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