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