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, struct VhostUserMsg *msg)
 150 {
 151         uint64_t features = 0;
 152
 153         rte_vhost_driver_get_features(dev->ifname, &features);
 154
 155         msg->payload.u64 = features;
 156         msg->size = sizeof(msg->payload.u64);
 157
 158         return features;
 159 }
 160
 161 /*
 162  * The queue number that we support are requested.
 163  */
 164 static uint32_t
 165 vhost_user_get_queue_num(struct virtio_net *dev, struct VhostUserMsg *msg)
 166 {
 167         uint32_t queue_num = 0;
 168
 169         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 170
 171         msg->payload.u64 = (uint64_t)queue_num;
 172         msg->size = sizeof(msg->payload.u64);
 173
 174         return queue_num;
 175 }
 176
 177 /*
 178  * We receive the negotiated features supported by us and the virtio device.
 179  */
 180 static int
 181 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 182 {
 183         uint64_t vhost_features = 0;
 184         struct rte_vdpa_device *vdpa_dev;
 185         int did = -1;
 186
 187         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 188         if (features & ~vhost_features) {
 189                 RTE_LOG(ERR, VHOST_CONFIG,
 190                         "(%d) received invalid negotiated features.\n",
 191                         dev->vid);
 192                 return -1;
 193         }
 194
 195         if (dev->flags & VIRTIO_DEV_RUNNING) {
 196                 if (dev->features == features)
 197                         return 0;
 198
 199                 /*
 200                  * Error out if master tries to change features while device is
 201                  * in running state. The exception being VHOST_F_LOG_ALL, which
 202                  * is enabled when the live-migration starts.
 203                  */
 204                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 205                         RTE_LOG(ERR, VHOST_CONFIG,
 206                                 "(%d) features changed while device is running.\n",
 207                                 dev->vid);
 208                         return -1;
 209                 }
 210
 211                 if (dev->notify_ops->features_changed)
 212                         dev->notify_ops->features_changed(dev->vid, features);
 213         }
 214
 215         dev->features = features;
 216         if (dev->features &
 217                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 218                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 219         } else {
 220                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 221         }
 222         VHOST_LOG_DEBUG(VHOST_CONFIG,
 223                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 224                 dev->vid,
 225                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 226                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 227
 228         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 229             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 230                 /*
 231                  * Remove all but first queue pair if MQ hasn't been
 232                  * negotiated. This is safe because the device is not
 233                  * running at this stage.
 234                  */
 235                 while (dev->nr_vring > 2) {
 236                         struct vhost_virtqueue *vq;
 237
 238                         vq = dev->virtqueue[--dev->nr_vring];
 239                         if (!vq)
 240                                 continue;
 241
 242                         dev->virtqueue[dev->nr_vring] = NULL;
 243                         cleanup_vq(vq, 1);
 244                         free_vq(dev, vq);
 245                 }
 246         }
 247
 248         did = dev->vdpa_dev_id;
 249         vdpa_dev = rte_vdpa_get_device(did);
 250         if (vdpa_dev && vdpa_dev->ops->set_features)
 251                 vdpa_dev->ops->set_features(dev->vid);
 252
 253         return 0;
 254 }
 255
 256 /*
 257  * The virtio device sends us the size of the descriptor ring.
 258  */
 259 static int
 260 vhost_user_set_vring_num(struct virtio_net *dev,
 261                          struct VhostUserMsg *msg)
 262 {
 263         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 264
 265         vq->size = msg->payload.state.num;
 266
 267         /* VIRTIO 1.0, 2.4 Virtqueues says:
 268          *
 269          *   Queue Size value is always a power of 2. The maximum Queue Size
 270          *   value is 32768.
 271          */
 272         if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
 273                 RTE_LOG(ERR, VHOST_CONFIG,
 274                         "invalid virtqueue size %u\n", vq->size);
 275                 return -1;
 276         }
 277
 278         if (dev->dequeue_zero_copy) {
 279                 vq->nr_zmbuf = 0;
 280                 vq->last_zmbuf_idx = 0;
 281                 vq->zmbuf_size = vq->size;
 282                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 283                                          sizeof(struct zcopy_mbuf), 0);
 284                 if (vq->zmbufs == NULL) {
 285                         RTE_LOG(WARNING, VHOST_CONFIG,
 286                                 "failed to allocate mem for zero copy; "
 287                                 "zero copy is force disabled\n");
 288                         dev->dequeue_zero_copy = 0;
 289                 }
 290                 TAILQ_INIT(&vq->zmbuf_list);
 291         }
 292
 293         if (vq_is_packed(dev)) {
 294                 vq->shadow_used_packed = rte_malloc(NULL,
 295                                 vq->size *
 296                                 sizeof(struct vring_used_elem_packed),
 297                                 RTE_CACHE_LINE_SIZE);
 298                 if (!vq->shadow_used_packed) {
 299                         RTE_LOG(ERR, VHOST_CONFIG,
 300                                         "failed to allocate memory for shadow used ring.\n");
 301                         return -1;
 302                 }
 303
 304         } else {
 305                 vq->shadow_used_split = rte_malloc(NULL,
 306                                 vq->size * sizeof(struct vring_used_elem),
 307                                 RTE_CACHE_LINE_SIZE);
 308                 if (!vq->shadow_used_split) {
 309                         RTE_LOG(ERR, VHOST_CONFIG,
 310                                         "failed to allocate memory for shadow used ring.\n");
 311                         return -1;
 312                 }
 313         }
 314
 315         vq->batch_copy_elems = rte_malloc(NULL,
 316                                 vq->size * sizeof(struct batch_copy_elem),
 317                                 RTE_CACHE_LINE_SIZE);
 318         if (!vq->batch_copy_elems) {
 319                 RTE_LOG(ERR, VHOST_CONFIG,
 320                         "failed to allocate memory for batching copy.\n");
 321                 return -1;
 322         }
 323
 324         return 0;
 325 }
 326
 327 /*
 328  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 329  * same numa node as the memory of vring descriptor.
 330  */
 331 #ifdef RTE_LIBRTE_VHOST_NUMA
 332 static struct virtio_net*
 333 numa_realloc(struct virtio_net *dev, int index)
 334 {
 335         int oldnode, newnode;
 336         struct virtio_net *old_dev;
 337         struct vhost_virtqueue *old_vq, *vq;
 338         struct zcopy_mbuf *new_zmbuf;
 339         struct vring_used_elem *new_shadow_used_split;
 340         struct vring_used_elem_packed *new_shadow_used_packed;
 341         struct batch_copy_elem *new_batch_copy_elems;
 342         int ret;
 343
 344         old_dev = dev;
 345         vq = old_vq = dev->virtqueue[index];
 346
 347         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 348                             MPOL_F_NODE | MPOL_F_ADDR);
 349
 350         /* check if we need to reallocate vq */
 351         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 352                              MPOL_F_NODE | MPOL_F_ADDR);
 353         if (ret) {
 354                 RTE_LOG(ERR, VHOST_CONFIG,
 355                         "Unable to get vq numa information.\n");
 356                 return dev;
 357         }
 358         if (oldnode != newnode) {
 359                 RTE_LOG(INFO, VHOST_CONFIG,
 360                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 361                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 362                 if (!vq)
 363                         return dev;
 364
 365                 memcpy(vq, old_vq, sizeof(*vq));
 366                 TAILQ_INIT(&vq->zmbuf_list);
 367
 368                 if (dev->dequeue_zero_copy) {
 369                         new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 370                                         sizeof(struct zcopy_mbuf), 0, newnode);
 371                         if (new_zmbuf) {
 372                                 rte_free(vq->zmbufs);
 373                                 vq->zmbufs = new_zmbuf;
 374                         }
 375                 }
 376
 377                 if (vq_is_packed(dev)) {
 378                         new_shadow_used_packed = rte_malloc_socket(NULL,
 379                                         vq->size *
 380                                         sizeof(struct vring_used_elem_packed),
 381                                         RTE_CACHE_LINE_SIZE,
 382                                         newnode);
 383                         if (new_shadow_used_packed) {
 384                                 rte_free(vq->shadow_used_packed);
 385                                 vq->shadow_used_packed = new_shadow_used_packed;
 386                         }
 387                 } else {
 388                         new_shadow_used_split = rte_malloc_socket(NULL,
 389                                         vq->size *
 390                                         sizeof(struct vring_used_elem),
 391                                         RTE_CACHE_LINE_SIZE,
 392                                         newnode);
 393                         if (new_shadow_used_split) {
 394                                 rte_free(vq->shadow_used_split);
 395                                 vq->shadow_used_split = new_shadow_used_split;
 396                         }
 397                 }
 398
 399                 new_batch_copy_elems = rte_malloc_socket(NULL,
 400                         vq->size * sizeof(struct batch_copy_elem),
 401                         RTE_CACHE_LINE_SIZE,
 402                         newnode);
 403                 if (new_batch_copy_elems) {
 404                         rte_free(vq->batch_copy_elems);
 405                         vq->batch_copy_elems = new_batch_copy_elems;
 406                 }
 407
 408                 rte_free(old_vq);
 409         }
 410
 411         /* check if we need to reallocate dev */
 412         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 413                             MPOL_F_NODE | MPOL_F_ADDR);
 414         if (ret) {
 415                 RTE_LOG(ERR, VHOST_CONFIG,
 416                         "Unable to get dev numa information.\n");
 417                 goto out;
 418         }
 419         if (oldnode != newnode) {
 420                 RTE_LOG(INFO, VHOST_CONFIG,
 421                         "reallocate dev from %d to %d node\n",
 422                         oldnode, newnode);
 423                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 424                 if (!dev) {
 425                         dev = old_dev;
 426                         goto out;
 427                 }
 428
 429                 memcpy(dev, old_dev, sizeof(*dev));
 430                 rte_free(old_dev);
 431         }
 432
 433 out:
 434         dev->virtqueue[index] = vq;
 435         vhost_devices[dev->vid] = dev;
 436
 437         if (old_vq != vq)
 438                 vhost_user_iotlb_init(dev, index);
 439
 440         return dev;
 441 }
 442 #else
 443 static struct virtio_net*
 444 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 445 {
 446         return dev;
 447 }
 448 #endif
 449
 450 /* Converts QEMU virtual address to Vhost virtual address. */
 451 static uint64_t
 452 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 453 {
 454         struct rte_vhost_mem_region *r;
 455         uint32_t i;
 456
 457         /* Find the region where the address lives. */
 458         for (i = 0; i < dev->mem->nregions; i++) {
 459                 r = &dev->mem->regions[i];
 460
 461                 if (qva >= r->guest_user_addr &&
 462                     qva <  r->guest_user_addr + r->size) {
 463
 464                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 465                                 *len = r->guest_user_addr + r->size - qva;
 466
 467                         return qva - r->guest_user_addr +
 468                                r->host_user_addr;
 469                 }
 470         }
 471         *len = 0;
 472
 473         return 0;
 474 }
 475
 476
 477 /*
 478  * Converts ring address to Vhost virtual address.
 479  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 480  * else it is a QEMU virtual address.
 481  */
 482 static uint64_t
 483 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 484                 uint64_t ra, uint64_t *size)
 485 {
 486         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 487                 uint64_t vva;
 488
 489                 vva = vhost_user_iotlb_cache_find(vq, ra,
 490                                         size, VHOST_ACCESS_RW);
 491                 if (!vva)
 492                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 493
 494                 return vva;
 495         }
 496
 497         return qva_to_vva(dev, ra, size);
 498 }
 499
 500 static struct virtio_net *
 501 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 502 {
 503         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 504         struct vhost_vring_addr *addr = &vq->ring_addrs;
 505         uint64_t len;
 506
 507         if (vq_is_packed(dev)) {
 508                 len = sizeof(struct vring_packed_desc) * vq->size;
 509                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 510                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 511                 vq->log_guest_addr = 0;
 512                 if (vq->desc_packed == NULL ||
 513                                 len != sizeof(struct vring_packed_desc) *
 514                                 vq->size) {
 515                         RTE_LOG(DEBUG, VHOST_CONFIG,
 516                                 "(%d) failed to map desc_packed ring.\n",
 517                                 dev->vid);
 518                         return dev;
 519                 }
 520
 521                 dev = numa_realloc(dev, vq_index);
 522                 vq = dev->virtqueue[vq_index];
 523                 addr = &vq->ring_addrs;
 524
 525                 len = sizeof(struct vring_packed_desc_event);
 526                 vq->driver_event = (struct vring_packed_desc_event *)
 527                                         (uintptr_t)ring_addr_to_vva(dev,
 528                                         vq, addr->avail_user_addr, &len);
 529                 if (vq->driver_event == NULL ||
 530                                 len != sizeof(struct vring_packed_desc_event)) {
 531                         RTE_LOG(DEBUG, VHOST_CONFIG,
 532                                 "(%d) failed to find driver area address.\n",
 533                                 dev->vid);
 534                         return dev;
 535                 }
 536
 537                 len = sizeof(struct vring_packed_desc_event);
 538                 vq->device_event = (struct vring_packed_desc_event *)
 539                                         (uintptr_t)ring_addr_to_vva(dev,
 540                                         vq, addr->used_user_addr, &len);
 541                 if (vq->device_event == NULL ||
 542                                 len != sizeof(struct vring_packed_desc_event)) {
 543                         RTE_LOG(DEBUG, VHOST_CONFIG,
 544                                 "(%d) failed to find device area address.\n",
 545                                 dev->vid);
 546                         return dev;
 547                 }
 548
 549                 return dev;
 550         }
 551
 552         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 553         if (vq->desc && vq->avail && vq->used)
 554                 return dev;
 555
 556         len = sizeof(struct vring_desc) * vq->size;
 557         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 558                         vq, addr->desc_user_addr, &len);
 559         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 560                 RTE_LOG(DEBUG, VHOST_CONFIG,
 561                         "(%d) failed to map desc ring.\n",
 562                         dev->vid);
 563                 return dev;
 564         }
 565
 566         dev = numa_realloc(dev, vq_index);
 567         vq = dev->virtqueue[vq_index];
 568         addr = &vq->ring_addrs;
 569
 570         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 571         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 572                         vq, addr->avail_user_addr, &len);
 573         if (vq->avail == 0 ||
 574                         len != sizeof(struct vring_avail) +
 575                         sizeof(uint16_t) * vq->size) {
 576                 RTE_LOG(DEBUG, VHOST_CONFIG,
 577                         "(%d) failed to map avail ring.\n",
 578                         dev->vid);
 579                 return dev;
 580         }
 581
 582         len = sizeof(struct vring_used) +
 583                 sizeof(struct vring_used_elem) * vq->size;
 584         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 585                         vq, addr->used_user_addr, &len);
 586         if (vq->used == 0 || len != sizeof(struct vring_used) +
 587                         sizeof(struct vring_used_elem) * vq->size) {
 588                 RTE_LOG(DEBUG, VHOST_CONFIG,
 589                         "(%d) failed to map used ring.\n",
 590                         dev->vid);
 591                 return dev;
 592         }
 593
 594         if (vq->last_used_idx != vq->used->idx) {
 595                 RTE_LOG(WARNING, VHOST_CONFIG,
 596                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 597                         "some packets maybe resent for Tx and dropped for Rx\n",
 598                         vq->last_used_idx, vq->used->idx);
 599                 vq->last_used_idx  = vq->used->idx;
 600                 vq->last_avail_idx = vq->used->idx;
 601         }
 602
 603         vq->log_guest_addr = addr->log_guest_addr;
 604
 605         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 606                         dev->vid, vq->desc);
 607         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 608                         dev->vid, vq->avail);
 609         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 610                         dev->vid, vq->used);
 611         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 612                         dev->vid, vq->log_guest_addr);
 613
 614         return dev;
 615 }
 616
 617 /*
 618  * The virtio device sends us the desc, used and avail ring addresses.
 619  * This function then converts these to our address space.
 620  */
 621 static int
 622 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg)
 623 {
 624         struct vhost_virtqueue *vq;
 625         struct vhost_vring_addr *addr = &msg->payload.addr;
 626         struct virtio_net *dev = *pdev;
 627
 628         if (dev->mem == NULL)
 629                 return -1;
 630
 631         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 632         vq = dev->virtqueue[msg->payload.addr.index];
 633
 634         /*
 635          * Rings addresses should not be interpreted as long as the ring is not
 636          * started and enabled
 637          */
 638         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 639
 640         vring_invalidate(dev, vq);
 641
 642         if (vq->enabled && (dev->features &
 643                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 644                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 645                 if (!dev)
 646                         return -1;
 647
 648                 *pdev = dev;
 649         }
 650
 651         return 0;
 652 }
 653
 654 /*
 655  * The virtio device sends us the available ring last used index.
 656  */
 657 static int
 658 vhost_user_set_vring_base(struct virtio_net *dev,
 659                           struct VhostUserMsg *msg)
 660 {
 661         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 662                         msg->payload.state.num;
 663         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 664                         msg->payload.state.num;
 665
 666         return 0;
 667 }
 668
 669 static int
 670 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 671                    uint64_t host_phys_addr, uint64_t size)
 672 {
 673         struct guest_page *page, *last_page;
 674
 675         if (dev->nr_guest_pages == dev->max_guest_pages) {
 676                 dev->max_guest_pages *= 2;
 677                 dev->guest_pages = realloc(dev->guest_pages,
 678                                         dev->max_guest_pages * sizeof(*page));
 679                 if (!dev->guest_pages) {
 680                         RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
 681                         return -1;
 682                 }
 683         }
 684
 685         if (dev->nr_guest_pages > 0) {
 686                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 687                 /* merge if the two pages are continuous */
 688                 if (host_phys_addr == last_page->host_phys_addr +
 689                                       last_page->size) {
 690                         last_page->size += size;
 691                         return 0;
 692                 }
 693         }
 694
 695         page = &dev->guest_pages[dev->nr_guest_pages++];
 696         page->guest_phys_addr = guest_phys_addr;
 697         page->host_phys_addr  = host_phys_addr;
 698         page->size = size;
 699
 700         return 0;
 701 }
 702
 703 static int
 704 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 705                 uint64_t page_size)
 706 {
 707         uint64_t reg_size = reg->size;
 708         uint64_t host_user_addr  = reg->host_user_addr;
 709         uint64_t guest_phys_addr = reg->guest_phys_addr;
 710         uint64_t host_phys_addr;
 711         uint64_t size;
 712
 713         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 714         size = page_size - (guest_phys_addr & (page_size - 1));
 715         size = RTE_MIN(size, reg_size);
 716
 717         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 718                 return -1;
 719
 720         host_user_addr  += size;
 721         guest_phys_addr += size;
 722         reg_size -= size;
 723
 724         while (reg_size > 0) {
 725                 size = RTE_MIN(reg_size, page_size);
 726                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 727                                                   host_user_addr);
 728                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 729                                 size) < 0)
 730                         return -1;
 731
 732                 host_user_addr  += size;
 733                 guest_phys_addr += size;
 734                 reg_size -= size;
 735         }
 736
 737         return 0;
 738 }
 739
 740 #ifdef RTE_LIBRTE_VHOST_DEBUG
 741 /* TODO: enable it only in debug mode? */
 742 static void
 743 dump_guest_pages(struct virtio_net *dev)
 744 {
 745         uint32_t i;
 746         struct guest_page *page;
 747
 748         for (i = 0; i < dev->nr_guest_pages; i++) {
 749                 page = &dev->guest_pages[i];
 750
 751                 RTE_LOG(INFO, VHOST_CONFIG,
 752                         "guest physical page region %u\n"
 753                         "\t guest_phys_addr: %" PRIx64 "\n"
 754                         "\t host_phys_addr : %" PRIx64 "\n"
 755                         "\t size           : %" PRIx64 "\n",
 756                         i,
 757                         page->guest_phys_addr,
 758                         page->host_phys_addr,
 759                         page->size);
 760         }
 761 }
 762 #else
 763 #define dump_guest_pages(dev)
 764 #endif
 765
 766 static bool
 767 vhost_memory_changed(struct VhostUserMemory *new,
 768                      struct rte_vhost_memory *old)
 769 {
 770         uint32_t i;
 771
 772         if (new->nregions != old->nregions)
 773                 return true;
 774
 775         for (i = 0; i < new->nregions; ++i) {
 776                 VhostUserMemoryRegion *new_r = &new->regions[i];
 777                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 778
 779                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 780                         return true;
 781                 if (new_r->memory_size != old_r->size)
 782                         return true;
 783                 if (new_r->userspace_addr != old_r->guest_user_addr)
 784                         return true;
 785         }
 786
 787         return false;
 788 }
 789
 790 static int
 791 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg)
 792 {
 793         struct virtio_net *dev = *pdev;
 794         struct VhostUserMemory memory = msg->payload.memory;
 795         struct rte_vhost_mem_region *reg;
 796         void *mmap_addr;
 797         uint64_t mmap_size;
 798         uint64_t mmap_offset;
 799         uint64_t alignment;
 800         uint32_t i;
 801         int populate;
 802         int fd;
 803
 804         if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
 805                 RTE_LOG(ERR, VHOST_CONFIG,
 806                         "too many memory regions (%u)\n", memory.nregions);
 807                 return -1;
 808         }
 809
 810         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 811                 RTE_LOG(INFO, VHOST_CONFIG,
 812                         "(%d) memory regions not changed\n", dev->vid);
 813
 814                 for (i = 0; i < memory.nregions; i++)
 815                         close(msg->fds[i]);
 816
 817                 return 0;
 818         }
 819
 820         if (dev->mem) {
 821                 free_mem_region(dev);
 822                 rte_free(dev->mem);
 823                 dev->mem = NULL;
 824         }
 825
 826         /* Flush IOTLB cache as previous HVAs are now invalid */
 827         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
 828                 for (i = 0; i < dev->nr_vring; i++)
 829                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
 830
 831         dev->nr_guest_pages = 0;
 832         if (!dev->guest_pages) {
 833                 dev->max_guest_pages = 8;
 834                 dev->guest_pages = malloc(dev->max_guest_pages *
 835                                                 sizeof(struct guest_page));
 836                 if (dev->guest_pages == NULL) {
 837                         RTE_LOG(ERR, VHOST_CONFIG,
 838                                 "(%d) failed to allocate memory "
 839                                 "for dev->guest_pages\n",
 840                                 dev->vid);
 841                         return -1;
 842                 }
 843         }
 844
 845         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 846                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 847         if (dev->mem == NULL) {
 848                 RTE_LOG(ERR, VHOST_CONFIG,
 849                         "(%d) failed to allocate memory for dev->mem\n",
 850                         dev->vid);
 851                 return -1;
 852         }
 853         dev->mem->nregions = memory.nregions;
 854
 855         for (i = 0; i < memory.nregions; i++) {
 856                 fd  = msg->fds[i];
 857                 reg = &dev->mem->regions[i];
 858
 859                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 860                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 861                 reg->size            = memory.regions[i].memory_size;
 862                 reg->fd              = fd;
 863
 864                 mmap_offset = memory.regions[i].mmap_offset;
 865
 866                 /* Check for memory_size + mmap_offset overflow */
 867                 if (mmap_offset >= -reg->size) {
 868                         RTE_LOG(ERR, VHOST_CONFIG,
 869                                 "mmap_offset (%#"PRIx64") and memory_size "
 870                                 "(%#"PRIx64") overflow\n",
 871                                 mmap_offset, reg->size);
 872                         goto err_mmap;
 873                 }
 874
 875                 mmap_size = reg->size + mmap_offset;
 876
 877                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 878                  * with length argument aligned with hugepagesz at older
 879                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 880                  * mmap() will fail with EINVAL.
 881                  *
 882                  * to avoid failure, make sure in caller to keep length
 883                  * aligned.
 884                  */
 885                 alignment = get_blk_size(fd);
 886                 if (alignment == (uint64_t)-1) {
 887                         RTE_LOG(ERR, VHOST_CONFIG,
 888                                 "couldn't get hugepage size through fstat\n");
 889                         goto err_mmap;
 890                 }
 891                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 892
 893                 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
 894                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 895                                  MAP_SHARED | populate, fd, 0);
 896
 897                 if (mmap_addr == MAP_FAILED) {
 898                         RTE_LOG(ERR, VHOST_CONFIG,
 899                                 "mmap region %u failed.\n", i);
 900                         goto err_mmap;
 901                 }
 902
 903                 reg->mmap_addr = mmap_addr;
 904                 reg->mmap_size = mmap_size;
 905                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 906                                       mmap_offset;
 907
 908                 if (dev->dequeue_zero_copy)
 909                         if (add_guest_pages(dev, reg, alignment) < 0) {
 910                                 RTE_LOG(ERR, VHOST_CONFIG,
 911                                         "adding guest pages to region %u failed.\n",
 912                                         i);
 913                                 goto err_mmap;
 914                         }
 915
 916                 RTE_LOG(INFO, VHOST_CONFIG,
 917                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 918                         "\t guest physical addr: 0x%" PRIx64 "\n"
 919                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 920                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 921                         "\t mmap addr : 0x%" PRIx64 "\n"
 922                         "\t mmap size : 0x%" PRIx64 "\n"
 923                         "\t mmap align: 0x%" PRIx64 "\n"
 924                         "\t mmap off  : 0x%" PRIx64 "\n",
 925                         i, reg->size,
 926                         reg->guest_phys_addr,
 927                         reg->guest_user_addr,
 928                         reg->host_user_addr,
 929                         (uint64_t)(uintptr_t)mmap_addr,
 930                         mmap_size,
 931                         alignment,
 932                         mmap_offset);
 933         }
 934
 935         for (i = 0; i < dev->nr_vring; i++) {
 936                 struct vhost_virtqueue *vq = dev->virtqueue[i];
 937
 938                 if (vq->desc || vq->avail || vq->used) {
 939                         /*
 940                          * If the memory table got updated, the ring addresses
 941                          * need to be translated again as virtual addresses have
 942                          * changed.
 943                          */
 944                         vring_invalidate(dev, vq);
 945
 946                         dev = translate_ring_addresses(dev, i);
 947                         if (!dev)
 948                                 return -1;
 949
 950                         *pdev = dev;
 951                 }
 952         }
 953
 954         dump_guest_pages(dev);
 955
 956         return 0;
 957
 958 err_mmap:
 959         free_mem_region(dev);
 960         rte_free(dev->mem);
 961         dev->mem = NULL;
 962         return -1;
 963 }
 964
 965 static bool
 966 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
 967 {
 968         bool rings_ok;
 969
 970         if (!vq)
 971                 return false;
 972
 973         if (vq_is_packed(dev))
 974                 rings_ok = !!vq->desc_packed;
 975         else
 976                 rings_ok = vq->desc && vq->avail && vq->used;
 977
 978         return rings_ok &&
 979                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 980                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 981 }
 982
 983 static int
 984 virtio_is_ready(struct virtio_net *dev)
 985 {
 986         struct vhost_virtqueue *vq;
 987         uint32_t i;
 988
 989         if (dev->nr_vring == 0)
 990                 return 0;
 991
 992         for (i = 0; i < dev->nr_vring; i++) {
 993                 vq = dev->virtqueue[i];
 994
 995                 if (!vq_is_ready(dev, vq))
 996                         return 0;
 997         }
 998
 999         RTE_LOG(INFO, VHOST_CONFIG,
1000                 "virtio is now ready for processing.\n");
1001         return 1;
1002 }
1003
1004 static void
1005 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *msg)
1006 {
1007         struct vhost_vring_file file;
1008         struct vhost_virtqueue *vq;
1009
1010         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1011         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1012                 file.fd = VIRTIO_INVALID_EVENTFD;
1013         else
1014                 file.fd = msg->fds[0];
1015         RTE_LOG(INFO, VHOST_CONFIG,
1016                 "vring call idx:%d file:%d\n", file.index, file.fd);
1017
1018         vq = dev->virtqueue[file.index];
1019         if (vq->callfd >= 0)
1020                 close(vq->callfd);
1021
1022         vq->callfd = file.fd;
1023 }
1024
1025 static int
1026 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg)
1027 {
1028         struct vhost_vring_file file;
1029         struct vhost_virtqueue *vq;
1030         struct virtio_net *dev = *pdev;
1031
1032         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1033         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1034                 file.fd = VIRTIO_INVALID_EVENTFD;
1035         else
1036                 file.fd = msg->fds[0];
1037         RTE_LOG(INFO, VHOST_CONFIG,
1038                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1039
1040         /* Interpret ring addresses only when ring is started. */
1041         dev = translate_ring_addresses(dev, file.index);
1042         if (!dev)
1043                 return -1;
1044
1045         *pdev = dev;
1046
1047         vq = dev->virtqueue[file.index];
1048
1049         /*
1050          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1051          * the ring starts already enabled. Otherwise, it is enabled via
1052          * the SET_VRING_ENABLE message.
1053          */
1054         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
1055                 vq->enabled = 1;
1056
1057         if (vq->kickfd >= 0)
1058                 close(vq->kickfd);
1059         vq->kickfd = file.fd;
1060         return 0;
1061 }
1062
1063 static void
1064 free_zmbufs(struct vhost_virtqueue *vq)
1065 {
1066         struct zcopy_mbuf *zmbuf, *next;
1067
1068         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
1069              zmbuf != NULL; zmbuf = next) {
1070                 next = TAILQ_NEXT(zmbuf, next);
1071
1072                 rte_pktmbuf_free(zmbuf->mbuf);
1073                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
1074         }
1075
1076         rte_free(vq->zmbufs);
1077 }
1078
1079 /*
1080  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1081  */
1082 static int
1083 vhost_user_get_vring_base(struct virtio_net *dev,
1084                           struct VhostUserMsg *msg)
1085 {
1086         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1087
1088         /* We have to stop the queue (virtio) if it is running. */
1089         vhost_destroy_device_notify(dev);
1090
1091         dev->flags &= ~VIRTIO_DEV_READY;
1092         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1093
1094         /* Here we are safe to get the last avail index */
1095         msg->payload.state.num = vq->last_avail_idx;
1096
1097         RTE_LOG(INFO, VHOST_CONFIG,
1098                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1099                 msg->payload.state.num);
1100         /*
1101          * Based on current qemu vhost-user implementation, this message is
1102          * sent and only sent in vhost_vring_stop.
1103          * TODO: cleanup the vring, it isn't usable since here.
1104          */
1105         if (vq->kickfd >= 0)
1106                 close(vq->kickfd);
1107
1108         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1109
1110         if (vq->callfd >= 0)
1111                 close(vq->callfd);
1112
1113         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1114
1115         if (dev->dequeue_zero_copy)
1116                 free_zmbufs(vq);
1117         if (vq_is_packed(dev)) {
1118                 rte_free(vq->shadow_used_packed);
1119                 vq->shadow_used_packed = NULL;
1120         } else {
1121                 rte_free(vq->shadow_used_split);
1122                 vq->shadow_used_split = NULL;
1123         }
1124
1125         rte_free(vq->batch_copy_elems);
1126         vq->batch_copy_elems = NULL;
1127
1128         msg->size = sizeof(msg->payload.state);
1129
1130         return 0;
1131 }
1132
1133 /*
1134  * when virtio queues are ready to work, qemu will send us to
1135  * enable the virtio queue pair.
1136  */
1137 static int
1138 vhost_user_set_vring_enable(struct virtio_net *dev,
1139                             struct VhostUserMsg *msg)
1140 {
1141         int enable = (int)msg->payload.state.num;
1142         int index = (int)msg->payload.state.index;
1143         struct rte_vdpa_device *vdpa_dev;
1144         int did = -1;
1145
1146         RTE_LOG(INFO, VHOST_CONFIG,
1147                 "set queue enable: %d to qp idx: %d\n",
1148                 enable, index);
1149
1150         did = dev->vdpa_dev_id;
1151         vdpa_dev = rte_vdpa_get_device(did);
1152         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1153                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1154
1155         if (dev->notify_ops->vring_state_changed)
1156                 dev->notify_ops->vring_state_changed(dev->vid,
1157                                 index, enable);
1158
1159         dev->virtqueue[index]->enabled = enable;
1160
1161         return 0;
1162 }
1163
1164 static void
1165 vhost_user_get_protocol_features(struct virtio_net *dev,
1166                                  struct VhostUserMsg *msg)
1167 {
1168         uint64_t features, protocol_features;
1169
1170         rte_vhost_driver_get_features(dev->ifname, &features);
1171         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1172
1173         /*
1174          * REPLY_ACK protocol feature is only mandatory for now
1175          * for IOMMU feature. If IOMMU is explicitly disabled by the
1176          * application, disable also REPLY_ACK feature for older buggy
1177          * Qemu versions (from v2.7.0 to v2.9.0).
1178          */
1179         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1180                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1181
1182         msg->payload.u64 = protocol_features;
1183         msg->size = sizeof(msg->payload.u64);
1184 }
1185
1186 static int
1187 vhost_user_set_protocol_features(struct virtio_net *dev,
1188                                  uint64_t protocol_features)
1189 {
1190         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES) {
1191                 RTE_LOG(ERR, VHOST_CONFIG,
1192                         "(%d) received invalid protocol features.\n",
1193                         dev->vid);
1194                 return -1;
1195         }
1196
1197         dev->protocol_features = protocol_features;
1198         return 0;
1199 }
1200
1201 static int
1202 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1203 {
1204         int fd = msg->fds[0];
1205         uint64_t size, off;
1206         void *addr;
1207
1208         if (fd < 0) {
1209                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1210                 return -1;
1211         }
1212
1213         if (msg->size != sizeof(VhostUserLog)) {
1214                 RTE_LOG(ERR, VHOST_CONFIG,
1215                         "invalid log base msg size: %"PRId32" != %d\n",
1216                         msg->size, (int)sizeof(VhostUserLog));
1217                 return -1;
1218         }
1219
1220         size = msg->payload.log.mmap_size;
1221         off  = msg->payload.log.mmap_offset;
1222
1223         /* Don't allow mmap_offset to point outside the mmap region */
1224         if (off > size) {
1225                 RTE_LOG(ERR, VHOST_CONFIG,
1226                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1227                         off, size);
1228                 return -1;
1229         }
1230
1231         RTE_LOG(INFO, VHOST_CONFIG,
1232                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1233                 size, off);
1234
1235         /*
1236          * mmap from 0 to workaround a hugepage mmap bug: mmap will
1237          * fail when offset is not page size aligned.
1238          */
1239         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1240         close(fd);
1241         if (addr == MAP_FAILED) {
1242                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1243                 return -1;
1244         }
1245
1246         /*
1247          * Free previously mapped log memory on occasionally
1248          * multiple VHOST_USER_SET_LOG_BASE.
1249          */
1250         if (dev->log_addr) {
1251                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1252         }
1253         dev->log_addr = (uint64_t)(uintptr_t)addr;
1254         dev->log_base = dev->log_addr + off;
1255         dev->log_size = size;
1256
1257         msg->size = sizeof(msg->payload.u64);
1258
1259         return 0;
1260 }
1261
1262 /*
1263  * An rarp packet is constructed and broadcasted to notify switches about
1264  * the new location of the migrated VM, so that packets from outside will
1265  * not be lost after migration.
1266  *
1267  * However, we don't actually "send" a rarp packet here, instead, we set
1268  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1269  */
1270 static int
1271 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1272 {
1273         uint8_t *mac = (uint8_t *)&msg->payload.u64;
1274         struct rte_vdpa_device *vdpa_dev;
1275         int did = -1;
1276
1277         RTE_LOG(DEBUG, VHOST_CONFIG,
1278                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1279                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1280         memcpy(dev->mac.addr_bytes, mac, 6);
1281
1282         /*
1283          * Set the flag to inject a RARP broadcast packet at
1284          * rte_vhost_dequeue_burst().
1285          *
1286          * rte_smp_wmb() is for making sure the mac is copied
1287          * before the flag is set.
1288          */
1289         rte_smp_wmb();
1290         rte_atomic16_set(&dev->broadcast_rarp, 1);
1291         did = dev->vdpa_dev_id;
1292         vdpa_dev = rte_vdpa_get_device(did);
1293         if (vdpa_dev && vdpa_dev->ops->migration_done)
1294                 vdpa_dev->ops->migration_done(dev->vid);
1295
1296         return 0;
1297 }
1298
1299 static int
1300 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1301 {
1302         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1303                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1304                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1305                                 msg->payload.u64);
1306
1307                 return -1;
1308         }
1309
1310         dev->mtu = msg->payload.u64;
1311
1312         return 0;
1313 }
1314
1315 static int
1316 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1317 {
1318         int fd = msg->fds[0];
1319
1320         if (fd < 0) {
1321                 RTE_LOG(ERR, VHOST_CONFIG,
1322                                 "Invalid file descriptor for slave channel (%d)\n",
1323                                 fd);
1324                 return -1;
1325         }
1326
1327         dev->slave_req_fd = fd;
1328
1329         return 0;
1330 }
1331
1332 static int
1333 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1334 {
1335         struct vhost_vring_addr *ra;
1336         uint64_t start, end;
1337
1338         start = imsg->iova;
1339         end = start + imsg->size;
1340
1341         ra = &vq->ring_addrs;
1342         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1343                 return 1;
1344         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1345                 return 1;
1346         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1347                 return 1;
1348
1349         return 0;
1350 }
1351
1352 static int
1353 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1354                                 struct vhost_iotlb_msg *imsg)
1355 {
1356         uint64_t istart, iend, vstart, vend;
1357
1358         istart = imsg->iova;
1359         iend = istart + imsg->size - 1;
1360
1361         vstart = (uintptr_t)vq->desc;
1362         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1363         if (vstart <= iend && istart <= vend)
1364                 return 1;
1365
1366         vstart = (uintptr_t)vq->avail;
1367         vend = vstart + sizeof(struct vring_avail);
1368         vend += sizeof(uint16_t) * vq->size - 1;
1369         if (vstart <= iend && istart <= vend)
1370                 return 1;
1371
1372         vstart = (uintptr_t)vq->used;
1373         vend = vstart + sizeof(struct vring_used);
1374         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1375         if (vstart <= iend && istart <= vend)
1376                 return 1;
1377
1378         return 0;
1379 }
1380
1381 static int
1382 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1383 {
1384         struct virtio_net *dev = *pdev;
1385         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1386         uint16_t i;
1387         uint64_t vva, len;
1388
1389         switch (imsg->type) {
1390         case VHOST_IOTLB_UPDATE:
1391                 len = imsg->size;
1392                 vva = qva_to_vva(dev, imsg->uaddr, &len);
1393                 if (!vva)
1394                         return -1;
1395
1396                 for (i = 0; i < dev->nr_vring; i++) {
1397                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1398
1399                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1400                                         len, imsg->perm);
1401
1402                         if (is_vring_iotlb_update(vq, imsg))
1403                                 *pdev = dev = translate_ring_addresses(dev, i);
1404                 }
1405                 break;
1406         case VHOST_IOTLB_INVALIDATE:
1407                 for (i = 0; i < dev->nr_vring; i++) {
1408                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1409
1410                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1411                                         imsg->size);
1412
1413                         if (is_vring_iotlb_invalidate(vq, imsg))
1414                                 vring_invalidate(dev, vq);
1415                 }
1416                 break;
1417         default:
1418                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1419                                 imsg->type);
1420                 return -1;
1421         }
1422
1423         return 0;
1424 }
1425
1426 /* return bytes# of read on success or negative val on failure. */
1427 static int
1428 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1429 {
1430         int ret;
1431
1432         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1433                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1434         if (ret <= 0)
1435                 return ret;
1436
1437         if (msg && msg->size) {
1438                 if (msg->size > sizeof(msg->payload)) {
1439                         RTE_LOG(ERR, VHOST_CONFIG,
1440                                 "invalid msg size: %d\n", msg->size);
1441                         return -1;
1442                 }
1443                 ret = read(sockfd, &msg->payload, msg->size);
1444                 if (ret <= 0)
1445                         return ret;
1446                 if (ret != (int)msg->size) {
1447                         RTE_LOG(ERR, VHOST_CONFIG,
1448                                 "read control message failed\n");
1449                         return -1;
1450                 }
1451         }
1452
1453         return ret;
1454 }
1455
1456 static int
1457 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1458 {
1459         if (!msg)
1460                 return 0;
1461
1462         return send_fd_message(sockfd, (char *)msg,
1463                 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1464 }
1465
1466 static int
1467 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1468 {
1469         if (!msg)
1470                 return 0;
1471
1472         msg->flags &= ~VHOST_USER_VERSION_MASK;
1473         msg->flags &= ~VHOST_USER_NEED_REPLY;
1474         msg->flags |= VHOST_USER_VERSION;
1475         msg->flags |= VHOST_USER_REPLY_MASK;
1476
1477         return send_vhost_message(sockfd, msg, NULL, 0);
1478 }
1479
1480 static int
1481 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg,
1482                          int *fds, int fd_num)
1483 {
1484         int ret;
1485
1486         if (msg->flags & VHOST_USER_NEED_REPLY)
1487                 rte_spinlock_lock(&dev->slave_req_lock);
1488
1489         ret = send_vhost_message(dev->slave_req_fd, msg, fds, fd_num);
1490         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
1491                 rte_spinlock_unlock(&dev->slave_req_lock);
1492
1493         return ret;
1494 }
1495
1496 /*
1497  * Allocate a queue pair if it hasn't been allocated yet
1498  */
1499 static int
1500 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
1501                         struct VhostUserMsg *msg)
1502 {
1503         uint16_t vring_idx;
1504
1505         switch (msg->request.master) {
1506         case VHOST_USER_SET_VRING_KICK:
1507         case VHOST_USER_SET_VRING_CALL:
1508         case VHOST_USER_SET_VRING_ERR:
1509                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1510                 break;
1511         case VHOST_USER_SET_VRING_NUM:
1512         case VHOST_USER_SET_VRING_BASE:
1513         case VHOST_USER_SET_VRING_ENABLE:
1514                 vring_idx = msg->payload.state.index;
1515                 break;
1516         case VHOST_USER_SET_VRING_ADDR:
1517                 vring_idx = msg->payload.addr.index;
1518                 break;
1519         default:
1520                 return 0;
1521         }
1522
1523         if (vring_idx >= VHOST_MAX_VRING) {
1524                 RTE_LOG(ERR, VHOST_CONFIG,
1525                         "invalid vring index: %u\n", vring_idx);
1526                 return -1;
1527         }
1528
1529         if (dev->virtqueue[vring_idx])
1530                 return 0;
1531
1532         return alloc_vring_queue(dev, vring_idx);
1533 }
1534
1535 static void
1536 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1537 {
1538         unsigned int i = 0;
1539         unsigned int vq_num = 0;
1540
1541         while (vq_num < dev->nr_vring) {
1542                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1543
1544                 if (vq) {
1545                         rte_spinlock_lock(&vq->access_lock);
1546                         vq_num++;
1547                 }
1548                 i++;
1549         }
1550 }
1551
1552 static void
1553 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1554 {
1555         unsigned int i = 0;
1556         unsigned int vq_num = 0;
1557
1558         while (vq_num < dev->nr_vring) {
1559                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1560
1561                 if (vq) {
1562                         rte_spinlock_unlock(&vq->access_lock);
1563                         vq_num++;
1564                 }
1565                 i++;
1566         }
1567 }
1568
1569 int
1570 vhost_user_msg_handler(int vid, int fd)
1571 {
1572         struct virtio_net *dev;
1573         struct VhostUserMsg msg;
1574         struct rte_vdpa_device *vdpa_dev;
1575         int did = -1;
1576         int ret;
1577         int unlock_required = 0;
1578         uint32_t skip_master = 0;
1579
1580         dev = get_device(vid);
1581         if (dev == NULL)
1582                 return -1;
1583
1584         if (!dev->notify_ops) {
1585                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1586                 if (!dev->notify_ops) {
1587                         RTE_LOG(ERR, VHOST_CONFIG,
1588                                 "failed to get callback ops for driver %s\n",
1589                                 dev->ifname);
1590                         return -1;
1591                 }
1592         }
1593
1594         ret = read_vhost_message(fd, &msg);
1595         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1596                 if (ret < 0)
1597                         RTE_LOG(ERR, VHOST_CONFIG,
1598                                 "vhost read message failed\n");
1599                 else if (ret == 0)
1600                         RTE_LOG(INFO, VHOST_CONFIG,
1601                                 "vhost peer closed\n");
1602                 else
1603                         RTE_LOG(ERR, VHOST_CONFIG,
1604                                 "vhost read incorrect message\n");
1605
1606                 return -1;
1607         }
1608
1609         ret = 0;
1610         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1611                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1612                         vhost_message_str[msg.request.master]);
1613         else
1614                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1615                         vhost_message_str[msg.request.master]);
1616
1617         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1618         if (ret < 0) {
1619                 RTE_LOG(ERR, VHOST_CONFIG,
1620                         "failed to alloc queue\n");
1621                 return -1;
1622         }
1623
1624         /*
1625          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1626          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1627          * and device is destroyed. destroy_device waits for queues to be
1628          * inactive, so it is safe. Otherwise taking the access_lock
1629          * would cause a dead lock.
1630          */
1631         switch (msg.request.master) {
1632         case VHOST_USER_SET_FEATURES:
1633         case VHOST_USER_SET_PROTOCOL_FEATURES:
1634         case VHOST_USER_SET_OWNER:
1635         case VHOST_USER_SET_MEM_TABLE:
1636         case VHOST_USER_SET_LOG_BASE:
1637         case VHOST_USER_SET_LOG_FD:
1638         case VHOST_USER_SET_VRING_NUM:
1639         case VHOST_USER_SET_VRING_ADDR:
1640         case VHOST_USER_SET_VRING_BASE:
1641         case VHOST_USER_SET_VRING_KICK:
1642         case VHOST_USER_SET_VRING_CALL:
1643         case VHOST_USER_SET_VRING_ERR:
1644         case VHOST_USER_SET_VRING_ENABLE:
1645         case VHOST_USER_SEND_RARP:
1646         case VHOST_USER_NET_SET_MTU:
1647         case VHOST_USER_SET_SLAVE_REQ_FD:
1648                 vhost_user_lock_all_queue_pairs(dev);
1649                 unlock_required = 1;
1650                 break;
1651         default:
1652                 break;
1653
1654         }
1655
1656         if (dev->extern_ops.pre_msg_handle) {
1657                 uint32_t need_reply;
1658
1659                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1660                                 (void *)&msg, &need_reply, &skip_master);
1661                 if (ret < 0)
1662                         goto skip_to_reply;
1663
1664                 if (need_reply)
1665                         send_vhost_reply(fd, &msg);
1666
1667                 if (skip_master)
1668                         goto skip_to_post_handle;
1669         }
1670
1671         switch (msg.request.master) {
1672         case VHOST_USER_GET_FEATURES:
1673                 vhost_user_get_features(dev, &msg);
1674                 send_vhost_reply(fd, &msg);
1675                 break;
1676         case VHOST_USER_SET_FEATURES:
1677                 ret = vhost_user_set_features(dev, msg.payload.u64);
1678                 break;
1679
1680         case VHOST_USER_GET_PROTOCOL_FEATURES:
1681                 vhost_user_get_protocol_features(dev, &msg);
1682                 send_vhost_reply(fd, &msg);
1683                 break;
1684         case VHOST_USER_SET_PROTOCOL_FEATURES:
1685                 ret = vhost_user_set_protocol_features(dev, msg.payload.u64);
1686                 break;
1687
1688         case VHOST_USER_SET_OWNER:
1689                 ret = vhost_user_set_owner();
1690                 break;
1691         case VHOST_USER_RESET_OWNER:
1692                 ret = vhost_user_reset_owner(dev);
1693                 break;
1694
1695         case VHOST_USER_SET_MEM_TABLE:
1696                 ret = vhost_user_set_mem_table(&dev, &msg);
1697                 break;
1698
1699         case VHOST_USER_SET_LOG_BASE:
1700                 ret = vhost_user_set_log_base(dev, &msg);
1701                 if (ret)
1702                         goto skip_to_reply;
1703                 /* it needs a reply */
1704                 send_vhost_reply(fd, &msg);
1705                 break;
1706         case VHOST_USER_SET_LOG_FD:
1707                 close(msg.fds[0]);
1708                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1709                 break;
1710
1711         case VHOST_USER_SET_VRING_NUM:
1712                 ret = vhost_user_set_vring_num(dev, &msg);
1713                 break;
1714         case VHOST_USER_SET_VRING_ADDR:
1715                 ret = vhost_user_set_vring_addr(&dev, &msg);
1716                 break;
1717         case VHOST_USER_SET_VRING_BASE:
1718                 ret = vhost_user_set_vring_base(dev, &msg);
1719                 break;
1720
1721         case VHOST_USER_GET_VRING_BASE:
1722                 ret = vhost_user_get_vring_base(dev, &msg);
1723                 if (ret)
1724                         goto skip_to_reply;
1725                 send_vhost_reply(fd, &msg);
1726                 break;
1727
1728         case VHOST_USER_SET_VRING_KICK:
1729                 ret = vhost_user_set_vring_kick(&dev, &msg);
1730                 break;
1731         case VHOST_USER_SET_VRING_CALL:
1732                 vhost_user_set_vring_call(dev, &msg);
1733                 break;
1734
1735         case VHOST_USER_SET_VRING_ERR:
1736                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1737                         close(msg.fds[0]);
1738                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1739                 break;
1740
1741         case VHOST_USER_GET_QUEUE_NUM:
1742                 vhost_user_get_queue_num(dev, &msg);
1743                 send_vhost_reply(fd, &msg);
1744                 break;
1745
1746         case VHOST_USER_SET_VRING_ENABLE:
1747                 ret = vhost_user_set_vring_enable(dev, &msg);
1748                 break;
1749         case VHOST_USER_SEND_RARP:
1750                 ret = vhost_user_send_rarp(dev, &msg);
1751                 break;
1752
1753         case VHOST_USER_NET_SET_MTU:
1754                 ret = vhost_user_net_set_mtu(dev, &msg);
1755                 break;
1756
1757         case VHOST_USER_SET_SLAVE_REQ_FD:
1758                 ret = vhost_user_set_req_fd(dev, &msg);
1759                 break;
1760
1761         case VHOST_USER_IOTLB_MSG:
1762                 ret = vhost_user_iotlb_msg(&dev, &msg);
1763                 break;
1764
1765         default:
1766                 ret = -1;
1767                 break;
1768         }
1769
1770 skip_to_post_handle:
1771         if (!ret && dev->extern_ops.post_msg_handle) {
1772                 uint32_t need_reply;
1773
1774                 ret = (*dev->extern_ops.post_msg_handle)(
1775                                 dev->vid, (void *)&msg, &need_reply);
1776                 if (ret < 0)
1777                         goto skip_to_reply;
1778
1779                 if (need_reply)
1780                         send_vhost_reply(fd, &msg);
1781         }
1782
1783 skip_to_reply:
1784         if (unlock_required)
1785                 vhost_user_unlock_all_queue_pairs(dev);
1786
1787         if (msg.flags & VHOST_USER_NEED_REPLY) {
1788                 msg.payload.u64 = !!ret;
1789                 msg.size = sizeof(msg.payload.u64);
1790                 send_vhost_reply(fd, &msg);
1791         } else if (ret) {
1792                 RTE_LOG(ERR, VHOST_CONFIG,
1793                         "vhost message handling failed.\n");
1794                 return -1;
1795         }
1796
1797         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1798                 dev->flags |= VIRTIO_DEV_READY;
1799
1800                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1801                         if (dev->dequeue_zero_copy) {
1802                                 RTE_LOG(INFO, VHOST_CONFIG,
1803                                                 "dequeue zero copy is enabled\n");
1804                         }
1805
1806                         if (dev->notify_ops->new_device(dev->vid) == 0)
1807                                 dev->flags |= VIRTIO_DEV_RUNNING;
1808                 }
1809         }
1810
1811         did = dev->vdpa_dev_id;
1812         vdpa_dev = rte_vdpa_get_device(did);
1813         if (vdpa_dev && virtio_is_ready(dev) &&
1814                         !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1815                         msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1816                 if (vdpa_dev->ops->dev_conf)
1817                         vdpa_dev->ops->dev_conf(dev->vid);
1818                 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1819                 if (vhost_user_host_notifier_ctrl(dev->vid, true) != 0) {
1820                         RTE_LOG(INFO, VHOST_CONFIG,
1821                                 "(%d) software relay is used for vDPA, performance may be low.\n",
1822                                 dev->vid);
1823                 }
1824         }
1825
1826         return 0;
1827 }
1828
1829 static int process_slave_message_reply(struct virtio_net *dev,
1830                                        const struct VhostUserMsg *msg)
1831 {
1832         struct VhostUserMsg msg_reply;
1833         int ret;
1834
1835         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
1836                 return 0;
1837
1838         if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
1839                 ret = -1;
1840                 goto out;
1841         }
1842
1843         if (msg_reply.request.slave != msg->request.slave) {
1844                 RTE_LOG(ERR, VHOST_CONFIG,
1845                         "Received unexpected msg type (%u), expected %u\n",
1846                         msg_reply.request.slave, msg->request.slave);
1847                 ret = -1;
1848                 goto out;
1849         }
1850
1851         ret = msg_reply.payload.u64 ? -1 : 0;
1852
1853 out:
1854         rte_spinlock_unlock(&dev->slave_req_lock);
1855         return ret;
1856 }
1857
1858 int
1859 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1860 {
1861         int ret;
1862         struct VhostUserMsg msg = {
1863                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1864                 .flags = VHOST_USER_VERSION,
1865                 .size = sizeof(msg.payload.iotlb),
1866                 .payload.iotlb = {
1867                         .iova = iova,
1868                         .perm = perm,
1869                         .type = VHOST_IOTLB_MISS,
1870                 },
1871         };
1872
1873         ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1874         if (ret < 0) {
1875                 RTE_LOG(ERR, VHOST_CONFIG,
1876                                 "Failed to send IOTLB miss message (%d)\n",
1877                                 ret);
1878                 return ret;
1879         }
1880
1881         return 0;
1882 }
1883
1884 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
1885                                                     int index, int fd,
1886                                                     uint64_t offset,
1887                                                     uint64_t size)
1888 {
1889         int *fdp = NULL;
1890         size_t fd_num = 0;
1891         int ret;
1892         struct VhostUserMsg msg = {
1893                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
1894                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
1895                 .size = sizeof(msg.payload.area),
1896                 .payload.area = {
1897                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
1898                         .size = size,
1899                         .offset = offset,
1900                 },
1901         };
1902
1903         if (fd < 0)
1904                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
1905         else {
1906                 fdp = &fd;
1907                 fd_num = 1;
1908         }
1909
1910         ret = send_vhost_slave_message(dev, &msg, fdp, fd_num);
1911         if (ret < 0) {
1912                 RTE_LOG(ERR, VHOST_CONFIG,
1913                         "Failed to set host notifier (%d)\n", ret);
1914                 return ret;
1915         }
1916
1917         return process_slave_message_reply(dev, &msg);
1918 }
1919
1920 int vhost_user_host_notifier_ctrl(int vid, bool enable)
1921 {
1922         struct virtio_net *dev;
1923         struct rte_vdpa_device *vdpa_dev;
1924         int vfio_device_fd, did, ret = 0;
1925         uint64_t offset, size;
1926         unsigned int i;
1927
1928         dev = get_device(vid);
1929         if (!dev)
1930                 return -ENODEV;
1931
1932         did = dev->vdpa_dev_id;
1933         if (did < 0)
1934                 return -EINVAL;
1935
1936         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
1937             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
1938             !(dev->protocol_features &
1939                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
1940             !(dev->protocol_features &
1941                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
1942             !(dev->protocol_features &
1943                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
1944                 return -ENOTSUP;
1945
1946         vdpa_dev = rte_vdpa_get_device(did);
1947         if (!vdpa_dev)
1948                 return -ENODEV;
1949
1950         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
1951         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
1952
1953         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
1954         if (vfio_device_fd < 0)
1955                 return -ENOTSUP;
1956
1957         if (enable) {
1958                 for (i = 0; i < dev->nr_vring; i++) {
1959                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
1960                                         &size) < 0) {
1961                                 ret = -ENOTSUP;
1962                                 goto disable;
1963                         }
1964
1965                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
1966                                         vfio_device_fd, offset, size) < 0) {
1967                                 ret = -EFAULT;
1968                                 goto disable;
1969                         }
1970                 }
1971         } else {
1972 disable:
1973                 for (i = 0; i < dev->nr_vring; i++) {
1974                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
1975                                         0, 0);
1976                 }
1977         }
1978
1979         return ret;
1980 }