lib/librte_vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2018 Intel Corporation
   3  */
   4
   5 /* Security model
   6  * --------------
   7  * The vhost-user protocol connection is an external interface, so it must be
   8  * robust against invalid inputs.
   9  *
  10  * This is important because the vhost-user master is only one step removed
  11  * from the guest.  Malicious guests that have escaped will then launch further
  12  * attacks from the vhost-user master.
  13  *
  14  * Even in deployments where guests are trusted, a bug in the vhost-user master
  15  * can still cause invalid messages to be sent.  Such messages must not
  16  * compromise the stability of the DPDK application by causing crashes, memory
  17  * corruption, or other problematic behavior.
  18  *
  19  * Do not assume received VhostUserMsg fields contain sensible values!
  20  */
  21
  22 #include <stdint.h>
  23 #include <stdio.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <fcntl.h>
  28 #include <sys/ioctl.h>
  29 #include <sys/mman.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/syscall.h>
  33 #include <assert.h>
  34 #ifdef RTE_LIBRTE_VHOST_NUMA
  35 #include <numaif.h>
  36 #endif
  37 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
  38 #include <linux/userfaultfd.h>
  39 #endif
  40 #ifdef F_ADD_SEALS /* if file sealing is supported, so is memfd */
  41 #include <linux/memfd.h>
  42 #define MEMFD_SUPPORTED
  43 #endif
  44
  45 #include <rte_common.h>
  46 #include <rte_malloc.h>
  47 #include <rte_log.h>
  48
  49 #include "iotlb.h"
  50 #include "vhost.h"
  51 #include "vhost_user.h"
  52
  53 #define VIRTIO_MIN_MTU 68
  54 #define VIRTIO_MAX_MTU 65535
  55
  56 #define INFLIGHT_ALIGNMENT      64
  57 #define INFLIGHT_VERSION        0x1
  58
  59 static const char *vhost_message_str[VHOST_USER_MAX] = {
  60         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  61         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  62         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  63         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  64         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  65         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  66         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  67         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  68         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  69         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  70         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  71         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  72         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  73         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  74         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  75         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  76         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  77         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  78         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  79         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  80         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  81         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  82         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  83         [VHOST_USER_CRYPTO_CREATE_SESS] = "VHOST_USER_CRYPTO_CREATE_SESS",
  84         [VHOST_USER_CRYPTO_CLOSE_SESS] = "VHOST_USER_CRYPTO_CLOSE_SESS",
  85         [VHOST_USER_POSTCOPY_ADVISE]  = "VHOST_USER_POSTCOPY_ADVISE",
  86         [VHOST_USER_POSTCOPY_LISTEN]  = "VHOST_USER_POSTCOPY_LISTEN",
  87         [VHOST_USER_POSTCOPY_END]  = "VHOST_USER_POSTCOPY_END",
  88         [VHOST_USER_GET_INFLIGHT_FD] = "VHOST_USER_GET_INFLIGHT_FD",
  89         [VHOST_USER_SET_INFLIGHT_FD] = "VHOST_USER_SET_INFLIGHT_FD",
  90         [VHOST_USER_SET_STATUS] = "VHOST_USER_SET_STATUS",
  91         [VHOST_USER_GET_STATUS] = "VHOST_USER_GET_STATUS",
  92 };
  93
  94 static int send_vhost_reply(int sockfd, struct VhostUserMsg *msg);
  95 static int read_vhost_message(int sockfd, struct VhostUserMsg *msg);
  96
  97 static void
  98 close_msg_fds(struct VhostUserMsg *msg)
  99 {
 100         int i;
 101
 102         for (i = 0; i < msg->fd_num; i++)
 103                 close(msg->fds[i]);
 104 }
 105
 106 /*
 107  * Ensure the expected number of FDs is received,
 108  * close all FDs and return an error if this is not the case.
 109  */
 110 static int
 111 validate_msg_fds(struct VhostUserMsg *msg, int expected_fds)
 112 {
 113         if (msg->fd_num == expected_fds)
 114                 return 0;
 115
 116         VHOST_LOG_CONFIG(ERR,
 117                 " Expect %d FDs for request %s, received %d\n",
 118                 expected_fds,
 119                 vhost_message_str[msg->request.master],
 120                 msg->fd_num);
 121
 122         close_msg_fds(msg);
 123
 124         return -1;
 125 }
 126
 127 static uint64_t
 128 get_blk_size(int fd)
 129 {
 130         struct stat stat;
 131         int ret;
 132
 133         ret = fstat(fd, &stat);
 134         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
 135 }
 136
 137 /*
 138  * Reclaim all the outstanding zmbufs for a virtqueue.
 139  */
 140 static void
 141 drain_zmbuf_list(struct vhost_virtqueue *vq)
 142 {
 143         struct zcopy_mbuf *zmbuf, *next;
 144
 145         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 146              zmbuf != NULL; zmbuf = next) {
 147                 next = TAILQ_NEXT(zmbuf, next);
 148
 149                 while (!mbuf_is_consumed(zmbuf->mbuf))
 150                         usleep(1000);
 151
 152                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 153                 restore_mbuf(zmbuf->mbuf);
 154                 rte_pktmbuf_free(zmbuf->mbuf);
 155                 put_zmbuf(zmbuf);
 156                 vq->nr_zmbuf -= 1;
 157         }
 158 }
 159
 160 static void
 161 free_mem_region(struct virtio_net *dev)
 162 {
 163         uint32_t i;
 164         struct rte_vhost_mem_region *reg;
 165         struct vhost_virtqueue *vq;
 166
 167         if (!dev || !dev->mem)
 168                 return;
 169
 170         if (dev->dequeue_zero_copy) {
 171                 for (i = 0; i < dev->nr_vring; i++) {
 172                         vq = dev->virtqueue[i];
 173                         if (vq)
 174                                 drain_zmbuf_list(vq);
 175                 }
 176         }
 177
 178         for (i = 0; i < dev->mem->nregions; i++) {
 179                 reg = &dev->mem->regions[i];
 180                 if (reg->host_user_addr) {
 181                         munmap(reg->mmap_addr, reg->mmap_size);
 182                         close(reg->fd);
 183                 }
 184         }
 185 }
 186
 187 void
 188 vhost_backend_cleanup(struct virtio_net *dev)
 189 {
 190         if (dev->mem) {
 191                 free_mem_region(dev);
 192                 rte_free(dev->mem);
 193                 dev->mem = NULL;
 194         }
 195
 196         rte_free(dev->guest_pages);
 197         dev->guest_pages = NULL;
 198
 199         if (dev->log_addr) {
 200                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 201                 dev->log_addr = 0;
 202         }
 203
 204         if (dev->inflight_info) {
 205                 if (dev->inflight_info->addr) {
 206                         munmap(dev->inflight_info->addr,
 207                                dev->inflight_info->size);
 208                         dev->inflight_info->addr = NULL;
 209                 }
 210
 211                 if (dev->inflight_info->fd >= 0) {
 212                         close(dev->inflight_info->fd);
 213                         dev->inflight_info->fd = -1;
 214                 }
 215
 216                 free(dev->inflight_info);
 217                 dev->inflight_info = NULL;
 218         }
 219
 220         if (dev->slave_req_fd >= 0) {
 221                 close(dev->slave_req_fd);
 222                 dev->slave_req_fd = -1;
 223         }
 224
 225         if (dev->postcopy_ufd >= 0) {
 226                 close(dev->postcopy_ufd);
 227                 dev->postcopy_ufd = -1;
 228         }
 229
 230         dev->postcopy_listening = 0;
 231 }
 232
 233 static void
 234 vhost_user_notify_queue_state(struct virtio_net *dev, uint16_t index,
 235                               int enable)
 236 {
 237         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
 238         struct vhost_virtqueue *vq = dev->virtqueue[index];
 239
 240         /* Configure guest notifications on enable */
 241         if (enable && vq->notif_enable != VIRTIO_UNINITIALIZED_NOTIF)
 242                 vhost_enable_guest_notification(dev, vq, vq->notif_enable);
 243
 244         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
 245                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
 246
 247         if (dev->notify_ops->vring_state_changed)
 248                 dev->notify_ops->vring_state_changed(dev->vid,
 249                                 index, enable);
 250 }
 251
 252 /*
 253  * This function just returns success at the moment unless
 254  * the device hasn't been initialised.
 255  */
 256 static int
 257 vhost_user_set_owner(struct virtio_net **pdev __rte_unused,
 258                         struct VhostUserMsg *msg,
 259                         int main_fd __rte_unused)
 260 {
 261         if (validate_msg_fds(msg, 0) != 0)
 262                 return RTE_VHOST_MSG_RESULT_ERR;
 263
 264         return RTE_VHOST_MSG_RESULT_OK;
 265 }
 266
 267 static int
 268 vhost_user_reset_owner(struct virtio_net **pdev,
 269                         struct VhostUserMsg *msg,
 270                         int main_fd __rte_unused)
 271 {
 272         struct virtio_net *dev = *pdev;
 273
 274         if (validate_msg_fds(msg, 0) != 0)
 275                 return RTE_VHOST_MSG_RESULT_ERR;
 276
 277         vhost_destroy_device_notify(dev);
 278
 279         cleanup_device(dev, 0);
 280         reset_device(dev);
 281         return RTE_VHOST_MSG_RESULT_OK;
 282 }
 283
 284 /*
 285  * The features that we support are requested.
 286  */
 287 static int
 288 vhost_user_get_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 289                         int main_fd __rte_unused)
 290 {
 291         struct virtio_net *dev = *pdev;
 292         uint64_t features = 0;
 293
 294         if (validate_msg_fds(msg, 0) != 0)
 295                 return RTE_VHOST_MSG_RESULT_ERR;
 296
 297         rte_vhost_driver_get_features(dev->ifname, &features);
 298
 299         msg->payload.u64 = features;
 300         msg->size = sizeof(msg->payload.u64);
 301         msg->fd_num = 0;
 302
 303         return RTE_VHOST_MSG_RESULT_REPLY;
 304 }
 305
 306 /*
 307  * The queue number that we support are requested.
 308  */
 309 static int
 310 vhost_user_get_queue_num(struct virtio_net **pdev, struct VhostUserMsg *msg,
 311                         int main_fd __rte_unused)
 312 {
 313         struct virtio_net *dev = *pdev;
 314         uint32_t queue_num = 0;
 315
 316         if (validate_msg_fds(msg, 0) != 0)
 317                 return RTE_VHOST_MSG_RESULT_ERR;
 318
 319         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 320
 321         msg->payload.u64 = (uint64_t)queue_num;
 322         msg->size = sizeof(msg->payload.u64);
 323         msg->fd_num = 0;
 324
 325         return RTE_VHOST_MSG_RESULT_REPLY;
 326 }
 327
 328 /*
 329  * We receive the negotiated features supported by us and the virtio device.
 330  */
 331 static int
 332 vhost_user_set_features(struct virtio_net **pdev, struct VhostUserMsg *msg,
 333                         int main_fd __rte_unused)
 334 {
 335         struct virtio_net *dev = *pdev;
 336         uint64_t features = msg->payload.u64;
 337         uint64_t vhost_features = 0;
 338         struct rte_vdpa_device *vdpa_dev;
 339
 340         if (validate_msg_fds(msg, 0) != 0)
 341                 return RTE_VHOST_MSG_RESULT_ERR;
 342
 343         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 344         if (features & ~vhost_features) {
 345                 VHOST_LOG_CONFIG(ERR,
 346                         "(%d) received invalid negotiated features.\n",
 347                         dev->vid);
 348                 dev->flags |= VIRTIO_DEV_FEATURES_FAILED;
 349                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
 350
 351                 return RTE_VHOST_MSG_RESULT_ERR;
 352         }
 353
 354         if (dev->flags & VIRTIO_DEV_RUNNING) {
 355                 if (dev->features == features)
 356                         return RTE_VHOST_MSG_RESULT_OK;
 357
 358                 /*
 359                  * Error out if master tries to change features while device is
 360                  * in running state. The exception being VHOST_F_LOG_ALL, which
 361                  * is enabled when the live-migration starts.
 362                  */
 363                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 364                         VHOST_LOG_CONFIG(ERR,
 365                                 "(%d) features changed while device is running.\n",
 366                                 dev->vid);
 367                         return RTE_VHOST_MSG_RESULT_ERR;
 368                 }
 369
 370                 if (dev->notify_ops->features_changed)
 371                         dev->notify_ops->features_changed(dev->vid, features);
 372         }
 373
 374         dev->features = features;
 375         if (dev->features &
 376                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 377                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 378         } else {
 379                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 380         }
 381         VHOST_LOG_CONFIG(INFO,
 382                 "negotiated Virtio features: 0x%" PRIx64 "\n", dev->features);
 383         VHOST_LOG_CONFIG(DEBUG,
 384                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 385                 dev->vid,
 386                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 387                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 388
 389         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 390             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 391                 /*
 392                  * Remove all but first queue pair if MQ hasn't been
 393                  * negotiated. This is safe because the device is not
 394                  * running at this stage.
 395                  */
 396                 while (dev->nr_vring > 2) {
 397                         struct vhost_virtqueue *vq;
 398
 399                         vq = dev->virtqueue[--dev->nr_vring];
 400                         if (!vq)
 401                                 continue;
 402
 403                         dev->virtqueue[dev->nr_vring] = NULL;
 404                         cleanup_vq(vq, 1);
 405                         cleanup_vq_inflight(dev, vq);
 406                         free_vq(dev, vq);
 407                 }
 408         }
 409
 410         vdpa_dev = dev->vdpa_dev;
 411         if (vdpa_dev)
 412                 vdpa_dev->ops->set_features(dev->vid);
 413
 414         dev->flags &= ~VIRTIO_DEV_FEATURES_FAILED;
 415         return RTE_VHOST_MSG_RESULT_OK;
 416 }
 417
 418 /*
 419  * The virtio device sends us the size of the descriptor ring.
 420  */
 421 static int
 422 vhost_user_set_vring_num(struct virtio_net **pdev,
 423                         struct VhostUserMsg *msg,
 424                         int main_fd __rte_unused)
 425 {
 426         struct virtio_net *dev = *pdev;
 427         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 428
 429         if (validate_msg_fds(msg, 0) != 0)
 430                 return RTE_VHOST_MSG_RESULT_ERR;
 431
 432         vq->size = msg->payload.state.num;
 433
 434         /* VIRTIO 1.0, 2.4 Virtqueues says:
 435          *
 436          *   Queue Size value is always a power of 2. The maximum Queue Size
 437          *   value is 32768.
 438          *
 439          * VIRTIO 1.1 2.7 Virtqueues says:
 440          *
 441          *   Packed virtqueues support up to 2^15 entries each.
 442          */
 443         if (!vq_is_packed(dev)) {
 444                 if (vq->size & (vq->size - 1)) {
 445                         VHOST_LOG_CONFIG(ERR,
 446                                 "invalid virtqueue size %u\n", vq->size);
 447                         return RTE_VHOST_MSG_RESULT_ERR;
 448                 }
 449         }
 450
 451         if (vq->size > 32768) {
 452                 VHOST_LOG_CONFIG(ERR,
 453                         "invalid virtqueue size %u\n", vq->size);
 454                 return RTE_VHOST_MSG_RESULT_ERR;
 455         }
 456
 457         if (dev->dequeue_zero_copy) {
 458                 vq->nr_zmbuf = 0;
 459                 vq->last_zmbuf_idx = 0;
 460                 vq->zmbuf_size = vq->size;
 461                 if (vq->zmbufs)
 462                         rte_free(vq->zmbufs);
 463                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 464                                          sizeof(struct zcopy_mbuf), 0);
 465                 if (vq->zmbufs == NULL) {
 466                         VHOST_LOG_CONFIG(WARNING,
 467                                 "failed to allocate mem for zero copy; "
 468                                 "zero copy is force disabled\n");
 469                         dev->dequeue_zero_copy = 0;
 470                 }
 471                 TAILQ_INIT(&vq->zmbuf_list);
 472         }
 473
 474         if (vq_is_packed(dev)) {
 475                 if (vq->shadow_used_packed)
 476                         rte_free(vq->shadow_used_packed);
 477                 vq->shadow_used_packed = rte_malloc(NULL,
 478                                 vq->size *
 479                                 sizeof(struct vring_used_elem_packed),
 480                                 RTE_CACHE_LINE_SIZE);
 481                 if (!vq->shadow_used_packed) {
 482                         VHOST_LOG_CONFIG(ERR,
 483                                         "failed to allocate memory for shadow used ring.\n");
 484                         return RTE_VHOST_MSG_RESULT_ERR;
 485                 }
 486
 487         } else {
 488                 if (vq->shadow_used_split)
 489                         rte_free(vq->shadow_used_split);
 490
 491                 vq->shadow_used_split = rte_malloc(NULL,
 492                                 vq->size * sizeof(struct vring_used_elem),
 493                                 RTE_CACHE_LINE_SIZE);
 494
 495                 if (!vq->shadow_used_split) {
 496                         VHOST_LOG_CONFIG(ERR,
 497                                         "failed to allocate memory for vq internal data.\n");
 498                         return RTE_VHOST_MSG_RESULT_ERR;
 499                 }
 500         }
 501
 502         if (vq->batch_copy_elems)
 503                 rte_free(vq->batch_copy_elems);
 504         vq->batch_copy_elems = rte_malloc(NULL,
 505                                 vq->size * sizeof(struct batch_copy_elem),
 506                                 RTE_CACHE_LINE_SIZE);
 507         if (!vq->batch_copy_elems) {
 508                 VHOST_LOG_CONFIG(ERR,
 509                         "failed to allocate memory for batching copy.\n");
 510                 return RTE_VHOST_MSG_RESULT_ERR;
 511         }
 512
 513         return RTE_VHOST_MSG_RESULT_OK;
 514 }
 515
 516 /*
 517  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 518  * same numa node as the memory of vring descriptor.
 519  */
 520 #ifdef RTE_LIBRTE_VHOST_NUMA
 521 static struct virtio_net*
 522 numa_realloc(struct virtio_net *dev, int index)
 523 {
 524         int oldnode, newnode;
 525         struct virtio_net *old_dev;
 526         struct vhost_virtqueue *old_vq, *vq;
 527         struct zcopy_mbuf *new_zmbuf;
 528         struct vring_used_elem *new_shadow_used_split;
 529         struct vring_used_elem_packed *new_shadow_used_packed;
 530         struct batch_copy_elem *new_batch_copy_elems;
 531         int ret;
 532
 533         if (dev->flags & VIRTIO_DEV_RUNNING)
 534                 return dev;
 535
 536         old_dev = dev;
 537         vq = old_vq = dev->virtqueue[index];
 538
 539         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 540                             MPOL_F_NODE | MPOL_F_ADDR);
 541
 542         /* check if we need to reallocate vq */
 543         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 544                              MPOL_F_NODE | MPOL_F_ADDR);
 545         if (ret) {
 546                 VHOST_LOG_CONFIG(ERR,
 547                         "Unable to get vq numa information.\n");
 548                 return dev;
 549         }
 550         if (oldnode != newnode) {
 551                 VHOST_LOG_CONFIG(INFO,
 552                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 553                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 554                 if (!vq)
 555                         return dev;
 556
 557                 memcpy(vq, old_vq, sizeof(*vq));
 558                 TAILQ_INIT(&vq->zmbuf_list);
 559
 560                 if (dev->dequeue_zero_copy) {
 561                         new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 562                                         sizeof(struct zcopy_mbuf), 0, newnode);
 563                         if (new_zmbuf) {
 564                                 rte_free(vq->zmbufs);
 565                                 vq->zmbufs = new_zmbuf;
 566                         }
 567                 }
 568
 569                 if (vq_is_packed(dev)) {
 570                         new_shadow_used_packed = rte_malloc_socket(NULL,
 571                                         vq->size *
 572                                         sizeof(struct vring_used_elem_packed),
 573                                         RTE_CACHE_LINE_SIZE,
 574                                         newnode);
 575                         if (new_shadow_used_packed) {
 576                                 rte_free(vq->shadow_used_packed);
 577                                 vq->shadow_used_packed = new_shadow_used_packed;
 578                         }
 579                 } else {
 580                         new_shadow_used_split = rte_malloc_socket(NULL,
 581                                         vq->size *
 582                                         sizeof(struct vring_used_elem),
 583                                         RTE_CACHE_LINE_SIZE,
 584                                         newnode);
 585                         if (new_shadow_used_split) {
 586                                 rte_free(vq->shadow_used_split);
 587                                 vq->shadow_used_split = new_shadow_used_split;
 588                         }
 589                 }
 590
 591                 new_batch_copy_elems = rte_malloc_socket(NULL,
 592                         vq->size * sizeof(struct batch_copy_elem),
 593                         RTE_CACHE_LINE_SIZE,
 594                         newnode);
 595                 if (new_batch_copy_elems) {
 596                         rte_free(vq->batch_copy_elems);
 597                         vq->batch_copy_elems = new_batch_copy_elems;
 598                 }
 599
 600                 rte_free(old_vq);
 601         }
 602
 603         /* check if we need to reallocate dev */
 604         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 605                             MPOL_F_NODE | MPOL_F_ADDR);
 606         if (ret) {
 607                 VHOST_LOG_CONFIG(ERR,
 608                         "Unable to get dev numa information.\n");
 609                 goto out;
 610         }
 611         if (oldnode != newnode) {
 612                 VHOST_LOG_CONFIG(INFO,
 613                         "reallocate dev from %d to %d node\n",
 614                         oldnode, newnode);
 615                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 616                 if (!dev) {
 617                         dev = old_dev;
 618                         goto out;
 619                 }
 620
 621                 memcpy(dev, old_dev, sizeof(*dev));
 622                 rte_free(old_dev);
 623         }
 624
 625 out:
 626         dev->virtqueue[index] = vq;
 627         vhost_devices[dev->vid] = dev;
 628
 629         if (old_vq != vq)
 630                 vhost_user_iotlb_init(dev, index);
 631
 632         return dev;
 633 }
 634 #else
 635 static struct virtio_net*
 636 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 637 {
 638         return dev;
 639 }
 640 #endif
 641
 642 /* Converts QEMU virtual address to Vhost virtual address. */
 643 static uint64_t
 644 qva_to_vva(struct virtio_net *dev, uint64_t qva, uint64_t *len)
 645 {
 646         struct rte_vhost_mem_region *r;
 647         uint32_t i;
 648
 649         if (unlikely(!dev || !dev->mem))
 650                 goto out_error;
 651
 652         /* Find the region where the address lives. */
 653         for (i = 0; i < dev->mem->nregions; i++) {
 654                 r = &dev->mem->regions[i];
 655
 656                 if (qva >= r->guest_user_addr &&
 657                     qva <  r->guest_user_addr + r->size) {
 658
 659                         if (unlikely(*len > r->guest_user_addr + r->size - qva))
 660                                 *len = r->guest_user_addr + r->size - qva;
 661
 662                         return qva - r->guest_user_addr +
 663                                r->host_user_addr;
 664                 }
 665         }
 666 out_error:
 667         *len = 0;
 668
 669         return 0;
 670 }
 671
 672
 673 /*
 674  * Converts ring address to Vhost virtual address.
 675  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 676  * else it is a QEMU virtual address.
 677  */
 678 static uint64_t
 679 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 680                 uint64_t ra, uint64_t *size)
 681 {
 682         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 683                 uint64_t vva;
 684
 685                 vhost_user_iotlb_rd_lock(vq);
 686                 vva = vhost_iova_to_vva(dev, vq, ra,
 687                                         size, VHOST_ACCESS_RW);
 688                 vhost_user_iotlb_rd_unlock(vq);
 689
 690                 return vva;
 691         }
 692
 693         return qva_to_vva(dev, ra, size);
 694 }
 695
 696 static uint64_t
 697 log_addr_to_gpa(struct virtio_net *dev, struct vhost_virtqueue *vq)
 698 {
 699         uint64_t log_gpa;
 700
 701         vhost_user_iotlb_rd_lock(vq);
 702         log_gpa = translate_log_addr(dev, vq, vq->ring_addrs.log_guest_addr);
 703         vhost_user_iotlb_rd_unlock(vq);
 704
 705         return log_gpa;
 706 }
 707
 708 static struct virtio_net *
 709 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 710 {
 711         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 712         struct vhost_vring_addr *addr = &vq->ring_addrs;
 713         uint64_t len, expected_len;
 714
 715         if (addr->flags & (1 << VHOST_VRING_F_LOG)) {
 716                 vq->log_guest_addr =
 717                         log_addr_to_gpa(dev, vq);
 718                 if (vq->log_guest_addr == 0) {
 719                         VHOST_LOG_CONFIG(DEBUG,
 720                                 "(%d) failed to map log_guest_addr.\n",
 721                                 dev->vid);
 722                         return dev;
 723                 }
 724         }
 725
 726         if (vq_is_packed(dev)) {
 727                 len = sizeof(struct vring_packed_desc) * vq->size;
 728                 vq->desc_packed = (struct vring_packed_desc *)(uintptr_t)
 729                         ring_addr_to_vva(dev, vq, addr->desc_user_addr, &len);
 730                 if (vq->desc_packed == NULL ||
 731                                 len != sizeof(struct vring_packed_desc) *
 732                                 vq->size) {
 733                         VHOST_LOG_CONFIG(DEBUG,
 734                                 "(%d) failed to map desc_packed ring.\n",
 735                                 dev->vid);
 736                         return dev;
 737                 }
 738
 739                 dev = numa_realloc(dev, vq_index);
 740                 vq = dev->virtqueue[vq_index];
 741                 addr = &vq->ring_addrs;
 742
 743                 len = sizeof(struct vring_packed_desc_event);
 744                 vq->driver_event = (struct vring_packed_desc_event *)
 745                                         (uintptr_t)ring_addr_to_vva(dev,
 746                                         vq, addr->avail_user_addr, &len);
 747                 if (vq->driver_event == NULL ||
 748                                 len != sizeof(struct vring_packed_desc_event)) {
 749                         VHOST_LOG_CONFIG(DEBUG,
 750                                 "(%d) failed to find driver area address.\n",
 751                                 dev->vid);
 752                         return dev;
 753                 }
 754
 755                 len = sizeof(struct vring_packed_desc_event);
 756                 vq->device_event = (struct vring_packed_desc_event *)
 757                                         (uintptr_t)ring_addr_to_vva(dev,
 758                                         vq, addr->used_user_addr, &len);
 759                 if (vq->device_event == NULL ||
 760                                 len != sizeof(struct vring_packed_desc_event)) {
 761                         VHOST_LOG_CONFIG(DEBUG,
 762                                 "(%d) failed to find device area address.\n",
 763                                 dev->vid);
 764                         return dev;
 765                 }
 766
 767                 vq->access_ok = 1;
 768                 return dev;
 769         }
 770
 771         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 772         if (vq->desc && vq->avail && vq->used)
 773                 return dev;
 774
 775         len = sizeof(struct vring_desc) * vq->size;
 776         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 777                         vq, addr->desc_user_addr, &len);
 778         if (vq->desc == 0 || len != sizeof(struct vring_desc) * vq->size) {
 779                 VHOST_LOG_CONFIG(DEBUG,
 780                         "(%d) failed to map desc ring.\n",
 781                         dev->vid);
 782                 return dev;
 783         }
 784
 785         dev = numa_realloc(dev, vq_index);
 786         vq = dev->virtqueue[vq_index];
 787         addr = &vq->ring_addrs;
 788
 789         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
 790         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 791                 len += sizeof(uint16_t);
 792         expected_len = len;
 793         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 794                         vq, addr->avail_user_addr, &len);
 795         if (vq->avail == 0 || len != expected_len) {
 796                 VHOST_LOG_CONFIG(DEBUG,
 797                         "(%d) failed to map avail ring.\n",
 798                         dev->vid);
 799                 return dev;
 800         }
 801
 802         len = sizeof(struct vring_used) +
 803                 sizeof(struct vring_used_elem) * vq->size;
 804         if (dev->features & (1ULL << VIRTIO_RING_F_EVENT_IDX))
 805                 len += sizeof(uint16_t);
 806         expected_len = len;
 807         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 808                         vq, addr->used_user_addr, &len);
 809         if (vq->used == 0 || len != expected_len) {
 810                 VHOST_LOG_CONFIG(DEBUG,
 811                         "(%d) failed to map used ring.\n",
 812                         dev->vid);
 813                 return dev;
 814         }
 815
 816         if (vq->last_used_idx != vq->used->idx) {
 817                 VHOST_LOG_CONFIG(WARNING,
 818                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 819                         "some packets maybe resent for Tx and dropped for Rx\n",
 820                         vq->last_used_idx, vq->used->idx);
 821                 vq->last_used_idx  = vq->used->idx;
 822                 vq->last_avail_idx = vq->used->idx;
 823         }
 824
 825         vq->access_ok = 1;
 826
 827         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address desc: %p\n",
 828                         dev->vid, vq->desc);
 829         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address avail: %p\n",
 830                         dev->vid, vq->avail);
 831         VHOST_LOG_CONFIG(DEBUG, "(%d) mapped address used: %p\n",
 832                         dev->vid, vq->used);
 833         VHOST_LOG_CONFIG(DEBUG, "(%d) log_guest_addr: %" PRIx64 "\n",
 834                         dev->vid, vq->log_guest_addr);
 835
 836         return dev;
 837 }
 838
 839 /*
 840  * The virtio device sends us the desc, used and avail ring addresses.
 841  * This function then converts these to our address space.
 842  */
 843 static int
 844 vhost_user_set_vring_addr(struct virtio_net **pdev, struct VhostUserMsg *msg,
 845                         int main_fd __rte_unused)
 846 {
 847         struct virtio_net *dev = *pdev;
 848         struct vhost_virtqueue *vq;
 849         struct vhost_vring_addr *addr = &msg->payload.addr;
 850         bool access_ok;
 851
 852         if (validate_msg_fds(msg, 0) != 0)
 853                 return RTE_VHOST_MSG_RESULT_ERR;
 854
 855         if (dev->mem == NULL)
 856                 return RTE_VHOST_MSG_RESULT_ERR;
 857
 858         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 859         vq = dev->virtqueue[msg->payload.addr.index];
 860
 861         access_ok = vq->access_ok;
 862
 863         /*
 864          * Rings addresses should not be interpreted as long as the ring is not
 865          * started and enabled
 866          */
 867         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 868
 869         vring_invalidate(dev, vq);
 870
 871         if ((vq->enabled && (dev->features &
 872                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) ||
 873                         access_ok) {
 874                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 875                 if (!dev)
 876                         return RTE_VHOST_MSG_RESULT_ERR;
 877
 878                 *pdev = dev;
 879         }
 880
 881         return RTE_VHOST_MSG_RESULT_OK;
 882 }
 883
 884 /*
 885  * The virtio device sends us the available ring last used index.
 886  */
 887 static int
 888 vhost_user_set_vring_base(struct virtio_net **pdev,
 889                         struct VhostUserMsg *msg,
 890                         int main_fd __rte_unused)
 891 {
 892         struct virtio_net *dev = *pdev;
 893         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 894         uint64_t val = msg->payload.state.num;
 895
 896         if (validate_msg_fds(msg, 0) != 0)
 897                 return RTE_VHOST_MSG_RESULT_ERR;
 898
 899         if (vq_is_packed(dev)) {
 900                 /*
 901                  * Bit[0:14]: avail index
 902                  * Bit[15]: avail wrap counter
 903                  */
 904                 vq->last_avail_idx = val & 0x7fff;
 905                 vq->avail_wrap_counter = !!(val & (0x1 << 15));
 906                 /*
 907                  * Set used index to same value as available one, as
 908                  * their values should be the same since ring processing
 909                  * was stopped at get time.
 910                  */
 911                 vq->last_used_idx = vq->last_avail_idx;
 912                 vq->used_wrap_counter = vq->avail_wrap_counter;
 913         } else {
 914                 vq->last_used_idx = msg->payload.state.num;
 915                 vq->last_avail_idx = msg->payload.state.num;
 916         }
 917
 918         return RTE_VHOST_MSG_RESULT_OK;
 919 }
 920
 921 static int
 922 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 923                    uint64_t host_phys_addr, uint64_t size)
 924 {
 925         struct guest_page *page, *last_page;
 926         struct guest_page *old_pages;
 927
 928         if (dev->nr_guest_pages == dev->max_guest_pages) {
 929                 dev->max_guest_pages *= 2;
 930                 old_pages = dev->guest_pages;
 931                 dev->guest_pages = rte_realloc(dev->guest_pages,
 932                                         dev->max_guest_pages * sizeof(*page),
 933                                         RTE_CACHE_LINE_SIZE);
 934                 if (dev->guest_pages == NULL) {
 935                         VHOST_LOG_CONFIG(ERR, "cannot realloc guest_pages\n");
 936                         rte_free(old_pages);
 937                         return -1;
 938                 }
 939         }
 940
 941         if (dev->nr_guest_pages > 0) {
 942                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 943                 /* merge if the two pages are continuous */
 944                 if (host_phys_addr == last_page->host_phys_addr +
 945                                       last_page->size) {
 946                         last_page->size += size;
 947                         return 0;
 948                 }
 949         }
 950
 951         page = &dev->guest_pages[dev->nr_guest_pages++];
 952         page->guest_phys_addr = guest_phys_addr;
 953         page->host_phys_addr  = host_phys_addr;
 954         page->size = size;
 955
 956         return 0;
 957 }
 958
 959 static int
 960 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 961                 uint64_t page_size)
 962 {
 963         uint64_t reg_size = reg->size;
 964         uint64_t host_user_addr  = reg->host_user_addr;
 965         uint64_t guest_phys_addr = reg->guest_phys_addr;
 966         uint64_t host_phys_addr;
 967         uint64_t size;
 968
 969         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 970         size = page_size - (guest_phys_addr & (page_size - 1));
 971         size = RTE_MIN(size, reg_size);
 972
 973         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 974                 return -1;
 975
 976         host_user_addr  += size;
 977         guest_phys_addr += size;
 978         reg_size -= size;
 979
 980         while (reg_size > 0) {
 981                 size = RTE_MIN(reg_size, page_size);
 982                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 983                                                   host_user_addr);
 984                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 985                                 size) < 0)
 986                         return -1;
 987
 988                 host_user_addr  += size;
 989                 guest_phys_addr += size;
 990                 reg_size -= size;
 991         }
 992
 993         /* sort guest page array if over binary search threshold */
 994         if (dev->nr_guest_pages >= VHOST_BINARY_SEARCH_THRESH) {
 995                 qsort((void *)dev->guest_pages, dev->nr_guest_pages,
 996                         sizeof(struct guest_page), guest_page_addrcmp);
 997         }
 998
 999         return 0;
1000 }
1001
1002 #ifdef RTE_LIBRTE_VHOST_DEBUG
1003 /* TODO: enable it only in debug mode? */
1004 static void
1005 dump_guest_pages(struct virtio_net *dev)
1006 {
1007         uint32_t i;
1008         struct guest_page *page;
1009
1010         for (i = 0; i < dev->nr_guest_pages; i++) {
1011                 page = &dev->guest_pages[i];
1012
1013                 VHOST_LOG_CONFIG(INFO,
1014                         "guest physical page region %u\n"
1015                         "\t guest_phys_addr: %" PRIx64 "\n"
1016                         "\t host_phys_addr : %" PRIx64 "\n"
1017                         "\t size           : %" PRIx64 "\n",
1018                         i,
1019                         page->guest_phys_addr,
1020                         page->host_phys_addr,
1021                         page->size);
1022         }
1023 }
1024 #else
1025 #define dump_guest_pages(dev)
1026 #endif
1027
1028 static bool
1029 vhost_memory_changed(struct VhostUserMemory *new,
1030                      struct rte_vhost_memory *old)
1031 {
1032         uint32_t i;
1033
1034         if (new->nregions != old->nregions)
1035                 return true;
1036
1037         for (i = 0; i < new->nregions; ++i) {
1038                 VhostUserMemoryRegion *new_r = &new->regions[i];
1039                 struct rte_vhost_mem_region *old_r = &old->regions[i];
1040
1041                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
1042                         return true;
1043                 if (new_r->memory_size != old_r->size)
1044                         return true;
1045                 if (new_r->userspace_addr != old_r->guest_user_addr)
1046                         return true;
1047         }
1048
1049         return false;
1050 }
1051
1052 static int
1053 vhost_user_set_mem_table(struct virtio_net **pdev, struct VhostUserMsg *msg,
1054                         int main_fd)
1055 {
1056         struct virtio_net *dev = *pdev;
1057         struct VhostUserMemory *memory = &msg->payload.memory;
1058         struct rte_vhost_mem_region *reg;
1059         void *mmap_addr;
1060         uint64_t mmap_size;
1061         uint64_t mmap_offset;
1062         uint64_t alignment;
1063         uint32_t i;
1064         int populate;
1065         int fd;
1066
1067         if (validate_msg_fds(msg, memory->nregions) != 0)
1068                 return RTE_VHOST_MSG_RESULT_ERR;
1069
1070         if (memory->nregions > VHOST_MEMORY_MAX_NREGIONS) {
1071                 VHOST_LOG_CONFIG(ERR,
1072                         "too many memory regions (%u)\n", memory->nregions);
1073                 return RTE_VHOST_MSG_RESULT_ERR;
1074         }
1075
1076         if (dev->mem && !vhost_memory_changed(memory, dev->mem)) {
1077                 VHOST_LOG_CONFIG(INFO,
1078                         "(%d) memory regions not changed\n", dev->vid);
1079
1080                 close_msg_fds(msg);
1081
1082                 return RTE_VHOST_MSG_RESULT_OK;
1083         }
1084
1085         if (dev->mem) {
1086                 if (dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) {
1087                         struct rte_vdpa_device *vdpa_dev = dev->vdpa_dev;
1088
1089                         if (vdpa_dev && vdpa_dev->ops->dev_close)
1090                                 vdpa_dev->ops->dev_close(dev->vid);
1091                         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1092                 }
1093                 free_mem_region(dev);
1094                 rte_free(dev->mem);
1095                 dev->mem = NULL;
1096         }
1097
1098         /* Flush IOTLB cache as previous HVAs are now invalid */
1099         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM))
1100                 for (i = 0; i < dev->nr_vring; i++)
1101                         vhost_user_iotlb_flush_all(dev->virtqueue[i]);
1102
1103         dev->nr_guest_pages = 0;
1104         if (dev->guest_pages == NULL) {
1105                 dev->max_guest_pages = 8;
1106                 dev->guest_pages = rte_zmalloc(NULL,
1107                                         dev->max_guest_pages *
1108                                         sizeof(struct guest_page),
1109                                         RTE_CACHE_LINE_SIZE);
1110                 if (dev->guest_pages == NULL) {
1111                         VHOST_LOG_CONFIG(ERR,
1112                                 "(%d) failed to allocate memory "
1113                                 "for dev->guest_pages\n",
1114                                 dev->vid);
1115                         return RTE_VHOST_MSG_RESULT_ERR;
1116                 }
1117         }
1118
1119         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
1120                 sizeof(struct rte_vhost_mem_region) * memory->nregions, 0);
1121         if (dev->mem == NULL) {
1122                 VHOST_LOG_CONFIG(ERR,
1123                         "(%d) failed to allocate memory for dev->mem\n",
1124                         dev->vid);
1125                 return RTE_VHOST_MSG_RESULT_ERR;
1126         }
1127         dev->mem->nregions = memory->nregions;
1128
1129         for (i = 0; i < memory->nregions; i++) {
1130                 fd  = msg->fds[i];
1131                 reg = &dev->mem->regions[i];
1132
1133                 reg->guest_phys_addr = memory->regions[i].guest_phys_addr;
1134                 reg->guest_user_addr = memory->regions[i].userspace_addr;
1135                 reg->size            = memory->regions[i].memory_size;
1136                 reg->fd              = fd;
1137
1138                 mmap_offset = memory->regions[i].mmap_offset;
1139
1140                 /* Check for memory_size + mmap_offset overflow */
1141                 if (mmap_offset >= -reg->size) {
1142                         VHOST_LOG_CONFIG(ERR,
1143                                 "mmap_offset (%#"PRIx64") and memory_size "
1144                                 "(%#"PRIx64") overflow\n",
1145                                 mmap_offset, reg->size);
1146                         goto err_mmap;
1147                 }
1148
1149                 mmap_size = reg->size + mmap_offset;
1150
1151                 /* mmap() without flag of MAP_ANONYMOUS, should be called
1152                  * with length argument aligned with hugepagesz at older
1153                  * longterm version Linux, like 2.6.32 and 3.2.72, or
1154                  * mmap() will fail with EINVAL.
1155                  *
1156                  * to avoid failure, make sure in caller to keep length
1157                  * aligned.
1158                  */
1159                 alignment = get_blk_size(fd);
1160                 if (alignment == (uint64_t)-1) {
1161                         VHOST_LOG_CONFIG(ERR,
1162                                 "couldn't get hugepage size through fstat\n");
1163                         goto err_mmap;
1164                 }
1165                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
1166                 if (mmap_size == 0) {
1167                         /*
1168                          * It could happen if initial mmap_size + alignment
1169                          * overflows the sizeof uint64, which could happen if
1170                          * either mmap_size or alignment value is wrong.
1171                          *
1172                          * mmap() kernel implementation would return an error,
1173                          * but better catch it before and provide useful info
1174                          * in the logs.
1175                          */
1176                         VHOST_LOG_CONFIG(ERR, "mmap size (0x%" PRIx64 ") "
1177                                         "or alignment (0x%" PRIx64 ") is invalid\n",
1178                                         reg->size + mmap_offset, alignment);
1179                         goto err_mmap;
1180                 }
1181
1182                 populate = (dev->dequeue_zero_copy || dev->async_copy) ?
1183                         MAP_POPULATE : 0;
1184                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
1185                                  MAP_SHARED | populate, fd, 0);
1186
1187                 if (mmap_addr == MAP_FAILED) {
1188                         VHOST_LOG_CONFIG(ERR,
1189                                 "mmap region %u failed.\n", i);
1190                         goto err_mmap;
1191                 }
1192
1193                 reg->mmap_addr = mmap_addr;
1194                 reg->mmap_size = mmap_size;
1195                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
1196                                       mmap_offset;
1197
1198                 if (dev->dequeue_zero_copy || dev->async_copy)
1199                         if (add_guest_pages(dev, reg, alignment) < 0) {
1200                                 VHOST_LOG_CONFIG(ERR,
1201                                         "adding guest pages to region %u failed.\n",
1202                                         i);
1203                                 goto err_mmap;
1204                         }
1205
1206                 VHOST_LOG_CONFIG(INFO,
1207                         "guest memory region %u, size: 0x%" PRIx64 "\n"
1208                         "\t guest physical addr: 0x%" PRIx64 "\n"
1209                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
1210                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
1211                         "\t mmap addr : 0x%" PRIx64 "\n"
1212                         "\t mmap size : 0x%" PRIx64 "\n"
1213                         "\t mmap align: 0x%" PRIx64 "\n"
1214                         "\t mmap off  : 0x%" PRIx64 "\n",
1215                         i, reg->size,
1216                         reg->guest_phys_addr,
1217                         reg->guest_user_addr,
1218                         reg->host_user_addr,
1219                         (uint64_t)(uintptr_t)mmap_addr,
1220                         mmap_size,
1221                         alignment,
1222                         mmap_offset);
1223
1224                 if (dev->postcopy_listening) {
1225                         /*
1226                          * We haven't a better way right now than sharing
1227                          * DPDK's virtual address with Qemu, so that Qemu can
1228                          * retrieve the region offset when handling userfaults.
1229                          */
1230                         memory->regions[i].userspace_addr =
1231                                 reg->host_user_addr;
1232                 }
1233         }
1234         if (dev->postcopy_listening) {
1235                 /* Send the addresses back to qemu */
1236                 msg->fd_num = 0;
1237                 send_vhost_reply(main_fd, msg);
1238
1239                 /* Wait for qemu to acknolwedge it's got the addresses
1240                  * we've got to wait before we're allowed to generate faults.
1241                  */
1242                 VhostUserMsg ack_msg;
1243                 if (read_vhost_message(main_fd, &ack_msg) <= 0) {
1244                         VHOST_LOG_CONFIG(ERR,
1245                                 "Failed to read qemu ack on postcopy set-mem-table\n");
1246                         goto err_mmap;
1247                 }
1248
1249                 if (validate_msg_fds(&ack_msg, 0) != 0)
1250                         goto err_mmap;
1251
1252                 if (ack_msg.request.master != VHOST_USER_SET_MEM_TABLE) {
1253                         VHOST_LOG_CONFIG(ERR,
1254                                 "Bad qemu ack on postcopy set-mem-table (%d)\n",
1255                                 ack_msg.request.master);
1256                         goto err_mmap;
1257                 }
1258
1259                 /* Now userfault register and we can use the memory */
1260                 for (i = 0; i < memory->nregions; i++) {
1261 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
1262                         reg = &dev->mem->regions[i];
1263                         struct uffdio_register reg_struct;
1264
1265                         /*
1266                          * Let's register all the mmap'ed area to ensure
1267                          * alignment on page boundary.
1268                          */
1269                         reg_struct.range.start =
1270                                 (uint64_t)(uintptr_t)reg->mmap_addr;
1271                         reg_struct.range.len = reg->mmap_size;
1272                         reg_struct.mode = UFFDIO_REGISTER_MODE_MISSING;
1273
1274                         if (ioctl(dev->postcopy_ufd, UFFDIO_REGISTER,
1275                                                 &reg_struct)) {
1276                                 VHOST_LOG_CONFIG(ERR,
1277                                         "Failed to register ufd for region %d: (ufd = %d) %s\n",
1278                                         i, dev->postcopy_ufd,
1279                                         strerror(errno));
1280                                 goto err_mmap;
1281                         }
1282                         VHOST_LOG_CONFIG(INFO,
1283                                 "\t userfaultfd registered for range : "
1284                                 "%" PRIx64 " - %" PRIx64 "\n",
1285                                 (uint64_t)reg_struct.range.start,
1286                                 (uint64_t)reg_struct.range.start +
1287                                 (uint64_t)reg_struct.range.len - 1);
1288 #else
1289                         goto err_mmap;
1290 #endif
1291                 }
1292         }
1293
1294         for (i = 0; i < dev->nr_vring; i++) {
1295                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1296
1297                 if (vq->desc || vq->avail || vq->used) {
1298                         /*
1299                          * If the memory table got updated, the ring addresses
1300                          * need to be translated again as virtual addresses have
1301                          * changed.
1302                          */
1303                         vring_invalidate(dev, vq);
1304
1305                         dev = translate_ring_addresses(dev, i);
1306                         if (!dev) {
1307                                 dev = *pdev;
1308                                 goto err_mmap;
1309                         }
1310
1311                         *pdev = dev;
1312                 }
1313         }
1314
1315         dump_guest_pages(dev);
1316
1317         return RTE_VHOST_MSG_RESULT_OK;
1318
1319 err_mmap:
1320         free_mem_region(dev);
1321         rte_free(dev->mem);
1322         dev->mem = NULL;
1323         return RTE_VHOST_MSG_RESULT_ERR;
1324 }
1325
1326 static bool
1327 vq_is_ready(struct virtio_net *dev, struct vhost_virtqueue *vq)
1328 {
1329         bool rings_ok;
1330
1331         if (!vq)
1332                 return false;
1333
1334         if (vq_is_packed(dev))
1335                 rings_ok = vq->desc_packed && vq->driver_event &&
1336                         vq->device_event;
1337         else
1338                 rings_ok = vq->desc && vq->avail && vq->used;
1339
1340         return rings_ok &&
1341                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1342                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD &&
1343                vq->enabled;
1344 }
1345
1346 #define VIRTIO_DEV_NUM_VQS_TO_BE_READY 2u
1347
1348 static int
1349 virtio_is_ready(struct virtio_net *dev)
1350 {
1351         struct vhost_virtqueue *vq;
1352         uint32_t i;
1353
1354         if (dev->flags & VIRTIO_DEV_READY)
1355                 return 1;
1356
1357         if (dev->nr_vring < VIRTIO_DEV_NUM_VQS_TO_BE_READY)
1358                 return 0;
1359
1360         for (i = 0; i < VIRTIO_DEV_NUM_VQS_TO_BE_READY; i++) {
1361                 vq = dev->virtqueue[i];
1362
1363                 if (!vq_is_ready(dev, vq))
1364                         return 0;
1365         }
1366
1367         /* If supported, ensure the frontend is really done with config */
1368         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_STATUS))
1369                 if (!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK))
1370                         return 0;
1371
1372         dev->flags |= VIRTIO_DEV_READY;
1373
1374         if (!(dev->flags & VIRTIO_DEV_RUNNING))
1375                 VHOST_LOG_CONFIG(INFO,
1376                         "virtio is now ready for processing.\n");
1377         return 1;
1378 }
1379
1380 static void *
1381 inflight_mem_alloc(const char *name, size_t size, int *fd)
1382 {
1383         void *ptr;
1384         int mfd = -1;
1385         char fname[20] = "/tmp/memfd-XXXXXX";
1386
1387         *fd = -1;
1388 #ifdef MEMFD_SUPPORTED
1389         mfd = memfd_create(name, MFD_CLOEXEC);
1390 #else
1391         RTE_SET_USED(name);
1392 #endif
1393         if (mfd == -1) {
1394                 mfd = mkstemp(fname);
1395                 if (mfd == -1) {
1396                         VHOST_LOG_CONFIG(ERR,
1397                                 "failed to get inflight buffer fd\n");
1398                         return NULL;
1399                 }
1400
1401                 unlink(fname);
1402         }
1403
1404         if (ftruncate(mfd, size) == -1) {
1405                 VHOST_LOG_CONFIG(ERR,
1406                         "failed to alloc inflight buffer\n");
1407                 close(mfd);
1408                 return NULL;
1409         }
1410
1411         ptr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, mfd, 0);
1412         if (ptr == MAP_FAILED) {
1413                 VHOST_LOG_CONFIG(ERR,
1414                         "failed to mmap inflight buffer\n");
1415                 close(mfd);
1416                 return NULL;
1417         }
1418
1419         *fd = mfd;
1420         return ptr;
1421 }
1422
1423 static uint32_t
1424 get_pervq_shm_size_split(uint16_t queue_size)
1425 {
1426         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_split) *
1427                                   queue_size + sizeof(uint64_t) +
1428                                   sizeof(uint16_t) * 4, INFLIGHT_ALIGNMENT);
1429 }
1430
1431 static uint32_t
1432 get_pervq_shm_size_packed(uint16_t queue_size)
1433 {
1434         return RTE_ALIGN_MUL_CEIL(sizeof(struct rte_vhost_inflight_desc_packed)
1435                                   * queue_size + sizeof(uint64_t) +
1436                                   sizeof(uint16_t) * 6 + sizeof(uint8_t) * 9,
1437                                   INFLIGHT_ALIGNMENT);
1438 }
1439
1440 static int
1441 vhost_user_get_inflight_fd(struct virtio_net **pdev,
1442                            VhostUserMsg *msg,
1443                            int main_fd __rte_unused)
1444 {
1445         struct rte_vhost_inflight_info_packed *inflight_packed;
1446         uint64_t pervq_inflight_size, mmap_size;
1447         uint16_t num_queues, queue_size;
1448         struct virtio_net *dev = *pdev;
1449         int fd, i, j;
1450         void *addr;
1451
1452         if (msg->size != sizeof(msg->payload.inflight)) {
1453                 VHOST_LOG_CONFIG(ERR,
1454                         "invalid get_inflight_fd message size is %d\n",
1455                         msg->size);
1456                 return RTE_VHOST_MSG_RESULT_ERR;
1457         }
1458
1459         if (dev->inflight_info == NULL) {
1460                 dev->inflight_info = calloc(1,
1461                                             sizeof(struct inflight_mem_info));
1462                 if (!dev->inflight_info) {
1463                         VHOST_LOG_CONFIG(ERR,
1464                                 "failed to alloc dev inflight area\n");
1465                         return RTE_VHOST_MSG_RESULT_ERR;
1466                 }
1467                 dev->inflight_info->fd = -1;
1468         }
1469
1470         num_queues = msg->payload.inflight.num_queues;
1471         queue_size = msg->payload.inflight.queue_size;
1472
1473         VHOST_LOG_CONFIG(INFO, "get_inflight_fd num_queues: %u\n",
1474                 msg->payload.inflight.num_queues);
1475         VHOST_LOG_CONFIG(INFO, "get_inflight_fd queue_size: %u\n",
1476                 msg->payload.inflight.queue_size);
1477
1478         if (vq_is_packed(dev))
1479                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1480         else
1481                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1482
1483         mmap_size = num_queues * pervq_inflight_size;
1484         addr = inflight_mem_alloc("vhost-inflight", mmap_size, &fd);
1485         if (!addr) {
1486                 VHOST_LOG_CONFIG(ERR,
1487                         "failed to alloc vhost inflight area\n");
1488                         msg->payload.inflight.mmap_size = 0;
1489                 return RTE_VHOST_MSG_RESULT_ERR;
1490         }
1491         memset(addr, 0, mmap_size);
1492
1493         if (dev->inflight_info->addr) {
1494                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1495                 dev->inflight_info->addr = NULL;
1496         }
1497
1498         if (dev->inflight_info->fd >= 0) {
1499                 close(dev->inflight_info->fd);
1500                 dev->inflight_info->fd = -1;
1501         }
1502
1503         dev->inflight_info->addr = addr;
1504         dev->inflight_info->size = msg->payload.inflight.mmap_size = mmap_size;
1505         dev->inflight_info->fd = msg->fds[0] = fd;
1506         msg->payload.inflight.mmap_offset = 0;
1507         msg->fd_num = 1;
1508
1509         if (vq_is_packed(dev)) {
1510                 for (i = 0; i < num_queues; i++) {
1511                         inflight_packed =
1512                                 (struct rte_vhost_inflight_info_packed *)addr;
1513                         inflight_packed->used_wrap_counter = 1;
1514                         inflight_packed->old_used_wrap_counter = 1;
1515                         for (j = 0; j < queue_size; j++)
1516                                 inflight_packed->desc[j].next = j + 1;
1517                         addr = (void *)((char *)addr + pervq_inflight_size);
1518                 }
1519         }
1520
1521         VHOST_LOG_CONFIG(INFO,
1522                 "send inflight mmap_size: %"PRIu64"\n",
1523                 msg->payload.inflight.mmap_size);
1524         VHOST_LOG_CONFIG(INFO,
1525                 "send inflight mmap_offset: %"PRIu64"\n",
1526                 msg->payload.inflight.mmap_offset);
1527         VHOST_LOG_CONFIG(INFO,
1528                 "send inflight fd: %d\n", msg->fds[0]);
1529
1530         return RTE_VHOST_MSG_RESULT_REPLY;
1531 }
1532
1533 static int
1534 vhost_user_set_inflight_fd(struct virtio_net **pdev, VhostUserMsg *msg,
1535                            int main_fd __rte_unused)
1536 {
1537         uint64_t mmap_size, mmap_offset;
1538         uint16_t num_queues, queue_size;
1539         struct virtio_net *dev = *pdev;
1540         uint32_t pervq_inflight_size;
1541         struct vhost_virtqueue *vq;
1542         void *addr;
1543         int fd, i;
1544
1545         fd = msg->fds[0];
1546         if (msg->size != sizeof(msg->payload.inflight) || fd < 0) {
1547                 VHOST_LOG_CONFIG(ERR,
1548                         "invalid set_inflight_fd message size is %d,fd is %d\n",
1549                         msg->size, fd);
1550                 return RTE_VHOST_MSG_RESULT_ERR;
1551         }
1552
1553         mmap_size = msg->payload.inflight.mmap_size;
1554         mmap_offset = msg->payload.inflight.mmap_offset;
1555         num_queues = msg->payload.inflight.num_queues;
1556         queue_size = msg->payload.inflight.queue_size;
1557
1558         if (vq_is_packed(dev))
1559                 pervq_inflight_size = get_pervq_shm_size_packed(queue_size);
1560         else
1561                 pervq_inflight_size = get_pervq_shm_size_split(queue_size);
1562
1563         VHOST_LOG_CONFIG(INFO,
1564                 "set_inflight_fd mmap_size: %"PRIu64"\n", mmap_size);
1565         VHOST_LOG_CONFIG(INFO,
1566                 "set_inflight_fd mmap_offset: %"PRIu64"\n", mmap_offset);
1567         VHOST_LOG_CONFIG(INFO,
1568                 "set_inflight_fd num_queues: %u\n", num_queues);
1569         VHOST_LOG_CONFIG(INFO,
1570                 "set_inflight_fd queue_size: %u\n", queue_size);
1571         VHOST_LOG_CONFIG(INFO,
1572                 "set_inflight_fd fd: %d\n", fd);
1573         VHOST_LOG_CONFIG(INFO,
1574                 "set_inflight_fd pervq_inflight_size: %d\n",
1575                 pervq_inflight_size);
1576
1577         if (!dev->inflight_info) {
1578                 dev->inflight_info = calloc(1,
1579                                             sizeof(struct inflight_mem_info));
1580                 if (dev->inflight_info == NULL) {
1581                         VHOST_LOG_CONFIG(ERR,
1582                                 "failed to alloc dev inflight area\n");
1583                         return RTE_VHOST_MSG_RESULT_ERR;
1584                 }
1585                 dev->inflight_info->fd = -1;
1586         }
1587
1588         if (dev->inflight_info->addr) {
1589                 munmap(dev->inflight_info->addr, dev->inflight_info->size);
1590                 dev->inflight_info->addr = NULL;
1591         }
1592
1593         addr = mmap(0, mmap_size, PROT_READ | PROT_WRITE, MAP_SHARED,
1594                     fd, mmap_offset);
1595         if (addr == MAP_FAILED) {
1596                 VHOST_LOG_CONFIG(ERR, "failed to mmap share memory.\n");
1597                 return RTE_VHOST_MSG_RESULT_ERR;
1598         }
1599
1600         if (dev->inflight_info->fd >= 0) {
1601                 close(dev->inflight_info->fd);
1602                 dev->inflight_info->fd = -1;
1603         }
1604
1605         dev->inflight_info->fd = fd;
1606         dev->inflight_info->addr = addr;
1607         dev->inflight_info->size = mmap_size;
1608
1609         for (i = 0; i < num_queues; i++) {
1610                 vq = dev->virtqueue[i];
1611                 if (vq_is_packed(dev)) {
1612                         vq->inflight_packed = addr;
1613                         vq->inflight_packed->desc_num = queue_size;
1614                 } else {
1615                         vq->inflight_split = addr;
1616                         vq->inflight_split->desc_num = queue_size;
1617                 }
1618                 addr = (void *)((char *)addr + pervq_inflight_size);
1619         }
1620
1621         return RTE_VHOST_MSG_RESULT_OK;
1622 }
1623
1624 static int
1625 vhost_user_set_vring_call(struct virtio_net **pdev, struct VhostUserMsg *msg,
1626                         int main_fd __rte_unused)
1627 {
1628         struct virtio_net *dev = *pdev;
1629         struct vhost_vring_file file;
1630         struct vhost_virtqueue *vq;
1631         int expected_fds;
1632
1633         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1634         if (validate_msg_fds(msg, expected_fds) != 0)
1635                 return RTE_VHOST_MSG_RESULT_ERR;
1636
1637         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1638         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1639                 file.fd = VIRTIO_INVALID_EVENTFD;
1640         else
1641                 file.fd = msg->fds[0];
1642         VHOST_LOG_CONFIG(INFO,
1643                 "vring call idx:%d file:%d\n", file.index, file.fd);
1644
1645         vq = dev->virtqueue[file.index];
1646
1647         if (vq->ready) {
1648                 vq->ready = 0;
1649                 vhost_user_notify_queue_state(dev, file.index, 0);
1650         }
1651
1652         if (vq->callfd >= 0)
1653                 close(vq->callfd);
1654
1655         vq->callfd = file.fd;
1656
1657         return RTE_VHOST_MSG_RESULT_OK;
1658 }
1659
1660 static int vhost_user_set_vring_err(struct virtio_net **pdev __rte_unused,
1661                         struct VhostUserMsg *msg,
1662                         int main_fd __rte_unused)
1663 {
1664         int expected_fds;
1665
1666         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1667         if (validate_msg_fds(msg, expected_fds) != 0)
1668                 return RTE_VHOST_MSG_RESULT_ERR;
1669
1670         if (!(msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1671                 close(msg->fds[0]);
1672         VHOST_LOG_CONFIG(INFO, "not implemented\n");
1673
1674         return RTE_VHOST_MSG_RESULT_OK;
1675 }
1676
1677 static int
1678 resubmit_desc_compare(const void *a, const void *b)
1679 {
1680         const struct rte_vhost_resubmit_desc *desc0 = a;
1681         const struct rte_vhost_resubmit_desc *desc1 = b;
1682
1683         if (desc1->counter > desc0->counter)
1684                 return 1;
1685
1686         return -1;
1687 }
1688
1689 static int
1690 vhost_check_queue_inflights_split(struct virtio_net *dev,
1691                                   struct vhost_virtqueue *vq)
1692 {
1693         uint16_t i;
1694         uint16_t resubmit_num = 0, last_io, num;
1695         struct vring_used *used = vq->used;
1696         struct rte_vhost_resubmit_info *resubmit;
1697         struct rte_vhost_inflight_info_split *inflight_split;
1698
1699         if (!(dev->protocol_features &
1700             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1701                 return RTE_VHOST_MSG_RESULT_OK;
1702
1703         /* The frontend may still not support the inflight feature
1704          * although we negotiate the protocol feature.
1705          */
1706         if ((!vq->inflight_split))
1707                 return RTE_VHOST_MSG_RESULT_OK;
1708
1709         if (!vq->inflight_split->version) {
1710                 vq->inflight_split->version = INFLIGHT_VERSION;
1711                 return RTE_VHOST_MSG_RESULT_OK;
1712         }
1713
1714         if (vq->resubmit_inflight)
1715                 return RTE_VHOST_MSG_RESULT_OK;
1716
1717         inflight_split = vq->inflight_split;
1718         vq->global_counter = 0;
1719         last_io = inflight_split->last_inflight_io;
1720
1721         if (inflight_split->used_idx != used->idx) {
1722                 inflight_split->desc[last_io].inflight = 0;
1723                 rte_smp_mb();
1724                 inflight_split->used_idx = used->idx;
1725         }
1726
1727         for (i = 0; i < inflight_split->desc_num; i++) {
1728                 if (inflight_split->desc[i].inflight == 1)
1729                         resubmit_num++;
1730         }
1731
1732         vq->last_avail_idx += resubmit_num;
1733
1734         if (resubmit_num) {
1735                 resubmit  = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1736                 if (!resubmit) {
1737                         VHOST_LOG_CONFIG(ERR,
1738                                 "failed to allocate memory for resubmit info.\n");
1739                         return RTE_VHOST_MSG_RESULT_ERR;
1740                 }
1741
1742                 resubmit->resubmit_list = calloc(resubmit_num,
1743                         sizeof(struct rte_vhost_resubmit_desc));
1744                 if (!resubmit->resubmit_list) {
1745                         VHOST_LOG_CONFIG(ERR,
1746                                 "failed to allocate memory for inflight desc.\n");
1747                         free(resubmit);
1748                         return RTE_VHOST_MSG_RESULT_ERR;
1749                 }
1750
1751                 num = 0;
1752                 for (i = 0; i < vq->inflight_split->desc_num; i++) {
1753                         if (vq->inflight_split->desc[i].inflight == 1) {
1754                                 resubmit->resubmit_list[num].index = i;
1755                                 resubmit->resubmit_list[num].counter =
1756                                         inflight_split->desc[i].counter;
1757                                 num++;
1758                         }
1759                 }
1760                 resubmit->resubmit_num = num;
1761
1762                 if (resubmit->resubmit_num > 1)
1763                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1764                               sizeof(struct rte_vhost_resubmit_desc),
1765                               resubmit_desc_compare);
1766
1767                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1768                 vq->resubmit_inflight = resubmit;
1769         }
1770
1771         return RTE_VHOST_MSG_RESULT_OK;
1772 }
1773
1774 static int
1775 vhost_check_queue_inflights_packed(struct virtio_net *dev,
1776                                    struct vhost_virtqueue *vq)
1777 {
1778         uint16_t i;
1779         uint16_t resubmit_num = 0, old_used_idx, num;
1780         struct rte_vhost_resubmit_info *resubmit;
1781         struct rte_vhost_inflight_info_packed *inflight_packed;
1782
1783         if (!(dev->protocol_features &
1784             (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)))
1785                 return RTE_VHOST_MSG_RESULT_OK;
1786
1787         /* The frontend may still not support the inflight feature
1788          * although we negotiate the protocol feature.
1789          */
1790         if ((!vq->inflight_packed))
1791                 return RTE_VHOST_MSG_RESULT_OK;
1792
1793         if (!vq->inflight_packed->version) {
1794                 vq->inflight_packed->version = INFLIGHT_VERSION;
1795                 return RTE_VHOST_MSG_RESULT_OK;
1796         }
1797
1798         if (vq->resubmit_inflight)
1799                 return RTE_VHOST_MSG_RESULT_OK;
1800
1801         inflight_packed = vq->inflight_packed;
1802         vq->global_counter = 0;
1803         old_used_idx = inflight_packed->old_used_idx;
1804
1805         if (inflight_packed->used_idx != old_used_idx) {
1806                 if (inflight_packed->desc[old_used_idx].inflight == 0) {
1807                         inflight_packed->old_used_idx =
1808                                 inflight_packed->used_idx;
1809                         inflight_packed->old_used_wrap_counter =
1810                                 inflight_packed->used_wrap_counter;
1811                         inflight_packed->old_free_head =
1812                                 inflight_packed->free_head;
1813                 } else {
1814                         inflight_packed->used_idx =
1815                                 inflight_packed->old_used_idx;
1816                         inflight_packed->used_wrap_counter =
1817                                 inflight_packed->old_used_wrap_counter;
1818                         inflight_packed->free_head =
1819                                 inflight_packed->old_free_head;
1820                 }
1821         }
1822
1823         for (i = 0; i < inflight_packed->desc_num; i++) {
1824                 if (inflight_packed->desc[i].inflight == 1)
1825                         resubmit_num++;
1826         }
1827
1828         if (resubmit_num) {
1829                 resubmit = calloc(1, sizeof(struct rte_vhost_resubmit_info));
1830                 if (resubmit == NULL) {
1831                         VHOST_LOG_CONFIG(ERR,
1832                                 "failed to allocate memory for resubmit info.\n");
1833                         return RTE_VHOST_MSG_RESULT_ERR;
1834                 }
1835
1836                 resubmit->resubmit_list = calloc(resubmit_num,
1837                         sizeof(struct rte_vhost_resubmit_desc));
1838                 if (resubmit->resubmit_list == NULL) {
1839                         VHOST_LOG_CONFIG(ERR,
1840                                 "failed to allocate memory for resubmit desc.\n");
1841                         free(resubmit);
1842                         return RTE_VHOST_MSG_RESULT_ERR;
1843                 }
1844
1845                 num = 0;
1846                 for (i = 0; i < inflight_packed->desc_num; i++) {
1847                         if (vq->inflight_packed->desc[i].inflight == 1) {
1848                                 resubmit->resubmit_list[num].index = i;
1849                                 resubmit->resubmit_list[num].counter =
1850                                         inflight_packed->desc[i].counter;
1851                                 num++;
1852                         }
1853                 }
1854                 resubmit->resubmit_num = num;
1855
1856                 if (resubmit->resubmit_num > 1)
1857                         qsort(resubmit->resubmit_list, resubmit->resubmit_num,
1858                               sizeof(struct rte_vhost_resubmit_desc),
1859                               resubmit_desc_compare);
1860
1861                 vq->global_counter = resubmit->resubmit_list[0].counter + 1;
1862                 vq->resubmit_inflight = resubmit;
1863         }
1864
1865         return RTE_VHOST_MSG_RESULT_OK;
1866 }
1867
1868 static int
1869 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *msg,
1870                         int main_fd __rte_unused)
1871 {
1872         struct virtio_net *dev = *pdev;
1873         struct vhost_vring_file file;
1874         struct vhost_virtqueue *vq;
1875         int expected_fds;
1876
1877         expected_fds = (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK) ? 0 : 1;
1878         if (validate_msg_fds(msg, expected_fds) != 0)
1879                 return RTE_VHOST_MSG_RESULT_ERR;
1880
1881         file.index = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1882         if (msg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
1883                 file.fd = VIRTIO_INVALID_EVENTFD;
1884         else
1885                 file.fd = msg->fds[0];
1886         VHOST_LOG_CONFIG(INFO,
1887                 "vring kick idx:%d file:%d\n", file.index, file.fd);
1888
1889         /* Interpret ring addresses only when ring is started. */
1890         dev = translate_ring_addresses(dev, file.index);
1891         if (!dev)
1892                 return RTE_VHOST_MSG_RESULT_ERR;
1893
1894         *pdev = dev;
1895
1896         vq = dev->virtqueue[file.index];
1897
1898         /*
1899          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
1900          * the ring starts already enabled. Otherwise, it is enabled via
1901          * the SET_VRING_ENABLE message.
1902          */
1903         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
1904                 vq->enabled = 1;
1905                 if (dev->notify_ops->vring_state_changed)
1906                         dev->notify_ops->vring_state_changed(
1907                                 dev->vid, file.index, 1);
1908         }
1909
1910         if (vq->ready) {
1911                 vq->ready = 0;
1912                 vhost_user_notify_queue_state(dev, file.index, 0);
1913         }
1914
1915         if (vq->kickfd >= 0)
1916                 close(vq->kickfd);
1917         vq->kickfd = file.fd;
1918
1919         if (vq_is_packed(dev)) {
1920                 if (vhost_check_queue_inflights_packed(dev, vq)) {
1921                         VHOST_LOG_CONFIG(ERR,
1922                                 "failed to inflights for vq: %d\n", file.index);
1923                         return RTE_VHOST_MSG_RESULT_ERR;
1924                 }
1925         } else {
1926                 if (vhost_check_queue_inflights_split(dev, vq)) {
1927                         VHOST_LOG_CONFIG(ERR,
1928                                 "failed to inflights for vq: %d\n", file.index);
1929                         return RTE_VHOST_MSG_RESULT_ERR;
1930                 }
1931         }
1932
1933         return RTE_VHOST_MSG_RESULT_OK;
1934 }
1935
1936 static void
1937 free_zmbufs(struct vhost_virtqueue *vq)
1938 {
1939         drain_zmbuf_list(vq);
1940
1941         rte_free(vq->zmbufs);
1942         vq->zmbufs = NULL;
1943 }
1944
1945 /*
1946  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
1947  */
1948 static int
1949 vhost_user_get_vring_base(struct virtio_net **pdev,
1950                         struct VhostUserMsg *msg,
1951                         int main_fd __rte_unused)
1952 {
1953         struct virtio_net *dev = *pdev;
1954         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
1955         uint64_t val;
1956
1957         if (validate_msg_fds(msg, 0) != 0)
1958                 return RTE_VHOST_MSG_RESULT_ERR;
1959
1960         /* We have to stop the queue (virtio) if it is running. */
1961         vhost_destroy_device_notify(dev);
1962
1963         dev->flags &= ~VIRTIO_DEV_READY;
1964         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
1965
1966         /* Here we are safe to get the indexes */
1967         if (vq_is_packed(dev)) {
1968                 /*
1969                  * Bit[0:14]: avail index
1970                  * Bit[15]: avail wrap counter
1971                  */
1972                 val = vq->last_avail_idx & 0x7fff;
1973                 val |= vq->avail_wrap_counter << 15;
1974                 msg->payload.state.num = val;
1975         } else {
1976                 msg->payload.state.num = vq->last_avail_idx;
1977         }
1978
1979         VHOST_LOG_CONFIG(INFO,
1980                 "vring base idx:%d file:%d\n", msg->payload.state.index,
1981                 msg->payload.state.num);
1982         /*
1983          * Based on current qemu vhost-user implementation, this message is
1984          * sent and only sent in vhost_vring_stop.
1985          * TODO: cleanup the vring, it isn't usable since here.
1986          */
1987         if (vq->kickfd >= 0)
1988                 close(vq->kickfd);
1989
1990         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
1991
1992         if (vq->callfd >= 0)
1993                 close(vq->callfd);
1994
1995         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1996
1997         vq->signalled_used_valid = false;
1998
1999         if (dev->dequeue_zero_copy)
2000                 free_zmbufs(vq);
2001         if (vq_is_packed(dev)) {
2002                 rte_free(vq->shadow_used_packed);
2003                 vq->shadow_used_packed = NULL;
2004         } else {
2005                 rte_free(vq->shadow_used_split);
2006                 vq->shadow_used_split = NULL;
2007                 if (vq->async_pkts_pending)
2008                         rte_free(vq->async_pkts_pending);
2009                 if (vq->async_pending_info)
2010                         rte_free(vq->async_pending_info);
2011                 vq->async_pkts_pending = NULL;
2012                 vq->async_pending_info = NULL;
2013         }
2014
2015         rte_free(vq->batch_copy_elems);
2016         vq->batch_copy_elems = NULL;
2017
2018         msg->size = sizeof(msg->payload.state);
2019         msg->fd_num = 0;
2020
2021         vring_invalidate(dev, vq);
2022
2023         return RTE_VHOST_MSG_RESULT_REPLY;
2024 }
2025
2026 /*
2027  * when virtio queues are ready to work, qemu will send us to
2028  * enable the virtio queue pair.
2029  */
2030 static int
2031 vhost_user_set_vring_enable(struct virtio_net **pdev,
2032                         struct VhostUserMsg *msg,
2033                         int main_fd __rte_unused)
2034 {
2035         struct virtio_net *dev = *pdev;
2036         int enable = (int)msg->payload.state.num;
2037         int index = (int)msg->payload.state.index;
2038
2039         if (validate_msg_fds(msg, 0) != 0)
2040                 return RTE_VHOST_MSG_RESULT_ERR;
2041
2042         VHOST_LOG_CONFIG(INFO,
2043                 "set queue enable: %d to qp idx: %d\n",
2044                 enable, index);
2045
2046         if (!enable && dev->virtqueue[index]->async_registered) {
2047                 if (dev->virtqueue[index]->async_pkts_inflight_n) {
2048                         VHOST_LOG_CONFIG(ERR, "failed to disable vring. "
2049                         "async inflight packets must be completed first\n");
2050                         return RTE_VHOST_MSG_RESULT_ERR;
2051                 }
2052         }
2053
2054         /* On disable, rings have to be stopped being processed. */
2055         if (!enable && dev->dequeue_zero_copy)
2056                 drain_zmbuf_list(dev->virtqueue[index]);
2057
2058         dev->virtqueue[index]->enabled = enable;
2059
2060         return RTE_VHOST_MSG_RESULT_OK;
2061 }
2062
2063 static int
2064 vhost_user_get_protocol_features(struct virtio_net **pdev,
2065                         struct VhostUserMsg *msg,
2066                         int main_fd __rte_unused)
2067 {
2068         struct virtio_net *dev = *pdev;
2069         uint64_t features, protocol_features;
2070
2071         if (validate_msg_fds(msg, 0) != 0)
2072                 return RTE_VHOST_MSG_RESULT_ERR;
2073
2074         rte_vhost_driver_get_features(dev->ifname, &features);
2075         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
2076
2077         msg->payload.u64 = protocol_features;
2078         msg->size = sizeof(msg->payload.u64);
2079         msg->fd_num = 0;
2080
2081         return RTE_VHOST_MSG_RESULT_REPLY;
2082 }
2083
2084 static int
2085 vhost_user_set_protocol_features(struct virtio_net **pdev,
2086                         struct VhostUserMsg *msg,
2087                         int main_fd __rte_unused)
2088 {
2089         struct virtio_net *dev = *pdev;
2090         uint64_t protocol_features = msg->payload.u64;
2091         uint64_t slave_protocol_features = 0;
2092
2093         if (validate_msg_fds(msg, 0) != 0)
2094                 return RTE_VHOST_MSG_RESULT_ERR;
2095
2096         rte_vhost_driver_get_protocol_features(dev->ifname,
2097                         &slave_protocol_features);
2098         if (protocol_features & ~slave_protocol_features) {
2099                 VHOST_LOG_CONFIG(ERR,
2100                         "(%d) received invalid protocol features.\n",
2101                         dev->vid);
2102                 return RTE_VHOST_MSG_RESULT_ERR;
2103         }
2104
2105         dev->protocol_features = protocol_features;
2106         VHOST_LOG_CONFIG(INFO,
2107                 "negotiated Vhost-user protocol features: 0x%" PRIx64 "\n",
2108                 dev->protocol_features);
2109
2110         return RTE_VHOST_MSG_RESULT_OK;
2111 }
2112
2113 static int
2114 vhost_user_set_log_base(struct virtio_net **pdev, struct VhostUserMsg *msg,
2115                         int main_fd __rte_unused)
2116 {
2117         struct virtio_net *dev = *pdev;
2118         int fd = msg->fds[0];
2119         uint64_t size, off;
2120         void *addr;
2121
2122         if (validate_msg_fds(msg, 1) != 0)
2123                 return RTE_VHOST_MSG_RESULT_ERR;
2124
2125         if (fd < 0) {
2126                 VHOST_LOG_CONFIG(ERR, "invalid log fd: %d\n", fd);
2127                 return RTE_VHOST_MSG_RESULT_ERR;
2128         }
2129
2130         if (msg->size != sizeof(VhostUserLog)) {
2131                 VHOST_LOG_CONFIG(ERR,
2132                         "invalid log base msg size: %"PRId32" != %d\n",
2133                         msg->size, (int)sizeof(VhostUserLog));
2134                 return RTE_VHOST_MSG_RESULT_ERR;
2135         }
2136
2137         size = msg->payload.log.mmap_size;
2138         off  = msg->payload.log.mmap_offset;
2139
2140         /* Check for mmap size and offset overflow. */
2141         if (off >= -size) {
2142                 VHOST_LOG_CONFIG(ERR,
2143                         "log offset %#"PRIx64" and log size %#"PRIx64" overflow\n",
2144                         off, size);
2145                 return RTE_VHOST_MSG_RESULT_ERR;
2146         }
2147
2148         VHOST_LOG_CONFIG(INFO,
2149                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
2150                 size, off);
2151
2152         /*
2153          * mmap from 0 to workaround a hugepage mmap bug: mmap will
2154          * fail when offset is not page size aligned.
2155          */
2156         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
2157         close(fd);
2158         if (addr == MAP_FAILED) {
2159                 VHOST_LOG_CONFIG(ERR, "mmap log base failed!\n");
2160                 return RTE_VHOST_MSG_RESULT_ERR;
2161         }
2162
2163         /*
2164          * Free previously mapped log memory on occasionally
2165          * multiple VHOST_USER_SET_LOG_BASE.
2166          */
2167         if (dev->log_addr) {
2168                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
2169         }
2170         dev->log_addr = (uint64_t)(uintptr_t)addr;
2171         dev->log_base = dev->log_addr + off;
2172         dev->log_size = size;
2173
2174         /*
2175          * The spec is not clear about it (yet), but QEMU doesn't expect
2176          * any payload in the reply.
2177          */
2178         msg->size = 0;
2179         msg->fd_num = 0;
2180
2181         return RTE_VHOST_MSG_RESULT_REPLY;
2182 }
2183
2184 static int vhost_user_set_log_fd(struct virtio_net **pdev __rte_unused,
2185                         struct VhostUserMsg *msg,
2186                         int main_fd __rte_unused)
2187 {
2188         if (validate_msg_fds(msg, 1) != 0)
2189                 return RTE_VHOST_MSG_RESULT_ERR;
2190
2191         close(msg->fds[0]);
2192         VHOST_LOG_CONFIG(INFO, "not implemented.\n");
2193
2194         return RTE_VHOST_MSG_RESULT_OK;
2195 }
2196
2197 /*
2198  * An rarp packet is constructed and broadcasted to notify switches about
2199  * the new location of the migrated VM, so that packets from outside will
2200  * not be lost after migration.
2201  *
2202  * However, we don't actually "send" a rarp packet here, instead, we set
2203  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
2204  */
2205 static int
2206 vhost_user_send_rarp(struct virtio_net **pdev, struct VhostUserMsg *msg,
2207                         int main_fd __rte_unused)
2208 {
2209         struct virtio_net *dev = *pdev;
2210         uint8_t *mac = (uint8_t *)&msg->payload.u64;
2211         struct rte_vdpa_device *vdpa_dev;
2212
2213         if (validate_msg_fds(msg, 0) != 0)
2214                 return RTE_VHOST_MSG_RESULT_ERR;
2215
2216         VHOST_LOG_CONFIG(DEBUG,
2217                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
2218                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
2219         memcpy(dev->mac.addr_bytes, mac, 6);
2220
2221         /*
2222          * Set the flag to inject a RARP broadcast packet at
2223          * rte_vhost_dequeue_burst().
2224          *
2225          * __ATOMIC_RELEASE ordering is for making sure the mac is
2226          * copied before the flag is set.
2227          */
2228         __atomic_store_n(&dev->broadcast_rarp, 1, __ATOMIC_RELEASE);
2229         vdpa_dev = dev->vdpa_dev;
2230         if (vdpa_dev && vdpa_dev->ops->migration_done)
2231                 vdpa_dev->ops->migration_done(dev->vid);
2232
2233         return RTE_VHOST_MSG_RESULT_OK;
2234 }
2235
2236 static int
2237 vhost_user_net_set_mtu(struct virtio_net **pdev, struct VhostUserMsg *msg,
2238                         int main_fd __rte_unused)
2239 {
2240         struct virtio_net *dev = *pdev;
2241
2242         if (validate_msg_fds(msg, 0) != 0)
2243                 return RTE_VHOST_MSG_RESULT_ERR;
2244
2245         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
2246                         msg->payload.u64 > VIRTIO_MAX_MTU) {
2247                 VHOST_LOG_CONFIG(ERR, "Invalid MTU size (%"PRIu64")\n",
2248                                 msg->payload.u64);
2249
2250                 return RTE_VHOST_MSG_RESULT_ERR;
2251         }
2252
2253         dev->mtu = msg->payload.u64;
2254
2255         return RTE_VHOST_MSG_RESULT_OK;
2256 }
2257
2258 static int
2259 vhost_user_set_req_fd(struct virtio_net **pdev, struct VhostUserMsg *msg,
2260                         int main_fd __rte_unused)
2261 {
2262         struct virtio_net *dev = *pdev;
2263         int fd = msg->fds[0];
2264
2265         if (validate_msg_fds(msg, 1) != 0)
2266                 return RTE_VHOST_MSG_RESULT_ERR;
2267
2268         if (fd < 0) {
2269                 VHOST_LOG_CONFIG(ERR,
2270                                 "Invalid file descriptor for slave channel (%d)\n",
2271                                 fd);
2272                 return RTE_VHOST_MSG_RESULT_ERR;
2273         }
2274
2275         if (dev->slave_req_fd >= 0)
2276                 close(dev->slave_req_fd);
2277
2278         dev->slave_req_fd = fd;
2279
2280         return RTE_VHOST_MSG_RESULT_OK;
2281 }
2282
2283 static int
2284 is_vring_iotlb_split(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2285 {
2286         struct vhost_vring_addr *ra;
2287         uint64_t start, end, len;
2288
2289         start = imsg->iova;
2290         end = start + imsg->size;
2291
2292         ra = &vq->ring_addrs;
2293         len = sizeof(struct vring_desc) * vq->size;
2294         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2295                 return 1;
2296
2297         len = sizeof(struct vring_avail) + sizeof(uint16_t) * vq->size;
2298         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2299                 return 1;
2300
2301         len = sizeof(struct vring_used) +
2302                sizeof(struct vring_used_elem) * vq->size;
2303         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2304                 return 1;
2305
2306         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2307                 len = sizeof(uint64_t);
2308                 if (ra->log_guest_addr < end &&
2309                     (ra->log_guest_addr + len) > start)
2310                         return 1;
2311         }
2312
2313         return 0;
2314 }
2315
2316 static int
2317 is_vring_iotlb_packed(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
2318 {
2319         struct vhost_vring_addr *ra;
2320         uint64_t start, end, len;
2321
2322         start = imsg->iova;
2323         end = start + imsg->size;
2324
2325         ra = &vq->ring_addrs;
2326         len = sizeof(struct vring_packed_desc) * vq->size;
2327         if (ra->desc_user_addr < end && (ra->desc_user_addr + len) > start)
2328                 return 1;
2329
2330         len = sizeof(struct vring_packed_desc_event);
2331         if (ra->avail_user_addr < end && (ra->avail_user_addr + len) > start)
2332                 return 1;
2333
2334         len = sizeof(struct vring_packed_desc_event);
2335         if (ra->used_user_addr < end && (ra->used_user_addr + len) > start)
2336                 return 1;
2337
2338         if (ra->flags & (1 << VHOST_VRING_F_LOG)) {
2339                 len = sizeof(uint64_t);
2340                 if (ra->log_guest_addr < end &&
2341                     (ra->log_guest_addr + len) > start)
2342                         return 1;
2343         }
2344
2345         return 0;
2346 }
2347
2348 static int is_vring_iotlb(struct virtio_net *dev,
2349                           struct vhost_virtqueue *vq,
2350                           struct vhost_iotlb_msg *imsg)
2351 {
2352         if (vq_is_packed(dev))
2353                 return is_vring_iotlb_packed(vq, imsg);
2354         else
2355                 return is_vring_iotlb_split(vq, imsg);
2356 }
2357
2358 static int
2359 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg,
2360                         int main_fd __rte_unused)
2361 {
2362         struct virtio_net *dev = *pdev;
2363         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
2364         uint16_t i;
2365         uint64_t vva, len;
2366
2367         if (validate_msg_fds(msg, 0) != 0)
2368                 return RTE_VHOST_MSG_RESULT_ERR;
2369
2370         switch (imsg->type) {
2371         case VHOST_IOTLB_UPDATE:
2372                 len = imsg->size;
2373                 vva = qva_to_vva(dev, imsg->uaddr, &len);
2374                 if (!vva)
2375                         return RTE_VHOST_MSG_RESULT_ERR;
2376
2377                 for (i = 0; i < dev->nr_vring; i++) {
2378                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2379
2380                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
2381                                         len, imsg->perm);
2382
2383                         if (is_vring_iotlb(dev, vq, imsg))
2384                                 *pdev = dev = translate_ring_addresses(dev, i);
2385                 }
2386                 break;
2387         case VHOST_IOTLB_INVALIDATE:
2388                 for (i = 0; i < dev->nr_vring; i++) {
2389                         struct vhost_virtqueue *vq = dev->virtqueue[i];
2390
2391                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
2392                                         imsg->size);
2393
2394                         if (is_vring_iotlb(dev, vq, imsg))
2395                                 vring_invalidate(dev, vq);
2396                 }
2397                 break;
2398         default:
2399                 VHOST_LOG_CONFIG(ERR, "Invalid IOTLB message type (%d)\n",
2400                                 imsg->type);
2401                 return RTE_VHOST_MSG_RESULT_ERR;
2402         }
2403
2404         return RTE_VHOST_MSG_RESULT_OK;
2405 }
2406
2407 static int
2408 vhost_user_set_postcopy_advise(struct virtio_net **pdev,
2409                         struct VhostUserMsg *msg,
2410                         int main_fd __rte_unused)
2411 {
2412         struct virtio_net *dev = *pdev;
2413 #ifdef RTE_LIBRTE_VHOST_POSTCOPY
2414         struct uffdio_api api_struct;
2415
2416         if (validate_msg_fds(msg, 0) != 0)
2417                 return RTE_VHOST_MSG_RESULT_ERR;
2418
2419         dev->postcopy_ufd = syscall(__NR_userfaultfd, O_CLOEXEC | O_NONBLOCK);
2420
2421         if (dev->postcopy_ufd == -1) {
2422                 VHOST_LOG_CONFIG(ERR, "Userfaultfd not available: %s\n",
2423                         strerror(errno));
2424                 return RTE_VHOST_MSG_RESULT_ERR;
2425         }
2426         api_struct.api = UFFD_API;
2427         api_struct.features = 0;
2428         if (ioctl(dev->postcopy_ufd, UFFDIO_API, &api_struct)) {
2429                 VHOST_LOG_CONFIG(ERR, "UFFDIO_API ioctl failure: %s\n",
2430                         strerror(errno));
2431                 close(dev->postcopy_ufd);
2432                 dev->postcopy_ufd = -1;
2433                 return RTE_VHOST_MSG_RESULT_ERR;
2434         }
2435         msg->fds[0] = dev->postcopy_ufd;
2436         msg->fd_num = 1;
2437
2438         return RTE_VHOST_MSG_RESULT_REPLY;
2439 #else
2440         dev->postcopy_ufd = -1;
2441         msg->fd_num = 0;
2442
2443         return RTE_VHOST_MSG_RESULT_ERR;
2444 #endif
2445 }
2446
2447 static int
2448 vhost_user_set_postcopy_listen(struct virtio_net **pdev,
2449                         struct VhostUserMsg *msg __rte_unused,
2450                         int main_fd __rte_unused)
2451 {
2452         struct virtio_net *dev = *pdev;
2453
2454         if (validate_msg_fds(msg, 0) != 0)
2455                 return RTE_VHOST_MSG_RESULT_ERR;
2456
2457         if (dev->mem && dev->mem->nregions) {
2458                 VHOST_LOG_CONFIG(ERR,
2459                         "Regions already registered at postcopy-listen\n");
2460                 return RTE_VHOST_MSG_RESULT_ERR;
2461         }
2462         dev->postcopy_listening = 1;
2463
2464         return RTE_VHOST_MSG_RESULT_OK;
2465 }
2466
2467 static int
2468 vhost_user_postcopy_end(struct virtio_net **pdev, struct VhostUserMsg *msg,
2469                         int main_fd __rte_unused)
2470 {
2471         struct virtio_net *dev = *pdev;
2472
2473         if (validate_msg_fds(msg, 0) != 0)
2474                 return RTE_VHOST_MSG_RESULT_ERR;
2475
2476         dev->postcopy_listening = 0;
2477         if (dev->postcopy_ufd >= 0) {
2478                 close(dev->postcopy_ufd);
2479                 dev->postcopy_ufd = -1;
2480         }
2481
2482         msg->payload.u64 = 0;
2483         msg->size = sizeof(msg->payload.u64);
2484         msg->fd_num = 0;
2485
2486         return RTE_VHOST_MSG_RESULT_REPLY;
2487 }
2488
2489 static int
2490 vhost_user_get_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2491                       int main_fd __rte_unused)
2492 {
2493         struct virtio_net *dev = *pdev;
2494
2495         if (validate_msg_fds(msg, 0) != 0)
2496                 return RTE_VHOST_MSG_RESULT_ERR;
2497
2498         msg->payload.u64 = dev->status;
2499         msg->size = sizeof(msg->payload.u64);
2500         msg->fd_num = 0;
2501
2502         return RTE_VHOST_MSG_RESULT_REPLY;
2503 }
2504
2505 static int
2506 vhost_user_set_status(struct virtio_net **pdev, struct VhostUserMsg *msg,
2507                         int main_fd __rte_unused)
2508 {
2509         struct virtio_net *dev = *pdev;
2510
2511         if (validate_msg_fds(msg, 0) != 0)
2512                 return RTE_VHOST_MSG_RESULT_ERR;
2513
2514         /* As per Virtio specification, the device status is 8bits long */
2515         if (msg->payload.u64 > UINT8_MAX) {
2516                 VHOST_LOG_CONFIG(ERR, "Invalid VHOST_USER_SET_STATUS payload 0x%" PRIx64 "\n",
2517                                 msg->payload.u64);
2518                 return RTE_VHOST_MSG_RESULT_ERR;
2519         }
2520
2521         dev->status = msg->payload.u64;
2522
2523         if ((dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK) &&
2524             (dev->flags & VIRTIO_DEV_FEATURES_FAILED)) {
2525                 VHOST_LOG_CONFIG(ERR, "FEATURES_OK bit is set but feature negotiation failed\n");
2526                 /*
2527                  * Clear the bit to let the driver know about the feature
2528                  * negotiation failure
2529                  */
2530                 dev->status &= ~VIRTIO_DEVICE_STATUS_FEATURES_OK;
2531         }
2532
2533         VHOST_LOG_CONFIG(INFO, "New device status(0x%08x):\n"
2534                         "\t-ACKNOWLEDGE: %u\n"
2535                         "\t-DRIVER: %u\n"
2536                         "\t-FEATURES_OK: %u\n"
2537                         "\t-DRIVER_OK: %u\n"
2538                         "\t-DEVICE_NEED_RESET: %u\n"
2539                         "\t-FAILED: %u\n",
2540                         dev->status,
2541                         !!(dev->status & VIRTIO_DEVICE_STATUS_ACK),
2542                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER),
2543                         !!(dev->status & VIRTIO_DEVICE_STATUS_FEATURES_OK),
2544                         !!(dev->status & VIRTIO_DEVICE_STATUS_DRIVER_OK),
2545                         !!(dev->status & VIRTIO_DEVICE_STATUS_DEV_NEED_RESET),
2546                         !!(dev->status & VIRTIO_DEVICE_STATUS_FAILED));
2547
2548         return RTE_VHOST_MSG_RESULT_OK;
2549 }
2550
2551 typedef int (*vhost_message_handler_t)(struct virtio_net **pdev,
2552                                         struct VhostUserMsg *msg,
2553                                         int main_fd);
2554 static vhost_message_handler_t vhost_message_handlers[VHOST_USER_MAX] = {
2555         [VHOST_USER_NONE] = NULL,
2556         [VHOST_USER_GET_FEATURES] = vhost_user_get_features,
2557         [VHOST_USER_SET_FEATURES] = vhost_user_set_features,
2558         [VHOST_USER_SET_OWNER] = vhost_user_set_owner,
2559         [VHOST_USER_RESET_OWNER] = vhost_user_reset_owner,
2560         [VHOST_USER_SET_MEM_TABLE] = vhost_user_set_mem_table,
2561         [VHOST_USER_SET_LOG_BASE] = vhost_user_set_log_base,
2562         [VHOST_USER_SET_LOG_FD] = vhost_user_set_log_fd,
2563         [VHOST_USER_SET_VRING_NUM] = vhost_user_set_vring_num,
2564         [VHOST_USER_SET_VRING_ADDR] = vhost_user_set_vring_addr,
2565         [VHOST_USER_SET_VRING_BASE] = vhost_user_set_vring_base,
2566         [VHOST_USER_GET_VRING_BASE] = vhost_user_get_vring_base,
2567         [VHOST_USER_SET_VRING_KICK] = vhost_user_set_vring_kick,
2568         [VHOST_USER_SET_VRING_CALL] = vhost_user_set_vring_call,
2569         [VHOST_USER_SET_VRING_ERR] = vhost_user_set_vring_err,
2570         [VHOST_USER_GET_PROTOCOL_FEATURES] = vhost_user_get_protocol_features,
2571         [VHOST_USER_SET_PROTOCOL_FEATURES] = vhost_user_set_protocol_features,
2572         [VHOST_USER_GET_QUEUE_NUM] = vhost_user_get_queue_num,
2573         [VHOST_USER_SET_VRING_ENABLE] = vhost_user_set_vring_enable,
2574         [VHOST_USER_SEND_RARP] = vhost_user_send_rarp,
2575         [VHOST_USER_NET_SET_MTU] = vhost_user_net_set_mtu,
2576         [VHOST_USER_SET_SLAVE_REQ_FD] = vhost_user_set_req_fd,
2577         [VHOST_USER_IOTLB_MSG] = vhost_user_iotlb_msg,
2578         [VHOST_USER_POSTCOPY_ADVISE] = vhost_user_set_postcopy_advise,
2579         [VHOST_USER_POSTCOPY_LISTEN] = vhost_user_set_postcopy_listen,
2580         [VHOST_USER_POSTCOPY_END] = vhost_user_postcopy_end,
2581         [VHOST_USER_GET_INFLIGHT_FD] = vhost_user_get_inflight_fd,
2582         [VHOST_USER_SET_INFLIGHT_FD] = vhost_user_set_inflight_fd,
2583         [VHOST_USER_SET_STATUS] = vhost_user_set_status,
2584         [VHOST_USER_GET_STATUS] = vhost_user_get_status,
2585 };
2586
2587 /* return bytes# of read on success or negative val on failure. */
2588 static int
2589 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
2590 {
2591         int ret;
2592
2593         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
2594                 msg->fds, VHOST_MEMORY_MAX_NREGIONS, &msg->fd_num);
2595         if (ret <= 0) {
2596                 return ret;
2597         } else if (ret != VHOST_USER_HDR_SIZE) {
2598                 VHOST_LOG_CONFIG(ERR, "Unexpected header size read\n");
2599                 close_msg_fds(msg);
2600                 return -1;
2601         }
2602
2603         if (msg->size) {
2604                 if (msg->size > sizeof(msg->payload)) {
2605                         VHOST_LOG_CONFIG(ERR,
2606                                 "invalid msg size: %d\n", msg->size);
2607                         return -1;
2608                 }
2609                 ret = read(sockfd, &msg->payload, msg->size);
2610                 if (ret <= 0)
2611                         return ret;
2612                 if (ret != (int)msg->size) {
2613                         VHOST_LOG_CONFIG(ERR,
2614                                 "read control message failed\n");
2615                         return -1;
2616                 }
2617         }
2618
2619         return ret;
2620 }
2621
2622 static int
2623 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
2624 {
2625         if (!msg)
2626                 return 0;
2627
2628         return send_fd_message(sockfd, (char *)msg,
2629                 VHOST_USER_HDR_SIZE + msg->size, msg->fds, msg->fd_num);
2630 }
2631
2632 static int
2633 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
2634 {
2635         if (!msg)
2636                 return 0;
2637
2638         msg->flags &= ~VHOST_USER_VERSION_MASK;
2639         msg->flags &= ~VHOST_USER_NEED_REPLY;
2640         msg->flags |= VHOST_USER_VERSION;
2641         msg->flags |= VHOST_USER_REPLY_MASK;
2642
2643         return send_vhost_message(sockfd, msg);
2644 }
2645
2646 static int
2647 send_vhost_slave_message(struct virtio_net *dev, struct VhostUserMsg *msg)
2648 {
2649         int ret;
2650
2651         if (msg->flags & VHOST_USER_NEED_REPLY)
2652                 rte_spinlock_lock(&dev->slave_req_lock);
2653
2654         ret = send_vhost_message(dev->slave_req_fd, msg);
2655         if (ret < 0 && (msg->flags & VHOST_USER_NEED_REPLY))
2656                 rte_spinlock_unlock(&dev->slave_req_lock);
2657
2658         return ret;
2659 }
2660
2661 /*
2662  * Allocate a queue pair if it hasn't been allocated yet
2663  */
2664 static int
2665 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev,
2666                         struct VhostUserMsg *msg)
2667 {
2668         uint32_t vring_idx;
2669
2670         switch (msg->request.master) {
2671         case VHOST_USER_SET_VRING_KICK:
2672         case VHOST_USER_SET_VRING_CALL:
2673         case VHOST_USER_SET_VRING_ERR:
2674                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
2675                 break;
2676         case VHOST_USER_SET_VRING_NUM:
2677         case VHOST_USER_SET_VRING_BASE:
2678         case VHOST_USER_SET_VRING_ENABLE:
2679                 vring_idx = msg->payload.state.index;
2680                 break;
2681         case VHOST_USER_SET_VRING_ADDR:
2682                 vring_idx = msg->payload.addr.index;
2683                 break;
2684         default:
2685                 return 0;
2686         }
2687
2688         if (vring_idx >= VHOST_MAX_VRING) {
2689                 VHOST_LOG_CONFIG(ERR,
2690                         "invalid vring index: %u\n", vring_idx);
2691                 return -1;
2692         }
2693
2694         if (dev->virtqueue[vring_idx])
2695                 return 0;
2696
2697         return alloc_vring_queue(dev, vring_idx);
2698 }
2699
2700 static void
2701 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
2702 {
2703         unsigned int i = 0;
2704         unsigned int vq_num = 0;
2705
2706         while (vq_num < dev->nr_vring) {
2707                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2708
2709                 if (vq) {
2710                         rte_spinlock_lock(&vq->access_lock);
2711                         vq_num++;
2712                 }
2713                 i++;
2714         }
2715 }
2716
2717 static void
2718 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
2719 {
2720         unsigned int i = 0;
2721         unsigned int vq_num = 0;
2722
2723         while (vq_num < dev->nr_vring) {
2724                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2725
2726                 if (vq) {
2727                         rte_spinlock_unlock(&vq->access_lock);
2728                         vq_num++;
2729                 }
2730                 i++;
2731         }
2732 }
2733
2734 int
2735 vhost_user_msg_handler(int vid, int fd)
2736 {
2737         struct virtio_net *dev;
2738         struct VhostUserMsg msg;
2739         struct rte_vdpa_device *vdpa_dev;
2740         int ret;
2741         int unlock_required = 0;
2742         bool handled;
2743         int request;
2744         uint32_t i;
2745
2746         dev = get_device(vid);
2747         if (dev == NULL)
2748                 return -1;
2749
2750         if (!dev->notify_ops) {
2751                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
2752                 if (!dev->notify_ops) {
2753                         VHOST_LOG_CONFIG(ERR,
2754                                 "failed to get callback ops for driver %s\n",
2755                                 dev->ifname);
2756                         return -1;
2757                 }
2758         }
2759
2760         ret = read_vhost_message(fd, &msg);
2761         if (ret <= 0) {
2762                 if (ret < 0)
2763                         VHOST_LOG_CONFIG(ERR,
2764                                 "vhost read message failed\n");
2765                 else
2766                         VHOST_LOG_CONFIG(INFO,
2767                                 "vhost peer closed\n");
2768
2769                 return -1;
2770         }
2771
2772         ret = 0;
2773         request = msg.request.master;
2774         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX &&
2775                         vhost_message_str[request]) {
2776                 if (request != VHOST_USER_IOTLB_MSG)
2777                         VHOST_LOG_CONFIG(INFO, "read message %s\n",
2778                                 vhost_message_str[request]);
2779                 else
2780                         VHOST_LOG_CONFIG(DEBUG, "read message %s\n",
2781                                 vhost_message_str[request]);
2782         } else {
2783                 VHOST_LOG_CONFIG(DEBUG, "External request %d\n", request);
2784         }
2785
2786         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
2787         if (ret < 0) {
2788                 VHOST_LOG_CONFIG(ERR,
2789                         "failed to alloc queue\n");
2790                 return -1;
2791         }
2792
2793         /*
2794          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
2795          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
2796          * and device is destroyed. destroy_device waits for queues to be
2797          * inactive, so it is safe. Otherwise taking the access_lock
2798          * would cause a dead lock.
2799          */
2800         switch (request) {
2801         case VHOST_USER_SET_FEATURES:
2802         case VHOST_USER_SET_PROTOCOL_FEATURES:
2803         case VHOST_USER_SET_OWNER:
2804         case VHOST_USER_SET_MEM_TABLE:
2805         case VHOST_USER_SET_LOG_BASE:
2806         case VHOST_USER_SET_LOG_FD:
2807         case VHOST_USER_SET_VRING_NUM:
2808         case VHOST_USER_SET_VRING_ADDR:
2809         case VHOST_USER_SET_VRING_BASE:
2810         case VHOST_USER_SET_VRING_KICK:
2811         case VHOST_USER_SET_VRING_CALL:
2812         case VHOST_USER_SET_VRING_ERR:
2813         case VHOST_USER_SET_VRING_ENABLE:
2814         case VHOST_USER_SEND_RARP:
2815         case VHOST_USER_NET_SET_MTU:
2816         case VHOST_USER_SET_SLAVE_REQ_FD:
2817                 if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2818                         vhost_user_lock_all_queue_pairs(dev);
2819                         unlock_required = 1;
2820                 }
2821                 break;
2822         default:
2823                 break;
2824
2825         }
2826
2827         handled = false;
2828         if (dev->extern_ops.pre_msg_handle) {
2829                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
2830                                 (void *)&msg);
2831                 switch (ret) {
2832                 case RTE_VHOST_MSG_RESULT_REPLY:
2833                         send_vhost_reply(fd, &msg);
2834                         /* Fall-through */
2835                 case RTE_VHOST_MSG_RESULT_ERR:
2836                 case RTE_VHOST_MSG_RESULT_OK:
2837                         handled = true;
2838                         goto skip_to_post_handle;
2839                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2840                 default:
2841                         break;
2842                 }
2843         }
2844
2845         if (request > VHOST_USER_NONE && request < VHOST_USER_MAX) {
2846                 if (!vhost_message_handlers[request])
2847                         goto skip_to_post_handle;
2848                 ret = vhost_message_handlers[request](&dev, &msg, fd);
2849
2850                 switch (ret) {
2851                 case RTE_VHOST_MSG_RESULT_ERR:
2852                         VHOST_LOG_CONFIG(ERR,
2853                                 "Processing %s failed.\n",
2854                                 vhost_message_str[request]);
2855                         handled = true;
2856                         break;
2857                 case RTE_VHOST_MSG_RESULT_OK:
2858                         VHOST_LOG_CONFIG(DEBUG,
2859                                 "Processing %s succeeded.\n",
2860                                 vhost_message_str[request]);
2861                         handled = true;
2862                         break;
2863                 case RTE_VHOST_MSG_RESULT_REPLY:
2864                         VHOST_LOG_CONFIG(DEBUG,
2865                                 "Processing %s succeeded and needs reply.\n",
2866                                 vhost_message_str[request]);
2867                         send_vhost_reply(fd, &msg);
2868                         handled = true;
2869                         break;
2870                 default:
2871                         break;
2872                 }
2873         }
2874
2875 skip_to_post_handle:
2876         if (ret != RTE_VHOST_MSG_RESULT_ERR &&
2877                         dev->extern_ops.post_msg_handle) {
2878                 ret = (*dev->extern_ops.post_msg_handle)(dev->vid,
2879                                 (void *)&msg);
2880                 switch (ret) {
2881                 case RTE_VHOST_MSG_RESULT_REPLY:
2882                         send_vhost_reply(fd, &msg);
2883                         /* Fall-through */
2884                 case RTE_VHOST_MSG_RESULT_ERR:
2885                 case RTE_VHOST_MSG_RESULT_OK:
2886                         handled = true;
2887                 case RTE_VHOST_MSG_RESULT_NOT_HANDLED:
2888                 default:
2889                         break;
2890                 }
2891         }
2892
2893         if (unlock_required)
2894                 vhost_user_unlock_all_queue_pairs(dev);
2895
2896         /* If message was not handled at this stage, treat it as an error */
2897         if (!handled) {
2898                 VHOST_LOG_CONFIG(ERR,
2899                         "vhost message (req: %d) was not handled.\n", request);
2900                 close_msg_fds(&msg);
2901                 ret = RTE_VHOST_MSG_RESULT_ERR;
2902         }
2903
2904         /*
2905          * If the request required a reply that was already sent,
2906          * this optional reply-ack won't be sent as the
2907          * VHOST_USER_NEED_REPLY was cleared in send_vhost_reply().
2908          */
2909         if (msg.flags & VHOST_USER_NEED_REPLY) {
2910                 msg.payload.u64 = ret == RTE_VHOST_MSG_RESULT_ERR;
2911                 msg.size = sizeof(msg.payload.u64);
2912                 msg.fd_num = 0;
2913                 send_vhost_reply(fd, &msg);
2914         } else if (ret == RTE_VHOST_MSG_RESULT_ERR) {
2915                 VHOST_LOG_CONFIG(ERR,
2916                         "vhost message handling failed.\n");
2917                 return -1;
2918         }
2919
2920         for (i = 0; i < dev->nr_vring; i++) {
2921                 struct vhost_virtqueue *vq = dev->virtqueue[i];
2922                 bool cur_ready = vq_is_ready(dev, vq);
2923
2924                 if (cur_ready != (vq && vq->ready)) {
2925                         vq->ready = cur_ready;
2926                         vhost_user_notify_queue_state(dev, i, cur_ready);
2927                 }
2928         }
2929
2930
2931         if (!virtio_is_ready(dev))
2932                 goto out;
2933
2934         /*
2935          * Virtio is now ready. If not done already, it is time
2936          * to notify the application it can process the rings and
2937          * configure the vDPA device if present.
2938          */
2939
2940         if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
2941                 if (dev->notify_ops->new_device(dev->vid) == 0)
2942                         dev->flags |= VIRTIO_DEV_RUNNING;
2943         }
2944
2945         vdpa_dev = dev->vdpa_dev;
2946         if (!vdpa_dev)
2947                 goto out;
2948
2949         if (!(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED)) {
2950                 if (vdpa_dev->ops->dev_conf(dev->vid))
2951                         VHOST_LOG_CONFIG(ERR,
2952                                          "Failed to configure vDPA device\n");
2953                 else
2954                         dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
2955         }
2956
2957 out:
2958         return 0;
2959 }
2960
2961 static int process_slave_message_reply(struct virtio_net *dev,
2962                                        const struct VhostUserMsg *msg)
2963 {
2964         struct VhostUserMsg msg_reply;
2965         int ret;
2966
2967         if ((msg->flags & VHOST_USER_NEED_REPLY) == 0)
2968                 return 0;
2969
2970         ret = read_vhost_message(dev->slave_req_fd, &msg_reply);
2971         if (ret <= 0) {
2972                 if (ret < 0)
2973                         VHOST_LOG_CONFIG(ERR,
2974                                 "vhost read slave message reply failed\n");
2975                 else
2976                         VHOST_LOG_CONFIG(INFO,
2977                                 "vhost peer closed\n");
2978                 ret = -1;
2979                 goto out;
2980         }
2981
2982         ret = 0;
2983         if (msg_reply.request.slave != msg->request.slave) {
2984                 VHOST_LOG_CONFIG(ERR,
2985                         "Received unexpected msg type (%u), expected %u\n",
2986                         msg_reply.request.slave, msg->request.slave);
2987                 ret = -1;
2988                 goto out;
2989         }
2990
2991         ret = msg_reply.payload.u64 ? -1 : 0;
2992
2993 out:
2994         rte_spinlock_unlock(&dev->slave_req_lock);
2995         return ret;
2996 }
2997
2998 int
2999 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
3000 {
3001         int ret;
3002         struct VhostUserMsg msg = {
3003                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
3004                 .flags = VHOST_USER_VERSION,
3005                 .size = sizeof(msg.payload.iotlb),
3006                 .payload.iotlb = {
3007                         .iova = iova,
3008                         .perm = perm,
3009                         .type = VHOST_IOTLB_MISS,
3010                 },
3011         };
3012
3013         ret = send_vhost_message(dev->slave_req_fd, &msg);
3014         if (ret < 0) {
3015                 VHOST_LOG_CONFIG(ERR,
3016                                 "Failed to send IOTLB miss message (%d)\n",
3017                                 ret);
3018                 return ret;
3019         }
3020
3021         return 0;
3022 }
3023
3024 static int
3025 vhost_user_slave_config_change(struct virtio_net *dev, bool need_reply)
3026 {
3027         int ret;
3028         struct VhostUserMsg msg = {
3029                 .request.slave = VHOST_USER_SLAVE_CONFIG_CHANGE_MSG,
3030                 .flags = VHOST_USER_VERSION,
3031                 .size = 0,
3032         };
3033
3034         if (need_reply)
3035                 msg.flags |= VHOST_USER_NEED_REPLY;
3036
3037         ret = send_vhost_slave_message(dev, &msg);
3038         if (ret < 0) {
3039                 VHOST_LOG_CONFIG(ERR,
3040                                 "Failed to send config change (%d)\n",
3041                                 ret);
3042                 return ret;
3043         }
3044
3045         return process_slave_message_reply(dev, &msg);
3046 }
3047
3048 int
3049 rte_vhost_slave_config_change(int vid, bool need_reply)
3050 {
3051         struct virtio_net *dev;
3052
3053         dev = get_device(vid);
3054         if (!dev)
3055                 return -ENODEV;
3056
3057         return vhost_user_slave_config_change(dev, need_reply);
3058 }
3059
3060 static int vhost_user_slave_set_vring_host_notifier(struct virtio_net *dev,
3061                                                     int index, int fd,
3062                                                     uint64_t offset,
3063                                                     uint64_t size)
3064 {
3065         int ret;
3066         struct VhostUserMsg msg = {
3067                 .request.slave = VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG,
3068                 .flags = VHOST_USER_VERSION | VHOST_USER_NEED_REPLY,
3069                 .size = sizeof(msg.payload.area),
3070                 .payload.area = {
3071                         .u64 = index & VHOST_USER_VRING_IDX_MASK,
3072                         .size = size,
3073                         .offset = offset,
3074                 },
3075         };
3076
3077         if (fd < 0)
3078                 msg.payload.area.u64 |= VHOST_USER_VRING_NOFD_MASK;
3079         else {
3080                 msg.fds[0] = fd;
3081                 msg.fd_num = 1;
3082         }
3083
3084         ret = send_vhost_slave_message(dev, &msg);
3085         if (ret < 0) {
3086                 VHOST_LOG_CONFIG(ERR,
3087                         "Failed to set host notifier (%d)\n", ret);
3088                 return ret;
3089         }
3090
3091         return process_slave_message_reply(dev, &msg);
3092 }
3093
3094 int rte_vhost_host_notifier_ctrl(int vid, uint16_t qid, bool enable)
3095 {
3096         struct virtio_net *dev;
3097         struct rte_vdpa_device *vdpa_dev;
3098         int vfio_device_fd, ret = 0;
3099         uint64_t offset, size;
3100         unsigned int i, q_start, q_last;
3101
3102         dev = get_device(vid);
3103         if (!dev)
3104                 return -ENODEV;
3105
3106         vdpa_dev = dev->vdpa_dev;
3107         if (vdpa_dev == NULL)
3108                 return -ENODEV;
3109
3110         if (!(dev->features & (1ULL << VIRTIO_F_VERSION_1)) ||
3111             !(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)) ||
3112             !(dev->protocol_features &
3113                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ)) ||
3114             !(dev->protocol_features &
3115                         (1ULL << VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD)) ||
3116             !(dev->protocol_features &
3117                         (1ULL << VHOST_USER_PROTOCOL_F_HOST_NOTIFIER)))
3118                 return -ENOTSUP;
3119
3120         if (qid == RTE_VHOST_QUEUE_ALL) {
3121                 q_start = 0;
3122                 q_last = dev->nr_vring - 1;
3123         } else {
3124                 if (qid >= dev->nr_vring)
3125                         return -EINVAL;
3126                 q_start = qid;
3127                 q_last = qid;
3128         }
3129
3130         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_vfio_device_fd, -ENOTSUP);
3131         RTE_FUNC_PTR_OR_ERR_RET(vdpa_dev->ops->get_notify_area, -ENOTSUP);
3132
3133         vfio_device_fd = vdpa_dev->ops->get_vfio_device_fd(vid);
3134         if (vfio_device_fd < 0)
3135                 return -ENOTSUP;
3136
3137         if (enable) {
3138                 for (i = q_start; i <= q_last; i++) {
3139                         if (vdpa_dev->ops->get_notify_area(vid, i, &offset,
3140                                         &size) < 0) {
3141                                 ret = -ENOTSUP;
3142                                 goto disable;
3143                         }
3144
3145                         if (vhost_user_slave_set_vring_host_notifier(dev, i,
3146                                         vfio_device_fd, offset, size) < 0) {
3147                                 ret = -EFAULT;
3148                                 goto disable;
3149                         }
3150                 }
3151         } else {
3152 disable:
3153                 for (i = q_start; i <= q_last; i++) {
3154                         vhost_user_slave_set_vring_host_notifier(dev, i, -1,
3155                                         0, 0);
3156                 }
3157         }
3158
3159         return ret;
3160 }