lib/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 #include <rte_vfio.h>
  49 #include <rte_errno.h>
  50
  51 #include "iotlb.h"
  52 #include "vhost.h"
  53 #include "vhost_user.h"
  54
  55 #define VIRTIO_MIN_MTU 68
  56 #define VIRTIO_MAX_MTU 65535
  57
  58 #define INFLIGHT_ALIGNMENT      64
  59 #define INFLIGHT_VERSION        0x1
  60
  61 static const char *vhost_message_str[VHOST_USER_MAX] = {
  62         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  63         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  64         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  65         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  66         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  67         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  68         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  69         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  70         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  71         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  72         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  73         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  74         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  75         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  76         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  77         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  78         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  79         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  80         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  81         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  82         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  83         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  84         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  85         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  86         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  87         [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
  88         [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
  89         [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
  90         [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
  91         [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
  92         [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
  93         [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
  94 };
  95
  96 static int send_vhost_reply(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx);
  97 static int read_vhost_message(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx);
  98
  99 static void
 100 close_msg_fds(struct vhu_msg_context *ctx)
 101 {
 102         int i;
 103
 104         for (i = 0; i < ctx->fd_num; i++) {
 105                 int fd = ctx->fds[i];
 106
 107                 if (fd == -1)
 108                         continue;
 109
 110                 ctx->fds[i] = -1;
 111                 close(fd);
 112         }
 113 }
 114
 115 /*
 116  * Ensure the expected number of FDs is received,
 117  * close all FDs and return an error if this is not the case.
 118  */
 119 static int
 120 validate_msg_fds(struct virtio_net *dev, struct vhu_msg_context *ctx, int expected_fds)
 121 {
 122         if (ctx->fd_num == expected_fds)
 123                 return 0;
 124
 125         VHOST_LOG_CONFIG(ERR, "(%s) expect %d FDs for request %s, received %d\n",
 126                 dev->ifname, expected_fds,
 127                 vhost_message_str[ctx->msg.request.master],
 128                 ctx->fd_num);
 129
 130         close_msg_fds(ctx);
 131
 132         return -1;
 133 }
 134
 135 static uint64_t
 136 get_blk_size(int fd)
 137 {
 138         struct stat stat;
 139         int ret;
 140
 141         ret = fstat(fd, &stat);
 142         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 143 }
 144
 145 static int
 146 async_dma_map(struct virtio_net *dev, struct rte_vhost_mem_region *region, bool do_map)
 147 {
 148         uint64_t host_iova;
 149         int ret = 0;
 150
 151         host_iova = rte_mem_virt2iova((void *)(uintptr_t)region->host_user_addr);
 152         if (do_map) {
 153                 /* Add mapped region into the default container of DPDK. */
 154                 ret = rte_vfio_container_dma_map(RTE_VFIO_DEFAULT_CONTAINER_FD,
 155                                                  region->host_user_addr,
 156                                                  host_iova,
 157                                                  region->size);
 158                 if (ret) {
 159                         /*
 160                          * DMA device may bind with kernel driver, in this case,
 161                          * we don't need to program IOMMU manually. However, if no
 162                          * device is bound with vfio/uio in DPDK, and vfio kernel
 163                          * module is loaded, the API will still be called and return
 164                          * with ENODEV/ENOSUP.
 165                          *
 166                          * DPDK vfio only returns ENODEV/ENOSUP in very similar
 167                          * situations(vfio either unsupported, or supported
 168                          * but no devices found). Either way, no mappings could be
 169                          * performed. We treat it as normal case in async path.
 170                          */
 171                         if (rte_errno == ENODEV || rte_errno == ENOTSUP)
 172                                 return 0;
 173
 174                         VHOST_LOG_CONFIG(ERR, "(%s) DMA engine map failed\n", dev->ifname);
 175                         /* DMA mapping errors won't stop VHST_USER_SET_MEM_TABLE. */
 176                         return 0;
 177                 }
 178
 179         } else {
 180                 /* Remove mapped region from the default container of DPDK. */
 181                 ret = rte_vfio_container_dma_unmap(RTE_VFIO_DEFAULT_CONTAINER_FD,
 182                                                    region->host_user_addr,
 183                                                    host_iova,
 184                                                    region->size);
 185                 if (ret) {
 186                         /* like DMA map, ignore the kernel driver case when unmap. */
 187                         if (rte_errno == EINVAL)
 188                                 return 0;
 189
 190                         VHOST_LOG_CONFIG(ERR, "(%s) DMA engine unmap failed\n", dev->ifname);
 191                         return ret;
 192                 }
 193         }
 194
 195         return ret;
 196 }
 197
 198 static void
 199 free_mem_region(struct virtio_net *dev)
 200 {
 201         uint32_t i;
 202         struct rte_vhost_mem_region *reg;
 203
 204         if (!dev || !dev->mem)
 205                 return;
 206
 207         for (i = 0; i < dev->mem->nregions; i++) {
 208                 reg = &dev->mem->regions[i];
 209                 if (reg->host_user_addr) {
 210                         if (dev->async_copy && rte_vfio_is_enabled("vfio"))
 211                                 async_dma_map(dev, reg, false);
 212
 213                         munmap(reg->mmap_addr, reg->mmap_size);
 214                         close(reg->fd);
 215                 }
 216         }
 217 }
 218
 219 void
 220 vhost_backend_cleanup(struct virtio_net *dev)
 221 {
 222         if (dev->mem) {
 223                 free_mem_region(dev);
 224                 rte_free(dev->mem);
 225                 dev->mem = NULL;
 226         }
 227
 228         rte_free(dev->guest_pages);
 229         dev->guest_pages = NULL;
 230
 231         if (dev->log_addr) {
 232                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 233                 dev->log_addr = 0;
 234         }
 235
 236         if (dev->inflight_info) {
 237                 if (dev->inflight_info->addr) {
 238                         munmap(dev->inflight_info->addr,
 239                                dev->inflight_info->size);
 240                         dev->inflight_info->addr = NULL;
 241                 }
 242
 243                 if (dev->inflight_info->fd >= 0) {
 244                         close(dev->inflight_info->fd);
 245                         dev->inflight_info->fd = -1;
 246                 }
 247
 248                 rte_free(dev->inflight_info);
 249                 dev->inflight_info = NULL;
 250         }
 251
 252         if (dev->slave_req_fd >= 0) {
 253                 close(dev->slave_req_fd);
 254                 dev->slave_req_fd = -1;
 255         }
 256
 257         if (dev->postcopy_ufd >= 0) {
 258                 close(dev->postcopy_ufd);
 259                 dev->postcopy_ufd = -1;
 260         }
 261
 262         dev->postcopy_listening = 0;
 263 }
 264
 265 static void
 266 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
 267                               int enable)
 268 {
 269         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
 270         struct vhost_virtqueue *vq = dev->virtqueue[index];
 271
 272         /* Configure guest notifications on enable */
 273         if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
 274                 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
 275
 276         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
 277                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
 278
 279         if (dev->notify_ops->vring_state_changed)
 280                 dev->notify_ops->vring_state_changed(dev->vid,
 281                                 index, enable);
 282 }
 283
 284 /*
 285  * This function just returns success at the moment unless
 286  * the device hasn't been initialised.
 287  */
 288 static int
 289 vhost_user_set_owner(struct virtio_net **pdev,
 290                         struct vhu_msg_context *ctx,
 291                         int main_fd __rte_unused)
 292 {
 293         struct virtio_net *dev = *pdev;
 294
 295         if (validate_msg_fds(dev, ctx, 0) != 0)
 296                 return RTE_VHOST_MSG_RESULT_ERR;
 297
 298         return RTE_VHOST_MSG_RESULT_OK;
 299 }
 300
 301 static int
 302 vhost_user_reset_owner(struct virtio_net **pdev,
 303                         struct vhu_msg_context *ctx,
 304                         int main_fd __rte_unused)
 305 {
 306         struct virtio_net *dev = *pdev;
 307
 308         if (validate_msg_fds(dev, ctx, 0) != 0)
 309                 return RTE_VHOST_MSG_RESULT_ERR;
 310
 311         vhost_destroy_device_notify(dev);
 312
 313         cleanup_device(dev, 0);
 314         reset_device(dev);
 315         return RTE_VHOST_MSG_RESULT_OK;
 316 }
 317
 318 /*
 319  * The features that we support are requested.
 320  */
 321 static int
 322 vhost_user_get_features(struct virtio_net **pdev,
 323                         struct vhu_msg_context *ctx,
 324                         int main_fd __rte_unused)
 325 {
 326         struct virtio_net *dev = *pdev;
 327         uint64_t features = 0;
 328
 329         if (validate_msg_fds(dev, ctx, 0) != 0)
 330                 return RTE_VHOST_MSG_RESULT_ERR;
 331
 332         rte_vhost_driver_get_features(dev->ifname, &features);
 333
 334         ctx->msg.payload.u64 = features;
 335         ctx->msg.size = sizeof(ctx->msg.payload.u64);
 336         ctx->fd_num = 0;
 337
 338         return RTE_VHOST_MSG_RESULT_REPLY;
 339 }
 340
 341 /*
 342  * The queue number that we support are requested.
 343  */
 344 static int
 345 vhost_user_get_queue_num(struct virtio_net **pdev,
 346                         struct vhu_msg_context *ctx,
 347                         int main_fd __rte_unused)
 348 {
 349         struct virtio_net *dev = *pdev;
 350         uint32_t queue_num = 0;
 351
 352         if (validate_msg_fds(dev, ctx, 0) != 0)
 353                 return RTE_VHOST_MSG_RESULT_ERR;
 354
 355         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 356
 357         ctx->msg.payload.u64 = (uint64_t)queue_num;
 358         ctx->msg.size = sizeof(ctx->msg.payload.u64);
 359         ctx->fd_num = 0;
 360
 361         return RTE_VHOST_MSG_RESULT_REPLY;
 362 }
 363
 364 /*
 365  * We receive the negotiated features supported by us and the virtio device.
 366  */
 367 static int
 368 vhost_user_set_features(struct virtio_net **pdev,
 369                         struct vhu_msg_context *ctx,
 370                         int main_fd __rte_unused)
 371 {
 372         struct virtio_net *dev = *pdev;
 373         uint64_t features = ctx->msg.payload.u64;
 374         uint64_t vhost_features = 0;
 375         struct rte_vdpa_device *vdpa_dev;
 376
 377         if (validate_msg_fds(dev, ctx, 0) != 0)
 378                 return RTE_VHOST_MSG_RESULT_ERR;
 379
 380         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 381         if (features & ~vhost_features) {
 382                 VHOST_LOG_CONFIG(ERR, "(%s) received invalid negotiated features.\n",
 383                         dev->ifname);
 384                 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
 385                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
 386
 387                 return RTE_VHOST_MSG_RESULT_ERR;
 388         }
 389
 390         if (dev->flags & VIRTIO_DEV_RUNNING) {
 391                 if (dev->features == features)
 392                         return RTE_VHOST_MSG_RESULT_OK;
 393
 394                 /*
 395                  * Error out if master tries to change features while device is
 396                  * in running state. The exception being VHOST_F_LOG_ALL, which
 397                  * is enabled when the live-migration starts.
 398                  */
 399                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 400                         VHOST_LOG_CONFIG(ERR, "(%s) features changed while device is running.\n",
 401                                 dev->ifname);
 402                         return RTE_VHOST_MSG_RESULT_ERR;
 403                 }
 404
 405                 if (dev->notify_ops->features_changed)
 406                         dev->notify_ops->features_changed(dev->vid, features);
 407         }
 408
 409         dev->features = features;
 410         if (dev->features &
 411                 ((1ULL << VIRTIO_NET_F_MRG_RXBUF) |
 412                  (1ULL << VIRTIO_F_VERSION_1) |
 413                  (1ULL << VIRTIO_F_RING_PACKED))) {
 414                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 415         } else {
 416                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 417         }
 418         VHOST_LOG_CONFIG(INFO, "(%s) negotiated Virtio features: 0x%" PRIx64 "\n",
 419                         dev->ifname, dev->features);
 420         VHOST_LOG_CONFIG(DEBUG, "(%s) mergeable RX buffers %s, virtio 1 %s\n",
 421                 dev->ifname,
 422                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 423                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 424
 425         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 426             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 427                 /*
 428                  * Remove all but first queue pair if MQ hasn't been
 429                  * negotiated. This is safe because the device is not
 430                  * running at this stage.
 431                  */
 432                 while (dev->nr_vring > 2) {
 433                         struct vhost_virtqueue *vq;
 434
 435                         vq = dev->virtqueue[--dev->nr_vring];
 436                         if (!vq)
 437                                 continue;
 438
 439                         dev->virtqueue[dev->nr_vring] = NULL;
 440                         cleanup_vq(vq, 1);
 441                         cleanup_vq_inflight(dev, vq);
 442                         free_vq(dev, vq);
 443                 }
 444         }
 445
 446         vdpa_dev = dev->vdpa_dev;
 447         if (vdpa_dev)
 448                 vdpa_dev->ops->set_features(dev->vid);
 449
 450         dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
 451         return RTE_VHOST_MSG_RESULT_OK;
 452 }
 453
 454 /*
 455  * The virtio device sends us the size of the descriptor ring.
 456  */
 457 static int
 458 vhost_user_set_vring_num(struct virtio_net **pdev,
 459                         struct vhu_msg_context *ctx,
 460                         int main_fd __rte_unused)
 461 {
 462         struct virtio_net *dev = *pdev;
 463         struct vhost_virtqueue *vq = dev->virtqueue[ctx->msg.payload.state.index];
 464
 465         if (validate_msg_fds(dev, ctx, 0) != 0)
 466                 return RTE_VHOST_MSG_RESULT_ERR;
 467
 468         if (ctx->msg.payload.state.num > 32768) {
 469                 VHOST_LOG_CONFIG(ERR, "(%s) invalid virtqueue size %u\n",
 470                                 dev->ifname, ctx->msg.payload.state.num);
 471                 return RTE_VHOST_MSG_RESULT_ERR;
 472         }
 473
 474         vq->size = ctx->msg.payload.state.num;
 475
 476         /* VIRTIO 1.0, 2.4 Virtqueues says:
 477          *
 478          *   Queue Size value is always a power of 2. The maximum Queue Size
 479          *   value is 32768.
 480          *
 481          * VIRTIO 1.1 2.7 Virtqueues says:
 482          *
 483          *   Packed virtqueues support up to 2^15 entries each.
 484          */
 485         if (!vq_is_packed(dev)) {
 486                 if (vq->size & (vq->size - 1)) {
 487                         VHOST_LOG_CONFIG(ERR, "(%s) invalid virtqueue size %u\n",
 488                                         dev->ifname, vq->size);
 489                         return RTE_VHOST_MSG_RESULT_ERR;
 490                 }
 491         }
 492
 493         if (vq_is_packed(dev)) {
 494                 rte_free(vq->shadow_used_packed);
 495                 vq->shadow_used_packed = rte_malloc_socket(NULL,
 496                                 vq->size *
 497                                 sizeof(struct vring_used_elem_packed),
 498                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 499                 if (!vq->shadow_used_packed) {
 500                         VHOST_LOG_CONFIG(ERR,
 501                                 "(%s) failed to allocate memory for shadow used ring.\n",
 502                                 dev->ifname);
 503                         return RTE_VHOST_MSG_RESULT_ERR;
 504                 }
 505
 506         } else {
 507                 rte_free(vq->shadow_used_split);
 508
 509                 vq->shadow_used_split = rte_malloc_socket(NULL,
 510                                 vq->size * sizeof(struct vring_used_elem),
 511                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 512
 513                 if (!vq->shadow_used_split) {
 514                         VHOST_LOG_CONFIG(ERR,
 515                                 "(%s) failed to allocate memory for vq internal data.\n",
 516                                 dev->ifname);
 517                         return RTE_VHOST_MSG_RESULT_ERR;
 518                 }
 519         }
 520
 521         rte_free(vq->batch_copy_elems);
 522         vq->batch_copy_elems = rte_malloc_socket(NULL,
 523                                 vq->size * sizeof(struct batch_copy_elem),
 524                                 RTE_CACHE_LINE_SIZE, vq->numa_node);
 525         if (!vq->batch_copy_elems) {
 526                 VHOST_LOG_CONFIG(ERR, "(%s) failed to allocate memory for batching copy.\n",
 527                         dev->ifname);
 528                 return RTE_VHOST_MSG_RESULT_ERR;
 529         }
 530
 531         return RTE_VHOST_MSG_RESULT_OK;
 532 }
 533
 534 /*
 535  * Reallocate virtio_dev, vhost_virtqueue and related data structures to
 536  * make them on the same numa node as the memory of vring descriptor.
 537  */
 538 #ifdef RTE_LIBRTE_VHOST_NUMA
 539 static struct virtio_net*
 540 numa_realloc(struct virtio_net *dev, int index)
 541 {
 542         int node, dev_node;
 543         struct virtio_net *old_dev;
 544         struct vhost_virtqueue *vq;
 545         struct batch_copy_elem *bce;
 546         struct guest_page *gp;
 547         struct rte_vhost_memory *mem;
 548         size_t mem_size;
 549         int ret;
 550
 551         old_dev = dev;
 552         vq = dev->virtqueue[index];
 553
 554         /*
 555          * If VQ is ready, it is too late to reallocate, it certainly already
 556          * happened anyway on VHOST_USER_SET_VRING_ADRR.
 557          */
 558         if (vq->ready)
 559                 return dev;
 560
 561         ret = get_mempolicy(&node, NULL, 0, vq->desc, MPOL_F_NODE | MPOL_F_ADDR);
 562         if (ret) {
 563                 VHOST_LOG_CONFIG(ERR, "(%s) unable to get virtqueue %d numa information.\n",
 564                                 dev->ifname, index);
 565                 return dev;
 566         }
 567
 568         if (node == vq->numa_node)
 569                 goto out_dev_realloc;
 570
 571         vq = rte_realloc_socket(vq, sizeof(*vq), 0, node);
 572         if (!vq) {
 573                 VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc virtqueue %d on node %d\n",
 574                                 dev->ifname, index, node);
 575                 return dev;
 576         }
 577
 578         if (vq != dev->virtqueue[index]) {
 579                 VHOST_LOG_CONFIG(INFO, "(%s) reallocated virtqueue on node %d\n",
 580                                 dev->ifname, node);
 581                 dev->virtqueue[index] = vq;
 582                 vhost_user_iotlb_init(dev, index);
 583         }
 584
 585         if (vq_is_packed(dev)) {
 586                 struct vring_used_elem_packed *sup;
 587
 588                 sup = rte_realloc_socket(vq->shadow_used_packed, vq->size * sizeof(*sup),
 589                                 RTE_CACHE_LINE_SIZE, node);
 590                 if (!sup) {
 591                         VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc shadow packed on node %d\n",
 592                                         dev->ifname, node);
 593                         return dev;
 594                 }
 595                 vq->shadow_used_packed = sup;
 596         } else {
 597                 struct vring_used_elem *sus;
 598
 599                 sus = rte_realloc_socket(vq->shadow_used_split, vq->size * sizeof(*sus),
 600                                 RTE_CACHE_LINE_SIZE, node);
 601                 if (!sus) {
 602                         VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc shadow split on node %d\n",
 603                                         dev->ifname, node);
 604                         return dev;
 605                 }
 606                 vq->shadow_used_split = sus;
 607         }
 608
 609         bce = rte_realloc_socket(vq->batch_copy_elems, vq->size * sizeof(*bce),
 610                         RTE_CACHE_LINE_SIZE, node);
 611         if (!bce) {
 612                 VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc batch copy elem on node %d\n",
 613                                 dev->ifname, node);
 614                 return dev;
 615         }
 616         vq->batch_copy_elems = bce;
 617
 618         if (vq->log_cache) {
 619                 struct log_cache_entry *lc;
 620
 621                 lc = rte_realloc_socket(vq->log_cache, sizeof(*lc) * VHOST_LOG_CACHE_NR, 0, node);
 622                 if (!lc) {
 623                         VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc log cache on node %d\n",
 624                                         dev->ifname, node);
 625                         return dev;
 626                 }
 627                 vq->log_cache = lc;
 628         }
 629
 630         if (vq->resubmit_inflight) {
 631                 struct rte_vhost_resubmit_info *ri;
 632
 633                 ri = rte_realloc_socket(vq->resubmit_inflight, sizeof(*ri), 0, node);
 634                 if (!ri) {
 635                         VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc resubmit inflight on node %d\n",
 636                                         dev->ifname, node);
 637                         return dev;
 638                 }
 639                 vq->resubmit_inflight = ri;
 640
 641                 if (ri->resubmit_list) {
 642                         struct rte_vhost_resubmit_desc *rd;
 643
 644                         rd = rte_realloc_socket(ri->resubmit_list, sizeof(*rd) * ri->resubmit_num,
 645                                         0, node);
 646                         if (!rd) {
 647                                 VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc resubmit list on node %d\n",
 648                                                 dev->ifname, node);
 649                                 return dev;
 650                         }
 651                         ri->resubmit_list = rd;
 652                 }
 653         }
 654
 655         vq->numa_node = node;
 656
 657 out_dev_realloc:
 658
 659         if (dev->flags & VIRTIO_DEV_RUNNING)
 660                 return dev;
 661
 662         ret = get_mempolicy(&dev_node, NULL, 0, dev, MPOL_F_NODE | MPOL_F_ADDR);
 663         if (ret) {
 664                 VHOST_LOG_CONFIG(ERR, "(%s) unable to get numa information.\n", dev->ifname);
 665                 return dev;
 666         }
 667
 668         if (dev_node == node)
 669                 return dev;
 670
 671         dev = rte_realloc_socket(old_dev, sizeof(*dev), 0, node);
 672         if (!dev) {
 673                 VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc dev on node %d\n",
 674                                 dev->ifname, node);
 675                 return old_dev;
 676         }
 677
 678         VHOST_LOG_CONFIG(INFO, "(%s) reallocated device on node %d\n", dev->ifname, node);
 679         vhost_devices[dev->vid] = dev;
 680
 681         mem_size = sizeof(struct rte_vhost_memory) +
 682                 sizeof(struct rte_vhost_mem_region) * dev->mem->nregions;
 683         mem = rte_realloc_socket(dev->mem, mem_size, 0, node);
 684         if (!mem) {
 685                 VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc mem table on node %d\n",
 686                                 dev->ifname, node);
 687                 return dev;
 688         }
 689         dev->mem = mem;
 690
 691         gp = rte_realloc_socket(dev->guest_pages, dev->max_guest_pages * sizeof(*gp),
 692                         RTE_CACHE_LINE_SIZE, node);
 693         if (!gp) {
 694                 VHOST_LOG_CONFIG(ERR, "(%s) failed to realloc guest pages on node %d\n",
 695                                 dev->ifname, node);
 696                 return dev;
 697         }
 698         dev->guest_pages = gp;
 699
 700         return dev;
 701 }
 702 #else
 703 static struct virtio_net*
 704 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 705 {
 706         return dev;
 707 }
 708 #endif
 709
 710 /* Converts QEMU virtual address to Vhost virtual address. */
 711 static uint64_t
 712 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 713 {
 714         struct rte_vhost_mem_region *r;
 715         uint32_t i;
 716
 717         if (unlikely(!dev || !dev->mem))
 718                 goto out_error;
 719
 720         /* Find the region where the address lives. */
 721         for (i = 0; i < dev->mem->nregions; i++) {
 722                 r = &dev->mem->regions[i];
 723
 724                 if (qva >= r->guest_user_addr &&
 725                     qva <  r->guest_user_addr + r->size) {
 726
 727                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 728                                 *len = r->guest_user_addr + r->size - qva;
 729
 730                         return qva - r->guest_user_addr +
 731                                r->host_user_addr;
 732                 }
 733         }
 734 out_error:
 735         *len = 0;
 736
 737         return 0;
 738 }
 739
 740
 741 /*
 742  * Converts ring address to Vhost virtual address.
 743  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 744  * else it is a QEMU virtual address.
 745  */
 746 static uint64_t
 747 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 748                 uint64_t ra, uint64_t *size)
 749 {
 750         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 751                 uint64_t vva;
 752
 753                 vhost_user_iotlb_rd_lock(vq);
 754                 vva = vhost_iova_to_vva(dev, vq, ra,
 755                                         size, VHOST_ACCESS_RW);
 756                 vhost_user_iotlb_rd_unlock(vq);
 757
 758                 return vva;
 759         }
 760
 761         return qva_to_vva(dev, ra, size);
 762 }
 763
 764 static uint64_t
 765 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 766 {
 767         uint64_t log_gpa;
 768
 769         vhost_user_iotlb_rd_lock(vq);
 770         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 771         vhost_user_iotlb_rd_unlock(vq);
 772
 773         return log_gpa;
 774 }
 775
 776 static struct virtio_net *
 777 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 778 {
 779         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 780         struct vhost_vring_addr *addr = &vq->ring_addrs;
 781         uint64_t len, expected_len;
 782
 783         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 784                 vq->log_guest_addr =
 785                         log_addr_to_gpa(dev, vq);
 786                 if (vq->log_guest_addr == 0) {
 787                         VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map log_guest_addr.\n",
 788                                 dev->ifname);
 789                         return dev;
 790                 }
 791         }
 792
 793         if (vq_is_packed(dev)) {
 794                 len = sizeof(struct vring_packed_desc) * vq->size;
 795                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 796                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 797                 if (vq->desc_packed == NULL ||
 798                                 len != sizeof(struct vring_packed_desc) *
 799                                 vq->size) {
 800                         VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map desc_packed ring.\n",
 801                                 dev->ifname);
 802                         return dev;
 803                 }
 804
 805                 dev = numa_realloc(dev, vq_index);
 806                 vq = dev->virtqueue[vq_index];
 807                 addr = &vq->ring_addrs;
 808
 809                 len = sizeof(struct vring_packed_desc_event);
 810                 vq->driver_event = (struct vring_packed_desc_event *)
 811                                         (uintptr_t)ring_addr_to_vva(dev,
 812                                         vq, addr->avail_user_addr, &len);
 813                 if (vq->driver_event == NULL ||
 814                                 len != sizeof(struct vring_packed_desc_event)) {
 815                         VHOST_LOG_CONFIG(DEBUG, "(%s) failed to find driver area address.\n",
 816                                 dev->ifname);
 817                         return dev;
 818                 }
 819
 820                 len = sizeof(struct vring_packed_desc_event);
 821                 vq->device_event = (struct vring_packed_desc_event *)
 822                                         (uintptr_t)ring_addr_to_vva(dev,
 823                                         vq, addr->used_user_addr, &len);
 824                 if (vq->device_event == NULL ||
 825                                 len != sizeof(struct vring_packed_desc_event)) {
 826                         VHOST_LOG_CONFIG(DEBUG, "(%s) failed to find device area address.\n",
 827                                 dev->ifname);
 828                         return dev;
 829                 }
 830
 831                 vq->access_ok = true;
 832                 return dev;
 833         }
 834
 835         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 836         if (vq->desc && vq->avail && vq->used)
 837                 return dev;
 838
 839         len = sizeof(struct vring_desc) * vq->size;
 840         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 841                         vq, addr->desc_user_addr, &len);
 842         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 843                 VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map desc ring.\n", dev->ifname);
 844                 return dev;
 845         }
 846
 847         dev = numa_realloc(dev, vq_index);
 848         vq = dev->virtqueue[vq_index];
 849         addr = &vq->ring_addrs;
 850
 851         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 852         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 853                 len += sizeof(uint16_t);
 854         expected_len = len;
 855         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 856                         vq, addr->avail_user_addr, &len);
 857         if (vq->avail == 0 || len != expected_len) {
 858                 VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map avail ring.\n", dev->ifname);
 859                 return dev;
 860         }
 861
 862         len = sizeof(struct vring_used) +
 863                 sizeof(struct vring_used_elem) * vq->size;
 864         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 865                 len += sizeof(uint16_t);
 866         expected_len = len;
 867         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 868                         vq, addr->used_user_addr, &len);
 869         if (vq->used == 0 || len != expected_len) {
 870                 VHOST_LOG_CONFIG(DEBUG, "(%s) failed to map used ring.\n", dev->ifname);
 871                 return dev;
 872         }
 873
 874         if (vq->last_used_idx != vq->used->idx) {
 875                 VHOST_LOG_CONFIG(WARNING, "(%s) last_used_idx (%u) and vq->used->idx (%u) mismatches;\n",
 876                         dev->ifname,
 877                         vq->last_used_idx, vq->used->idx);
 878                 vq->last_used_idx  = vq->used->idx;
 879                 vq->last_avail_idx = vq->used->idx;
 880                 VHOST_LOG_CONFIG(WARNING, "(%s) some packets maybe resent for Tx and dropped for Rx\n",
 881                         dev->ifname);
 882         }
 883
 884         vq->access_ok = true;
 885
 886         VHOST_LOG_CONFIG(DEBUG, "(%s) mapped address desc: %p\n", dev->ifname, vq->desc);
 887         VHOST_LOG_CONFIG(DEBUG, "(%s) mapped address avail: %p\n", dev->ifname, vq->avail);
 888         VHOST_LOG_CONFIG(DEBUG, "(%s) mapped address used: %p\n", dev->ifname, vq->used);
 889         VHOST_LOG_CONFIG(DEBUG, "(%s) log_guest_addr: %" PRIx64 "\n",
 890                         dev->ifname, vq->log_guest_addr);
 891
 892         return dev;
 893 }
 894
 895 /*
 896  * The virtio device sends us the desc, used and avail ring addresses.
 897  * This function then converts these to our address space.
 898  */
 899 static int
 900 vhost_user_set_vring_addr(struct virtio_net **pdev,
 901                         struct vhu_msg_context *ctx,
 902                         int main_fd __rte_unused)
 903 {
 904         struct virtio_net *dev = *pdev;
 905         struct vhost_virtqueue *vq;
 906         struct vhost_vring_addr *addr = &ctx->msg.payload.addr;
 907         bool access_ok;
 908
 909         if (validate_msg_fds(dev, ctx, 0) != 0)
 910                 return RTE_VHOST_MSG_RESULT_ERR;
 911
 912         if (dev->mem == NULL)
 913                 return RTE_VHOST_MSG_RESULT_ERR;
 914
 915         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 916         vq = dev->virtqueue[ctx->msg.payload.addr.index];
 917
 918         access_ok = vq->access_ok;
 919
 920         /*
 921          * Rings addresses should not be interpreted as long as the ring is not
 922          * started and enabled
 923          */
 924         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 925
 926         vring_invalidate(dev, vq);
 927
 928         if ((vq->enabled && (dev->features &
 929                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 930                         access_ok) {
 931                 dev = translate_ring_addresses(dev, ctx->msg.payload.addr.index);
 932                 if (!dev)
 933                         return RTE_VHOST_MSG_RESULT_ERR;
 934
 935                 *pdev = dev;
 936         }
 937
 938         return RTE_VHOST_MSG_RESULT_OK;
 939 }
 940
 941 /*
 942  * The virtio device sends us the available ring last used index.
 943  */
 944 static int
 945 vhost_user_set_vring_base(struct virtio_net **pdev,
 946                         struct vhu_msg_context *ctx,
 947                         int main_fd __rte_unused)
 948 {
 949         struct virtio_net *dev = *pdev;
 950         struct vhost_virtqueue *vq = dev->virtqueue[ctx->msg.payload.state.index];
 951         uint64_t val = ctx->msg.payload.state.num;
 952
 953         if (validate_msg_fds(dev, ctx, 0) != 0)
 954                 return RTE_VHOST_MSG_RESULT_ERR;
 955
 956         if (vq_is_packed(dev)) {
 957                 /*
 958                  * Bit[0:14]: avail index
 959                  * Bit[15]: avail wrap counter
 960                  */
 961                 vq->last_avail_idx = val & 0x7fff;
 962                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 963                 /*
 964                  * Set used index to same value as available one, as
 965                  * their values should be the same since ring processing
 966                  * was stopped at get time.
 967                  */
 968                 vq->last_used_idx = vq->last_avail_idx;
 969                 vq->used_wrap_counter = vq->avail_wrap_counter;
 970         } else {
 971                 vq->last_used_idx = ctx->msg.payload.state.num;
 972                 vq->last_avail_idx = ctx->msg.payload.state.num;
 973         }
 974
 975         VHOST_LOG_CONFIG(INFO,
 976                 "(%s) vring base idx:%u last_used_idx:%u last_avail_idx:%u.\n",
 977                 dev->ifname, ctx->msg.payload.state.index, vq->last_used_idx,
 978                 vq->last_avail_idx);
 979
 980         return RTE_VHOST_MSG_RESULT_OK;
 981 }
 982
 983 static int
 984 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 985                    uint64_t host_phys_addr, uint64_t size)
 986 {
 987         struct guest_page *page, *last_page;
 988         struct guest_page *old_pages;
 989
 990         if (dev->nr_guest_pages == dev->max_guest_pages) {
 991                 dev->max_guest_pages *= 2;
 992                 old_pages = dev->guest_pages;
 993                 dev->guest_pages = rte_realloc(dev->guest_pages,
 994                                         dev->max_guest_pages * sizeof(*page),
 995                                         RTE_CACHE_LINE_SIZE);
 996                 if (dev->guest_pages == NULL) {
 997                         VHOST_LOG_CONFIG(ERR, "(%s) cannot realloc guest_pages\n", dev->ifname);
 998                         rte_free(old_pages);
 999                         return -1;
1000                 }
1001         }
1002
1003         if (dev->nr_guest_pages > 0) {
1004                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
1005                 /* merge if the two pages are continuous */
1006                 if (host_phys_addr == last_page->host_phys_addr +
1007                                       last_page->size) {
1008                         last_page->size += size;
1009                         return 0;
1010                 }
1011         }
1012
1013         page = &dev->guest_pages[dev->nr_guest_pages++];
1014         page->guest_phys_addr = guest_phys_addr;
1015         page->host_phys_addr  = host_phys_addr;
1016         page->size = size;
1017
1018         return 0;
1019 }
1020
1021 static int
1022 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
1023                 uint64_t page_size)
1024 {
1025         uint64_t reg_size = reg->size;
1026         uint64_t host_user_addr  = reg->host_user_addr;
1027         uint64_t guest_phys_addr = reg->guest_phys_addr;
1028         uint64_t host_phys_addr;
1029         uint64_t size;
1030
1031         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
1032         size = page_size - (guest_phys_addr & (page_size - 1));
1033         size = RTE_MIN(size, reg_size);
1034
1035         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
1036                 return -1;
1037
1038         host_user_addr  += size;
1039         guest_phys_addr += size;
1040         reg_size -= size;
1041
1042         while (reg_size > 0) {
1043                 size = RTE_MIN(reg_size, page_size);
1044                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
1045                                                   host_user_addr);
1046                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
1047                                 size) < 0)
1048                         return -1;
1049
1050                 host_user_addr  += size;
1051                 guest_phys_addr += size;
1052                 reg_size -= size;
1053         }
1054
1055         /* sort guest page array if over binary search threshold */
1056         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
1057                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
1058                         sizeof(struct guest_page), guest_page_addrcmp);
1059         }
1060
1061         return 0;
1062 }
1063
1064 #ifdef RTE_LIBRTE_VHOST_DEBUG
1065 /* TODO: enable it only in debug mode? */
1066 static void
1067 dump_guest_pages(struct virtio_net *dev)
1068 {
1069         uint32_t i;
1070         struct guest_page *page;
1071
1072         for (i = 0; i < dev->nr_guest_pages; i++) {
1073                 page = &dev->guest_pages[i];
1074
1075                 VHOST_LOG_CONFIG(INFO, "(%s) guest physical page region %u\n",
1076                                 dev->ifname, i);
1077                 VHOST_LOG_CONFIG(INFO, "(%s)\tguest_phys_addr: %" PRIx64 "\n",
1078                                 dev->ifname, page->guest_phys_addr);
1079                 VHOST_LOG_CONFIG(INFO, "(%s)\thost_phys_addr : %" PRIx64 "\n",
1080                                 dev->ifname, page->host_phys_addr);
1081                 VHOST_LOG_CONFIG(INFO, "(%s)\tsize           : %" PRIx64 "\n",
1082                                 dev->ifname, page->size);
1083         }
1084 }
1085 #else
1086 #define dump_guest_pages(dev)
1087 #endif
1088
1089 static bool
1090 vhost_memory_changed(struct VhostUserMemory *new,
1091                      struct rte_vhost_memory *old)
1092 {
1093         uint32_t i;
1094
1095         if (new->nregions != old->nregions)
1096                 return true;
1097
1098         for (i = 0; i < new->nregions; ++i) {
1099                 VhostUserMemoryRegion *new_r = &new->regions[i];
1100                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1101
1102                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1103                         return true;
1104                 if (new_r->memory_size != old_r->size)
1105                         return true;
1106                 if (new_r->userspace_addr != old_r->guest_user_addr)
1107                         return true;
1108         }
1109
1110         return false;
1111 }
1112
1113 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1114 static int
1115 vhost_user_postcopy_region_register(struct virtio_net *dev,
1116                 struct rte_vhost_mem_region *reg)
1117 {
1118         struct uffdio_register reg_struct;
1119
1120         /*
1121          * Let's register all the mmapped area to ensure
1122          * alignment on page boundary.
1123          */
1124         reg_struct.range.start = (uint64_t)(uintptr_t)reg->mmap_addr;
1125         reg_struct.range.len = reg->mmap_size;
1126         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1127
1128         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1129                                 &reg_struct)) {
1130                 VHOST_LOG_CONFIG(ERR, "(%s) failed to register ufd for region "
1131                                 "%" PRIx64 " - %" PRIx64 " (ufd = %d) %s\n",
1132                                 dev->ifname,
1133                                 (uint64_t)reg_struct.range.start,
1134                                 (uint64_t)reg_struct.range.start +
1135                                 (uint64_t)reg_struct.range.len - 1,
1136                                 dev->postcopy_ufd,
1137                                 strerror(errno));
1138                 return -1;
1139         }
1140
1141         VHOST_LOG_CONFIG(INFO,
1142                         "(%s)\t userfaultfd registered for range : %" PRIx64 " - %" PRIx64 "\n",
1143                         dev->ifname,
1144                         (uint64_t)reg_struct.range.start,
1145                         (uint64_t)reg_struct.range.start +
1146                         (uint64_t)reg_struct.range.len - 1);
1147
1148         return 0;
1149 }
1150 #else
1151 static int
1152 vhost_user_postcopy_region_register(struct virtio_net *dev __rte_unused,
1153                 struct rte_vhost_mem_region *reg __rte_unused)
1154 {
1155         return -1;
1156 }
1157 #endif
1158
1159 static int
1160 vhost_user_postcopy_register(struct virtio_net *dev, int main_fd,
1161                 struct vhu_msg_context *ctx)
1162 {
1163         struct VhostUserMemory *memory;
1164         struct rte_vhost_mem_region *reg;
1165         struct vhu_msg_context ack_ctx;
1166         uint32_t i;
1167
1168         if (!dev->postcopy_listening)
1169                 return 0;
1170
1171         /*
1172          * We haven't a better way right now than sharing
1173          * DPDK's virtual address with Qemu, so that Qemu can
1174          * retrieve the region offset when handling userfaults.
1175          */
1176         memory = &ctx->msg.payload.memory;
1177         for (i = 0; i < memory->nregions; i++) {
1178                 reg = &dev->mem->regions[i];
1179                 memory->regions[i].userspace_addr = reg->host_user_addr;
1180         }
1181
1182         /* Send the addresses back to qemu */
1183         ctx->fd_num = 0;
1184         send_vhost_reply(dev, main_fd, ctx);
1185
1186         /* Wait for qemu to acknowledge it got the addresses
1187          * we've got to wait before we're allowed to generate faults.
1188          */
1189         if (read_vhost_message(dev, main_fd, &ack_ctx) <= 0) {
1190                 VHOST_LOG_CONFIG(ERR, "(%s) failed to read qemu ack on postcopy set-mem-table\n",
1191                                 dev->ifname);
1192                 return -1;
1193         }
1194
1195         if (validate_msg_fds(dev, &ack_ctx, 0) != 0)
1196                 return -1;
1197
1198         if (ack_ctx.msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1199                 VHOST_LOG_CONFIG(ERR, "(%s) bad qemu ack on postcopy set-mem-table (%d)\n",
1200                                 dev->ifname, ack_ctx.msg.request.master);
1201                 return -1;
1202         }
1203
1204         /* Now userfault register and we can use the memory */
1205         for (i = 0; i < memory->nregions; i++) {
1206                 reg = &dev->mem->regions[i];
1207                 if (vhost_user_postcopy_region_register(dev, reg) < 0)
1208                         return -1;
1209         }
1210
1211         return 0;
1212 }
1213
1214 static int
1215 vhost_user_mmap_region(struct virtio_net *dev,
1216                 struct rte_vhost_mem_region *region,
1217                 uint64_t mmap_offset)
1218 {
1219         void *mmap_addr;
1220         uint64_t mmap_size;
1221         uint64_t alignment;
1222         int populate;
1223         int ret;
1224
1225         /* Check for memory_size + mmap_offset overflow */
1226         if (mmap_offset >= -region->size) {
1227                 VHOST_LOG_CONFIG(ERR, "(%s) mmap_offset (%#"PRIx64") and memory_size (%#"PRIx64") overflow\n",
1228                                 dev->ifname, mmap_offset, region->size);
1229                 return -1;
1230         }
1231
1232         mmap_size = region->size + mmap_offset;
1233
1234         /* mmap() without flag of MAP_ANONYMOUS, should be called with length
1235          * argument aligned with hugepagesz at older longterm version Linux,
1236          * like 2.6.32 and 3.2.72, or mmap() will fail with EINVAL.
1237          *
1238          * To avoid failure, make sure in caller to keep length aligned.
1239          */
1240         alignment = get_blk_size(region->fd);
1241         if (alignment == (uint64_t)-1) {
1242                 VHOST_LOG_CONFIG(ERR, "(%s) couldn't get hugepage size through fstat\n",
1243                                 dev->ifname);
1244                 return -1;
1245         }
1246         mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1247         if (mmap_size == 0) {
1248                 /*
1249                  * It could happen if initial mmap_size + alignment overflows
1250                  * the sizeof uint64, which could happen if either mmap_size or
1251                  * alignment value is wrong.
1252                  *
1253                  * mmap() kernel implementation would return an error, but
1254                  * better catch it before and provide useful info in the logs.
1255                  */
1256                 VHOST_LOG_CONFIG(ERR, "(%s) mmap size (0x%" PRIx64 ") or alignment (0x%" PRIx64 ") is invalid\n",
1257                                 dev->ifname, region->size + mmap_offset, alignment);
1258                 return -1;
1259         }
1260
1261         populate = dev->async_copy ? MAP_POPULATE : 0;
1262         mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1263                         MAP_SHARED | populate, region->fd, 0);
1264
1265         if (mmap_addr == MAP_FAILED) {
1266                 VHOST_LOG_CONFIG(ERR, "(%s) mmap failed (%s).\n", dev->ifname, strerror(errno));
1267                 return -1;
1268         }
1269
1270         region->mmap_addr = mmap_addr;
1271         region->mmap_size = mmap_size;
1272         region->host_user_addr = (uint64_t)(uintptr_t)mmap_addr + mmap_offset;
1273
1274         if (dev->async_copy) {
1275                 if (add_guest_pages(dev, region, alignment) < 0) {
1276                         VHOST_LOG_CONFIG(ERR, "(%s) adding guest pages to region failed.\n",
1277                                         dev->ifname);
1278                         return -1;
1279                 }
1280
1281                 if (rte_vfio_is_enabled("vfio")) {
1282                         ret = async_dma_map(dev, region, true);
1283                         if (ret) {
1284                                 VHOST_LOG_CONFIG(ERR,
1285                                         "(%s) configure IOMMU for DMA engine failed\n",
1286                                         dev->ifname);
1287                                 return -1;
1288                         }
1289                 }
1290         }
1291
1292         VHOST_LOG_CONFIG(INFO, "(%s) guest memory region size: 0x%" PRIx64 "\n",
1293                         dev->ifname, region->size);
1294         VHOST_LOG_CONFIG(INFO, "(%s)\t guest physical addr: 0x%" PRIx64 "\n",
1295                         dev->ifname, region->guest_phys_addr);
1296         VHOST_LOG_CONFIG(INFO, "(%s)\t guest virtual  addr: 0x%" PRIx64 "\n",
1297                         dev->ifname, region->guest_user_addr);
1298         VHOST_LOG_CONFIG(INFO, "(%s)\t host  virtual  addr: 0x%" PRIx64 "\n",
1299                         dev->ifname, region->host_user_addr);
1300         VHOST_LOG_CONFIG(INFO, "(%s)\t mmap addr : 0x%" PRIx64 "\n",
1301                         dev->ifname, (uint64_t)(uintptr_t)mmap_addr);
1302         VHOST_LOG_CONFIG(INFO, "(%s)\t mmap size : 0x%" PRIx64 "\n",
1303                         dev->ifname, mmap_size);
1304         VHOST_LOG_CONFIG(INFO, "(%s)\t mmap align: 0x%" PRIx64 "\n",
1305                         dev->ifname, alignment);
1306         VHOST_LOG_CONFIG(INFO, "(%s)\t mmap off  : 0x%" PRIx64 "\n",
1307                         dev->ifname, mmap_offset);
1308
1309         return 0;
1310 }
1311
1312 static int
1313 vhost_user_set_mem_table(struct virtio_net **pdev,
1314                         struct vhu_msg_context *ctx,
1315                         int main_fd)
1316 {
1317         struct virtio_net *dev = *pdev;
1318         struct VhostUserMemory *memory = &ctx->msg.payload.memory;
1319         struct rte_vhost_mem_region *reg;
1320         int numa_node = SOCKET_ID_ANY;
1321         uint64_t mmap_offset;
1322         uint32_t i;
1323         bool async_notify = false;
1324
1325         if (validate_msg_fds(dev, ctx, memory->nregions) != 0)
1326                 return RTE_VHOST_MSG_RESULT_ERR;
1327
1328         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1329                 VHOST_LOG_CONFIG(ERR, "(%s) too many memory regions (%u)\n",
1330                                 dev->ifname, memory->nregions);
1331                 goto close_msg_fds;
1332         }
1333
1334         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1335                 VHOST_LOG_CONFIG(INFO, "(%s) memory regions not changed\n", dev->ifname);
1336
1337                 close_msg_fds(ctx);
1338
1339                 return RTE_VHOST_MSG_RESULT_OK;
1340         }
1341
1342         if (dev->mem) {
1343                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1344                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1345
1346                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1347                                 vdpa_dev->ops->dev_close(dev->vid);
1348                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1349                 }
1350
1351                 /* notify the vhost application to stop DMA transfers */
1352                 if (dev->async_copy && dev->notify_ops->vring_state_changed) {
1353                         for (i = 0; i < dev->nr_vring; i++) {
1354                                 dev->notify_ops->vring_state_changed(dev->vid,
1355                                                 i, 0);
1356                         }
1357                         async_notify = true;
1358                 }
1359
1360                 free_mem_region(dev);
1361                 rte_free(dev->mem);
1362                 dev->mem = NULL;
1363         }
1364
1365         /* Flush IOTLB cache as previous HVAs are now invalid */
1366         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1367                 for (i = 0; i < dev->nr_vring; i++)
1368                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1369
1370         /*
1371          * If VQ 0 has already been allocated, try to allocate on the same
1372          * NUMA node. It can be reallocated later in numa_realloc().
1373          */
1374         if (dev->nr_vring > 0)
1375                 numa_node = dev->virtqueue[0]->numa_node;
1376
1377         dev->nr_guest_pages = 0;
1378         if (dev->guest_pages == NULL) {
1379                 dev->max_guest_pages = 8;
1380                 dev->guest_pages = rte_zmalloc_socket(NULL,
1381                                         dev->max_guest_pages *
1382                                         sizeof(struct guest_page),
1383                                         RTE_CACHE_LINE_SIZE,
1384                                         numa_node);
1385                 if (dev->guest_pages == NULL) {
1386                         VHOST_LOG_CONFIG(ERR,
1387                                 "(%s) failed to allocate memory for dev->guest_pages\n",
1388                                 dev->ifname);
1389                         goto close_msg_fds;
1390                 }
1391         }
1392
1393         dev->mem = rte_zmalloc_socket("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1394                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0, numa_node);
1395         if (dev->mem == NULL) {
1396                 VHOST_LOG_CONFIG(ERR,
1397                         "(%s) failed to allocate memory for dev->mem\n",
1398                         dev->ifname);
1399                 goto free_guest_pages;
1400         }
1401
1402         for (i = 0; i < memory->nregions; i++) {
1403                 reg = &dev->mem->regions[i];
1404
1405                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1406                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1407                 reg->size            = memory->regions[i].memory_size;
1408                 reg->fd              = ctx->fds[i];
1409
1410                 /*
1411                  * Assign invalid file descriptor value to avoid double
1412                  * closing on error path.
1413                  */
1414                 ctx->fds[i] = -1;
1415
1416                 mmap_offset = memory->regions[i].mmap_offset;
1417
1418                 if (vhost_user_mmap_region(dev, reg, mmap_offset) < 0) {
1419                         VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap region %u\n", dev->ifname, i);
1420                         goto free_mem_table;
1421                 }
1422
1423                 dev->mem->nregions++;
1424         }
1425
1426         if (vhost_user_postcopy_register(dev, main_fd, ctx) < 0)
1427                 goto free_mem_table;
1428
1429         for (i = 0; i < dev->nr_vring; i++) {
1430                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1431
1432                 if (!vq)
1433                         continue;
1434
1435                 if (vq->desc || vq->avail || vq->used) {
1436                         /*
1437                          * If the memory table got updated, the ring addresses
1438                          * need to be translated again as virtual addresses have
1439                          * changed.
1440                          */
1441                         vring_invalidate(dev, vq);
1442
1443                         dev = translate_ring_addresses(dev, i);
1444                         if (!dev) {
1445                                 dev = *pdev;
1446                                 goto free_mem_table;
1447                         }
1448
1449                         *pdev = dev;
1450                 }
1451         }
1452
1453         dump_guest_pages(dev);
1454
1455         if (async_notify) {
1456                 for (i = 0; i < dev->nr_vring; i++)
1457                         dev->notify_ops->vring_state_changed(dev->vid, i, 1);
1458         }
1459
1460         return RTE_VHOST_MSG_RESULT_OK;
1461
1462 free_mem_table:
1463         free_mem_region(dev);
1464         rte_free(dev->mem);
1465         dev->mem = NULL;
1466
1467 free_guest_pages:
1468         rte_free(dev->guest_pages);
1469         dev->guest_pages = NULL;
1470 close_msg_fds:
1471         close_msg_fds(ctx);
1472         return RTE_VHOST_MSG_RESULT_ERR;
1473 }
1474
1475 static bool
1476 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1477 {
1478         bool rings_ok;
1479
1480         if (!vq)
1481                 return false;
1482
1483         if (vq_is_packed(dev))
1484                 rings_ok = vq->desc_packed && vq->driver_event &&
1485                         vq->device_event;
1486         else
1487                 rings_ok = vq->desc && vq->avail && vq->used;
1488
1489         return rings_ok &&
1490                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1491                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1492                vq->enabled;
1493 }
1494
1495 #define VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY 2u
1496
1497 static int
1498 virtio_is_ready(struct virtio_net *dev)
1499 {
1500         struct vhost_virtqueue *vq;
1501         uint32_t i, nr_vring = dev->nr_vring;
1502
1503         if (dev->flags & VIRTIO_DEV_READY)
1504                 return 1;
1505
1506         if (!dev->nr_vring)
1507                 return 0;
1508
1509         if (dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) {
1510                 nr_vring = VIRTIO_BUILTIN_NUM_VQS_TO_BE_READY;
1511
1512                 if (dev->nr_vring < nr_vring)
1513                         return 0;
1514         }
1515
1516         for (i = 0; i < nr_vring; i++) {
1517                 vq = dev->virtqueue[i];
1518
1519                 if (!vq_is_ready(dev, vq))
1520                         return 0;
1521         }
1522
1523         /* If supported, ensure the frontend is really done with config */
1524         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1525                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1526                         return 0;
1527
1528         dev->flags |= VIRTIO_DEV_READY;
1529
1530         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1531                 VHOST_LOG_CONFIG(INFO, "(%s) virtio is now ready for processing.\n", dev->ifname);
1532         return 1;
1533 }
1534
1535 static void *
1536 inflight_mem_alloc(struct virtio_net *dev, const char *name, size_t size, int *fd)
1537 {
1538         void *ptr;
1539         int mfd = -1;
1540         char fname[20] = "/tmp/memfd-XXXXXX";
1541
1542         *fd = -1;
1543 #ifdef MEMFD_SUPPORTED
1544         mfd = memfd_create(name, MFD_CLOEXEC);
1545 #else
1546         RTE_SET_USED(name);
1547 #endif
1548         if (mfd == -1) {
1549                 mfd = mkstemp(fname);
1550                 if (mfd == -1) {
1551                         VHOST_LOG_CONFIG(ERR, "(%s) failed to get inflight buffer fd\n",
1552                                         dev->ifname);
1553                         return NULL;
1554                 }
1555
1556                 unlink(fname);
1557         }
1558
1559         if (ftruncate(mfd, size) == -1) {
1560                 VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc inflight buffer\n", dev->ifname);
1561                 close(mfd);
1562                 return NULL;
1563         }
1564
1565         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1566         if (ptr == MAP_FAILED) {
1567                 VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap inflight buffer\n", dev->ifname);
1568                 close(mfd);
1569                 return NULL;
1570         }
1571
1572         *fd = mfd;
1573         return ptr;
1574 }
1575
1576 static uint32_t
1577 get_pervq_shm_size_split(uint16_t queue_size)
1578 {
1579         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1580                                   queue_size + sizeof(uint64_t) +
1581                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1582 }
1583
1584 static uint32_t
1585 get_pervq_shm_size_packed(uint16_t queue_size)
1586 {
1587         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1588                                   * queue_size + sizeof(uint64_t) +
1589                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1590                                   INFLIGHT_ALIGNMENT);
1591 }
1592
1593 static int
1594 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1595                            struct vhu_msg_context *ctx,
1596                            int main_fd __rte_unused)
1597 {
1598         struct rte_vhost_inflight_info_packed *inflight_packed;
1599         uint64_t pervq_inflight_size, mmap_size;
1600         uint16_t num_queues, queue_size;
1601         struct virtio_net *dev = *pdev;
1602         int fd, i, j;
1603         int numa_node = SOCKET_ID_ANY;
1604         void *addr;
1605
1606         if (ctx->msg.size != sizeof(ctx->msg.payload.inflight)) {
1607                 VHOST_LOG_CONFIG(ERR, "(%s) invalid get_inflight_fd message size is %d\n",
1608                         dev->ifname, ctx->msg.size);
1609                 return RTE_VHOST_MSG_RESULT_ERR;
1610         }
1611
1612         /*
1613          * If VQ 0 has already been allocated, try to allocate on the same
1614          * NUMA node. It can be reallocated later in numa_realloc().
1615          */
1616         if (dev->nr_vring > 0)
1617                 numa_node = dev->virtqueue[0]->numa_node;
1618
1619         if (dev->inflight_info == NULL) {
1620                 dev->inflight_info = rte_zmalloc_socket("inflight_info",
1621                                 sizeof(struct inflight_mem_info), 0, numa_node);
1622                 if (!dev->inflight_info) {
1623                         VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc dev inflight area\n",
1624                                         dev->ifname);
1625                         return RTE_VHOST_MSG_RESULT_ERR;
1626                 }
1627                 dev->inflight_info->fd = -1;
1628         }
1629
1630         num_queues = ctx->msg.payload.inflight.num_queues;
1631         queue_size = ctx->msg.payload.inflight.queue_size;
1632
1633         VHOST_LOG_CONFIG(INFO, "(%s) get_inflight_fd num_queues: %u\n",
1634                 dev->ifname, ctx->msg.payload.inflight.num_queues);
1635         VHOST_LOG_CONFIG(INFO, "(%s) get_inflight_fd queue_size: %u\n",
1636                 dev->ifname, ctx->msg.payload.inflight.queue_size);
1637
1638         if (vq_is_packed(dev))
1639                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1640         else
1641                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1642
1643         mmap_size = num_queues * pervq_inflight_size;
1644         addr = inflight_mem_alloc(dev, "vhost-inflight", mmap_size, &fd);
1645         if (!addr) {
1646                 VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc vhost inflight area\n", dev->ifname);
1647                         ctx->msg.payload.inflight.mmap_size = 0;
1648                 return RTE_VHOST_MSG_RESULT_ERR;
1649         }
1650         memset(addr, 0, mmap_size);
1651
1652         if (dev->inflight_info->addr) {
1653                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1654                 dev->inflight_info->addr = NULL;
1655         }
1656
1657         if (dev->inflight_info->fd >= 0) {
1658                 close(dev->inflight_info->fd);
1659                 dev->inflight_info->fd = -1;
1660         }
1661
1662         dev->inflight_info->addr = addr;
1663         dev->inflight_info->size = ctx->msg.payload.inflight.mmap_size = mmap_size;
1664         dev->inflight_info->fd = ctx->fds[0] = fd;
1665         ctx->msg.payload.inflight.mmap_offset = 0;
1666         ctx->fd_num = 1;
1667
1668         if (vq_is_packed(dev)) {
1669                 for (i = 0; i < num_queues; i++) {
1670                         inflight_packed =
1671                                 (struct rte_vhost_inflight_info_packed *)addr;
1672                         inflight_packed->used_wrap_counter = 1;
1673                         inflight_packed->old_used_wrap_counter = 1;
1674                         for (j = 0; j < queue_size; j++)
1675                                 inflight_packed->desc[j].next = j + 1;
1676                         addr = (void *)((char *)addr + pervq_inflight_size);
1677                 }
1678         }
1679
1680         VHOST_LOG_CONFIG(INFO, "(%s) send inflight mmap_size: %"PRIu64"\n",
1681                         dev->ifname, ctx->msg.payload.inflight.mmap_size);
1682         VHOST_LOG_CONFIG(INFO, "(%s) send inflight mmap_offset: %"PRIu64"\n",
1683                         dev->ifname, ctx->msg.payload.inflight.mmap_offset);
1684         VHOST_LOG_CONFIG(INFO, "(%s) send inflight fd: %d\n", dev->ifname, ctx->fds[0]);
1685
1686         return RTE_VHOST_MSG_RESULT_REPLY;
1687 }
1688
1689 static int
1690 vhost_user_set_inflight_fd(struct virtio_net **pdev,
1691                            struct vhu_msg_context *ctx,
1692                            int main_fd __rte_unused)
1693 {
1694         uint64_t mmap_size, mmap_offset;
1695         uint16_t num_queues, queue_size;
1696         struct virtio_net *dev = *pdev;
1697         uint32_t pervq_inflight_size;
1698         struct vhost_virtqueue *vq;
1699         void *addr;
1700         int fd, i;
1701         int numa_node = SOCKET_ID_ANY;
1702
1703         fd = ctx->fds[0];
1704         if (ctx->msg.size != sizeof(ctx->msg.payload.inflight) || fd < 0) {
1705                 VHOST_LOG_CONFIG(ERR, "(%s) invalid set_inflight_fd message size is %d,fd is %d\n",
1706                         dev->ifname, ctx->msg.size, fd);
1707                 return RTE_VHOST_MSG_RESULT_ERR;
1708         }
1709
1710         mmap_size = ctx->msg.payload.inflight.mmap_size;
1711         mmap_offset = ctx->msg.payload.inflight.mmap_offset;
1712         num_queues = ctx->msg.payload.inflight.num_queues;
1713         queue_size = ctx->msg.payload.inflight.queue_size;
1714
1715         if (vq_is_packed(dev))
1716                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1717         else
1718                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1719
1720         VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd mmap_size: %"PRIu64"\n",
1721                         dev->ifname, mmap_size);
1722         VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd mmap_offset: %"PRIu64"\n",
1723                         dev->ifname, mmap_offset);
1724         VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd num_queues: %u\n", dev->ifname, num_queues);
1725         VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd queue_size: %u\n", dev->ifname, queue_size);
1726         VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd fd: %d\n", dev->ifname, fd);
1727         VHOST_LOG_CONFIG(INFO, "(%s) set_inflight_fd pervq_inflight_size: %d\n",
1728                         dev->ifname, pervq_inflight_size);
1729
1730         /*
1731          * If VQ 0 has already been allocated, try to allocate on the same
1732          * NUMA node. It can be reallocated later in numa_realloc().
1733          */
1734         if (dev->nr_vring > 0)
1735                 numa_node = dev->virtqueue[0]->numa_node;
1736
1737         if (!dev->inflight_info) {
1738                 dev->inflight_info = rte_zmalloc_socket("inflight_info",
1739                                 sizeof(struct inflight_mem_info), 0, numa_node);
1740                 if (dev->inflight_info == NULL) {
1741                         VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc dev inflight area\n",
1742                                         dev->ifname);
1743                         return RTE_VHOST_MSG_RESULT_ERR;
1744                 }
1745                 dev->inflight_info->fd = -1;
1746         }
1747
1748         if (dev->inflight_info->addr) {
1749                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1750                 dev->inflight_info->addr = NULL;
1751         }
1752
1753         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1754                     fd, mmap_offset);
1755         if (addr == MAP_FAILED) {
1756                 VHOST_LOG_CONFIG(ERR, "(%s) failed to mmap share memory.\n", dev->ifname);
1757                 return RTE_VHOST_MSG_RESULT_ERR;
1758         }
1759
1760         if (dev->inflight_info->fd >= 0) {
1761                 close(dev->inflight_info->fd);
1762                 dev->inflight_info->fd = -1;
1763         }
1764
1765         dev->inflight_info->fd = fd;
1766         dev->inflight_info->addr = addr;
1767         dev->inflight_info->size = mmap_size;
1768
1769         for (i = 0; i < num_queues; i++) {
1770                 vq = dev->virtqueue[i];
1771                 if (!vq)
1772                         continue;
1773
1774                 if (vq_is_packed(dev)) {
1775                         vq->inflight_packed = addr;
1776                         vq->inflight_packed->desc_num = queue_size;
1777                 } else {
1778                         vq->inflight_split = addr;
1779                         vq->inflight_split->desc_num = queue_size;
1780                 }
1781                 addr = (void *)((char *)addr + pervq_inflight_size);
1782         }
1783
1784         return RTE_VHOST_MSG_RESULT_OK;
1785 }
1786
1787 static int
1788 vhost_user_set_vring_call(struct virtio_net **pdev,
1789                         struct vhu_msg_context *ctx,
1790                         int main_fd __rte_unused)
1791 {
1792         struct virtio_net *dev = *pdev;
1793         struct vhost_vring_file file;
1794         struct vhost_virtqueue *vq;
1795         int expected_fds;
1796
1797         expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1798         if (validate_msg_fds(dev, ctx, expected_fds) != 0)
1799                 return RTE_VHOST_MSG_RESULT_ERR;
1800
1801         file.index = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
1802         if (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1803                 file.fd = VIRTIO_INVALID_EVENTFD;
1804         else
1805                 file.fd = ctx->fds[0];
1806         VHOST_LOG_CONFIG(INFO, "(%s) vring call idx:%d file:%d\n",
1807                         dev->ifname, file.index, file.fd);
1808
1809         vq = dev->virtqueue[file.index];
1810
1811         if (vq->ready) {
1812                 vq->ready = false;
1813                 vhost_user_notify_queue_state(dev, file.index, 0);
1814         }
1815
1816         if (vq->callfd >= 0)
1817                 close(vq->callfd);
1818
1819         vq->callfd = file.fd;
1820
1821         return RTE_VHOST_MSG_RESULT_OK;
1822 }
1823
1824 static int vhost_user_set_vring_err(struct virtio_net **pdev,
1825                         struct vhu_msg_context *ctx,
1826                         int main_fd __rte_unused)
1827 {
1828         struct virtio_net *dev = *pdev;
1829         int expected_fds;
1830
1831         expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1832         if (validate_msg_fds(dev, ctx, expected_fds) != 0)
1833                 return RTE_VHOST_MSG_RESULT_ERR;
1834
1835         if (!(ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1836                 close(ctx->fds[0]);
1837         VHOST_LOG_CONFIG(INFO, "(%s) not implemented\n", dev->ifname);
1838
1839         return RTE_VHOST_MSG_RESULT_OK;
1840 }
1841
1842 static int
1843 resubmit_desc_compare(const void *a, const void *b)
1844 {
1845         const struct rte_vhost_resubmit_desc *desc0 = a;
1846         const struct rte_vhost_resubmit_desc *desc1 = b;
1847
1848         if (desc1->counter > desc0->counter)
1849                 return 1;
1850
1851         return -1;
1852 }
1853
1854 static int
1855 vhost_check_queue_inflights_split(struct virtio_net *dev,
1856                                   struct vhost_virtqueue *vq)
1857 {
1858         uint16_t i;
1859         uint16_t resubmit_num = 0, last_io, num;
1860         struct vring_used *used = vq->used;
1861         struct rte_vhost_resubmit_info *resubmit;
1862         struct rte_vhost_inflight_info_split *inflight_split;
1863
1864         if (!(dev->protocol_features &
1865             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1866                 return RTE_VHOST_MSG_RESULT_OK;
1867
1868         /* The frontend may still not support the inflight feature
1869          * although we negotiate the protocol feature.
1870          */
1871         if ((!vq->inflight_split))
1872                 return RTE_VHOST_MSG_RESULT_OK;
1873
1874         if (!vq->inflight_split->version) {
1875                 vq->inflight_split->version = INFLIGHT_VERSION;
1876                 return RTE_VHOST_MSG_RESULT_OK;
1877         }
1878
1879         if (vq->resubmit_inflight)
1880                 return RTE_VHOST_MSG_RESULT_OK;
1881
1882         inflight_split = vq->inflight_split;
1883         vq->global_counter = 0;
1884         last_io = inflight_split->last_inflight_io;
1885
1886         if (inflight_split->used_idx != used->idx) {
1887                 inflight_split->desc[last_io].inflight = 0;
1888                 rte_atomic_thread_fence(__ATOMIC_SEQ_CST);
1889                 inflight_split->used_idx = used->idx;
1890         }
1891
1892         for (i = 0; i < inflight_split->desc_num; i++) {
1893                 if (inflight_split->desc[i].inflight == 1)
1894                         resubmit_num++;
1895         }
1896
1897         vq->last_avail_idx += resubmit_num;
1898
1899         if (resubmit_num) {
1900                 resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1901                                 0, vq->numa_node);
1902                 if (!resubmit) {
1903                         VHOST_LOG_CONFIG(ERR,
1904                                         "(%s) failed to allocate memory for resubmit info.\n",
1905                                         dev->ifname);
1906                         return RTE_VHOST_MSG_RESULT_ERR;
1907                 }
1908
1909                 resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
1910                                 resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
1911                                 0, vq->numa_node);
1912                 if (!resubmit->resubmit_list) {
1913                         VHOST_LOG_CONFIG(ERR,
1914                                         "(%s) failed to allocate memory for inflight desc.\n",
1915                                         dev->ifname);
1916                         rte_free(resubmit);
1917                         return RTE_VHOST_MSG_RESULT_ERR;
1918                 }
1919
1920                 num = 0;
1921                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1922                         if (vq->inflight_split->desc[i].inflight == 1) {
1923                                 resubmit->resubmit_list[num].index = i;
1924                                 resubmit->resubmit_list[num].counter =
1925                                         inflight_split->desc[i].counter;
1926                                 num++;
1927                         }
1928                 }
1929                 resubmit->resubmit_num = num;
1930
1931                 if (resubmit->resubmit_num > 1)
1932                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1933                               sizeof(struct rte_vhost_resubmit_desc),
1934                               resubmit_desc_compare);
1935
1936                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1937                 vq->resubmit_inflight = resubmit;
1938         }
1939
1940         return RTE_VHOST_MSG_RESULT_OK;
1941 }
1942
1943 static int
1944 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1945                                    struct vhost_virtqueue *vq)
1946 {
1947         uint16_t i;
1948         uint16_t resubmit_num = 0, old_used_idx, num;
1949         struct rte_vhost_resubmit_info *resubmit;
1950         struct rte_vhost_inflight_info_packed *inflight_packed;
1951
1952         if (!(dev->protocol_features &
1953             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1954                 return RTE_VHOST_MSG_RESULT_OK;
1955
1956         /* The frontend may still not support the inflight feature
1957          * although we negotiate the protocol feature.
1958          */
1959         if ((!vq->inflight_packed))
1960                 return RTE_VHOST_MSG_RESULT_OK;
1961
1962         if (!vq->inflight_packed->version) {
1963                 vq->inflight_packed->version = INFLIGHT_VERSION;
1964                 return RTE_VHOST_MSG_RESULT_OK;
1965         }
1966
1967         if (vq->resubmit_inflight)
1968                 return RTE_VHOST_MSG_RESULT_OK;
1969
1970         inflight_packed = vq->inflight_packed;
1971         vq->global_counter = 0;
1972         old_used_idx = inflight_packed->old_used_idx;
1973
1974         if (inflight_packed->used_idx != old_used_idx) {
1975                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1976                         inflight_packed->old_used_idx =
1977                                 inflight_packed->used_idx;
1978                         inflight_packed->old_used_wrap_counter =
1979                                 inflight_packed->used_wrap_counter;
1980                         inflight_packed->old_free_head =
1981                                 inflight_packed->free_head;
1982                 } else {
1983                         inflight_packed->used_idx =
1984                                 inflight_packed->old_used_idx;
1985                         inflight_packed->used_wrap_counter =
1986                                 inflight_packed->old_used_wrap_counter;
1987                         inflight_packed->free_head =
1988                                 inflight_packed->old_free_head;
1989                 }
1990         }
1991
1992         for (i = 0; i < inflight_packed->desc_num; i++) {
1993                 if (inflight_packed->desc[i].inflight == 1)
1994                         resubmit_num++;
1995         }
1996
1997         if (resubmit_num) {
1998                 resubmit = rte_zmalloc_socket("resubmit", sizeof(struct rte_vhost_resubmit_info),
1999                                 0, vq->numa_node);
2000                 if (resubmit == NULL) {
2001                         VHOST_LOG_CONFIG(ERR,
2002                                         "(%s) failed to allocate memory for resubmit info.\n",
2003                                         dev->ifname);
2004                         return RTE_VHOST_MSG_RESULT_ERR;
2005                 }
2006
2007                 resubmit->resubmit_list = rte_zmalloc_socket("resubmit_list",
2008                                 resubmit_num * sizeof(struct rte_vhost_resubmit_desc),
2009                                 0, vq->numa_node);
2010                 if (resubmit->resubmit_list == NULL) {
2011                         VHOST_LOG_CONFIG(ERR,
2012                                         "(%s) failed to allocate memory for resubmit desc.\n",
2013                                         dev->ifname);
2014                         rte_free(resubmit);
2015                         return RTE_VHOST_MSG_RESULT_ERR;
2016                 }
2017
2018                 num = 0;
2019                 for (i = 0; i < inflight_packed->desc_num; i++) {
2020                         if (vq->inflight_packed->desc[i].inflight == 1) {
2021                                 resubmit->resubmit_list[num].index = i;
2022                                 resubmit->resubmit_list[num].counter =
2023                                         inflight_packed->desc[i].counter;
2024                                 num++;
2025                         }
2026                 }
2027                 resubmit->resubmit_num = num;
2028
2029                 if (resubmit->resubmit_num > 1)
2030                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
2031                               sizeof(struct rte_vhost_resubmit_desc),
2032                               resubmit_desc_compare);
2033
2034                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
2035                 vq->resubmit_inflight = resubmit;
2036         }
2037
2038         return RTE_VHOST_MSG_RESULT_OK;
2039 }
2040
2041 static int
2042 vhost_user_set_vring_kick(struct virtio_net **pdev,
2043                         struct vhu_msg_context *ctx,
2044                         int main_fd __rte_unused)
2045 {
2046         struct virtio_net *dev = *pdev;
2047         struct vhost_vring_file file;
2048         struct vhost_virtqueue *vq;
2049         int expected_fds;
2050
2051         expected_fds = (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
2052         if (validate_msg_fds(dev, ctx, expected_fds) != 0)
2053                 return RTE_VHOST_MSG_RESULT_ERR;
2054
2055         file.index = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
2056         if (ctx->msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK)
2057                 file.fd = VIRTIO_INVALID_EVENTFD;
2058         else
2059                 file.fd = ctx->fds[0];
2060         VHOST_LOG_CONFIG(INFO, "(%s) vring kick idx:%d file:%d\n",
2061                         dev->ifname, file.index, file.fd);
2062
2063         /* Interpret ring addresses only when ring is started. */
2064         dev = translate_ring_addresses(dev, file.index);
2065         if (!dev) {
2066                 if (file.fd != VIRTIO_INVALID_EVENTFD)
2067                         close(file.fd);
2068
2069                 return RTE_VHOST_MSG_RESULT_ERR;
2070         }
2071
2072         *pdev = dev;
2073
2074         vq = dev->virtqueue[file.index];
2075
2076         /*
2077          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
2078          * the ring starts already enabled. Otherwise, it is enabled via
2079          * the SET_VRING_ENABLE message.
2080          */
2081         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
2082                 vq->enabled = true;
2083         }
2084
2085         if (vq->ready) {
2086                 vq->ready = false;
2087                 vhost_user_notify_queue_state(dev, file.index, 0);
2088         }
2089
2090         if (vq->kickfd >= 0)
2091                 close(vq->kickfd);
2092         vq->kickfd = file.fd;
2093
2094         if (vq_is_packed(dev)) {
2095                 if (vhost_check_queue_inflights_packed(dev, vq)) {
2096                         VHOST_LOG_CONFIG(ERR, "(%s) failed to inflights for vq: %d\n",
2097                                         dev->ifname, file.index);
2098                         return RTE_VHOST_MSG_RESULT_ERR;
2099                 }
2100         } else {
2101                 if (vhost_check_queue_inflights_split(dev, vq)) {
2102                         VHOST_LOG_CONFIG(ERR, "(%s) failed to inflights for vq: %d\n",
2103                                         dev->ifname, file.index);
2104                         return RTE_VHOST_MSG_RESULT_ERR;
2105                 }
2106         }
2107
2108         return RTE_VHOST_MSG_RESULT_OK;
2109 }
2110
2111 /*
2112  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
2113  */
2114 static int
2115 vhost_user_get_vring_base(struct virtio_net **pdev,
2116                         struct vhu_msg_context *ctx,
2117                         int main_fd __rte_unused)
2118 {
2119         struct virtio_net *dev = *pdev;
2120         struct vhost_virtqueue *vq = dev->virtqueue[ctx->msg.payload.state.index];
2121         uint64_t val;
2122
2123         if (validate_msg_fds(dev, ctx, 0) != 0)
2124                 return RTE_VHOST_MSG_RESULT_ERR;
2125
2126         /* We have to stop the queue (virtio) if it is running. */
2127         vhost_destroy_device_notify(dev);
2128
2129         dev->flags &= ~VIRTIO_DEV_READY;
2130         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
2131
2132         /* Here we are safe to get the indexes */
2133         if (vq_is_packed(dev)) {
2134                 /*
2135                  * Bit[0:14]: avail index
2136                  * Bit[15]: avail wrap counter
2137                  */
2138                 val = vq->last_avail_idx & 0x7fff;
2139                 val |= vq->avail_wrap_counter << 15;
2140                 ctx->msg.payload.state.num = val;
2141         } else {
2142                 ctx->msg.payload.state.num = vq->last_avail_idx;
2143         }
2144
2145         VHOST_LOG_CONFIG(INFO, "(%s) vring base idx:%d file:%d\n",
2146                         dev->ifname, ctx->msg.payload.state.index,
2147                         ctx->msg.payload.state.num);
2148         /*
2149          * Based on current qemu vhost-user implementation, this message is
2150          * sent and only sent in vhost_vring_stop.
2151          * TODO: cleanup the vring, it isn't usable since here.
2152          */
2153         if (vq->kickfd >= 0)
2154                 close(vq->kickfd);
2155
2156         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
2157
2158         if (vq->callfd >= 0)
2159                 close(vq->callfd);
2160
2161         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
2162
2163         vq->signalled_used_valid = false;
2164
2165         if (vq_is_packed(dev)) {
2166                 rte_free(vq->shadow_used_packed);
2167                 vq->shadow_used_packed = NULL;
2168         } else {
2169                 rte_free(vq->shadow_used_split);
2170                 vq->shadow_used_split = NULL;
2171         }
2172
2173         rte_free(vq->batch_copy_elems);
2174         vq->batch_copy_elems = NULL;
2175
2176         rte_free(vq->log_cache);
2177         vq->log_cache = NULL;
2178
2179         ctx->msg.size = sizeof(ctx->msg.payload.state);
2180         ctx->fd_num = 0;
2181
2182         vhost_user_iotlb_flush_all(vq);
2183
2184         vring_invalidate(dev, vq);
2185
2186         return RTE_VHOST_MSG_RESULT_REPLY;
2187 }
2188
2189 /*
2190  * when virtio queues are ready to work, qemu will send us to
2191  * enable the virtio queue pair.
2192  */
2193 static int
2194 vhost_user_set_vring_enable(struct virtio_net **pdev,
2195                         struct vhu_msg_context *ctx,
2196                         int main_fd __rte_unused)
2197 {
2198         struct virtio_net *dev = *pdev;
2199         bool enable = !!ctx->msg.payload.state.num;
2200         int index = (int)ctx->msg.payload.state.index;
2201
2202         if (validate_msg_fds(dev, ctx, 0) != 0)
2203                 return RTE_VHOST_MSG_RESULT_ERR;
2204
2205         VHOST_LOG_CONFIG(INFO, "(%s) set queue enable: %d to qp idx: %d\n",
2206                         dev->ifname, enable, index);
2207
2208         if (enable && dev->virtqueue[index]->async) {
2209                 if (dev->virtqueue[index]->async->pkts_inflight_n) {
2210                         VHOST_LOG_CONFIG(ERR,
2211                                 "(%s) failed to enable vring. Inflight packets must be completed first\n",
2212                                 dev->ifname);
2213                         return RTE_VHOST_MSG_RESULT_ERR;
2214                 }
2215         }
2216
2217         dev->virtqueue[index]->enabled = enable;
2218
2219         return RTE_VHOST_MSG_RESULT_OK;
2220 }
2221
2222 static int
2223 vhost_user_get_protocol_features(struct virtio_net **pdev,
2224                         struct vhu_msg_context *ctx,
2225                         int main_fd __rte_unused)
2226 {
2227         struct virtio_net *dev = *pdev;
2228         uint64_t features, protocol_features;
2229
2230         if (validate_msg_fds(dev, ctx, 0) != 0)
2231                 return RTE_VHOST_MSG_RESULT_ERR;
2232
2233         rte_vhost_driver_get_features(dev->ifname, &features);
2234         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2235
2236         ctx->msg.payload.u64 = protocol_features;
2237         ctx->msg.size = sizeof(ctx->msg.payload.u64);
2238         ctx->fd_num = 0;
2239
2240         return RTE_VHOST_MSG_RESULT_REPLY;
2241 }
2242
2243 static int
2244 vhost_user_set_protocol_features(struct virtio_net **pdev,
2245                         struct vhu_msg_context *ctx,
2246                         int main_fd __rte_unused)
2247 {
2248         struct virtio_net *dev = *pdev;
2249         uint64_t protocol_features = ctx->msg.payload.u64;
2250         uint64_t slave_protocol_features = 0;
2251
2252         if (validate_msg_fds(dev, ctx, 0) != 0)
2253                 return RTE_VHOST_MSG_RESULT_ERR;
2254
2255         rte_vhost_driver_get_protocol_features(dev->ifname,
2256                         &slave_protocol_features);
2257         if (protocol_features & ~slave_protocol_features) {
2258                 VHOST_LOG_CONFIG(ERR, "(%s) received invalid protocol features.\n", dev->ifname);
2259                 return RTE_VHOST_MSG_RESULT_ERR;
2260         }
2261
2262         dev->protocol_features = protocol_features;
2263         VHOST_LOG_CONFIG(INFO, "(%s) negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2264                 dev->ifname, dev->protocol_features);
2265
2266         return RTE_VHOST_MSG_RESULT_OK;
2267 }
2268
2269 static int
2270 vhost_user_set_log_base(struct virtio_net **pdev,
2271                         struct vhu_msg_context *ctx,
2272                         int main_fd __rte_unused)
2273 {
2274         struct virtio_net *dev = *pdev;
2275         int fd = ctx->fds[0];
2276         uint64_t size, off;
2277         void *addr;
2278         uint32_t i;
2279
2280         if (validate_msg_fds(dev, ctx, 1) != 0)
2281                 return RTE_VHOST_MSG_RESULT_ERR;
2282
2283         if (fd < 0) {
2284                 VHOST_LOG_CONFIG(ERR, "(%s) invalid log fd: %d\n", dev->ifname, fd);
2285                 return RTE_VHOST_MSG_RESULT_ERR;
2286         }
2287
2288         if (ctx->msg.size != sizeof(VhostUserLog)) {
2289                 VHOST_LOG_CONFIG(ERR, "(%s) invalid log base msg size: %"PRId32" != %d\n",
2290                         dev->ifname, ctx->msg.size, (int)sizeof(VhostUserLog));
2291                 goto close_msg_fds;
2292         }
2293
2294         size = ctx->msg.payload.log.mmap_size;
2295         off  = ctx->msg.payload.log.mmap_offset;
2296
2297         /* Check for mmap size and offset overflow. */
2298         if (off >= -size) {
2299                 VHOST_LOG_CONFIG(ERR,
2300                                 "(%s) log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2301                                 dev->ifname, off, size);
2302                 goto close_msg_fds;
2303         }
2304
2305         VHOST_LOG_CONFIG(INFO, "(%s) log mmap size: %"PRId64", offset: %"PRId64"\n",
2306                         dev->ifname, size, off);
2307
2308         /*
2309          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2310          * fail when offset is not page size aligned.
2311          */
2312         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2313         close(fd);
2314         if (addr == MAP_FAILED) {
2315                 VHOST_LOG_CONFIG(ERR, "(%s) mmap log base failed!\n", dev->ifname);
2316                 return RTE_VHOST_MSG_RESULT_ERR;
2317         }
2318
2319         /*
2320          * Free previously mapped log memory on occasionally
2321          * multiple VHOST_USER_SET_LOG_BASE.
2322          */
2323         if (dev->log_addr) {
2324                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2325         }
2326         dev->log_addr = (uint64_t)(uintptr_t)addr;
2327         dev->log_base = dev->log_addr + off;
2328         dev->log_size = size;
2329
2330         for (i = 0; i < dev->nr_vring; i++) {
2331                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2332
2333                 rte_free(vq->log_cache);
2334                 vq->log_cache = NULL;
2335                 vq->log_cache_nb_elem = 0;
2336                 vq->log_cache = rte_malloc_socket("vq log cache",
2337                                 sizeof(struct log_cache_entry) * VHOST_LOG_CACHE_NR,
2338                                 0, vq->numa_node);
2339                 /*
2340                  * If log cache alloc fail, don't fail migration, but no
2341                  * caching will be done, which will impact performance
2342                  */
2343                 if (!vq->log_cache)
2344                         VHOST_LOG_CONFIG(ERR, "(%s) failed to allocate VQ logging cache\n",
2345                                         dev->ifname);
2346         }
2347
2348         /*
2349          * The spec is not clear about it (yet), but QEMU doesn't expect
2350          * any payload in the reply.
2351          */
2352         ctx->msg.size = 0;
2353         ctx->fd_num = 0;
2354
2355         return RTE_VHOST_MSG_RESULT_REPLY;
2356
2357 close_msg_fds:
2358         close_msg_fds(ctx);
2359         return RTE_VHOST_MSG_RESULT_ERR;
2360 }
2361
2362 static int vhost_user_set_log_fd(struct virtio_net **pdev,
2363                         struct vhu_msg_context *ctx,
2364                         int main_fd __rte_unused)
2365 {
2366         struct virtio_net *dev = *pdev;
2367
2368         if (validate_msg_fds(dev, ctx, 1) != 0)
2369                 return RTE_VHOST_MSG_RESULT_ERR;
2370
2371         close(ctx->fds[0]);
2372         VHOST_LOG_CONFIG(INFO, "(%s) not implemented.\n", dev->ifname);
2373
2374         return RTE_VHOST_MSG_RESULT_OK;
2375 }
2376
2377 /*
2378  * An rarp packet is constructed and broadcasted to notify switches about
2379  * the new location of the migrated VM, so that packets from outside will
2380  * not be lost after migration.
2381  *
2382  * However, we don't actually "send" a rarp packet here, instead, we set
2383  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2384  */
2385 static int
2386 vhost_user_send_rarp(struct virtio_net **pdev,
2387                         struct vhu_msg_context *ctx,
2388                         int main_fd __rte_unused)
2389 {
2390         struct virtio_net *dev = *pdev;
2391         uint8_t *mac = (uint8_t *)&ctx->msg.payload.u64;
2392         struct rte_vdpa_device *vdpa_dev;
2393
2394         if (validate_msg_fds(dev, ctx, 0) != 0)
2395                 return RTE_VHOST_MSG_RESULT_ERR;
2396
2397         VHOST_LOG_CONFIG(DEBUG, "(%s) MAC: " RTE_ETHER_ADDR_PRT_FMT "\n",
2398                 dev->ifname, mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2399         memcpy(dev->mac.addr_bytes, mac, 6);
2400
2401         /*
2402          * Set the flag to inject a RARP broadcast packet at
2403          * rte_vhost_dequeue_burst().
2404          *
2405          * __ATOMIC_RELEASE ordering is for making sure the mac is
2406          * copied before the flag is set.
2407          */
2408         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2409         vdpa_dev = dev->vdpa_dev;
2410         if (vdpa_dev && vdpa_dev->ops->migration_done)
2411                 vdpa_dev->ops->migration_done(dev->vid);
2412
2413         return RTE_VHOST_MSG_RESULT_OK;
2414 }
2415
2416 static int
2417 vhost_user_net_set_mtu(struct virtio_net **pdev,
2418                         struct vhu_msg_context *ctx,
2419                         int main_fd __rte_unused)
2420 {
2421         struct virtio_net *dev = *pdev;
2422
2423         if (validate_msg_fds(dev, ctx, 0) != 0)
2424                 return RTE_VHOST_MSG_RESULT_ERR;
2425
2426         if (ctx->msg.payload.u64 < VIRTIO_MIN_MTU ||
2427                         ctx->msg.payload.u64 > VIRTIO_MAX_MTU) {
2428                 VHOST_LOG_CONFIG(ERR, "(%s) invalid MTU size (%"PRIu64")\n",
2429                                 dev->ifname, ctx->msg.payload.u64);
2430
2431                 return RTE_VHOST_MSG_RESULT_ERR;
2432         }
2433
2434         dev->mtu = ctx->msg.payload.u64;
2435
2436         return RTE_VHOST_MSG_RESULT_OK;
2437 }
2438
2439 static int
2440 vhost_user_set_req_fd(struct virtio_net **pdev,
2441                         struct vhu_msg_context *ctx,
2442                         int main_fd __rte_unused)
2443 {
2444         struct virtio_net *dev = *pdev;
2445         int fd = ctx->fds[0];
2446
2447         if (validate_msg_fds(dev, ctx, 1) != 0)
2448                 return RTE_VHOST_MSG_RESULT_ERR;
2449
2450         if (fd < 0) {
2451                 VHOST_LOG_CONFIG(ERR, "(%s) invalid file descriptor for slave channel (%d)\n",
2452                                 dev->ifname, fd);
2453                 return RTE_VHOST_MSG_RESULT_ERR;
2454         }
2455
2456         if (dev->slave_req_fd >= 0)
2457                 close(dev->slave_req_fd);
2458
2459         dev->slave_req_fd = fd;
2460
2461         return RTE_VHOST_MSG_RESULT_OK;
2462 }
2463
2464 static int
2465 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2466 {
2467         struct vhost_vring_addr *ra;
2468         uint64_t start, end, len;
2469
2470         start = imsg->iova;
2471         end = start + imsg->size;
2472
2473         ra = &vq->ring_addrs;
2474         len = sizeof(struct vring_desc) * vq->size;
2475         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2476                 return 1;
2477
2478         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2479         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2480                 return 1;
2481
2482         len = sizeof(struct vring_used) +
2483                sizeof(struct vring_used_elem) * vq->size;
2484         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2485                 return 1;
2486
2487         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2488                 len = sizeof(uint64_t);
2489                 if (ra->log_guest_addr < end &&
2490                     (ra->log_guest_addr + len) > start)
2491                         return 1;
2492         }
2493
2494         return 0;
2495 }
2496
2497 static int
2498 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2499 {
2500         struct vhost_vring_addr *ra;
2501         uint64_t start, end, len;
2502
2503         start = imsg->iova;
2504         end = start + imsg->size;
2505
2506         ra = &vq->ring_addrs;
2507         len = sizeof(struct vring_packed_desc) * vq->size;
2508         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2509                 return 1;
2510
2511         len = sizeof(struct vring_packed_desc_event);
2512         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2513                 return 1;
2514
2515         len = sizeof(struct vring_packed_desc_event);
2516         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2517                 return 1;
2518
2519         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2520                 len = sizeof(uint64_t);
2521                 if (ra->log_guest_addr < end &&
2522                     (ra->log_guest_addr + len) > start)
2523                         return 1;
2524         }
2525
2526         return 0;
2527 }
2528
2529 static int is_vring_iotlb(struct virtio_net *dev,
2530                           struct vhost_virtqueue *vq,
2531                           struct vhost_iotlb_msg *imsg)
2532 {
2533         if (vq_is_packed(dev))
2534                 return is_vring_iotlb_packed(vq, imsg);
2535         else
2536                 return is_vring_iotlb_split(vq, imsg);
2537 }
2538
2539 static int
2540 vhost_user_iotlb_msg(struct virtio_net **pdev,
2541                         struct vhu_msg_context *ctx,
2542                         int main_fd __rte_unused)
2543 {
2544         struct virtio_net *dev = *pdev;
2545         struct vhost_iotlb_msg *imsg = &ctx->msg.payload.iotlb;
2546         uint16_t i;
2547         uint64_t vva, len;
2548
2549         if (validate_msg_fds(dev, ctx, 0) != 0)
2550                 return RTE_VHOST_MSG_RESULT_ERR;
2551
2552         switch (imsg->type) {
2553         case VHOST_IOTLB_UPDATE:
2554                 len = imsg->size;
2555                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2556                 if (!vva)
2557                         return RTE_VHOST_MSG_RESULT_ERR;
2558
2559                 for (i = 0; i < dev->nr_vring; i++) {
2560                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2561
2562                         if (!vq)
2563                                 continue;
2564
2565                         vhost_user_iotlb_cache_insert(dev, vq, imsg->iova, vva,
2566                                         len, imsg->perm);
2567
2568                         if (is_vring_iotlb(dev, vq, imsg))
2569                                 *pdev = dev = translate_ring_addresses(dev, i);
2570                 }
2571                 break;
2572         case VHOST_IOTLB_INVALIDATE:
2573                 for (i = 0; i < dev->nr_vring; i++) {
2574                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2575
2576                         if (!vq)
2577                                 continue;
2578
2579                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2580                                         imsg->size);
2581
2582                         if (is_vring_iotlb(dev, vq, imsg))
2583                                 vring_invalidate(dev, vq);
2584                 }
2585                 break;
2586         default:
2587                 VHOST_LOG_CONFIG(ERR, "(%s) invalid IOTLB message type (%d)\n",
2588                                 dev->ifname, imsg->type);
2589                 return RTE_VHOST_MSG_RESULT_ERR;
2590         }
2591
2592         return RTE_VHOST_MSG_RESULT_OK;
2593 }
2594
2595 static int
2596 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2597                         struct vhu_msg_context *ctx,
2598                         int main_fd __rte_unused)
2599 {
2600         struct virtio_net *dev = *pdev;
2601 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2602         struct uffdio_api api_struct;
2603
2604         if (validate_msg_fds(dev, ctx, 0) != 0)
2605                 return RTE_VHOST_MSG_RESULT_ERR;
2606
2607         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2608
2609         if (dev->postcopy_ufd == -1) {
2610                 VHOST_LOG_CONFIG(ERR, "(%s) userfaultfd not available: %s\n",
2611                         dev->ifname, strerror(errno));
2612                 return RTE_VHOST_MSG_RESULT_ERR;
2613         }
2614         api_struct.api = UFFD_API;
2615         api_struct.features = 0;
2616         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2617                 VHOST_LOG_CONFIG(ERR, "(%s) UFFDIO_API ioctl failure: %s\n",
2618                         dev->ifname, strerror(errno));
2619                 close(dev->postcopy_ufd);
2620                 dev->postcopy_ufd = -1;
2621                 return RTE_VHOST_MSG_RESULT_ERR;
2622         }
2623         ctx->fds[0] = dev->postcopy_ufd;
2624         ctx->fd_num = 1;
2625
2626         return RTE_VHOST_MSG_RESULT_REPLY;
2627 #else
2628         dev->postcopy_ufd = -1;
2629         ctx->fd_num = 0;
2630
2631         return RTE_VHOST_MSG_RESULT_ERR;
2632 #endif
2633 }
2634
2635 static int
2636 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2637                         struct vhu_msg_context *ctx __rte_unused,
2638                         int main_fd __rte_unused)
2639 {
2640         struct virtio_net *dev = *pdev;
2641
2642         if (validate_msg_fds(dev, ctx, 0) != 0)
2643                 return RTE_VHOST_MSG_RESULT_ERR;
2644
2645         if (dev->mem && dev->mem->nregions) {
2646                 VHOST_LOG_CONFIG(ERR, "(%s) regions already registered at postcopy-listen\n",
2647                                 dev->ifname);
2648                 return RTE_VHOST_MSG_RESULT_ERR;
2649         }
2650         dev->postcopy_listening = 1;
2651
2652         return RTE_VHOST_MSG_RESULT_OK;
2653 }
2654
2655 static int
2656 vhost_user_postcopy_end(struct virtio_net **pdev,
2657                         struct vhu_msg_context *ctx,
2658                         int main_fd __rte_unused)
2659 {
2660         struct virtio_net *dev = *pdev;
2661
2662         if (validate_msg_fds(dev, ctx, 0) != 0)
2663                 return RTE_VHOST_MSG_RESULT_ERR;
2664
2665         dev->postcopy_listening = 0;
2666         if (dev->postcopy_ufd >= 0) {
2667                 close(dev->postcopy_ufd);
2668                 dev->postcopy_ufd = -1;
2669         }
2670
2671         ctx->msg.payload.u64 = 0;
2672         ctx->msg.size = sizeof(ctx->msg.payload.u64);
2673         ctx->fd_num = 0;
2674
2675         return RTE_VHOST_MSG_RESULT_REPLY;
2676 }
2677
2678 static int
2679 vhost_user_get_status(struct virtio_net **pdev,
2680                       struct vhu_msg_context *ctx,
2681                       int main_fd __rte_unused)
2682 {
2683         struct virtio_net *dev = *pdev;
2684
2685         if (validate_msg_fds(dev, ctx, 0) != 0)
2686                 return RTE_VHOST_MSG_RESULT_ERR;
2687
2688         ctx->msg.payload.u64 = dev->status;
2689         ctx->msg.size = sizeof(ctx->msg.payload.u64);
2690         ctx->fd_num = 0;
2691
2692         return RTE_VHOST_MSG_RESULT_REPLY;
2693 }
2694
2695 static int
2696 vhost_user_set_status(struct virtio_net **pdev,
2697                         struct vhu_msg_context *ctx,
2698                         int main_fd __rte_unused)
2699 {
2700         struct virtio_net *dev = *pdev;
2701
2702         if (validate_msg_fds(dev, ctx, 0) != 0)
2703                 return RTE_VHOST_MSG_RESULT_ERR;
2704
2705         /* As per Virtio specification, the device status is 8bits long */
2706         if (ctx->msg.payload.u64 > UINT8_MAX) {
2707                 VHOST_LOG_CONFIG(ERR, "(%s) invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2708                                 dev->ifname, ctx->msg.payload.u64);
2709                 return RTE_VHOST_MSG_RESULT_ERR;
2710         }
2711
2712         dev->status = ctx->msg.payload.u64;
2713
2714         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2715             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2716                 VHOST_LOG_CONFIG(ERR,
2717                                 "(%s) FEATURES_OK bit is set but feature negotiation failed\n",
2718                                 dev->ifname);
2719                 /*
2720                  * Clear the bit to let the driver know about the feature
2721                  * negotiation failure
2722                  */
2723                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2724         }
2725
2726         VHOST_LOG_CONFIG(INFO, "(%s) new device status(0x%08x):\n", dev->ifname,
2727                         dev->status);
2728         VHOST_LOG_CONFIG(INFO, "(%s)\t-RESET: %u\n", dev->ifname,
2729                         (dev->status == VIRTIO_DEVICE_STATUS_RESET));
2730         VHOST_LOG_CONFIG(INFO, "(%s)\t-ACKNOWLEDGE: %u\n", dev->ifname,
2731                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK));
2732         VHOST_LOG_CONFIG(INFO, "(%s)\t-DRIVER: %u\n", dev->ifname,
2733                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER));
2734         VHOST_LOG_CONFIG(INFO, "(%s)\t-FEATURES_OK: %u\n", dev->ifname,
2735                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK));
2736         VHOST_LOG_CONFIG(INFO, "(%s)\t-DRIVER_OK: %u\n", dev->ifname,
2737                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK));
2738         VHOST_LOG_CONFIG(INFO, "(%s)\t-DEVICE_NEED_RESET: %u\n", dev->ifname,
2739                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET));
2740         VHOST_LOG_CONFIG(INFO, "(%s)\t-FAILED: %u\n", dev->ifname,
2741                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2742
2743         return RTE_VHOST_MSG_RESULT_OK;
2744 }
2745
2746 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2747                                         struct vhu_msg_context *ctx,
2748                                         int main_fd);
2749
2750 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2751         [VHOST_USER_NONE] = NULL,
2752         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2753         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2754         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2755         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2756         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2757         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2758         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2759         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2760         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2761         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2762         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2763         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2764         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2765         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2766         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2767         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2768         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2769         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2770         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2771         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2772         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2773         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2774         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2775         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2776         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2777         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2778         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2779         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2780         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2781 };
2782
2783 /* return bytes# of read on success or negative val on failure. */
2784 static int
2785 read_vhost_message(struct virtio_net *dev, int sockfd, struct  vhu_msg_context *ctx)
2786 {
2787         int ret;
2788
2789         ret = read_fd_message(dev->ifname, sockfd, (char *)&ctx->msg, VHOST_USER_HDR_SIZE,
2790                 ctx->fds, VHOST_MEMORY_MAX_NREGIONS, &ctx->fd_num);
2791         if (ret <= 0) {
2792                 return ret;
2793         } else if (ret != VHOST_USER_HDR_SIZE) {
2794                 VHOST_LOG_CONFIG(ERR, "(%s) Unexpected header size read\n", dev->ifname);
2795                 close_msg_fds(ctx);
2796                 return -1;
2797         }
2798
2799         if (ctx->msg.size) {
2800                 if (ctx->msg.size > sizeof(ctx->msg.payload)) {
2801                         VHOST_LOG_CONFIG(ERR, "(%s) invalid msg size: %d\n",
2802                                         dev->ifname, ctx->msg.size);
2803                         return -1;
2804                 }
2805                 ret = read(sockfd, &ctx->msg.payload, ctx->msg.size);
2806                 if (ret <= 0)
2807                         return ret;
2808                 if (ret != (int)ctx->msg.size) {
2809                         VHOST_LOG_CONFIG(ERR, "(%s) read control message failed\n", dev->ifname);
2810                         return -1;
2811                 }
2812         }
2813
2814         return ret;
2815 }
2816
2817 static int
2818 send_vhost_message(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx)
2819 {
2820         if (!ctx)
2821                 return 0;
2822
2823         return send_fd_message(dev->ifname, sockfd, (char *)&ctx->msg,
2824                 VHOST_USER_HDR_SIZE + ctx->msg.size, ctx->fds, ctx->fd_num);
2825 }
2826
2827 static int
2828 send_vhost_reply(struct virtio_net *dev, int sockfd, struct vhu_msg_context *ctx)
2829 {
2830         if (!ctx)
2831                 return 0;
2832
2833         ctx->msg.flags &= ~VHOST_USER_VERSION_MASK;
2834         ctx->msg.flags &= ~VHOST_USER_NEED_REPLY;
2835         ctx->msg.flags |= VHOST_USER_VERSION;
2836         ctx->msg.flags |= VHOST_USER_REPLY_MASK;
2837
2838         return send_vhost_message(dev, sockfd, ctx);
2839 }
2840
2841 static int
2842 send_vhost_slave_message(struct virtio_net *dev,
2843                 struct vhu_msg_context *ctx)
2844 {
2845         int ret;
2846
2847         if (ctx->msg.flags & VHOST_USER_NEED_REPLY)
2848                 rte_spinlock_lock(&dev->slave_req_lock);
2849
2850         ret = send_vhost_message(dev, dev->slave_req_fd, ctx);
2851         if (ret < 0 && (ctx->msg.flags & VHOST_USER_NEED_REPLY))
2852                 rte_spinlock_unlock(&dev->slave_req_lock);
2853
2854         return ret;
2855 }
2856
2857 /*
2858  * Allocate a queue pair if it hasn't been allocated yet
2859  */
2860 static int
2861 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2862                         struct vhu_msg_context *ctx)
2863 {
2864         uint32_t vring_idx;
2865
2866         switch (ctx->msg.request.master) {
2867         case VHOST_USER_SET_VRING_KICK:
2868         case VHOST_USER_SET_VRING_CALL:
2869         case VHOST_USER_SET_VRING_ERR:
2870                 vring_idx = ctx->msg.payload.u64 & VHOST_USER_VRING_IDX_MASK;
2871                 break;
2872         case VHOST_USER_SET_VRING_NUM:
2873         case VHOST_USER_SET_VRING_BASE:
2874         case VHOST_USER_GET_VRING_BASE:
2875         case VHOST_USER_SET_VRING_ENABLE:
2876                 vring_idx = ctx->msg.payload.state.index;
2877                 break;
2878         case VHOST_USER_SET_VRING_ADDR:
2879                 vring_idx = ctx->msg.payload.addr.index;
2880                 break;
2881         default:
2882                 return 0;
2883         }
2884
2885         if (vring_idx >= VHOST_MAX_VRING) {
2886                 VHOST_LOG_CONFIG(ERR, "(%s) invalid vring index: %u\n", dev->ifname, vring_idx);
2887                 return -1;
2888         }
2889
2890         if (dev->virtqueue[vring_idx])
2891                 return 0;
2892
2893         return alloc_vring_queue(dev, vring_idx);
2894 }
2895
2896 static void
2897 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2898 {
2899         unsigned int i = 0;
2900         unsigned int vq_num = 0;
2901
2902         while (vq_num < dev->nr_vring) {
2903                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2904
2905                 if (vq) {
2906                         rte_spinlock_lock(&vq->access_lock);
2907                         vq_num++;
2908                 }
2909                 i++;
2910         }
2911 }
2912
2913 static void
2914 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2915 {
2916         unsigned int i = 0;
2917         unsigned int vq_num = 0;
2918
2919         while (vq_num < dev->nr_vring) {
2920                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2921
2922                 if (vq) {
2923                         rte_spinlock_unlock(&vq->access_lock);
2924                         vq_num++;
2925                 }
2926                 i++;
2927         }
2928 }
2929
2930 int
2931 vhost_user_msg_handler(int vid, int fd)
2932 {
2933         struct virtio_net *dev;
2934         struct vhu_msg_context ctx;
2935         struct rte_vdpa_device *vdpa_dev;
2936         int ret;
2937         int unlock_required = 0;
2938         bool handled;
2939         int request;
2940         uint32_t i;
2941
2942         dev = get_device(vid);
2943         if (dev == NULL)
2944                 return -1;
2945
2946         if (!dev->notify_ops) {
2947                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2948                 if (!dev->notify_ops) {
2949                         VHOST_LOG_CONFIG(ERR, "(%s) failed to get callback ops for driver\n",
2950                                 dev->ifname);
2951                         return -1;
2952                 }
2953         }
2954
2955         ret = read_vhost_message(dev, fd, &ctx);
2956         if (ret <= 0) {
2957                 if (ret < 0)
2958                         VHOST_LOG_CONFIG(ERR, "(%s) vhost read message failed\n", dev->ifname);
2959                 else
2960                         VHOST_LOG_CONFIG(INFO, "(%s) vhost peer closed\n", dev->ifname);
2961
2962                 return -1;
2963         }
2964
2965         ret = 0;
2966         request = ctx.msg.request.master;
2967         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2968                         vhost_message_str[request]) {
2969                 if (request != VHOST_USER_IOTLB_MSG)
2970                         VHOST_LOG_CONFIG(INFO, "(%s) read message %s\n",
2971                                 dev->ifname, vhost_message_str[request]);
2972                 else
2973                         VHOST_LOG_CONFIG(DEBUG, "(%s) read message %s\n",
2974                                 dev->ifname, vhost_message_str[request]);
2975         } else {
2976                 VHOST_LOG_CONFIG(DEBUG, "(%s) external request %d\n", dev->ifname, request);
2977         }
2978
2979         ret = vhost_user_check_and_alloc_queue_pair(dev, &ctx);
2980         if (ret < 0) {
2981                 VHOST_LOG_CONFIG(ERR, "(%s) failed to alloc queue\n", dev->ifname);
2982                 return -1;
2983         }
2984
2985         /*
2986          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2987          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2988          * and device is destroyed. destroy_device waits for queues to be
2989          * inactive, so it is safe. Otherwise taking the access_lock
2990          * would cause a dead lock.
2991          */
2992         switch (request) {
2993         case VHOST_USER_SET_FEATURES:
2994         case VHOST_USER_SET_PROTOCOL_FEATURES:
2995         case VHOST_USER_SET_OWNER:
2996         case VHOST_USER_SET_MEM_TABLE:
2997         case VHOST_USER_SET_LOG_BASE:
2998         case VHOST_USER_SET_LOG_FD:
2999         case VHOST_USER_SET_VRING_NUM:
3000         case VHOST_USER_SET_VRING_ADDR:
3001         case VHOST_USER_SET_VRING_BASE:
3002         case VHOST_USER_SET_VRING_KICK:
3003         case VHOST_USER_SET_VRING_CALL:
3004         case VHOST_USER_SET_VRING_ERR:
3005         case VHOST_USER_SET_VRING_ENABLE:
3006         case VHOST_USER_SEND_RARP:
3007         case VHOST_USER_NET_SET_MTU:
3008         case VHOST_USER_SET_SLAVE_REQ_FD:
3009                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3010                         vhost_user_lock_all_queue_pairs(dev);
3011                         unlock_required = 1;
3012                 }
3013                 break;
3014         default:
3015                 break;
3016
3017         }
3018
3019         handled = false;
3020         if (dev->extern_ops.pre_msg_handle) {
3021                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
3022                                 (void *)&ctx.msg);
3023                 switch (ret) {
3024                 case RTE_VHOST_MSG_RESULT_REPLY:
3025                         send_vhost_reply(dev, fd, &ctx);
3026                         /* Fall-through */
3027                 case RTE_VHOST_MSG_RESULT_ERR:
3028                 case RTE_VHOST_MSG_RESULT_OK:
3029                         handled = true;
3030                         goto skip_to_post_handle;
3031                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3032                 default:
3033                         break;
3034                 }
3035         }
3036
3037         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
3038                 if (!vhost_message_handlers[request])
3039                         goto skip_to_post_handle;
3040                 ret = vhost_message_handlers[request](&dev, &ctx, fd);
3041
3042                 switch (ret) {
3043                 case RTE_VHOST_MSG_RESULT_ERR:
3044                         VHOST_LOG_CONFIG(ERR, "(%s) processing %s failed.\n",
3045                                         dev->ifname, vhost_message_str[request]);
3046                         handled = true;
3047                         break;
3048                 case RTE_VHOST_MSG_RESULT_OK:
3049                         VHOST_LOG_CONFIG(DEBUG, "(%s) processing %s succeeded.\n",
3050                                         dev->ifname, vhost_message_str[request]);
3051                         handled = true;
3052                         break;
3053                 case RTE_VHOST_MSG_RESULT_REPLY:
3054                         VHOST_LOG_CONFIG(DEBUG, "(%s) processing %s succeeded and needs reply.\n",
3055                                         dev->ifname, vhost_message_str[request]);
3056                         send_vhost_reply(dev, fd, &ctx);
3057                         handled = true;
3058                         break;
3059                 default:
3060                         break;
3061                 }
3062         }
3063
3064 skip_to_post_handle:
3065         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
3066                         dev->extern_ops.post_msg_handle) {
3067                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
3068                                 (void *)&ctx.msg);
3069                 switch (ret) {
3070                 case RTE_VHOST_MSG_RESULT_REPLY:
3071                         send_vhost_reply(dev, fd, &ctx);
3072                         /* Fall-through */
3073                 case RTE_VHOST_MSG_RESULT_ERR:
3074                 case RTE_VHOST_MSG_RESULT_OK:
3075                         handled = true;
3076                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
3077                 default:
3078                         break;
3079                 }
3080         }
3081
3082         /* If message was not handled at this stage, treat it as an error */
3083         if (!handled) {
3084                 VHOST_LOG_CONFIG(ERR, "(%s) vhost message (req: %d) was not handled.\n",
3085                                 dev->ifname, request);
3086                 close_msg_fds(&ctx);
3087                 ret = RTE_VHOST_MSG_RESULT_ERR;
3088         }
3089
3090         /*
3091          * If the request required a reply that was already sent,
3092          * this optional reply-ack won't be sent as the
3093          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
3094          */
3095         if (ctx.msg.flags & VHOST_USER_NEED_REPLY) {
3096                 ctx.msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
3097                 ctx.msg.size = sizeof(ctx.msg.payload.u64);
3098                 ctx.fd_num = 0;
3099                 send_vhost_reply(dev, fd, &ctx);
3100         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
3101                 VHOST_LOG_CONFIG(ERR, "(%s) vhost message handling failed.\n", dev->ifname);
3102                 return -1;
3103         }
3104
3105         for (i = 0; i < dev->nr_vring; i++) {
3106                 struct vhost_virtqueue *vq = dev->virtqueue[i];
3107                 bool cur_ready = vq_is_ready(dev, vq);
3108
3109                 if (cur_ready != (vq && vq->ready)) {
3110                         vq->ready = cur_ready;
3111                         vhost_user_notify_queue_state(dev, i, cur_ready);
3112                 }
3113         }
3114
3115         if (unlock_required)
3116                 vhost_user_unlock_all_queue_pairs(dev);
3117
3118         if (!virtio_is_ready(dev))
3119                 goto out;
3120
3121         /*
3122          * Virtio is now ready. If not done already, it is time
3123          * to notify the application it can process the rings and
3124          * configure the vDPA device if present.
3125          */
3126
3127         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
3128                 if (dev->notify_ops->new_device(dev->vid) == 0)
3129                         dev->flags |= VIRTIO_DEV_RUNNING;
3130         }
3131
3132         vdpa_dev = dev->vdpa_dev;
3133         if (!vdpa_dev)
3134                 goto out;
3135
3136         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
3137                 if (vdpa_dev->ops->dev_conf(dev->vid))
3138                         VHOST_LOG_CONFIG(ERR, "(%s) failed to configure vDPA device\n",
3139                                         dev->ifname);
3140                 else
3141                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
3142         }
3143
3144 out:
3145         return 0;
3146 }
3147
3148 static int process_slave_message_reply(struct virtio_net *dev,
3149                                        const struct vhu_msg_context *ctx)
3150 {
3151         struct vhu_msg_context msg_reply;
3152         int ret;
3153
3154         if ((ctx->msg.flags & VHOST_USER_NEED_REPLY) == 0)
3155                 return 0;
3156
3157         ret = read_vhost_message(dev, dev->slave_req_fd, &msg_reply);
3158         if (ret <= 0) {
3159                 if (ret < 0)
3160                         VHOST_LOG_CONFIG(ERR, "(%s) vhost read slave message reply failed\n",
3161                                         dev->ifname);
3162                 else
3163                         VHOST_LOG_CONFIG(INFO, "(%s) vhost peer closed\n", dev->ifname);
3164                 ret = -1;
3165                 goto out;
3166         }
3167
3168         ret = 0;
3169         if (msg_reply.msg.request.slave != ctx->msg.request.slave) {
3170                 VHOST_LOG_CONFIG(ERR, "(%s) received unexpected msg type (%u), expected %u\n",
3171                                 dev->ifname, msg_reply.msg.request.slave, ctx->msg.request.slave);
3172                 ret = -1;
3173                 goto out;
3174         }
3175
3176         ret = msg_reply.msg.payload.u64 ? -1 : 0;
3177
3178 out:
3179         rte_spinlock_unlock(&dev->slave_req_lock);
3180         return ret;
3181 }
3182
3183 int
3184 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3185 {
3186         int ret;
3187         struct vhu_msg_context ctx = {
3188                 .msg = {
3189                         .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3190                         .flags = VHOST_USER_VERSION,
3191                         .size = sizeof(ctx.msg.payload.iotlb),
3192                         .payload.iotlb = {
3193                                 .iova = iova,
3194                                 .perm = perm,
3195                                 .type = VHOST_IOTLB_MISS,
3196                         },
3197                 },
3198         };
3199
3200         ret = send_vhost_message(dev, dev->slave_req_fd, &ctx);
3201         if (ret < 0) {
3202                 VHOST_LOG_CONFIG(ERR, "(%s) failed to send IOTLB miss message (%d)\n",
3203                                 dev->ifname, ret);
3204                 return ret;
3205         }
3206
3207         return 0;
3208 }
3209
3210 static int
3211 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3212 {
3213         int ret;
3214         struct vhu_msg_context ctx = {
3215                 .msg = {
3216                         .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3217                         .flags = VHOST_USER_VERSION,
3218                         .size = 0,
3219                 }
3220         };
3221
3222         if (need_reply)
3223                 ctx.msg.flags |= VHOST_USER_NEED_REPLY;
3224
3225         ret = send_vhost_slave_message(dev, &ctx);
3226         if (ret < 0) {
3227                 VHOST_LOG_CONFIG(ERR, "(%s) failed to send config change (%d)\n",
3228                                 dev->ifname, ret);
3229                 return ret;
3230         }
3231
3232         return process_slave_message_reply(dev, &ctx);
3233 }
3234
3235 int
3236 rte_vhost_slave_config_change(int vid, bool need_reply)
3237 {
3238         struct virtio_net *dev;
3239
3240         dev = get_device(vid);
3241         if (!dev)
3242                 return -ENODEV;
3243
3244         return vhost_user_slave_config_change(dev, need_reply);
3245 }
3246
3247 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3248                                                     int index, int fd,
3249                                                     uint64_t offset,
3250                                                     uint64_t size)
3251 {
3252         int ret;
3253         struct vhu_msg_context ctx = {
3254                 .msg = {
3255                         .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3256                         .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3257                         .size = sizeof(ctx.msg.payload.area),
3258                         .payload.area = {
3259                                 .u64 = index & VHOST_USER_VRING_IDX_MASK,
3260                                 .size = size,
3261                                 .offset = offset,
3262                         },
3263                 },
3264         };
3265
3266         if (fd < 0)
3267                 ctx.msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3268         else {
3269                 ctx.fds[0] = fd;
3270                 ctx.fd_num = 1;
3271         }
3272
3273         ret = send_vhost_slave_message(dev, &ctx);
3274         if (ret < 0) {
3275                 VHOST_LOG_CONFIG(ERR, "(%s) failed to set host notifier (%d)\n",
3276                                 dev->ifname, ret);
3277                 return ret;
3278         }
3279
3280         return process_slave_message_reply(dev, &ctx);
3281 }
3282
3283 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3284 {
3285         struct virtio_net *dev;
3286         struct rte_vdpa_device *vdpa_dev;
3287         int vfio_device_fd, ret = 0;
3288         uint64_t offset, size;
3289         unsigned int i, q_start, q_last;
3290
3291         dev = get_device(vid);
3292         if (!dev)
3293                 return -ENODEV;
3294
3295         vdpa_dev = dev->vdpa_dev;
3296         if (vdpa_dev == NULL)
3297                 return -ENODEV;
3298
3299         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3300             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3301             !(dev->protocol_features &
3302                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3303             !(dev->protocol_features &
3304                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3305             !(dev->protocol_features &
3306                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3307                 return -ENOTSUP;
3308
3309         if (qid == RTE_VHOST_QUEUE_ALL) {
3310                 q_start = 0;
3311                 q_last = dev->nr_vring - 1;
3312         } else {
3313                 if (qid >= dev->nr_vring)
3314                         return -EINVAL;
3315                 q_start = qid;
3316                 q_last = qid;
3317         }
3318
3319         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3320         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3321
3322         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3323         if (vfio_device_fd < 0)
3324                 return -ENOTSUP;
3325
3326         if (enable) {
3327                 for (i = q_start; i <= q_last; i++) {
3328                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3329                                         &size) < 0) {
3330                                 ret = -ENOTSUP;
3331                                 goto disable;
3332                         }
3333
3334                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3335                                         vfio_device_fd, offset, size) < 0) {
3336                                 ret = -EFAULT;
3337                                 goto disable;
3338                         }
3339                 }
3340         } else {
3341 disable:
3342                 for (i = q_start; i <= q_last; i++) {
3343                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3344                                         0, 0);
3345                 }
3346         }
3347
3348         return ret;
3349 }