lib/librte_vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 /* Security model
   6  * --------------
   7  * The vhost-user protocol connection is an external interface, so it must be
   8  * robust against invalid inputs.
   9  *
  10  * This is important because the vhost-user master is only one step removed
  11  * from the guest.  Malicious guests that have escaped will then launch further
  12  * attacks from the vhost-user master.
  13  *
  14  * Even in deployments where guests are trusted, a bug in the vhost-user master
  15  * can still cause invalid messages to be sent.  Such messages must not
  16  * compromise the stability of the DPDK application by causing crashes, memory
  17  * corruption, or other problematic behavior.
  18  *
  19  * Do not assume received VhostUserMsg fields contain sensible values!
  20  */
  21
  22 #include <stdint.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <fcntl.h>
  28 #include <sys/ioctl.h>
  29 #include <sys/mman.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/syscall.h>
  33 #include <assert.h>
  34 #ifdef RTE_LIBRTE_VHOST_NUMA
  35 #include <numaif.h>
  36 #endif
  37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
  38 #include <linux/userfaultfd.h>
  39 #endif
  40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
  41 #include <linux/memfd.h>
  42 #define MEMFD_SUPPORTED
  43 #endif
  44
  45 #include <rte_common.h>
  46 #include <rte_malloc.h>
  47 #include <rte_log.h>
  48
  49 #include "iotlb.h"
  50 #include "vhost.h"
  51 #include "vhost_user.h"
  52
  53 #define VIRTIO_MIN_MTU 68
  54 #define VIRTIO_MAX_MTU 65535
  55
  56 #define INFLIGHT_ALIGNMENT      64
  57 #define INFLIGHT_VERSION        0x1
  58
  59 static const char *vhost_message_str[VHOST_USER_MAX] = {
  60         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  61         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  62         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  63         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  64         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  65         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  66         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  67         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  68         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  69         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  70         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  71         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  72         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  73         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  74         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  75         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  76         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  77         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  78         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  79         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  80         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  81         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  82         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  83         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  84         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  85         [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
  86         [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
  87         [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
  88         [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
  89         [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
  90         [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
  91         [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
  92 };
  93
  94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
  95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
  96
  97 static void
  98 close_msg_fds(struct VhostUserMsg *msg)
  99 {
 100         int i;
 101
 102         for (i = 0; i < msg->fd_num; i++) {
 103                 int fd = msg->fds[i];
 104
 105                 if (fd == -1)
 106                         continue;
 107
 108                 msg->fds[i] = -1;
 109                 close(fd);
 110         }
 111 }
 112
 113 /*
 114  * Ensure the expected number of FDs is received,
 115  * close all FDs and return an error if this is not the case.
 116  */
 117 static int
 118 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
 119 {
 120         if (msg->fd_num == expected_fds)
 121                 return 0;
 122
 123         VHOST_LOG_CONFIG(ERR,
 124                 " Expect %d FDs for request %s, received %d\n",
 125                 expected_fds,
 126                 vhost_message_str[msg->request.master],
 127                 msg->fd_num);
 128
 129         close_msg_fds(msg);
 130
 131         return -1;
 132 }
 133
 134 static uint64_t
 135 get_blk_size(int fd)
 136 {
 137         struct stat stat;
 138         int ret;
 139
 140         ret = fstat(fd, &stat);
 141         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 142 }
 143
 144 static void
 145 free_mem_region(struct virtio_net *dev)
 146 {
 147         uint32_t i;
 148         struct rte_vhost_mem_region *reg;
 149
 150         if (!dev || !dev->mem)
 151                 return;
 152
 153         for (i = 0; i < dev->mem->nregions; i++) {
 154                 reg = &dev->mem->regions[i];
 155                 if (reg->host_user_addr) {
 156                         munmap(reg->mmap_addr, reg->mmap_size);
 157                         close(reg->fd);
 158                 }
 159         }
 160 }
 161
 162 void
 163 vhost_backend_cleanup(struct virtio_net *dev)
 164 {
 165         if (dev->mem) {
 166                 free_mem_region(dev);
 167                 rte_free(dev->mem);
 168                 dev->mem = NULL;
 169         }
 170
 171         rte_free(dev->guest_pages);
 172         dev->guest_pages = NULL;
 173
 174         if (dev->log_addr) {
 175                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 176                 dev->log_addr = 0;
 177         }
 178
 179         if (dev->inflight_info) {
 180                 if (dev->inflight_info->addr) {
 181                         munmap(dev->inflight_info->addr,
 182                                dev->inflight_info->size);
 183                         dev->inflight_info->addr = NULL;
 184                 }
 185
 186                 if (dev->inflight_info->fd >= 0) {
 187                         close(dev->inflight_info->fd);
 188                         dev->inflight_info->fd = -1;
 189                 }
 190
 191                 free(dev->inflight_info);
 192                 dev->inflight_info = NULL;
 193         }
 194
 195         if (dev->slave_req_fd >= 0) {
 196                 close(dev->slave_req_fd);
 197                 dev->slave_req_fd = -1;
 198         }
 199
 200         if (dev->postcopy_ufd >= 0) {
 201                 close(dev->postcopy_ufd);
 202                 dev->postcopy_ufd = -1;
 203         }
 204
 205         dev->postcopy_listening = 0;
 206 }
 207
 208 static void
 209 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
 210                               int enable)
 211 {
 212         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
 213         struct vhost_virtqueue *vq = dev->virtqueue[index];
 214
 215         /* Configure guest notifications on enable */
 216         if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
 217                 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
 218
 219         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
 220                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
 221
 222         if (dev->notify_ops->vring_state_changed)
 223                 dev->notify_ops->vring_state_changed(dev->vid,
 224                                 index, enable);
 225 }
 226
 227 /*
 228  * This function just returns success at the moment unless
 229  * the device hasn't been initialised.
 230  */
 231 static int
 232 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
 233                         struct VhostUserMsg *msg,
 234                         int main_fd __rte_unused)
 235 {
 236         if (validate_msg_fds(msg, 0) != 0)
 237                 return RTE_VHOST_MSG_RESULT_ERR;
 238
 239         return RTE_VHOST_MSG_RESULT_OK;
 240 }
 241
 242 static int
 243 vhost_user_reset_owner(struct virtio_net **pdev,
 244                         struct VhostUserMsg *msg,
 245                         int main_fd __rte_unused)
 246 {
 247         struct virtio_net *dev = *pdev;
 248
 249         if (validate_msg_fds(msg, 0) != 0)
 250                 return RTE_VHOST_MSG_RESULT_ERR;
 251
 252         vhost_destroy_device_notify(dev);
 253
 254         cleanup_device(dev, 0);
 255         reset_device(dev);
 256         return RTE_VHOST_MSG_RESULT_OK;
 257 }
 258
 259 /*
 260  * The features that we support are requested.
 261  */
 262 static int
 263 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 264                         int main_fd __rte_unused)
 265 {
 266         struct virtio_net *dev = *pdev;
 267         uint64_t features = 0;
 268
 269         if (validate_msg_fds(msg, 0) != 0)
 270                 return RTE_VHOST_MSG_RESULT_ERR;
 271
 272         rte_vhost_driver_get_features(dev->ifname, &features);
 273
 274         msg->payload.u64 = features;
 275         msg->size = sizeof(msg->payload.u64);
 276         msg->fd_num = 0;
 277
 278         return RTE_VHOST_MSG_RESULT_REPLY;
 279 }
 280
 281 /*
 282  * The queue number that we support are requested.
 283  */
 284 static int
 285 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
 286                         int main_fd __rte_unused)
 287 {
 288         struct virtio_net *dev = *pdev;
 289         uint32_t queue_num = 0;
 290
 291         if (validate_msg_fds(msg, 0) != 0)
 292                 return RTE_VHOST_MSG_RESULT_ERR;
 293
 294         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 295
 296         msg->payload.u64 = (uint64_t)queue_num;
 297         msg->size = sizeof(msg->payload.u64);
 298         msg->fd_num = 0;
 299
 300         return RTE_VHOST_MSG_RESULT_REPLY;
 301 }
 302
 303 /*
 304  * We receive the negotiated features supported by us and the virtio device.
 305  */
 306 static int
 307 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 308                         int main_fd __rte_unused)
 309 {
 310         struct virtio_net *dev = *pdev;
 311         uint64_t features = msg->payload.u64;
 312         uint64_t vhost_features = 0;
 313         struct rte_vdpa_device *vdpa_dev;
 314
 315         if (validate_msg_fds(msg, 0) != 0)
 316                 return RTE_VHOST_MSG_RESULT_ERR;
 317
 318         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 319         if (features & ~vhost_features) {
 320                 VHOST_LOG_CONFIG(ERR,
 321                         "(%d) received invalid negotiated features.\n",
 322                         dev->vid);
 323                 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
 324                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
 325
 326                 return RTE_VHOST_MSG_RESULT_ERR;
 327         }
 328
 329         if (dev->flags & VIRTIO_DEV_RUNNING) {
 330                 if (dev->features == features)
 331                         return RTE_VHOST_MSG_RESULT_OK;
 332
 333                 /*
 334                  * Error out if master tries to change features while device is
 335                  * in running state. The exception being VHOST_F_LOG_ALL, which
 336                  * is enabled when the live-migration starts.
 337                  */
 338                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 339                         VHOST_LOG_CONFIG(ERR,
 340                                 "(%d) features changed while device is running.\n",
 341                                 dev->vid);
 342                         return RTE_VHOST_MSG_RESULT_ERR;
 343                 }
 344
 345                 if (dev->notify_ops->features_changed)
 346                         dev->notify_ops->features_changed(dev->vid, features);
 347         }
 348
 349         dev->features = features;
 350         if (dev->features &
 351                 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
 352                  (1ULL << VIRTIO_F_VERSION_1) |
 353                  (1ULL << VIRTIO_F_RING_PACKED))) {
 354                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 355         } else {
 356                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 357         }
 358         VHOST_LOG_CONFIG(INFO,
 359                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 360         VHOST_LOG_CONFIG(DEBUG,
 361                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 362                 dev->vid,
 363                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 364                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 365
 366         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 367             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 368                 /*
 369                  * Remove all but first queue pair if MQ hasn't been
 370                  * negotiated. This is safe because the device is not
 371                  * running at this stage.
 372                  */
 373                 while (dev->nr_vring > 2) {
 374                         struct vhost_virtqueue *vq;
 375
 376                         vq = dev->virtqueue[--dev->nr_vring];
 377                         if (!vq)
 378                                 continue;
 379
 380                         dev->virtqueue[dev->nr_vring] = NULL;
 381                         cleanup_vq(vq, 1);
 382                         cleanup_vq_inflight(dev, vq);
 383                         free_vq(dev, vq);
 384                 }
 385         }
 386
 387         vdpa_dev = dev->vdpa_dev;
 388         if (vdpa_dev)
 389                 vdpa_dev->ops->set_features(dev->vid);
 390
 391         dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
 392         return RTE_VHOST_MSG_RESULT_OK;
 393 }
 394
 395 /*
 396  * The virtio device sends us the size of the descriptor ring.
 397  */
 398 static int
 399 vhost_user_set_vring_num(struct virtio_net **pdev,
 400                         struct VhostUserMsg *msg,
 401                         int main_fd __rte_unused)
 402 {
 403         struct virtio_net *dev = *pdev;
 404         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 405
 406         if (validate_msg_fds(msg, 0) != 0)
 407                 return RTE_VHOST_MSG_RESULT_ERR;
 408
 409         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 and vhost_virtqueue data structure to make them on the
 477  * 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                 rte_free(old_vq);
 549         }
 550
 551         /* check if we need to reallocate dev */
 552         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 553                             MPOL_F_NODE | MPOL_F_ADDR);
 554         if (ret) {
 555                 VHOST_LOG_CONFIG(ERR,
 556                         "Unable to get dev numa information.\n");
 557                 goto out;
 558         }
 559         if (oldnode != newnode) {
 560                 VHOST_LOG_CONFIG(INFO,
 561                         "reallocate dev from %d to %d node\n",
 562                         oldnode, newnode);
 563                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 564                 if (!dev) {
 565                         dev = old_dev;
 566                         goto out;
 567                 }
 568
 569                 memcpy(dev, old_dev, sizeof(*dev));
 570                 rte_free(old_dev);
 571         }
 572
 573 out:
 574         dev->virtqueue[index] = vq;
 575         vhost_devices[dev->vid] = dev;
 576
 577         if (old_vq != vq)
 578                 vhost_user_iotlb_init(dev, index);
 579
 580         return dev;
 581 }
 582 #else
 583 static struct virtio_net*
 584 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 585 {
 586         return dev;
 587 }
 588 #endif
 589
 590 /* Converts QEMU virtual address to Vhost virtual address. */
 591 static uint64_t
 592 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 593 {
 594         struct rte_vhost_mem_region *r;
 595         uint32_t i;
 596
 597         if (unlikely(!dev || !dev->mem))
 598                 goto out_error;
 599
 600         /* Find the region where the address lives. */
 601         for (i = 0; i < dev->mem->nregions; i++) {
 602                 r = &dev->mem->regions[i];
 603
 604                 if (qva >= r->guest_user_addr &&
 605                     qva <  r->guest_user_addr + r->size) {
 606
 607                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 608                                 *len = r->guest_user_addr + r->size - qva;
 609
 610                         return qva - r->guest_user_addr +
 611                                r->host_user_addr;
 612                 }
 613         }
 614 out_error:
 615         *len = 0;
 616
 617         return 0;
 618 }
 619
 620
 621 /*
 622  * Converts ring address to Vhost virtual address.
 623  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 624  * else it is a QEMU virtual address.
 625  */
 626 static uint64_t
 627 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 628                 uint64_t ra, uint64_t *size)
 629 {
 630         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 631                 uint64_t vva;
 632
 633                 vhost_user_iotlb_rd_lock(vq);
 634                 vva = vhost_iova_to_vva(dev, vq, ra,
 635                                         size, VHOST_ACCESS_RW);
 636                 vhost_user_iotlb_rd_unlock(vq);
 637
 638                 return vva;
 639         }
 640
 641         return qva_to_vva(dev, ra, size);
 642 }
 643
 644 static uint64_t
 645 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 646 {
 647         uint64_t log_gpa;
 648
 649         vhost_user_iotlb_rd_lock(vq);
 650         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 651         vhost_user_iotlb_rd_unlock(vq);
 652
 653         return log_gpa;
 654 }
 655
 656 static struct virtio_net *
 657 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 658 {
 659         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 660         struct vhost_vring_addr *addr = &vq->ring_addrs;
 661         uint64_t len, expected_len;
 662
 663         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 664                 vq->log_guest_addr =
 665                         log_addr_to_gpa(dev, vq);
 666                 if (vq->log_guest_addr == 0) {
 667                         VHOST_LOG_CONFIG(DEBUG,
 668                                 "(%d) failed to map log_guest_addr.\n",
 669                                 dev->vid);
 670                         return dev;
 671                 }
 672         }
 673
 674         if (vq_is_packed(dev)) {
 675                 len = sizeof(struct vring_packed_desc) * vq->size;
 676                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 677                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 678                 if (vq->desc_packed == NULL ||
 679                                 len != sizeof(struct vring_packed_desc) *
 680                                 vq->size) {
 681                         VHOST_LOG_CONFIG(DEBUG,
 682                                 "(%d) failed to map desc_packed ring.\n",
 683                                 dev->vid);
 684                         return dev;
 685                 }
 686
 687                 dev = numa_realloc(dev, vq_index);
 688                 vq = dev->virtqueue[vq_index];
 689                 addr = &vq->ring_addrs;
 690
 691                 len = sizeof(struct vring_packed_desc_event);
 692                 vq->driver_event = (struct vring_packed_desc_event *)
 693                                         (uintptr_t)ring_addr_to_vva(dev,
 694                                         vq, addr->avail_user_addr, &len);
 695                 if (vq->driver_event == NULL ||
 696                                 len != sizeof(struct vring_packed_desc_event)) {
 697                         VHOST_LOG_CONFIG(DEBUG,
 698                                 "(%d) failed to find driver area address.\n",
 699                                 dev->vid);
 700                         return dev;
 701                 }
 702
 703                 len = sizeof(struct vring_packed_desc_event);
 704                 vq->device_event = (struct vring_packed_desc_event *)
 705                                         (uintptr_t)ring_addr_to_vva(dev,
 706                                         vq, addr->used_user_addr, &len);
 707                 if (vq->device_event == NULL ||
 708                                 len != sizeof(struct vring_packed_desc_event)) {
 709                         VHOST_LOG_CONFIG(DEBUG,
 710                                 "(%d) failed to find device area address.\n",
 711                                 dev->vid);
 712                         return dev;
 713                 }
 714
 715                 vq->access_ok = true;
 716                 return dev;
 717         }
 718
 719         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 720         if (vq->desc && vq->avail && vq->used)
 721                 return dev;
 722
 723         len = sizeof(struct vring_desc) * vq->size;
 724         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 725                         vq, addr->desc_user_addr, &len);
 726         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 727                 VHOST_LOG_CONFIG(DEBUG,
 728                         "(%d) failed to map desc ring.\n",
 729                         dev->vid);
 730                 return dev;
 731         }
 732
 733         dev = numa_realloc(dev, vq_index);
 734         vq = dev->virtqueue[vq_index];
 735         addr = &vq->ring_addrs;
 736
 737         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 738         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 739                 len += sizeof(uint16_t);
 740         expected_len = len;
 741         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 742                         vq, addr->avail_user_addr, &len);
 743         if (vq->avail == 0 || len != expected_len) {
 744                 VHOST_LOG_CONFIG(DEBUG,
 745                         "(%d) failed to map avail ring.\n",
 746                         dev->vid);
 747                 return dev;
 748         }
 749
 750         len = sizeof(struct vring_used) +
 751                 sizeof(struct vring_used_elem) * vq->size;
 752         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 753                 len += sizeof(uint16_t);
 754         expected_len = len;
 755         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 756                         vq, addr->used_user_addr, &len);
 757         if (vq->used == 0 || len != expected_len) {
 758                 VHOST_LOG_CONFIG(DEBUG,
 759                         "(%d) failed to map used ring.\n",
 760                         dev->vid);
 761                 return dev;
 762         }
 763
 764         if (vq->last_used_idx != vq->used->idx) {
 765                 VHOST_LOG_CONFIG(WARNING,
 766                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 767                         "some packets maybe resent for Tx and dropped for Rx\n",
 768                         vq->last_used_idx, vq->used->idx);
 769                 vq->last_used_idx  = vq->used->idx;
 770                 vq->last_avail_idx = vq->used->idx;
 771         }
 772
 773         vq->access_ok = true;
 774
 775         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 776                         dev->vid, vq->desc);
 777         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 778                         dev->vid, vq->avail);
 779         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 780                         dev->vid, vq->used);
 781         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 782                         dev->vid, vq->log_guest_addr);
 783
 784         return dev;
 785 }
 786
 787 /*
 788  * The virtio device sends us the desc, used and avail ring addresses.
 789  * This function then converts these to our address space.
 790  */
 791 static int
 792 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 793                         int main_fd __rte_unused)
 794 {
 795         struct virtio_net *dev = *pdev;
 796         struct vhost_virtqueue *vq;
 797         struct vhost_vring_addr *addr = &msg->payload.addr;
 798         bool access_ok;
 799
 800         if (validate_msg_fds(msg, 0) != 0)
 801                 return RTE_VHOST_MSG_RESULT_ERR;
 802
 803         if (dev->mem == NULL)
 804                 return RTE_VHOST_MSG_RESULT_ERR;
 805
 806         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 807         vq = dev->virtqueue[msg->payload.addr.index];
 808
 809         access_ok = vq->access_ok;
 810
 811         /*
 812          * Rings addresses should not be interpreted as long as the ring is not
 813          * started and enabled
 814          */
 815         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 816
 817         vring_invalidate(dev, vq);
 818
 819         if ((vq->enabled && (dev->features &
 820                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 821                         access_ok) {
 822                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 823                 if (!dev)
 824                         return RTE_VHOST_MSG_RESULT_ERR;
 825
 826                 *pdev = dev;
 827         }
 828
 829         return RTE_VHOST_MSG_RESULT_OK;
 830 }
 831
 832 /*
 833  * The virtio device sends us the available ring last used index.
 834  */
 835 static int
 836 vhost_user_set_vring_base(struct virtio_net **pdev,
 837                         struct VhostUserMsg *msg,
 838                         int main_fd __rte_unused)
 839 {
 840         struct virtio_net *dev = *pdev;
 841         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 842         uint64_t val = msg->payload.state.num;
 843
 844         if (validate_msg_fds(msg, 0) != 0)
 845                 return RTE_VHOST_MSG_RESULT_ERR;
 846
 847         if (vq_is_packed(dev)) {
 848                 /*
 849                  * Bit[0:14]: avail index
 850                  * Bit[15]: avail wrap counter
 851                  */
 852                 vq->last_avail_idx = val & 0x7fff;
 853                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 854                 /*
 855                  * Set used index to same value as available one, as
 856                  * their values should be the same since ring processing
 857                  * was stopped at get time.
 858                  */
 859                 vq->last_used_idx = vq->last_avail_idx;
 860                 vq->used_wrap_counter = vq->avail_wrap_counter;
 861         } else {
 862                 vq->last_used_idx = msg->payload.state.num;
 863                 vq->last_avail_idx = msg->payload.state.num;
 864         }
 865
 866         return RTE_VHOST_MSG_RESULT_OK;
 867 }
 868
 869 static int
 870 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 871                    uint64_t host_phys_addr, uint64_t size)
 872 {
 873         struct guest_page *page, *last_page;
 874         struct guest_page *old_pages;
 875
 876         if (dev->nr_guest_pages == dev->max_guest_pages) {
 877                 dev->max_guest_pages *= 2;
 878                 old_pages = dev->guest_pages;
 879                 dev->guest_pages = rte_realloc(dev->guest_pages,
 880                                         dev->max_guest_pages * sizeof(*page),
 881                                         RTE_CACHE_LINE_SIZE);
 882                 if (dev->guest_pages == NULL) {
 883                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 884                         rte_free(old_pages);
 885                         return -1;
 886                 }
 887         }
 888
 889         if (dev->nr_guest_pages > 0) {
 890                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 891                 /* merge if the two pages are continuous */
 892                 if (host_phys_addr == last_page->host_phys_addr +
 893                                       last_page->size) {
 894                         last_page->size += size;
 895                         return 0;
 896                 }
 897         }
 898
 899         page = &dev->guest_pages[dev->nr_guest_pages++];
 900         page->guest_phys_addr = guest_phys_addr;
 901         page->host_phys_addr  = host_phys_addr;
 902         page->size = size;
 903
 904         return 0;
 905 }
 906
 907 static int
 908 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 909                 uint64_t page_size)
 910 {
 911         uint64_t reg_size = reg->size;
 912         uint64_t host_user_addr  = reg->host_user_addr;
 913         uint64_t guest_phys_addr = reg->guest_phys_addr;
 914         uint64_t host_phys_addr;
 915         uint64_t size;
 916
 917         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 918         size = page_size - (guest_phys_addr & (page_size - 1));
 919         size = RTE_MIN(size, reg_size);
 920
 921         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 922                 return -1;
 923
 924         host_user_addr  += size;
 925         guest_phys_addr += size;
 926         reg_size -= size;
 927
 928         while (reg_size > 0) {
 929                 size = RTE_MIN(reg_size, page_size);
 930                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 931                                                   host_user_addr);
 932                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 933                                 size) < 0)
 934                         return -1;
 935
 936                 host_user_addr  += size;
 937                 guest_phys_addr += size;
 938                 reg_size -= size;
 939         }
 940
 941         /* sort guest page array if over binary search threshold */
 942         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 943                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 944                         sizeof(struct guest_page), guest_page_addrcmp);
 945         }
 946
 947         return 0;
 948 }
 949
 950 #ifdef RTE_LIBRTE_VHOST_DEBUG
 951 /* TODO: enable it only in debug mode? */
 952 static void
 953 dump_guest_pages(struct virtio_net *dev)
 954 {
 955         uint32_t i;
 956         struct guest_page *page;
 957
 958         for (i = 0; i < dev->nr_guest_pages; i++) {
 959                 page = &dev->guest_pages[i];
 960
 961                 VHOST_LOG_CONFIG(INFO,
 962                         "guest physical page region %u\n"
 963                         "\t guest_phys_addr: %" PRIx64 "\n"
 964                         "\t host_phys_addr : %" PRIx64 "\n"
 965                         "\t size           : %" PRIx64 "\n",
 966                         i,
 967                         page->guest_phys_addr,
 968                         page->host_phys_addr,
 969                         page->size);
 970         }
 971 }
 972 #else
 973 #define dump_guest_pages(dev)
 974 #endif
 975
 976 static bool
 977 vhost_memory_changed(struct VhostUserMemory *new,
 978                      struct rte_vhost_memory *old)
 979 {
 980         uint32_t i;
 981
 982         if (new->nregions != old->nregions)
 983                 return true;
 984
 985         for (i = 0; i < new->nregions; ++i) {
 986                 VhostUserMemoryRegion *new_r = &new->regions[i];
 987                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 988
 989                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 990                         return true;
 991                 if (new_r->memory_size != old_r->size)
 992                         return true;
 993                 if (new_r->userspace_addr != old_r->guest_user_addr)
 994                         return true;
 995         }
 996
 997         return false;
 998 }
 999
1000 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1001 static int
1002 vhost_user_postcopy_region_register(struct virtio_net *dev,
1003                 struct rte_vhost_mem_region *reg)
1004 {
1005         struct uffdio_register reg_struct;
1006
1007         /*
1008          * Let's register all the mmap'ed area to ensure
1009          * alignment on page boundary.
1010          */
1011         reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1012         reg_struct.range.len = reg->mmap_size;
1013         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1014
1015         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1016                                 &reg_struct)) {
1017                 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1018                                 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1019                                 (uint64_t)reg_struct.range.start,
1020                                 (uint64_t)reg_struct.range.start +
1021                                 (uint64_t)reg_struct.range.len - 1,
1022                                 dev->postcopy_ufd,
1023                                 strerror(errno));
1024                 return -1;
1025         }
1026
1027         VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1028                         (uint64_t)reg_struct.range.start,
1029                         (uint64_t)reg_struct.range.start +
1030                         (uint64_t)reg_struct.range.len - 1);
1031
1032         return 0;
1033 }
1034 #else
1035 static int
1036 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1037                 struct rte_vhost_mem_region *reg __rte_unused)
1038 {
1039         return -1;
1040 }
1041 #endif
1042
1043 static int
1044 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1045                 struct VhostUserMsg *msg)
1046 {
1047         struct VhostUserMemory *memory;
1048         struct rte_vhost_mem_region *reg;
1049         VhostUserMsg ack_msg;
1050         uint32_t i;
1051
1052         if (!dev->postcopy_listening)
1053                 return 0;
1054
1055         /*
1056          * We haven't a better way right now than sharing
1057          * DPDK's virtual address with Qemu, so that Qemu can
1058          * retrieve the region offset when handling userfaults.
1059          */
1060         memory = &msg->payload.memory;
1061         for (i = 0; i < memory->nregions; i++) {
1062                 reg = &dev->mem->regions[i];
1063                 memory->regions[i].userspace_addr = reg->host_user_addr;
1064         }
1065
1066         /* Send the addresses back to qemu */
1067         msg->fd_num = 0;
1068         send_vhost_reply(main_fd, msg);
1069
1070         /* Wait for qemu to acknolwedge it's got the addresses
1071          * we've got to wait before we're allowed to generate faults.
1072          */
1073         if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1074                 VHOST_LOG_CONFIG(ERR,
1075                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1076                 return -1;
1077         }
1078
1079         if (validate_msg_fds(&ack_msg, 0) != 0)
1080                 return -1;
1081
1082         if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1083                 VHOST_LOG_CONFIG(ERR,
1084                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1085                                 ack_msg.request.master);
1086                 return -1;
1087         }
1088
1089         /* Now userfault register and we can use the memory */
1090         for (i = 0; i < memory->nregions; i++) {
1091                 reg = &dev->mem->regions[i];
1092                 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1093                         return -1;
1094         }
1095
1096         return 0;
1097 }
1098
1099 static int
1100 vhost_user_mmap_region(struct virtio_net *dev,
1101                 struct rte_vhost_mem_region *region,
1102                 uint64_t mmap_offset)
1103 {
1104         void *mmap_addr;
1105         uint64_t mmap_size;
1106         uint64_t alignment;
1107         int populate;
1108
1109         /* Check for memory_size + mmap_offset overflow */
1110         if (mmap_offset >= -region->size) {
1111                 VHOST_LOG_CONFIG(ERR,
1112                                 "mmap_offset (%#"PRIx64") and memory_size "
1113                                 "(%#"PRIx64") overflow\n",
1114                                 mmap_offset, region->size);
1115                 return -1;
1116         }
1117
1118         mmap_size = region->size + mmap_offset;
1119
1120         /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1121          * argument aligned with hugepagesz at older longterm version Linux,
1122          * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1123          *
1124          * To avoid failure, make sure in caller to keep length aligned.
1125          */
1126         alignment = get_blk_size(region->fd);
1127         if (alignment == (uint64_t)-1) {
1128                 VHOST_LOG_CONFIG(ERR,
1129                                 "couldn't get hugepage size through fstat\n");
1130                 return -1;
1131         }
1132         mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1133         if (mmap_size == 0) {
1134                 /*
1135                  * It could happen if initial mmap_size + alignment overflows
1136                  * the sizeof uint64, which could happen if either mmap_size or
1137                  * alignment value is wrong.
1138                  *
1139                  * mmap() kernel implementation would return an error, but
1140                  * better catch it before and provide useful info in the logs.
1141                  */
1142                 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1143                                 "or alignment (0x%" PRIx64 ") is invalid\n",
1144                                 region->size + mmap_offset, alignment);
1145                 return -1;
1146         }
1147
1148         populate = dev->async_copy ? MAP_POPULATE : 0;
1149         mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1150                         MAP_SHARED | populate, region->fd, 0);
1151
1152         if (mmap_addr == MAP_FAILED) {
1153                 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1154                 return -1;
1155         }
1156
1157         region->mmap_addr = mmap_addr;
1158         region->mmap_size = mmap_size;
1159         region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1160
1161         if (dev->async_copy)
1162                 if (add_guest_pages(dev, region, alignment) < 0) {
1163                         VHOST_LOG_CONFIG(ERR,
1164                                         "adding guest pages to region failed.\n");
1165                         return -1;
1166                 }
1167
1168         VHOST_LOG_CONFIG(INFO,
1169                         "guest memory region size: 0x%" PRIx64 "\n"
1170                         "\t guest physical addr: 0x%" PRIx64 "\n"
1171                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1172                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1173                         "\t mmap addr : 0x%" PRIx64 "\n"
1174                         "\t mmap size : 0x%" PRIx64 "\n"
1175                         "\t mmap align: 0x%" PRIx64 "\n"
1176                         "\t mmap off  : 0x%" PRIx64 "\n",
1177                         region->size,
1178                         region->guest_phys_addr,
1179                         region->guest_user_addr,
1180                         region->host_user_addr,
1181                         (uint64_t)(uintptr_t)mmap_addr,
1182                         mmap_size,
1183                         alignment,
1184                         mmap_offset);
1185
1186         return 0;
1187 }
1188
1189 static int
1190 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1191                         int main_fd)
1192 {
1193         struct virtio_net *dev = *pdev;
1194         struct VhostUserMemory *memory = &msg->payload.memory;
1195         struct rte_vhost_mem_region *reg;
1196
1197         uint64_t mmap_offset;
1198         uint32_t i;
1199
1200         if (validate_msg_fds(msg, memory->nregions) != 0)
1201                 return RTE_VHOST_MSG_RESULT_ERR;
1202
1203         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1204                 VHOST_LOG_CONFIG(ERR,
1205                         "too many memory regions (%u)\n", memory->nregions);
1206                 goto close_msg_fds;
1207         }
1208
1209         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1210                 VHOST_LOG_CONFIG(INFO,
1211                         "(%d) memory regions not changed\n", dev->vid);
1212
1213                 close_msg_fds(msg);
1214
1215                 return RTE_VHOST_MSG_RESULT_OK;
1216         }
1217
1218         if (dev->mem) {
1219                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1220                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1221
1222                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1223                                 vdpa_dev->ops->dev_close(dev->vid);
1224                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1225                 }
1226                 free_mem_region(dev);
1227                 rte_free(dev->mem);
1228                 dev->mem = NULL;
1229         }
1230
1231         /* Flush IOTLB cache as previous HVAs are now invalid */
1232         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1233                 for (i = 0; i < dev->nr_vring; i++)
1234                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1235
1236         dev->nr_guest_pages = 0;
1237         if (dev->guest_pages == NULL) {
1238                 dev->max_guest_pages = 8;
1239                 dev->guest_pages = rte_zmalloc(NULL,
1240                                         dev->max_guest_pages *
1241                                         sizeof(struct guest_page),
1242                                         RTE_CACHE_LINE_SIZE);
1243                 if (dev->guest_pages == NULL) {
1244                         VHOST_LOG_CONFIG(ERR,
1245                                 "(%d) failed to allocate memory "
1246                                 "for dev->guest_pages\n",
1247                                 dev->vid);
1248                         goto close_msg_fds;
1249                 }
1250         }
1251
1252         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1253                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1254         if (dev->mem == NULL) {
1255                 VHOST_LOG_CONFIG(ERR,
1256                         "(%d) failed to allocate memory for dev->mem\n",
1257                         dev->vid);
1258                 goto free_guest_pages;
1259         }
1260
1261         for (i = 0; i < memory->nregions; i++) {
1262                 reg = &dev->mem->regions[i];
1263
1264                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1265                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1266                 reg->size            = memory->regions[i].memory_size;
1267                 reg->fd              = msg->fds[i];
1268
1269                 /*
1270                  * Assign invalid file descriptor value to avoid double
1271                  * closing on error path.
1272                  */
1273                 msg->fds[i] = -1;
1274
1275                 mmap_offset = memory->regions[i].mmap_offset;
1276
1277                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1278                         VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1279                         goto free_mem_table;
1280                 }
1281
1282                 dev->mem->nregions++;
1283         }
1284
1285         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1286                 goto free_mem_table;
1287
1288         for (i = 0; i < dev->nr_vring; i++) {
1289                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1290
1291                 if (!vq)
1292                         continue;
1293
1294                 if (vq->desc || vq->avail || vq->used) {
1295                         /*
1296                          * If the memory table got updated, the ring addresses
1297                          * need to be translated again as virtual addresses have
1298                          * changed.
1299                          */
1300                         vring_invalidate(dev, vq);
1301
1302                         dev = translate_ring_addresses(dev, i);
1303                         if (!dev) {
1304                                 dev = *pdev;
1305                                 goto free_mem_table;
1306                         }
1307
1308                         *pdev = dev;
1309                 }
1310         }
1311
1312         dump_guest_pages(dev);
1313
1314         return RTE_VHOST_MSG_RESULT_OK;
1315
1316 free_mem_table:
1317         free_mem_region(dev);
1318         rte_free(dev->mem);
1319         dev->mem = NULL;
1320 free_guest_pages:
1321         rte_free(dev->guest_pages);
1322         dev->guest_pages = NULL;
1323 close_msg_fds:
1324         close_msg_fds(msg);
1325         return RTE_VHOST_MSG_RESULT_ERR;
1326 }
1327
1328 static bool
1329 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1330 {
1331         bool rings_ok;
1332
1333         if (!vq)
1334                 return false;
1335
1336         if (vq_is_packed(dev))
1337                 rings_ok = vq->desc_packed && vq->driver_event &&
1338                         vq->device_event;
1339         else
1340                 rings_ok = vq->desc && vq->avail && vq->used;
1341
1342         return rings_ok &&
1343                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1344                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1345                vq->enabled;
1346 }
1347
1348 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1349
1350 static int
1351 virtio_is_ready(struct virtio_net *dev)
1352 {
1353         struct vhost_virtqueue *vq;
1354         uint32_t i, nr_vring = dev->nr_vring;
1355
1356         if (dev->flags & VIRTIO_DEV_READY)
1357                 return 1;
1358
1359         if (!dev->nr_vring)
1360                 return 0;
1361
1362         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1363                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1364
1365                 if (dev->nr_vring < nr_vring)
1366                         return 0;
1367         }
1368
1369         for (i = 0; i < nr_vring; i++) {
1370                 vq = dev->virtqueue[i];
1371
1372                 if (!vq_is_ready(dev, vq))
1373                         return 0;
1374         }
1375
1376         /* If supported, ensure the frontend is really done with config */
1377         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1378                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1379                         return 0;
1380
1381         dev->flags |= VIRTIO_DEV_READY;
1382
1383         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1384                 VHOST_LOG_CONFIG(INFO,
1385                         "virtio is now ready for processing.\n");
1386         return 1;
1387 }
1388
1389 static void *
1390 inflight_mem_alloc(const char *name, size_t size, int *fd)
1391 {
1392         void *ptr;
1393         int mfd = -1;
1394         char fname[20] = "/tmp/memfd-XXXXXX";
1395
1396         *fd = -1;
1397 #ifdef MEMFD_SUPPORTED
1398         mfd = memfd_create(name, MFD_CLOEXEC);
1399 #else
1400         RTE_SET_USED(name);
1401 #endif
1402         if (mfd == -1) {
1403                 mfd = mkstemp(fname);
1404                 if (mfd == -1) {
1405                         VHOST_LOG_CONFIG(ERR,
1406                                 "failed to get inflight buffer fd\n");
1407                         return NULL;
1408                 }
1409
1410                 unlink(fname);
1411         }
1412
1413         if (ftruncate(mfd, size) == -1) {
1414                 VHOST_LOG_CONFIG(ERR,
1415                         "failed to alloc inflight buffer\n");
1416                 close(mfd);
1417                 return NULL;
1418         }
1419
1420         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1421         if (ptr == MAP_FAILED) {
1422                 VHOST_LOG_CONFIG(ERR,
1423                         "failed to mmap inflight buffer\n");
1424                 close(mfd);
1425                 return NULL;
1426         }
1427
1428         *fd = mfd;
1429         return ptr;
1430 }
1431
1432 static uint32_t
1433 get_pervq_shm_size_split(uint16_t queue_size)
1434 {
1435         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1436                                   queue_size + sizeof(uint64_t) +
1437                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1438 }
1439
1440 static uint32_t
1441 get_pervq_shm_size_packed(uint16_t queue_size)
1442 {
1443         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1444                                   * queue_size + sizeof(uint64_t) +
1445                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1446                                   INFLIGHT_ALIGNMENT);
1447 }
1448
1449 static int
1450 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1451                            VhostUserMsg *msg,
1452                            int main_fd __rte_unused)
1453 {
1454         struct rte_vhost_inflight_info_packed *inflight_packed;
1455         uint64_t pervq_inflight_size, mmap_size;
1456         uint16_t num_queues, queue_size;
1457         struct virtio_net *dev = *pdev;
1458         int fd, i, j;
1459         void *addr;
1460
1461         if (msg->size != sizeof(msg->payload.inflight)) {
1462                 VHOST_LOG_CONFIG(ERR,
1463                         "invalid get_inflight_fd message size is %d\n",
1464                         msg->size);
1465                 return RTE_VHOST_MSG_RESULT_ERR;
1466         }
1467
1468         if (dev->inflight_info == NULL) {
1469                 dev->inflight_info = calloc(1,
1470                                             sizeof(struct inflight_mem_info));
1471                 if (!dev->inflight_info) {
1472                         VHOST_LOG_CONFIG(ERR,
1473                                 "failed to alloc dev inflight area\n");
1474                         return RTE_VHOST_MSG_RESULT_ERR;
1475                 }
1476                 dev->inflight_info->fd = -1;
1477         }
1478
1479         num_queues = msg->payload.inflight.num_queues;
1480         queue_size = msg->payload.inflight.queue_size;
1481
1482         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1483                 msg->payload.inflight.num_queues);
1484         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1485                 msg->payload.inflight.queue_size);
1486
1487         if (vq_is_packed(dev))
1488                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1489         else
1490                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1491
1492         mmap_size = num_queues * pervq_inflight_size;
1493         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1494         if (!addr) {
1495                 VHOST_LOG_CONFIG(ERR,
1496                         "failed to alloc vhost inflight area\n");
1497                         msg->payload.inflight.mmap_size = 0;
1498                 return RTE_VHOST_MSG_RESULT_ERR;
1499         }
1500         memset(addr, 0, mmap_size);
1501
1502         if (dev->inflight_info->addr) {
1503                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1504                 dev->inflight_info->addr = NULL;
1505         }
1506
1507         if (dev->inflight_info->fd >= 0) {
1508                 close(dev->inflight_info->fd);
1509                 dev->inflight_info->fd = -1;
1510         }
1511
1512         dev->inflight_info->addr = addr;
1513         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1514         dev->inflight_info->fd = msg->fds[0] = fd;
1515         msg->payload.inflight.mmap_offset = 0;
1516         msg->fd_num = 1;
1517
1518         if (vq_is_packed(dev)) {
1519                 for (i = 0; i < num_queues; i++) {
1520                         inflight_packed =
1521                                 (struct rte_vhost_inflight_info_packed *)addr;
1522                         inflight_packed->used_wrap_counter = 1;
1523                         inflight_packed->old_used_wrap_counter = 1;
1524                         for (j = 0; j < queue_size; j++)
1525                                 inflight_packed->desc[j].next = j + 1;
1526                         addr = (void *)((char *)addr + pervq_inflight_size);
1527                 }
1528         }
1529
1530         VHOST_LOG_CONFIG(INFO,
1531                 "send inflight mmap_size: %"PRIu64"\n",
1532                 msg->payload.inflight.mmap_size);
1533         VHOST_LOG_CONFIG(INFO,
1534                 "send inflight mmap_offset: %"PRIu64"\n",
1535                 msg->payload.inflight.mmap_offset);
1536         VHOST_LOG_CONFIG(INFO,
1537                 "send inflight fd: %d\n", msg->fds[0]);
1538
1539         return RTE_VHOST_MSG_RESULT_REPLY;
1540 }
1541
1542 static int
1543 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1544                            int main_fd __rte_unused)
1545 {
1546         uint64_t mmap_size, mmap_offset;
1547         uint16_t num_queues, queue_size;
1548         struct virtio_net *dev = *pdev;
1549         uint32_t pervq_inflight_size;
1550         struct vhost_virtqueue *vq;
1551         void *addr;
1552         int fd, i;
1553
1554         fd = msg->fds[0];
1555         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1556                 VHOST_LOG_CONFIG(ERR,
1557                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1558                         msg->size, fd);
1559                 return RTE_VHOST_MSG_RESULT_ERR;
1560         }
1561
1562         mmap_size = msg->payload.inflight.mmap_size;
1563         mmap_offset = msg->payload.inflight.mmap_offset;
1564         num_queues = msg->payload.inflight.num_queues;
1565         queue_size = msg->payload.inflight.queue_size;
1566
1567         if (vq_is_packed(dev))
1568                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1569         else
1570                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1571
1572         VHOST_LOG_CONFIG(INFO,
1573                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1574         VHOST_LOG_CONFIG(INFO,
1575                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1576         VHOST_LOG_CONFIG(INFO,
1577                 "set_inflight_fd num_queues: %u\n", num_queues);
1578         VHOST_LOG_CONFIG(INFO,
1579                 "set_inflight_fd queue_size: %u\n", queue_size);
1580         VHOST_LOG_CONFIG(INFO,
1581                 "set_inflight_fd fd: %d\n", fd);
1582         VHOST_LOG_CONFIG(INFO,
1583                 "set_inflight_fd pervq_inflight_size: %d\n",
1584                 pervq_inflight_size);
1585
1586         if (!dev->inflight_info) {
1587                 dev->inflight_info = calloc(1,
1588                                             sizeof(struct inflight_mem_info));
1589                 if (dev->inflight_info == NULL) {
1590                         VHOST_LOG_CONFIG(ERR,
1591                                 "failed to alloc dev inflight area\n");
1592                         return RTE_VHOST_MSG_RESULT_ERR;
1593                 }
1594                 dev->inflight_info->fd = -1;
1595         }
1596
1597         if (dev->inflight_info->addr) {
1598                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1599                 dev->inflight_info->addr = NULL;
1600         }
1601
1602         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1603                     fd, mmap_offset);
1604         if (addr == MAP_FAILED) {
1605                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1606                 return RTE_VHOST_MSG_RESULT_ERR;
1607         }
1608
1609         if (dev->inflight_info->fd >= 0) {
1610                 close(dev->inflight_info->fd);
1611                 dev->inflight_info->fd = -1;
1612         }
1613
1614         dev->inflight_info->fd = fd;
1615         dev->inflight_info->addr = addr;
1616         dev->inflight_info->size = mmap_size;
1617
1618         for (i = 0; i < num_queues; i++) {
1619                 vq = dev->virtqueue[i];
1620                 if (!vq)
1621                         continue;
1622
1623                 if (vq_is_packed(dev)) {
1624                         vq->inflight_packed = addr;
1625                         vq->inflight_packed->desc_num = queue_size;
1626                 } else {
1627                         vq->inflight_split = addr;
1628                         vq->inflight_split->desc_num = queue_size;
1629                 }
1630                 addr = (void *)((char *)addr + pervq_inflight_size);
1631         }
1632
1633         return RTE_VHOST_MSG_RESULT_OK;
1634 }
1635
1636 static int
1637 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1638                         int main_fd __rte_unused)
1639 {
1640         struct virtio_net *dev = *pdev;
1641         struct vhost_vring_file file;
1642         struct vhost_virtqueue *vq;
1643         int expected_fds;
1644
1645         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1646         if (validate_msg_fds(msg, expected_fds) != 0)
1647                 return RTE_VHOST_MSG_RESULT_ERR;
1648
1649         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1650         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1651                 file.fd = VIRTIO_INVALID_EVENTFD;
1652         else
1653                 file.fd = msg->fds[0];
1654         VHOST_LOG_CONFIG(INFO,
1655                 "vring call idx:%d file:%d\n", file.index, file.fd);
1656
1657         vq = dev->virtqueue[file.index];
1658
1659         if (vq->ready) {
1660                 vq->ready = false;
1661                 vhost_user_notify_queue_state(dev, file.index, 0);
1662         }
1663
1664         if (vq->callfd >= 0)
1665                 close(vq->callfd);
1666
1667         vq->callfd = file.fd;
1668
1669         return RTE_VHOST_MSG_RESULT_OK;
1670 }
1671
1672 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1673                         struct VhostUserMsg *msg,
1674                         int main_fd __rte_unused)
1675 {
1676         int expected_fds;
1677
1678         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1679         if (validate_msg_fds(msg, expected_fds) != 0)
1680                 return RTE_VHOST_MSG_RESULT_ERR;
1681
1682         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1683                 close(msg->fds[0]);
1684         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1685
1686         return RTE_VHOST_MSG_RESULT_OK;
1687 }
1688
1689 static int
1690 resubmit_desc_compare(const void *a, const void *b)
1691 {
1692         const struct rte_vhost_resubmit_desc *desc0 = a;
1693         const struct rte_vhost_resubmit_desc *desc1 = b;
1694
1695         if (desc1->counter > desc0->counter)
1696                 return 1;
1697
1698         return -1;
1699 }
1700
1701 static int
1702 vhost_check_queue_inflights_split(struct virtio_net *dev,
1703                                   struct vhost_virtqueue *vq)
1704 {
1705         uint16_t i;
1706         uint16_t resubmit_num = 0, last_io, num;
1707         struct vring_used *used = vq->used;
1708         struct rte_vhost_resubmit_info *resubmit;
1709         struct rte_vhost_inflight_info_split *inflight_split;
1710
1711         if (!(dev->protocol_features &
1712             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1713                 return RTE_VHOST_MSG_RESULT_OK;
1714
1715         /* The frontend may still not support the inflight feature
1716          * although we negotiate the protocol feature.
1717          */
1718         if ((!vq->inflight_split))
1719                 return RTE_VHOST_MSG_RESULT_OK;
1720
1721         if (!vq->inflight_split->version) {
1722                 vq->inflight_split->version = INFLIGHT_VERSION;
1723                 return RTE_VHOST_MSG_RESULT_OK;
1724         }
1725
1726         if (vq->resubmit_inflight)
1727                 return RTE_VHOST_MSG_RESULT_OK;
1728
1729         inflight_split = vq->inflight_split;
1730         vq->global_counter = 0;
1731         last_io = inflight_split->last_inflight_io;
1732
1733         if (inflight_split->used_idx != used->idx) {
1734                 inflight_split->desc[last_io].inflight = 0;
1735                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1736                 inflight_split->used_idx = used->idx;
1737         }
1738
1739         for (i = 0; i < inflight_split->desc_num; i++) {
1740                 if (inflight_split->desc[i].inflight == 1)
1741                         resubmit_num++;
1742         }
1743
1744         vq->last_avail_idx += resubmit_num;
1745
1746         if (resubmit_num) {
1747                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1748                 if (!resubmit) {
1749                         VHOST_LOG_CONFIG(ERR,
1750                                 "failed to allocate memory for resubmit info.\n");
1751                         return RTE_VHOST_MSG_RESULT_ERR;
1752                 }
1753
1754                 resubmit->resubmit_list = calloc(resubmit_num,
1755                         sizeof(struct rte_vhost_resubmit_desc));
1756                 if (!resubmit->resubmit_list) {
1757                         VHOST_LOG_CONFIG(ERR,
1758                                 "failed to allocate memory for inflight desc.\n");
1759                         free(resubmit);
1760                         return RTE_VHOST_MSG_RESULT_ERR;
1761                 }
1762
1763                 num = 0;
1764                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1765                         if (vq->inflight_split->desc[i].inflight == 1) {
1766                                 resubmit->resubmit_list[num].index = i;
1767                                 resubmit->resubmit_list[num].counter =
1768                                         inflight_split->desc[i].counter;
1769                                 num++;
1770                         }
1771                 }
1772                 resubmit->resubmit_num = num;
1773
1774                 if (resubmit->resubmit_num > 1)
1775                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1776                               sizeof(struct rte_vhost_resubmit_desc),
1777                               resubmit_desc_compare);
1778
1779                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1780                 vq->resubmit_inflight = resubmit;
1781         }
1782
1783         return RTE_VHOST_MSG_RESULT_OK;
1784 }
1785
1786 static int
1787 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1788                                    struct vhost_virtqueue *vq)
1789 {
1790         uint16_t i;
1791         uint16_t resubmit_num = 0, old_used_idx, num;
1792         struct rte_vhost_resubmit_info *resubmit;
1793         struct rte_vhost_inflight_info_packed *inflight_packed;
1794
1795         if (!(dev->protocol_features &
1796             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1797                 return RTE_VHOST_MSG_RESULT_OK;
1798
1799         /* The frontend may still not support the inflight feature
1800          * although we negotiate the protocol feature.
1801          */
1802         if ((!vq->inflight_packed))
1803                 return RTE_VHOST_MSG_RESULT_OK;
1804
1805         if (!vq->inflight_packed->version) {
1806                 vq->inflight_packed->version = INFLIGHT_VERSION;
1807                 return RTE_VHOST_MSG_RESULT_OK;
1808         }
1809
1810         if (vq->resubmit_inflight)
1811                 return RTE_VHOST_MSG_RESULT_OK;
1812
1813         inflight_packed = vq->inflight_packed;
1814         vq->global_counter = 0;
1815         old_used_idx = inflight_packed->old_used_idx;
1816
1817         if (inflight_packed->used_idx != old_used_idx) {
1818                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1819                         inflight_packed->old_used_idx =
1820                                 inflight_packed->used_idx;
1821                         inflight_packed->old_used_wrap_counter =
1822                                 inflight_packed->used_wrap_counter;
1823                         inflight_packed->old_free_head =
1824                                 inflight_packed->free_head;
1825                 } else {
1826                         inflight_packed->used_idx =
1827                                 inflight_packed->old_used_idx;
1828                         inflight_packed->used_wrap_counter =
1829                                 inflight_packed->old_used_wrap_counter;
1830                         inflight_packed->free_head =
1831                                 inflight_packed->old_free_head;
1832                 }
1833         }
1834
1835         for (i = 0; i < inflight_packed->desc_num; i++) {
1836                 if (inflight_packed->desc[i].inflight == 1)
1837                         resubmit_num++;
1838         }
1839
1840         if (resubmit_num) {
1841                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1842                 if (resubmit == NULL) {
1843                         VHOST_LOG_CONFIG(ERR,
1844                                 "failed to allocate memory for resubmit info.\n");
1845                         return RTE_VHOST_MSG_RESULT_ERR;
1846                 }
1847
1848                 resubmit->resubmit_list = calloc(resubmit_num,
1849                         sizeof(struct rte_vhost_resubmit_desc));
1850                 if (resubmit->resubmit_list == NULL) {
1851                         VHOST_LOG_CONFIG(ERR,
1852                                 "failed to allocate memory for resubmit desc.\n");
1853                         free(resubmit);
1854                         return RTE_VHOST_MSG_RESULT_ERR;
1855                 }
1856
1857                 num = 0;
1858                 for (i = 0; i < inflight_packed->desc_num; i++) {
1859                         if (vq->inflight_packed->desc[i].inflight == 1) {
1860                                 resubmit->resubmit_list[num].index = i;
1861                                 resubmit->resubmit_list[num].counter =
1862                                         inflight_packed->desc[i].counter;
1863                                 num++;
1864                         }
1865                 }
1866                 resubmit->resubmit_num = num;
1867
1868                 if (resubmit->resubmit_num > 1)
1869                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1870                               sizeof(struct rte_vhost_resubmit_desc),
1871                               resubmit_desc_compare);
1872
1873                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1874                 vq->resubmit_inflight = resubmit;
1875         }
1876
1877         return RTE_VHOST_MSG_RESULT_OK;
1878 }
1879
1880 static int
1881 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1882                         int main_fd __rte_unused)
1883 {
1884         struct virtio_net *dev = *pdev;
1885         struct vhost_vring_file file;
1886         struct vhost_virtqueue *vq;
1887         int expected_fds;
1888
1889         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1890         if (validate_msg_fds(msg, expected_fds) != 0)
1891                 return RTE_VHOST_MSG_RESULT_ERR;
1892
1893         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1894         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1895                 file.fd = VIRTIO_INVALID_EVENTFD;
1896         else
1897                 file.fd = msg->fds[0];
1898         VHOST_LOG_CONFIG(INFO,
1899                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1900
1901         /* Interpret ring addresses only when ring is started. */
1902         dev = translate_ring_addresses(dev, file.index);
1903         if (!dev) {
1904                 if (file.fd != VIRTIO_INVALID_EVENTFD)
1905                         close(file.fd);
1906
1907                 return RTE_VHOST_MSG_RESULT_ERR;
1908         }
1909
1910         *pdev = dev;
1911
1912         vq = dev->virtqueue[file.index];
1913
1914         /*
1915          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1916          * the ring starts already enabled. Otherwise, it is enabled via
1917          * the SET_VRING_ENABLE message.
1918          */
1919         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1920                 vq->enabled = true;
1921                 if (dev->notify_ops->vring_state_changed)
1922                         dev->notify_ops->vring_state_changed(
1923                                 dev->vid, file.index, 1);
1924         }
1925
1926         if (vq->ready) {
1927                 vq->ready = false;
1928                 vhost_user_notify_queue_state(dev, file.index, 0);
1929         }
1930
1931         if (vq->kickfd >= 0)
1932                 close(vq->kickfd);
1933         vq->kickfd = file.fd;
1934
1935         if (vq_is_packed(dev)) {
1936                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1937                         VHOST_LOG_CONFIG(ERR,
1938                                 "failed to inflights for vq: %d\n", file.index);
1939                         return RTE_VHOST_MSG_RESULT_ERR;
1940                 }
1941         } else {
1942                 if (vhost_check_queue_inflights_split(dev, vq)) {
1943                         VHOST_LOG_CONFIG(ERR,
1944                                 "failed to inflights for vq: %d\n", file.index);
1945                         return RTE_VHOST_MSG_RESULT_ERR;
1946                 }
1947         }
1948
1949         return RTE_VHOST_MSG_RESULT_OK;
1950 }
1951
1952 /*
1953  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1954  */
1955 static int
1956 vhost_user_get_vring_base(struct virtio_net **pdev,
1957                         struct VhostUserMsg *msg,
1958                         int main_fd __rte_unused)
1959 {
1960         struct virtio_net *dev = *pdev;
1961         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1962         uint64_t val;
1963
1964         if (validate_msg_fds(msg, 0) != 0)
1965                 return RTE_VHOST_MSG_RESULT_ERR;
1966
1967         /* We have to stop the queue (virtio) if it is running. */
1968         vhost_destroy_device_notify(dev);
1969
1970         dev->flags &= ~VIRTIO_DEV_READY;
1971         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1972
1973         /* Here we are safe to get the indexes */
1974         if (vq_is_packed(dev)) {
1975                 /*
1976                  * Bit[0:14]: avail index
1977                  * Bit[15]: avail wrap counter
1978                  */
1979                 val = vq->last_avail_idx & 0x7fff;
1980                 val |= vq->avail_wrap_counter << 15;
1981                 msg->payload.state.num = val;
1982         } else {
1983                 msg->payload.state.num = vq->last_avail_idx;
1984         }
1985
1986         VHOST_LOG_CONFIG(INFO,
1987                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1988                 msg->payload.state.num);
1989         /*
1990          * Based on current qemu vhost-user implementation, this message is
1991          * sent and only sent in vhost_vring_stop.
1992          * TODO: cleanup the vring, it isn't usable since here.
1993          */
1994         if (vq->kickfd >= 0)
1995                 close(vq->kickfd);
1996
1997         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1998
1999         if (vq->callfd >= 0)
2000                 close(vq->callfd);
2001
2002         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2003
2004         vq->signalled_used_valid = false;
2005
2006         if (vq_is_packed(dev)) {
2007                 rte_free(vq->shadow_used_packed);
2008                 vq->shadow_used_packed = NULL;
2009         } else {
2010                 rte_free(vq->shadow_used_split);
2011                 vq->shadow_used_split = NULL;
2012
2013                 if (vq->async_pkts_info)
2014                         rte_free(vq->async_pkts_info);
2015                 if (vq->async_descs_split)
2016                         rte_free(vq->async_descs_split);
2017                 vq->async_pkts_info = NULL;
2018                 vq->async_descs_split = NULL;
2019         }
2020
2021         rte_free(vq->batch_copy_elems);
2022         vq->batch_copy_elems = NULL;
2023
2024         rte_free(vq->log_cache);
2025         vq->log_cache = NULL;
2026
2027         msg->size = sizeof(msg->payload.state);
2028         msg->fd_num = 0;
2029
2030         vring_invalidate(dev, vq);
2031
2032         return RTE_VHOST_MSG_RESULT_REPLY;
2033 }
2034
2035 /*
2036  * when virtio queues are ready to work, qemu will send us to
2037  * enable the virtio queue pair.
2038  */
2039 static int
2040 vhost_user_set_vring_enable(struct virtio_net **pdev,
2041                         struct VhostUserMsg *msg,
2042                         int main_fd __rte_unused)
2043 {
2044         struct virtio_net *dev = *pdev;
2045         bool enable = !!msg->payload.state.num;
2046         int index = (int)msg->payload.state.index;
2047
2048         if (validate_msg_fds(msg, 0) != 0)
2049                 return RTE_VHOST_MSG_RESULT_ERR;
2050
2051         VHOST_LOG_CONFIG(INFO,
2052                 "set queue enable: %d to qp idx: %d\n",
2053                 enable, index);
2054
2055         if (enable && dev->virtqueue[index]->async_registered) {
2056                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2057                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2058                         "async inflight packets must be completed first\n");
2059                         return RTE_VHOST_MSG_RESULT_ERR;
2060                 }
2061         }
2062
2063         dev->virtqueue[index]->enabled = enable;
2064
2065         return RTE_VHOST_MSG_RESULT_OK;
2066 }
2067
2068 static int
2069 vhost_user_get_protocol_features(struct virtio_net **pdev,
2070                         struct VhostUserMsg *msg,
2071                         int main_fd __rte_unused)
2072 {
2073         struct virtio_net *dev = *pdev;
2074         uint64_t features, protocol_features;
2075
2076         if (validate_msg_fds(msg, 0) != 0)
2077                 return RTE_VHOST_MSG_RESULT_ERR;
2078
2079         rte_vhost_driver_get_features(dev->ifname, &features);
2080         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2081
2082         msg->payload.u64 = protocol_features;
2083         msg->size = sizeof(msg->payload.u64);
2084         msg->fd_num = 0;
2085
2086         return RTE_VHOST_MSG_RESULT_REPLY;
2087 }
2088
2089 static int
2090 vhost_user_set_protocol_features(struct virtio_net **pdev,
2091                         struct VhostUserMsg *msg,
2092                         int main_fd __rte_unused)
2093 {
2094         struct virtio_net *dev = *pdev;
2095         uint64_t protocol_features = msg->payload.u64;
2096         uint64_t slave_protocol_features = 0;
2097
2098         if (validate_msg_fds(msg, 0) != 0)
2099                 return RTE_VHOST_MSG_RESULT_ERR;
2100
2101         rte_vhost_driver_get_protocol_features(dev->ifname,
2102                         &slave_protocol_features);
2103         if (protocol_features & ~slave_protocol_features) {
2104                 VHOST_LOG_CONFIG(ERR,
2105                         "(%d) received invalid protocol features.\n",
2106                         dev->vid);
2107                 return RTE_VHOST_MSG_RESULT_ERR;
2108         }
2109
2110         dev->protocol_features = protocol_features;
2111         VHOST_LOG_CONFIG(INFO,
2112                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2113                 dev->protocol_features);
2114
2115         return RTE_VHOST_MSG_RESULT_OK;
2116 }
2117
2118 static int
2119 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2120                         int main_fd __rte_unused)
2121 {
2122         struct virtio_net *dev = *pdev;
2123         int fd = msg->fds[0];
2124         uint64_t size, off;
2125         void *addr;
2126         uint32_t i;
2127
2128         if (validate_msg_fds(msg, 1) != 0)
2129                 return RTE_VHOST_MSG_RESULT_ERR;
2130
2131         if (fd < 0) {
2132                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2133                 return RTE_VHOST_MSG_RESULT_ERR;
2134         }
2135
2136         if (msg->size != sizeof(VhostUserLog)) {
2137                 VHOST_LOG_CONFIG(ERR,
2138                         "invalid log base msg size: %"PRId32" != %d\n",
2139                         msg->size, (int)sizeof(VhostUserLog));
2140                 goto close_msg_fds;
2141         }
2142
2143         size = msg->payload.log.mmap_size;
2144         off  = msg->payload.log.mmap_offset;
2145
2146         /* Check for mmap size and offset overflow. */
2147         if (off >= -size) {
2148                 VHOST_LOG_CONFIG(ERR,
2149                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2150                         off, size);
2151                 goto close_msg_fds;
2152         }
2153
2154         VHOST_LOG_CONFIG(INFO,
2155                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2156                 size, off);
2157
2158         /*
2159          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2160          * fail when offset is not page size aligned.
2161          */
2162         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2163         close(fd);
2164         if (addr == MAP_FAILED) {
2165                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2166                 return RTE_VHOST_MSG_RESULT_ERR;
2167         }
2168
2169         /*
2170          * Free previously mapped log memory on occasionally
2171          * multiple VHOST_USER_SET_LOG_BASE.
2172          */
2173         if (dev->log_addr) {
2174                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2175         }
2176         dev->log_addr = (uint64_t)(uintptr_t)addr;
2177         dev->log_base = dev->log_addr + off;
2178         dev->log_size = size;
2179
2180         for (i = 0; i < dev->nr_vring; i++) {
2181                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2182
2183                 rte_free(vq->log_cache);
2184                 vq->log_cache = NULL;
2185                 vq->log_cache_nb_elem = 0;
2186                 vq->log_cache = rte_zmalloc("vq log cache",
2187                                 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2188                                 0);
2189                 /*
2190                  * If log cache alloc fail, don't fail migration, but no
2191                  * caching will be done, which will impact performance
2192                  */
2193                 if (!vq->log_cache)
2194                         VHOST_LOG_CONFIG(ERR, "Failed to allocate VQ logging cache\n");
2195         }
2196
2197         /*
2198          * The spec is not clear about it (yet), but QEMU doesn't expect
2199          * any payload in the reply.
2200          */
2201         msg->size = 0;
2202         msg->fd_num = 0;
2203
2204         return RTE_VHOST_MSG_RESULT_REPLY;
2205
2206 close_msg_fds:
2207         close_msg_fds(msg);
2208         return RTE_VHOST_MSG_RESULT_ERR;
2209 }
2210
2211 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2212                         struct VhostUserMsg *msg,
2213                         int main_fd __rte_unused)
2214 {
2215         if (validate_msg_fds(msg, 1) != 0)
2216                 return RTE_VHOST_MSG_RESULT_ERR;
2217
2218         close(msg->fds[0]);
2219         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2220
2221         return RTE_VHOST_MSG_RESULT_OK;
2222 }
2223
2224 /*
2225  * An rarp packet is constructed and broadcasted to notify switches about
2226  * the new location of the migrated VM, so that packets from outside will
2227  * not be lost after migration.
2228  *
2229  * However, we don't actually "send" a rarp packet here, instead, we set
2230  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2231  */
2232 static int
2233 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2234                         int main_fd __rte_unused)
2235 {
2236         struct virtio_net *dev = *pdev;
2237         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2238         struct rte_vdpa_device *vdpa_dev;
2239
2240         if (validate_msg_fds(msg, 0) != 0)
2241                 return RTE_VHOST_MSG_RESULT_ERR;
2242
2243         VHOST_LOG_CONFIG(DEBUG,
2244                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2245                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2246         memcpy(dev->mac.addr_bytes, mac, 6);
2247
2248         /*
2249          * Set the flag to inject a RARP broadcast packet at
2250          * rte_vhost_dequeue_burst().
2251          *
2252          * __ATOMIC_RELEASE ordering is for making sure the mac is
2253          * copied before the flag is set.
2254          */
2255         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2256         vdpa_dev = dev->vdpa_dev;
2257         if (vdpa_dev && vdpa_dev->ops->migration_done)
2258                 vdpa_dev->ops->migration_done(dev->vid);
2259
2260         return RTE_VHOST_MSG_RESULT_OK;
2261 }
2262
2263 static int
2264 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2265                         int main_fd __rte_unused)
2266 {
2267         struct virtio_net *dev = *pdev;
2268
2269         if (validate_msg_fds(msg, 0) != 0)
2270                 return RTE_VHOST_MSG_RESULT_ERR;
2271
2272         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2273                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2274                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2275                                 msg->payload.u64);
2276
2277                 return RTE_VHOST_MSG_RESULT_ERR;
2278         }
2279
2280         dev->mtu = msg->payload.u64;
2281
2282         return RTE_VHOST_MSG_RESULT_OK;
2283 }
2284
2285 static int
2286 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2287                         int main_fd __rte_unused)
2288 {
2289         struct virtio_net *dev = *pdev;
2290         int fd = msg->fds[0];
2291
2292         if (validate_msg_fds(msg, 1) != 0)
2293                 return RTE_VHOST_MSG_RESULT_ERR;
2294
2295         if (fd < 0) {
2296                 VHOST_LOG_CONFIG(ERR,
2297                                 "Invalid file descriptor for slave channel (%d)\n",
2298                                 fd);
2299                 return RTE_VHOST_MSG_RESULT_ERR;
2300         }
2301
2302         if (dev->slave_req_fd >= 0)
2303                 close(dev->slave_req_fd);
2304
2305         dev->slave_req_fd = fd;
2306
2307         return RTE_VHOST_MSG_RESULT_OK;
2308 }
2309
2310 static int
2311 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2312 {
2313         struct vhost_vring_addr *ra;
2314         uint64_t start, end, len;
2315
2316         start = imsg->iova;
2317         end = start + imsg->size;
2318
2319         ra = &vq->ring_addrs;
2320         len = sizeof(struct vring_desc) * vq->size;
2321         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2322                 return 1;
2323
2324         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2325         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2326                 return 1;
2327
2328         len = sizeof(struct vring_used) +
2329                sizeof(struct vring_used_elem) * vq->size;
2330         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2331                 return 1;
2332
2333         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2334                 len = sizeof(uint64_t);
2335                 if (ra->log_guest_addr < end &&
2336                     (ra->log_guest_addr + len) > start)
2337                         return 1;
2338         }
2339
2340         return 0;
2341 }
2342
2343 static int
2344 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2345 {
2346         struct vhost_vring_addr *ra;
2347         uint64_t start, end, len;
2348
2349         start = imsg->iova;
2350         end = start + imsg->size;
2351
2352         ra = &vq->ring_addrs;
2353         len = sizeof(struct vring_packed_desc) * vq->size;
2354         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2355                 return 1;
2356
2357         len = sizeof(struct vring_packed_desc_event);
2358         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2359                 return 1;
2360
2361         len = sizeof(struct vring_packed_desc_event);
2362         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2363                 return 1;
2364
2365         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2366                 len = sizeof(uint64_t);
2367                 if (ra->log_guest_addr < end &&
2368                     (ra->log_guest_addr + len) > start)
2369                         return 1;
2370         }
2371
2372         return 0;
2373 }
2374
2375 static int is_vring_iotlb(struct virtio_net *dev,
2376                           struct vhost_virtqueue *vq,
2377                           struct vhost_iotlb_msg *imsg)
2378 {
2379         if (vq_is_packed(dev))
2380                 return is_vring_iotlb_packed(vq, imsg);
2381         else
2382                 return is_vring_iotlb_split(vq, imsg);
2383 }
2384
2385 static int
2386 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2387                         int main_fd __rte_unused)
2388 {
2389         struct virtio_net *dev = *pdev;
2390         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2391         uint16_t i;
2392         uint64_t vva, len;
2393
2394         if (validate_msg_fds(msg, 0) != 0)
2395                 return RTE_VHOST_MSG_RESULT_ERR;
2396
2397         switch (imsg->type) {
2398         case VHOST_IOTLB_UPDATE:
2399                 len = imsg->size;
2400                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2401                 if (!vva)
2402                         return RTE_VHOST_MSG_RESULT_ERR;
2403
2404                 for (i = 0; i < dev->nr_vring; i++) {
2405                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2406
2407                         if (!vq)
2408                                 continue;
2409
2410                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2411                                         len, imsg->perm);
2412
2413                         if (is_vring_iotlb(dev, vq, imsg))
2414                                 *pdev = dev = translate_ring_addresses(dev, i);
2415                 }
2416                 break;
2417         case VHOST_IOTLB_INVALIDATE:
2418                 for (i = 0; i < dev->nr_vring; i++) {
2419                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2420
2421                         if (!vq)
2422                                 continue;
2423
2424                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2425                                         imsg->size);
2426
2427                         if (is_vring_iotlb(dev, vq, imsg))
2428                                 vring_invalidate(dev, vq);
2429                 }
2430                 break;
2431         default:
2432                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2433                                 imsg->type);
2434                 return RTE_VHOST_MSG_RESULT_ERR;
2435         }
2436
2437         return RTE_VHOST_MSG_RESULT_OK;
2438 }
2439
2440 static int
2441 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2442                         struct VhostUserMsg *msg,
2443                         int main_fd __rte_unused)
2444 {
2445         struct virtio_net *dev = *pdev;
2446 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2447         struct uffdio_api api_struct;
2448
2449         if (validate_msg_fds(msg, 0) != 0)
2450                 return RTE_VHOST_MSG_RESULT_ERR;
2451
2452         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2453
2454         if (dev->postcopy_ufd == -1) {
2455                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2456                         strerror(errno));
2457                 return RTE_VHOST_MSG_RESULT_ERR;
2458         }
2459         api_struct.api = UFFD_API;
2460         api_struct.features = 0;
2461         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2462                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2463                         strerror(errno));
2464                 close(dev->postcopy_ufd);
2465                 dev->postcopy_ufd = -1;
2466                 return RTE_VHOST_MSG_RESULT_ERR;
2467         }
2468         msg->fds[0] = dev->postcopy_ufd;
2469         msg->fd_num = 1;
2470
2471         return RTE_VHOST_MSG_RESULT_REPLY;
2472 #else
2473         dev->postcopy_ufd = -1;
2474         msg->fd_num = 0;
2475
2476         return RTE_VHOST_MSG_RESULT_ERR;
2477 #endif
2478 }
2479
2480 static int
2481 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2482                         struct VhostUserMsg *msg __rte_unused,
2483                         int main_fd __rte_unused)
2484 {
2485         struct virtio_net *dev = *pdev;
2486
2487         if (validate_msg_fds(msg, 0) != 0)
2488                 return RTE_VHOST_MSG_RESULT_ERR;
2489
2490         if (dev->mem && dev->mem->nregions) {
2491                 VHOST_LOG_CONFIG(ERR,
2492                         "Regions already registered at postcopy-listen\n");
2493                 return RTE_VHOST_MSG_RESULT_ERR;
2494         }
2495         dev->postcopy_listening = 1;
2496
2497         return RTE_VHOST_MSG_RESULT_OK;
2498 }
2499
2500 static int
2501 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2502                         int main_fd __rte_unused)
2503 {
2504         struct virtio_net *dev = *pdev;
2505
2506         if (validate_msg_fds(msg, 0) != 0)
2507                 return RTE_VHOST_MSG_RESULT_ERR;
2508
2509         dev->postcopy_listening = 0;
2510         if (dev->postcopy_ufd >= 0) {
2511                 close(dev->postcopy_ufd);
2512                 dev->postcopy_ufd = -1;
2513         }
2514
2515         msg->payload.u64 = 0;
2516         msg->size = sizeof(msg->payload.u64);
2517         msg->fd_num = 0;
2518
2519         return RTE_VHOST_MSG_RESULT_REPLY;
2520 }
2521
2522 static int
2523 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2524                       int main_fd __rte_unused)
2525 {
2526         struct virtio_net *dev = *pdev;
2527
2528         if (validate_msg_fds(msg, 0) != 0)
2529                 return RTE_VHOST_MSG_RESULT_ERR;
2530
2531         msg->payload.u64 = dev->status;
2532         msg->size = sizeof(msg->payload.u64);
2533         msg->fd_num = 0;
2534
2535         return RTE_VHOST_MSG_RESULT_REPLY;
2536 }
2537
2538 static int
2539 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2540                         int main_fd __rte_unused)
2541 {
2542         struct virtio_net *dev = *pdev;
2543
2544         if (validate_msg_fds(msg, 0) != 0)
2545                 return RTE_VHOST_MSG_RESULT_ERR;
2546
2547         /* As per Virtio specification, the device status is 8bits long */
2548         if (msg->payload.u64 > UINT8_MAX) {
2549                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2550                                 msg->payload.u64);
2551                 return RTE_VHOST_MSG_RESULT_ERR;
2552         }
2553
2554         dev->status = msg->payload.u64;
2555
2556         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2557             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2558                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2559                 /*
2560                  * Clear the bit to let the driver know about the feature
2561                  * negotiation failure
2562                  */
2563                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2564         }
2565
2566         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2567                         "\t-RESET: %u\n"
2568                         "\t-ACKNOWLEDGE: %u\n"
2569                         "\t-DRIVER: %u\n"
2570                         "\t-FEATURES_OK: %u\n"
2571                         "\t-DRIVER_OK: %u\n"
2572                         "\t-DEVICE_NEED_RESET: %u\n"
2573                         "\t-FAILED: %u\n",
2574                         dev->status,
2575                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2576                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2577                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2578                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2579                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2580                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2581                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2582
2583         return RTE_VHOST_MSG_RESULT_OK;
2584 }
2585
2586 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2587                                         struct VhostUserMsg *msg,
2588                                         int main_fd);
2589 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2590         [VHOST_USER_NONE] = NULL,
2591         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2592         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2593         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2594         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2595         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2596         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2597         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2598         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2599         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2600         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2601         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2602         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2603         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2604         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2605         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2606         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2607         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2608         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2609         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2610         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2611         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2612         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2613         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2614         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2615         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2616         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2617         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2618         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2619         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2620 };
2621
2622 /* return bytes# of read on success or negative val on failure. */
2623 static int
2624 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2625 {
2626         int ret;
2627
2628         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2629                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2630         if (ret <= 0) {
2631                 return ret;
2632         } else if (ret != VHOST_USER_HDR_SIZE) {
2633                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2634                 close_msg_fds(msg);
2635                 return -1;
2636         }
2637
2638         if (msg->size) {
2639                 if (msg->size > sizeof(msg->payload)) {
2640                         VHOST_LOG_CONFIG(ERR,
2641                                 "invalid msg size: %d\n", msg->size);
2642                         return -1;
2643                 }
2644                 ret = read(sockfd, &msg->payload, msg->size);
2645                 if (ret <= 0)
2646                         return ret;
2647                 if (ret != (int)msg->size) {
2648                         VHOST_LOG_CONFIG(ERR,
2649                                 "read control message failed\n");
2650                         return -1;
2651                 }
2652         }
2653
2654         return ret;
2655 }
2656
2657 static int
2658 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2659 {
2660         if (!msg)
2661                 return 0;
2662
2663         return send_fd_message(sockfd, (char *)msg,
2664                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2665 }
2666
2667 static int
2668 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2669 {
2670         if (!msg)
2671                 return 0;
2672
2673         msg->flags &= ~VHOST_USER_VERSION_MASK;
2674         msg->flags &= ~VHOST_USER_NEED_REPLY;
2675         msg->flags |= VHOST_USER_VERSION;
2676         msg->flags |= VHOST_USER_REPLY_MASK;
2677
2678         return send_vhost_message(sockfd, msg);
2679 }
2680
2681 static int
2682 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2683 {
2684         int ret;
2685
2686         if (msg->flags & VHOST_USER_NEED_REPLY)
2687                 rte_spinlock_lock(&dev->slave_req_lock);
2688
2689         ret = send_vhost_message(dev->slave_req_fd, msg);
2690         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2691                 rte_spinlock_unlock(&dev->slave_req_lock);
2692
2693         return ret;
2694 }
2695
2696 /*
2697  * Allocate a queue pair if it hasn't been allocated yet
2698  */
2699 static int
2700 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2701                         struct VhostUserMsg *msg)
2702 {
2703         uint32_t vring_idx;
2704
2705         switch (msg->request.master) {
2706         case VHOST_USER_SET_VRING_KICK:
2707         case VHOST_USER_SET_VRING_CALL:
2708         case VHOST_USER_SET_VRING_ERR:
2709                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2710                 break;
2711         case VHOST_USER_SET_VRING_NUM:
2712         case VHOST_USER_SET_VRING_BASE:
2713         case VHOST_USER_SET_VRING_ENABLE:
2714                 vring_idx = msg->payload.state.index;
2715                 break;
2716         case VHOST_USER_SET_VRING_ADDR:
2717                 vring_idx = msg->payload.addr.index;
2718                 break;
2719         default:
2720                 return 0;
2721         }
2722
2723         if (vring_idx >= VHOST_MAX_VRING) {
2724                 VHOST_LOG_CONFIG(ERR,
2725                         "invalid vring index: %u\n", vring_idx);
2726                 return -1;
2727         }
2728
2729         if (dev->virtqueue[vring_idx])
2730                 return 0;
2731
2732         return alloc_vring_queue(dev, vring_idx);
2733 }
2734
2735 static void
2736 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2737 {
2738         unsigned int i = 0;
2739         unsigned int vq_num = 0;
2740
2741         while (vq_num < dev->nr_vring) {
2742                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2743
2744                 if (vq) {
2745                         rte_spinlock_lock(&vq->access_lock);
2746                         vq_num++;
2747                 }
2748                 i++;
2749         }
2750 }
2751
2752 static void
2753 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2754 {
2755         unsigned int i = 0;
2756         unsigned int vq_num = 0;
2757
2758         while (vq_num < dev->nr_vring) {
2759                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2760
2761                 if (vq) {
2762                         rte_spinlock_unlock(&vq->access_lock);
2763                         vq_num++;
2764                 }
2765                 i++;
2766         }
2767 }
2768
2769 int
2770 vhost_user_msg_handler(int vid, int fd)
2771 {
2772         struct virtio_net *dev;
2773         struct VhostUserMsg msg;
2774         struct rte_vdpa_device *vdpa_dev;
2775         int ret;
2776         int unlock_required = 0;
2777         bool handled;
2778         int request;
2779         uint32_t i;
2780
2781         dev = get_device(vid);
2782         if (dev == NULL)
2783                 return -1;
2784
2785         if (!dev->notify_ops) {
2786                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2787                 if (!dev->notify_ops) {
2788                         VHOST_LOG_CONFIG(ERR,
2789                                 "failed to get callback ops for driver %s\n",
2790                                 dev->ifname);
2791                         return -1;
2792                 }
2793         }
2794
2795         ret = read_vhost_message(fd, &msg);
2796         if (ret <= 0) {
2797                 if (ret < 0)
2798                         VHOST_LOG_CONFIG(ERR,
2799                                 "vhost read message failed\n");
2800                 else
2801                         VHOST_LOG_CONFIG(INFO,
2802                                 "vhost peer closed\n");
2803
2804                 return -1;
2805         }
2806
2807         ret = 0;
2808         request = msg.request.master;
2809         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2810                         vhost_message_str[request]) {
2811                 if (request != VHOST_USER_IOTLB_MSG)
2812                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2813                                 vhost_message_str[request]);
2814                 else
2815                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2816                                 vhost_message_str[request]);
2817         } else {
2818                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2819         }
2820
2821         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2822         if (ret < 0) {
2823                 VHOST_LOG_CONFIG(ERR,
2824                         "failed to alloc queue\n");
2825                 return -1;
2826         }
2827
2828         /*
2829          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2830          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2831          * and device is destroyed. destroy_device waits for queues to be
2832          * inactive, so it is safe. Otherwise taking the access_lock
2833          * would cause a dead lock.
2834          */
2835         switch (request) {
2836         case VHOST_USER_SET_FEATURES:
2837         case VHOST_USER_SET_PROTOCOL_FEATURES:
2838         case VHOST_USER_SET_OWNER:
2839         case VHOST_USER_SET_MEM_TABLE:
2840         case VHOST_USER_SET_LOG_BASE:
2841         case VHOST_USER_SET_LOG_FD:
2842         case VHOST_USER_SET_VRING_NUM:
2843         case VHOST_USER_SET_VRING_ADDR:
2844         case VHOST_USER_SET_VRING_BASE:
2845         case VHOST_USER_SET_VRING_KICK:
2846         case VHOST_USER_SET_VRING_CALL:
2847         case VHOST_USER_SET_VRING_ERR:
2848         case VHOST_USER_SET_VRING_ENABLE:
2849         case VHOST_USER_SEND_RARP:
2850         case VHOST_USER_NET_SET_MTU:
2851         case VHOST_USER_SET_SLAVE_REQ_FD:
2852                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2853                         vhost_user_lock_all_queue_pairs(dev);
2854                         unlock_required = 1;
2855                 }
2856                 break;
2857         default:
2858                 break;
2859
2860         }
2861
2862         handled = false;
2863         if (dev->extern_ops.pre_msg_handle) {
2864                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2865                                 (void *)&msg);
2866                 switch (ret) {
2867                 case RTE_VHOST_MSG_RESULT_REPLY:
2868                         send_vhost_reply(fd, &msg);
2869                         /* Fall-through */
2870                 case RTE_VHOST_MSG_RESULT_ERR:
2871                 case RTE_VHOST_MSG_RESULT_OK:
2872                         handled = true;
2873                         goto skip_to_post_handle;
2874                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2875                 default:
2876                         break;
2877                 }
2878         }
2879
2880         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2881                 if (!vhost_message_handlers[request])
2882                         goto skip_to_post_handle;
2883                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2884
2885                 switch (ret) {
2886                 case RTE_VHOST_MSG_RESULT_ERR:
2887                         VHOST_LOG_CONFIG(ERR,
2888                                 "Processing %s failed.\n",
2889                                 vhost_message_str[request]);
2890                         handled = true;
2891                         break;
2892                 case RTE_VHOST_MSG_RESULT_OK:
2893                         VHOST_LOG_CONFIG(DEBUG,
2894                                 "Processing %s succeeded.\n",
2895                                 vhost_message_str[request]);
2896                         handled = true;
2897                         break;
2898                 case RTE_VHOST_MSG_RESULT_REPLY:
2899                         VHOST_LOG_CONFIG(DEBUG,
2900                                 "Processing %s succeeded and needs reply.\n",
2901                                 vhost_message_str[request]);
2902                         send_vhost_reply(fd, &msg);
2903                         handled = true;
2904                         break;
2905                 default:
2906                         break;
2907                 }
2908         }
2909
2910 skip_to_post_handle:
2911         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2912                         dev->extern_ops.post_msg_handle) {
2913                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2914                                 (void *)&msg);
2915                 switch (ret) {
2916                 case RTE_VHOST_MSG_RESULT_REPLY:
2917                         send_vhost_reply(fd, &msg);
2918                         /* Fall-through */
2919                 case RTE_VHOST_MSG_RESULT_ERR:
2920                 case RTE_VHOST_MSG_RESULT_OK:
2921                         handled = true;
2922                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2923                 default:
2924                         break;
2925                 }
2926         }
2927
2928         if (unlock_required)
2929                 vhost_user_unlock_all_queue_pairs(dev);
2930
2931         /* If message was not handled at this stage, treat it as an error */
2932         if (!handled) {
2933                 VHOST_LOG_CONFIG(ERR,
2934                         "vhost message (req: %d) was not handled.\n", request);
2935                 close_msg_fds(&msg);
2936                 ret = RTE_VHOST_MSG_RESULT_ERR;
2937         }
2938
2939         /*
2940          * If the request required a reply that was already sent,
2941          * this optional reply-ack won't be sent as the
2942          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2943          */
2944         if (msg.flags & VHOST_USER_NEED_REPLY) {
2945                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2946                 msg.size = sizeof(msg.payload.u64);
2947                 msg.fd_num = 0;
2948                 send_vhost_reply(fd, &msg);
2949         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2950                 VHOST_LOG_CONFIG(ERR,
2951                         "vhost message handling failed.\n");
2952                 return -1;
2953         }
2954
2955         for (i = 0; i < dev->nr_vring; i++) {
2956                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2957                 bool cur_ready = vq_is_ready(dev, vq);
2958
2959                 if (cur_ready != (vq && vq->ready)) {
2960                         vq->ready = cur_ready;
2961                         vhost_user_notify_queue_state(dev, i, cur_ready);
2962                 }
2963         }
2964
2965
2966         if (!virtio_is_ready(dev))
2967                 goto out;
2968
2969         /*
2970          * Virtio is now ready. If not done already, it is time
2971          * to notify the application it can process the rings and
2972          * configure the vDPA device if present.
2973          */
2974
2975         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2976                 if (dev->notify_ops->new_device(dev->vid) == 0)
2977                         dev->flags |= VIRTIO_DEV_RUNNING;
2978         }
2979
2980         vdpa_dev = dev->vdpa_dev;
2981         if (!vdpa_dev)
2982                 goto out;
2983
2984         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2985                 if (vdpa_dev->ops->dev_conf(dev->vid))
2986                         VHOST_LOG_CONFIG(ERR,
2987                                          "Failed to configure vDPA device\n");
2988                 else
2989                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2990         }
2991
2992 out:
2993         return 0;
2994 }
2995
2996 static int process_slave_message_reply(struct virtio_net *dev,
2997                                        const struct VhostUserMsg *msg)
2998 {
2999         struct VhostUserMsg msg_reply;
3000         int ret;
3001
3002         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
3003                 return 0;
3004
3005         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
3006         if (ret <= 0) {
3007                 if (ret < 0)
3008                         VHOST_LOG_CONFIG(ERR,
3009                                 "vhost read slave message reply failed\n");
3010                 else
3011                         VHOST_LOG_CONFIG(INFO,
3012                                 "vhost peer closed\n");
3013                 ret = -1;
3014                 goto out;
3015         }
3016
3017         ret = 0;
3018         if (msg_reply.request.slave != msg->request.slave) {
3019                 VHOST_LOG_CONFIG(ERR,
3020                         "Received unexpected msg type (%u), expected %u\n",
3021                         msg_reply.request.slave, msg->request.slave);
3022                 ret = -1;
3023                 goto out;
3024         }
3025
3026         ret = msg_reply.payload.u64 ? -1 : 0;
3027
3028 out:
3029         rte_spinlock_unlock(&dev->slave_req_lock);
3030         return ret;
3031 }
3032
3033 int
3034 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3035 {
3036         int ret;
3037         struct VhostUserMsg msg = {
3038                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3039                 .flags = VHOST_USER_VERSION,
3040                 .size = sizeof(msg.payload.iotlb),
3041                 .payload.iotlb = {
3042                         .iova = iova,
3043                         .perm = perm,
3044                         .type = VHOST_IOTLB_MISS,
3045                 },
3046         };
3047
3048         ret = send_vhost_message(dev->slave_req_fd, &msg);
3049         if (ret < 0) {
3050                 VHOST_LOG_CONFIG(ERR,
3051                                 "Failed to send IOTLB miss message (%d)\n",
3052                                 ret);
3053                 return ret;
3054         }
3055
3056         return 0;
3057 }
3058
3059 static int
3060 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3061 {
3062         int ret;
3063         struct VhostUserMsg msg = {
3064                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3065                 .flags = VHOST_USER_VERSION,
3066                 .size = 0,
3067         };
3068
3069         if (need_reply)
3070                 msg.flags |= VHOST_USER_NEED_REPLY;
3071
3072         ret = send_vhost_slave_message(dev, &msg);
3073         if (ret < 0) {
3074                 VHOST_LOG_CONFIG(ERR,
3075                                 "Failed to send config change (%d)\n",
3076                                 ret);
3077                 return ret;
3078         }
3079
3080         return process_slave_message_reply(dev, &msg);
3081 }
3082
3083 int
3084 rte_vhost_slave_config_change(int vid, bool need_reply)
3085 {
3086         struct virtio_net *dev;
3087
3088         dev = get_device(vid);
3089         if (!dev)
3090                 return -ENODEV;
3091
3092         return vhost_user_slave_config_change(dev, need_reply);
3093 }
3094
3095 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3096                                                     int index, int fd,
3097                                                     uint64_t offset,
3098                                                     uint64_t size)
3099 {
3100         int ret;
3101         struct VhostUserMsg msg = {
3102                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3103                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3104                 .size = sizeof(msg.payload.area),
3105                 .payload.area = {
3106                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3107                         .size = size,
3108                         .offset = offset,
3109                 },
3110         };
3111
3112         if (fd < 0)
3113                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3114         else {
3115                 msg.fds[0] = fd;
3116                 msg.fd_num = 1;
3117         }
3118
3119         ret = send_vhost_slave_message(dev, &msg);
3120         if (ret < 0) {
3121                 VHOST_LOG_CONFIG(ERR,
3122                         "Failed to set host notifier (%d)\n", ret);
3123                 return ret;
3124         }
3125
3126         return process_slave_message_reply(dev, &msg);
3127 }
3128
3129 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3130 {
3131         struct virtio_net *dev;
3132         struct rte_vdpa_device *vdpa_dev;
3133         int vfio_device_fd, ret = 0;
3134         uint64_t offset, size;
3135         unsigned int i, q_start, q_last;
3136
3137         dev = get_device(vid);
3138         if (!dev)
3139                 return -ENODEV;
3140
3141         vdpa_dev = dev->vdpa_dev;
3142         if (vdpa_dev == NULL)
3143                 return -ENODEV;
3144
3145         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3146             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3147             !(dev->protocol_features &
3148                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3149             !(dev->protocol_features &
3150                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3151             !(dev->protocol_features &
3152                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3153                 return -ENOTSUP;
3154
3155         if (qid == RTE_VHOST_QUEUE_ALL) {
3156                 q_start = 0;
3157                 q_last = dev->nr_vring - 1;
3158         } else {
3159                 if (qid >= dev->nr_vring)
3160                         return -EINVAL;
3161                 q_start = qid;
3162                 q_last = qid;
3163         }
3164
3165         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3166         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3167
3168         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3169         if (vfio_device_fd < 0)
3170                 return -ENOTSUP;
3171
3172         if (enable) {
3173                 for (i = q_start; i <= q_last; i++) {
3174                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3175                                         &size) < 0) {
3176                                 ret = -ENOTSUP;
3177                                 goto disable;
3178                         }
3179
3180                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3181                                         vfio_device_fd, offset, size) < 0) {
3182                                 ret = -EFAULT;
3183                                 goto disable;
3184                         }
3185                 }
3186         } else {
3187 disable:
3188                 for (i = q_start; i <= q_last; i++) {
3189                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3190                                         0, 0);
3191                 }
3192         }
3193
3194         return ret;
3195 }