lib/librte_vhost/vhost_user.c

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