr.index];
 572
 573         /*
 574          * Rings addresses should not be interpreted as long as the ring is not
 575          * started and enabled
 576          */
 577         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 578
 579         vring_invalidate(dev, vq);
 580
 581         if (vq->enabled && (dev->features &
 582                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 583                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 584                 if (!dev)
 585                         return -1;
 586
 587                 *pdev = dev;
 588         }
 589
 590         return 0;
 591 }
 592
 593 /*
 594  * The virtio device sends us the available ring last used index.
 595  */
 596 static int
 597 vhost_user_set_vring_base(struct virtio_net *dev,
 598                           VhostUserMsg *msg)
 599 {
 600         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 601                         msg->payload.state.num;
 602         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 603                         msg->payload.state.num;
 604
 605         return 0;
 606 }
 607
 608 static int
 609 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 610                    uint64_t host_phys_addr, uint64_t size)
 611 {
 612         struct guest_page *page, *last_page;
 613
 614         if (dev->nr_guest_pages == dev->max_guest_pages) {
 615                 dev->max_guest_pages *= 2;
 616                 dev->guest_pages = realloc(dev->guest_pages,
 617                                         dev->max_guest_pages * sizeof(*page));
 618                 if (!dev->guest_pages) {
 619                         RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
 620                         return -1;
 621                 }
 622         }
 623
 624         if (dev->nr_guest_pages > 0) {
 625                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 626                 /* merge if the two pages are continuous */
 627                 if (host_phys_addr == last_page->host_phys_addr +
 628                                       last_page->size) {
 629                         last_page->size += size;
 630                         return 0;
 631                 }
 632         }
 633
 634         page = &dev->guest_pages[dev->nr_guest_pages++];
 635         page->guest_phys_addr = guest_phys_addr;
 636         page->host_phys_addr  = host_phys_addr;
 637         page->size = size;
 638
 639         return 0;
 640 }
 641
 642 static int
 643 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 644                 uint64_t page_size)
 645 {
 646         uint64_t reg_size = reg->size;
 647         uint64_t host_user_addr  = reg->host_user_addr;
 648         uint64_t guest_phys_addr = reg->guest_phys_addr;
 649         uint64_t host_phys_addr;
 650         uint64_t size;
 651
 652         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 653         size = page_size - (guest_phys_addr & (page_size - 1));
 654         size = RTE_MIN(size, reg_size);
 655
 656         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 657                 return -1;
 658
 659         host_user_addr  += size;
 660         guest_phys_addr += size;
 661         reg_size -= size;
 662
 663         while (reg_size > 0) {
 664                 size = RTE_MIN(reg_size, page_size);
 665                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 666                                                   host_user_addr);
 667                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 668                                 size) < 0)
 669                         return -1;
 670
 671                 host_user_addr  += size;
 672                 guest_phys_addr += size;
 673                 reg_size -= size;
 674         }
 675
 676         return 0;
 677 }
 678
 679 #ifdef RTE_LIBRTE_VHOST_DEBUG
 680 /* TODO: enable it only in debug mode? */
 681 static void
 682 dump_guest_pages(struct virtio_net *dev)
 683 {
 684         uint32_t i;
 685         struct guest_page *page;
 686
 687         for (i = 0; i < dev->nr_guest_pages; i++) {
 688                 page = &dev->guest_pages[i];
 689
 690                 RTE_LOG(INFO, VHOST_CONFIG,
 691                         "guest physical page region %u\n"
 692                         "\t guest_phys_addr: %" PRIx64 "\n"
 693                         "\t host_phys_addr : %" PRIx64 "\n"
 694                         "\t size           : %" PRIx64 "\n",
 695                         i,
 696                         page->guest_phys_addr,
 697                         page->host_phys_addr,
 698                         page->size);
 699         }
 700 }
 701 #else
 702 #define dump_guest_pages(dev)
 703 #endif
 704
 705 static bool
 706 vhost_memory_changed(struct VhostUserMemory *new,
 707                      struct rte_vhost_memory *old)
 708 {
 709         uint32_t i;
 710
 711         if (new->nregions != old->nregions)
 712                 return true;
 713
 714         for (i = 0; i < new->nregions; ++i) {
 715                 VhostUserMemoryRegion *new_r = &new->regions[i];
 716                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 717
 718                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 719                         return true;
 720                 if (new_r->memory_size != old_r->size)
 721                         return true;
 722                 if (new_r->userspace_addr != old_r->guest_user_addr)
 723                         return true;
 724         }
 725
 726         return false;
 727 }
 728
 729 static int
 730 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 731 {
 732         struct virtio_net *dev = *pdev;
 733         struct VhostUserMemory memory = pmsg->payload.memory;
 734         struct rte_vhost_mem_region *reg;
 735         void *mmap_addr;
 736         uint64_t mmap_size;
 737         uint64_t mmap_offset;
 738         uint64_t alignment;
 739         uint32_t i;
 740         int populate;
 741         int fd;
 742
 743         if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
 744                 RTE_LOG(ERR, VHOST_CONFIG,
 745                         "too many memory regions (%u)\n", memory.nregions);
 746                 return -1;
 747         }
 748
 749         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 750                 RTE_LOG(INFO, VHOST_CONFIG,
 751                         "(%d) memory regions not changed\n", dev->vid);
 752
 753                 for (i = 0; i < memory.nregions; i++)
 754                         close(pmsg->fds[i]);
 755
 756                 return 0;
 757         }
 758
 759         if (dev->mem) {
 760                 free_mem_region(dev);
 761                 rte_free(dev->mem);
 762                 dev->mem = NULL;
 763         }
 764
 765         dev->nr_guest_pages = 0;
 766         if (!dev->guest_pages) {
 767                 dev->max_guest_pages = 8;
 768                 dev->guest_pages = malloc(dev->max_guest_pages *
 769                                                 sizeof(struct guest_page));
 770                 if (dev->guest_pages == NULL) {
 771                         RTE_LOG(ERR, VHOST_CONFIG,
 772                                 "(%d) failed to allocate memory "
 773                                 "for dev->guest_pages\n",
 774                                 dev->vid);
 775                         return -1;
 776                 }
 777         }
 778
 779         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 780                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 781         if (dev->mem == NULL) {
 782                 RTE_LOG(ERR, VHOST_CONFIG,
 783                         "(%d) failed to allocate memory for dev->mem\n",
 784                         dev->vid);
 785                 return -1;
 786         }
 787         dev->mem->nregions = memory.nregions;
 788
 789         for (i = 0; i < memory.nregions; i++) {
 790                 fd  = pmsg->fds[i];
 791                 reg = &dev->mem->regions[i];
 792
 793                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 794                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 795                 reg->size            = memory.regions[i].memory_size;
 796                 reg->fd              = fd;
 797
 798                 mmap_offset = memory.regions[i].mmap_offset;
 799
 800                 /* Check for memory_size + mmap_offset overflow */
 801                 if (mmap_offset >= -reg->size) {
 802                         RTE_LOG(ERR, VHOST_CONFIG,
 803                                 "mmap_offset (%#"PRIx64") and memory_size "
 804                                 "(%#"PRIx64") overflow\n",
 805                                 mmap_offset, reg->size);
 806                         goto err_mmap;
 807                 }
 808
 809                 mmap_size = reg->size + mmap_offset;
 810
 811                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 812                  * with length argument aligned with hugepagesz at older
 813                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 814                  * mmap() will fail with EINVAL.
 815                  *
 816                  * to avoid failure, make sure in caller to keep length
 817                  * aligned.
 818                  */
 819                 alignment = get_blk_size(fd);
 820                 if (alignment == (uint64_t)-1) {
 821                         RTE_LOG(ERR, VHOST_CONFIG,
 822                                 "couldn't get hugepage size through fstat\n");
 823                         goto err_mmap;
 824                 }
 825                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 826
 827                 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
 828                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 829                                  MAP_SHARED | populate, fd, 0);
 830
 831                 if (mmap_addr == MAP_FAILED) {
 832                         RTE_LOG(ERR, VHOST_CONFIG,
 833                                 "mmap region %u failed.\n", i);
 834                         goto err_mmap;
 835                 }
 836
 837                 reg->mmap_addr = mmap_addr;
 838                 reg->mmap_size = mmap_size;
 839                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 840                                       mmap_offset;
 841
 842                 if (dev->dequeue_zero_copy)
 843                         if (add_guest_pages(dev, reg, alignment) < 0) {
 844                                 RTE_LOG(ERR, VHOST_CONFIG,
 845                                         "adding guest pages to region %u failed.\n",
 846                                         i);
 847                                 goto err_mmap;
 848                         }
 849
 850                 RTE_LOG(INFO, VHOST_CONFIG,
 851                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 852                         "\t guest physical addr: 0x%" PRIx64 "\n"
 853                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 854                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 855                         "\t mmap addr : 0x%" PRIx64 "\n"
 856                         "\t mmap size : 0x%" PRIx64 "\n"
 857                         "\t mmap align: 0x%" PRIx64 "\n"
 858                         "\t mmap off  : 0x%" PRIx64 "\n",
 859                         i, reg->size,
 860                         reg->guest_phys_addr,
 861                         reg->guest_user_addr,
 862                         reg->host_user_addr,
 863                         (uint64_t)(uintptr_t)mmap_addr,
 864                         mmap_size,
 865                         alignment,
 866                         mmap_offset);
 867         }
 868
 869         for (i = 0; i < dev->nr_vring; i++) {
 870                 struct vhost_virtqueue *vq = dev->virtqueue[i];
 871
 872                 if (vq->desc || vq->avail || vq->used) {
 873                         /*
 874                          * If the memory table got updated, the ring addresses
 875                          * need to be translated again as virtual addresses have
 876                          * changed.
 877                          */
 878                         vring_invalidate(dev, vq);
 879
 880                         dev = translate_ring_addresses(dev, i);
 881                         if (!dev)
 882                                 return -1;
 883
 884                         *pdev = dev;
 885                 }
 886         }
 887
 888         dump_guest_pages(dev);
 889
 890         return 0;
 891
 892 err_mmap:
 893         free_mem_region(dev);
 894         rte_free(dev->mem);
 895         dev->mem = NULL;
 896         return -1;
 897 }
 898
 899 static bool
 900 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
 901 {
 902         bool rings_ok;
 903
 904         if (!vq)
 905                 return false;
 906
 907         if (vq_is_packed(dev))
 908                 rings_ok = !!vq->desc_packed;
 909         else
 910                 rings_ok = vq->desc && vq->avail && vq->used;
 911
 912         return rings_ok &&
 913                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 914                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 915 }
 916
 917 static int
 918 virtio_is_ready(struct virtio_net *dev)
 919 {
 920         struct vhost_virtqueue *vq;
 921         uint32_t i;
 922
 923         if (dev->nr_vring == 0)
 924                 return 0;
 925
 926         for (i = 0; i < dev->nr_vring; i++) {
 927                 vq = dev->virtqueue[i];
 928
 929                 if (!vq_is_ready(dev, vq))
 930                         return 0;
 931         }
 932
 933         RTE_LOG(INFO, VHOST_CONFIG,
 934                 "virtio is now ready for processing.\n");
 935         return 1;
 936 }
 937
 938 static void
 939 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 940 {
 941         struct vhost_vring_file file;
 942         struct vhost_virtqueue *vq;
 943
 944         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 945         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 946                 file.fd = VIRTIO_INVALID_EVENTFD;
 947         else
 948                 file.fd = pmsg->fds[0];
 949         RTE_LOG(INFO, VHOST_CONFIG,
 950                 "vring call idx:%d file:%d\n", file.index, file.fd);
 951
 952         vq = dev->virtqueue[file.index];
 953         if (vq->callfd >= 0)
 954                 close(vq->callfd);
 955
 956         vq->callfd = file.fd;
 957 }
 958
 959 static void
 960 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 961 {
 962         struct vhost_vring_file file;
 963         struct vhost_virtqueue *vq;
 964         struct virtio_net *dev = *pdev;
 965
 966         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 967         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 968                 file.fd = VIRTIO_INVALID_EVENTFD;
 969         else
 970                 file.fd = pmsg->fds[0];
 971         RTE_LOG(INFO, VHOST_CONFIG,
 972                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 973
 974         /* Interpret ring addresses only when ring is started. */
 975         dev = translate_ring_addresses(dev, file.index);
 976         if (!dev)
 977                 return;
 978
 979         *pdev = dev;
 980
 981         vq = dev->virtqueue[file.index];
 982
 983         /*
 984          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 985          * the ring starts already enabled. Otherwise, it is enabled via
 986          * the SET_VRING_ENABLE message.
 987          */
 988         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 989                 vq->enabled = 1;
 990
 991         if (vq->kickfd >= 0)
 992                 close(vq->kickfd);
 993         vq->kickfd = file.fd;
 994 }
 995
 996 static void
 997 free_zmbufs(struct vhost_virtqueue *vq)
 998 {
 999         struct zcopy_mbuf *zmbuf, *next;
1000
1001         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1002              zmbuf != NULL; zmbuf = next) {
1003                 next = TAILQ_NEXT(zmbuf, next);
1004
1005                 rte_pktmbuf_free(zmbuf->mbuf);
1006                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1007         }
1008
1009         rte_free(vq->zmbufs);
1010 }
1011
1012 /*
1013  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1014  */
1015 static int
1016 vhost_user_get_vring_base(struct virtio_net *dev,
1017                           VhostUserMsg *msg)
1018 {
1019         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1020
1021         /* We have to stop the queue (virtio) if it is running. */
1022         vhost_destroy_device_notify(dev);
1023
1024         dev->flags &= ~VIRTIO_DEV_READY;
1025         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1026
1027         /* Here we are safe to get the last avail index */
1028         msg->payload.state.num = vq->last_avail_idx;
1029
1030         RTE_LOG(INFO, VHOST_CONFIG,
1031                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1032                 msg->payload.state.num);
1033         /*
1034          * Based on current qemu vhost-user implementation, this message is
1035          * sent and only sent in vhost_vring_stop.
1036          * TODO: cleanup the vring, it isn't usable since here.
1037          */
1038         if (vq->kickfd >= 0)
1039                 close(vq->kickfd);
1040
1041         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1042
1043         if (vq->callfd >= 0)
1044                 close(vq->callfd);
1045
1046         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1047
1048         if (dev->dequeue_zero_copy)
1049                 free_zmbufs(vq);
1050         rte_free(vq->shadow_used_ring);
1051         vq->shadow_used_ring = NULL;
1052
1053         rte_free(vq->batch_copy_elems);
1054         vq->batch_copy_elems = NULL;
1055
1056         return 0;
1057 }
1058
1059 /*
1060  * when virtio queues are ready to work, qemu will send us to
1061  * enable the virtio queue pair.
1062  */
1063 static int
1064 vhost_user_set_vring_enable(struct virtio_net *dev,
1065                             VhostUserMsg *msg)
1066 {
1067         int enable = (int)msg->payload.state.num;
1068         int index = (int)msg->payload.state.index;
1069         struct rte_vdpa_device *vdpa_dev;
1070         int did = -1;
1071
1072         RTE_LOG(INFO, VHOST_CONFIG,
1073                 "set queue enable: %d to qp idx: %d\n",
1074                 enable, index);
1075
1076         did = dev->vdpa_dev_id;
1077         vdpa_dev = rte_vdpa_get_device(did);
1078         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1079                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1080
1081         if (dev->notify_ops->vring_state_changed)
1082                 dev->notify_ops->vring_state_changed(dev->vid,
1083                                 index, enable);
1084
1085         dev->virtqueue[index]->enabled = enable;
1086
1087         return 0;
1088 }
1089
1090 static void
1091 vhost_user_get_protocol_features(struct virtio_net *dev,
1092                                  struct VhostUserMsg *msg)
1093 {
1094         uint64_t features, protocol_features;
1095
1096         rte_vhost_driver_get_features(dev->ifname, &features);
1097         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1098
1099         /*
1100          * REPLY_ACK protocol feature is only mandatory for now
1101          * for IOMMU feature. If IOMMU is explicitly disabled by the
1102          * application, disable also REPLY_ACK feature for older buggy
1103          * Qemu versions (from v2.7.0 to v2.9.0).
1104          */
1105         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1106                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1107
1108         msg->payload.u64 = protocol_features;
1109         msg->size = sizeof(msg->payload.u64);
1110 }
1111
1112 static void
1113 vhost_user_set_protocol_features(struct virtio_net *dev,
1114                                  uint64_t protocol_features)
1115 {
1116         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1117                 return;
1118
1119         dev->protocol_features = protocol_features;
1120 }
1121
1122 static int
1123 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1124 {
1125         int fd = msg->fds[0];
1126         uint64_t size, off;
1127         void *addr;
1128
1129         if (fd < 0) {
1130                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1131                 return -1;
1132         }
1133
1134         if (msg->size != sizeof(VhostUserLog)) {
1135                 RTE_LOG(ERR, VHOST_CONFIG,
1136                         "invalid log base msg size: %"PRId32" != %d\n",
1137                         msg->size, (int)sizeof(VhostUserLog));
1138                 return -1;
1139         }
1140
1141         size = msg->payload.log.mmap_size;
1142         off  = msg->payload.log.mmap_offset;
1143
1144         /* Don't allow mmap_offset to point outside the mmap region */
1145         if (off > size) {
1146                 RTE_LOG(ERR, VHOST_CONFIG,
1147                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1148                         off, size);
1149                 return -1;
1150         }
1151
1152         RTE_LOG(INFO, VHOST_CONFIG,
1153                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1154                 size, off);
1155
1156         /*
1157          * mmap from 0 to workaround a hugepage mmap bug: mmap will
1158          * fail when offset is not page size aligned.
1159          */
1160         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1161         close(fd);
1162         if (addr == MAP_FAILED) {
1163                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1164                 return -1;
1165         }
1166
1167         /*
1168          * Free previously mapped log memory on occasionally
1169          * multiple VHOST_USER_SET_LOG_BASE.
1170          */
1171         if (dev->log_addr) {
1172                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1173         }
1174         dev->log_addr = (uint64_t)(uintptr_t)addr;
1175         dev->log_base = dev->log_addr + off;
1176         dev->log_size = size;
1177
1178         return 0;
1179 }
1180
1181 /*
1182  * An rarp packet is constructed and broadcasted to notify switches about
1183  * the new location of the migrated VM, so that packets from outside will
1184  * not be lost after migration.
1185  *
1186  * However, we don't actually "send" a rarp packet here, instead, we set
1187  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1188  */
1189 static int
1190 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1191 {
1192         uint8_t *mac = (uint8_t *)&msg->payload.u64;
1193         struct rte_vdpa_device *vdpa_dev;
1194         int did = -1;
1195
1196         RTE_LOG(DEBUG, VHOST_CONFIG,
1197                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1198                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1199         memcpy(dev->mac.addr_bytes, mac, 6);
1200
1201         /*
1202          * Set the flag to inject a RARP broadcast packet at
1203          * rte_vhost_dequeue_burst().
1204          *
1205          * rte_smp_wmb() is for making sure the mac is copied
1206          * before the flag is set.
1207          */
1208         rte_smp_wmb();
1209         rte_atomic16_set(&dev->broadcast_rarp, 1);
1210         did = dev->vdpa_dev_id;
1211         vdpa_dev = rte_vdpa_get_device(did);
1212         if (vdpa_dev && vdpa_dev->ops->migration_done)
1213                 vdpa_dev->ops->migration_done(dev->vid);
1214
1215         return 0;
1216 }
1217
1218 static int
1219 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1220 {
1221         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1222                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1223                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1224                                 msg->payload.u64);
1225
1226                 return -1;
1227         }
1228
1229         dev->mtu = msg->payload.u64;
1230
1231         return 0;
1232 }
1233
1234 static int
1235 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1236 {
1237         int fd = msg->fds[0];
1238
1239         if (fd < 0) {
1240                 RTE_LOG(ERR, VHOST_CONFIG,
1241                                 "Invalid file descriptor for slave channel (%d)\n",
1242                                 fd);
1243                 return -1;
1244         }
1245
1246         dev->slave_req_fd = fd;
1247
1248         return 0;
1249 }
1250
1251 static int
1252 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1253 {
1254         struct vhost_vring_addr *ra;
1255         uint64_t start, end;
1256
1257         start = imsg->iova;
1258         end = start + imsg->size;
1259
1260         ra = &vq->ring_addrs;
1261         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1262                 return 1;
1263         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1264                 return 1;
1265         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1266                 return 1;
1267
1268         return 0;
1269 }
1270
1271 static int
1272 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1273                                 struct vhost_iotlb_msg *imsg)
1274 {
1275         uint64_t istart, iend, vstart, vend;
1276
1277         istart = imsg->iova;
1278         iend = istart + imsg->size - 1;
1279
1280         vstart = (uintptr_t)vq->desc;
1281         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1282         if (vstart <= iend && istart <= vend)
1283                 return 1;
1284
1285         vstart = (uintptr_t)vq->avail;
1286         vend = vstart + sizeof(struct vring_avail);
1287         vend += sizeof(uint16_t) * vq->size - 1;
1288         if (vstart <= iend && istart <= vend)
1289                 return 1;
1290
1291         vstart = (uintptr_t)vq->used;
1292         vend = vstart + sizeof(struct vring_used);
1293         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1294         if (vstart <= iend && istart <= vend)
1295                 return 1;
1296
1297         return 0;
1298 }
1299
1300 static int
1301 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1302 {
1303         struct virtio_net *dev = *pdev;
1304         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1305         uint16_t i;
1306         uint64_t vva, len;
1307
1308         switch (imsg->type) {
1309         case VHOST_IOTLB_UPDATE:
1310                 len = imsg->size;
1311                 vva = qva_to_vva(dev, imsg->uaddr, &len);
1312                 if (!vva)
1313                         return -1;
1314
1315                 for (i = 0; i < dev->nr_vring; i++) {
1316                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1317
1318                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1319                                         len, imsg->perm);
1320
1321                         if (is_vring_iotlb_update(vq, imsg))
1322                                 *pdev = dev = translate_ring_addresses(dev, i);
1323                 }
1324                 break;
1325         case VHOST_IOTLB_INVALIDATE:
1326                 for (i = 0; i < dev->nr_vring; i++) {
1327                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1328
1329                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1330                                         imsg->size);
1331
1332                         if (is_vring_iotlb_invalidate(vq, imsg))
1333                                 vring_invalidate(dev, vq);
1334                 }
1335                 break;
1336         default:
1337                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1338                                 imsg->type);
1339                 return -1;
1340         }
1341
1342         return 0;
1343 }
1344
1345 /* return bytes# of read on success or negative val on failure. */
1346 static int
1347 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1348 {
1349         int ret;
1350
1351         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1352                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1353         if (ret <= 0)
1354                 return ret;
1355
1356         if (msg && msg->size) {
1357                 if (msg->size > sizeof(msg->payload)) {
1358                         RTE_LOG(ERR, VHOST_CONFIG,
1359                                 "invalid msg size: %d\n", msg->size);
1360                         return -1;
1361                 }
1362                 ret = read(sockfd, &msg->payload, msg->size);
1363                 if (ret <= 0)
1364                         return ret;
1365                 if (ret != (int)msg->size) {
1366                         RTE_LOG(ERR, VHOST_CONFIG,
1367                                 "read control message failed\n");
1368                         return -1;
1369                 }
1370         }
1371
1372         return ret;
1373 }
1374
1375 static int
1376 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1377 {
1378         if (!msg)
1379                 return 0;
1380
1381         return send_fd_message(sockfd, (char *)msg,
1382                 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1383 }
1384
1385 static int
1386 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1387 {
1388         if (!msg)
1389                 return 0;
1390
1391         msg->flags &= ~VHOST_USER_VERSION_MASK;
1392         msg->flags &= ~VHOST_USER_NEED_REPLY;
1393         msg->flags |= VHOST_USER_VERSION;
1394         msg->flags |= VHOST_USER_REPLY_MASK;
1395
1396         return send_vhost_message(sockfd, msg, NULL, 0);
1397 }
1398
1399 static int
1400 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1401                          int *fds, int fd_num)
1402 {
1403         int ret;
1404
1405         if (msg->flags & VHOST_USER_NEED_REPLY)
1406                 rte_spinlock_lock(&dev->slave_req_lock);
1407
1408         ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1409         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1410                 rte_spinlock_unlock(&dev->slave_req_lock);
1411
1412         return ret;
1413 }
1414
1415 /*
1416  * Allocate a queue pair if it hasn't been allocated yet
1417  */
1418 static int
1419 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1420 {
1421         uint16_t vring_idx;
1422
1423         switch (msg->request.master) {
1424         case VHOST_USER_SET_VRING_KICK:
1425         case VHOST_USER_SET_VRING_CALL:
1426         case VHOST_USER_SET_VRING_ERR:
1427                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1428                 break;
1429         case VHOST_USER_SET_VRING_NUM:
1430         case VHOST_USER_SET_VRING_BASE:
1431         case VHOST_USER_SET_VRING_ENABLE:
1432                 vring_idx = msg->payload.state.index;
1433                 break;
1434         case VHOST_USER_SET_VRING_ADDR:
1435                 vring_idx = msg->payload.addr.index;
1436                 break;
1437         default:
1438                 return 0;
1439         }
1440
1441         if (vring_idx >= VHOST_MAX_VRING) {
1442                 RTE_LOG(ERR, VHOST_CONFIG,
1443                         "invalid vring index: %u\n", vring_idx);
1444                 return -1;
1445         }
1446
1447         if (dev->virtqueue[vring_idx])
1448                 return 0;
1449
1450         return alloc_vring_queue(dev, vring_idx);
1451 }
1452
1453 static void
1454 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1455 {
1456         unsigned int i = 0;
1457         unsigned int vq_num = 0;
1458
1459         while (vq_num < dev->nr_vring) {
1460                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1461
1462                 if (vq) {
1463                         rte_spinlock_lock(&vq->access_lock);
1464                         vq_num++;
1465                 }
1466                 i++;
1467         }
1468 }
1469
1470 static void
1471 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1472 {
1473         unsigned int i = 0;
1474         unsigned int vq_num = 0;
1475
1476         while (vq_num < dev->nr_vring) {
1477                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1478
1479                 if (vq) {
1480                         rte_spinlock_unlock(&vq->access_lock);
1481                         vq_num++;
1482                 }
1483                 i++;
1484         }
1485 }
1486
1487 int
1488 vhost_user_msg_handler(int vid, int fd)
1489 {
1490         struct virtio_net *dev;
1491         struct VhostUserMsg msg;
1492         struct rte_vdpa_device *vdpa_dev;
1493         int did = -1;
1494         int ret;
1495         int unlock_required = 0;
1496         uint32_t skip_master = 0;
1497
1498         dev = get_device(vid);
1499         if (dev == NULL)
1500                 return -1;
1501
1502         if (!dev->notify_ops) {
1503                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1504                 if (!dev->notify_ops) {
1505                         RTE_LOG(ERR, VHOST_CONFIG,
1506                                 "failed to get callback ops for driver %s\n",
1507                                 dev->ifname);
1508                         return -1;
1509                 }
1510         }
1511
1512         ret = read_vhost_message(fd, &msg);
1513         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1514                 if (ret < 0)
1515                         RTE_LOG(ERR, VHOST_CONFIG,
1516                                 "vhost read message failed\n");
1517                 else if (ret == 0)
1518                         RTE_LOG(INFO, VHOST_CONFIG,
1519                                 "vhost peer closed\n");
1520                 else
1521                         RTE_LOG(ERR, VHOST_CONFIG,
1522                                 "vhost read incorrect message\n");
1523
1524                 return -1;
1525         }
1526
1527         ret = 0;
1528         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1529                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1530                         vhost_message_str[msg.request.master]);
1531         else
1532                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1533                         vhost_message_str[msg.request.master]);
1534
1535         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1536         if (ret < 0) {
1537                 RTE_LOG(ERR, VHOST_CONFIG,
1538                         "failed to alloc queue\n");
1539                 return -1;
1540         }
1541
1542         /*
1543          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1544          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1545          * and device is destroyed. destroy_device waits for queues to be
1546          * inactive, so it is safe. Otherwise taking the access_lock
1547          * would cause a dead lock.
1548          */
1549         switch (msg.request.master) {
1550         case VHOST_USER_SET_FEATURES:
1551         case VHOST_USER_SET_PROTOCOL_FEATURES:
1552         case VHOST_USER_SET_OWNER:
1553         case VHOST_USER_SET_MEM_TABLE:
1554         case VHOST_USER_SET_LOG_BASE:
1555         case VHOST_USER_SET_LOG_FD:
1556         case VHOST_USER_SET_VRING_NUM:
1557         case VHOST_USER_SET_VRING_ADDR:
1558         case VHOST_USER_SET_VRING_BASE:
1559         case VHOST_USER_SET_VRING_KICK:
1560         case VHOST_USER_SET_VRING_CALL:
1561         case VHOST_USER_SET_VRING_ERR:
1562         case VHOST_USER_SET_VRING_ENABLE:
1563         case VHOST_USER_SEND_RARP:
1564         case VHOST_USER_NET_SET_MTU:
1565         case VHOST_USER_SET_SLAVE_REQ_FD:
1566                 vhost_user_lock_all_queue_pairs(dev);
1567                 unlock_required = 1;
1568                 break;
1569         default:
1570                 break;
1571
1572         }
1573
1574         if (dev->extern_ops.pre_msg_handle) {
1575                 uint32_t need_reply;
1576
1577                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1578                                 (void *)&msg, &need_reply, &skip_master);
1579                 if (ret < 0)
1580                         goto skip_to_reply;
1581
1582                 if (need_reply)
1583                         send_vhost_reply(fd, &msg);
1584
1585                 if (skip_master)
1586                         goto skip_to_post_handle;
1587         }
1588
1589         switch (msg.request.master) {
1590         case VHOST_USER_GET_FEATURES:
1591                 msg.payload.u64 = vhost_user_get_features(dev);
1592                 msg.size = sizeof(msg.payload.u64);
1593                 send_vhost_reply(fd, &msg);
1594                 break;
1595         case VHOST_USER_SET_FEATURES:
1596                 ret = vhost_user_set_features(dev, msg.payload.u64);
1597                 if (ret)
1598                         return -1;
1599                 break;
1600
1601         case VHOST_USER_GET_PROTOCOL_FEATURES:
1602                 vhost_user_get_protocol_features(dev, &msg);
1603                 send_vhost_reply(fd, &msg);
1604                 break;
1605         case VHOST_USER_SET_PROTOCOL_FEATURES:
1606                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1607                 break;
1608
1609         case VHOST_USER_SET_OWNER:
1610                 vhost_user_set_owner();
1611                 break;
1612         case VHOST_USER_RESET_OWNER:
1613                 vhost_user_reset_owner(dev);
1614                 break;
1615
1616         case VHOST_USER_SET_MEM_TABLE:
1617                 ret = vhost_user_set_mem_table(&dev, &msg);
1618                 break;
1619
1620         case VHOST_USER_SET_LOG_BASE:
1621                 vhost_user_set_log_base(dev, &msg);
1622
1623                 /* it needs a reply */
1624                 msg.size = sizeof(msg.payload.u64);
1625                 send_vhost_reply(fd, &msg);
1626                 break;
1627         case VHOST_USER_SET_LOG_FD:
1628                 close(msg.fds[0]);
1629                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1630                 break;
1631
1632         case VHOST_USER_SET_VRING_NUM:
1633                 vhost_user_set_vring_num(dev, &msg);
1634                 break;
1635         case VHOST_USER_SET_VRING_ADDR:
1636                 vhost_user_set_vring_addr(&dev, &msg);
1637                 break;
1638         case VHOST_USER_SET_VRING_BASE:
1639                 vhost_user_set_vring_base(dev, &msg);
1640                 break;
1641
1642         case VHOST_USER_GET_VRING_BASE:
1643                 vhost_user_get_vring_base(dev, &msg);
1644                 msg.size = sizeof(msg.payload.state);
1645                 send_vhost_reply(fd, &msg);
1646                 break;
1647
1648         case VHOST_USER_SET_VRING_KICK:
1649                 vhost_user_set_vring_kick(&dev, &msg);
1650                 break;
1651         case VHOST_USER_SET_VRING_CALL:
1652                 vhost_user_set_vring_call(dev, &msg);
1653                 break;
1654
1655         case VHOST_USER_SET_VRING_ERR:
1656                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1657                         close(msg.fds[0]);
1658                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1659                 break;
1660
1661         case VHOST_USER_GET_QUEUE_NUM:
1662                 msg.payload.u64 = (uint64_t)vhost_user_get_queue_num(dev);
1663                 msg.size = sizeof(msg.payload.u64);
1664                 send_vhost_reply(fd, &msg);
1665                 break;
1666
1667         case VHOST_USER_SET_VRING_ENABLE:
1668                 vhost_user_set_vring_enable(dev, &msg);
1669                 break;
1670         case VHOST_USER_SEND_RARP:
1671                 vhost_user_send_rarp(dev, &msg);
1672                 break;
1673
1674         case VHOST_USER_NET_SET_MTU:
1675                 ret = vhost_user_net_set_mtu(dev, &msg);
1676                 break;
1677
1678         case VHOST_USER_SET_SLAVE_REQ_FD:
1679                 ret = vhost_user_set_req_fd(dev, &msg);
1680                 break;
1681
1682         case VHOST_USER_IOTLB_MSG:
1683                 ret = vhost_user_iotlb_msg(&dev, &msg);
1684                 break;
1685
1686         default:
1687                 ret = -1;
1688                 break;
1689         }
1690
1691 skip_to_post_handle:
1692         if (dev->extern_ops.post_msg_handle) {
1693                 uint32_t need_reply;
1694
1695                 ret = (*dev->extern_ops.post_msg_handle)(
1696                                 dev->vid, (void *)&msg, &need_reply);
1697                 if (ret < 0)
1698                         goto skip_to_reply;
1699
1700                 if (need_reply)
1701                         send_vhost_reply(fd, &msg);
1702         }
1703
1704 skip_to_reply:
1705         if (unlock_required)
1706                 vhost_user_unlock_all_queue_pairs(dev);
1707
1708         if (msg.flags & VHOST_USER_NEED_REPLY) {
1709                 msg.payload.u64 = !!ret;
1710                 msg.size = sizeof(msg.payload.u64);
1711                 send_vhost_reply(fd, &msg);
1712         }
1713
1714         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1715                 dev->flags |= VIRTIO_DEV_READY;
1716
1717                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1718                         if (dev->dequeue_zero_copy) {
1719                                 RTE_LOG(INFO, VHOST_CONFIG,
1720                                                 "dequeue zero copy is enabled\n");
1721                         }
1722
1723                         if (dev->notify_ops->new_device(dev->vid) == 0)
1724                                 dev->flags |= VIRTIO_DEV_RUNNING;
1725                 }
1726         }
1727
1728         did = dev->vdpa_dev_id;
1729         vdpa_dev = rte_vdpa_get_device(did);
1730         if (vdpa_dev && virtio_is_ready(dev) &&
1731                         !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1732                         msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1733                 if (vdpa_dev->ops->dev_conf)
1734                         vdpa_dev->ops->dev_conf(dev->vid);
1735                 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1736                 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1737                         RTE_LOG(INFO, VHOST_CONFIG,
1738                                 "(%d) software relay is used for vDPA, performance may be low.\n",
1739                                 dev->vid);
1740                 }
1741         }
1742
1743         return 0;
1744 }
1745
1746 static int process_slave_message_reply(struct virtio_net *dev,
1747                                        const VhostUserMsg *msg)
1748 {
1749         VhostUserMsg msg_reply;
1750         int ret;
1751
1752         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1753                 return 0;
1754
1755         if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1756                 ret = -1;
1757                 goto out;
1758         }
1759
1760         if (msg_reply.request.slave != msg->request.slave) {
1761                 RTE_LOG(ERR, VHOST_CONFIG,
1762                         "Received unexpected msg type (%u), expected %u\n",
1763                         msg_reply.request.slave, msg->request.slave);
1764                 ret = -1;
1765                 goto out;
1766         }
1767
1768         ret = msg_reply.payload.u64 ? -1 : 0;
1769
1770 out:
1771         rte_spinlock_unlock(&dev->slave_req_lock);
1772         return ret;
1773 }
1774
1775 int
1776 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1777 {
1778         int ret;
1779         struct VhostUserMsg msg = {
1780                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1781                 .flags = VHOST_USER_VERSION,
1782                 .size = sizeof(msg.payload.iotlb),
1783                 .payload.iotlb = {
1784                         .iova = iova,
1785                         .perm = perm,
1786                         .type = VHOST_IOTLB_MISS,
1787                 },
1788         };
1789
1790         ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1791         if (ret < 0) {
1792                 RTE_LOG(ERR, VHOST_CONFIG,
1793                                 "Failed to send IOTLB miss message (%d)\n",
1794                                 ret);
1795                 return ret;
1796         }
1797
1798         return 0;
1799 }
1800
1801 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1802                                                     int index, int fd,
1803                                                     uint64_t offset,
1804                                                     uint64_t size)
1805 {
1806         int *fdp = NULL;
1807         size_t fd_num = 0;
1808         int ret;
1809         struct VhostUserMsg msg = {
1810                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1811                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1812                 .size = sizeof(msg.payload.area),
1813                 .payload.area = {
1814                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
1815                         .size = size,
1816                         .offset = offset,
1817                 },
1818         };
1819
1820         if (fd < 0)
1821                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1822         else {
1823                 fdp = &fd;
1824                 fd_num = 1;
1825         }
1826
1827         ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1828         if (ret < 0) {
1829                 RTE_LOG(ERR, VHOST_CONFIG,
1830                         "Failed to set host notifier (%d)\n", ret);
1831                 return ret;
1832         }
1833
1834         return process_slave_message_reply(dev, &msg);
1835 }
1836
1837 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1838 {
1839         struct virtio_net *dev;
1840         struct rte_vdpa_device *vdpa_dev;
1841         int vfio_device_fd, did, ret = 0;
1842         uint64_t offset, size;
1843         unsigned int i;
1844
1845         dev = get_device(vid);
1846         if (!dev)
1847                 return -ENODEV;
1848
1849         did = dev->vdpa_dev_id;
1850         if (did < 0)
1851                 return -EINVAL;
1852
1853         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1854             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1855             !(dev->protocol_features &
1856                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1857             !(dev->protocol_features &
1858                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1859             !(dev->protocol_features &
1860                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1861                 return -ENOTSUP;
1862
1863         vdpa_dev = rte_vdpa_get_device(did);
1864         if (!vdpa_dev)
1865                 return -ENODEV;
1866
1867         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1868         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1869
1870         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1871         if (vfio_device_fd < 0)
1872                 return -ENOTSUP;
1873
1874         if (enable) {
1875                 for (i = 0; i < dev->nr_vring; i++) {
1876                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1877                                         &size) < 0) {
1878                                 ret = -ENOTSUP;
1879                                 goto disable;
1880                         }
1881
1882                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
1883                                         vfio_device_fd, offset, size) < 0) {
1884                                 ret = -EFAULT;
1885                                 goto disable;
1886                         }
1887                 }
1888         } else {
1889 disable:
1890                 for (i = 0; i < dev->nr_vring; i++) {
1891                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
1892                                         0, 0);
1893                 }
1894         }
1895
1896         return ret;
1897 }