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 };
  91
  92 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
  93 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
  94
  95 static void
  96 close_msg_fds(struct VhostUserMsg *msg)
  97 {
  98         int i;
  99
 100         for (i = 0; i < msg->fd_num; i++)
 101                 close(msg->fds[i]);
 102 }
 103
 104 /*
 105  * Ensure the expected number of FDs is received,
 106  * close all FDs and return an error if this is not the case.
 107  */
 108 static int
 109 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
 110 {
 111         if (msg->fd_num == expected_fds)
 112                 return 0;
 113
 114         RTE_LOG(ERR, VHOST_CONFIG,
 115                 " Expect %d FDs for request %s, received %d\n",
 116                 expected_fds,
 117                 vhost_message_str[msg->request.master],
 118                 msg->fd_num);
 119
 120         close_msg_fds(msg);
 121
 122         return -1;
 123 }
 124
 125 static uint64_t
 126 get_blk_size(int fd)
 127 {
 128         struct stat stat;
 129         int ret;
 130
 131         ret = fstat(fd, &stat);
 132         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 133 }
 134
 135 /*
 136  * Reclaim all the outstanding zmbufs for a virtqueue.
 137  */
 138 static void
 139 drain_zmbuf_list(struct vhost_virtqueue *vq)
 140 {
 141         struct zcopy_mbuf *zmbuf, *next;
 142
 143         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 144              zmbuf != NULL; zmbuf = next) {
 145                 next = TAILQ_NEXT(zmbuf, next);
 146
 147                 while (!mbuf_is_consumed(zmbuf->mbuf))
 148                         usleep(1000);
 149
 150                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 151                 restore_mbuf(zmbuf->mbuf);
 152                 rte_pktmbuf_free(zmbuf->mbuf);
 153                 put_zmbuf(zmbuf);
 154                 vq->nr_zmbuf -= 1;
 155         }
 156 }
 157
 158 static void
 159 free_mem_region(struct virtio_net *dev)
 160 {
 161         uint32_t i;
 162         struct rte_vhost_mem_region *reg;
 163         struct vhost_virtqueue *vq;
 164
 165         if (!dev || !dev->mem)
 166                 return;
 167
 168         if (dev->dequeue_zero_copy) {
 169                 for (i = 0; i < dev->nr_vring; i++) {
 170                         vq = dev->virtqueue[i];
 171                         if (vq)
 172                                 drain_zmbuf_list(vq);
 173                 }
 174         }
 175
 176         for (i = 0; i < dev->mem->nregions; i++) {
 177                 reg = &dev->mem->regions[i];
 178                 if (reg->host_user_addr) {
 179                         munmap(reg->mmap_addr, reg->mmap_size);
 180                         close(reg->fd);
 181                 }
 182         }
 183 }
 184
 185 void
 186 vhost_backend_cleanup(struct virtio_net *dev)
 187 {
 188         if (dev->mem) {
 189                 free_mem_region(dev);
 190                 rte_free(dev->mem);
 191                 dev->mem = NULL;
 192         }
 193
 194         free(dev->guest_pages);
 195         dev->guest_pages = NULL;
 196
 197         if (dev->log_addr) {
 198                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 199                 dev->log_addr = 0;
 200         }
 201
 202         if (dev->inflight_info) {
 203                 if (dev->inflight_info->addr) {
 204                         munmap(dev->inflight_info->addr,
 205                                dev->inflight_info->size);
 206                         dev->inflight_info->addr = NULL;
 207                 }
 208
 209                 if (dev->inflight_info->fd > 0) {
 210                         close(dev->inflight_info->fd);
 211                         dev->inflight_info->fd = -1;
 212                 }
 213
 214                 free(dev->inflight_info);
 215                 dev->inflight_info = NULL;
 216         }
 217
 218         if (dev->slave_req_fd >= 0) {
 219                 close(dev->slave_req_fd);
 220                 dev->slave_req_fd = -1;
 221         }
 222
 223         if (dev->postcopy_ufd >= 0) {
 224                 close(dev->postcopy_ufd);
 225                 dev->postcopy_ufd = -1;
 226         }
 227
 228         dev->postcopy_listening = 0;
 229 }
 230
 231 /*
 232  * This function just returns success at the moment unless
 233  * the device hasn't been initialised.
 234  */
 235 static int
 236 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
 237                         struct VhostUserMsg *msg,
 238                         int main_fd __rte_unused)
 239 {
 240         if (validate_msg_fds(msg, 0) != 0)
 241                 return RTE_VHOST_MSG_RESULT_ERR;
 242
 243         return RTE_VHOST_MSG_RESULT_OK;
 244 }
 245
 246 static int
 247 vhost_user_reset_owner(struct virtio_net **pdev,
 248                         struct VhostUserMsg *msg,
 249                         int main_fd __rte_unused)
 250 {
 251         struct virtio_net *dev = *pdev;
 252
 253         if (validate_msg_fds(msg, 0) != 0)
 254                 return RTE_VHOST_MSG_RESULT_ERR;
 255
 256         vhost_destroy_device_notify(dev);
 257
 258         cleanup_device(dev, 0);
 259         reset_device(dev);
 260         return RTE_VHOST_MSG_RESULT_OK;
 261 }
 262
 263 /*
 264  * The features that we support are requested.
 265  */
 266 static int
 267 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 268                         int main_fd __rte_unused)
 269 {
 270         struct virtio_net *dev = *pdev;
 271         uint64_t features = 0;
 272
 273         if (validate_msg_fds(msg, 0) != 0)
 274                 return RTE_VHOST_MSG_RESULT_ERR;
 275
 276         rte_vhost_driver_get_features(dev->ifname, &features);
 277
 278         msg->payload.u64 = features;
 279         msg->size = sizeof(msg->payload.u64);
 280         msg->fd_num = 0;
 281
 282         return RTE_VHOST_MSG_RESULT_REPLY;
 283 }
 284
 285 /*
 286  * The queue number that we support are requested.
 287  */
 288 static int
 289 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
 290                         int main_fd __rte_unused)
 291 {
 292         struct virtio_net *dev = *pdev;
 293         uint32_t queue_num = 0;
 294
 295         if (validate_msg_fds(msg, 0) != 0)
 296                 return RTE_VHOST_MSG_RESULT_ERR;
 297
 298         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 299
 300         msg->payload.u64 = (uint64_t)queue_num;
 301         msg->size = sizeof(msg->payload.u64);
 302         msg->fd_num = 0;
 303
 304         return RTE_VHOST_MSG_RESULT_REPLY;
 305 }
 306
 307 /*
 308  * We receive the negotiated features supported by us and the virtio device.
 309  */
 310 static int
 311 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 312                         int main_fd __rte_unused)
 313 {
 314         struct virtio_net *dev = *pdev;
 315         uint64_t features = msg->payload.u64;
 316         uint64_t vhost_features = 0;
 317         struct rte_vdpa_device *vdpa_dev;
 318         int did = -1;
 319
 320         if (validate_msg_fds(msg, 0) != 0)
 321                 return RTE_VHOST_MSG_RESULT_ERR;
 322
 323         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 324         if (features & ~vhost_features) {
 325                 RTE_LOG(ERR, VHOST_CONFIG,
 326                         "(%d) received invalid negotiated features.\n",
 327                         dev->vid);
 328                 return RTE_VHOST_MSG_RESULT_ERR;
 329         }
 330
 331         if (dev->flags & VIRTIO_DEV_RUNNING) {
 332                 if (dev->features == features)
 333                         return RTE_VHOST_MSG_RESULT_OK;
 334
 335                 /*
 336                  * Error out if master tries to change features while device is
 337                  * in running state. The exception being VHOST_F_LOG_ALL, which
 338                  * is enabled when the live-migration starts.
 339                  */
 340                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 341                         RTE_LOG(ERR, VHOST_CONFIG,
 342                                 "(%d) features changed while device is running.\n",
 343                                 dev->vid);
 344                         return RTE_VHOST_MSG_RESULT_ERR;
 345                 }
 346
 347                 if (dev->notify_ops->features_changed)
 348                         dev->notify_ops->features_changed(dev->vid, features);
 349         }
 350
 351         dev->features = features;
 352         if (dev->features &
 353                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 354                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 355         } else {
 356                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 357         }
 358         RTE_LOG(INFO, VHOST_CONFIG,
 359                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 360         VHOST_LOG_DEBUG(VHOST_CONFIG,
 361                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 362                 dev->vid,
 363                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 364                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 365
 366         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 367             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 368                 /*
 369                  * Remove all but first queue pair if MQ hasn't been
 370                  * negotiated. This is safe because the device is not
 371                  * running at this stage.
 372                  */
 373                 while (dev->nr_vring > 2) {
 374                         struct vhost_virtqueue *vq;
 375
 376                         vq = dev->virtqueue[--dev->nr_vring];
 377                         if (!vq)
 378                                 continue;
 379
 380                         dev->virtqueue[dev->nr_vring] = NULL;
 381                         cleanup_vq(vq, 1);
 382                         cleanup_vq_inflight(dev, vq);
 383                         free_vq(dev, vq);
 384                 }
 385         }
 386
 387         did = dev->vdpa_dev_id;
 388         vdpa_dev = rte_vdpa_get_device(did);
 389         if (vdpa_dev && vdpa_dev->ops->set_features)
 390                 vdpa_dev->ops->set_features(dev->vid);
 391
 392         return RTE_VHOST_MSG_RESULT_OK;
 393 }
 394
 395 /*
 396  * The virtio device sends us the size of the descriptor ring.
 397  */
 398 static int
 399 vhost_user_set_vring_num(struct virtio_net **pdev,
 400                         struct VhostUserMsg *msg,
 401                         int main_fd __rte_unused)
 402 {
 403         struct virtio_net *dev = *pdev;
 404         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 405
 406         if (validate_msg_fds(msg, 0) != 0)
 407                 return RTE_VHOST_MSG_RESULT_ERR;
 408
 409         vq->size = msg->payload.state.num;
 410
 411         /* VIRTIO 1.0, 2.4 Virtqueues says:
 412          *
 413          *   Queue Size value is always a power of 2. The maximum Queue Size
 414          *   value is 32768.
 415          *
 416          * VIRTIO 1.1 2.7 Virtqueues says:
 417          *
 418          *   Packed virtqueues support up to 2^15 entries each.
 419          */
 420         if (!vq_is_packed(dev)) {
 421                 if (vq->size & (vq->size - 1)) {
 422                         RTE_LOG(ERR, VHOST_CONFIG,
 423                                 "invalid virtqueue size %u\n", vq->size);
 424                         return RTE_VHOST_MSG_RESULT_ERR;
 425                 }
 426         }
 427
 428         if (vq->size > 32768) {
 429                 RTE_LOG(ERR, VHOST_CONFIG,
 430                         "invalid virtqueue size %u\n", vq->size);
 431                 return RTE_VHOST_MSG_RESULT_ERR;
 432         }
 433
 434         if (dev->dequeue_zero_copy) {
 435                 vq->nr_zmbuf = 0;
 436                 vq->last_zmbuf_idx = 0;
 437                 vq->zmbuf_size = vq->size;
 438                 if (vq->zmbufs)
 439                         rte_free(vq->zmbufs);
 440                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 441                                          sizeof(struct zcopy_mbuf), 0);
 442                 if (vq->zmbufs == NULL) {
 443                         RTE_LOG(WARNING, VHOST_CONFIG,
 444                                 "failed to allocate mem for zero copy; "
 445                                 "zero copy is force disabled\n");
 446                         dev->dequeue_zero_copy = 0;
 447                 }
 448                 TAILQ_INIT(&vq->zmbuf_list);
 449         }
 450
 451         if (vq_is_packed(dev)) {
 452                 if (vq->shadow_used_packed)
 453                         rte_free(vq->shadow_used_packed);
 454                 vq->shadow_used_packed = rte_malloc(NULL,
 455                                 vq->size *
 456                                 sizeof(struct vring_used_elem_packed),
 457                                 RTE_CACHE_LINE_SIZE);
 458                 if (!vq->shadow_used_packed) {
 459                         RTE_LOG(ERR, VHOST_CONFIG,
 460                                         "failed to allocate memory for shadow used ring.\n");
 461                         return RTE_VHOST_MSG_RESULT_ERR;
 462                 }
 463
 464         } else {
 465                 if (vq->shadow_used_split)
 466                         rte_free(vq->shadow_used_split);
 467                 vq->shadow_used_split = rte_malloc(NULL,
 468                                 vq->size * sizeof(struct vring_used_elem),
 469                                 RTE_CACHE_LINE_SIZE);
 470                 if (!vq->shadow_used_split) {
 471                         RTE_LOG(ERR, VHOST_CONFIG,
 472                                         "failed to allocate memory for shadow used ring.\n");
 473                         return RTE_VHOST_MSG_RESULT_ERR;
 474                 }
 475         }
 476
 477         if (vq->batch_copy_elems)
 478                 rte_free(vq->batch_copy_elems);
 479         vq->batch_copy_elems = rte_malloc(NULL,
 480                                 vq->size * sizeof(struct batch_copy_elem),
 481                                 RTE_CACHE_LINE_SIZE);
 482         if (!vq->batch_copy_elems) {
 483                 RTE_LOG(ERR, VHOST_CONFIG,
 484                         "failed to allocate memory for batching copy.\n");
 485                 return RTE_VHOST_MSG_RESULT_ERR;
 486         }
 487
 488         return RTE_VHOST_MSG_RESULT_OK;
 489 }
 490
 491 /*
 492  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 493  * same numa node as the memory of vring descriptor.
 494  */
 495 #ifdef RTE_LIBRTE_VHOST_NUMA
 496 static struct virtio_net*
 497 numa_realloc(struct virtio_net *dev, int index)
 498 {
 499         int oldnode, newnode;
 500         struct virtio_net *old_dev;
 501         struct vhost_virtqueue *old_vq, *vq;
 502         struct zcopy_mbuf *new_zmbuf;
 503         struct vring_used_elem *new_shadow_used_split;
 504         struct vring_used_elem_packed *new_shadow_used_packed;
 505         struct batch_copy_elem *new_batch_copy_elems;
 506         int ret;
 507
 508         if (dev->flags & VIRTIO_DEV_RUNNING)
 509                 return dev;
 510
 511         old_dev = dev;
 512         vq = old_vq = dev->virtqueue[index];
 513
 514         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 515                             MPOL_F_NODE | MPOL_F_ADDR);
 516
 517         /* check if we need to reallocate vq */
 518         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 519                              MPOL_F_NODE | MPOL_F_ADDR);
 520         if (ret) {
 521                 RTE_LOG(ERR, VHOST_CONFIG,
 522                         "Unable to get vq numa information.\n");
 523                 return dev;
 524         }
 525         if (oldnode != newnode) {
 526                 RTE_LOG(INFO, VHOST_CONFIG,
 527                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 528                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 529                 if (!vq)
 530                         return dev;
 531
 532                 memcpy(vq, old_vq, sizeof(*vq));
 533                 TAILQ_INIT(&vq->zmbuf_list);
 534
 535                 if (dev->dequeue_zero_copy) {
 536                         new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 537                                         sizeof(struct zcopy_mbuf), 0, newnode);
 538                         if (new_zmbuf) {
 539                                 rte_free(vq->zmbufs);
 540                                 vq->zmbufs = new_zmbuf;
 541                         }
 542                 }
 543
 544                 if (vq_is_packed(dev)) {
 545                         new_shadow_used_packed = rte_malloc_socket(NULL,
 546                                         vq->size *
 547                                         sizeof(struct vring_used_elem_packed),
 548                                         RTE_CACHE_LINE_SIZE,
 549                                         newnode);
 550                         if (new_shadow_used_packed) {
 551                                 rte_free(vq->shadow_used_packed);
 552                                 vq->shadow_used_packed = new_shadow_used_packed;
 553                         }
 554                 } else {
 555                         new_shadow_used_split = rte_malloc_socket(NULL,
 556                                         vq->size *
 557                                         sizeof(struct vring_used_elem),
 558                                         RTE_CACHE_LINE_SIZE,
 559                                         newnode);
 560                         if (new_shadow_used_split) {
 561                                 rte_free(vq->shadow_used_split);
 562                                 vq->shadow_used_split = new_shadow_used_split;
 563                         }
 564                 }
 565
 566                 new_batch_copy_elems = rte_malloc_socket(NULL,
 567                         vq->size * sizeof(struct batch_copy_elem),
 568                         RTE_CACHE_LINE_SIZE,
 569                         newnode);
 570                 if (new_batch_copy_elems) {
 571                         rte_free(vq->batch_copy_elems);
 572                         vq->batch_copy_elems = new_batch_copy_elems;
 573                 }
 574
 575                 rte_free(old_vq);
 576         }
 577
 578         /* check if we need to reallocate dev */
 579         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 580                             MPOL_F_NODE | MPOL_F_ADDR);
 581         if (ret) {
 582                 RTE_LOG(ERR, VHOST_CONFIG,
 583                         "Unable to get dev numa information.\n");
 584                 goto out;
 585         }
 586         if (oldnode != newnode) {
 587                 RTE_LOG(INFO, VHOST_CONFIG,
 588                         "reallocate dev from %d to %d node\n",
 589                         oldnode, newnode);
 590                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 591                 if (!dev) {
 592                         dev = old_dev;
 593                         goto out;
 594                 }
 595
 596                 memcpy(dev, old_dev, sizeof(*dev));
 597                 rte_free(old_dev);
 598         }
 599
 600 out:
 601         dev->virtqueue[index] = vq;
 602         vhost_devices[dev->vid] = dev;
 603
 604         if (old_vq != vq)
 605                 vhost_user_iotlb_init(dev, index);
 606
 607         return dev;
 608 }
 609 #else
 610 static struct virtio_net*
 611 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 612 {
 613         return dev;
 614 }
 615 #endif
 616
 617 /* Converts QEMU virtual address to Vhost virtual address. */
 618 static uint64_t
 619 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 620 {
 621         struct rte_vhost_mem_region *r;
 622         uint32_t i;
 623
 624         if (unlikely(!dev || !dev->mem))
 625                 goto out_error;
 626
 627         /* Find the region where the address lives. */
 628         for (i = 0; i < dev->mem->nregions; i++) {
 629                 r = &dev->mem->regions[i];
 630
 631                 if (qva >= r->guest_user_addr &&
 632                     qva <  r->guest_user_addr + r->size) {
 633
 634                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 635                                 *len = r->guest_user_addr + r->size - qva;
 636
 637                         return qva - r->guest_user_addr +
 638                                r->host_user_addr;
 639                 }
 640         }
 641 out_error:
 642         *len = 0;
 643
 644         return 0;
 645 }
 646
 647
 648 /*
 649  * Converts ring address to Vhost virtual address.
 650  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 651  * else it is a QEMU virtual address.
 652  */
 653 static uint64_t
 654 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 655                 uint64_t ra, uint64_t *size)
 656 {
 657         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 658                 uint64_t vva;
 659                 uint64_t req_size = *size;
 660
 661                 vva = vhost_user_iotlb_cache_find(vq, ra,
 662                                         size, VHOST_ACCESS_RW);
 663                 if (req_size != *size)
 664                         vhost_user_iotlb_miss(dev, (ra + *size),
 665                                               VHOST_ACCESS_RW);
 666
 667                 return vva;
 668         }
 669
 670         return qva_to_vva(dev, ra, size);
 671 }
 672
 673 /*
 674  * Converts vring log address to GPA
 675  * If IOMMU is enabled, the log address is IOVA
 676  * If IOMMU not enabled, the log address is already GPA
 677  */
 678 static uint64_t
 679 translate_log_addr(struct virtio_net *dev, struct vhost_virtqueue *vq,
 680                 uint64_t log_addr)
 681 {
 682         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 683                 const uint64_t exp_size = sizeof(struct vring_used) +
 684                         sizeof(struct vring_used_elem) * vq->size;
 685                 uint64_t hva, gpa;
 686                 uint64_t size = exp_size;
 687
 688                 hva = vhost_iova_to_vva(dev, vq, log_addr,
 689                                         &size, VHOST_ACCESS_RW);
 690                 if (size != exp_size)
 691                         return 0;
 692
 693                 gpa = hva_to_gpa(dev, hva, exp_size);
 694                 if (!gpa) {
 695                         RTE_LOG(ERR, VHOST_CONFIG,
 696                                 "VQ: Failed to find GPA for log_addr: 0x%" PRIx64 " hva: 0x%" PRIx64 "\n",
 697                                 log_addr, hva);
 698                         return 0;
 699                 }
 700                 return gpa;
 701
 702         } else
 703                 return log_addr;
 704 }
 705
 706 static struct virtio_net *
 707 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 708 {
 709         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 710         struct vhost_vring_addr *addr = &vq->ring_addrs;
 711         uint64_t len, expected_len;
 712
 713         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 714                 vq->log_guest_addr =
 715                         translate_log_addr(dev, vq, addr->log_guest_addr);
 716                 if (vq->log_guest_addr == 0) {
 717                         RTE_LOG(DEBUG, VHOST_CONFIG,
 718                                 "(%d) failed to map log_guest_addr.\n",
 719                                 dev->vid);
 720                         return dev;
 721                 }
 722         }
 723
 724         if (vq_is_packed(dev)) {
 725                 len = sizeof(struct vring_packed_desc) * vq->size;
 726                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 727                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 728                 if (vq->desc_packed == NULL ||
 729                                 len != sizeof(struct vring_packed_desc) *
 730                                 vq->size) {
 731                         RTE_LOG(DEBUG, VHOST_CONFIG,
 732                                 "(%d) failed to map desc_packed ring.\n",
 733                                 dev->vid);
 734                         return dev;
 735                 }
 736
 737                 dev = numa_realloc(dev, vq_index);
 738                 vq = dev->virtqueue[vq_index];
 739                 addr = &vq->ring_addrs;
 740
 741                 len = sizeof(struct vring_packed_desc_event);
 742                 vq->driver_event = (struct vring_packed_desc_event *)
 743                                         (uintptr_t)ring_addr_to_vva(dev,
 744                                         vq, addr->avail_user_addr, &len);
 745                 if (vq->driver_event == NULL ||
 746                                 len != sizeof(struct vring_packed_desc_event)) {
 747                         RTE_LOG(DEBUG, VHOST_CONFIG,
 748                                 "(%d) failed to find driver area address.\n",
 749                                 dev->vid);
 750                         return dev;
 751                 }
 752
 753                 len = sizeof(struct vring_packed_desc_event);
 754                 vq->device_event = (struct vring_packed_desc_event *)
 755                                         (uintptr_t)ring_addr_to_vva(dev,
 756                                         vq, addr->used_user_addr, &len);
 757                 if (vq->device_event == NULL ||
 758                                 len != sizeof(struct vring_packed_desc_event)) {
 759                         RTE_LOG(DEBUG, VHOST_CONFIG,
 760                                 "(%d) failed to find device area address.\n",
 761                                 dev->vid);
 762                         return dev;
 763                 }
 764
 765                 vq->access_ok = 1;
 766                 return dev;
 767         }
 768
 769         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 770         if (vq->desc && vq->avail && vq->used)
 771                 return dev;
 772
 773         len = sizeof(struct vring_desc) * vq->size;
 774         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 775                         vq, addr->desc_user_addr, &len);
 776         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 777                 RTE_LOG(DEBUG, VHOST_CONFIG,
 778                         "(%d) failed to map desc ring.\n",
 779                         dev->vid);
 780                 return dev;
 781         }
 782
 783         dev = numa_realloc(dev, vq_index);
 784         vq = dev->virtqueue[vq_index];
 785         addr = &vq->ring_addrs;
 786
 787         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 788         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 789                 len += sizeof(uint16_t);
 790         expected_len = len;
 791         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 792                         vq, addr->avail_user_addr, &len);
 793         if (vq->avail == 0 || len != expected_len) {
 794                 RTE_LOG(DEBUG, VHOST_CONFIG,
 795                         "(%d) failed to map avail ring.\n",
 796                         dev->vid);
 797                 return dev;
 798         }
 799
 800         len = sizeof(struct vring_used) +
 801                 sizeof(struct vring_used_elem) * vq->size;
 802         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 803                 len += sizeof(uint16_t);
 804         expected_len = len;
 805         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 806                         vq, addr->used_user_addr, &len);
 807         if (vq->used == 0 || len != expected_len) {
 808                 RTE_LOG(DEBUG, VHOST_CONFIG,
 809                         "(%d) failed to map used ring.\n",
 810                         dev->vid);
 811                 return dev;
 812         }
 813
 814         if (vq->last_used_idx != vq->used->idx) {
 815                 RTE_LOG(WARNING, VHOST_CONFIG,
 816                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 817                         "some packets maybe resent for Tx and dropped for Rx\n",
 818                         vq->last_used_idx, vq->used->idx);
 819                 vq->last_used_idx  = vq->used->idx;
 820                 vq->last_avail_idx = vq->used->idx;
 821         }
 822
 823         vq->access_ok = 1;
 824
 825         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 826                         dev->vid, vq->desc);
 827         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 828                         dev->vid, vq->avail);
 829         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 830                         dev->vid, vq->used);
 831         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 832                         dev->vid, vq->log_guest_addr);
 833
 834         return dev;
 835 }
 836
 837 /*
 838  * The virtio device sends us the desc, used and avail ring addresses.
 839  * This function then converts these to our address space.
 840  */
 841 static int
 842 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 843                         int main_fd __rte_unused)
 844 {
 845         struct virtio_net *dev = *pdev;
 846         struct vhost_virtqueue *vq;
 847         struct vhost_vring_addr *addr = &msg->payload.addr;
 848         bool access_ok;
 849
 850         if (validate_msg_fds(msg, 0) != 0)
 851                 return RTE_VHOST_MSG_RESULT_ERR;
 852
 853         if (dev->mem == NULL)
 854                 return RTE_VHOST_MSG_RESULT_ERR;
 855
 856         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 857         vq = dev->virtqueue[msg->payload.addr.index];
 858
 859         access_ok = vq->access_ok;
 860
 861         /*
 862          * Rings addresses should not be interpreted as long as the ring is not
 863          * started and enabled
 864          */
 865         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 866
 867         vring_invalidate(dev, vq);
 868
 869         if ((vq->enabled && (dev->features &
 870                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 871                         access_ok) {
 872                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 873                 if (!dev)
 874                         return RTE_VHOST_MSG_RESULT_ERR;
 875
 876                 *pdev = dev;
 877         }
 878
 879         return RTE_VHOST_MSG_RESULT_OK;
 880 }
 881
 882 /*
 883  * The virtio device sends us the available ring last used index.
 884  */
 885 static int
 886 vhost_user_set_vring_base(struct virtio_net **pdev,
 887                         struct VhostUserMsg *msg,
 888                         int main_fd __rte_unused)
 889 {
 890         struct virtio_net *dev = *pdev;
 891         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 892         uint64_t val = msg->payload.state.num;
 893
 894         if (validate_msg_fds(msg, 0) != 0)
 895                 return RTE_VHOST_MSG_RESULT_ERR;
 896
 897         if (vq_is_packed(dev)) {
 898                 /*
 899                  * Bit[0:14]: avail index
 900                  * Bit[15]: avail wrap counter
 901                  */
 902                 vq->last_avail_idx = val & 0x7fff;
 903                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 904                 /*
 905                  * Set used index to same value as available one, as
 906                  * their values should be the same since ring processing
 907                  * was stopped at get time.
 908                  */
 909                 vq->last_used_idx = vq->last_avail_idx;
 910                 vq->used_wrap_counter = vq->avail_wrap_counter;
 911         } else {
 912                 vq->last_used_idx = msg->payload.state.num;
 913                 vq->last_avail_idx = msg->payload.state.num;
 914         }
 915
 916         return RTE_VHOST_MSG_RESULT_OK;
 917 }
 918
 919 static int
 920 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 921                    uint64_t host_phys_addr, uint64_t size)
 922 {
 923         struct guest_page *page, *last_page;
 924         struct guest_page *old_pages;
 925
 926         if (dev->nr_guest_pages == dev->max_guest_pages) {
 927                 dev->max_guest_pages *= 2;
 928                 old_pages = dev->guest_pages;
 929                 dev->guest_pages = realloc(dev->guest_pages,
 930                                         dev->max_guest_pages * sizeof(*page));
 931                 if (!dev->guest_pages) {
 932                         RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
 933                         free(old_pages);
 934                         return -1;
 935                 }
 936         }
 937
 938         if (dev->nr_guest_pages > 0) {
 939                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 940                 /* merge if the two pages are continuous */
 941                 if (host_phys_addr == last_page->host_phys_addr +
 942                                       last_page->size) {
 943                         last_page->size += size;
 944                         return 0;
 945                 }
 946         }
 947
 948         page = &dev->guest_pages[dev->nr_guest_pages++];
 949         page->guest_phys_addr = guest_phys_addr;
 950         page->host_phys_addr  = host_phys_addr;
 951         page->size = size;
 952
 953         return 0;
 954 }
 955
 956 static int
 957 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 958                 uint64_t page_size)
 959 {
 960         uint64_t reg_size = reg->size;
 961         uint64_t host_user_addr  = reg->host_user_addr;
 962         uint64_t guest_phys_addr = reg->guest_phys_addr;
 963         uint64_t host_phys_addr;
 964         uint64_t size;
 965
 966         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 967         size = page_size - (guest_phys_addr & (page_size - 1));
 968         size = RTE_MIN(size, reg_size);
 969
 970         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 971                 return -1;
 972
 973         host_user_addr  += size;
 974         guest_phys_addr += size;
 975         reg_size -= size;
 976
 977         while (reg_size > 0) {
 978                 size = RTE_MIN(reg_size, page_size);
 979                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 980                                                   host_user_addr);
 981                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 982                                 size) < 0)
 983                         return -1;
 984
 985                 host_user_addr  += size;
 986                 guest_phys_addr += size;
 987                 reg_size -= size;
 988         }
 989
 990         return 0;
 991 }
 992
 993 #ifdef RTE_LIBRTE_VHOST_DEBUG
 994 /* TODO: enable it only in debug mode? */
 995 static void
 996 dump_guest_pages(struct virtio_net *dev)
 997 {
 998         uint32_t i;
 999         struct guest_page *page;
1000
1001         for (i = 0; i < dev->nr_guest_pages; i++) {
1002                 page = &dev->guest_pages[i];
1003
1004                 RTE_LOG(INFO, VHOST_CONFIG,
1005                         "guest physical page region %u\n"
1006                         "\t guest_phys_addr: %" PRIx64 "\n"
1007                         "\t host_phys_addr : %" PRIx64 "\n"
1008                         "\t size           : %" PRIx64 "\n",
1009                         i,
1010                         page->guest_phys_addr,
1011                         page->host_phys_addr,
1012                         page->size);
1013         }
1014 }
1015 #else
1016 #define dump_guest_pages(dev)
1017 #endif
1018
1019 static bool
1020 vhost_memory_changed(struct VhostUserMemory *new,
1021                      struct rte_vhost_memory *old)
1022 {
1023         uint32_t i;
1024
1025         if (new->nregions != old->nregions)
1026                 return true;
1027
1028         for (i = 0; i < new->nregions; ++i) {
1029                 VhostUserMemoryRegion *new_r = &new->regions[i];
1030                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1031
1032                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1033                         return true;
1034                 if (new_r->memory_size != old_r->size)
1035                         return true;
1036                 if (new_r->userspace_addr != old_r->guest_user_addr)
1037                         return true;
1038         }
1039
1040         return false;
1041 }
1042
1043 static int
1044 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1045                         int main_fd)
1046 {
1047         struct virtio_net *dev = *pdev;
1048         struct VhostUserMemory *memory = &msg->payload.memory;
1049         struct rte_vhost_mem_region *reg;
1050         void *mmap_addr;
1051         uint64_t mmap_size;
1052         uint64_t mmap_offset;
1053         uint64_t alignment;
1054         uint32_t i;
1055         int populate;
1056         int fd;
1057
1058         if (validate_msg_fds(msg, memory->nregions) != 0)
1059                 return RTE_VHOST_MSG_RESULT_ERR;
1060
1061         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1062                 RTE_LOG(ERR, VHOST_CONFIG,
1063                         "too many memory regions (%u)\n", memory->nregions);
1064                 return RTE_VHOST_MSG_RESULT_ERR;
1065         }
1066
1067         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1068                 RTE_LOG(INFO, VHOST_CONFIG,
1069                         "(%d) memory regions not changed\n", dev->vid);
1070
1071                 close_msg_fds(msg);
1072
1073                 return RTE_VHOST_MSG_RESULT_OK;
1074         }
1075
1076         if (dev->mem) {
1077                 free_mem_region(dev);
1078                 rte_free(dev->mem);
1079                 dev->mem = NULL;
1080         }
1081
1082         /* Flush IOTLB cache as previous HVAs are now invalid */
1083         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1084                 for (i = 0; i < dev->nr_vring; i++)
1085                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1086
1087         dev->nr_guest_pages = 0;
1088         if (!dev->guest_pages) {
1089                 dev->max_guest_pages = 8;
1090                 dev->guest_pages = malloc(dev->max_guest_pages *
1091                                                 sizeof(struct guest_page));
1092                 if (dev->guest_pages == NULL) {
1093                         RTE_LOG(ERR, VHOST_CONFIG,
1094                                 "(%d) failed to allocate memory "
1095                                 "for dev->guest_pages\n",
1096                                 dev->vid);
1097                         return RTE_VHOST_MSG_RESULT_ERR;
1098                 }
1099         }
1100
1101         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1102                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1103         if (dev->mem == NULL) {
1104                 RTE_LOG(ERR, VHOST_CONFIG,
1105                         "(%d) failed to allocate memory for dev->mem\n",
1106                         dev->vid);
1107                 return RTE_VHOST_MSG_RESULT_ERR;
1108         }
1109         dev->mem->nregions = memory->nregions;
1110
1111         for (i = 0; i < memory->nregions; i++) {
1112                 fd  = msg->fds[i];
1113                 reg = &dev->mem->regions[i];
1114
1115                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1116                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1117                 reg->size            = memory->regions[i].memory_size;
1118                 reg->fd              = fd;
1119
1120                 mmap_offset = memory->regions[i].mmap_offset;
1121
1122                 /* Check for memory_size + mmap_offset overflow */
1123                 if (mmap_offset >= -reg->size) {
1124                         RTE_LOG(ERR, VHOST_CONFIG,
1125                                 "mmap_offset (%#"PRIx64") and memory_size "
1126                                 "(%#"PRIx64") overflow\n",
1127                                 mmap_offset, reg->size);
1128                         goto err_mmap;
1129                 }
1130
1131                 mmap_size = reg->size + mmap_offset;
1132
1133                 /* mmap() without flag of MAP_ANONYMOUS, should be called
1134                  * with length argument aligned with hugepagesz at older
1135                  * longterm version Linux, like 2.6.32 and 3.2.72, or
1136                  * mmap() will fail with EINVAL.
1137                  *
1138                  * to avoid failure, make sure in caller to keep length
1139                  * aligned.
1140                  */
1141                 alignment = get_blk_size(fd);
1142                 if (alignment == (uint64_t)-1) {
1143                         RTE_LOG(ERR, VHOST_CONFIG,
1144                                 "couldn't get hugepage size through fstat\n");
1145                         goto err_mmap;
1146                 }
1147                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1148
1149                 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
1150                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1151                                  MAP_SHARED | populate, fd, 0);
1152
1153                 if (mmap_addr == MAP_FAILED) {
1154                         RTE_LOG(ERR, VHOST_CONFIG,
1155                                 "mmap region %u failed.\n", i);
1156                         goto err_mmap;
1157                 }
1158
1159                 reg->mmap_addr = mmap_addr;
1160                 reg->mmap_size = mmap_size;
1161                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1162                                       mmap_offset;
1163
1164                 if (dev->dequeue_zero_copy)
1165                         if (add_guest_pages(dev, reg, alignment) < 0) {
1166                                 RTE_LOG(ERR, VHOST_CONFIG,
1167                                         "adding guest pages to region %u failed.\n",
1168                                         i);
1169                                 goto err_mmap;
1170                         }
1171
1172                 RTE_LOG(INFO, VHOST_CONFIG,
1173                         "guest memory region %u, size: 0x%" PRIx64 "\n"
1174                         "\t guest physical addr: 0x%" PRIx64 "\n"
1175                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1176                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1177                         "\t mmap addr : 0x%" PRIx64 "\n"
1178                         "\t mmap size : 0x%" PRIx64 "\n"
1179                         "\t mmap align: 0x%" PRIx64 "\n"
1180                         "\t mmap off  : 0x%" PRIx64 "\n",
1181                         i, reg->size,
1182                         reg->guest_phys_addr,
1183                         reg->guest_user_addr,
1184                         reg->host_user_addr,
1185                         (uint64_t)(uintptr_t)mmap_addr,
1186                         mmap_size,
1187                         alignment,
1188                         mmap_offset);
1189
1190                 if (dev->postcopy_listening) {
1191                         /*
1192                          * We haven't a better way right now than sharing
1193                          * DPDK's virtual address with Qemu, so that Qemu can
1194                          * retrieve the region offset when handling userfaults.
1195                          */
1196                         memory->regions[i].userspace_addr =
1197                                 reg->host_user_addr;
1198                 }
1199         }
1200         if (dev->postcopy_listening) {
1201                 /* Send the addresses back to qemu */
1202                 msg->fd_num = 0;
1203                 send_vhost_reply(main_fd, msg);
1204
1205                 /* Wait for qemu to acknolwedge it's got the addresses
1206                  * we've got to wait before we're allowed to generate faults.
1207                  */
1208                 VhostUserMsg ack_msg;
1209                 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1210                         RTE_LOG(ERR, VHOST_CONFIG,
1211                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1212                         goto err_mmap;
1213                 }
1214
1215                 if (validate_msg_fds(&ack_msg, 0) != 0)
1216                         goto err_mmap;
1217
1218                 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1219                         RTE_LOG(ERR, VHOST_CONFIG,
1220                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1221                                 ack_msg.request.master);
1222                         goto err_mmap;
1223                 }
1224
1225                 /* Now userfault register and we can use the memory */
1226                 for (i = 0; i < memory->nregions; i++) {
1227 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1228                         reg = &dev->mem->regions[i];
1229                         struct uffdio_register reg_struct;
1230
1231                         /*
1232                          * Let's register all the mmap'ed area to ensure
1233                          * alignment on page boundary.
1234                          */
1235                         reg_struct.range.start =
1236                                 (uint64_t)(uintptr_t)reg->mmap_addr;
1237                         reg_struct.range.len = reg->mmap_size;
1238                         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1239
1240                         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1241                                                 &reg_struct)) {
1242                                 RTE_LOG(ERR, VHOST_CONFIG,
1243                                         "Failed to register ufd for region %d: (ufd = %d) %s\n",
1244                                         i, dev->postcopy_ufd,
1245                                         strerror(errno));
1246                                 goto err_mmap;
1247                         }
1248                         RTE_LOG(INFO, VHOST_CONFIG,
1249                                 "\t userfaultfd registered for range : "
1250                                 "%" PRIx64 " - %" PRIx64 "\n",
1251                                 (uint64_t)reg_struct.range.start,
1252                                 (uint64_t)reg_struct.range.start +
1253                                 (uint64_t)reg_struct.range.len - 1);
1254 #else
1255                         goto err_mmap;
1256 #endif
1257                 }
1258         }
1259
1260         for (i = 0; i < dev->nr_vring; i++) {
1261                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1262
1263                 if (vq->desc || vq->avail || vq->used) {
1264                         /*
1265                          * If the memory table got updated, the ring addresses
1266                          * need to be translated again as virtual addresses have
1267                          * changed.
1268                          */
1269                         vring_invalidate(dev, vq);
1270
1271                         dev = translate_ring_addresses(dev, i);
1272                         if (!dev) {
1273                                 dev = *pdev;
1274                                 goto err_mmap;
1275                         }
1276
1277                         *pdev = dev;
1278                 }
1279         }
1280
1281         dump_guest_pages(dev);
1282
1283         return RTE_VHOST_MSG_RESULT_OK;
1284
1285 err_mmap:
1286         free_mem_region(dev);
1287         rte_free(dev->mem);
1288         dev->mem = NULL;
1289         return RTE_VHOST_MSG_RESULT_ERR;
1290 }
1291
1292 static bool
1293 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1294 {
1295         bool rings_ok;
1296
1297         if (!vq)
1298                 return false;
1299
1300         if (vq_is_packed(dev))
1301                 rings_ok = !!vq->desc_packed;
1302         else
1303                 rings_ok = vq->desc && vq->avail && vq->used;
1304
1305         return rings_ok &&
1306                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1307                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1308 }
1309
1310 static int
1311 virtio_is_ready(struct virtio_net *dev)
1312 {
1313         struct vhost_virtqueue *vq;
1314         uint32_t i;
1315
1316         if (dev->nr_vring == 0)
1317                 return 0;
1318
1319         for (i = 0; i < dev->nr_vring; i++) {
1320                 vq = dev->virtqueue[i];
1321
1322                 if (!vq_is_ready(dev, vq))
1323                         return 0;
1324         }
1325
1326         RTE_LOG(INFO, VHOST_CONFIG,
1327                 "virtio is now ready for processing.\n");
1328         return 1;
1329 }
1330
1331 static void *
1332 inflight_mem_alloc(const char *name, size_t size, int *fd)
1333 {
1334         void *ptr;
1335         int mfd = -1;
1336         char fname[20] = "/tmp/memfd-XXXXXX";
1337
1338         *fd = -1;
1339 #ifdef MEMFD_SUPPORTED
1340         mfd = memfd_create(name, MFD_CLOEXEC);
1341 #else
1342         RTE_SET_USED(name);
1343 #endif
1344         if (mfd == -1) {
1345                 mfd = mkstemp(fname);
1346                 if (mfd == -1) {
1347                         RTE_LOG(ERR, VHOST_CONFIG,
1348                                 "failed to get inflight buffer fd\n");
1349                         return NULL;
1350                 }
1351
1352                 unlink(fname);
1353         }
1354
1355         if (ftruncate(mfd, size) == -1) {
1356                 RTE_LOG(ERR, VHOST_CONFIG,
1357                         "failed to alloc inflight buffer\n");
1358                 close(mfd);
1359                 return NULL;
1360         }
1361
1362         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1363         if (ptr == MAP_FAILED) {
1364                 RTE_LOG(ERR, VHOST_CONFIG,
1365                         "failed to mmap inflight buffer\n");
1366                 close(mfd);
1367                 return NULL;
1368         }
1369
1370         *fd = mfd;
1371         return ptr;
1372 }
1373
1374 static uint32_t
1375 get_pervq_shm_size_split(uint16_t queue_size)
1376 {
1377         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1378                                   queue_size + sizeof(uint64_t) +
1379                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1380 }
1381
1382 static uint32_t
1383 get_pervq_shm_size_packed(uint16_t queue_size)
1384 {
1385         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1386                                   * queue_size + sizeof(uint64_t) +
1387                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1388                                   INFLIGHT_ALIGNMENT);
1389 }
1390
1391 static int
1392 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1393                            VhostUserMsg *msg,
1394                            int main_fd __rte_unused)
1395 {
1396         struct rte_vhost_inflight_info_packed *inflight_packed;
1397         uint64_t pervq_inflight_size, mmap_size;
1398         uint16_t num_queues, queue_size;
1399         struct virtio_net *dev = *pdev;
1400         int fd, i, j;
1401         void *addr;
1402
1403         if (msg->size != sizeof(msg->payload.inflight)) {
1404                 RTE_LOG(ERR, VHOST_CONFIG,
1405                         "invalid get_inflight_fd message size is %d\n",
1406                         msg->size);
1407                 return RTE_VHOST_MSG_RESULT_ERR;
1408         }
1409
1410         if (dev->inflight_info == NULL) {
1411                 dev->inflight_info = calloc(1,
1412                                             sizeof(struct inflight_mem_info));
1413                 if (!dev->inflight_info) {
1414                         RTE_LOG(ERR, VHOST_CONFIG,
1415                                 "failed to alloc dev inflight area\n");
1416                         return RTE_VHOST_MSG_RESULT_ERR;
1417                 }
1418         }
1419
1420         num_queues = msg->payload.inflight.num_queues;
1421         queue_size = msg->payload.inflight.queue_size;
1422
1423         RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd num_queues: %u\n",
1424                 msg->payload.inflight.num_queues);
1425         RTE_LOG(INFO, VHOST_CONFIG, "get_inflight_fd queue_size: %u\n",
1426                 msg->payload.inflight.queue_size);
1427
1428         if (vq_is_packed(dev))
1429                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1430         else
1431                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1432
1433         mmap_size = num_queues * pervq_inflight_size;
1434         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1435         if (!addr) {
1436                 RTE_LOG(ERR, VHOST_CONFIG,
1437                         "failed to alloc vhost inflight area\n");
1438                         msg->payload.inflight.mmap_size = 0;
1439                 return RTE_VHOST_MSG_RESULT_ERR;
1440         }
1441         memset(addr, 0, mmap_size);
1442
1443         dev->inflight_info->addr = addr;
1444         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1445         dev->inflight_info->fd = msg->fds[0] = fd;
1446         msg->payload.inflight.mmap_offset = 0;
1447         msg->fd_num = 1;
1448
1449         if (vq_is_packed(dev)) {
1450                 for (i = 0; i < num_queues; i++) {
1451                         inflight_packed =
1452                                 (struct rte_vhost_inflight_info_packed *)addr;
1453                         inflight_packed->used_wrap_counter = 1;
1454                         inflight_packed->old_used_wrap_counter = 1;
1455                         for (j = 0; j < queue_size; j++)
1456                                 inflight_packed->desc[j].next = j + 1;
1457                         addr = (void *)((char *)addr + pervq_inflight_size);
1458                 }
1459         }
1460
1461         RTE_LOG(INFO, VHOST_CONFIG,
1462                 "send inflight mmap_size: %"PRIu64"\n",
1463                 msg->payload.inflight.mmap_size);
1464         RTE_LOG(INFO, VHOST_CONFIG,
1465                 "send inflight mmap_offset: %"PRIu64"\n",
1466                 msg->payload.inflight.mmap_offset);
1467         RTE_LOG(INFO, VHOST_CONFIG,
1468                 "send inflight fd: %d\n", msg->fds[0]);
1469
1470         return RTE_VHOST_MSG_RESULT_REPLY;
1471 }
1472
1473 static int
1474 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1475                            int main_fd __rte_unused)
1476 {
1477         uint64_t mmap_size, mmap_offset;
1478         uint16_t num_queues, queue_size;
1479         struct virtio_net *dev = *pdev;
1480         uint32_t pervq_inflight_size;
1481         struct vhost_virtqueue *vq;
1482         void *addr;
1483         int fd, i;
1484
1485         fd = msg->fds[0];
1486         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1487                 RTE_LOG(ERR, VHOST_CONFIG,
1488                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1489                         msg->size, fd);
1490                 return RTE_VHOST_MSG_RESULT_ERR;
1491         }
1492
1493         mmap_size = msg->payload.inflight.mmap_size;
1494         mmap_offset = msg->payload.inflight.mmap_offset;
1495         num_queues = msg->payload.inflight.num_queues;
1496         queue_size = msg->payload.inflight.queue_size;
1497
1498         if (vq_is_packed(dev))
1499                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1500         else
1501                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1502
1503         RTE_LOG(INFO, VHOST_CONFIG,
1504                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1505         RTE_LOG(INFO, VHOST_CONFIG,
1506                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1507         RTE_LOG(INFO, VHOST_CONFIG,
1508                 "set_inflight_fd num_queues: %u\n", num_queues);
1509         RTE_LOG(INFO, VHOST_CONFIG,
1510                 "set_inflight_fd queue_size: %u\n", queue_size);
1511         RTE_LOG(INFO, VHOST_CONFIG,
1512                 "set_inflight_fd fd: %d\n", fd);
1513         RTE_LOG(INFO, VHOST_CONFIG,
1514                 "set_inflight_fd pervq_inflight_size: %d\n",
1515                 pervq_inflight_size);
1516
1517         if (!dev->inflight_info) {
1518                 dev->inflight_info = calloc(1,
1519                                             sizeof(struct inflight_mem_info));
1520                 if (dev->inflight_info == NULL) {
1521                         RTE_LOG(ERR, VHOST_CONFIG,
1522                                 "failed to alloc dev inflight area\n");
1523                         return RTE_VHOST_MSG_RESULT_ERR;
1524                 }
1525         }
1526
1527         if (dev->inflight_info->addr)
1528                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1529
1530         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1531                     fd, mmap_offset);
1532         if (addr == MAP_FAILED) {
1533                 RTE_LOG(ERR, VHOST_CONFIG, "failed to mmap share memory.\n");
1534                 return RTE_VHOST_MSG_RESULT_ERR;
1535         }
1536
1537         if (dev->inflight_info->fd)
1538                 close(dev->inflight_info->fd);
1539
1540         dev->inflight_info->fd = fd;
1541         dev->inflight_info->addr = addr;
1542         dev->inflight_info->size = mmap_size;
1543
1544         for (i = 0; i < num_queues; i++) {
1545                 vq = dev->virtqueue[i];
1546                 if (vq_is_packed(dev)) {
1547                         vq->inflight_packed = addr;
1548                         vq->inflight_packed->desc_num = queue_size;
1549                 } else {
1550                         vq->inflight_split = addr;
1551                         vq->inflight_split->desc_num = queue_size;
1552                 }
1553                 addr = (void *)((char *)addr + pervq_inflight_size);
1554         }
1555
1556         return RTE_VHOST_MSG_RESULT_OK;
1557 }
1558
1559 static int
1560 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1561                         int main_fd __rte_unused)
1562 {
1563         struct virtio_net *dev = *pdev;
1564         struct vhost_vring_file file;
1565         struct vhost_virtqueue *vq;
1566
1567         if (validate_msg_fds(msg, 1) != 0)
1568                 return RTE_VHOST_MSG_RESULT_ERR;
1569
1570         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1571         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1572                 file.fd = VIRTIO_INVALID_EVENTFD;
1573         else
1574                 file.fd = msg->fds[0];
1575         RTE_LOG(INFO, VHOST_CONFIG,
1576                 "vring call idx:%d file:%d\n", file.index, file.fd);
1577
1578         vq = dev->virtqueue[file.index];
1579         if (vq->callfd >= 0)
1580                 close(vq->callfd);
1581
1582         vq->callfd = file.fd;
1583
1584         return RTE_VHOST_MSG_RESULT_OK;
1585 }
1586
1587 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1588                         struct VhostUserMsg *msg,
1589                         int main_fd __rte_unused)
1590 {
1591         if (validate_msg_fds(msg, 1) != 0)
1592                 return RTE_VHOST_MSG_RESULT_ERR;
1593
1594         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1595                 close(msg->fds[0]);
1596         RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1597
1598         return RTE_VHOST_MSG_RESULT_OK;
1599 }
1600
1601 static int
1602 resubmit_desc_compare(const void *a, const void *b)
1603 {
1604         const struct rte_vhost_resubmit_desc *desc0 = a;
1605         const struct rte_vhost_resubmit_desc *desc1 = b;
1606
1607         if (desc1->counter > desc0->counter)
1608                 return 1;
1609
1610         return -1;
1611 }
1612
1613 static int
1614 vhost_check_queue_inflights_split(struct virtio_net *dev,
1615                                   struct vhost_virtqueue *vq)
1616 {
1617         uint16_t i;
1618         uint16_t resubmit_num = 0, last_io, num;
1619         struct vring_used *used = vq->used;
1620         struct rte_vhost_resubmit_info *resubmit;
1621         struct rte_vhost_inflight_info_split *inflight_split;
1622
1623         if (!(dev->protocol_features &
1624             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1625                 return RTE_VHOST_MSG_RESULT_OK;
1626
1627         if ((!vq->inflight_split))
1628                 return RTE_VHOST_MSG_RESULT_ERR;
1629
1630         if (!vq->inflight_split->version) {
1631                 vq->inflight_split->version = INFLIGHT_VERSION;
1632                 return RTE_VHOST_MSG_RESULT_OK;
1633         }
1634
1635         if (vq->resubmit_inflight)
1636                 return RTE_VHOST_MSG_RESULT_OK;
1637
1638         inflight_split = vq->inflight_split;
1639         vq->global_counter = 0;
1640         last_io = inflight_split->last_inflight_io;
1641
1642         if (inflight_split->used_idx != used->idx) {
1643                 inflight_split->desc[last_io].inflight = 0;
1644                 rte_smp_mb();
1645                 inflight_split->used_idx = used->idx;
1646         }
1647
1648         for (i = 0; i < inflight_split->desc_num; i++) {
1649                 if (inflight_split->desc[i].inflight == 1)
1650                         resubmit_num++;
1651         }
1652
1653         vq->last_avail_idx += resubmit_num;
1654
1655         if (resubmit_num) {
1656                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1657                 if (!resubmit) {
1658                         RTE_LOG(ERR, VHOST_CONFIG,
1659                                 "failed to allocate memory for resubmit info.\n");
1660                         return RTE_VHOST_MSG_RESULT_ERR;
1661                 }
1662
1663                 resubmit->resubmit_list = calloc(resubmit_num,
1664                         sizeof(struct rte_vhost_resubmit_desc));
1665                 if (!resubmit->resubmit_list) {
1666                         RTE_LOG(ERR, VHOST_CONFIG,
1667                                 "failed to allocate memory for inflight desc.\n");
1668                         free(resubmit);
1669                         return RTE_VHOST_MSG_RESULT_ERR;
1670                 }
1671
1672                 num = 0;
1673                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1674                         if (vq->inflight_split->desc[i].inflight == 1) {
1675                                 resubmit->resubmit_list[num].index = i;
1676                                 resubmit->resubmit_list[num].counter =
1677                                         inflight_split->desc[i].counter;
1678                                 num++;
1679                         }
1680                 }
1681                 resubmit->resubmit_num = num;
1682
1683                 if (resubmit->resubmit_num > 1)
1684                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1685                               sizeof(struct rte_vhost_resubmit_desc),
1686                               resubmit_desc_compare);
1687
1688                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1689                 vq->resubmit_inflight = resubmit;
1690         }
1691
1692         return RTE_VHOST_MSG_RESULT_OK;
1693 }
1694
1695 static int
1696 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1697                                    struct vhost_virtqueue *vq)
1698 {
1699         uint16_t i;
1700         uint16_t resubmit_num = 0, old_used_idx, num;
1701         struct rte_vhost_resubmit_info *resubmit;
1702         struct rte_vhost_inflight_info_packed *inflight_packed;
1703
1704         if (!(dev->protocol_features &
1705             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1706                 return RTE_VHOST_MSG_RESULT_OK;
1707
1708         if ((!vq->inflight_packed))
1709                 return RTE_VHOST_MSG_RESULT_ERR;
1710
1711         if (!vq->inflight_packed->version) {
1712                 vq->inflight_packed->version = INFLIGHT_VERSION;
1713                 return RTE_VHOST_MSG_RESULT_OK;
1714         }
1715
1716         if (vq->resubmit_inflight)
1717                 return RTE_VHOST_MSG_RESULT_OK;
1718
1719         inflight_packed = vq->inflight_packed;
1720         vq->global_counter = 0;
1721         old_used_idx = inflight_packed->old_used_idx;
1722
1723         if (inflight_packed->used_idx != old_used_idx) {
1724                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1725                         inflight_packed->old_used_idx =
1726                                 inflight_packed->used_idx;
1727                         inflight_packed->old_used_wrap_counter =
1728                                 inflight_packed->used_wrap_counter;
1729                         inflight_packed->old_free_head =
1730                                 inflight_packed->free_head;
1731                 } else {
1732                         inflight_packed->used_idx =
1733                                 inflight_packed->old_used_idx;
1734                         inflight_packed->used_wrap_counter =
1735                                 inflight_packed->old_used_wrap_counter;
1736                         inflight_packed->free_head =
1737                                 inflight_packed->old_free_head;
1738                 }
1739         }
1740
1741         for (i = 0; i < inflight_packed->desc_num; i++) {
1742                 if (inflight_packed->desc[i].inflight == 1)
1743                         resubmit_num++;
1744         }
1745
1746         if (resubmit_num) {
1747                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1748                 if (resubmit == NULL) {
1749                         RTE_LOG(ERR, VHOST_CONFIG,
1750                                 "failed to allocate memory for resubmit info.\n");
1751                         return RTE_VHOST_MSG_RESULT_ERR;
1752                 }
1753
1754                 resubmit->resubmit_list = calloc(resubmit_num,
1755                         sizeof(struct rte_vhost_resubmit_desc));
1756                 if (resubmit->resubmit_list == NULL) {
1757                         RTE_LOG(ERR, VHOST_CONFIG,
1758                                 "failed to allocate memory for resubmit desc.\n");
1759                         free(resubmit);
1760                         return RTE_VHOST_MSG_RESULT_ERR;
1761                 }
1762
1763                 num = 0;
1764                 for (i = 0; i < inflight_packed->desc_num; i++) {
1765                         if (vq->inflight_packed->desc[i].inflight == 1) {
1766                                 resubmit->resubmit_list[num].index = i;
1767                                 resubmit->resubmit_list[num].counter =
1768                                         inflight_packed->desc[i].counter;
1769                                 num++;
1770                         }
1771                 }
1772                 resubmit->resubmit_num = num;
1773
1774                 if (resubmit->resubmit_num > 1)
1775                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1776                               sizeof(struct rte_vhost_resubmit_desc),
1777                               resubmit_desc_compare);
1778
1779                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1780                 vq->resubmit_inflight = resubmit;
1781         }
1782
1783         return RTE_VHOST_MSG_RESULT_OK;
1784 }
1785
1786 static int
1787 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1788                         int main_fd __rte_unused)
1789 {
1790         struct virtio_net *dev = *pdev;
1791         struct vhost_vring_file file;
1792         struct vhost_virtqueue *vq;
1793
1794         if (validate_msg_fds(msg, 1) != 0)
1795                 return RTE_VHOST_MSG_RESULT_ERR;
1796
1797         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1798         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1799                 file.fd = VIRTIO_INVALID_EVENTFD;
1800         else
1801                 file.fd = msg->fds[0];
1802         RTE_LOG(INFO, VHOST_CONFIG,
1803                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1804
1805         /* Interpret ring addresses only when ring is started. */
1806         dev = translate_ring_addresses(dev, file.index);
1807         if (!dev)
1808                 return RTE_VHOST_MSG_RESULT_ERR;
1809
1810         *pdev = dev;
1811
1812         vq = dev->virtqueue[file.index];
1813
1814         /*
1815          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1816          * the ring starts already enabled. Otherwise, it is enabled via
1817          * the SET_VRING_ENABLE message.
1818          */
1819         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1820                 vq->enabled = 1;
1821                 if (dev->notify_ops->vring_state_changed)
1822                         dev->notify_ops->vring_state_changed(
1823                                 dev->vid, file.index, 1);
1824         }
1825
1826         if (vq->kickfd >= 0)
1827                 close(vq->kickfd);
1828         vq->kickfd = file.fd;
1829
1830         if (vq_is_packed(dev)) {
1831                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1832                         RTE_LOG(ERR, VHOST_CONFIG,
1833                                 "failed to inflights for vq: %d\n", file.index);
1834                         return RTE_VHOST_MSG_RESULT_ERR;
1835                 }
1836         } else {
1837                 if (vhost_check_queue_inflights_split(dev, vq)) {
1838                         RTE_LOG(ERR, VHOST_CONFIG,
1839                                 "failed to inflights for vq: %d\n", file.index);
1840                         return RTE_VHOST_MSG_RESULT_ERR;
1841                 }
1842         }
1843
1844         return RTE_VHOST_MSG_RESULT_OK;
1845 }
1846
1847 static void
1848 free_zmbufs(struct vhost_virtqueue *vq)
1849 {
1850         drain_zmbuf_list(vq);
1851
1852         rte_free(vq->zmbufs);
1853 }
1854
1855 /*
1856  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1857  */
1858 static int
1859 vhost_user_get_vring_base(struct virtio_net **pdev,
1860                         struct VhostUserMsg *msg,
1861                         int main_fd __rte_unused)
1862 {
1863         struct virtio_net *dev = *pdev;
1864         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1865         uint64_t val;
1866
1867         if (validate_msg_fds(msg, 0) != 0)
1868                 return RTE_VHOST_MSG_RESULT_ERR;
1869
1870         /* We have to stop the queue (virtio) if it is running. */
1871         vhost_destroy_device_notify(dev);
1872
1873         dev->flags &= ~VIRTIO_DEV_READY;
1874         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1875
1876         /* Here we are safe to get the indexes */
1877         if (vq_is_packed(dev)) {
1878                 /*
1879                  * Bit[0:14]: avail index
1880                  * Bit[15]: avail wrap counter
1881                  */
1882                 val = vq->last_avail_idx & 0x7fff;
1883                 val |= vq->avail_wrap_counter << 15;
1884                 msg->payload.state.num = val;
1885         } else {
1886                 msg->payload.state.num = vq->last_avail_idx;
1887         }
1888
1889         RTE_LOG(INFO, VHOST_CONFIG,
1890                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1891                 msg->payload.state.num);
1892         /*
1893          * Based on current qemu vhost-user implementation, this message is
1894          * sent and only sent in vhost_vring_stop.
1895          * TODO: cleanup the vring, it isn't usable since here.
1896          */
1897         if (vq->kickfd >= 0)
1898                 close(vq->kickfd);
1899
1900         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1901
1902         if (vq->callfd >= 0)
1903                 close(vq->callfd);
1904
1905         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1906
1907         vq->signalled_used_valid = false;
1908
1909         if (dev->dequeue_zero_copy)
1910                 free_zmbufs(vq);
1911         if (vq_is_packed(dev)) {
1912                 rte_free(vq->shadow_used_packed);
1913                 vq->shadow_used_packed = NULL;
1914         } else {
1915                 rte_free(vq->shadow_used_split);
1916                 vq->shadow_used_split = NULL;
1917         }
1918
1919         rte_free(vq->batch_copy_elems);
1920         vq->batch_copy_elems = NULL;
1921
1922         msg->size = sizeof(msg->payload.state);
1923         msg->fd_num = 0;
1924
1925         vring_invalidate(dev, vq);
1926
1927         return RTE_VHOST_MSG_RESULT_REPLY;
1928 }
1929
1930 /*
1931  * when virtio queues are ready to work, qemu will send us to
1932  * enable the virtio queue pair.
1933  */
1934 static int
1935 vhost_user_set_vring_enable(struct virtio_net **pdev,
1936                         struct VhostUserMsg *msg,
1937                         int main_fd __rte_unused)
1938 {
1939         struct virtio_net *dev = *pdev;
1940         int enable = (int)msg->payload.state.num;
1941         int index = (int)msg->payload.state.index;
1942         struct rte_vdpa_device *vdpa_dev;
1943         int did = -1;
1944
1945         if (validate_msg_fds(msg, 0) != 0)
1946                 return RTE_VHOST_MSG_RESULT_ERR;
1947
1948         RTE_LOG(INFO, VHOST_CONFIG,
1949                 "set queue enable: %d to qp idx: %d\n",
1950                 enable, index);
1951
1952         did = dev->vdpa_dev_id;
1953         vdpa_dev = rte_vdpa_get_device(did);
1954         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1955                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1956
1957         if (dev->notify_ops->vring_state_changed)
1958                 dev->notify_ops->vring_state_changed(dev->vid,
1959                                 index, enable);
1960
1961         /* On disable, rings have to be stopped being processed. */
1962         if (!enable && dev->dequeue_zero_copy)
1963                 drain_zmbuf_list(dev->virtqueue[index]);
1964
1965         dev->virtqueue[index]->enabled = enable;
1966
1967         return RTE_VHOST_MSG_RESULT_OK;
1968 }
1969
1970 static int
1971 vhost_user_get_protocol_features(struct virtio_net **pdev,
1972                         struct VhostUserMsg *msg,
1973                         int main_fd __rte_unused)
1974 {
1975         struct virtio_net *dev = *pdev;
1976         uint64_t features, protocol_features;
1977
1978         if (validate_msg_fds(msg, 0) != 0)
1979                 return RTE_VHOST_MSG_RESULT_ERR;
1980
1981         rte_vhost_driver_get_features(dev->ifname, &features);
1982         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1983
1984         /*
1985          * REPLY_ACK protocol feature is only mandatory for now
1986          * for IOMMU feature. If IOMMU is explicitly disabled by the
1987          * application, disable also REPLY_ACK feature for older buggy
1988          * Qemu versions (from v2.7.0 to v2.9.0).
1989          */
1990         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1991                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1992
1993         msg->payload.u64 = protocol_features;
1994         msg->size = sizeof(msg->payload.u64);
1995         msg->fd_num = 0;
1996
1997         return RTE_VHOST_MSG_RESULT_REPLY;
1998 }
1999
2000 static int
2001 vhost_user_set_protocol_features(struct virtio_net **pdev,
2002                         struct VhostUserMsg *msg,
2003                         int main_fd __rte_unused)
2004 {
2005         struct virtio_net *dev = *pdev;
2006         uint64_t protocol_features = msg->payload.u64;
2007         uint64_t slave_protocol_features = 0;
2008
2009         if (validate_msg_fds(msg, 0) != 0)
2010                 return RTE_VHOST_MSG_RESULT_ERR;
2011
2012         rte_vhost_driver_get_protocol_features(dev->ifname,
2013                         &slave_protocol_features);
2014         if (protocol_features & ~slave_protocol_features) {
2015                 RTE_LOG(ERR, VHOST_CONFIG,
2016                         "(%d) received invalid protocol features.\n",
2017                         dev->vid);
2018                 return RTE_VHOST_MSG_RESULT_ERR;
2019         }
2020
2021         dev->protocol_features = protocol_features;
2022         RTE_LOG(INFO, VHOST_CONFIG,
2023                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2024                 dev->protocol_features);
2025
2026         return RTE_VHOST_MSG_RESULT_OK;
2027 }
2028
2029 static int
2030 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2031                         int main_fd __rte_unused)
2032 {
2033         struct virtio_net *dev = *pdev;
2034         int fd = msg->fds[0];
2035         uint64_t size, off;
2036         void *addr;
2037
2038         if (validate_msg_fds(msg, 1) != 0)
2039                 return RTE_VHOST_MSG_RESULT_ERR;
2040
2041         if (fd < 0) {
2042                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
2043                 return RTE_VHOST_MSG_RESULT_ERR;
2044         }
2045
2046         if (msg->size != sizeof(VhostUserLog)) {
2047                 RTE_LOG(ERR, VHOST_CONFIG,
2048                         "invalid log base msg size: %"PRId32" != %d\n",
2049                         msg->size, (int)sizeof(VhostUserLog));
2050                 return RTE_VHOST_MSG_RESULT_ERR;
2051         }
2052
2053         size = msg->payload.log.mmap_size;
2054         off  = msg->payload.log.mmap_offset;
2055
2056         /* Don't allow mmap_offset to point outside the mmap region */
2057         if (off > size) {
2058                 RTE_LOG(ERR, VHOST_CONFIG,
2059                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
2060                         off, size);
2061                 return RTE_VHOST_MSG_RESULT_ERR;
2062         }
2063
2064         RTE_LOG(INFO, VHOST_CONFIG,
2065                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2066                 size, off);
2067
2068         /*
2069          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2070          * fail when offset is not page size aligned.
2071          */
2072         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2073         close(fd);
2074         if (addr == MAP_FAILED) {
2075                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
2076                 return RTE_VHOST_MSG_RESULT_ERR;
2077         }
2078
2079         /*
2080          * Free previously mapped log memory on occasionally
2081          * multiple VHOST_USER_SET_LOG_BASE.
2082          */
2083         if (dev->log_addr) {
2084                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2085         }
2086         dev->log_addr = (uint64_t)(uintptr_t)addr;
2087         dev->log_base = dev->log_addr + off;
2088         dev->log_size = size;
2089
2090         /*
2091          * The spec is not clear about it (yet), but QEMU doesn't expect
2092          * any payload in the reply.
2093          */
2094         msg->size = 0;
2095         msg->fd_num = 0;
2096
2097         return RTE_VHOST_MSG_RESULT_REPLY;
2098 }
2099
2100 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2101                         struct VhostUserMsg *msg,
2102                         int main_fd __rte_unused)
2103 {
2104         if (validate_msg_fds(msg, 1) != 0)
2105                 return RTE_VHOST_MSG_RESULT_ERR;
2106
2107         close(msg->fds[0]);
2108         RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
2109
2110         return RTE_VHOST_MSG_RESULT_OK;
2111 }
2112
2113 /*
2114  * An rarp packet is constructed and broadcasted to notify switches about
2115  * the new location of the migrated VM, so that packets from outside will
2116  * not be lost after migration.
2117  *
2118  * However, we don't actually "send" a rarp packet here, instead, we set
2119  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2120  */
2121 static int
2122 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2123                         int main_fd __rte_unused)
2124 {
2125         struct virtio_net *dev = *pdev;
2126         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2127         struct rte_vdpa_device *vdpa_dev;
2128         int did = -1;
2129
2130         if (validate_msg_fds(msg, 0) != 0)
2131                 return RTE_VHOST_MSG_RESULT_ERR;
2132
2133         RTE_LOG(DEBUG, VHOST_CONFIG,
2134                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2135                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2136         memcpy(dev->mac.addr_bytes, mac, 6);
2137
2138         /*
2139          * Set the flag to inject a RARP broadcast packet at
2140          * rte_vhost_dequeue_burst().
2141          *
2142          * rte_smp_wmb() is for making sure the mac is copied
2143          * before the flag is set.
2144          */
2145         rte_smp_wmb();
2146         rte_atomic16_set(&dev->broadcast_rarp, 1);
2147         did = dev->vdpa_dev_id;
2148         vdpa_dev = rte_vdpa_get_device(did);
2149         if (vdpa_dev && vdpa_dev->ops->migration_done)
2150                 vdpa_dev->ops->migration_done(dev->vid);
2151
2152         return RTE_VHOST_MSG_RESULT_OK;
2153 }
2154
2155 static int
2156 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2157                         int main_fd __rte_unused)
2158 {
2159         struct virtio_net *dev = *pdev;
2160
2161         if (validate_msg_fds(msg, 0) != 0)
2162                 return RTE_VHOST_MSG_RESULT_ERR;
2163
2164         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2165                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2166                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
2167                                 msg->payload.u64);
2168
2169                 return RTE_VHOST_MSG_RESULT_ERR;
2170         }
2171
2172         dev->mtu = msg->payload.u64;
2173
2174         return RTE_VHOST_MSG_RESULT_OK;
2175 }
2176
2177 static int
2178 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2179                         int main_fd __rte_unused)
2180 {
2181         struct virtio_net *dev = *pdev;
2182         int fd = msg->fds[0];
2183
2184         if (validate_msg_fds(msg, 1) != 0)
2185                 return RTE_VHOST_MSG_RESULT_ERR;
2186
2187         if (fd < 0) {
2188                 RTE_LOG(ERR, VHOST_CONFIG,
2189                                 "Invalid file descriptor for slave channel (%d)\n",
2190                                 fd);
2191                 return RTE_VHOST_MSG_RESULT_ERR;
2192         }
2193
2194         if (dev->slave_req_fd >= 0)
2195                 close(dev->slave_req_fd);
2196
2197         dev->slave_req_fd = fd;
2198
2199         return RTE_VHOST_MSG_RESULT_OK;
2200 }
2201
2202 static int
2203 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2204 {
2205         struct vhost_vring_addr *ra;
2206         uint64_t start, end, len;
2207
2208         start = imsg->iova;
2209         end = start + imsg->size;
2210
2211         ra = &vq->ring_addrs;
2212         len = sizeof(struct vring_desc) * vq->size;
2213         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2214                 return 1;
2215
2216         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2217         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2218                 return 1;
2219
2220         len = sizeof(struct vring_used) +
2221                sizeof(struct vring_used_elem) * vq->size;
2222         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2223                 return 1;
2224
2225         return 0;
2226 }
2227
2228 static int
2229 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2230 {
2231         struct vhost_vring_addr *ra;
2232         uint64_t start, end, len;
2233
2234         start = imsg->iova;
2235         end = start + imsg->size;
2236
2237         ra = &vq->ring_addrs;
2238         len = sizeof(struct vring_packed_desc) * vq->size;
2239         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2240                 return 1;
2241
2242         len = sizeof(struct vring_packed_desc_event);
2243         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2244                 return 1;
2245
2246         len = sizeof(struct vring_packed_desc_event);
2247         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2248                 return 1;
2249
2250         return 0;
2251 }
2252
2253 static int is_vring_iotlb(struct virtio_net *dev,
2254                           struct vhost_virtqueue *vq,
2255                           struct vhost_iotlb_msg *imsg)
2256 {
2257         if (vq_is_packed(dev))
2258                 return is_vring_iotlb_packed(vq, imsg);
2259         else
2260                 return is_vring_iotlb_split(vq, imsg);
2261 }
2262
2263 static int
2264 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2265                         int main_fd __rte_unused)
2266 {
2267         struct virtio_net *dev = *pdev;
2268         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2269         uint16_t i;
2270         uint64_t vva, len;
2271
2272         if (validate_msg_fds(msg, 0) != 0)
2273                 return RTE_VHOST_MSG_RESULT_ERR;
2274
2275         switch (imsg->type) {
2276         case VHOST_IOTLB_UPDATE:
2277                 len = imsg->size;
2278                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2279                 if (!vva)
2280                         return RTE_VHOST_MSG_RESULT_ERR;
2281
2282                 for (i = 0; i < dev->nr_vring; i++) {
2283                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2284
2285                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2286                                         len, imsg->perm);
2287
2288                         if (is_vring_iotlb(dev, vq, imsg))
2289                                 *pdev = dev = translate_ring_addresses(dev, i);
2290                 }
2291                 break;
2292         case VHOST_IOTLB_INVALIDATE:
2293                 for (i = 0; i < dev->nr_vring; i++) {
2294                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2295
2296                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2297                                         imsg->size);
2298
2299                         if (is_vring_iotlb(dev, vq, imsg))
2300                                 vring_invalidate(dev, vq);
2301                 }
2302                 break;
2303         default:
2304                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
2305                                 imsg->type);
2306                 return RTE_VHOST_MSG_RESULT_ERR;
2307         }
2308
2309         return RTE_VHOST_MSG_RESULT_OK;
2310 }
2311
2312 static int
2313 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2314                         struct VhostUserMsg *msg,
2315                         int main_fd __rte_unused)
2316 {
2317         struct virtio_net *dev = *pdev;
2318 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2319         struct uffdio_api api_struct;
2320
2321         if (validate_msg_fds(msg, 0) != 0)
2322                 return RTE_VHOST_MSG_RESULT_ERR;
2323
2324         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2325
2326         if (dev->postcopy_ufd == -1) {
2327                 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
2328                         strerror(errno));
2329                 return RTE_VHOST_MSG_RESULT_ERR;
2330         }
2331         api_struct.api = UFFD_API;
2332         api_struct.features = 0;
2333         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2334                 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
2335                         strerror(errno));
2336                 close(dev->postcopy_ufd);
2337                 dev->postcopy_ufd = -1;
2338                 return RTE_VHOST_MSG_RESULT_ERR;
2339         }
2340         msg->fds[0] = dev->postcopy_ufd;
2341         msg->fd_num = 1;
2342
2343         return RTE_VHOST_MSG_RESULT_REPLY;
2344 #else
2345         dev->postcopy_ufd = -1;
2346         msg->fd_num = 0;
2347
2348         return RTE_VHOST_MSG_RESULT_ERR;
2349 #endif
2350 }
2351
2352 static int
2353 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2354                         struct VhostUserMsg *msg __rte_unused,
2355                         int main_fd __rte_unused)
2356 {
2357         struct virtio_net *dev = *pdev;
2358
2359         if (validate_msg_fds(msg, 0) != 0)
2360                 return RTE_VHOST_MSG_RESULT_ERR;
2361
2362         if (dev->mem && dev->mem->nregions) {
2363                 RTE_LOG(ERR, VHOST_CONFIG,
2364                         "Regions already registered at postcopy-listen\n");
2365                 return RTE_VHOST_MSG_RESULT_ERR;
2366         }
2367         dev->postcopy_listening = 1;
2368
2369         return RTE_VHOST_MSG_RESULT_OK;
2370 }
2371
2372 static int
2373 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2374                         int main_fd __rte_unused)
2375 {
2376         struct virtio_net *dev = *pdev;
2377
2378         if (validate_msg_fds(msg, 0) != 0)
2379                 return RTE_VHOST_MSG_RESULT_ERR;
2380
2381         dev->postcopy_listening = 0;
2382         if (dev->postcopy_ufd >= 0) {
2383                 close(dev->postcopy_ufd);
2384                 dev->postcopy_ufd = -1;
2385         }
2386
2387         msg->payload.u64 = 0;
2388         msg->size = sizeof(msg->payload.u64);
2389         msg->fd_num = 0;
2390
2391         return RTE_VHOST_MSG_RESULT_REPLY;
2392 }
2393
2394 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2395                                         struct VhostUserMsg *msg,
2396                                         int main_fd);
2397 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2398         [VHOST_USER_NONE] = NULL,
2399         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2400         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2401         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2402         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2403         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2404         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2405         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2406         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2407         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2408         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2409         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2410         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2411         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2412         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2413         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2414         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2415         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2416         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2417         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2418         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2419         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2420         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2421         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2422         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2423         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2424         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2425         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2426 };
2427
2428 /* return bytes# of read on success or negative val on failure. */
2429 static int
2430 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2431 {
2432         int ret;
2433
2434         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2435                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2436         if (ret <= 0)
2437                 return ret;
2438
2439         if (msg->size) {
2440                 if (msg->size > sizeof(msg->payload)) {
2441                         RTE_LOG(ERR, VHOST_CONFIG,
2442                                 "invalid msg size: %d\n", msg->size);
2443                         return -1;
2444                 }
2445                 ret = read(sockfd, &msg->payload, msg->size);
2446                 if (ret <= 0)
2447                         return ret;
2448                 if (ret != (int)msg->size) {
2449                         RTE_LOG(ERR, VHOST_CONFIG,
2450                                 "read control message failed\n");
2451                         return -1;
2452                 }
2453         }
2454
2455         return ret;
2456 }
2457
2458 static int
2459 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2460 {
2461         if (!msg)
2462                 return 0;
2463
2464         return send_fd_message(sockfd, (char *)msg,
2465                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2466 }
2467
2468 static int
2469 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2470 {
2471         if (!msg)
2472                 return 0;
2473
2474         msg->flags &= ~VHOST_USER_VERSION_MASK;
2475         msg->flags &= ~VHOST_USER_NEED_REPLY;
2476         msg->flags |= VHOST_USER_VERSION;
2477         msg->flags |= VHOST_USER_REPLY_MASK;
2478
2479         return send_vhost_message(sockfd, msg);
2480 }
2481
2482 static int
2483 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2484 {
2485         int ret;
2486
2487         if (msg->flags & VHOST_USER_NEED_REPLY)
2488                 rte_spinlock_lock(&dev->slave_req_lock);
2489
2490         ret = send_vhost_message(dev->slave_req_fd, msg);
2491         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2492                 rte_spinlock_unlock(&dev->slave_req_lock);
2493
2494         return ret;
2495 }
2496
2497 /*
2498  * Allocate a queue pair if it hasn't been allocated yet
2499  */
2500 static int
2501 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2502                         struct VhostUserMsg *msg)
2503 {
2504         uint16_t vring_idx;
2505
2506         switch (msg->request.master) {
2507         case VHOST_USER_SET_VRING_KICK:
2508         case VHOST_USER_SET_VRING_CALL:
2509         case VHOST_USER_SET_VRING_ERR:
2510                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2511                 break;
2512         case VHOST_USER_SET_VRING_NUM:
2513         case VHOST_USER_SET_VRING_BASE:
2514         case VHOST_USER_SET_VRING_ENABLE:
2515                 vring_idx = msg->payload.state.index;
2516                 break;
2517         case VHOST_USER_SET_VRING_ADDR:
2518                 vring_idx = msg->payload.addr.index;
2519                 break;
2520         default:
2521                 return 0;
2522         }
2523
2524         if (vring_idx >= VHOST_MAX_VRING) {
2525                 RTE_LOG(ERR, VHOST_CONFIG,
2526                         "invalid vring index: %u\n", vring_idx);
2527                 return -1;
2528         }
2529
2530         if (dev->virtqueue[vring_idx])
2531                 return 0;
2532
2533         return alloc_vring_queue(dev, vring_idx);
2534 }
2535
2536 static void
2537 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2538 {
2539         unsigned int i = 0;
2540         unsigned int vq_num = 0;
2541
2542         while (vq_num < dev->nr_vring) {
2543                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2544
2545                 if (vq) {
2546                         rte_spinlock_lock(&vq->access_lock);
2547                         vq_num++;
2548                 }
2549                 i++;
2550         }
2551 }
2552
2553 static void
2554 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2555 {
2556         unsigned int i = 0;
2557         unsigned int vq_num = 0;
2558
2559         while (vq_num < dev->nr_vring) {
2560                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2561
2562                 if (vq) {
2563                         rte_spinlock_unlock(&vq->access_lock);
2564                         vq_num++;
2565                 }
2566                 i++;
2567         }
2568 }
2569
2570 int
2571 vhost_user_msg_handler(int vid, int fd)
2572 {
2573         struct virtio_net *dev;
2574         struct VhostUserMsg msg;
2575         struct rte_vdpa_device *vdpa_dev;
2576         int did = -1;
2577         int ret;
2578         int unlock_required = 0;
2579         bool handled;
2580         int request;
2581
2582         dev = get_device(vid);
2583         if (dev == NULL)
2584                 return -1;
2585
2586         if (!dev->notify_ops) {
2587                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2588                 if (!dev->notify_ops) {
2589                         RTE_LOG(ERR, VHOST_CONFIG,
2590                                 "failed to get callback ops for driver %s\n",
2591                                 dev->ifname);
2592                         return -1;
2593                 }
2594         }
2595
2596         ret = read_vhost_message(fd, &msg);
2597         if (ret <= 0) {
2598                 if (ret < 0)
2599                         RTE_LOG(ERR, VHOST_CONFIG,
2600                                 "vhost read message failed\n");
2601                 else
2602                         RTE_LOG(INFO, VHOST_CONFIG,
2603                                 "vhost peer closed\n");
2604
2605                 return -1;
2606         }
2607
2608         ret = 0;
2609         request = msg.request.master;
2610         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2611                         vhost_message_str[request]) {
2612                 if (request != VHOST_USER_IOTLB_MSG)
2613                         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
2614                                 vhost_message_str[request]);
2615                 else
2616                         RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
2617                                 vhost_message_str[request]);
2618         } else {
2619                 RTE_LOG(DEBUG, VHOST_CONFIG, "External request %d\n", request);
2620         }
2621
2622         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2623         if (ret < 0) {
2624                 RTE_LOG(ERR, VHOST_CONFIG,
2625                         "failed to alloc queue\n");
2626                 return -1;
2627         }
2628
2629         /*
2630          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2631          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2632          * and device is destroyed. destroy_device waits for queues to be
2633          * inactive, so it is safe. Otherwise taking the access_lock
2634          * would cause a dead lock.
2635          */
2636         switch (request) {
2637         case VHOST_USER_SET_FEATURES:
2638         case VHOST_USER_SET_PROTOCOL_FEATURES:
2639         case VHOST_USER_SET_OWNER:
2640         case VHOST_USER_SET_MEM_TABLE:
2641         case VHOST_USER_SET_LOG_BASE:
2642         case VHOST_USER_SET_LOG_FD:
2643         case VHOST_USER_SET_VRING_NUM:
2644         case VHOST_USER_SET_VRING_ADDR:
2645         case VHOST_USER_SET_VRING_BASE:
2646         case VHOST_USER_SET_VRING_KICK:
2647         case VHOST_USER_SET_VRING_CALL:
2648         case VHOST_USER_SET_VRING_ERR:
2649         case VHOST_USER_SET_VRING_ENABLE:
2650         case VHOST_USER_SEND_RARP:
2651         case VHOST_USER_NET_SET_MTU:
2652         case VHOST_USER_SET_SLAVE_REQ_FD:
2653                 vhost_user_lock_all_queue_pairs(dev);
2654                 unlock_required = 1;
2655                 break;
2656         default:
2657                 break;
2658
2659         }
2660
2661         handled = false;
2662         if (dev->extern_ops.pre_msg_handle) {
2663                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2664                                 (void *)&msg);
2665                 switch (ret) {
2666                 case RTE_VHOST_MSG_RESULT_REPLY:
2667                         send_vhost_reply(fd, &msg);
2668                         /* Fall-through */
2669                 case RTE_VHOST_MSG_RESULT_ERR:
2670                 case RTE_VHOST_MSG_RESULT_OK:
2671                         handled = true;
2672                         goto skip_to_post_handle;
2673                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2674                 default:
2675                         break;
2676                 }
2677         }
2678
2679         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2680                 if (!vhost_message_handlers[request])
2681                         goto skip_to_post_handle;
2682                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2683
2684                 switch (ret) {
2685                 case RTE_VHOST_MSG_RESULT_ERR:
2686                         RTE_LOG(ERR, VHOST_CONFIG,
2687                                 "Processing %s failed.\n",
2688                                 vhost_message_str[request]);
2689                         handled = true;
2690                         break;
2691                 case RTE_VHOST_MSG_RESULT_OK:
2692                         RTE_LOG(DEBUG, VHOST_CONFIG,
2693                                 "Processing %s succeeded.\n",
2694                                 vhost_message_str[request]);
2695                         handled = true;
2696                         break;
2697                 case RTE_VHOST_MSG_RESULT_REPLY:
2698                         RTE_LOG(DEBUG, VHOST_CONFIG,
2699                                 "Processing %s succeeded and needs reply.\n",
2700                                 vhost_message_str[request]);
2701                         send_vhost_reply(fd, &msg);
2702                         handled = true;
2703                         break;
2704                 default:
2705                         break;
2706                 }
2707         }
2708
2709 skip_to_post_handle:
2710         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2711                         dev->extern_ops.post_msg_handle) {
2712                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2713                                 (void *)&msg);
2714                 switch (ret) {
2715                 case RTE_VHOST_MSG_RESULT_REPLY:
2716                         send_vhost_reply(fd, &msg);
2717                         /* Fall-through */
2718                 case RTE_VHOST_MSG_RESULT_ERR:
2719                 case RTE_VHOST_MSG_RESULT_OK:
2720                         handled = true;
2721                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2722                 default:
2723                         break;
2724                 }
2725         }
2726
2727         if (unlock_required)
2728                 vhost_user_unlock_all_queue_pairs(dev);
2729
2730         /* If message was not handled at this stage, treat it as an error */
2731         if (!handled) {
2732                 RTE_LOG(ERR, VHOST_CONFIG,
2733                         "vhost message (req: %d) was not handled.\n", request);
2734                 close_msg_fds(&msg);
2735                 ret = RTE_VHOST_MSG_RESULT_ERR;
2736         }
2737
2738         /*
2739          * If the request required a reply that was already sent,
2740          * this optional reply-ack won't be sent as the
2741          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2742          */
2743         if (msg.flags & VHOST_USER_NEED_REPLY) {
2744                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2745                 msg.size = sizeof(msg.payload.u64);
2746                 msg.fd_num = 0;
2747                 send_vhost_reply(fd, &msg);
2748         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2749                 RTE_LOG(ERR, VHOST_CONFIG,
2750                         "vhost message handling failed.\n");
2751                 return -1;
2752         }
2753
2754         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2755                 dev->flags |= VIRTIO_DEV_READY;
2756
2757                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2758                         if (dev->dequeue_zero_copy) {
2759                                 RTE_LOG(INFO, VHOST_CONFIG,
2760                                                 "dequeue zero copy is enabled\n");
2761                         }
2762
2763                         if (dev->notify_ops->new_device(dev->vid) == 0)
2764                                 dev->flags |= VIRTIO_DEV_RUNNING;
2765                 }
2766         }
2767
2768         did = dev->vdpa_dev_id;
2769         vdpa_dev = rte_vdpa_get_device(did);
2770         if (vdpa_dev && virtio_is_ready(dev) &&
2771                         !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2772                         msg.request.master == VHOST_USER_SET_VRING_CALL) {
2773                 if (vdpa_dev->ops->dev_conf)
2774                         vdpa_dev->ops->dev_conf(dev->vid);
2775                 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2776         }
2777
2778         return 0;
2779 }
2780
2781 static int process_slave_message_reply(struct virtio_net *dev,
2782                                        const struct VhostUserMsg *msg)
2783 {
2784         struct VhostUserMsg msg_reply;
2785         int ret;
2786
2787         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2788                 return 0;
2789
2790         if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2791                 ret = -1;
2792                 goto out;
2793         }
2794
2795         if (msg_reply.request.slave != msg->request.slave) {
2796                 RTE_LOG(ERR, VHOST_CONFIG,
2797                         "Received unexpected msg type (%u), expected %u\n",
2798                         msg_reply.request.slave, msg->request.slave);
2799                 ret = -1;
2800                 goto out;
2801         }
2802
2803         ret = msg_reply.payload.u64 ? -1 : 0;
2804
2805 out:
2806         rte_spinlock_unlock(&dev->slave_req_lock);
2807         return ret;
2808 }
2809
2810 int
2811 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2812 {
2813         int ret;
2814         struct VhostUserMsg msg = {
2815                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2816                 .flags = VHOST_USER_VERSION,
2817                 .size = sizeof(msg.payload.iotlb),
2818                 .payload.iotlb = {
2819                         .iova = iova,
2820                         .perm = perm,
2821                         .type = VHOST_IOTLB_MISS,
2822                 },
2823         };
2824
2825         ret = send_vhost_message(dev->slave_req_fd, &msg);
2826         if (ret < 0) {
2827                 RTE_LOG(ERR, VHOST_CONFIG,
2828                                 "Failed to send IOTLB miss message (%d)\n",
2829                                 ret);
2830                 return ret;
2831         }
2832
2833         return 0;
2834 }
2835
2836 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2837                                                     int index, int fd,
2838                                                     uint64_t offset,
2839                                                     uint64_t size)
2840 {
2841         int ret;
2842         struct VhostUserMsg msg = {
2843                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2844                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2845                 .size = sizeof(msg.payload.area),
2846                 .payload.area = {
2847                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
2848                         .size = size,
2849                         .offset = offset,
2850                 },
2851         };
2852
2853         if (fd < 0)
2854                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2855         else {
2856                 msg.fds[0] = fd;
2857                 msg.fd_num = 1;
2858         }
2859
2860         ret = send_vhost_slave_message(dev, &msg);
2861         if (ret < 0) {
2862                 RTE_LOG(ERR, VHOST_CONFIG,
2863                         "Failed to set host notifier (%d)\n", ret);
2864                 return ret;
2865         }
2866
2867         return process_slave_message_reply(dev, &msg);
2868 }
2869
2870 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2871 {
2872         struct virtio_net *dev;
2873         struct rte_vdpa_device *vdpa_dev;
2874         int vfio_device_fd, did, ret = 0;
2875         uint64_t offset, size;
2876         unsigned int i;
2877
2878         dev = get_device(vid);
2879         if (!dev)
2880                 return -ENODEV;
2881
2882         did = dev->vdpa_dev_id;
2883         if (did < 0)
2884                 return -EINVAL;
2885
2886         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2887             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2888             !(dev->protocol_features &
2889                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2890             !(dev->protocol_features &
2891                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2892             !(dev->protocol_features &
2893                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2894                 return -ENOTSUP;
2895
2896         vdpa_dev = rte_vdpa_get_device(did);
2897         if (!vdpa_dev)
2898                 return -ENODEV;
2899
2900         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2901         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2902
2903         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2904         if (vfio_device_fd < 0)
2905                 return -ENOTSUP;
2906
2907         if (enable) {
2908                 for (i = 0; i < dev->nr_vring; i++) {
2909                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2910                                         &size) < 0) {
2911                                 ret = -ENOTSUP;
2912                                 goto disable;
2913                         }
2914
2915                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
2916                                         vfio_device_fd, offset, size) < 0) {
2917                                 ret = -EFAULT;
2918                                 goto disable;
2919                         }
2920                 }
2921         } else {
2922 disable:
2923                 for (i = 0; i < dev->nr_vring; i++) {
2924                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
2925                                         0, 0);
2926                 }
2927         }
2928
2929         return ret;
2930 }