lib/vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 /* Security model
   6  * --------------
   7  * The vhost-user protocol connection is an external interface, so it must be
   8  * robust against invalid inputs.
   9  *
  10  * This is important because the vhost-user master is only one step removed
  11  * from the guest.  Malicious guests that have escaped will then launch further
  12  * attacks from the vhost-user master.
  13  *
  14  * Even in deployments where guests are trusted, a bug in the vhost-user master
  15  * can still cause invalid messages to be sent.  Such messages must not
  16  * compromise the stability of the DPDK application by causing crashes, memory
  17  * corruption, or other problematic behavior.
  18  *
  19  * Do not assume received VhostUserMsg fields contain sensible values!
  20  */
  21
  22 #include <stdint.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <fcntl.h>
  28 #include <sys/ioctl.h>
  29 #include <sys/mman.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/syscall.h>
  33 #include <assert.h>
  34 #ifdef RTE_LIBRTE_VHOST_NUMA
  35 #include <numaif.h>
  36 #endif
  37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
  38 #include <linux/userfaultfd.h>
  39 #endif
  40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
  41 #include <linux/memfd.h>
  42 #define MEMFD_SUPPORTED
  43 #endif
  44
  45 #include <rte_common.h>
  46 #include <rte_malloc.h>
  47 #include <rte_log.h>
  48
  49 #include "iotlb.h"
  50 #include "vhost.h"
  51 #include "vhost_user.h"
  52
  53 #define VIRTIO_MIN_MTU 68
  54 #define VIRTIO_MAX_MTU 65535
  55
  56 #define INFLIGHT_ALIGNMENT      64
  57 #define INFLIGHT_VERSION        0x1
  58
  59 static const char *vhost_message_str[VHOST_USER_MAX] = {
  60         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  61         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  62         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  63         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  64         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  65         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  66         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  67         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  68         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  69         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  70         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  71         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  72         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  73         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  74         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  75         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  76         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  77         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  78         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  79         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  80         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  81         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  82         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  83         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  84         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  85         [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
  86         [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
  87         [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
  88         [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
  89         [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
  90         [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
  91         [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
  92 };
  93
  94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
  95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
  96
  97 static void
  98 close_msg_fds(struct VhostUserMsg *msg)
  99 {
 100         int i;
 101
 102         for (i = 0; i < msg->fd_num; i++) {
 103                 int fd = msg->fds[i];
 104
 105                 if (fd == -1)
 106                         continue;
 107
 108                 msg->fds[i] = -1;
 109                 close(fd);
 110         }
 111 }
 112
 113 /*
 114  * Ensure the expected number of FDs is received,
 115  * close all FDs and return an error if this is not the case.
 116  */
 117 static int
 118 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
 119 {
 120         if (msg->fd_num == expected_fds)
 121                 return 0;
 122
 123         VHOST_LOG_CONFIG(ERR,
 124                 " Expect %d FDs for request %s, received %d\n",
 125                 expected_fds,
 126                 vhost_message_str[msg->request.master],
 127                 msg->fd_num);
 128
 129         close_msg_fds(msg);
 130
 131         return -1;
 132 }
 133
 134 static uint64_t
 135 get_blk_size(int fd)
 136 {
 137         struct stat stat;
 138         int ret;
 139
 140         ret = fstat(fd, &stat);
 141         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 142 }
 143
 144 static void
 145 free_mem_region(struct virtio_net *dev)
 146 {
 147         uint32_t i;
 148         struct rte_vhost_mem_region *reg;
 149
 150         if (!dev || !dev->mem)
 151                 return;
 152
 153         for (i = 0; i < dev->mem->nregions; i++) {
 154                 reg = &dev->mem->regions[i];
 155                 if (reg->host_user_addr) {
 156                         munmap(reg->mmap_addr, reg->mmap_size);
 157                         close(reg->fd);
 158                 }
 159         }
 160 }
 161
 162 void
 163 vhost_backend_cleanup(struct virtio_net *dev)
 164 {
 165         if (dev->mem) {
 166                 free_mem_region(dev);
 167                 rte_free(dev->mem);
 168                 dev->mem = NULL;
 169         }
 170
 171         rte_free(dev->guest_pages);
 172         dev->guest_pages = NULL;
 173
 174         if (dev->log_addr) {
 175                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 176                 dev->log_addr = 0;
 177         }
 178
 179         if (dev->inflight_info) {
 180                 if (dev->inflight_info->addr) {
 181                         munmap(dev->inflight_info->addr,
 182                                dev->inflight_info->size);
 183                         dev->inflight_info->addr = NULL;
 184                 }
 185
 186                 if (dev->inflight_info->fd >= 0) {
 187                         close(dev->inflight_info->fd);
 188                         dev->inflight_info->fd = -1;
 189                 }
 190
 191                 free(dev->inflight_info);
 192                 dev->inflight_info = NULL;
 193         }
 194
 195         if (dev->slave_req_fd >= 0) {
 196                 close(dev->slave_req_fd);
 197                 dev->slave_req_fd = -1;
 198         }
 199
 200         if (dev->postcopy_ufd >= 0) {
 201                 close(dev->postcopy_ufd);
 202                 dev->postcopy_ufd = -1;
 203         }
 204
 205         dev->postcopy_listening = 0;
 206 }
 207
 208 static void
 209 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
 210                               int enable)
 211 {
 212         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
 213         struct vhost_virtqueue *vq = dev->virtqueue[index];
 214
 215         /* Configure guest notifications on enable */
 216         if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
 217                 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
 218
 219         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
 220                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
 221
 222         if (dev->notify_ops->vring_state_changed)
 223                 dev->notify_ops->vring_state_changed(dev->vid,
 224                                 index, enable);
 225 }
 226
 227 /*
 228  * This function just returns success at the moment unless
 229  * the device hasn't been initialised.
 230  */
 231 static int
 232 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
 233                         struct VhostUserMsg *msg,
 234                         int main_fd __rte_unused)
 235 {
 236         if (validate_msg_fds(msg, 0) != 0)
 237                 return RTE_VHOST_MSG_RESULT_ERR;
 238
 239         return RTE_VHOST_MSG_RESULT_OK;
 240 }
 241
 242 static int
 243 vhost_user_reset_owner(struct virtio_net **pdev,
 244                         struct VhostUserMsg *msg,
 245                         int main_fd __rte_unused)
 246 {
 247         struct virtio_net *dev = *pdev;
 248
 249         if (validate_msg_fds(msg, 0) != 0)
 250                 return RTE_VHOST_MSG_RESULT_ERR;
 251
 252         vhost_destroy_device_notify(dev);
 253
 254         cleanup_device(dev, 0);
 255         reset_device(dev);
 256         return RTE_VHOST_MSG_RESULT_OK;
 257 }
 258
 259 /*
 260  * The features that we support are requested.
 261  */
 262 static int
 263 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 264                         int main_fd __rte_unused)
 265 {
 266         struct virtio_net *dev = *pdev;
 267         uint64_t features = 0;
 268
 269         if (validate_msg_fds(msg, 0) != 0)
 270                 return RTE_VHOST_MSG_RESULT_ERR;
 271
 272         rte_vhost_driver_get_features(dev->ifname, &features);
 273
 274         msg->payload.u64 = features;
 275         msg->size = sizeof(msg->payload.u64);
 276         msg->fd_num = 0;
 277
 278         return RTE_VHOST_MSG_RESULT_REPLY;
 279 }
 280
 281 /*
 282  * The queue number that we support are requested.
 283  */
 284 static int
 285 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
 286                         int main_fd __rte_unused)
 287 {
 288         struct virtio_net *dev = *pdev;
 289         uint32_t queue_num = 0;
 290
 291         if (validate_msg_fds(msg, 0) != 0)
 292                 return RTE_VHOST_MSG_RESULT_ERR;
 293
 294         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 295
 296         msg->payload.u64 = (uint64_t)queue_num;
 297         msg->size = sizeof(msg->payload.u64);
 298         msg->fd_num = 0;
 299
 300         return RTE_VHOST_MSG_RESULT_REPLY;
 301 }
 302
 303 /*
 304  * We receive the negotiated features supported by us and the virtio device.
 305  */
 306 static int
 307 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 308                         int main_fd __rte_unused)
 309 {
 310         struct virtio_net *dev = *pdev;
 311         uint64_t features = msg->payload.u64;
 312         uint64_t vhost_features = 0;
 313         struct rte_vdpa_device *vdpa_dev;
 314
 315         if (validate_msg_fds(msg, 0) != 0)
 316                 return RTE_VHOST_MSG_RESULT_ERR;
 317
 318         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 319         if (features & ~vhost_features) {
 320                 VHOST_LOG_CONFIG(ERR,
 321                         "(%d) received invalid negotiated features.\n",
 322                         dev->vid);
 323                 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
 324                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
 325
 326                 return RTE_VHOST_MSG_RESULT_ERR;
 327         }
 328
 329         if (dev->flags & VIRTIO_DEV_RUNNING) {
 330                 if (dev->features == features)
 331                         return RTE_VHOST_MSG_RESULT_OK;
 332
 333                 /*
 334                  * Error out if master tries to change features while device is
 335                  * in running state. The exception being VHOST_F_LOG_ALL, which
 336                  * is enabled when the live-migration starts.
 337                  */
 338                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 339                         VHOST_LOG_CONFIG(ERR,
 340                                 "(%d) features changed while device is running.\n",
 341                                 dev->vid);
 342                         return RTE_VHOST_MSG_RESULT_ERR;
 343                 }
 344
 345                 if (dev->notify_ops->features_changed)
 346                         dev->notify_ops->features_changed(dev->vid, features);
 347         }
 348
 349         dev->features = features;
 350         if (dev->features &
 351                 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
 352                  (1ULL << VIRTIO_F_VERSION_1) |
 353                  (1ULL << VIRTIO_F_RING_PACKED))) {
 354                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 355         } else {
 356                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 357         }
 358         VHOST_LOG_CONFIG(INFO,
 359                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 360         VHOST_LOG_CONFIG(DEBUG,
 361                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 362                 dev->vid,
 363                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 364                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 365
 366         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 367             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 368                 /*
 369                  * Remove all but first queue pair if MQ hasn't been
 370                  * negotiated. This is safe because the device is not
 371                  * running at this stage.
 372                  */
 373                 while (dev->nr_vring > 2) {
 374                         struct vhost_virtqueue *vq;
 375
 376                         vq = dev->virtqueue[--dev->nr_vring];
 377                         if (!vq)
 378                                 continue;
 379
 380                         dev->virtqueue[dev->nr_vring] = NULL;
 381                         cleanup_vq(vq, 1);
 382                         cleanup_vq_inflight(dev, vq);
 383                         free_vq(dev, vq);
 384                 }
 385         }
 386
 387         vdpa_dev = dev->vdpa_dev;
 388         if (vdpa_dev)
 389                 vdpa_dev->ops->set_features(dev->vid);
 390
 391         dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
 392         return RTE_VHOST_MSG_RESULT_OK;
 393 }
 394
 395 /*
 396  * The virtio device sends us the size of the descriptor ring.
 397  */
 398 static int
 399 vhost_user_set_vring_num(struct virtio_net **pdev,
 400                         struct VhostUserMsg *msg,
 401                         int main_fd __rte_unused)
 402 {
 403         struct virtio_net *dev = *pdev;
 404         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 405
 406         if (validate_msg_fds(msg, 0) != 0)
 407                 return RTE_VHOST_MSG_RESULT_ERR;
 408
 409         if (msg->payload.state.num > 32768) {
 410                 VHOST_LOG_CONFIG(ERR, "invalid virtqueue size %u\n", msg->payload.state.num);
 411                 return RTE_VHOST_MSG_RESULT_ERR;
 412         }
 413
 414         vq->size = msg->payload.state.num;
 415
 416         /* VIRTIO 1.0, 2.4 Virtqueues says:
 417          *
 418          *   Queue Size value is always a power of 2. The maximum Queue Size
 419          *   value is 32768.
 420          *
 421          * VIRTIO 1.1 2.7 Virtqueues says:
 422          *
 423          *   Packed virtqueues support up to 2^15 entries each.
 424          */
 425         if (!vq_is_packed(dev)) {
 426                 if (vq->size & (vq->size - 1)) {
 427                         VHOST_LOG_CONFIG(ERR,
 428                                 "invalid virtqueue size %u\n", vq->size);
 429                         return RTE_VHOST_MSG_RESULT_ERR;
 430                 }
 431         }
 432
 433         if (vq_is_packed(dev)) {
 434                 if (vq->shadow_used_packed)
 435                         rte_free(vq->shadow_used_packed);
 436                 vq->shadow_used_packed = rte_malloc(NULL,
 437                                 vq->size *
 438                                 sizeof(struct vring_used_elem_packed),
 439                                 RTE_CACHE_LINE_SIZE);
 440                 if (!vq->shadow_used_packed) {
 441                         VHOST_LOG_CONFIG(ERR,
 442                                         "failed to allocate memory for shadow used ring.\n");
 443                         return RTE_VHOST_MSG_RESULT_ERR;
 444                 }
 445
 446         } else {
 447                 if (vq->shadow_used_split)
 448                         rte_free(vq->shadow_used_split);
 449
 450                 vq->shadow_used_split = rte_malloc(NULL,
 451                                 vq->size * sizeof(struct vring_used_elem),
 452                                 RTE_CACHE_LINE_SIZE);
 453
 454                 if (!vq->shadow_used_split) {
 455                         VHOST_LOG_CONFIG(ERR,
 456                                         "failed to allocate memory for vq internal data.\n");
 457                         return RTE_VHOST_MSG_RESULT_ERR;
 458                 }
 459         }
 460
 461         if (vq->batch_copy_elems)
 462                 rte_free(vq->batch_copy_elems);
 463         vq->batch_copy_elems = rte_malloc(NULL,
 464                                 vq->size * sizeof(struct batch_copy_elem),
 465                                 RTE_CACHE_LINE_SIZE);
 466         if (!vq->batch_copy_elems) {
 467                 VHOST_LOG_CONFIG(ERR,
 468                         "failed to allocate memory for batching copy.\n");
 469                 return RTE_VHOST_MSG_RESULT_ERR;
 470         }
 471
 472         return RTE_VHOST_MSG_RESULT_OK;
 473 }
 474
 475 /*
 476  * Reallocate virtio_dev, vhost_virtqueue and related data structures to
 477  * make them on the same numa node as the memory of vring descriptor.
 478  */
 479 #ifdef RTE_LIBRTE_VHOST_NUMA
 480 static struct virtio_net*
 481 numa_realloc(struct virtio_net *dev, int index)
 482 {
 483         int node, dev_node;
 484         struct virtio_net *old_dev;
 485         struct vhost_virtqueue *vq;
 486         struct batch_copy_elem *bce;
 487         struct guest_page *gp;
 488         struct rte_vhost_memory *mem;
 489         size_t mem_size;
 490         int ret;
 491
 492         old_dev = dev;
 493         vq = dev->virtqueue[index];
 494
 495         /*
 496          * If VQ is ready, it is too late to reallocate, it certainly already
 497          * happened anyway on VHOST_USER_SET_VRING_ADRR.
 498          */
 499         if (vq->ready)
 500                 return dev;
 501
 502         ret = get_mempolicy(&node, NULL, 0, vq->desc, MPOL_F_NODE | MPOL_F_ADDR);
 503         if (ret) {
 504                 VHOST_LOG_CONFIG(ERR, "Unable to get virtqueue %d numa information.\n", index);
 505                 return dev;
 506         }
 507
 508         vq = rte_realloc_socket(vq, sizeof(*vq), 0, node);
 509         if (!vq) {
 510                 VHOST_LOG_CONFIG(ERR, "Failed to realloc virtqueue %d on node %d\n",
 511                                 index, node);
 512                 return dev;
 513         }
 514
 515         if (vq != dev->virtqueue[index]) {
 516                 VHOST_LOG_CONFIG(INFO, "reallocated virtqueue on node %d\n", node);
 517                 dev->virtqueue[index] = vq;
 518                 vhost_user_iotlb_init(dev, index);
 519         }
 520
 521         if (vq_is_packed(dev)) {
 522                 struct vring_used_elem_packed *sup;
 523
 524                 sup = rte_realloc_socket(vq->shadow_used_packed, vq->size * sizeof(*sup),
 525                                 RTE_CACHE_LINE_SIZE, node);
 526                 if (!sup) {
 527                         VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow packed on node %d\n", node);
 528                         return dev;
 529                 }
 530                 vq->shadow_used_packed = sup;
 531         } else {
 532                 struct vring_used_elem *sus;
 533
 534                 sus = rte_realloc_socket(vq->shadow_used_split, vq->size * sizeof(*sus),
 535                                 RTE_CACHE_LINE_SIZE, node);
 536                 if (!sus) {
 537                         VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow split on node %d\n", node);
 538                         return dev;
 539                 }
 540                 vq->shadow_used_split = sus;
 541         }
 542
 543         bce = rte_realloc_socket(vq->batch_copy_elems, vq->size * sizeof(*bce),
 544                         RTE_CACHE_LINE_SIZE, node);
 545         if (!bce) {
 546                 VHOST_LOG_CONFIG(ERR, "Failed to realloc batch copy elem on node %d\n", node);
 547                 return dev;
 548         }
 549         vq->batch_copy_elems = bce;
 550
 551         if (vq->log_cache) {
 552                 struct log_cache_entry *lc;
 553
 554                 lc = rte_realloc_socket(vq->log_cache, sizeof(*lc) * VHOST_LOG_CACHE_NR, 0, node);
 555                 if (!lc) {
 556                         VHOST_LOG_CONFIG(ERR, "Failed to realloc log cache on node %d\n", node);
 557                         return dev;
 558                 }
 559                 vq->log_cache = lc;
 560         }
 561
 562         if (dev->flags & VIRTIO_DEV_RUNNING)
 563                 return dev;
 564
 565         ret = get_mempolicy(&dev_node, NULL, 0, dev, MPOL_F_NODE | MPOL_F_ADDR);
 566         if (ret) {
 567                 VHOST_LOG_CONFIG(ERR, "Unable to get Virtio dev %d numa information.\n", dev->vid);
 568                 return dev;
 569         }
 570
 571         if (dev_node == node)
 572                 return dev;
 573
 574         dev = rte_realloc_socket(old_dev, sizeof(*dev), 0, node);
 575         if (!dev) {
 576                 VHOST_LOG_CONFIG(ERR, "Failed to realloc dev on node %d\n", node);
 577                 return old_dev;
 578         }
 579
 580         VHOST_LOG_CONFIG(INFO, "reallocated device on node %d\n", node);
 581         vhost_devices[dev->vid] = dev;
 582
 583         mem_size = sizeof(struct rte_vhost_memory) +
 584                 sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
 585         mem = rte_realloc_socket(dev->mem, mem_size, 0, node);
 586         if (!mem) {
 587                 VHOST_LOG_CONFIG(ERR, "Failed to realloc mem table on node %d\n", node);
 588                 return dev;
 589         }
 590         dev->mem = mem;
 591
 592         gp = rte_realloc_socket(dev->guest_pages, dev->max_guest_pages * sizeof(*gp),
 593                         RTE_CACHE_LINE_SIZE, node);
 594         if (!gp) {
 595                 VHOST_LOG_CONFIG(ERR, "Failed to realloc guest pages on node %d\n", node);
 596                 return dev;
 597         }
 598         dev->guest_pages = gp;
 599
 600         return dev;
 601 }
 602 #else
 603 static struct virtio_net*
 604 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 605 {
 606         return dev;
 607 }
 608 #endif
 609
 610 /* Converts QEMU virtual address to Vhost virtual address. */
 611 static uint64_t
 612 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 613 {
 614         struct rte_vhost_mem_region *r;
 615         uint32_t i;
 616
 617         if (unlikely(!dev || !dev->mem))
 618                 goto out_error;
 619
 620         /* Find the region where the address lives. */
 621         for (i = 0; i < dev->mem->nregions; i++) {
 622                 r = &dev->mem->regions[i];
 623
 624                 if (qva >= r->guest_user_addr &&
 625                     qva <  r->guest_user_addr + r->size) {
 626
 627                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 628                                 *len = r->guest_user_addr + r->size - qva;
 629
 630                         return qva - r->guest_user_addr +
 631                                r->host_user_addr;
 632                 }
 633         }
 634 out_error:
 635         *len = 0;
 636
 637         return 0;
 638 }
 639
 640
 641 /*
 642  * Converts ring address to Vhost virtual address.
 643  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 644  * else it is a QEMU virtual address.
 645  */
 646 static uint64_t
 647 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 648                 uint64_t ra, uint64_t *size)
 649 {
 650         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 651                 uint64_t vva;
 652
 653                 vhost_user_iotlb_rd_lock(vq);
 654                 vva = vhost_iova_to_vva(dev, vq, ra,
 655                                         size, VHOST_ACCESS_RW);
 656                 vhost_user_iotlb_rd_unlock(vq);
 657
 658                 return vva;
 659         }
 660
 661         return qva_to_vva(dev, ra, size);
 662 }
 663
 664 static uint64_t
 665 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 666 {
 667         uint64_t log_gpa;
 668
 669         vhost_user_iotlb_rd_lock(vq);
 670         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 671         vhost_user_iotlb_rd_unlock(vq);
 672
 673         return log_gpa;
 674 }
 675
 676 static struct virtio_net *
 677 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 678 {
 679         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 680         struct vhost_vring_addr *addr = &vq->ring_addrs;
 681         uint64_t len, expected_len;
 682
 683         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 684                 vq->log_guest_addr =
 685                         log_addr_to_gpa(dev, vq);
 686                 if (vq->log_guest_addr == 0) {
 687                         VHOST_LOG_CONFIG(DEBUG,
 688                                 "(%d) failed to map log_guest_addr.\n",
 689                                 dev->vid);
 690                         return dev;
 691                 }
 692         }
 693
 694         if (vq_is_packed(dev)) {
 695                 len = sizeof(struct vring_packed_desc) * vq->size;
 696                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 697                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 698                 if (vq->desc_packed == NULL ||
 699                                 len != sizeof(struct vring_packed_desc) *
 700                                 vq->size) {
 701                         VHOST_LOG_CONFIG(DEBUG,
 702                                 "(%d) failed to map desc_packed ring.\n",
 703                                 dev->vid);
 704                         return dev;
 705                 }
 706
 707                 dev = numa_realloc(dev, vq_index);
 708                 vq = dev->virtqueue[vq_index];
 709                 addr = &vq->ring_addrs;
 710
 711                 len = sizeof(struct vring_packed_desc_event);
 712                 vq->driver_event = (struct vring_packed_desc_event *)
 713                                         (uintptr_t)ring_addr_to_vva(dev,
 714                                         vq, addr->avail_user_addr, &len);
 715                 if (vq->driver_event == NULL ||
 716                                 len != sizeof(struct vring_packed_desc_event)) {
 717                         VHOST_LOG_CONFIG(DEBUG,
 718                                 "(%d) failed to find driver area address.\n",
 719                                 dev->vid);
 720                         return dev;
 721                 }
 722
 723                 len = sizeof(struct vring_packed_desc_event);
 724                 vq->device_event = (struct vring_packed_desc_event *)
 725                                         (uintptr_t)ring_addr_to_vva(dev,
 726                                         vq, addr->used_user_addr, &len);
 727                 if (vq->device_event == NULL ||
 728                                 len != sizeof(struct vring_packed_desc_event)) {
 729                         VHOST_LOG_CONFIG(DEBUG,
 730                                 "(%d) failed to find device area address.\n",
 731                                 dev->vid);
 732                         return dev;
 733                 }
 734
 735                 vq->access_ok = true;
 736                 return dev;
 737         }
 738
 739         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 740         if (vq->desc && vq->avail && vq->used)
 741                 return dev;
 742
 743         len = sizeof(struct vring_desc) * vq->size;
 744         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 745                         vq, addr->desc_user_addr, &len);
 746         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 747                 VHOST_LOG_CONFIG(DEBUG,
 748                         "(%d) failed to map desc ring.\n",
 749                         dev->vid);
 750                 return dev;
 751         }
 752
 753         dev = numa_realloc(dev, vq_index);
 754         vq = dev->virtqueue[vq_index];
 755         addr = &vq->ring_addrs;
 756
 757         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 758         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 759                 len += sizeof(uint16_t);
 760         expected_len = len;
 761         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 762                         vq, addr->avail_user_addr, &len);
 763         if (vq->avail == 0 || len != expected_len) {
 764                 VHOST_LOG_CONFIG(DEBUG,
 765                         "(%d) failed to map avail ring.\n",
 766                         dev->vid);
 767                 return dev;
 768         }
 769
 770         len = sizeof(struct vring_used) +
 771                 sizeof(struct vring_used_elem) * vq->size;
 772         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 773                 len += sizeof(uint16_t);
 774         expected_len = len;
 775         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 776                         vq, addr->used_user_addr, &len);
 777         if (vq->used == 0 || len != expected_len) {
 778                 VHOST_LOG_CONFIG(DEBUG,
 779                         "(%d) failed to map used ring.\n",
 780                         dev->vid);
 781                 return dev;
 782         }
 783
 784         if (vq->last_used_idx != vq->used->idx) {
 785                 VHOST_LOG_CONFIG(WARNING,
 786                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 787                         "some packets maybe resent for Tx and dropped for Rx\n",
 788                         vq->last_used_idx, vq->used->idx);
 789                 vq->last_used_idx  = vq->used->idx;
 790                 vq->last_avail_idx = vq->used->idx;
 791         }
 792
 793         vq->access_ok = true;
 794
 795         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 796                         dev->vid, vq->desc);
 797         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 798                         dev->vid, vq->avail);
 799         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 800                         dev->vid, vq->used);
 801         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 802                         dev->vid, vq->log_guest_addr);
 803
 804         return dev;
 805 }
 806
 807 /*
 808  * The virtio device sends us the desc, used and avail ring addresses.
 809  * This function then converts these to our address space.
 810  */
 811 static int
 812 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 813                         int main_fd __rte_unused)
 814 {
 815         struct virtio_net *dev = *pdev;
 816         struct vhost_virtqueue *vq;
 817         struct vhost_vring_addr *addr = &msg->payload.addr;
 818         bool access_ok;
 819
 820         if (validate_msg_fds(msg, 0) != 0)
 821                 return RTE_VHOST_MSG_RESULT_ERR;
 822
 823         if (dev->mem == NULL)
 824                 return RTE_VHOST_MSG_RESULT_ERR;
 825
 826         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 827         vq = dev->virtqueue[msg->payload.addr.index];
 828
 829         access_ok = vq->access_ok;
 830
 831         /*
 832          * Rings addresses should not be interpreted as long as the ring is not
 833          * started and enabled
 834          */
 835         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 836
 837         vring_invalidate(dev, vq);
 838
 839         if ((vq->enabled && (dev->features &
 840                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 841                         access_ok) {
 842                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 843                 if (!dev)
 844                         return RTE_VHOST_MSG_RESULT_ERR;
 845
 846                 *pdev = dev;
 847         }
 848
 849         return RTE_VHOST_MSG_RESULT_OK;
 850 }
 851
 852 /*
 853  * The virtio device sends us the available ring last used index.
 854  */
 855 static int
 856 vhost_user_set_vring_base(struct virtio_net **pdev,
 857                         struct VhostUserMsg *msg,
 858                         int main_fd __rte_unused)
 859 {
 860         struct virtio_net *dev = *pdev;
 861         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 862         uint64_t val = msg->payload.state.num;
 863
 864         if (validate_msg_fds(msg, 0) != 0)
 865                 return RTE_VHOST_MSG_RESULT_ERR;
 866
 867         if (vq_is_packed(dev)) {
 868                 /*
 869                  * Bit[0:14]: avail index
 870                  * Bit[15]: avail wrap counter
 871                  */
 872                 vq->last_avail_idx = val & 0x7fff;
 873                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 874                 /*
 875                  * Set used index to same value as available one, as
 876                  * their values should be the same since ring processing
 877                  * was stopped at get time.
 878                  */
 879                 vq->last_used_idx = vq->last_avail_idx;
 880                 vq->used_wrap_counter = vq->avail_wrap_counter;
 881         } else {
 882                 vq->last_used_idx = msg->payload.state.num;
 883                 vq->last_avail_idx = msg->payload.state.num;
 884         }
 885
 886         return RTE_VHOST_MSG_RESULT_OK;
 887 }
 888
 889 static int
 890 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 891                    uint64_t host_phys_addr, uint64_t size)
 892 {
 893         struct guest_page *page, *last_page;
 894         struct guest_page *old_pages;
 895
 896         if (dev->nr_guest_pages == dev->max_guest_pages) {
 897                 dev->max_guest_pages *= 2;
 898                 old_pages = dev->guest_pages;
 899                 dev->guest_pages = rte_realloc(dev->guest_pages,
 900                                         dev->max_guest_pages * sizeof(*page),
 901                                         RTE_CACHE_LINE_SIZE);
 902                 if (dev->guest_pages == NULL) {
 903                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 904                         rte_free(old_pages);
 905                         return -1;
 906                 }
 907         }
 908
 909         if (dev->nr_guest_pages > 0) {
 910                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 911                 /* merge if the two pages are continuous */
 912                 if (host_phys_addr == last_page->host_phys_addr +
 913                                       last_page->size) {
 914                         last_page->size += size;
 915                         return 0;
 916                 }
 917         }
 918
 919         page = &dev->guest_pages[dev->nr_guest_pages++];
 920         page->guest_phys_addr = guest_phys_addr;
 921         page->host_phys_addr  = host_phys_addr;
 922         page->size = size;
 923
 924         return 0;
 925 }
 926
 927 static int
 928 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 929                 uint64_t page_size)
 930 {
 931         uint64_t reg_size = reg->size;
 932         uint64_t host_user_addr  = reg->host_user_addr;
 933         uint64_t guest_phys_addr = reg->guest_phys_addr;
 934         uint64_t host_phys_addr;
 935         uint64_t size;
 936
 937         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 938         size = page_size - (guest_phys_addr & (page_size - 1));
 939         size = RTE_MIN(size, reg_size);
 940
 941         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 942                 return -1;
 943
 944         host_user_addr  += size;
 945         guest_phys_addr += size;
 946         reg_size -= size;
 947
 948         while (reg_size > 0) {
 949                 size = RTE_MIN(reg_size, page_size);
 950                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 951                                                   host_user_addr);
 952                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 953                                 size) < 0)
 954                         return -1;
 955
 956                 host_user_addr  += size;
 957                 guest_phys_addr += size;
 958                 reg_size -= size;
 959         }
 960
 961         /* sort guest page array if over binary search threshold */
 962         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 963                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 964                         sizeof(struct guest_page), guest_page_addrcmp);
 965         }
 966
 967         return 0;
 968 }
 969
 970 #ifdef RTE_LIBRTE_VHOST_DEBUG
 971 /* TODO: enable it only in debug mode? */
 972 static void
 973 dump_guest_pages(struct virtio_net *dev)
 974 {
 975         uint32_t i;
 976         struct guest_page *page;
 977
 978         for (i = 0; i < dev->nr_guest_pages; i++) {
 979                 page = &dev->guest_pages[i];
 980
 981                 VHOST_LOG_CONFIG(INFO,
 982                         "guest physical page region %u\n"
 983                         "\t guest_phys_addr: %" PRIx64 "\n"
 984                         "\t host_phys_addr : %" PRIx64 "\n"
 985                         "\t size           : %" PRIx64 "\n",
 986                         i,
 987                         page->guest_phys_addr,
 988                         page->host_phys_addr,
 989                         page->size);
 990         }
 991 }
 992 #else
 993 #define dump_guest_pages(dev)
 994 #endif
 995
 996 static bool
 997 vhost_memory_changed(struct VhostUserMemory *new,
 998                      struct rte_vhost_memory *old)
 999 {
1000         uint32_t i;
1001
1002         if (new->nregions != old->nregions)
1003                 return true;
1004
1005         for (i = 0; i < new->nregions; ++i) {
1006                 VhostUserMemoryRegion *new_r = &new->regions[i];
1007                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1008
1009                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1010                         return true;
1011                 if (new_r->memory_size != old_r->size)
1012                         return true;
1013                 if (new_r->userspace_addr != old_r->guest_user_addr)
1014                         return true;
1015         }
1016
1017         return false;
1018 }
1019
1020 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1021 static int
1022 vhost_user_postcopy_region_register(struct virtio_net *dev,
1023                 struct rte_vhost_mem_region *reg)
1024 {
1025         struct uffdio_register reg_struct;
1026
1027         /*
1028          * Let's register all the mmap'ed area to ensure
1029          * alignment on page boundary.
1030          */
1031         reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1032         reg_struct.range.len = reg->mmap_size;
1033         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1034
1035         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1036                                 &reg_struct)) {
1037                 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1038                                 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1039                                 (uint64_t)reg_struct.range.start,
1040                                 (uint64_t)reg_struct.range.start +
1041                                 (uint64_t)reg_struct.range.len - 1,
1042                                 dev->postcopy_ufd,
1043                                 strerror(errno));
1044                 return -1;
1045         }
1046
1047         VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1048                         (uint64_t)reg_struct.range.start,
1049                         (uint64_t)reg_struct.range.start +
1050                         (uint64_t)reg_struct.range.len - 1);
1051
1052         return 0;
1053 }
1054 #else
1055 static int
1056 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1057                 struct rte_vhost_mem_region *reg __rte_unused)
1058 {
1059         return -1;
1060 }
1061 #endif
1062
1063 static int
1064 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1065                 struct VhostUserMsg *msg)
1066 {
1067         struct VhostUserMemory *memory;
1068         struct rte_vhost_mem_region *reg;
1069         VhostUserMsg ack_msg;
1070         uint32_t i;
1071
1072         if (!dev->postcopy_listening)
1073                 return 0;
1074
1075         /*
1076          * We haven't a better way right now than sharing
1077          * DPDK's virtual address with Qemu, so that Qemu can
1078          * retrieve the region offset when handling userfaults.
1079          */
1080         memory = &msg->payload.memory;
1081         for (i = 0; i < memory->nregions; i++) {
1082                 reg = &dev->mem->regions[i];
1083                 memory->regions[i].userspace_addr = reg->host_user_addr;
1084         }
1085
1086         /* Send the addresses back to qemu */
1087         msg->fd_num = 0;
1088         send_vhost_reply(main_fd, msg);
1089
1090         /* Wait for qemu to acknolwedge it's got the addresses
1091          * we've got to wait before we're allowed to generate faults.
1092          */
1093         if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1094                 VHOST_LOG_CONFIG(ERR,
1095                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1096                 return -1;
1097         }
1098
1099         if (validate_msg_fds(&ack_msg, 0) != 0)
1100                 return -1;
1101
1102         if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1103                 VHOST_LOG_CONFIG(ERR,
1104                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1105                                 ack_msg.request.master);
1106                 return -1;
1107         }
1108
1109         /* Now userfault register and we can use the memory */
1110         for (i = 0; i < memory->nregions; i++) {
1111                 reg = &dev->mem->regions[i];
1112                 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1113                         return -1;
1114         }
1115
1116         return 0;
1117 }
1118
1119 static int
1120 vhost_user_mmap_region(struct virtio_net *dev,
1121                 struct rte_vhost_mem_region *region,
1122                 uint64_t mmap_offset)
1123 {
1124         void *mmap_addr;
1125         uint64_t mmap_size;
1126         uint64_t alignment;
1127         int populate;
1128
1129         /* Check for memory_size + mmap_offset overflow */
1130         if (mmap_offset >= -region->size) {
1131                 VHOST_LOG_CONFIG(ERR,
1132                                 "mmap_offset (%#"PRIx64") and memory_size "
1133                                 "(%#"PRIx64") overflow\n",
1134                                 mmap_offset, region->size);
1135                 return -1;
1136         }
1137
1138         mmap_size = region->size + mmap_offset;
1139
1140         /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1141          * argument aligned with hugepagesz at older longterm version Linux,
1142          * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1143          *
1144          * To avoid failure, make sure in caller to keep length aligned.
1145          */
1146         alignment = get_blk_size(region->fd);
1147         if (alignment == (uint64_t)-1) {
1148                 VHOST_LOG_CONFIG(ERR,
1149                                 "couldn't get hugepage size through fstat\n");
1150                 return -1;
1151         }
1152         mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1153         if (mmap_size == 0) {
1154                 /*
1155                  * It could happen if initial mmap_size + alignment overflows
1156                  * the sizeof uint64, which could happen if either mmap_size or
1157                  * alignment value is wrong.
1158                  *
1159                  * mmap() kernel implementation would return an error, but
1160                  * better catch it before and provide useful info in the logs.
1161                  */
1162                 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1163                                 "or alignment (0x%" PRIx64 ") is invalid\n",
1164                                 region->size + mmap_offset, alignment);
1165                 return -1;
1166         }
1167
1168         populate = dev->async_copy ? MAP_POPULATE : 0;
1169         mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1170                         MAP_SHARED | populate, region->fd, 0);
1171
1172         if (mmap_addr == MAP_FAILED) {
1173                 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1174                 return -1;
1175         }
1176
1177         region->mmap_addr = mmap_addr;
1178         region->mmap_size = mmap_size;
1179         region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1180
1181         if (dev->async_copy)
1182                 if (add_guest_pages(dev, region, alignment) < 0) {
1183                         VHOST_LOG_CONFIG(ERR,
1184                                         "adding guest pages to region failed.\n");
1185                         return -1;
1186                 }
1187
1188         VHOST_LOG_CONFIG(INFO,
1189                         "guest memory region size: 0x%" PRIx64 "\n"
1190                         "\t guest physical addr: 0x%" PRIx64 "\n"
1191                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1192                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1193                         "\t mmap addr : 0x%" PRIx64 "\n"
1194                         "\t mmap size : 0x%" PRIx64 "\n"
1195                         "\t mmap align: 0x%" PRIx64 "\n"
1196                         "\t mmap off  : 0x%" PRIx64 "\n",
1197                         region->size,
1198                         region->guest_phys_addr,
1199                         region->guest_user_addr,
1200                         region->host_user_addr,
1201                         (uint64_t)(uintptr_t)mmap_addr,
1202                         mmap_size,
1203                         alignment,
1204                         mmap_offset);
1205
1206         return 0;
1207 }
1208
1209 static int
1210 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1211                         int main_fd)
1212 {
1213         struct virtio_net *dev = *pdev;
1214         struct VhostUserMemory *memory = &msg->payload.memory;
1215         struct rte_vhost_mem_region *reg;
1216
1217         uint64_t mmap_offset;
1218         uint32_t i;
1219
1220         if (validate_msg_fds(msg, memory->nregions) != 0)
1221                 return RTE_VHOST_MSG_RESULT_ERR;
1222
1223         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1224                 VHOST_LOG_CONFIG(ERR,
1225                         "too many memory regions (%u)\n", memory->nregions);
1226                 goto close_msg_fds;
1227         }
1228
1229         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1230                 VHOST_LOG_CONFIG(INFO,
1231                         "(%d) memory regions not changed\n", dev->vid);
1232
1233                 close_msg_fds(msg);
1234
1235                 return RTE_VHOST_MSG_RESULT_OK;
1236         }
1237
1238         if (dev->mem) {
1239                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1240                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1241
1242                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1243                                 vdpa_dev->ops->dev_close(dev->vid);
1244                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1245                 }
1246                 free_mem_region(dev);
1247                 rte_free(dev->mem);
1248                 dev->mem = NULL;
1249         }
1250
1251         /* Flush IOTLB cache as previous HVAs are now invalid */
1252         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1253                 for (i = 0; i < dev->nr_vring; i++)
1254                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1255
1256         dev->nr_guest_pages = 0;
1257         if (dev->guest_pages == NULL) {
1258                 dev->max_guest_pages = 8;
1259                 dev->guest_pages = rte_zmalloc(NULL,
1260                                         dev->max_guest_pages *
1261                                         sizeof(struct guest_page),
1262                                         RTE_CACHE_LINE_SIZE);
1263                 if (dev->guest_pages == NULL) {
1264                         VHOST_LOG_CONFIG(ERR,
1265                                 "(%d) failed to allocate memory "
1266                                 "for dev->guest_pages\n",
1267                                 dev->vid);
1268                         goto close_msg_fds;
1269                 }
1270         }
1271
1272         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1273                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1274         if (dev->mem == NULL) {
1275                 VHOST_LOG_CONFIG(ERR,
1276                         "(%d) failed to allocate memory for dev->mem\n",
1277                         dev->vid);
1278                 goto free_guest_pages;
1279         }
1280
1281         for (i = 0; i < memory->nregions; i++) {
1282                 reg = &dev->mem->regions[i];
1283
1284                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1285                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1286                 reg->size            = memory->regions[i].memory_size;
1287                 reg->fd              = msg->fds[i];
1288
1289                 /*
1290                  * Assign invalid file descriptor value to avoid double
1291                  * closing on error path.
1292                  */
1293                 msg->fds[i] = -1;
1294
1295                 mmap_offset = memory->regions[i].mmap_offset;
1296
1297                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1298                         VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1299                         goto free_mem_table;
1300                 }
1301
1302                 dev->mem->nregions++;
1303         }
1304
1305         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1306                 goto free_mem_table;
1307
1308         for (i = 0; i < dev->nr_vring; i++) {
1309                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1310
1311                 if (!vq)
1312                         continue;
1313
1314                 if (vq->desc || vq->avail || vq->used) {
1315                         /*
1316                          * If the memory table got updated, the ring addresses
1317                          * need to be translated again as virtual addresses have
1318                          * changed.
1319                          */
1320                         vring_invalidate(dev, vq);
1321
1322                         dev = translate_ring_addresses(dev, i);
1323                         if (!dev) {
1324                                 dev = *pdev;
1325                                 goto free_mem_table;
1326                         }
1327
1328                         *pdev = dev;
1329                 }
1330         }
1331
1332         dump_guest_pages(dev);
1333
1334         return RTE_VHOST_MSG_RESULT_OK;
1335
1336 free_mem_table:
1337         free_mem_region(dev);
1338         rte_free(dev->mem);
1339         dev->mem = NULL;
1340 free_guest_pages:
1341         rte_free(dev->guest_pages);
1342         dev->guest_pages = NULL;
1343 close_msg_fds:
1344         close_msg_fds(msg);
1345         return RTE_VHOST_MSG_RESULT_ERR;
1346 }
1347
1348 static bool
1349 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1350 {
1351         bool rings_ok;
1352
1353         if (!vq)
1354                 return false;
1355
1356         if (vq_is_packed(dev))
1357                 rings_ok = vq->desc_packed && vq->driver_event &&
1358                         vq->device_event;
1359         else
1360                 rings_ok = vq->desc && vq->avail && vq->used;
1361
1362         return rings_ok &&
1363                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1364                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1365                vq->enabled;
1366 }
1367
1368 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1369
1370 static int
1371 virtio_is_ready(struct virtio_net *dev)
1372 {
1373         struct vhost_virtqueue *vq;
1374         uint32_t i, nr_vring = dev->nr_vring;
1375
1376         if (dev->flags & VIRTIO_DEV_READY)
1377                 return 1;
1378
1379         if (!dev->nr_vring)
1380                 return 0;
1381
1382         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1383                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1384
1385                 if (dev->nr_vring < nr_vring)
1386                         return 0;
1387         }
1388
1389         for (i = 0; i < nr_vring; i++) {
1390                 vq = dev->virtqueue[i];
1391
1392                 if (!vq_is_ready(dev, vq))
1393                         return 0;
1394         }
1395
1396         /* If supported, ensure the frontend is really done with config */
1397         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1398                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1399                         return 0;
1400
1401         dev->flags |= VIRTIO_DEV_READY;
1402
1403         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1404                 VHOST_LOG_CONFIG(INFO,
1405                         "virtio is now ready for processing.\n");
1406         return 1;
1407 }
1408
1409 static void *
1410 inflight_mem_alloc(const char *name, size_t size, int *fd)
1411 {
1412         void *ptr;
1413         int mfd = -1;
1414         char fname[20] = "/tmp/memfd-XXXXXX";
1415
1416         *fd = -1;
1417 #ifdef MEMFD_SUPPORTED
1418         mfd = memfd_create(name, MFD_CLOEXEC);
1419 #else
1420         RTE_SET_USED(name);
1421 #endif
1422         if (mfd == -1) {
1423                 mfd = mkstemp(fname);
1424                 if (mfd == -1) {
1425                         VHOST_LOG_CONFIG(ERR,
1426                                 "failed to get inflight buffer fd\n");
1427                         return NULL;
1428                 }
1429
1430                 unlink(fname);
1431         }
1432
1433         if (ftruncate(mfd, size) == -1) {
1434                 VHOST_LOG_CONFIG(ERR,
1435                         "failed to alloc inflight buffer\n");
1436                 close(mfd);
1437                 return NULL;
1438         }
1439
1440         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1441         if (ptr == MAP_FAILED) {
1442                 VHOST_LOG_CONFIG(ERR,
1443                         "failed to mmap inflight buffer\n");
1444                 close(mfd);
1445                 return NULL;
1446         }
1447
1448         *fd = mfd;
1449         return ptr;
1450 }
1451
1452 static uint32_t
1453 get_pervq_shm_size_split(uint16_t queue_size)
1454 {
1455         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1456                                   queue_size + sizeof(uint64_t) +
1457                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1458 }
1459
1460 static uint32_t
1461 get_pervq_shm_size_packed(uint16_t queue_size)
1462 {
1463         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1464                                   * queue_size + sizeof(uint64_t) +
1465                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1466                                   INFLIGHT_ALIGNMENT);
1467 }
1468
1469 static int
1470 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1471                            VhostUserMsg *msg,
1472                            int main_fd __rte_unused)
1473 {
1474         struct rte_vhost_inflight_info_packed *inflight_packed;
1475         uint64_t pervq_inflight_size, mmap_size;
1476         uint16_t num_queues, queue_size;
1477         struct virtio_net *dev = *pdev;
1478         int fd, i, j;
1479         void *addr;
1480
1481         if (msg->size != sizeof(msg->payload.inflight)) {
1482                 VHOST_LOG_CONFIG(ERR,
1483                         "invalid get_inflight_fd message size is %d\n",
1484                         msg->size);
1485                 return RTE_VHOST_MSG_RESULT_ERR;
1486         }
1487
1488         if (dev->inflight_info == NULL) {
1489                 dev->inflight_info = calloc(1,
1490                                             sizeof(struct inflight_mem_info));
1491                 if (!dev->inflight_info) {
1492                         VHOST_LOG_CONFIG(ERR,
1493                                 "failed to alloc dev inflight area\n");
1494                         return RTE_VHOST_MSG_RESULT_ERR;
1495                 }
1496                 dev->inflight_info->fd = -1;
1497         }
1498
1499         num_queues = msg->payload.inflight.num_queues;
1500         queue_size = msg->payload.inflight.queue_size;
1501
1502         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1503                 msg->payload.inflight.num_queues);
1504         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1505                 msg->payload.inflight.queue_size);
1506
1507         if (vq_is_packed(dev))
1508                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1509         else
1510                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1511
1512         mmap_size = num_queues * pervq_inflight_size;
1513         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1514         if (!addr) {
1515                 VHOST_LOG_CONFIG(ERR,
1516                         "failed to alloc vhost inflight area\n");
1517                         msg->payload.inflight.mmap_size = 0;
1518                 return RTE_VHOST_MSG_RESULT_ERR;
1519         }
1520         memset(addr, 0, mmap_size);
1521
1522         if (dev->inflight_info->addr) {
1523                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1524                 dev->inflight_info->addr = NULL;
1525         }
1526
1527         if (dev->inflight_info->fd >= 0) {
1528                 close(dev->inflight_info->fd);
1529                 dev->inflight_info->fd = -1;
1530         }
1531
1532         dev->inflight_info->addr = addr;
1533         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1534         dev->inflight_info->fd = msg->fds[0] = fd;
1535         msg->payload.inflight.mmap_offset = 0;
1536         msg->fd_num = 1;
1537
1538         if (vq_is_packed(dev)) {
1539                 for (i = 0; i < num_queues; i++) {
1540                         inflight_packed =
1541                                 (struct rte_vhost_inflight_info_packed *)addr;
1542                         inflight_packed->used_wrap_counter = 1;
1543                         inflight_packed->old_used_wrap_counter = 1;
1544                         for (j = 0; j < queue_size; j++)
1545                                 inflight_packed->desc[j].next = j + 1;
1546                         addr = (void *)((char *)addr + pervq_inflight_size);
1547                 }
1548         }
1549
1550         VHOST_LOG_CONFIG(INFO,
1551                 "send inflight mmap_size: %"PRIu64"\n",
1552                 msg->payload.inflight.mmap_size);
1553         VHOST_LOG_CONFIG(INFO,
1554                 "send inflight mmap_offset: %"PRIu64"\n",
1555                 msg->payload.inflight.mmap_offset);
1556         VHOST_LOG_CONFIG(INFO,
1557                 "send inflight fd: %d\n", msg->fds[0]);
1558
1559         return RTE_VHOST_MSG_RESULT_REPLY;
1560 }
1561
1562 static int
1563 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1564                            int main_fd __rte_unused)
1565 {
1566         uint64_t mmap_size, mmap_offset;
1567         uint16_t num_queues, queue_size;
1568         struct virtio_net *dev = *pdev;
1569         uint32_t pervq_inflight_size;
1570         struct vhost_virtqueue *vq;
1571         void *addr;
1572         int fd, i;
1573
1574         fd = msg->fds[0];
1575         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1576                 VHOST_LOG_CONFIG(ERR,
1577                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1578                         msg->size, fd);
1579                 return RTE_VHOST_MSG_RESULT_ERR;
1580         }
1581
1582         mmap_size = msg->payload.inflight.mmap_size;
1583         mmap_offset = msg->payload.inflight.mmap_offset;
1584         num_queues = msg->payload.inflight.num_queues;
1585         queue_size = msg->payload.inflight.queue_size;
1586
1587         if (vq_is_packed(dev))
1588                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1589         else
1590                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1591
1592         VHOST_LOG_CONFIG(INFO,
1593                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1594         VHOST_LOG_CONFIG(INFO,
1595                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1596         VHOST_LOG_CONFIG(INFO,
1597                 "set_inflight_fd num_queues: %u\n", num_queues);
1598         VHOST_LOG_CONFIG(INFO,
1599                 "set_inflight_fd queue_size: %u\n", queue_size);
1600         VHOST_LOG_CONFIG(INFO,
1601                 "set_inflight_fd fd: %d\n", fd);
1602         VHOST_LOG_CONFIG(INFO,
1603                 "set_inflight_fd pervq_inflight_size: %d\n",
1604                 pervq_inflight_size);
1605
1606         if (!dev->inflight_info) {
1607                 dev->inflight_info = calloc(1,
1608                                             sizeof(struct inflight_mem_info));
1609                 if (dev->inflight_info == NULL) {
1610                         VHOST_LOG_CONFIG(ERR,
1611                                 "failed to alloc dev inflight area\n");
1612                         return RTE_VHOST_MSG_RESULT_ERR;
1613                 }
1614                 dev->inflight_info->fd = -1;
1615         }
1616
1617         if (dev->inflight_info->addr) {
1618                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1619                 dev->inflight_info->addr = NULL;
1620         }
1621
1622         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1623                     fd, mmap_offset);
1624         if (addr == MAP_FAILED) {
1625                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1626                 return RTE_VHOST_MSG_RESULT_ERR;
1627         }
1628
1629         if (dev->inflight_info->fd >= 0) {
1630                 close(dev->inflight_info->fd);
1631                 dev->inflight_info->fd = -1;
1632         }
1633
1634         dev->inflight_info->fd = fd;
1635         dev->inflight_info->addr = addr;
1636         dev->inflight_info->size = mmap_size;
1637
1638         for (i = 0; i < num_queues; i++) {
1639                 vq = dev->virtqueue[i];
1640                 if (!vq)
1641                         continue;
1642
1643                 if (vq_is_packed(dev)) {
1644                         vq->inflight_packed = addr;
1645                         vq->inflight_packed->desc_num = queue_size;
1646                 } else {
1647                         vq->inflight_split = addr;
1648                         vq->inflight_split->desc_num = queue_size;
1649                 }
1650                 addr = (void *)((char *)addr + pervq_inflight_size);
1651         }
1652
1653         return RTE_VHOST_MSG_RESULT_OK;
1654 }
1655
1656 static int
1657 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1658                         int main_fd __rte_unused)
1659 {
1660         struct virtio_net *dev = *pdev;
1661         struct vhost_vring_file file;
1662         struct vhost_virtqueue *vq;
1663         int expected_fds;
1664
1665         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1666         if (validate_msg_fds(msg, expected_fds) != 0)
1667                 return RTE_VHOST_MSG_RESULT_ERR;
1668
1669         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1670         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1671                 file.fd = VIRTIO_INVALID_EVENTFD;
1672         else
1673                 file.fd = msg->fds[0];
1674         VHOST_LOG_CONFIG(INFO,
1675                 "vring call idx:%d file:%d\n", file.index, file.fd);
1676
1677         vq = dev->virtqueue[file.index];
1678
1679         if (vq->ready) {
1680                 vq->ready = false;
1681                 vhost_user_notify_queue_state(dev, file.index, 0);
1682         }
1683
1684         if (vq->callfd >= 0)
1685                 close(vq->callfd);
1686
1687         vq->callfd = file.fd;
1688
1689         return RTE_VHOST_MSG_RESULT_OK;
1690 }
1691
1692 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1693                         struct VhostUserMsg *msg,
1694                         int main_fd __rte_unused)
1695 {
1696         int expected_fds;
1697
1698         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1699         if (validate_msg_fds(msg, expected_fds) != 0)
1700                 return RTE_VHOST_MSG_RESULT_ERR;
1701
1702         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1703                 close(msg->fds[0]);
1704         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1705
1706         return RTE_VHOST_MSG_RESULT_OK;
1707 }
1708
1709 static int
1710 resubmit_desc_compare(const void *a, const void *b)
1711 {
1712         const struct rte_vhost_resubmit_desc *desc0 = a;
1713         const struct rte_vhost_resubmit_desc *desc1 = b;
1714
1715         if (desc1->counter > desc0->counter)
1716                 return 1;
1717
1718         return -1;
1719 }
1720
1721 static int
1722 vhost_check_queue_inflights_split(struct virtio_net *dev,
1723                                   struct vhost_virtqueue *vq)
1724 {
1725         uint16_t i;
1726         uint16_t resubmit_num = 0, last_io, num;
1727         struct vring_used *used = vq->used;
1728         struct rte_vhost_resubmit_info *resubmit;
1729         struct rte_vhost_inflight_info_split *inflight_split;
1730
1731         if (!(dev->protocol_features &
1732             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1733                 return RTE_VHOST_MSG_RESULT_OK;
1734
1735         /* The frontend may still not support the inflight feature
1736          * although we negotiate the protocol feature.
1737          */
1738         if ((!vq->inflight_split))
1739                 return RTE_VHOST_MSG_RESULT_OK;
1740
1741         if (!vq->inflight_split->version) {
1742                 vq->inflight_split->version = INFLIGHT_VERSION;
1743                 return RTE_VHOST_MSG_RESULT_OK;
1744         }
1745
1746         if (vq->resubmit_inflight)
1747                 return RTE_VHOST_MSG_RESULT_OK;
1748
1749         inflight_split = vq->inflight_split;
1750         vq->global_counter = 0;
1751         last_io = inflight_split->last_inflight_io;
1752
1753         if (inflight_split->used_idx != used->idx) {
1754                 inflight_split->desc[last_io].inflight = 0;
1755                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1756                 inflight_split->used_idx = used->idx;
1757         }
1758
1759         for (i = 0; i < inflight_split->desc_num; i++) {
1760                 if (inflight_split->desc[i].inflight == 1)
1761                         resubmit_num++;
1762         }
1763
1764         vq->last_avail_idx += resubmit_num;
1765
1766         if (resubmit_num) {
1767                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1768                 if (!resubmit) {
1769                         VHOST_LOG_CONFIG(ERR,
1770                                 "failed to allocate memory for resubmit info.\n");
1771                         return RTE_VHOST_MSG_RESULT_ERR;
1772                 }
1773
1774                 resubmit->resubmit_list = calloc(resubmit_num,
1775                         sizeof(struct rte_vhost_resubmit_desc));
1776                 if (!resubmit->resubmit_list) {
1777                         VHOST_LOG_CONFIG(ERR,
1778                                 "failed to allocate memory for inflight desc.\n");
1779                         free(resubmit);
1780                         return RTE_VHOST_MSG_RESULT_ERR;
1781                 }
1782
1783                 num = 0;
1784                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1785                         if (vq->inflight_split->desc[i].inflight == 1) {
1786                                 resubmit->resubmit_list[num].index = i;
1787                                 resubmit->resubmit_list[num].counter =
1788                                         inflight_split->desc[i].counter;
1789                                 num++;
1790                         }
1791                 }
1792                 resubmit->resubmit_num = num;
1793
1794                 if (resubmit->resubmit_num > 1)
1795                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1796                               sizeof(struct rte_vhost_resubmit_desc),
1797                               resubmit_desc_compare);
1798
1799                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1800                 vq->resubmit_inflight = resubmit;
1801         }
1802
1803         return RTE_VHOST_MSG_RESULT_OK;
1804 }
1805
1806 static int
1807 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1808                                    struct vhost_virtqueue *vq)
1809 {
1810         uint16_t i;
1811         uint16_t resubmit_num = 0, old_used_idx, num;
1812         struct rte_vhost_resubmit_info *resubmit;
1813         struct rte_vhost_inflight_info_packed *inflight_packed;
1814
1815         if (!(dev->protocol_features &
1816             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1817                 return RTE_VHOST_MSG_RESULT_OK;
1818
1819         /* The frontend may still not support the inflight feature
1820          * although we negotiate the protocol feature.
1821          */
1822         if ((!vq->inflight_packed))
1823                 return RTE_VHOST_MSG_RESULT_OK;
1824
1825         if (!vq->inflight_packed->version) {
1826                 vq->inflight_packed->version = INFLIGHT_VERSION;
1827                 return RTE_VHOST_MSG_RESULT_OK;
1828         }
1829
1830         if (vq->resubmit_inflight)
1831                 return RTE_VHOST_MSG_RESULT_OK;
1832
1833         inflight_packed = vq->inflight_packed;
1834         vq->global_counter = 0;
1835         old_used_idx = inflight_packed->old_used_idx;
1836
1837         if (inflight_packed->used_idx != old_used_idx) {
1838                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1839                         inflight_packed->old_used_idx =
1840                                 inflight_packed->used_idx;
1841                         inflight_packed->old_used_wrap_counter =
1842                                 inflight_packed->used_wrap_counter;
1843                         inflight_packed->old_free_head =
1844                                 inflight_packed->free_head;
1845                 } else {
1846                         inflight_packed->used_idx =
1847                                 inflight_packed->old_used_idx;
1848                         inflight_packed->used_wrap_counter =
1849                                 inflight_packed->old_used_wrap_counter;
1850                         inflight_packed->free_head =
1851                                 inflight_packed->old_free_head;
1852                 }
1853         }
1854
1855         for (i = 0; i < inflight_packed->desc_num; i++) {
1856                 if (inflight_packed->desc[i].inflight == 1)
1857                         resubmit_num++;
1858         }
1859
1860         if (resubmit_num) {
1861                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1862                 if (resubmit == NULL) {
1863                         VHOST_LOG_CONFIG(ERR,
1864                                 "failed to allocate memory for resubmit info.\n");
1865                         return RTE_VHOST_MSG_RESULT_ERR;
1866                 }
1867
1868                 resubmit->resubmit_list = calloc(resubmit_num,
1869                         sizeof(struct rte_vhost_resubmit_desc));
1870                 if (resubmit->resubmit_list == NULL) {
1871                         VHOST_LOG_CONFIG(ERR,
1872                                 "failed to allocate memory for resubmit desc.\n");
1873                         free(resubmit);
1874                         return RTE_VHOST_MSG_RESULT_ERR;
1875                 }
1876
1877                 num = 0;
1878                 for (i = 0; i < inflight_packed->desc_num; i++) {
1879                         if (vq->inflight_packed->desc[i].inflight == 1) {
1880                                 resubmit->resubmit_list[num].index = i;
1881                                 resubmit->resubmit_list[num].counter =
1882                                         inflight_packed->desc[i].counter;
1883                                 num++;
1884                         }
1885                 }
1886                 resubmit->resubmit_num = num;
1887
1888                 if (resubmit->resubmit_num > 1)
1889                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1890                               sizeof(struct rte_vhost_resubmit_desc),
1891                               resubmit_desc_compare);
1892
1893                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1894                 vq->resubmit_inflight = resubmit;
1895         }
1896
1897         return RTE_VHOST_MSG_RESULT_OK;
1898 }
1899
1900 static int
1901 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1902                         int main_fd __rte_unused)
1903 {
1904         struct virtio_net *dev = *pdev;
1905         struct vhost_vring_file file;
1906         struct vhost_virtqueue *vq;
1907         int expected_fds;
1908
1909         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1910         if (validate_msg_fds(msg, expected_fds) != 0)
1911                 return RTE_VHOST_MSG_RESULT_ERR;
1912
1913         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1914         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1915                 file.fd = VIRTIO_INVALID_EVENTFD;
1916         else
1917                 file.fd = msg->fds[0];
1918         VHOST_LOG_CONFIG(INFO,
1919                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1920
1921         /* Interpret ring addresses only when ring is started. */
1922         dev = translate_ring_addresses(dev, file.index);
1923         if (!dev) {
1924                 if (file.fd != VIRTIO_INVALID_EVENTFD)
1925                         close(file.fd);
1926
1927                 return RTE_VHOST_MSG_RESULT_ERR;
1928         }
1929
1930         *pdev = dev;
1931
1932         vq = dev->virtqueue[file.index];
1933
1934         /*
1935          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1936          * the ring starts already enabled. Otherwise, it is enabled via
1937          * the SET_VRING_ENABLE message.
1938          */
1939         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1940                 vq->enabled = true;
1941         }
1942
1943         if (vq->ready) {
1944                 vq->ready = false;
1945                 vhost_user_notify_queue_state(dev, file.index, 0);
1946         }
1947
1948         if (vq->kickfd >= 0)
1949                 close(vq->kickfd);
1950         vq->kickfd = file.fd;
1951
1952         if (vq_is_packed(dev)) {
1953                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1954                         VHOST_LOG_CONFIG(ERR,
1955                                 "failed to inflights for vq: %d\n", file.index);
1956                         return RTE_VHOST_MSG_RESULT_ERR;
1957                 }
1958         } else {
1959                 if (vhost_check_queue_inflights_split(dev, vq)) {
1960                         VHOST_LOG_CONFIG(ERR,
1961                                 "failed to inflights for vq: %d\n", file.index);
1962                         return RTE_VHOST_MSG_RESULT_ERR;
1963                 }
1964         }
1965
1966         return RTE_VHOST_MSG_RESULT_OK;
1967 }
1968
1969 /*
1970  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1971  */
1972 static int
1973 vhost_user_get_vring_base(struct virtio_net **pdev,
1974                         struct VhostUserMsg *msg,
1975                         int main_fd __rte_unused)
1976 {
1977         struct virtio_net *dev = *pdev;
1978         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1979         uint64_t val;
1980
1981         if (validate_msg_fds(msg, 0) != 0)
1982                 return RTE_VHOST_MSG_RESULT_ERR;
1983
1984         /* We have to stop the queue (virtio) if it is running. */
1985         vhost_destroy_device_notify(dev);
1986
1987         dev->flags &= ~VIRTIO_DEV_READY;
1988         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1989
1990         /* Here we are safe to get the indexes */
1991         if (vq_is_packed(dev)) {
1992                 /*
1993                  * Bit[0:14]: avail index
1994                  * Bit[15]: avail wrap counter
1995                  */
1996                 val = vq->last_avail_idx & 0x7fff;
1997                 val |= vq->avail_wrap_counter << 15;
1998                 msg->payload.state.num = val;
1999         } else {
2000                 msg->payload.state.num = vq->last_avail_idx;
2001         }
2002
2003         VHOST_LOG_CONFIG(INFO,
2004                 "vring base idx:%d file:%d\n", msg->payload.state.index,
2005                 msg->payload.state.num);
2006         /*
2007          * Based on current qemu vhost-user implementation, this message is
2008          * sent and only sent in vhost_vring_stop.
2009          * TODO: cleanup the vring, it isn't usable since here.
2010          */
2011         if (vq->kickfd >= 0)
2012                 close(vq->kickfd);
2013
2014         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2015
2016         if (vq->callfd >= 0)
2017                 close(vq->callfd);
2018
2019         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2020
2021         vq->signalled_used_valid = false;
2022
2023         if (vq_is_packed(dev)) {
2024                 rte_free(vq->shadow_used_packed);
2025                 vq->shadow_used_packed = NULL;
2026         } else {
2027                 rte_free(vq->shadow_used_split);
2028                 vq->shadow_used_split = NULL;
2029         }
2030
2031         rte_free(vq->batch_copy_elems);
2032         vq->batch_copy_elems = NULL;
2033
2034         rte_free(vq->log_cache);
2035         vq->log_cache = NULL;
2036
2037         msg->size = sizeof(msg->payload.state);
2038         msg->fd_num = 0;
2039
2040         vring_invalidate(dev, vq);
2041
2042         return RTE_VHOST_MSG_RESULT_REPLY;
2043 }
2044
2045 /*
2046  * when virtio queues are ready to work, qemu will send us to
2047  * enable the virtio queue pair.
2048  */
2049 static int
2050 vhost_user_set_vring_enable(struct virtio_net **pdev,
2051                         struct VhostUserMsg *msg,
2052                         int main_fd __rte_unused)
2053 {
2054         struct virtio_net *dev = *pdev;
2055         bool enable = !!msg->payload.state.num;
2056         int index = (int)msg->payload.state.index;
2057
2058         if (validate_msg_fds(msg, 0) != 0)
2059                 return RTE_VHOST_MSG_RESULT_ERR;
2060
2061         VHOST_LOG_CONFIG(INFO,
2062                 "set queue enable: %d to qp idx: %d\n",
2063                 enable, index);
2064
2065         if (enable && dev->virtqueue[index]->async_registered) {
2066                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2067                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2068                         "async inflight packets must be completed first\n");
2069                         return RTE_VHOST_MSG_RESULT_ERR;
2070                 }
2071         }
2072
2073         dev->virtqueue[index]->enabled = enable;
2074
2075         return RTE_VHOST_MSG_RESULT_OK;
2076 }
2077
2078 static int
2079 vhost_user_get_protocol_features(struct virtio_net **pdev,
2080                         struct VhostUserMsg *msg,
2081                         int main_fd __rte_unused)
2082 {
2083         struct virtio_net *dev = *pdev;
2084         uint64_t features, protocol_features;
2085
2086         if (validate_msg_fds(msg, 0) != 0)
2087                 return RTE_VHOST_MSG_RESULT_ERR;
2088
2089         rte_vhost_driver_get_features(dev->ifname, &features);
2090         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2091
2092         msg->payload.u64 = protocol_features;
2093         msg->size = sizeof(msg->payload.u64);
2094         msg->fd_num = 0;
2095
2096         return RTE_VHOST_MSG_RESULT_REPLY;
2097 }
2098
2099 static int
2100 vhost_user_set_protocol_features(struct virtio_net **pdev,
2101                         struct VhostUserMsg *msg,
2102                         int main_fd __rte_unused)
2103 {
2104         struct virtio_net *dev = *pdev;
2105         uint64_t protocol_features = msg->payload.u64;
2106         uint64_t slave_protocol_features = 0;
2107
2108         if (validate_msg_fds(msg, 0) != 0)
2109                 return RTE_VHOST_MSG_RESULT_ERR;
2110
2111         rte_vhost_driver_get_protocol_features(dev->ifname,
2112                         &slave_protocol_features);
2113         if (protocol_features & ~slave_protocol_features) {
2114                 VHOST_LOG_CONFIG(ERR,
2115                         "(%d) received invalid protocol features.\n",
2116                         dev->vid);
2117                 return RTE_VHOST_MSG_RESULT_ERR;
2118         }
2119
2120         dev->protocol_features = protocol_features;
2121         VHOST_LOG_CONFIG(INFO,
2122                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2123                 dev->protocol_features);
2124
2125         return RTE_VHOST_MSG_RESULT_OK;
2126 }
2127
2128 static int
2129 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2130                         int main_fd __rte_unused)
2131 {
2132         struct virtio_net *dev = *pdev;
2133         int fd = msg->fds[0];
2134         uint64_t size, off;
2135         void *addr;
2136         uint32_t i;
2137
2138         if (validate_msg_fds(msg, 1) != 0)
2139                 return RTE_VHOST_MSG_RESULT_ERR;
2140
2141         if (fd < 0) {
2142                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2143                 return RTE_VHOST_MSG_RESULT_ERR;
2144         }
2145
2146         if (msg->size != sizeof(VhostUserLog)) {
2147                 VHOST_LOG_CONFIG(ERR,
2148                         "invalid log base msg size: %"PRId32" != %d\n",
2149                         msg->size, (int)sizeof(VhostUserLog));
2150                 goto close_msg_fds;
2151         }
2152
2153         size = msg->payload.log.mmap_size;
2154         off  = msg->payload.log.mmap_offset;
2155
2156         /* Check for mmap size and offset overflow. */
2157         if (off >= -size) {
2158                 VHOST_LOG_CONFIG(ERR,
2159                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2160                         off, size);
2161                 goto close_msg_fds;
2162         }
2163
2164         VHOST_LOG_CONFIG(INFO,
2165                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2166                 size, off);
2167
2168         /*
2169          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2170          * fail when offset is not page size aligned.
2171          */
2172         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2173         close(fd);
2174         if (addr == MAP_FAILED) {
2175                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2176                 return RTE_VHOST_MSG_RESULT_ERR;
2177         }
2178
2179         /*
2180          * Free previously mapped log memory on occasionally
2181          * multiple VHOST_USER_SET_LOG_BASE.
2182          */
2183         if (dev->log_addr) {
2184                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2185         }
2186         dev->log_addr = (uint64_t)(uintptr_t)addr;
2187         dev->log_base = dev->log_addr + off;
2188         dev->log_size = size;
2189
2190         for (i = 0; i < dev->nr_vring; i++) {
2191                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2192
2193                 rte_free(vq->log_cache);
2194                 vq->log_cache = NULL;
2195                 vq->log_cache_nb_elem = 0;
2196                 vq->log_cache = rte_zmalloc("vq log cache",
2197                                 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2198                                 0);
2199                 /*
2200                  * If log cache alloc fail, don't fail migration, but no
2201                  * caching will be done, which will impact performance
2202                  */
2203                 if (!vq->log_cache)
2204                         VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2205         }
2206
2207         /*
2208          * The spec is not clear about it (yet), but QEMU doesn't expect
2209          * any payload in the reply.
2210          */
2211         msg->size = 0;
2212         msg->fd_num = 0;
2213
2214         return RTE_VHOST_MSG_RESULT_REPLY;
2215
2216 close_msg_fds:
2217         close_msg_fds(msg);
2218         return RTE_VHOST_MSG_RESULT_ERR;
2219 }
2220
2221 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2222                         struct VhostUserMsg *msg,
2223                         int main_fd __rte_unused)
2224 {
2225         if (validate_msg_fds(msg, 1) != 0)
2226                 return RTE_VHOST_MSG_RESULT_ERR;
2227
2228         close(msg->fds[0]);
2229         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2230
2231         return RTE_VHOST_MSG_RESULT_OK;
2232 }
2233
2234 /*
2235  * An rarp packet is constructed and broadcasted to notify switches about
2236  * the new location of the migrated VM, so that packets from outside will
2237  * not be lost after migration.
2238  *
2239  * However, we don't actually "send" a rarp packet here, instead, we set
2240  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2241  */
2242 static int
2243 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2244                         int main_fd __rte_unused)
2245 {
2246         struct virtio_net *dev = *pdev;
2247         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2248         struct rte_vdpa_device *vdpa_dev;
2249
2250         if (validate_msg_fds(msg, 0) != 0)
2251                 return RTE_VHOST_MSG_RESULT_ERR;
2252
2253         VHOST_LOG_CONFIG(DEBUG,
2254                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2255                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2256         memcpy(dev->mac.addr_bytes, mac, 6);
2257
2258         /*
2259          * Set the flag to inject a RARP broadcast packet at
2260          * rte_vhost_dequeue_burst().
2261          *
2262          * __ATOMIC_RELEASE ordering is for making sure the mac is
2263          * copied before the flag is set.
2264          */
2265         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2266         vdpa_dev = dev->vdpa_dev;
2267         if (vdpa_dev && vdpa_dev->ops->migration_done)
2268                 vdpa_dev->ops->migration_done(dev->vid);
2269
2270         return RTE_VHOST_MSG_RESULT_OK;
2271 }
2272
2273 static int
2274 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2275                         int main_fd __rte_unused)
2276 {
2277         struct virtio_net *dev = *pdev;
2278
2279         if (validate_msg_fds(msg, 0) != 0)
2280                 return RTE_VHOST_MSG_RESULT_ERR;
2281
2282         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2283                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2284                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2285                                 msg->payload.u64);
2286
2287                 return RTE_VHOST_MSG_RESULT_ERR;
2288         }
2289
2290         dev->mtu = msg->payload.u64;
2291
2292         return RTE_VHOST_MSG_RESULT_OK;
2293 }
2294
2295 static int
2296 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2297                         int main_fd __rte_unused)
2298 {
2299         struct virtio_net *dev = *pdev;
2300         int fd = msg->fds[0];
2301
2302         if (validate_msg_fds(msg, 1) != 0)
2303                 return RTE_VHOST_MSG_RESULT_ERR;
2304
2305         if (fd < 0) {
2306                 VHOST_LOG_CONFIG(ERR,
2307                                 "Invalid file descriptor for slave channel (%d)\n",
2308                                 fd);
2309                 return RTE_VHOST_MSG_RESULT_ERR;
2310         }
2311
2312         if (dev->slave_req_fd >= 0)
2313                 close(dev->slave_req_fd);
2314
2315         dev->slave_req_fd = fd;
2316
2317         return RTE_VHOST_MSG_RESULT_OK;
2318 }
2319
2320 static int
2321 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2322 {
2323         struct vhost_vring_addr *ra;
2324         uint64_t start, end, len;
2325
2326         start = imsg->iova;
2327         end = start + imsg->size;
2328
2329         ra = &vq->ring_addrs;
2330         len = sizeof(struct vring_desc) * vq->size;
2331         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2332                 return 1;
2333
2334         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2335         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2336                 return 1;
2337
2338         len = sizeof(struct vring_used) +
2339                sizeof(struct vring_used_elem) * vq->size;
2340         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2341                 return 1;
2342
2343         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2344                 len = sizeof(uint64_t);
2345                 if (ra->log_guest_addr < end &&
2346                     (ra->log_guest_addr + len) > start)
2347                         return 1;
2348         }
2349
2350         return 0;
2351 }
2352
2353 static int
2354 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2355 {
2356         struct vhost_vring_addr *ra;
2357         uint64_t start, end, len;
2358
2359         start = imsg->iova;
2360         end = start + imsg->size;
2361
2362         ra = &vq->ring_addrs;
2363         len = sizeof(struct vring_packed_desc) * vq->size;
2364         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2365                 return 1;
2366
2367         len = sizeof(struct vring_packed_desc_event);
2368         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2369                 return 1;
2370
2371         len = sizeof(struct vring_packed_desc_event);
2372         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2373                 return 1;
2374
2375         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2376                 len = sizeof(uint64_t);
2377                 if (ra->log_guest_addr < end &&
2378                     (ra->log_guest_addr + len) > start)
2379                         return 1;
2380         }
2381
2382         return 0;
2383 }
2384
2385 static int is_vring_iotlb(struct virtio_net *dev,
2386                           struct vhost_virtqueue *vq,
2387                           struct vhost_iotlb_msg *imsg)
2388 {
2389         if (vq_is_packed(dev))
2390                 return is_vring_iotlb_packed(vq, imsg);
2391         else
2392                 return is_vring_iotlb_split(vq, imsg);
2393 }
2394
2395 static int
2396 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2397                         int main_fd __rte_unused)
2398 {
2399         struct virtio_net *dev = *pdev;
2400         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2401         uint16_t i;
2402         uint64_t vva, len;
2403
2404         if (validate_msg_fds(msg, 0) != 0)
2405                 return RTE_VHOST_MSG_RESULT_ERR;
2406
2407         switch (imsg->type) {
2408         case VHOST_IOTLB_UPDATE:
2409                 len = imsg->size;
2410                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2411                 if (!vva)
2412                         return RTE_VHOST_MSG_RESULT_ERR;
2413
2414                 for (i = 0; i < dev->nr_vring; i++) {
2415                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2416
2417                         if (!vq)
2418                                 continue;
2419
2420                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2421                                         len, imsg->perm);
2422
2423                         if (is_vring_iotlb(dev, vq, imsg))
2424                                 *pdev = dev = translate_ring_addresses(dev, i);
2425                 }
2426                 break;
2427         case VHOST_IOTLB_INVALIDATE:
2428                 for (i = 0; i < dev->nr_vring; i++) {
2429                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2430
2431                         if (!vq)
2432                                 continue;
2433
2434                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2435                                         imsg->size);
2436
2437                         if (is_vring_iotlb(dev, vq, imsg))
2438                                 vring_invalidate(dev, vq);
2439                 }
2440                 break;
2441         default:
2442                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2443                                 imsg->type);
2444                 return RTE_VHOST_MSG_RESULT_ERR;
2445         }
2446
2447         return RTE_VHOST_MSG_RESULT_OK;
2448 }
2449
2450 static int
2451 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2452                         struct VhostUserMsg *msg,
2453                         int main_fd __rte_unused)
2454 {
2455         struct virtio_net *dev = *pdev;
2456 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2457         struct uffdio_api api_struct;
2458
2459         if (validate_msg_fds(msg, 0) != 0)
2460                 return RTE_VHOST_MSG_RESULT_ERR;
2461
2462         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2463
2464         if (dev->postcopy_ufd == -1) {
2465                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2466                         strerror(errno));
2467                 return RTE_VHOST_MSG_RESULT_ERR;
2468         }
2469         api_struct.api = UFFD_API;
2470         api_struct.features = 0;
2471         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2472                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2473                         strerror(errno));
2474                 close(dev->postcopy_ufd);
2475                 dev->postcopy_ufd = -1;
2476                 return RTE_VHOST_MSG_RESULT_ERR;
2477         }
2478         msg->fds[0] = dev->postcopy_ufd;
2479         msg->fd_num = 1;
2480
2481         return RTE_VHOST_MSG_RESULT_REPLY;
2482 #else
2483         dev->postcopy_ufd = -1;
2484         msg->fd_num = 0;
2485
2486         return RTE_VHOST_MSG_RESULT_ERR;
2487 #endif
2488 }
2489
2490 static int
2491 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2492                         struct VhostUserMsg *msg __rte_unused,
2493                         int main_fd __rte_unused)
2494 {
2495         struct virtio_net *dev = *pdev;
2496
2497         if (validate_msg_fds(msg, 0) != 0)
2498                 return RTE_VHOST_MSG_RESULT_ERR;
2499
2500         if (dev->mem && dev->mem->nregions) {
2501                 VHOST_LOG_CONFIG(ERR,
2502                         "Regions already registered at postcopy-listen\n");
2503                 return RTE_VHOST_MSG_RESULT_ERR;
2504         }
2505         dev->postcopy_listening = 1;
2506
2507         return RTE_VHOST_MSG_RESULT_OK;
2508 }
2509
2510 static int
2511 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2512                         int main_fd __rte_unused)
2513 {
2514         struct virtio_net *dev = *pdev;
2515
2516         if (validate_msg_fds(msg, 0) != 0)
2517                 return RTE_VHOST_MSG_RESULT_ERR;
2518
2519         dev->postcopy_listening = 0;
2520         if (dev->postcopy_ufd >= 0) {
2521                 close(dev->postcopy_ufd);
2522                 dev->postcopy_ufd = -1;
2523         }
2524
2525         msg->payload.u64 = 0;
2526         msg->size = sizeof(msg->payload.u64);
2527         msg->fd_num = 0;
2528
2529         return RTE_VHOST_MSG_RESULT_REPLY;
2530 }
2531
2532 static int
2533 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2534                       int main_fd __rte_unused)
2535 {
2536         struct virtio_net *dev = *pdev;
2537
2538         if (validate_msg_fds(msg, 0) != 0)
2539                 return RTE_VHOST_MSG_RESULT_ERR;
2540
2541         msg->payload.u64 = dev->status;
2542         msg->size = sizeof(msg->payload.u64);
2543         msg->fd_num = 0;
2544
2545         return RTE_VHOST_MSG_RESULT_REPLY;
2546 }
2547
2548 static int
2549 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2550                         int main_fd __rte_unused)
2551 {
2552         struct virtio_net *dev = *pdev;
2553
2554         if (validate_msg_fds(msg, 0) != 0)
2555                 return RTE_VHOST_MSG_RESULT_ERR;
2556
2557         /* As per Virtio specification, the device status is 8bits long */
2558         if (msg->payload.u64 > UINT8_MAX) {
2559                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2560                                 msg->payload.u64);
2561                 return RTE_VHOST_MSG_RESULT_ERR;
2562         }
2563
2564         dev->status = msg->payload.u64;
2565
2566         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2567             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2568                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2569                 /*
2570                  * Clear the bit to let the driver know about the feature
2571                  * negotiation failure
2572                  */
2573                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2574         }
2575
2576         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2577                         "\t-RESET: %u\n"
2578                         "\t-ACKNOWLEDGE: %u\n"
2579                         "\t-DRIVER: %u\n"
2580                         "\t-FEATURES_OK: %u\n"
2581                         "\t-DRIVER_OK: %u\n"
2582                         "\t-DEVICE_NEED_RESET: %u\n"
2583                         "\t-FAILED: %u\n",
2584                         dev->status,
2585                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2586                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2587                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2588                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2589                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2590                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2591                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2592
2593         return RTE_VHOST_MSG_RESULT_OK;
2594 }
2595
2596 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2597                                         struct VhostUserMsg *msg,
2598                                         int main_fd);
2599 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2600         [VHOST_USER_NONE] = NULL,
2601         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2602         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2603         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2604         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2605         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2606         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2607         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2608         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2609         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2610         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2611         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2612         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2613         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2614         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2615         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2616         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2617         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2618         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2619         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2620         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2621         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2622         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2623         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2624         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2625         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2626         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2627         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2628         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2629         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2630 };
2631
2632 /* return bytes# of read on success or negative val on failure. */
2633 static int
2634 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2635 {
2636         int ret;
2637
2638         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2639                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2640         if (ret <= 0) {
2641                 return ret;
2642         } else if (ret != VHOST_USER_HDR_SIZE) {
2643                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2644                 close_msg_fds(msg);
2645                 return -1;
2646         }
2647
2648         if (msg->size) {
2649                 if (msg->size > sizeof(msg->payload)) {
2650                         VHOST_LOG_CONFIG(ERR,
2651                                 "invalid msg size: %d\n", msg->size);
2652                         return -1;
2653                 }
2654                 ret = read(sockfd, &msg->payload, msg->size);
2655                 if (ret <= 0)
2656                         return ret;
2657                 if (ret != (int)msg->size) {
2658                         VHOST_LOG_CONFIG(ERR,
2659                                 "read control message failed\n");
2660                         return -1;
2661                 }
2662         }
2663
2664         return ret;
2665 }
2666
2667 static int
2668 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2669 {
2670         if (!msg)
2671                 return 0;
2672
2673         return send_fd_message(sockfd, (char *)msg,
2674                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2675 }
2676
2677 static int
2678 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2679 {
2680         if (!msg)
2681                 return 0;
2682
2683         msg->flags &= ~VHOST_USER_VERSION_MASK;
2684         msg->flags &= ~VHOST_USER_NEED_REPLY;
2685         msg->flags |= VHOST_USER_VERSION;
2686         msg->flags |= VHOST_USER_REPLY_MASK;
2687
2688         return send_vhost_message(sockfd, msg);
2689 }
2690
2691 static int
2692 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2693 {
2694         int ret;
2695
2696         if (msg->flags & VHOST_USER_NEED_REPLY)
2697                 rte_spinlock_lock(&dev->slave_req_lock);
2698
2699         ret = send_vhost_message(dev->slave_req_fd, msg);
2700         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2701                 rte_spinlock_unlock(&dev->slave_req_lock);
2702
2703         return ret;
2704 }
2705
2706 /*
2707  * Allocate a queue pair if it hasn't been allocated yet
2708  */
2709 static int
2710 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2711                         struct VhostUserMsg *msg)
2712 {
2713         uint32_t vring_idx;
2714
2715         switch (msg->request.master) {
2716         case VHOST_USER_SET_VRING_KICK:
2717         case VHOST_USER_SET_VRING_CALL:
2718         case VHOST_USER_SET_VRING_ERR:
2719                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2720                 break;
2721         case VHOST_USER_SET_VRING_NUM:
2722         case VHOST_USER_SET_VRING_BASE:
2723         case VHOST_USER_SET_VRING_ENABLE:
2724                 vring_idx = msg->payload.state.index;
2725                 break;
2726         case VHOST_USER_SET_VRING_ADDR:
2727                 vring_idx = msg->payload.addr.index;
2728                 break;
2729         default:
2730                 return 0;
2731         }
2732
2733         if (vring_idx >= VHOST_MAX_VRING) {
2734                 VHOST_LOG_CONFIG(ERR,
2735                         "invalid vring index: %u\n", vring_idx);
2736                 return -1;
2737         }
2738
2739         if (dev->virtqueue[vring_idx])
2740                 return 0;
2741
2742         return alloc_vring_queue(dev, vring_idx);
2743 }
2744
2745 static void
2746 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2747 {
2748         unsigned int i = 0;
2749         unsigned int vq_num = 0;
2750
2751         while (vq_num < dev->nr_vring) {
2752                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2753
2754                 if (vq) {
2755                         rte_spinlock_lock(&vq->access_lock);
2756                         vq_num++;
2757                 }
2758                 i++;
2759         }
2760 }
2761
2762 static void
2763 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2764 {
2765         unsigned int i = 0;
2766         unsigned int vq_num = 0;
2767
2768         while (vq_num < dev->nr_vring) {
2769                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2770
2771                 if (vq) {
2772                         rte_spinlock_unlock(&vq->access_lock);
2773                         vq_num++;
2774                 }
2775                 i++;
2776         }
2777 }
2778
2779 int
2780 vhost_user_msg_handler(int vid, int fd)
2781 {
2782         struct virtio_net *dev;
2783         struct VhostUserMsg msg;
2784         struct rte_vdpa_device *vdpa_dev;
2785         int ret;
2786         int unlock_required = 0;
2787         bool handled;
2788         int request;
2789         uint32_t i;
2790
2791         dev = get_device(vid);
2792         if (dev == NULL)
2793                 return -1;
2794
2795         if (!dev->notify_ops) {
2796                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2797                 if (!dev->notify_ops) {
2798                         VHOST_LOG_CONFIG(ERR,
2799                                 "failed to get callback ops for driver %s\n",
2800                                 dev->ifname);
2801                         return -1;
2802                 }
2803         }
2804
2805         ret = read_vhost_message(fd, &msg);
2806         if (ret <= 0) {
2807                 if (ret < 0)
2808                         VHOST_LOG_CONFIG(ERR,
2809                                 "vhost read message failed\n");
2810                 else
2811                         VHOST_LOG_CONFIG(INFO,
2812                                 "vhost peer closed\n");
2813
2814                 return -1;
2815         }
2816
2817         ret = 0;
2818         request = msg.request.master;
2819         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2820                         vhost_message_str[request]) {
2821                 if (request != VHOST_USER_IOTLB_MSG)
2822                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2823                                 vhost_message_str[request]);
2824                 else
2825                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2826                                 vhost_message_str[request]);
2827         } else {
2828                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2829         }
2830
2831         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2832         if (ret < 0) {
2833                 VHOST_LOG_CONFIG(ERR,
2834                         "failed to alloc queue\n");
2835                 return -1;
2836         }
2837
2838         /*
2839          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2840          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2841          * and device is destroyed. destroy_device waits for queues to be
2842          * inactive, so it is safe. Otherwise taking the access_lock
2843          * would cause a dead lock.
2844          */
2845         switch (request) {
2846         case VHOST_USER_SET_FEATURES:
2847         case VHOST_USER_SET_PROTOCOL_FEATURES:
2848         case VHOST_USER_SET_OWNER:
2849         case VHOST_USER_SET_MEM_TABLE:
2850         case VHOST_USER_SET_LOG_BASE:
2851         case VHOST_USER_SET_LOG_FD:
2852         case VHOST_USER_SET_VRING_NUM:
2853         case VHOST_USER_SET_VRING_ADDR:
2854         case VHOST_USER_SET_VRING_BASE:
2855         case VHOST_USER_SET_VRING_KICK:
2856         case VHOST_USER_SET_VRING_CALL:
2857         case VHOST_USER_SET_VRING_ERR:
2858         case VHOST_USER_SET_VRING_ENABLE:
2859         case VHOST_USER_SEND_RARP:
2860         case VHOST_USER_NET_SET_MTU:
2861         case VHOST_USER_SET_SLAVE_REQ_FD:
2862                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2863                         vhost_user_lock_all_queue_pairs(dev);
2864                         unlock_required = 1;
2865                 }
2866                 break;
2867         default:
2868                 break;
2869
2870         }
2871
2872         handled = false;
2873         if (dev->extern_ops.pre_msg_handle) {
2874                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2875                                 (void *)&msg);
2876                 switch (ret) {
2877                 case RTE_VHOST_MSG_RESULT_REPLY:
2878                         send_vhost_reply(fd, &msg);
2879                         /* Fall-through */
2880                 case RTE_VHOST_MSG_RESULT_ERR:
2881                 case RTE_VHOST_MSG_RESULT_OK:
2882                         handled = true;
2883                         goto skip_to_post_handle;
2884                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2885                 default:
2886                         break;
2887                 }
2888         }
2889
2890         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2891                 if (!vhost_message_handlers[request])
2892                         goto skip_to_post_handle;
2893                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2894
2895                 switch (ret) {
2896                 case RTE_VHOST_MSG_RESULT_ERR:
2897                         VHOST_LOG_CONFIG(ERR,
2898                                 "Processing %s failed.\n",
2899                                 vhost_message_str[request]);
2900                         handled = true;
2901                         break;
2902                 case RTE_VHOST_MSG_RESULT_OK:
2903                         VHOST_LOG_CONFIG(DEBUG,
2904                                 "Processing %s succeeded.\n",
2905                                 vhost_message_str[request]);
2906                         handled = true;
2907                         break;
2908                 case RTE_VHOST_MSG_RESULT_REPLY:
2909                         VHOST_LOG_CONFIG(DEBUG,
2910                                 "Processing %s succeeded and needs reply.\n",
2911                                 vhost_message_str[request]);
2912                         send_vhost_reply(fd, &msg);
2913                         handled = true;
2914                         break;
2915                 default:
2916                         break;
2917                 }
2918         }
2919
2920 skip_to_post_handle:
2921         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2922                         dev->extern_ops.post_msg_handle) {
2923                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2924                                 (void *)&msg);
2925                 switch (ret) {
2926                 case RTE_VHOST_MSG_RESULT_REPLY:
2927                         send_vhost_reply(fd, &msg);
2928                         /* Fall-through */
2929                 case RTE_VHOST_MSG_RESULT_ERR:
2930                 case RTE_VHOST_MSG_RESULT_OK:
2931                         handled = true;
2932                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2933                 default:
2934                         break;
2935                 }
2936         }
2937
2938         if (unlock_required)
2939                 vhost_user_unlock_all_queue_pairs(dev);
2940
2941         /* If message was not handled at this stage, treat it as an error */
2942         if (!handled) {
2943                 VHOST_LOG_CONFIG(ERR,
2944                         "vhost message (req: %d) was not handled.\n", request);
2945                 close_msg_fds(&msg);
2946                 ret = RTE_VHOST_MSG_RESULT_ERR;
2947         }
2948
2949         /*
2950          * If the request required a reply that was already sent,
2951          * this optional reply-ack won't be sent as the
2952          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2953          */
2954         if (msg.flags & VHOST_USER_NEED_REPLY) {
2955                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2956                 msg.size = sizeof(msg.payload.u64);
2957                 msg.fd_num = 0;
2958                 send_vhost_reply(fd, &msg);
2959         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2960                 VHOST_LOG_CONFIG(ERR,
2961                         "vhost message handling failed.\n");
2962                 return -1;
2963         }
2964
2965         for (i = 0; i < dev->nr_vring; i++) {
2966                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2967                 bool cur_ready = vq_is_ready(dev, vq);
2968
2969                 if (cur_ready != (vq && vq->ready)) {
2970                         vq->ready = cur_ready;
2971                         vhost_user_notify_queue_state(dev, i, cur_ready);
2972                 }
2973         }
2974
2975
2976         if (!virtio_is_ready(dev))
2977                 goto out;
2978
2979         /*
2980          * Virtio is now ready. If not done already, it is time
2981          * to notify the application it can process the rings and
2982          * configure the vDPA device if present.
2983          */
2984
2985         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2986                 if (dev->notify_ops->new_device(dev->vid) == 0)
2987                         dev->flags |= VIRTIO_DEV_RUNNING;
2988         }
2989
2990         vdpa_dev = dev->vdpa_dev;
2991         if (!vdpa_dev)
2992                 goto out;
2993
2994         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2995                 if (vdpa_dev->ops->dev_conf(dev->vid))
2996                         VHOST_LOG_CONFIG(ERR,
2997                                          "Failed to configure vDPA device\n");
2998                 else
2999                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3000         }
3001
3002 out:
3003         return 0;
3004 }
3005
3006 static int process_slave_message_reply(struct virtio_net *dev,
3007                                        const struct VhostUserMsg *msg)
3008 {
3009         struct VhostUserMsg msg_reply;
3010         int ret;
3011
3012         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3013                 return 0;
3014
3015         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3016         if (ret <= 0) {
3017                 if (ret < 0)
3018                         VHOST_LOG_CONFIG(ERR,
3019                                 "vhost read slave message reply failed\n");
3020                 else
3021                         VHOST_LOG_CONFIG(INFO,
3022                                 "vhost peer closed\n");
3023                 ret = -1;
3024                 goto out;
3025         }
3026
3027         ret = 0;
3028         if (msg_reply.request.slave != msg->request.slave) {
3029                 VHOST_LOG_CONFIG(ERR,
3030                         "Received unexpected msg type (%u), expected %u\n",
3031                         msg_reply.request.slave, msg->request.slave);
3032                 ret = -1;
3033                 goto out;
3034         }
3035
3036         ret = msg_reply.payload.u64 ? -1 : 0;
3037
3038 out:
3039         rte_spinlock_unlock(&dev->slave_req_lock);
3040         return ret;
3041 }
3042
3043 int
3044 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3045 {
3046         int ret;
3047         struct VhostUserMsg msg = {
3048                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3049                 .flags = VHOST_USER_VERSION,
3050                 .size = sizeof(msg.payload.iotlb),
3051                 .payload.iotlb = {
3052                         .iova = iova,
3053                         .perm = perm,
3054                         .type = VHOST_IOTLB_MISS,
3055                 },
3056         };
3057
3058         ret = send_vhost_message(dev->slave_req_fd, &msg);
3059         if (ret < 0) {
3060                 VHOST_LOG_CONFIG(ERR,
3061                                 "Failed to send IOTLB miss message (%d)\n",
3062                                 ret);
3063                 return ret;
3064         }
3065
3066         return 0;
3067 }
3068
3069 static int
3070 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3071 {
3072         int ret;
3073         struct VhostUserMsg msg = {
3074                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3075                 .flags = VHOST_USER_VERSION,
3076                 .size = 0,
3077         };
3078
3079         if (need_reply)
3080                 msg.flags |= VHOST_USER_NEED_REPLY;
3081
3082         ret = send_vhost_slave_message(dev, &msg);
3083         if (ret < 0) {
3084                 VHOST_LOG_CONFIG(ERR,
3085                                 "Failed to send config change (%d)\n",
3086                                 ret);
3087                 return ret;
3088         }
3089
3090         return process_slave_message_reply(dev, &msg);
3091 }
3092
3093 int
3094 rte_vhost_slave_config_change(int vid, bool need_reply)
3095 {
3096         struct virtio_net *dev;
3097
3098         dev = get_device(vid);
3099         if (!dev)
3100                 return -ENODEV;
3101
3102         return vhost_user_slave_config_change(dev, need_reply);
3103 }
3104
3105 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3106                                                     int index, int fd,
3107                                                     uint64_t offset,
3108                                                     uint64_t size)
3109 {
3110         int ret;
3111         struct VhostUserMsg msg = {
3112                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3113                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3114                 .size = sizeof(msg.payload.area),
3115                 .payload.area = {
3116                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3117                         .size = size,
3118                         .offset = offset,
3119                 },
3120         };
3121
3122         if (fd < 0)
3123                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3124         else {
3125                 msg.fds[0] = fd;
3126                 msg.fd_num = 1;
3127         }
3128
3129         ret = send_vhost_slave_message(dev, &msg);
3130         if (ret < 0) {
3131                 VHOST_LOG_CONFIG(ERR,
3132                         "Failed to set host notifier (%d)\n", ret);
3133                 return ret;
3134         }
3135
3136         return process_slave_message_reply(dev, &msg);
3137 }
3138
3139 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3140 {
3141         struct virtio_net *dev;
3142         struct rte_vdpa_device *vdpa_dev;
3143         int vfio_device_fd, ret = 0;
3144         uint64_t offset, size;
3145         unsigned int i, q_start, q_last;
3146
3147         dev = get_device(vid);
3148         if (!dev)
3149                 return -ENODEV;
3150
3151         vdpa_dev = dev->vdpa_dev;
3152         if (vdpa_dev == NULL)
3153                 return -ENODEV;
3154
3155         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3156             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3157             !(dev->protocol_features &
3158                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3159             !(dev->protocol_features &
3160                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3161             !(dev->protocol_features &
3162                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3163                 return -ENOTSUP;
3164
3165         if (qid == RTE_VHOST_QUEUE_ALL) {
3166                 q_start = 0;
3167                 q_last = dev->nr_vring - 1;
3168         } else {
3169                 if (qid >= dev->nr_vring)
3170                         return -EINVAL;
3171                 q_start = qid;
3172                 q_last = qid;
3173         }
3174
3175         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3176         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3177
3178         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3179         if (vfio_device_fd < 0)
3180                 return -ENOTSUP;
3181
3182         if (enable) {
3183                 for (i = q_start; i <= q_last; i++) {
3184                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3185                                         &size) < 0) {
3186                                 ret = -ENOTSUP;
3187                                 goto disable;
3188                         }
3189
3190                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3191                                         vfio_device_fd, offset, size) < 0) {
3192                                 ret = -EFAULT;
3193                                 goto disable;
3194                         }
3195                 }
3196         } else {
3197 disable:
3198                 for (i = q_start; i <= q_last; i++) {
3199                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3200                                         0, 0);
3201                 }
3202         }
3203
3204         return ret;
3205 }