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         int expected_fds;
1567
1568         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1569         if (validate_msg_fds(msg, expected_fds) != 0)
1570                 return RTE_VHOST_MSG_RESULT_ERR;
1571
1572         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1573         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1574                 file.fd = VIRTIO_INVALID_EVENTFD;
1575         else
1576                 file.fd = msg->fds[0];
1577         RTE_LOG(INFO, VHOST_CONFIG,
1578                 "vring call idx:%d file:%d\n", file.index, file.fd);
1579
1580         vq = dev->virtqueue[file.index];
1581         if (vq->callfd >= 0)
1582                 close(vq->callfd);
1583
1584         vq->callfd = file.fd;
1585
1586         return RTE_VHOST_MSG_RESULT_OK;
1587 }
1588
1589 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1590                         struct VhostUserMsg *msg,
1591                         int main_fd __rte_unused)
1592 {
1593         int expected_fds;
1594
1595         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1596         if (validate_msg_fds(msg, expected_fds) != 0)
1597                 return RTE_VHOST_MSG_RESULT_ERR;
1598
1599         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1600                 close(msg->fds[0]);
1601         RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1602
1603         return RTE_VHOST_MSG_RESULT_OK;
1604 }
1605
1606 static int
1607 resubmit_desc_compare(const void *a, const void *b)
1608 {
1609         const struct rte_vhost_resubmit_desc *desc0 = a;
1610         const struct rte_vhost_resubmit_desc *desc1 = b;
1611
1612         if (desc1->counter > desc0->counter)
1613                 return 1;
1614
1615         return -1;
1616 }
1617
1618 static int
1619 vhost_check_queue_inflights_split(struct virtio_net *dev,
1620                                   struct vhost_virtqueue *vq)
1621 {
1622         uint16_t i;
1623         uint16_t resubmit_num = 0, last_io, num;
1624         struct vring_used *used = vq->used;
1625         struct rte_vhost_resubmit_info *resubmit;
1626         struct rte_vhost_inflight_info_split *inflight_split;
1627
1628         if (!(dev->protocol_features &
1629             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1630                 return RTE_VHOST_MSG_RESULT_OK;
1631
1632         if ((!vq->inflight_split))
1633                 return RTE_VHOST_MSG_RESULT_ERR;
1634
1635         if (!vq->inflight_split->version) {
1636                 vq->inflight_split->version = INFLIGHT_VERSION;
1637                 return RTE_VHOST_MSG_RESULT_OK;
1638         }
1639
1640         if (vq->resubmit_inflight)
1641                 return RTE_VHOST_MSG_RESULT_OK;
1642
1643         inflight_split = vq->inflight_split;
1644         vq->global_counter = 0;
1645         last_io = inflight_split->last_inflight_io;
1646
1647         if (inflight_split->used_idx != used->idx) {
1648                 inflight_split->desc[last_io].inflight = 0;
1649                 rte_smp_mb();
1650                 inflight_split->used_idx = used->idx;
1651         }
1652
1653         for (i = 0; i < inflight_split->desc_num; i++) {
1654                 if (inflight_split->desc[i].inflight == 1)
1655                         resubmit_num++;
1656         }
1657
1658         vq->last_avail_idx += resubmit_num;
1659
1660         if (resubmit_num) {
1661                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1662                 if (!resubmit) {
1663                         RTE_LOG(ERR, VHOST_CONFIG,
1664                                 "failed to allocate memory for resubmit info.\n");
1665                         return RTE_VHOST_MSG_RESULT_ERR;
1666                 }
1667
1668                 resubmit->resubmit_list = calloc(resubmit_num,
1669                         sizeof(struct rte_vhost_resubmit_desc));
1670                 if (!resubmit->resubmit_list) {
1671                         RTE_LOG(ERR, VHOST_CONFIG,
1672                                 "failed to allocate memory for inflight desc.\n");
1673                         free(resubmit);
1674                         return RTE_VHOST_MSG_RESULT_ERR;
1675                 }
1676
1677                 num = 0;
1678                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1679                         if (vq->inflight_split->desc[i].inflight == 1) {
1680                                 resubmit->resubmit_list[num].index = i;
1681                                 resubmit->resubmit_list[num].counter =
1682                                         inflight_split->desc[i].counter;
1683                                 num++;
1684                         }
1685                 }
1686                 resubmit->resubmit_num = num;
1687
1688                 if (resubmit->resubmit_num > 1)
1689                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1690                               sizeof(struct rte_vhost_resubmit_desc),
1691                               resubmit_desc_compare);
1692
1693                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1694                 vq->resubmit_inflight = resubmit;
1695         }
1696
1697         return RTE_VHOST_MSG_RESULT_OK;
1698 }
1699
1700 static int
1701 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1702                                    struct vhost_virtqueue *vq)
1703 {
1704         uint16_t i;
1705         uint16_t resubmit_num = 0, old_used_idx, num;
1706         struct rte_vhost_resubmit_info *resubmit;
1707         struct rte_vhost_inflight_info_packed *inflight_packed;
1708
1709         if (!(dev->protocol_features &
1710             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1711                 return RTE_VHOST_MSG_RESULT_OK;
1712
1713         if ((!vq->inflight_packed))
1714                 return RTE_VHOST_MSG_RESULT_ERR;
1715
1716         if (!vq->inflight_packed->version) {
1717                 vq->inflight_packed->version = INFLIGHT_VERSION;
1718                 return RTE_VHOST_MSG_RESULT_OK;
1719         }
1720
1721         if (vq->resubmit_inflight)
1722                 return RTE_VHOST_MSG_RESULT_OK;
1723
1724         inflight_packed = vq->inflight_packed;
1725         vq->global_counter = 0;
1726         old_used_idx = inflight_packed->old_used_idx;
1727
1728         if (inflight_packed->used_idx != old_used_idx) {
1729                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1730                         inflight_packed->old_used_idx =
1731                                 inflight_packed->used_idx;
1732                         inflight_packed->old_used_wrap_counter =
1733                                 inflight_packed->used_wrap_counter;
1734                         inflight_packed->old_free_head =
1735                                 inflight_packed->free_head;
1736                 } else {
1737                         inflight_packed->used_idx =
1738                                 inflight_packed->old_used_idx;
1739                         inflight_packed->used_wrap_counter =
1740                                 inflight_packed->old_used_wrap_counter;
1741                         inflight_packed->free_head =
1742                                 inflight_packed->old_free_head;
1743                 }
1744         }
1745
1746         for (i = 0; i < inflight_packed->desc_num; i++) {
1747                 if (inflight_packed->desc[i].inflight == 1)
1748                         resubmit_num++;
1749         }
1750
1751         if (resubmit_num) {
1752                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1753                 if (resubmit == NULL) {
1754                         RTE_LOG(ERR, VHOST_CONFIG,
1755                                 "failed to allocate memory for resubmit info.\n");
1756                         return RTE_VHOST_MSG_RESULT_ERR;
1757                 }
1758
1759                 resubmit->resubmit_list = calloc(resubmit_num,
1760                         sizeof(struct rte_vhost_resubmit_desc));
1761                 if (resubmit->resubmit_list == NULL) {
1762                         RTE_LOG(ERR, VHOST_CONFIG,
1763                                 "failed to allocate memory for resubmit desc.\n");
1764                         free(resubmit);
1765                         return RTE_VHOST_MSG_RESULT_ERR;
1766                 }
1767
1768                 num = 0;
1769                 for (i = 0; i < inflight_packed->desc_num; i++) {
1770                         if (vq->inflight_packed->desc[i].inflight == 1) {
1771                                 resubmit->resubmit_list[num].index = i;
1772                                 resubmit->resubmit_list[num].counter =
1773                                         inflight_packed->desc[i].counter;
1774                                 num++;
1775                         }
1776                 }
1777                 resubmit->resubmit_num = num;
1778
1779                 if (resubmit->resubmit_num > 1)
1780                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1781                               sizeof(struct rte_vhost_resubmit_desc),
1782                               resubmit_desc_compare);
1783
1784                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1785                 vq->resubmit_inflight = resubmit;
1786         }
1787
1788         return RTE_VHOST_MSG_RESULT_OK;
1789 }
1790
1791 static int
1792 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1793                         int main_fd __rte_unused)
1794 {
1795         struct virtio_net *dev = *pdev;
1796         struct vhost_vring_file file;
1797         struct vhost_virtqueue *vq;
1798         int expected_fds;
1799
1800         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1801         if (validate_msg_fds(msg, expected_fds) != 0)
1802                 return RTE_VHOST_MSG_RESULT_ERR;
1803
1804         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1805         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1806                 file.fd = VIRTIO_INVALID_EVENTFD;
1807         else
1808                 file.fd = msg->fds[0];
1809         RTE_LOG(INFO, VHOST_CONFIG,
1810                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1811
1812         /* Interpret ring addresses only when ring is started. */
1813         dev = translate_ring_addresses(dev, file.index);
1814         if (!dev)
1815                 return RTE_VHOST_MSG_RESULT_ERR;
1816
1817         *pdev = dev;
1818
1819         vq = dev->virtqueue[file.index];
1820
1821         /*
1822          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1823          * the ring starts already enabled. Otherwise, it is enabled via
1824          * the SET_VRING_ENABLE message.
1825          */
1826         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1827                 vq->enabled = 1;
1828                 if (dev->notify_ops->vring_state_changed)
1829                         dev->notify_ops->vring_state_changed(
1830                                 dev->vid, file.index, 1);
1831         }
1832
1833         if (vq->kickfd >= 0)
1834                 close(vq->kickfd);
1835         vq->kickfd = file.fd;
1836
1837         if (vq_is_packed(dev)) {
1838                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1839                         RTE_LOG(ERR, VHOST_CONFIG,
1840                                 "failed to inflights for vq: %d\n", file.index);
1841                         return RTE_VHOST_MSG_RESULT_ERR;
1842                 }
1843         } else {
1844                 if (vhost_check_queue_inflights_split(dev, vq)) {
1845                         RTE_LOG(ERR, VHOST_CONFIG,
1846                                 "failed to inflights for vq: %d\n", file.index);
1847                         return RTE_VHOST_MSG_RESULT_ERR;
1848                 }
1849         }
1850
1851         return RTE_VHOST_MSG_RESULT_OK;
1852 }
1853
1854 static void
1855 free_zmbufs(struct vhost_virtqueue *vq)
1856 {
1857         drain_zmbuf_list(vq);
1858
1859         rte_free(vq->zmbufs);
1860 }
1861
1862 /*
1863  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1864  */
1865 static int
1866 vhost_user_get_vring_base(struct virtio_net **pdev,
1867                         struct VhostUserMsg *msg,
1868                         int main_fd __rte_unused)
1869 {
1870         struct virtio_net *dev = *pdev;
1871         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1872         uint64_t val;
1873
1874         if (validate_msg_fds(msg, 0) != 0)
1875                 return RTE_VHOST_MSG_RESULT_ERR;
1876
1877         /* We have to stop the queue (virtio) if it is running. */
1878         vhost_destroy_device_notify(dev);
1879
1880         dev->flags &= ~VIRTIO_DEV_READY;
1881         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1882
1883         /* Here we are safe to get the indexes */
1884         if (vq_is_packed(dev)) {
1885                 /*
1886                  * Bit[0:14]: avail index
1887                  * Bit[15]: avail wrap counter
1888                  */
1889                 val = vq->last_avail_idx & 0x7fff;
1890                 val |= vq->avail_wrap_counter << 15;
1891                 msg->payload.state.num = val;
1892         } else {
1893                 msg->payload.state.num = vq->last_avail_idx;
1894         }
1895
1896         RTE_LOG(INFO, VHOST_CONFIG,
1897                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1898                 msg->payload.state.num);
1899         /*
1900          * Based on current qemu vhost-user implementation, this message is
1901          * sent and only sent in vhost_vring_stop.
1902          * TODO: cleanup the vring, it isn't usable since here.
1903          */
1904         if (vq->kickfd >= 0)
1905                 close(vq->kickfd);
1906
1907         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1908
1909         if (vq->callfd >= 0)
1910                 close(vq->callfd);
1911
1912         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1913
1914         vq->signalled_used_valid = false;
1915
1916         if (dev->dequeue_zero_copy)
1917                 free_zmbufs(vq);
1918         if (vq_is_packed(dev)) {
1919                 rte_free(vq->shadow_used_packed);
1920                 vq->shadow_used_packed = NULL;
1921         } else {
1922                 rte_free(vq->shadow_used_split);
1923                 vq->shadow_used_split = NULL;
1924         }
1925
1926         rte_free(vq->batch_copy_elems);
1927         vq->batch_copy_elems = NULL;
1928
1929         msg->size = sizeof(msg->payload.state);
1930         msg->fd_num = 0;
1931
1932         vring_invalidate(dev, vq);
1933
1934         return RTE_VHOST_MSG_RESULT_REPLY;
1935 }
1936
1937 /*
1938  * when virtio queues are ready to work, qemu will send us to
1939  * enable the virtio queue pair.
1940  */
1941 static int
1942 vhost_user_set_vring_enable(struct virtio_net **pdev,
1943                         struct VhostUserMsg *msg,
1944                         int main_fd __rte_unused)
1945 {
1946         struct virtio_net *dev = *pdev;
1947         int enable = (int)msg->payload.state.num;
1948         int index = (int)msg->payload.state.index;
1949         struct rte_vdpa_device *vdpa_dev;
1950         int did = -1;
1951
1952         if (validate_msg_fds(msg, 0) != 0)
1953                 return RTE_VHOST_MSG_RESULT_ERR;
1954
1955         RTE_LOG(INFO, VHOST_CONFIG,
1956                 "set queue enable: %d to qp idx: %d\n",
1957                 enable, index);
1958
1959         did = dev->vdpa_dev_id;
1960         vdpa_dev = rte_vdpa_get_device(did);
1961         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1962                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1963
1964         if (dev->notify_ops->vring_state_changed)
1965                 dev->notify_ops->vring_state_changed(dev->vid,
1966                                 index, enable);
1967
1968         /* On disable, rings have to be stopped being processed. */
1969         if (!enable && dev->dequeue_zero_copy)
1970                 drain_zmbuf_list(dev->virtqueue[index]);
1971
1972         dev->virtqueue[index]->enabled = enable;
1973
1974         return RTE_VHOST_MSG_RESULT_OK;
1975 }
1976
1977 static int
1978 vhost_user_get_protocol_features(struct virtio_net **pdev,
1979                         struct VhostUserMsg *msg,
1980                         int main_fd __rte_unused)
1981 {
1982         struct virtio_net *dev = *pdev;
1983         uint64_t features, protocol_features;
1984
1985         if (validate_msg_fds(msg, 0) != 0)
1986                 return RTE_VHOST_MSG_RESULT_ERR;
1987
1988         rte_vhost_driver_get_features(dev->ifname, &features);
1989         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1990
1991         /*
1992          * REPLY_ACK protocol feature is only mandatory for now
1993          * for IOMMU feature. If IOMMU is explicitly disabled by the
1994          * application, disable also REPLY_ACK feature for older buggy
1995          * Qemu versions (from v2.7.0 to v2.9.0).
1996          */
1997         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1998                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1999
2000         msg->payload.u64 = protocol_features;
2001         msg->size = sizeof(msg->payload.u64);
2002         msg->fd_num = 0;
2003
2004         return RTE_VHOST_MSG_RESULT_REPLY;
2005 }
2006
2007 static int
2008 vhost_user_set_protocol_features(struct virtio_net **pdev,
2009                         struct VhostUserMsg *msg,
2010                         int main_fd __rte_unused)
2011 {
2012         struct virtio_net *dev = *pdev;
2013         uint64_t protocol_features = msg->payload.u64;
2014         uint64_t slave_protocol_features = 0;
2015
2016         if (validate_msg_fds(msg, 0) != 0)
2017                 return RTE_VHOST_MSG_RESULT_ERR;
2018
2019         rte_vhost_driver_get_protocol_features(dev->ifname,
2020                         &slave_protocol_features);
2021         if (protocol_features & ~slave_protocol_features) {
2022                 RTE_LOG(ERR, VHOST_CONFIG,
2023                         "(%d) received invalid protocol features.\n",
2024                         dev->vid);
2025                 return RTE_VHOST_MSG_RESULT_ERR;
2026         }
2027
2028         dev->protocol_features = protocol_features;
2029         RTE_LOG(INFO, VHOST_CONFIG,
2030                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2031                 dev->protocol_features);
2032
2033         return RTE_VHOST_MSG_RESULT_OK;
2034 }
2035
2036 static int
2037 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2038                         int main_fd __rte_unused)
2039 {
2040         struct virtio_net *dev = *pdev;
2041         int fd = msg->fds[0];
2042         uint64_t size, off;
2043         void *addr;
2044
2045         if (validate_msg_fds(msg, 1) != 0)
2046                 return RTE_VHOST_MSG_RESULT_ERR;
2047
2048         if (fd < 0) {
2049                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
2050                 return RTE_VHOST_MSG_RESULT_ERR;
2051         }
2052
2053         if (msg->size != sizeof(VhostUserLog)) {
2054                 RTE_LOG(ERR, VHOST_CONFIG,
2055                         "invalid log base msg size: %"PRId32" != %d\n",
2056                         msg->size, (int)sizeof(VhostUserLog));
2057                 return RTE_VHOST_MSG_RESULT_ERR;
2058         }
2059
2060         size = msg->payload.log.mmap_size;
2061         off  = msg->payload.log.mmap_offset;
2062
2063         /* Don't allow mmap_offset to point outside the mmap region */
2064         if (off > size) {
2065                 RTE_LOG(ERR, VHOST_CONFIG,
2066                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
2067                         off, size);
2068                 return RTE_VHOST_MSG_RESULT_ERR;
2069         }
2070
2071         RTE_LOG(INFO, VHOST_CONFIG,
2072                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2073                 size, off);
2074
2075         /*
2076          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2077          * fail when offset is not page size aligned.
2078          */
2079         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2080         close(fd);
2081         if (addr == MAP_FAILED) {
2082                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
2083                 return RTE_VHOST_MSG_RESULT_ERR;
2084         }
2085
2086         /*
2087          * Free previously mapped log memory on occasionally
2088          * multiple VHOST_USER_SET_LOG_BASE.
2089          */
2090         if (dev->log_addr) {
2091                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2092         }
2093         dev->log_addr = (uint64_t)(uintptr_t)addr;
2094         dev->log_base = dev->log_addr + off;
2095         dev->log_size = size;
2096
2097         /*
2098          * The spec is not clear about it (yet), but QEMU doesn't expect
2099          * any payload in the reply.
2100          */
2101         msg->size = 0;
2102         msg->fd_num = 0;
2103
2104         return RTE_VHOST_MSG_RESULT_REPLY;
2105 }
2106
2107 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2108                         struct VhostUserMsg *msg,
2109                         int main_fd __rte_unused)
2110 {
2111         if (validate_msg_fds(msg, 1) != 0)
2112                 return RTE_VHOST_MSG_RESULT_ERR;
2113
2114         close(msg->fds[0]);
2115         RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
2116
2117         return RTE_VHOST_MSG_RESULT_OK;
2118 }
2119
2120 /*
2121  * An rarp packet is constructed and broadcasted to notify switches about
2122  * the new location of the migrated VM, so that packets from outside will
2123  * not be lost after migration.
2124  *
2125  * However, we don't actually "send" a rarp packet here, instead, we set
2126  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2127  */
2128 static int
2129 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2130                         int main_fd __rte_unused)
2131 {
2132         struct virtio_net *dev = *pdev;
2133         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2134         struct rte_vdpa_device *vdpa_dev;
2135         int did = -1;
2136
2137         if (validate_msg_fds(msg, 0) != 0)
2138                 return RTE_VHOST_MSG_RESULT_ERR;
2139
2140         RTE_LOG(DEBUG, VHOST_CONFIG,
2141                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2142                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2143         memcpy(dev->mac.addr_bytes, mac, 6);
2144
2145         /*
2146          * Set the flag to inject a RARP broadcast packet at
2147          * rte_vhost_dequeue_burst().
2148          *
2149          * rte_smp_wmb() is for making sure the mac is copied
2150          * before the flag is set.
2151          */
2152         rte_smp_wmb();
2153         rte_atomic16_set(&dev->broadcast_rarp, 1);
2154         did = dev->vdpa_dev_id;
2155         vdpa_dev = rte_vdpa_get_device(did);
2156         if (vdpa_dev && vdpa_dev->ops->migration_done)
2157                 vdpa_dev->ops->migration_done(dev->vid);
2158
2159         return RTE_VHOST_MSG_RESULT_OK;
2160 }
2161
2162 static int
2163 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2164                         int main_fd __rte_unused)
2165 {
2166         struct virtio_net *dev = *pdev;
2167
2168         if (validate_msg_fds(msg, 0) != 0)
2169                 return RTE_VHOST_MSG_RESULT_ERR;
2170
2171         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2172                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2173                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
2174                                 msg->payload.u64);
2175
2176                 return RTE_VHOST_MSG_RESULT_ERR;
2177         }
2178
2179         dev->mtu = msg->payload.u64;
2180
2181         return RTE_VHOST_MSG_RESULT_OK;
2182 }
2183
2184 static int
2185 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2186                         int main_fd __rte_unused)
2187 {
2188         struct virtio_net *dev = *pdev;
2189         int fd = msg->fds[0];
2190
2191         if (validate_msg_fds(msg, 1) != 0)
2192                 return RTE_VHOST_MSG_RESULT_ERR;
2193
2194         if (fd < 0) {
2195                 RTE_LOG(ERR, VHOST_CONFIG,
2196                                 "Invalid file descriptor for slave channel (%d)\n",
2197                                 fd);
2198                 return RTE_VHOST_MSG_RESULT_ERR;
2199         }
2200
2201         if (dev->slave_req_fd >= 0)
2202                 close(dev->slave_req_fd);
2203
2204         dev->slave_req_fd = fd;
2205
2206         return RTE_VHOST_MSG_RESULT_OK;
2207 }
2208
2209 static int
2210 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2211 {
2212         struct vhost_vring_addr *ra;
2213         uint64_t start, end, len;
2214
2215         start = imsg->iova;
2216         end = start + imsg->size;
2217
2218         ra = &vq->ring_addrs;
2219         len = sizeof(struct vring_desc) * vq->size;
2220         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2221                 return 1;
2222
2223         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2224         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2225                 return 1;
2226
2227         len = sizeof(struct vring_used) +
2228                sizeof(struct vring_used_elem) * vq->size;
2229         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2230                 return 1;
2231
2232         return 0;
2233 }
2234
2235 static int
2236 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2237 {
2238         struct vhost_vring_addr *ra;
2239         uint64_t start, end, len;
2240
2241         start = imsg->iova;
2242         end = start + imsg->size;
2243
2244         ra = &vq->ring_addrs;
2245         len = sizeof(struct vring_packed_desc) * vq->size;
2246         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2247                 return 1;
2248
2249         len = sizeof(struct vring_packed_desc_event);
2250         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2251                 return 1;
2252
2253         len = sizeof(struct vring_packed_desc_event);
2254         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2255                 return 1;
2256
2257         return 0;
2258 }
2259
2260 static int is_vring_iotlb(struct virtio_net *dev,
2261                           struct vhost_virtqueue *vq,
2262                           struct vhost_iotlb_msg *imsg)
2263 {
2264         if (vq_is_packed(dev))
2265                 return is_vring_iotlb_packed(vq, imsg);
2266         else
2267                 return is_vring_iotlb_split(vq, imsg);
2268 }
2269
2270 static int
2271 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2272                         int main_fd __rte_unused)
2273 {
2274         struct virtio_net *dev = *pdev;
2275         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2276         uint16_t i;
2277         uint64_t vva, len;
2278
2279         if (validate_msg_fds(msg, 0) != 0)
2280                 return RTE_VHOST_MSG_RESULT_ERR;
2281
2282         switch (imsg->type) {
2283         case VHOST_IOTLB_UPDATE:
2284                 len = imsg->size;
2285                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2286                 if (!vva)
2287                         return RTE_VHOST_MSG_RESULT_ERR;
2288
2289                 for (i = 0; i < dev->nr_vring; i++) {
2290                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2291
2292                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2293                                         len, imsg->perm);
2294
2295                         if (is_vring_iotlb(dev, vq, imsg))
2296                                 *pdev = dev = translate_ring_addresses(dev, i);
2297                 }
2298                 break;
2299         case VHOST_IOTLB_INVALIDATE:
2300                 for (i = 0; i < dev->nr_vring; i++) {
2301                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2302
2303                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2304                                         imsg->size);
2305
2306                         if (is_vring_iotlb(dev, vq, imsg))
2307                                 vring_invalidate(dev, vq);
2308                 }
2309                 break;
2310         default:
2311                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
2312                                 imsg->type);
2313                 return RTE_VHOST_MSG_RESULT_ERR;
2314         }
2315
2316         return RTE_VHOST_MSG_RESULT_OK;
2317 }
2318
2319 static int
2320 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2321                         struct VhostUserMsg *msg,
2322                         int main_fd __rte_unused)
2323 {
2324         struct virtio_net *dev = *pdev;
2325 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2326         struct uffdio_api api_struct;
2327
2328         if (validate_msg_fds(msg, 0) != 0)
2329                 return RTE_VHOST_MSG_RESULT_ERR;
2330
2331         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2332
2333         if (dev->postcopy_ufd == -1) {
2334                 RTE_LOG(ERR, VHOST_CONFIG, "Userfaultfd not available: %s\n",
2335                         strerror(errno));
2336                 return RTE_VHOST_MSG_RESULT_ERR;
2337         }
2338         api_struct.api = UFFD_API;
2339         api_struct.features = 0;
2340         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2341                 RTE_LOG(ERR, VHOST_CONFIG, "UFFDIO_API ioctl failure: %s\n",
2342                         strerror(errno));
2343                 close(dev->postcopy_ufd);
2344                 dev->postcopy_ufd = -1;
2345                 return RTE_VHOST_MSG_RESULT_ERR;
2346         }
2347         msg->fds[0] = dev->postcopy_ufd;
2348         msg->fd_num = 1;
2349
2350         return RTE_VHOST_MSG_RESULT_REPLY;
2351 #else
2352         dev->postcopy_ufd = -1;
2353         msg->fd_num = 0;
2354
2355         return RTE_VHOST_MSG_RESULT_ERR;
2356 #endif
2357 }
2358
2359 static int
2360 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2361                         struct VhostUserMsg *msg __rte_unused,
2362                         int main_fd __rte_unused)
2363 {
2364         struct virtio_net *dev = *pdev;
2365
2366         if (validate_msg_fds(msg, 0) != 0)
2367                 return RTE_VHOST_MSG_RESULT_ERR;
2368
2369         if (dev->mem && dev->mem->nregions) {
2370                 RTE_LOG(ERR, VHOST_CONFIG,
2371                         "Regions already registered at postcopy-listen\n");
2372                 return RTE_VHOST_MSG_RESULT_ERR;
2373         }
2374         dev->postcopy_listening = 1;
2375
2376         return RTE_VHOST_MSG_RESULT_OK;
2377 }
2378
2379 static int
2380 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2381                         int main_fd __rte_unused)
2382 {
2383         struct virtio_net *dev = *pdev;
2384
2385         if (validate_msg_fds(msg, 0) != 0)
2386                 return RTE_VHOST_MSG_RESULT_ERR;
2387
2388         dev->postcopy_listening = 0;
2389         if (dev->postcopy_ufd >= 0) {
2390                 close(dev->postcopy_ufd);
2391                 dev->postcopy_ufd = -1;
2392         }
2393
2394         msg->payload.u64 = 0;
2395         msg->size = sizeof(msg->payload.u64);
2396         msg->fd_num = 0;
2397
2398         return RTE_VHOST_MSG_RESULT_REPLY;
2399 }
2400
2401 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2402                                         struct VhostUserMsg *msg,
2403                                         int main_fd);
2404 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2405         [VHOST_USER_NONE] = NULL,
2406         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2407         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2408         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2409         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2410         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2411         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2412         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2413         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2414         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2415         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2416         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2417         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2418         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2419         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2420         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2421         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2422         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2423         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2424         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2425         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2426         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2427         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2428         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2429         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2430         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2431         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2432         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2433 };
2434
2435 /* return bytes# of read on success or negative val on failure. */
2436 static int
2437 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2438 {
2439         int ret;
2440
2441         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2442                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2443         if (ret <= 0)
2444                 return ret;
2445
2446         if (msg->size) {
2447                 if (msg->size > sizeof(msg->payload)) {
2448                         RTE_LOG(ERR, VHOST_CONFIG,
2449                                 "invalid msg size: %d\n", msg->size);
2450                         return -1;
2451                 }
2452                 ret = read(sockfd, &msg->payload, msg->size);
2453                 if (ret <= 0)
2454                         return ret;
2455                 if (ret != (int)msg->size) {
2456                         RTE_LOG(ERR, VHOST_CONFIG,
2457                                 "read control message failed\n");
2458                         return -1;
2459                 }
2460         }
2461
2462         return ret;
2463 }
2464
2465 static int
2466 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2467 {
2468         if (!msg)
2469                 return 0;
2470
2471         return send_fd_message(sockfd, (char *)msg,
2472                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2473 }
2474
2475 static int
2476 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2477 {
2478         if (!msg)
2479                 return 0;
2480
2481         msg->flags &= ~VHOST_USER_VERSION_MASK;
2482         msg->flags &= ~VHOST_USER_NEED_REPLY;
2483         msg->flags |= VHOST_USER_VERSION;
2484         msg->flags |= VHOST_USER_REPLY_MASK;
2485
2486         return send_vhost_message(sockfd, msg);
2487 }
2488
2489 static int
2490 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2491 {
2492         int ret;
2493
2494         if (msg->flags & VHOST_USER_NEED_REPLY)
2495                 rte_spinlock_lock(&dev->slave_req_lock);
2496
2497         ret = send_vhost_message(dev->slave_req_fd, msg);
2498         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2499                 rte_spinlock_unlock(&dev->slave_req_lock);
2500
2501         return ret;
2502 }
2503
2504 /*
2505  * Allocate a queue pair if it hasn't been allocated yet
2506  */
2507 static int
2508 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2509                         struct VhostUserMsg *msg)
2510 {
2511         uint16_t vring_idx;
2512
2513         switch (msg->request.master) {
2514         case VHOST_USER_SET_VRING_KICK:
2515         case VHOST_USER_SET_VRING_CALL:
2516         case VHOST_USER_SET_VRING_ERR:
2517                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2518                 break;
2519         case VHOST_USER_SET_VRING_NUM:
2520         case VHOST_USER_SET_VRING_BASE:
2521         case VHOST_USER_SET_VRING_ENABLE:
2522                 vring_idx = msg->payload.state.index;
2523                 break;
2524         case VHOST_USER_SET_VRING_ADDR:
2525                 vring_idx = msg->payload.addr.index;
2526                 break;
2527         default:
2528                 return 0;
2529         }
2530
2531         if (vring_idx >= VHOST_MAX_VRING) {
2532                 RTE_LOG(ERR, VHOST_CONFIG,
2533                         "invalid vring index: %u\n", vring_idx);
2534                 return -1;
2535         }
2536
2537         if (dev->virtqueue[vring_idx])
2538                 return 0;
2539
2540         return alloc_vring_queue(dev, vring_idx);
2541 }
2542
2543 static void
2544 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2545 {
2546         unsigned int i = 0;
2547         unsigned int vq_num = 0;
2548
2549         while (vq_num < dev->nr_vring) {
2550                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2551
2552                 if (vq) {
2553                         rte_spinlock_lock(&vq->access_lock);
2554                         vq_num++;
2555                 }
2556                 i++;
2557         }
2558 }
2559
2560 static void
2561 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2562 {
2563         unsigned int i = 0;
2564         unsigned int vq_num = 0;
2565
2566         while (vq_num < dev->nr_vring) {
2567                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2568
2569                 if (vq) {
2570                         rte_spinlock_unlock(&vq->access_lock);
2571                         vq_num++;
2572                 }
2573                 i++;
2574         }
2575 }
2576
2577 int
2578 vhost_user_msg_handler(int vid, int fd)
2579 {
2580         struct virtio_net *dev;
2581         struct VhostUserMsg msg;
2582         struct rte_vdpa_device *vdpa_dev;
2583         int did = -1;
2584         int ret;
2585         int unlock_required = 0;
2586         bool handled;
2587         int request;
2588
2589         dev = get_device(vid);
2590         if (dev == NULL)
2591                 return -1;
2592
2593         if (!dev->notify_ops) {
2594                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2595                 if (!dev->notify_ops) {
2596                         RTE_LOG(ERR, VHOST_CONFIG,
2597                                 "failed to get callback ops for driver %s\n",
2598                                 dev->ifname);
2599                         return -1;
2600                 }
2601         }
2602
2603         ret = read_vhost_message(fd, &msg);
2604         if (ret <= 0) {
2605                 if (ret < 0)
2606                         RTE_LOG(ERR, VHOST_CONFIG,
2607                                 "vhost read message failed\n");
2608                 else
2609                         RTE_LOG(INFO, VHOST_CONFIG,
2610                                 "vhost peer closed\n");
2611
2612                 return -1;
2613         }
2614
2615         ret = 0;
2616         request = msg.request.master;
2617         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2618                         vhost_message_str[request]) {
2619                 if (request != VHOST_USER_IOTLB_MSG)
2620                         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
2621                                 vhost_message_str[request]);
2622                 else
2623                         RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
2624                                 vhost_message_str[request]);
2625         } else {
2626                 RTE_LOG(DEBUG, VHOST_CONFIG, "External request %d\n", request);
2627         }
2628
2629         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2630         if (ret < 0) {
2631                 RTE_LOG(ERR, VHOST_CONFIG,
2632                         "failed to alloc queue\n");
2633                 return -1;
2634         }
2635
2636         /*
2637          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2638          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2639          * and device is destroyed. destroy_device waits for queues to be
2640          * inactive, so it is safe. Otherwise taking the access_lock
2641          * would cause a dead lock.
2642          */
2643         switch (request) {
2644         case VHOST_USER_SET_FEATURES:
2645         case VHOST_USER_SET_PROTOCOL_FEATURES:
2646         case VHOST_USER_SET_OWNER:
2647         case VHOST_USER_SET_MEM_TABLE:
2648         case VHOST_USER_SET_LOG_BASE:
2649         case VHOST_USER_SET_LOG_FD:
2650         case VHOST_USER_SET_VRING_NUM:
2651         case VHOST_USER_SET_VRING_ADDR:
2652         case VHOST_USER_SET_VRING_BASE:
2653         case VHOST_USER_SET_VRING_KICK:
2654         case VHOST_USER_SET_VRING_CALL:
2655         case VHOST_USER_SET_VRING_ERR:
2656         case VHOST_USER_SET_VRING_ENABLE:
2657         case VHOST_USER_SEND_RARP:
2658         case VHOST_USER_NET_SET_MTU:
2659         case VHOST_USER_SET_SLAVE_REQ_FD:
2660                 vhost_user_lock_all_queue_pairs(dev);
2661                 unlock_required = 1;
2662                 break;
2663         default:
2664                 break;
2665
2666         }
2667
2668         handled = false;
2669         if (dev->extern_ops.pre_msg_handle) {
2670                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2671                                 (void *)&msg);
2672                 switch (ret) {
2673                 case RTE_VHOST_MSG_RESULT_REPLY:
2674                         send_vhost_reply(fd, &msg);
2675                         /* Fall-through */
2676                 case RTE_VHOST_MSG_RESULT_ERR:
2677                 case RTE_VHOST_MSG_RESULT_OK:
2678                         handled = true;
2679                         goto skip_to_post_handle;
2680                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2681                 default:
2682                         break;
2683                 }
2684         }
2685
2686         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2687                 if (!vhost_message_handlers[request])
2688                         goto skip_to_post_handle;
2689                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2690
2691                 switch (ret) {
2692                 case RTE_VHOST_MSG_RESULT_ERR:
2693                         RTE_LOG(ERR, VHOST_CONFIG,
2694                                 "Processing %s failed.\n",
2695                                 vhost_message_str[request]);
2696                         handled = true;
2697                         break;
2698                 case RTE_VHOST_MSG_RESULT_OK:
2699                         RTE_LOG(DEBUG, VHOST_CONFIG,
2700                                 "Processing %s succeeded.\n",
2701                                 vhost_message_str[request]);
2702                         handled = true;
2703                         break;
2704                 case RTE_VHOST_MSG_RESULT_REPLY:
2705                         RTE_LOG(DEBUG, VHOST_CONFIG,
2706                                 "Processing %s succeeded and needs reply.\n",
2707                                 vhost_message_str[request]);
2708                         send_vhost_reply(fd, &msg);
2709                         handled = true;
2710                         break;
2711                 default:
2712                         break;
2713                 }
2714         }
2715
2716 skip_to_post_handle:
2717         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2718                         dev->extern_ops.post_msg_handle) {
2719                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2720                                 (void *)&msg);
2721                 switch (ret) {
2722                 case RTE_VHOST_MSG_RESULT_REPLY:
2723                         send_vhost_reply(fd, &msg);
2724                         /* Fall-through */
2725                 case RTE_VHOST_MSG_RESULT_ERR:
2726                 case RTE_VHOST_MSG_RESULT_OK:
2727                         handled = true;
2728                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2729                 default:
2730                         break;
2731                 }
2732         }
2733
2734         if (unlock_required)
2735                 vhost_user_unlock_all_queue_pairs(dev);
2736
2737         /* If message was not handled at this stage, treat it as an error */
2738         if (!handled) {
2739                 RTE_LOG(ERR, VHOST_CONFIG,
2740                         "vhost message (req: %d) was not handled.\n", request);
2741                 close_msg_fds(&msg);
2742                 ret = RTE_VHOST_MSG_RESULT_ERR;
2743         }
2744
2745         /*
2746          * If the request required a reply that was already sent,
2747          * this optional reply-ack won't be sent as the
2748          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2749          */
2750         if (msg.flags & VHOST_USER_NEED_REPLY) {
2751                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2752                 msg.size = sizeof(msg.payload.u64);
2753                 msg.fd_num = 0;
2754                 send_vhost_reply(fd, &msg);
2755         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2756                 RTE_LOG(ERR, VHOST_CONFIG,
2757                         "vhost message handling failed.\n");
2758                 return -1;
2759         }
2760
2761         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2762                 dev->flags |= VIRTIO_DEV_READY;
2763
2764                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2765                         if (dev->dequeue_zero_copy) {
2766                                 RTE_LOG(INFO, VHOST_CONFIG,
2767                                                 "dequeue zero copy is enabled\n");
2768                         }
2769
2770                         if (dev->notify_ops->new_device(dev->vid) == 0)
2771                                 dev->flags |= VIRTIO_DEV_RUNNING;
2772                 }
2773         }
2774
2775         did = dev->vdpa_dev_id;
2776         vdpa_dev = rte_vdpa_get_device(did);
2777         if (vdpa_dev && virtio_is_ready(dev) &&
2778                         !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2779                         msg.request.master == VHOST_USER_SET_VRING_CALL) {
2780                 if (vdpa_dev->ops->dev_conf)
2781                         vdpa_dev->ops->dev_conf(dev->vid);
2782                 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2783         }
2784
2785         return 0;
2786 }
2787
2788 static int process_slave_message_reply(struct virtio_net *dev,
2789                                        const struct VhostUserMsg *msg)
2790 {
2791         struct VhostUserMsg msg_reply;
2792         int ret;
2793
2794         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2795                 return 0;
2796
2797         if (read_vhost_message(dev->slave_req_fd, &msg_reply) < 0) {
2798                 ret = -1;
2799                 goto out;
2800         }
2801
2802         if (msg_reply.request.slave != msg->request.slave) {
2803                 RTE_LOG(ERR, VHOST_CONFIG,
2804                         "Received unexpected msg type (%u), expected %u\n",
2805                         msg_reply.request.slave, msg->request.slave);
2806                 ret = -1;
2807                 goto out;
2808         }
2809
2810         ret = msg_reply.payload.u64 ? -1 : 0;
2811
2812 out:
2813         rte_spinlock_unlock(&dev->slave_req_lock);
2814         return ret;
2815 }
2816
2817 int
2818 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2819 {
2820         int ret;
2821         struct VhostUserMsg msg = {
2822                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2823                 .flags = VHOST_USER_VERSION,
2824                 .size = sizeof(msg.payload.iotlb),
2825                 .payload.iotlb = {
2826                         .iova = iova,
2827                         .perm = perm,
2828                         .type = VHOST_IOTLB_MISS,
2829                 },
2830         };
2831
2832         ret = send_vhost_message(dev->slave_req_fd, &msg);
2833         if (ret < 0) {
2834                 RTE_LOG(ERR, VHOST_CONFIG,
2835                                 "Failed to send IOTLB miss message (%d)\n",
2836                                 ret);
2837                 return ret;
2838         }
2839
2840         return 0;
2841 }
2842
2843 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2844                                                     int index, int fd,
2845                                                     uint64_t offset,
2846                                                     uint64_t size)
2847 {
2848         int ret;
2849         struct VhostUserMsg msg = {
2850                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2851                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2852                 .size = sizeof(msg.payload.area),
2853                 .payload.area = {
2854                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
2855                         .size = size,
2856                         .offset = offset,
2857                 },
2858         };
2859
2860         if (fd < 0)
2861                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2862         else {
2863                 msg.fds[0] = fd;
2864                 msg.fd_num = 1;
2865         }
2866
2867         ret = send_vhost_slave_message(dev, &msg);
2868         if (ret < 0) {
2869                 RTE_LOG(ERR, VHOST_CONFIG,
2870                         "Failed to set host notifier (%d)\n", ret);
2871                 return ret;
2872         }
2873
2874         return process_slave_message_reply(dev, &msg);
2875 }
2876
2877 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2878 {
2879         struct virtio_net *dev;
2880         struct rte_vdpa_device *vdpa_dev;
2881         int vfio_device_fd, did, ret = 0;
2882         uint64_t offset, size;
2883         unsigned int i;
2884
2885         dev = get_device(vid);
2886         if (!dev)
2887                 return -ENODEV;
2888
2889         did = dev->vdpa_dev_id;
2890         if (did < 0)
2891                 return -EINVAL;
2892
2893         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2894             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2895             !(dev->protocol_features &
2896                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2897             !(dev->protocol_features &
2898                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2899             !(dev->protocol_features &
2900                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2901                 return -ENOTSUP;
2902
2903         vdpa_dev = rte_vdpa_get_device(did);
2904         if (!vdpa_dev)
2905                 return -ENODEV;
2906
2907         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2908         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2909
2910         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2911         if (vfio_device_fd < 0)
2912                 return -ENOTSUP;
2913
2914         if (enable) {
2915                 for (i = 0; i < dev->nr_vring; i++) {
2916                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2917                                         &size) < 0) {
2918                                 ret = -ENOTSUP;
2919                                 goto disable;
2920                         }
2921
2922                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
2923                                         vfio_device_fd, offset, size) < 0) {
2924                                 ret = -EFAULT;
2925                                 goto disable;
2926                         }
2927                 }
2928         } else {
2929 disable:
2930                 for (i = 0; i < dev->nr_vring; i++) {
2931                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
2932                                         0, 0);
2933                 }
2934         }
2935
2936         return ret;
2937 }