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(dev, 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         if (vq_is_packed(dev)) {
 286                 vq->shadow_used_packed = rte_malloc(NULL,
 287                                 vq->size *
 288                                 sizeof(struct vring_used_elem_packed),
 289                                 RTE_CACHE_LINE_SIZE);
 290                 if (!vq->shadow_used_packed) {
 291                         RTE_LOG(ERR, VHOST_CONFIG,
 292                                         "failed to allocate memory for shadow used ring.\n");
 293                         return -1;
 294                 }
 295
 296         } else {
 297                 vq->shadow_used_split = rte_malloc(NULL,
 298                                 vq->size * sizeof(struct vring_used_elem),
 299                                 RTE_CACHE_LINE_SIZE);
 300                 if (!vq->shadow_used_split) {
 301                         RTE_LOG(ERR, VHOST_CONFIG,
 302                                         "failed to allocate memory for shadow used ring.\n");
 303                         return -1;
 304                 }
 305         }
 306
 307         vq->batch_copy_elems = rte_malloc(NULL,
 308                                 vq->size * sizeof(struct batch_copy_elem),
 309                                 RTE_CACHE_LINE_SIZE);
 310         if (!vq->batch_copy_elems) {
 311                 RTE_LOG(ERR, VHOST_CONFIG,
 312                         "failed to allocate memory for batching copy.\n");
 313                 return -1;
 314         }
 315
 316         return 0;
 317 }
 318
 319 /*
 320  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 321  * same numa node as the memory of vring descriptor.
 322  */
 323 #ifdef RTE_LIBRTE_VHOST_NUMA
 324 static struct virtio_net*
 325 numa_realloc(struct virtio_net *dev, int index)
 326 {
 327         int oldnode, newnode;
 328         struct virtio_net *old_dev;
 329         struct vhost_virtqueue *old_vq, *vq;
 330         struct zcopy_mbuf *new_zmbuf;
 331         struct vring_used_elem *new_shadow_used_split;
 332         struct vring_used_elem_packed *new_shadow_used_packed;
 333         struct batch_copy_elem *new_batch_copy_elems;
 334         int ret;
 335
 336         old_dev = dev;
 337         vq = old_vq = dev->virtqueue[index];
 338
 339         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 340                             MPOL_F_NODE | MPOL_F_ADDR);
 341
 342         /* check if we need to reallocate vq */
 343         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 344                              MPOL_F_NODE | MPOL_F_ADDR);
 345         if (ret) {
 346                 RTE_LOG(ERR, VHOST_CONFIG,
 347                         "Unable to get vq numa information.\n");
 348                 return dev;
 349         }
 350         if (oldnode != newnode) {
 351                 RTE_LOG(INFO, VHOST_CONFIG,
 352                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 353                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 354                 if (!vq)
 355                         return dev;
 356
 357                 memcpy(vq, old_vq, sizeof(*vq));
 358                 TAILQ_INIT(&vq->zmbuf_list);
 359
 360                 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 361                         sizeof(struct zcopy_mbuf), 0, newnode);
 362                 if (new_zmbuf) {
 363                         rte_free(vq->zmbufs);
 364                         vq->zmbufs = new_zmbuf;
 365                 }
 366
 367                 if (vq_is_packed(dev)) {
 368                         new_shadow_used_packed = rte_malloc_socket(NULL,
 369                                         vq->size *
 370                                         sizeof(struct vring_used_elem_packed),
 371                                         RTE_CACHE_LINE_SIZE,
 372                                         newnode);
 373                         if (new_shadow_used_packed) {
 374                                 rte_free(vq->shadow_used_packed);
 375                                 vq->shadow_used_packed = new_shadow_used_packed;
 376                         }
 377                 } else {
 378                         new_shadow_used_split = rte_malloc_socket(NULL,
 379                                         vq->size *
 380                                         sizeof(struct vring_used_elem),
 381                                         RTE_CACHE_LINE_SIZE,
 382                                         newnode);
 383                         if (new_shadow_used_split) {
 384                                 rte_free(vq->shadow_used_split);
 385                                 vq->shadow_used_split = new_shadow_used_split;
 386                         }
 387                 }
 388
 389                 new_batch_copy_elems = rte_malloc_socket(NULL,
 390                         vq->size * sizeof(struct batch_copy_elem),
 391                         RTE_CACHE_LINE_SIZE,
 392                         newnode);
 393                 if (new_batch_copy_elems) {
 394                         rte_free(vq->batch_copy_elems);
 395                         vq->batch_copy_elems = new_batch_copy_elems;
 396                 }
 397
 398                 rte_free(old_vq);
 399         }
 400
 401         /* check if we need to reallocate dev */
 402         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 403                             MPOL_F_NODE | MPOL_F_ADDR);
 404         if (ret) {
 405                 RTE_LOG(ERR, VHOST_CONFIG,
 406                         "Unable to get dev numa information.\n");
 407                 goto out;
 408         }
 409         if (oldnode != newnode) {
 410                 RTE_LOG(INFO, VHOST_CONFIG,
 411                         "reallocate dev from %d to %d node\n",
 412                         oldnode, newnode);
 413                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 414                 if (!dev) {
 415                         dev = old_dev;
 416                         goto out;
 417                 }
 418
 419                 memcpy(dev, old_dev, sizeof(*dev));
 420                 rte_free(old_dev);
 421         }
 422
 423 out:
 424         dev->virtqueue[index] = vq;
 425         vhost_devices[dev->vid] = dev;
 426
 427         if (old_vq != vq)
 428                 vhost_user_iotlb_init(dev, index);
 429
 430         return dev;
 431 }
 432 #else
 433 static struct virtio_net*
 434 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 435 {
 436         return dev;
 437 }
 438 #endif
 439
 440 /* Converts QEMU virtual address to Vhost virtual address. */
 441 static uint64_t
 442 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 443 {
 444         struct rte_vhost_mem_region *r;
 445         uint32_t i;
 446
 447         /* Find the region where the address lives. */
 448         for (i = 0; i < dev->mem->nregions; i++) {
 449                 r = &dev->mem->regions[i];
 450
 451                 if (qva >= r->guest_user_addr &&
 452                     qva <  r->guest_user_addr + r->size) {
 453
 454                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 455                                 *len = r->guest_user_addr + r->size - qva;
 456
 457                         return qva - r->guest_user_addr +
 458                                r->host_user_addr;
 459                 }
 460         }
 461         *len = 0;
 462
 463         return 0;
 464 }
 465
 466
 467 /*
 468  * Converts ring address to Vhost virtual address.
 469  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 470  * else it is a QEMU virtual address.
 471  */
 472 static uint64_t
 473 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 474                 uint64_t ra, uint64_t *size)
 475 {
 476         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 477                 uint64_t vva;
 478
 479                 vva = vhost_user_iotlb_cache_find(vq, ra,
 480                                         size, VHOST_ACCESS_RW);
 481                 if (!vva)
 482                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 483
 484                 return vva;
 485         }
 486
 487         return qva_to_vva(dev, ra, size);
 488 }
 489
 490 static struct virtio_net *
 491 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 492 {
 493         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 494         struct vhost_vring_addr *addr = &vq->ring_addrs;
 495         uint64_t len;
 496
 497         if (vq_is_packed(dev)) {
 498                 len = sizeof(struct vring_packed_desc) * vq->size;
 499                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 500                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 501                 vq->log_guest_addr = 0;
 502                 if (vq->desc_packed == NULL ||
 503                                 len != sizeof(struct vring_packed_desc) *
 504                                 vq->size) {
 505                         RTE_LOG(DEBUG, VHOST_CONFIG,
 506                                 "(%d) failed to map desc_packed ring.\n",
 507                                 dev->vid);
 508                         return dev;
 509                 }
 510
 511                 dev = numa_realloc(dev, vq_index);
 512                 vq = dev->virtqueue[vq_index];
 513                 addr = &vq->ring_addrs;
 514
 515                 return dev;
 516         }
 517
 518         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 519         if (vq->desc && vq->avail && vq->used)
 520                 return dev;
 521
 522         len = sizeof(struct vring_desc) * vq->size;
 523         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 524                         vq, addr->desc_user_addr, &len);
 525         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 526                 RTE_LOG(DEBUG, VHOST_CONFIG,
 527                         "(%d) failed to map desc ring.\n",
 528                         dev->vid);
 529                 return dev;
 530         }
 531
 532         dev = numa_realloc(dev, vq_index);
 533         vq = dev->virtqueue[vq_index];
 534         addr = &vq->ring_addrs;
 535
 536         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 537         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 538                         vq, addr->avail_user_addr, &len);
 539         if (vq->avail == 0 ||
 540                         len != sizeof(struct vring_avail) +
 541                         sizeof(uint16_t) * vq->size) {
 542                 RTE_LOG(DEBUG, VHOST_CONFIG,
 543                         "(%d) failed to map avail ring.\n",
 544                         dev->vid);
 545                 return dev;
 546         }
 547
 548         len = sizeof(struct vring_used) +
 549                 sizeof(struct vring_used_elem) * vq->size;
 550         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 551                         vq, addr->used_user_addr, &len);
 552         if (vq->used == 0 || len != sizeof(struct vring_used) +
 553                         sizeof(struct vring_used_elem) * vq->size) {
 554                 RTE_LOG(DEBUG, VHOST_CONFIG,
 555                         "(%d) failed to map used ring.\n",
 556                         dev->vid);
 557                 return dev;
 558         }
 559
 560         if (vq->last_used_idx != vq->used->idx) {
 561                 RTE_LOG(WARNING, VHOST_CONFIG,
 562                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 563                         "some packets maybe resent for Tx and dropped for Rx\n",
 564                         vq->last_used_idx, vq->used->idx);
 565                 vq->last_used_idx  = vq->used->idx;
 566                 vq->last_avail_idx = vq->used->idx;
 567         }
 568
 569         vq->log_guest_addr = addr->log_guest_addr;
 570
 571         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 572                         dev->vid, vq->desc);
 573         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 574                         dev->vid, vq->avail);
 575         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 576                         dev->vid, vq->used);
 577         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 578                         dev->vid, vq->log_guest_addr);
 579
 580         return dev;
 581 }
 582
 583 /*
 584  * The virtio device sends us the desc, used and avail ring addresses.
 585  * This function then converts these to our address space.
 586  */
 587 static int
 588 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 589 {
 590         struct vhost_virtqueue *vq;
 591         struct vhost_vring_addr *addr = &msg->payload.addr;
 592         struct virtio_net *dev = *pdev;
 593
 594         if (dev->mem == NULL)
 595                 return -1;
 596
 597         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 598         vq = dev->virtqueue[msg->payload.addr.index];
 599
 600         /*
 601          * Rings addresses should not be interpreted as long as the ring is not
 602          * started and enabled
 603          */
 604         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 605
 606         vring_invalidate(dev, vq);
 607
 608         if (vq->enabled && (dev->features &
 609                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 610                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 611                 if (!dev)
 612                         return -1;
 613
 614                 *pdev = dev;
 615         }
 616
 617         return 0;
 618 }
 619
 620 /*
 621  * The virtio device sends us the available ring last used index.
 622  */
 623 static int
 624 vhost_user_set_vring_base(struct virtio_net *dev,
 625                           VhostUserMsg *msg)
 626 {
 627         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 628                         msg->payload.state.num;
 629         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 630                         msg->payload.state.num;
 631
 632         return 0;
 633 }
 634
 635 static int
 636 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 637                    uint64_t host_phys_addr, uint64_t size)
 638 {
 639         struct guest_page *page, *last_page;
 640
 641         if (dev->nr_guest_pages == dev->max_guest_pages) {
 642                 dev->max_guest_pages *= 2;
 643                 dev->guest_pages = realloc(dev->guest_pages,
 644                                         dev->max_guest_pages * sizeof(*page));
 645                 if (!dev->guest_pages) {
 646                         RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
 647                         return -1;
 648                 }
 649         }
 650
 651         if (dev->nr_guest_pages > 0) {
 652                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 653                 /* merge if the two pages are continuous */
 654                 if (host_phys_addr == last_page->host_phys_addr +
 655                                       last_page->size) {
 656                         last_page->size += size;
 657                         return 0;
 658                 }
 659         }
 660
 661         page = &dev->guest_pages[dev->nr_guest_pages++];
 662         page->guest_phys_addr = guest_phys_addr;
 663         page->host_phys_addr  = host_phys_addr;
 664         page->size = size;
 665
 666         return 0;
 667 }
 668
 669 static int
 670 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 671                 uint64_t page_size)
 672 {
 673         uint64_t reg_size = reg->size;
 674         uint64_t host_user_addr  = reg->host_user_addr;
 675         uint64_t guest_phys_addr = reg->guest_phys_addr;
 676         uint64_t host_phys_addr;
 677         uint64_t size;
 678
 679         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 680         size = page_size - (guest_phys_addr & (page_size - 1));
 681         size = RTE_MIN(size, reg_size);
 682
 683         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 684                 return -1;
 685
 686         host_user_addr  += size;
 687         guest_phys_addr += size;
 688         reg_size -= size;
 689
 690         while (reg_size > 0) {
 691                 size = RTE_MIN(reg_size, page_size);
 692                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 693                                                   host_user_addr);
 694                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 695                                 size) < 0)
 696                         return -1;
 697
 698                 host_user_addr  += size;
 699                 guest_phys_addr += size;
 700                 reg_size -= size;
 701         }
 702
 703         return 0;
 704 }
 705
 706 #ifdef RTE_LIBRTE_VHOST_DEBUG
 707 /* TODO: enable it only in debug mode? */
 708 static void
 709 dump_guest_pages(struct virtio_net *dev)
 710 {
 711         uint32_t i;
 712         struct guest_page *page;
 713
 714         for (i = 0; i < dev->nr_guest_pages; i++) {
 715                 page = &dev->guest_pages[i];
 716
 717                 RTE_LOG(INFO, VHOST_CONFIG,
 718                         "guest physical page region %u\n"
 719                         "\t guest_phys_addr: %" PRIx64 "\n"
 720                         "\t host_phys_addr : %" PRIx64 "\n"
 721                         "\t size           : %" PRIx64 "\n",
 722                         i,
 723                         page->guest_phys_addr,
 724                         page->host_phys_addr,
 725                         page->size);
 726         }
 727 }
 728 #else
 729 #define dump_guest_pages(dev)
 730 #endif
 731
 732 static bool
 733 vhost_memory_changed(struct VhostUserMemory *new,
 734                      struct rte_vhost_memory *old)
 735 {
 736         uint32_t i;
 737
 738         if (new->nregions != old->nregions)
 739                 return true;
 740
 741         for (i = 0; i < new->nregions; ++i) {
 742                 VhostUserMemoryRegion *new_r = &new->regions[i];
 743                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 744
 745                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 746                         return true;
 747                 if (new_r->memory_size != old_r->size)
 748                         return true;
 749                 if (new_r->userspace_addr != old_r->guest_user_addr)
 750                         return true;
 751         }
 752
 753         return false;
 754 }
 755
 756 static int
 757 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 758 {
 759         struct virtio_net *dev = *pdev;
 760         struct VhostUserMemory memory = pmsg->payload.memory;
 761         struct rte_vhost_mem_region *reg;
 762         void *mmap_addr;
 763         uint64_t mmap_size;
 764         uint64_t mmap_offset;
 765         uint64_t alignment;
 766         uint32_t i;
 767         int populate;
 768         int fd;
 769
 770         if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
 771                 RTE_LOG(ERR, VHOST_CONFIG,
 772                         "too many memory regions (%u)\n", memory.nregions);
 773                 return -1;
 774         }
 775
 776         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 777                 RTE_LOG(INFO, VHOST_CONFIG,
 778                         "(%d) memory regions not changed\n", dev->vid);
 779
 780                 for (i = 0; i < memory.nregions; i++)
 781                         close(pmsg->fds[i]);
 782
 783                 return 0;
 784         }
 785
 786         if (dev->mem) {
 787                 free_mem_region(dev);
 788                 rte_free(dev->mem);
 789                 dev->mem = NULL;
 790         }
 791
 792         dev->nr_guest_pages = 0;
 793         if (!dev->guest_pages) {
 794                 dev->max_guest_pages = 8;
 795                 dev->guest_pages = malloc(dev->max_guest_pages *
 796                                                 sizeof(struct guest_page));
 797                 if (dev->guest_pages == NULL) {
 798                         RTE_LOG(ERR, VHOST_CONFIG,
 799                                 "(%d) failed to allocate memory "
 800                                 "for dev->guest_pages\n",
 801                                 dev->vid);
 802                         return -1;
 803                 }
 804         }
 805
 806         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 807                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 808         if (dev->mem == NULL) {
 809                 RTE_LOG(ERR, VHOST_CONFIG,
 810                         "(%d) failed to allocate memory for dev->mem\n",
 811                         dev->vid);
 812                 return -1;
 813         }
 814         dev->mem->nregions = memory.nregions;
 815
 816         for (i = 0; i < memory.nregions; i++) {
 817                 fd  = pmsg->fds[i];
 818                 reg = &dev->mem->regions[i];
 819
 820                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 821                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 822                 reg->size            = memory.regions[i].memory_size;
 823                 reg->fd              = fd;
 824
 825                 mmap_offset = memory.regions[i].mmap_offset;
 826
 827                 /* Check for memory_size + mmap_offset overflow */
 828                 if (mmap_offset >= -reg->size) {
 829                         RTE_LOG(ERR, VHOST_CONFIG,
 830                                 "mmap_offset (%#"PRIx64") and memory_size "
 831                                 "(%#"PRIx64") overflow\n",
 832                                 mmap_offset, reg->size);
 833                         goto err_mmap;
 834                 }
 835
 836                 mmap_size = reg->size + mmap_offset;
 837
 838                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 839                  * with length argument aligned with hugepagesz at older
 840                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 841                  * mmap() will fail with EINVAL.
 842                  *
 843                  * to avoid failure, make sure in caller to keep length
 844                  * aligned.
 845                  */
 846                 alignment = get_blk_size(fd);
 847                 if (alignment == (uint64_t)-1) {
 848                         RTE_LOG(ERR, VHOST_CONFIG,
 849                                 "couldn't get hugepage size through fstat\n");
 850                         goto err_mmap;
 851                 }
 852                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 853
 854                 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
 855                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 856                                  MAP_SHARED | populate, fd, 0);
 857
 858                 if (mmap_addr == MAP_FAILED) {
 859                         RTE_LOG(ERR, VHOST_CONFIG,
 860                                 "mmap region %u failed.\n", i);
 861                         goto err_mmap;
 862                 }
 863
 864                 reg->mmap_addr = mmap_addr;
 865                 reg->mmap_size = mmap_size;
 866                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 867                                       mmap_offset;
 868
 869                 if (dev->dequeue_zero_copy)
 870                         if (add_guest_pages(dev, reg, alignment) < 0) {
 871                                 RTE_LOG(ERR, VHOST_CONFIG,
 872                                         "adding guest pages to region %u failed.\n",
 873                                         i);
 874                                 goto err_mmap;
 875                         }
 876
 877                 RTE_LOG(INFO, VHOST_CONFIG,
 878                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 879                         "\t guest physical addr: 0x%" PRIx64 "\n"
 880                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 881                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 882                         "\t mmap addr : 0x%" PRIx64 "\n"
 883                         "\t mmap size : 0x%" PRIx64 "\n"
 884                         "\t mmap align: 0x%" PRIx64 "\n"
 885                         "\t mmap off  : 0x%" PRIx64 "\n",
 886                         i, reg->size,
 887                         reg->guest_phys_addr,
 888                         reg->guest_user_addr,
 889                         reg->host_user_addr,
 890                         (uint64_t)(uintptr_t)mmap_addr,
 891                         mmap_size,
 892                         alignment,
 893                         mmap_offset);
 894         }
 895
 896         for (i = 0; i < dev->nr_vring; i++) {
 897                 struct vhost_virtqueue *vq = dev->virtqueue[i];
 898
 899                 if (vq->desc || vq->avail || vq->used) {
 900                         /*
 901                          * If the memory table got updated, the ring addresses
 902                          * need to be translated again as virtual addresses have
 903                          * changed.
 904                          */
 905                         vring_invalidate(dev, vq);
 906
 907                         dev = translate_ring_addresses(dev, i);
 908                         if (!dev)
 909                                 return -1;
 910
 911                         *pdev = dev;
 912                 }
 913         }
 914
 915         dump_guest_pages(dev);
 916
 917         return 0;
 918
 919 err_mmap:
 920         free_mem_region(dev);
 921         rte_free(dev->mem);
 922         dev->mem = NULL;
 923         return -1;
 924 }
 925
 926 static bool
 927 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
 928 {
 929         bool rings_ok;
 930
 931         if (!vq)
 932                 return false;
 933
 934         if (vq_is_packed(dev))
 935                 rings_ok = !!vq->desc_packed;
 936         else
 937                 rings_ok = vq->desc && vq->avail && vq->used;
 938
 939         return rings_ok &&
 940                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 941                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 942 }
 943
 944 static int
 945 virtio_is_ready(struct virtio_net *dev)
 946 {
 947         struct vhost_virtqueue *vq;
 948         uint32_t i;
 949
 950         if (dev->nr_vring == 0)
 951                 return 0;
 952
 953         for (i = 0; i < dev->nr_vring; i++) {
 954                 vq = dev->virtqueue[i];
 955
 956                 if (!vq_is_ready(dev, vq))
 957                         return 0;
 958         }
 959
 960         RTE_LOG(INFO, VHOST_CONFIG,
 961                 "virtio is now ready for processing.\n");
 962         return 1;
 963 }
 964
 965 static void
 966 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 967 {
 968         struct vhost_vring_file file;
 969         struct vhost_virtqueue *vq;
 970
 971         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 972         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 973                 file.fd = VIRTIO_INVALID_EVENTFD;
 974         else
 975                 file.fd = pmsg->fds[0];
 976         RTE_LOG(INFO, VHOST_CONFIG,
 977                 "vring call idx:%d file:%d\n", file.index, file.fd);
 978
 979         vq = dev->virtqueue[file.index];
 980         if (vq->callfd >= 0)
 981                 close(vq->callfd);
 982
 983         vq->callfd = file.fd;
 984 }
 985
 986 static void
 987 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 988 {
 989         struct vhost_vring_file file;
 990         struct vhost_virtqueue *vq;
 991         struct virtio_net *dev = *pdev;
 992
 993         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 994         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 995                 file.fd = VIRTIO_INVALID_EVENTFD;
 996         else
 997                 file.fd = pmsg->fds[0];
 998         RTE_LOG(INFO, VHOST_CONFIG,
 999                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1000
1001         /* Interpret ring addresses only when ring is started. */
1002         dev = translate_ring_addresses(dev, file.index);
1003         if (!dev)
1004                 return;
1005
1006         *pdev = dev;
1007
1008         vq = dev->virtqueue[file.index];
1009
1010         /*
1011          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1012          * the ring starts already enabled. Otherwise, it is enabled via
1013          * the SET_VRING_ENABLE message.
1014          */
1015         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1016                 vq->enabled = 1;
1017
1018         if (vq->kickfd >= 0)
1019                 close(vq->kickfd);
1020         vq->kickfd = file.fd;
1021 }
1022
1023 static void
1024 free_zmbufs(struct vhost_virtqueue *vq)
1025 {
1026         struct zcopy_mbuf *zmbuf, *next;
1027
1028         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1029              zmbuf != NULL; zmbuf = next) {
1030                 next = TAILQ_NEXT(zmbuf, next);
1031
1032                 rte_pktmbuf_free(zmbuf->mbuf);
1033                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1034         }
1035
1036         rte_free(vq->zmbufs);
1037 }
1038
1039 /*
1040  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1041  */
1042 static int
1043 vhost_user_get_vring_base(struct virtio_net *dev,
1044                           VhostUserMsg *msg)
1045 {
1046         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1047
1048         /* We have to stop the queue (virtio) if it is running. */
1049         vhost_destroy_device_notify(dev);
1050
1051         dev->flags &= ~VIRTIO_DEV_READY;
1052         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1053
1054         /* Here we are safe to get the last avail index */
1055         msg->payload.state.num = vq->last_avail_idx;
1056
1057         RTE_LOG(INFO, VHOST_CONFIG,
1058                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1059                 msg->payload.state.num);
1060         /*
1061          * Based on current qemu vhost-user implementation, this message is
1062          * sent and only sent in vhost_vring_stop.
1063          * TODO: cleanup the vring, it isn't usable since here.
1064          */
1065         if (vq->kickfd >= 0)
1066                 close(vq->kickfd);
1067
1068         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1069
1070         if (vq->callfd >= 0)
1071                 close(vq->callfd);
1072
1073         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1074
1075         if (dev->dequeue_zero_copy)
1076                 free_zmbufs(vq);
1077         if (vq_is_packed(dev)) {
1078                 rte_free(vq->shadow_used_packed);
1079                 vq->shadow_used_packed = NULL;
1080         } else {
1081                 rte_free(vq->shadow_used_split);
1082                 vq->shadow_used_split = NULL;
1083         }
1084
1085         rte_free(vq->batch_copy_elems);
1086         vq->batch_copy_elems = NULL;
1087
1088         return 0;
1089 }
1090
1091 /*
1092  * when virtio queues are ready to work, qemu will send us to
1093  * enable the virtio queue pair.
1094  */
1095 static int
1096 vhost_user_set_vring_enable(struct virtio_net *dev,
1097                             VhostUserMsg *msg)
1098 {
1099         int enable = (int)msg->payload.state.num;
1100         int index = (int)msg->payload.state.index;
1101         struct rte_vdpa_device *vdpa_dev;
1102         int did = -1;
1103
1104         RTE_LOG(INFO, VHOST_CONFIG,
1105                 "set queue enable: %d to qp idx: %d\n",
1106                 enable, index);
1107
1108         did = dev->vdpa_dev_id;
1109         vdpa_dev = rte_vdpa_get_device(did);
1110         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1111                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1112
1113         if (dev->notify_ops->vring_state_changed)
1114                 dev->notify_ops->vring_state_changed(dev->vid,
1115                                 index, enable);
1116
1117         dev->virtqueue[index]->enabled = enable;
1118
1119         return 0;
1120 }
1121
1122 static void
1123 vhost_user_get_protocol_features(struct virtio_net *dev,
1124                                  struct VhostUserMsg *msg)
1125 {
1126         uint64_t features, protocol_features;
1127
1128         rte_vhost_driver_get_features(dev->ifname, &features);
1129         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1130
1131         /*
1132          * REPLY_ACK protocol feature is only mandatory for now
1133          * for IOMMU feature. If IOMMU is explicitly disabled by the
1134          * application, disable also REPLY_ACK feature for older buggy
1135          * Qemu versions (from v2.7.0 to v2.9.0).
1136          */
1137         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1138                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1139
1140         msg->payload.u64 = protocol_features;
1141         msg->size = sizeof(msg->payload.u64);
1142 }
1143
1144 static void
1145 vhost_user_set_protocol_features(struct virtio_net *dev,
1146                                  uint64_t protocol_features)
1147 {
1148         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1149                 return;
1150
1151         dev->protocol_features = protocol_features;
1152 }
1153
1154 static int
1155 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1156 {
1157         int fd = msg->fds[0];
1158         uint64_t size, off;
1159         void *addr;
1160
1161         if (fd < 0) {
1162                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1163                 return -1;
1164         }
1165
1166         if (msg->size != sizeof(VhostUserLog)) {
1167                 RTE_LOG(ERR, VHOST_CONFIG,
1168                         "invalid log base msg size: %"PRId32" != %d\n",
1169                         msg->size, (int)sizeof(VhostUserLog));
1170                 return -1;
1171         }
1172
1173         size = msg->payload.log.mmap_size;
1174         off  = msg->payload.log.mmap_offset;
1175
1176         /* Don't allow mmap_offset to point outside the mmap region */
1177         if (off > size) {
1178                 RTE_LOG(ERR, VHOST_CONFIG,
1179                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1180                         off, size);
1181                 return -1;
1182         }
1183
1184         RTE_LOG(INFO, VHOST_CONFIG,
1185                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1186                 size, off);
1187
1188         /*
1189          * mmap from 0 to workaround a hugepage mmap bug: mmap will
1190          * fail when offset is not page size aligned.
1191          */
1192         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1193         close(fd);
1194         if (addr == MAP_FAILED) {
1195                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1196                 return -1;
1197         }
1198
1199         /*
1200          * Free previously mapped log memory on occasionally
1201          * multiple VHOST_USER_SET_LOG_BASE.
1202          */
1203         if (dev->log_addr) {
1204                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1205         }
1206         dev->log_addr = (uint64_t)(uintptr_t)addr;
1207         dev->log_base = dev->log_addr + off;
1208         dev->log_size = size;
1209
1210         return 0;
1211 }
1212
1213 /*
1214  * An rarp packet is constructed and broadcasted to notify switches about
1215  * the new location of the migrated VM, so that packets from outside will
1216  * not be lost after migration.
1217  *
1218  * However, we don't actually "send" a rarp packet here, instead, we set
1219  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1220  */
1221 static int
1222 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1223 {
1224         uint8_t *mac = (uint8_t *)&msg->payload.u64;
1225         struct rte_vdpa_device *vdpa_dev;
1226         int did = -1;
1227
1228         RTE_LOG(DEBUG, VHOST_CONFIG,
1229                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1230                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1231         memcpy(dev->mac.addr_bytes, mac, 6);
1232
1233         /*
1234          * Set the flag to inject a RARP broadcast packet at
1235          * rte_vhost_dequeue_burst().
1236          *
1237          * rte_smp_wmb() is for making sure the mac is copied
1238          * before the flag is set.
1239          */
1240         rte_smp_wmb();
1241         rte_atomic16_set(&dev->broadcast_rarp, 1);
1242         did = dev->vdpa_dev_id;
1243         vdpa_dev = rte_vdpa_get_device(did);
1244         if (vdpa_dev && vdpa_dev->ops->migration_done)
1245                 vdpa_dev->ops->migration_done(dev->vid);
1246
1247         return 0;
1248 }
1249
1250 static int
1251 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1252 {
1253         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1254                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1255                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1256                                 msg->payload.u64);
1257
1258                 return -1;
1259         }
1260
1261         dev->mtu = msg->payload.u64;
1262
1263         return 0;
1264 }
1265
1266 static int
1267 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1268 {
1269         int fd = msg->fds[0];
1270
1271         if (fd < 0) {
1272                 RTE_LOG(ERR, VHOST_CONFIG,
1273                                 "Invalid file descriptor for slave channel (%d)\n",
1274                                 fd);
1275                 return -1;
1276         }
1277
1278         dev->slave_req_fd = fd;
1279
1280         return 0;
1281 }
1282
1283 static int
1284 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1285 {
1286         struct vhost_vring_addr *ra;
1287         uint64_t start, end;
1288
1289         start = imsg->iova;
1290         end = start + imsg->size;
1291
1292         ra = &vq->ring_addrs;
1293         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1294                 return 1;
1295         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1296                 return 1;
1297         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1298                 return 1;
1299
1300         return 0;
1301 }
1302
1303 static int
1304 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1305                                 struct vhost_iotlb_msg *imsg)
1306 {
1307         uint64_t istart, iend, vstart, vend;
1308
1309         istart = imsg->iova;
1310         iend = istart + imsg->size - 1;
1311
1312         vstart = (uintptr_t)vq->desc;
1313         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1314         if (vstart <= iend && istart <= vend)
1315                 return 1;
1316
1317         vstart = (uintptr_t)vq->avail;
1318         vend = vstart + sizeof(struct vring_avail);
1319         vend += sizeof(uint16_t) * vq->size - 1;
1320         if (vstart <= iend && istart <= vend)
1321                 return 1;
1322
1323         vstart = (uintptr_t)vq->used;
1324         vend = vstart + sizeof(struct vring_used);
1325         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1326         if (vstart <= iend && istart <= vend)
1327                 return 1;
1328
1329         return 0;
1330 }
1331
1332 static int
1333 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1334 {
1335         struct virtio_net *dev = *pdev;
1336         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1337         uint16_t i;
1338         uint64_t vva, len;
1339
1340         switch (imsg->type) {
1341         case VHOST_IOTLB_UPDATE:
1342                 len = imsg->size;
1343                 vva = qva_to_vva(dev, imsg->uaddr, &len);
1344                 if (!vva)
1345                         return -1;
1346
1347                 for (i = 0; i < dev->nr_vring; i++) {
1348                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1349
1350                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1351                                         len, imsg->perm);
1352
1353                         if (is_vring_iotlb_update(vq, imsg))
1354                                 *pdev = dev = translate_ring_addresses(dev, i);
1355                 }
1356                 break;
1357         case VHOST_IOTLB_INVALIDATE:
1358                 for (i = 0; i < dev->nr_vring; i++) {
1359                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1360
1361                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1362                                         imsg->size);
1363
1364                         if (is_vring_iotlb_invalidate(vq, imsg))
1365                                 vring_invalidate(dev, vq);
1366                 }
1367                 break;
1368         default:
1369                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1370                                 imsg->type);
1371                 return -1;
1372         }
1373
1374         return 0;
1375 }
1376
1377 /* return bytes# of read on success or negative val on failure. */
1378 static int
1379 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1380 {
1381         int ret;
1382
1383         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1384                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1385         if (ret <= 0)
1386                 return ret;
1387
1388         if (msg && msg->size) {
1389                 if (msg->size > sizeof(msg->payload)) {
1390                         RTE_LOG(ERR, VHOST_CONFIG,
1391                                 "invalid msg size: %d\n", msg->size);
1392                         return -1;
1393                 }
1394                 ret = read(sockfd, &msg->payload, msg->size);
1395                 if (ret <= 0)
1396                         return ret;
1397                 if (ret != (int)msg->size) {
1398                         RTE_LOG(ERR, VHOST_CONFIG,
1399                                 "read control message failed\n");
1400                         return -1;
1401                 }
1402         }
1403
1404         return ret;
1405 }
1406
1407 static int
1408 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1409 {
1410         if (!msg)
1411                 return 0;
1412
1413         return send_fd_message(sockfd, (char *)msg,
1414                 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1415 }
1416
1417 static int
1418 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1419 {
1420         if (!msg)
1421                 return 0;
1422
1423         msg->flags &= ~VHOST_USER_VERSION_MASK;
1424         msg->flags &= ~VHOST_USER_NEED_REPLY;
1425         msg->flags |= VHOST_USER_VERSION;
1426         msg->flags |= VHOST_USER_REPLY_MASK;
1427
1428         return send_vhost_message(sockfd, msg, NULL, 0);
1429 }
1430
1431 static int
1432 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1433                          int *fds, int fd_num)
1434 {
1435         int ret;
1436
1437         if (msg->flags & VHOST_USER_NEED_REPLY)
1438                 rte_spinlock_lock(&dev->slave_req_lock);
1439
1440         ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1441         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1442                 rte_spinlock_unlock(&dev->slave_req_lock);
1443
1444         return ret;
1445 }
1446
1447 /*
1448  * Allocate a queue pair if it hasn't been allocated yet
1449  */
1450 static int
1451 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1452 {
1453         uint16_t vring_idx;
1454
1455         switch (msg->request.master) {
1456         case VHOST_USER_SET_VRING_KICK:
1457         case VHOST_USER_SET_VRING_CALL:
1458         case VHOST_USER_SET_VRING_ERR:
1459                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1460                 break;
1461         case VHOST_USER_SET_VRING_NUM:
1462         case VHOST_USER_SET_VRING_BASE:
1463         case VHOST_USER_SET_VRING_ENABLE:
1464                 vring_idx = msg->payload.state.index;
1465                 break;
1466         case VHOST_USER_SET_VRING_ADDR:
1467                 vring_idx = msg->payload.addr.index;
1468                 break;
1469         default:
1470                 return 0;
1471         }
1472
1473         if (vring_idx >= VHOST_MAX_VRING) {
1474                 RTE_LOG(ERR, VHOST_CONFIG,
1475                         "invalid vring index: %u\n", vring_idx);
1476                 return -1;
1477         }
1478
1479         if (dev->virtqueue[vring_idx])
1480                 return 0;
1481
1482         return alloc_vring_queue(dev, vring_idx);
1483 }
1484
1485 static void
1486 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1487 {
1488         unsigned int i = 0;
1489         unsigned int vq_num = 0;
1490
1491         while (vq_num < dev->nr_vring) {
1492                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1493
1494                 if (vq) {
1495                         rte_spinlock_lock(&vq->access_lock);
1496                         vq_num++;
1497                 }
1498                 i++;
1499         }
1500 }
1501
1502 static void
1503 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1504 {
1505         unsigned int i = 0;
1506         unsigned int vq_num = 0;
1507
1508         while (vq_num < dev->nr_vring) {
1509                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1510
1511                 if (vq) {
1512                         rte_spinlock_unlock(&vq->access_lock);
1513                         vq_num++;
1514                 }
1515                 i++;
1516         }
1517 }
1518
1519 int
1520 vhost_user_msg_handler(int vid, int fd)
1521 {
1522         struct virtio_net *dev;
1523         struct VhostUserMsg msg;
1524         struct rte_vdpa_device *vdpa_dev;
1525         int did = -1;
1526         int ret;
1527         int unlock_required = 0;
1528         uint32_t skip_master = 0;
1529
1530         dev = get_device(vid);
1531         if (dev == NULL)
1532                 return -1;
1533
1534         if (!dev->notify_ops) {
1535                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1536                 if (!dev->notify_ops) {
1537                         RTE_LOG(ERR, VHOST_CONFIG,
1538                                 "failed to get callback ops for driver %s\n",
1539                                 dev->ifname);
1540                         return -1;
1541                 }
1542         }
1543
1544         ret = read_vhost_message(fd, &msg);
1545         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1546                 if (ret < 0)
1547                         RTE_LOG(ERR, VHOST_CONFIG,
1548                                 "vhost read message failed\n");
1549                 else if (ret == 0)
1550                         RTE_LOG(INFO, VHOST_CONFIG,
1551                                 "vhost peer closed\n");
1552                 else
1553                         RTE_LOG(ERR, VHOST_CONFIG,
1554                                 "vhost read incorrect message\n");
1555
1556                 return -1;
1557         }
1558
1559         ret = 0;
1560         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1561                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1562                         vhost_message_str[msg.request.master]);
1563         else
1564                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1565                         vhost_message_str[msg.request.master]);
1566
1567         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1568         if (ret < 0) {
1569                 RTE_LOG(ERR, VHOST_CONFIG,
1570                         "failed to alloc queue\n");
1571                 return -1;
1572         }
1573
1574         /*
1575          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1576          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1577          * and device is destroyed. destroy_device waits for queues to be
1578          * inactive, so it is safe. Otherwise taking the access_lock
1579          * would cause a dead lock.
1580          */
1581         switch (msg.request.master) {
1582         case VHOST_USER_SET_FEATURES:
1583         case VHOST_USER_SET_PROTOCOL_FEATURES:
1584         case VHOST_USER_SET_OWNER:
1585         case VHOST_USER_SET_MEM_TABLE:
1586         case VHOST_USER_SET_LOG_BASE:
1587         case VHOST_USER_SET_LOG_FD:
1588         case VHOST_USER_SET_VRING_NUM:
1589         case VHOST_USER_SET_VRING_ADDR:
1590         case VHOST_USER_SET_VRING_BASE:
1591         case VHOST_USER_SET_VRING_KICK:
1592         case VHOST_USER_SET_VRING_CALL:
1593         case VHOST_USER_SET_VRING_ERR:
1594         case VHOST_USER_SET_VRING_ENABLE:
1595         case VHOST_USER_SEND_RARP:
1596         case VHOST_USER_NET_SET_MTU:
1597         case VHOST_USER_SET_SLAVE_REQ_FD:
1598                 vhost_user_lock_all_queue_pairs(dev);
1599                 unlock_required = 1;
1600                 break;
1601         default:
1602                 break;
1603
1604         }
1605
1606         if (dev->extern_ops.pre_msg_handle) {
1607                 uint32_t need_reply;
1608
1609                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1610                                 (void *)&msg, &need_reply, &skip_master);
1611                 if (ret < 0)
1612                         goto skip_to_reply;
1613
1614                 if (need_reply)
1615                         send_vhost_reply(fd, &msg);
1616
1617                 if (skip_master)
1618                         goto skip_to_post_handle;
1619         }
1620
1621         switch (msg.request.master) {
1622         case VHOST_USER_GET_FEATURES:
1623                 msg.payload.u64 = vhost_user_get_features(dev);
1624                 msg.size = sizeof(msg.payload.u64);
1625                 send_vhost_reply(fd, &msg);
1626                 break;
1627         case VHOST_USER_SET_FEATURES:
1628                 ret = vhost_user_set_features(dev, msg.payload.u64);
1629                 if (ret)
1630                         return -1;
1631                 break;
1632
1633         case VHOST_USER_GET_PROTOCOL_FEATURES:
1634                 vhost_user_get_protocol_features(dev, &msg);
1635                 send_vhost_reply(fd, &msg);
1636                 break;
1637         case VHOST_USER_SET_PROTOCOL_FEATURES:
1638                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1639                 break;
1640
1641         case VHOST_USER_SET_OWNER:
1642                 vhost_user_set_owner();
1643                 break;
1644         case VHOST_USER_RESET_OWNER:
1645                 vhost_user_reset_owner(dev);
1646                 break;
1647
1648         case VHOST_USER_SET_MEM_TABLE:
1649                 ret = vhost_user_set_mem_table(&dev, &msg);
1650                 break;
1651
1652         case VHOST_USER_SET_LOG_BASE:
1653                 vhost_user_set_log_base(dev, &msg);
1654
1655                 /* it needs a reply */
1656                 msg.size = sizeof(msg.payload.u64);
1657                 send_vhost_reply(fd, &msg);
1658                 break;
1659         case VHOST_USER_SET_LOG_FD:
1660                 close(msg.fds[0]);
1661                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1662                 break;
1663
1664         case VHOST_USER_SET_VRING_NUM:
1665                 vhost_user_set_vring_num(dev, &msg);
1666                 break;
1667         case VHOST_USER_SET_VRING_ADDR:
1668                 vhost_user_set_vring_addr(&dev, &msg);
1669                 break;
1670         case VHOST_USER_SET_VRING_BASE:
1671                 vhost_user_set_vring_base(dev, &msg);
1672                 break;
1673
1674         case VHOST_USER_GET_VRING_BASE:
1675                 vhost_user_get_vring_base(dev, &msg);
1676                 msg.size = sizeof(msg.payload.state);
1677                 send_vhost_reply(fd, &msg);
1678                 break;
1679
1680         case VHOST_USER_SET_VRING_KICK:
1681                 vhost_user_set_vring_kick(&dev, &msg);
1682                 break;
1683         case VHOST_USER_SET_VRING_CALL:
1684                 vhost_user_set_vring_call(dev, &msg);
1685                 break;
1686
1687         case VHOST_USER_SET_VRING_ERR:
1688                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1689                         close(msg.fds[0]);
1690                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1691                 break;
1692
1693         case VHOST_USER_GET_QUEUE_NUM:
1694                 msg.payload.u64 = (uint64_t)vhost_user_get_queue_num(dev);
1695                 msg.size = sizeof(msg.payload.u64);
1696                 send_vhost_reply(fd, &msg);
1697                 break;
1698
1699         case VHOST_USER_SET_VRING_ENABLE:
1700                 vhost_user_set_vring_enable(dev, &msg);
1701                 break;
1702         case VHOST_USER_SEND_RARP:
1703                 vhost_user_send_rarp(dev, &msg);
1704                 break;
1705
1706         case VHOST_USER_NET_SET_MTU:
1707                 ret = vhost_user_net_set_mtu(dev, &msg);
1708                 break;
1709
1710         case VHOST_USER_SET_SLAVE_REQ_FD:
1711                 ret = vhost_user_set_req_fd(dev, &msg);
1712                 break;
1713
1714         case VHOST_USER_IOTLB_MSG:
1715                 ret = vhost_user_iotlb_msg(&dev, &msg);
1716                 break;
1717
1718         default:
1719                 ret = -1;
1720                 break;
1721         }
1722
1723 skip_to_post_handle:
1724         if (dev->extern_ops.post_msg_handle) {
1725                 uint32_t need_reply;
1726
1727                 ret = (*dev->extern_ops.post_msg_handle)(
1728                                 dev->vid, (void *)&msg, &need_reply);
1729                 if (ret < 0)
1730                         goto skip_to_reply;
1731
1732                 if (need_reply)
1733                         send_vhost_reply(fd, &msg);
1734         }
1735
1736 skip_to_reply:
1737         if (unlock_required)
1738                 vhost_user_unlock_all_queue_pairs(dev);
1739
1740         if (msg.flags & VHOST_USER_NEED_REPLY) {
1741                 msg.payload.u64 = !!ret;
1742                 msg.size = sizeof(msg.payload.u64);
1743                 send_vhost_reply(fd, &msg);
1744         }
1745
1746         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1747                 dev->flags |= VIRTIO_DEV_READY;
1748
1749                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1750                         if (dev->dequeue_zero_copy) {
1751                                 RTE_LOG(INFO, VHOST_CONFIG,
1752                                                 "dequeue zero copy is enabled\n");
1753                         }
1754
1755                         if (dev->notify_ops->new_device(dev->vid) == 0)
1756                                 dev->flags |= VIRTIO_DEV_RUNNING;
1757                 }
1758         }
1759
1760         did = dev->vdpa_dev_id;
1761         vdpa_dev = rte_vdpa_get_device(did);
1762         if (vdpa_dev && virtio_is_ready(dev) &&
1763                         !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1764                         msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1765                 if (vdpa_dev->ops->dev_conf)
1766                         vdpa_dev->ops->dev_conf(dev->vid);
1767                 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1768                 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1769                         RTE_LOG(INFO, VHOST_CONFIG,
1770                                 "(%d) software relay is used for vDPA, performance may be low.\n",
1771                                 dev->vid);
1772                 }
1773         }
1774
1775         return 0;
1776 }
1777
1778 static int process_slave_message_reply(struct virtio_net *dev,
1779                                        const VhostUserMsg *msg)
1780 {
1781         VhostUserMsg msg_reply;
1782         int ret;
1783
1784         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1785                 return 0;
1786
1787         if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1788                 ret = -1;
1789                 goto out;
1790         }
1791
1792         if (msg_reply.request.slave != msg->request.slave) {
1793                 RTE_LOG(ERR, VHOST_CONFIG,
1794                         "Received unexpected msg type (%u), expected %u\n",
1795                         msg_reply.request.slave, msg->request.slave);
1796                 ret = -1;
1797                 goto out;
1798         }
1799
1800         ret = msg_reply.payload.u64 ? -1 : 0;
1801
1802 out:
1803         rte_spinlock_unlock(&dev->slave_req_lock);
1804         return ret;
1805 }
1806
1807 int
1808 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1809 {
1810         int ret;
1811         struct VhostUserMsg msg = {
1812                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1813                 .flags = VHOST_USER_VERSION,
1814                 .size = sizeof(msg.payload.iotlb),
1815                 .payload.iotlb = {
1816                         .iova = iova,
1817                         .perm = perm,
1818                         .type = VHOST_IOTLB_MISS,
1819                 },
1820         };
1821
1822         ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1823         if (ret < 0) {
1824                 RTE_LOG(ERR, VHOST_CONFIG,
1825                                 "Failed to send IOTLB miss message (%d)\n",
1826                                 ret);
1827                 return ret;
1828         }
1829
1830         return 0;
1831 }
1832
1833 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1834                                                     int index, int fd,
1835                                                     uint64_t offset,
1836                                                     uint64_t size)
1837 {
1838         int *fdp = NULL;
1839         size_t fd_num = 0;
1840         int ret;
1841         struct VhostUserMsg msg = {
1842                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1843                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1844                 .size = sizeof(msg.payload.area),
1845                 .payload.area = {
1846                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
1847                         .size = size,
1848                         .offset = offset,
1849                 },
1850         };
1851
1852         if (fd < 0)
1853                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1854         else {
1855                 fdp = &fd;
1856                 fd_num = 1;
1857         }
1858
1859         ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1860         if (ret < 0) {
1861                 RTE_LOG(ERR, VHOST_CONFIG,
1862                         "Failed to set host notifier (%d)\n", ret);
1863                 return ret;
1864         }
1865
1866         return process_slave_message_reply(dev, &msg);
1867 }
1868
1869 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1870 {
1871         struct virtio_net *dev;
1872         struct rte_vdpa_device *vdpa_dev;
1873         int vfio_device_fd, did, ret = 0;
1874         uint64_t offset, size;
1875         unsigned int i;
1876
1877         dev = get_device(vid);
1878         if (!dev)
1879                 return -ENODEV;
1880
1881         did = dev->vdpa_dev_id;
1882         if (did < 0)
1883                 return -EINVAL;
1884
1885         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1886             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1887             !(dev->protocol_features &
1888                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1889             !(dev->protocol_features &
1890                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1891             !(dev->protocol_features &
1892                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1893                 return -ENOTSUP;
1894
1895         vdpa_dev = rte_vdpa_get_device(did);
1896         if (!vdpa_dev)
1897                 return -ENODEV;
1898
1899         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1900         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1901
1902         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1903         if (vfio_device_fd < 0)
1904                 return -ENOTSUP;
1905
1906         if (enable) {
1907                 for (i = 0; i < dev->nr_vring; i++) {
1908                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1909                                         &size) < 0) {
1910                                 ret = -ENOTSUP;
1911                                 goto disable;
1912                         }
1913
1914                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
1915                                         vfio_device_fd, offset, size) < 0) {
1916                                 ret = -EFAULT;
1917                                 goto disable;
1918                         }
1919                 }
1920         } else {
1921 disable:
1922                 for (i = 0; i < dev->nr_vring; i++) {
1923                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
1924                                         0, 0);
1925                 }
1926         }
1927
1928         return ret;
1929 }