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                 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(NULL,
 437                                 vq->size *
 438                                 sizeof(struct vring_used_elem_packed),
 439                                 RTE_CACHE_LINE_SIZE);
 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(NULL,
 451                                 vq->size * sizeof(struct vring_used_elem),
 452                                 RTE_CACHE_LINE_SIZE);
 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(NULL,
 464                                 vq->size * sizeof(struct batch_copy_elem),
 465                                 RTE_CACHE_LINE_SIZE);
 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 oldnode, newnode;
 484         struct virtio_net *old_dev;
 485         struct vhost_virtqueue *old_vq, *vq;
 486         struct vring_used_elem *new_shadow_used_split;
 487         struct vring_used_elem_packed *new_shadow_used_packed;
 488         struct batch_copy_elem *new_batch_copy_elems;
 489         int ret;
 490
 491         if (dev->flags & VIRTIO_DEV_RUNNING)
 492                 return dev;
 493
 494         old_dev = dev;
 495         vq = old_vq = dev->virtqueue[index];
 496
 497         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 498                             MPOL_F_NODE | MPOL_F_ADDR);
 499
 500         /* check if we need to reallocate vq */
 501         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 502                              MPOL_F_NODE | MPOL_F_ADDR);
 503         if (ret) {
 504                 VHOST_LOG_CONFIG(ERR,
 505                         "Unable to get vq numa information.\n");
 506                 return dev;
 507         }
 508         if (oldnode != newnode) {
 509                 VHOST_LOG_CONFIG(INFO,
 510                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 511                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 512                 if (!vq)
 513                         return dev;
 514
 515                 memcpy(vq, old_vq, sizeof(*vq));
 516
 517                 if (vq_is_packed(dev)) {
 518                         new_shadow_used_packed = rte_malloc_socket(NULL,
 519                                         vq->size *
 520                                         sizeof(struct vring_used_elem_packed),
 521                                         RTE_CACHE_LINE_SIZE,
 522                                         newnode);
 523                         if (new_shadow_used_packed) {
 524                                 rte_free(vq->shadow_used_packed);
 525                                 vq->shadow_used_packed = new_shadow_used_packed;
 526                         }
 527                 } else {
 528                         new_shadow_used_split = rte_malloc_socket(NULL,
 529                                         vq->size *
 530                                         sizeof(struct vring_used_elem),
 531                                         RTE_CACHE_LINE_SIZE,
 532                                         newnode);
 533                         if (new_shadow_used_split) {
 534                                 rte_free(vq->shadow_used_split);
 535                                 vq->shadow_used_split = new_shadow_used_split;
 536                         }
 537                 }
 538
 539                 new_batch_copy_elems = rte_malloc_socket(NULL,
 540                         vq->size * sizeof(struct batch_copy_elem),
 541                         RTE_CACHE_LINE_SIZE,
 542                         newnode);
 543                 if (new_batch_copy_elems) {
 544                         rte_free(vq->batch_copy_elems);
 545                         vq->batch_copy_elems = new_batch_copy_elems;
 546                 }
 547
 548                 if (vq->log_cache) {
 549                         struct log_cache_entry *log_cache;
 550
 551                         log_cache = rte_realloc_socket(vq->log_cache,
 552                                         sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
 553                                         0, newnode);
 554                         if (log_cache)
 555                                 vq->log_cache = log_cache;
 556                 }
 557
 558                 rte_free(old_vq);
 559         }
 560
 561         /* check if we need to reallocate dev */
 562         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 563                             MPOL_F_NODE | MPOL_F_ADDR);
 564         if (ret) {
 565                 VHOST_LOG_CONFIG(ERR,
 566                         "Unable to get dev numa information.\n");
 567                 goto out;
 568         }
 569         if (oldnode != newnode) {
 570                 struct rte_vhost_memory *old_mem;
 571                 struct guest_page *old_gp;
 572                 ssize_t mem_size, gp_size;
 573
 574                 VHOST_LOG_CONFIG(INFO,
 575                         "reallocate dev from %d to %d node\n",
 576                         oldnode, newnode);
 577                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 578                 if (!dev) {
 579                         dev = old_dev;
 580                         goto out;
 581                 }
 582
 583                 memcpy(dev, old_dev, sizeof(*dev));
 584                 rte_free(old_dev);
 585
 586                 mem_size = sizeof(struct rte_vhost_memory) +
 587                         sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
 588                 old_mem = dev->mem;
 589                 dev->mem = rte_malloc_socket(NULL, mem_size, 0, newnode);
 590                 if (!dev->mem) {
 591                         dev->mem = old_mem;
 592                         goto out;
 593                 }
 594
 595                 memcpy(dev->mem, old_mem, mem_size);
 596                 rte_free(old_mem);
 597
 598                 gp_size = dev->max_guest_pages * sizeof(*dev->guest_pages);
 599                 old_gp = dev->guest_pages;
 600                 dev->guest_pages = rte_malloc_socket(NULL, gp_size, RTE_CACHE_LINE_SIZE, newnode);
 601                 if (!dev->guest_pages) {
 602                         dev->guest_pages = old_gp;
 603                         goto out;
 604                 }
 605
 606                 memcpy(dev->guest_pages, old_gp, gp_size);
 607                 rte_free(old_gp);
 608         }
 609
 610 out:
 611         dev->virtqueue[index] = vq;
 612         vhost_devices[dev->vid] = dev;
 613
 614         if (old_vq != vq)
 615                 vhost_user_iotlb_init(dev, index);
 616
 617         return dev;
 618 }
 619 #else
 620 static struct virtio_net*
 621 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 622 {
 623         return dev;
 624 }
 625 #endif
 626
 627 /* Converts QEMU virtual address to Vhost virtual address. */
 628 static uint64_t
 629 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 630 {
 631         struct rte_vhost_mem_region *r;
 632         uint32_t i;
 633
 634         if (unlikely(!dev || !dev->mem))
 635                 goto out_error;
 636
 637         /* Find the region where the address lives. */
 638         for (i = 0; i < dev->mem->nregions; i++) {
 639                 r = &dev->mem->regions[i];
 640
 641                 if (qva >= r->guest_user_addr &&
 642                     qva <  r->guest_user_addr + r->size) {
 643
 644                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 645                                 *len = r->guest_user_addr + r->size - qva;
 646
 647                         return qva - r->guest_user_addr +
 648                                r->host_user_addr;
 649                 }
 650         }
 651 out_error:
 652         *len = 0;
 653
 654         return 0;
 655 }
 656
 657
 658 /*
 659  * Converts ring address to Vhost virtual address.
 660  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 661  * else it is a QEMU virtual address.
 662  */
 663 static uint64_t
 664 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 665                 uint64_t ra, uint64_t *size)
 666 {
 667         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 668                 uint64_t vva;
 669
 670                 vhost_user_iotlb_rd_lock(vq);
 671                 vva = vhost_iova_to_vva(dev, vq, ra,
 672                                         size, VHOST_ACCESS_RW);
 673                 vhost_user_iotlb_rd_unlock(vq);
 674
 675                 return vva;
 676         }
 677
 678         return qva_to_vva(dev, ra, size);
 679 }
 680
 681 static uint64_t
 682 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 683 {
 684         uint64_t log_gpa;
 685
 686         vhost_user_iotlb_rd_lock(vq);
 687         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 688         vhost_user_iotlb_rd_unlock(vq);
 689
 690         return log_gpa;
 691 }
 692
 693 static struct virtio_net *
 694 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 695 {
 696         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 697         struct vhost_vring_addr *addr = &vq->ring_addrs;
 698         uint64_t len, expected_len;
 699
 700         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 701                 vq->log_guest_addr =
 702                         log_addr_to_gpa(dev, vq);
 703                 if (vq->log_guest_addr == 0) {
 704                         VHOST_LOG_CONFIG(DEBUG,
 705                                 "(%d) failed to map log_guest_addr.\n",
 706                                 dev->vid);
 707                         return dev;
 708                 }
 709         }
 710
 711         if (vq_is_packed(dev)) {
 712                 len = sizeof(struct vring_packed_desc) * vq->size;
 713                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 714                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 715                 if (vq->desc_packed == NULL ||
 716                                 len != sizeof(struct vring_packed_desc) *
 717                                 vq->size) {
 718                         VHOST_LOG_CONFIG(DEBUG,
 719                                 "(%d) failed to map desc_packed ring.\n",
 720                                 dev->vid);
 721                         return dev;
 722                 }
 723
 724                 dev = numa_realloc(dev, vq_index);
 725                 vq = dev->virtqueue[vq_index];
 726                 addr = &vq->ring_addrs;
 727
 728                 len = sizeof(struct vring_packed_desc_event);
 729                 vq->driver_event = (struct vring_packed_desc_event *)
 730                                         (uintptr_t)ring_addr_to_vva(dev,
 731                                         vq, addr->avail_user_addr, &len);
 732                 if (vq->driver_event == NULL ||
 733                                 len != sizeof(struct vring_packed_desc_event)) {
 734                         VHOST_LOG_CONFIG(DEBUG,
 735                                 "(%d) failed to find driver area address.\n",
 736                                 dev->vid);
 737                         return dev;
 738                 }
 739
 740                 len = sizeof(struct vring_packed_desc_event);
 741                 vq->device_event = (struct vring_packed_desc_event *)
 742                                         (uintptr_t)ring_addr_to_vva(dev,
 743                                         vq, addr->used_user_addr, &len);
 744                 if (vq->device_event == NULL ||
 745                                 len != sizeof(struct vring_packed_desc_event)) {
 746                         VHOST_LOG_CONFIG(DEBUG,
 747                                 "(%d) failed to find device area address.\n",
 748                                 dev->vid);
 749                         return dev;
 750                 }
 751
 752                 vq->access_ok = true;
 753                 return dev;
 754         }
 755
 756         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 757         if (vq->desc && vq->avail && vq->used)
 758                 return dev;
 759
 760         len = sizeof(struct vring_desc) * vq->size;
 761         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 762                         vq, addr->desc_user_addr, &len);
 763         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 764                 VHOST_LOG_CONFIG(DEBUG,
 765                         "(%d) failed to map desc ring.\n",
 766                         dev->vid);
 767                 return dev;
 768         }
 769
 770         dev = numa_realloc(dev, vq_index);
 771         vq = dev->virtqueue[vq_index];
 772         addr = &vq->ring_addrs;
 773
 774         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 775         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 776                 len += sizeof(uint16_t);
 777         expected_len = len;
 778         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 779                         vq, addr->avail_user_addr, &len);
 780         if (vq->avail == 0 || len != expected_len) {
 781                 VHOST_LOG_CONFIG(DEBUG,
 782                         "(%d) failed to map avail ring.\n",
 783                         dev->vid);
 784                 return dev;
 785         }
 786
 787         len = sizeof(struct vring_used) +
 788                 sizeof(struct vring_used_elem) * vq->size;
 789         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 790                 len += sizeof(uint16_t);
 791         expected_len = len;
 792         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 793                         vq, addr->used_user_addr, &len);
 794         if (vq->used == 0 || len != expected_len) {
 795                 VHOST_LOG_CONFIG(DEBUG,
 796                         "(%d) failed to map used ring.\n",
 797                         dev->vid);
 798                 return dev;
 799         }
 800
 801         if (vq->last_used_idx != vq->used->idx) {
 802                 VHOST_LOG_CONFIG(WARNING,
 803                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 804                         "some packets maybe resent for Tx and dropped for Rx\n",
 805                         vq->last_used_idx, vq->used->idx);
 806                 vq->last_used_idx  = vq->used->idx;
 807                 vq->last_avail_idx = vq->used->idx;
 808         }
 809
 810         vq->access_ok = true;
 811
 812         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 813                         dev->vid, vq->desc);
 814         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 815                         dev->vid, vq->avail);
 816         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 817                         dev->vid, vq->used);
 818         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 819                         dev->vid, vq->log_guest_addr);
 820
 821         return dev;
 822 }
 823
 824 /*
 825  * The virtio device sends us the desc, used and avail ring addresses.
 826  * This function then converts these to our address space.
 827  */
 828 static int
 829 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 830                         int main_fd __rte_unused)
 831 {
 832         struct virtio_net *dev = *pdev;
 833         struct vhost_virtqueue *vq;
 834         struct vhost_vring_addr *addr = &msg->payload.addr;
 835         bool access_ok;
 836
 837         if (validate_msg_fds(msg, 0) != 0)
 838                 return RTE_VHOST_MSG_RESULT_ERR;
 839
 840         if (dev->mem == NULL)
 841                 return RTE_VHOST_MSG_RESULT_ERR;
 842
 843         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 844         vq = dev->virtqueue[msg->payload.addr.index];
 845
 846         access_ok = vq->access_ok;
 847
 848         /*
 849          * Rings addresses should not be interpreted as long as the ring is not
 850          * started and enabled
 851          */
 852         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 853
 854         vring_invalidate(dev, vq);
 855
 856         if ((vq->enabled && (dev->features &
 857                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 858                         access_ok) {
 859                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 860                 if (!dev)
 861                         return RTE_VHOST_MSG_RESULT_ERR;
 862
 863                 *pdev = dev;
 864         }
 865
 866         return RTE_VHOST_MSG_RESULT_OK;
 867 }
 868
 869 /*
 870  * The virtio device sends us the available ring last used index.
 871  */
 872 static int
 873 vhost_user_set_vring_base(struct virtio_net **pdev,
 874                         struct VhostUserMsg *msg,
 875                         int main_fd __rte_unused)
 876 {
 877         struct virtio_net *dev = *pdev;
 878         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 879         uint64_t val = msg->payload.state.num;
 880
 881         if (validate_msg_fds(msg, 0) != 0)
 882                 return RTE_VHOST_MSG_RESULT_ERR;
 883
 884         if (vq_is_packed(dev)) {
 885                 /*
 886                  * Bit[0:14]: avail index
 887                  * Bit[15]: avail wrap counter
 888                  */
 889                 vq->last_avail_idx = val & 0x7fff;
 890                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 891                 /*
 892                  * Set used index to same value as available one, as
 893                  * their values should be the same since ring processing
 894                  * was stopped at get time.
 895                  */
 896                 vq->last_used_idx = vq->last_avail_idx;
 897                 vq->used_wrap_counter = vq->avail_wrap_counter;
 898         } else {
 899                 vq->last_used_idx = msg->payload.state.num;
 900                 vq->last_avail_idx = msg->payload.state.num;
 901         }
 902
 903         return RTE_VHOST_MSG_RESULT_OK;
 904 }
 905
 906 static int
 907 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 908                    uint64_t host_phys_addr, uint64_t size)
 909 {
 910         struct guest_page *page, *last_page;
 911         struct guest_page *old_pages;
 912
 913         if (dev->nr_guest_pages == dev->max_guest_pages) {
 914                 dev->max_guest_pages *= 2;
 915                 old_pages = dev->guest_pages;
 916                 dev->guest_pages = rte_realloc(dev->guest_pages,
 917                                         dev->max_guest_pages * sizeof(*page),
 918                                         RTE_CACHE_LINE_SIZE);
 919                 if (dev->guest_pages == NULL) {
 920                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 921                         rte_free(old_pages);
 922                         return -1;
 923                 }
 924         }
 925
 926         if (dev->nr_guest_pages > 0) {
 927                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 928                 /* merge if the two pages are continuous */
 929                 if (host_phys_addr == last_page->host_phys_addr +
 930                                       last_page->size) {
 931                         last_page->size += size;
 932                         return 0;
 933                 }
 934         }
 935
 936         page = &dev->guest_pages[dev->nr_guest_pages++];
 937         page->guest_phys_addr = guest_phys_addr;
 938         page->host_phys_addr  = host_phys_addr;
 939         page->size = size;
 940
 941         return 0;
 942 }
 943
 944 static int
 945 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 946                 uint64_t page_size)
 947 {
 948         uint64_t reg_size = reg->size;
 949         uint64_t host_user_addr  = reg->host_user_addr;
 950         uint64_t guest_phys_addr = reg->guest_phys_addr;
 951         uint64_t host_phys_addr;
 952         uint64_t size;
 953
 954         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 955         size = page_size - (guest_phys_addr & (page_size - 1));
 956         size = RTE_MIN(size, reg_size);
 957
 958         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 959                 return -1;
 960
 961         host_user_addr  += size;
 962         guest_phys_addr += size;
 963         reg_size -= size;
 964
 965         while (reg_size > 0) {
 966                 size = RTE_MIN(reg_size, page_size);
 967                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 968                                                   host_user_addr);
 969                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 970                                 size) < 0)
 971                         return -1;
 972
 973                 host_user_addr  += size;
 974                 guest_phys_addr += size;
 975                 reg_size -= size;
 976         }
 977
 978         /* sort guest page array if over binary search threshold */
 979         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 980                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 981                         sizeof(struct guest_page), guest_page_addrcmp);
 982         }
 983
 984         return 0;
 985 }
 986
 987 #ifdef RTE_LIBRTE_VHOST_DEBUG
 988 /* TODO: enable it only in debug mode? */
 989 static void
 990 dump_guest_pages(struct virtio_net *dev)
 991 {
 992         uint32_t i;
 993         struct guest_page *page;
 994
 995         for (i = 0; i < dev->nr_guest_pages; i++) {
 996                 page = &dev->guest_pages[i];
 997
 998                 VHOST_LOG_CONFIG(INFO,
 999                         "guest physical page region %u\n"
1000                         "\t guest_phys_addr: %" PRIx64 "\n"
1001                         "\t host_phys_addr : %" PRIx64 "\n"
1002                         "\t size           : %" PRIx64 "\n",
1003                         i,
1004                         page->guest_phys_addr,
1005                         page->host_phys_addr,
1006                         page->size);
1007         }
1008 }
1009 #else
1010 #define dump_guest_pages(dev)
1011 #endif
1012
1013 static bool
1014 vhost_memory_changed(struct VhostUserMemory *new,
1015                      struct rte_vhost_memory *old)
1016 {
1017         uint32_t i;
1018
1019         if (new->nregions != old->nregions)
1020                 return true;
1021
1022         for (i = 0; i < new->nregions; ++i) {
1023                 VhostUserMemoryRegion *new_r = &new->regions[i];
1024                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1025
1026                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1027                         return true;
1028                 if (new_r->memory_size != old_r->size)
1029                         return true;
1030                 if (new_r->userspace_addr != old_r->guest_user_addr)
1031                         return true;
1032         }
1033
1034         return false;
1035 }
1036
1037 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1038 static int
1039 vhost_user_postcopy_region_register(struct virtio_net *dev,
1040                 struct rte_vhost_mem_region *reg)
1041 {
1042         struct uffdio_register reg_struct;
1043
1044         /*
1045          * Let's register all the mmap'ed area to ensure
1046          * alignment on page boundary.
1047          */
1048         reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1049         reg_struct.range.len = reg->mmap_size;
1050         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1051
1052         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1053                                 &reg_struct)) {
1054                 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1055                                 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1056                                 (uint64_t)reg_struct.range.start,
1057                                 (uint64_t)reg_struct.range.start +
1058                                 (uint64_t)reg_struct.range.len - 1,
1059                                 dev->postcopy_ufd,
1060                                 strerror(errno));
1061                 return -1;
1062         }
1063
1064         VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1065                         (uint64_t)reg_struct.range.start,
1066                         (uint64_t)reg_struct.range.start +
1067                         (uint64_t)reg_struct.range.len - 1);
1068
1069         return 0;
1070 }
1071 #else
1072 static int
1073 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1074                 struct rte_vhost_mem_region *reg __rte_unused)
1075 {
1076         return -1;
1077 }
1078 #endif
1079
1080 static int
1081 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1082                 struct VhostUserMsg *msg)
1083 {
1084         struct VhostUserMemory *memory;
1085         struct rte_vhost_mem_region *reg;
1086         VhostUserMsg ack_msg;
1087         uint32_t i;
1088
1089         if (!dev->postcopy_listening)
1090                 return 0;
1091
1092         /*
1093          * We haven't a better way right now than sharing
1094          * DPDK's virtual address with Qemu, so that Qemu can
1095          * retrieve the region offset when handling userfaults.
1096          */
1097         memory = &msg->payload.memory;
1098         for (i = 0; i < memory->nregions; i++) {
1099                 reg = &dev->mem->regions[i];
1100                 memory->regions[i].userspace_addr = reg->host_user_addr;
1101         }
1102
1103         /* Send the addresses back to qemu */
1104         msg->fd_num = 0;
1105         send_vhost_reply(main_fd, msg);
1106
1107         /* Wait for qemu to acknolwedge it's got the addresses
1108          * we've got to wait before we're allowed to generate faults.
1109          */
1110         if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1111                 VHOST_LOG_CONFIG(ERR,
1112                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1113                 return -1;
1114         }
1115
1116         if (validate_msg_fds(&ack_msg, 0) != 0)
1117                 return -1;
1118
1119         if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1120                 VHOST_LOG_CONFIG(ERR,
1121                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1122                                 ack_msg.request.master);
1123                 return -1;
1124         }
1125
1126         /* Now userfault register and we can use the memory */
1127         for (i = 0; i < memory->nregions; i++) {
1128                 reg = &dev->mem->regions[i];
1129                 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1130                         return -1;
1131         }
1132
1133         return 0;
1134 }
1135
1136 static int
1137 vhost_user_mmap_region(struct virtio_net *dev,
1138                 struct rte_vhost_mem_region *region,
1139                 uint64_t mmap_offset)
1140 {
1141         void *mmap_addr;
1142         uint64_t mmap_size;
1143         uint64_t alignment;
1144         int populate;
1145
1146         /* Check for memory_size + mmap_offset overflow */
1147         if (mmap_offset >= -region->size) {
1148                 VHOST_LOG_CONFIG(ERR,
1149                                 "mmap_offset (%#"PRIx64") and memory_size "
1150                                 "(%#"PRIx64") overflow\n",
1151                                 mmap_offset, region->size);
1152                 return -1;
1153         }
1154
1155         mmap_size = region->size + mmap_offset;
1156
1157         /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1158          * argument aligned with hugepagesz at older longterm version Linux,
1159          * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1160          *
1161          * To avoid failure, make sure in caller to keep length aligned.
1162          */
1163         alignment = get_blk_size(region->fd);
1164         if (alignment == (uint64_t)-1) {
1165                 VHOST_LOG_CONFIG(ERR,
1166                                 "couldn't get hugepage size through fstat\n");
1167                 return -1;
1168         }
1169         mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1170         if (mmap_size == 0) {
1171                 /*
1172                  * It could happen if initial mmap_size + alignment overflows
1173                  * the sizeof uint64, which could happen if either mmap_size or
1174                  * alignment value is wrong.
1175                  *
1176                  * mmap() kernel implementation would return an error, but
1177                  * better catch it before and provide useful info in the logs.
1178                  */
1179                 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1180                                 "or alignment (0x%" PRIx64 ") is invalid\n",
1181                                 region->size + mmap_offset, alignment);
1182                 return -1;
1183         }
1184
1185         populate = dev->async_copy ? MAP_POPULATE : 0;
1186         mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1187                         MAP_SHARED | populate, region->fd, 0);
1188
1189         if (mmap_addr == MAP_FAILED) {
1190                 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1191                 return -1;
1192         }
1193
1194         region->mmap_addr = mmap_addr;
1195         region->mmap_size = mmap_size;
1196         region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1197
1198         if (dev->async_copy)
1199                 if (add_guest_pages(dev, region, alignment) < 0) {
1200                         VHOST_LOG_CONFIG(ERR,
1201                                         "adding guest pages to region failed.\n");
1202                         return -1;
1203                 }
1204
1205         VHOST_LOG_CONFIG(INFO,
1206                         "guest memory region size: 0x%" PRIx64 "\n"
1207                         "\t guest physical addr: 0x%" PRIx64 "\n"
1208                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1209                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1210                         "\t mmap addr : 0x%" PRIx64 "\n"
1211                         "\t mmap size : 0x%" PRIx64 "\n"
1212                         "\t mmap align: 0x%" PRIx64 "\n"
1213                         "\t mmap off  : 0x%" PRIx64 "\n",
1214                         region->size,
1215                         region->guest_phys_addr,
1216                         region->guest_user_addr,
1217                         region->host_user_addr,
1218                         (uint64_t)(uintptr_t)mmap_addr,
1219                         mmap_size,
1220                         alignment,
1221                         mmap_offset);
1222
1223         return 0;
1224 }
1225
1226 static int
1227 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1228                         int main_fd)
1229 {
1230         struct virtio_net *dev = *pdev;
1231         struct VhostUserMemory *memory = &msg->payload.memory;
1232         struct rte_vhost_mem_region *reg;
1233
1234         uint64_t mmap_offset;
1235         uint32_t i;
1236
1237         if (validate_msg_fds(msg, memory->nregions) != 0)
1238                 return RTE_VHOST_MSG_RESULT_ERR;
1239
1240         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1241                 VHOST_LOG_CONFIG(ERR,
1242                         "too many memory regions (%u)\n", memory->nregions);
1243                 goto close_msg_fds;
1244         }
1245
1246         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1247                 VHOST_LOG_CONFIG(INFO,
1248                         "(%d) memory regions not changed\n", dev->vid);
1249
1250                 close_msg_fds(msg);
1251
1252                 return RTE_VHOST_MSG_RESULT_OK;
1253         }
1254
1255         if (dev->mem) {
1256                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1257                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1258
1259                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1260                                 vdpa_dev->ops->dev_close(dev->vid);
1261                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1262                 }
1263                 free_mem_region(dev);
1264                 rte_free(dev->mem);
1265                 dev->mem = NULL;
1266         }
1267
1268         /* Flush IOTLB cache as previous HVAs are now invalid */
1269         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1270                 for (i = 0; i < dev->nr_vring; i++)
1271                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1272
1273         dev->nr_guest_pages = 0;
1274         if (dev->guest_pages == NULL) {
1275                 dev->max_guest_pages = 8;
1276                 dev->guest_pages = rte_zmalloc(NULL,
1277                                         dev->max_guest_pages *
1278                                         sizeof(struct guest_page),
1279                                         RTE_CACHE_LINE_SIZE);
1280                 if (dev->guest_pages == NULL) {
1281                         VHOST_LOG_CONFIG(ERR,
1282                                 "(%d) failed to allocate memory "
1283                                 "for dev->guest_pages\n",
1284                                 dev->vid);
1285                         goto close_msg_fds;
1286                 }
1287         }
1288
1289         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1290                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1291         if (dev->mem == NULL) {
1292                 VHOST_LOG_CONFIG(ERR,
1293                         "(%d) failed to allocate memory for dev->mem\n",
1294                         dev->vid);
1295                 goto free_guest_pages;
1296         }
1297
1298         for (i = 0; i < memory->nregions; i++) {
1299                 reg = &dev->mem->regions[i];
1300
1301                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1302                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1303                 reg->size            = memory->regions[i].memory_size;
1304                 reg->fd              = msg->fds[i];
1305
1306                 /*
1307                  * Assign invalid file descriptor value to avoid double
1308                  * closing on error path.
1309                  */
1310                 msg->fds[i] = -1;
1311
1312                 mmap_offset = memory->regions[i].mmap_offset;
1313
1314                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1315                         VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1316                         goto free_mem_table;
1317                 }
1318
1319                 dev->mem->nregions++;
1320         }
1321
1322         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1323                 goto free_mem_table;
1324
1325         for (i = 0; i < dev->nr_vring; i++) {
1326                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1327
1328                 if (!vq)
1329                         continue;
1330
1331                 if (vq->desc || vq->avail || vq->used) {
1332                         /*
1333                          * If the memory table got updated, the ring addresses
1334                          * need to be translated again as virtual addresses have
1335                          * changed.
1336                          */
1337                         vring_invalidate(dev, vq);
1338
1339                         dev = translate_ring_addresses(dev, i);
1340                         if (!dev) {
1341                                 dev = *pdev;
1342                                 goto free_mem_table;
1343                         }
1344
1345                         *pdev = dev;
1346                 }
1347         }
1348
1349         dump_guest_pages(dev);
1350
1351         return RTE_VHOST_MSG_RESULT_OK;
1352
1353 free_mem_table:
1354         free_mem_region(dev);
1355         rte_free(dev->mem);
1356         dev->mem = NULL;
1357 free_guest_pages:
1358         rte_free(dev->guest_pages);
1359         dev->guest_pages = NULL;
1360 close_msg_fds:
1361         close_msg_fds(msg);
1362         return RTE_VHOST_MSG_RESULT_ERR;
1363 }
1364
1365 static bool
1366 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1367 {
1368         bool rings_ok;
1369
1370         if (!vq)
1371                 return false;
1372
1373         if (vq_is_packed(dev))
1374                 rings_ok = vq->desc_packed && vq->driver_event &&
1375                         vq->device_event;
1376         else
1377                 rings_ok = vq->desc && vq->avail && vq->used;
1378
1379         return rings_ok &&
1380                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1381                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1382                vq->enabled;
1383 }
1384
1385 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1386
1387 static int
1388 virtio_is_ready(struct virtio_net *dev)
1389 {
1390         struct vhost_virtqueue *vq;
1391         uint32_t i, nr_vring = dev->nr_vring;
1392
1393         if (dev->flags & VIRTIO_DEV_READY)
1394                 return 1;
1395
1396         if (!dev->nr_vring)
1397                 return 0;
1398
1399         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1400                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1401
1402                 if (dev->nr_vring < nr_vring)
1403                         return 0;
1404         }
1405
1406         for (i = 0; i < nr_vring; i++) {
1407                 vq = dev->virtqueue[i];
1408
1409                 if (!vq_is_ready(dev, vq))
1410                         return 0;
1411         }
1412
1413         /* If supported, ensure the frontend is really done with config */
1414         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1415                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1416                         return 0;
1417
1418         dev->flags |= VIRTIO_DEV_READY;
1419
1420         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1421                 VHOST_LOG_CONFIG(INFO,
1422                         "virtio is now ready for processing.\n");
1423         return 1;
1424 }
1425
1426 static void *
1427 inflight_mem_alloc(const char *name, size_t size, int *fd)
1428 {
1429         void *ptr;
1430         int mfd = -1;
1431         char fname[20] = "/tmp/memfd-XXXXXX";
1432
1433         *fd = -1;
1434 #ifdef MEMFD_SUPPORTED
1435         mfd = memfd_create(name, MFD_CLOEXEC);
1436 #else
1437         RTE_SET_USED(name);
1438 #endif
1439         if (mfd == -1) {
1440                 mfd = mkstemp(fname);
1441                 if (mfd == -1) {
1442                         VHOST_LOG_CONFIG(ERR,
1443                                 "failed to get inflight buffer fd\n");
1444                         return NULL;
1445                 }
1446
1447                 unlink(fname);
1448         }
1449
1450         if (ftruncate(mfd, size) == -1) {
1451                 VHOST_LOG_CONFIG(ERR,
1452                         "failed to alloc inflight buffer\n");
1453                 close(mfd);
1454                 return NULL;
1455         }
1456
1457         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1458         if (ptr == MAP_FAILED) {
1459                 VHOST_LOG_CONFIG(ERR,
1460                         "failed to mmap inflight buffer\n");
1461                 close(mfd);
1462                 return NULL;
1463         }
1464
1465         *fd = mfd;
1466         return ptr;
1467 }
1468
1469 static uint32_t
1470 get_pervq_shm_size_split(uint16_t queue_size)
1471 {
1472         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1473                                   queue_size + sizeof(uint64_t) +
1474                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1475 }
1476
1477 static uint32_t
1478 get_pervq_shm_size_packed(uint16_t queue_size)
1479 {
1480         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1481                                   * queue_size + sizeof(uint64_t) +
1482                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1483                                   INFLIGHT_ALIGNMENT);
1484 }
1485
1486 static int
1487 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1488                            VhostUserMsg *msg,
1489                            int main_fd __rte_unused)
1490 {
1491         struct rte_vhost_inflight_info_packed *inflight_packed;
1492         uint64_t pervq_inflight_size, mmap_size;
1493         uint16_t num_queues, queue_size;
1494         struct virtio_net *dev = *pdev;
1495         int fd, i, j;
1496         void *addr;
1497
1498         if (msg->size != sizeof(msg->payload.inflight)) {
1499                 VHOST_LOG_CONFIG(ERR,
1500                         "invalid get_inflight_fd message size is %d\n",
1501                         msg->size);
1502                 return RTE_VHOST_MSG_RESULT_ERR;
1503         }
1504
1505         if (dev->inflight_info == NULL) {
1506                 dev->inflight_info = calloc(1,
1507                                             sizeof(struct inflight_mem_info));
1508                 if (!dev->inflight_info) {
1509                         VHOST_LOG_CONFIG(ERR,
1510                                 "failed to alloc dev inflight area\n");
1511                         return RTE_VHOST_MSG_RESULT_ERR;
1512                 }
1513                 dev->inflight_info->fd = -1;
1514         }
1515
1516         num_queues = msg->payload.inflight.num_queues;
1517         queue_size = msg->payload.inflight.queue_size;
1518
1519         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1520                 msg->payload.inflight.num_queues);
1521         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1522                 msg->payload.inflight.queue_size);
1523
1524         if (vq_is_packed(dev))
1525                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1526         else
1527                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1528
1529         mmap_size = num_queues * pervq_inflight_size;
1530         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1531         if (!addr) {
1532                 VHOST_LOG_CONFIG(ERR,
1533                         "failed to alloc vhost inflight area\n");
1534                         msg->payload.inflight.mmap_size = 0;
1535                 return RTE_VHOST_MSG_RESULT_ERR;
1536         }
1537         memset(addr, 0, mmap_size);
1538
1539         if (dev->inflight_info->addr) {
1540                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1541                 dev->inflight_info->addr = NULL;
1542         }
1543
1544         if (dev->inflight_info->fd >= 0) {
1545                 close(dev->inflight_info->fd);
1546                 dev->inflight_info->fd = -1;
1547         }
1548
1549         dev->inflight_info->addr = addr;
1550         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1551         dev->inflight_info->fd = msg->fds[0] = fd;
1552         msg->payload.inflight.mmap_offset = 0;
1553         msg->fd_num = 1;
1554
1555         if (vq_is_packed(dev)) {
1556                 for (i = 0; i < num_queues; i++) {
1557                         inflight_packed =
1558                                 (struct rte_vhost_inflight_info_packed *)addr;
1559                         inflight_packed->used_wrap_counter = 1;
1560                         inflight_packed->old_used_wrap_counter = 1;
1561                         for (j = 0; j < queue_size; j++)
1562                                 inflight_packed->desc[j].next = j + 1;
1563                         addr = (void *)((char *)addr + pervq_inflight_size);
1564                 }
1565         }
1566
1567         VHOST_LOG_CONFIG(INFO,
1568                 "send inflight mmap_size: %"PRIu64"\n",
1569                 msg->payload.inflight.mmap_size);
1570         VHOST_LOG_CONFIG(INFO,
1571                 "send inflight mmap_offset: %"PRIu64"\n",
1572                 msg->payload.inflight.mmap_offset);
1573         VHOST_LOG_CONFIG(INFO,
1574                 "send inflight fd: %d\n", msg->fds[0]);
1575
1576         return RTE_VHOST_MSG_RESULT_REPLY;
1577 }
1578
1579 static int
1580 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1581                            int main_fd __rte_unused)
1582 {
1583         uint64_t mmap_size, mmap_offset;
1584         uint16_t num_queues, queue_size;
1585         struct virtio_net *dev = *pdev;
1586         uint32_t pervq_inflight_size;
1587         struct vhost_virtqueue *vq;
1588         void *addr;
1589         int fd, i;
1590
1591         fd = msg->fds[0];
1592         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1593                 VHOST_LOG_CONFIG(ERR,
1594                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1595                         msg->size, fd);
1596                 return RTE_VHOST_MSG_RESULT_ERR;
1597         }
1598
1599         mmap_size = msg->payload.inflight.mmap_size;
1600         mmap_offset = msg->payload.inflight.mmap_offset;
1601         num_queues = msg->payload.inflight.num_queues;
1602         queue_size = msg->payload.inflight.queue_size;
1603
1604         if (vq_is_packed(dev))
1605                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1606         else
1607                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1608
1609         VHOST_LOG_CONFIG(INFO,
1610                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1611         VHOST_LOG_CONFIG(INFO,
1612                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1613         VHOST_LOG_CONFIG(INFO,
1614                 "set_inflight_fd num_queues: %u\n", num_queues);
1615         VHOST_LOG_CONFIG(INFO,
1616                 "set_inflight_fd queue_size: %u\n", queue_size);
1617         VHOST_LOG_CONFIG(INFO,
1618                 "set_inflight_fd fd: %d\n", fd);
1619         VHOST_LOG_CONFIG(INFO,
1620                 "set_inflight_fd pervq_inflight_size: %d\n",
1621                 pervq_inflight_size);
1622
1623         if (!dev->inflight_info) {
1624                 dev->inflight_info = calloc(1,
1625                                             sizeof(struct inflight_mem_info));
1626                 if (dev->inflight_info == NULL) {
1627                         VHOST_LOG_CONFIG(ERR,
1628                                 "failed to alloc dev inflight area\n");
1629                         return RTE_VHOST_MSG_RESULT_ERR;
1630                 }
1631                 dev->inflight_info->fd = -1;
1632         }
1633
1634         if (dev->inflight_info->addr) {
1635                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1636                 dev->inflight_info->addr = NULL;
1637         }
1638
1639         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1640                     fd, mmap_offset);
1641         if (addr == MAP_FAILED) {
1642                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1643                 return RTE_VHOST_MSG_RESULT_ERR;
1644         }
1645
1646         if (dev->inflight_info->fd >= 0) {
1647                 close(dev->inflight_info->fd);
1648                 dev->inflight_info->fd = -1;
1649         }
1650
1651         dev->inflight_info->fd = fd;
1652         dev->inflight_info->addr = addr;
1653         dev->inflight_info->size = mmap_size;
1654
1655         for (i = 0; i < num_queues; i++) {
1656                 vq = dev->virtqueue[i];
1657                 if (!vq)
1658                         continue;
1659
1660                 if (vq_is_packed(dev)) {
1661                         vq->inflight_packed = addr;
1662                         vq->inflight_packed->desc_num = queue_size;
1663                 } else {
1664                         vq->inflight_split = addr;
1665                         vq->inflight_split->desc_num = queue_size;
1666                 }
1667                 addr = (void *)((char *)addr + pervq_inflight_size);
1668         }
1669
1670         return RTE_VHOST_MSG_RESULT_OK;
1671 }
1672
1673 static int
1674 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1675                         int main_fd __rte_unused)
1676 {
1677         struct virtio_net *dev = *pdev;
1678         struct vhost_vring_file file;
1679         struct vhost_virtqueue *vq;
1680         int expected_fds;
1681
1682         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1683         if (validate_msg_fds(msg, expected_fds) != 0)
1684                 return RTE_VHOST_MSG_RESULT_ERR;
1685
1686         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1687         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1688                 file.fd = VIRTIO_INVALID_EVENTFD;
1689         else
1690                 file.fd = msg->fds[0];
1691         VHOST_LOG_CONFIG(INFO,
1692                 "vring call idx:%d file:%d\n", file.index, file.fd);
1693
1694         vq = dev->virtqueue[file.index];
1695
1696         if (vq->ready) {
1697                 vq->ready = false;
1698                 vhost_user_notify_queue_state(dev, file.index, 0);
1699         }
1700
1701         if (vq->callfd >= 0)
1702                 close(vq->callfd);
1703
1704         vq->callfd = file.fd;
1705
1706         return RTE_VHOST_MSG_RESULT_OK;
1707 }
1708
1709 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1710                         struct VhostUserMsg *msg,
1711                         int main_fd __rte_unused)
1712 {
1713         int expected_fds;
1714
1715         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1716         if (validate_msg_fds(msg, expected_fds) != 0)
1717                 return RTE_VHOST_MSG_RESULT_ERR;
1718
1719         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1720                 close(msg->fds[0]);
1721         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1722
1723         return RTE_VHOST_MSG_RESULT_OK;
1724 }
1725
1726 static int
1727 resubmit_desc_compare(const void *a, const void *b)
1728 {
1729         const struct rte_vhost_resubmit_desc *desc0 = a;
1730         const struct rte_vhost_resubmit_desc *desc1 = b;
1731
1732         if (desc1->counter > desc0->counter)
1733                 return 1;
1734
1735         return -1;
1736 }
1737
1738 static int
1739 vhost_check_queue_inflights_split(struct virtio_net *dev,
1740                                   struct vhost_virtqueue *vq)
1741 {
1742         uint16_t i;
1743         uint16_t resubmit_num = 0, last_io, num;
1744         struct vring_used *used = vq->used;
1745         struct rte_vhost_resubmit_info *resubmit;
1746         struct rte_vhost_inflight_info_split *inflight_split;
1747
1748         if (!(dev->protocol_features &
1749             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1750                 return RTE_VHOST_MSG_RESULT_OK;
1751
1752         /* The frontend may still not support the inflight feature
1753          * although we negotiate the protocol feature.
1754          */
1755         if ((!vq->inflight_split))
1756                 return RTE_VHOST_MSG_RESULT_OK;
1757
1758         if (!vq->inflight_split->version) {
1759                 vq->inflight_split->version = INFLIGHT_VERSION;
1760                 return RTE_VHOST_MSG_RESULT_OK;
1761         }
1762
1763         if (vq->resubmit_inflight)
1764                 return RTE_VHOST_MSG_RESULT_OK;
1765
1766         inflight_split = vq->inflight_split;
1767         vq->global_counter = 0;
1768         last_io = inflight_split->last_inflight_io;
1769
1770         if (inflight_split->used_idx != used->idx) {
1771                 inflight_split->desc[last_io].inflight = 0;
1772                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1773                 inflight_split->used_idx = used->idx;
1774         }
1775
1776         for (i = 0; i < inflight_split->desc_num; i++) {
1777                 if (inflight_split->desc[i].inflight == 1)
1778                         resubmit_num++;
1779         }
1780
1781         vq->last_avail_idx += resubmit_num;
1782
1783         if (resubmit_num) {
1784                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1785                 if (!resubmit) {
1786                         VHOST_LOG_CONFIG(ERR,
1787                                 "failed to allocate memory for resubmit info.\n");
1788                         return RTE_VHOST_MSG_RESULT_ERR;
1789                 }
1790
1791                 resubmit->resubmit_list = calloc(resubmit_num,
1792                         sizeof(struct rte_vhost_resubmit_desc));
1793                 if (!resubmit->resubmit_list) {
1794                         VHOST_LOG_CONFIG(ERR,
1795                                 "failed to allocate memory for inflight desc.\n");
1796                         free(resubmit);
1797                         return RTE_VHOST_MSG_RESULT_ERR;
1798                 }
1799
1800                 num = 0;
1801                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1802                         if (vq->inflight_split->desc[i].inflight == 1) {
1803                                 resubmit->resubmit_list[num].index = i;
1804                                 resubmit->resubmit_list[num].counter =
1805                                         inflight_split->desc[i].counter;
1806                                 num++;
1807                         }
1808                 }
1809                 resubmit->resubmit_num = num;
1810
1811                 if (resubmit->resubmit_num > 1)
1812                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1813                               sizeof(struct rte_vhost_resubmit_desc),
1814                               resubmit_desc_compare);
1815
1816                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1817                 vq->resubmit_inflight = resubmit;
1818         }
1819
1820         return RTE_VHOST_MSG_RESULT_OK;
1821 }
1822
1823 static int
1824 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1825                                    struct vhost_virtqueue *vq)
1826 {
1827         uint16_t i;
1828         uint16_t resubmit_num = 0, old_used_idx, num;
1829         struct rte_vhost_resubmit_info *resubmit;
1830         struct rte_vhost_inflight_info_packed *inflight_packed;
1831
1832         if (!(dev->protocol_features &
1833             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1834                 return RTE_VHOST_MSG_RESULT_OK;
1835
1836         /* The frontend may still not support the inflight feature
1837          * although we negotiate the protocol feature.
1838          */
1839         if ((!vq->inflight_packed))
1840                 return RTE_VHOST_MSG_RESULT_OK;
1841
1842         if (!vq->inflight_packed->version) {
1843                 vq->inflight_packed->version = INFLIGHT_VERSION;
1844                 return RTE_VHOST_MSG_RESULT_OK;
1845         }
1846
1847         if (vq->resubmit_inflight)
1848                 return RTE_VHOST_MSG_RESULT_OK;
1849
1850         inflight_packed = vq->inflight_packed;
1851         vq->global_counter = 0;
1852         old_used_idx = inflight_packed->old_used_idx;
1853
1854         if (inflight_packed->used_idx != old_used_idx) {
1855                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1856                         inflight_packed->old_used_idx =
1857                                 inflight_packed->used_idx;
1858                         inflight_packed->old_used_wrap_counter =
1859                                 inflight_packed->used_wrap_counter;
1860                         inflight_packed->old_free_head =
1861                                 inflight_packed->free_head;
1862                 } else {
1863                         inflight_packed->used_idx =
1864                                 inflight_packed->old_used_idx;
1865                         inflight_packed->used_wrap_counter =
1866                                 inflight_packed->old_used_wrap_counter;
1867                         inflight_packed->free_head =
1868                                 inflight_packed->old_free_head;
1869                 }
1870         }
1871
1872         for (i = 0; i < inflight_packed->desc_num; i++) {
1873                 if (inflight_packed->desc[i].inflight == 1)
1874                         resubmit_num++;
1875         }
1876
1877         if (resubmit_num) {
1878                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1879                 if (resubmit == NULL) {
1880                         VHOST_LOG_CONFIG(ERR,
1881                                 "failed to allocate memory for resubmit info.\n");
1882                         return RTE_VHOST_MSG_RESULT_ERR;
1883                 }
1884
1885                 resubmit->resubmit_list = calloc(resubmit_num,
1886                         sizeof(struct rte_vhost_resubmit_desc));
1887                 if (resubmit->resubmit_list == NULL) {
1888                         VHOST_LOG_CONFIG(ERR,
1889                                 "failed to allocate memory for resubmit desc.\n");
1890                         free(resubmit);
1891                         return RTE_VHOST_MSG_RESULT_ERR;
1892                 }
1893
1894                 num = 0;
1895                 for (i = 0; i < inflight_packed->desc_num; i++) {
1896                         if (vq->inflight_packed->desc[i].inflight == 1) {
1897                                 resubmit->resubmit_list[num].index = i;
1898                                 resubmit->resubmit_list[num].counter =
1899                                         inflight_packed->desc[i].counter;
1900                                 num++;
1901                         }
1902                 }
1903                 resubmit->resubmit_num = num;
1904
1905                 if (resubmit->resubmit_num > 1)
1906                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1907                               sizeof(struct rte_vhost_resubmit_desc),
1908                               resubmit_desc_compare);
1909
1910                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1911                 vq->resubmit_inflight = resubmit;
1912         }
1913
1914         return RTE_VHOST_MSG_RESULT_OK;
1915 }
1916
1917 static int
1918 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1919                         int main_fd __rte_unused)
1920 {
1921         struct virtio_net *dev = *pdev;
1922         struct vhost_vring_file file;
1923         struct vhost_virtqueue *vq;
1924         int expected_fds;
1925
1926         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1927         if (validate_msg_fds(msg, expected_fds) != 0)
1928                 return RTE_VHOST_MSG_RESULT_ERR;
1929
1930         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1931         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1932                 file.fd = VIRTIO_INVALID_EVENTFD;
1933         else
1934                 file.fd = msg->fds[0];
1935         VHOST_LOG_CONFIG(INFO,
1936                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1937
1938         /* Interpret ring addresses only when ring is started. */
1939         dev = translate_ring_addresses(dev, file.index);
1940         if (!dev) {
1941                 if (file.fd != VIRTIO_INVALID_EVENTFD)
1942                         close(file.fd);
1943
1944                 return RTE_VHOST_MSG_RESULT_ERR;
1945         }
1946
1947         *pdev = dev;
1948
1949         vq = dev->virtqueue[file.index];
1950
1951         /*
1952          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1953          * the ring starts already enabled. Otherwise, it is enabled via
1954          * the SET_VRING_ENABLE message.
1955          */
1956         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1957                 vq->enabled = true;
1958         }
1959
1960         if (vq->ready) {
1961                 vq->ready = false;
1962                 vhost_user_notify_queue_state(dev, file.index, 0);
1963         }
1964
1965         if (vq->kickfd >= 0)
1966                 close(vq->kickfd);
1967         vq->kickfd = file.fd;
1968
1969         if (vq_is_packed(dev)) {
1970                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1971                         VHOST_LOG_CONFIG(ERR,
1972                                 "failed to inflights for vq: %d\n", file.index);
1973                         return RTE_VHOST_MSG_RESULT_ERR;
1974                 }
1975         } else {
1976                 if (vhost_check_queue_inflights_split(dev, vq)) {
1977                         VHOST_LOG_CONFIG(ERR,
1978                                 "failed to inflights for vq: %d\n", file.index);
1979                         return RTE_VHOST_MSG_RESULT_ERR;
1980                 }
1981         }
1982
1983         return RTE_VHOST_MSG_RESULT_OK;
1984 }
1985
1986 /*
1987  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1988  */
1989 static int
1990 vhost_user_get_vring_base(struct virtio_net **pdev,
1991                         struct VhostUserMsg *msg,
1992                         int main_fd __rte_unused)
1993 {
1994         struct virtio_net *dev = *pdev;
1995         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1996         uint64_t val;
1997
1998         if (validate_msg_fds(msg, 0) != 0)
1999                 return RTE_VHOST_MSG_RESULT_ERR;
2000
2001         /* We have to stop the queue (virtio) if it is running. */
2002         vhost_destroy_device_notify(dev);
2003
2004         dev->flags &= ~VIRTIO_DEV_READY;
2005         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
2006
2007         /* Here we are safe to get the indexes */
2008         if (vq_is_packed(dev)) {
2009                 /*
2010                  * Bit[0:14]: avail index
2011                  * Bit[15]: avail wrap counter
2012                  */
2013                 val = vq->last_avail_idx & 0x7fff;
2014                 val |= vq->avail_wrap_counter << 15;
2015                 msg->payload.state.num = val;
2016         } else {
2017                 msg->payload.state.num = vq->last_avail_idx;
2018         }
2019
2020         VHOST_LOG_CONFIG(INFO,
2021                 "vring base idx:%d file:%d\n", msg->payload.state.index,
2022                 msg->payload.state.num);
2023         /*
2024          * Based on current qemu vhost-user implementation, this message is
2025          * sent and only sent in vhost_vring_stop.
2026          * TODO: cleanup the vring, it isn't usable since here.
2027          */
2028         if (vq->kickfd >= 0)
2029                 close(vq->kickfd);
2030
2031         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2032
2033         if (vq->callfd >= 0)
2034                 close(vq->callfd);
2035
2036         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2037
2038         vq->signalled_used_valid = false;
2039
2040         if (vq_is_packed(dev)) {
2041                 rte_free(vq->shadow_used_packed);
2042                 vq->shadow_used_packed = NULL;
2043         } else {
2044                 rte_free(vq->shadow_used_split);
2045                 vq->shadow_used_split = NULL;
2046         }
2047
2048         rte_free(vq->batch_copy_elems);
2049         vq->batch_copy_elems = NULL;
2050
2051         rte_free(vq->log_cache);
2052         vq->log_cache = NULL;
2053
2054         msg->size = sizeof(msg->payload.state);
2055         msg->fd_num = 0;
2056
2057         vring_invalidate(dev, vq);
2058
2059         return RTE_VHOST_MSG_RESULT_REPLY;
2060 }
2061
2062 /*
2063  * when virtio queues are ready to work, qemu will send us to
2064  * enable the virtio queue pair.
2065  */
2066 static int
2067 vhost_user_set_vring_enable(struct virtio_net **pdev,
2068                         struct VhostUserMsg *msg,
2069                         int main_fd __rte_unused)
2070 {
2071         struct virtio_net *dev = *pdev;
2072         bool enable = !!msg->payload.state.num;
2073         int index = (int)msg->payload.state.index;
2074
2075         if (validate_msg_fds(msg, 0) != 0)
2076                 return RTE_VHOST_MSG_RESULT_ERR;
2077
2078         VHOST_LOG_CONFIG(INFO,
2079                 "set queue enable: %d to qp idx: %d\n",
2080                 enable, index);
2081
2082         if (enable && dev->virtqueue[index]->async_registered) {
2083                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2084                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2085                         "async inflight packets must be completed first\n");
2086                         return RTE_VHOST_MSG_RESULT_ERR;
2087                 }
2088         }
2089
2090         dev->virtqueue[index]->enabled = enable;
2091
2092         return RTE_VHOST_MSG_RESULT_OK;
2093 }
2094
2095 static int
2096 vhost_user_get_protocol_features(struct virtio_net **pdev,
2097                         struct VhostUserMsg *msg,
2098                         int main_fd __rte_unused)
2099 {
2100         struct virtio_net *dev = *pdev;
2101         uint64_t features, protocol_features;
2102
2103         if (validate_msg_fds(msg, 0) != 0)
2104                 return RTE_VHOST_MSG_RESULT_ERR;
2105
2106         rte_vhost_driver_get_features(dev->ifname, &features);
2107         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2108
2109         msg->payload.u64 = protocol_features;
2110         msg->size = sizeof(msg->payload.u64);
2111         msg->fd_num = 0;
2112
2113         return RTE_VHOST_MSG_RESULT_REPLY;
2114 }
2115
2116 static int
2117 vhost_user_set_protocol_features(struct virtio_net **pdev,
2118                         struct VhostUserMsg *msg,
2119                         int main_fd __rte_unused)
2120 {
2121         struct virtio_net *dev = *pdev;
2122         uint64_t protocol_features = msg->payload.u64;
2123         uint64_t slave_protocol_features = 0;
2124
2125         if (validate_msg_fds(msg, 0) != 0)
2126                 return RTE_VHOST_MSG_RESULT_ERR;
2127
2128         rte_vhost_driver_get_protocol_features(dev->ifname,
2129                         &slave_protocol_features);
2130         if (protocol_features & ~slave_protocol_features) {
2131                 VHOST_LOG_CONFIG(ERR,
2132                         "(%d) received invalid protocol features.\n",
2133                         dev->vid);
2134                 return RTE_VHOST_MSG_RESULT_ERR;
2135         }
2136
2137         dev->protocol_features = protocol_features;
2138         VHOST_LOG_CONFIG(INFO,
2139                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2140                 dev->protocol_features);
2141
2142         return RTE_VHOST_MSG_RESULT_OK;
2143 }
2144
2145 static int
2146 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2147                         int main_fd __rte_unused)
2148 {
2149         struct virtio_net *dev = *pdev;
2150         int fd = msg->fds[0];
2151         uint64_t size, off;
2152         void *addr;
2153         uint32_t i;
2154
2155         if (validate_msg_fds(msg, 1) != 0)
2156                 return RTE_VHOST_MSG_RESULT_ERR;
2157
2158         if (fd < 0) {
2159                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2160                 return RTE_VHOST_MSG_RESULT_ERR;
2161         }
2162
2163         if (msg->size != sizeof(VhostUserLog)) {
2164                 VHOST_LOG_CONFIG(ERR,
2165                         "invalid log base msg size: %"PRId32" != %d\n",
2166                         msg->size, (int)sizeof(VhostUserLog));
2167                 goto close_msg_fds;
2168         }
2169
2170         size = msg->payload.log.mmap_size;
2171         off  = msg->payload.log.mmap_offset;
2172
2173         /* Check for mmap size and offset overflow. */
2174         if (off >= -size) {
2175                 VHOST_LOG_CONFIG(ERR,
2176                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2177                         off, size);
2178                 goto close_msg_fds;
2179         }
2180
2181         VHOST_LOG_CONFIG(INFO,
2182                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2183                 size, off);
2184
2185         /*
2186          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2187          * fail when offset is not page size aligned.
2188          */
2189         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2190         close(fd);
2191         if (addr == MAP_FAILED) {
2192                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2193                 return RTE_VHOST_MSG_RESULT_ERR;
2194         }
2195
2196         /*
2197          * Free previously mapped log memory on occasionally
2198          * multiple VHOST_USER_SET_LOG_BASE.
2199          */
2200         if (dev->log_addr) {
2201                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2202         }
2203         dev->log_addr = (uint64_t)(uintptr_t)addr;
2204         dev->log_base = dev->log_addr + off;
2205         dev->log_size = size;
2206
2207         for (i = 0; i < dev->nr_vring; i++) {
2208                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2209
2210                 rte_free(vq->log_cache);
2211                 vq->log_cache = NULL;
2212                 vq->log_cache_nb_elem = 0;
2213                 vq->log_cache = rte_zmalloc("vq log cache",
2214                                 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2215                                 0);
2216                 /*
2217                  * If log cache alloc fail, don't fail migration, but no
2218                  * caching will be done, which will impact performance
2219                  */
2220                 if (!vq->log_cache)
2221                         VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2222         }
2223
2224         /*
2225          * The spec is not clear about it (yet), but QEMU doesn't expect
2226          * any payload in the reply.
2227          */
2228         msg->size = 0;
2229         msg->fd_num = 0;
2230
2231         return RTE_VHOST_MSG_RESULT_REPLY;
2232
2233 close_msg_fds:
2234         close_msg_fds(msg);
2235         return RTE_VHOST_MSG_RESULT_ERR;
2236 }
2237
2238 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2239                         struct VhostUserMsg *msg,
2240                         int main_fd __rte_unused)
2241 {
2242         if (validate_msg_fds(msg, 1) != 0)
2243                 return RTE_VHOST_MSG_RESULT_ERR;
2244
2245         close(msg->fds[0]);
2246         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2247
2248         return RTE_VHOST_MSG_RESULT_OK;
2249 }
2250
2251 /*
2252  * An rarp packet is constructed and broadcasted to notify switches about
2253  * the new location of the migrated VM, so that packets from outside will
2254  * not be lost after migration.
2255  *
2256  * However, we don't actually "send" a rarp packet here, instead, we set
2257  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2258  */
2259 static int
2260 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2261                         int main_fd __rte_unused)
2262 {
2263         struct virtio_net *dev = *pdev;
2264         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2265         struct rte_vdpa_device *vdpa_dev;
2266
2267         if (validate_msg_fds(msg, 0) != 0)
2268                 return RTE_VHOST_MSG_RESULT_ERR;
2269
2270         VHOST_LOG_CONFIG(DEBUG,
2271                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2272                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2273         memcpy(dev->mac.addr_bytes, mac, 6);
2274
2275         /*
2276          * Set the flag to inject a RARP broadcast packet at
2277          * rte_vhost_dequeue_burst().
2278          *
2279          * __ATOMIC_RELEASE ordering is for making sure the mac is
2280          * copied before the flag is set.
2281          */
2282         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2283         vdpa_dev = dev->vdpa_dev;
2284         if (vdpa_dev && vdpa_dev->ops->migration_done)
2285                 vdpa_dev->ops->migration_done(dev->vid);
2286
2287         return RTE_VHOST_MSG_RESULT_OK;
2288 }
2289
2290 static int
2291 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2292                         int main_fd __rte_unused)
2293 {
2294         struct virtio_net *dev = *pdev;
2295
2296         if (validate_msg_fds(msg, 0) != 0)
2297                 return RTE_VHOST_MSG_RESULT_ERR;
2298
2299         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2300                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2301                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2302                                 msg->payload.u64);
2303
2304                 return RTE_VHOST_MSG_RESULT_ERR;
2305         }
2306
2307         dev->mtu = msg->payload.u64;
2308
2309         return RTE_VHOST_MSG_RESULT_OK;
2310 }
2311
2312 static int
2313 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2314                         int main_fd __rte_unused)
2315 {
2316         struct virtio_net *dev = *pdev;
2317         int fd = msg->fds[0];
2318
2319         if (validate_msg_fds(msg, 1) != 0)
2320                 return RTE_VHOST_MSG_RESULT_ERR;
2321
2322         if (fd < 0) {
2323                 VHOST_LOG_CONFIG(ERR,
2324                                 "Invalid file descriptor for slave channel (%d)\n",
2325                                 fd);
2326                 return RTE_VHOST_MSG_RESULT_ERR;
2327         }
2328
2329         if (dev->slave_req_fd >= 0)
2330                 close(dev->slave_req_fd);
2331
2332         dev->slave_req_fd = fd;
2333
2334         return RTE_VHOST_MSG_RESULT_OK;
2335 }
2336
2337 static int
2338 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2339 {
2340         struct vhost_vring_addr *ra;
2341         uint64_t start, end, len;
2342
2343         start = imsg->iova;
2344         end = start + imsg->size;
2345
2346         ra = &vq->ring_addrs;
2347         len = sizeof(struct vring_desc) * vq->size;
2348         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2349                 return 1;
2350
2351         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2352         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2353                 return 1;
2354
2355         len = sizeof(struct vring_used) +
2356                sizeof(struct vring_used_elem) * vq->size;
2357         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2358                 return 1;
2359
2360         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2361                 len = sizeof(uint64_t);
2362                 if (ra->log_guest_addr < end &&
2363                     (ra->log_guest_addr + len) > start)
2364                         return 1;
2365         }
2366
2367         return 0;
2368 }
2369
2370 static int
2371 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2372 {
2373         struct vhost_vring_addr *ra;
2374         uint64_t start, end, len;
2375
2376         start = imsg->iova;
2377         end = start + imsg->size;
2378
2379         ra = &vq->ring_addrs;
2380         len = sizeof(struct vring_packed_desc) * vq->size;
2381         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2382                 return 1;
2383
2384         len = sizeof(struct vring_packed_desc_event);
2385         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2386                 return 1;
2387
2388         len = sizeof(struct vring_packed_desc_event);
2389         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2390                 return 1;
2391
2392         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2393                 len = sizeof(uint64_t);
2394                 if (ra->log_guest_addr < end &&
2395                     (ra->log_guest_addr + len) > start)
2396                         return 1;
2397         }
2398
2399         return 0;
2400 }
2401
2402 static int is_vring_iotlb(struct virtio_net *dev,
2403                           struct vhost_virtqueue *vq,
2404                           struct vhost_iotlb_msg *imsg)
2405 {
2406         if (vq_is_packed(dev))
2407                 return is_vring_iotlb_packed(vq, imsg);
2408         else
2409                 return is_vring_iotlb_split(vq, imsg);
2410 }
2411
2412 static int
2413 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2414                         int main_fd __rte_unused)
2415 {
2416         struct virtio_net *dev = *pdev;
2417         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2418         uint16_t i;
2419         uint64_t vva, len;
2420
2421         if (validate_msg_fds(msg, 0) != 0)
2422                 return RTE_VHOST_MSG_RESULT_ERR;
2423
2424         switch (imsg->type) {
2425         case VHOST_IOTLB_UPDATE:
2426                 len = imsg->size;
2427                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2428                 if (!vva)
2429                         return RTE_VHOST_MSG_RESULT_ERR;
2430
2431                 for (i = 0; i < dev->nr_vring; i++) {
2432                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2433
2434                         if (!vq)
2435                                 continue;
2436
2437                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2438                                         len, imsg->perm);
2439
2440                         if (is_vring_iotlb(dev, vq, imsg))
2441                                 *pdev = dev = translate_ring_addresses(dev, i);
2442                 }
2443                 break;
2444         case VHOST_IOTLB_INVALIDATE:
2445                 for (i = 0; i < dev->nr_vring; i++) {
2446                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2447
2448                         if (!vq)
2449                                 continue;
2450
2451                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2452                                         imsg->size);
2453
2454                         if (is_vring_iotlb(dev, vq, imsg))
2455                                 vring_invalidate(dev, vq);
2456                 }
2457                 break;
2458         default:
2459                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2460                                 imsg->type);
2461                 return RTE_VHOST_MSG_RESULT_ERR;
2462         }
2463
2464         return RTE_VHOST_MSG_RESULT_OK;
2465 }
2466
2467 static int
2468 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2469                         struct VhostUserMsg *msg,
2470                         int main_fd __rte_unused)
2471 {
2472         struct virtio_net *dev = *pdev;
2473 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2474         struct uffdio_api api_struct;
2475
2476         if (validate_msg_fds(msg, 0) != 0)
2477                 return RTE_VHOST_MSG_RESULT_ERR;
2478
2479         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2480
2481         if (dev->postcopy_ufd == -1) {
2482                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2483                         strerror(errno));
2484                 return RTE_VHOST_MSG_RESULT_ERR;
2485         }
2486         api_struct.api = UFFD_API;
2487         api_struct.features = 0;
2488         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2489                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2490                         strerror(errno));
2491                 close(dev->postcopy_ufd);
2492                 dev->postcopy_ufd = -1;
2493                 return RTE_VHOST_MSG_RESULT_ERR;
2494         }
2495         msg->fds[0] = dev->postcopy_ufd;
2496         msg->fd_num = 1;
2497
2498         return RTE_VHOST_MSG_RESULT_REPLY;
2499 #else
2500         dev->postcopy_ufd = -1;
2501         msg->fd_num = 0;
2502
2503         return RTE_VHOST_MSG_RESULT_ERR;
2504 #endif
2505 }
2506
2507 static int
2508 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2509                         struct VhostUserMsg *msg __rte_unused,
2510                         int main_fd __rte_unused)
2511 {
2512         struct virtio_net *dev = *pdev;
2513
2514         if (validate_msg_fds(msg, 0) != 0)
2515                 return RTE_VHOST_MSG_RESULT_ERR;
2516
2517         if (dev->mem && dev->mem->nregions) {
2518                 VHOST_LOG_CONFIG(ERR,
2519                         "Regions already registered at postcopy-listen\n");
2520                 return RTE_VHOST_MSG_RESULT_ERR;
2521         }
2522         dev->postcopy_listening = 1;
2523
2524         return RTE_VHOST_MSG_RESULT_OK;
2525 }
2526
2527 static int
2528 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2529                         int main_fd __rte_unused)
2530 {
2531         struct virtio_net *dev = *pdev;
2532
2533         if (validate_msg_fds(msg, 0) != 0)
2534                 return RTE_VHOST_MSG_RESULT_ERR;
2535
2536         dev->postcopy_listening = 0;
2537         if (dev->postcopy_ufd >= 0) {
2538                 close(dev->postcopy_ufd);
2539                 dev->postcopy_ufd = -1;
2540         }
2541
2542         msg->payload.u64 = 0;
2543         msg->size = sizeof(msg->payload.u64);
2544         msg->fd_num = 0;
2545
2546         return RTE_VHOST_MSG_RESULT_REPLY;
2547 }
2548
2549 static int
2550 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2551                       int main_fd __rte_unused)
2552 {
2553         struct virtio_net *dev = *pdev;
2554
2555         if (validate_msg_fds(msg, 0) != 0)
2556                 return RTE_VHOST_MSG_RESULT_ERR;
2557
2558         msg->payload.u64 = dev->status;
2559         msg->size = sizeof(msg->payload.u64);
2560         msg->fd_num = 0;
2561
2562         return RTE_VHOST_MSG_RESULT_REPLY;
2563 }
2564
2565 static int
2566 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2567                         int main_fd __rte_unused)
2568 {
2569         struct virtio_net *dev = *pdev;
2570
2571         if (validate_msg_fds(msg, 0) != 0)
2572                 return RTE_VHOST_MSG_RESULT_ERR;
2573
2574         /* As per Virtio specification, the device status is 8bits long */
2575         if (msg->payload.u64 > UINT8_MAX) {
2576                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2577                                 msg->payload.u64);
2578                 return RTE_VHOST_MSG_RESULT_ERR;
2579         }
2580
2581         dev->status = msg->payload.u64;
2582
2583         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2584             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2585                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2586                 /*
2587                  * Clear the bit to let the driver know about the feature
2588                  * negotiation failure
2589                  */
2590                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2591         }
2592
2593         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2594                         "\t-RESET: %u\n"
2595                         "\t-ACKNOWLEDGE: %u\n"
2596                         "\t-DRIVER: %u\n"
2597                         "\t-FEATURES_OK: %u\n"
2598                         "\t-DRIVER_OK: %u\n"
2599                         "\t-DEVICE_NEED_RESET: %u\n"
2600                         "\t-FAILED: %u\n",
2601                         dev->status,
2602                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2603                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2604                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2605                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2606                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2607                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2608                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2609
2610         return RTE_VHOST_MSG_RESULT_OK;
2611 }
2612
2613 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2614                                         struct VhostUserMsg *msg,
2615                                         int main_fd);
2616 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2617         [VHOST_USER_NONE] = NULL,
2618         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2619         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2620         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2621         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2622         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2623         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2624         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2625         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2626         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2627         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2628         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2629         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2630         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2631         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2632         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2633         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2634         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2635         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2636         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2637         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2638         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2639         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2640         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2641         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2642         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2643         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2644         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2645         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2646         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2647 };
2648
2649 /* return bytes# of read on success or negative val on failure. */
2650 static int
2651 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2652 {
2653         int ret;
2654
2655         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2656                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2657         if (ret <= 0) {
2658                 return ret;
2659         } else if (ret != VHOST_USER_HDR_SIZE) {
2660                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2661                 close_msg_fds(msg);
2662                 return -1;
2663         }
2664
2665         if (msg->size) {
2666                 if (msg->size > sizeof(msg->payload)) {
2667                         VHOST_LOG_CONFIG(ERR,
2668                                 "invalid msg size: %d\n", msg->size);
2669                         return -1;
2670                 }
2671                 ret = read(sockfd, &msg->payload, msg->size);
2672                 if (ret <= 0)
2673                         return ret;
2674                 if (ret != (int)msg->size) {
2675                         VHOST_LOG_CONFIG(ERR,
2676                                 "read control message failed\n");
2677                         return -1;
2678                 }
2679         }
2680
2681         return ret;
2682 }
2683
2684 static int
2685 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2686 {
2687         if (!msg)
2688                 return 0;
2689
2690         return send_fd_message(sockfd, (char *)msg,
2691                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2692 }
2693
2694 static int
2695 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2696 {
2697         if (!msg)
2698                 return 0;
2699
2700         msg->flags &= ~VHOST_USER_VERSION_MASK;
2701         msg->flags &= ~VHOST_USER_NEED_REPLY;
2702         msg->flags |= VHOST_USER_VERSION;
2703         msg->flags |= VHOST_USER_REPLY_MASK;
2704
2705         return send_vhost_message(sockfd, msg);
2706 }
2707
2708 static int
2709 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2710 {
2711         int ret;
2712
2713         if (msg->flags & VHOST_USER_NEED_REPLY)
2714                 rte_spinlock_lock(&dev->slave_req_lock);
2715
2716         ret = send_vhost_message(dev->slave_req_fd, msg);
2717         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2718                 rte_spinlock_unlock(&dev->slave_req_lock);
2719
2720         return ret;
2721 }
2722
2723 /*
2724  * Allocate a queue pair if it hasn't been allocated yet
2725  */
2726 static int
2727 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2728                         struct VhostUserMsg *msg)
2729 {
2730         uint32_t vring_idx;
2731
2732         switch (msg->request.master) {
2733         case VHOST_USER_SET_VRING_KICK:
2734         case VHOST_USER_SET_VRING_CALL:
2735         case VHOST_USER_SET_VRING_ERR:
2736                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2737                 break;
2738         case VHOST_USER_SET_VRING_NUM:
2739         case VHOST_USER_SET_VRING_BASE:
2740         case VHOST_USER_SET_VRING_ENABLE:
2741                 vring_idx = msg->payload.state.index;
2742                 break;
2743         case VHOST_USER_SET_VRING_ADDR:
2744                 vring_idx = msg->payload.addr.index;
2745                 break;
2746         default:
2747                 return 0;
2748         }
2749
2750         if (vring_idx >= VHOST_MAX_VRING) {
2751                 VHOST_LOG_CONFIG(ERR,
2752                         "invalid vring index: %u\n", vring_idx);
2753                 return -1;
2754         }
2755
2756         if (dev->virtqueue[vring_idx])
2757                 return 0;
2758
2759         return alloc_vring_queue(dev, vring_idx);
2760 }
2761
2762 static void
2763 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2764 {
2765         unsigned int i = 0;
2766         unsigned int vq_num = 0;
2767
2768         while (vq_num < dev->nr_vring) {
2769                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2770
2771                 if (vq) {
2772                         rte_spinlock_lock(&vq->access_lock);
2773                         vq_num++;
2774                 }
2775                 i++;
2776         }
2777 }
2778
2779 static void
2780 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2781 {
2782         unsigned int i = 0;
2783         unsigned int vq_num = 0;
2784
2785         while (vq_num < dev->nr_vring) {
2786                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2787
2788                 if (vq) {
2789                         rte_spinlock_unlock(&vq->access_lock);
2790                         vq_num++;
2791                 }
2792                 i++;
2793         }
2794 }
2795
2796 int
2797 vhost_user_msg_handler(int vid, int fd)
2798 {
2799         struct virtio_net *dev;
2800         struct VhostUserMsg msg;
2801         struct rte_vdpa_device *vdpa_dev;
2802         int ret;
2803         int unlock_required = 0;
2804         bool handled;
2805         int request;
2806         uint32_t i;
2807
2808         dev = get_device(vid);
2809         if (dev == NULL)
2810                 return -1;
2811
2812         if (!dev->notify_ops) {
2813                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2814                 if (!dev->notify_ops) {
2815                         VHOST_LOG_CONFIG(ERR,
2816                                 "failed to get callback ops for driver %s\n",
2817                                 dev->ifname);
2818                         return -1;
2819                 }
2820         }
2821
2822         ret = read_vhost_message(fd, &msg);
2823         if (ret <= 0) {
2824                 if (ret < 0)
2825                         VHOST_LOG_CONFIG(ERR,
2826                                 "vhost read message failed\n");
2827                 else
2828                         VHOST_LOG_CONFIG(INFO,
2829                                 "vhost peer closed\n");
2830
2831                 return -1;
2832         }
2833
2834         ret = 0;
2835         request = msg.request.master;
2836         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2837                         vhost_message_str[request]) {
2838                 if (request != VHOST_USER_IOTLB_MSG)
2839                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2840                                 vhost_message_str[request]);
2841                 else
2842                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2843                                 vhost_message_str[request]);
2844         } else {
2845                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2846         }
2847
2848         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2849         if (ret < 0) {
2850                 VHOST_LOG_CONFIG(ERR,
2851                         "failed to alloc queue\n");
2852                 return -1;
2853         }
2854
2855         /*
2856          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2857          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2858          * and device is destroyed. destroy_device waits for queues to be
2859          * inactive, so it is safe. Otherwise taking the access_lock
2860          * would cause a dead lock.
2861          */
2862         switch (request) {
2863         case VHOST_USER_SET_FEATURES:
2864         case VHOST_USER_SET_PROTOCOL_FEATURES:
2865         case VHOST_USER_SET_OWNER:
2866         case VHOST_USER_SET_MEM_TABLE:
2867         case VHOST_USER_SET_LOG_BASE:
2868         case VHOST_USER_SET_LOG_FD:
2869         case VHOST_USER_SET_VRING_NUM:
2870         case VHOST_USER_SET_VRING_ADDR:
2871         case VHOST_USER_SET_VRING_BASE:
2872         case VHOST_USER_SET_VRING_KICK:
2873         case VHOST_USER_SET_VRING_CALL:
2874         case VHOST_USER_SET_VRING_ERR:
2875         case VHOST_USER_SET_VRING_ENABLE:
2876         case VHOST_USER_SEND_RARP:
2877         case VHOST_USER_NET_SET_MTU:
2878         case VHOST_USER_SET_SLAVE_REQ_FD:
2879                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2880                         vhost_user_lock_all_queue_pairs(dev);
2881                         unlock_required = 1;
2882                 }
2883                 break;
2884         default:
2885                 break;
2886
2887         }
2888
2889         handled = false;
2890         if (dev->extern_ops.pre_msg_handle) {
2891                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2892                                 (void *)&msg);
2893                 switch (ret) {
2894                 case RTE_VHOST_MSG_RESULT_REPLY:
2895                         send_vhost_reply(fd, &msg);
2896                         /* Fall-through */
2897                 case RTE_VHOST_MSG_RESULT_ERR:
2898                 case RTE_VHOST_MSG_RESULT_OK:
2899                         handled = true;
2900                         goto skip_to_post_handle;
2901                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2902                 default:
2903                         break;
2904                 }
2905         }
2906
2907         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2908                 if (!vhost_message_handlers[request])
2909                         goto skip_to_post_handle;
2910                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2911
2912                 switch (ret) {
2913                 case RTE_VHOST_MSG_RESULT_ERR:
2914                         VHOST_LOG_CONFIG(ERR,
2915                                 "Processing %s failed.\n",
2916                                 vhost_message_str[request]);
2917                         handled = true;
2918                         break;
2919                 case RTE_VHOST_MSG_RESULT_OK:
2920                         VHOST_LOG_CONFIG(DEBUG,
2921                                 "Processing %s succeeded.\n",
2922                                 vhost_message_str[request]);
2923                         handled = true;
2924                         break;
2925                 case RTE_VHOST_MSG_RESULT_REPLY:
2926                         VHOST_LOG_CONFIG(DEBUG,
2927                                 "Processing %s succeeded and needs reply.\n",
2928                                 vhost_message_str[request]);
2929                         send_vhost_reply(fd, &msg);
2930                         handled = true;
2931                         break;
2932                 default:
2933                         break;
2934                 }
2935         }
2936
2937 skip_to_post_handle:
2938         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2939                         dev->extern_ops.post_msg_handle) {
2940                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2941                                 (void *)&msg);
2942                 switch (ret) {
2943                 case RTE_VHOST_MSG_RESULT_REPLY:
2944                         send_vhost_reply(fd, &msg);
2945                         /* Fall-through */
2946                 case RTE_VHOST_MSG_RESULT_ERR:
2947                 case RTE_VHOST_MSG_RESULT_OK:
2948                         handled = true;
2949                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2950                 default:
2951                         break;
2952                 }
2953         }
2954
2955         if (unlock_required)
2956                 vhost_user_unlock_all_queue_pairs(dev);
2957
2958         /* If message was not handled at this stage, treat it as an error */
2959         if (!handled) {
2960                 VHOST_LOG_CONFIG(ERR,
2961                         "vhost message (req: %d) was not handled.\n", request);
2962                 close_msg_fds(&msg);
2963                 ret = RTE_VHOST_MSG_RESULT_ERR;
2964         }
2965
2966         /*
2967          * If the request required a reply that was already sent,
2968          * this optional reply-ack won't be sent as the
2969          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2970          */
2971         if (msg.flags & VHOST_USER_NEED_REPLY) {
2972                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2973                 msg.size = sizeof(msg.payload.u64);
2974                 msg.fd_num = 0;
2975                 send_vhost_reply(fd, &msg);
2976         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2977                 VHOST_LOG_CONFIG(ERR,
2978                         "vhost message handling failed.\n");
2979                 return -1;
2980         }
2981
2982         for (i = 0; i < dev->nr_vring; i++) {
2983                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2984                 bool cur_ready = vq_is_ready(dev, vq);
2985
2986                 if (cur_ready != (vq && vq->ready)) {
2987                         vq->ready = cur_ready;
2988                         vhost_user_notify_queue_state(dev, i, cur_ready);
2989                 }
2990         }
2991
2992
2993         if (!virtio_is_ready(dev))
2994                 goto out;
2995
2996         /*
2997          * Virtio is now ready. If not done already, it is time
2998          * to notify the application it can process the rings and
2999          * configure the vDPA device if present.
3000          */
3001
3002         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
3003                 if (dev->notify_ops->new_device(dev->vid) == 0)
3004                         dev->flags |= VIRTIO_DEV_RUNNING;
3005         }
3006
3007         vdpa_dev = dev->vdpa_dev;
3008         if (!vdpa_dev)
3009                 goto out;
3010
3011         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3012                 if (vdpa_dev->ops->dev_conf(dev->vid))
3013                         VHOST_LOG_CONFIG(ERR,
3014                                          "Failed to configure vDPA device\n");
3015                 else
3016                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3017         }
3018
3019 out:
3020         return 0;
3021 }
3022
3023 static int process_slave_message_reply(struct virtio_net *dev,
3024                                        const struct VhostUserMsg *msg)
3025 {
3026         struct VhostUserMsg msg_reply;
3027         int ret;
3028
3029         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3030                 return 0;
3031
3032         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3033         if (ret <= 0) {
3034                 if (ret < 0)
3035                         VHOST_LOG_CONFIG(ERR,
3036                                 "vhost read slave message reply failed\n");
3037                 else
3038                         VHOST_LOG_CONFIG(INFO,
3039                                 "vhost peer closed\n");
3040                 ret = -1;
3041                 goto out;
3042         }
3043
3044         ret = 0;
3045         if (msg_reply.request.slave != msg->request.slave) {
3046                 VHOST_LOG_CONFIG(ERR,
3047                         "Received unexpected msg type (%u), expected %u\n",
3048                         msg_reply.request.slave, msg->request.slave);
3049                 ret = -1;
3050                 goto out;
3051         }
3052
3053         ret = msg_reply.payload.u64 ? -1 : 0;
3054
3055 out:
3056         rte_spinlock_unlock(&dev->slave_req_lock);
3057         return ret;
3058 }
3059
3060 int
3061 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3062 {
3063         int ret;
3064         struct VhostUserMsg msg = {
3065                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3066                 .flags = VHOST_USER_VERSION,
3067                 .size = sizeof(msg.payload.iotlb),
3068                 .payload.iotlb = {
3069                         .iova = iova,
3070                         .perm = perm,
3071                         .type = VHOST_IOTLB_MISS,
3072                 },
3073         };
3074
3075         ret = send_vhost_message(dev->slave_req_fd, &msg);
3076         if (ret < 0) {
3077                 VHOST_LOG_CONFIG(ERR,
3078                                 "Failed to send IOTLB miss message (%d)\n",
3079                                 ret);
3080                 return ret;
3081         }
3082
3083         return 0;
3084 }
3085
3086 static int
3087 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3088 {
3089         int ret;
3090         struct VhostUserMsg msg = {
3091                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3092                 .flags = VHOST_USER_VERSION,
3093                 .size = 0,
3094         };
3095
3096         if (need_reply)
3097                 msg.flags |= VHOST_USER_NEED_REPLY;
3098
3099         ret = send_vhost_slave_message(dev, &msg);
3100         if (ret < 0) {
3101                 VHOST_LOG_CONFIG(ERR,
3102                                 "Failed to send config change (%d)\n",
3103                                 ret);
3104                 return ret;
3105         }
3106
3107         return process_slave_message_reply(dev, &msg);
3108 }
3109
3110 int
3111 rte_vhost_slave_config_change(int vid, bool need_reply)
3112 {
3113         struct virtio_net *dev;
3114
3115         dev = get_device(vid);
3116         if (!dev)
3117                 return -ENODEV;
3118
3119         return vhost_user_slave_config_change(dev, need_reply);
3120 }
3121
3122 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3123                                                     int index, int fd,
3124                                                     uint64_t offset,
3125                                                     uint64_t size)
3126 {
3127         int ret;
3128         struct VhostUserMsg msg = {
3129                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3130                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3131                 .size = sizeof(msg.payload.area),
3132                 .payload.area = {
3133                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3134                         .size = size,
3135                         .offset = offset,
3136                 },
3137         };
3138
3139         if (fd < 0)
3140                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3141         else {
3142                 msg.fds[0] = fd;
3143                 msg.fd_num = 1;
3144         }
3145
3146         ret = send_vhost_slave_message(dev, &msg);
3147         if (ret < 0) {
3148                 VHOST_LOG_CONFIG(ERR,
3149                         "Failed to set host notifier (%d)\n", ret);
3150                 return ret;
3151         }
3152
3153         return process_slave_message_reply(dev, &msg);
3154 }
3155
3156 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3157 {
3158         struct virtio_net *dev;
3159         struct rte_vdpa_device *vdpa_dev;
3160         int vfio_device_fd, ret = 0;
3161         uint64_t offset, size;
3162         unsigned int i, q_start, q_last;
3163
3164         dev = get_device(vid);
3165         if (!dev)
3166                 return -ENODEV;
3167
3168         vdpa_dev = dev->vdpa_dev;
3169         if (vdpa_dev == NULL)
3170                 return -ENODEV;
3171
3172         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3173             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3174             !(dev->protocol_features &
3175                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3176             !(dev->protocol_features &
3177                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3178             !(dev->protocol_features &
3179                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3180                 return -ENOTSUP;
3181
3182         if (qid == RTE_VHOST_QUEUE_ALL) {
3183                 q_start = 0;
3184                 q_last = dev->nr_vring - 1;
3185         } else {
3186                 if (qid >= dev->nr_vring)
3187                         return -EINVAL;
3188                 q_start = qid;
3189                 q_last = qid;
3190         }
3191
3192         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3193         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3194
3195         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3196         if (vfio_device_fd < 0)
3197                 return -ENOTSUP;
3198
3199         if (enable) {
3200                 for (i = q_start; i <= q_last; i++) {
3201                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3202                                         &size) < 0) {
3203                                 ret = -ENOTSUP;
3204                                 goto disable;
3205                         }
3206
3207                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3208                                         vfio_device_fd, offset, size) < 0) {
3209                                 ret = -EFAULT;
3210                                 goto disable;
3211                         }
3212                 }
3213         } else {
3214 disable:
3215                 for (i = q_start; i <= q_last; i++) {
3216                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3217                                         0, 0);
3218                 }
3219         }
3220
3221         return ret;
3222 }