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_mmap_region(struct virtio_net *dev,
1102                 struct rte_vhost_mem_region *region,
1103                 uint64_t mmap_offset)
1104 {
1105         void *mmap_addr;
1106         uint64_t mmap_size;
1107         uint64_t alignment;
1108         int populate;
1109
1110         /* Check for memory_size + mmap_offset overflow */
1111         if (mmap_offset >= -region->size) {
1112                 VHOST_LOG_CONFIG(ERR,
1113                                 "mmap_offset (%#"PRIx64") and memory_size "
1114                                 "(%#"PRIx64") overflow\n",
1115                                 mmap_offset, region->size);
1116                 return -1;
1117         }
1118
1119         mmap_size = region->size + mmap_offset;
1120
1121         /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1122          * argument aligned with hugepagesz at older longterm version Linux,
1123          * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1124          *
1125          * To avoid failure, make sure in caller to keep length aligned.
1126          */
1127         alignment = get_blk_size(region->fd);
1128         if (alignment == (uint64_t)-1) {
1129                 VHOST_LOG_CONFIG(ERR,
1130                                 "couldn't get hugepage size through fstat\n");
1131                 return -1;
1132         }
1133         mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1134         if (mmap_size == 0) {
1135                 /*
1136                  * It could happen if initial mmap_size + alignment overflows
1137                  * the sizeof uint64, which could happen if either mmap_size or
1138                  * alignment value is wrong.
1139                  *
1140                  * mmap() kernel implementation would return an error, but
1141                  * better catch it before and provide useful info in the logs.
1142                  */
1143                 VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1144                                 "or alignment (0x%" PRIx64 ") is invalid\n",
1145                                 region->size + mmap_offset, alignment);
1146                 return -1;
1147         }
1148
1149         populate = dev->async_copy ? MAP_POPULATE : 0;
1150         mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1151                         MAP_SHARED | populate, region->fd, 0);
1152
1153         if (mmap_addr == MAP_FAILED) {
1154                 VHOST_LOG_CONFIG(ERR, "mmap failed (%s).\n", strerror(errno));
1155                 return -1;
1156         }
1157
1158         region->mmap_addr = mmap_addr;
1159         region->mmap_size = mmap_size;
1160         region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1161
1162         if (dev->async_copy)
1163                 if (add_guest_pages(dev, region, alignment) < 0) {
1164                         VHOST_LOG_CONFIG(ERR,
1165                                         "adding guest pages to region failed.\n");
1166                         return -1;
1167                 }
1168
1169         VHOST_LOG_CONFIG(INFO,
1170                         "guest memory region size: 0x%" PRIx64 "\n"
1171                         "\t guest physical addr: 0x%" PRIx64 "\n"
1172                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1173                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1174                         "\t mmap addr : 0x%" PRIx64 "\n"
1175                         "\t mmap size : 0x%" PRIx64 "\n"
1176                         "\t mmap align: 0x%" PRIx64 "\n"
1177                         "\t mmap off  : 0x%" PRIx64 "\n",
1178                         region->size,
1179                         region->guest_phys_addr,
1180                         region->guest_user_addr,
1181                         region->host_user_addr,
1182                         (uint64_t)(uintptr_t)mmap_addr,
1183                         mmap_size,
1184                         alignment,
1185                         mmap_offset);
1186
1187         return 0;
1188 }
1189
1190 static int
1191 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1192                         int main_fd)
1193 {
1194         struct virtio_net *dev = *pdev;
1195         struct VhostUserMemory *memory = &msg->payload.memory;
1196         struct rte_vhost_mem_region *reg;
1197
1198         uint64_t mmap_offset;
1199         uint32_t i;
1200
1201         if (validate_msg_fds(msg, memory->nregions) != 0)
1202                 return RTE_VHOST_MSG_RESULT_ERR;
1203
1204         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1205                 VHOST_LOG_CONFIG(ERR,
1206                         "too many memory regions (%u)\n", memory->nregions);
1207                 goto close_msg_fds;
1208         }
1209
1210         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1211                 VHOST_LOG_CONFIG(INFO,
1212                         "(%d) memory regions not changed\n", dev->vid);
1213
1214                 close_msg_fds(msg);
1215
1216                 return RTE_VHOST_MSG_RESULT_OK;
1217         }
1218
1219         if (dev->mem) {
1220                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1221                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1222
1223                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1224                                 vdpa_dev->ops->dev_close(dev->vid);
1225                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1226                 }
1227                 free_mem_region(dev);
1228                 rte_free(dev->mem);
1229                 dev->mem = NULL;
1230         }
1231
1232         /* Flush IOTLB cache as previous HVAs are now invalid */
1233         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1234                 for (i = 0; i < dev->nr_vring; i++)
1235                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1236
1237         dev->nr_guest_pages = 0;
1238         if (dev->guest_pages == NULL) {
1239                 dev->max_guest_pages = 8;
1240                 dev->guest_pages = rte_zmalloc(NULL,
1241                                         dev->max_guest_pages *
1242                                         sizeof(struct guest_page),
1243                                         RTE_CACHE_LINE_SIZE);
1244                 if (dev->guest_pages == NULL) {
1245                         VHOST_LOG_CONFIG(ERR,
1246                                 "(%d) failed to allocate memory "
1247                                 "for dev->guest_pages\n",
1248                                 dev->vid);
1249                         goto close_msg_fds;
1250                 }
1251         }
1252
1253         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1254                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1255         if (dev->mem == NULL) {
1256                 VHOST_LOG_CONFIG(ERR,
1257                         "(%d) failed to allocate memory for dev->mem\n",
1258                         dev->vid);
1259                 goto free_guest_pages;
1260         }
1261
1262         for (i = 0; i < memory->nregions; i++) {
1263                 reg = &dev->mem->regions[i];
1264
1265                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1266                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1267                 reg->size            = memory->regions[i].memory_size;
1268                 reg->fd              = msg->fds[i];
1269
1270                 /*
1271                  * Assign invalid file descriptor value to avoid double
1272                  * closing on error path.
1273                  */
1274                 msg->fds[i] = -1;
1275
1276                 mmap_offset = memory->regions[i].mmap_offset;
1277
1278                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1279                         VHOST_LOG_CONFIG(ERR, "Failed to mmap region %u\n", i);
1280                         goto free_mem_table;
1281                 }
1282
1283                 dev->mem->nregions++;
1284         }
1285
1286         if (vhost_user_postcopy_register(dev, main_fd, msg) < 0)
1287                 goto free_mem_table;
1288
1289         for (i = 0; i < dev->nr_vring; i++) {
1290                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1291
1292                 if (!vq)
1293                         continue;
1294
1295                 if (vq->desc || vq->avail || vq->used) {
1296                         /*
1297                          * If the memory table got updated, the ring addresses
1298                          * need to be translated again as virtual addresses have
1299                          * changed.
1300                          */
1301                         vring_invalidate(dev, vq);
1302
1303                         dev = translate_ring_addresses(dev, i);
1304                         if (!dev) {
1305                                 dev = *pdev;
1306                                 goto free_mem_table;
1307                         }
1308
1309                         *pdev = dev;
1310                 }
1311         }
1312
1313         dump_guest_pages(dev);
1314
1315         return RTE_VHOST_MSG_RESULT_OK;
1316
1317 free_mem_table:
1318         free_mem_region(dev);
1319         rte_free(dev->mem);
1320         dev->mem = NULL;
1321 free_guest_pages:
1322         rte_free(dev->guest_pages);
1323         dev->guest_pages = NULL;
1324 close_msg_fds:
1325         close_msg_fds(msg);
1326         return RTE_VHOST_MSG_RESULT_ERR;
1327 }
1328
1329 static bool
1330 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1331 {
1332         bool rings_ok;
1333
1334         if (!vq)
1335                 return false;
1336
1337         if (vq_is_packed(dev))
1338                 rings_ok = vq->desc_packed && vq->driver_event &&
1339                         vq->device_event;
1340         else
1341                 rings_ok = vq->desc && vq->avail && vq->used;
1342
1343         return rings_ok &&
1344                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1345                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1346                vq->enabled;
1347 }
1348
1349 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1350
1351 static int
1352 virtio_is_ready(struct virtio_net *dev)
1353 {
1354         struct vhost_virtqueue *vq;
1355         uint32_t i, nr_vring = dev->nr_vring;
1356
1357         if (dev->flags & VIRTIO_DEV_READY)
1358                 return 1;
1359
1360         if (!dev->nr_vring)
1361                 return 0;
1362
1363         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1364                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1365
1366                 if (dev->nr_vring < nr_vring)
1367                         return 0;
1368         }
1369
1370         for (i = 0; i < nr_vring; i++) {
1371                 vq = dev->virtqueue[i];
1372
1373                 if (!vq_is_ready(dev, vq))
1374                         return 0;
1375         }
1376
1377         /* If supported, ensure the frontend is really done with config */
1378         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1379                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1380                         return 0;
1381
1382         dev->flags |= VIRTIO_DEV_READY;
1383
1384         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1385                 VHOST_LOG_CONFIG(INFO,
1386                         "virtio is now ready for processing.\n");
1387         return 1;
1388 }
1389
1390 static void *
1391 inflight_mem_alloc(const char *name, size_t size, int *fd)
1392 {
1393         void *ptr;
1394         int mfd = -1;
1395         char fname[20] = "/tmp/memfd-XXXXXX";
1396
1397         *fd = -1;
1398 #ifdef MEMFD_SUPPORTED
1399         mfd = memfd_create(name, MFD_CLOEXEC);
1400 #else
1401         RTE_SET_USED(name);
1402 #endif
1403         if (mfd == -1) {
1404                 mfd = mkstemp(fname);
1405                 if (mfd == -1) {
1406                         VHOST_LOG_CONFIG(ERR,
1407                                 "failed to get inflight buffer fd\n");
1408                         return NULL;
1409                 }
1410
1411                 unlink(fname);
1412         }
1413
1414         if (ftruncate(mfd, size) == -1) {
1415                 VHOST_LOG_CONFIG(ERR,
1416                         "failed to alloc inflight buffer\n");
1417                 close(mfd);
1418                 return NULL;
1419         }
1420
1421         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1422         if (ptr == MAP_FAILED) {
1423                 VHOST_LOG_CONFIG(ERR,
1424                         "failed to mmap inflight buffer\n");
1425                 close(mfd);
1426                 return NULL;
1427         }
1428
1429         *fd = mfd;
1430         return ptr;
1431 }
1432
1433 static uint32_t
1434 get_pervq_shm_size_split(uint16_t queue_size)
1435 {
1436         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1437                                   queue_size + sizeof(uint64_t) +
1438                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1439 }
1440
1441 static uint32_t
1442 get_pervq_shm_size_packed(uint16_t queue_size)
1443 {
1444         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1445                                   * queue_size + sizeof(uint64_t) +
1446                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1447                                   INFLIGHT_ALIGNMENT);
1448 }
1449
1450 static int
1451 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1452                            VhostUserMsg *msg,
1453                            int main_fd __rte_unused)
1454 {
1455         struct rte_vhost_inflight_info_packed *inflight_packed;
1456         uint64_t pervq_inflight_size, mmap_size;
1457         uint16_t num_queues, queue_size;
1458         struct virtio_net *dev = *pdev;
1459         int fd, i, j;
1460         void *addr;
1461
1462         if (msg->size != sizeof(msg->payload.inflight)) {
1463                 VHOST_LOG_CONFIG(ERR,
1464                         "invalid get_inflight_fd message size is %d\n",
1465                         msg->size);
1466                 return RTE_VHOST_MSG_RESULT_ERR;
1467         }
1468
1469         if (dev->inflight_info == NULL) {
1470                 dev->inflight_info = calloc(1,
1471                                             sizeof(struct inflight_mem_info));
1472                 if (!dev->inflight_info) {
1473                         VHOST_LOG_CONFIG(ERR,
1474                                 "failed to alloc dev inflight area\n");
1475                         return RTE_VHOST_MSG_RESULT_ERR;
1476                 }
1477                 dev->inflight_info->fd = -1;
1478         }
1479
1480         num_queues = msg->payload.inflight.num_queues;
1481         queue_size = msg->payload.inflight.queue_size;
1482
1483         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1484                 msg->payload.inflight.num_queues);
1485         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1486                 msg->payload.inflight.queue_size);
1487
1488         if (vq_is_packed(dev))
1489                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1490         else
1491                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1492
1493         mmap_size = num_queues * pervq_inflight_size;
1494         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1495         if (!addr) {
1496                 VHOST_LOG_CONFIG(ERR,
1497                         "failed to alloc vhost inflight area\n");
1498                         msg->payload.inflight.mmap_size = 0;
1499                 return RTE_VHOST_MSG_RESULT_ERR;
1500         }
1501         memset(addr, 0, mmap_size);
1502
1503         if (dev->inflight_info->addr) {
1504                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1505                 dev->inflight_info->addr = NULL;
1506         }
1507
1508         if (dev->inflight_info->fd >= 0) {
1509                 close(dev->inflight_info->fd);
1510                 dev->inflight_info->fd = -1;
1511         }
1512
1513         dev->inflight_info->addr = addr;
1514         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1515         dev->inflight_info->fd = msg->fds[0] = fd;
1516         msg->payload.inflight.mmap_offset = 0;
1517         msg->fd_num = 1;
1518
1519         if (vq_is_packed(dev)) {
1520                 for (i = 0; i < num_queues; i++) {
1521                         inflight_packed =
1522                                 (struct rte_vhost_inflight_info_packed *)addr;
1523                         inflight_packed->used_wrap_counter = 1;
1524                         inflight_packed->old_used_wrap_counter = 1;
1525                         for (j = 0; j < queue_size; j++)
1526                                 inflight_packed->desc[j].next = j + 1;
1527                         addr = (void *)((char *)addr + pervq_inflight_size);
1528                 }
1529         }
1530
1531         VHOST_LOG_CONFIG(INFO,
1532                 "send inflight mmap_size: %"PRIu64"\n",
1533                 msg->payload.inflight.mmap_size);
1534         VHOST_LOG_CONFIG(INFO,
1535                 "send inflight mmap_offset: %"PRIu64"\n",
1536                 msg->payload.inflight.mmap_offset);
1537         VHOST_LOG_CONFIG(INFO,
1538                 "send inflight fd: %d\n", msg->fds[0]);
1539
1540         return RTE_VHOST_MSG_RESULT_REPLY;
1541 }
1542
1543 static int
1544 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1545                            int main_fd __rte_unused)
1546 {
1547         uint64_t mmap_size, mmap_offset;
1548         uint16_t num_queues, queue_size;
1549         struct virtio_net *dev = *pdev;
1550         uint32_t pervq_inflight_size;
1551         struct vhost_virtqueue *vq;
1552         void *addr;
1553         int fd, i;
1554
1555         fd = msg->fds[0];
1556         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1557                 VHOST_LOG_CONFIG(ERR,
1558                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1559                         msg->size, fd);
1560                 return RTE_VHOST_MSG_RESULT_ERR;
1561         }
1562
1563         mmap_size = msg->payload.inflight.mmap_size;
1564         mmap_offset = msg->payload.inflight.mmap_offset;
1565         num_queues = msg->payload.inflight.num_queues;
1566         queue_size = msg->payload.inflight.queue_size;
1567
1568         if (vq_is_packed(dev))
1569                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1570         else
1571                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1572
1573         VHOST_LOG_CONFIG(INFO,
1574                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1575         VHOST_LOG_CONFIG(INFO,
1576                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1577         VHOST_LOG_CONFIG(INFO,
1578                 "set_inflight_fd num_queues: %u\n", num_queues);
1579         VHOST_LOG_CONFIG(INFO,
1580                 "set_inflight_fd queue_size: %u\n", queue_size);
1581         VHOST_LOG_CONFIG(INFO,
1582                 "set_inflight_fd fd: %d\n", fd);
1583         VHOST_LOG_CONFIG(INFO,
1584                 "set_inflight_fd pervq_inflight_size: %d\n",
1585                 pervq_inflight_size);
1586
1587         if (!dev->inflight_info) {
1588                 dev->inflight_info = calloc(1,
1589                                             sizeof(struct inflight_mem_info));
1590                 if (dev->inflight_info == NULL) {
1591                         VHOST_LOG_CONFIG(ERR,
1592                                 "failed to alloc dev inflight area\n");
1593                         return RTE_VHOST_MSG_RESULT_ERR;
1594                 }
1595                 dev->inflight_info->fd = -1;
1596         }
1597
1598         if (dev->inflight_info->addr) {
1599                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1600                 dev->inflight_info->addr = NULL;
1601         }
1602
1603         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1604                     fd, mmap_offset);
1605         if (addr == MAP_FAILED) {
1606                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1607                 return RTE_VHOST_MSG_RESULT_ERR;
1608         }
1609
1610         if (dev->inflight_info->fd >= 0) {
1611                 close(dev->inflight_info->fd);
1612                 dev->inflight_info->fd = -1;
1613         }
1614
1615         dev->inflight_info->fd = fd;
1616         dev->inflight_info->addr = addr;
1617         dev->inflight_info->size = mmap_size;
1618
1619         for (i = 0; i < num_queues; i++) {
1620                 vq = dev->virtqueue[i];
1621                 if (!vq)
1622                         continue;
1623
1624                 if (vq_is_packed(dev)) {
1625                         vq->inflight_packed = addr;
1626                         vq->inflight_packed->desc_num = queue_size;
1627                 } else {
1628                         vq->inflight_split = addr;
1629                         vq->inflight_split->desc_num = queue_size;
1630                 }
1631                 addr = (void *)((char *)addr + pervq_inflight_size);
1632         }
1633
1634         return RTE_VHOST_MSG_RESULT_OK;
1635 }
1636
1637 static int
1638 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1639                         int main_fd __rte_unused)
1640 {
1641         struct virtio_net *dev = *pdev;
1642         struct vhost_vring_file file;
1643         struct vhost_virtqueue *vq;
1644         int expected_fds;
1645
1646         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1647         if (validate_msg_fds(msg, expected_fds) != 0)
1648                 return RTE_VHOST_MSG_RESULT_ERR;
1649
1650         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1651         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1652                 file.fd = VIRTIO_INVALID_EVENTFD;
1653         else
1654                 file.fd = msg->fds[0];
1655         VHOST_LOG_CONFIG(INFO,
1656                 "vring call idx:%d file:%d\n", file.index, file.fd);
1657
1658         vq = dev->virtqueue[file.index];
1659
1660         if (vq->ready) {
1661                 vq->ready = 0;
1662                 vhost_user_notify_queue_state(dev, file.index, 0);
1663         }
1664
1665         if (vq->callfd >= 0)
1666                 close(vq->callfd);
1667
1668         vq->callfd = file.fd;
1669
1670         return RTE_VHOST_MSG_RESULT_OK;
1671 }
1672
1673 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1674                         struct VhostUserMsg *msg,
1675                         int main_fd __rte_unused)
1676 {
1677         int expected_fds;
1678
1679         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1680         if (validate_msg_fds(msg, expected_fds) != 0)
1681                 return RTE_VHOST_MSG_RESULT_ERR;
1682
1683         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1684                 close(msg->fds[0]);
1685         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1686
1687         return RTE_VHOST_MSG_RESULT_OK;
1688 }
1689
1690 static int
1691 resubmit_desc_compare(const void *a, const void *b)
1692 {
1693         const struct rte_vhost_resubmit_desc *desc0 = a;
1694         const struct rte_vhost_resubmit_desc *desc1 = b;
1695
1696         if (desc1->counter > desc0->counter)
1697                 return 1;
1698
1699         return -1;
1700 }
1701
1702 static int
1703 vhost_check_queue_inflights_split(struct virtio_net *dev,
1704                                   struct vhost_virtqueue *vq)
1705 {
1706         uint16_t i;
1707         uint16_t resubmit_num = 0, last_io, num;
1708         struct vring_used *used = vq->used;
1709         struct rte_vhost_resubmit_info *resubmit;
1710         struct rte_vhost_inflight_info_split *inflight_split;
1711
1712         if (!(dev->protocol_features &
1713             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1714                 return RTE_VHOST_MSG_RESULT_OK;
1715
1716         /* The frontend may still not support the inflight feature
1717          * although we negotiate the protocol feature.
1718          */
1719         if ((!vq->inflight_split))
1720                 return RTE_VHOST_MSG_RESULT_OK;
1721
1722         if (!vq->inflight_split->version) {
1723                 vq->inflight_split->version = INFLIGHT_VERSION;
1724                 return RTE_VHOST_MSG_RESULT_OK;
1725         }
1726
1727         if (vq->resubmit_inflight)
1728                 return RTE_VHOST_MSG_RESULT_OK;
1729
1730         inflight_split = vq->inflight_split;
1731         vq->global_counter = 0;
1732         last_io = inflight_split->last_inflight_io;
1733
1734         if (inflight_split->used_idx != used->idx) {
1735                 inflight_split->desc[last_io].inflight = 0;
1736                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1737                 inflight_split->used_idx = used->idx;
1738         }
1739
1740         for (i = 0; i < inflight_split->desc_num; i++) {
1741                 if (inflight_split->desc[i].inflight == 1)
1742                         resubmit_num++;
1743         }
1744
1745         vq->last_avail_idx += resubmit_num;
1746
1747         if (resubmit_num) {
1748                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1749                 if (!resubmit) {
1750                         VHOST_LOG_CONFIG(ERR,
1751                                 "failed to allocate memory for resubmit info.\n");
1752                         return RTE_VHOST_MSG_RESULT_ERR;
1753                 }
1754
1755                 resubmit->resubmit_list = calloc(resubmit_num,
1756                         sizeof(struct rte_vhost_resubmit_desc));
1757                 if (!resubmit->resubmit_list) {
1758                         VHOST_LOG_CONFIG(ERR,
1759                                 "failed to allocate memory for inflight desc.\n");
1760                         free(resubmit);
1761                         return RTE_VHOST_MSG_RESULT_ERR;
1762                 }
1763
1764                 num = 0;
1765                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1766                         if (vq->inflight_split->desc[i].inflight == 1) {
1767                                 resubmit->resubmit_list[num].index = i;
1768                                 resubmit->resubmit_list[num].counter =
1769                                         inflight_split->desc[i].counter;
1770                                 num++;
1771                         }
1772                 }
1773                 resubmit->resubmit_num = num;
1774
1775                 if (resubmit->resubmit_num > 1)
1776                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1777                               sizeof(struct rte_vhost_resubmit_desc),
1778                               resubmit_desc_compare);
1779
1780                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1781                 vq->resubmit_inflight = resubmit;
1782         }
1783
1784         return RTE_VHOST_MSG_RESULT_OK;
1785 }
1786
1787 static int
1788 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1789                                    struct vhost_virtqueue *vq)
1790 {
1791         uint16_t i;
1792         uint16_t resubmit_num = 0, old_used_idx, num;
1793         struct rte_vhost_resubmit_info *resubmit;
1794         struct rte_vhost_inflight_info_packed *inflight_packed;
1795
1796         if (!(dev->protocol_features &
1797             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1798                 return RTE_VHOST_MSG_RESULT_OK;
1799
1800         /* The frontend may still not support the inflight feature
1801          * although we negotiate the protocol feature.
1802          */
1803         if ((!vq->inflight_packed))
1804                 return RTE_VHOST_MSG_RESULT_OK;
1805
1806         if (!vq->inflight_packed->version) {
1807                 vq->inflight_packed->version = INFLIGHT_VERSION;
1808                 return RTE_VHOST_MSG_RESULT_OK;
1809         }
1810
1811         if (vq->resubmit_inflight)
1812                 return RTE_VHOST_MSG_RESULT_OK;
1813
1814         inflight_packed = vq->inflight_packed;
1815         vq->global_counter = 0;
1816         old_used_idx = inflight_packed->old_used_idx;
1817
1818         if (inflight_packed->used_idx != old_used_idx) {
1819                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1820                         inflight_packed->old_used_idx =
1821                                 inflight_packed->used_idx;
1822                         inflight_packed->old_used_wrap_counter =
1823                                 inflight_packed->used_wrap_counter;
1824                         inflight_packed->old_free_head =
1825                                 inflight_packed->free_head;
1826                 } else {
1827                         inflight_packed->used_idx =
1828                                 inflight_packed->old_used_idx;
1829                         inflight_packed->used_wrap_counter =
1830                                 inflight_packed->old_used_wrap_counter;
1831                         inflight_packed->free_head =
1832                                 inflight_packed->old_free_head;
1833                 }
1834         }
1835
1836         for (i = 0; i < inflight_packed->desc_num; i++) {
1837                 if (inflight_packed->desc[i].inflight == 1)
1838                         resubmit_num++;
1839         }
1840
1841         if (resubmit_num) {
1842                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1843                 if (resubmit == NULL) {
1844                         VHOST_LOG_CONFIG(ERR,
1845                                 "failed to allocate memory for resubmit info.\n");
1846                         return RTE_VHOST_MSG_RESULT_ERR;
1847                 }
1848
1849                 resubmit->resubmit_list = calloc(resubmit_num,
1850                         sizeof(struct rte_vhost_resubmit_desc));
1851                 if (resubmit->resubmit_list == NULL) {
1852                         VHOST_LOG_CONFIG(ERR,
1853                                 "failed to allocate memory for resubmit desc.\n");
1854                         free(resubmit);
1855                         return RTE_VHOST_MSG_RESULT_ERR;
1856                 }
1857
1858                 num = 0;
1859                 for (i = 0; i < inflight_packed->desc_num; i++) {
1860                         if (vq->inflight_packed->desc[i].inflight == 1) {
1861                                 resubmit->resubmit_list[num].index = i;
1862                                 resubmit->resubmit_list[num].counter =
1863                                         inflight_packed->desc[i].counter;
1864                                 num++;
1865                         }
1866                 }
1867                 resubmit->resubmit_num = num;
1868
1869                 if (resubmit->resubmit_num > 1)
1870                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1871                               sizeof(struct rte_vhost_resubmit_desc),
1872                               resubmit_desc_compare);
1873
1874                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1875                 vq->resubmit_inflight = resubmit;
1876         }
1877
1878         return RTE_VHOST_MSG_RESULT_OK;
1879 }
1880
1881 static int
1882 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1883                         int main_fd __rte_unused)
1884 {
1885         struct virtio_net *dev = *pdev;
1886         struct vhost_vring_file file;
1887         struct vhost_virtqueue *vq;
1888         int expected_fds;
1889
1890         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1891         if (validate_msg_fds(msg, expected_fds) != 0)
1892                 return RTE_VHOST_MSG_RESULT_ERR;
1893
1894         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1895         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1896                 file.fd = VIRTIO_INVALID_EVENTFD;
1897         else
1898                 file.fd = msg->fds[0];
1899         VHOST_LOG_CONFIG(INFO,
1900                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1901
1902         /* Interpret ring addresses only when ring is started. */
1903         dev = translate_ring_addresses(dev, file.index);
1904         if (!dev) {
1905                 if (file.fd != VIRTIO_INVALID_EVENTFD)
1906                         close(file.fd);
1907
1908                 return RTE_VHOST_MSG_RESULT_ERR;
1909         }
1910
1911         *pdev = dev;
1912
1913         vq = dev->virtqueue[file.index];
1914
1915         /*
1916          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1917          * the ring starts already enabled. Otherwise, it is enabled via
1918          * the SET_VRING_ENABLE message.
1919          */
1920         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1921                 vq->enabled = 1;
1922                 if (dev->notify_ops->vring_state_changed)
1923                         dev->notify_ops->vring_state_changed(
1924                                 dev->vid, file.index, 1);
1925         }
1926
1927         if (vq->ready) {
1928                 vq->ready = 0;
1929                 vhost_user_notify_queue_state(dev, file.index, 0);
1930         }
1931
1932         if (vq->kickfd >= 0)
1933                 close(vq->kickfd);
1934         vq->kickfd = file.fd;
1935
1936         if (vq_is_packed(dev)) {
1937                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1938                         VHOST_LOG_CONFIG(ERR,
1939                                 "failed to inflights for vq: %d\n", file.index);
1940                         return RTE_VHOST_MSG_RESULT_ERR;
1941                 }
1942         } else {
1943                 if (vhost_check_queue_inflights_split(dev, vq)) {
1944                         VHOST_LOG_CONFIG(ERR,
1945                                 "failed to inflights for vq: %d\n", file.index);
1946                         return RTE_VHOST_MSG_RESULT_ERR;
1947                 }
1948         }
1949
1950         return RTE_VHOST_MSG_RESULT_OK;
1951 }
1952
1953 /*
1954  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1955  */
1956 static int
1957 vhost_user_get_vring_base(struct virtio_net **pdev,
1958                         struct VhostUserMsg *msg,
1959                         int main_fd __rte_unused)
1960 {
1961         struct virtio_net *dev = *pdev;
1962         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1963         uint64_t val;
1964
1965         if (validate_msg_fds(msg, 0) != 0)
1966                 return RTE_VHOST_MSG_RESULT_ERR;
1967
1968         /* We have to stop the queue (virtio) if it is running. */
1969         vhost_destroy_device_notify(dev);
1970
1971         dev->flags &= ~VIRTIO_DEV_READY;
1972         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1973
1974         /* Here we are safe to get the indexes */
1975         if (vq_is_packed(dev)) {
1976                 /*
1977                  * Bit[0:14]: avail index
1978                  * Bit[15]: avail wrap counter
1979                  */
1980                 val = vq->last_avail_idx & 0x7fff;
1981                 val |= vq->avail_wrap_counter << 15;
1982                 msg->payload.state.num = val;
1983         } else {
1984                 msg->payload.state.num = vq->last_avail_idx;
1985         }
1986
1987         VHOST_LOG_CONFIG(INFO,
1988                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1989                 msg->payload.state.num);
1990         /*
1991          * Based on current qemu vhost-user implementation, this message is
1992          * sent and only sent in vhost_vring_stop.
1993          * TODO: cleanup the vring, it isn't usable since here.
1994          */
1995         if (vq->kickfd >= 0)
1996                 close(vq->kickfd);
1997
1998         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1999
2000         if (vq->callfd >= 0)
2001                 close(vq->callfd);
2002
2003         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2004
2005         vq->signalled_used_valid = false;
2006
2007         if (vq_is_packed(dev)) {
2008                 rte_free(vq->shadow_used_packed);
2009                 vq->shadow_used_packed = NULL;
2010         } else {
2011                 rte_free(vq->shadow_used_split);
2012                 vq->shadow_used_split = NULL;
2013
2014                 if (vq->async_pkts_info)
2015                         rte_free(vq->async_pkts_info);
2016                 if (vq->async_descs_split)
2017                         rte_free(vq->async_descs_split);
2018                 vq->async_pkts_info = NULL;
2019                 vq->async_descs_split = NULL;
2020         }
2021
2022         rte_free(vq->batch_copy_elems);
2023         vq->batch_copy_elems = NULL;
2024
2025         msg->size = sizeof(msg->payload.state);
2026         msg->fd_num = 0;
2027
2028         vring_invalidate(dev, vq);
2029
2030         return RTE_VHOST_MSG_RESULT_REPLY;
2031 }
2032
2033 /*
2034  * when virtio queues are ready to work, qemu will send us to
2035  * enable the virtio queue pair.
2036  */
2037 static int
2038 vhost_user_set_vring_enable(struct virtio_net **pdev,
2039                         struct VhostUserMsg *msg,
2040                         int main_fd __rte_unused)
2041 {
2042         struct virtio_net *dev = *pdev;
2043         int enable = (int)msg->payload.state.num;
2044         int index = (int)msg->payload.state.index;
2045
2046         if (validate_msg_fds(msg, 0) != 0)
2047                 return RTE_VHOST_MSG_RESULT_ERR;
2048
2049         VHOST_LOG_CONFIG(INFO,
2050                 "set queue enable: %d to qp idx: %d\n",
2051                 enable, index);
2052
2053         if (enable && dev->virtqueue[index]->async_registered) {
2054                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2055                         VHOST_LOG_CONFIG(ERR, "failed to enable vring. "
2056                         "async inflight packets must be completed first\n");
2057                         return RTE_VHOST_MSG_RESULT_ERR;
2058                 }
2059         }
2060
2061         dev->virtqueue[index]->enabled = enable;
2062
2063         return RTE_VHOST_MSG_RESULT_OK;
2064 }
2065
2066 static int
2067 vhost_user_get_protocol_features(struct virtio_net **pdev,
2068                         struct VhostUserMsg *msg,
2069                         int main_fd __rte_unused)
2070 {
2071         struct virtio_net *dev = *pdev;
2072         uint64_t features, protocol_features;
2073
2074         if (validate_msg_fds(msg, 0) != 0)
2075                 return RTE_VHOST_MSG_RESULT_ERR;
2076
2077         rte_vhost_driver_get_features(dev->ifname, &features);
2078         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2079
2080         msg->payload.u64 = protocol_features;
2081         msg->size = sizeof(msg->payload.u64);
2082         msg->fd_num = 0;
2083
2084         return RTE_VHOST_MSG_RESULT_REPLY;
2085 }
2086
2087 static int
2088 vhost_user_set_protocol_features(struct virtio_net **pdev,
2089                         struct VhostUserMsg *msg,
2090                         int main_fd __rte_unused)
2091 {
2092         struct virtio_net *dev = *pdev;
2093         uint64_t protocol_features = msg->payload.u64;
2094         uint64_t slave_protocol_features = 0;
2095
2096         if (validate_msg_fds(msg, 0) != 0)
2097                 return RTE_VHOST_MSG_RESULT_ERR;
2098
2099         rte_vhost_driver_get_protocol_features(dev->ifname,
2100                         &slave_protocol_features);
2101         if (protocol_features & ~slave_protocol_features) {
2102                 VHOST_LOG_CONFIG(ERR,
2103                         "(%d) received invalid protocol features.\n",
2104                         dev->vid);
2105                 return RTE_VHOST_MSG_RESULT_ERR;
2106         }
2107
2108         dev->protocol_features = protocol_features;
2109         VHOST_LOG_CONFIG(INFO,
2110                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2111                 dev->protocol_features);
2112
2113         return RTE_VHOST_MSG_RESULT_OK;
2114 }
2115
2116 static int
2117 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2118                         int main_fd __rte_unused)
2119 {
2120         struct virtio_net *dev = *pdev;
2121         int fd = msg->fds[0];
2122         uint64_t size, off;
2123         void *addr;
2124
2125         if (validate_msg_fds(msg, 1) != 0)
2126                 return RTE_VHOST_MSG_RESULT_ERR;
2127
2128         if (fd < 0) {
2129                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2130                 return RTE_VHOST_MSG_RESULT_ERR;
2131         }
2132
2133         if (msg->size != sizeof(VhostUserLog)) {
2134                 VHOST_LOG_CONFIG(ERR,
2135                         "invalid log base msg size: %"PRId32" != %d\n",
2136                         msg->size, (int)sizeof(VhostUserLog));
2137                 goto close_msg_fds;
2138         }
2139
2140         size = msg->payload.log.mmap_size;
2141         off  = msg->payload.log.mmap_offset;
2142
2143         /* Check for mmap size and offset overflow. */
2144         if (off >= -size) {
2145                 VHOST_LOG_CONFIG(ERR,
2146                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2147                         off, size);
2148                 goto close_msg_fds;
2149         }
2150
2151         VHOST_LOG_CONFIG(INFO,
2152                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2153                 size, off);
2154
2155         /*
2156          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2157          * fail when offset is not page size aligned.
2158          */
2159         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2160         close(fd);
2161         if (addr == MAP_FAILED) {
2162                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2163                 return RTE_VHOST_MSG_RESULT_ERR;
2164         }
2165
2166         /*
2167          * Free previously mapped log memory on occasionally
2168          * multiple VHOST_USER_SET_LOG_BASE.
2169          */
2170         if (dev->log_addr) {
2171                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2172         }
2173         dev->log_addr = (uint64_t)(uintptr_t)addr;
2174         dev->log_base = dev->log_addr + off;
2175         dev->log_size = size;
2176
2177         /*
2178          * The spec is not clear about it (yet), but QEMU doesn't expect
2179          * any payload in the reply.
2180          */
2181         msg->size = 0;
2182         msg->fd_num = 0;
2183
2184         return RTE_VHOST_MSG_RESULT_REPLY;
2185
2186 close_msg_fds:
2187         close_msg_fds(msg);
2188         return RTE_VHOST_MSG_RESULT_ERR;
2189 }
2190
2191 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2192                         struct VhostUserMsg *msg,
2193                         int main_fd __rte_unused)
2194 {
2195         if (validate_msg_fds(msg, 1) != 0)
2196                 return RTE_VHOST_MSG_RESULT_ERR;
2197
2198         close(msg->fds[0]);
2199         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2200
2201         return RTE_VHOST_MSG_RESULT_OK;
2202 }
2203
2204 /*
2205  * An rarp packet is constructed and broadcasted to notify switches about
2206  * the new location of the migrated VM, so that packets from outside will
2207  * not be lost after migration.
2208  *
2209  * However, we don't actually "send" a rarp packet here, instead, we set
2210  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2211  */
2212 static int
2213 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2214                         int main_fd __rte_unused)
2215 {
2216         struct virtio_net *dev = *pdev;
2217         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2218         struct rte_vdpa_device *vdpa_dev;
2219
2220         if (validate_msg_fds(msg, 0) != 0)
2221                 return RTE_VHOST_MSG_RESULT_ERR;
2222
2223         VHOST_LOG_CONFIG(DEBUG,
2224                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2225                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2226         memcpy(dev->mac.addr_bytes, mac, 6);
2227
2228         /*
2229          * Set the flag to inject a RARP broadcast packet at
2230          * rte_vhost_dequeue_burst().
2231          *
2232          * __ATOMIC_RELEASE ordering is for making sure the mac is
2233          * copied before the flag is set.
2234          */
2235         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2236         vdpa_dev = dev->vdpa_dev;
2237         if (vdpa_dev && vdpa_dev->ops->migration_done)
2238                 vdpa_dev->ops->migration_done(dev->vid);
2239
2240         return RTE_VHOST_MSG_RESULT_OK;
2241 }
2242
2243 static int
2244 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2245                         int main_fd __rte_unused)
2246 {
2247         struct virtio_net *dev = *pdev;
2248
2249         if (validate_msg_fds(msg, 0) != 0)
2250                 return RTE_VHOST_MSG_RESULT_ERR;
2251
2252         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2253                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2254                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2255                                 msg->payload.u64);
2256
2257                 return RTE_VHOST_MSG_RESULT_ERR;
2258         }
2259
2260         dev->mtu = msg->payload.u64;
2261
2262         return RTE_VHOST_MSG_RESULT_OK;
2263 }
2264
2265 static int
2266 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2267                         int main_fd __rte_unused)
2268 {
2269         struct virtio_net *dev = *pdev;
2270         int fd = msg->fds[0];
2271
2272         if (validate_msg_fds(msg, 1) != 0)
2273                 return RTE_VHOST_MSG_RESULT_ERR;
2274
2275         if (fd < 0) {
2276                 VHOST_LOG_CONFIG(ERR,
2277                                 "Invalid file descriptor for slave channel (%d)\n",
2278                                 fd);
2279                 return RTE_VHOST_MSG_RESULT_ERR;
2280         }
2281
2282         if (dev->slave_req_fd >= 0)
2283                 close(dev->slave_req_fd);
2284
2285         dev->slave_req_fd = fd;
2286
2287         return RTE_VHOST_MSG_RESULT_OK;
2288 }
2289
2290 static int
2291 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2292 {
2293         struct vhost_vring_addr *ra;
2294         uint64_t start, end, len;
2295
2296         start = imsg->iova;
2297         end = start + imsg->size;
2298
2299         ra = &vq->ring_addrs;
2300         len = sizeof(struct vring_desc) * vq->size;
2301         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2302                 return 1;
2303
2304         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2305         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2306                 return 1;
2307
2308         len = sizeof(struct vring_used) +
2309                sizeof(struct vring_used_elem) * vq->size;
2310         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2311                 return 1;
2312
2313         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2314                 len = sizeof(uint64_t);
2315                 if (ra->log_guest_addr < end &&
2316                     (ra->log_guest_addr + len) > start)
2317                         return 1;
2318         }
2319
2320         return 0;
2321 }
2322
2323 static int
2324 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2325 {
2326         struct vhost_vring_addr *ra;
2327         uint64_t start, end, len;
2328
2329         start = imsg->iova;
2330         end = start + imsg->size;
2331
2332         ra = &vq->ring_addrs;
2333         len = sizeof(struct vring_packed_desc) * vq->size;
2334         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2335                 return 1;
2336
2337         len = sizeof(struct vring_packed_desc_event);
2338         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2339                 return 1;
2340
2341         len = sizeof(struct vring_packed_desc_event);
2342         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2343                 return 1;
2344
2345         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2346                 len = sizeof(uint64_t);
2347                 if (ra->log_guest_addr < end &&
2348                     (ra->log_guest_addr + len) > start)
2349                         return 1;
2350         }
2351
2352         return 0;
2353 }
2354
2355 static int is_vring_iotlb(struct virtio_net *dev,
2356                           struct vhost_virtqueue *vq,
2357                           struct vhost_iotlb_msg *imsg)
2358 {
2359         if (vq_is_packed(dev))
2360                 return is_vring_iotlb_packed(vq, imsg);
2361         else
2362                 return is_vring_iotlb_split(vq, imsg);
2363 }
2364
2365 static int
2366 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2367                         int main_fd __rte_unused)
2368 {
2369         struct virtio_net *dev = *pdev;
2370         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2371         uint16_t i;
2372         uint64_t vva, len;
2373
2374         if (validate_msg_fds(msg, 0) != 0)
2375                 return RTE_VHOST_MSG_RESULT_ERR;
2376
2377         switch (imsg->type) {
2378         case VHOST_IOTLB_UPDATE:
2379                 len = imsg->size;
2380                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2381                 if (!vva)
2382                         return RTE_VHOST_MSG_RESULT_ERR;
2383
2384                 for (i = 0; i < dev->nr_vring; i++) {
2385                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2386
2387                         if (!vq)
2388                                 continue;
2389
2390                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2391                                         len, imsg->perm);
2392
2393                         if (is_vring_iotlb(dev, vq, imsg))
2394                                 *pdev = dev = translate_ring_addresses(dev, i);
2395                 }
2396                 break;
2397         case VHOST_IOTLB_INVALIDATE:
2398                 for (i = 0; i < dev->nr_vring; i++) {
2399                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2400
2401                         if (!vq)
2402                                 continue;
2403
2404                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2405                                         imsg->size);
2406
2407                         if (is_vring_iotlb(dev, vq, imsg))
2408                                 vring_invalidate(dev, vq);
2409                 }
2410                 break;
2411         default:
2412                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2413                                 imsg->type);
2414                 return RTE_VHOST_MSG_RESULT_ERR;
2415         }
2416
2417         return RTE_VHOST_MSG_RESULT_OK;
2418 }
2419
2420 static int
2421 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2422                         struct VhostUserMsg *msg,
2423                         int main_fd __rte_unused)
2424 {
2425         struct virtio_net *dev = *pdev;
2426 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2427         struct uffdio_api api_struct;
2428
2429         if (validate_msg_fds(msg, 0) != 0)
2430                 return RTE_VHOST_MSG_RESULT_ERR;
2431
2432         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2433
2434         if (dev->postcopy_ufd == -1) {
2435                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2436                         strerror(errno));
2437                 return RTE_VHOST_MSG_RESULT_ERR;
2438         }
2439         api_struct.api = UFFD_API;
2440         api_struct.features = 0;
2441         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2442                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2443                         strerror(errno));
2444                 close(dev->postcopy_ufd);
2445                 dev->postcopy_ufd = -1;
2446                 return RTE_VHOST_MSG_RESULT_ERR;
2447         }
2448         msg->fds[0] = dev->postcopy_ufd;
2449         msg->fd_num = 1;
2450
2451         return RTE_VHOST_MSG_RESULT_REPLY;
2452 #else
2453         dev->postcopy_ufd = -1;
2454         msg->fd_num = 0;
2455
2456         return RTE_VHOST_MSG_RESULT_ERR;
2457 #endif
2458 }
2459
2460 static int
2461 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2462                         struct VhostUserMsg *msg __rte_unused,
2463                         int main_fd __rte_unused)
2464 {
2465         struct virtio_net *dev = *pdev;
2466
2467         if (validate_msg_fds(msg, 0) != 0)
2468                 return RTE_VHOST_MSG_RESULT_ERR;
2469
2470         if (dev->mem && dev->mem->nregions) {
2471                 VHOST_LOG_CONFIG(ERR,
2472                         "Regions already registered at postcopy-listen\n");
2473                 return RTE_VHOST_MSG_RESULT_ERR;
2474         }
2475         dev->postcopy_listening = 1;
2476
2477         return RTE_VHOST_MSG_RESULT_OK;
2478 }
2479
2480 static int
2481 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2482                         int main_fd __rte_unused)
2483 {
2484         struct virtio_net *dev = *pdev;
2485
2486         if (validate_msg_fds(msg, 0) != 0)
2487                 return RTE_VHOST_MSG_RESULT_ERR;
2488
2489         dev->postcopy_listening = 0;
2490         if (dev->postcopy_ufd >= 0) {
2491                 close(dev->postcopy_ufd);
2492                 dev->postcopy_ufd = -1;
2493         }
2494
2495         msg->payload.u64 = 0;
2496         msg->size = sizeof(msg->payload.u64);
2497         msg->fd_num = 0;
2498
2499         return RTE_VHOST_MSG_RESULT_REPLY;
2500 }
2501
2502 static int
2503 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2504                       int main_fd __rte_unused)
2505 {
2506         struct virtio_net *dev = *pdev;
2507
2508         if (validate_msg_fds(msg, 0) != 0)
2509                 return RTE_VHOST_MSG_RESULT_ERR;
2510
2511         msg->payload.u64 = dev->status;
2512         msg->size = sizeof(msg->payload.u64);
2513         msg->fd_num = 0;
2514
2515         return RTE_VHOST_MSG_RESULT_REPLY;
2516 }
2517
2518 static int
2519 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2520                         int main_fd __rte_unused)
2521 {
2522         struct virtio_net *dev = *pdev;
2523
2524         if (validate_msg_fds(msg, 0) != 0)
2525                 return RTE_VHOST_MSG_RESULT_ERR;
2526
2527         /* As per Virtio specification, the device status is 8bits long */
2528         if (msg->payload.u64 > UINT8_MAX) {
2529                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2530                                 msg->payload.u64);
2531                 return RTE_VHOST_MSG_RESULT_ERR;
2532         }
2533
2534         dev->status = msg->payload.u64;
2535
2536         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2537             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2538                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2539                 /*
2540                  * Clear the bit to let the driver know about the feature
2541                  * negotiation failure
2542                  */
2543                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2544         }
2545
2546         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2547                         "\t-RESET: %u\n"
2548                         "\t-ACKNOWLEDGE: %u\n"
2549                         "\t-DRIVER: %u\n"
2550                         "\t-FEATURES_OK: %u\n"
2551                         "\t-DRIVER_OK: %u\n"
2552                         "\t-DEVICE_NEED_RESET: %u\n"
2553                         "\t-FAILED: %u\n",
2554                         dev->status,
2555                         (dev->status == VIRTIO_DEVICE_STATUS_RESET),
2556                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2557                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2558                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2559                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2560                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2561                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2562
2563         return RTE_VHOST_MSG_RESULT_OK;
2564 }
2565
2566 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2567                                         struct VhostUserMsg *msg,
2568                                         int main_fd);
2569 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2570         [VHOST_USER_NONE] = NULL,
2571         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2572         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2573         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2574         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2575         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2576         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2577         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2578         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2579         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2580         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2581         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2582         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2583         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2584         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2585         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2586         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2587         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2588         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2589         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2590         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2591         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2592         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2593         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2594         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2595         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2596         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2597         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2598         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2599         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2600 };
2601
2602 /* return bytes# of read on success or negative val on failure. */
2603 static int
2604 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2605 {
2606         int ret;
2607
2608         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2609                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2610         if (ret <= 0) {
2611                 return ret;
2612         } else if (ret != VHOST_USER_HDR_SIZE) {
2613                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2614                 close_msg_fds(msg);
2615                 return -1;
2616         }
2617
2618         if (msg->size) {
2619                 if (msg->size > sizeof(msg->payload)) {
2620                         VHOST_LOG_CONFIG(ERR,
2621                                 "invalid msg size: %d\n", msg->size);
2622                         return -1;
2623                 }
2624                 ret = read(sockfd, &msg->payload, msg->size);
2625                 if (ret <= 0)
2626                         return ret;
2627                 if (ret != (int)msg->size) {
2628                         VHOST_LOG_CONFIG(ERR,
2629                                 "read control message failed\n");
2630                         return -1;
2631                 }
2632         }
2633
2634         return ret;
2635 }
2636
2637 static int
2638 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2639 {
2640         if (!msg)
2641                 return 0;
2642
2643         return send_fd_message(sockfd, (char *)msg,
2644                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2645 }
2646
2647 static int
2648 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2649 {
2650         if (!msg)
2651                 return 0;
2652
2653         msg->flags &= ~VHOST_USER_VERSION_MASK;
2654         msg->flags &= ~VHOST_USER_NEED_REPLY;
2655         msg->flags |= VHOST_USER_VERSION;
2656         msg->flags |= VHOST_USER_REPLY_MASK;
2657
2658         return send_vhost_message(sockfd, msg);
2659 }
2660
2661 static int
2662 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2663 {
2664         int ret;
2665
2666         if (msg->flags & VHOST_USER_NEED_REPLY)
2667                 rte_spinlock_lock(&dev->slave_req_lock);
2668
2669         ret = send_vhost_message(dev->slave_req_fd, msg);
2670         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2671                 rte_spinlock_unlock(&dev->slave_req_lock);
2672
2673         return ret;
2674 }
2675
2676 /*
2677  * Allocate a queue pair if it hasn't been allocated yet
2678  */
2679 static int
2680 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2681                         struct VhostUserMsg *msg)
2682 {
2683         uint32_t vring_idx;
2684
2685         switch (msg->request.master) {
2686         case VHOST_USER_SET_VRING_KICK:
2687         case VHOST_USER_SET_VRING_CALL:
2688         case VHOST_USER_SET_VRING_ERR:
2689                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2690                 break;
2691         case VHOST_USER_SET_VRING_NUM:
2692         case VHOST_USER_SET_VRING_BASE:
2693         case VHOST_USER_SET_VRING_ENABLE:
2694                 vring_idx = msg->payload.state.index;
2695                 break;
2696         case VHOST_USER_SET_VRING_ADDR:
2697                 vring_idx = msg->payload.addr.index;
2698                 break;
2699         default:
2700                 return 0;
2701         }
2702
2703         if (vring_idx >= VHOST_MAX_VRING) {
2704                 VHOST_LOG_CONFIG(ERR,
2705                         "invalid vring index: %u\n", vring_idx);
2706                 return -1;
2707         }
2708
2709         if (dev->virtqueue[vring_idx])
2710                 return 0;
2711
2712         return alloc_vring_queue(dev, vring_idx);
2713 }
2714
2715 static void
2716 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2717 {
2718         unsigned int i = 0;
2719         unsigned int vq_num = 0;
2720
2721         while (vq_num < dev->nr_vring) {
2722                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2723
2724                 if (vq) {
2725                         rte_spinlock_lock(&vq->access_lock);
2726                         vq_num++;
2727                 }
2728                 i++;
2729         }
2730 }
2731
2732 static void
2733 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2734 {
2735         unsigned int i = 0;
2736         unsigned int vq_num = 0;
2737
2738         while (vq_num < dev->nr_vring) {
2739                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2740
2741                 if (vq) {
2742                         rte_spinlock_unlock(&vq->access_lock);
2743                         vq_num++;
2744                 }
2745                 i++;
2746         }
2747 }
2748
2749 int
2750 vhost_user_msg_handler(int vid, int fd)
2751 {
2752         struct virtio_net *dev;
2753         struct VhostUserMsg msg;
2754         struct rte_vdpa_device *vdpa_dev;
2755         int ret;
2756         int unlock_required = 0;
2757         bool handled;
2758         int request;
2759         uint32_t i;
2760
2761         dev = get_device(vid);
2762         if (dev == NULL)
2763                 return -1;
2764
2765         if (!dev->notify_ops) {
2766                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2767                 if (!dev->notify_ops) {
2768                         VHOST_LOG_CONFIG(ERR,
2769                                 "failed to get callback ops for driver %s\n",
2770                                 dev->ifname);
2771                         return -1;
2772                 }
2773         }
2774
2775         ret = read_vhost_message(fd, &msg);
2776         if (ret <= 0) {
2777                 if (ret < 0)
2778                         VHOST_LOG_CONFIG(ERR,
2779                                 "vhost read message failed\n");
2780                 else
2781                         VHOST_LOG_CONFIG(INFO,
2782                                 "vhost peer closed\n");
2783
2784                 return -1;
2785         }
2786
2787         ret = 0;
2788         request = msg.request.master;
2789         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2790                         vhost_message_str[request]) {
2791                 if (request != VHOST_USER_IOTLB_MSG)
2792                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2793                                 vhost_message_str[request]);
2794                 else
2795                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2796                                 vhost_message_str[request]);
2797         } else {
2798                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2799         }
2800
2801         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2802         if (ret < 0) {
2803                 VHOST_LOG_CONFIG(ERR,
2804                         "failed to alloc queue\n");
2805                 return -1;
2806         }
2807
2808         /*
2809          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2810          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2811          * and device is destroyed. destroy_device waits for queues to be
2812          * inactive, so it is safe. Otherwise taking the access_lock
2813          * would cause a dead lock.
2814          */
2815         switch (request) {
2816         case VHOST_USER_SET_FEATURES:
2817         case VHOST_USER_SET_PROTOCOL_FEATURES:
2818         case VHOST_USER_SET_OWNER:
2819         case VHOST_USER_SET_MEM_TABLE:
2820         case VHOST_USER_SET_LOG_BASE:
2821         case VHOST_USER_SET_LOG_FD:
2822         case VHOST_USER_SET_VRING_NUM:
2823         case VHOST_USER_SET_VRING_ADDR:
2824         case VHOST_USER_SET_VRING_BASE:
2825         case VHOST_USER_SET_VRING_KICK:
2826         case VHOST_USER_SET_VRING_CALL:
2827         case VHOST_USER_SET_VRING_ERR:
2828         case VHOST_USER_SET_VRING_ENABLE:
2829         case VHOST_USER_SEND_RARP:
2830         case VHOST_USER_NET_SET_MTU:
2831         case VHOST_USER_SET_SLAVE_REQ_FD:
2832                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2833                         vhost_user_lock_all_queue_pairs(dev);
2834                         unlock_required = 1;
2835                 }
2836                 break;
2837         default:
2838                 break;
2839
2840         }
2841
2842         handled = false;
2843         if (dev->extern_ops.pre_msg_handle) {
2844                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2845                                 (void *)&msg);
2846                 switch (ret) {
2847                 case RTE_VHOST_MSG_RESULT_REPLY:
2848                         send_vhost_reply(fd, &msg);
2849                         /* Fall-through */
2850                 case RTE_VHOST_MSG_RESULT_ERR:
2851                 case RTE_VHOST_MSG_RESULT_OK:
2852                         handled = true;
2853                         goto skip_to_post_handle;
2854                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2855                 default:
2856                         break;
2857                 }
2858         }
2859
2860         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2861                 if (!vhost_message_handlers[request])
2862                         goto skip_to_post_handle;
2863                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2864
2865                 switch (ret) {
2866                 case RTE_VHOST_MSG_RESULT_ERR:
2867                         VHOST_LOG_CONFIG(ERR,
2868                                 "Processing %s failed.\n",
2869                                 vhost_message_str[request]);
2870                         handled = true;
2871                         break;
2872                 case RTE_VHOST_MSG_RESULT_OK:
2873                         VHOST_LOG_CONFIG(DEBUG,
2874                                 "Processing %s succeeded.\n",
2875                                 vhost_message_str[request]);
2876                         handled = true;
2877                         break;
2878                 case RTE_VHOST_MSG_RESULT_REPLY:
2879                         VHOST_LOG_CONFIG(DEBUG,
2880                                 "Processing %s succeeded and needs reply.\n",
2881                                 vhost_message_str[request]);
2882                         send_vhost_reply(fd, &msg);
2883                         handled = true;
2884                         break;
2885                 default:
2886                         break;
2887                 }
2888         }
2889
2890 skip_to_post_handle:
2891         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2892                         dev->extern_ops.post_msg_handle) {
2893                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2894                                 (void *)&msg);
2895                 switch (ret) {
2896                 case RTE_VHOST_MSG_RESULT_REPLY:
2897                         send_vhost_reply(fd, &msg);
2898                         /* Fall-through */
2899                 case RTE_VHOST_MSG_RESULT_ERR:
2900                 case RTE_VHOST_MSG_RESULT_OK:
2901                         handled = true;
2902                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2903                 default:
2904                         break;
2905                 }
2906         }
2907
2908         if (unlock_required)
2909                 vhost_user_unlock_all_queue_pairs(dev);
2910
2911         /* If message was not handled at this stage, treat it as an error */
2912         if (!handled) {
2913                 VHOST_LOG_CONFIG(ERR,
2914                         "vhost message (req: %d) was not handled.\n", request);
2915                 close_msg_fds(&msg);
2916                 ret = RTE_VHOST_MSG_RESULT_ERR;
2917         }
2918
2919         /*
2920          * If the request required a reply that was already sent,
2921          * this optional reply-ack won't be sent as the
2922          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2923          */
2924         if (msg.flags & VHOST_USER_NEED_REPLY) {
2925                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2926                 msg.size = sizeof(msg.payload.u64);
2927                 msg.fd_num = 0;
2928                 send_vhost_reply(fd, &msg);
2929         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2930                 VHOST_LOG_CONFIG(ERR,
2931                         "vhost message handling failed.\n");
2932                 return -1;
2933         }
2934
2935         for (i = 0; i < dev->nr_vring; i++) {
2936                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2937                 bool cur_ready = vq_is_ready(dev, vq);
2938
2939                 if (cur_ready != (vq && vq->ready)) {
2940                         vq->ready = cur_ready;
2941                         vhost_user_notify_queue_state(dev, i, cur_ready);
2942                 }
2943         }
2944
2945
2946         if (!virtio_is_ready(dev))
2947                 goto out;
2948
2949         /*
2950          * Virtio is now ready. If not done already, it is time
2951          * to notify the application it can process the rings and
2952          * configure the vDPA device if present.
2953          */
2954
2955         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2956                 if (dev->notify_ops->new_device(dev->vid) == 0)
2957                         dev->flags |= VIRTIO_DEV_RUNNING;
2958         }
2959
2960         vdpa_dev = dev->vdpa_dev;
2961         if (!vdpa_dev)
2962                 goto out;
2963
2964         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2965                 if (vdpa_dev->ops->dev_conf(dev->vid))
2966                         VHOST_LOG_CONFIG(ERR,
2967                                          "Failed to configure vDPA device\n");
2968                 else
2969                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2970         }
2971
2972 out:
2973         return 0;
2974 }
2975
2976 static int process_slave_message_reply(struct virtio_net *dev,
2977                                        const struct VhostUserMsg *msg)
2978 {
2979         struct VhostUserMsg msg_reply;
2980         int ret;
2981
2982         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2983                 return 0;
2984
2985         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
2986         if (ret <= 0) {
2987                 if (ret < 0)
2988                         VHOST_LOG_CONFIG(ERR,
2989                                 "vhost read slave message reply failed\n");
2990                 else
2991                         VHOST_LOG_CONFIG(INFO,
2992                                 "vhost peer closed\n");
2993                 ret = -1;
2994                 goto out;
2995         }
2996
2997         ret = 0;
2998         if (msg_reply.request.slave != msg->request.slave) {
2999                 VHOST_LOG_CONFIG(ERR,
3000                         "Received unexpected msg type (%u), expected %u\n",
3001                         msg_reply.request.slave, msg->request.slave);
3002                 ret = -1;
3003                 goto out;
3004         }
3005
3006         ret = msg_reply.payload.u64 ? -1 : 0;
3007
3008 out:
3009         rte_spinlock_unlock(&dev->slave_req_lock);
3010         return ret;
3011 }
3012
3013 int
3014 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3015 {
3016         int ret;
3017         struct VhostUserMsg msg = {
3018                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3019                 .flags = VHOST_USER_VERSION,
3020                 .size = sizeof(msg.payload.iotlb),
3021                 .payload.iotlb = {
3022                         .iova = iova,
3023                         .perm = perm,
3024                         .type = VHOST_IOTLB_MISS,
3025                 },
3026         };
3027
3028         ret = send_vhost_message(dev->slave_req_fd, &msg);
3029         if (ret < 0) {
3030                 VHOST_LOG_CONFIG(ERR,
3031                                 "Failed to send IOTLB miss message (%d)\n",
3032                                 ret);
3033                 return ret;
3034         }
3035
3036         return 0;
3037 }
3038
3039 static int
3040 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3041 {
3042         int ret;
3043         struct VhostUserMsg msg = {
3044                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3045                 .flags = VHOST_USER_VERSION,
3046                 .size = 0,
3047         };
3048
3049         if (need_reply)
3050                 msg.flags |= VHOST_USER_NEED_REPLY;
3051
3052         ret = send_vhost_slave_message(dev, &msg);
3053         if (ret < 0) {
3054                 VHOST_LOG_CONFIG(ERR,
3055                                 "Failed to send config change (%d)\n",
3056                                 ret);
3057                 return ret;
3058         }
3059
3060         return process_slave_message_reply(dev, &msg);
3061 }
3062
3063 int
3064 rte_vhost_slave_config_change(int vid, bool need_reply)
3065 {
3066         struct virtio_net *dev;
3067
3068         dev = get_device(vid);
3069         if (!dev)
3070                 return -ENODEV;
3071
3072         return vhost_user_slave_config_change(dev, need_reply);
3073 }
3074
3075 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3076                                                     int index, int fd,
3077                                                     uint64_t offset,
3078                                                     uint64_t size)
3079 {
3080         int ret;
3081         struct VhostUserMsg msg = {
3082                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3083                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3084                 .size = sizeof(msg.payload.area),
3085                 .payload.area = {
3086                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3087                         .size = size,
3088                         .offset = offset,
3089                 },
3090         };
3091
3092         if (fd < 0)
3093                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3094         else {
3095                 msg.fds[0] = fd;
3096                 msg.fd_num = 1;
3097         }
3098
3099         ret = send_vhost_slave_message(dev, &msg);
3100         if (ret < 0) {
3101                 VHOST_LOG_CONFIG(ERR,
3102                         "Failed to set host notifier (%d)\n", ret);
3103                 return ret;
3104         }
3105
3106         return process_slave_message_reply(dev, &msg);
3107 }
3108
3109 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3110 {
3111         struct virtio_net *dev;
3112         struct rte_vdpa_device *vdpa_dev;
3113         int vfio_device_fd, ret = 0;
3114         uint64_t offset, size;
3115         unsigned int i, q_start, q_last;
3116
3117         dev = get_device(vid);
3118         if (!dev)
3119                 return -ENODEV;
3120
3121         vdpa_dev = dev->vdpa_dev;
3122         if (vdpa_dev == NULL)
3123                 return -ENODEV;
3124
3125         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3126             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3127             !(dev->protocol_features &
3128                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3129             !(dev->protocol_features &
3130                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3131             !(dev->protocol_features &
3132                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3133                 return -ENOTSUP;
3134
3135         if (qid == RTE_VHOST_QUEUE_ALL) {
3136                 q_start = 0;
3137                 q_last = dev->nr_vring - 1;
3138         } else {
3139                 if (qid >= dev->nr_vring)
3140                         return -EINVAL;
3141                 q_start = qid;
3142                 q_last = qid;
3143         }
3144
3145         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3146         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3147
3148         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3149         if (vfio_device_fd < 0)
3150                 return -ENOTSUP;
3151
3152         if (enable) {
3153                 for (i = q_start; i <= q_last; i++) {
3154                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3155                                         &size) < 0) {
3156                                 ret = -ENOTSUP;
3157                                 goto disable;
3158                         }
3159
3160                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3161                                         vfio_device_fd, offset, size) < 0) {
3162                                 ret = -EFAULT;
3163                                 goto disable;
3164                         }
3165                 }
3166         } else {
3167 disable:
3168                 for (i = q_start; i <= q_last; i++) {
3169                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3170                                         0, 0);
3171                 }
3172         }
3173
3174         return ret;
3175 }