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