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