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