lib/vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 /* Security model
   6  * --------------
   7  * The vhost-user protocol connection is an external interface, so it must be
   8  * robust against invalid inputs.
   9  *
  10  * This is important because the vhost-user master is only one step removed
  11  * from the guest.  Malicious guests that have escaped will then launch further
  12  * attacks from the vhost-user master.
  13  *
  14  * Even in deployments where guests are trusted, a bug in the vhost-user master
  15  * can still cause invalid messages to be sent.  Such messages must not
  16  * compromise the stability of the DPDK application by causing crashes, memory
  17  * corruption, or other problematic behavior.
  18  *
  19  * Do not assume received VhostUserMsg fields contain sensible values!
  20  */
  21
  22 #include <stdint.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <fcntl.h>
  28 #include <sys/ioctl.h>
  29 #include <sys/mman.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/syscall.h>
  33 #include <assert.h>
  34 #ifdef RTE_LIBRTE_VHOST_NUMA
  35 #include <numaif.h>
  36 #endif
  37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
  38 #include <linux/userfaultfd.h>
  39 #endif
  40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
  41 #include <linux/memfd.h>
  42 #define MEMFD_SUPPORTED
  43 #endif
  44
  45 #include <rte_common.h>
  46 #include <rte_malloc.h>
  47 #include <rte_log.h>
  48
  49 #include "iotlb.h"
  50 #include "vhost.h"
  51 #include "vhost_user.h"
  52
  53 #define VIRTIO_MIN_MTU 68
  54 #define VIRTIO_MAX_MTU 65535
  55
  56 #define INFLIGHT_ALIGNMENT      64
  57 #define INFLIGHT_VERSION        0x1
  58
  59 static const char *vhost_message_str[VHOST_USER_MAX] = {
  60         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  61         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  62         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  63         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  64         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  65         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  66         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  67         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  68         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  69         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  70         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  71         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  72         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  73         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  74         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  75         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  76         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  77         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  78         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  79         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  80         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  81         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  82         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  83         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  84         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  85         [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
  86         [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
  87         [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
  88         [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
  89         [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
  90         [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
  91         [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
  92 };
  93
  94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
  95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
  96
  97 static void
  98 close_msg_fds(struct VhostUserMsg *msg)
  99 {
 100         int i;
 101
 102         for (i = 0; i < msg->fd_num; i++) {
 103                 int fd = msg->fds[i];
 104
 105                 if (fd == -1)
 106                         continue;
 107
 108                 msg->fds[i] = -1;
 109                 close(fd);
 110         }
 111 }
 112
 113 /*
 114  * Ensure the expected number of FDs is received,
 115  * close all FDs and return an error if this is not the case.
 116  */
 117 static int
 118 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
 119 {
 120         if (msg->fd_num == expected_fds)
 121                 return 0;
 122
 123         VHOST_LOG_CONFIG(ERR,
 124                 " Expect %d FDs for request %s, received %d\n",
 125                 expected_fds,
 126                 vhost_message_str[msg->request.master],
 127                 msg->fd_num);
 128
 129         close_msg_fds(msg);
 130
 131         return -1;
 132 }
 133
 134 static uint64_t
 135 get_blk_size(int fd)
 136 {
 137         struct stat stat;
 138         int ret;
 139
 140         ret = fstat(fd, &stat);
 141         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 142 }
 143
 144 static void
 145 free_mem_region(struct virtio_net *dev)
 146 {
 147         uint32_t i;
 148         struct rte_vhost_mem_region *reg;
 149
 150         if (!dev || !dev->mem)
 151                 return;
 152
 153         for (i = 0; i < dev->mem->nregions; i++) {
 154                 reg = &dev->mem->regions[i];
 155                 if (reg->host_user_addr) {
 156                         munmap(reg->mmap_addr, reg->mmap_size);
 157                         close(reg->fd);
 158                 }
 159         }
 160 }
 161
 162 void
 163 vhost_backend_cleanup(struct virtio_net *dev)
 164 {
 165         if (dev->mem) {
 166                 free_mem_region(dev);
 167                 rte_free(dev->mem);
 168                 dev->mem = NULL;
 169         }
 170
 171         rte_free(dev->guest_pages);
 172         dev->guest_pages = NULL;
 173
 174         if (dev->log_addr) {
 175                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 176                 dev->log_addr = 0;
 177         }
 178
 179         if (dev->inflight_info) {
 180                 if (dev->inflight_info->addr) {
 181                         munmap(dev->inflight_info->addr,
 182                                dev->inflight_info->size);
 183                         dev->inflight_info->addr = NULL;
 184                 }
 185
 186                 if (dev->inflight_info->fd >= 0) {
 187                         close(dev->inflight_info->fd);
 188                         dev->inflight_info->fd = -1;
 189                 }
 190
 191                 rte_free(dev->inflight_info);
 192                 dev->inflight_info = NULL;
 193         }
 194
 195         if (dev->slave_req_fd >= 0) {
 196                 close(dev->slave_req_fd);
 197                 dev->slave_req_fd = -1;
 198         }
 199
 200         if (dev->postcopy_ufd >= 0) {
 201                 close(dev->postcopy_ufd);
 202                 dev->postcopy_ufd = -1;
 203         }
 204
 205         dev->postcopy_listening = 0;
 206 }
 207
 208 static void
 209 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
 210                               int enable)
 211 {
 212         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
 213         struct vhost_virtqueue *vq = dev->virtqueue[index];
 214
 215         /* Configure guest notifications on enable */
 216         if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
 217                 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
 218
 219         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
 220                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
 221
 222         if (dev->notify_ops->vring_state_changed)
 223                 dev->notify_ops->vring_state_changed(dev->vid,
 224                                 index, enable);
 225 }
 226
 227 /*
 228  * This function just returns success at the moment unless
 229  * the device hasn't been initialised.
 230  */
 231 static int
 232 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
 233                         struct VhostUserMsg *msg,
 234                         int main_fd __rte_unused)
 235 {
 236         if (validate_msg_fds(msg, 0) != 0)
 237                 return RTE_VHOST_MSG_RESULT_ERR;
 238
 239         return RTE_VHOST_MSG_RESULT_OK;
 240 }
 241
 242 static int
 243 vhost_user_reset_owner(struct virtio_net **pdev,
 244                         struct VhostUserMsg *msg,
 245                         int main_fd __rte_unused)
 246 {
 247         struct virtio_net *dev = *pdev;
 248
 249         if (validate_msg_fds(msg, 0) != 0)
 250                 return RTE_VHOST_MSG_RESULT_ERR;
 251
 252         vhost_destroy_device_notify(dev);
 253
 254         cleanup_device(dev, 0);
 255         reset_device(dev);
 256         return RTE_VHOST_MSG_RESULT_OK;
 257 }
 258
 259 /*
 260  * The features that we support are requested.
 261  */
 262 static int
 263 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 264                         int main_fd __rte_unused)
 265 {
 266         struct virtio_net *dev = *pdev;
 267         uint64_t features = 0;
 268
 269         if (validate_msg_fds(msg, 0) != 0)
 270                 return RTE_VHOST_MSG_RESULT_ERR;
 271
 272         rte_vhost_driver_get_features(dev->ifname, &features);
 273
 274         msg->payload.u64 = features;
 275         msg->size = sizeof(msg->payload.u64);
 276         msg->fd_num = 0;
 277
 278         return RTE_VHOST_MSG_RESULT_REPLY;
 279 }
 280
 281 /*
 282  * The queue number that we support are requested.
 283  */
 284 static int
 285 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
 286                         int main_fd __rte_unused)
 287 {
 288         struct virtio_net *dev = *pdev;
 289         uint32_t queue_num = 0;
 290
 291         if (validate_msg_fds(msg, 0) != 0)
 292                 return RTE_VHOST_MSG_RESULT_ERR;
 293
 294         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 295
 296         msg->payload.u64 = (uint64_t)queue_num;
 297         msg->size = sizeof(msg->payload.u64);
 298         msg->fd_num = 0;
 299
 300         return RTE_VHOST_MSG_RESULT_REPLY;
 301 }
 302
 303 /*
 304  * We receive the negotiated features supported by us and the virtio device.
 305  */
 306 static int
 307 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 308                         int main_fd __rte_unused)
 309 {
 310         struct virtio_net *dev = *pdev;
 311         uint64_t features = msg->payload.u64;
 312         uint64_t vhost_features = 0;
 313         struct rte_vdpa_device *vdpa_dev;
 314
 315         if (validate_msg_fds(msg, 0) != 0)
 316                 return RTE_VHOST_MSG_RESULT_ERR;
 317
 318         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 319         if (features & ~vhost_features) {
 320                 VHOST_LOG_CONFIG(ERR,
 321                         "(%d) received invalid negotiated features.\n",
 322                         dev->vid);
 323                 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
 324                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
 325
 326                 return RTE_VHOST_MSG_RESULT_ERR;
 327         }
 328
 329         if (dev->flags & VIRTIO_DEV_RUNNING) {
 330                 if (dev->features == features)
 331                         return RTE_VHOST_MSG_RESULT_OK;
 332
 333                 /*
 334                  * Error out if master tries to change features while device is
 335                  * in running state. The exception being VHOST_F_LOG_ALL, which
 336                  * is enabled when the live-migration starts.
 337                  */
 338                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 339                         VHOST_LOG_CONFIG(ERR,
 340                                 "(%d) features changed while device is running.\n",
 341                                 dev->vid);
 342                         return RTE_VHOST_MSG_RESULT_ERR;
 343                 }
 344
 345                 if (dev->notify_ops->features_changed)
 346                         dev->notify_ops->features_changed(dev->vid, features);
 347         }
 348
 349         dev->features = features;
 350         if (dev->features &
 351                 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
 352                  (1ULL << VIRTIO_F_VERSION_1) |
 353                  (1ULL << VIRTIO_F_RING_PACKED))) {
 354                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 355         } else {
 356                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 357         }
 358         VHOST_LOG_CONFIG(INFO,
 359                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 360         VHOST_LOG_CONFIG(DEBUG,
 361                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 362                 dev->vid,
 363                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 364                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 365
 366         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 367             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 368                 /*
 369                  * Remove all but first queue pair if MQ hasn't been
 370                  * negotiated. This is safe because the device is not
 371                  * running at this stage.
 372                  */
 373                 while (dev->nr_vring > 2) {
 374                         struct vhost_virtqueue *vq;
 375
 376                         vq = dev->virtqueue[--dev->nr_vring];
 377                         if (!vq)
 378                                 continue;
 379
 380                         dev->virtqueue[dev->nr_vring] = NULL;
 381                         cleanup_vq(vq, 1);
 382                         cleanup_vq_inflight(dev, vq);
 383                         free_vq(dev, vq);
 384                 }
 385         }
 386
 387         vdpa_dev = dev->vdpa_dev;
 388         if (vdpa_dev)
 389                 vdpa_dev->ops->set_features(dev->vid);
 390
 391         dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
 392         return RTE_VHOST_MSG_RESULT_OK;
 393 }
 394
 395 /*
 396  * The virtio device sends us the size of the descriptor ring.
 397  */
 398 static int
 399 vhost_user_set_vring_num(struct virtio_net **pdev,
 400                         struct VhostUserMsg *msg,
 401                         int main_fd __rte_unused)
 402 {
 403         struct virtio_net *dev = *pdev;
 404         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 405
 406         if (validate_msg_fds(msg, 0) != 0)
 407                 return RTE_VHOST_MSG_RESULT_ERR;
 408
 409         if (msg->payload.state.num > 32768) {
 410                 VHOST_LOG_CONFIG(ERR, "invalid virtqueue size %u\n", msg->payload.state.num);
 411                 return RTE_VHOST_MSG_RESULT_ERR;
 412         }
 413
 414         vq->size = msg->payload.state.num;
 415
 416         /* VIRTIO 1.0, 2.4 Virtqueues says:
 417          *
 418          *   Queue Size value is always a power of 2. The maximum Queue Size
 419          *   value is 32768.
 420          *
 421          * VIRTIO 1.1 2.7 Virtqueues says:
 422          *
 423          *   Packed virtqueues support up to 2^15 entries each.
 424          */
 425         if (!vq_is_packed(dev)) {
 426                 if (vq->size & (vq->size - 1)) {
 427                         VHOST_LOG_CONFIG(ERR,
 428                                 "invalid virtqueue size %u\n", vq->size);
 429                         return RTE_VHOST_MSG_RESULT_ERR;
 430                 }
 431         }
 432
 433         if (vq_is_packed(dev)) {
 434                 if (vq->shadow_used_packed)
 435                         rte_free(vq->shadow_used_packed);
 436                 vq->shadow_used_packed = rte_malloc_socket(NULL,
 437                                 vq->size *
 438                                 sizeof(struct vring_used_elem_packed),
 439                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 440                 if (!vq->shadow_used_packed) {
 441                         VHOST_LOG_CONFIG(ERR,
 442                                         "failed to allocate memory for shadow used ring.\n");
 443                         return RTE_VHOST_MSG_RESULT_ERR;
 444                 }
 445
 446         } else {
 447                 if (vq->shadow_used_split)
 448                         rte_free(vq->shadow_used_split);
 449
 450                 vq->shadow_used_split = rte_malloc_socket(NULL,
 451                                 vq->size * sizeof(struct vring_used_elem),
 452                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 453
 454                 if (!vq->shadow_used_split) {
 455                         VHOST_LOG_CONFIG(ERR,
 456                                         "failed to allocate memory for vq internal data.\n");
 457                         return RTE_VHOST_MSG_RESULT_ERR;
 458                 }
 459         }
 460
 461         if (vq->batch_copy_elems)
 462                 rte_free(vq->batch_copy_elems);
 463         vq->batch_copy_elems = rte_malloc_socket(NULL,
 464                                 vq->size * sizeof(struct batch_copy_elem),
 465                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 466         if (!vq->batch_copy_elems) {
 467                 VHOST_LOG_CONFIG(ERR,
 468                         "failed to allocate memory for batching copy.\n");
 469                 return RTE_VHOST_MSG_RESULT_ERR;
 470         }
 471
 472         return RTE_VHOST_MSG_RESULT_OK;
 473 }
 474
 475 /*
 476  * Reallocate virtio_dev, vhost_virtqueue and related data structures to
 477  * make them on the same numa node as the memory of vring descriptor.
 478  */
 479 #ifdef RTE_LIBRTE_VHOST_NUMA
 480 static struct virtio_net*
 481 numa_realloc(struct virtio_net *dev, int index)
 482 {
 483         int node, dev_node;
 484         struct virtio_net *old_dev;
 485         struct vhost_virtqueue *vq;
 486         struct batch_copy_elem *bce;
 487         struct guest_page *gp;
 488         struct rte_vhost_memory *mem;
 489         size_t mem_size;
 490         int ret;
 491
 492         old_dev = dev;
 493         vq = dev->virtqueue[index];
 494
 495         /*
 496          * If VQ is ready, it is too late to reallocate, it certainly already
 497          * happened anyway on VHOST_USER_SET_VRING_ADRR.
 498          */
 499         if (vq->ready)
 500                 return dev;
 501
 502         ret = get_mempolicy(&node, NULL, 0, vq->desc, MPOL_F_NODE | MPOL_F_ADDR);
 503         if (ret) {
 504                 VHOST_LOG_CONFIG(ERR, "Unable to get virtqueue %d numa information.\n", index);
 505                 return dev;
 506         }
 507
 508         if (node == vq->numa_node)
 509                 goto out_dev_realloc;
 510
 511         vq = rte_realloc_socket(vq, sizeof(*vq), 0, node);
 512         if (!vq) {
 513                 VHOST_LOG_CONFIG(ERR, "Failed to realloc virtqueue %d on node %d\n",
 514                                 index, node);
 515                 return dev;
 516         }
 517
 518         if (vq != dev->virtqueue[index]) {
 519                 VHOST_LOG_CONFIG(INFO, "reallocated virtqueue on node %d\n", node);
 520                 dev->virtqueue[index] = vq;
 521                 vhost_user_iotlb_init(dev, index);
 522         }
 523
 524         if (vq_is_packed(dev)) {
 525                 struct vring_used_elem_packed *sup;
 526
 527                 sup = rte_realloc_socket(vq->shadow_used_packed, vq->size * sizeof(*sup),
 528                                 RTE_CACHE_LINE_SIZE, node);
 529                 if (!sup) {
 530                         VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow packed on node %d\n", node);
 531                         return dev;
 532                 }
 533                 vq->shadow_used_packed = sup;
 534         } else {
 535                 struct vring_used_elem *sus;
 536
 537                 sus = rte_realloc_socket(vq->shadow_used_split, vq->size * sizeof(*sus),
 538                                 RTE_CACHE_LINE_SIZE, node);
 539                 if (!sus) {
 540                         VHOST_LOG_CONFIG(ERR, "Failed to realloc shadow split on node %d\n", node);
 541                         return dev;
 542                 }
 543                 vq->shadow_used_split = sus;
 544         }
 545
 546         bce = rte_realloc_socket(vq->batch_copy_elems, vq->size * sizeof(*bce),
 547                         RTE_CACHE_LINE_SIZE, node);
 548         if (!bce) {
 549                 VHOST_LOG_CONFIG(ERR, "Failed to realloc batch copy elem on node %d\n", node);
 550                 return dev;
 551         }
 552         vq->batch_copy_elems = bce;
 553
 554         if (vq->log_cache) {
 555                 struct log_cache_entry *lc;
 556
 557                 lc = rte_realloc_socket(vq->log_cache, sizeof(*lc) * VHOST_LOG_CACHE_NR, 0, node);
 558                 if (!lc) {
 559                         VHOST_LOG_CONFIG(ERR, "Failed to realloc log cache on node %d\n", node);
 560                         return dev;
 561                 }
 562                 vq->log_cache = lc;
 563         }
 564
 565         if (vq->resubmit_inflight) {
 566                 struct rte_vhost_resubmit_info *ri;
 567
 568                 ri = rte_realloc_socket(vq->resubmit_inflight, sizeof(*ri), 0, node);
 569                 if (!ri) {
 570                         VHOST_LOG_CONFIG(ERR, "Failed to realloc resubmit inflight on node %d\n",
 571                                         node);
 572                         return dev;
 573                 }
 574                 vq->resubmit_inflight = ri;
 575
 576                 if (ri->resubmit_list) {
 577                         struct rte_vhost_resubmit_desc *rd;
 578
 579                         rd = rte_realloc_socket(ri->resubmit_list, sizeof(*rd) * ri->resubmit_num,
 580                                         0, node);
 581                         if (!rd) {
 582                                 VHOST_LOG_CONFIG(ERR, "Failed to realloc resubmit list on node %d\n",
 583                                                 node);
 584                                 return dev;
 585                         }
 586                         ri->resubmit_list = rd;
 587                 }
 588         }
 589
 590         vq->numa_node = node;
 591
 592 out_dev_realloc:
 593
 594         if (dev->flags & VIRTIO_DEV_RUNNING)
 595                 return dev;
 596
 597         ret = get_mempolicy(&dev_node, NULL, 0, dev, MPOL_F_NODE | MPOL_F_ADDR);
 598         if (ret) {
 599                 VHOST_LOG_CONFIG(ERR, "Unable to get Virtio dev %d numa information.\n", dev->vid);
 600                 return dev;
 601         }
 602
 603         if (dev_node == node)
 604                 return dev;
 605
 606         dev = rte_realloc_socket(old_dev, sizeof(*dev), 0, node);
 607         if (!dev) {
 608                 VHOST_LOG_CONFIG(ERR, "Failed to realloc dev on node %d\n", node);
 609                 return old_dev;
 610         }
 611
 612         VHOST_LOG_CONFIG(INFO, "reallocated device on node %d\n", node);
 613         vhost_devices[dev->vid] = dev;
 614
 615         mem_size = sizeof(struct rte_vhost_memory) +
 616                 sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
 617         mem = rte_realloc_socket(dev->mem, mem_size, 0, node);
 618         if (!mem) {
 619                 VHOST_LOG_CONFIG(ERR, "Failed to realloc mem table on node %d\n", node);
 620                 return dev;
 621         }
 622         dev->mem = mem;
 623
 624         gp = rte_realloc_socket(dev->guest_pages, dev->max_guest_pages * sizeof(*gp),
 625                         RTE_CACHE_LINE_SIZE, node);
 626         if (!gp) {
 627                 VHOST_LOG_CONFIG(ERR, "Failed to realloc guest pages on node %d\n", node);
 628                 return dev;
 629         }
 630         dev->guest_pages = gp;
 631
 632         return dev;
 633 }
 634 #else
 635 static struct virtio_net*
 636 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 637 {
 638         return dev;
 639 }
 640 #endif
 641
 642 /* Converts QEMU virtual address to Vhost virtual address. */
 643 static uint64_t
 644 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 645 {
 646         struct rte_vhost_mem_region *r;
 647         uint32_t i;
 648
 649         if (unlikely(!dev || !dev->mem))
 650                 goto out_error;
 651
 652         /* Find the region where the address lives. */
 653         for (i = 0; i < dev->mem->nregions; i++) {
 654                 r = &dev->mem->regions[i];
 655
 656                 if (qva >= r->guest_user_addr &&
 657                     qva <  r->guest_user_addr + r->size) {
 658
 659                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 660                                 *len = r->guest_user_addr + r->size - qva;
 661
 662                         return qva - r->guest_user_addr +
 663                                r->host_user_addr;
 664                 }
 665         }
 666 out_error:
 667         *len = 0;
 668
 669         return 0;
 670 }
 671
 672
 673 /*
 674  * Converts ring address to Vhost virtual address.
 675  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 676  * else it is a QEMU virtual address.
 677  */
 678 static uint64_t
 679 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 680                 uint64_t ra, uint64_t *size)
 681 {
 682         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 683                 uint64_t vva;
 684
 685                 vhost_user_iotlb_rd_lock(vq);
 686                 vva = vhost_iova_to_vva(dev, vq, ra,
 687                                         size, VHOST_ACCESS_RW);
 688                 vhost_user_iotlb_rd_unlock(vq);
 689
 690                 return vva;
 691         }
 692
 693         return qva_to_vva(dev, ra, size);
 694 }
 695
 696 static uint64_t
 697 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 698 {
 699         uint64_t log_gpa;
 700
 701         vhost_user_iotlb_rd_lock(vq);
 702         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 703         vhost_user_iotlb_rd_unlock(vq);
 704
 705         return log_gpa;
 706 }
 707
 708 static struct virtio_net *
 709 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 710 {
 711         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 712         struct vhost_vring_addr *addr = &vq->ring_addrs;
 713         uint64_t len, expected_len;
 714
 715         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 716                 vq->log_guest_addr =
 717                         log_addr_to_gpa(dev, vq);
 718                 if (vq->log_guest_addr == 0) {
 719                         VHOST_LOG_CONFIG(DEBUG,
 720                                 "(%d) failed to map log_guest_addr.\n",
 721                                 dev->vid);
 722                         return dev;
 723                 }
 724         }
 725
 726         if (vq_is_packed(dev)) {
 727                 len = sizeof(struct vring_packed_desc) * vq->size;
 728                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 729                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 730                 if (vq->desc_packed == NULL ||
 731                                 len != sizeof(struct vring_packed_desc) *
 732                                 vq->size) {
 733                         VHOST_LOG_CONFIG(DEBUG,
 734                                 "(%d) failed to map desc_packed ring.\n",
 735                                 dev->vid);
 736                         return dev;
 737                 }
 738
 739                 dev = numa_realloc(dev, vq_index);
 740                 vq = dev->virtqueue[vq_index];
 741                 addr = &vq->ring_addrs;
 742
 743                 len = sizeof(struct vring_packed_desc_event);
 744                 vq->driver_event = (struct vring_packed_desc_event *)
 745                                         (uintptr_t)ring_addr_to_vva(dev,
 746                                         vq, addr->avail_user_addr, &len);
 747                 if (vq->driver_event == NULL ||
 748                                 len != sizeof(struct vring_packed_desc_event)) {
 749                         VHOST_LOG_CONFIG(DEBUG,
 750                                 "(%d) failed to find driver area address.\n",
 751                                 dev->vid);
 752                         return dev;
 753                 }
 754
 755                 len = sizeof(struct vring_packed_desc_event);
 756                 vq->device_event = (struct vring_packed_desc_event *)
 757                                         (uintptr_t)ring_addr_to_vva(dev,
 758                                         vq, addr->used_user_addr, &len);
 759                 if (vq->device_event == NULL ||
 760                                 len != sizeof(struct vring_packed_desc_event)) {
 761                         VHOST_LOG_CONFIG(DEBUG,
 762                                 "(%d) failed to find device area address.\n",
 763                                 dev->vid);
 764                         return dev;
 765                 }
 766
 767                 vq->access_ok = true;
 768                 return dev;
 769         }
 770
 771         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 772         if (vq->desc && vq->avail && vq->used)
 773                 return dev;
 774
 775         len = sizeof(struct vring_desc) * vq->size;
 776         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 777                         vq, addr->desc_user_addr, &len);
 778         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 779                 VHOST_LOG_CONFIG(DEBUG,
 780                         "(%d) failed to map desc ring.\n",
 781                         dev->vid);
 782                 return dev;
 783         }
 784
 785         dev = numa_realloc(dev, vq_index);
 786         vq = dev->virtqueue[vq_index];
 787         addr = &vq->ring_addrs;
 788
 789         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 790         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 791                 len += sizeof(uint16_t);
 792         expected_len = len;
 793         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 794                         vq, addr->avail_user_addr, &len);
 795         if (vq->avail == 0 || len != expected_len) {
 796                 VHOST_LOG_CONFIG(DEBUG,
 797                         "(%d) failed to map avail ring.\n",
 798                         dev->vid);
 799                 return dev;
 800         }
 801
 802         len = sizeof(struct vring_used) +
 803                 sizeof(struct vring_used_elem) * vq->size;
 804         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 805                 len += sizeof(uint16_t);
 806         expected_len = len;
 807         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 808                         vq, addr->used_user_addr, &len);
 809         if (vq->used == 0 || len != expected_len) {
 810                 VHOST_LOG_CONFIG(DEBUG,
 811                         "(%d) failed to map used ring.\n",
 812                         dev->vid);
 813                 return dev;
 814         }
 815
 816         if (vq->last_used_idx != vq->used->idx) {
 817                 VHOST_LOG_CONFIG(WARNING,
 818                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 819                         "some packets maybe resent for Tx and dropped for Rx\n",
 820                         vq->last_used_idx, vq->used->idx);
 821                 vq->last_used_idx  = vq->used->idx;
 822                 vq->last_avail_idx = vq->used->idx;
 823         }
 824
 825         vq->access_ok = true;
 826
 827         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 828                         dev->vid, vq->desc);
 829         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 830                         dev->vid, vq->avail);
 831         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 832                         dev->vid, vq->used);
 833         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 834                         dev->vid, vq->log_guest_addr);
 835
 836         return dev;
 837 }
 838
 839 /*
 840  * The virtio device sends us the desc, used and avail ring addresses.
 841  * This function then converts these to our address space.
 842  */
 843 static int
 844 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 845                         int main_fd __rte_unused)
 846 {
 847         struct virtio_net *dev = *pdev;
 848         struct vhost_virtqueue *vq;
 849         struct vhost_vring_addr *addr = &msg->payload.addr;
 850         bool access_ok;
 851
 852         if (validate_msg_fds(msg, 0) != 0)
 853                 return RTE_VHOST_MSG_RESULT_ERR;
 854
 855         if (dev->mem == NULL)
 856                 return RTE_VHOST_MSG_RESULT_ERR;
 857
 858         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 859         vq = dev->virtqueue[msg->payload.addr.index];
 860
 861         access_ok = vq->access_ok;
 862
 863         /*
 864          * Rings addresses should not be interpreted as long as the ring is not
 865          * started and enabled
 866          */
 867         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 868
 869         vring_invalidate(dev, vq);
 870
 871         if ((vq->enabled && (dev->features &
 872                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 873                         access_ok) {
 874                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 875                 if (!dev)
 876                         return RTE_VHOST_MSG_RESULT_ERR;
 877
 878                 *pdev = dev;
 879         }
 880
 881         return RTE_VHOST_MSG_RESULT_OK;
 882 }
 883
 884 /*
 885  * The virtio device sends us the available ring last used index.
 886  */
 887 static int
 888 vhost_user_set_vring_base(struct virtio_net **pdev,
 889                         struct VhostUserMsg *msg,
 890                         int main_fd __rte_unused)
 891 {
 892         struct virtio_net *dev = *pdev;
 893         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 894         uint64_t val = msg->payload.state.num;
 895
 896         if (validate_msg_fds(msg, 0) != 0)
 897                 return RTE_VHOST_MSG_RESULT_ERR;
 898
 899         if (vq_is_packed(dev)) {
 900                 /*
 901                  * Bit[0:14]: avail index
 902                  * Bit[15]: avail wrap counter
 903                  */
 904                 vq->last_avail_idx = val & 0x7fff;
 905                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 906                 /*
 907                  * Set used index to same value as available one, as
 908                  * their values should be the same since ring processing
 909                  * was stopped at get time.
 910                  */
 911                 vq->last_used_idx = vq->last_avail_idx;
 912                 vq->used_wrap_counter = vq->avail_wrap_counter;
 913         } else {
 914                 vq->last_used_idx = msg->payload.state.num;
 915                 vq->last_avail_idx = msg->payload.state.num;
 916         }
 917
 918         return RTE_VHOST_MSG_RESULT_OK;
 919 }
 920
 921 static int
 922 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 923                    uint64_t host_phys_addr, uint64_t size)
 924 {
 925         struct guest_page *page, *last_page;
 926         struct guest_page *old_pages;
 927
 928         if (dev->nr_guest_pages == dev->max_guest_pages) {
 929                 dev->max_guest_pages *= 2;
 930                 old_pages = dev->guest_pages;
 931                 dev->guest_pages = rte_realloc(dev->guest_pages,
 932                                         dev->max_guest_pages * sizeof(*page),
 933                                         RTE_CACHE_LINE_SIZE);
 934                 if (dev->guest_pages == NULL) {
 935                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 936                         rte_free(old_pages);
 937                         return -1;
 938                 }
 939         }
 940
 941         if (dev->nr_guest_pages > 0) {
 942                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 943                 /* merge if the two pages are continuous */
 944                 if (host_phys_addr == last_page->host_phys_addr +
 945                                       last_page->size) {
 946                         last_page->size += size;
 947                         return 0;
 948                 }
 949         }
 950
 951         page = &dev->guest_pages[dev->nr_guest_pages++];
 952         page->guest_phys_addr = guest_phys_addr;
 953         page->host_phys_addr  = host_phys_addr;
 954         page->size = size;
 955
 956         return 0;
 957 }
 958
 959 static int
 960 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 961                 uint64_t page_size)
 962 {
 963         uint64_t reg_size = reg->size;
 964         uint64_t host_user_addr  = reg->host_user_addr;
 965         uint64_t guest_phys_addr = reg->guest_phys_addr;
 966         uint64_t host_phys_addr;
 967         uint64_t size;
 968
 969         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 970         size = page_size - (guest_phys_addr & (page_size - 1));
 971         size = RTE_MIN(size, reg_size);
 972
 973         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 974                 return -1;
 975
 976         host_user_addr  += size;
 977         guest_phys_addr += size;
 978         reg_size -= size;
 979
 980         while (reg_size > 0) {
 981                 size = RTE_MIN(reg_size, page_size);
 982                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 983                                                   host_user_addr);
 984                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 985                                 size) < 0)
 986                         return -1;
 987
 988                 host_user_addr  += size;
 989                 guest_phys_addr += size;
 990                 reg_size -= size;
 991         }
 992
 993         /* sort guest page array if over binary search threshold */
 994         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 995                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 996                         sizeof(struct guest_page), guest_page_addrcmp);
 997         }
 998
 999         return 0;
1000 }
1001
1002 #ifdef RTE_LIBRTE_VHOST_DEBUG
1003 /* TODO: enable it only in debug mode? */
1004 static void
1005 dump_guest_pages(struct virtio_net *dev)
1006 {
1007         uint32_t i;
1008         struct guest_page *page;
1009
1010         for (i = 0; i < dev->nr_guest_pages; i++) {
1011                 page = &dev->guest_pages[i];
1012
1013                 VHOST_LOG_CONFIG(INFO,
1014                         "guest physical page region %u\n"
1015                         "\t guest_phys_addr: %" PRIx64 "\n"
1016                         "\t host_phys_addr : %" PRIx64 "\n"
1017                         "\t size           : %" PRIx64 "\n",
1018                         i,
1019                         page->guest_phys_addr,
1020                         page->host_phys_addr,
1021                         page->size);
1022         }
1023 }
1024 #else
1025 #define dump_guest_pages(dev)
1026 #endif
1027
1028 static bool
1029 vhost_memory_changed(struct VhostUserMemory *new,
1030                      struct rte_vhost_memory *old)
1031 {
1032         uint32_t i;
1033
1034         if (new->nregions != old->nregions)
1035                 return true;
1036
1037         for (i = 0; i < new->nregions; ++i) {
1038                 VhostUserMemoryRegion *new_r = &new->regions[i];
1039                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1040
1041                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1042                         return true;
1043                 if (new_r->memory_size != old_r->size)
1044                         return true;
1045                 if (new_r->userspace_addr != old_r->guest_user_addr)
1046                         return true;
1047         }
1048
1049         return false;
1050 }
1051
1052 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1053 static int
1054 vhost_user_postcopy_region_register(struct virtio_net *dev,
1055                 struct rte_vhost_mem_region *reg)
1056 {
1057         struct uffdio_register reg_struct;
1058
1059         /*
1060          * Let's register all the mmap'ed area to ensure
1061          * alignment on page boundary.
1062          */
1063         reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1064         reg_struct.range.len = reg->mmap_size;
1065         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1066
1067         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1068                                 &reg_struct)) {
1069                 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1070                                 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1071                                 (uint64_t)reg_struct.range.start,
1072                                 (uint64_t)reg_struct.range.start +
1073                                 (uint64_t)reg_struct.range.len - 1,
1074                                 dev->postcopy_ufd,
1075                                 strerror(errno));
1076                 return -1;
1077         }
1078
1079         VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1080                         (uint64_t)reg_struct.range.start,
1081                         (uint64_t)reg_struct.range.start +
1082                         (uint64_t)reg_struct.range.len - 1);
1083
1084         return 0;
1085 }
1086 #else
1087 static int
1088 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1089                 struct rte_vhost_mem_region *reg __rte_unused)
1090 {
1091         return -1;
1092 }
1093 #endif
1094
1095 static int
1096 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1097                 struct VhostUserMsg *msg)
1098 {
1099         struct VhostUserMemory *memory;
1100         struct rte_vhost_mem_region *reg;
1101         VhostUserMsg ack_msg;
1102         uint32_t i;
1103
1104         if (!dev->postcopy_listening)
1105                 return 0;
1106
1107         /*
1108          * We haven't a better way right now than sharing
1109          * DPDK's virtual address with Qemu, so that Qemu can
1110          * retrieve the region offset when handling userfaults.
1111          */
1112         memory = &msg->payload.memory;
1113         for (i = 0; i < memory->nregions; i++) {
1114                 reg = &dev->mem->regions[i];
1115                 memory->regions[i].userspace_addr = reg->host_user_addr;
1116         }
1117
1118         /* Send the addresses back to qemu */
1119         msg->fd_num = 0;
1120         send_vhost_reply(main_fd, msg);
1121
1122         /* Wait for qemu to acknolwedge it's got the addresses
1123          * we've got to wait before we're allowed to generate faults.
1124          */
1125         if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1126                 VHOST_LOG_CONFIG(ERR,
1127                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1128                 return -1;
1129         }
1130
1131         if (validate_msg_fds(&ack_msg, 0) != 0)
1132                 return -1;
1133
1134         if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1135                 VHOST_LOG_CONFIG(ERR,
1136                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1137                                 ack_msg.request.master);
1138                 return -1;
1139         }
1140
1141         /* Now userfault register and we can use the memory */
1142         for (i = 0; i < memory->nregions; i++) {
1143                 reg = &dev->mem->regions[i];
1144                 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1145                         return -1;
1146         }
1147
1148         return 0;
1149 }
1150
1151 static int
1152 vhost_user_mmap_region(struct virtio_net *dev,
1153                 struct rte_vhost_mem_region *region,
1154                 uint64_t mmap_offset)
1155 {
1156         void *mmap_addr;
1157         uint64_t mmap_size;
1158         uint64_t alignment;
1159         int populate;
1160
1161         /* Check for memory_size + mmap_offset overflow */
1162         if (mmap_offset >= -region->size) {
1163                 VHOST_LOG_CONFIG(ERR,
1164                                 "mmap_offset (%#"PRIx64") and memory_size "
1165                                 "(%#"PRIx64") overflow\n",
1166                                 mmap_offset, region->size);
1167                 return -1;
1168         }
1169
1170         mmap_size = region->size + mmap_offset;
1171
1172         /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1173          * argument aligned with hugepagesz at older longterm version Linux,
1174          * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1175          *
1176          * To avoid failure, make sure in caller to keep length aligned.
1177          */
1178         alignment = get_blk_size(region->fd);
1179         if (alignment == (uint64_t)-1) {
1180                 VHOST_LOG_CONFIG(ERR,
1181                                 "couldn't get hugepage size through fstat\n");
1182                 return -1;
1183         }
1184         mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1185         if (mmap_size == 0) {
1186                 /*
1187                  * It could happen if initial mmap_size + alignment overflows
1188                  * the sizeof uint64, which could happen if either mmap_size or
1189                  * alignment value is wrong.
1190                  *
1191                  * mmap() kernel implementation would return an error, but
1192                  * better catch it before and provide useful info in the logs.
1193                  */
1194                 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1195                                 "or alignment (0x%" PRIx64 ") is invalid\n",
1196                                 region->size + mmap_offset, alignment);
1197                 return -1;
1198         }
1199
1200         populate = dev->async_copy ? MAP_POPULATE : 0;
1201         mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1202                         MAP_SHARED | populate, region->fd, 0);
1203
1204         if (mmap_addr == MAP_FAILED) {
1205                 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1206                 return -1;
1207         }
1208
1209         region->mmap_addr = mmap_addr;
1210         region->mmap_size = mmap_size;
1211         region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1212
1213         if (dev->async_copy)
1214                 if (add_guest_pages(dev, region, alignment) < 0) {
1215                         VHOST_LOG_CONFIG(ERR,
1216                                         "adding guest pages to region failed.\n");
1217                         return -1;
1218                 }
1219
1220         VHOST_LOG_CONFIG(INFO,
1221                         "guest memory region size: 0x%" PRIx64 "\n"
1222                         "\t guest physical addr: 0x%" PRIx64 "\n"
1223                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1224                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1225                         "\t mmap addr : 0x%" PRIx64 "\n"
1226                         "\t mmap size : 0x%" PRIx64 "\n"
1227                         "\t mmap align: 0x%" PRIx64 "\n"
1228                         "\t mmap off  : 0x%" PRIx64 "\n",
1229                         region->size,
1230                         region->guest_phys_addr,
1231                         region->guest_user_addr,
1232                         region->host_user_addr,
1233                         (uint64_t)(uintptr_t)mmap_addr,
1234                         mmap_size,
1235                         alignment,
1236                         mmap_offset);
1237
1238         return 0;
1239 }
1240
1241 static int
1242 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1243                         int main_fd)
1244 {
1245         struct virtio_net *dev = *pdev;
1246         struct VhostUserMemory *memory = &msg->payload.memory;
1247         struct rte_vhost_mem_region *reg;
1248         int numa_node = SOCKET_ID_ANY;
1249         uint64_t mmap_offset;
1250         uint32_t i;
1251         bool async_notify = false;
1252
1253         if (validate_msg_fds(msg, memory->nregions) != 0)
1254                 return RTE_VHOST_MSG_RESULT_ERR;
1255
1256         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1257                 VHOST_LOG_CONFIG(ERR,
1258                         "too many memory regions (%u)\n", memory->nregions);
1259                 goto close_msg_fds;
1260         }
1261
1262         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1263                 VHOST_LOG_CONFIG(INFO,
1264                         "(%d) memory regions not changed\n", dev->vid);
1265
1266                 close_msg_fds(msg);
1267
1268                 return RTE_VHOST_MSG_RESULT_OK;
1269         }
1270
1271         if (dev->mem) {
1272                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1273                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1274
1275                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1276                                 vdpa_dev->ops->dev_close(dev->vid);
1277                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1278                 }
1279
1280                 /* notify the vhost application to stop DMA transfers */
1281                 if (dev->async_copy && dev->notify_ops->vring_state_changed) {
1282                         for (i = 0; i < dev->nr_vring; i++) {
1283                                 dev->notify_ops->vring_state_changed(dev->vid,
1284                                                 i, 0);
1285                         }
1286                         async_notify = true;
1287                 }
1288
1289                 free_mem_region(dev);
1290                 rte_free(dev->mem);
1291                 dev->mem = NULL;
1292         }
1293
1294         /* Flush IOTLB cache as previous HVAs are now invalid */
1295         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1296                 for (i = 0; i < dev->nr_vring; i++)
1297                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1298
1299         /*
1300          * If VQ 0 has already been allocated, try to allocate on the same
1301          * NUMA node. It can be reallocated later in numa_realloc().
1302          */
1303         if (dev->nr_vring > 0)
1304                 numa_node = dev->virtqueue[0]->numa_node;
1305
1306         dev->nr_guest_pages = 0;
1307         if (dev->guest_pages == NULL) {
1308                 dev->max_guest_pages = 8;
1309                 dev->guest_pages = rte_zmalloc_socket(NULL,
1310                                         dev->max_guest_pages *
1311                                         sizeof(struct guest_page),
1312                                         RTE_CACHE_LINE_SIZE,
1313                                         numa_node);
1314                 if (dev->guest_pages == NULL) {
1315                         VHOST_LOG_CONFIG(ERR,
1316                                 "(%d) failed to allocate memory "
1317                                 "for dev->guest_pages\n",
1318                                 dev->vid);
1319                         goto close_msg_fds;
1320                 }
1321         }
1322
1323         dev->mem = rte_zmalloc_socket("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1324                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0, numa_node);
1325         if (dev->mem == NULL) {
1326                 VHOST_LOG_CONFIG(ERR,
1327                         "(%d) failed to allocate memory for dev->mem\n",
1328                         dev->vid);
1329                 goto free_guest_pages;
1330         }
1331
1332         for (i = 0; i < memory->nregions; i++) {
1333                 reg = &dev->mem->regions[i];
1334
1335                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1336                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1337                 reg->size            = memory->regions[i].memory_size;
1338                 reg->fd              = msg->fds[i];
1339
1340                 /*
1341                  * Assign invalid file descriptor value to avoid double
1342                  * closing on error path.
1343                  */
1344                 msg->fds[i] = -1;
1345
1346                 mmap_offset = memory->regions[i].mmap_offset;
1347
1348                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1349                         VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1350                         goto free_mem_table;
1351                 }
1352
1353                 dev->mem->nregions++;
1354         }
1355
1356         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1357                 goto free_mem_table;
1358
1359         for (i = 0; i < dev->nr_vring; i++) {
1360                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1361
1362                 if (!vq)
1363                         continue;
1364
1365                 if (vq->desc || vq->avail || vq->used) {
1366                         /*
1367                          * If the memory table got updated, the ring addresses
1368                          * need to be translated again as virtual addresses have
1369                          * changed.
1370                          */
1371                         vring_invalidate(dev, vq);
1372
1373                         dev = translate_ring_addresses(dev, i);
1374                         if (!dev) {
1375                                 dev = *pdev;
1376                                 goto free_mem_table;
1377                         }
1378
1379                         *pdev = dev;
1380                 }
1381         }
1382
1383         dump_guest_pages(dev);
1384
1385         if (async_notify) {
1386                 for (i = 0; i < dev->nr_vring; i++)
1387                         dev->notify_ops->vring_state_changed(dev->vid, i, 1);
1388         }
1389
1390         return RTE_VHOST_MSG_RESULT_OK;
1391
1392 free_mem_table:
1393         free_mem_region(dev);
1394         rte_free(dev->mem);
1395         dev->mem = NULL;
1396 free_guest_pages:
1397         rte_free(dev->guest_pages);
1398         dev->guest_pages = NULL;
1399 close_msg_fds:
1400         close_msg_fds(msg);
1401         return RTE_VHOST_MSG_RESULT_ERR;
1402 }
1403
1404 static bool
1405 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1406 {
1407         bool rings_ok;
1408
1409         if (!vq)
1410                 return false;
1411
1412         if (vq_is_packed(dev))
1413                 rings_ok = vq->desc_packed && vq->driver_event &&
1414                         vq->device_event;
1415         else
1416                 rings_ok = vq->desc && vq->avail && vq->used;
1417
1418         return rings_ok &&
1419                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1420                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1421                vq->enabled;
1422 }
1423
1424 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1425
1426 static int
1427 virtio_is_ready(struct virtio_net *dev)
1428 {
1429         struct vhost_virtqueue *vq;
1430         uint32_t i, nr_vring = dev->nr_vring;
1431
1432         if (dev->flags & VIRTIO_DEV_READY)
1433                 return 1;
1434
1435         if (!dev->nr_vring)
1436                 return 0;
1437
1438         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1439                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1440
1441                 if (dev->nr_vring < nr_vring)
1442                         return 0;
1443         }
1444
1445         for (i = 0; i < nr_vring; i++) {
1446                 vq = dev->virtqueue[i];
1447
1448                 if (!vq_is_ready(dev, vq))
1449                         return 0;
1450         }
1451
1452         /* If supported, ensure the frontend is really done with config */
1453         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1454                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1455                         return 0;
1456
1457         dev->flags |= VIRTIO_DEV_READY;
1458
1459         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1460                 VHOST_LOG_CONFIG(INFO,
1461                         "virtio is now ready for processing.\n");
1462         return 1;
1463 }
1464
1465 static void *
1466 inflight_mem_alloc(const char *name, size_t size, int *fd)
1467 {
1468         void *ptr;
1469         int mfd = -1;
1470         char fname[20] = "/tmp/memfd-XXXXXX";
1471
1472         *fd = -1;
1473 #ifdef MEMFD_SUPPORTED
1474         mfd = memfd_create(name, MFD_CLOEXEC);
1475 #else
1476         RTE_SET_USED(name);
1477 #endif
1478         if (mfd == -1) {
1479                 mfd = mkstemp(fname);
1480                 if (mfd == -1) {
1481                         VHOST_LOG_CONFIG(ERR,
1482                                 "failed to get inflight buffer fd\n");
1483                         return NULL;
1484                 }
1485
1486                 unlink(fname);
1487         }
1488
1489         if (ftruncate(mfd, size) == -1) {
1490                 VHOST_LOG_CONFIG(ERR,
1491                         "failed to alloc inflight buffer\n");
1492                 close(mfd);
1493                 return NULL;
1494         }
1495
1496         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1497         if (ptr == MAP_FAILED) {
1498                 VHOST_LOG_CONFIG(ERR,
1499                         "failed to mmap inflight buffer\n");
1500                 close(mfd);
1501                 return NULL;
1502         }
1503
1504         *fd = mfd;
1505         return ptr;
1506 }
1507
1508 static uint32_t
1509 get_pervq_shm_size_split(uint16_t queue_size)
1510 {
1511         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1512                                   queue_size + sizeof(uint64_t) +
1513                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1514 }
1515
1516 static uint32_t
1517 get_pervq_shm_size_packed(uint16_t queue_size)
1518 {
1519         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1520                                   * queue_size + sizeof(uint64_t) +
1521                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1522                                   INFLIGHT_ALIGNMENT);
1523 }
1524
1525 static int
1526 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1527                            VhostUserMsg *msg,
1528                            int main_fd __rte_unused)
1529 {
1530         struct rte_vhost_inflight_info_packed *inflight_packed;
1531         uint64_t pervq_inflight_size, mmap_size;
1532         uint16_t num_queues, queue_size;
1533         struct virtio_net *dev = *pdev;
1534         int fd, i, j;
1535         int numa_node = SOCKET_ID_ANY;
1536         void *addr;
1537
1538         if (msg->size != sizeof(msg->payload.inflight)) {
1539                 VHOST_LOG_CONFIG(ERR,
1540                         "invalid get_inflight_fd message size is %d\n",
1541                         msg->size);
1542                 return RTE_VHOST_MSG_RESULT_ERR;
1543         }
1544
1545         /*
1546          * If VQ 0 has already been allocated, try to allocate on the same
1547          * NUMA node. It can be reallocated later in numa_realloc().
1548          */
1549         if (dev->nr_vring > 0)
1550                 numa_node = dev->virtqueue[0]->numa_node;
1551
1552         if (dev->inflight_info == NULL) {
1553                 dev->inflight_info = rte_zmalloc_socket("inflight_info",
1554                                 sizeof(struct inflight_mem_info), 0, numa_node);
1555                 if (!dev->inflight_info) {
1556                         VHOST_LOG_CONFIG(ERR,
1557                                 "failed to alloc dev inflight area\n");
1558                         return RTE_VHOST_MSG_RESULT_ERR;
1559                 }
1560                 dev->inflight_info->fd = -1;
1561         }
1562
1563         num_queues = msg->payload.inflight.num_queues;
1564         queue_size = msg->payload.inflight.queue_size;
1565
1566         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1567                 msg->payload.inflight.num_queues);
1568         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1569                 msg->payload.inflight.queue_size);
1570
1571         if (vq_is_packed(dev))
1572                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1573         else
1574                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1575
1576         mmap_size = num_queues * pervq_inflight_size;
1577         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1578         if (!addr) {
1579                 VHOST_LOG_CONFIG(ERR,
1580                         "failed to alloc vhost inflight area\n");
1581                         msg->payload.inflight.mmap_size = 0;
1582                 return RTE_VHOST_MSG_RESULT_ERR;
1583         }
1584         memset(addr, 0, mmap_size);
1585
1586         if (dev->inflight_info->addr) {
1587                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1588                 dev->inflight_info->addr = NULL;
1589         }
1590
1591         if (dev->inflight_info->fd >= 0) {
1592                 close(dev->inflight_info->fd);
1593                 dev->inflight_info->fd = -1;
1594         }
1595
1596         dev->inflight_info->addr = addr;
1597         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1598         dev->inflight_info->fd = msg->fds[0] = fd;
1599         msg->payload.inflight.mmap_offset = 0;
1600         msg->fd_num = 1;
1601
1602         if (vq_is_packed(dev)) {
1603                 for (i = 0; i < num_queues; i++) {
1604                         inflight_packed =
1605                                 (struct rte_vhost_inflight_info_packed *)addr;
1606                         inflight_packed->used_wrap_counter = 1;
1607                         inflight_packed->old_used_wrap_counter = 1;
1608                         for (j = 0; j < queue_size; j++)
1609                                 inflight_packed->desc[j].next = j + 1;
1610                         addr = (void *)((char *)addr + pervq_inflight_size);
1611                 }
1612         }
1613
1614         VHOST_LOG_CONFIG(INFO,
1615                 "send inflight mmap_size: %"PRIu64"\n",
1616                 msg->payload.inflight.mmap_size);
1617         VHOST_LOG_CONFIG(INFO,
1618                 "send inflight mmap_offset: %"PRIu64"\n",
1619                 msg->payload.inflight.mmap_offset);
1620         VHOST_LOG_CONFIG(INFO,
1621                 "send inflight fd: %d\n", msg->fds[0]);
1622
1623         return RTE_VHOST_MSG_RESULT_REPLY;
1624 }
1625
1626 static int
1627 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1628                            int main_fd __rte_unused)
1629 {
1630         uint64_t mmap_size, mmap_offset;
1631         uint16_t num_queues, queue_size;
1632         struct virtio_net *dev = *pdev;
1633         uint32_t pervq_inflight_size;
1634         struct vhost_virtqueue *vq;
1635         void *addr;
1636         int fd, i;
1637         int numa_node = SOCKET_ID_ANY;
1638
1639         fd = msg->fds[0];
1640         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1641                 VHOST_LOG_CONFIG(ERR,
1642                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1643                         msg->size, fd);
1644                 return RTE_VHOST_MSG_RESULT_ERR;
1645         }
1646
1647         mmap_size = msg->payload.inflight.mmap_size;
1648         mmap_offset = msg->payload.inflight.mmap_offset;
1649         num_queues = msg->payload.inflight.num_queues;
1650         queue_size = msg->payload.inflight.queue_size;
1651
1652         if (vq_is_packed(dev))
1653                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1654         else
1655                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1656
1657         VHOST_LOG_CONFIG(INFO,
1658                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1659         VHOST_LOG_CONFIG(INFO,
1660                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1661         VHOST_LOG_CONFIG(INFO,
1662                 "set_inflight_fd num_queues: %u\n", num_queues);
1663         VHOST_LOG_CONFIG(INFO,
1664                 "set_inflight_fd queue_size: %u\n", queue_size);
1665         VHOST_LOG_CONFIG(INFO,
1666                 "set_inflight_fd fd: %d\n", fd);
1667         VHOST_LOG_CONFIG(INFO,
1668                 "set_inflight_fd pervq_inflight_size: %d\n",
1669                 pervq_inflight_size);
1670
1671         /*
1672          * If VQ 0 has already been allocated, try to allocate on the same
1673          * NUMA node. It can be reallocated later in numa_realloc().
1674          */
1675         if (dev->nr_vring > 0)
1676                 numa_node = dev->virtqueue[0]->numa_node;
1677
1678         if (!dev->inflight_info) {
1679                 dev->inflight_info = rte_zmalloc_socket("inflight_info",
1680                                 sizeof(struct inflight_mem_info), 0, numa_node);
1681                 if (dev->inflight_info == NULL) {
1682                         VHOST_LOG_CONFIG(ERR,
1683                                 "failed to alloc dev inflight area\n");
1684                         return RTE_VHOST_MSG_RESULT_ERR;
1685                 }
1686                 dev->inflight_info->fd = -1;
1687         }
1688
1689         if (dev->inflight_info->addr) {
1690                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1691                 dev->inflight_info->addr = NULL;
1692         }
1693
1694         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1695                     fd, mmap_offset);
1696         if (addr == MAP_FAILED) {
1697                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1698                 return RTE_VHOST_MSG_RESULT_ERR;
1699         }
1700
1701         if (dev->inflight_info->fd >= 0) {
1702                 close(dev->inflight_info->fd);
1703                 dev->inflight_info->fd = -1;
1704         }
1705
1706         dev->inflight_info->fd = fd;
1707         dev->inflight_info->addr = addr;
1708         dev->inflight_info->size = mmap_size;
1709
1710         for (i = 0; i < num_queues; i++) {
1711                 vq = dev->virtqueue[i];
1712                 if (!vq)
1713                         continue;
1714
1715                 if (vq_is_packed(dev)) {
1716                         vq->inflight_packed = addr;
1717                         vq->inflight_packed->desc_num = queue_size;
1718                 } else {
1719                         vq->inflight_split = addr;
1720                         vq->inflight_split->desc_num = queue_size;
1721                 }
1722                 addr = (void *)((char *)addr + pervq_inflight_size);
1723         }
1724
1725         return RTE_VHOST_MSG_RESULT_OK;
1726 }
1727
1728 static int
1729 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1730                         int main_fd __rte_unused)
1731 {
1732         struct virtio_net *dev = *pdev;
1733         struct vhost_vring_file file;
1734         struct vhost_virtqueue *vq;
1735         int expected_fds;
1736
1737         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1738         if (validate_msg_fds(msg, expected_fds) != 0)
1739                 return RTE_VHOST_MSG_RESULT_ERR;
1740
1741         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1742         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1743                 file.fd = VIRTIO_INVALID_EVENTFD;
1744         else
1745                 file.fd = msg->fds[0];
1746         VHOST_LOG_CONFIG(INFO,
1747                 "vring call idx:%d file:%d\n", file.index, file.fd);
1748
1749         vq = dev->virtqueue[file.index];
1750
1751         if (vq->ready) {
1752                 vq->ready = false;
1753                 vhost_user_notify_queue_state(dev, file.index, 0);
1754         }
1755
1756         if (vq->callfd >= 0)
1757                 close(vq->callfd);
1758
1759         vq->callfd = file.fd;
1760
1761         return RTE_VHOST_MSG_RESULT_OK;
1762 }
1763
1764 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1765                         struct VhostUserMsg *msg,
1766                         int main_fd __rte_unused)
1767 {
1768         int expected_fds;
1769
1770         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1771         if (validate_msg_fds(msg, expected_fds) != 0)
1772                 return RTE_VHOST_MSG_RESULT_ERR;
1773
1774         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1775                 close(msg->fds[0]);
1776         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1777
1778         return RTE_VHOST_MSG_RESULT_OK;
1779 }
1780
1781 static int
1782 resubmit_desc_compare(const void *a, const void *b)
1783 {
1784         const struct rte_vhost_resubmit_desc *desc0 = a;
1785         const struct rte_vhost_resubmit_desc *desc1 = b;
1786
1787         if (desc1->counter > desc0->counter)
1788                 return 1;
1789
1790         return -1;
1791 }
1792
1793 static int
1794 vhost_check_queue_inflights_split(struct virtio_net *dev,
1795                                   struct vhost_virtqueue *vq)
1796 {
1797         uint16_t i;
1798         uint16_t resubmit_num = 0, last_io, num;
1799         struct vring_used *used = vq->used;
1800         struct rte_vhost_resubmit_info *resubmit;
1801         struct rte_vhost_inflight_info_split *inflight_split;
1802
1803         if (!(dev->protocol_features &
1804             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1805                 return RTE_VHOST_MSG_RESULT_OK;
1806
1807         /* The frontend may still not support the inflight feature
1808          * although we negotiate the protocol feature.
1809          */
1810         if ((!vq->inflight_split))
1811                 return RTE_VHOST_MSG_RESULT_OK;
1812
1813         if (!vq->inflight_split->version) {
1814                 vq->inflight_split->version = INFLIGHT_VERSION;
1815                 return RTE_VHOST_MSG_RESULT_OK;
1816         }
1817
1818         if (vq->resubmit_inflight)
1819                 return RTE_VHOST_MSG_RESULT_OK;
1820
1821         inflight_split = vq->inflight_split;
1822         vq->global_counter = 0;
1823         last_io = inflight_split->last_inflight_io;
1824
1825         if (inflight_split->used_idx != used->idx) {
1826                 inflight_split->desc[last_io].inflight = 0;
1827                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1828                 inflight_split->used_idx = used->idx;
1829         }
1830
1831         for (i = 0; i < inflight_split->desc_num; i++) {
1832                 if (inflight_split->desc[i].inflight == 1)
1833                         resubmit_num++;
1834         }
1835
1836         vq->last_avail_idx += resubmit_num;
1837
1838         if (resubmit_num) {
1839                 resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1840                                 0, vq->numa_node);
1841                 if (!resubmit) {
1842                         VHOST_LOG_CONFIG(ERR,
1843                                 "failed to allocate memory for resubmit info.\n");
1844                         return RTE_VHOST_MSG_RESULT_ERR;
1845                 }
1846
1847                 resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
1848                                 resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
1849                                 0, vq->numa_node);
1850                 if (!resubmit->resubmit_list) {
1851                         VHOST_LOG_CONFIG(ERR,
1852                                 "failed to allocate memory for inflight desc.\n");
1853                         rte_free(resubmit);
1854                         return RTE_VHOST_MSG_RESULT_ERR;
1855                 }
1856
1857                 num = 0;
1858                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1859                         if (vq->inflight_split->desc[i].inflight == 1) {
1860                                 resubmit->resubmit_list[num].index = i;
1861                                 resubmit->resubmit_list[num].counter =
1862                                         inflight_split->desc[i].counter;
1863                                 num++;
1864                         }
1865                 }
1866                 resubmit->resubmit_num = num;
1867
1868                 if (resubmit->resubmit_num > 1)
1869                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1870                               sizeof(struct rte_vhost_resubmit_desc),
1871                               resubmit_desc_compare);
1872
1873                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1874                 vq->resubmit_inflight = resubmit;
1875         }
1876
1877         return RTE_VHOST_MSG_RESULT_OK;
1878 }
1879
1880 static int
1881 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1882                                    struct vhost_virtqueue *vq)
1883 {
1884         uint16_t i;
1885         uint16_t resubmit_num = 0, old_used_idx, num;
1886         struct rte_vhost_resubmit_info *resubmit;
1887         struct rte_vhost_inflight_info_packed *inflight_packed;
1888
1889         if (!(dev->protocol_features &
1890             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1891                 return RTE_VHOST_MSG_RESULT_OK;
1892
1893         /* The frontend may still not support the inflight feature
1894          * although we negotiate the protocol feature.
1895          */
1896         if ((!vq->inflight_packed))
1897                 return RTE_VHOST_MSG_RESULT_OK;
1898
1899         if (!vq->inflight_packed->version) {
1900                 vq->inflight_packed->version = INFLIGHT_VERSION;
1901                 return RTE_VHOST_MSG_RESULT_OK;
1902         }
1903
1904         if (vq->resubmit_inflight)
1905                 return RTE_VHOST_MSG_RESULT_OK;
1906
1907         inflight_packed = vq->inflight_packed;
1908         vq->global_counter = 0;
1909         old_used_idx = inflight_packed->old_used_idx;
1910
1911         if (inflight_packed->used_idx != old_used_idx) {
1912                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1913                         inflight_packed->old_used_idx =
1914                                 inflight_packed->used_idx;
1915                         inflight_packed->old_used_wrap_counter =
1916                                 inflight_packed->used_wrap_counter;
1917                         inflight_packed->old_free_head =
1918                                 inflight_packed->free_head;
1919                 } else {
1920                         inflight_packed->used_idx =
1921                                 inflight_packed->old_used_idx;
1922                         inflight_packed->used_wrap_counter =
1923                                 inflight_packed->old_used_wrap_counter;
1924                         inflight_packed->free_head =
1925                                 inflight_packed->old_free_head;
1926                 }
1927         }
1928
1929         for (i = 0; i < inflight_packed->desc_num; i++) {
1930                 if (inflight_packed->desc[i].inflight == 1)
1931                         resubmit_num++;
1932         }
1933
1934         if (resubmit_num) {
1935                 resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1936                                 0, vq->numa_node);
1937                 if (resubmit == NULL) {
1938                         VHOST_LOG_CONFIG(ERR,
1939                                 "failed to allocate memory for resubmit info.\n");
1940                         return RTE_VHOST_MSG_RESULT_ERR;
1941                 }
1942
1943                 resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
1944                                 resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
1945                                 0, vq->numa_node);
1946                 if (resubmit->resubmit_list == NULL) {
1947                         VHOST_LOG_CONFIG(ERR,
1948                                 "failed to allocate memory for resubmit desc.\n");
1949                         rte_free(resubmit);
1950                         return RTE_VHOST_MSG_RESULT_ERR;
1951                 }
1952
1953                 num = 0;
1954                 for (i = 0; i < inflight_packed->desc_num; i++) {
1955                         if (vq->inflight_packed->desc[i].inflight == 1) {
1956                                 resubmit->resubmit_list[num].index = i;
1957                                 resubmit->resubmit_list[num].counter =
1958                                         inflight_packed->desc[i].counter;
1959                                 num++;
1960                         }
1961                 }
1962                 resubmit->resubmit_num = num;
1963
1964                 if (resubmit->resubmit_num > 1)
1965                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1966                               sizeof(struct rte_vhost_resubmit_desc),
1967                               resubmit_desc_compare);
1968
1969                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1970                 vq->resubmit_inflight = resubmit;
1971         }
1972
1973         return RTE_VHOST_MSG_RESULT_OK;
1974 }
1975
1976 static int
1977 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1978                         int main_fd __rte_unused)
1979 {
1980         struct virtio_net *dev = *pdev;
1981         struct vhost_vring_file file;
1982         struct vhost_virtqueue *vq;
1983         int expected_fds;
1984
1985         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1986         if (validate_msg_fds(msg, expected_fds) != 0)
1987                 return RTE_VHOST_MSG_RESULT_ERR;
1988
1989         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1990         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1991                 file.fd = VIRTIO_INVALID_EVENTFD;
1992         else
1993                 file.fd = msg->fds[0];
1994         VHOST_LOG_CONFIG(INFO,
1995                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1996
1997         /* Interpret ring addresses only when ring is started. */
1998         dev = translate_ring_addresses(dev, file.index);
1999         if (!dev) {
2000                 if (file.fd != VIRTIO_INVALID_EVENTFD)
2001                         close(file.fd);
2002
2003                 return RTE_VHOST_MSG_RESULT_ERR;
2004         }
2005
2006         *pdev = dev;
2007
2008         vq = dev->virtqueue[file.index];
2009
2010         /*
2011          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
2012          * the ring starts already enabled. Otherwise, it is enabled via
2013          * the SET_VRING_ENABLE message.
2014          */
2015         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
2016                 vq->enabled = true;
2017         }
2018
2019         if (vq->ready) {
2020                 vq->ready = false;
2021                 vhost_user_notify_queue_state(dev, file.index, 0);
2022         }
2023
2024         if (vq->kickfd >= 0)
2025                 close(vq->kickfd);
2026         vq->kickfd = file.fd;
2027
2028         if (vq_is_packed(dev)) {
2029                 if (vhost_check_queue_inflights_packed(dev, vq)) {
2030                         VHOST_LOG_CONFIG(ERR,
2031                                 "failed to inflights for vq: %d\n", file.index);
2032                         return RTE_VHOST_MSG_RESULT_ERR;
2033                 }
2034         } else {
2035                 if (vhost_check_queue_inflights_split(dev, vq)) {
2036                         VHOST_LOG_CONFIG(ERR,
2037                                 "failed to inflights for vq: %d\n", file.index);
2038                         return RTE_VHOST_MSG_RESULT_ERR;
2039                 }
2040         }
2041
2042         return RTE_VHOST_MSG_RESULT_OK;
2043 }
2044
2045 /*
2046  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
2047  */
2048 static int
2049 vhost_user_get_vring_base(struct virtio_net **pdev,
2050                         struct VhostUserMsg *msg,
2051                         int main_fd __rte_unused)
2052 {
2053         struct virtio_net *dev = *pdev;
2054         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
2055         uint64_t val;
2056
2057         if (validate_msg_fds(msg, 0) != 0)
2058                 return RTE_VHOST_MSG_RESULT_ERR;
2059
2060         /* We have to stop the queue (virtio) if it is running. */
2061         vhost_destroy_device_notify(dev);
2062
2063         dev->flags &= ~VIRTIO_DEV_READY;
2064         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
2065
2066         /* Here we are safe to get the indexes */
2067         if (vq_is_packed(dev)) {
2068                 /*
2069                  * Bit[0:14]: avail index
2070                  * Bit[15]: avail wrap counter
2071                  */
2072                 val = vq->last_avail_idx & 0x7fff;
2073                 val |= vq->avail_wrap_counter << 15;
2074                 msg->payload.state.num = val;
2075         } else {
2076                 msg->payload.state.num = vq->last_avail_idx;
2077         }
2078
2079         VHOST_LOG_CONFIG(INFO,
2080                 "vring base idx:%d file:%d\n", msg->payload.state.index,
2081                 msg->payload.state.num);
2082         /*
2083          * Based on current qemu vhost-user implementation, this message is
2084          * sent and only sent in vhost_vring_stop.
2085          * TODO: cleanup the vring, it isn't usable since here.
2086          */
2087         if (vq->kickfd >= 0)
2088                 close(vq->kickfd);
2089
2090         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2091
2092         if (vq->callfd >= 0)
2093                 close(vq->callfd);
2094
2095         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2096
2097         vq->signalled_used_valid = false;
2098
2099         if (vq_is_packed(dev)) {
2100                 rte_free(vq->shadow_used_packed);
2101                 vq->shadow_used_packed = NULL;
2102         } else {
2103                 rte_free(vq->shadow_used_split);
2104                 vq->shadow_used_split = NULL;
2105         }
2106
2107         rte_free(vq->batch_copy_elems);
2108         vq->batch_copy_elems = NULL;
2109
2110         rte_free(vq->log_cache);
2111         vq->log_cache = NULL;
2112
2113         msg->size = sizeof(msg->payload.state);
2114         msg->fd_num = 0;
2115
2116         vhost_user_iotlb_flush_all(vq);
2117
2118         vring_invalidate(dev, vq);
2119
2120         return RTE_VHOST_MSG_RESULT_REPLY;
2121 }
2122
2123 /*
2124  * when virtio queues are ready to work, qemu will send us to
2125  * enable the virtio queue pair.
2126  */
2127 static int
2128 vhost_user_set_vring_enable(struct virtio_net **pdev,
2129                         struct VhostUserMsg *msg,
2130                         int main_fd __rte_unused)
2131 {
2132         struct virtio_net *dev = *pdev;
2133         bool enable = !!msg->payload.state.num;
2134         int index = (int)msg->payload.state.index;
2135
2136         if (validate_msg_fds(msg, 0) != 0)
2137                 return RTE_VHOST_MSG_RESULT_ERR;
2138
2139         VHOST_LOG_CONFIG(INFO,
2140                 "set queue enable: %d to qp idx: %d\n",
2141                 enable, index);
2142
2143         if (enable && dev->virtqueue[index]->async_registered) {
2144                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2145                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2146                         "async inflight packets must be completed first\n");
2147                         return RTE_VHOST_MSG_RESULT_ERR;
2148                 }
2149         }
2150
2151         dev->virtqueue[index]->enabled = enable;
2152
2153         return RTE_VHOST_MSG_RESULT_OK;
2154 }
2155
2156 static int
2157 vhost_user_get_protocol_features(struct virtio_net **pdev,
2158                         struct VhostUserMsg *msg,
2159                         int main_fd __rte_unused)
2160 {
2161         struct virtio_net *dev = *pdev;
2162         uint64_t features, protocol_features;
2163
2164         if (validate_msg_fds(msg, 0) != 0)
2165                 return RTE_VHOST_MSG_RESULT_ERR;
2166
2167         rte_vhost_driver_get_features(dev->ifname, &features);
2168         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2169
2170         msg->payload.u64 = protocol_features;
2171         msg->size = sizeof(msg->payload.u64);
2172         msg->fd_num = 0;
2173
2174         return RTE_VHOST_MSG_RESULT_REPLY;
2175 }
2176
2177 static int
2178 vhost_user_set_protocol_features(struct virtio_net **pdev,
2179                         struct VhostUserMsg *msg,
2180                         int main_fd __rte_unused)
2181 {
2182         struct virtio_net *dev = *pdev;
2183         uint64_t protocol_features = msg->payload.u64;
2184         uint64_t slave_protocol_features = 0;
2185
2186         if (validate_msg_fds(msg, 0) != 0)
2187                 return RTE_VHOST_MSG_RESULT_ERR;
2188
2189         rte_vhost_driver_get_protocol_features(dev->ifname,
2190                         &slave_protocol_features);
2191         if (protocol_features & ~slave_protocol_features) {
2192                 VHOST_LOG_CONFIG(ERR,
2193                         "(%d) received invalid protocol features.\n",
2194                         dev->vid);
2195                 return RTE_VHOST_MSG_RESULT_ERR;
2196         }
2197
2198         dev->protocol_features = protocol_features;
2199         VHOST_LOG_CONFIG(INFO,
2200                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2201                 dev->protocol_features);
2202
2203         return RTE_VHOST_MSG_RESULT_OK;
2204 }
2205
2206 static int
2207 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2208                         int main_fd __rte_unused)
2209 {
2210         struct virtio_net *dev = *pdev;
2211         int fd = msg->fds[0];
2212         uint64_t size, off;
2213         void *addr;
2214         uint32_t i;
2215
2216         if (validate_msg_fds(msg, 1) != 0)
2217                 return RTE_VHOST_MSG_RESULT_ERR;
2218
2219         if (fd < 0) {
2220                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2221                 return RTE_VHOST_MSG_RESULT_ERR;
2222         }
2223
2224         if (msg->size != sizeof(VhostUserLog)) {
2225                 VHOST_LOG_CONFIG(ERR,
2226                         "invalid log base msg size: %"PRId32" != %d\n",
2227                         msg->size, (int)sizeof(VhostUserLog));
2228                 goto close_msg_fds;
2229         }
2230
2231         size = msg->payload.log.mmap_size;
2232         off  = msg->payload.log.mmap_offset;
2233
2234         /* Check for mmap size and offset overflow. */
2235         if (off >= -size) {
2236                 VHOST_LOG_CONFIG(ERR,
2237                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2238                         off, size);
2239                 goto close_msg_fds;
2240         }
2241
2242         VHOST_LOG_CONFIG(INFO,
2243                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2244                 size, off);
2245
2246         /*
2247          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2248          * fail when offset is not page size aligned.
2249          */
2250         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2251         close(fd);
2252         if (addr == MAP_FAILED) {
2253                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2254                 return RTE_VHOST_MSG_RESULT_ERR;
2255         }
2256
2257         /*
2258          * Free previously mapped log memory on occasionally
2259          * multiple VHOST_USER_SET_LOG_BASE.
2260          */
2261         if (dev->log_addr) {
2262                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2263         }
2264         dev->log_addr = (uint64_t)(uintptr_t)addr;
2265         dev->log_base = dev->log_addr + off;
2266         dev->log_size = size;
2267
2268         for (i = 0; i < dev->nr_vring; i++) {
2269                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2270
2271                 rte_free(vq->log_cache);
2272                 vq->log_cache = NULL;
2273                 vq->log_cache_nb_elem = 0;
2274                 vq->log_cache = rte_malloc_socket("vq log cache",
2275                                 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2276                                 0, vq->numa_node);
2277                 /*
2278                  * If log cache alloc fail, don't fail migration, but no
2279                  * caching will be done, which will impact performance
2280                  */
2281                 if (!vq->log_cache)
2282                         VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2283         }
2284
2285         /*
2286          * The spec is not clear about it (yet), but QEMU doesn't expect
2287          * any payload in the reply.
2288          */
2289         msg->size = 0;
2290         msg->fd_num = 0;
2291
2292         return RTE_VHOST_MSG_RESULT_REPLY;
2293
2294 close_msg_fds:
2295         close_msg_fds(msg);
2296         return RTE_VHOST_MSG_RESULT_ERR;
2297 }
2298
2299 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2300                         struct VhostUserMsg *msg,
2301                         int main_fd __rte_unused)
2302 {
2303         if (validate_msg_fds(msg, 1) != 0)
2304                 return RTE_VHOST_MSG_RESULT_ERR;
2305
2306         close(msg->fds[0]);
2307         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2308
2309         return RTE_VHOST_MSG_RESULT_OK;
2310 }
2311
2312 /*
2313  * An rarp packet is constructed and broadcasted to notify switches about
2314  * the new location of the migrated VM, so that packets from outside will
2315  * not be lost after migration.
2316  *
2317  * However, we don't actually "send" a rarp packet here, instead, we set
2318  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2319  */
2320 static int
2321 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2322                         int main_fd __rte_unused)
2323 {
2324         struct virtio_net *dev = *pdev;
2325         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2326         struct rte_vdpa_device *vdpa_dev;
2327
2328         if (validate_msg_fds(msg, 0) != 0)
2329                 return RTE_VHOST_MSG_RESULT_ERR;
2330
2331         VHOST_LOG_CONFIG(DEBUG,
2332                 ":: mac: " RTE_ETHER_ADDR_PRT_FMT "\n",
2333                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2334         memcpy(dev->mac.addr_bytes, mac, 6);
2335
2336         /*
2337          * Set the flag to inject a RARP broadcast packet at
2338          * rte_vhost_dequeue_burst().
2339          *
2340          * __ATOMIC_RELEASE ordering is for making sure the mac is
2341          * copied before the flag is set.
2342          */
2343         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2344         vdpa_dev = dev->vdpa_dev;
2345         if (vdpa_dev && vdpa_dev->ops->migration_done)
2346                 vdpa_dev->ops->migration_done(dev->vid);
2347
2348         return RTE_VHOST_MSG_RESULT_OK;
2349 }
2350
2351 static int
2352 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2353                         int main_fd __rte_unused)
2354 {
2355         struct virtio_net *dev = *pdev;
2356
2357         if (validate_msg_fds(msg, 0) != 0)
2358                 return RTE_VHOST_MSG_RESULT_ERR;
2359
2360         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2361                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2362                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2363                                 msg->payload.u64);
2364
2365                 return RTE_VHOST_MSG_RESULT_ERR;
2366         }
2367
2368         dev->mtu = msg->payload.u64;
2369
2370         return RTE_VHOST_MSG_RESULT_OK;
2371 }
2372
2373 static int
2374 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2375                         int main_fd __rte_unused)
2376 {
2377         struct virtio_net *dev = *pdev;
2378         int fd = msg->fds[0];
2379
2380         if (validate_msg_fds(msg, 1) != 0)
2381                 return RTE_VHOST_MSG_RESULT_ERR;
2382
2383         if (fd < 0) {
2384                 VHOST_LOG_CONFIG(ERR,
2385                                 "Invalid file descriptor for slave channel (%d)\n",
2386                                 fd);
2387                 return RTE_VHOST_MSG_RESULT_ERR;
2388         }
2389
2390         if (dev->slave_req_fd >= 0)
2391                 close(dev->slave_req_fd);
2392
2393         dev->slave_req_fd = fd;
2394
2395         return RTE_VHOST_MSG_RESULT_OK;
2396 }
2397
2398 static int
2399 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2400 {
2401         struct vhost_vring_addr *ra;
2402         uint64_t start, end, len;
2403
2404         start = imsg->iova;
2405         end = start + imsg->size;
2406
2407         ra = &vq->ring_addrs;
2408         len = sizeof(struct vring_desc) * vq->size;
2409         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2410                 return 1;
2411
2412         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2413         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2414                 return 1;
2415
2416         len = sizeof(struct vring_used) +
2417                sizeof(struct vring_used_elem) * vq->size;
2418         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2419                 return 1;
2420
2421         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2422                 len = sizeof(uint64_t);
2423                 if (ra->log_guest_addr < end &&
2424                     (ra->log_guest_addr + len) > start)
2425                         return 1;
2426         }
2427
2428         return 0;
2429 }
2430
2431 static int
2432 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2433 {
2434         struct vhost_vring_addr *ra;
2435         uint64_t start, end, len;
2436
2437         start = imsg->iova;
2438         end = start + imsg->size;
2439
2440         ra = &vq->ring_addrs;
2441         len = sizeof(struct vring_packed_desc) * vq->size;
2442         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2443                 return 1;
2444
2445         len = sizeof(struct vring_packed_desc_event);
2446         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2447                 return 1;
2448
2449         len = sizeof(struct vring_packed_desc_event);
2450         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2451                 return 1;
2452
2453         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2454                 len = sizeof(uint64_t);
2455                 if (ra->log_guest_addr < end &&
2456                     (ra->log_guest_addr + len) > start)
2457                         return 1;
2458         }
2459
2460         return 0;
2461 }
2462
2463 static int is_vring_iotlb(struct virtio_net *dev,
2464                           struct vhost_virtqueue *vq,
2465                           struct vhost_iotlb_msg *imsg)
2466 {
2467         if (vq_is_packed(dev))
2468                 return is_vring_iotlb_packed(vq, imsg);
2469         else
2470                 return is_vring_iotlb_split(vq, imsg);
2471 }
2472
2473 static int
2474 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2475                         int main_fd __rte_unused)
2476 {
2477         struct virtio_net *dev = *pdev;
2478         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2479         uint16_t i;
2480         uint64_t vva, len;
2481
2482         if (validate_msg_fds(msg, 0) != 0)
2483                 return RTE_VHOST_MSG_RESULT_ERR;
2484
2485         switch (imsg->type) {
2486         case VHOST_IOTLB_UPDATE:
2487                 len = imsg->size;
2488                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2489                 if (!vva)
2490                         return RTE_VHOST_MSG_RESULT_ERR;
2491
2492                 for (i = 0; i < dev->nr_vring; i++) {
2493                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2494
2495                         if (!vq)
2496                                 continue;
2497
2498                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2499                                         len, imsg->perm);
2500
2501                         if (is_vring_iotlb(dev, vq, imsg))
2502                                 *pdev = dev = translate_ring_addresses(dev, i);
2503                 }
2504                 break;
2505         case VHOST_IOTLB_INVALIDATE:
2506                 for (i = 0; i < dev->nr_vring; i++) {
2507                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2508
2509                         if (!vq)
2510                                 continue;
2511
2512                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2513                                         imsg->size);
2514
2515                         if (is_vring_iotlb(dev, vq, imsg))
2516                                 vring_invalidate(dev, vq);
2517                 }
2518                 break;
2519         default:
2520                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2521                                 imsg->type);
2522                 return RTE_VHOST_MSG_RESULT_ERR;
2523         }
2524
2525         return RTE_VHOST_MSG_RESULT_OK;
2526 }
2527
2528 static int
2529 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2530                         struct VhostUserMsg *msg,
2531                         int main_fd __rte_unused)
2532 {
2533         struct virtio_net *dev = *pdev;
2534 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2535         struct uffdio_api api_struct;
2536
2537         if (validate_msg_fds(msg, 0) != 0)
2538                 return RTE_VHOST_MSG_RESULT_ERR;
2539
2540         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2541
2542         if (dev->postcopy_ufd == -1) {
2543                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2544                         strerror(errno));
2545                 return RTE_VHOST_MSG_RESULT_ERR;
2546         }
2547         api_struct.api = UFFD_API;
2548         api_struct.features = 0;
2549         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2550                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2551                         strerror(errno));
2552                 close(dev->postcopy_ufd);
2553                 dev->postcopy_ufd = -1;
2554                 return RTE_VHOST_MSG_RESULT_ERR;
2555         }
2556         msg->fds[0] = dev->postcopy_ufd;
2557         msg->fd_num = 1;
2558
2559         return RTE_VHOST_MSG_RESULT_REPLY;
2560 #else
2561         dev->postcopy_ufd = -1;
2562         msg->fd_num = 0;
2563
2564         return RTE_VHOST_MSG_RESULT_ERR;
2565 #endif
2566 }
2567
2568 static int
2569 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2570                         struct VhostUserMsg *msg __rte_unused,
2571                         int main_fd __rte_unused)
2572 {
2573         struct virtio_net *dev = *pdev;
2574
2575         if (validate_msg_fds(msg, 0) != 0)
2576                 return RTE_VHOST_MSG_RESULT_ERR;
2577
2578         if (dev->mem && dev->mem->nregions) {
2579                 VHOST_LOG_CONFIG(ERR,
2580                         "Regions already registered at postcopy-listen\n");
2581                 return RTE_VHOST_MSG_RESULT_ERR;
2582         }
2583         dev->postcopy_listening = 1;
2584
2585         return RTE_VHOST_MSG_RESULT_OK;
2586 }
2587
2588 static int
2589 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2590                         int main_fd __rte_unused)
2591 {
2592         struct virtio_net *dev = *pdev;
2593
2594         if (validate_msg_fds(msg, 0) != 0)
2595                 return RTE_VHOST_MSG_RESULT_ERR;
2596
2597         dev->postcopy_listening = 0;
2598         if (dev->postcopy_ufd >= 0) {
2599                 close(dev->postcopy_ufd);
2600                 dev->postcopy_ufd = -1;
2601         }
2602
2603         msg->payload.u64 = 0;
2604         msg->size = sizeof(msg->payload.u64);
2605         msg->fd_num = 0;
2606
2607         return RTE_VHOST_MSG_RESULT_REPLY;
2608 }
2609
2610 static int
2611 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2612                       int main_fd __rte_unused)
2613 {
2614         struct virtio_net *dev = *pdev;
2615
2616         if (validate_msg_fds(msg, 0) != 0)
2617                 return RTE_VHOST_MSG_RESULT_ERR;
2618
2619         msg->payload.u64 = dev->status;
2620         msg->size = sizeof(msg->payload.u64);
2621         msg->fd_num = 0;
2622
2623         return RTE_VHOST_MSG_RESULT_REPLY;
2624 }
2625
2626 static int
2627 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2628                         int main_fd __rte_unused)
2629 {
2630         struct virtio_net *dev = *pdev;
2631
2632         if (validate_msg_fds(msg, 0) != 0)
2633                 return RTE_VHOST_MSG_RESULT_ERR;
2634
2635         /* As per Virtio specification, the device status is 8bits long */
2636         if (msg->payload.u64 > UINT8_MAX) {
2637                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2638                                 msg->payload.u64);
2639                 return RTE_VHOST_MSG_RESULT_ERR;
2640         }
2641
2642         dev->status = msg->payload.u64;
2643
2644         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2645             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2646                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2647                 /*
2648                  * Clear the bit to let the driver know about the feature
2649                  * negotiation failure
2650                  */
2651                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2652         }
2653
2654         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2655                         "\t-RESET: %u\n"
2656                         "\t-ACKNOWLEDGE: %u\n"
2657                         "\t-DRIVER: %u\n"
2658                         "\t-FEATURES_OK: %u\n"
2659                         "\t-DRIVER_OK: %u\n"
2660                         "\t-DEVICE_NEED_RESET: %u\n"
2661                         "\t-FAILED: %u\n",
2662                         dev->status,
2663                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2664                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2665                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2666                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2667                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2668                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2669                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2670
2671         return RTE_VHOST_MSG_RESULT_OK;
2672 }
2673
2674 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2675                                         struct VhostUserMsg *msg,
2676                                         int main_fd);
2677 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2678         [VHOST_USER_NONE] = NULL,
2679         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2680         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2681         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2682         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2683         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2684         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2685         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2686         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2687         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2688         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2689         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2690         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2691         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2692         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2693         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2694         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2695         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2696         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2697         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2698         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2699         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2700         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2701         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2702         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2703         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2704         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2705         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2706         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2707         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2708 };
2709
2710 /* return bytes# of read on success or negative val on failure. */
2711 static int
2712 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2713 {
2714         int ret;
2715
2716         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2717                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2718         if (ret <= 0) {
2719                 return ret;
2720         } else if (ret != VHOST_USER_HDR_SIZE) {
2721                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2722                 close_msg_fds(msg);
2723                 return -1;
2724         }
2725
2726         if (msg->size) {
2727                 if (msg->size > sizeof(msg->payload)) {
2728                         VHOST_LOG_CONFIG(ERR,
2729                                 "invalid msg size: %d\n", msg->size);
2730                         return -1;
2731                 }
2732                 ret = read(sockfd, &msg->payload, msg->size);
2733                 if (ret <= 0)
2734                         return ret;
2735                 if (ret != (int)msg->size) {
2736                         VHOST_LOG_CONFIG(ERR,
2737                                 "read control message failed\n");
2738                         return -1;
2739                 }
2740         }
2741
2742         return ret;
2743 }
2744
2745 static int
2746 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2747 {
2748         if (!msg)
2749                 return 0;
2750
2751         return send_fd_message(sockfd, (char *)msg,
2752                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2753 }
2754
2755 static int
2756 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2757 {
2758         if (!msg)
2759                 return 0;
2760
2761         msg->flags &= ~VHOST_USER_VERSION_MASK;
2762         msg->flags &= ~VHOST_USER_NEED_REPLY;
2763         msg->flags |= VHOST_USER_VERSION;
2764         msg->flags |= VHOST_USER_REPLY_MASK;
2765
2766         return send_vhost_message(sockfd, msg);
2767 }
2768
2769 static int
2770 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2771 {
2772         int ret;
2773
2774         if (msg->flags & VHOST_USER_NEED_REPLY)
2775                 rte_spinlock_lock(&dev->slave_req_lock);
2776
2777         ret = send_vhost_message(dev->slave_req_fd, msg);
2778         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2779                 rte_spinlock_unlock(&dev->slave_req_lock);
2780
2781         return ret;
2782 }
2783
2784 /*
2785  * Allocate a queue pair if it hasn't been allocated yet
2786  */
2787 static int
2788 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2789                         struct VhostUserMsg *msg)
2790 {
2791         uint32_t vring_idx;
2792
2793         switch (msg->request.master) {
2794         case VHOST_USER_SET_VRING_KICK:
2795         case VHOST_USER_SET_VRING_CALL:
2796         case VHOST_USER_SET_VRING_ERR:
2797                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2798                 break;
2799         case VHOST_USER_SET_VRING_NUM:
2800         case VHOST_USER_SET_VRING_BASE:
2801         case VHOST_USER_GET_VRING_BASE:
2802         case VHOST_USER_SET_VRING_ENABLE:
2803                 vring_idx = msg->payload.state.index;
2804                 break;
2805         case VHOST_USER_SET_VRING_ADDR:
2806                 vring_idx = msg->payload.addr.index;
2807                 break;
2808         default:
2809                 return 0;
2810         }
2811
2812         if (vring_idx >= VHOST_MAX_VRING) {
2813                 VHOST_LOG_CONFIG(ERR,
2814                         "invalid vring index: %u\n", vring_idx);
2815                 return -1;
2816         }
2817
2818         if (dev->virtqueue[vring_idx])
2819                 return 0;
2820
2821         return alloc_vring_queue(dev, vring_idx);
2822 }
2823
2824 static void
2825 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2826 {
2827         unsigned int i = 0;
2828         unsigned int vq_num = 0;
2829
2830         while (vq_num < dev->nr_vring) {
2831                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2832
2833                 if (vq) {
2834                         rte_spinlock_lock(&vq->access_lock);
2835                         vq_num++;
2836                 }
2837                 i++;
2838         }
2839 }
2840
2841 static void
2842 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2843 {
2844         unsigned int i = 0;
2845         unsigned int vq_num = 0;
2846
2847         while (vq_num < dev->nr_vring) {
2848                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2849
2850                 if (vq) {
2851                         rte_spinlock_unlock(&vq->access_lock);
2852                         vq_num++;
2853                 }
2854                 i++;
2855         }
2856 }
2857
2858 int
2859 vhost_user_msg_handler(int vid, int fd)
2860 {
2861         struct virtio_net *dev;
2862         struct VhostUserMsg msg;
2863         struct rte_vdpa_device *vdpa_dev;
2864         int ret;
2865         int unlock_required = 0;
2866         bool handled;
2867         int request;
2868         uint32_t i;
2869
2870         dev = get_device(vid);
2871         if (dev == NULL)
2872                 return -1;
2873
2874         if (!dev->notify_ops) {
2875                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2876                 if (!dev->notify_ops) {
2877                         VHOST_LOG_CONFIG(ERR,
2878                                 "failed to get callback ops for driver %s\n",
2879                                 dev->ifname);
2880                         return -1;
2881                 }
2882         }
2883
2884         ret = read_vhost_message(fd, &msg);
2885         if (ret <= 0) {
2886                 if (ret < 0)
2887                         VHOST_LOG_CONFIG(ERR,
2888                                 "vhost read message failed\n");
2889                 else
2890                         VHOST_LOG_CONFIG(INFO,
2891                                 "vhost peer closed\n");
2892
2893                 return -1;
2894         }
2895
2896         ret = 0;
2897         request = msg.request.master;
2898         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2899                         vhost_message_str[request]) {
2900                 if (request != VHOST_USER_IOTLB_MSG)
2901                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2902                                 vhost_message_str[request]);
2903                 else
2904                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2905                                 vhost_message_str[request]);
2906         } else {
2907                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2908         }
2909
2910         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2911         if (ret < 0) {
2912                 VHOST_LOG_CONFIG(ERR,
2913                         "failed to alloc queue\n");
2914                 return -1;
2915         }
2916
2917         /*
2918          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2919          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2920          * and device is destroyed. destroy_device waits for queues to be
2921          * inactive, so it is safe. Otherwise taking the access_lock
2922          * would cause a dead lock.
2923          */
2924         switch (request) {
2925         case VHOST_USER_SET_FEATURES:
2926         case VHOST_USER_SET_PROTOCOL_FEATURES:
2927         case VHOST_USER_SET_OWNER:
2928         case VHOST_USER_SET_MEM_TABLE:
2929         case VHOST_USER_SET_LOG_BASE:
2930         case VHOST_USER_SET_LOG_FD:
2931         case VHOST_USER_SET_VRING_NUM:
2932         case VHOST_USER_SET_VRING_ADDR:
2933         case VHOST_USER_SET_VRING_BASE:
2934         case VHOST_USER_SET_VRING_KICK:
2935         case VHOST_USER_SET_VRING_CALL:
2936         case VHOST_USER_SET_VRING_ERR:
2937         case VHOST_USER_SET_VRING_ENABLE:
2938         case VHOST_USER_SEND_RARP:
2939         case VHOST_USER_NET_SET_MTU:
2940         case VHOST_USER_SET_SLAVE_REQ_FD:
2941                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2942                         vhost_user_lock_all_queue_pairs(dev);
2943                         unlock_required = 1;
2944                 }
2945                 break;
2946         default:
2947                 break;
2948
2949         }
2950
2951         handled = false;
2952         if (dev->extern_ops.pre_msg_handle) {
2953                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2954                                 (void *)&msg);
2955                 switch (ret) {
2956                 case RTE_VHOST_MSG_RESULT_REPLY:
2957                         send_vhost_reply(fd, &msg);
2958                         /* Fall-through */
2959                 case RTE_VHOST_MSG_RESULT_ERR:
2960                 case RTE_VHOST_MSG_RESULT_OK:
2961                         handled = true;
2962                         goto skip_to_post_handle;
2963                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2964                 default:
2965                         break;
2966                 }
2967         }
2968
2969         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2970                 if (!vhost_message_handlers[request])
2971                         goto skip_to_post_handle;
2972                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2973
2974                 switch (ret) {
2975                 case RTE_VHOST_MSG_RESULT_ERR:
2976                         VHOST_LOG_CONFIG(ERR,
2977                                 "Processing %s failed.\n",
2978                                 vhost_message_str[request]);
2979                         handled = true;
2980                         break;
2981                 case RTE_VHOST_MSG_RESULT_OK:
2982                         VHOST_LOG_CONFIG(DEBUG,
2983                                 "Processing %s succeeded.\n",
2984                                 vhost_message_str[request]);
2985                         handled = true;
2986                         break;
2987                 case RTE_VHOST_MSG_RESULT_REPLY:
2988                         VHOST_LOG_CONFIG(DEBUG,
2989                                 "Processing %s succeeded and needs reply.\n",
2990                                 vhost_message_str[request]);
2991                         send_vhost_reply(fd, &msg);
2992                         handled = true;
2993                         break;
2994                 default:
2995                         break;
2996                 }
2997         }
2998
2999 skip_to_post_handle:
3000         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
3001                         dev->extern_ops.post_msg_handle) {
3002                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
3003                                 (void *)&msg);
3004                 switch (ret) {
3005                 case RTE_VHOST_MSG_RESULT_REPLY:
3006                         send_vhost_reply(fd, &msg);
3007                         /* Fall-through */
3008                 case RTE_VHOST_MSG_RESULT_ERR:
3009                 case RTE_VHOST_MSG_RESULT_OK:
3010                         handled = true;
3011                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3012                 default:
3013                         break;
3014                 }
3015         }
3016
3017         /* If message was not handled at this stage, treat it as an error */
3018         if (!handled) {
3019                 VHOST_LOG_CONFIG(ERR,
3020                         "vhost message (req: %d) was not handled.\n", request);
3021                 close_msg_fds(&msg);
3022                 ret = RTE_VHOST_MSG_RESULT_ERR;
3023         }
3024
3025         /*
3026          * If the request required a reply that was already sent,
3027          * this optional reply-ack won't be sent as the
3028          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
3029          */
3030         if (msg.flags & VHOST_USER_NEED_REPLY) {
3031                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
3032                 msg.size = sizeof(msg.payload.u64);
3033                 msg.fd_num = 0;
3034                 send_vhost_reply(fd, &msg);
3035         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
3036                 VHOST_LOG_CONFIG(ERR,
3037                         "vhost message handling failed.\n");
3038                 return -1;
3039         }
3040
3041         for (i = 0; i < dev->nr_vring; i++) {
3042                 struct vhost_virtqueue *vq = dev->virtqueue[i];
3043                 bool cur_ready = vq_is_ready(dev, vq);
3044
3045                 if (cur_ready != (vq && vq->ready)) {
3046                         vq->ready = cur_ready;
3047                         vhost_user_notify_queue_state(dev, i, cur_ready);
3048                 }
3049         }
3050
3051         if (unlock_required)
3052                 vhost_user_unlock_all_queue_pairs(dev);
3053
3054         if (!virtio_is_ready(dev))
3055                 goto out;
3056
3057         /*
3058          * Virtio is now ready. If not done already, it is time
3059          * to notify the application it can process the rings and
3060          * configure the vDPA device if present.
3061          */
3062
3063         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
3064                 if (dev->notify_ops->new_device(dev->vid) == 0)
3065                         dev->flags |= VIRTIO_DEV_RUNNING;
3066         }
3067
3068         vdpa_dev = dev->vdpa_dev;
3069         if (!vdpa_dev)
3070                 goto out;
3071
3072         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3073                 if (vdpa_dev->ops->dev_conf(dev->vid))
3074                         VHOST_LOG_CONFIG(ERR,
3075                                          "Failed to configure vDPA device\n");
3076                 else
3077                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3078         }
3079
3080 out:
3081         return 0;
3082 }
3083
3084 static int process_slave_message_reply(struct virtio_net *dev,
3085                                        const struct VhostUserMsg *msg)
3086 {
3087         struct VhostUserMsg msg_reply;
3088         int ret;
3089
3090         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3091                 return 0;
3092
3093         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3094         if (ret <= 0) {
3095                 if (ret < 0)
3096                         VHOST_LOG_CONFIG(ERR,
3097                                 "vhost read slave message reply failed\n");
3098                 else
3099                         VHOST_LOG_CONFIG(INFO,
3100                                 "vhost peer closed\n");
3101                 ret = -1;
3102                 goto out;
3103         }
3104
3105         ret = 0;
3106         if (msg_reply.request.slave != msg->request.slave) {
3107                 VHOST_LOG_CONFIG(ERR,
3108                         "Received unexpected msg type (%u), expected %u\n",
3109                         msg_reply.request.slave, msg->request.slave);
3110                 ret = -1;
3111                 goto out;
3112         }
3113
3114         ret = msg_reply.payload.u64 ? -1 : 0;
3115
3116 out:
3117         rte_spinlock_unlock(&dev->slave_req_lock);
3118         return ret;
3119 }
3120
3121 int
3122 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3123 {
3124         int ret;
3125         struct VhostUserMsg msg = {
3126                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3127                 .flags = VHOST_USER_VERSION,
3128                 .size = sizeof(msg.payload.iotlb),
3129                 .payload.iotlb = {
3130                         .iova = iova,
3131                         .perm = perm,
3132                         .type = VHOST_IOTLB_MISS,
3133                 },
3134         };
3135
3136         ret = send_vhost_message(dev->slave_req_fd, &msg);
3137         if (ret < 0) {
3138                 VHOST_LOG_CONFIG(ERR,
3139                                 "Failed to send IOTLB miss message (%d)\n",
3140                                 ret);
3141                 return ret;
3142         }
3143
3144         return 0;
3145 }
3146
3147 static int
3148 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3149 {
3150         int ret;
3151         struct VhostUserMsg msg = {
3152                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3153                 .flags = VHOST_USER_VERSION,
3154                 .size = 0,
3155         };
3156
3157         if (need_reply)
3158                 msg.flags |= VHOST_USER_NEED_REPLY;
3159
3160         ret = send_vhost_slave_message(dev, &msg);
3161         if (ret < 0) {
3162                 VHOST_LOG_CONFIG(ERR,
3163                                 "Failed to send config change (%d)\n",
3164                                 ret);
3165                 return ret;
3166         }
3167
3168         return process_slave_message_reply(dev, &msg);
3169 }
3170
3171 int
3172 rte_vhost_slave_config_change(int vid, bool need_reply)
3173 {
3174         struct virtio_net *dev;
3175
3176         dev = get_device(vid);
3177         if (!dev)
3178                 return -ENODEV;
3179
3180         return vhost_user_slave_config_change(dev, need_reply);
3181 }
3182
3183 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3184                                                     int index, int fd,
3185                                                     uint64_t offset,
3186                                                     uint64_t size)
3187 {
3188         int ret;
3189         struct VhostUserMsg msg = {
3190                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3191                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3192                 .size = sizeof(msg.payload.area),
3193                 .payload.area = {
3194                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3195                         .size = size,
3196                         .offset = offset,
3197                 },
3198         };
3199
3200         if (fd < 0)
3201                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3202         else {
3203                 msg.fds[0] = fd;
3204                 msg.fd_num = 1;
3205         }
3206
3207         ret = send_vhost_slave_message(dev, &msg);
3208         if (ret < 0) {
3209                 VHOST_LOG_CONFIG(ERR,
3210                         "Failed to set host notifier (%d)\n", ret);
3211                 return ret;
3212         }
3213
3214         return process_slave_message_reply(dev, &msg);
3215 }
3216
3217 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3218 {
3219         struct virtio_net *dev;
3220         struct rte_vdpa_device *vdpa_dev;
3221         int vfio_device_fd, ret = 0;
3222         uint64_t offset, size;
3223         unsigned int i, q_start, q_last;
3224
3225         dev = get_device(vid);
3226         if (!dev)
3227                 return -ENODEV;
3228
3229         vdpa_dev = dev->vdpa_dev;
3230         if (vdpa_dev == NULL)
3231                 return -ENODEV;
3232
3233         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3234             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3235             !(dev->protocol_features &
3236                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3237             !(dev->protocol_features &
3238                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3239             !(dev->protocol_features &
3240                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3241                 return -ENOTSUP;
3242
3243         if (qid == RTE_VHOST_QUEUE_ALL) {
3244                 q_start = 0;
3245                 q_last = dev->nr_vring - 1;
3246         } else {
3247                 if (qid >= dev->nr_vring)
3248                         return -EINVAL;
3249                 q_start = qid;
3250                 q_last = qid;
3251         }
3252
3253         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3254         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3255
3256         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3257         if (vfio_device_fd < 0)
3258                 return -ENOTSUP;
3259
3260         if (enable) {
3261                 for (i = q_start; i <= q_last; i++) {
3262                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3263                                         &size) < 0) {
3264                                 ret = -ENOTSUP;
3265                                 goto disable;
3266                         }
3267
3268                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3269                                         vfio_device_fd, offset, size) < 0) {
3270                                 ret = -EFAULT;
3271                                 goto disable;
3272                         }
3273                 }
3274         } else {
3275 disable:
3276                 for (i = q_start; i <= q_last; i++) {
3277                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3278                                         0, 0);
3279                 }
3280         }
3281
3282         return ret;
3283 }