lib/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 <fcntl.h>
  28 #include <sys/ioctl.h>
  29 #include <sys/mman.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/syscall.h>
  33 #include <assert.h>
  34 #ifdef RTE_LIBRTE_VHOST_NUMA
  35 #include <numaif.h>
  36 #endif
  37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
  38 #include <linux/userfaultfd.h>
  39 #endif
  40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
  41 #include <linux/memfd.h>
  42 #define MEMFD_SUPPORTED
  43 #endif
  44
  45 #include <rte_common.h>
  46 #include <rte_malloc.h>
  47 #include <rte_log.h>
  48
  49 #include "iotlb.h"
  50 #include "vhost.h"
  51 #include "vhost_user.h"
  52
  53 #define VIRTIO_MIN_MTU 68
  54 #define VIRTIO_MAX_MTU 65535
  55
  56 #define INFLIGHT_ALIGNMENT      64
  57 #define INFLIGHT_VERSION        0x1
  58
  59 static const char *vhost_message_str[VHOST_USER_MAX] = {
  60         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  61         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  62         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  63         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  64         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  65         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  66         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  67         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  68         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  69         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  70         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  71         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  72         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  73         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  74         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  75         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  76         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  77         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  78         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  79         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  80         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  81         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  82         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  83         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  84         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  85         [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
  86         [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
  87         [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
  88         [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
  89         [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
  90         [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
  91         [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
  92 };
  93
  94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
  95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
  96
  97 static void
  98 close_msg_fds(struct VhostUserMsg *msg)
  99 {
 100         int i;
 101
 102         for (i = 0; i < msg->fd_num; i++) {
 103                 int fd = msg->fds[i];
 104
 105                 if (fd == -1)
 106                         continue;
 107
 108                 msg->fds[i] = -1;
 109                 close(fd);
 110         }
 111 }
 112
 113 /*
 114  * Ensure the expected number of FDs is received,
 115  * close all FDs and return an error if this is not the case.
 116  */
 117 static int
 118 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
 119 {
 120         if (msg->fd_num == expected_fds)
 121                 return 0;
 122
 123         VHOST_LOG_CONFIG(ERR,
 124                 " Expect %d FDs for request %s, received %d\n",
 125                 expected_fds,
 126                 vhost_message_str[msg->request.master],
 127                 msg->fd_num);
 128
 129         close_msg_fds(msg);
 130
 131         return -1;
 132 }
 133
 134 static uint64_t
 135 get_blk_size(int fd)
 136 {
 137         struct stat stat;
 138         int ret;
 139
 140         ret = fstat(fd, &stat);
 141         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 142 }
 143
 144 static void
 145 free_mem_region(struct virtio_net *dev)
 146 {
 147         uint32_t i;
 148         struct rte_vhost_mem_region *reg;
 149
 150         if (!dev || !dev->mem)
 151                 return;
 152
 153         for (i = 0; i < dev->mem->nregions; i++) {
 154                 reg = &dev->mem->regions[i];
 155                 if (reg->host_user_addr) {
 156                         munmap(reg->mmap_addr, reg->mmap_size);
 157                         close(reg->fd);
 158                 }
 159         }
 160 }
 161
 162 void
 163 vhost_backend_cleanup(struct virtio_net *dev)
 164 {
 165         if (dev->mem) {
 166                 free_mem_region(dev);
 167                 rte_free(dev->mem);
 168                 dev->mem = NULL;
 169         }
 170
 171         rte_free(dev->guest_pages);
 172         dev->guest_pages = NULL;
 173
 174         if (dev->log_addr) {
 175                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 176                 dev->log_addr = 0;
 177         }
 178
 179         if (dev->inflight_info) {
 180                 if (dev->inflight_info->addr) {
 181                         munmap(dev->inflight_info->addr,
 182                                dev->inflight_info->size);
 183                         dev->inflight_info->addr = NULL;
 184                 }
 185
 186                 if (dev->inflight_info->fd >= 0) {
 187                         close(dev->inflight_info->fd);
 188                         dev->inflight_info->fd = -1;
 189                 }
 190
 191                 free(dev->inflight_info);
 192                 dev->inflight_info = NULL;
 193         }
 194
 195         if (dev->slave_req_fd >= 0) {
 196                 close(dev->slave_req_fd);
 197                 dev->slave_req_fd = -1;
 198         }
 199
 200         if (dev->postcopy_ufd >= 0) {
 201                 close(dev->postcopy_ufd);
 202                 dev->postcopy_ufd = -1;
 203         }
 204
 205         dev->postcopy_listening = 0;
 206 }
 207
 208 static void
 209 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
 210                               int enable)
 211 {
 212         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
 213         struct vhost_virtqueue *vq = dev->virtqueue[index];
 214
 215         /* Configure guest notifications on enable */
 216         if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
 217                 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
 218
 219         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
 220                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
 221
 222         if (dev->notify_ops->vring_state_changed)
 223                 dev->notify_ops->vring_state_changed(dev->vid,
 224                                 index, enable);
 225 }
 226
 227 /*
 228  * This function just returns success at the moment unless
 229  * the device hasn't been initialised.
 230  */
 231 static int
 232 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
 233                         struct VhostUserMsg *msg,
 234                         int main_fd __rte_unused)
 235 {
 236         if (validate_msg_fds(msg, 0) != 0)
 237                 return RTE_VHOST_MSG_RESULT_ERR;
 238
 239         return RTE_VHOST_MSG_RESULT_OK;
 240 }
 241
 242 static int
 243 vhost_user_reset_owner(struct virtio_net **pdev,
 244                         struct VhostUserMsg *msg,
 245                         int main_fd __rte_unused)
 246 {
 247         struct virtio_net *dev = *pdev;
 248
 249         if (validate_msg_fds(msg, 0) != 0)
 250                 return RTE_VHOST_MSG_RESULT_ERR;
 251
 252         vhost_destroy_device_notify(dev);
 253
 254         cleanup_device(dev, 0);
 255         reset_device(dev);
 256         return RTE_VHOST_MSG_RESULT_OK;
 257 }
 258
 259 /*
 260  * The features that we support are requested.
 261  */
 262 static int
 263 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 264                         int main_fd __rte_unused)
 265 {
 266         struct virtio_net *dev = *pdev;
 267         uint64_t features = 0;
 268
 269         if (validate_msg_fds(msg, 0) != 0)
 270                 return RTE_VHOST_MSG_RESULT_ERR;
 271
 272         rte_vhost_driver_get_features(dev->ifname, &features);
 273
 274         msg->payload.u64 = features;
 275         msg->size = sizeof(msg->payload.u64);
 276         msg->fd_num = 0;
 277
 278         return RTE_VHOST_MSG_RESULT_REPLY;
 279 }
 280
 281 /*
 282  * The queue number that we support are requested.
 283  */
 284 static int
 285 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
 286                         int main_fd __rte_unused)
 287 {
 288         struct virtio_net *dev = *pdev;
 289         uint32_t queue_num = 0;
 290
 291         if (validate_msg_fds(msg, 0) != 0)
 292                 return RTE_VHOST_MSG_RESULT_ERR;
 293
 294         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 295
 296         msg->payload.u64 = (uint64_t)queue_num;
 297         msg->size = sizeof(msg->payload.u64);
 298         msg->fd_num = 0;
 299
 300         return RTE_VHOST_MSG_RESULT_REPLY;
 301 }
 302
 303 /*
 304  * We receive the negotiated features supported by us and the virtio device.
 305  */
 306 static int
 307 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 308                         int main_fd __rte_unused)
 309 {
 310         struct virtio_net *dev = *pdev;
 311         uint64_t features = msg->payload.u64;
 312         uint64_t vhost_features = 0;
 313         struct rte_vdpa_device *vdpa_dev;
 314
 315         if (validate_msg_fds(msg, 0) != 0)
 316                 return RTE_VHOST_MSG_RESULT_ERR;
 317
 318         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 319         if (features & ~vhost_features) {
 320                 VHOST_LOG_CONFIG(ERR,
 321                         "(%d) received invalid negotiated features.\n",
 322                         dev->vid);
 323                 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
 324                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
 325
 326                 return RTE_VHOST_MSG_RESULT_ERR;
 327         }
 328
 329         if (dev->flags & VIRTIO_DEV_RUNNING) {
 330                 if (dev->features == features)
 331                         return RTE_VHOST_MSG_RESULT_OK;
 332
 333                 /*
 334                  * Error out if master tries to change features while device is
 335                  * in running state. The exception being VHOST_F_LOG_ALL, which
 336                  * is enabled when the live-migration starts.
 337                  */
 338                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 339                         VHOST_LOG_CONFIG(ERR,
 340                                 "(%d) features changed while device is running.\n",
 341                                 dev->vid);
 342                         return RTE_VHOST_MSG_RESULT_ERR;
 343                 }
 344
 345                 if (dev->notify_ops->features_changed)
 346                         dev->notify_ops->features_changed(dev->vid, features);
 347         }
 348
 349         dev->features = features;
 350         if (dev->features &
 351                 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
 352                  (1ULL << VIRTIO_F_VERSION_1) |
 353                  (1ULL << VIRTIO_F_RING_PACKED))) {
 354                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 355         } else {
 356                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 357         }
 358         VHOST_LOG_CONFIG(INFO,
 359                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 360         VHOST_LOG_CONFIG(DEBUG,
 361                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 362                 dev->vid,
 363                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 364                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 365
 366         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 367             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 368                 /*
 369                  * Remove all but first queue pair if MQ hasn't been
 370                  * negotiated. This is safe because the device is not
 371                  * running at this stage.
 372                  */
 373                 while (dev->nr_vring > 2) {
 374                         struct vhost_virtqueue *vq;
 375
 376                         vq = dev->virtqueue[--dev->nr_vring];
 377                         if (!vq)
 378                                 continue;
 379
 380                         dev->virtqueue[dev->nr_vring] = NULL;
 381                         cleanup_vq(vq, 1);
 382                         cleanup_vq_inflight(dev, vq);
 383                         free_vq(dev, vq);
 384                 }
 385         }
 386
 387         vdpa_dev = dev->vdpa_dev;
 388         if (vdpa_dev)
 389                 vdpa_dev->ops->set_features(dev->vid);
 390
 391         dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
 392         return RTE_VHOST_MSG_RESULT_OK;
 393 }
 394
 395 /*
 396  * The virtio device sends us the size of the descriptor ring.
 397  */
 398 static int
 399 vhost_user_set_vring_num(struct virtio_net **pdev,
 400                         struct VhostUserMsg *msg,
 401                         int main_fd __rte_unused)
 402 {
 403         struct virtio_net *dev = *pdev;
 404         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 405
 406         if (validate_msg_fds(msg, 0) != 0)
 407                 return RTE_VHOST_MSG_RESULT_ERR;
 408
 409         if (msg->payload.state.num > 32768) {
 410                 VHOST_LOG_CONFIG(ERR, "invalid virtqueue size %u\n", msg->payload.state.num);
 411                 return RTE_VHOST_MSG_RESULT_ERR;
 412         }
 413
 414         vq->size = msg->payload.state.num;
 415
 416         /* VIRTIO 1.0, 2.4 Virtqueues says:
 417          *
 418          *   Queue Size value is always a power of 2. The maximum Queue Size
 419          *   value is 32768.
 420          *
 421          * VIRTIO 1.1 2.7 Virtqueues says:
 422          *
 423          *   Packed virtqueues support up to 2^15 entries each.
 424          */
 425         if (!vq_is_packed(dev)) {
 426                 if (vq->size & (vq->size - 1)) {
 427                         VHOST_LOG_CONFIG(ERR,
 428                                 "invalid virtqueue size %u\n", vq->size);
 429                         return RTE_VHOST_MSG_RESULT_ERR;
 430                 }
 431         }
 432
 433         if (vq_is_packed(dev)) {
 434                 if (vq->shadow_used_packed)
 435                         rte_free(vq->shadow_used_packed);
 436                 vq->shadow_used_packed = rte_malloc_socket(NULL,
 437                                 vq->size *
 438                                 sizeof(struct vring_used_elem_packed),
 439                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 440                 if (!vq->shadow_used_packed) {
 441                         VHOST_LOG_CONFIG(ERR,
 442                                         "failed to allocate memory for shadow used ring.\n");
 443                         return RTE_VHOST_MSG_RESULT_ERR;
 444                 }
 445
 446         } else {
 447                 if (vq->shadow_used_split)
 448                         rte_free(vq->shadow_used_split);
 449
 450                 vq->shadow_used_split = rte_malloc_socket(NULL,
 451                                 vq->size * sizeof(struct vring_used_elem),
 452                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 453
 454                 if (!vq->shadow_used_split) {
 455                         VHOST_LOG_CONFIG(ERR,
 456                                         "failed to allocate memory for vq internal data.\n");
 457                         return RTE_VHOST_MSG_RESULT_ERR;
 458                 }
 459         }
 460
 461         if (vq->batch_copy_elems)
 462                 rte_free(vq->batch_copy_elems);
 463         vq->batch_copy_elems = rte_malloc_socket(NULL,
 464                                 vq->size * sizeof(struct batch_copy_elem),
 465                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 466         if (!vq->batch_copy_elems) {
 467                 VHOST_LOG_CONFIG(ERR,
 468                         "failed to allocate memory for batching copy.\n");
 469                 return RTE_VHOST_MSG_RESULT_ERR;
 470         }
 471
 472         return RTE_VHOST_MSG_RESULT_OK;
 473 }
 474
 475 /*
 476  * Reallocate virtio_dev, vhost_virtqueue and related data structures to
 477  * make them on the same numa node as the memory of vring descriptor.
 478  */
 479 #ifdef RTE_LIBRTE_VHOST_NUMA
 480 static struct virtio_net*
 481 numa_realloc(struct virtio_net *dev, int index)
 482 {
 483         int node, dev_node;
 484         struct virtio_net *old_dev;
 485         struct vhost_virtqueue *vq;
 486         struct batch_copy_elem *bce;
 487         struct guest_page *gp;
 488         struct rte_vhost_memory *mem;
 489         size_t mem_size;
 490         int ret;
 491
 492         old_dev = dev;
 493         vq = dev->virtqueue[index];
 494
 495         /*
 496          * If VQ is ready, it is too late to reallocate, it certainly already
 497          * happened anyway on VHOST_USER_SET_VRING_ADRR.
 498          */
 499         if (vq->ready)
 500                 return dev;
 501
 502         ret = get_mempolicy(&node, NULL, 0, vq->desc, MPOL_F_NODE | MPOL_F_ADDR);
 503         if (ret) {
 504                 VHOST_LOG_CONFIG(ERR, "Unable to get virtqueue %d numa information.\n", index);
 505                 return dev;
 506         }
 507
 508         if (node == vq->numa_node)
 509                 goto out_dev_realloc;
 510
 511         vq = rte_realloc_socket(vq, sizeof(*vq), 0, node);
 512         if (!vq) {
 513                 VHOST_LOG_CONFIG(ERR, "Failed to realloc virtqueue %d on node %d\n",
 514                                 index, node);
 515                 return dev;
 516         }
 517
 518         if (vq != dev->virtqueue[index]) {
 519                 VHOST_LOG_CONFIG(INFO, "reallocated virtqueue on node %d\n", node);
 520                 dev->virtqueue[index] = vq;
 521                 vhost_user_iotlb_init(dev, index);
 522         }
 523
 524         if (vq_is_packed(dev)) {
 525                 struct vring_used_elem_packed *sup;
 526
 527                 sup = rte_realloc_socket(vq->shadow_used_packed, vq->size * sizeof(*sup),
 528                                 RTE_CACHE_LINE_SIZE, node);
 529                 if (!sup) {
 530                         VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow packed on node %d\n", node);
 531                         return dev;
 532                 }
 533                 vq->shadow_used_packed = sup;
 534         } else {
 535                 struct vring_used_elem *sus;
 536
 537                 sus = rte_realloc_socket(vq->shadow_used_split, vq->size * sizeof(*sus),
 538                                 RTE_CACHE_LINE_SIZE, node);
 539                 if (!sus) {
 540                         VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow split on node %d\n", node);
 541                         return dev;
 542                 }
 543                 vq->shadow_used_split = sus;
 544         }
 545
 546         bce = rte_realloc_socket(vq->batch_copy_elems, vq->size * sizeof(*bce),
 547                         RTE_CACHE_LINE_SIZE, node);
 548         if (!bce) {
 549                 VHOST_LOG_CONFIG(ERR, "Failed to realloc batch copy elem on node %d\n", node);
 550                 return dev;
 551         }
 552         vq->batch_copy_elems = bce;
 553
 554         if (vq->log_cache) {
 555                 struct log_cache_entry *lc;
 556
 557                 lc = rte_realloc_socket(vq->log_cache, sizeof(*lc) * VHOST_LOG_CACHE_NR, 0, node);
 558                 if (!lc) {
 559                         VHOST_LOG_CONFIG(ERR, "Failed to realloc log cache on node %d\n", node);
 560                         return dev;
 561                 }
 562                 vq->log_cache = lc;
 563         }
 564
 565         vq->numa_node = node;
 566
 567 out_dev_realloc:
 568
 569         if (dev->flags & VIRTIO_DEV_RUNNING)
 570                 return dev;
 571
 572         ret = get_mempolicy(&dev_node, NULL, 0, dev, MPOL_F_NODE | MPOL_F_ADDR);
 573         if (ret) {
 574                 VHOST_LOG_CONFIG(ERR, "Unable to get Virtio dev %d numa information.\n", dev->vid);
 575                 return dev;
 576         }
 577
 578         if (dev_node == node)
 579                 return dev;
 580
 581         dev = rte_realloc_socket(old_dev, sizeof(*dev), 0, node);
 582         if (!dev) {
 583                 VHOST_LOG_CONFIG(ERR, "Failed to realloc dev on node %d\n", node);
 584                 return old_dev;
 585         }
 586
 587         VHOST_LOG_CONFIG(INFO, "reallocated device on node %d\n", node);
 588         vhost_devices[dev->vid] = dev;
 589
 590         mem_size = sizeof(struct rte_vhost_memory) +
 591                 sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
 592         mem = rte_realloc_socket(dev->mem, mem_size, 0, node);
 593         if (!mem) {
 594                 VHOST_LOG_CONFIG(ERR, "Failed to realloc mem table on node %d\n", node);
 595                 return dev;
 596         }
 597         dev->mem = mem;
 598
 599         gp = rte_realloc_socket(dev->guest_pages, dev->max_guest_pages * sizeof(*gp),
 600                         RTE_CACHE_LINE_SIZE, node);
 601         if (!gp) {
 602                 VHOST_LOG_CONFIG(ERR, "Failed to realloc guest pages on node %d\n", node);
 603                 return dev;
 604         }
 605         dev->guest_pages = gp;
 606
 607         return dev;
 608 }
 609 #else
 610 static struct virtio_net*
 611 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 612 {
 613         return dev;
 614 }
 615 #endif
 616
 617 /* Converts QEMU virtual address to Vhost virtual address. */
 618 static uint64_t
 619 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 620 {
 621         struct rte_vhost_mem_region *r;
 622         uint32_t i;
 623
 624         if (unlikely(!dev || !dev->mem))
 625                 goto out_error;
 626
 627         /* Find the region where the address lives. */
 628         for (i = 0; i < dev->mem->nregions; i++) {
 629                 r = &dev->mem->regions[i];
 630
 631                 if (qva >= r->guest_user_addr &&
 632                     qva <  r->guest_user_addr + r->size) {
 633
 634                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 635                                 *len = r->guest_user_addr + r->size - qva;
 636
 637                         return qva - r->guest_user_addr +
 638                                r->host_user_addr;
 639                 }
 640         }
 641 out_error:
 642         *len = 0;
 643
 644         return 0;
 645 }
 646
 647
 648 /*
 649  * Converts ring address to Vhost virtual address.
 650  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 651  * else it is a QEMU virtual address.
 652  */
 653 static uint64_t
 654 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 655                 uint64_t ra, uint64_t *size)
 656 {
 657         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 658                 uint64_t vva;
 659
 660                 vhost_user_iotlb_rd_lock(vq);
 661                 vva = vhost_iova_to_vva(dev, vq, ra,
 662                                         size, VHOST_ACCESS_RW);
 663                 vhost_user_iotlb_rd_unlock(vq);
 664
 665                 return vva;
 666         }
 667
 668         return qva_to_vva(dev, ra, size);
 669 }
 670
 671 static uint64_t
 672 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 673 {
 674         uint64_t log_gpa;
 675
 676         vhost_user_iotlb_rd_lock(vq);
 677         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 678         vhost_user_iotlb_rd_unlock(vq);
 679
 680         return log_gpa;
 681 }
 682
 683 static struct virtio_net *
 684 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 685 {
 686         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 687         struct vhost_vring_addr *addr = &vq->ring_addrs;
 688         uint64_t len, expected_len;
 689
 690         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 691                 vq->log_guest_addr =
 692                         log_addr_to_gpa(dev, vq);
 693                 if (vq->log_guest_addr == 0) {
 694                         VHOST_LOG_CONFIG(DEBUG,
 695                                 "(%d) failed to map log_guest_addr.\n",
 696                                 dev->vid);
 697                         return dev;
 698                 }
 699         }
 700
 701         if (vq_is_packed(dev)) {
 702                 len = sizeof(struct vring_packed_desc) * vq->size;
 703                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 704                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 705                 if (vq->desc_packed == NULL ||
 706                                 len != sizeof(struct vring_packed_desc) *
 707                                 vq->size) {
 708                         VHOST_LOG_CONFIG(DEBUG,
 709                                 "(%d) failed to map desc_packed ring.\n",
 710                                 dev->vid);
 711                         return dev;
 712                 }
 713
 714                 dev = numa_realloc(dev, vq_index);
 715                 vq = dev->virtqueue[vq_index];
 716                 addr = &vq->ring_addrs;
 717
 718                 len = sizeof(struct vring_packed_desc_event);
 719                 vq->driver_event = (struct vring_packed_desc_event *)
 720                                         (uintptr_t)ring_addr_to_vva(dev,
 721                                         vq, addr->avail_user_addr, &len);
 722                 if (vq->driver_event == NULL ||
 723                                 len != sizeof(struct vring_packed_desc_event)) {
 724                         VHOST_LOG_CONFIG(DEBUG,
 725                                 "(%d) failed to find driver area address.\n",
 726                                 dev->vid);
 727                         return dev;
 728                 }
 729
 730                 len = sizeof(struct vring_packed_desc_event);
 731                 vq->device_event = (struct vring_packed_desc_event *)
 732                                         (uintptr_t)ring_addr_to_vva(dev,
 733                                         vq, addr->used_user_addr, &len);
 734                 if (vq->device_event == NULL ||
 735                                 len != sizeof(struct vring_packed_desc_event)) {
 736                         VHOST_LOG_CONFIG(DEBUG,
 737                                 "(%d) failed to find device area address.\n",
 738                                 dev->vid);
 739                         return dev;
 740                 }
 741
 742                 vq->access_ok = true;
 743                 return dev;
 744         }
 745
 746         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 747         if (vq->desc && vq->avail && vq->used)
 748                 return dev;
 749
 750         len = sizeof(struct vring_desc) * vq->size;
 751         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 752                         vq, addr->desc_user_addr, &len);
 753         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 754                 VHOST_LOG_CONFIG(DEBUG,
 755                         "(%d) failed to map desc ring.\n",
 756                         dev->vid);
 757                 return dev;
 758         }
 759
 760         dev = numa_realloc(dev, vq_index);
 761         vq = dev->virtqueue[vq_index];
 762         addr = &vq->ring_addrs;
 763
 764         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 765         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 766                 len += sizeof(uint16_t);
 767         expected_len = len;
 768         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 769                         vq, addr->avail_user_addr, &len);
 770         if (vq->avail == 0 || len != expected_len) {
 771                 VHOST_LOG_CONFIG(DEBUG,
 772                         "(%d) failed to map avail ring.\n",
 773                         dev->vid);
 774                 return dev;
 775         }
 776
 777         len = sizeof(struct vring_used) +
 778                 sizeof(struct vring_used_elem) * vq->size;
 779         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 780                 len += sizeof(uint16_t);
 781         expected_len = len;
 782         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 783                         vq, addr->used_user_addr, &len);
 784         if (vq->used == 0 || len != expected_len) {
 785                 VHOST_LOG_CONFIG(DEBUG,
 786                         "(%d) failed to map used ring.\n",
 787                         dev->vid);
 788                 return dev;
 789         }
 790
 791         if (vq->last_used_idx != vq->used->idx) {
 792                 VHOST_LOG_CONFIG(WARNING,
 793                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 794                         "some packets maybe resent for Tx and dropped for Rx\n",
 795                         vq->last_used_idx, vq->used->idx);
 796                 vq->last_used_idx  = vq->used->idx;
 797                 vq->last_avail_idx = vq->used->idx;
 798         }
 799
 800         vq->access_ok = true;
 801
 802         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 803                         dev->vid, vq->desc);
 804         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 805                         dev->vid, vq->avail);
 806         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 807                         dev->vid, vq->used);
 808         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 809                         dev->vid, vq->log_guest_addr);
 810
 811         return dev;
 812 }
 813
 814 /*
 815  * The virtio device sends us the desc, used and avail ring addresses.
 816  * This function then converts these to our address space.
 817  */
 818 static int
 819 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 820                         int main_fd __rte_unused)
 821 {
 822         struct virtio_net *dev = *pdev;
 823         struct vhost_virtqueue *vq;
 824         struct vhost_vring_addr *addr = &msg->payload.addr;
 825         bool access_ok;
 826
 827         if (validate_msg_fds(msg, 0) != 0)
 828                 return RTE_VHOST_MSG_RESULT_ERR;
 829
 830         if (dev->mem == NULL)
 831                 return RTE_VHOST_MSG_RESULT_ERR;
 832
 833         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 834         vq = dev->virtqueue[msg->payload.addr.index];
 835
 836         access_ok = vq->access_ok;
 837
 838         /*
 839          * Rings addresses should not be interpreted as long as the ring is not
 840          * started and enabled
 841          */
 842         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 843
 844         vring_invalidate(dev, vq);
 845
 846         if ((vq->enabled && (dev->features &
 847                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 848                         access_ok) {
 849                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 850                 if (!dev)
 851                         return RTE_VHOST_MSG_RESULT_ERR;
 852
 853                 *pdev = dev;
 854         }
 855
 856         return RTE_VHOST_MSG_RESULT_OK;
 857 }
 858
 859 /*
 860  * The virtio device sends us the available ring last used index.
 861  */
 862 static int
 863 vhost_user_set_vring_base(struct virtio_net **pdev,
 864                         struct VhostUserMsg *msg,
 865                         int main_fd __rte_unused)
 866 {
 867         struct virtio_net *dev = *pdev;
 868         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 869         uint64_t val = msg->payload.state.num;
 870
 871         if (validate_msg_fds(msg, 0) != 0)
 872                 return RTE_VHOST_MSG_RESULT_ERR;
 873
 874         if (vq_is_packed(dev)) {
 875                 /*
 876                  * Bit[0:14]: avail index
 877                  * Bit[15]: avail wrap counter
 878                  */
 879                 vq->last_avail_idx = val & 0x7fff;
 880                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 881                 /*
 882                  * Set used index to same value as available one, as
 883                  * their values should be the same since ring processing
 884                  * was stopped at get time.
 885                  */
 886                 vq->last_used_idx = vq->last_avail_idx;
 887                 vq->used_wrap_counter = vq->avail_wrap_counter;
 888         } else {
 889                 vq->last_used_idx = msg->payload.state.num;
 890                 vq->last_avail_idx = msg->payload.state.num;
 891         }
 892
 893         return RTE_VHOST_MSG_RESULT_OK;
 894 }
 895
 896 static int
 897 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 898                    uint64_t host_phys_addr, uint64_t size)
 899 {
 900         struct guest_page *page, *last_page;
 901         struct guest_page *old_pages;
 902
 903         if (dev->nr_guest_pages == dev->max_guest_pages) {
 904                 dev->max_guest_pages *= 2;
 905                 old_pages = dev->guest_pages;
 906                 dev->guest_pages = rte_realloc(dev->guest_pages,
 907                                         dev->max_guest_pages * sizeof(*page),
 908                                         RTE_CACHE_LINE_SIZE);
 909                 if (dev->guest_pages == NULL) {
 910                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 911                         rte_free(old_pages);
 912                         return -1;
 913                 }
 914         }
 915
 916         if (dev->nr_guest_pages > 0) {
 917                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 918                 /* merge if the two pages are continuous */
 919                 if (host_phys_addr == last_page->host_phys_addr +
 920                                       last_page->size) {
 921                         last_page->size += size;
 922                         return 0;
 923                 }
 924         }
 925
 926         page = &dev->guest_pages[dev->nr_guest_pages++];
 927         page->guest_phys_addr = guest_phys_addr;
 928         page->host_phys_addr  = host_phys_addr;
 929         page->size = size;
 930
 931         return 0;
 932 }
 933
 934 static int
 935 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 936                 uint64_t page_size)
 937 {
 938         uint64_t reg_size = reg->size;
 939         uint64_t host_user_addr  = reg->host_user_addr;
 940         uint64_t guest_phys_addr = reg->guest_phys_addr;
 941         uint64_t host_phys_addr;
 942         uint64_t size;
 943
 944         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 945         size = page_size - (guest_phys_addr & (page_size - 1));
 946         size = RTE_MIN(size, reg_size);
 947
 948         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 949                 return -1;
 950
 951         host_user_addr  += size;
 952         guest_phys_addr += size;
 953         reg_size -= size;
 954
 955         while (reg_size > 0) {
 956                 size = RTE_MIN(reg_size, page_size);
 957                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 958                                                   host_user_addr);
 959                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 960                                 size) < 0)
 961                         return -1;
 962
 963                 host_user_addr  += size;
 964                 guest_phys_addr += size;
 965                 reg_size -= size;
 966         }
 967
 968         /* sort guest page array if over binary search threshold */
 969         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 970                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 971                         sizeof(struct guest_page), guest_page_addrcmp);
 972         }
 973
 974         return 0;
 975 }
 976
 977 #ifdef RTE_LIBRTE_VHOST_DEBUG
 978 /* TODO: enable it only in debug mode? */
 979 static void
 980 dump_guest_pages(struct virtio_net *dev)
 981 {
 982         uint32_t i;
 983         struct guest_page *page;
 984
 985         for (i = 0; i < dev->nr_guest_pages; i++) {
 986                 page = &dev->guest_pages[i];
 987
 988                 VHOST_LOG_CONFIG(INFO,
 989                         "guest physical page region %u\n"
 990                         "\t guest_phys_addr: %" PRIx64 "\n"
 991                         "\t host_phys_addr : %" PRIx64 "\n"
 992                         "\t size           : %" PRIx64 "\n",
 993                         i,
 994                         page->guest_phys_addr,
 995                         page->host_phys_addr,
 996                         page->size);
 997         }
 998 }
 999 #else
1000 #define dump_guest_pages(dev)
1001 #endif
1002
1003 static bool
1004 vhost_memory_changed(struct VhostUserMemory *new,
1005                      struct rte_vhost_memory *old)
1006 {
1007         uint32_t i;
1008
1009         if (new->nregions != old->nregions)
1010                 return true;
1011
1012         for (i = 0; i < new->nregions; ++i) {
1013                 VhostUserMemoryRegion *new_r = &new->regions[i];
1014                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1015
1016                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1017                         return true;
1018                 if (new_r->memory_size != old_r->size)
1019                         return true;
1020                 if (new_r->userspace_addr != old_r->guest_user_addr)
1021                         return true;
1022         }
1023
1024         return false;
1025 }
1026
1027 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1028 static int
1029 vhost_user_postcopy_region_register(struct virtio_net *dev,
1030                 struct rte_vhost_mem_region *reg)
1031 {
1032         struct uffdio_register reg_struct;
1033
1034         /*
1035          * Let's register all the mmap'ed area to ensure
1036          * alignment on page boundary.
1037          */
1038         reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1039         reg_struct.range.len = reg->mmap_size;
1040         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1041
1042         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1043                                 &reg_struct)) {
1044                 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1045                                 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1046                                 (uint64_t)reg_struct.range.start,
1047                                 (uint64_t)reg_struct.range.start +
1048                                 (uint64_t)reg_struct.range.len - 1,
1049                                 dev->postcopy_ufd,
1050                                 strerror(errno));
1051                 return -1;
1052         }
1053
1054         VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1055                         (uint64_t)reg_struct.range.start,
1056                         (uint64_t)reg_struct.range.start +
1057                         (uint64_t)reg_struct.range.len - 1);
1058
1059         return 0;
1060 }
1061 #else
1062 static int
1063 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1064                 struct rte_vhost_mem_region *reg __rte_unused)
1065 {
1066         return -1;
1067 }
1068 #endif
1069
1070 static int
1071 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1072                 struct VhostUserMsg *msg)
1073 {
1074         struct VhostUserMemory *memory;
1075         struct rte_vhost_mem_region *reg;
1076         VhostUserMsg ack_msg;
1077         uint32_t i;
1078
1079         if (!dev->postcopy_listening)
1080                 return 0;
1081
1082         /*
1083          * We haven't a better way right now than sharing
1084          * DPDK's virtual address with Qemu, so that Qemu can
1085          * retrieve the region offset when handling userfaults.
1086          */
1087         memory = &msg->payload.memory;
1088         for (i = 0; i < memory->nregions; i++) {
1089                 reg = &dev->mem->regions[i];
1090                 memory->regions[i].userspace_addr = reg->host_user_addr;
1091         }
1092
1093         /* Send the addresses back to qemu */
1094         msg->fd_num = 0;
1095         send_vhost_reply(main_fd, msg);
1096
1097         /* Wait for qemu to acknolwedge it's got the addresses
1098          * we've got to wait before we're allowed to generate faults.
1099          */
1100         if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1101                 VHOST_LOG_CONFIG(ERR,
1102                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1103                 return -1;
1104         }
1105
1106         if (validate_msg_fds(&ack_msg, 0) != 0)
1107                 return -1;
1108
1109         if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1110                 VHOST_LOG_CONFIG(ERR,
1111                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1112                                 ack_msg.request.master);
1113                 return -1;
1114         }
1115
1116         /* Now userfault register and we can use the memory */
1117         for (i = 0; i < memory->nregions; i++) {
1118                 reg = &dev->mem->regions[i];
1119                 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1120                         return -1;
1121         }
1122
1123         return 0;
1124 }
1125
1126 static int
1127 vhost_user_mmap_region(struct virtio_net *dev,
1128                 struct rte_vhost_mem_region *region,
1129                 uint64_t mmap_offset)
1130 {
1131         void *mmap_addr;
1132         uint64_t mmap_size;
1133         uint64_t alignment;
1134         int populate;
1135
1136         /* Check for memory_size + mmap_offset overflow */
1137         if (mmap_offset >= -region->size) {
1138                 VHOST_LOG_CONFIG(ERR,
1139                                 "mmap_offset (%#"PRIx64") and memory_size "
1140                                 "(%#"PRIx64") overflow\n",
1141                                 mmap_offset, region->size);
1142                 return -1;
1143         }
1144
1145         mmap_size = region->size + mmap_offset;
1146
1147         /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1148          * argument aligned with hugepagesz at older longterm version Linux,
1149          * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1150          *
1151          * To avoid failure, make sure in caller to keep length aligned.
1152          */
1153         alignment = get_blk_size(region->fd);
1154         if (alignment == (uint64_t)-1) {
1155                 VHOST_LOG_CONFIG(ERR,
1156                                 "couldn't get hugepage size through fstat\n");
1157                 return -1;
1158         }
1159         mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1160         if (mmap_size == 0) {
1161                 /*
1162                  * It could happen if initial mmap_size + alignment overflows
1163                  * the sizeof uint64, which could happen if either mmap_size or
1164                  * alignment value is wrong.
1165                  *
1166                  * mmap() kernel implementation would return an error, but
1167                  * better catch it before and provide useful info in the logs.
1168                  */
1169                 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1170                                 "or alignment (0x%" PRIx64 ") is invalid\n",
1171                                 region->size + mmap_offset, alignment);
1172                 return -1;
1173         }
1174
1175         populate = dev->async_copy ? MAP_POPULATE : 0;
1176         mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1177                         MAP_SHARED | populate, region->fd, 0);
1178
1179         if (mmap_addr == MAP_FAILED) {
1180                 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1181                 return -1;
1182         }
1183
1184         region->mmap_addr = mmap_addr;
1185         region->mmap_size = mmap_size;
1186         region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1187
1188         if (dev->async_copy)
1189                 if (add_guest_pages(dev, region, alignment) < 0) {
1190                         VHOST_LOG_CONFIG(ERR,
1191                                         "adding guest pages to region failed.\n");
1192                         return -1;
1193                 }
1194
1195         VHOST_LOG_CONFIG(INFO,
1196                         "guest memory region size: 0x%" PRIx64 "\n"
1197                         "\t guest physical addr: 0x%" PRIx64 "\n"
1198                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1199                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1200                         "\t mmap addr : 0x%" PRIx64 "\n"
1201                         "\t mmap size : 0x%" PRIx64 "\n"
1202                         "\t mmap align: 0x%" PRIx64 "\n"
1203                         "\t mmap off  : 0x%" PRIx64 "\n",
1204                         region->size,
1205                         region->guest_phys_addr,
1206                         region->guest_user_addr,
1207                         region->host_user_addr,
1208                         (uint64_t)(uintptr_t)mmap_addr,
1209                         mmap_size,
1210                         alignment,
1211                         mmap_offset);
1212
1213         return 0;
1214 }
1215
1216 static int
1217 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1218                         int main_fd)
1219 {
1220         struct virtio_net *dev = *pdev;
1221         struct VhostUserMemory *memory = &msg->payload.memory;
1222         struct rte_vhost_mem_region *reg;
1223         int numa_node = SOCKET_ID_ANY;
1224         uint64_t mmap_offset;
1225         uint32_t i;
1226
1227         if (validate_msg_fds(msg, memory->nregions) != 0)
1228                 return RTE_VHOST_MSG_RESULT_ERR;
1229
1230         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1231                 VHOST_LOG_CONFIG(ERR,
1232                         "too many memory regions (%u)\n", memory->nregions);
1233                 goto close_msg_fds;
1234         }
1235
1236         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1237                 VHOST_LOG_CONFIG(INFO,
1238                         "(%d) memory regions not changed\n", dev->vid);
1239
1240                 close_msg_fds(msg);
1241
1242                 return RTE_VHOST_MSG_RESULT_OK;
1243         }
1244
1245         if (dev->mem) {
1246                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1247                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1248
1249                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1250                                 vdpa_dev->ops->dev_close(dev->vid);
1251                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1252                 }
1253                 free_mem_region(dev);
1254                 rte_free(dev->mem);
1255                 dev->mem = NULL;
1256         }
1257
1258         /* Flush IOTLB cache as previous HVAs are now invalid */
1259         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1260                 for (i = 0; i < dev->nr_vring; i++)
1261                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1262
1263         /*
1264          * If VQ 0 has already been allocated, try to allocate on the same
1265          * NUMA node. It can be reallocated later in numa_realloc().
1266          */
1267         if (dev->nr_vring > 0)
1268                 numa_node = dev->virtqueue[0]->numa_node;
1269
1270         dev->nr_guest_pages = 0;
1271         if (dev->guest_pages == NULL) {
1272                 dev->max_guest_pages = 8;
1273                 dev->guest_pages = rte_zmalloc_socket(NULL,
1274                                         dev->max_guest_pages *
1275                                         sizeof(struct guest_page),
1276                                         RTE_CACHE_LINE_SIZE,
1277                                         numa_node);
1278                 if (dev->guest_pages == NULL) {
1279                         VHOST_LOG_CONFIG(ERR,
1280                                 "(%d) failed to allocate memory "
1281                                 "for dev->guest_pages\n",
1282                                 dev->vid);
1283                         goto close_msg_fds;
1284                 }
1285         }
1286
1287         dev->mem = rte_zmalloc_socket("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1288                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0, numa_node);
1289         if (dev->mem == NULL) {
1290                 VHOST_LOG_CONFIG(ERR,
1291                         "(%d) failed to allocate memory for dev->mem\n",
1292                         dev->vid);
1293                 goto free_guest_pages;
1294         }
1295
1296         for (i = 0; i < memory->nregions; i++) {
1297                 reg = &dev->mem->regions[i];
1298
1299                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1300                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1301                 reg->size            = memory->regions[i].memory_size;
1302                 reg->fd              = msg->fds[i];
1303
1304                 /*
1305                  * Assign invalid file descriptor value to avoid double
1306                  * closing on error path.
1307                  */
1308                 msg->fds[i] = -1;
1309
1310                 mmap_offset = memory->regions[i].mmap_offset;
1311
1312                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1313                         VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1314                         goto free_mem_table;
1315                 }
1316
1317                 dev->mem->nregions++;
1318         }
1319
1320         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1321                 goto free_mem_table;
1322
1323         for (i = 0; i < dev->nr_vring; i++) {
1324                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1325
1326                 if (!vq)
1327                         continue;
1328
1329                 if (vq->desc || vq->avail || vq->used) {
1330                         /*
1331                          * If the memory table got updated, the ring addresses
1332                          * need to be translated again as virtual addresses have
1333                          * changed.
1334                          */
1335                         vring_invalidate(dev, vq);
1336
1337                         dev = translate_ring_addresses(dev, i);
1338                         if (!dev) {
1339                                 dev = *pdev;
1340                                 goto free_mem_table;
1341                         }
1342
1343                         *pdev = dev;
1344                 }
1345         }
1346
1347         dump_guest_pages(dev);
1348
1349         return RTE_VHOST_MSG_RESULT_OK;
1350
1351 free_mem_table:
1352         free_mem_region(dev);
1353         rte_free(dev->mem);
1354         dev->mem = NULL;
1355 free_guest_pages:
1356         rte_free(dev->guest_pages);
1357         dev->guest_pages = NULL;
1358 close_msg_fds:
1359         close_msg_fds(msg);
1360         return RTE_VHOST_MSG_RESULT_ERR;
1361 }
1362
1363 static bool
1364 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1365 {
1366         bool rings_ok;
1367
1368         if (!vq)
1369                 return false;
1370
1371         if (vq_is_packed(dev))
1372                 rings_ok = vq->desc_packed && vq->driver_event &&
1373                         vq->device_event;
1374         else
1375                 rings_ok = vq->desc && vq->avail && vq->used;
1376
1377         return rings_ok &&
1378                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1379                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1380                vq->enabled;
1381 }
1382
1383 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1384
1385 static int
1386 virtio_is_ready(struct virtio_net *dev)
1387 {
1388         struct vhost_virtqueue *vq;
1389         uint32_t i, nr_vring = dev->nr_vring;
1390
1391         if (dev->flags & VIRTIO_DEV_READY)
1392                 return 1;
1393
1394         if (!dev->nr_vring)
1395                 return 0;
1396
1397         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1398                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1399
1400                 if (dev->nr_vring < nr_vring)
1401                         return 0;
1402         }
1403
1404         for (i = 0; i < nr_vring; i++) {
1405                 vq = dev->virtqueue[i];
1406
1407                 if (!vq_is_ready(dev, vq))
1408                         return 0;
1409         }
1410
1411         /* If supported, ensure the frontend is really done with config */
1412         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1413                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1414                         return 0;
1415
1416         dev->flags |= VIRTIO_DEV_READY;
1417
1418         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1419                 VHOST_LOG_CONFIG(INFO,
1420                         "virtio is now ready for processing.\n");
1421         return 1;
1422 }
1423
1424 static void *
1425 inflight_mem_alloc(const char *name, size_t size, int *fd)
1426 {
1427         void *ptr;
1428         int mfd = -1;
1429         char fname[20] = "/tmp/memfd-XXXXXX";
1430
1431         *fd = -1;
1432 #ifdef MEMFD_SUPPORTED
1433         mfd = memfd_create(name, MFD_CLOEXEC);
1434 #else
1435         RTE_SET_USED(name);
1436 #endif
1437         if (mfd == -1) {
1438                 mfd = mkstemp(fname);
1439                 if (mfd == -1) {
1440                         VHOST_LOG_CONFIG(ERR,
1441                                 "failed to get inflight buffer fd\n");
1442                         return NULL;
1443                 }
1444
1445                 unlink(fname);
1446         }
1447
1448         if (ftruncate(mfd, size) == -1) {
1449                 VHOST_LOG_CONFIG(ERR,
1450                         "failed to alloc inflight buffer\n");
1451                 close(mfd);
1452                 return NULL;
1453         }
1454
1455         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1456         if (ptr == MAP_FAILED) {
1457                 VHOST_LOG_CONFIG(ERR,
1458                         "failed to mmap inflight buffer\n");
1459                 close(mfd);
1460                 return NULL;
1461         }
1462
1463         *fd = mfd;
1464         return ptr;
1465 }
1466
1467 static uint32_t
1468 get_pervq_shm_size_split(uint16_t queue_size)
1469 {
1470         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1471                                   queue_size + sizeof(uint64_t) +
1472                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1473 }
1474
1475 static uint32_t
1476 get_pervq_shm_size_packed(uint16_t queue_size)
1477 {
1478         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1479                                   * queue_size + sizeof(uint64_t) +
1480                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1481                                   INFLIGHT_ALIGNMENT);
1482 }
1483
1484 static int
1485 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1486                            VhostUserMsg *msg,
1487                            int main_fd __rte_unused)
1488 {
1489         struct rte_vhost_inflight_info_packed *inflight_packed;
1490         uint64_t pervq_inflight_size, mmap_size;
1491         uint16_t num_queues, queue_size;
1492         struct virtio_net *dev = *pdev;
1493         int fd, i, j;
1494         void *addr;
1495
1496         if (msg->size != sizeof(msg->payload.inflight)) {
1497                 VHOST_LOG_CONFIG(ERR,
1498                         "invalid get_inflight_fd message size is %d\n",
1499                         msg->size);
1500                 return RTE_VHOST_MSG_RESULT_ERR;
1501         }
1502
1503         if (dev->inflight_info == NULL) {
1504                 dev->inflight_info = calloc(1,
1505                                             sizeof(struct inflight_mem_info));
1506                 if (!dev->inflight_info) {
1507                         VHOST_LOG_CONFIG(ERR,
1508                                 "failed to alloc dev inflight area\n");
1509                         return RTE_VHOST_MSG_RESULT_ERR;
1510                 }
1511                 dev->inflight_info->fd = -1;
1512         }
1513
1514         num_queues = msg->payload.inflight.num_queues;
1515         queue_size = msg->payload.inflight.queue_size;
1516
1517         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1518                 msg->payload.inflight.num_queues);
1519         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1520                 msg->payload.inflight.queue_size);
1521
1522         if (vq_is_packed(dev))
1523                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1524         else
1525                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1526
1527         mmap_size = num_queues * pervq_inflight_size;
1528         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1529         if (!addr) {
1530                 VHOST_LOG_CONFIG(ERR,
1531                         "failed to alloc vhost inflight area\n");
1532                         msg->payload.inflight.mmap_size = 0;
1533                 return RTE_VHOST_MSG_RESULT_ERR;
1534         }
1535         memset(addr, 0, mmap_size);
1536
1537         if (dev->inflight_info->addr) {
1538                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1539                 dev->inflight_info->addr = NULL;
1540         }
1541
1542         if (dev->inflight_info->fd >= 0) {
1543                 close(dev->inflight_info->fd);
1544                 dev->inflight_info->fd = -1;
1545         }
1546
1547         dev->inflight_info->addr = addr;
1548         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1549         dev->inflight_info->fd = msg->fds[0] = fd;
1550         msg->payload.inflight.mmap_offset = 0;
1551         msg->fd_num = 1;
1552
1553         if (vq_is_packed(dev)) {
1554                 for (i = 0; i < num_queues; i++) {
1555                         inflight_packed =
1556                                 (struct rte_vhost_inflight_info_packed *)addr;
1557                         inflight_packed->used_wrap_counter = 1;
1558                         inflight_packed->old_used_wrap_counter = 1;
1559                         for (j = 0; j < queue_size; j++)
1560                                 inflight_packed->desc[j].next = j + 1;
1561                         addr = (void *)((char *)addr + pervq_inflight_size);
1562                 }
1563         }
1564
1565         VHOST_LOG_CONFIG(INFO,
1566                 "send inflight mmap_size: %"PRIu64"\n",
1567                 msg->payload.inflight.mmap_size);
1568         VHOST_LOG_CONFIG(INFO,
1569                 "send inflight mmap_offset: %"PRIu64"\n",
1570                 msg->payload.inflight.mmap_offset);
1571         VHOST_LOG_CONFIG(INFO,
1572                 "send inflight fd: %d\n", msg->fds[0]);
1573
1574         return RTE_VHOST_MSG_RESULT_REPLY;
1575 }
1576
1577 static int
1578 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1579                            int main_fd __rte_unused)
1580 {
1581         uint64_t mmap_size, mmap_offset;
1582         uint16_t num_queues, queue_size;
1583         struct virtio_net *dev = *pdev;
1584         uint32_t pervq_inflight_size;
1585         struct vhost_virtqueue *vq;
1586         void *addr;
1587         int fd, i;
1588
1589         fd = msg->fds[0];
1590         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1591                 VHOST_LOG_CONFIG(ERR,
1592                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1593                         msg->size, fd);
1594                 return RTE_VHOST_MSG_RESULT_ERR;
1595         }
1596
1597         mmap_size = msg->payload.inflight.mmap_size;
1598         mmap_offset = msg->payload.inflight.mmap_offset;
1599         num_queues = msg->payload.inflight.num_queues;
1600         queue_size = msg->payload.inflight.queue_size;
1601
1602         if (vq_is_packed(dev))
1603                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1604         else
1605                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1606
1607         VHOST_LOG_CONFIG(INFO,
1608                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1609         VHOST_LOG_CONFIG(INFO,
1610                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1611         VHOST_LOG_CONFIG(INFO,
1612                 "set_inflight_fd num_queues: %u\n", num_queues);
1613         VHOST_LOG_CONFIG(INFO,
1614                 "set_inflight_fd queue_size: %u\n", queue_size);
1615         VHOST_LOG_CONFIG(INFO,
1616                 "set_inflight_fd fd: %d\n", fd);
1617         VHOST_LOG_CONFIG(INFO,
1618                 "set_inflight_fd pervq_inflight_size: %d\n",
1619                 pervq_inflight_size);
1620
1621         if (!dev->inflight_info) {
1622                 dev->inflight_info = calloc(1,
1623                                             sizeof(struct inflight_mem_info));
1624                 if (dev->inflight_info == NULL) {
1625                         VHOST_LOG_CONFIG(ERR,
1626                                 "failed to alloc dev inflight area\n");
1627                         return RTE_VHOST_MSG_RESULT_ERR;
1628                 }
1629                 dev->inflight_info->fd = -1;
1630         }
1631
1632         if (dev->inflight_info->addr) {
1633                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1634                 dev->inflight_info->addr = NULL;
1635         }
1636
1637         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1638                     fd, mmap_offset);
1639         if (addr == MAP_FAILED) {
1640                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1641                 return RTE_VHOST_MSG_RESULT_ERR;
1642         }
1643
1644         if (dev->inflight_info->fd >= 0) {
1645                 close(dev->inflight_info->fd);
1646                 dev->inflight_info->fd = -1;
1647         }
1648
1649         dev->inflight_info->fd = fd;
1650         dev->inflight_info->addr = addr;
1651         dev->inflight_info->size = mmap_size;
1652
1653         for (i = 0; i < num_queues; i++) {
1654                 vq = dev->virtqueue[i];
1655                 if (!vq)
1656                         continue;
1657
1658                 if (vq_is_packed(dev)) {
1659                         vq->inflight_packed = addr;
1660                         vq->inflight_packed->desc_num = queue_size;
1661                 } else {
1662                         vq->inflight_split = addr;
1663                         vq->inflight_split->desc_num = queue_size;
1664                 }
1665                 addr = (void *)((char *)addr + pervq_inflight_size);
1666         }
1667
1668         return RTE_VHOST_MSG_RESULT_OK;
1669 }
1670
1671 static int
1672 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1673                         int main_fd __rte_unused)
1674 {
1675         struct virtio_net *dev = *pdev;
1676         struct vhost_vring_file file;
1677         struct vhost_virtqueue *vq;
1678         int expected_fds;
1679
1680         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1681         if (validate_msg_fds(msg, expected_fds) != 0)
1682                 return RTE_VHOST_MSG_RESULT_ERR;
1683
1684         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1685         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1686                 file.fd = VIRTIO_INVALID_EVENTFD;
1687         else
1688                 file.fd = msg->fds[0];
1689         VHOST_LOG_CONFIG(INFO,
1690                 "vring call idx:%d file:%d\n", file.index, file.fd);
1691
1692         vq = dev->virtqueue[file.index];
1693
1694         if (vq->ready) {
1695                 vq->ready = false;
1696                 vhost_user_notify_queue_state(dev, file.index, 0);
1697         }
1698
1699         if (vq->callfd >= 0)
1700                 close(vq->callfd);
1701
1702         vq->callfd = file.fd;
1703
1704         return RTE_VHOST_MSG_RESULT_OK;
1705 }
1706
1707 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1708                         struct VhostUserMsg *msg,
1709                         int main_fd __rte_unused)
1710 {
1711         int expected_fds;
1712
1713         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1714         if (validate_msg_fds(msg, expected_fds) != 0)
1715                 return RTE_VHOST_MSG_RESULT_ERR;
1716
1717         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1718                 close(msg->fds[0]);
1719         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1720
1721         return RTE_VHOST_MSG_RESULT_OK;
1722 }
1723
1724 static int
1725 resubmit_desc_compare(const void *a, const void *b)
1726 {
1727         const struct rte_vhost_resubmit_desc *desc0 = a;
1728         const struct rte_vhost_resubmit_desc *desc1 = b;
1729
1730         if (desc1->counter > desc0->counter)
1731                 return 1;
1732
1733         return -1;
1734 }
1735
1736 static int
1737 vhost_check_queue_inflights_split(struct virtio_net *dev,
1738                                   struct vhost_virtqueue *vq)
1739 {
1740         uint16_t i;
1741         uint16_t resubmit_num = 0, last_io, num;
1742         struct vring_used *used = vq->used;
1743         struct rte_vhost_resubmit_info *resubmit;
1744         struct rte_vhost_inflight_info_split *inflight_split;
1745
1746         if (!(dev->protocol_features &
1747             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1748                 return RTE_VHOST_MSG_RESULT_OK;
1749
1750         /* The frontend may still not support the inflight feature
1751          * although we negotiate the protocol feature.
1752          */
1753         if ((!vq->inflight_split))
1754                 return RTE_VHOST_MSG_RESULT_OK;
1755
1756         if (!vq->inflight_split->version) {
1757                 vq->inflight_split->version = INFLIGHT_VERSION;
1758                 return RTE_VHOST_MSG_RESULT_OK;
1759         }
1760
1761         if (vq->resubmit_inflight)
1762                 return RTE_VHOST_MSG_RESULT_OK;
1763
1764         inflight_split = vq->inflight_split;
1765         vq->global_counter = 0;
1766         last_io = inflight_split->last_inflight_io;
1767
1768         if (inflight_split->used_idx != used->idx) {
1769                 inflight_split->desc[last_io].inflight = 0;
1770                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1771                 inflight_split->used_idx = used->idx;
1772         }
1773
1774         for (i = 0; i < inflight_split->desc_num; i++) {
1775                 if (inflight_split->desc[i].inflight == 1)
1776                         resubmit_num++;
1777         }
1778
1779         vq->last_avail_idx += resubmit_num;
1780
1781         if (resubmit_num) {
1782                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1783                 if (!resubmit) {
1784                         VHOST_LOG_CONFIG(ERR,
1785                                 "failed to allocate memory for resubmit info.\n");
1786                         return RTE_VHOST_MSG_RESULT_ERR;
1787                 }
1788
1789                 resubmit->resubmit_list = calloc(resubmit_num,
1790                         sizeof(struct rte_vhost_resubmit_desc));
1791                 if (!resubmit->resubmit_list) {
1792                         VHOST_LOG_CONFIG(ERR,
1793                                 "failed to allocate memory for inflight desc.\n");
1794                         free(resubmit);
1795                         return RTE_VHOST_MSG_RESULT_ERR;
1796                 }
1797
1798                 num = 0;
1799                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1800                         if (vq->inflight_split->desc[i].inflight == 1) {
1801                                 resubmit->resubmit_list[num].index = i;
1802                                 resubmit->resubmit_list[num].counter =
1803                                         inflight_split->desc[i].counter;
1804                                 num++;
1805                         }
1806                 }
1807                 resubmit->resubmit_num = num;
1808
1809                 if (resubmit->resubmit_num > 1)
1810                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1811                               sizeof(struct rte_vhost_resubmit_desc),
1812                               resubmit_desc_compare);
1813
1814                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1815                 vq->resubmit_inflight = resubmit;
1816         }
1817
1818         return RTE_VHOST_MSG_RESULT_OK;
1819 }
1820
1821 static int
1822 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1823                                    struct vhost_virtqueue *vq)
1824 {
1825         uint16_t i;
1826         uint16_t resubmit_num = 0, old_used_idx, num;
1827         struct rte_vhost_resubmit_info *resubmit;
1828         struct rte_vhost_inflight_info_packed *inflight_packed;
1829
1830         if (!(dev->protocol_features &
1831             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1832                 return RTE_VHOST_MSG_RESULT_OK;
1833
1834         /* The frontend may still not support the inflight feature
1835          * although we negotiate the protocol feature.
1836          */
1837         if ((!vq->inflight_packed))
1838                 return RTE_VHOST_MSG_RESULT_OK;
1839
1840         if (!vq->inflight_packed->version) {
1841                 vq->inflight_packed->version = INFLIGHT_VERSION;
1842                 return RTE_VHOST_MSG_RESULT_OK;
1843         }
1844
1845         if (vq->resubmit_inflight)
1846                 return RTE_VHOST_MSG_RESULT_OK;
1847
1848         inflight_packed = vq->inflight_packed;
1849         vq->global_counter = 0;
1850         old_used_idx = inflight_packed->old_used_idx;
1851
1852         if (inflight_packed->used_idx != old_used_idx) {
1853                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1854                         inflight_packed->old_used_idx =
1855                                 inflight_packed->used_idx;
1856                         inflight_packed->old_used_wrap_counter =
1857                                 inflight_packed->used_wrap_counter;
1858                         inflight_packed->old_free_head =
1859                                 inflight_packed->free_head;
1860                 } else {
1861                         inflight_packed->used_idx =
1862                                 inflight_packed->old_used_idx;
1863                         inflight_packed->used_wrap_counter =
1864                                 inflight_packed->old_used_wrap_counter;
1865                         inflight_packed->free_head =
1866                                 inflight_packed->old_free_head;
1867                 }
1868         }
1869
1870         for (i = 0; i < inflight_packed->desc_num; i++) {
1871                 if (inflight_packed->desc[i].inflight == 1)
1872                         resubmit_num++;
1873         }
1874
1875         if (resubmit_num) {
1876                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1877                 if (resubmit == NULL) {
1878                         VHOST_LOG_CONFIG(ERR,
1879                                 "failed to allocate memory for resubmit info.\n");
1880                         return RTE_VHOST_MSG_RESULT_ERR;
1881                 }
1882
1883                 resubmit->resubmit_list = calloc(resubmit_num,
1884                         sizeof(struct rte_vhost_resubmit_desc));
1885                 if (resubmit->resubmit_list == NULL) {
1886                         VHOST_LOG_CONFIG(ERR,
1887                                 "failed to allocate memory for resubmit desc.\n");
1888                         free(resubmit);
1889                         return RTE_VHOST_MSG_RESULT_ERR;
1890                 }
1891
1892                 num = 0;
1893                 for (i = 0; i < inflight_packed->desc_num; i++) {
1894                         if (vq->inflight_packed->desc[i].inflight == 1) {
1895                                 resubmit->resubmit_list[num].index = i;
1896                                 resubmit->resubmit_list[num].counter =
1897                                         inflight_packed->desc[i].counter;
1898                                 num++;
1899                         }
1900                 }
1901                 resubmit->resubmit_num = num;
1902
1903                 if (resubmit->resubmit_num > 1)
1904                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1905                               sizeof(struct rte_vhost_resubmit_desc),
1906                               resubmit_desc_compare);
1907
1908                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1909                 vq->resubmit_inflight = resubmit;
1910         }
1911
1912         return RTE_VHOST_MSG_RESULT_OK;
1913 }
1914
1915 static int
1916 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1917                         int main_fd __rte_unused)
1918 {
1919         struct virtio_net *dev = *pdev;
1920         struct vhost_vring_file file;
1921         struct vhost_virtqueue *vq;
1922         int expected_fds;
1923
1924         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1925         if (validate_msg_fds(msg, expected_fds) != 0)
1926                 return RTE_VHOST_MSG_RESULT_ERR;
1927
1928         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1929         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1930                 file.fd = VIRTIO_INVALID_EVENTFD;
1931         else
1932                 file.fd = msg->fds[0];
1933         VHOST_LOG_CONFIG(INFO,
1934                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1935
1936         /* Interpret ring addresses only when ring is started. */
1937         dev = translate_ring_addresses(dev, file.index);
1938         if (!dev) {
1939                 if (file.fd != VIRTIO_INVALID_EVENTFD)
1940                         close(file.fd);
1941
1942                 return RTE_VHOST_MSG_RESULT_ERR;
1943         }
1944
1945         *pdev = dev;
1946
1947         vq = dev->virtqueue[file.index];
1948
1949         /*
1950          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1951          * the ring starts already enabled. Otherwise, it is enabled via
1952          * the SET_VRING_ENABLE message.
1953          */
1954         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1955                 vq->enabled = true;
1956         }
1957
1958         if (vq->ready) {
1959                 vq->ready = false;
1960                 vhost_user_notify_queue_state(dev, file.index, 0);
1961         }
1962
1963         if (vq->kickfd >= 0)
1964                 close(vq->kickfd);
1965         vq->kickfd = file.fd;
1966
1967         if (vq_is_packed(dev)) {
1968                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1969                         VHOST_LOG_CONFIG(ERR,
1970                                 "failed to inflights for vq: %d\n", file.index);
1971                         return RTE_VHOST_MSG_RESULT_ERR;
1972                 }
1973         } else {
1974                 if (vhost_check_queue_inflights_split(dev, vq)) {
1975                         VHOST_LOG_CONFIG(ERR,
1976                                 "failed to inflights for vq: %d\n", file.index);
1977                         return RTE_VHOST_MSG_RESULT_ERR;
1978                 }
1979         }
1980
1981         return RTE_VHOST_MSG_RESULT_OK;
1982 }
1983
1984 /*
1985  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1986  */
1987 static int
1988 vhost_user_get_vring_base(struct virtio_net **pdev,
1989                         struct VhostUserMsg *msg,
1990                         int main_fd __rte_unused)
1991 {
1992         struct virtio_net *dev = *pdev;
1993         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1994         uint64_t val;
1995
1996         if (validate_msg_fds(msg, 0) != 0)
1997                 return RTE_VHOST_MSG_RESULT_ERR;
1998
1999         /* We have to stop the queue (virtio) if it is running. */
2000         vhost_destroy_device_notify(dev);
2001
2002         dev->flags &= ~VIRTIO_DEV_READY;
2003         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
2004
2005         /* Here we are safe to get the indexes */
2006         if (vq_is_packed(dev)) {
2007                 /*
2008                  * Bit[0:14]: avail index
2009                  * Bit[15]: avail wrap counter
2010                  */
2011                 val = vq->last_avail_idx & 0x7fff;
2012                 val |= vq->avail_wrap_counter << 15;
2013                 msg->payload.state.num = val;
2014         } else {
2015                 msg->payload.state.num = vq->last_avail_idx;
2016         }
2017
2018         VHOST_LOG_CONFIG(INFO,
2019                 "vring base idx:%d file:%d\n", msg->payload.state.index,
2020                 msg->payload.state.num);
2021         /*
2022          * Based on current qemu vhost-user implementation, this message is
2023          * sent and only sent in vhost_vring_stop.
2024          * TODO: cleanup the vring, it isn't usable since here.
2025          */
2026         if (vq->kickfd >= 0)
2027                 close(vq->kickfd);
2028
2029         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2030
2031         if (vq->callfd >= 0)
2032                 close(vq->callfd);
2033
2034         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2035
2036         vq->signalled_used_valid = false;
2037
2038         if (vq_is_packed(dev)) {
2039                 rte_free(vq->shadow_used_packed);
2040                 vq->shadow_used_packed = NULL;
2041         } else {
2042                 rte_free(vq->shadow_used_split);
2043                 vq->shadow_used_split = NULL;
2044         }
2045
2046         rte_free(vq->batch_copy_elems);
2047         vq->batch_copy_elems = NULL;
2048
2049         rte_free(vq->log_cache);
2050         vq->log_cache = NULL;
2051
2052         msg->size = sizeof(msg->payload.state);
2053         msg->fd_num = 0;
2054
2055         vring_invalidate(dev, vq);
2056
2057         return RTE_VHOST_MSG_RESULT_REPLY;
2058 }
2059
2060 /*
2061  * when virtio queues are ready to work, qemu will send us to
2062  * enable the virtio queue pair.
2063  */
2064 static int
2065 vhost_user_set_vring_enable(struct virtio_net **pdev,
2066                         struct VhostUserMsg *msg,
2067                         int main_fd __rte_unused)
2068 {
2069         struct virtio_net *dev = *pdev;
2070         bool enable = !!msg->payload.state.num;
2071         int index = (int)msg->payload.state.index;
2072
2073         if (validate_msg_fds(msg, 0) != 0)
2074                 return RTE_VHOST_MSG_RESULT_ERR;
2075
2076         VHOST_LOG_CONFIG(INFO,
2077                 "set queue enable: %d to qp idx: %d\n",
2078                 enable, index);
2079
2080         if (enable && dev->virtqueue[index]->async_registered) {
2081                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2082                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2083                         "async inflight packets must be completed first\n");
2084                         return RTE_VHOST_MSG_RESULT_ERR;
2085                 }
2086         }
2087
2088         dev->virtqueue[index]->enabled = enable;
2089
2090         return RTE_VHOST_MSG_RESULT_OK;
2091 }
2092
2093 static int
2094 vhost_user_get_protocol_features(struct virtio_net **pdev,
2095                         struct VhostUserMsg *msg,
2096                         int main_fd __rte_unused)
2097 {
2098         struct virtio_net *dev = *pdev;
2099         uint64_t features, protocol_features;
2100
2101         if (validate_msg_fds(msg, 0) != 0)
2102                 return RTE_VHOST_MSG_RESULT_ERR;
2103
2104         rte_vhost_driver_get_features(dev->ifname, &features);
2105         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2106
2107         msg->payload.u64 = protocol_features;
2108         msg->size = sizeof(msg->payload.u64);
2109         msg->fd_num = 0;
2110
2111         return RTE_VHOST_MSG_RESULT_REPLY;
2112 }
2113
2114 static int
2115 vhost_user_set_protocol_features(struct virtio_net **pdev,
2116                         struct VhostUserMsg *msg,
2117                         int main_fd __rte_unused)
2118 {
2119         struct virtio_net *dev = *pdev;
2120         uint64_t protocol_features = msg->payload.u64;
2121         uint64_t slave_protocol_features = 0;
2122
2123         if (validate_msg_fds(msg, 0) != 0)
2124                 return RTE_VHOST_MSG_RESULT_ERR;
2125
2126         rte_vhost_driver_get_protocol_features(dev->ifname,
2127                         &slave_protocol_features);
2128         if (protocol_features & ~slave_protocol_features) {
2129                 VHOST_LOG_CONFIG(ERR,
2130                         "(%d) received invalid protocol features.\n",
2131                         dev->vid);
2132                 return RTE_VHOST_MSG_RESULT_ERR;
2133         }
2134
2135         dev->protocol_features = protocol_features;
2136         VHOST_LOG_CONFIG(INFO,
2137                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2138                 dev->protocol_features);
2139
2140         return RTE_VHOST_MSG_RESULT_OK;
2141 }
2142
2143 static int
2144 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2145                         int main_fd __rte_unused)
2146 {
2147         struct virtio_net *dev = *pdev;
2148         int fd = msg->fds[0];
2149         uint64_t size, off;
2150         void *addr;
2151         uint32_t i;
2152
2153         if (validate_msg_fds(msg, 1) != 0)
2154                 return RTE_VHOST_MSG_RESULT_ERR;
2155
2156         if (fd < 0) {
2157                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2158                 return RTE_VHOST_MSG_RESULT_ERR;
2159         }
2160
2161         if (msg->size != sizeof(VhostUserLog)) {
2162                 VHOST_LOG_CONFIG(ERR,
2163                         "invalid log base msg size: %"PRId32" != %d\n",
2164                         msg->size, (int)sizeof(VhostUserLog));
2165                 goto close_msg_fds;
2166         }
2167
2168         size = msg->payload.log.mmap_size;
2169         off  = msg->payload.log.mmap_offset;
2170
2171         /* Check for mmap size and offset overflow. */
2172         if (off >= -size) {
2173                 VHOST_LOG_CONFIG(ERR,
2174                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2175                         off, size);
2176                 goto close_msg_fds;
2177         }
2178
2179         VHOST_LOG_CONFIG(INFO,
2180                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2181                 size, off);
2182
2183         /*
2184          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2185          * fail when offset is not page size aligned.
2186          */
2187         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2188         close(fd);
2189         if (addr == MAP_FAILED) {
2190                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2191                 return RTE_VHOST_MSG_RESULT_ERR;
2192         }
2193
2194         /*
2195          * Free previously mapped log memory on occasionally
2196          * multiple VHOST_USER_SET_LOG_BASE.
2197          */
2198         if (dev->log_addr) {
2199                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2200         }
2201         dev->log_addr = (uint64_t)(uintptr_t)addr;
2202         dev->log_base = dev->log_addr + off;
2203         dev->log_size = size;
2204
2205         for (i = 0; i < dev->nr_vring; i++) {
2206                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2207
2208                 rte_free(vq->log_cache);
2209                 vq->log_cache = NULL;
2210                 vq->log_cache_nb_elem = 0;
2211                 vq->log_cache = rte_malloc_socket("vq log cache",
2212                                 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2213                                 0, vq->numa_node);
2214                 /*
2215                  * If log cache alloc fail, don't fail migration, but no
2216                  * caching will be done, which will impact performance
2217                  */
2218                 if (!vq->log_cache)
2219                         VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2220         }
2221
2222         /*
2223          * The spec is not clear about it (yet), but QEMU doesn't expect
2224          * any payload in the reply.
2225          */
2226         msg->size = 0;
2227         msg->fd_num = 0;
2228
2229         return RTE_VHOST_MSG_RESULT_REPLY;
2230
2231 close_msg_fds:
2232         close_msg_fds(msg);
2233         return RTE_VHOST_MSG_RESULT_ERR;
2234 }
2235
2236 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2237                         struct VhostUserMsg *msg,
2238                         int main_fd __rte_unused)
2239 {
2240         if (validate_msg_fds(msg, 1) != 0)
2241                 return RTE_VHOST_MSG_RESULT_ERR;
2242
2243         close(msg->fds[0]);
2244         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2245
2246         return RTE_VHOST_MSG_RESULT_OK;
2247 }
2248
2249 /*
2250  * An rarp packet is constructed and broadcasted to notify switches about
2251  * the new location of the migrated VM, so that packets from outside will
2252  * not be lost after migration.
2253  *
2254  * However, we don't actually "send" a rarp packet here, instead, we set
2255  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2256  */
2257 static int
2258 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2259                         int main_fd __rte_unused)
2260 {
2261         struct virtio_net *dev = *pdev;
2262         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2263         struct rte_vdpa_device *vdpa_dev;
2264
2265         if (validate_msg_fds(msg, 0) != 0)
2266                 return RTE_VHOST_MSG_RESULT_ERR;
2267
2268         VHOST_LOG_CONFIG(DEBUG,
2269                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2270                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2271         memcpy(dev->mac.addr_bytes, mac, 6);
2272
2273         /*
2274          * Set the flag to inject a RARP broadcast packet at
2275          * rte_vhost_dequeue_burst().
2276          *
2277          * __ATOMIC_RELEASE ordering is for making sure the mac is
2278          * copied before the flag is set.
2279          */
2280         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2281         vdpa_dev = dev->vdpa_dev;
2282         if (vdpa_dev && vdpa_dev->ops->migration_done)
2283                 vdpa_dev->ops->migration_done(dev->vid);
2284
2285         return RTE_VHOST_MSG_RESULT_OK;
2286 }
2287
2288 static int
2289 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2290                         int main_fd __rte_unused)
2291 {
2292         struct virtio_net *dev = *pdev;
2293
2294         if (validate_msg_fds(msg, 0) != 0)
2295                 return RTE_VHOST_MSG_RESULT_ERR;
2296
2297         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2298                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2299                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2300                                 msg->payload.u64);
2301
2302                 return RTE_VHOST_MSG_RESULT_ERR;
2303         }
2304
2305         dev->mtu = msg->payload.u64;
2306
2307         return RTE_VHOST_MSG_RESULT_OK;
2308 }
2309
2310 static int
2311 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2312                         int main_fd __rte_unused)
2313 {
2314         struct virtio_net *dev = *pdev;
2315         int fd = msg->fds[0];
2316
2317         if (validate_msg_fds(msg, 1) != 0)
2318                 return RTE_VHOST_MSG_RESULT_ERR;
2319
2320         if (fd < 0) {
2321                 VHOST_LOG_CONFIG(ERR,
2322                                 "Invalid file descriptor for slave channel (%d)\n",
2323                                 fd);
2324                 return RTE_VHOST_MSG_RESULT_ERR;
2325         }
2326
2327         if (dev->slave_req_fd >= 0)
2328                 close(dev->slave_req_fd);
2329
2330         dev->slave_req_fd = fd;
2331
2332         return RTE_VHOST_MSG_RESULT_OK;
2333 }
2334
2335 static int
2336 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2337 {
2338         struct vhost_vring_addr *ra;
2339         uint64_t start, end, len;
2340
2341         start = imsg->iova;
2342         end = start + imsg->size;
2343
2344         ra = &vq->ring_addrs;
2345         len = sizeof(struct vring_desc) * vq->size;
2346         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2347                 return 1;
2348
2349         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2350         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2351                 return 1;
2352
2353         len = sizeof(struct vring_used) +
2354                sizeof(struct vring_used_elem) * vq->size;
2355         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2356                 return 1;
2357
2358         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2359                 len = sizeof(uint64_t);
2360                 if (ra->log_guest_addr < end &&
2361                     (ra->log_guest_addr + len) > start)
2362                         return 1;
2363         }
2364
2365         return 0;
2366 }
2367
2368 static int
2369 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2370 {
2371         struct vhost_vring_addr *ra;
2372         uint64_t start, end, len;
2373
2374         start = imsg->iova;
2375         end = start + imsg->size;
2376
2377         ra = &vq->ring_addrs;
2378         len = sizeof(struct vring_packed_desc) * vq->size;
2379         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2380                 return 1;
2381
2382         len = sizeof(struct vring_packed_desc_event);
2383         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2384                 return 1;
2385
2386         len = sizeof(struct vring_packed_desc_event);
2387         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2388                 return 1;
2389
2390         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2391                 len = sizeof(uint64_t);
2392                 if (ra->log_guest_addr < end &&
2393                     (ra->log_guest_addr + len) > start)
2394                         return 1;
2395         }
2396
2397         return 0;
2398 }
2399
2400 static int is_vring_iotlb(struct virtio_net *dev,
2401                           struct vhost_virtqueue *vq,
2402                           struct vhost_iotlb_msg *imsg)
2403 {
2404         if (vq_is_packed(dev))
2405                 return is_vring_iotlb_packed(vq, imsg);
2406         else
2407                 return is_vring_iotlb_split(vq, imsg);
2408 }
2409
2410 static int
2411 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2412                         int main_fd __rte_unused)
2413 {
2414         struct virtio_net *dev = *pdev;
2415         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2416         uint16_t i;
2417         uint64_t vva, len;
2418
2419         if (validate_msg_fds(msg, 0) != 0)
2420                 return RTE_VHOST_MSG_RESULT_ERR;
2421
2422         switch (imsg->type) {
2423         case VHOST_IOTLB_UPDATE:
2424                 len = imsg->size;
2425                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2426                 if (!vva)
2427                         return RTE_VHOST_MSG_RESULT_ERR;
2428
2429                 for (i = 0; i < dev->nr_vring; i++) {
2430                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2431
2432                         if (!vq)
2433                                 continue;
2434
2435                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2436                                         len, imsg->perm);
2437
2438                         if (is_vring_iotlb(dev, vq, imsg))
2439                                 *pdev = dev = translate_ring_addresses(dev, i);
2440                 }
2441                 break;
2442         case VHOST_IOTLB_INVALIDATE:
2443                 for (i = 0; i < dev->nr_vring; i++) {
2444                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2445
2446                         if (!vq)
2447                                 continue;
2448
2449                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2450                                         imsg->size);
2451
2452                         if (is_vring_iotlb(dev, vq, imsg))
2453                                 vring_invalidate(dev, vq);
2454                 }
2455                 break;
2456         default:
2457                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2458                                 imsg->type);
2459                 return RTE_VHOST_MSG_RESULT_ERR;
2460         }
2461
2462         return RTE_VHOST_MSG_RESULT_OK;
2463 }
2464
2465 static int
2466 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2467                         struct VhostUserMsg *msg,
2468                         int main_fd __rte_unused)
2469 {
2470         struct virtio_net *dev = *pdev;
2471 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2472         struct uffdio_api api_struct;
2473
2474         if (validate_msg_fds(msg, 0) != 0)
2475                 return RTE_VHOST_MSG_RESULT_ERR;
2476
2477         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2478
2479         if (dev->postcopy_ufd == -1) {
2480                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2481                         strerror(errno));
2482                 return RTE_VHOST_MSG_RESULT_ERR;
2483         }
2484         api_struct.api = UFFD_API;
2485         api_struct.features = 0;
2486         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2487                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2488                         strerror(errno));
2489                 close(dev->postcopy_ufd);
2490                 dev->postcopy_ufd = -1;
2491                 return RTE_VHOST_MSG_RESULT_ERR;
2492         }
2493         msg->fds[0] = dev->postcopy_ufd;
2494         msg->fd_num = 1;
2495
2496         return RTE_VHOST_MSG_RESULT_REPLY;
2497 #else
2498         dev->postcopy_ufd = -1;
2499         msg->fd_num = 0;
2500
2501         return RTE_VHOST_MSG_RESULT_ERR;
2502 #endif
2503 }
2504
2505 static int
2506 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2507                         struct VhostUserMsg *msg __rte_unused,
2508                         int main_fd __rte_unused)
2509 {
2510         struct virtio_net *dev = *pdev;
2511
2512         if (validate_msg_fds(msg, 0) != 0)
2513                 return RTE_VHOST_MSG_RESULT_ERR;
2514
2515         if (dev->mem && dev->mem->nregions) {
2516                 VHOST_LOG_CONFIG(ERR,
2517                         "Regions already registered at postcopy-listen\n");
2518                 return RTE_VHOST_MSG_RESULT_ERR;
2519         }
2520         dev->postcopy_listening = 1;
2521
2522         return RTE_VHOST_MSG_RESULT_OK;
2523 }
2524
2525 static int
2526 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2527                         int main_fd __rte_unused)
2528 {
2529         struct virtio_net *dev = *pdev;
2530
2531         if (validate_msg_fds(msg, 0) != 0)
2532                 return RTE_VHOST_MSG_RESULT_ERR;
2533
2534         dev->postcopy_listening = 0;
2535         if (dev->postcopy_ufd >= 0) {
2536                 close(dev->postcopy_ufd);
2537                 dev->postcopy_ufd = -1;
2538         }
2539
2540         msg->payload.u64 = 0;
2541         msg->size = sizeof(msg->payload.u64);
2542         msg->fd_num = 0;
2543
2544         return RTE_VHOST_MSG_RESULT_REPLY;
2545 }
2546
2547 static int
2548 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2549                       int main_fd __rte_unused)
2550 {
2551         struct virtio_net *dev = *pdev;
2552
2553         if (validate_msg_fds(msg, 0) != 0)
2554                 return RTE_VHOST_MSG_RESULT_ERR;
2555
2556         msg->payload.u64 = dev->status;
2557         msg->size = sizeof(msg->payload.u64);
2558         msg->fd_num = 0;
2559
2560         return RTE_VHOST_MSG_RESULT_REPLY;
2561 }
2562
2563 static int
2564 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2565                         int main_fd __rte_unused)
2566 {
2567         struct virtio_net *dev = *pdev;
2568
2569         if (validate_msg_fds(msg, 0) != 0)
2570                 return RTE_VHOST_MSG_RESULT_ERR;
2571
2572         /* As per Virtio specification, the device status is 8bits long */
2573         if (msg->payload.u64 > UINT8_MAX) {
2574                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2575                                 msg->payload.u64);
2576                 return RTE_VHOST_MSG_RESULT_ERR;
2577         }
2578
2579         dev->status = msg->payload.u64;
2580
2581         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2582             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2583                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2584                 /*
2585                  * Clear the bit to let the driver know about the feature
2586                  * negotiation failure
2587                  */
2588                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2589         }
2590
2591         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2592                         "\t-RESET: %u\n"
2593                         "\t-ACKNOWLEDGE: %u\n"
2594                         "\t-DRIVER: %u\n"
2595                         "\t-FEATURES_OK: %u\n"
2596                         "\t-DRIVER_OK: %u\n"
2597                         "\t-DEVICE_NEED_RESET: %u\n"
2598                         "\t-FAILED: %u\n",
2599                         dev->status,
2600                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2601                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2602                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2603                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2604                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2605                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2606                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2607
2608         return RTE_VHOST_MSG_RESULT_OK;
2609 }
2610
2611 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2612                                         struct VhostUserMsg *msg,
2613                                         int main_fd);
2614 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2615         [VHOST_USER_NONE] = NULL,
2616         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2617         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2618         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2619         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2620         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2621         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2622         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2623         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2624         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2625         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2626         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2627         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2628         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2629         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2630         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2631         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2632         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2633         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2634         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2635         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2636         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2637         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2638         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2639         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2640         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2641         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2642         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2643         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2644         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2645 };
2646
2647 /* return bytes# of read on success or negative val on failure. */
2648 static int
2649 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2650 {
2651         int ret;
2652
2653         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2654                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2655         if (ret <= 0) {
2656                 return ret;
2657         } else if (ret != VHOST_USER_HDR_SIZE) {
2658                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2659                 close_msg_fds(msg);
2660                 return -1;
2661         }
2662
2663         if (msg->size) {
2664                 if (msg->size > sizeof(msg->payload)) {
2665                         VHOST_LOG_CONFIG(ERR,
2666                                 "invalid msg size: %d\n", msg->size);
2667                         return -1;
2668                 }
2669                 ret = read(sockfd, &msg->payload, msg->size);
2670                 if (ret <= 0)
2671                         return ret;
2672                 if (ret != (int)msg->size) {
2673                         VHOST_LOG_CONFIG(ERR,
2674                                 "read control message failed\n");
2675                         return -1;
2676                 }
2677         }
2678
2679         return ret;
2680 }
2681
2682 static int
2683 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2684 {
2685         if (!msg)
2686                 return 0;
2687
2688         return send_fd_message(sockfd, (char *)msg,
2689                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2690 }
2691
2692 static int
2693 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2694 {
2695         if (!msg)
2696                 return 0;
2697
2698         msg->flags &= ~VHOST_USER_VERSION_MASK;
2699         msg->flags &= ~VHOST_USER_NEED_REPLY;
2700         msg->flags |= VHOST_USER_VERSION;
2701         msg->flags |= VHOST_USER_REPLY_MASK;
2702
2703         return send_vhost_message(sockfd, msg);
2704 }
2705
2706 static int
2707 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2708 {
2709         int ret;
2710
2711         if (msg->flags & VHOST_USER_NEED_REPLY)
2712                 rte_spinlock_lock(&dev->slave_req_lock);
2713
2714         ret = send_vhost_message(dev->slave_req_fd, msg);
2715         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2716                 rte_spinlock_unlock(&dev->slave_req_lock);
2717
2718         return ret;
2719 }
2720
2721 /*
2722  * Allocate a queue pair if it hasn't been allocated yet
2723  */
2724 static int
2725 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2726                         struct VhostUserMsg *msg)
2727 {
2728         uint32_t vring_idx;
2729
2730         switch (msg->request.master) {
2731         case VHOST_USER_SET_VRING_KICK:
2732         case VHOST_USER_SET_VRING_CALL:
2733         case VHOST_USER_SET_VRING_ERR:
2734                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2735                 break;
2736         case VHOST_USER_SET_VRING_NUM:
2737         case VHOST_USER_SET_VRING_BASE:
2738         case VHOST_USER_SET_VRING_ENABLE:
2739                 vring_idx = msg->payload.state.index;
2740                 break;
2741         case VHOST_USER_SET_VRING_ADDR:
2742                 vring_idx = msg->payload.addr.index;
2743                 break;
2744         default:
2745                 return 0;
2746         }
2747
2748         if (vring_idx >= VHOST_MAX_VRING) {
2749                 VHOST_LOG_CONFIG(ERR,
2750                         "invalid vring index: %u\n", vring_idx);
2751                 return -1;
2752         }
2753
2754         if (dev->virtqueue[vring_idx])
2755                 return 0;
2756
2757         return alloc_vring_queue(dev, vring_idx);
2758 }
2759
2760 static void
2761 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2762 {
2763         unsigned int i = 0;
2764         unsigned int vq_num = 0;
2765
2766         while (vq_num < dev->nr_vring) {
2767                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2768
2769                 if (vq) {
2770                         rte_spinlock_lock(&vq->access_lock);
2771                         vq_num++;
2772                 }
2773                 i++;
2774         }
2775 }
2776
2777 static void
2778 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2779 {
2780         unsigned int i = 0;
2781         unsigned int vq_num = 0;
2782
2783         while (vq_num < dev->nr_vring) {
2784                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2785
2786                 if (vq) {
2787                         rte_spinlock_unlock(&vq->access_lock);
2788                         vq_num++;
2789                 }
2790                 i++;
2791         }
2792 }
2793
2794 int
2795 vhost_user_msg_handler(int vid, int fd)
2796 {
2797         struct virtio_net *dev;
2798         struct VhostUserMsg msg;
2799         struct rte_vdpa_device *vdpa_dev;
2800         int ret;
2801         int unlock_required = 0;
2802         bool handled;
2803         int request;
2804         uint32_t i;
2805
2806         dev = get_device(vid);
2807         if (dev == NULL)
2808                 return -1;
2809
2810         if (!dev->notify_ops) {
2811                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2812                 if (!dev->notify_ops) {
2813                         VHOST_LOG_CONFIG(ERR,
2814                                 "failed to get callback ops for driver %s\n",
2815                                 dev->ifname);
2816                         return -1;
2817                 }
2818         }
2819
2820         ret = read_vhost_message(fd, &msg);
2821         if (ret <= 0) {
2822                 if (ret < 0)
2823                         VHOST_LOG_CONFIG(ERR,
2824                                 "vhost read message failed\n");
2825                 else
2826                         VHOST_LOG_CONFIG(INFO,
2827                                 "vhost peer closed\n");
2828
2829                 return -1;
2830         }
2831
2832         ret = 0;
2833         request = msg.request.master;
2834         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2835                         vhost_message_str[request]) {
2836                 if (request != VHOST_USER_IOTLB_MSG)
2837                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2838                                 vhost_message_str[request]);
2839                 else
2840                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2841                                 vhost_message_str[request]);
2842         } else {
2843                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2844         }
2845
2846         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2847         if (ret < 0) {
2848                 VHOST_LOG_CONFIG(ERR,
2849                         "failed to alloc queue\n");
2850                 return -1;
2851         }
2852
2853         /*
2854          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2855          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2856          * and device is destroyed. destroy_device waits for queues to be
2857          * inactive, so it is safe. Otherwise taking the access_lock
2858          * would cause a dead lock.
2859          */
2860         switch (request) {
2861         case VHOST_USER_SET_FEATURES:
2862         case VHOST_USER_SET_PROTOCOL_FEATURES:
2863         case VHOST_USER_SET_OWNER:
2864         case VHOST_USER_SET_MEM_TABLE:
2865         case VHOST_USER_SET_LOG_BASE:
2866         case VHOST_USER_SET_LOG_FD:
2867         case VHOST_USER_SET_VRING_NUM:
2868         case VHOST_USER_SET_VRING_ADDR:
2869         case VHOST_USER_SET_VRING_BASE:
2870         case VHOST_USER_SET_VRING_KICK:
2871         case VHOST_USER_SET_VRING_CALL:
2872         case VHOST_USER_SET_VRING_ERR:
2873         case VHOST_USER_SET_VRING_ENABLE:
2874         case VHOST_USER_SEND_RARP:
2875         case VHOST_USER_NET_SET_MTU:
2876         case VHOST_USER_SET_SLAVE_REQ_FD:
2877                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2878                         vhost_user_lock_all_queue_pairs(dev);
2879                         unlock_required = 1;
2880                 }
2881                 break;
2882         default:
2883                 break;
2884
2885         }
2886
2887         handled = false;
2888         if (dev->extern_ops.pre_msg_handle) {
2889                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2890                                 (void *)&msg);
2891                 switch (ret) {
2892                 case RTE_VHOST_MSG_RESULT_REPLY:
2893                         send_vhost_reply(fd, &msg);
2894                         /* Fall-through */
2895                 case RTE_VHOST_MSG_RESULT_ERR:
2896                 case RTE_VHOST_MSG_RESULT_OK:
2897                         handled = true;
2898                         goto skip_to_post_handle;
2899                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2900                 default:
2901                         break;
2902                 }
2903         }
2904
2905         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2906                 if (!vhost_message_handlers[request])
2907                         goto skip_to_post_handle;
2908                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2909
2910                 switch (ret) {
2911                 case RTE_VHOST_MSG_RESULT_ERR:
2912                         VHOST_LOG_CONFIG(ERR,
2913                                 "Processing %s failed.\n",
2914                                 vhost_message_str[request]);
2915                         handled = true;
2916                         break;
2917                 case RTE_VHOST_MSG_RESULT_OK:
2918                         VHOST_LOG_CONFIG(DEBUG,
2919                                 "Processing %s succeeded.\n",
2920                                 vhost_message_str[request]);
2921                         handled = true;
2922                         break;
2923                 case RTE_VHOST_MSG_RESULT_REPLY:
2924                         VHOST_LOG_CONFIG(DEBUG,
2925                                 "Processing %s succeeded and needs reply.\n",
2926                                 vhost_message_str[request]);
2927                         send_vhost_reply(fd, &msg);
2928                         handled = true;
2929                         break;
2930                 default:
2931                         break;
2932                 }
2933         }
2934
2935 skip_to_post_handle:
2936         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2937                         dev->extern_ops.post_msg_handle) {
2938                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2939                                 (void *)&msg);
2940                 switch (ret) {
2941                 case RTE_VHOST_MSG_RESULT_REPLY:
2942                         send_vhost_reply(fd, &msg);
2943                         /* Fall-through */
2944                 case RTE_VHOST_MSG_RESULT_ERR:
2945                 case RTE_VHOST_MSG_RESULT_OK:
2946                         handled = true;
2947                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2948                 default:
2949                         break;
2950                 }
2951         }
2952
2953         if (unlock_required)
2954                 vhost_user_unlock_all_queue_pairs(dev);
2955
2956         /* If message was not handled at this stage, treat it as an error */
2957         if (!handled) {
2958                 VHOST_LOG_CONFIG(ERR,
2959                         "vhost message (req: %d) was not handled.\n", request);
2960                 close_msg_fds(&msg);
2961                 ret = RTE_VHOST_MSG_RESULT_ERR;
2962         }
2963
2964         /*
2965          * If the request required a reply that was already sent,
2966          * this optional reply-ack won't be sent as the
2967          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2968          */
2969         if (msg.flags & VHOST_USER_NEED_REPLY) {
2970                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2971                 msg.size = sizeof(msg.payload.u64);
2972                 msg.fd_num = 0;
2973                 send_vhost_reply(fd, &msg);
2974         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2975                 VHOST_LOG_CONFIG(ERR,
2976                         "vhost message handling failed.\n");
2977                 return -1;
2978         }
2979
2980         for (i = 0; i < dev->nr_vring; i++) {
2981                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2982                 bool cur_ready = vq_is_ready(dev, vq);
2983
2984                 if (cur_ready != (vq && vq->ready)) {
2985                         vq->ready = cur_ready;
2986                         vhost_user_notify_queue_state(dev, i, cur_ready);
2987                 }
2988         }
2989
2990
2991         if (!virtio_is_ready(dev))
2992                 goto out;
2993
2994         /*
2995          * Virtio is now ready. If not done already, it is time
2996          * to notify the application it can process the rings and
2997          * configure the vDPA device if present.
2998          */
2999
3000         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
3001                 if (dev->notify_ops->new_device(dev->vid) == 0)
3002                         dev->flags |= VIRTIO_DEV_RUNNING;
3003         }
3004
3005         vdpa_dev = dev->vdpa_dev;
3006         if (!vdpa_dev)
3007                 goto out;
3008
3009         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3010                 if (vdpa_dev->ops->dev_conf(dev->vid))
3011                         VHOST_LOG_CONFIG(ERR,
3012                                          "Failed to configure vDPA device\n");
3013                 else
3014                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3015         }
3016
3017 out:
3018         return 0;
3019 }
3020
3021 static int process_slave_message_reply(struct virtio_net *dev,
3022                                        const struct VhostUserMsg *msg)
3023 {
3024         struct VhostUserMsg msg_reply;
3025         int ret;
3026
3027         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3028                 return 0;
3029
3030         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3031         if (ret <= 0) {
3032                 if (ret < 0)
3033                         VHOST_LOG_CONFIG(ERR,
3034                                 "vhost read slave message reply failed\n");
3035                 else
3036                         VHOST_LOG_CONFIG(INFO,
3037                                 "vhost peer closed\n");
3038                 ret = -1;
3039                 goto out;
3040         }
3041
3042         ret = 0;
3043         if (msg_reply.request.slave != msg->request.slave) {
3044                 VHOST_LOG_CONFIG(ERR,
3045                         "Received unexpected msg type (%u), expected %u\n",
3046                         msg_reply.request.slave, msg->request.slave);
3047                 ret = -1;
3048                 goto out;
3049         }
3050
3051         ret = msg_reply.payload.u64 ? -1 : 0;
3052
3053 out:
3054         rte_spinlock_unlock(&dev->slave_req_lock);
3055         return ret;
3056 }
3057
3058 int
3059 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3060 {
3061         int ret;
3062         struct VhostUserMsg msg = {
3063                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3064                 .flags = VHOST_USER_VERSION,
3065                 .size = sizeof(msg.payload.iotlb),
3066                 .payload.iotlb = {
3067                         .iova = iova,
3068                         .perm = perm,
3069                         .type = VHOST_IOTLB_MISS,
3070                 },
3071         };
3072
3073         ret = send_vhost_message(dev->slave_req_fd, &msg);
3074         if (ret < 0) {
3075                 VHOST_LOG_CONFIG(ERR,
3076                                 "Failed to send IOTLB miss message (%d)\n",
3077                                 ret);
3078                 return ret;
3079         }
3080
3081         return 0;
3082 }
3083
3084 static int
3085 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3086 {
3087         int ret;
3088         struct VhostUserMsg msg = {
3089                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3090                 .flags = VHOST_USER_VERSION,
3091                 .size = 0,
3092         };
3093
3094         if (need_reply)
3095                 msg.flags |= VHOST_USER_NEED_REPLY;
3096
3097         ret = send_vhost_slave_message(dev, &msg);
3098         if (ret < 0) {
3099                 VHOST_LOG_CONFIG(ERR,
3100                                 "Failed to send config change (%d)\n",
3101                                 ret);
3102                 return ret;
3103         }
3104
3105         return process_slave_message_reply(dev, &msg);
3106 }
3107
3108 int
3109 rte_vhost_slave_config_change(int vid, bool need_reply)
3110 {
3111         struct virtio_net *dev;
3112
3113         dev = get_device(vid);
3114         if (!dev)
3115                 return -ENODEV;
3116
3117         return vhost_user_slave_config_change(dev, need_reply);
3118 }
3119
3120 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3121                                                     int index, int fd,
3122                                                     uint64_t offset,
3123                                                     uint64_t size)
3124 {
3125         int ret;
3126         struct VhostUserMsg msg = {
3127                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3128                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3129                 .size = sizeof(msg.payload.area),
3130                 .payload.area = {
3131                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3132                         .size = size,
3133                         .offset = offset,
3134                 },
3135         };
3136
3137         if (fd < 0)
3138                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3139         else {
3140                 msg.fds[0] = fd;
3141                 msg.fd_num = 1;
3142         }
3143
3144         ret = send_vhost_slave_message(dev, &msg);
3145         if (ret < 0) {
3146                 VHOST_LOG_CONFIG(ERR,
3147                         "Failed to set host notifier (%d)\n", ret);
3148                 return ret;
3149         }
3150
3151         return process_slave_message_reply(dev, &msg);
3152 }
3153
3154 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3155 {
3156         struct virtio_net *dev;
3157         struct rte_vdpa_device *vdpa_dev;
3158         int vfio_device_fd, ret = 0;
3159         uint64_t offset, size;
3160         unsigned int i, q_start, q_last;
3161
3162         dev = get_device(vid);
3163         if (!dev)
3164                 return -ENODEV;
3165
3166         vdpa_dev = dev->vdpa_dev;
3167         if (vdpa_dev == NULL)
3168                 return -ENODEV;
3169
3170         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3171             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3172             !(dev->protocol_features &
3173                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3174             !(dev->protocol_features &
3175                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3176             !(dev->protocol_features &
3177                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3178                 return -ENOTSUP;
3179
3180         if (qid == RTE_VHOST_QUEUE_ALL) {
3181                 q_start = 0;
3182                 q_last = dev->nr_vring - 1;
3183         } else {
3184                 if (qid >= dev->nr_vring)
3185                         return -EINVAL;
3186                 q_start = qid;
3187                 q_last = qid;
3188         }
3189
3190         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3191         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3192
3193         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3194         if (vfio_device_fd < 0)
3195                 return -ENOTSUP;
3196
3197         if (enable) {
3198                 for (i = q_start; i <= q_last; i++) {
3199                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3200                                         &size) < 0) {
3201                                 ret = -ENOTSUP;
3202                                 goto disable;
3203                         }
3204
3205                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3206                                         vfio_device_fd, offset, size) < 0) {
3207                                 ret = -EFAULT;
3208                                 goto disable;
3209                         }
3210                 }
3211         } else {
3212 disable:
3213                 for (i = q_start; i <= q_last; i++) {
3214                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3215                                         0, 0);
3216                 }
3217         }
3218
3219         return ret;
3220 }