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