lib/librte_vhost/vhost_user.c

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