lib/librte_vhost/vhost_user.c

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