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