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         VHOST_LOG_CONFIG(ERR,
 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         rte_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                 VHOST_LOG_CONFIG(ERR,
 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                         VHOST_LOG_CONFIG(ERR,
 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         VHOST_LOG_CONFIG(INFO,
 359                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 360         VHOST_LOG_CONFIG(DEBUG,
 361                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 362                 dev->vid,
 363                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 364                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 365
 366         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 367             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 368                 /*
 369                  * Remove all but first queue pair if MQ hasn't been
 370                  * negotiated. This is safe because the device is not
 371                  * running at this stage.
 372                  */
 373                 while (dev->nr_vring > 2) {
 374                         struct vhost_virtqueue *vq;
 375
 376                         vq = dev->virtqueue[--dev->nr_vring];
 377                         if (!vq)
 378                                 continue;
 379
 380                         dev->virtqueue[dev->nr_vring] = NULL;
 381                         cleanup_vq(vq, 1);
 382                         cleanup_vq_inflight(dev, vq);
 383                         free_vq(dev, vq);
 384                 }
 385         }
 386
 387         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                         VHOST_LOG_CONFIG(ERR,
 423                                 "invalid virtqueue size %u\n", vq->size);
 424                         return RTE_VHOST_MSG_RESULT_ERR;
 425                 }
 426         }
 427
 428         if (vq->size > 32768) {
 429                 VHOST_LOG_CONFIG(ERR,
 430                         "invalid virtqueue size %u\n", vq->size);
 431                 return RTE_VHOST_MSG_RESULT_ERR;
 432         }
 433
 434         if (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                         VHOST_LOG_CONFIG(WARNING,
 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                         VHOST_LOG_CONFIG(ERR,
 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                         VHOST_LOG_CONFIG(ERR,
 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                 VHOST_LOG_CONFIG(ERR,
 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                 VHOST_LOG_CONFIG(ERR,
 522                         "Unable to get vq numa information.\n");
 523                 return dev;
 524         }
 525         if (oldnode != newnode) {
 526                 VHOST_LOG_CONFIG(INFO,
 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                 VHOST_LOG_CONFIG(ERR,
 583                         "Unable to get dev numa information.\n");
 584                 goto out;
 585         }
 586         if (oldnode != newnode) {
 587                 VHOST_LOG_CONFIG(INFO,
 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
 660                 vhost_user_iotlb_rd_lock(vq);
 661                 vva = vhost_iova_to_vva(dev, vq, ra,
 662                                         size, VHOST_ACCESS_RW);
 663                 vhost_user_iotlb_rd_unlock(vq);
 664
 665                 return vva;
 666         }
 667
 668         return qva_to_vva(dev, ra, size);
 669 }
 670
 671 static uint64_t
 672 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 673 {
 674         uint64_t log_gpa;
 675
 676         vhost_user_iotlb_rd_lock(vq);
 677         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 678         vhost_user_iotlb_rd_unlock(vq);
 679
 680         return log_gpa;
 681 }
 682
 683 static struct virtio_net *
 684 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 685 {
 686         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 687         struct vhost_vring_addr *addr = &vq->ring_addrs;
 688         uint64_t len, expected_len;
 689
 690         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 691                 vq->log_guest_addr =
 692                         log_addr_to_gpa(dev, vq);
 693                 if (vq->log_guest_addr == 0) {
 694                         VHOST_LOG_CONFIG(DEBUG,
 695                                 "(%d) failed to map log_guest_addr.\n",
 696                                 dev->vid);
 697                         return dev;
 698                 }
 699         }
 700
 701         if (vq_is_packed(dev)) {
 702                 len = sizeof(struct vring_packed_desc) * vq->size;
 703                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 704                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 705                 if (vq->desc_packed == NULL ||
 706                                 len != sizeof(struct vring_packed_desc) *
 707                                 vq->size) {
 708                         VHOST_LOG_CONFIG(DEBUG,
 709                                 "(%d) failed to map desc_packed ring.\n",
 710                                 dev->vid);
 711                         return dev;
 712                 }
 713
 714                 dev = numa_realloc(dev, vq_index);
 715                 vq = dev->virtqueue[vq_index];
 716                 addr = &vq->ring_addrs;
 717
 718                 len = sizeof(struct vring_packed_desc_event);
 719                 vq->driver_event = (struct vring_packed_desc_event *)
 720                                         (uintptr_t)ring_addr_to_vva(dev,
 721                                         vq, addr->avail_user_addr, &len);
 722                 if (vq->driver_event == NULL ||
 723                                 len != sizeof(struct vring_packed_desc_event)) {
 724                         VHOST_LOG_CONFIG(DEBUG,
 725                                 "(%d) failed to find driver area address.\n",
 726                                 dev->vid);
 727                         return dev;
 728                 }
 729
 730                 len = sizeof(struct vring_packed_desc_event);
 731                 vq->device_event = (struct vring_packed_desc_event *)
 732                                         (uintptr_t)ring_addr_to_vva(dev,
 733                                         vq, addr->used_user_addr, &len);
 734                 if (vq->device_event == NULL ||
 735                                 len != sizeof(struct vring_packed_desc_event)) {
 736                         VHOST_LOG_CONFIG(DEBUG,
 737                                 "(%d) failed to find device area address.\n",
 738                                 dev->vid);
 739                         return dev;
 740                 }
 741
 742                 vq->access_ok = 1;
 743                 return dev;
 744         }
 745
 746         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 747         if (vq->desc && vq->avail && vq->used)
 748                 return dev;
 749
 750         len = sizeof(struct vring_desc) * vq->size;
 751         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 752                         vq, addr->desc_user_addr, &len);
 753         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 754                 VHOST_LOG_CONFIG(DEBUG,
 755                         "(%d) failed to map desc ring.\n",
 756                         dev->vid);
 757                 return dev;
 758         }
 759
 760         dev = numa_realloc(dev, vq_index);
 761         vq = dev->virtqueue[vq_index];
 762         addr = &vq->ring_addrs;
 763
 764         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 765         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 766                 len += sizeof(uint16_t);
 767         expected_len = len;
 768         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 769                         vq, addr->avail_user_addr, &len);
 770         if (vq->avail == 0 || len != expected_len) {
 771                 VHOST_LOG_CONFIG(DEBUG,
 772                         "(%d) failed to map avail ring.\n",
 773                         dev->vid);
 774                 return dev;
 775         }
 776
 777         len = sizeof(struct vring_used) +
 778                 sizeof(struct vring_used_elem) * vq->size;
 779         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 780                 len += sizeof(uint16_t);
 781         expected_len = len;
 782         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 783                         vq, addr->used_user_addr, &len);
 784         if (vq->used == 0 || len != expected_len) {
 785                 VHOST_LOG_CONFIG(DEBUG,
 786                         "(%d) failed to map used ring.\n",
 787                         dev->vid);
 788                 return dev;
 789         }
 790
 791         if (vq->last_used_idx != vq->used->idx) {
 792                 VHOST_LOG_CONFIG(WARNING,
 793                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 794                         "some packets maybe resent for Tx and dropped for Rx\n",
 795                         vq->last_used_idx, vq->used->idx);
 796                 vq->last_used_idx  = vq->used->idx;
 797                 vq->last_avail_idx = vq->used->idx;
 798         }
 799
 800         vq->access_ok = 1;
 801
 802         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 803                         dev->vid, vq->desc);
 804         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 805                         dev->vid, vq->avail);
 806         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 807                         dev->vid, vq->used);
 808         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 809                         dev->vid, vq->log_guest_addr);
 810
 811         return dev;
 812 }
 813
 814 /*
 815  * The virtio device sends us the desc, used and avail ring addresses.
 816  * This function then converts these to our address space.
 817  */
 818 static int
 819 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 820                         int main_fd __rte_unused)
 821 {
 822         struct virtio_net *dev = *pdev;
 823         struct vhost_virtqueue *vq;
 824         struct vhost_vring_addr *addr = &msg->payload.addr;
 825         bool access_ok;
 826
 827         if (validate_msg_fds(msg, 0) != 0)
 828                 return RTE_VHOST_MSG_RESULT_ERR;
 829
 830         if (dev->mem == NULL)
 831                 return RTE_VHOST_MSG_RESULT_ERR;
 832
 833         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 834         vq = dev->virtqueue[msg->payload.addr.index];
 835
 836         access_ok = vq->access_ok;
 837
 838         /*
 839          * Rings addresses should not be interpreted as long as the ring is not
 840          * started and enabled
 841          */
 842         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 843
 844         vring_invalidate(dev, vq);
 845
 846         if ((vq->enabled && (dev->features &
 847                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 848                         access_ok) {
 849                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 850                 if (!dev)
 851                         return RTE_VHOST_MSG_RESULT_ERR;
 852
 853                 *pdev = dev;
 854         }
 855
 856         return RTE_VHOST_MSG_RESULT_OK;
 857 }
 858
 859 /*
 860  * The virtio device sends us the available ring last used index.
 861  */
 862 static int
 863 vhost_user_set_vring_base(struct virtio_net **pdev,
 864                         struct VhostUserMsg *msg,
 865                         int main_fd __rte_unused)
 866 {
 867         struct virtio_net *dev = *pdev;
 868         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 869         uint64_t val = msg->payload.state.num;
 870
 871         if (validate_msg_fds(msg, 0) != 0)
 872                 return RTE_VHOST_MSG_RESULT_ERR;
 873
 874         if (vq_is_packed(dev)) {
 875                 /*
 876                  * Bit[0:14]: avail index
 877                  * Bit[15]: avail wrap counter
 878                  */
 879                 vq->last_avail_idx = val & 0x7fff;
 880                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 881                 /*
 882                  * Set used index to same value as available one, as
 883                  * their values should be the same since ring processing
 884                  * was stopped at get time.
 885                  */
 886                 vq->last_used_idx = vq->last_avail_idx;
 887                 vq->used_wrap_counter = vq->avail_wrap_counter;
 888         } else {
 889                 vq->last_used_idx = msg->payload.state.num;
 890                 vq->last_avail_idx = msg->payload.state.num;
 891         }
 892
 893         return RTE_VHOST_MSG_RESULT_OK;
 894 }
 895
 896 static int
 897 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 898                    uint64_t host_phys_addr, uint64_t size)
 899 {
 900         struct guest_page *page, *last_page;
 901         struct guest_page *old_pages;
 902
 903         if (dev->nr_guest_pages == dev->max_guest_pages) {
 904                 dev->max_guest_pages *= 2;
 905                 old_pages = dev->guest_pages;
 906                 dev->guest_pages = rte_realloc(dev->guest_pages,
 907                                         dev->max_guest_pages * sizeof(*page),
 908                                         RTE_CACHE_LINE_SIZE);
 909                 if (dev->guest_pages == NULL) {
 910                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 911                         rte_free(old_pages);
 912                         return -1;
 913                 }
 914         }
 915
 916         if (dev->nr_guest_pages > 0) {
 917                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 918                 /* merge if the two pages are continuous */
 919                 if (host_phys_addr == last_page->host_phys_addr +
 920                                       last_page->size) {
 921                         last_page->size += size;
 922                         return 0;
 923                 }
 924         }
 925
 926         page = &dev->guest_pages[dev->nr_guest_pages++];
 927         page->guest_phys_addr = guest_phys_addr;
 928         page->host_phys_addr  = host_phys_addr;
 929         page->size = size;
 930
 931         return 0;
 932 }
 933
 934 static int
 935 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 936                 uint64_t page_size)
 937 {
 938         uint64_t reg_size = reg->size;
 939         uint64_t host_user_addr  = reg->host_user_addr;
 940         uint64_t guest_phys_addr = reg->guest_phys_addr;
 941         uint64_t host_phys_addr;
 942         uint64_t size;
 943
 944         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 945         size = page_size - (guest_phys_addr & (page_size - 1));
 946         size = RTE_MIN(size, reg_size);
 947
 948         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 949                 return -1;
 950
 951         host_user_addr  += size;
 952         guest_phys_addr += size;
 953         reg_size -= size;
 954
 955         while (reg_size > 0) {
 956                 size = RTE_MIN(reg_size, page_size);
 957                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 958                                                   host_user_addr);
 959                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 960                                 size) < 0)
 961                         return -1;
 962
 963                 host_user_addr  += size;
 964                 guest_phys_addr += size;
 965                 reg_size -= size;
 966         }
 967
 968         /* sort guest page array if over binary search threshold */
 969         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 970                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 971                         sizeof(struct guest_page), guest_page_addrcmp);
 972         }
 973
 974         return 0;
 975 }
 976
 977 #ifdef RTE_LIBRTE_VHOST_DEBUG
 978 /* TODO: enable it only in debug mode? */
 979 static void
 980 dump_guest_pages(struct virtio_net *dev)
 981 {
 982         uint32_t i;
 983         struct guest_page *page;
 984
 985         for (i = 0; i < dev->nr_guest_pages; i++) {
 986                 page = &dev->guest_pages[i];
 987
 988                 VHOST_LOG_CONFIG(INFO,
 989                         "guest physical page region %u\n"
 990                         "\t guest_phys_addr: %" PRIx64 "\n"
 991                         "\t host_phys_addr : %" PRIx64 "\n"
 992                         "\t size           : %" PRIx64 "\n",
 993                         i,
 994                         page->guest_phys_addr,
 995                         page->host_phys_addr,
 996                         page->size);
 997         }
 998 }
 999 #else
1000 #define dump_guest_pages(dev)
1001 #endif
1002
1003 static bool
1004 vhost_memory_changed(struct VhostUserMemory *new,
1005                      struct rte_vhost_memory *old)
1006 {
1007         uint32_t i;
1008
1009         if (new->nregions != old->nregions)
1010                 return true;
1011
1012         for (i = 0; i < new->nregions; ++i) {
1013                 VhostUserMemoryRegion *new_r = &new->regions[i];
1014                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1015
1016                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1017                         return true;
1018                 if (new_r->memory_size != old_r->size)
1019                         return true;
1020                 if (new_r->userspace_addr != old_r->guest_user_addr)
1021                         return true;
1022         }
1023
1024         return false;
1025 }
1026
1027 static int
1028 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1029                         int main_fd)
1030 {
1031         struct virtio_net *dev = *pdev;
1032         struct VhostUserMemory *memory = &msg->payload.memory;
1033         struct rte_vhost_mem_region *reg;
1034         void *mmap_addr;
1035         uint64_t mmap_size;
1036         uint64_t mmap_offset;
1037         uint64_t alignment;
1038         uint32_t i;
1039         int populate;
1040         int fd;
1041
1042         if (validate_msg_fds(msg, memory->nregions) != 0)
1043                 return RTE_VHOST_MSG_RESULT_ERR;
1044
1045         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1046                 VHOST_LOG_CONFIG(ERR,
1047                         "too many memory regions (%u)\n", memory->nregions);
1048                 return RTE_VHOST_MSG_RESULT_ERR;
1049         }
1050
1051         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1052                 VHOST_LOG_CONFIG(INFO,
1053                         "(%d) memory regions not changed\n", dev->vid);
1054
1055                 close_msg_fds(msg);
1056
1057                 return RTE_VHOST_MSG_RESULT_OK;
1058         }
1059
1060         if (dev->mem) {
1061                 free_mem_region(dev);
1062                 rte_free(dev->mem);
1063                 dev->mem = NULL;
1064         }
1065
1066         /* Flush IOTLB cache as previous HVAs are now invalid */
1067         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1068                 for (i = 0; i < dev->nr_vring; i++)
1069                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1070
1071         dev->nr_guest_pages = 0;
1072         if (dev->guest_pages == NULL) {
1073                 dev->max_guest_pages = 8;
1074                 dev->guest_pages = rte_zmalloc(NULL,
1075                                         dev->max_guest_pages *
1076                                         sizeof(struct guest_page),
1077                                         RTE_CACHE_LINE_SIZE);
1078                 if (dev->guest_pages == NULL) {
1079                         VHOST_LOG_CONFIG(ERR,
1080                                 "(%d) failed to allocate memory "
1081                                 "for dev->guest_pages\n",
1082                                 dev->vid);
1083                         return RTE_VHOST_MSG_RESULT_ERR;
1084                 }
1085         }
1086
1087         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1088                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1089         if (dev->mem == NULL) {
1090                 VHOST_LOG_CONFIG(ERR,
1091                         "(%d) failed to allocate memory for dev->mem\n",
1092                         dev->vid);
1093                 return RTE_VHOST_MSG_RESULT_ERR;
1094         }
1095         dev->mem->nregions = memory->nregions;
1096
1097         for (i = 0; i < memory->nregions; i++) {
1098                 fd  = msg->fds[i];
1099                 reg = &dev->mem->regions[i];
1100
1101                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1102                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1103                 reg->size            = memory->regions[i].memory_size;
1104                 reg->fd              = fd;
1105
1106                 mmap_offset = memory->regions[i].mmap_offset;
1107
1108                 /* Check for memory_size + mmap_offset overflow */
1109                 if (mmap_offset >= -reg->size) {
1110                         VHOST_LOG_CONFIG(ERR,
1111                                 "mmap_offset (%#"PRIx64") and memory_size "
1112                                 "(%#"PRIx64") overflow\n",
1113                                 mmap_offset, reg->size);
1114                         goto err_mmap;
1115                 }
1116
1117                 mmap_size = reg->size + mmap_offset;
1118
1119                 /* mmap() without flag of MAP_ANONYMOUS, should be called
1120                  * with length argument aligned with hugepagesz at older
1121                  * longterm version Linux, like 2.6.32 and 3.2.72, or
1122                  * mmap() will fail with EINVAL.
1123                  *
1124                  * to avoid failure, make sure in caller to keep length
1125                  * aligned.
1126                  */
1127                 alignment = get_blk_size(fd);
1128                 if (alignment == (uint64_t)-1) {
1129                         VHOST_LOG_CONFIG(ERR,
1130                                 "couldn't get hugepage size through fstat\n");
1131                         goto err_mmap;
1132                 }
1133                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1134                 if (mmap_size == 0) {
1135                         /*
1136                          * It could happen if initial mmap_size + alignment
1137                          * overflows the sizeof uint64, which could happen if
1138                          * either mmap_size or alignment value is wrong.
1139                          *
1140                          * mmap() kernel implementation would return an error,
1141                          * but better catch it before and provide useful info
1142                          * in the logs.
1143                          */
1144                         VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1145                                         "or alignment (0x%" PRIx64 ") is invalid\n",
1146                                         reg->size + mmap_offset, alignment);
1147                         goto err_mmap;
1148                 }
1149
1150                 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
1151                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1152                                  MAP_SHARED | populate, fd, 0);
1153
1154                 if (mmap_addr == MAP_FAILED) {
1155                         VHOST_LOG_CONFIG(ERR,
1156                                 "mmap region %u failed.\n", i);
1157                         goto err_mmap;
1158                 }
1159
1160                 reg->mmap_addr = mmap_addr;
1161                 reg->mmap_size = mmap_size;
1162                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1163                                       mmap_offset;
1164
1165                 if (dev->dequeue_zero_copy)
1166                         if (add_guest_pages(dev, reg, alignment) < 0) {
1167                                 VHOST_LOG_CONFIG(ERR,
1168                                         "adding guest pages to region %u failed.\n",
1169                                         i);
1170                                 goto err_mmap;
1171                         }
1172
1173                 VHOST_LOG_CONFIG(INFO,
1174                         "guest memory region %u, size: 0x%" PRIx64 "\n"
1175                         "\t guest physical addr: 0x%" PRIx64 "\n"
1176                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1177                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1178                         "\t mmap addr : 0x%" PRIx64 "\n"
1179                         "\t mmap size : 0x%" PRIx64 "\n"
1180                         "\t mmap align: 0x%" PRIx64 "\n"
1181                         "\t mmap off  : 0x%" PRIx64 "\n",
1182                         i, reg->size,
1183                         reg->guest_phys_addr,
1184                         reg->guest_user_addr,
1185                         reg->host_user_addr,
1186                         (uint64_t)(uintptr_t)mmap_addr,
1187                         mmap_size,
1188                         alignment,
1189                         mmap_offset);
1190
1191                 if (dev->postcopy_listening) {
1192                         /*
1193                          * We haven't a better way right now than sharing
1194                          * DPDK's virtual address with Qemu, so that Qemu can
1195                          * retrieve the region offset when handling userfaults.
1196                          */
1197                         memory->regions[i].userspace_addr =
1198                                 reg->host_user_addr;
1199                 }
1200         }
1201         if (dev->postcopy_listening) {
1202                 /* Send the addresses back to qemu */
1203                 msg->fd_num = 0;
1204                 send_vhost_reply(main_fd, msg);
1205
1206                 /* Wait for qemu to acknolwedge it's got the addresses
1207                  * we've got to wait before we're allowed to generate faults.
1208                  */
1209                 VhostUserMsg ack_msg;
1210                 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1211                         VHOST_LOG_CONFIG(ERR,
1212                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1213                         goto err_mmap;
1214                 }
1215
1216                 if (validate_msg_fds(&ack_msg, 0) != 0)
1217                         goto err_mmap;
1218
1219                 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1220                         VHOST_LOG_CONFIG(ERR,
1221                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1222                                 ack_msg.request.master);
1223                         goto err_mmap;
1224                 }
1225
1226                 /* Now userfault register and we can use the memory */
1227                 for (i = 0; i < memory->nregions; i++) {
1228 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1229                         reg = &dev->mem->regions[i];
1230                         struct uffdio_register reg_struct;
1231
1232                         /*
1233                          * Let's register all the mmap'ed area to ensure
1234                          * alignment on page boundary.
1235                          */
1236                         reg_struct.range.start =
1237                                 (uint64_t)(uintptr_t)reg->mmap_addr;
1238                         reg_struct.range.len = reg->mmap_size;
1239                         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1240
1241                         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1242                                                 &reg_struct)) {
1243                                 VHOST_LOG_CONFIG(ERR,
1244                                         "Failed to register ufd for region %d: (ufd = %d) %s\n",
1245                                         i, dev->postcopy_ufd,
1246                                         strerror(errno));
1247                                 goto err_mmap;
1248                         }
1249                         VHOST_LOG_CONFIG(INFO,
1250                                 "\t userfaultfd registered for range : "
1251                                 "%" PRIx64 " - %" PRIx64 "\n",
1252                                 (uint64_t)reg_struct.range.start,
1253                                 (uint64_t)reg_struct.range.start +
1254                                 (uint64_t)reg_struct.range.len - 1);
1255 #else
1256                         goto err_mmap;
1257 #endif
1258                 }
1259         }
1260
1261         for (i = 0; i < dev->nr_vring; i++) {
1262                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1263
1264                 if (vq->desc || vq->avail || vq->used) {
1265                         /*
1266                          * If the memory table got updated, the ring addresses
1267                          * need to be translated again as virtual addresses have
1268                          * changed.
1269                          */
1270                         vring_invalidate(dev, vq);
1271
1272                         dev = translate_ring_addresses(dev, i);
1273                         if (!dev) {
1274                                 dev = *pdev;
1275                                 goto err_mmap;
1276                         }
1277
1278                         *pdev = dev;
1279                 }
1280         }
1281
1282         dump_guest_pages(dev);
1283
1284         return RTE_VHOST_MSG_RESULT_OK;
1285
1286 err_mmap:
1287         free_mem_region(dev);
1288         rte_free(dev->mem);
1289         dev->mem = NULL;
1290         return RTE_VHOST_MSG_RESULT_ERR;
1291 }
1292
1293 static bool
1294 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1295 {
1296         bool rings_ok;
1297
1298         if (!vq)
1299                 return false;
1300
1301         if (vq_is_packed(dev))
1302                 rings_ok = vq->desc_packed && vq->driver_event &&
1303                         vq->device_event;
1304         else
1305                 rings_ok = vq->desc && vq->avail && vq->used;
1306
1307         return rings_ok &&
1308                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1309                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
1310 }
1311
1312 static int
1313 virtio_is_ready(struct virtio_net *dev)
1314 {
1315         struct vhost_virtqueue *vq;
1316         uint32_t i;
1317
1318         if (dev->nr_vring == 0)
1319                 return 0;
1320
1321         for (i = 0; i < dev->nr_vring; i++) {
1322                 vq = dev->virtqueue[i];
1323
1324                 if (!vq_is_ready(dev, vq))
1325                         return 0;
1326         }
1327
1328         VHOST_LOG_CONFIG(INFO,
1329                 "virtio is now ready for processing.\n");
1330         return 1;
1331 }
1332
1333 static void *
1334 inflight_mem_alloc(const char *name, size_t size, int *fd)
1335 {
1336         void *ptr;
1337         int mfd = -1;
1338         char fname[20] = "/tmp/memfd-XXXXXX";
1339
1340         *fd = -1;
1341 #ifdef MEMFD_SUPPORTED
1342         mfd = memfd_create(name, MFD_CLOEXEC);
1343 #else
1344         RTE_SET_USED(name);
1345 #endif
1346         if (mfd == -1) {
1347                 mfd = mkstemp(fname);
1348                 if (mfd == -1) {
1349                         VHOST_LOG_CONFIG(ERR,
1350                                 "failed to get inflight buffer fd\n");
1351                         return NULL;
1352                 }
1353
1354                 unlink(fname);
1355         }
1356
1357         if (ftruncate(mfd, size) == -1) {
1358                 VHOST_LOG_CONFIG(ERR,
1359                         "failed to alloc inflight buffer\n");
1360                 close(mfd);
1361                 return NULL;
1362         }
1363
1364         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1365         if (ptr == MAP_FAILED) {
1366                 VHOST_LOG_CONFIG(ERR,
1367                         "failed to mmap inflight buffer\n");
1368                 close(mfd);
1369                 return NULL;
1370         }
1371
1372         *fd = mfd;
1373         return ptr;
1374 }
1375
1376 static uint32_t
1377 get_pervq_shm_size_split(uint16_t queue_size)
1378 {
1379         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1380                                   queue_size + sizeof(uint64_t) +
1381                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1382 }
1383
1384 static uint32_t
1385 get_pervq_shm_size_packed(uint16_t queue_size)
1386 {
1387         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1388                                   * queue_size + sizeof(uint64_t) +
1389                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1390                                   INFLIGHT_ALIGNMENT);
1391 }
1392
1393 static int
1394 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1395                            VhostUserMsg *msg,
1396                            int main_fd __rte_unused)
1397 {
1398         struct rte_vhost_inflight_info_packed *inflight_packed;
1399         uint64_t pervq_inflight_size, mmap_size;
1400         uint16_t num_queues, queue_size;
1401         struct virtio_net *dev = *pdev;
1402         int fd, i, j;
1403         void *addr;
1404
1405         if (msg->size != sizeof(msg->payload.inflight)) {
1406                 VHOST_LOG_CONFIG(ERR,
1407                         "invalid get_inflight_fd message size is %d\n",
1408                         msg->size);
1409                 return RTE_VHOST_MSG_RESULT_ERR;
1410         }
1411
1412         if (dev->inflight_info == NULL) {
1413                 dev->inflight_info = calloc(1,
1414                                             sizeof(struct inflight_mem_info));
1415                 if (!dev->inflight_info) {
1416                         VHOST_LOG_CONFIG(ERR,
1417                                 "failed to alloc dev inflight area\n");
1418                         return RTE_VHOST_MSG_RESULT_ERR;
1419                 }
1420         }
1421
1422         num_queues = msg->payload.inflight.num_queues;
1423         queue_size = msg->payload.inflight.queue_size;
1424
1425         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1426                 msg->payload.inflight.num_queues);
1427         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1428                 msg->payload.inflight.queue_size);
1429
1430         if (vq_is_packed(dev))
1431                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1432         else
1433                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1434
1435         mmap_size = num_queues * pervq_inflight_size;
1436         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1437         if (!addr) {
1438                 VHOST_LOG_CONFIG(ERR,
1439                         "failed to alloc vhost inflight area\n");
1440                         msg->payload.inflight.mmap_size = 0;
1441                 return RTE_VHOST_MSG_RESULT_ERR;
1442         }
1443         memset(addr, 0, mmap_size);
1444
1445         if (dev->inflight_info->addr) {
1446                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1447                 dev->inflight_info->addr = NULL;
1448         }
1449
1450         dev->inflight_info->addr = addr;
1451         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1452         dev->inflight_info->fd = msg->fds[0] = fd;
1453         msg->payload.inflight.mmap_offset = 0;
1454         msg->fd_num = 1;
1455
1456         if (vq_is_packed(dev)) {
1457                 for (i = 0; i < num_queues; i++) {
1458                         inflight_packed =
1459                                 (struct rte_vhost_inflight_info_packed *)addr;
1460                         inflight_packed->used_wrap_counter = 1;
1461                         inflight_packed->old_used_wrap_counter = 1;
1462                         for (j = 0; j < queue_size; j++)
1463                                 inflight_packed->desc[j].next = j + 1;
1464                         addr = (void *)((char *)addr + pervq_inflight_size);
1465                 }
1466         }
1467
1468         VHOST_LOG_CONFIG(INFO,
1469                 "send inflight mmap_size: %"PRIu64"\n",
1470                 msg->payload.inflight.mmap_size);
1471         VHOST_LOG_CONFIG(INFO,
1472                 "send inflight mmap_offset: %"PRIu64"\n",
1473                 msg->payload.inflight.mmap_offset);
1474         VHOST_LOG_CONFIG(INFO,
1475                 "send inflight fd: %d\n", msg->fds[0]);
1476
1477         return RTE_VHOST_MSG_RESULT_REPLY;
1478 }
1479
1480 static int
1481 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1482                            int main_fd __rte_unused)
1483 {
1484         uint64_t mmap_size, mmap_offset;
1485         uint16_t num_queues, queue_size;
1486         struct virtio_net *dev = *pdev;
1487         uint32_t pervq_inflight_size;
1488         struct vhost_virtqueue *vq;
1489         void *addr;
1490         int fd, i;
1491
1492         fd = msg->fds[0];
1493         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1494                 VHOST_LOG_CONFIG(ERR,
1495                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1496                         msg->size, fd);
1497                 return RTE_VHOST_MSG_RESULT_ERR;
1498         }
1499
1500         mmap_size = msg->payload.inflight.mmap_size;
1501         mmap_offset = msg->payload.inflight.mmap_offset;
1502         num_queues = msg->payload.inflight.num_queues;
1503         queue_size = msg->payload.inflight.queue_size;
1504
1505         if (vq_is_packed(dev))
1506                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1507         else
1508                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1509
1510         VHOST_LOG_CONFIG(INFO,
1511                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1512         VHOST_LOG_CONFIG(INFO,
1513                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1514         VHOST_LOG_CONFIG(INFO,
1515                 "set_inflight_fd num_queues: %u\n", num_queues);
1516         VHOST_LOG_CONFIG(INFO,
1517                 "set_inflight_fd queue_size: %u\n", queue_size);
1518         VHOST_LOG_CONFIG(INFO,
1519                 "set_inflight_fd fd: %d\n", fd);
1520         VHOST_LOG_CONFIG(INFO,
1521                 "set_inflight_fd pervq_inflight_size: %d\n",
1522                 pervq_inflight_size);
1523
1524         if (!dev->inflight_info) {
1525                 dev->inflight_info = calloc(1,
1526                                             sizeof(struct inflight_mem_info));
1527                 if (dev->inflight_info == NULL) {
1528                         VHOST_LOG_CONFIG(ERR,
1529                                 "failed to alloc dev inflight area\n");
1530                         return RTE_VHOST_MSG_RESULT_ERR;
1531                 }
1532         }
1533
1534         if (dev->inflight_info->addr) {
1535                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1536                 dev->inflight_info->addr = NULL;
1537         }
1538
1539         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1540                     fd, mmap_offset);
1541         if (addr == MAP_FAILED) {
1542                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1543                 return RTE_VHOST_MSG_RESULT_ERR;
1544         }
1545
1546         if (dev->inflight_info->fd)
1547                 close(dev->inflight_info->fd);
1548
1549         dev->inflight_info->fd = fd;
1550         dev->inflight_info->addr = addr;
1551         dev->inflight_info->size = mmap_size;
1552
1553         for (i = 0; i < num_queues; i++) {
1554                 vq = dev->virtqueue[i];
1555                 if (vq_is_packed(dev)) {
1556                         vq->inflight_packed = addr;
1557                         vq->inflight_packed->desc_num = queue_size;
1558                 } else {
1559                         vq->inflight_split = addr;
1560                         vq->inflight_split->desc_num = queue_size;
1561                 }
1562                 addr = (void *)((char *)addr + pervq_inflight_size);
1563         }
1564
1565         return RTE_VHOST_MSG_RESULT_OK;
1566 }
1567
1568 static int
1569 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1570                         int main_fd __rte_unused)
1571 {
1572         struct virtio_net *dev = *pdev;
1573         struct vhost_vring_file file;
1574         struct vhost_virtqueue *vq;
1575         int expected_fds;
1576
1577         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1578         if (validate_msg_fds(msg, expected_fds) != 0)
1579                 return RTE_VHOST_MSG_RESULT_ERR;
1580
1581         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1582         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1583                 file.fd = VIRTIO_INVALID_EVENTFD;
1584         else
1585                 file.fd = msg->fds[0];
1586         VHOST_LOG_CONFIG(INFO,
1587                 "vring call idx:%d file:%d\n", file.index, file.fd);
1588
1589         vq = dev->virtqueue[file.index];
1590         if (vq->callfd >= 0)
1591                 close(vq->callfd);
1592
1593         vq->callfd = file.fd;
1594
1595         return RTE_VHOST_MSG_RESULT_OK;
1596 }
1597
1598 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1599                         struct VhostUserMsg *msg,
1600                         int main_fd __rte_unused)
1601 {
1602         int expected_fds;
1603
1604         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1605         if (validate_msg_fds(msg, expected_fds) != 0)
1606                 return RTE_VHOST_MSG_RESULT_ERR;
1607
1608         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1609                 close(msg->fds[0]);
1610         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1611
1612         return RTE_VHOST_MSG_RESULT_OK;
1613 }
1614
1615 static int
1616 resubmit_desc_compare(const void *a, const void *b)
1617 {
1618         const struct rte_vhost_resubmit_desc *desc0 = a;
1619         const struct rte_vhost_resubmit_desc *desc1 = b;
1620
1621         if (desc1->counter > desc0->counter)
1622                 return 1;
1623
1624         return -1;
1625 }
1626
1627 static int
1628 vhost_check_queue_inflights_split(struct virtio_net *dev,
1629                                   struct vhost_virtqueue *vq)
1630 {
1631         uint16_t i;
1632         uint16_t resubmit_num = 0, last_io, num;
1633         struct vring_used *used = vq->used;
1634         struct rte_vhost_resubmit_info *resubmit;
1635         struct rte_vhost_inflight_info_split *inflight_split;
1636
1637         if (!(dev->protocol_features &
1638             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1639                 return RTE_VHOST_MSG_RESULT_OK;
1640
1641         /* The frontend may still not support the inflight feature
1642          * although we negotiate the protocol feature.
1643          */
1644         if ((!vq->inflight_split))
1645                 return RTE_VHOST_MSG_RESULT_OK;
1646
1647         if (!vq->inflight_split->version) {
1648                 vq->inflight_split->version = INFLIGHT_VERSION;
1649                 return RTE_VHOST_MSG_RESULT_OK;
1650         }
1651
1652         if (vq->resubmit_inflight)
1653                 return RTE_VHOST_MSG_RESULT_OK;
1654
1655         inflight_split = vq->inflight_split;
1656         vq->global_counter = 0;
1657         last_io = inflight_split->last_inflight_io;
1658
1659         if (inflight_split->used_idx != used->idx) {
1660                 inflight_split->desc[last_io].inflight = 0;
1661                 rte_smp_mb();
1662                 inflight_split->used_idx = used->idx;
1663         }
1664
1665         for (i = 0; i < inflight_split->desc_num; i++) {
1666                 if (inflight_split->desc[i].inflight == 1)
1667                         resubmit_num++;
1668         }
1669
1670         vq->last_avail_idx += resubmit_num;
1671
1672         if (resubmit_num) {
1673                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1674                 if (!resubmit) {
1675                         VHOST_LOG_CONFIG(ERR,
1676                                 "failed to allocate memory for resubmit info.\n");
1677                         return RTE_VHOST_MSG_RESULT_ERR;
1678                 }
1679
1680                 resubmit->resubmit_list = calloc(resubmit_num,
1681                         sizeof(struct rte_vhost_resubmit_desc));
1682                 if (!resubmit->resubmit_list) {
1683                         VHOST_LOG_CONFIG(ERR,
1684                                 "failed to allocate memory for inflight desc.\n");
1685                         free(resubmit);
1686                         return RTE_VHOST_MSG_RESULT_ERR;
1687                 }
1688
1689                 num = 0;
1690                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1691                         if (vq->inflight_split->desc[i].inflight == 1) {
1692                                 resubmit->resubmit_list[num].index = i;
1693                                 resubmit->resubmit_list[num].counter =
1694                                         inflight_split->desc[i].counter;
1695                                 num++;
1696                         }
1697                 }
1698                 resubmit->resubmit_num = num;
1699
1700                 if (resubmit->resubmit_num > 1)
1701                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1702                               sizeof(struct rte_vhost_resubmit_desc),
1703                               resubmit_desc_compare);
1704
1705                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1706                 vq->resubmit_inflight = resubmit;
1707         }
1708
1709         return RTE_VHOST_MSG_RESULT_OK;
1710 }
1711
1712 static int
1713 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1714                                    struct vhost_virtqueue *vq)
1715 {
1716         uint16_t i;
1717         uint16_t resubmit_num = 0, old_used_idx, num;
1718         struct rte_vhost_resubmit_info *resubmit;
1719         struct rte_vhost_inflight_info_packed *inflight_packed;
1720
1721         if (!(dev->protocol_features &
1722             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1723                 return RTE_VHOST_MSG_RESULT_OK;
1724
1725         /* The frontend may still not support the inflight feature
1726          * although we negotiate the protocol feature.
1727          */
1728         if ((!vq->inflight_packed))
1729                 return RTE_VHOST_MSG_RESULT_OK;
1730
1731         if (!vq->inflight_packed->version) {
1732                 vq->inflight_packed->version = INFLIGHT_VERSION;
1733                 return RTE_VHOST_MSG_RESULT_OK;
1734         }
1735
1736         if (vq->resubmit_inflight)
1737                 return RTE_VHOST_MSG_RESULT_OK;
1738
1739         inflight_packed = vq->inflight_packed;
1740         vq->global_counter = 0;
1741         old_used_idx = inflight_packed->old_used_idx;
1742
1743         if (inflight_packed->used_idx != old_used_idx) {
1744                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1745                         inflight_packed->old_used_idx =
1746                                 inflight_packed->used_idx;
1747                         inflight_packed->old_used_wrap_counter =
1748                                 inflight_packed->used_wrap_counter;
1749                         inflight_packed->old_free_head =
1750                                 inflight_packed->free_head;
1751                 } else {
1752                         inflight_packed->used_idx =
1753                                 inflight_packed->old_used_idx;
1754                         inflight_packed->used_wrap_counter =
1755                                 inflight_packed->old_used_wrap_counter;
1756                         inflight_packed->free_head =
1757                                 inflight_packed->old_free_head;
1758                 }
1759         }
1760
1761         for (i = 0; i < inflight_packed->desc_num; i++) {
1762                 if (inflight_packed->desc[i].inflight == 1)
1763                         resubmit_num++;
1764         }
1765
1766         if (resubmit_num) {
1767                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1768                 if (resubmit == NULL) {
1769                         VHOST_LOG_CONFIG(ERR,
1770                                 "failed to allocate memory for resubmit info.\n");
1771                         return RTE_VHOST_MSG_RESULT_ERR;
1772                 }
1773
1774                 resubmit->resubmit_list = calloc(resubmit_num,
1775                         sizeof(struct rte_vhost_resubmit_desc));
1776                 if (resubmit->resubmit_list == NULL) {
1777                         VHOST_LOG_CONFIG(ERR,
1778                                 "failed to allocate memory for resubmit desc.\n");
1779                         free(resubmit);
1780                         return RTE_VHOST_MSG_RESULT_ERR;
1781                 }
1782
1783                 num = 0;
1784                 for (i = 0; i < inflight_packed->desc_num; i++) {
1785                         if (vq->inflight_packed->desc[i].inflight == 1) {
1786                                 resubmit->resubmit_list[num].index = i;
1787                                 resubmit->resubmit_list[num].counter =
1788                                         inflight_packed->desc[i].counter;
1789                                 num++;
1790                         }
1791                 }
1792                 resubmit->resubmit_num = num;
1793
1794                 if (resubmit->resubmit_num > 1)
1795                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1796                               sizeof(struct rte_vhost_resubmit_desc),
1797                               resubmit_desc_compare);
1798
1799                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1800                 vq->resubmit_inflight = resubmit;
1801         }
1802
1803         return RTE_VHOST_MSG_RESULT_OK;
1804 }
1805
1806 static int
1807 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1808                         int main_fd __rte_unused)
1809 {
1810         struct virtio_net *dev = *pdev;
1811         struct vhost_vring_file file;
1812         struct vhost_virtqueue *vq;
1813         int expected_fds;
1814
1815         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1816         if (validate_msg_fds(msg, expected_fds) != 0)
1817                 return RTE_VHOST_MSG_RESULT_ERR;
1818
1819         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1820         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1821                 file.fd = VIRTIO_INVALID_EVENTFD;
1822         else
1823                 file.fd = msg->fds[0];
1824         VHOST_LOG_CONFIG(INFO,
1825                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1826
1827         /* Interpret ring addresses only when ring is started. */
1828         dev = translate_ring_addresses(dev, file.index);
1829         if (!dev)
1830                 return RTE_VHOST_MSG_RESULT_ERR;
1831
1832         *pdev = dev;
1833
1834         vq = dev->virtqueue[file.index];
1835
1836         /*
1837          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1838          * the ring starts already enabled. Otherwise, it is enabled via
1839          * the SET_VRING_ENABLE message.
1840          */
1841         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1842                 vq->enabled = 1;
1843                 if (dev->notify_ops->vring_state_changed)
1844                         dev->notify_ops->vring_state_changed(
1845                                 dev->vid, file.index, 1);
1846         }
1847
1848         if (vq->kickfd >= 0)
1849                 close(vq->kickfd);
1850         vq->kickfd = file.fd;
1851
1852         if (vq_is_packed(dev)) {
1853                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1854                         VHOST_LOG_CONFIG(ERR,
1855                                 "failed to inflights for vq: %d\n", file.index);
1856                         return RTE_VHOST_MSG_RESULT_ERR;
1857                 }
1858         } else {
1859                 if (vhost_check_queue_inflights_split(dev, vq)) {
1860                         VHOST_LOG_CONFIG(ERR,
1861                                 "failed to inflights for vq: %d\n", file.index);
1862                         return RTE_VHOST_MSG_RESULT_ERR;
1863                 }
1864         }
1865
1866         return RTE_VHOST_MSG_RESULT_OK;
1867 }
1868
1869 static void
1870 free_zmbufs(struct vhost_virtqueue *vq)
1871 {
1872         drain_zmbuf_list(vq);
1873
1874         rte_free(vq->zmbufs);
1875 }
1876
1877 /*
1878  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1879  */
1880 static int
1881 vhost_user_get_vring_base(struct virtio_net **pdev,
1882                         struct VhostUserMsg *msg,
1883                         int main_fd __rte_unused)
1884 {
1885         struct virtio_net *dev = *pdev;
1886         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1887         uint64_t val;
1888
1889         if (validate_msg_fds(msg, 0) != 0)
1890                 return RTE_VHOST_MSG_RESULT_ERR;
1891
1892         /* We have to stop the queue (virtio) if it is running. */
1893         vhost_destroy_device_notify(dev);
1894
1895         dev->flags &= ~VIRTIO_DEV_READY;
1896         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1897
1898         /* Here we are safe to get the indexes */
1899         if (vq_is_packed(dev)) {
1900                 /*
1901                  * Bit[0:14]: avail index
1902                  * Bit[15]: avail wrap counter
1903                  */
1904                 val = vq->last_avail_idx & 0x7fff;
1905                 val |= vq->avail_wrap_counter << 15;
1906                 msg->payload.state.num = val;
1907         } else {
1908                 msg->payload.state.num = vq->last_avail_idx;
1909         }
1910
1911         VHOST_LOG_CONFIG(INFO,
1912                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1913                 msg->payload.state.num);
1914         /*
1915          * Based on current qemu vhost-user implementation, this message is
1916          * sent and only sent in vhost_vring_stop.
1917          * TODO: cleanup the vring, it isn't usable since here.
1918          */
1919         if (vq->kickfd >= 0)
1920                 close(vq->kickfd);
1921
1922         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1923
1924         if (vq->callfd >= 0)
1925                 close(vq->callfd);
1926
1927         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1928
1929         vq->signalled_used_valid = false;
1930
1931         if (dev->dequeue_zero_copy)
1932                 free_zmbufs(vq);
1933         if (vq_is_packed(dev)) {
1934                 rte_free(vq->shadow_used_packed);
1935                 vq->shadow_used_packed = NULL;
1936         } else {
1937                 rte_free(vq->shadow_used_split);
1938                 vq->shadow_used_split = NULL;
1939         }
1940
1941         rte_free(vq->batch_copy_elems);
1942         vq->batch_copy_elems = NULL;
1943
1944         msg->size = sizeof(msg->payload.state);
1945         msg->fd_num = 0;
1946
1947         vring_invalidate(dev, vq);
1948
1949         return RTE_VHOST_MSG_RESULT_REPLY;
1950 }
1951
1952 /*
1953  * when virtio queues are ready to work, qemu will send us to
1954  * enable the virtio queue pair.
1955  */
1956 static int
1957 vhost_user_set_vring_enable(struct virtio_net **pdev,
1958                         struct VhostUserMsg *msg,
1959                         int main_fd __rte_unused)
1960 {
1961         struct virtio_net *dev = *pdev;
1962         int enable = (int)msg->payload.state.num;
1963         int index = (int)msg->payload.state.index;
1964         struct rte_vdpa_device *vdpa_dev;
1965         int did = -1;
1966
1967         if (validate_msg_fds(msg, 0) != 0)
1968                 return RTE_VHOST_MSG_RESULT_ERR;
1969
1970         VHOST_LOG_CONFIG(INFO,
1971                 "set queue enable: %d to qp idx: %d\n",
1972                 enable, index);
1973
1974         did = dev->vdpa_dev_id;
1975         vdpa_dev = rte_vdpa_get_device(did);
1976         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1977                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1978
1979         if (dev->notify_ops->vring_state_changed)
1980                 dev->notify_ops->vring_state_changed(dev->vid,
1981                                 index, enable);
1982
1983         /* On disable, rings have to be stopped being processed. */
1984         if (!enable && dev->dequeue_zero_copy)
1985                 drain_zmbuf_list(dev->virtqueue[index]);
1986
1987         dev->virtqueue[index]->enabled = enable;
1988
1989         return RTE_VHOST_MSG_RESULT_OK;
1990 }
1991
1992 static int
1993 vhost_user_get_protocol_features(struct virtio_net **pdev,
1994                         struct VhostUserMsg *msg,
1995                         int main_fd __rte_unused)
1996 {
1997         struct virtio_net *dev = *pdev;
1998         uint64_t features, protocol_features;
1999
2000         if (validate_msg_fds(msg, 0) != 0)
2001                 return RTE_VHOST_MSG_RESULT_ERR;
2002
2003         rte_vhost_driver_get_features(dev->ifname, &features);
2004         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2005
2006         /*
2007          * REPLY_ACK protocol feature is only mandatory for now
2008          * for IOMMU feature. If IOMMU is explicitly disabled by the
2009          * application, disable also REPLY_ACK feature for older buggy
2010          * Qemu versions (from v2.7.0 to v2.9.0).
2011          */
2012         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
2013                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
2014
2015         msg->payload.u64 = protocol_features;
2016         msg->size = sizeof(msg->payload.u64);
2017         msg->fd_num = 0;
2018
2019         return RTE_VHOST_MSG_RESULT_REPLY;
2020 }
2021
2022 static int
2023 vhost_user_set_protocol_features(struct virtio_net **pdev,
2024                         struct VhostUserMsg *msg,
2025                         int main_fd __rte_unused)
2026 {
2027         struct virtio_net *dev = *pdev;
2028         uint64_t protocol_features = msg->payload.u64;
2029         uint64_t slave_protocol_features = 0;
2030
2031         if (validate_msg_fds(msg, 0) != 0)
2032                 return RTE_VHOST_MSG_RESULT_ERR;
2033
2034         rte_vhost_driver_get_protocol_features(dev->ifname,
2035                         &slave_protocol_features);
2036         if (protocol_features & ~slave_protocol_features) {
2037                 VHOST_LOG_CONFIG(ERR,
2038                         "(%d) received invalid protocol features.\n",
2039                         dev->vid);
2040                 return RTE_VHOST_MSG_RESULT_ERR;
2041         }
2042
2043         dev->protocol_features = protocol_features;
2044         VHOST_LOG_CONFIG(INFO,
2045                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2046                 dev->protocol_features);
2047
2048         return RTE_VHOST_MSG_RESULT_OK;
2049 }
2050
2051 static int
2052 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2053                         int main_fd __rte_unused)
2054 {
2055         struct virtio_net *dev = *pdev;
2056         int fd = msg->fds[0];
2057         uint64_t size, off;
2058         void *addr;
2059
2060         if (validate_msg_fds(msg, 1) != 0)
2061                 return RTE_VHOST_MSG_RESULT_ERR;
2062
2063         if (fd < 0) {
2064                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2065                 return RTE_VHOST_MSG_RESULT_ERR;
2066         }
2067
2068         if (msg->size != sizeof(VhostUserLog)) {
2069                 VHOST_LOG_CONFIG(ERR,
2070                         "invalid log base msg size: %"PRId32" != %d\n",
2071                         msg->size, (int)sizeof(VhostUserLog));
2072                 return RTE_VHOST_MSG_RESULT_ERR;
2073         }
2074
2075         size = msg->payload.log.mmap_size;
2076         off  = msg->payload.log.mmap_offset;
2077
2078         /* Check for mmap size and offset overflow. */
2079         if (off >= -size) {
2080                 VHOST_LOG_CONFIG(ERR,
2081                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2082                         off, size);
2083                 return RTE_VHOST_MSG_RESULT_ERR;
2084         }
2085
2086         VHOST_LOG_CONFIG(INFO,
2087                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2088                 size, off);
2089
2090         /*
2091          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2092          * fail when offset is not page size aligned.
2093          */
2094         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2095         close(fd);
2096         if (addr == MAP_FAILED) {
2097                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2098                 return RTE_VHOST_MSG_RESULT_ERR;
2099         }
2100
2101         /*
2102          * Free previously mapped log memory on occasionally
2103          * multiple VHOST_USER_SET_LOG_BASE.
2104          */
2105         if (dev->log_addr) {
2106                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2107         }
2108         dev->log_addr = (uint64_t)(uintptr_t)addr;
2109         dev->log_base = dev->log_addr + off;
2110         dev->log_size = size;
2111
2112         /*
2113          * The spec is not clear about it (yet), but QEMU doesn't expect
2114          * any payload in the reply.
2115          */
2116         msg->size = 0;
2117         msg->fd_num = 0;
2118
2119         return RTE_VHOST_MSG_RESULT_REPLY;
2120 }
2121
2122 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2123                         struct VhostUserMsg *msg,
2124                         int main_fd __rte_unused)
2125 {
2126         if (validate_msg_fds(msg, 1) != 0)
2127                 return RTE_VHOST_MSG_RESULT_ERR;
2128
2129         close(msg->fds[0]);
2130         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2131
2132         return RTE_VHOST_MSG_RESULT_OK;
2133 }
2134
2135 /*
2136  * An rarp packet is constructed and broadcasted to notify switches about
2137  * the new location of the migrated VM, so that packets from outside will
2138  * not be lost after migration.
2139  *
2140  * However, we don't actually "send" a rarp packet here, instead, we set
2141  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2142  */
2143 static int
2144 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2145                         int main_fd __rte_unused)
2146 {
2147         struct virtio_net *dev = *pdev;
2148         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2149         struct rte_vdpa_device *vdpa_dev;
2150         int did = -1;
2151
2152         if (validate_msg_fds(msg, 0) != 0)
2153                 return RTE_VHOST_MSG_RESULT_ERR;
2154
2155         VHOST_LOG_CONFIG(DEBUG,
2156                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2157                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2158         memcpy(dev->mac.addr_bytes, mac, 6);
2159
2160         /*
2161          * Set the flag to inject a RARP broadcast packet at
2162          * rte_vhost_dequeue_burst().
2163          *
2164          * __ATOMIC_RELEASE ordering is for making sure the mac is
2165          * copied before the flag is set.
2166          */
2167         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2168         did = dev->vdpa_dev_id;
2169         vdpa_dev = rte_vdpa_get_device(did);
2170         if (vdpa_dev && vdpa_dev->ops->migration_done)
2171                 vdpa_dev->ops->migration_done(dev->vid);
2172
2173         return RTE_VHOST_MSG_RESULT_OK;
2174 }
2175
2176 static int
2177 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2178                         int main_fd __rte_unused)
2179 {
2180         struct virtio_net *dev = *pdev;
2181
2182         if (validate_msg_fds(msg, 0) != 0)
2183                 return RTE_VHOST_MSG_RESULT_ERR;
2184
2185         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2186                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2187                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2188                                 msg->payload.u64);
2189
2190                 return RTE_VHOST_MSG_RESULT_ERR;
2191         }
2192
2193         dev->mtu = msg->payload.u64;
2194
2195         return RTE_VHOST_MSG_RESULT_OK;
2196 }
2197
2198 static int
2199 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2200                         int main_fd __rte_unused)
2201 {
2202         struct virtio_net *dev = *pdev;
2203         int fd = msg->fds[0];
2204
2205         if (validate_msg_fds(msg, 1) != 0)
2206                 return RTE_VHOST_MSG_RESULT_ERR;
2207
2208         if (fd < 0) {
2209                 VHOST_LOG_CONFIG(ERR,
2210                                 "Invalid file descriptor for slave channel (%d)\n",
2211                                 fd);
2212                 return RTE_VHOST_MSG_RESULT_ERR;
2213         }
2214
2215         if (dev->slave_req_fd >= 0)
2216                 close(dev->slave_req_fd);
2217
2218         dev->slave_req_fd = fd;
2219
2220         return RTE_VHOST_MSG_RESULT_OK;
2221 }
2222
2223 static int
2224 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2225 {
2226         struct vhost_vring_addr *ra;
2227         uint64_t start, end, len;
2228
2229         start = imsg->iova;
2230         end = start + imsg->size;
2231
2232         ra = &vq->ring_addrs;
2233         len = sizeof(struct vring_desc) * vq->size;
2234         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2235                 return 1;
2236
2237         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2238         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2239                 return 1;
2240
2241         len = sizeof(struct vring_used) +
2242                sizeof(struct vring_used_elem) * vq->size;
2243         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2244                 return 1;
2245
2246         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2247                 len = sizeof(uint64_t);
2248                 if (ra->log_guest_addr < end &&
2249                     (ra->log_guest_addr + len) > start)
2250                         return 1;
2251         }
2252
2253         return 0;
2254 }
2255
2256 static int
2257 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2258 {
2259         struct vhost_vring_addr *ra;
2260         uint64_t start, end, len;
2261
2262         start = imsg->iova;
2263         end = start + imsg->size;
2264
2265         ra = &vq->ring_addrs;
2266         len = sizeof(struct vring_packed_desc) * vq->size;
2267         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2268                 return 1;
2269
2270         len = sizeof(struct vring_packed_desc_event);
2271         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2272                 return 1;
2273
2274         len = sizeof(struct vring_packed_desc_event);
2275         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2276                 return 1;
2277
2278         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2279                 len = sizeof(uint64_t);
2280                 if (ra->log_guest_addr < end &&
2281                     (ra->log_guest_addr + len) > start)
2282                         return 1;
2283         }
2284
2285         return 0;
2286 }
2287
2288 static int is_vring_iotlb(struct virtio_net *dev,
2289                           struct vhost_virtqueue *vq,
2290                           struct vhost_iotlb_msg *imsg)
2291 {
2292         if (vq_is_packed(dev))
2293                 return is_vring_iotlb_packed(vq, imsg);
2294         else
2295                 return is_vring_iotlb_split(vq, imsg);
2296 }
2297
2298 static int
2299 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2300                         int main_fd __rte_unused)
2301 {
2302         struct virtio_net *dev = *pdev;
2303         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2304         uint16_t i;
2305         uint64_t vva, len;
2306
2307         if (validate_msg_fds(msg, 0) != 0)
2308                 return RTE_VHOST_MSG_RESULT_ERR;
2309
2310         switch (imsg->type) {
2311         case VHOST_IOTLB_UPDATE:
2312                 len = imsg->size;
2313                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2314                 if (!vva)
2315                         return RTE_VHOST_MSG_RESULT_ERR;
2316
2317                 for (i = 0; i < dev->nr_vring; i++) {
2318                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2319
2320                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2321                                         len, imsg->perm);
2322
2323                         if (is_vring_iotlb(dev, vq, imsg))
2324                                 *pdev = dev = translate_ring_addresses(dev, i);
2325                 }
2326                 break;
2327         case VHOST_IOTLB_INVALIDATE:
2328                 for (i = 0; i < dev->nr_vring; i++) {
2329                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2330
2331                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2332                                         imsg->size);
2333
2334                         if (is_vring_iotlb(dev, vq, imsg))
2335                                 vring_invalidate(dev, vq);
2336                 }
2337                 break;
2338         default:
2339                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2340                                 imsg->type);
2341                 return RTE_VHOST_MSG_RESULT_ERR;
2342         }
2343
2344         return RTE_VHOST_MSG_RESULT_OK;
2345 }
2346
2347 static int
2348 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2349                         struct VhostUserMsg *msg,
2350                         int main_fd __rte_unused)
2351 {
2352         struct virtio_net *dev = *pdev;
2353 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2354         struct uffdio_api api_struct;
2355
2356         if (validate_msg_fds(msg, 0) != 0)
2357                 return RTE_VHOST_MSG_RESULT_ERR;
2358
2359         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2360
2361         if (dev->postcopy_ufd == -1) {
2362                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2363                         strerror(errno));
2364                 return RTE_VHOST_MSG_RESULT_ERR;
2365         }
2366         api_struct.api = UFFD_API;
2367         api_struct.features = 0;
2368         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2369                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2370                         strerror(errno));
2371                 close(dev->postcopy_ufd);
2372                 dev->postcopy_ufd = -1;
2373                 return RTE_VHOST_MSG_RESULT_ERR;
2374         }
2375         msg->fds[0] = dev->postcopy_ufd;
2376         msg->fd_num = 1;
2377
2378         return RTE_VHOST_MSG_RESULT_REPLY;
2379 #else
2380         dev->postcopy_ufd = -1;
2381         msg->fd_num = 0;
2382
2383         return RTE_VHOST_MSG_RESULT_ERR;
2384 #endif
2385 }
2386
2387 static int
2388 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2389                         struct VhostUserMsg *msg __rte_unused,
2390                         int main_fd __rte_unused)
2391 {
2392         struct virtio_net *dev = *pdev;
2393
2394         if (validate_msg_fds(msg, 0) != 0)
2395                 return RTE_VHOST_MSG_RESULT_ERR;
2396
2397         if (dev->mem && dev->mem->nregions) {
2398                 VHOST_LOG_CONFIG(ERR,
2399                         "Regions already registered at postcopy-listen\n");
2400                 return RTE_VHOST_MSG_RESULT_ERR;
2401         }
2402         dev->postcopy_listening = 1;
2403
2404         return RTE_VHOST_MSG_RESULT_OK;
2405 }
2406
2407 static int
2408 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2409                         int main_fd __rte_unused)
2410 {
2411         struct virtio_net *dev = *pdev;
2412
2413         if (validate_msg_fds(msg, 0) != 0)
2414                 return RTE_VHOST_MSG_RESULT_ERR;
2415
2416         dev->postcopy_listening = 0;
2417         if (dev->postcopy_ufd >= 0) {
2418                 close(dev->postcopy_ufd);
2419                 dev->postcopy_ufd = -1;
2420         }
2421
2422         msg->payload.u64 = 0;
2423         msg->size = sizeof(msg->payload.u64);
2424         msg->fd_num = 0;
2425
2426         return RTE_VHOST_MSG_RESULT_REPLY;
2427 }
2428
2429 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2430                                         struct VhostUserMsg *msg,
2431                                         int main_fd);
2432 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2433         [VHOST_USER_NONE] = NULL,
2434         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2435         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2436         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2437         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2438         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2439         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2440         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2441         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2442         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2443         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2444         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2445         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2446         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2447         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2448         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2449         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2450         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2451         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2452         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2453         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2454         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2455         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2456         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2457         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2458         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2459         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2460         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2461 };
2462
2463 /* return bytes# of read on success or negative val on failure. */
2464 static int
2465 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2466 {
2467         int ret;
2468
2469         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2470                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2471         if (ret <= 0) {
2472                 return ret;
2473         } else if (ret != VHOST_USER_HDR_SIZE) {
2474                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2475                 close_msg_fds(msg);
2476                 return -1;
2477         }
2478
2479         if (msg->size) {
2480                 if (msg->size > sizeof(msg->payload)) {
2481                         VHOST_LOG_CONFIG(ERR,
2482                                 "invalid msg size: %d\n", msg->size);
2483                         return -1;
2484                 }
2485                 ret = read(sockfd, &msg->payload, msg->size);
2486                 if (ret <= 0)
2487                         return ret;
2488                 if (ret != (int)msg->size) {
2489                         VHOST_LOG_CONFIG(ERR,
2490                                 "read control message failed\n");
2491                         return -1;
2492                 }
2493         }
2494
2495         return ret;
2496 }
2497
2498 static int
2499 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2500 {
2501         if (!msg)
2502                 return 0;
2503
2504         return send_fd_message(sockfd, (char *)msg,
2505                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2506 }
2507
2508 static int
2509 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2510 {
2511         if (!msg)
2512                 return 0;
2513
2514         msg->flags &= ~VHOST_USER_VERSION_MASK;
2515         msg->flags &= ~VHOST_USER_NEED_REPLY;
2516         msg->flags |= VHOST_USER_VERSION;
2517         msg->flags |= VHOST_USER_REPLY_MASK;
2518
2519         return send_vhost_message(sockfd, msg);
2520 }
2521
2522 static int
2523 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2524 {
2525         int ret;
2526
2527         if (msg->flags & VHOST_USER_NEED_REPLY)
2528                 rte_spinlock_lock(&dev->slave_req_lock);
2529
2530         ret = send_vhost_message(dev->slave_req_fd, msg);
2531         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2532                 rte_spinlock_unlock(&dev->slave_req_lock);
2533
2534         return ret;
2535 }
2536
2537 /*
2538  * Allocate a queue pair if it hasn't been allocated yet
2539  */
2540 static int
2541 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2542                         struct VhostUserMsg *msg)
2543 {
2544         uint32_t vring_idx;
2545
2546         switch (msg->request.master) {
2547         case VHOST_USER_SET_VRING_KICK:
2548         case VHOST_USER_SET_VRING_CALL:
2549         case VHOST_USER_SET_VRING_ERR:
2550                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2551                 break;
2552         case VHOST_USER_SET_VRING_NUM:
2553         case VHOST_USER_SET_VRING_BASE:
2554         case VHOST_USER_SET_VRING_ENABLE:
2555                 vring_idx = msg->payload.state.index;
2556                 break;
2557         case VHOST_USER_SET_VRING_ADDR:
2558                 vring_idx = msg->payload.addr.index;
2559                 break;
2560         default:
2561                 return 0;
2562         }
2563
2564         if (vring_idx >= VHOST_MAX_VRING) {
2565                 VHOST_LOG_CONFIG(ERR,
2566                         "invalid vring index: %u\n", vring_idx);
2567                 return -1;
2568         }
2569
2570         if (dev->virtqueue[vring_idx])
2571                 return 0;
2572
2573         return alloc_vring_queue(dev, vring_idx);
2574 }
2575
2576 static void
2577 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2578 {
2579         unsigned int i = 0;
2580         unsigned int vq_num = 0;
2581
2582         while (vq_num < dev->nr_vring) {
2583                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2584
2585                 if (vq) {
2586                         rte_spinlock_lock(&vq->access_lock);
2587                         vq_num++;
2588                 }
2589                 i++;
2590         }
2591 }
2592
2593 static void
2594 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2595 {
2596         unsigned int i = 0;
2597         unsigned int vq_num = 0;
2598
2599         while (vq_num < dev->nr_vring) {
2600                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2601
2602                 if (vq) {
2603                         rte_spinlock_unlock(&vq->access_lock);
2604                         vq_num++;
2605                 }
2606                 i++;
2607         }
2608 }
2609
2610 int
2611 vhost_user_msg_handler(int vid, int fd)
2612 {
2613         struct virtio_net *dev;
2614         struct VhostUserMsg msg;
2615         struct rte_vdpa_device *vdpa_dev;
2616         int did = -1;
2617         int ret;
2618         int unlock_required = 0;
2619         bool handled;
2620         int request;
2621
2622         dev = get_device(vid);
2623         if (dev == NULL)
2624                 return -1;
2625
2626         if (!dev->notify_ops) {
2627                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2628                 if (!dev->notify_ops) {
2629                         VHOST_LOG_CONFIG(ERR,
2630                                 "failed to get callback ops for driver %s\n",
2631                                 dev->ifname);
2632                         return -1;
2633                 }
2634         }
2635
2636         ret = read_vhost_message(fd, &msg);
2637         if (ret <= 0) {
2638                 if (ret < 0)
2639                         VHOST_LOG_CONFIG(ERR,
2640                                 "vhost read message failed\n");
2641                 else
2642                         VHOST_LOG_CONFIG(INFO,
2643                                 "vhost peer closed\n");
2644
2645                 return -1;
2646         }
2647
2648         ret = 0;
2649         request = msg.request.master;
2650         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2651                         vhost_message_str[request]) {
2652                 if (request != VHOST_USER_IOTLB_MSG)
2653                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2654                                 vhost_message_str[request]);
2655                 else
2656                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2657                                 vhost_message_str[request]);
2658         } else {
2659                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2660         }
2661
2662         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2663         if (ret < 0) {
2664                 VHOST_LOG_CONFIG(ERR,
2665                         "failed to alloc queue\n");
2666                 return -1;
2667         }
2668
2669         /*
2670          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2671          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2672          * and device is destroyed. destroy_device waits for queues to be
2673          * inactive, so it is safe. Otherwise taking the access_lock
2674          * would cause a dead lock.
2675          */
2676         switch (request) {
2677         case VHOST_USER_SET_FEATURES:
2678         case VHOST_USER_SET_PROTOCOL_FEATURES:
2679         case VHOST_USER_SET_OWNER:
2680         case VHOST_USER_SET_MEM_TABLE:
2681         case VHOST_USER_SET_LOG_BASE:
2682         case VHOST_USER_SET_LOG_FD:
2683         case VHOST_USER_SET_VRING_NUM:
2684         case VHOST_USER_SET_VRING_ADDR:
2685         case VHOST_USER_SET_VRING_BASE:
2686         case VHOST_USER_SET_VRING_KICK:
2687         case VHOST_USER_SET_VRING_CALL:
2688         case VHOST_USER_SET_VRING_ERR:
2689         case VHOST_USER_SET_VRING_ENABLE:
2690         case VHOST_USER_SEND_RARP:
2691         case VHOST_USER_NET_SET_MTU:
2692         case VHOST_USER_SET_SLAVE_REQ_FD:
2693                 vhost_user_lock_all_queue_pairs(dev);
2694                 unlock_required = 1;
2695                 break;
2696         default:
2697                 break;
2698
2699         }
2700
2701         handled = false;
2702         if (dev->extern_ops.pre_msg_handle) {
2703                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2704                                 (void *)&msg);
2705                 switch (ret) {
2706                 case RTE_VHOST_MSG_RESULT_REPLY:
2707                         send_vhost_reply(fd, &msg);
2708                         /* Fall-through */
2709                 case RTE_VHOST_MSG_RESULT_ERR:
2710                 case RTE_VHOST_MSG_RESULT_OK:
2711                         handled = true;
2712                         goto skip_to_post_handle;
2713                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2714                 default:
2715                         break;
2716                 }
2717         }
2718
2719         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2720                 if (!vhost_message_handlers[request])
2721                         goto skip_to_post_handle;
2722                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2723
2724                 switch (ret) {
2725                 case RTE_VHOST_MSG_RESULT_ERR:
2726                         VHOST_LOG_CONFIG(ERR,
2727                                 "Processing %s failed.\n",
2728                                 vhost_message_str[request]);
2729                         handled = true;
2730                         break;
2731                 case RTE_VHOST_MSG_RESULT_OK:
2732                         VHOST_LOG_CONFIG(DEBUG,
2733                                 "Processing %s succeeded.\n",
2734                                 vhost_message_str[request]);
2735                         handled = true;
2736                         break;
2737                 case RTE_VHOST_MSG_RESULT_REPLY:
2738                         VHOST_LOG_CONFIG(DEBUG,
2739                                 "Processing %s succeeded and needs reply.\n",
2740                                 vhost_message_str[request]);
2741                         send_vhost_reply(fd, &msg);
2742                         handled = true;
2743                         break;
2744                 default:
2745                         break;
2746                 }
2747         }
2748
2749 skip_to_post_handle:
2750         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2751                         dev->extern_ops.post_msg_handle) {
2752                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2753                                 (void *)&msg);
2754                 switch (ret) {
2755                 case RTE_VHOST_MSG_RESULT_REPLY:
2756                         send_vhost_reply(fd, &msg);
2757                         /* Fall-through */
2758                 case RTE_VHOST_MSG_RESULT_ERR:
2759                 case RTE_VHOST_MSG_RESULT_OK:
2760                         handled = true;
2761                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2762                 default:
2763                         break;
2764                 }
2765         }
2766
2767         if (unlock_required)
2768                 vhost_user_unlock_all_queue_pairs(dev);
2769
2770         /* If message was not handled at this stage, treat it as an error */
2771         if (!handled) {
2772                 VHOST_LOG_CONFIG(ERR,
2773                         "vhost message (req: %d) was not handled.\n", request);
2774                 close_msg_fds(&msg);
2775                 ret = RTE_VHOST_MSG_RESULT_ERR;
2776         }
2777
2778         /*
2779          * If the request required a reply that was already sent,
2780          * this optional reply-ack won't be sent as the
2781          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2782          */
2783         if (msg.flags & VHOST_USER_NEED_REPLY) {
2784                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2785                 msg.size = sizeof(msg.payload.u64);
2786                 msg.fd_num = 0;
2787                 send_vhost_reply(fd, &msg);
2788         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2789                 VHOST_LOG_CONFIG(ERR,
2790                         "vhost message handling failed.\n");
2791                 return -1;
2792         }
2793
2794         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
2795                 dev->flags |= VIRTIO_DEV_READY;
2796
2797                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2798                         if (dev->dequeue_zero_copy) {
2799                                 VHOST_LOG_CONFIG(INFO,
2800                                                 "dequeue zero copy is enabled\n");
2801                         }
2802
2803                         if (dev->notify_ops->new_device(dev->vid) == 0)
2804                                 dev->flags |= VIRTIO_DEV_RUNNING;
2805                 }
2806         }
2807
2808         did = dev->vdpa_dev_id;
2809         vdpa_dev = rte_vdpa_get_device(did);
2810         if (vdpa_dev && virtio_is_ready(dev) &&
2811                         !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
2812                         msg.request.master == VHOST_USER_SET_VRING_CALL) {
2813                 if (vdpa_dev->ops->dev_conf)
2814                         vdpa_dev->ops->dev_conf(dev->vid);
2815                 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2816         }
2817
2818         return 0;
2819 }
2820
2821 static int process_slave_message_reply(struct virtio_net *dev,
2822                                        const struct VhostUserMsg *msg)
2823 {
2824         struct VhostUserMsg msg_reply;
2825         int ret;
2826
2827         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2828                 return 0;
2829
2830         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
2831         if (ret <= 0) {
2832                 if (ret < 0)
2833                         VHOST_LOG_CONFIG(ERR,
2834                                 "vhost read slave message reply failed\n");
2835                 else
2836                         VHOST_LOG_CONFIG(INFO,
2837                                 "vhost peer closed\n");
2838                 ret = -1;
2839                 goto out;
2840         }
2841
2842         ret = 0;
2843         if (msg_reply.request.slave != msg->request.slave) {
2844                 VHOST_LOG_CONFIG(ERR,
2845                         "Received unexpected msg type (%u), expected %u\n",
2846                         msg_reply.request.slave, msg->request.slave);
2847                 ret = -1;
2848                 goto out;
2849         }
2850
2851         ret = msg_reply.payload.u64 ? -1 : 0;
2852
2853 out:
2854         rte_spinlock_unlock(&dev->slave_req_lock);
2855         return ret;
2856 }
2857
2858 int
2859 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
2860 {
2861         int ret;
2862         struct VhostUserMsg msg = {
2863                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
2864                 .flags = VHOST_USER_VERSION,
2865                 .size = sizeof(msg.payload.iotlb),
2866                 .payload.iotlb = {
2867                         .iova = iova,
2868                         .perm = perm,
2869                         .type = VHOST_IOTLB_MISS,
2870                 },
2871         };
2872
2873         ret = send_vhost_message(dev->slave_req_fd, &msg);
2874         if (ret < 0) {
2875                 VHOST_LOG_CONFIG(ERR,
2876                                 "Failed to send IOTLB miss message (%d)\n",
2877                                 ret);
2878                 return ret;
2879         }
2880
2881         return 0;
2882 }
2883
2884 static int
2885 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
2886 {
2887         int ret;
2888         struct VhostUserMsg msg = {
2889                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
2890                 .flags = VHOST_USER_VERSION,
2891                 .size = 0,
2892         };
2893
2894         if (need_reply)
2895                 msg.flags |= VHOST_USER_NEED_REPLY;
2896
2897         ret = send_vhost_slave_message(dev, &msg);
2898         if (ret < 0) {
2899                 VHOST_LOG_CONFIG(ERR,
2900                                 "Failed to send config change (%d)\n",
2901                                 ret);
2902                 return ret;
2903         }
2904
2905         return process_slave_message_reply(dev, &msg);
2906 }
2907
2908 int
2909 rte_vhost_slave_config_change(int vid, bool need_reply)
2910 {
2911         struct virtio_net *dev;
2912
2913         dev = get_device(vid);
2914         if (!dev)
2915                 return -ENODEV;
2916
2917         return vhost_user_slave_config_change(dev, need_reply);
2918 }
2919
2920 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
2921                                                     int index, int fd,
2922                                                     uint64_t offset,
2923                                                     uint64_t size)
2924 {
2925         int ret;
2926         struct VhostUserMsg msg = {
2927                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
2928                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
2929                 .size = sizeof(msg.payload.area),
2930                 .payload.area = {
2931                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
2932                         .size = size,
2933                         .offset = offset,
2934                 },
2935         };
2936
2937         if (fd < 0)
2938                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
2939         else {
2940                 msg.fds[0] = fd;
2941                 msg.fd_num = 1;
2942         }
2943
2944         ret = send_vhost_slave_message(dev, &msg);
2945         if (ret < 0) {
2946                 VHOST_LOG_CONFIG(ERR,
2947                         "Failed to set host notifier (%d)\n", ret);
2948                 return ret;
2949         }
2950
2951         return process_slave_message_reply(dev, &msg);
2952 }
2953
2954 int rte_vhost_host_notifier_ctrl(int vid, bool enable)
2955 {
2956         struct virtio_net *dev;
2957         struct rte_vdpa_device *vdpa_dev;
2958         int vfio_device_fd, did, ret = 0;
2959         uint64_t offset, size;
2960         unsigned int i;
2961
2962         dev = get_device(vid);
2963         if (!dev)
2964                 return -ENODEV;
2965
2966         did = dev->vdpa_dev_id;
2967         if (did < 0)
2968                 return -EINVAL;
2969
2970         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
2971             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
2972             !(dev->protocol_features &
2973                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
2974             !(dev->protocol_features &
2975                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
2976             !(dev->protocol_features &
2977                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
2978                 return -ENOTSUP;
2979
2980         vdpa_dev = rte_vdpa_get_device(did);
2981         if (!vdpa_dev)
2982                 return -ENODEV;
2983
2984         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
2985         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
2986
2987         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
2988         if (vfio_device_fd < 0)
2989                 return -ENOTSUP;
2990
2991         if (enable) {
2992                 for (i = 0; i < dev->nr_vring; i++) {
2993                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
2994                                         &size) < 0) {
2995                                 ret = -ENOTSUP;
2996                                 goto disable;
2997                         }
2998
2999                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3000                                         vfio_device_fd, offset, size) < 0) {
3001                                 ret = -EFAULT;
3002                                 goto disable;
3003                         }
3004                 }
3005         } else {
3006 disable:
3007                 for (i = 0; i < dev->nr_vring; i++) {
3008                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3009                                         0, 0);
3010                 }
3011         }
3012
3013         return ret;
3014 }