lib/librte_vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 /* Security model
   6  * --------------
   7  * The vhost-user protocol connection is an external interface, so it must be
   8  * robust against invalid inputs.
   9  *
  10  * This is important because the vhost-user master is only one step removed
  11  * from the guest.  Malicious guests that have escaped will then launch further
  12  * attacks from the vhost-user master.
  13  *
  14  * Even in deployments where guests are trusted, a bug in the vhost-user master
  15  * can still cause invalid messages to be sent.  Such messages must not
  16  * compromise the stability of the DPDK application by causing crashes, memory
  17  * corruption, or other problematic behavior.
  18  *
  19  * Do not assume received VhostUserMsg fields contain sensible values!
  20  */
  21
  22 #include <stdint.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <fcntl.h>
  28 #include <sys/ioctl.h>
  29 #include <sys/mman.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/syscall.h>
  33 #include <assert.h>
  34 #ifdef RTE_LIBRTE_VHOST_NUMA
  35 #include <numaif.h>
  36 #endif
  37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
  38 #include <linux/userfaultfd.h>
  39 #endif
  40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
  41 #include <linux/memfd.h>
  42 #define MEMFD_SUPPORTED
  43 #endif
  44
  45 #include <rte_common.h>
  46 #include <rte_malloc.h>
  47 #include <rte_log.h>
  48
  49 #include "iotlb.h"
  50 #include "vhost.h"
  51 #include "vhost_user.h"
  52
  53 #define VIRTIO_MIN_MTU 68
  54 #define VIRTIO_MAX_MTU 65535
  55
  56 #define INFLIGHT_ALIGNMENT      64
  57 #define INFLIGHT_VERSION        0x1
  58
  59 static const char *vhost_message_str[VHOST_USER_MAX] = {
  60         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  61         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  62         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  63         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  64         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  65         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  66         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  67         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  68         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  69         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  70         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  71         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  72         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  73         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  74         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  75         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  76         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  77         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  78         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  79         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  80         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  81         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  82         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  83         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  84         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  85         [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
  86         [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
  87         [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
  88         [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
  89         [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
  90         [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
  91         [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
  92 };
  93
  94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
  95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
  96
  97 static void
  98 close_msg_fds(struct VhostUserMsg *msg)
  99 {
 100         int i;
 101
 102         for (i = 0; i < msg->fd_num; i++) {
 103                 int fd = msg->fds[i];
 104
 105                 if (fd == -1)
 106                         continue;
 107
 108                 msg->fds[i] = -1;
 109                 close(fd);
 110         }
 111 }
 112
 113 /*
 114  * Ensure the expected number of FDs is received,
 115  * close all FDs and return an error if this is not the case.
 116  */
 117 static int
 118 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
 119 {
 120         if (msg->fd_num == expected_fds)
 121                 return 0;
 122
 123         VHOST_LOG_CONFIG(ERR,
 124                 " Expect %d FDs for request %s, received %d\n",
 125                 expected_fds,
 126                 vhost_message_str[msg->request.master],
 127                 msg->fd_num);
 128
 129         close_msg_fds(msg);
 130
 131         return -1;
 132 }
 133
 134 static uint64_t
 135 get_blk_size(int fd)
 136 {
 137         struct stat stat;
 138         int ret;
 139
 140         ret = fstat(fd, &stat);
 141         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 142 }
 143
 144 static void
 145 free_mem_region(struct virtio_net *dev)
 146 {
 147         uint32_t i;
 148         struct rte_vhost_mem_region *reg;
 149
 150         if (!dev || !dev->mem)
 151                 return;
 152
 153         for (i = 0; i < dev->mem->nregions; i++) {
 154                 reg = &dev->mem->regions[i];
 155                 if (reg->host_user_addr) {
 156                         munmap(reg->mmap_addr, reg->mmap_size);
 157                         close(reg->fd);
 158                 }
 159         }
 160 }
 161
 162 void
 163 vhost_backend_cleanup(struct virtio_net *dev)
 164 {
 165         if (dev->mem) {
 166                 free_mem_region(dev);
 167                 rte_free(dev->mem);
 168                 dev->mem = NULL;
 169         }
 170
 171         rte_free(dev->guest_pages);
 172         dev->guest_pages = NULL;
 173
 174         if (dev->log_addr) {
 175                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 176                 dev->log_addr = 0;
 177         }
 178
 179         if (dev->inflight_info) {
 180                 if (dev->inflight_info->addr) {
 181                         munmap(dev->inflight_info->addr,
 182                                dev->inflight_info->size);
 183                         dev->inflight_info->addr = NULL;
 184                 }
 185
 186                 if (dev->inflight_info->fd >= 0) {
 187                         close(dev->inflight_info->fd);
 188                         dev->inflight_info->fd = -1;
 189                 }
 190
 191                 free(dev->inflight_info);
 192                 dev->inflight_info = NULL;
 193         }
 194
 195         if (dev->slave_req_fd >= 0) {
 196                 close(dev->slave_req_fd);
 197                 dev->slave_req_fd = -1;
 198         }
 199
 200         if (dev->postcopy_ufd >= 0) {
 201                 close(dev->postcopy_ufd);
 202                 dev->postcopy_ufd = -1;
 203         }
 204
 205         dev->postcopy_listening = 0;
 206 }
 207
 208 static void
 209 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
 210                               int enable)
 211 {
 212         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
 213         struct vhost_virtqueue *vq = dev->virtqueue[index];
 214
 215         /* Configure guest notifications on enable */
 216         if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
 217                 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
 218
 219         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
 220                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
 221
 222         if (dev->notify_ops->vring_state_changed)
 223                 dev->notify_ops->vring_state_changed(dev->vid,
 224                                 index, enable);
 225 }
 226
 227 /*
 228  * This function just returns success at the moment unless
 229  * the device hasn't been initialised.
 230  */
 231 static int
 232 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
 233                         struct VhostUserMsg *msg,
 234                         int main_fd __rte_unused)
 235 {
 236         if (validate_msg_fds(msg, 0) != 0)
 237                 return RTE_VHOST_MSG_RESULT_ERR;
 238
 239         return RTE_VHOST_MSG_RESULT_OK;
 240 }
 241
 242 static int
 243 vhost_user_reset_owner(struct virtio_net **pdev,
 244                         struct VhostUserMsg *msg,
 245                         int main_fd __rte_unused)
 246 {
 247         struct virtio_net *dev = *pdev;
 248
 249         if (validate_msg_fds(msg, 0) != 0)
 250                 return RTE_VHOST_MSG_RESULT_ERR;
 251
 252         vhost_destroy_device_notify(dev);
 253
 254         cleanup_device(dev, 0);
 255         reset_device(dev);
 256         return RTE_VHOST_MSG_RESULT_OK;
 257 }
 258
 259 /*
 260  * The features that we support are requested.
 261  */
 262 static int
 263 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 264                         int main_fd __rte_unused)
 265 {
 266         struct virtio_net *dev = *pdev;
 267         uint64_t features = 0;
 268
 269         if (validate_msg_fds(msg, 0) != 0)
 270                 return RTE_VHOST_MSG_RESULT_ERR;
 271
 272         rte_vhost_driver_get_features(dev->ifname, &features);
 273
 274         msg->payload.u64 = features;
 275         msg->size = sizeof(msg->payload.u64);
 276         msg->fd_num = 0;
 277
 278         return RTE_VHOST_MSG_RESULT_REPLY;
 279 }
 280
 281 /*
 282  * The queue number that we support are requested.
 283  */
 284 static int
 285 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
 286                         int main_fd __rte_unused)
 287 {
 288         struct virtio_net *dev = *pdev;
 289         uint32_t queue_num = 0;
 290
 291         if (validate_msg_fds(msg, 0) != 0)
 292                 return RTE_VHOST_MSG_RESULT_ERR;
 293
 294         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 295
 296         msg->payload.u64 = (uint64_t)queue_num;
 297         msg->size = sizeof(msg->payload.u64);
 298         msg->fd_num = 0;
 299
 300         return RTE_VHOST_MSG_RESULT_REPLY;
 301 }
 302
 303 /*
 304  * We receive the negotiated features supported by us and the virtio device.
 305  */
 306 static int
 307 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 308                         int main_fd __rte_unused)
 309 {
 310         struct virtio_net *dev = *pdev;
 311         uint64_t features = msg->payload.u64;
 312         uint64_t vhost_features = 0;
 313         struct rte_vdpa_device *vdpa_dev;
 314
 315         if (validate_msg_fds(msg, 0) != 0)
 316                 return RTE_VHOST_MSG_RESULT_ERR;
 317
 318         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 319         if (features & ~vhost_features) {
 320                 VHOST_LOG_CONFIG(ERR,
 321                         "(%d) received invalid negotiated features.\n",
 322                         dev->vid);
 323                 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
 324                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
 325
 326                 return RTE_VHOST_MSG_RESULT_ERR;
 327         }
 328
 329         if (dev->flags & VIRTIO_DEV_RUNNING) {
 330                 if (dev->features == features)
 331                         return RTE_VHOST_MSG_RESULT_OK;
 332
 333                 /*
 334                  * Error out if master tries to change features while device is
 335                  * in running state. The exception being VHOST_F_LOG_ALL, which
 336                  * is enabled when the live-migration starts.
 337                  */
 338                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 339                         VHOST_LOG_CONFIG(ERR,
 340                                 "(%d) features changed while device is running.\n",
 341                                 dev->vid);
 342                         return RTE_VHOST_MSG_RESULT_ERR;
 343                 }
 344
 345                 if (dev->notify_ops->features_changed)
 346                         dev->notify_ops->features_changed(dev->vid, features);
 347         }
 348
 349         dev->features = features;
 350         if (dev->features &
 351                 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
 352                  (1ULL << VIRTIO_F_VERSION_1) |
 353                  (1ULL << VIRTIO_F_RING_PACKED))) {
 354                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 355         } else {
 356                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 357         }
 358         VHOST_LOG_CONFIG(INFO,
 359                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 360         VHOST_LOG_CONFIG(DEBUG,
 361                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 362                 dev->vid,
 363                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 364                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 365
 366         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 367             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 368                 /*
 369                  * Remove all but first queue pair if MQ hasn't been
 370                  * negotiated. This is safe because the device is not
 371                  * running at this stage.
 372                  */
 373                 while (dev->nr_vring > 2) {
 374                         struct vhost_virtqueue *vq;
 375
 376                         vq = dev->virtqueue[--dev->nr_vring];
 377                         if (!vq)
 378                                 continue;
 379
 380                         dev->virtqueue[dev->nr_vring] = NULL;
 381                         cleanup_vq(vq, 1);
 382                         cleanup_vq_inflight(dev, vq);
 383                         free_vq(dev, vq);
 384                 }
 385         }
 386
 387         vdpa_dev = dev->vdpa_dev;
 388         if (vdpa_dev)
 389                 vdpa_dev->ops->set_features(dev->vid);
 390
 391         dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
 392         return RTE_VHOST_MSG_RESULT_OK;
 393 }
 394
 395 /*
 396  * The virtio device sends us the size of the descriptor ring.
 397  */
 398 static int
 399 vhost_user_set_vring_num(struct virtio_net **pdev,
 400                         struct VhostUserMsg *msg,
 401                         int main_fd __rte_unused)
 402 {
 403         struct virtio_net *dev = *pdev;
 404         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 405
 406         if (validate_msg_fds(msg, 0) != 0)
 407                 return RTE_VHOST_MSG_RESULT_ERR;
 408
 409         vq->size = msg->payload.state.num;
 410
 411         /* VIRTIO 1.0, 2.4 Virtqueues says:
 412          *
 413          *   Queue Size value is always a power of 2. The maximum Queue Size
 414          *   value is 32768.
 415          *
 416          * VIRTIO 1.1 2.7 Virtqueues says:
 417          *
 418          *   Packed virtqueues support up to 2^15 entries each.
 419          */
 420         if (!vq_is_packed(dev)) {
 421                 if (vq->size & (vq->size - 1)) {
 422                         VHOST_LOG_CONFIG(ERR,
 423                                 "invalid virtqueue size %u\n", vq->size);
 424                         return RTE_VHOST_MSG_RESULT_ERR;
 425                 }
 426         }
 427
 428         if (vq->size > 32768) {
 429                 VHOST_LOG_CONFIG(ERR,
 430                         "invalid virtqueue size %u\n", vq->size);
 431                 return RTE_VHOST_MSG_RESULT_ERR;
 432         }
 433
 434         if (vq_is_packed(dev)) {
 435                 if (vq->shadow_used_packed)
 436                         rte_free(vq->shadow_used_packed);
 437                 vq->shadow_used_packed = rte_malloc(NULL,
 438                                 vq->size *
 439                                 sizeof(struct vring_used_elem_packed),
 440                                 RTE_CACHE_LINE_SIZE);
 441                 if (!vq->shadow_used_packed) {
 442                         VHOST_LOG_CONFIG(ERR,
 443                                         "failed to allocate memory for shadow used ring.\n");
 444                         return RTE_VHOST_MSG_RESULT_ERR;
 445                 }
 446
 447         } else {
 448                 if (vq->shadow_used_split)
 449                         rte_free(vq->shadow_used_split);
 450
 451                 vq->shadow_used_split = rte_malloc(NULL,
 452                                 vq->size * sizeof(struct vring_used_elem),
 453                                 RTE_CACHE_LINE_SIZE);
 454
 455                 if (!vq->shadow_used_split) {
 456                         VHOST_LOG_CONFIG(ERR,
 457                                         "failed to allocate memory for vq internal data.\n");
 458                         return RTE_VHOST_MSG_RESULT_ERR;
 459                 }
 460         }
 461
 462         if (vq->batch_copy_elems)
 463                 rte_free(vq->batch_copy_elems);
 464         vq->batch_copy_elems = rte_malloc(NULL,
 465                                 vq->size * sizeof(struct batch_copy_elem),
 466                                 RTE_CACHE_LINE_SIZE);
 467         if (!vq->batch_copy_elems) {
 468                 VHOST_LOG_CONFIG(ERR,
 469                         "failed to allocate memory for batching copy.\n");
 470                 return RTE_VHOST_MSG_RESULT_ERR;
 471         }
 472
 473         return RTE_VHOST_MSG_RESULT_OK;
 474 }
 475
 476 /*
 477  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 478  * same numa node as the memory of vring descriptor.
 479  */
 480 #ifdef RTE_LIBRTE_VHOST_NUMA
 481 static struct virtio_net*
 482 numa_realloc(struct virtio_net *dev, int index)
 483 {
 484         int oldnode, newnode;
 485         struct virtio_net *old_dev;
 486         struct vhost_virtqueue *old_vq, *vq;
 487         struct vring_used_elem *new_shadow_used_split;
 488         struct vring_used_elem_packed *new_shadow_used_packed;
 489         struct batch_copy_elem *new_batch_copy_elems;
 490         int ret;
 491
 492         if (dev->flags & VIRTIO_DEV_RUNNING)
 493                 return dev;
 494
 495         old_dev = dev;
 496         vq = old_vq = dev->virtqueue[index];
 497
 498         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 499                             MPOL_F_NODE | MPOL_F_ADDR);
 500
 501         /* check if we need to reallocate vq */
 502         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 503                              MPOL_F_NODE | MPOL_F_ADDR);
 504         if (ret) {
 505                 VHOST_LOG_CONFIG(ERR,
 506                         "Unable to get vq numa information.\n");
 507                 return dev;
 508         }
 509         if (oldnode != newnode) {
 510                 VHOST_LOG_CONFIG(INFO,
 511                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 512                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 513                 if (!vq)
 514                         return dev;
 515
 516                 memcpy(vq, old_vq, sizeof(*vq));
 517
 518                 if (vq_is_packed(dev)) {
 519                         new_shadow_used_packed = rte_malloc_socket(NULL,
 520                                         vq->size *
 521                                         sizeof(struct vring_used_elem_packed),
 522                                         RTE_CACHE_LINE_SIZE,
 523                                         newnode);
 524                         if (new_shadow_used_packed) {
 525                                 rte_free(vq->shadow_used_packed);
 526                                 vq->shadow_used_packed = new_shadow_used_packed;
 527                         }
 528                 } else {
 529                         new_shadow_used_split = rte_malloc_socket(NULL,
 530                                         vq->size *
 531                                         sizeof(struct vring_used_elem),
 532                                         RTE_CACHE_LINE_SIZE,
 533                                         newnode);
 534                         if (new_shadow_used_split) {
 535                                 rte_free(vq->shadow_used_split);
 536                                 vq->shadow_used_split = new_shadow_used_split;
 537                         }
 538                 }
 539
 540                 new_batch_copy_elems = rte_malloc_socket(NULL,
 541                         vq->size * sizeof(struct batch_copy_elem),
 542                         RTE_CACHE_LINE_SIZE,
 543                         newnode);
 544                 if (new_batch_copy_elems) {
 545                         rte_free(vq->batch_copy_elems);
 546                         vq->batch_copy_elems = new_batch_copy_elems;
 547                 }
 548
 549                 rte_free(old_vq);
 550         }
 551
 552         /* check if we need to reallocate dev */
 553         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 554                             MPOL_F_NODE | MPOL_F_ADDR);
 555         if (ret) {
 556                 VHOST_LOG_CONFIG(ERR,
 557                         "Unable to get dev numa information.\n");
 558                 goto out;
 559         }
 560         if (oldnode != newnode) {
 561                 VHOST_LOG_CONFIG(INFO,
 562                         "reallocate dev from %d to %d node\n",
 563                         oldnode, newnode);
 564                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 565                 if (!dev) {
 566                         dev = old_dev;
 567                         goto out;
 568                 }
 569
 570                 memcpy(dev, old_dev, sizeof(*dev));
 571                 rte_free(old_dev);
 572         }
 573
 574 out:
 575         dev->virtqueue[index] = vq;
 576         vhost_devices[dev->vid] = dev;
 577
 578         if (old_vq != vq)
 579                 vhost_user_iotlb_init(dev, index);
 580
 581         return dev;
 582 }
 583 #else
 584 static struct virtio_net*
 585 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 586 {
 587         return dev;
 588 }
 589 #endif
 590
 591 /* Converts QEMU virtual address to Vhost virtual address. */
 592 static uint64_t
 593 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 594 {
 595         struct rte_vhost_mem_region *r;
 596         uint32_t i;
 597
 598         if (unlikely(!dev || !dev->mem))
 599                 goto out_error;
 600
 601         /* Find the region where the address lives. */
 602         for (i = 0; i < dev->mem->nregions; i++) {
 603                 r = &dev->mem->regions[i];
 604
 605                 if (qva >= r->guest_user_addr &&
 606                     qva <  r->guest_user_addr + r->size) {
 607
 608                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 609                                 *len = r->guest_user_addr + r->size - qva;
 610
 611                         return qva - r->guest_user_addr +
 612                                r->host_user_addr;
 613                 }
 614         }
 615 out_error:
 616         *len = 0;
 617
 618         return 0;
 619 }
 620
 621
 622 /*
 623  * Converts ring address to Vhost virtual address.
 624  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 625  * else it is a QEMU virtual address.
 626  */
 627 static uint64_t
 628 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 629                 uint64_t ra, uint64_t *size)
 630 {
 631         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 632                 uint64_t vva;
 633
 634                 vhost_user_iotlb_rd_lock(vq);
 635                 vva = vhost_iova_to_vva(dev, vq, ra,
 636                                         size, VHOST_ACCESS_RW);
 637                 vhost_user_iotlb_rd_unlock(vq);
 638
 639                 return vva;
 640         }
 641
 642         return qva_to_vva(dev, ra, size);
 643 }
 644
 645 static uint64_t
 646 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 647 {
 648         uint64_t log_gpa;
 649
 650         vhost_user_iotlb_rd_lock(vq);
 651         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 652         vhost_user_iotlb_rd_unlock(vq);
 653
 654         return log_gpa;
 655 }
 656
 657 static struct virtio_net *
 658 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 659 {
 660         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 661         struct vhost_vring_addr *addr = &vq->ring_addrs;
 662         uint64_t len, expected_len;
 663
 664         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 665                 vq->log_guest_addr =
 666                         log_addr_to_gpa(dev, vq);
 667                 if (vq->log_guest_addr == 0) {
 668                         VHOST_LOG_CONFIG(DEBUG,
 669                                 "(%d) failed to map log_guest_addr.\n",
 670                                 dev->vid);
 671                         return dev;
 672                 }
 673         }
 674
 675         if (vq_is_packed(dev)) {
 676                 len = sizeof(struct vring_packed_desc) * vq->size;
 677                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 678                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 679                 if (vq->desc_packed == NULL ||
 680                                 len != sizeof(struct vring_packed_desc) *
 681                                 vq->size) {
 682                         VHOST_LOG_CONFIG(DEBUG,
 683                                 "(%d) failed to map desc_packed ring.\n",
 684                                 dev->vid);
 685                         return dev;
 686                 }
 687
 688                 dev = numa_realloc(dev, vq_index);
 689                 vq = dev->virtqueue[vq_index];
 690                 addr = &vq->ring_addrs;
 691
 692                 len = sizeof(struct vring_packed_desc_event);
 693                 vq->driver_event = (struct vring_packed_desc_event *)
 694                                         (uintptr_t)ring_addr_to_vva(dev,
 695                                         vq, addr->avail_user_addr, &len);
 696                 if (vq->driver_event == NULL ||
 697                                 len != sizeof(struct vring_packed_desc_event)) {
 698                         VHOST_LOG_CONFIG(DEBUG,
 699                                 "(%d) failed to find driver area address.\n",
 700                                 dev->vid);
 701                         return dev;
 702                 }
 703
 704                 len = sizeof(struct vring_packed_desc_event);
 705                 vq->device_event = (struct vring_packed_desc_event *)
 706                                         (uintptr_t)ring_addr_to_vva(dev,
 707                                         vq, addr->used_user_addr, &len);
 708                 if (vq->device_event == NULL ||
 709                                 len != sizeof(struct vring_packed_desc_event)) {
 710                         VHOST_LOG_CONFIG(DEBUG,
 711                                 "(%d) failed to find device area address.\n",
 712                                 dev->vid);
 713                         return dev;
 714                 }
 715
 716                 vq->access_ok = 1;
 717                 return dev;
 718         }
 719
 720         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 721         if (vq->desc && vq->avail && vq->used)
 722                 return dev;
 723
 724         len = sizeof(struct vring_desc) * vq->size;
 725         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 726                         vq, addr->desc_user_addr, &len);
 727         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 728                 VHOST_LOG_CONFIG(DEBUG,
 729                         "(%d) failed to map desc ring.\n",
 730                         dev->vid);
 731                 return dev;
 732         }
 733
 734         dev = numa_realloc(dev, vq_index);
 735         vq = dev->virtqueue[vq_index];
 736         addr = &vq->ring_addrs;
 737
 738         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 739         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 740                 len += sizeof(uint16_t);
 741         expected_len = len;
 742         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 743                         vq, addr->avail_user_addr, &len);
 744         if (vq->avail == 0 || len != expected_len) {
 745                 VHOST_LOG_CONFIG(DEBUG,
 746                         "(%d) failed to map avail ring.\n",
 747                         dev->vid);
 748                 return dev;
 749         }
 750
 751         len = sizeof(struct vring_used) +
 752                 sizeof(struct vring_used_elem) * vq->size;
 753         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 754                 len += sizeof(uint16_t);
 755         expected_len = len;
 756         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 757                         vq, addr->used_user_addr, &len);
 758         if (vq->used == 0 || len != expected_len) {
 759                 VHOST_LOG_CONFIG(DEBUG,
 760                         "(%d) failed to map used ring.\n",
 761                         dev->vid);
 762                 return dev;
 763         }
 764
 765         if (vq->last_used_idx != vq->used->idx) {
 766                 VHOST_LOG_CONFIG(WARNING,
 767                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 768                         "some packets maybe resent for Tx and dropped for Rx\n",
 769                         vq->last_used_idx, vq->used->idx);
 770                 vq->last_used_idx  = vq->used->idx;
 771                 vq->last_avail_idx = vq->used->idx;
 772         }
 773
 774         vq->access_ok = 1;
 775
 776         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 777                         dev->vid, vq->desc);
 778         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 779                         dev->vid, vq->avail);
 780         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 781                         dev->vid, vq->used);
 782         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 783                         dev->vid, vq->log_guest_addr);
 784
 785         return dev;
 786 }
 787
 788 /*
 789  * The virtio device sends us the desc, used and avail ring addresses.
 790  * This function then converts these to our address space.
 791  */
 792 static int
 793 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 794                         int main_fd __rte_unused)
 795 {
 796         struct virtio_net *dev = *pdev;
 797         struct vhost_virtqueue *vq;
 798         struct vhost_vring_addr *addr = &msg->payload.addr;
 799         bool access_ok;
 800
 801         if (validate_msg_fds(msg, 0) != 0)
 802                 return RTE_VHOST_MSG_RESULT_ERR;
 803
 804         if (dev->mem == NULL)
 805                 return RTE_VHOST_MSG_RESULT_ERR;
 806
 807         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 808         vq = dev->virtqueue[msg->payload.addr.index];
 809
 810         access_ok = vq->access_ok;
 811
 812         /*
 813          * Rings addresses should not be interpreted as long as the ring is not
 814          * started and enabled
 815          */
 816         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 817
 818         vring_invalidate(dev, vq);
 819
 820         if ((vq->enabled && (dev->features &
 821                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 822                         access_ok) {
 823                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 824                 if (!dev)
 825                         return RTE_VHOST_MSG_RESULT_ERR;
 826
 827                 *pdev = dev;
 828         }
 829
 830         return RTE_VHOST_MSG_RESULT_OK;
 831 }
 832
 833 /*
 834  * The virtio device sends us the available ring last used index.
 835  */
 836 static int
 837 vhost_user_set_vring_base(struct virtio_net **pdev,
 838                         struct VhostUserMsg *msg,
 839                         int main_fd __rte_unused)
 840 {
 841         struct virtio_net *dev = *pdev;
 842         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 843         uint64_t val = msg->payload.state.num;
 844
 845         if (validate_msg_fds(msg, 0) != 0)
 846                 return RTE_VHOST_MSG_RESULT_ERR;
 847
 848         if (vq_is_packed(dev)) {
 849                 /*
 850                  * Bit[0:14]: avail index
 851                  * Bit[15]: avail wrap counter
 852                  */
 853                 vq->last_avail_idx = val & 0x7fff;
 854                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 855                 /*
 856                  * Set used index to same value as available one, as
 857                  * their values should be the same since ring processing
 858                  * was stopped at get time.
 859                  */
 860                 vq->last_used_idx = vq->last_avail_idx;
 861                 vq->used_wrap_counter = vq->avail_wrap_counter;
 862         } else {
 863                 vq->last_used_idx = msg->payload.state.num;
 864                 vq->last_avail_idx = msg->payload.state.num;
 865         }
 866
 867         return RTE_VHOST_MSG_RESULT_OK;
 868 }
 869
 870 static int
 871 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 872                    uint64_t host_phys_addr, uint64_t size)
 873 {
 874         struct guest_page *page, *last_page;
 875         struct guest_page *old_pages;
 876
 877         if (dev->nr_guest_pages == dev->max_guest_pages) {
 878                 dev->max_guest_pages *= 2;
 879                 old_pages = dev->guest_pages;
 880                 dev->guest_pages = rte_realloc(dev->guest_pages,
 881                                         dev->max_guest_pages * sizeof(*page),
 882                                         RTE_CACHE_LINE_SIZE);
 883                 if (dev->guest_pages == NULL) {
 884                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 885                         rte_free(old_pages);
 886                         return -1;
 887                 }
 888         }
 889
 890         if (dev->nr_guest_pages > 0) {
 891                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 892                 /* merge if the two pages are continuous */
 893                 if (host_phys_addr == last_page->host_phys_addr +
 894                                       last_page->size) {
 895                         last_page->size += size;
 896                         return 0;
 897                 }
 898         }
 899
 900         page = &dev->guest_pages[dev->nr_guest_pages++];
 901         page->guest_phys_addr = guest_phys_addr;
 902         page->host_phys_addr  = host_phys_addr;
 903         page->size = size;
 904
 905         return 0;
 906 }
 907
 908 static int
 909 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 910                 uint64_t page_size)
 911 {
 912         uint64_t reg_size = reg->size;
 913         uint64_t host_user_addr  = reg->host_user_addr;
 914         uint64_t guest_phys_addr = reg->guest_phys_addr;
 915         uint64_t host_phys_addr;
 916         uint64_t size;
 917
 918         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 919         size = page_size - (guest_phys_addr & (page_size - 1));
 920         size = RTE_MIN(size, reg_size);
 921
 922         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 923                 return -1;
 924
 925         host_user_addr  += size;
 926         guest_phys_addr += size;
 927         reg_size -= size;
 928
 929         while (reg_size > 0) {
 930                 size = RTE_MIN(reg_size, page_size);
 931                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 932                                                   host_user_addr);
 933                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 934                                 size) < 0)
 935                         return -1;
 936
 937                 host_user_addr  += size;
 938                 guest_phys_addr += size;
 939                 reg_size -= size;
 940         }
 941
 942         /* sort guest page array if over binary search threshold */
 943         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 944                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 945                         sizeof(struct guest_page), guest_page_addrcmp);
 946         }
 947
 948         return 0;
 949 }
 950
 951 #ifdef RTE_LIBRTE_VHOST_DEBUG
 952 /* TODO: enable it only in debug mode? */
 953 static void
 954 dump_guest_pages(struct virtio_net *dev)
 955 {
 956         uint32_t i;
 957         struct guest_page *page;
 958
 959         for (i = 0; i < dev->nr_guest_pages; i++) {
 960                 page = &dev->guest_pages[i];
 961
 962                 VHOST_LOG_CONFIG(INFO,
 963                         "guest physical page region %u\n"
 964                         "\t guest_phys_addr: %" PRIx64 "\n"
 965                         "\t host_phys_addr : %" PRIx64 "\n"
 966                         "\t size           : %" PRIx64 "\n",
 967                         i,
 968                         page->guest_phys_addr,
 969                         page->host_phys_addr,
 970                         page->size);
 971         }
 972 }
 973 #else
 974 #define dump_guest_pages(dev)
 975 #endif
 976
 977 static bool
 978 vhost_memory_changed(struct VhostUserMemory *new,
 979                      struct rte_vhost_memory *old)
 980 {
 981         uint32_t i;
 982
 983         if (new->nregions != old->nregions)
 984                 return true;
 985
 986         for (i = 0; i < new->nregions; ++i) {
 987                 VhostUserMemoryRegion *new_r = &new->regions[i];
 988                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 989
 990                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 991                         return true;
 992                 if (new_r->memory_size != old_r->size)
 993                         return true;
 994                 if (new_r->userspace_addr != old_r->guest_user_addr)
 995                         return true;
 996         }
 997
 998         return false;
 999 }
1000
1001 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1002 static int
1003 vhost_user_postcopy_region_register(struct virtio_net *dev,
1004                 struct rte_vhost_mem_region *reg)
1005 {
1006         struct uffdio_register reg_struct;
1007
1008         /*
1009          * Let's register all the mmap'ed area to ensure
1010          * alignment on page boundary.
1011          */
1012         reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1013         reg_struct.range.len = reg->mmap_size;
1014         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1015
1016         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1017                                 &reg_struct)) {
1018                 VHOST_LOG_CONFIG(ERR, "Failed to register ufd for region "
1019                                 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1020                                 (uint64_t)reg_struct.range.start,
1021                                 (uint64_t)reg_struct.range.start +
1022                                 (uint64_t)reg_struct.range.len - 1,
1023                                 dev->postcopy_ufd,
1024                                 strerror(errno));
1025                 return -1;
1026         }
1027
1028         VHOST_LOG_CONFIG(INFO, "\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1029                         (uint64_t)reg_struct.range.start,
1030                         (uint64_t)reg_struct.range.start +
1031                         (uint64_t)reg_struct.range.len - 1);
1032
1033         return 0;
1034 }
1035 #else
1036 static int
1037 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1038                 struct rte_vhost_mem_region *reg __rte_unused)
1039 {
1040         return -1;
1041 }
1042 #endif
1043
1044 static int
1045 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1046                 struct VhostUserMsg *msg)
1047 {
1048         struct VhostUserMemory *memory;
1049         struct rte_vhost_mem_region *reg;
1050         VhostUserMsg ack_msg;
1051         uint32_t i;
1052
1053         if (!dev->postcopy_listening)
1054                 return 0;
1055
1056         /*
1057          * We haven't a better way right now than sharing
1058          * DPDK's virtual address with Qemu, so that Qemu can
1059          * retrieve the region offset when handling userfaults.
1060          */
1061         memory = &msg->payload.memory;
1062         for (i = 0; i < memory->nregions; i++) {
1063                 reg = &dev->mem->regions[i];
1064                 memory->regions[i].userspace_addr = reg->host_user_addr;
1065         }
1066
1067         /* Send the addresses back to qemu */
1068         msg->fd_num = 0;
1069         send_vhost_reply(main_fd, msg);
1070
1071         /* Wait for qemu to acknolwedge it's got the addresses
1072          * we've got to wait before we're allowed to generate faults.
1073          */
1074         if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1075                 VHOST_LOG_CONFIG(ERR,
1076                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1077                 return -1;
1078         }
1079
1080         if (validate_msg_fds(&ack_msg, 0) != 0)
1081                 return -1;
1082
1083         if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1084                 VHOST_LOG_CONFIG(ERR,
1085                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1086                                 ack_msg.request.master);
1087                 return -1;
1088         }
1089
1090         /* Now userfault register and we can use the memory */
1091         for (i = 0; i < memory->nregions; i++) {
1092                 reg = &dev->mem->regions[i];
1093                 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1094                         return -1;
1095         }
1096
1097         return 0;
1098 }
1099
1100 static int
1101 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1102                         int main_fd)
1103 {
1104         struct virtio_net *dev = *pdev;
1105         struct VhostUserMemory *memory = &msg->payload.memory;
1106         struct rte_vhost_mem_region *reg;
1107         void *mmap_addr;
1108         uint64_t mmap_size;
1109         uint64_t mmap_offset;
1110         uint64_t alignment;
1111         uint32_t i;
1112         int populate;
1113
1114         if (validate_msg_fds(msg, memory->nregions) != 0)
1115                 return RTE_VHOST_MSG_RESULT_ERR;
1116
1117         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1118                 VHOST_LOG_CONFIG(ERR,
1119                         "too many memory regions (%u)\n", memory->nregions);
1120                 goto close_msg_fds;
1121         }
1122
1123         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1124                 VHOST_LOG_CONFIG(INFO,
1125                         "(%d) memory regions not changed\n", dev->vid);
1126
1127                 close_msg_fds(msg);
1128
1129                 return RTE_VHOST_MSG_RESULT_OK;
1130         }
1131
1132         if (dev->mem) {
1133                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1134                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1135
1136                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1137                                 vdpa_dev->ops->dev_close(dev->vid);
1138                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1139                 }
1140                 free_mem_region(dev);
1141                 rte_free(dev->mem);
1142                 dev->mem = NULL;
1143         }
1144
1145         /* Flush IOTLB cache as previous HVAs are now invalid */
1146         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1147                 for (i = 0; i < dev->nr_vring; i++)
1148                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1149
1150         dev->nr_guest_pages = 0;
1151         if (dev->guest_pages == NULL) {
1152                 dev->max_guest_pages = 8;
1153                 dev->guest_pages = rte_zmalloc(NULL,
1154                                         dev->max_guest_pages *
1155                                         sizeof(struct guest_page),
1156                                         RTE_CACHE_LINE_SIZE);
1157                 if (dev->guest_pages == NULL) {
1158                         VHOST_LOG_CONFIG(ERR,
1159                                 "(%d) failed to allocate memory "
1160                                 "for dev->guest_pages\n",
1161                                 dev->vid);
1162                         goto close_msg_fds;
1163                 }
1164         }
1165
1166         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1167                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1168         if (dev->mem == NULL) {
1169                 VHOST_LOG_CONFIG(ERR,
1170                         "(%d) failed to allocate memory for dev->mem\n",
1171                         dev->vid);
1172                 goto free_guest_pages;
1173         }
1174         dev->mem->nregions = memory->nregions;
1175
1176         for (i = 0; i < memory->nregions; i++) {
1177                 reg = &dev->mem->regions[i];
1178
1179                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1180                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1181                 reg->size            = memory->regions[i].memory_size;
1182                 reg->fd              = msg->fds[i];
1183
1184                 /*
1185                  * Assign invalid file descriptor value to avoid double
1186                  * closing on error path.
1187                  */
1188                 msg->fds[i] = -1;
1189
1190                 mmap_offset = memory->regions[i].mmap_offset;
1191
1192                 /* Check for memory_size + mmap_offset overflow */
1193                 if (mmap_offset >= -reg->size) {
1194                         VHOST_LOG_CONFIG(ERR,
1195                                 "mmap_offset (%#"PRIx64") and memory_size "
1196                                 "(%#"PRIx64") overflow\n",
1197                                 mmap_offset, reg->size);
1198                         goto free_mem_table;
1199                 }
1200
1201                 mmap_size = reg->size + mmap_offset;
1202
1203                 /* mmap() without flag of MAP_ANONYMOUS, should be called
1204                  * with length argument aligned with hugepagesz at older
1205                  * longterm version Linux, like 2.6.32 and 3.2.72, or
1206                  * mmap() will fail with EINVAL.
1207                  *
1208                  * to avoid failure, make sure in caller to keep length
1209                  * aligned.
1210                  */
1211                 alignment = get_blk_size(reg->fd);
1212                 if (alignment == (uint64_t)-1) {
1213                         VHOST_LOG_CONFIG(ERR,
1214                                 "couldn't get hugepage size through fstat\n");
1215                         goto free_mem_table;
1216                 }
1217                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1218                 if (mmap_size == 0) {
1219                         /*
1220                          * It could happen if initial mmap_size + alignment
1221                          * overflows the sizeof uint64, which could happen if
1222                          * either mmap_size or alignment value is wrong.
1223                          *
1224                          * mmap() kernel implementation would return an error,
1225                          * but better catch it before and provide useful info
1226                          * in the logs.
1227                          */
1228                         VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1229                                         "or alignment (0x%" PRIx64 ") is invalid\n",
1230                                         reg->size + mmap_offset, alignment);
1231                         goto free_mem_table;
1232                 }
1233
1234                 populate = dev->async_copy ? MAP_POPULATE : 0;
1235                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1236                                  MAP_SHARED | populate, reg->fd, 0);
1237
1238                 if (mmap_addr == MAP_FAILED) {
1239                         VHOST_LOG_CONFIG(ERR,
1240                                 "mmap region %u failed.\n", i);
1241                         goto free_mem_table;
1242                 }
1243
1244                 reg->mmap_addr = mmap_addr;
1245                 reg->mmap_size = mmap_size;
1246                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1247                                       mmap_offset;
1248
1249                 if (dev->async_copy)
1250                         if (add_guest_pages(dev, reg, alignment) < 0) {
1251                                 VHOST_LOG_CONFIG(ERR,
1252                                         "adding guest pages to region %u failed.\n",
1253                                         i);
1254                                 goto free_mem_table;
1255                         }
1256
1257                 VHOST_LOG_CONFIG(INFO,
1258                         "guest memory region %u, size: 0x%" PRIx64 "\n"
1259                         "\t guest physical addr: 0x%" PRIx64 "\n"
1260                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1261                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1262                         "\t mmap addr : 0x%" PRIx64 "\n"
1263                         "\t mmap size : 0x%" PRIx64 "\n"
1264                         "\t mmap align: 0x%" PRIx64 "\n"
1265                         "\t mmap off  : 0x%" PRIx64 "\n",
1266                         i, reg->size,
1267                         reg->guest_phys_addr,
1268                         reg->guest_user_addr,
1269                         reg->host_user_addr,
1270                         (uint64_t)(uintptr_t)mmap_addr,
1271                         mmap_size,
1272                         alignment,
1273                         mmap_offset);
1274         }
1275
1276         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1277                 goto free_mem_table;
1278
1279         for (i = 0; i < dev->nr_vring; i++) {
1280                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1281
1282                 if (!vq)
1283                         continue;
1284
1285                 if (vq->desc || vq->avail || vq->used) {
1286                         /*
1287                          * If the memory table got updated, the ring addresses
1288                          * need to be translated again as virtual addresses have
1289                          * changed.
1290                          */
1291                         vring_invalidate(dev, vq);
1292
1293                         dev = translate_ring_addresses(dev, i);
1294                         if (!dev) {
1295                                 dev = *pdev;
1296                                 goto free_mem_table;
1297                         }
1298
1299                         *pdev = dev;
1300                 }
1301         }
1302
1303         dump_guest_pages(dev);
1304
1305         return RTE_VHOST_MSG_RESULT_OK;
1306
1307 free_mem_table:
1308         free_mem_region(dev);
1309         rte_free(dev->mem);
1310         dev->mem = NULL;
1311 free_guest_pages:
1312         rte_free(dev->guest_pages);
1313         dev->guest_pages = NULL;
1314 close_msg_fds:
1315         close_msg_fds(msg);
1316         return RTE_VHOST_MSG_RESULT_ERR;
1317 }
1318
1319 static bool
1320 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1321 {
1322         bool rings_ok;
1323
1324         if (!vq)
1325                 return false;
1326
1327         if (vq_is_packed(dev))
1328                 rings_ok = vq->desc_packed && vq->driver_event &&
1329                         vq->device_event;
1330         else
1331                 rings_ok = vq->desc && vq->avail && vq->used;
1332
1333         return rings_ok &&
1334                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1335                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1336                vq->enabled;
1337 }
1338
1339 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1340
1341 static int
1342 virtio_is_ready(struct virtio_net *dev)
1343 {
1344         struct vhost_virtqueue *vq;
1345         uint32_t i, nr_vring = dev->nr_vring;
1346
1347         if (dev->flags & VIRTIO_DEV_READY)
1348                 return 1;
1349
1350         if (!dev->nr_vring)
1351                 return 0;
1352
1353         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1354                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1355
1356                 if (dev->nr_vring < nr_vring)
1357                         return 0;
1358         }
1359
1360         for (i = 0; i < nr_vring; i++) {
1361                 vq = dev->virtqueue[i];
1362
1363                 if (!vq_is_ready(dev, vq))
1364                         return 0;
1365         }
1366
1367         /* If supported, ensure the frontend is really done with config */
1368         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1369                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1370                         return 0;
1371
1372         dev->flags |= VIRTIO_DEV_READY;
1373
1374         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1375                 VHOST_LOG_CONFIG(INFO,
1376                         "virtio is now ready for processing.\n");
1377         return 1;
1378 }
1379
1380 static void *
1381 inflight_mem_alloc(const char *name, size_t size, int *fd)
1382 {
1383         void *ptr;
1384         int mfd = -1;
1385         char fname[20] = "/tmp/memfd-XXXXXX";
1386
1387         *fd = -1;
1388 #ifdef MEMFD_SUPPORTED
1389         mfd = memfd_create(name, MFD_CLOEXEC);
1390 #else
1391         RTE_SET_USED(name);
1392 #endif
1393         if (mfd == -1) {
1394                 mfd = mkstemp(fname);
1395                 if (mfd == -1) {
1396                         VHOST_LOG_CONFIG(ERR,
1397                                 "failed to get inflight buffer fd\n");
1398                         return NULL;
1399                 }
1400
1401                 unlink(fname);
1402         }
1403
1404         if (ftruncate(mfd, size) == -1) {
1405                 VHOST_LOG_CONFIG(ERR,
1406                         "failed to alloc inflight buffer\n");
1407                 close(mfd);
1408                 return NULL;
1409         }
1410
1411         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1412         if (ptr == MAP_FAILED) {
1413                 VHOST_LOG_CONFIG(ERR,
1414                         "failed to mmap inflight buffer\n");
1415                 close(mfd);
1416                 return NULL;
1417         }
1418
1419         *fd = mfd;
1420         return ptr;
1421 }
1422
1423 static uint32_t
1424 get_pervq_shm_size_split(uint16_t queue_size)
1425 {
1426         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1427                                   queue_size + sizeof(uint64_t) +
1428                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1429 }
1430
1431 static uint32_t
1432 get_pervq_shm_size_packed(uint16_t queue_size)
1433 {
1434         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1435                                   * queue_size + sizeof(uint64_t) +
1436                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1437                                   INFLIGHT_ALIGNMENT);
1438 }
1439
1440 static int
1441 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1442                            VhostUserMsg *msg,
1443                            int main_fd __rte_unused)
1444 {
1445         struct rte_vhost_inflight_info_packed *inflight_packed;
1446         uint64_t pervq_inflight_size, mmap_size;
1447         uint16_t num_queues, queue_size;
1448         struct virtio_net *dev = *pdev;
1449         int fd, i, j;
1450         void *addr;
1451
1452         if (msg->size != sizeof(msg->payload.inflight)) {
1453                 VHOST_LOG_CONFIG(ERR,
1454                         "invalid get_inflight_fd message size is %d\n",
1455                         msg->size);
1456                 return RTE_VHOST_MSG_RESULT_ERR;
1457         }
1458
1459         if (dev->inflight_info == NULL) {
1460                 dev->inflight_info = calloc(1,
1461                                             sizeof(struct inflight_mem_info));
1462                 if (!dev->inflight_info) {
1463                         VHOST_LOG_CONFIG(ERR,
1464                                 "failed to alloc dev inflight area\n");
1465                         return RTE_VHOST_MSG_RESULT_ERR;
1466                 }
1467                 dev->inflight_info->fd = -1;
1468         }
1469
1470         num_queues = msg->payload.inflight.num_queues;
1471         queue_size = msg->payload.inflight.queue_size;
1472
1473         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1474                 msg->payload.inflight.num_queues);
1475         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1476                 msg->payload.inflight.queue_size);
1477
1478         if (vq_is_packed(dev))
1479                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1480         else
1481                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1482
1483         mmap_size = num_queues * pervq_inflight_size;
1484         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1485         if (!addr) {
1486                 VHOST_LOG_CONFIG(ERR,
1487                         "failed to alloc vhost inflight area\n");
1488                         msg->payload.inflight.mmap_size = 0;
1489                 return RTE_VHOST_MSG_RESULT_ERR;
1490         }
1491         memset(addr, 0, mmap_size);
1492
1493         if (dev->inflight_info->addr) {
1494                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1495                 dev->inflight_info->addr = NULL;
1496         }
1497
1498         if (dev->inflight_info->fd >= 0) {
1499                 close(dev->inflight_info->fd);
1500                 dev->inflight_info->fd = -1;
1501         }
1502
1503         dev->inflight_info->addr = addr;
1504         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1505         dev->inflight_info->fd = msg->fds[0] = fd;
1506         msg->payload.inflight.mmap_offset = 0;
1507         msg->fd_num = 1;
1508
1509         if (vq_is_packed(dev)) {
1510                 for (i = 0; i < num_queues; i++) {
1511                         inflight_packed =
1512                                 (struct rte_vhost_inflight_info_packed *)addr;
1513                         inflight_packed->used_wrap_counter = 1;
1514                         inflight_packed->old_used_wrap_counter = 1;
1515                         for (j = 0; j < queue_size; j++)
1516                                 inflight_packed->desc[j].next = j + 1;
1517                         addr = (void *)((char *)addr + pervq_inflight_size);
1518                 }
1519         }
1520
1521         VHOST_LOG_CONFIG(INFO,
1522                 "send inflight mmap_size: %"PRIu64"\n",
1523                 msg->payload.inflight.mmap_size);
1524         VHOST_LOG_CONFIG(INFO,
1525                 "send inflight mmap_offset: %"PRIu64"\n",
1526                 msg->payload.inflight.mmap_offset);
1527         VHOST_LOG_CONFIG(INFO,
1528                 "send inflight fd: %d\n", msg->fds[0]);
1529
1530         return RTE_VHOST_MSG_RESULT_REPLY;
1531 }
1532
1533 static int
1534 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1535                            int main_fd __rte_unused)
1536 {
1537         uint64_t mmap_size, mmap_offset;
1538         uint16_t num_queues, queue_size;
1539         struct virtio_net *dev = *pdev;
1540         uint32_t pervq_inflight_size;
1541         struct vhost_virtqueue *vq;
1542         void *addr;
1543         int fd, i;
1544
1545         fd = msg->fds[0];
1546         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1547                 VHOST_LOG_CONFIG(ERR,
1548                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1549                         msg->size, fd);
1550                 return RTE_VHOST_MSG_RESULT_ERR;
1551         }
1552
1553         mmap_size = msg->payload.inflight.mmap_size;
1554         mmap_offset = msg->payload.inflight.mmap_offset;
1555         num_queues = msg->payload.inflight.num_queues;
1556         queue_size = msg->payload.inflight.queue_size;
1557
1558         if (vq_is_packed(dev))
1559                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1560         else
1561                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1562
1563         VHOST_LOG_CONFIG(INFO,
1564                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1565         VHOST_LOG_CONFIG(INFO,
1566                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1567         VHOST_LOG_CONFIG(INFO,
1568                 "set_inflight_fd num_queues: %u\n", num_queues);
1569         VHOST_LOG_CONFIG(INFO,
1570                 "set_inflight_fd queue_size: %u\n", queue_size);
1571         VHOST_LOG_CONFIG(INFO,
1572                 "set_inflight_fd fd: %d\n", fd);
1573         VHOST_LOG_CONFIG(INFO,
1574                 "set_inflight_fd pervq_inflight_size: %d\n",
1575                 pervq_inflight_size);
1576
1577         if (!dev->inflight_info) {
1578                 dev->inflight_info = calloc(1,
1579                                             sizeof(struct inflight_mem_info));
1580                 if (dev->inflight_info == NULL) {
1581                         VHOST_LOG_CONFIG(ERR,
1582                                 "failed to alloc dev inflight area\n");
1583                         return RTE_VHOST_MSG_RESULT_ERR;
1584                 }
1585                 dev->inflight_info->fd = -1;
1586         }
1587
1588         if (dev->inflight_info->addr) {
1589                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1590                 dev->inflight_info->addr = NULL;
1591         }
1592
1593         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1594                     fd, mmap_offset);
1595         if (addr == MAP_FAILED) {
1596                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1597                 return RTE_VHOST_MSG_RESULT_ERR;
1598         }
1599
1600         if (dev->inflight_info->fd >= 0) {
1601                 close(dev->inflight_info->fd);
1602                 dev->inflight_info->fd = -1;
1603         }
1604
1605         dev->inflight_info->fd = fd;
1606         dev->inflight_info->addr = addr;
1607         dev->inflight_info->size = mmap_size;
1608
1609         for (i = 0; i < num_queues; i++) {
1610                 vq = dev->virtqueue[i];
1611                 if (!vq)
1612                         continue;
1613
1614                 if (vq_is_packed(dev)) {
1615                         vq->inflight_packed = addr;
1616                         vq->inflight_packed->desc_num = queue_size;
1617                 } else {
1618                         vq->inflight_split = addr;
1619                         vq->inflight_split->desc_num = queue_size;
1620                 }
1621                 addr = (void *)((char *)addr + pervq_inflight_size);
1622         }
1623
1624         return RTE_VHOST_MSG_RESULT_OK;
1625 }
1626
1627 static int
1628 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1629                         int main_fd __rte_unused)
1630 {
1631         struct virtio_net *dev = *pdev;
1632         struct vhost_vring_file file;
1633         struct vhost_virtqueue *vq;
1634         int expected_fds;
1635
1636         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1637         if (validate_msg_fds(msg, expected_fds) != 0)
1638                 return RTE_VHOST_MSG_RESULT_ERR;
1639
1640         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1641         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1642                 file.fd = VIRTIO_INVALID_EVENTFD;
1643         else
1644                 file.fd = msg->fds[0];
1645         VHOST_LOG_CONFIG(INFO,
1646                 "vring call idx:%d file:%d\n", file.index, file.fd);
1647
1648         vq = dev->virtqueue[file.index];
1649
1650         if (vq->ready) {
1651                 vq->ready = 0;
1652                 vhost_user_notify_queue_state(dev, file.index, 0);
1653         }
1654
1655         if (vq->callfd >= 0)
1656                 close(vq->callfd);
1657
1658         vq->callfd = file.fd;
1659
1660         return RTE_VHOST_MSG_RESULT_OK;
1661 }
1662
1663 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1664                         struct VhostUserMsg *msg,
1665                         int main_fd __rte_unused)
1666 {
1667         int expected_fds;
1668
1669         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1670         if (validate_msg_fds(msg, expected_fds) != 0)
1671                 return RTE_VHOST_MSG_RESULT_ERR;
1672
1673         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1674                 close(msg->fds[0]);
1675         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1676
1677         return RTE_VHOST_MSG_RESULT_OK;
1678 }
1679
1680 static int
1681 resubmit_desc_compare(const void *a, const void *b)
1682 {
1683         const struct rte_vhost_resubmit_desc *desc0 = a;
1684         const struct rte_vhost_resubmit_desc *desc1 = b;
1685
1686         if (desc1->counter > desc0->counter)
1687                 return 1;
1688
1689         return -1;
1690 }
1691
1692 static int
1693 vhost_check_queue_inflights_split(struct virtio_net *dev,
1694                                   struct vhost_virtqueue *vq)
1695 {
1696         uint16_t i;
1697         uint16_t resubmit_num = 0, last_io, num;
1698         struct vring_used *used = vq->used;
1699         struct rte_vhost_resubmit_info *resubmit;
1700         struct rte_vhost_inflight_info_split *inflight_split;
1701
1702         if (!(dev->protocol_features &
1703             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1704                 return RTE_VHOST_MSG_RESULT_OK;
1705
1706         /* The frontend may still not support the inflight feature
1707          * although we negotiate the protocol feature.
1708          */
1709         if ((!vq->inflight_split))
1710                 return RTE_VHOST_MSG_RESULT_OK;
1711
1712         if (!vq->inflight_split->version) {
1713                 vq->inflight_split->version = INFLIGHT_VERSION;
1714                 return RTE_VHOST_MSG_RESULT_OK;
1715         }
1716
1717         if (vq->resubmit_inflight)
1718                 return RTE_VHOST_MSG_RESULT_OK;
1719
1720         inflight_split = vq->inflight_split;
1721         vq->global_counter = 0;
1722         last_io = inflight_split->last_inflight_io;
1723
1724         if (inflight_split->used_idx != used->idx) {
1725                 inflight_split->desc[last_io].inflight = 0;
1726                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1727                 inflight_split->used_idx = used->idx;
1728         }
1729
1730         for (i = 0; i < inflight_split->desc_num; i++) {
1731                 if (inflight_split->desc[i].inflight == 1)
1732                         resubmit_num++;
1733         }
1734
1735         vq->last_avail_idx += resubmit_num;
1736
1737         if (resubmit_num) {
1738                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1739                 if (!resubmit) {
1740                         VHOST_LOG_CONFIG(ERR,
1741                                 "failed to allocate memory for resubmit info.\n");
1742                         return RTE_VHOST_MSG_RESULT_ERR;
1743                 }
1744
1745                 resubmit->resubmit_list = calloc(resubmit_num,
1746                         sizeof(struct rte_vhost_resubmit_desc));
1747                 if (!resubmit->resubmit_list) {
1748                         VHOST_LOG_CONFIG(ERR,
1749                                 "failed to allocate memory for inflight desc.\n");
1750                         free(resubmit);
1751                         return RTE_VHOST_MSG_RESULT_ERR;
1752                 }
1753
1754                 num = 0;
1755                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1756                         if (vq->inflight_split->desc[i].inflight == 1) {
1757                                 resubmit->resubmit_list[num].index = i;
1758                                 resubmit->resubmit_list[num].counter =
1759                                         inflight_split->desc[i].counter;
1760                                 num++;
1761                         }
1762                 }
1763                 resubmit->resubmit_num = num;
1764
1765                 if (resubmit->resubmit_num > 1)
1766                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1767                               sizeof(struct rte_vhost_resubmit_desc),
1768                               resubmit_desc_compare);
1769
1770                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1771                 vq->resubmit_inflight = resubmit;
1772         }
1773
1774         return RTE_VHOST_MSG_RESULT_OK;
1775 }
1776
1777 static int
1778 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1779                                    struct vhost_virtqueue *vq)
1780 {
1781         uint16_t i;
1782         uint16_t resubmit_num = 0, old_used_idx, num;
1783         struct rte_vhost_resubmit_info *resubmit;
1784         struct rte_vhost_inflight_info_packed *inflight_packed;
1785
1786         if (!(dev->protocol_features &
1787             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1788                 return RTE_VHOST_MSG_RESULT_OK;
1789
1790         /* The frontend may still not support the inflight feature
1791          * although we negotiate the protocol feature.
1792          */
1793         if ((!vq->inflight_packed))
1794                 return RTE_VHOST_MSG_RESULT_OK;
1795
1796         if (!vq->inflight_packed->version) {
1797                 vq->inflight_packed->version = INFLIGHT_VERSION;
1798                 return RTE_VHOST_MSG_RESULT_OK;
1799         }
1800
1801         if (vq->resubmit_inflight)
1802                 return RTE_VHOST_MSG_RESULT_OK;
1803
1804         inflight_packed = vq->inflight_packed;
1805         vq->global_counter = 0;
1806         old_used_idx = inflight_packed->old_used_idx;
1807
1808         if (inflight_packed->used_idx != old_used_idx) {
1809                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1810                         inflight_packed->old_used_idx =
1811                                 inflight_packed->used_idx;
1812                         inflight_packed->old_used_wrap_counter =
1813                                 inflight_packed->used_wrap_counter;
1814                         inflight_packed->old_free_head =
1815                                 inflight_packed->free_head;
1816                 } else {
1817                         inflight_packed->used_idx =
1818                                 inflight_packed->old_used_idx;
1819                         inflight_packed->used_wrap_counter =
1820                                 inflight_packed->old_used_wrap_counter;
1821                         inflight_packed->free_head =
1822                                 inflight_packed->old_free_head;
1823                 }
1824         }
1825
1826         for (i = 0; i < inflight_packed->desc_num; i++) {
1827                 if (inflight_packed->desc[i].inflight == 1)
1828                         resubmit_num++;
1829         }
1830
1831         if (resubmit_num) {
1832                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1833                 if (resubmit == NULL) {
1834                         VHOST_LOG_CONFIG(ERR,
1835                                 "failed to allocate memory for resubmit info.\n");
1836                         return RTE_VHOST_MSG_RESULT_ERR;
1837                 }
1838
1839                 resubmit->resubmit_list = calloc(resubmit_num,
1840                         sizeof(struct rte_vhost_resubmit_desc));
1841                 if (resubmit->resubmit_list == NULL) {
1842                         VHOST_LOG_CONFIG(ERR,
1843                                 "failed to allocate memory for resubmit desc.\n");
1844                         free(resubmit);
1845                         return RTE_VHOST_MSG_RESULT_ERR;
1846                 }
1847
1848                 num = 0;
1849                 for (i = 0; i < inflight_packed->desc_num; i++) {
1850                         if (vq->inflight_packed->desc[i].inflight == 1) {
1851                                 resubmit->resubmit_list[num].index = i;
1852                                 resubmit->resubmit_list[num].counter =
1853                                         inflight_packed->desc[i].counter;
1854                                 num++;
1855                         }
1856                 }
1857                 resubmit->resubmit_num = num;
1858
1859                 if (resubmit->resubmit_num > 1)
1860                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1861                               sizeof(struct rte_vhost_resubmit_desc),
1862                               resubmit_desc_compare);
1863
1864                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1865                 vq->resubmit_inflight = resubmit;
1866         }
1867
1868         return RTE_VHOST_MSG_RESULT_OK;
1869 }
1870
1871 static int
1872 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1873                         int main_fd __rte_unused)
1874 {
1875         struct virtio_net *dev = *pdev;
1876         struct vhost_vring_file file;
1877         struct vhost_virtqueue *vq;
1878         int expected_fds;
1879
1880         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1881         if (validate_msg_fds(msg, expected_fds) != 0)
1882                 return RTE_VHOST_MSG_RESULT_ERR;
1883
1884         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1885         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1886                 file.fd = VIRTIO_INVALID_EVENTFD;
1887         else
1888                 file.fd = msg->fds[0];
1889         VHOST_LOG_CONFIG(INFO,
1890                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1891
1892         /* Interpret ring addresses only when ring is started. */
1893         dev = translate_ring_addresses(dev, file.index);
1894         if (!dev) {
1895                 if (file.fd != VIRTIO_INVALID_EVENTFD)
1896                         close(file.fd);
1897
1898                 return RTE_VHOST_MSG_RESULT_ERR;
1899         }
1900
1901         *pdev = dev;
1902
1903         vq = dev->virtqueue[file.index];
1904
1905         /*
1906          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1907          * the ring starts already enabled. Otherwise, it is enabled via
1908          * the SET_VRING_ENABLE message.
1909          */
1910         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1911                 vq->enabled = 1;
1912                 if (dev->notify_ops->vring_state_changed)
1913                         dev->notify_ops->vring_state_changed(
1914                                 dev->vid, file.index, 1);
1915         }
1916
1917         if (vq->ready) {
1918                 vq->ready = 0;
1919                 vhost_user_notify_queue_state(dev, file.index, 0);
1920         }
1921
1922         if (vq->kickfd >= 0)
1923                 close(vq->kickfd);
1924         vq->kickfd = file.fd;
1925
1926         if (vq_is_packed(dev)) {
1927                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1928                         VHOST_LOG_CONFIG(ERR,
1929                                 "failed to inflights for vq: %d\n", file.index);
1930                         return RTE_VHOST_MSG_RESULT_ERR;
1931                 }
1932         } else {
1933                 if (vhost_check_queue_inflights_split(dev, vq)) {
1934                         VHOST_LOG_CONFIG(ERR,
1935                                 "failed to inflights for vq: %d\n", file.index);
1936                         return RTE_VHOST_MSG_RESULT_ERR;
1937                 }
1938         }
1939
1940         return RTE_VHOST_MSG_RESULT_OK;
1941 }
1942
1943 /*
1944  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1945  */
1946 static int
1947 vhost_user_get_vring_base(struct virtio_net **pdev,
1948                         struct VhostUserMsg *msg,
1949                         int main_fd __rte_unused)
1950 {
1951         struct virtio_net *dev = *pdev;
1952         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1953         uint64_t val;
1954
1955         if (validate_msg_fds(msg, 0) != 0)
1956                 return RTE_VHOST_MSG_RESULT_ERR;
1957
1958         /* We have to stop the queue (virtio) if it is running. */
1959         vhost_destroy_device_notify(dev);
1960
1961         dev->flags &= ~VIRTIO_DEV_READY;
1962         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1963
1964         /* Here we are safe to get the indexes */
1965         if (vq_is_packed(dev)) {
1966                 /*
1967                  * Bit[0:14]: avail index
1968                  * Bit[15]: avail wrap counter
1969                  */
1970                 val = vq->last_avail_idx & 0x7fff;
1971                 val |= vq->avail_wrap_counter << 15;
1972                 msg->payload.state.num = val;
1973         } else {
1974                 msg->payload.state.num = vq->last_avail_idx;
1975         }
1976
1977         VHOST_LOG_CONFIG(INFO,
1978                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1979                 msg->payload.state.num);
1980         /*
1981          * Based on current qemu vhost-user implementation, this message is
1982          * sent and only sent in vhost_vring_stop.
1983          * TODO: cleanup the vring, it isn't usable since here.
1984          */
1985         if (vq->kickfd >= 0)
1986                 close(vq->kickfd);
1987
1988         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1989
1990         if (vq->callfd >= 0)
1991                 close(vq->callfd);
1992
1993         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1994
1995         vq->signalled_used_valid = false;
1996
1997         if (vq_is_packed(dev)) {
1998                 rte_free(vq->shadow_used_packed);
1999                 vq->shadow_used_packed = NULL;
2000         } else {
2001                 rte_free(vq->shadow_used_split);
2002                 vq->shadow_used_split = NULL;
2003                 if (vq->async_pkts_pending)
2004                         rte_free(vq->async_pkts_pending);
2005                 if (vq->async_pkts_info)
2006                         rte_free(vq->async_pkts_info);
2007                 vq->async_pkts_pending = NULL;
2008                 vq->async_pkts_info = NULL;
2009         }
2010
2011         rte_free(vq->batch_copy_elems);
2012         vq->batch_copy_elems = NULL;
2013
2014         msg->size = sizeof(msg->payload.state);
2015         msg->fd_num = 0;
2016
2017         vring_invalidate(dev, vq);
2018
2019         return RTE_VHOST_MSG_RESULT_REPLY;
2020 }
2021
2022 /*
2023  * when virtio queues are ready to work, qemu will send us to
2024  * enable the virtio queue pair.
2025  */
2026 static int
2027 vhost_user_set_vring_enable(struct virtio_net **pdev,
2028                         struct VhostUserMsg *msg,
2029                         int main_fd __rte_unused)
2030 {
2031         struct virtio_net *dev = *pdev;
2032         int enable = (int)msg->payload.state.num;
2033         int index = (int)msg->payload.state.index;
2034
2035         if (validate_msg_fds(msg, 0) != 0)
2036                 return RTE_VHOST_MSG_RESULT_ERR;
2037
2038         VHOST_LOG_CONFIG(INFO,
2039                 "set queue enable: %d to qp idx: %d\n",
2040                 enable, index);
2041
2042         if (enable && dev->virtqueue[index]->async_registered) {
2043                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2044                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2045                         "async inflight packets must be completed first\n");
2046                         return RTE_VHOST_MSG_RESULT_ERR;
2047                 }
2048         }
2049
2050         dev->virtqueue[index]->enabled = enable;
2051
2052         return RTE_VHOST_MSG_RESULT_OK;
2053 }
2054
2055 static int
2056 vhost_user_get_protocol_features(struct virtio_net **pdev,
2057                         struct VhostUserMsg *msg,
2058                         int main_fd __rte_unused)
2059 {
2060         struct virtio_net *dev = *pdev;
2061         uint64_t features, protocol_features;
2062
2063         if (validate_msg_fds(msg, 0) != 0)
2064                 return RTE_VHOST_MSG_RESULT_ERR;
2065
2066         rte_vhost_driver_get_features(dev->ifname, &features);
2067         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2068
2069         msg->payload.u64 = protocol_features;
2070         msg->size = sizeof(msg->payload.u64);
2071         msg->fd_num = 0;
2072
2073         return RTE_VHOST_MSG_RESULT_REPLY;
2074 }
2075
2076 static int
2077 vhost_user_set_protocol_features(struct virtio_net **pdev,
2078                         struct VhostUserMsg *msg,
2079                         int main_fd __rte_unused)
2080 {
2081         struct virtio_net *dev = *pdev;
2082         uint64_t protocol_features = msg->payload.u64;
2083         uint64_t slave_protocol_features = 0;
2084
2085         if (validate_msg_fds(msg, 0) != 0)
2086                 return RTE_VHOST_MSG_RESULT_ERR;
2087
2088         rte_vhost_driver_get_protocol_features(dev->ifname,
2089                         &slave_protocol_features);
2090         if (protocol_features & ~slave_protocol_features) {
2091                 VHOST_LOG_CONFIG(ERR,
2092                         "(%d) received invalid protocol features.\n",
2093                         dev->vid);
2094                 return RTE_VHOST_MSG_RESULT_ERR;
2095         }
2096
2097         dev->protocol_features = protocol_features;
2098         VHOST_LOG_CONFIG(INFO,
2099                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2100                 dev->protocol_features);
2101
2102         return RTE_VHOST_MSG_RESULT_OK;
2103 }
2104
2105 static int
2106 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2107                         int main_fd __rte_unused)
2108 {
2109         struct virtio_net *dev = *pdev;
2110         int fd = msg->fds[0];
2111         uint64_t size, off;
2112         void *addr;
2113
2114         if (validate_msg_fds(msg, 1) != 0)
2115                 return RTE_VHOST_MSG_RESULT_ERR;
2116
2117         if (fd < 0) {
2118                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2119                 return RTE_VHOST_MSG_RESULT_ERR;
2120         }
2121
2122         if (msg->size != sizeof(VhostUserLog)) {
2123                 VHOST_LOG_CONFIG(ERR,
2124                         "invalid log base msg size: %"PRId32" != %d\n",
2125                         msg->size, (int)sizeof(VhostUserLog));
2126                 goto close_msg_fds;
2127         }
2128
2129         size = msg->payload.log.mmap_size;
2130         off  = msg->payload.log.mmap_offset;
2131
2132         /* Check for mmap size and offset overflow. */
2133         if (off >= -size) {
2134                 VHOST_LOG_CONFIG(ERR,
2135                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2136                         off, size);
2137                 goto close_msg_fds;
2138         }
2139
2140         VHOST_LOG_CONFIG(INFO,
2141                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2142                 size, off);
2143
2144         /*
2145          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2146          * fail when offset is not page size aligned.
2147          */
2148         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2149         close(fd);
2150         if (addr == MAP_FAILED) {
2151                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2152                 return RTE_VHOST_MSG_RESULT_ERR;
2153         }
2154
2155         /*
2156          * Free previously mapped log memory on occasionally
2157          * multiple VHOST_USER_SET_LOG_BASE.
2158          */
2159         if (dev->log_addr) {
2160                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2161         }
2162         dev->log_addr = (uint64_t)(uintptr_t)addr;
2163         dev->log_base = dev->log_addr + off;
2164         dev->log_size = size;
2165
2166         /*
2167          * The spec is not clear about it (yet), but QEMU doesn't expect
2168          * any payload in the reply.
2169          */
2170         msg->size = 0;
2171         msg->fd_num = 0;
2172
2173         return RTE_VHOST_MSG_RESULT_REPLY;
2174
2175 close_msg_fds:
2176         close_msg_fds(msg);
2177         return RTE_VHOST_MSG_RESULT_ERR;
2178 }
2179
2180 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2181                         struct VhostUserMsg *msg,
2182                         int main_fd __rte_unused)
2183 {
2184         if (validate_msg_fds(msg, 1) != 0)
2185                 return RTE_VHOST_MSG_RESULT_ERR;
2186
2187         close(msg->fds[0]);
2188         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2189
2190         return RTE_VHOST_MSG_RESULT_OK;
2191 }
2192
2193 /*
2194  * An rarp packet is constructed and broadcasted to notify switches about
2195  * the new location of the migrated VM, so that packets from outside will
2196  * not be lost after migration.
2197  *
2198  * However, we don't actually "send" a rarp packet here, instead, we set
2199  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2200  */
2201 static int
2202 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2203                         int main_fd __rte_unused)
2204 {
2205         struct virtio_net *dev = *pdev;
2206         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2207         struct rte_vdpa_device *vdpa_dev;
2208
2209         if (validate_msg_fds(msg, 0) != 0)
2210                 return RTE_VHOST_MSG_RESULT_ERR;
2211
2212         VHOST_LOG_CONFIG(DEBUG,
2213                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2214                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2215         memcpy(dev->mac.addr_bytes, mac, 6);
2216
2217         /*
2218          * Set the flag to inject a RARP broadcast packet at
2219          * rte_vhost_dequeue_burst().
2220          *
2221          * __ATOMIC_RELEASE ordering is for making sure the mac is
2222          * copied before the flag is set.
2223          */
2224         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2225         vdpa_dev = dev->vdpa_dev;
2226         if (vdpa_dev && vdpa_dev->ops->migration_done)
2227                 vdpa_dev->ops->migration_done(dev->vid);
2228
2229         return RTE_VHOST_MSG_RESULT_OK;
2230 }
2231
2232 static int
2233 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2234                         int main_fd __rte_unused)
2235 {
2236         struct virtio_net *dev = *pdev;
2237
2238         if (validate_msg_fds(msg, 0) != 0)
2239                 return RTE_VHOST_MSG_RESULT_ERR;
2240
2241         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2242                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2243                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2244                                 msg->payload.u64);
2245
2246                 return RTE_VHOST_MSG_RESULT_ERR;
2247         }
2248
2249         dev->mtu = msg->payload.u64;
2250
2251         return RTE_VHOST_MSG_RESULT_OK;
2252 }
2253
2254 static int
2255 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2256                         int main_fd __rte_unused)
2257 {
2258         struct virtio_net *dev = *pdev;
2259         int fd = msg->fds[0];
2260
2261         if (validate_msg_fds(msg, 1) != 0)
2262                 return RTE_VHOST_MSG_RESULT_ERR;
2263
2264         if (fd < 0) {
2265                 VHOST_LOG_CONFIG(ERR,
2266                                 "Invalid file descriptor for slave channel (%d)\n",
2267                                 fd);
2268                 return RTE_VHOST_MSG_RESULT_ERR;
2269         }
2270
2271         if (dev->slave_req_fd >= 0)
2272                 close(dev->slave_req_fd);
2273
2274         dev->slave_req_fd = fd;
2275
2276         return RTE_VHOST_MSG_RESULT_OK;
2277 }
2278
2279 static int
2280 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2281 {
2282         struct vhost_vring_addr *ra;
2283         uint64_t start, end, len;
2284
2285         start = imsg->iova;
2286         end = start + imsg->size;
2287
2288         ra = &vq->ring_addrs;
2289         len = sizeof(struct vring_desc) * vq->size;
2290         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2291                 return 1;
2292
2293         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2294         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2295                 return 1;
2296
2297         len = sizeof(struct vring_used) +
2298                sizeof(struct vring_used_elem) * vq->size;
2299         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2300                 return 1;
2301
2302         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2303                 len = sizeof(uint64_t);
2304                 if (ra->log_guest_addr < end &&
2305                     (ra->log_guest_addr + len) > start)
2306                         return 1;
2307         }
2308
2309         return 0;
2310 }
2311
2312 static int
2313 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2314 {
2315         struct vhost_vring_addr *ra;
2316         uint64_t start, end, len;
2317
2318         start = imsg->iova;
2319         end = start + imsg->size;
2320
2321         ra = &vq->ring_addrs;
2322         len = sizeof(struct vring_packed_desc) * vq->size;
2323         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2324                 return 1;
2325
2326         len = sizeof(struct vring_packed_desc_event);
2327         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2328                 return 1;
2329
2330         len = sizeof(struct vring_packed_desc_event);
2331         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2332                 return 1;
2333
2334         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2335                 len = sizeof(uint64_t);
2336                 if (ra->log_guest_addr < end &&
2337                     (ra->log_guest_addr + len) > start)
2338                         return 1;
2339         }
2340
2341         return 0;
2342 }
2343
2344 static int is_vring_iotlb(struct virtio_net *dev,
2345                           struct vhost_virtqueue *vq,
2346                           struct vhost_iotlb_msg *imsg)
2347 {
2348         if (vq_is_packed(dev))
2349                 return is_vring_iotlb_packed(vq, imsg);
2350         else
2351                 return is_vring_iotlb_split(vq, imsg);
2352 }
2353
2354 static int
2355 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2356                         int main_fd __rte_unused)
2357 {
2358         struct virtio_net *dev = *pdev;
2359         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2360         uint16_t i;
2361         uint64_t vva, len;
2362
2363         if (validate_msg_fds(msg, 0) != 0)
2364                 return RTE_VHOST_MSG_RESULT_ERR;
2365
2366         switch (imsg->type) {
2367         case VHOST_IOTLB_UPDATE:
2368                 len = imsg->size;
2369                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2370                 if (!vva)
2371                         return RTE_VHOST_MSG_RESULT_ERR;
2372
2373                 for (i = 0; i < dev->nr_vring; i++) {
2374                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2375
2376                         if (!vq)
2377                                 continue;
2378
2379                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2380                                         len, imsg->perm);
2381
2382                         if (is_vring_iotlb(dev, vq, imsg))
2383                                 *pdev = dev = translate_ring_addresses(dev, i);
2384                 }
2385                 break;
2386         case VHOST_IOTLB_INVALIDATE:
2387                 for (i = 0; i < dev->nr_vring; i++) {
2388                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2389
2390                         if (!vq)
2391                                 continue;
2392
2393                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2394                                         imsg->size);
2395
2396                         if (is_vring_iotlb(dev, vq, imsg))
2397                                 vring_invalidate(dev, vq);
2398                 }
2399                 break;
2400         default:
2401                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2402                                 imsg->type);
2403                 return RTE_VHOST_MSG_RESULT_ERR;
2404         }
2405
2406         return RTE_VHOST_MSG_RESULT_OK;
2407 }
2408
2409 static int
2410 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2411                         struct VhostUserMsg *msg,
2412                         int main_fd __rte_unused)
2413 {
2414         struct virtio_net *dev = *pdev;
2415 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2416         struct uffdio_api api_struct;
2417
2418         if (validate_msg_fds(msg, 0) != 0)
2419                 return RTE_VHOST_MSG_RESULT_ERR;
2420
2421         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2422
2423         if (dev->postcopy_ufd == -1) {
2424                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2425                         strerror(errno));
2426                 return RTE_VHOST_MSG_RESULT_ERR;
2427         }
2428         api_struct.api = UFFD_API;
2429         api_struct.features = 0;
2430         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2431                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2432                         strerror(errno));
2433                 close(dev->postcopy_ufd);
2434                 dev->postcopy_ufd = -1;
2435                 return RTE_VHOST_MSG_RESULT_ERR;
2436         }
2437         msg->fds[0] = dev->postcopy_ufd;
2438         msg->fd_num = 1;
2439
2440         return RTE_VHOST_MSG_RESULT_REPLY;
2441 #else
2442         dev->postcopy_ufd = -1;
2443         msg->fd_num = 0;
2444
2445         return RTE_VHOST_MSG_RESULT_ERR;
2446 #endif
2447 }
2448
2449 static int
2450 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2451                         struct VhostUserMsg *msg __rte_unused,
2452                         int main_fd __rte_unused)
2453 {
2454         struct virtio_net *dev = *pdev;
2455
2456         if (validate_msg_fds(msg, 0) != 0)
2457                 return RTE_VHOST_MSG_RESULT_ERR;
2458
2459         if (dev->mem && dev->mem->nregions) {
2460                 VHOST_LOG_CONFIG(ERR,
2461                         "Regions already registered at postcopy-listen\n");
2462                 return RTE_VHOST_MSG_RESULT_ERR;
2463         }
2464         dev->postcopy_listening = 1;
2465
2466         return RTE_VHOST_MSG_RESULT_OK;
2467 }
2468
2469 static int
2470 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2471                         int main_fd __rte_unused)
2472 {
2473         struct virtio_net *dev = *pdev;
2474
2475         if (validate_msg_fds(msg, 0) != 0)
2476                 return RTE_VHOST_MSG_RESULT_ERR;
2477
2478         dev->postcopy_listening = 0;
2479         if (dev->postcopy_ufd >= 0) {
2480                 close(dev->postcopy_ufd);
2481                 dev->postcopy_ufd = -1;
2482         }
2483
2484         msg->payload.u64 = 0;
2485         msg->size = sizeof(msg->payload.u64);
2486         msg->fd_num = 0;
2487
2488         return RTE_VHOST_MSG_RESULT_REPLY;
2489 }
2490
2491 static int
2492 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2493                       int main_fd __rte_unused)
2494 {
2495         struct virtio_net *dev = *pdev;
2496
2497         if (validate_msg_fds(msg, 0) != 0)
2498                 return RTE_VHOST_MSG_RESULT_ERR;
2499
2500         msg->payload.u64 = dev->status;
2501         msg->size = sizeof(msg->payload.u64);
2502         msg->fd_num = 0;
2503
2504         return RTE_VHOST_MSG_RESULT_REPLY;
2505 }
2506
2507 static int
2508 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2509                         int main_fd __rte_unused)
2510 {
2511         struct virtio_net *dev = *pdev;
2512
2513         if (validate_msg_fds(msg, 0) != 0)
2514                 return RTE_VHOST_MSG_RESULT_ERR;
2515
2516         /* As per Virtio specification, the device status is 8bits long */
2517         if (msg->payload.u64 > UINT8_MAX) {
2518                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2519                                 msg->payload.u64);
2520                 return RTE_VHOST_MSG_RESULT_ERR;
2521         }
2522
2523         dev->status = msg->payload.u64;
2524
2525         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2526             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2527                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2528                 /*
2529                  * Clear the bit to let the driver know about the feature
2530                  * negotiation failure
2531                  */
2532                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2533         }
2534
2535         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2536                         "\t-RESET: %u\n"
2537                         "\t-ACKNOWLEDGE: %u\n"
2538                         "\t-DRIVER: %u\n"
2539                         "\t-FEATURES_OK: %u\n"
2540                         "\t-DRIVER_OK: %u\n"
2541                         "\t-DEVICE_NEED_RESET: %u\n"
2542                         "\t-FAILED: %u\n",
2543                         dev->status,
2544                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2545                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2546                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2547                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2548                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2549                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2550                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2551
2552         return RTE_VHOST_MSG_RESULT_OK;
2553 }
2554
2555 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2556                                         struct VhostUserMsg *msg,
2557                                         int main_fd);
2558 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2559         [VHOST_USER_NONE] = NULL,
2560         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2561         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2562         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2563         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2564         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2565         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2566         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2567         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2568         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2569         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2570         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2571         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2572         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2573         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2574         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2575         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2576         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2577         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2578         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2579         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2580         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2581         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2582         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2583         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2584         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2585         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2586         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2587         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2588         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2589 };
2590
2591 /* return bytes# of read on success or negative val on failure. */
2592 static int
2593 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2594 {
2595         int ret;
2596
2597         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2598                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2599         if (ret <= 0) {
2600                 return ret;
2601         } else if (ret != VHOST_USER_HDR_SIZE) {
2602                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2603                 close_msg_fds(msg);
2604                 return -1;
2605         }
2606
2607         if (msg->size) {
2608                 if (msg->size > sizeof(msg->payload)) {
2609                         VHOST_LOG_CONFIG(ERR,
2610                                 "invalid msg size: %d\n", msg->size);
2611                         return -1;
2612                 }
2613                 ret = read(sockfd, &msg->payload, msg->size);
2614                 if (ret <= 0)
2615                         return ret;
2616                 if (ret != (int)msg->size) {
2617                         VHOST_LOG_CONFIG(ERR,
2618                                 "read control message failed\n");
2619                         return -1;
2620                 }
2621         }
2622
2623         return ret;
2624 }
2625
2626 static int
2627 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2628 {
2629         if (!msg)
2630                 return 0;
2631
2632         return send_fd_message(sockfd, (char *)msg,
2633                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2634 }
2635
2636 static int
2637 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2638 {
2639         if (!msg)
2640                 return 0;
2641
2642         msg->flags &= ~VHOST_USER_VERSION_MASK;
2643         msg->flags &= ~VHOST_USER_NEED_REPLY;
2644         msg->flags |= VHOST_USER_VERSION;
2645         msg->flags |= VHOST_USER_REPLY_MASK;
2646
2647         return send_vhost_message(sockfd, msg);
2648 }
2649
2650 static int
2651 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2652 {
2653         int ret;
2654
2655         if (msg->flags & VHOST_USER_NEED_REPLY)
2656                 rte_spinlock_lock(&dev->slave_req_lock);
2657
2658         ret = send_vhost_message(dev->slave_req_fd, msg);
2659         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2660                 rte_spinlock_unlock(&dev->slave_req_lock);
2661
2662         return ret;
2663 }
2664
2665 /*
2666  * Allocate a queue pair if it hasn't been allocated yet
2667  */
2668 static int
2669 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2670                         struct VhostUserMsg *msg)
2671 {
2672         uint32_t vring_idx;
2673
2674         switch (msg->request.master) {
2675         case VHOST_USER_SET_VRING_KICK:
2676         case VHOST_USER_SET_VRING_CALL:
2677         case VHOST_USER_SET_VRING_ERR:
2678                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2679                 break;
2680         case VHOST_USER_SET_VRING_NUM:
2681         case VHOST_USER_SET_VRING_BASE:
2682         case VHOST_USER_SET_VRING_ENABLE:
2683                 vring_idx = msg->payload.state.index;
2684                 break;
2685         case VHOST_USER_SET_VRING_ADDR:
2686                 vring_idx = msg->payload.addr.index;
2687                 break;
2688         default:
2689                 return 0;
2690         }
2691
2692         if (vring_idx >= VHOST_MAX_VRING) {
2693                 VHOST_LOG_CONFIG(ERR,
2694                         "invalid vring index: %u\n", vring_idx);
2695                 return -1;
2696         }
2697
2698         if (dev->virtqueue[vring_idx])
2699                 return 0;
2700
2701         return alloc_vring_queue(dev, vring_idx);
2702 }
2703
2704 static void
2705 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2706 {
2707         unsigned int i = 0;
2708         unsigned int vq_num = 0;
2709
2710         while (vq_num < dev->nr_vring) {
2711                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2712
2713                 if (vq) {
2714                         rte_spinlock_lock(&vq->access_lock);
2715                         vq_num++;
2716                 }
2717                 i++;
2718         }
2719 }
2720
2721 static void
2722 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2723 {
2724         unsigned int i = 0;
2725         unsigned int vq_num = 0;
2726
2727         while (vq_num < dev->nr_vring) {
2728                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2729
2730                 if (vq) {
2731                         rte_spinlock_unlock(&vq->access_lock);
2732                         vq_num++;
2733                 }
2734                 i++;
2735         }
2736 }
2737
2738 int
2739 vhost_user_msg_handler(int vid, int fd)
2740 {
2741         struct virtio_net *dev;
2742         struct VhostUserMsg msg;
2743         struct rte_vdpa_device *vdpa_dev;
2744         int ret;
2745         int unlock_required = 0;
2746         bool handled;
2747         int request;
2748         uint32_t i;
2749
2750         dev = get_device(vid);
2751         if (dev == NULL)
2752                 return -1;
2753
2754         if (!dev->notify_ops) {
2755                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2756                 if (!dev->notify_ops) {
2757                         VHOST_LOG_CONFIG(ERR,
2758                                 "failed to get callback ops for driver %s\n",
2759                                 dev->ifname);
2760                         return -1;
2761                 }
2762         }
2763
2764         ret = read_vhost_message(fd, &msg);
2765         if (ret <= 0) {
2766                 if (ret < 0)
2767                         VHOST_LOG_CONFIG(ERR,
2768                                 "vhost read message failed\n");
2769                 else
2770                         VHOST_LOG_CONFIG(INFO,
2771                                 "vhost peer closed\n");
2772
2773                 return -1;
2774         }
2775
2776         ret = 0;
2777         request = msg.request.master;
2778         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2779                         vhost_message_str[request]) {
2780                 if (request != VHOST_USER_IOTLB_MSG)
2781                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2782                                 vhost_message_str[request]);
2783                 else
2784                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2785                                 vhost_message_str[request]);
2786         } else {
2787                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2788         }
2789
2790         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2791         if (ret < 0) {
2792                 VHOST_LOG_CONFIG(ERR,
2793                         "failed to alloc queue\n");
2794                 return -1;
2795         }
2796
2797         /*
2798          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2799          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2800          * and device is destroyed. destroy_device waits for queues to be
2801          * inactive, so it is safe. Otherwise taking the access_lock
2802          * would cause a dead lock.
2803          */
2804         switch (request) {
2805         case VHOST_USER_SET_FEATURES:
2806         case VHOST_USER_SET_PROTOCOL_FEATURES:
2807         case VHOST_USER_SET_OWNER:
2808         case VHOST_USER_SET_MEM_TABLE:
2809         case VHOST_USER_SET_LOG_BASE:
2810         case VHOST_USER_SET_LOG_FD:
2811         case VHOST_USER_SET_VRING_NUM:
2812         case VHOST_USER_SET_VRING_ADDR:
2813         case VHOST_USER_SET_VRING_BASE:
2814         case VHOST_USER_SET_VRING_KICK:
2815         case VHOST_USER_SET_VRING_CALL:
2816         case VHOST_USER_SET_VRING_ERR:
2817         case VHOST_USER_SET_VRING_ENABLE:
2818         case VHOST_USER_SEND_RARP:
2819         case VHOST_USER_NET_SET_MTU:
2820         case VHOST_USER_SET_SLAVE_REQ_FD:
2821                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2822                         vhost_user_lock_all_queue_pairs(dev);
2823                         unlock_required = 1;
2824                 }
2825                 break;
2826         default:
2827                 break;
2828
2829         }
2830
2831         handled = false;
2832         if (dev->extern_ops.pre_msg_handle) {
2833                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2834                                 (void *)&msg);
2835                 switch (ret) {
2836                 case RTE_VHOST_MSG_RESULT_REPLY:
2837                         send_vhost_reply(fd, &msg);
2838                         /* Fall-through */
2839                 case RTE_VHOST_MSG_RESULT_ERR:
2840                 case RTE_VHOST_MSG_RESULT_OK:
2841                         handled = true;
2842                         goto skip_to_post_handle;
2843                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2844                 default:
2845                         break;
2846                 }
2847         }
2848
2849         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2850                 if (!vhost_message_handlers[request])
2851                         goto skip_to_post_handle;
2852                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2853
2854                 switch (ret) {
2855                 case RTE_VHOST_MSG_RESULT_ERR:
2856                         VHOST_LOG_CONFIG(ERR,
2857                                 "Processing %s failed.\n",
2858                                 vhost_message_str[request]);
2859                         handled = true;
2860                         break;
2861                 case RTE_VHOST_MSG_RESULT_OK:
2862                         VHOST_LOG_CONFIG(DEBUG,
2863                                 "Processing %s succeeded.\n",
2864                                 vhost_message_str[request]);
2865                         handled = true;
2866                         break;
2867                 case RTE_VHOST_MSG_RESULT_REPLY:
2868                         VHOST_LOG_CONFIG(DEBUG,
2869                                 "Processing %s succeeded and needs reply.\n",
2870                                 vhost_message_str[request]);
2871                         send_vhost_reply(fd, &msg);
2872                         handled = true;
2873                         break;
2874                 default:
2875                         break;
2876                 }
2877         }
2878
2879 skip_to_post_handle:
2880         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2881                         dev->extern_ops.post_msg_handle) {
2882                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2883                                 (void *)&msg);
2884                 switch (ret) {
2885                 case RTE_VHOST_MSG_RESULT_REPLY:
2886                         send_vhost_reply(fd, &msg);
2887                         /* Fall-through */
2888                 case RTE_VHOST_MSG_RESULT_ERR:
2889                 case RTE_VHOST_MSG_RESULT_OK:
2890                         handled = true;
2891                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2892                 default:
2893                         break;
2894                 }
2895         }
2896
2897         if (unlock_required)
2898                 vhost_user_unlock_all_queue_pairs(dev);
2899
2900         /* If message was not handled at this stage, treat it as an error */
2901         if (!handled) {
2902                 VHOST_LOG_CONFIG(ERR,
2903                         "vhost message (req: %d) was not handled.\n", request);
2904                 close_msg_fds(&msg);
2905                 ret = RTE_VHOST_MSG_RESULT_ERR;
2906         }
2907
2908         /*
2909          * If the request required a reply that was already sent,
2910          * this optional reply-ack won't be sent as the
2911          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2912          */
2913         if (msg.flags & VHOST_USER_NEED_REPLY) {
2914                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2915                 msg.size = sizeof(msg.payload.u64);
2916                 msg.fd_num = 0;
2917                 send_vhost_reply(fd, &msg);
2918         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2919                 VHOST_LOG_CONFIG(ERR,
2920                         "vhost message handling failed.\n");
2921                 return -1;
2922         }
2923
2924         for (i = 0; i < dev->nr_vring; i++) {
2925                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2926                 bool cur_ready = vq_is_ready(dev, vq);
2927
2928                 if (cur_ready != (vq && vq->ready)) {
2929                         vq->ready = cur_ready;
2930                         vhost_user_notify_queue_state(dev, i, cur_ready);
2931                 }
2932         }
2933
2934
2935         if (!virtio_is_ready(dev))
2936                 goto out;
2937
2938         /*
2939          * Virtio is now ready. If not done already, it is time
2940          * to notify the application it can process the rings and
2941          * configure the vDPA device if present.
2942          */
2943
2944         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2945                 if (dev->notify_ops->new_device(dev->vid) == 0)
2946                         dev->flags |= VIRTIO_DEV_RUNNING;
2947         }
2948
2949         vdpa_dev = dev->vdpa_dev;
2950         if (!vdpa_dev)
2951                 goto out;
2952
2953         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2954                 if (vdpa_dev->ops->dev_conf(dev->vid))
2955                         VHOST_LOG_CONFIG(ERR,
2956                                          "Failed to configure vDPA device\n");
2957                 else
2958                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2959         }
2960
2961 out:
2962         return 0;
2963 }
2964
2965 static int process_slave_message_reply(struct virtio_net *dev,
2966                                        const struct VhostUserMsg *msg)
2967 {
2968         struct VhostUserMsg msg_reply;
2969         int ret;
2970
2971         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2972                 return 0;
2973
2974         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
2975         if (ret <= 0) {
2976                 if (ret < 0)
2977                         VHOST_LOG_CONFIG(ERR,
2978                                 "vhost read slave message reply failed\n");
2979                 else
2980                         VHOST_LOG_CONFIG(INFO,
2981                                 "vhost peer closed\n");
2982                 ret = -1;
2983                 goto out;
2984         }
2985
2986         ret = 0;
2987         if (msg_reply.request.slave != msg->request.slave) {
2988                 VHOST_LOG_CONFIG(ERR,
2989                         "Received unexpected msg type (%u), expected %u\n",
2990                         msg_reply.request.slave, msg->request.slave);
2991                 ret = -1;
2992                 goto out;
2993         }
2994
2995         ret = msg_reply.payload.u64 ? -1 : 0;
2996
2997 out:
2998         rte_spinlock_unlock(&dev->slave_req_lock);
2999         return ret;
3000 }
3001
3002 int
3003 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3004 {
3005         int ret;
3006         struct VhostUserMsg msg = {
3007                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3008                 .flags = VHOST_USER_VERSION,
3009                 .size = sizeof(msg.payload.iotlb),
3010                 .payload.iotlb = {
3011                         .iova = iova,
3012                         .perm = perm,
3013                         .type = VHOST_IOTLB_MISS,
3014                 },
3015         };
3016
3017         ret = send_vhost_message(dev->slave_req_fd, &msg);
3018         if (ret < 0) {
3019                 VHOST_LOG_CONFIG(ERR,
3020                                 "Failed to send IOTLB miss message (%d)\n",
3021                                 ret);
3022                 return ret;
3023         }
3024
3025         return 0;
3026 }
3027
3028 static int
3029 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3030 {
3031         int ret;
3032         struct VhostUserMsg msg = {
3033                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3034                 .flags = VHOST_USER_VERSION,
3035                 .size = 0,
3036         };
3037
3038         if (need_reply)
3039                 msg.flags |= VHOST_USER_NEED_REPLY;
3040
3041         ret = send_vhost_slave_message(dev, &msg);
3042         if (ret < 0) {
3043                 VHOST_LOG_CONFIG(ERR,
3044                                 "Failed to send config change (%d)\n",
3045                                 ret);
3046                 return ret;
3047         }
3048
3049         return process_slave_message_reply(dev, &msg);
3050 }
3051
3052 int
3053 rte_vhost_slave_config_change(int vid, bool need_reply)
3054 {
3055         struct virtio_net *dev;
3056
3057         dev = get_device(vid);
3058         if (!dev)
3059                 return -ENODEV;
3060
3061         return vhost_user_slave_config_change(dev, need_reply);
3062 }
3063
3064 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3065                                                     int index, int fd,
3066                                                     uint64_t offset,
3067                                                     uint64_t size)
3068 {
3069         int ret;
3070         struct VhostUserMsg msg = {
3071                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3072                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3073                 .size = sizeof(msg.payload.area),
3074                 .payload.area = {
3075                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3076                         .size = size,
3077                         .offset = offset,
3078                 },
3079         };
3080
3081         if (fd < 0)
3082                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3083         else {
3084                 msg.fds[0] = fd;
3085                 msg.fd_num = 1;
3086         }
3087
3088         ret = send_vhost_slave_message(dev, &msg);
3089         if (ret < 0) {
3090                 VHOST_LOG_CONFIG(ERR,
3091                         "Failed to set host notifier (%d)\n", ret);
3092                 return ret;
3093         }
3094
3095         return process_slave_message_reply(dev, &msg);
3096 }
3097
3098 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3099 {
3100         struct virtio_net *dev;
3101         struct rte_vdpa_device *vdpa_dev;
3102         int vfio_device_fd, ret = 0;
3103         uint64_t offset, size;
3104         unsigned int i, q_start, q_last;
3105
3106         dev = get_device(vid);
3107         if (!dev)
3108                 return -ENODEV;
3109
3110         vdpa_dev = dev->vdpa_dev;
3111         if (vdpa_dev == NULL)
3112                 return -ENODEV;
3113
3114         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3115             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3116             !(dev->protocol_features &
3117                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3118             !(dev->protocol_features &
3119                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3120             !(dev->protocol_features &
3121                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3122                 return -ENOTSUP;
3123
3124         if (qid == RTE_VHOST_QUEUE_ALL) {
3125                 q_start = 0;
3126                 q_last = dev->nr_vring - 1;
3127         } else {
3128                 if (qid >= dev->nr_vring)
3129                         return -EINVAL;
3130                 q_start = qid;
3131                 q_last = qid;
3132         }
3133
3134         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3135         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3136
3137         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3138         if (vfio_device_fd < 0)
3139                 return -ENOTSUP;
3140
3141         if (enable) {
3142                 for (i = q_start; i <= q_last; i++) {
3143                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3144                                         &size) < 0) {
3145                                 ret = -ENOTSUP;
3146                                 goto disable;
3147                         }
3148
3149                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3150                                         vfio_device_fd, offset, size) < 0) {
3151                                 ret = -EFAULT;
3152                                 goto disable;
3153                         }
3154                 }
3155         } else {
3156 disable:
3157                 for (i = q_start; i <= q_last; i++) {
3158                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3159                                         0, 0);
3160                 }
3161         }
3162
3163         return ret;
3164 }