lib/vhost/vhost_user.c

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