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(NULL,
 437                                 vq->size *
 438                                 sizeof(struct vring_used_elem_packed),
 439                                 RTE_CACHE_LINE_SIZE);
 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(NULL,
 451                                 vq->size * sizeof(struct vring_used_elem),
 452                                 RTE_CACHE_LINE_SIZE);
 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(NULL,
 464                                 vq->size * sizeof(struct batch_copy_elem),
 465                                 RTE_CACHE_LINE_SIZE);
 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 oldnode, newnode;
 484         struct virtio_net *old_dev;
 485         struct vhost_virtqueue *old_vq, *vq;
 486         struct vring_used_elem *new_shadow_used_split;
 487         struct vring_used_elem_packed *new_shadow_used_packed;
 488         struct batch_copy_elem *new_batch_copy_elems;
 489         int ret;
 490
 491         if (dev->flags & VIRTIO_DEV_RUNNING)
 492                 return dev;
 493
 494         old_dev = dev;
 495         vq = old_vq = dev->virtqueue[index];
 496
 497         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 498                             MPOL_F_NODE | MPOL_F_ADDR);
 499
 500         /* check if we need to reallocate vq */
 501         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 502                              MPOL_F_NODE | MPOL_F_ADDR);
 503         if (ret) {
 504                 VHOST_LOG_CONFIG(ERR,
 505                         "Unable to get vq numa information.\n");
 506                 return dev;
 507         }
 508         if (oldnode != newnode) {
 509                 VHOST_LOG_CONFIG(INFO,
 510                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 511                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 512                 if (!vq)
 513                         return dev;
 514
 515                 memcpy(vq, old_vq, sizeof(*vq));
 516
 517                 if (vq_is_packed(dev)) {
 518                         new_shadow_used_packed = rte_malloc_socket(NULL,
 519                                         vq->size *
 520                                         sizeof(struct vring_used_elem_packed),
 521                                         RTE_CACHE_LINE_SIZE,
 522                                         newnode);
 523                         if (new_shadow_used_packed) {
 524                                 rte_free(vq->shadow_used_packed);
 525                                 vq->shadow_used_packed = new_shadow_used_packed;
 526                         }
 527                 } else {
 528                         new_shadow_used_split = rte_malloc_socket(NULL,
 529                                         vq->size *
 530                                         sizeof(struct vring_used_elem),
 531                                         RTE_CACHE_LINE_SIZE,
 532                                         newnode);
 533                         if (new_shadow_used_split) {
 534                                 rte_free(vq->shadow_used_split);
 535                                 vq->shadow_used_split = new_shadow_used_split;
 536                         }
 537                 }
 538
 539                 new_batch_copy_elems = rte_malloc_socket(NULL,
 540                         vq->size * sizeof(struct batch_copy_elem),
 541                         RTE_CACHE_LINE_SIZE,
 542                         newnode);
 543                 if (new_batch_copy_elems) {
 544                         rte_free(vq->batch_copy_elems);
 545                         vq->batch_copy_elems = new_batch_copy_elems;
 546                 }
 547
 548                 rte_free(old_vq);
 549         }
 550
 551         /* check if we need to reallocate dev */
 552         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 553                             MPOL_F_NODE | MPOL_F_ADDR);
 554         if (ret) {
 555                 VHOST_LOG_CONFIG(ERR,
 556                         "Unable to get dev numa information.\n");
 557                 goto out;
 558         }
 559         if (oldnode != newnode) {
 560                 struct rte_vhost_memory *old_mem;
 561                 struct guest_page *old_gp;
 562                 ssize_t mem_size, gp_size;
 563
 564                 VHOST_LOG_CONFIG(INFO,
 565                         "reallocate dev from %d to %d node\n",
 566                         oldnode, newnode);
 567                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 568                 if (!dev) {
 569                         dev = old_dev;
 570                         goto out;
 571                 }
 572
 573                 memcpy(dev, old_dev, sizeof(*dev));
 574                 rte_free(old_dev);
 575
 576                 mem_size = sizeof(struct rte_vhost_memory) +
 577                         sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
 578                 old_mem = dev->mem;
 579                 dev->mem = rte_malloc_socket(NULL, mem_size, 0, newnode);
 580                 if (!dev->mem) {
 581                         dev->mem = old_mem;
 582                         goto out;
 583                 }
 584
 585                 memcpy(dev->mem, old_mem, mem_size);
 586                 rte_free(old_mem);
 587
 588                 gp_size = dev->max_guest_pages * sizeof(*dev->guest_pages);
 589                 old_gp = dev->guest_pages;
 590                 dev->guest_pages = rte_malloc_socket(NULL, gp_size, RTE_CACHE_LINE_SIZE, newnode);
 591                 if (!dev->guest_pages) {
 592                         dev->guest_pages = old_gp;
 593                         goto out;
 594                 }
 595
 596                 memcpy(dev->guest_pages, old_gp, gp_size);
 597                 rte_free(old_gp);
 598         }
 599
 600 out:
 601         dev->virtqueue[index] = vq;
 602         vhost_devices[dev->vid] = dev;
 603
 604         if (old_vq != vq)
 605                 vhost_user_iotlb_init(dev, index);
 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
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         dev->nr_guest_pages = 0;
1264         if (dev->guest_pages == NULL) {
1265                 dev->max_guest_pages = 8;
1266                 dev->guest_pages = rte_zmalloc(NULL,
1267                                         dev->max_guest_pages *
1268                                         sizeof(struct guest_page),
1269                                         RTE_CACHE_LINE_SIZE);
1270                 if (dev->guest_pages == NULL) {
1271                         VHOST_LOG_CONFIG(ERR,
1272                                 "(%d) failed to allocate memory "
1273                                 "for dev->guest_pages\n",
1274                                 dev->vid);
1275                         goto close_msg_fds;
1276                 }
1277         }
1278
1279         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1280                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1281         if (dev->mem == NULL) {
1282                 VHOST_LOG_CONFIG(ERR,
1283                         "(%d) failed to allocate memory for dev->mem\n",
1284                         dev->vid);
1285                 goto free_guest_pages;
1286         }
1287
1288         for (i = 0; i < memory->nregions; i++) {
1289                 reg = &dev->mem->regions[i];
1290
1291                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1292                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1293                 reg->size            = memory->regions[i].memory_size;
1294                 reg->fd              = msg->fds[i];
1295
1296                 /*
1297                  * Assign invalid file descriptor value to avoid double
1298                  * closing on error path.
1299                  */
1300                 msg->fds[i] = -1;
1301
1302                 mmap_offset = memory->regions[i].mmap_offset;
1303
1304                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1305                         VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1306                         goto free_mem_table;
1307                 }
1308
1309                 dev->mem->nregions++;
1310         }
1311
1312         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1313                 goto free_mem_table;
1314
1315         for (i = 0; i < dev->nr_vring; i++) {
1316                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1317
1318                 if (!vq)
1319                         continue;
1320
1321                 if (vq->desc || vq->avail || vq->used) {
1322                         /*
1323                          * If the memory table got updated, the ring addresses
1324                          * need to be translated again as virtual addresses have
1325                          * changed.
1326                          */
1327                         vring_invalidate(dev, vq);
1328
1329                         dev = translate_ring_addresses(dev, i);
1330                         if (!dev) {
1331                                 dev = *pdev;
1332                                 goto free_mem_table;
1333                         }
1334
1335                         *pdev = dev;
1336                 }
1337         }
1338
1339         dump_guest_pages(dev);
1340
1341         return RTE_VHOST_MSG_RESULT_OK;
1342
1343 free_mem_table:
1344         free_mem_region(dev);
1345         rte_free(dev->mem);
1346         dev->mem = NULL;
1347 free_guest_pages:
1348         rte_free(dev->guest_pages);
1349         dev->guest_pages = NULL;
1350 close_msg_fds:
1351         close_msg_fds(msg);
1352         return RTE_VHOST_MSG_RESULT_ERR;
1353 }
1354
1355 static bool
1356 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1357 {
1358         bool rings_ok;
1359
1360         if (!vq)
1361                 return false;
1362
1363         if (vq_is_packed(dev))
1364                 rings_ok = vq->desc_packed && vq->driver_event &&
1365                         vq->device_event;
1366         else
1367                 rings_ok = vq->desc && vq->avail && vq->used;
1368
1369         return rings_ok &&
1370                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1371                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1372                vq->enabled;
1373 }
1374
1375 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1376
1377 static int
1378 virtio_is_ready(struct virtio_net *dev)
1379 {
1380         struct vhost_virtqueue *vq;
1381         uint32_t i, nr_vring = dev->nr_vring;
1382
1383         if (dev->flags & VIRTIO_DEV_READY)
1384                 return 1;
1385
1386         if (!dev->nr_vring)
1387                 return 0;
1388
1389         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1390                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1391
1392                 if (dev->nr_vring < nr_vring)
1393                         return 0;
1394         }
1395
1396         for (i = 0; i < nr_vring; i++) {
1397                 vq = dev->virtqueue[i];
1398
1399                 if (!vq_is_ready(dev, vq))
1400                         return 0;
1401         }
1402
1403         /* If supported, ensure the frontend is really done with config */
1404         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1405                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1406                         return 0;
1407
1408         dev->flags |= VIRTIO_DEV_READY;
1409
1410         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1411                 VHOST_LOG_CONFIG(INFO,
1412                         "virtio is now ready for processing.\n");
1413         return 1;
1414 }
1415
1416 static void *
1417 inflight_mem_alloc(const char *name, size_t size, int *fd)
1418 {
1419         void *ptr;
1420         int mfd = -1;
1421         char fname[20] = "/tmp/memfd-XXXXXX";
1422
1423         *fd = -1;
1424 #ifdef MEMFD_SUPPORTED
1425         mfd = memfd_create(name, MFD_CLOEXEC);
1426 #else
1427         RTE_SET_USED(name);
1428 #endif
1429         if (mfd == -1) {
1430                 mfd = mkstemp(fname);
1431                 if (mfd == -1) {
1432                         VHOST_LOG_CONFIG(ERR,
1433                                 "failed to get inflight buffer fd\n");
1434                         return NULL;
1435                 }
1436
1437                 unlink(fname);
1438         }
1439
1440         if (ftruncate(mfd, size) == -1) {
1441                 VHOST_LOG_CONFIG(ERR,
1442                         "failed to alloc inflight buffer\n");
1443                 close(mfd);
1444                 return NULL;
1445         }
1446
1447         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1448         if (ptr == MAP_FAILED) {
1449                 VHOST_LOG_CONFIG(ERR,
1450                         "failed to mmap inflight buffer\n");
1451                 close(mfd);
1452                 return NULL;
1453         }
1454
1455         *fd = mfd;
1456         return ptr;
1457 }
1458
1459 static uint32_t
1460 get_pervq_shm_size_split(uint16_t queue_size)
1461 {
1462         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1463                                   queue_size + sizeof(uint64_t) +
1464                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1465 }
1466
1467 static uint32_t
1468 get_pervq_shm_size_packed(uint16_t queue_size)
1469 {
1470         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1471                                   * queue_size + sizeof(uint64_t) +
1472                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1473                                   INFLIGHT_ALIGNMENT);
1474 }
1475
1476 static int
1477 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1478                            VhostUserMsg *msg,
1479                            int main_fd __rte_unused)
1480 {
1481         struct rte_vhost_inflight_info_packed *inflight_packed;
1482         uint64_t pervq_inflight_size, mmap_size;
1483         uint16_t num_queues, queue_size;
1484         struct virtio_net *dev = *pdev;
1485         int fd, i, j;
1486         void *addr;
1487
1488         if (msg->size != sizeof(msg->payload.inflight)) {
1489                 VHOST_LOG_CONFIG(ERR,
1490                         "invalid get_inflight_fd message size is %d\n",
1491                         msg->size);
1492                 return RTE_VHOST_MSG_RESULT_ERR;
1493         }
1494
1495         if (dev->inflight_info == NULL) {
1496                 dev->inflight_info = calloc(1,
1497                                             sizeof(struct inflight_mem_info));
1498                 if (!dev->inflight_info) {
1499                         VHOST_LOG_CONFIG(ERR,
1500                                 "failed to alloc dev inflight area\n");
1501                         return RTE_VHOST_MSG_RESULT_ERR;
1502                 }
1503                 dev->inflight_info->fd = -1;
1504         }
1505
1506         num_queues = msg->payload.inflight.num_queues;
1507         queue_size = msg->payload.inflight.queue_size;
1508
1509         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1510                 msg->payload.inflight.num_queues);
1511         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1512                 msg->payload.inflight.queue_size);
1513
1514         if (vq_is_packed(dev))
1515                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1516         else
1517                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1518
1519         mmap_size = num_queues * pervq_inflight_size;
1520         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1521         if (!addr) {
1522                 VHOST_LOG_CONFIG(ERR,
1523                         "failed to alloc vhost inflight area\n");
1524                         msg->payload.inflight.mmap_size = 0;
1525                 return RTE_VHOST_MSG_RESULT_ERR;
1526         }
1527         memset(addr, 0, mmap_size);
1528
1529         if (dev->inflight_info->addr) {
1530                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1531                 dev->inflight_info->addr = NULL;
1532         }
1533
1534         if (dev->inflight_info->fd >= 0) {
1535                 close(dev->inflight_info->fd);
1536                 dev->inflight_info->fd = -1;
1537         }
1538
1539         dev->inflight_info->addr = addr;
1540         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1541         dev->inflight_info->fd = msg->fds[0] = fd;
1542         msg->payload.inflight.mmap_offset = 0;
1543         msg->fd_num = 1;
1544
1545         if (vq_is_packed(dev)) {
1546                 for (i = 0; i < num_queues; i++) {
1547                         inflight_packed =
1548                                 (struct rte_vhost_inflight_info_packed *)addr;
1549                         inflight_packed->used_wrap_counter = 1;
1550                         inflight_packed->old_used_wrap_counter = 1;
1551                         for (j = 0; j < queue_size; j++)
1552                                 inflight_packed->desc[j].next = j + 1;
1553                         addr = (void *)((char *)addr + pervq_inflight_size);
1554                 }
1555         }
1556
1557         VHOST_LOG_CONFIG(INFO,
1558                 "send inflight mmap_size: %"PRIu64"\n",
1559                 msg->payload.inflight.mmap_size);
1560         VHOST_LOG_CONFIG(INFO,
1561                 "send inflight mmap_offset: %"PRIu64"\n",
1562                 msg->payload.inflight.mmap_offset);
1563         VHOST_LOG_CONFIG(INFO,
1564                 "send inflight fd: %d\n", msg->fds[0]);
1565
1566         return RTE_VHOST_MSG_RESULT_REPLY;
1567 }
1568
1569 static int
1570 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1571                            int main_fd __rte_unused)
1572 {
1573         uint64_t mmap_size, mmap_offset;
1574         uint16_t num_queues, queue_size;
1575         struct virtio_net *dev = *pdev;
1576         uint32_t pervq_inflight_size;
1577         struct vhost_virtqueue *vq;
1578         void *addr;
1579         int fd, i;
1580
1581         fd = msg->fds[0];
1582         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1583                 VHOST_LOG_CONFIG(ERR,
1584                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1585                         msg->size, fd);
1586                 return RTE_VHOST_MSG_RESULT_ERR;
1587         }
1588
1589         mmap_size = msg->payload.inflight.mmap_size;
1590         mmap_offset = msg->payload.inflight.mmap_offset;
1591         num_queues = msg->payload.inflight.num_queues;
1592         queue_size = msg->payload.inflight.queue_size;
1593
1594         if (vq_is_packed(dev))
1595                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1596         else
1597                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1598
1599         VHOST_LOG_CONFIG(INFO,
1600                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1601         VHOST_LOG_CONFIG(INFO,
1602                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1603         VHOST_LOG_CONFIG(INFO,
1604                 "set_inflight_fd num_queues: %u\n", num_queues);
1605         VHOST_LOG_CONFIG(INFO,
1606                 "set_inflight_fd queue_size: %u\n", queue_size);
1607         VHOST_LOG_CONFIG(INFO,
1608                 "set_inflight_fd fd: %d\n", fd);
1609         VHOST_LOG_CONFIG(INFO,
1610                 "set_inflight_fd pervq_inflight_size: %d\n",
1611                 pervq_inflight_size);
1612
1613         if (!dev->inflight_info) {
1614                 dev->inflight_info = calloc(1,
1615                                             sizeof(struct inflight_mem_info));
1616                 if (dev->inflight_info == NULL) {
1617                         VHOST_LOG_CONFIG(ERR,
1618                                 "failed to alloc dev inflight area\n");
1619                         return RTE_VHOST_MSG_RESULT_ERR;
1620                 }
1621                 dev->inflight_info->fd = -1;
1622         }
1623
1624         if (dev->inflight_info->addr) {
1625                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1626                 dev->inflight_info->addr = NULL;
1627         }
1628
1629         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1630                     fd, mmap_offset);
1631         if (addr == MAP_FAILED) {
1632                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1633                 return RTE_VHOST_MSG_RESULT_ERR;
1634         }
1635
1636         if (dev->inflight_info->fd >= 0) {
1637                 close(dev->inflight_info->fd);
1638                 dev->inflight_info->fd = -1;
1639         }
1640
1641         dev->inflight_info->fd = fd;
1642         dev->inflight_info->addr = addr;
1643         dev->inflight_info->size = mmap_size;
1644
1645         for (i = 0; i < num_queues; i++) {
1646                 vq = dev->virtqueue[i];
1647                 if (!vq)
1648                         continue;
1649
1650                 if (vq_is_packed(dev)) {
1651                         vq->inflight_packed = addr;
1652                         vq->inflight_packed->desc_num = queue_size;
1653                 } else {
1654                         vq->inflight_split = addr;
1655                         vq->inflight_split->desc_num = queue_size;
1656                 }
1657                 addr = (void *)((char *)addr + pervq_inflight_size);
1658         }
1659
1660         return RTE_VHOST_MSG_RESULT_OK;
1661 }
1662
1663 static int
1664 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1665                         int main_fd __rte_unused)
1666 {
1667         struct virtio_net *dev = *pdev;
1668         struct vhost_vring_file file;
1669         struct vhost_virtqueue *vq;
1670         int expected_fds;
1671
1672         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1673         if (validate_msg_fds(msg, expected_fds) != 0)
1674                 return RTE_VHOST_MSG_RESULT_ERR;
1675
1676         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1677         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1678                 file.fd = VIRTIO_INVALID_EVENTFD;
1679         else
1680                 file.fd = msg->fds[0];
1681         VHOST_LOG_CONFIG(INFO,
1682                 "vring call idx:%d file:%d\n", file.index, file.fd);
1683
1684         vq = dev->virtqueue[file.index];
1685
1686         if (vq->ready) {
1687                 vq->ready = false;
1688                 vhost_user_notify_queue_state(dev, file.index, 0);
1689         }
1690
1691         if (vq->callfd >= 0)
1692                 close(vq->callfd);
1693
1694         vq->callfd = file.fd;
1695
1696         return RTE_VHOST_MSG_RESULT_OK;
1697 }
1698
1699 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1700                         struct VhostUserMsg *msg,
1701                         int main_fd __rte_unused)
1702 {
1703         int expected_fds;
1704
1705         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1706         if (validate_msg_fds(msg, expected_fds) != 0)
1707                 return RTE_VHOST_MSG_RESULT_ERR;
1708
1709         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1710                 close(msg->fds[0]);
1711         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1712
1713         return RTE_VHOST_MSG_RESULT_OK;
1714 }
1715
1716 static int
1717 resubmit_desc_compare(const void *a, const void *b)
1718 {
1719         const struct rte_vhost_resubmit_desc *desc0 = a;
1720         const struct rte_vhost_resubmit_desc *desc1 = b;
1721
1722         if (desc1->counter > desc0->counter)
1723                 return 1;
1724
1725         return -1;
1726 }
1727
1728 static int
1729 vhost_check_queue_inflights_split(struct virtio_net *dev,
1730                                   struct vhost_virtqueue *vq)
1731 {
1732         uint16_t i;
1733         uint16_t resubmit_num = 0, last_io, num;
1734         struct vring_used *used = vq->used;
1735         struct rte_vhost_resubmit_info *resubmit;
1736         struct rte_vhost_inflight_info_split *inflight_split;
1737
1738         if (!(dev->protocol_features &
1739             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1740                 return RTE_VHOST_MSG_RESULT_OK;
1741
1742         /* The frontend may still not support the inflight feature
1743          * although we negotiate the protocol feature.
1744          */
1745         if ((!vq->inflight_split))
1746                 return RTE_VHOST_MSG_RESULT_OK;
1747
1748         if (!vq->inflight_split->version) {
1749                 vq->inflight_split->version = INFLIGHT_VERSION;
1750                 return RTE_VHOST_MSG_RESULT_OK;
1751         }
1752
1753         if (vq->resubmit_inflight)
1754                 return RTE_VHOST_MSG_RESULT_OK;
1755
1756         inflight_split = vq->inflight_split;
1757         vq->global_counter = 0;
1758         last_io = inflight_split->last_inflight_io;
1759
1760         if (inflight_split->used_idx != used->idx) {
1761                 inflight_split->desc[last_io].inflight = 0;
1762                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1763                 inflight_split->used_idx = used->idx;
1764         }
1765
1766         for (i = 0; i < inflight_split->desc_num; i++) {
1767                 if (inflight_split->desc[i].inflight == 1)
1768                         resubmit_num++;
1769         }
1770
1771         vq->last_avail_idx += resubmit_num;
1772
1773         if (resubmit_num) {
1774                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1775                 if (!resubmit) {
1776                         VHOST_LOG_CONFIG(ERR,
1777                                 "failed to allocate memory for resubmit info.\n");
1778                         return RTE_VHOST_MSG_RESULT_ERR;
1779                 }
1780
1781                 resubmit->resubmit_list = calloc(resubmit_num,
1782                         sizeof(struct rte_vhost_resubmit_desc));
1783                 if (!resubmit->resubmit_list) {
1784                         VHOST_LOG_CONFIG(ERR,
1785                                 "failed to allocate memory for inflight desc.\n");
1786                         free(resubmit);
1787                         return RTE_VHOST_MSG_RESULT_ERR;
1788                 }
1789
1790                 num = 0;
1791                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1792                         if (vq->inflight_split->desc[i].inflight == 1) {
1793                                 resubmit->resubmit_list[num].index = i;
1794                                 resubmit->resubmit_list[num].counter =
1795                                         inflight_split->desc[i].counter;
1796                                 num++;
1797                         }
1798                 }
1799                 resubmit->resubmit_num = num;
1800
1801                 if (resubmit->resubmit_num > 1)
1802                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1803                               sizeof(struct rte_vhost_resubmit_desc),
1804                               resubmit_desc_compare);
1805
1806                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1807                 vq->resubmit_inflight = resubmit;
1808         }
1809
1810         return RTE_VHOST_MSG_RESULT_OK;
1811 }
1812
1813 static int
1814 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1815                                    struct vhost_virtqueue *vq)
1816 {
1817         uint16_t i;
1818         uint16_t resubmit_num = 0, old_used_idx, num;
1819         struct rte_vhost_resubmit_info *resubmit;
1820         struct rte_vhost_inflight_info_packed *inflight_packed;
1821
1822         if (!(dev->protocol_features &
1823             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1824                 return RTE_VHOST_MSG_RESULT_OK;
1825
1826         /* The frontend may still not support the inflight feature
1827          * although we negotiate the protocol feature.
1828          */
1829         if ((!vq->inflight_packed))
1830                 return RTE_VHOST_MSG_RESULT_OK;
1831
1832         if (!vq->inflight_packed->version) {
1833                 vq->inflight_packed->version = INFLIGHT_VERSION;
1834                 return RTE_VHOST_MSG_RESULT_OK;
1835         }
1836
1837         if (vq->resubmit_inflight)
1838                 return RTE_VHOST_MSG_RESULT_OK;
1839
1840         inflight_packed = vq->inflight_packed;
1841         vq->global_counter = 0;
1842         old_used_idx = inflight_packed->old_used_idx;
1843
1844         if (inflight_packed->used_idx != old_used_idx) {
1845                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1846                         inflight_packed->old_used_idx =
1847                                 inflight_packed->used_idx;
1848                         inflight_packed->old_used_wrap_counter =
1849                                 inflight_packed->used_wrap_counter;
1850                         inflight_packed->old_free_head =
1851                                 inflight_packed->free_head;
1852                 } else {
1853                         inflight_packed->used_idx =
1854                                 inflight_packed->old_used_idx;
1855                         inflight_packed->used_wrap_counter =
1856                                 inflight_packed->old_used_wrap_counter;
1857                         inflight_packed->free_head =
1858                                 inflight_packed->old_free_head;
1859                 }
1860         }
1861
1862         for (i = 0; i < inflight_packed->desc_num; i++) {
1863                 if (inflight_packed->desc[i].inflight == 1)
1864                         resubmit_num++;
1865         }
1866
1867         if (resubmit_num) {
1868                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1869                 if (resubmit == NULL) {
1870                         VHOST_LOG_CONFIG(ERR,
1871                                 "failed to allocate memory for resubmit info.\n");
1872                         return RTE_VHOST_MSG_RESULT_ERR;
1873                 }
1874
1875                 resubmit->resubmit_list = calloc(resubmit_num,
1876                         sizeof(struct rte_vhost_resubmit_desc));
1877                 if (resubmit->resubmit_list == NULL) {
1878                         VHOST_LOG_CONFIG(ERR,
1879                                 "failed to allocate memory for resubmit desc.\n");
1880                         free(resubmit);
1881                         return RTE_VHOST_MSG_RESULT_ERR;
1882                 }
1883
1884                 num = 0;
1885                 for (i = 0; i < inflight_packed->desc_num; i++) {
1886                         if (vq->inflight_packed->desc[i].inflight == 1) {
1887                                 resubmit->resubmit_list[num].index = i;
1888                                 resubmit->resubmit_list[num].counter =
1889                                         inflight_packed->desc[i].counter;
1890                                 num++;
1891                         }
1892                 }
1893                 resubmit->resubmit_num = num;
1894
1895                 if (resubmit->resubmit_num > 1)
1896                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1897                               sizeof(struct rte_vhost_resubmit_desc),
1898                               resubmit_desc_compare);
1899
1900                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1901                 vq->resubmit_inflight = resubmit;
1902         }
1903
1904         return RTE_VHOST_MSG_RESULT_OK;
1905 }
1906
1907 static int
1908 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1909                         int main_fd __rte_unused)
1910 {
1911         struct virtio_net *dev = *pdev;
1912         struct vhost_vring_file file;
1913         struct vhost_virtqueue *vq;
1914         int expected_fds;
1915
1916         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1917         if (validate_msg_fds(msg, expected_fds) != 0)
1918                 return RTE_VHOST_MSG_RESULT_ERR;
1919
1920         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1921         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1922                 file.fd = VIRTIO_INVALID_EVENTFD;
1923         else
1924                 file.fd = msg->fds[0];
1925         VHOST_LOG_CONFIG(INFO,
1926                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1927
1928         /* Interpret ring addresses only when ring is started. */
1929         dev = translate_ring_addresses(dev, file.index);
1930         if (!dev) {
1931                 if (file.fd != VIRTIO_INVALID_EVENTFD)
1932                         close(file.fd);
1933
1934                 return RTE_VHOST_MSG_RESULT_ERR;
1935         }
1936
1937         *pdev = dev;
1938
1939         vq = dev->virtqueue[file.index];
1940
1941         /*
1942          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1943          * the ring starts already enabled. Otherwise, it is enabled via
1944          * the SET_VRING_ENABLE message.
1945          */
1946         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1947                 vq->enabled = true;
1948         }
1949
1950         if (vq->ready) {
1951                 vq->ready = false;
1952                 vhost_user_notify_queue_state(dev, file.index, 0);
1953         }
1954
1955         if (vq->kickfd >= 0)
1956                 close(vq->kickfd);
1957         vq->kickfd = file.fd;
1958
1959         if (vq_is_packed(dev)) {
1960                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1961                         VHOST_LOG_CONFIG(ERR,
1962                                 "failed to inflights for vq: %d\n", file.index);
1963                         return RTE_VHOST_MSG_RESULT_ERR;
1964                 }
1965         } else {
1966                 if (vhost_check_queue_inflights_split(dev, vq)) {
1967                         VHOST_LOG_CONFIG(ERR,
1968                                 "failed to inflights for vq: %d\n", file.index);
1969                         return RTE_VHOST_MSG_RESULT_ERR;
1970                 }
1971         }
1972
1973         return RTE_VHOST_MSG_RESULT_OK;
1974 }
1975
1976 /*
1977  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1978  */
1979 static int
1980 vhost_user_get_vring_base(struct virtio_net **pdev,
1981                         struct VhostUserMsg *msg,
1982                         int main_fd __rte_unused)
1983 {
1984         struct virtio_net *dev = *pdev;
1985         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1986         uint64_t val;
1987
1988         if (validate_msg_fds(msg, 0) != 0)
1989                 return RTE_VHOST_MSG_RESULT_ERR;
1990
1991         /* We have to stop the queue (virtio) if it is running. */
1992         vhost_destroy_device_notify(dev);
1993
1994         dev->flags &= ~VIRTIO_DEV_READY;
1995         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1996
1997         /* Here we are safe to get the indexes */
1998         if (vq_is_packed(dev)) {
1999                 /*
2000                  * Bit[0:14]: avail index
2001                  * Bit[15]: avail wrap counter
2002                  */
2003                 val = vq->last_avail_idx & 0x7fff;
2004                 val |= vq->avail_wrap_counter << 15;
2005                 msg->payload.state.num = val;
2006         } else {
2007                 msg->payload.state.num = vq->last_avail_idx;
2008         }
2009
2010         VHOST_LOG_CONFIG(INFO,
2011                 "vring base idx:%d file:%d\n", msg->payload.state.index,
2012                 msg->payload.state.num);
2013         /*
2014          * Based on current qemu vhost-user implementation, this message is
2015          * sent and only sent in vhost_vring_stop.
2016          * TODO: cleanup the vring, it isn't usable since here.
2017          */
2018         if (vq->kickfd >= 0)
2019                 close(vq->kickfd);
2020
2021         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2022
2023         if (vq->callfd >= 0)
2024                 close(vq->callfd);
2025
2026         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2027
2028         vq->signalled_used_valid = false;
2029
2030         if (vq_is_packed(dev)) {
2031                 rte_free(vq->shadow_used_packed);
2032                 vq->shadow_used_packed = NULL;
2033         } else {
2034                 rte_free(vq->shadow_used_split);
2035                 vq->shadow_used_split = NULL;
2036         }
2037
2038         rte_free(vq->batch_copy_elems);
2039         vq->batch_copy_elems = NULL;
2040
2041         rte_free(vq->log_cache);
2042         vq->log_cache = NULL;
2043
2044         msg->size = sizeof(msg->payload.state);
2045         msg->fd_num = 0;
2046
2047         vring_invalidate(dev, vq);
2048
2049         return RTE_VHOST_MSG_RESULT_REPLY;
2050 }
2051
2052 /*
2053  * when virtio queues are ready to work, qemu will send us to
2054  * enable the virtio queue pair.
2055  */
2056 static int
2057 vhost_user_set_vring_enable(struct virtio_net **pdev,
2058                         struct VhostUserMsg *msg,
2059                         int main_fd __rte_unused)
2060 {
2061         struct virtio_net *dev = *pdev;
2062         bool enable = !!msg->payload.state.num;
2063         int index = (int)msg->payload.state.index;
2064
2065         if (validate_msg_fds(msg, 0) != 0)
2066                 return RTE_VHOST_MSG_RESULT_ERR;
2067
2068         VHOST_LOG_CONFIG(INFO,
2069                 "set queue enable: %d to qp idx: %d\n",
2070                 enable, index);
2071
2072         if (enable && dev->virtqueue[index]->async_registered) {
2073                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2074                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2075                         "async inflight packets must be completed first\n");
2076                         return RTE_VHOST_MSG_RESULT_ERR;
2077                 }
2078         }
2079
2080         dev->virtqueue[index]->enabled = enable;
2081
2082         return RTE_VHOST_MSG_RESULT_OK;
2083 }
2084
2085 static int
2086 vhost_user_get_protocol_features(struct virtio_net **pdev,
2087                         struct VhostUserMsg *msg,
2088                         int main_fd __rte_unused)
2089 {
2090         struct virtio_net *dev = *pdev;
2091         uint64_t features, protocol_features;
2092
2093         if (validate_msg_fds(msg, 0) != 0)
2094                 return RTE_VHOST_MSG_RESULT_ERR;
2095
2096         rte_vhost_driver_get_features(dev->ifname, &features);
2097         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2098
2099         msg->payload.u64 = protocol_features;
2100         msg->size = sizeof(msg->payload.u64);
2101         msg->fd_num = 0;
2102
2103         return RTE_VHOST_MSG_RESULT_REPLY;
2104 }
2105
2106 static int
2107 vhost_user_set_protocol_features(struct virtio_net **pdev,
2108                         struct VhostUserMsg *msg,
2109                         int main_fd __rte_unused)
2110 {
2111         struct virtio_net *dev = *pdev;
2112         uint64_t protocol_features = msg->payload.u64;
2113         uint64_t slave_protocol_features = 0;
2114
2115         if (validate_msg_fds(msg, 0) != 0)
2116                 return RTE_VHOST_MSG_RESULT_ERR;
2117
2118         rte_vhost_driver_get_protocol_features(dev->ifname,
2119                         &slave_protocol_features);
2120         if (protocol_features & ~slave_protocol_features) {
2121                 VHOST_LOG_CONFIG(ERR,
2122                         "(%d) received invalid protocol features.\n",
2123                         dev->vid);
2124                 return RTE_VHOST_MSG_RESULT_ERR;
2125         }
2126
2127         dev->protocol_features = protocol_features;
2128         VHOST_LOG_CONFIG(INFO,
2129                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2130                 dev->protocol_features);
2131
2132         return RTE_VHOST_MSG_RESULT_OK;
2133 }
2134
2135 static int
2136 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2137                         int main_fd __rte_unused)
2138 {
2139         struct virtio_net *dev = *pdev;
2140         int fd = msg->fds[0];
2141         uint64_t size, off;
2142         void *addr;
2143         uint32_t i;
2144
2145         if (validate_msg_fds(msg, 1) != 0)
2146                 return RTE_VHOST_MSG_RESULT_ERR;
2147
2148         if (fd < 0) {
2149                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2150                 return RTE_VHOST_MSG_RESULT_ERR;
2151         }
2152
2153         if (msg->size != sizeof(VhostUserLog)) {
2154                 VHOST_LOG_CONFIG(ERR,
2155                         "invalid log base msg size: %"PRId32" != %d\n",
2156                         msg->size, (int)sizeof(VhostUserLog));
2157                 goto close_msg_fds;
2158         }
2159
2160         size = msg->payload.log.mmap_size;
2161         off  = msg->payload.log.mmap_offset;
2162
2163         /* Check for mmap size and offset overflow. */
2164         if (off >= -size) {
2165                 VHOST_LOG_CONFIG(ERR,
2166                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2167                         off, size);
2168                 goto close_msg_fds;
2169         }
2170
2171         VHOST_LOG_CONFIG(INFO,
2172                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2173                 size, off);
2174
2175         /*
2176          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2177          * fail when offset is not page size aligned.
2178          */
2179         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2180         close(fd);
2181         if (addr == MAP_FAILED) {
2182                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2183                 return RTE_VHOST_MSG_RESULT_ERR;
2184         }
2185
2186         /*
2187          * Free previously mapped log memory on occasionally
2188          * multiple VHOST_USER_SET_LOG_BASE.
2189          */
2190         if (dev->log_addr) {
2191                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2192         }
2193         dev->log_addr = (uint64_t)(uintptr_t)addr;
2194         dev->log_base = dev->log_addr + off;
2195         dev->log_size = size;
2196
2197         for (i = 0; i < dev->nr_vring; i++) {
2198                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2199
2200                 rte_free(vq->log_cache);
2201                 vq->log_cache = NULL;
2202                 vq->log_cache_nb_elem = 0;
2203                 vq->log_cache = rte_zmalloc("vq log cache",
2204                                 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2205                                 0);
2206                 /*
2207                  * If log cache alloc fail, don't fail migration, but no
2208                  * caching will be done, which will impact performance
2209                  */
2210                 if (!vq->log_cache)
2211                         VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2212         }
2213
2214         /*
2215          * The spec is not clear about it (yet), but QEMU doesn't expect
2216          * any payload in the reply.
2217          */
2218         msg->size = 0;
2219         msg->fd_num = 0;
2220
2221         return RTE_VHOST_MSG_RESULT_REPLY;
2222
2223 close_msg_fds:
2224         close_msg_fds(msg);
2225         return RTE_VHOST_MSG_RESULT_ERR;
2226 }
2227
2228 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2229                         struct VhostUserMsg *msg,
2230                         int main_fd __rte_unused)
2231 {
2232         if (validate_msg_fds(msg, 1) != 0)
2233                 return RTE_VHOST_MSG_RESULT_ERR;
2234
2235         close(msg->fds[0]);
2236         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2237
2238         return RTE_VHOST_MSG_RESULT_OK;
2239 }
2240
2241 /*
2242  * An rarp packet is constructed and broadcasted to notify switches about
2243  * the new location of the migrated VM, so that packets from outside will
2244  * not be lost after migration.
2245  *
2246  * However, we don't actually "send" a rarp packet here, instead, we set
2247  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2248  */
2249 static int
2250 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2251                         int main_fd __rte_unused)
2252 {
2253         struct virtio_net *dev = *pdev;
2254         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2255         struct rte_vdpa_device *vdpa_dev;
2256
2257         if (validate_msg_fds(msg, 0) != 0)
2258                 return RTE_VHOST_MSG_RESULT_ERR;
2259
2260         VHOST_LOG_CONFIG(DEBUG,
2261                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2262                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2263         memcpy(dev->mac.addr_bytes, mac, 6);
2264
2265         /*
2266          * Set the flag to inject a RARP broadcast packet at
2267          * rte_vhost_dequeue_burst().
2268          *
2269          * __ATOMIC_RELEASE ordering is for making sure the mac is
2270          * copied before the flag is set.
2271          */
2272         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2273         vdpa_dev = dev->vdpa_dev;
2274         if (vdpa_dev && vdpa_dev->ops->migration_done)
2275                 vdpa_dev->ops->migration_done(dev->vid);
2276
2277         return RTE_VHOST_MSG_RESULT_OK;
2278 }
2279
2280 static int
2281 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2282                         int main_fd __rte_unused)
2283 {
2284         struct virtio_net *dev = *pdev;
2285
2286         if (validate_msg_fds(msg, 0) != 0)
2287                 return RTE_VHOST_MSG_RESULT_ERR;
2288
2289         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2290                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2291                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2292                                 msg->payload.u64);
2293
2294                 return RTE_VHOST_MSG_RESULT_ERR;
2295         }
2296
2297         dev->mtu = msg->payload.u64;
2298
2299         return RTE_VHOST_MSG_RESULT_OK;
2300 }
2301
2302 static int
2303 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2304                         int main_fd __rte_unused)
2305 {
2306         struct virtio_net *dev = *pdev;
2307         int fd = msg->fds[0];
2308
2309         if (validate_msg_fds(msg, 1) != 0)
2310                 return RTE_VHOST_MSG_RESULT_ERR;
2311
2312         if (fd < 0) {
2313                 VHOST_LOG_CONFIG(ERR,
2314                                 "Invalid file descriptor for slave channel (%d)\n",
2315                                 fd);
2316                 return RTE_VHOST_MSG_RESULT_ERR;
2317         }
2318
2319         if (dev->slave_req_fd >= 0)
2320                 close(dev->slave_req_fd);
2321
2322         dev->slave_req_fd = fd;
2323
2324         return RTE_VHOST_MSG_RESULT_OK;
2325 }
2326
2327 static int
2328 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2329 {
2330         struct vhost_vring_addr *ra;
2331         uint64_t start, end, len;
2332
2333         start = imsg->iova;
2334         end = start + imsg->size;
2335
2336         ra = &vq->ring_addrs;
2337         len = sizeof(struct vring_desc) * vq->size;
2338         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2339                 return 1;
2340
2341         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2342         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2343                 return 1;
2344
2345         len = sizeof(struct vring_used) +
2346                sizeof(struct vring_used_elem) * vq->size;
2347         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2348                 return 1;
2349
2350         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2351                 len = sizeof(uint64_t);
2352                 if (ra->log_guest_addr < end &&
2353                     (ra->log_guest_addr + len) > start)
2354                         return 1;
2355         }
2356
2357         return 0;
2358 }
2359
2360 static int
2361 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2362 {
2363         struct vhost_vring_addr *ra;
2364         uint64_t start, end, len;
2365
2366         start = imsg->iova;
2367         end = start + imsg->size;
2368
2369         ra = &vq->ring_addrs;
2370         len = sizeof(struct vring_packed_desc) * vq->size;
2371         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2372                 return 1;
2373
2374         len = sizeof(struct vring_packed_desc_event);
2375         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2376                 return 1;
2377
2378         len = sizeof(struct vring_packed_desc_event);
2379         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2380                 return 1;
2381
2382         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2383                 len = sizeof(uint64_t);
2384                 if (ra->log_guest_addr < end &&
2385                     (ra->log_guest_addr + len) > start)
2386                         return 1;
2387         }
2388
2389         return 0;
2390 }
2391
2392 static int is_vring_iotlb(struct virtio_net *dev,
2393                           struct vhost_virtqueue *vq,
2394                           struct vhost_iotlb_msg *imsg)
2395 {
2396         if (vq_is_packed(dev))
2397                 return is_vring_iotlb_packed(vq, imsg);
2398         else
2399                 return is_vring_iotlb_split(vq, imsg);
2400 }
2401
2402 static int
2403 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2404                         int main_fd __rte_unused)
2405 {
2406         struct virtio_net *dev = *pdev;
2407         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2408         uint16_t i;
2409         uint64_t vva, len;
2410
2411         if (validate_msg_fds(msg, 0) != 0)
2412                 return RTE_VHOST_MSG_RESULT_ERR;
2413
2414         switch (imsg->type) {
2415         case VHOST_IOTLB_UPDATE:
2416                 len = imsg->size;
2417                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2418                 if (!vva)
2419                         return RTE_VHOST_MSG_RESULT_ERR;
2420
2421                 for (i = 0; i < dev->nr_vring; i++) {
2422                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2423
2424                         if (!vq)
2425                                 continue;
2426
2427                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2428                                         len, imsg->perm);
2429
2430                         if (is_vring_iotlb(dev, vq, imsg))
2431                                 *pdev = dev = translate_ring_addresses(dev, i);
2432                 }
2433                 break;
2434         case VHOST_IOTLB_INVALIDATE:
2435                 for (i = 0; i < dev->nr_vring; i++) {
2436                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2437
2438                         if (!vq)
2439                                 continue;
2440
2441                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2442                                         imsg->size);
2443
2444                         if (is_vring_iotlb(dev, vq, imsg))
2445                                 vring_invalidate(dev, vq);
2446                 }
2447                 break;
2448         default:
2449                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2450                                 imsg->type);
2451                 return RTE_VHOST_MSG_RESULT_ERR;
2452         }
2453
2454         return RTE_VHOST_MSG_RESULT_OK;
2455 }
2456
2457 static int
2458 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2459                         struct VhostUserMsg *msg,
2460                         int main_fd __rte_unused)
2461 {
2462         struct virtio_net *dev = *pdev;
2463 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2464         struct uffdio_api api_struct;
2465
2466         if (validate_msg_fds(msg, 0) != 0)
2467                 return RTE_VHOST_MSG_RESULT_ERR;
2468
2469         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2470
2471         if (dev->postcopy_ufd == -1) {
2472                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2473                         strerror(errno));
2474                 return RTE_VHOST_MSG_RESULT_ERR;
2475         }
2476         api_struct.api = UFFD_API;
2477         api_struct.features = 0;
2478         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2479                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2480                         strerror(errno));
2481                 close(dev->postcopy_ufd);
2482                 dev->postcopy_ufd = -1;
2483                 return RTE_VHOST_MSG_RESULT_ERR;
2484         }
2485         msg->fds[0] = dev->postcopy_ufd;
2486         msg->fd_num = 1;
2487
2488         return RTE_VHOST_MSG_RESULT_REPLY;
2489 #else
2490         dev->postcopy_ufd = -1;
2491         msg->fd_num = 0;
2492
2493         return RTE_VHOST_MSG_RESULT_ERR;
2494 #endif
2495 }
2496
2497 static int
2498 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2499                         struct VhostUserMsg *msg __rte_unused,
2500                         int main_fd __rte_unused)
2501 {
2502         struct virtio_net *dev = *pdev;
2503
2504         if (validate_msg_fds(msg, 0) != 0)
2505                 return RTE_VHOST_MSG_RESULT_ERR;
2506
2507         if (dev->mem && dev->mem->nregions) {
2508                 VHOST_LOG_CONFIG(ERR,
2509                         "Regions already registered at postcopy-listen\n");
2510                 return RTE_VHOST_MSG_RESULT_ERR;
2511         }
2512         dev->postcopy_listening = 1;
2513
2514         return RTE_VHOST_MSG_RESULT_OK;
2515 }
2516
2517 static int
2518 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2519                         int main_fd __rte_unused)
2520 {
2521         struct virtio_net *dev = *pdev;
2522
2523         if (validate_msg_fds(msg, 0) != 0)
2524                 return RTE_VHOST_MSG_RESULT_ERR;
2525
2526         dev->postcopy_listening = 0;
2527         if (dev->postcopy_ufd >= 0) {
2528                 close(dev->postcopy_ufd);
2529                 dev->postcopy_ufd = -1;
2530         }
2531
2532         msg->payload.u64 = 0;
2533         msg->size = sizeof(msg->payload.u64);
2534         msg->fd_num = 0;
2535
2536         return RTE_VHOST_MSG_RESULT_REPLY;
2537 }
2538
2539 static int
2540 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2541                       int main_fd __rte_unused)
2542 {
2543         struct virtio_net *dev = *pdev;
2544
2545         if (validate_msg_fds(msg, 0) != 0)
2546                 return RTE_VHOST_MSG_RESULT_ERR;
2547
2548         msg->payload.u64 = dev->status;
2549         msg->size = sizeof(msg->payload.u64);
2550         msg->fd_num = 0;
2551
2552         return RTE_VHOST_MSG_RESULT_REPLY;
2553 }
2554
2555 static int
2556 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2557                         int main_fd __rte_unused)
2558 {
2559         struct virtio_net *dev = *pdev;
2560
2561         if (validate_msg_fds(msg, 0) != 0)
2562                 return RTE_VHOST_MSG_RESULT_ERR;
2563
2564         /* As per Virtio specification, the device status is 8bits long */
2565         if (msg->payload.u64 > UINT8_MAX) {
2566                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2567                                 msg->payload.u64);
2568                 return RTE_VHOST_MSG_RESULT_ERR;
2569         }
2570
2571         dev->status = msg->payload.u64;
2572
2573         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2574             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2575                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2576                 /*
2577                  * Clear the bit to let the driver know about the feature
2578                  * negotiation failure
2579                  */
2580                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2581         }
2582
2583         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2584                         "\t-RESET: %u\n"
2585                         "\t-ACKNOWLEDGE: %u\n"
2586                         "\t-DRIVER: %u\n"
2587                         "\t-FEATURES_OK: %u\n"
2588                         "\t-DRIVER_OK: %u\n"
2589                         "\t-DEVICE_NEED_RESET: %u\n"
2590                         "\t-FAILED: %u\n",
2591                         dev->status,
2592                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2593                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2594                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2595                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2596                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2597                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2598                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2599
2600         return RTE_VHOST_MSG_RESULT_OK;
2601 }
2602
2603 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2604                                         struct VhostUserMsg *msg,
2605                                         int main_fd);
2606 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2607         [VHOST_USER_NONE] = NULL,
2608         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2609         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2610         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2611         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2612         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2613         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2614         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2615         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2616         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2617         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2618         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2619         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2620         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2621         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2622         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2623         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2624         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2625         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2626         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2627         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2628         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2629         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2630         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2631         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2632         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2633         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2634         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2635         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2636         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2637 };
2638
2639 /* return bytes# of read on success or negative val on failure. */
2640 static int
2641 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2642 {
2643         int ret;
2644
2645         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2646                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2647         if (ret <= 0) {
2648                 return ret;
2649         } else if (ret != VHOST_USER_HDR_SIZE) {
2650                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2651                 close_msg_fds(msg);
2652                 return -1;
2653         }
2654
2655         if (msg->size) {
2656                 if (msg->size > sizeof(msg->payload)) {
2657                         VHOST_LOG_CONFIG(ERR,
2658                                 "invalid msg size: %d\n", msg->size);
2659                         return -1;
2660                 }
2661                 ret = read(sockfd, &msg->payload, msg->size);
2662                 if (ret <= 0)
2663                         return ret;
2664                 if (ret != (int)msg->size) {
2665                         VHOST_LOG_CONFIG(ERR,
2666                                 "read control message failed\n");
2667                         return -1;
2668                 }
2669         }
2670
2671         return ret;
2672 }
2673
2674 static int
2675 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2676 {
2677         if (!msg)
2678                 return 0;
2679
2680         return send_fd_message(sockfd, (char *)msg,
2681                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2682 }
2683
2684 static int
2685 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2686 {
2687         if (!msg)
2688                 return 0;
2689
2690         msg->flags &= ~VHOST_USER_VERSION_MASK;
2691         msg->flags &= ~VHOST_USER_NEED_REPLY;
2692         msg->flags |= VHOST_USER_VERSION;
2693         msg->flags |= VHOST_USER_REPLY_MASK;
2694
2695         return send_vhost_message(sockfd, msg);
2696 }
2697
2698 static int
2699 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2700 {
2701         int ret;
2702
2703         if (msg->flags & VHOST_USER_NEED_REPLY)
2704                 rte_spinlock_lock(&dev->slave_req_lock);
2705
2706         ret = send_vhost_message(dev->slave_req_fd, msg);
2707         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2708                 rte_spinlock_unlock(&dev->slave_req_lock);
2709
2710         return ret;
2711 }
2712
2713 /*
2714  * Allocate a queue pair if it hasn't been allocated yet
2715  */
2716 static int
2717 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2718                         struct VhostUserMsg *msg)
2719 {
2720         uint32_t vring_idx;
2721
2722         switch (msg->request.master) {
2723         case VHOST_USER_SET_VRING_KICK:
2724         case VHOST_USER_SET_VRING_CALL:
2725         case VHOST_USER_SET_VRING_ERR:
2726                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2727                 break;
2728         case VHOST_USER_SET_VRING_NUM:
2729         case VHOST_USER_SET_VRING_BASE:
2730         case VHOST_USER_SET_VRING_ENABLE:
2731                 vring_idx = msg->payload.state.index;
2732                 break;
2733         case VHOST_USER_SET_VRING_ADDR:
2734                 vring_idx = msg->payload.addr.index;
2735                 break;
2736         default:
2737                 return 0;
2738         }
2739
2740         if (vring_idx >= VHOST_MAX_VRING) {
2741                 VHOST_LOG_CONFIG(ERR,
2742                         "invalid vring index: %u\n", vring_idx);
2743                 return -1;
2744         }
2745
2746         if (dev->virtqueue[vring_idx])
2747                 return 0;
2748
2749         return alloc_vring_queue(dev, vring_idx);
2750 }
2751
2752 static void
2753 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2754 {
2755         unsigned int i = 0;
2756         unsigned int vq_num = 0;
2757
2758         while (vq_num < dev->nr_vring) {
2759                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2760
2761                 if (vq) {
2762                         rte_spinlock_lock(&vq->access_lock);
2763                         vq_num++;
2764                 }
2765                 i++;
2766         }
2767 }
2768
2769 static void
2770 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2771 {
2772         unsigned int i = 0;
2773         unsigned int vq_num = 0;
2774
2775         while (vq_num < dev->nr_vring) {
2776                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2777
2778                 if (vq) {
2779                         rte_spinlock_unlock(&vq->access_lock);
2780                         vq_num++;
2781                 }
2782                 i++;
2783         }
2784 }
2785
2786 int
2787 vhost_user_msg_handler(int vid, int fd)
2788 {
2789         struct virtio_net *dev;
2790         struct VhostUserMsg msg;
2791         struct rte_vdpa_device *vdpa_dev;
2792         int ret;
2793         int unlock_required = 0;
2794         bool handled;
2795         int request;
2796         uint32_t i;
2797
2798         dev = get_device(vid);
2799         if (dev == NULL)
2800                 return -1;
2801
2802         if (!dev->notify_ops) {
2803                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2804                 if (!dev->notify_ops) {
2805                         VHOST_LOG_CONFIG(ERR,
2806                                 "failed to get callback ops for driver %s\n",
2807                                 dev->ifname);
2808                         return -1;
2809                 }
2810         }
2811
2812         ret = read_vhost_message(fd, &msg);
2813         if (ret <= 0) {
2814                 if (ret < 0)
2815                         VHOST_LOG_CONFIG(ERR,
2816                                 "vhost read message failed\n");
2817                 else
2818                         VHOST_LOG_CONFIG(INFO,
2819                                 "vhost peer closed\n");
2820
2821                 return -1;
2822         }
2823
2824         ret = 0;
2825         request = msg.request.master;
2826         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2827                         vhost_message_str[request]) {
2828                 if (request != VHOST_USER_IOTLB_MSG)
2829                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2830                                 vhost_message_str[request]);
2831                 else
2832                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2833                                 vhost_message_str[request]);
2834         } else {
2835                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2836         }
2837
2838         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2839         if (ret < 0) {
2840                 VHOST_LOG_CONFIG(ERR,
2841                         "failed to alloc queue\n");
2842                 return -1;
2843         }
2844
2845         /*
2846          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2847          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2848          * and device is destroyed. destroy_device waits for queues to be
2849          * inactive, so it is safe. Otherwise taking the access_lock
2850          * would cause a dead lock.
2851          */
2852         switch (request) {
2853         case VHOST_USER_SET_FEATURES:
2854         case VHOST_USER_SET_PROTOCOL_FEATURES:
2855         case VHOST_USER_SET_OWNER:
2856         case VHOST_USER_SET_MEM_TABLE:
2857         case VHOST_USER_SET_LOG_BASE:
2858         case VHOST_USER_SET_LOG_FD:
2859         case VHOST_USER_SET_VRING_NUM:
2860         case VHOST_USER_SET_VRING_ADDR:
2861         case VHOST_USER_SET_VRING_BASE:
2862         case VHOST_USER_SET_VRING_KICK:
2863         case VHOST_USER_SET_VRING_CALL:
2864         case VHOST_USER_SET_VRING_ERR:
2865         case VHOST_USER_SET_VRING_ENABLE:
2866         case VHOST_USER_SEND_RARP:
2867         case VHOST_USER_NET_SET_MTU:
2868         case VHOST_USER_SET_SLAVE_REQ_FD:
2869                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2870                         vhost_user_lock_all_queue_pairs(dev);
2871                         unlock_required = 1;
2872                 }
2873                 break;
2874         default:
2875                 break;
2876
2877         }
2878
2879         handled = false;
2880         if (dev->extern_ops.pre_msg_handle) {
2881                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2882                                 (void *)&msg);
2883                 switch (ret) {
2884                 case RTE_VHOST_MSG_RESULT_REPLY:
2885                         send_vhost_reply(fd, &msg);
2886                         /* Fall-through */
2887                 case RTE_VHOST_MSG_RESULT_ERR:
2888                 case RTE_VHOST_MSG_RESULT_OK:
2889                         handled = true;
2890                         goto skip_to_post_handle;
2891                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2892                 default:
2893                         break;
2894                 }
2895         }
2896
2897         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2898                 if (!vhost_message_handlers[request])
2899                         goto skip_to_post_handle;
2900                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2901
2902                 switch (ret) {
2903                 case RTE_VHOST_MSG_RESULT_ERR:
2904                         VHOST_LOG_CONFIG(ERR,
2905                                 "Processing %s failed.\n",
2906                                 vhost_message_str[request]);
2907                         handled = true;
2908                         break;
2909                 case RTE_VHOST_MSG_RESULT_OK:
2910                         VHOST_LOG_CONFIG(DEBUG,
2911                                 "Processing %s succeeded.\n",
2912                                 vhost_message_str[request]);
2913                         handled = true;
2914                         break;
2915                 case RTE_VHOST_MSG_RESULT_REPLY:
2916                         VHOST_LOG_CONFIG(DEBUG,
2917                                 "Processing %s succeeded and needs reply.\n",
2918                                 vhost_message_str[request]);
2919                         send_vhost_reply(fd, &msg);
2920                         handled = true;
2921                         break;
2922                 default:
2923                         break;
2924                 }
2925         }
2926
2927 skip_to_post_handle:
2928         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2929                         dev->extern_ops.post_msg_handle) {
2930                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2931                                 (void *)&msg);
2932                 switch (ret) {
2933                 case RTE_VHOST_MSG_RESULT_REPLY:
2934                         send_vhost_reply(fd, &msg);
2935                         /* Fall-through */
2936                 case RTE_VHOST_MSG_RESULT_ERR:
2937                 case RTE_VHOST_MSG_RESULT_OK:
2938                         handled = true;
2939                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2940                 default:
2941                         break;
2942                 }
2943         }
2944
2945         if (unlock_required)
2946                 vhost_user_unlock_all_queue_pairs(dev);
2947
2948         /* If message was not handled at this stage, treat it as an error */
2949         if (!handled) {
2950                 VHOST_LOG_CONFIG(ERR,
2951                         "vhost message (req: %d) was not handled.\n", request);
2952                 close_msg_fds(&msg);
2953                 ret = RTE_VHOST_MSG_RESULT_ERR;
2954         }
2955
2956         /*
2957          * If the request required a reply that was already sent,
2958          * this optional reply-ack won't be sent as the
2959          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2960          */
2961         if (msg.flags & VHOST_USER_NEED_REPLY) {
2962                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2963                 msg.size = sizeof(msg.payload.u64);
2964                 msg.fd_num = 0;
2965                 send_vhost_reply(fd, &msg);
2966         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2967                 VHOST_LOG_CONFIG(ERR,
2968                         "vhost message handling failed.\n");
2969                 return -1;
2970         }
2971
2972         for (i = 0; i < dev->nr_vring; i++) {
2973                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2974                 bool cur_ready = vq_is_ready(dev, vq);
2975
2976                 if (cur_ready != (vq && vq->ready)) {
2977                         vq->ready = cur_ready;
2978                         vhost_user_notify_queue_state(dev, i, cur_ready);
2979                 }
2980         }
2981
2982
2983         if (!virtio_is_ready(dev))
2984                 goto out;
2985
2986         /*
2987          * Virtio is now ready. If not done already, it is time
2988          * to notify the application it can process the rings and
2989          * configure the vDPA device if present.
2990          */
2991
2992         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2993                 if (dev->notify_ops->new_device(dev->vid) == 0)
2994                         dev->flags |= VIRTIO_DEV_RUNNING;
2995         }
2996
2997         vdpa_dev = dev->vdpa_dev;
2998         if (!vdpa_dev)
2999                 goto out;
3000
3001         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3002                 if (vdpa_dev->ops->dev_conf(dev->vid))
3003                         VHOST_LOG_CONFIG(ERR,
3004                                          "Failed to configure vDPA device\n");
3005                 else
3006                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3007         }
3008
3009 out:
3010         return 0;
3011 }
3012
3013 static int process_slave_message_reply(struct virtio_net *dev,
3014                                        const struct VhostUserMsg *msg)
3015 {
3016         struct VhostUserMsg msg_reply;
3017         int ret;
3018
3019         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3020                 return 0;
3021
3022         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3023         if (ret <= 0) {
3024                 if (ret < 0)
3025                         VHOST_LOG_CONFIG(ERR,
3026                                 "vhost read slave message reply failed\n");
3027                 else
3028                         VHOST_LOG_CONFIG(INFO,
3029                                 "vhost peer closed\n");
3030                 ret = -1;
3031                 goto out;
3032         }
3033
3034         ret = 0;
3035         if (msg_reply.request.slave != msg->request.slave) {
3036                 VHOST_LOG_CONFIG(ERR,
3037                         "Received unexpected msg type (%u), expected %u\n",
3038                         msg_reply.request.slave, msg->request.slave);
3039                 ret = -1;
3040                 goto out;
3041         }
3042
3043         ret = msg_reply.payload.u64 ? -1 : 0;
3044
3045 out:
3046         rte_spinlock_unlock(&dev->slave_req_lock);
3047         return ret;
3048 }
3049
3050 int
3051 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3052 {
3053         int ret;
3054         struct VhostUserMsg msg = {
3055                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3056                 .flags = VHOST_USER_VERSION,
3057                 .size = sizeof(msg.payload.iotlb),
3058                 .payload.iotlb = {
3059                         .iova = iova,
3060                         .perm = perm,
3061                         .type = VHOST_IOTLB_MISS,
3062                 },
3063         };
3064
3065         ret = send_vhost_message(dev->slave_req_fd, &msg);
3066         if (ret < 0) {
3067                 VHOST_LOG_CONFIG(ERR,
3068                                 "Failed to send IOTLB miss message (%d)\n",
3069                                 ret);
3070                 return ret;
3071         }
3072
3073         return 0;
3074 }
3075
3076 static int
3077 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3078 {
3079         int ret;
3080         struct VhostUserMsg msg = {
3081                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3082                 .flags = VHOST_USER_VERSION,
3083                 .size = 0,
3084         };
3085
3086         if (need_reply)
3087                 msg.flags |= VHOST_USER_NEED_REPLY;
3088
3089         ret = send_vhost_slave_message(dev, &msg);
3090         if (ret < 0) {
3091                 VHOST_LOG_CONFIG(ERR,
3092                                 "Failed to send config change (%d)\n",
3093                                 ret);
3094                 return ret;
3095         }
3096
3097         return process_slave_message_reply(dev, &msg);
3098 }
3099
3100 int
3101 rte_vhost_slave_config_change(int vid, bool need_reply)
3102 {
3103         struct virtio_net *dev;
3104
3105         dev = get_device(vid);
3106         if (!dev)
3107                 return -ENODEV;
3108
3109         return vhost_user_slave_config_change(dev, need_reply);
3110 }
3111
3112 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3113                                                     int index, int fd,
3114                                                     uint64_t offset,
3115                                                     uint64_t size)
3116 {
3117         int ret;
3118         struct VhostUserMsg msg = {
3119                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3120                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3121                 .size = sizeof(msg.payload.area),
3122                 .payload.area = {
3123                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3124                         .size = size,
3125                         .offset = offset,
3126                 },
3127         };
3128
3129         if (fd < 0)
3130                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3131         else {
3132                 msg.fds[0] = fd;
3133                 msg.fd_num = 1;
3134         }
3135
3136         ret = send_vhost_slave_message(dev, &msg);
3137         if (ret < 0) {
3138                 VHOST_LOG_CONFIG(ERR,
3139                         "Failed to set host notifier (%d)\n", ret);
3140                 return ret;
3141         }
3142
3143         return process_slave_message_reply(dev, &msg);
3144 }
3145
3146 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3147 {
3148         struct virtio_net *dev;
3149         struct rte_vdpa_device *vdpa_dev;
3150         int vfio_device_fd, ret = 0;
3151         uint64_t offset, size;
3152         unsigned int i, q_start, q_last;
3153
3154         dev = get_device(vid);
3155         if (!dev)
3156                 return -ENODEV;
3157
3158         vdpa_dev = dev->vdpa_dev;
3159         if (vdpa_dev == NULL)
3160                 return -ENODEV;
3161
3162         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3163             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3164             !(dev->protocol_features &
3165                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3166             !(dev->protocol_features &
3167                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3168             !(dev->protocol_features &
3169                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3170                 return -ENOTSUP;
3171
3172         if (qid == RTE_VHOST_QUEUE_ALL) {
3173                 q_start = 0;
3174                 q_last = dev->nr_vring - 1;
3175         } else {
3176                 if (qid >= dev->nr_vring)
3177                         return -EINVAL;
3178                 q_start = qid;
3179                 q_last = qid;
3180         }
3181
3182         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3183         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3184
3185         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3186         if (vfio_device_fd < 0)
3187                 return -ENOTSUP;
3188
3189         if (enable) {
3190                 for (i = q_start; i <= q_last; i++) {
3191                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3192                                         &size) < 0) {
3193                                 ret = -ENOTSUP;
3194                                 goto disable;
3195                         }
3196
3197                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3198                                         vfio_device_fd, offset, size) < 0) {
3199                                 ret = -EFAULT;
3200                                 goto disable;
3201                         }
3202                 }
3203         } else {
3204 disable:
3205                 for (i = q_start; i <= q_last; i++) {
3206                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3207                                         0, 0);
3208                 }
3209         }
3210
3211         return ret;
3212 }