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 <sys/mman.h>
  28 #include <sys/types.h>
  29 #include <sys/stat.h>
  30 #include <assert.h>
  31 #ifdef RTE_LIBRTE_VHOST_NUMA
  32 #include <numaif.h>
  33 #endif
  34
  35 #include <rte_common.h>
  36 #include <rte_malloc.h>
  37 #include <rte_log.h>
  38
  39 #include "iotlb.h"
  40 #include "vhost.h"
  41 #include "vhost_user.h"
  42
  43 #define VIRTIO_MIN_MTU 68
  44 #define VIRTIO_MAX_MTU 65535
  45
  46 static const char *vhost_message_str[VHOST_USER_MAX] = {
  47         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  48         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  49         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  50         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  51         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  52         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  53         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  54         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  55         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  56         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  57         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  58         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  59         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  60         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  61         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  62         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  63         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  64         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  65         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  66         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  67         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  68         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  69         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  70 };
  71
  72 static uint64_t
  73 get_blk_size(int fd)
  74 {
  75         struct stat stat;
  76         int ret;
  77
  78         ret = fstat(fd, &stat);
  79         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  80 }
  81
  82 static void
  83 free_mem_region(struct virtio_net *dev)
  84 {
  85         uint32_t i;
  86         struct rte_vhost_mem_region *reg;
  87
  88         if (!dev || !dev->mem)
  89                 return;
  90
  91         for (i = 0; i < dev->mem->nregions; i++) {
  92                 reg = &dev->mem->regions[i];
  93                 if (reg->host_user_addr) {
  94                         munmap(reg->mmap_addr, reg->mmap_size);
  95                         close(reg->fd);
  96                 }
  97         }
  98 }
  99
 100 void
 101 vhost_backend_cleanup(struct virtio_net *dev)
 102 {
 103         if (dev->mem) {
 104                 free_mem_region(dev);
 105                 rte_free(dev->mem);
 106                 dev->mem = NULL;
 107         }
 108
 109         free(dev->guest_pages);
 110         dev->guest_pages = NULL;
 111
 112         if (dev->log_addr) {
 113                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 114                 dev->log_addr = 0;
 115         }
 116
 117         if (dev->slave_req_fd >= 0) {
 118                 close(dev->slave_req_fd);
 119                 dev->slave_req_fd = -1;
 120         }
 121 }
 122
 123 /*
 124  * This function just returns success at the moment unless
 125  * the device hasn't been initialised.
 126  */
 127 static int
 128 vhost_user_set_owner(void)
 129 {
 130         return 0;
 131 }
 132
 133 static int
 134 vhost_user_reset_owner(struct virtio_net *dev)
 135 {
 136         struct rte_vdpa_device *vdpa_dev;
 137         int did = -1;
 138
 139         if (dev->flags & VIRTIO_DEV_RUNNING) {
 140                 did = dev->vdpa_dev_id;
 141                 vdpa_dev = rte_vdpa_get_device(did);
 142                 if (vdpa_dev && vdpa_dev->ops->dev_close)
 143                         vdpa_dev->ops->dev_close(dev->vid);
 144                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 145                 dev->notify_ops->destroy_device(dev->vid);
 146         }
 147
 148         cleanup_device(dev, 0);
 149         reset_device(dev);
 150         return 0;
 151 }
 152
 153 /*
 154  * The features that we support are requested.
 155  */
 156 static uint64_t
 157 vhost_user_get_features(struct virtio_net *dev)
 158 {
 159         uint64_t features = 0;
 160
 161         rte_vhost_driver_get_features(dev->ifname, &features);
 162         return features;
 163 }
 164
 165 /*
 166  * The queue number that we support are requested.
 167  */
 168 static uint32_t
 169 vhost_user_get_queue_num(struct virtio_net *dev)
 170 {
 171         uint32_t queue_num = 0;
 172
 173         rte_vhost_driver_get_queue_num(dev->ifname, &queue_num);
 174         return queue_num;
 175 }
 176
 177 /*
 178  * We receive the negotiated features supported by us and the virtio device.
 179  */
 180 static int
 181 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 182 {
 183         uint64_t vhost_features = 0;
 184         struct rte_vdpa_device *vdpa_dev;
 185         int did = -1;
 186
 187         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 188         if (features & ~vhost_features) {
 189                 RTE_LOG(ERR, VHOST_CONFIG,
 190                         "(%d) received invalid negotiated features.\n",
 191                         dev->vid);
 192                 return -1;
 193         }
 194
 195         if (dev->flags & VIRTIO_DEV_RUNNING) {
 196                 if (dev->features == features)
 197                         return 0;
 198
 199                 /*
 200                  * Error out if master tries to change features while device is
 201                  * in running state. The exception being VHOST_F_LOG_ALL, which
 202                  * is enabled when the live-migration starts.
 203                  */
 204                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 205                         RTE_LOG(ERR, VHOST_CONFIG,
 206                                 "(%d) features changed while device is running.\n",
 207                                 dev->vid);
 208                         return -1;
 209                 }
 210
 211                 if (dev->notify_ops->features_changed)
 212                         dev->notify_ops->features_changed(dev->vid, features);
 213         }
 214
 215         did = dev->vdpa_dev_id;
 216         vdpa_dev = rte_vdpa_get_device(did);
 217         if (vdpa_dev && vdpa_dev->ops->set_features)
 218                 vdpa_dev->ops->set_features(dev->vid);
 219
 220         dev->features = features;
 221         if (dev->features &
 222                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 223                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 224         } else {
 225                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 226         }
 227         VHOST_LOG_DEBUG(VHOST_CONFIG,
 228                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 229                 dev->vid,
 230                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 231                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 232
 233         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 234             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 235                 /*
 236                  * Remove all but first queue pair if MQ hasn't been
 237                  * negotiated. This is safe because the device is not
 238                  * running at this stage.
 239                  */
 240                 while (dev->nr_vring > 2) {
 241                         struct vhost_virtqueue *vq;
 242
 243                         vq = dev->virtqueue[--dev->nr_vring];
 244                         if (!vq)
 245                                 continue;
 246
 247                         dev->virtqueue[dev->nr_vring] = NULL;
 248                         cleanup_vq(vq, 1);
 249                         free_vq(vq);
 250                 }
 251         }
 252
 253         return 0;
 254 }
 255
 256 /*
 257  * The virtio device sends us the size of the descriptor ring.
 258  */
 259 static int
 260 vhost_user_set_vring_num(struct virtio_net *dev,
 261                          VhostUserMsg *msg)
 262 {
 263         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 264
 265         vq->size = msg->payload.state.num;
 266
 267         /* VIRTIO 1.0, 2.4 Virtqueues says:
 268          *
 269          *   Queue Size value is always a power of 2. The maximum Queue Size
 270          *   value is 32768.
 271          */
 272         if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
 273                 RTE_LOG(ERR, VHOST_CONFIG,
 274                         "invalid virtqueue size %u\n", vq->size);
 275                 return -1;
 276         }
 277
 278         if (dev->dequeue_zero_copy) {
 279                 vq->nr_zmbuf = 0;
 280                 vq->last_zmbuf_idx = 0;
 281                 vq->zmbuf_size = vq->size;
 282                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 283                                          sizeof(struct zcopy_mbuf), 0);
 284                 if (vq->zmbufs == NULL) {
 285                         RTE_LOG(WARNING, VHOST_CONFIG,
 286                                 "failed to allocate mem for zero copy; "
 287                                 "zero copy is force disabled\n");
 288                         dev->dequeue_zero_copy = 0;
 289                 }
 290                 TAILQ_INIT(&vq->zmbuf_list);
 291         }
 292
 293         vq->shadow_used_ring = rte_malloc(NULL,
 294                                 vq->size * sizeof(struct vring_used_elem),
 295                                 RTE_CACHE_LINE_SIZE);
 296         if (!vq->shadow_used_ring) {
 297                 RTE_LOG(ERR, VHOST_CONFIG,
 298                         "failed to allocate memory for shadow used ring.\n");
 299                 return -1;
 300         }
 301
 302         vq->batch_copy_elems = rte_malloc(NULL,
 303                                 vq->size * sizeof(struct batch_copy_elem),
 304                                 RTE_CACHE_LINE_SIZE);
 305         if (!vq->batch_copy_elems) {
 306                 RTE_LOG(ERR, VHOST_CONFIG,
 307                         "failed to allocate memory for batching copy.\n");
 308                 return -1;
 309         }
 310
 311         return 0;
 312 }
 313
 314 /*
 315  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 316  * same numa node as the memory of vring descriptor.
 317  */
 318 #ifdef RTE_LIBRTE_VHOST_NUMA
 319 static struct virtio_net*
 320 numa_realloc(struct virtio_net *dev, int index)
 321 {
 322         int oldnode, newnode;
 323         struct virtio_net *old_dev;
 324         struct vhost_virtqueue *old_vq, *vq;
 325         struct zcopy_mbuf *new_zmbuf;
 326         struct vring_used_elem *new_shadow_used_ring;
 327         struct batch_copy_elem *new_batch_copy_elems;
 328         int ret;
 329
 330         old_dev = dev;
 331         vq = old_vq = dev->virtqueue[index];
 332
 333         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 334                             MPOL_F_NODE | MPOL_F_ADDR);
 335
 336         /* check if we need to reallocate vq */
 337         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 338                              MPOL_F_NODE | MPOL_F_ADDR);
 339         if (ret) {
 340                 RTE_LOG(ERR, VHOST_CONFIG,
 341                         "Unable to get vq numa information.\n");
 342                 return dev;
 343         }
 344         if (oldnode != newnode) {
 345                 RTE_LOG(INFO, VHOST_CONFIG,
 346                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 347                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 348                 if (!vq)
 349                         return dev;
 350
 351                 memcpy(vq, old_vq, sizeof(*vq));
 352                 TAILQ_INIT(&vq->zmbuf_list);
 353
 354                 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 355                         sizeof(struct zcopy_mbuf), 0, newnode);
 356                 if (new_zmbuf) {
 357                         rte_free(vq->zmbufs);
 358                         vq->zmbufs = new_zmbuf;
 359                 }
 360
 361                 new_shadow_used_ring = rte_malloc_socket(NULL,
 362                         vq->size * sizeof(struct vring_used_elem),
 363                         RTE_CACHE_LINE_SIZE,
 364                         newnode);
 365                 if (new_shadow_used_ring) {
 366                         rte_free(vq->shadow_used_ring);
 367                         vq->shadow_used_ring = new_shadow_used_ring;
 368                 }
 369
 370                 new_batch_copy_elems = rte_malloc_socket(NULL,
 371                         vq->size * sizeof(struct batch_copy_elem),
 372                         RTE_CACHE_LINE_SIZE,
 373                         newnode);
 374                 if (new_batch_copy_elems) {
 375                         rte_free(vq->batch_copy_elems);
 376                         vq->batch_copy_elems = new_batch_copy_elems;
 377                 }
 378
 379                 rte_free(old_vq);
 380         }
 381
 382         /* check if we need to reallocate dev */
 383         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 384                             MPOL_F_NODE | MPOL_F_ADDR);
 385         if (ret) {
 386                 RTE_LOG(ERR, VHOST_CONFIG,
 387                         "Unable to get dev numa information.\n");
 388                 goto out;
 389         }
 390         if (oldnode != newnode) {
 391                 RTE_LOG(INFO, VHOST_CONFIG,
 392                         "reallocate dev from %d to %d node\n",
 393                         oldnode, newnode);
 394                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 395                 if (!dev) {
 396                         dev = old_dev;
 397                         goto out;
 398                 }
 399
 400                 memcpy(dev, old_dev, sizeof(*dev));
 401                 rte_free(old_dev);
 402         }
 403
 404 out:
 405         dev->virtqueue[index] = vq;
 406         vhost_devices[dev->vid] = dev;
 407
 408         if (old_vq != vq)
 409                 vhost_user_iotlb_init(dev, index);
 410
 411         return dev;
 412 }
 413 #else
 414 static struct virtio_net*
 415 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 416 {
 417         return dev;
 418 }
 419 #endif
 420
 421 /* Converts QEMU virtual address to Vhost virtual address. */
 422 static uint64_t
 423 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 424 {
 425         struct rte_vhost_mem_region *reg;
 426         uint32_t i;
 427
 428         /* Find the region where the address lives. */
 429         for (i = 0; i < dev->mem->nregions; i++) {
 430                 reg = &dev->mem->regions[i];
 431
 432                 if (qva >= reg->guest_user_addr &&
 433                     qva <  reg->guest_user_addr + reg->size) {
 434                         return qva - reg->guest_user_addr +
 435                                reg->host_user_addr;
 436                 }
 437         }
 438
 439         return 0;
 440 }
 441
 442
 443 /*
 444  * Converts ring address to Vhost virtual address.
 445  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 446  * else it is a QEMU virtual address.
 447  */
 448 static uint64_t
 449 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 450                 uint64_t ra, uint64_t size)
 451 {
 452         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 453                 uint64_t vva;
 454
 455                 vva = vhost_user_iotlb_cache_find(vq, ra,
 456                                         &size, VHOST_ACCESS_RW);
 457                 if (!vva)
 458                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 459
 460                 return vva;
 461         }
 462
 463         return qva_to_vva(dev, ra);
 464 }
 465
 466 static struct virtio_net *
 467 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 468 {
 469         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 470         struct vhost_vring_addr *addr = &vq->ring_addrs;
 471
 472         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 473         if (vq->desc && vq->avail && vq->used)
 474                 return dev;
 475
 476         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 477                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 478         if (vq->desc == 0) {
 479                 RTE_LOG(DEBUG, VHOST_CONFIG,
 480                         "(%d) failed to find desc ring address.\n",
 481                         dev->vid);
 482                 return dev;
 483         }
 484
 485         dev = numa_realloc(dev, vq_index);
 486         vq = dev->virtqueue[vq_index];
 487         addr = &vq->ring_addrs;
 488
 489         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 490                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 491         if (vq->avail == 0) {
 492                 RTE_LOG(DEBUG, VHOST_CONFIG,
 493                         "(%d) failed to find avail ring address.\n",
 494                         dev->vid);
 495                 return dev;
 496         }
 497
 498         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 499                         vq, addr->used_user_addr, sizeof(struct vring_used));
 500         if (vq->used == 0) {
 501                 RTE_LOG(DEBUG, VHOST_CONFIG,
 502                         "(%d) failed to find used ring address.\n",
 503                         dev->vid);
 504                 return dev;
 505         }
 506
 507         if (vq->last_used_idx != vq->used->idx) {
 508                 RTE_LOG(WARNING, VHOST_CONFIG,
 509                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 510                         "some packets maybe resent for Tx and dropped for Rx\n",
 511                         vq->last_used_idx, vq->used->idx);
 512                 vq->last_used_idx  = vq->used->idx;
 513                 vq->last_avail_idx = vq->used->idx;
 514         }
 515
 516         vq->log_guest_addr = addr->log_guest_addr;
 517
 518         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 519                         dev->vid, vq->desc);
 520         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 521                         dev->vid, vq->avail);
 522         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 523                         dev->vid, vq->used);
 524         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 525                         dev->vid, vq->log_guest_addr);
 526
 527         return dev;
 528 }
 529
 530 /*
 531  * The virtio device sends us the desc, used and avail ring addresses.
 532  * This function then converts these to our address space.
 533  */
 534 static int
 535 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 536 {
 537         struct vhost_virtqueue *vq;
 538         struct vhost_vring_addr *addr = &msg->payload.addr;
 539         struct virtio_net *dev = *pdev;
 540
 541         if (dev->mem == NULL)
 542                 return -1;
 543
 544         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 545         vq = dev->virtqueue[msg->payload.addr.index];
 546
 547         /*
 548          * Rings addresses should not be interpreted as long as the ring is not
 549          * started and enabled
 550          */
 551         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 552
 553         vring_invalidate(dev, vq);
 554
 555         if (vq->enabled && (dev->features &
 556                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 557                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 558                 if (!dev)
 559                         return -1;
 560
 561                 *pdev = dev;
 562         }
 563
 564         return 0;
 565 }
 566
 567 /*
 568  * The virtio device sends us the available ring last used index.
 569  */
 570 static int
 571 vhost_user_set_vring_base(struct virtio_net *dev,
 572                           VhostUserMsg *msg)
 573 {
 574         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 575                         msg->payload.state.num;
 576         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 577                         msg->payload.state.num;
 578
 579         return 0;
 580 }
 581
 582 static int
 583 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 584                    uint64_t host_phys_addr, uint64_t size)
 585 {
 586         struct guest_page *page, *last_page;
 587
 588         if (dev->nr_guest_pages == dev->max_guest_pages) {
 589                 dev->max_guest_pages *= 2;
 590                 dev->guest_pages = realloc(dev->guest_pages,
 591                                         dev->max_guest_pages * sizeof(*page));
 592                 if (!dev->guest_pages) {
 593                         RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
 594                         return -1;
 595                 }
 596         }
 597
 598         if (dev->nr_guest_pages > 0) {
 599                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 600                 /* merge if the two pages are continuous */
 601                 if (host_phys_addr == last_page->host_phys_addr +
 602                                       last_page->size) {
 603                         last_page->size += size;
 604                         return 0;
 605                 }
 606         }
 607
 608         page = &dev->guest_pages[dev->nr_guest_pages++];
 609         page->guest_phys_addr = guest_phys_addr;
 610         page->host_phys_addr  = host_phys_addr;
 611         page->size = size;
 612
 613         return 0;
 614 }
 615
 616 static int
 617 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 618                 uint64_t page_size)
 619 {
 620         uint64_t reg_size = reg->size;
 621         uint64_t host_user_addr  = reg->host_user_addr;
 622         uint64_t guest_phys_addr = reg->guest_phys_addr;
 623         uint64_t host_phys_addr;
 624         uint64_t size;
 625
 626         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 627         size = page_size - (guest_phys_addr & (page_size - 1));
 628         size = RTE_MIN(size, reg_size);
 629
 630         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 631                 return -1;
 632
 633         host_user_addr  += size;
 634         guest_phys_addr += size;
 635         reg_size -= size;
 636
 637         while (reg_size > 0) {
 638                 size = RTE_MIN(reg_size, page_size);
 639                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 640                                                   host_user_addr);
 641                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 642                                 size) < 0)
 643                         return -1;
 644
 645                 host_user_addr  += size;
 646                 guest_phys_addr += size;
 647                 reg_size -= size;
 648         }
 649
 650         return 0;
 651 }
 652
 653 #ifdef RTE_LIBRTE_VHOST_DEBUG
 654 /* TODO: enable it only in debug mode? */
 655 static void
 656 dump_guest_pages(struct virtio_net *dev)
 657 {
 658         uint32_t i;
 659         struct guest_page *page;
 660
 661         for (i = 0; i < dev->nr_guest_pages; i++) {
 662                 page = &dev->guest_pages[i];
 663
 664                 RTE_LOG(INFO, VHOST_CONFIG,
 665                         "guest physical page region %u\n"
 666                         "\t guest_phys_addr: %" PRIx64 "\n"
 667                         "\t host_phys_addr : %" PRIx64 "\n"
 668                         "\t size           : %" PRIx64 "\n",
 669                         i,
 670                         page->guest_phys_addr,
 671                         page->host_phys_addr,
 672                         page->size);
 673         }
 674 }
 675 #else
 676 #define dump_guest_pages(dev)
 677 #endif
 678
 679 static bool
 680 vhost_memory_changed(struct VhostUserMemory *new,
 681                      struct rte_vhost_memory *old)
 682 {
 683         uint32_t i;
 684
 685         if (new->nregions != old->nregions)
 686                 return true;
 687
 688         for (i = 0; i < new->nregions; ++i) {
 689                 VhostUserMemoryRegion *new_r = &new->regions[i];
 690                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 691
 692                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 693                         return true;
 694                 if (new_r->memory_size != old_r->size)
 695                         return true;
 696                 if (new_r->userspace_addr != old_r->guest_user_addr)
 697                         return true;
 698         }
 699
 700         return false;
 701 }
 702
 703 static int
 704 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 705 {
 706         struct VhostUserMemory memory = pmsg->payload.memory;
 707         struct rte_vhost_mem_region *reg;
 708         void *mmap_addr;
 709         uint64_t mmap_size;
 710         uint64_t mmap_offset;
 711         uint64_t alignment;
 712         uint32_t i;
 713         int populate;
 714         int fd;
 715
 716         if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
 717                 RTE_LOG(ERR, VHOST_CONFIG,
 718                         "too many memory regions (%u)\n", memory.nregions);
 719                 return -1;
 720         }
 721
 722         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 723                 RTE_LOG(INFO, VHOST_CONFIG,
 724                         "(%d) memory regions not changed\n", dev->vid);
 725
 726                 for (i = 0; i < memory.nregions; i++)
 727                         close(pmsg->fds[i]);
 728
 729                 return 0;
 730         }
 731
 732         if (dev->mem) {
 733                 free_mem_region(dev);
 734                 rte_free(dev->mem);
 735                 dev->mem = NULL;
 736         }
 737
 738         dev->nr_guest_pages = 0;
 739         if (!dev->guest_pages) {
 740                 dev->max_guest_pages = 8;
 741                 dev->guest_pages = malloc(dev->max_guest_pages *
 742                                                 sizeof(struct guest_page));
 743                 if (dev->guest_pages == NULL) {
 744                         RTE_LOG(ERR, VHOST_CONFIG,
 745                                 "(%d) failed to allocate memory "
 746                                 "for dev->guest_pages\n",
 747                                 dev->vid);
 748                         return -1;
 749                 }
 750         }
 751
 752         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 753                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 754         if (dev->mem == NULL) {
 755                 RTE_LOG(ERR, VHOST_CONFIG,
 756                         "(%d) failed to allocate memory for dev->mem\n",
 757                         dev->vid);
 758                 return -1;
 759         }
 760         dev->mem->nregions = memory.nregions;
 761
 762         for (i = 0; i < memory.nregions; i++) {
 763                 fd  = pmsg->fds[i];
 764                 reg = &dev->mem->regions[i];
 765
 766                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 767                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 768                 reg->size            = memory.regions[i].memory_size;
 769                 reg->fd              = fd;
 770
 771                 mmap_offset = memory.regions[i].mmap_offset;
 772
 773                 /* Check for memory_size + mmap_offset overflow */
 774                 if (mmap_offset >= -reg->size) {
 775                         RTE_LOG(ERR, VHOST_CONFIG,
 776                                 "mmap_offset (%#"PRIx64") and memory_size "
 777                                 "(%#"PRIx64") overflow\n",
 778                                 mmap_offset, reg->size);
 779                         goto err_mmap;
 780                 }
 781
 782                 mmap_size = reg->size + mmap_offset;
 783
 784                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 785                  * with length argument aligned with hugepagesz at older
 786                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 787                  * mmap() will fail with EINVAL.
 788                  *
 789                  * to avoid failure, make sure in caller to keep length
 790                  * aligned.
 791                  */
 792                 alignment = get_blk_size(fd);
 793                 if (alignment == (uint64_t)-1) {
 794                         RTE_LOG(ERR, VHOST_CONFIG,
 795                                 "couldn't get hugepage size through fstat\n");
 796                         goto err_mmap;
 797                 }
 798                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 799
 800                 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
 801                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 802                                  MAP_SHARED | populate, fd, 0);
 803
 804                 if (mmap_addr == MAP_FAILED) {
 805                         RTE_LOG(ERR, VHOST_CONFIG,
 806                                 "mmap region %u failed.\n", i);
 807                         goto err_mmap;
 808                 }
 809
 810                 reg->mmap_addr = mmap_addr;
 811                 reg->mmap_size = mmap_size;
 812                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 813                                       mmap_offset;
 814
 815                 if (dev->dequeue_zero_copy)
 816                         if (add_guest_pages(dev, reg, alignment) < 0) {
 817                                 RTE_LOG(ERR, VHOST_CONFIG,
 818                                         "adding guest pages to region %u failed.\n",
 819                                         i);
 820                                 goto err_mmap;
 821                         }
 822
 823                 RTE_LOG(INFO, VHOST_CONFIG,
 824                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 825                         "\t guest physical addr: 0x%" PRIx64 "\n"
 826                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 827                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 828                         "\t mmap addr : 0x%" PRIx64 "\n"
 829                         "\t mmap size : 0x%" PRIx64 "\n"
 830                         "\t mmap align: 0x%" PRIx64 "\n"
 831                         "\t mmap off  : 0x%" PRIx64 "\n",
 832                         i, reg->size,
 833                         reg->guest_phys_addr,
 834                         reg->guest_user_addr,
 835                         reg->host_user_addr,
 836                         (uint64_t)(uintptr_t)mmap_addr,
 837                         mmap_size,
 838                         alignment,
 839                         mmap_offset);
 840         }
 841
 842         dump_guest_pages(dev);
 843
 844         return 0;
 845
 846 err_mmap:
 847         free_mem_region(dev);
 848         rte_free(dev->mem);
 849         dev->mem = NULL;
 850         return -1;
 851 }
 852
 853 static int
 854 vq_is_ready(struct vhost_virtqueue *vq)
 855 {
 856         return vq && vq->desc && vq->avail && vq->used &&
 857                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 858                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 859 }
 860
 861 static int
 862 virtio_is_ready(struct virtio_net *dev)
 863 {
 864         struct vhost_virtqueue *vq;
 865         uint32_t i;
 866
 867         if (dev->nr_vring == 0)
 868                 return 0;
 869
 870         for (i = 0; i < dev->nr_vring; i++) {
 871                 vq = dev->virtqueue[i];
 872
 873                 if (!vq_is_ready(vq))
 874                         return 0;
 875         }
 876
 877         RTE_LOG(INFO, VHOST_CONFIG,
 878                 "virtio is now ready for processing.\n");
 879         return 1;
 880 }
 881
 882 static void
 883 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 884 {
 885         struct vhost_vring_file file;
 886         struct vhost_virtqueue *vq;
 887
 888         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 889         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 890                 file.fd = VIRTIO_INVALID_EVENTFD;
 891         else
 892                 file.fd = pmsg->fds[0];
 893         RTE_LOG(INFO, VHOST_CONFIG,
 894                 "vring call idx:%d file:%d\n", file.index, file.fd);
 895
 896         vq = dev->virtqueue[file.index];
 897         if (vq->callfd >= 0)
 898                 close(vq->callfd);
 899
 900         vq->callfd = file.fd;
 901 }
 902
 903 static void
 904 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 905 {
 906         struct vhost_vring_file file;
 907         struct vhost_virtqueue *vq;
 908         struct virtio_net *dev = *pdev;
 909
 910         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 911         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 912                 file.fd = VIRTIO_INVALID_EVENTFD;
 913         else
 914                 file.fd = pmsg->fds[0];
 915         RTE_LOG(INFO, VHOST_CONFIG,
 916                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 917
 918         /* Interpret ring addresses only when ring is started. */
 919         dev = translate_ring_addresses(dev, file.index);
 920         if (!dev)
 921                 return;
 922
 923         *pdev = dev;
 924
 925         vq = dev->virtqueue[file.index];
 926
 927         /*
 928          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 929          * the ring starts already enabled. Otherwise, it is enabled via
 930          * the SET_VRING_ENABLE message.
 931          */
 932         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 933                 vq->enabled = 1;
 934
 935         if (vq->kickfd >= 0)
 936                 close(vq->kickfd);
 937         vq->kickfd = file.fd;
 938 }
 939
 940 static void
 941 free_zmbufs(struct vhost_virtqueue *vq)
 942 {
 943         struct zcopy_mbuf *zmbuf, *next;
 944
 945         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 946              zmbuf != NULL; zmbuf = next) {
 947                 next = TAILQ_NEXT(zmbuf, next);
 948
 949                 rte_pktmbuf_free(zmbuf->mbuf);
 950                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 951         }
 952
 953         rte_free(vq->zmbufs);
 954 }
 955
 956 /*
 957  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 958  */
 959 static int
 960 vhost_user_get_vring_base(struct virtio_net *dev,
 961                           VhostUserMsg *msg)
 962 {
 963         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 964         struct rte_vdpa_device *vdpa_dev;
 965         int did = -1;
 966
 967         /* We have to stop the queue (virtio) if it is running. */
 968         if (dev->flags & VIRTIO_DEV_RUNNING) {
 969                 did = dev->vdpa_dev_id;
 970                 vdpa_dev = rte_vdpa_get_device(did);
 971                 if (vdpa_dev && vdpa_dev->ops->dev_close)
 972                         vdpa_dev->ops->dev_close(dev->vid);
 973                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 974                 dev->notify_ops->destroy_device(dev->vid);
 975         }
 976
 977         dev->flags &= ~VIRTIO_DEV_READY;
 978         dev->flags &= ~VIRTIO_DEV_VDPA_CONFIGURED;
 979
 980         /* Here we are safe to get the last avail index */
 981         msg->payload.state.num = vq->last_avail_idx;
 982
 983         RTE_LOG(INFO, VHOST_CONFIG,
 984                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 985                 msg->payload.state.num);
 986         /*
 987          * Based on current qemu vhost-user implementation, this message is
 988          * sent and only sent in vhost_vring_stop.
 989          * TODO: cleanup the vring, it isn't usable since here.
 990          */
 991         if (vq->kickfd >= 0)
 992                 close(vq->kickfd);
 993
 994         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 995
 996         if (vq->callfd >= 0)
 997                 close(vq->callfd);
 998
 999         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
1000
1001         if (dev->dequeue_zero_copy)
1002                 free_zmbufs(vq);
1003         rte_free(vq->shadow_used_ring);
1004         vq->shadow_used_ring = NULL;
1005
1006         rte_free(vq->batch_copy_elems);
1007         vq->batch_copy_elems = NULL;
1008
1009         return 0;
1010 }
1011
1012 /*
1013  * when virtio queues are ready to work, qemu will send us to
1014  * enable the virtio queue pair.
1015  */
1016 static int
1017 vhost_user_set_vring_enable(struct virtio_net *dev,
1018                             VhostUserMsg *msg)
1019 {
1020         int enable = (int)msg->payload.state.num;
1021         int index = (int)msg->payload.state.index;
1022         struct rte_vdpa_device *vdpa_dev;
1023         int did = -1;
1024
1025         RTE_LOG(INFO, VHOST_CONFIG,
1026                 "set queue enable: %d to qp idx: %d\n",
1027                 enable, index);
1028
1029         did = dev->vdpa_dev_id;
1030         vdpa_dev = rte_vdpa_get_device(did);
1031         if (vdpa_dev && vdpa_dev->ops->set_vring_state)
1032                 vdpa_dev->ops->set_vring_state(dev->vid, index, enable);
1033
1034         if (dev->notify_ops->vring_state_changed)
1035                 dev->notify_ops->vring_state_changed(dev->vid,
1036                                 index, enable);
1037
1038         dev->virtqueue[index]->enabled = enable;
1039
1040         return 0;
1041 }
1042
1043 static void
1044 vhost_user_get_protocol_features(struct virtio_net *dev,
1045                                  struct VhostUserMsg *msg)
1046 {
1047         uint64_t features, protocol_features;
1048
1049         rte_vhost_driver_get_features(dev->ifname, &features);
1050         rte_vhost_driver_get_protocol_features(dev->ifname, &protocol_features);
1051
1052         /*
1053          * REPLY_ACK protocol feature is only mandatory for now
1054          * for IOMMU feature. If IOMMU is explicitly disabled by the
1055          * application, disable also REPLY_ACK feature for older buggy
1056          * Qemu versions (from v2.7.0 to v2.9.0).
1057          */
1058         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1059                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1060
1061         msg->payload.u64 = protocol_features;
1062         msg->size = sizeof(msg->payload.u64);
1063 }
1064
1065 static void
1066 vhost_user_set_protocol_features(struct virtio_net *dev,
1067                                  uint64_t protocol_features)
1068 {
1069         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1070                 return;
1071
1072         dev->protocol_features = protocol_features;
1073 }
1074
1075 static int
1076 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1077 {
1078         int fd = msg->fds[0];
1079         uint64_t size, off;
1080         void *addr;
1081
1082         if (fd < 0) {
1083                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1084                 return -1;
1085         }
1086
1087         if (msg->size != sizeof(VhostUserLog)) {
1088                 RTE_LOG(ERR, VHOST_CONFIG,
1089                         "invalid log base msg size: %"PRId32" != %d\n",
1090                         msg->size, (int)sizeof(VhostUserLog));
1091                 return -1;
1092         }
1093
1094         size = msg->payload.log.mmap_size;
1095         off  = msg->payload.log.mmap_offset;
1096
1097         /* Don't allow mmap_offset to point outside the mmap region */
1098         if (off > size) {
1099                 RTE_LOG(ERR, VHOST_CONFIG,
1100                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1101                         off, size);
1102                 return -1;
1103         }
1104
1105         RTE_LOG(INFO, VHOST_CONFIG,
1106                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1107                 size, off);
1108
1109         /*
1110          * mmap from 0 to workaround a hugepage mmap bug: mmap will
1111          * fail when offset is not page size aligned.
1112          */
1113         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1114         close(fd);
1115         if (addr == MAP_FAILED) {
1116                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1117                 return -1;
1118         }
1119
1120         /*
1121          * Free previously mapped log memory on occasionally
1122          * multiple VHOST_USER_SET_LOG_BASE.
1123          */
1124         if (dev->log_addr) {
1125                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1126         }
1127         dev->log_addr = (uint64_t)(uintptr_t)addr;
1128         dev->log_base = dev->log_addr + off;
1129         dev->log_size = size;
1130
1131         return 0;
1132 }
1133
1134 /*
1135  * An rarp packet is constructed and broadcasted to notify switches about
1136  * the new location of the migrated VM, so that packets from outside will
1137  * not be lost after migration.
1138  *
1139  * However, we don't actually "send" a rarp packet here, instead, we set
1140  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1141  */
1142 static int
1143 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1144 {
1145         uint8_t *mac = (uint8_t *)&msg->payload.u64;
1146         struct rte_vdpa_device *vdpa_dev;
1147         int did = -1;
1148
1149         RTE_LOG(DEBUG, VHOST_CONFIG,
1150                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1151                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1152         memcpy(dev->mac.addr_bytes, mac, 6);
1153
1154         /*
1155          * Set the flag to inject a RARP broadcast packet at
1156          * rte_vhost_dequeue_burst().
1157          *
1158          * rte_smp_wmb() is for making sure the mac is copied
1159          * before the flag is set.
1160          */
1161         rte_smp_wmb();
1162         rte_atomic16_set(&dev->broadcast_rarp, 1);
1163         did = dev->vdpa_dev_id;
1164         vdpa_dev = rte_vdpa_get_device(did);
1165         if (vdpa_dev && vdpa_dev->ops->migration_done)
1166                 vdpa_dev->ops->migration_done(dev->vid);
1167
1168         return 0;
1169 }
1170
1171 static int
1172 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1173 {
1174         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1175                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1176                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1177                                 msg->payload.u64);
1178
1179                 return -1;
1180         }
1181
1182         dev->mtu = msg->payload.u64;
1183
1184         return 0;
1185 }
1186
1187 static int
1188 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1189 {
1190         int fd = msg->fds[0];
1191
1192         if (fd < 0) {
1193                 RTE_LOG(ERR, VHOST_CONFIG,
1194                                 "Invalid file descriptor for slave channel (%d)\n",
1195                                 fd);
1196                 return -1;
1197         }
1198
1199         dev->slave_req_fd = fd;
1200
1201         return 0;
1202 }
1203
1204 static int
1205 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1206 {
1207         struct vhost_vring_addr *ra;
1208         uint64_t start, end;
1209
1210         start = imsg->iova;
1211         end = start + imsg->size;
1212
1213         ra = &vq->ring_addrs;
1214         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1215                 return 1;
1216         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1217                 return 1;
1218         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1219                 return 1;
1220
1221         return 0;
1222 }
1223
1224 static int
1225 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1226                                 struct vhost_iotlb_msg *imsg)
1227 {
1228         uint64_t istart, iend, vstart, vend;
1229
1230         istart = imsg->iova;
1231         iend = istart + imsg->size - 1;
1232
1233         vstart = (uintptr_t)vq->desc;
1234         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1235         if (vstart <= iend && istart <= vend)
1236                 return 1;
1237
1238         vstart = (uintptr_t)vq->avail;
1239         vend = vstart + sizeof(struct vring_avail);
1240         vend += sizeof(uint16_t) * vq->size - 1;
1241         if (vstart <= iend && istart <= vend)
1242                 return 1;
1243
1244         vstart = (uintptr_t)vq->used;
1245         vend = vstart + sizeof(struct vring_used);
1246         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1247         if (vstart <= iend && istart <= vend)
1248                 return 1;
1249
1250         return 0;
1251 }
1252
1253 static int
1254 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1255 {
1256         struct virtio_net *dev = *pdev;
1257         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1258         uint16_t i;
1259         uint64_t vva;
1260
1261         switch (imsg->type) {
1262         case VHOST_IOTLB_UPDATE:
1263                 vva = qva_to_vva(dev, imsg->uaddr);
1264                 if (!vva)
1265                         return -1;
1266
1267                 for (i = 0; i < dev->nr_vring; i++) {
1268                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1269
1270                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1271                                         imsg->size, imsg->perm);
1272
1273                         if (is_vring_iotlb_update(vq, imsg))
1274                                 *pdev = dev = translate_ring_addresses(dev, i);
1275                 }
1276                 break;
1277         case VHOST_IOTLB_INVALIDATE:
1278                 for (i = 0; i < dev->nr_vring; i++) {
1279                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1280
1281                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1282                                         imsg->size);
1283
1284                         if (is_vring_iotlb_invalidate(vq, imsg))
1285                                 vring_invalidate(dev, vq);
1286                 }
1287                 break;
1288         default:
1289                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1290                                 imsg->type);
1291                 return -1;
1292         }
1293
1294         return 0;
1295 }
1296
1297 /* return bytes# of read on success or negative val on failure. */
1298 static int
1299 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1300 {
1301         int ret;
1302
1303         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1304                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1305         if (ret <= 0)
1306                 return ret;
1307
1308         if (msg && msg->size) {
1309                 if (msg->size > sizeof(msg->payload)) {
1310                         RTE_LOG(ERR, VHOST_CONFIG,
1311                                 "invalid msg size: %d\n", msg->size);
1312                         return -1;
1313                 }
1314                 ret = read(sockfd, &msg->payload, msg->size);
1315                 if (ret <= 0)
1316                         return ret;
1317                 if (ret != (int)msg->size) {
1318                         RTE_LOG(ERR, VHOST_CONFIG,
1319                                 "read control message failed\n");
1320                         return -1;
1321                 }
1322         }
1323
1324         return ret;
1325 }
1326
1327 static int
1328 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1329 {
1330         if (!msg)
1331                 return 0;
1332
1333         return send_fd_message(sockfd, (char *)msg,
1334                 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1335 }
1336
1337 static int
1338 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1339 {
1340         if (!msg)
1341                 return 0;
1342
1343         msg->flags &= ~VHOST_USER_VERSION_MASK;
1344         msg->flags &= ~VHOST_USER_NEED_REPLY;
1345         msg->flags |= VHOST_USER_VERSION;
1346         msg->flags |= VHOST_USER_REPLY_MASK;
1347
1348         return send_vhost_message(sockfd, msg, NULL, 0);
1349 }
1350
1351 /*
1352  * Allocate a queue pair if it hasn't been allocated yet
1353  */
1354 static int
1355 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1356 {
1357         uint16_t vring_idx;
1358
1359         switch (msg->request.master) {
1360         case VHOST_USER_SET_VRING_KICK:
1361         case VHOST_USER_SET_VRING_CALL:
1362         case VHOST_USER_SET_VRING_ERR:
1363                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1364                 break;
1365         case VHOST_USER_SET_VRING_NUM:
1366         case VHOST_USER_SET_VRING_BASE:
1367         case VHOST_USER_SET_VRING_ENABLE:
1368                 vring_idx = msg->payload.state.index;
1369                 break;
1370         case VHOST_USER_SET_VRING_ADDR:
1371                 vring_idx = msg->payload.addr.index;
1372                 break;
1373         default:
1374                 return 0;
1375         }
1376
1377         if (vring_idx >= VHOST_MAX_VRING) {
1378                 RTE_LOG(ERR, VHOST_CONFIG,
1379                         "invalid vring index: %u\n", vring_idx);
1380                 return -1;
1381         }
1382
1383         if (dev->virtqueue[vring_idx])
1384                 return 0;
1385
1386         return alloc_vring_queue(dev, vring_idx);
1387 }
1388
1389 static void
1390 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1391 {
1392         unsigned int i = 0;
1393         unsigned int vq_num = 0;
1394
1395         while (vq_num < dev->nr_vring) {
1396                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1397
1398                 if (vq) {
1399                         rte_spinlock_lock(&vq->access_lock);
1400                         vq_num++;
1401                 }
1402                 i++;
1403         }
1404 }
1405
1406 static void
1407 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1408 {
1409         unsigned int i = 0;
1410         unsigned int vq_num = 0;
1411
1412         while (vq_num < dev->nr_vring) {
1413                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1414
1415                 if (vq) {
1416                         rte_spinlock_unlock(&vq->access_lock);
1417                         vq_num++;
1418                 }
1419                 i++;
1420         }
1421 }
1422
1423 int
1424 vhost_user_msg_handler(int vid, int fd)
1425 {
1426         struct virtio_net *dev;
1427         struct VhostUserMsg msg;
1428         struct rte_vdpa_device *vdpa_dev;
1429         int did = -1;
1430         int ret;
1431         int unlock_required = 0;
1432         uint32_t skip_master = 0;
1433
1434         dev = get_device(vid);
1435         if (dev == NULL)
1436                 return -1;
1437
1438         if (!dev->notify_ops) {
1439                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1440                 if (!dev->notify_ops) {
1441                         RTE_LOG(ERR, VHOST_CONFIG,
1442                                 "failed to get callback ops for driver %s\n",
1443                                 dev->ifname);
1444                         return -1;
1445                 }
1446         }
1447
1448         ret = read_vhost_message(fd, &msg);
1449         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1450                 if (ret < 0)
1451                         RTE_LOG(ERR, VHOST_CONFIG,
1452                                 "vhost read message failed\n");
1453                 else if (ret == 0)
1454                         RTE_LOG(INFO, VHOST_CONFIG,
1455                                 "vhost peer closed\n");
1456                 else
1457                         RTE_LOG(ERR, VHOST_CONFIG,
1458                                 "vhost read incorrect message\n");
1459
1460                 return -1;
1461         }
1462
1463         ret = 0;
1464         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1465                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1466                         vhost_message_str[msg.request.master]);
1467         else
1468                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1469                         vhost_message_str[msg.request.master]);
1470
1471         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1472         if (ret < 0) {
1473                 RTE_LOG(ERR, VHOST_CONFIG,
1474                         "failed to alloc queue\n");
1475                 return -1;
1476         }
1477
1478         /*
1479          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1480          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1481          * and device is destroyed. destroy_device waits for queues to be
1482          * inactive, so it is safe. Otherwise taking the access_lock
1483          * would cause a dead lock.
1484          */
1485         switch (msg.request.master) {
1486         case VHOST_USER_SET_FEATURES:
1487         case VHOST_USER_SET_PROTOCOL_FEATURES:
1488         case VHOST_USER_SET_OWNER:
1489         case VHOST_USER_SET_MEM_TABLE:
1490         case VHOST_USER_SET_LOG_BASE:
1491         case VHOST_USER_SET_LOG_FD:
1492         case VHOST_USER_SET_VRING_NUM:
1493         case VHOST_USER_SET_VRING_ADDR:
1494         case VHOST_USER_SET_VRING_BASE:
1495         case VHOST_USER_SET_VRING_KICK:
1496         case VHOST_USER_SET_VRING_CALL:
1497         case VHOST_USER_SET_VRING_ERR:
1498         case VHOST_USER_SET_VRING_ENABLE:
1499         case VHOST_USER_SEND_RARP:
1500         case VHOST_USER_NET_SET_MTU:
1501         case VHOST_USER_SET_SLAVE_REQ_FD:
1502                 vhost_user_lock_all_queue_pairs(dev);
1503                 unlock_required = 1;
1504                 break;
1505         default:
1506                 break;
1507
1508         }
1509
1510         if (dev->extern_ops.pre_msg_handle) {
1511                 uint32_t need_reply;
1512
1513                 ret = (*dev->extern_ops.pre_msg_handle)(dev->vid,
1514                                 (void *)&msg, &need_reply, &skip_master);
1515                 if (ret < 0)
1516                         goto skip_to_reply;
1517
1518                 if (need_reply)
1519                         send_vhost_reply(fd, &msg);
1520
1521                 if (skip_master)
1522                         goto skip_to_post_handle;
1523         }
1524
1525         switch (msg.request.master) {
1526         case VHOST_USER_GET_FEATURES:
1527                 msg.payload.u64 = vhost_user_get_features(dev);
1528                 msg.size = sizeof(msg.payload.u64);
1529                 send_vhost_reply(fd, &msg);
1530                 break;
1531         case VHOST_USER_SET_FEATURES:
1532                 ret = vhost_user_set_features(dev, msg.payload.u64);
1533                 if (ret)
1534                         return -1;
1535                 break;
1536
1537         case VHOST_USER_GET_PROTOCOL_FEATURES:
1538                 vhost_user_get_protocol_features(dev, &msg);
1539                 send_vhost_reply(fd, &msg);
1540                 break;
1541         case VHOST_USER_SET_PROTOCOL_FEATURES:
1542                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1543                 break;
1544
1545         case VHOST_USER_SET_OWNER:
1546                 vhost_user_set_owner();
1547                 break;
1548         case VHOST_USER_RESET_OWNER:
1549                 vhost_user_reset_owner(dev);
1550                 break;
1551
1552         case VHOST_USER_SET_MEM_TABLE:
1553                 ret = vhost_user_set_mem_table(dev, &msg);
1554                 break;
1555
1556         case VHOST_USER_SET_LOG_BASE:
1557                 vhost_user_set_log_base(dev, &msg);
1558
1559                 /* it needs a reply */
1560                 msg.size = sizeof(msg.payload.u64);
1561                 send_vhost_reply(fd, &msg);
1562                 break;
1563         case VHOST_USER_SET_LOG_FD:
1564                 close(msg.fds[0]);
1565                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1566                 break;
1567
1568         case VHOST_USER_SET_VRING_NUM:
1569                 vhost_user_set_vring_num(dev, &msg);
1570                 break;
1571         case VHOST_USER_SET_VRING_ADDR:
1572                 vhost_user_set_vring_addr(&dev, &msg);
1573                 break;
1574         case VHOST_USER_SET_VRING_BASE:
1575                 vhost_user_set_vring_base(dev, &msg);
1576                 break;
1577
1578         case VHOST_USER_GET_VRING_BASE:
1579                 vhost_user_get_vring_base(dev, &msg);
1580                 msg.size = sizeof(msg.payload.state);
1581                 send_vhost_reply(fd, &msg);
1582                 break;
1583
1584         case VHOST_USER_SET_VRING_KICK:
1585                 vhost_user_set_vring_kick(&dev, &msg);
1586                 break;
1587         case VHOST_USER_SET_VRING_CALL:
1588                 vhost_user_set_vring_call(dev, &msg);
1589                 break;
1590
1591         case VHOST_USER_SET_VRING_ERR:
1592                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1593                         close(msg.fds[0]);
1594                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1595                 break;
1596
1597         case VHOST_USER_GET_QUEUE_NUM:
1598                 msg.payload.u64 = (uint64_t)vhost_user_get_queue_num(dev);
1599                 msg.size = sizeof(msg.payload.u64);
1600                 send_vhost_reply(fd, &msg);
1601                 break;
1602
1603         case VHOST_USER_SET_VRING_ENABLE:
1604                 vhost_user_set_vring_enable(dev, &msg);
1605                 break;
1606         case VHOST_USER_SEND_RARP:
1607                 vhost_user_send_rarp(dev, &msg);
1608                 break;
1609
1610         case VHOST_USER_NET_SET_MTU:
1611                 ret = vhost_user_net_set_mtu(dev, &msg);
1612                 break;
1613
1614         case VHOST_USER_SET_SLAVE_REQ_FD:
1615                 ret = vhost_user_set_req_fd(dev, &msg);
1616                 break;
1617
1618         case VHOST_USER_IOTLB_MSG:
1619                 ret = vhost_user_iotlb_msg(&dev, &msg);
1620                 break;
1621
1622         default:
1623                 ret = -1;
1624                 break;
1625         }
1626
1627 skip_to_post_handle:
1628         if (dev->extern_ops.post_msg_handle) {
1629                 uint32_t need_reply;
1630
1631                 ret = (*dev->extern_ops.post_msg_handle)(
1632                                 dev->vid, (void *)&msg, &need_reply);
1633                 if (ret < 0)
1634                         goto skip_to_reply;
1635
1636                 if (need_reply)
1637                         send_vhost_reply(fd, &msg);
1638         }
1639
1640 skip_to_reply:
1641         if (unlock_required)
1642                 vhost_user_unlock_all_queue_pairs(dev);
1643
1644         if (msg.flags & VHOST_USER_NEED_REPLY) {
1645                 msg.payload.u64 = !!ret;
1646                 msg.size = sizeof(msg.payload.u64);
1647                 send_vhost_reply(fd, &msg);
1648         }
1649
1650         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1651                 dev->flags |= VIRTIO_DEV_READY;
1652
1653                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1654                         if (dev->dequeue_zero_copy) {
1655                                 RTE_LOG(INFO, VHOST_CONFIG,
1656                                                 "dequeue zero copy is enabled\n");
1657                         }
1658
1659                         if (dev->notify_ops->new_device(dev->vid) == 0)
1660                                 dev->flags |= VIRTIO_DEV_RUNNING;
1661                 }
1662         }
1663
1664         did = dev->vdpa_dev_id;
1665         vdpa_dev = rte_vdpa_get_device(did);
1666         if (vdpa_dev && virtio_is_ready(dev) &&
1667                         !(dev->flags & VIRTIO_DEV_VDPA_CONFIGURED) &&
1668                         msg.request.master == VHOST_USER_SET_VRING_ENABLE) {
1669                 if (vdpa_dev->ops->dev_conf)
1670                         vdpa_dev->ops->dev_conf(dev->vid);
1671                 dev->flags |= VIRTIO_DEV_VDPA_CONFIGURED;
1672         }
1673
1674         return 0;
1675 }
1676
1677 int
1678 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1679 {
1680         int ret;
1681         struct VhostUserMsg msg = {
1682                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1683                 .flags = VHOST_USER_VERSION,
1684                 .size = sizeof(msg.payload.iotlb),
1685                 .payload.iotlb = {
1686                         .iova = iova,
1687                         .perm = perm,
1688                         .type = VHOST_IOTLB_MISS,
1689                 },
1690         };
1691
1692         ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1693         if (ret < 0) {
1694                 RTE_LOG(ERR, VHOST_CONFIG,
1695                                 "Failed to send IOTLB miss message (%d)\n",
1696                                 ret);
1697                 return ret;
1698         }
1699
1700         return 0;
1701 }