lib/librte_vhost/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 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         if (dev->flags & VIRTIO_DEV_RUNNING) {
 137                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 138                 dev->notify_ops->destroy_device(dev->vid);
 139         }
 140
 141         cleanup_device(dev, 0);
 142         reset_device(dev);
 143         return 0;
 144 }
 145
 146 /*
 147  * The features that we support are requested.
 148  */
 149 static uint64_t
 150 vhost_user_get_features(struct virtio_net *dev)
 151 {
 152         uint64_t features = 0;
 153
 154         rte_vhost_driver_get_features(dev->ifname, &features);
 155         return features;
 156 }
 157
 158 /*
 159  * We receive the negotiated features supported by us and the virtio device.
 160  */
 161 static int
 162 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 163 {
 164         uint64_t vhost_features = 0;
 165
 166         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 167         if (features & ~vhost_features) {
 168                 RTE_LOG(ERR, VHOST_CONFIG,
 169                         "(%d) received invalid negotiated features.\n",
 170                         dev->vid);
 171                 return -1;
 172         }
 173
 174         if (dev->flags & VIRTIO_DEV_RUNNING) {
 175                 if (dev->features == features)
 176                         return 0;
 177
 178                 /*
 179                  * Error out if master tries to change features while device is
 180                  * in running state. The exception being VHOST_F_LOG_ALL, which
 181                  * is enabled when the live-migration starts.
 182                  */
 183                 if ((dev->features ^ features) & ~(1ULL << VHOST_F_LOG_ALL)) {
 184                         RTE_LOG(ERR, VHOST_CONFIG,
 185                                 "(%d) features changed while device is running.\n",
 186                                 dev->vid);
 187                         return -1;
 188                 }
 189
 190                 if (dev->notify_ops->features_changed)
 191                         dev->notify_ops->features_changed(dev->vid, features);
 192         }
 193
 194         dev->features = features;
 195         if (dev->features &
 196                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 197                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 198         } else {
 199                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 200         }
 201         VHOST_LOG_DEBUG(VHOST_CONFIG,
 202                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 203                 dev->vid,
 204                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 205                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 206
 207         if ((dev->flags & VIRTIO_DEV_BUILTIN_VIRTIO_NET) &&
 208             !(dev->features & (1ULL << VIRTIO_NET_F_MQ))) {
 209                 /*
 210                  * Remove all but first queue pair if MQ hasn't been
 211                  * negotiated. This is safe because the device is not
 212                  * running at this stage.
 213                  */
 214                 while (dev->nr_vring > 2) {
 215                         struct vhost_virtqueue *vq;
 216
 217                         vq = dev->virtqueue[--dev->nr_vring];
 218                         if (!vq)
 219                                 continue;
 220
 221                         dev->virtqueue[dev->nr_vring] = NULL;
 222                         cleanup_vq(vq, 1);
 223                         free_vq(vq);
 224                 }
 225         }
 226
 227         return 0;
 228 }
 229
 230 /*
 231  * The virtio device sends us the size of the descriptor ring.
 232  */
 233 static int
 234 vhost_user_set_vring_num(struct virtio_net *dev,
 235                          VhostUserMsg *msg)
 236 {
 237         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 238
 239         vq->size = msg->payload.state.num;
 240
 241         /* VIRTIO 1.0, 2.4 Virtqueues says:
 242          *
 243          *   Queue Size value is always a power of 2. The maximum Queue Size
 244          *   value is 32768.
 245          */
 246         if ((vq->size & (vq->size - 1)) || vq->size > 32768) {
 247                 RTE_LOG(ERR, VHOST_CONFIG,
 248                         "invalid virtqueue size %u\n", vq->size);
 249                 return -1;
 250         }
 251
 252         if (dev->dequeue_zero_copy) {
 253                 vq->nr_zmbuf = 0;
 254                 vq->last_zmbuf_idx = 0;
 255                 vq->zmbuf_size = vq->size;
 256                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 257                                          sizeof(struct zcopy_mbuf), 0);
 258                 if (vq->zmbufs == NULL) {
 259                         RTE_LOG(WARNING, VHOST_CONFIG,
 260                                 "failed to allocate mem for zero copy; "
 261                                 "zero copy is force disabled\n");
 262                         dev->dequeue_zero_copy = 0;
 263                 }
 264                 TAILQ_INIT(&vq->zmbuf_list);
 265         }
 266
 267         vq->shadow_used_ring = rte_malloc(NULL,
 268                                 vq->size * sizeof(struct vring_used_elem),
 269                                 RTE_CACHE_LINE_SIZE);
 270         if (!vq->shadow_used_ring) {
 271                 RTE_LOG(ERR, VHOST_CONFIG,
 272                         "failed to allocate memory for shadow used ring.\n");
 273                 return -1;
 274         }
 275
 276         vq->batch_copy_elems = rte_malloc(NULL,
 277                                 vq->size * sizeof(struct batch_copy_elem),
 278                                 RTE_CACHE_LINE_SIZE);
 279         if (!vq->batch_copy_elems) {
 280                 RTE_LOG(ERR, VHOST_CONFIG,
 281                         "failed to allocate memory for batching copy.\n");
 282                 return -1;
 283         }
 284
 285         return 0;
 286 }
 287
 288 /*
 289  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 290  * same numa node as the memory of vring descriptor.
 291  */
 292 #ifdef RTE_LIBRTE_VHOST_NUMA
 293 static struct virtio_net*
 294 numa_realloc(struct virtio_net *dev, int index)
 295 {
 296         int oldnode, newnode;
 297         struct virtio_net *old_dev;
 298         struct vhost_virtqueue *old_vq, *vq;
 299         struct zcopy_mbuf *new_zmbuf;
 300         struct vring_used_elem *new_shadow_used_ring;
 301         struct batch_copy_elem *new_batch_copy_elems;
 302         int ret;
 303
 304         old_dev = dev;
 305         vq = old_vq = dev->virtqueue[index];
 306
 307         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 308                             MPOL_F_NODE | MPOL_F_ADDR);
 309
 310         /* check if we need to reallocate vq */
 311         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 312                              MPOL_F_NODE | MPOL_F_ADDR);
 313         if (ret) {
 314                 RTE_LOG(ERR, VHOST_CONFIG,
 315                         "Unable to get vq numa information.\n");
 316                 return dev;
 317         }
 318         if (oldnode != newnode) {
 319                 RTE_LOG(INFO, VHOST_CONFIG,
 320                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 321                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 322                 if (!vq)
 323                         return dev;
 324
 325                 memcpy(vq, old_vq, sizeof(*vq));
 326                 TAILQ_INIT(&vq->zmbuf_list);
 327
 328                 new_zmbuf = rte_malloc_socket(NULL, vq->zmbuf_size *
 329                         sizeof(struct zcopy_mbuf), 0, newnode);
 330                 if (new_zmbuf) {
 331                         rte_free(vq->zmbufs);
 332                         vq->zmbufs = new_zmbuf;
 333                 }
 334
 335                 new_shadow_used_ring = rte_malloc_socket(NULL,
 336                         vq->size * sizeof(struct vring_used_elem),
 337                         RTE_CACHE_LINE_SIZE,
 338                         newnode);
 339                 if (new_shadow_used_ring) {
 340                         rte_free(vq->shadow_used_ring);
 341                         vq->shadow_used_ring = new_shadow_used_ring;
 342                 }
 343
 344                 new_batch_copy_elems = rte_malloc_socket(NULL,
 345                         vq->size * sizeof(struct batch_copy_elem),
 346                         RTE_CACHE_LINE_SIZE,
 347                         newnode);
 348                 if (new_batch_copy_elems) {
 349                         rte_free(vq->batch_copy_elems);
 350                         vq->batch_copy_elems = new_batch_copy_elems;
 351                 }
 352
 353                 rte_free(old_vq);
 354         }
 355
 356         /* check if we need to reallocate dev */
 357         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 358                             MPOL_F_NODE | MPOL_F_ADDR);
 359         if (ret) {
 360                 RTE_LOG(ERR, VHOST_CONFIG,
 361                         "Unable to get dev numa information.\n");
 362                 goto out;
 363         }
 364         if (oldnode != newnode) {
 365                 RTE_LOG(INFO, VHOST_CONFIG,
 366                         "reallocate dev from %d to %d node\n",
 367                         oldnode, newnode);
 368                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 369                 if (!dev) {
 370                         dev = old_dev;
 371                         goto out;
 372                 }
 373
 374                 memcpy(dev, old_dev, sizeof(*dev));
 375                 rte_free(old_dev);
 376         }
 377
 378 out:
 379         dev->virtqueue[index] = vq;
 380         vhost_devices[dev->vid] = dev;
 381
 382         if (old_vq != vq)
 383                 vhost_user_iotlb_init(dev, index);
 384
 385         return dev;
 386 }
 387 #else
 388 static struct virtio_net*
 389 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 390 {
 391         return dev;
 392 }
 393 #endif
 394
 395 /* Converts QEMU virtual address to Vhost virtual address. */
 396 static uint64_t
 397 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 398 {
 399         struct rte_vhost_mem_region *reg;
 400         uint32_t i;
 401
 402         /* Find the region where the address lives. */
 403         for (i = 0; i < dev->mem->nregions; i++) {
 404                 reg = &dev->mem->regions[i];
 405
 406                 if (qva >= reg->guest_user_addr &&
 407                     qva <  reg->guest_user_addr + reg->size) {
 408                         return qva - reg->guest_user_addr +
 409                                reg->host_user_addr;
 410                 }
 411         }
 412
 413         return 0;
 414 }
 415
 416
 417 /*
 418  * Converts ring address to Vhost virtual address.
 419  * If IOMMU is enabled, the ring address is a guest IO virtual address,
 420  * else it is a QEMU virtual address.
 421  */
 422 static uint64_t
 423 ring_addr_to_vva(struct virtio_net *dev, struct vhost_virtqueue *vq,
 424                 uint64_t ra, uint64_t size)
 425 {
 426         if (dev->features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)) {
 427                 uint64_t vva;
 428
 429                 vva = vhost_user_iotlb_cache_find(vq, ra,
 430                                         &size, VHOST_ACCESS_RW);
 431                 if (!vva)
 432                         vhost_user_iotlb_miss(dev, ra, VHOST_ACCESS_RW);
 433
 434                 return vva;
 435         }
 436
 437         return qva_to_vva(dev, ra);
 438 }
 439
 440 static struct virtio_net *
 441 translate_ring_addresses(struct virtio_net *dev, int vq_index)
 442 {
 443         struct vhost_virtqueue *vq = dev->virtqueue[vq_index];
 444         struct vhost_vring_addr *addr = &vq->ring_addrs;
 445
 446         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 447         if (vq->desc && vq->avail && vq->used)
 448                 return dev;
 449
 450         vq->desc = (struct vring_desc *)(uintptr_t)ring_addr_to_vva(dev,
 451                         vq, addr->desc_user_addr, sizeof(struct vring_desc));
 452         if (vq->desc == 0) {
 453                 RTE_LOG(DEBUG, VHOST_CONFIG,
 454                         "(%d) failed to find desc ring address.\n",
 455                         dev->vid);
 456                 return dev;
 457         }
 458
 459         dev = numa_realloc(dev, vq_index);
 460         vq = dev->virtqueue[vq_index];
 461         addr = &vq->ring_addrs;
 462
 463         vq->avail = (struct vring_avail *)(uintptr_t)ring_addr_to_vva(dev,
 464                         vq, addr->avail_user_addr, sizeof(struct vring_avail));
 465         if (vq->avail == 0) {
 466                 RTE_LOG(DEBUG, VHOST_CONFIG,
 467                         "(%d) failed to find avail ring address.\n",
 468                         dev->vid);
 469                 return dev;
 470         }
 471
 472         vq->used = (struct vring_used *)(uintptr_t)ring_addr_to_vva(dev,
 473                         vq, addr->used_user_addr, sizeof(struct vring_used));
 474         if (vq->used == 0) {
 475                 RTE_LOG(DEBUG, VHOST_CONFIG,
 476                         "(%d) failed to find used ring address.\n",
 477                         dev->vid);
 478                 return dev;
 479         }
 480
 481         if (vq->last_used_idx != vq->used->idx) {
 482                 RTE_LOG(WARNING, VHOST_CONFIG,
 483                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 484                         "some packets maybe resent for Tx and dropped for Rx\n",
 485                         vq->last_used_idx, vq->used->idx);
 486                 vq->last_used_idx  = vq->used->idx;
 487                 vq->last_avail_idx = vq->used->idx;
 488         }
 489
 490         vq->log_guest_addr = addr->log_guest_addr;
 491
 492         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 493                         dev->vid, vq->desc);
 494         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 495                         dev->vid, vq->avail);
 496         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 497                         dev->vid, vq->used);
 498         VHOST_LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 499                         dev->vid, vq->log_guest_addr);
 500
 501         return dev;
 502 }
 503
 504 /*
 505  * The virtio device sends us the desc, used and avail ring addresses.
 506  * This function then converts these to our address space.
 507  */
 508 static int
 509 vhost_user_set_vring_addr(struct virtio_net **pdev, VhostUserMsg *msg)
 510 {
 511         struct vhost_virtqueue *vq;
 512         struct vhost_vring_addr *addr = &msg->payload.addr;
 513         struct virtio_net *dev = *pdev;
 514
 515         if (dev->mem == NULL)
 516                 return -1;
 517
 518         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 519         vq = dev->virtqueue[msg->payload.addr.index];
 520
 521         /*
 522          * Rings addresses should not be interpreted as long as the ring is not
 523          * started and enabled
 524          */
 525         memcpy(&vq->ring_addrs, addr, sizeof(*addr));
 526
 527         vring_invalidate(dev, vq);
 528
 529         if (vq->enabled && (dev->features &
 530                                 (1ULL << VHOST_USER_F_PROTOCOL_FEATURES))) {
 531                 dev = translate_ring_addresses(dev, msg->payload.addr.index);
 532                 if (!dev)
 533                         return -1;
 534
 535                 *pdev = dev;
 536         }
 537
 538         return 0;
 539 }
 540
 541 /*
 542  * The virtio device sends us the available ring last used index.
 543  */
 544 static int
 545 vhost_user_set_vring_base(struct virtio_net *dev,
 546                           VhostUserMsg *msg)
 547 {
 548         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 549                         msg->payload.state.num;
 550         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 551                         msg->payload.state.num;
 552
 553         return 0;
 554 }
 555
 556 static int
 557 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 558                    uint64_t host_phys_addr, uint64_t size)
 559 {
 560         struct guest_page *page, *last_page;
 561
 562         if (dev->nr_guest_pages == dev->max_guest_pages) {
 563                 dev->max_guest_pages *= 2;
 564                 dev->guest_pages = realloc(dev->guest_pages,
 565                                         dev->max_guest_pages * sizeof(*page));
 566                 if (!dev->guest_pages) {
 567                         RTE_LOG(ERR, VHOST_CONFIG, "cannot realloc guest_pages\n");
 568                         return -1;
 569                 }
 570         }
 571
 572         if (dev->nr_guest_pages > 0) {
 573                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 574                 /* merge if the two pages are continuous */
 575                 if (host_phys_addr == last_page->host_phys_addr +
 576                                       last_page->size) {
 577                         last_page->size += size;
 578                         return 0;
 579                 }
 580         }
 581
 582         page = &dev->guest_pages[dev->nr_guest_pages++];
 583         page->guest_phys_addr = guest_phys_addr;
 584         page->host_phys_addr  = host_phys_addr;
 585         page->size = size;
 586
 587         return 0;
 588 }
 589
 590 static int
 591 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 592                 uint64_t page_size)
 593 {
 594         uint64_t reg_size = reg->size;
 595         uint64_t host_user_addr  = reg->host_user_addr;
 596         uint64_t guest_phys_addr = reg->guest_phys_addr;
 597         uint64_t host_phys_addr;
 598         uint64_t size;
 599
 600         host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)host_user_addr);
 601         size = page_size - (guest_phys_addr & (page_size - 1));
 602         size = RTE_MIN(size, reg_size);
 603
 604         if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size) < 0)
 605                 return -1;
 606
 607         host_user_addr  += size;
 608         guest_phys_addr += size;
 609         reg_size -= size;
 610
 611         while (reg_size > 0) {
 612                 size = RTE_MIN(reg_size, page_size);
 613                 host_phys_addr = rte_mem_virt2iova((void *)(uintptr_t)
 614                                                   host_user_addr);
 615                 if (add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 616                                 size) < 0)
 617                         return -1;
 618
 619                 host_user_addr  += size;
 620                 guest_phys_addr += size;
 621                 reg_size -= size;
 622         }
 623
 624         return 0;
 625 }
 626
 627 #ifdef RTE_LIBRTE_VHOST_DEBUG
 628 /* TODO: enable it only in debug mode? */
 629 static void
 630 dump_guest_pages(struct virtio_net *dev)
 631 {
 632         uint32_t i;
 633         struct guest_page *page;
 634
 635         for (i = 0; i < dev->nr_guest_pages; i++) {
 636                 page = &dev->guest_pages[i];
 637
 638                 RTE_LOG(INFO, VHOST_CONFIG,
 639                         "guest physical page region %u\n"
 640                         "\t guest_phys_addr: %" PRIx64 "\n"
 641                         "\t host_phys_addr : %" PRIx64 "\n"
 642                         "\t size           : %" PRIx64 "\n",
 643                         i,
 644                         page->guest_phys_addr,
 645                         page->host_phys_addr,
 646                         page->size);
 647         }
 648 }
 649 #else
 650 #define dump_guest_pages(dev)
 651 #endif
 652
 653 static bool
 654 vhost_memory_changed(struct VhostUserMemory *new,
 655                      struct rte_vhost_memory *old)
 656 {
 657         uint32_t i;
 658
 659         if (new->nregions != old->nregions)
 660                 return true;
 661
 662         for (i = 0; i < new->nregions; ++i) {
 663                 VhostUserMemoryRegion *new_r = &new->regions[i];
 664                 struct rte_vhost_mem_region *old_r = &old->regions[i];
 665
 666                 if (new_r->guest_phys_addr != old_r->guest_phys_addr)
 667                         return true;
 668                 if (new_r->memory_size != old_r->size)
 669                         return true;
 670                 if (new_r->userspace_addr != old_r->guest_user_addr)
 671                         return true;
 672         }
 673
 674         return false;
 675 }
 676
 677 static int
 678 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 679 {
 680         struct VhostUserMemory memory = pmsg->payload.memory;
 681         struct rte_vhost_mem_region *reg;
 682         void *mmap_addr;
 683         uint64_t mmap_size;
 684         uint64_t mmap_offset;
 685         uint64_t alignment;
 686         uint32_t i;
 687         int populate;
 688         int fd;
 689
 690         if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
 691                 RTE_LOG(ERR, VHOST_CONFIG,
 692                         "too many memory regions (%u)\n", memory.nregions);
 693                 return -1;
 694         }
 695
 696         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 697                 RTE_LOG(INFO, VHOST_CONFIG,
 698                         "(%d) memory regions not changed\n", dev->vid);
 699
 700                 for (i = 0; i < memory.nregions; i++)
 701                         close(pmsg->fds[i]);
 702
 703                 return 0;
 704         }
 705
 706         if (dev->mem) {
 707                 free_mem_region(dev);
 708                 rte_free(dev->mem);
 709                 dev->mem = NULL;
 710         }
 711
 712         dev->nr_guest_pages = 0;
 713         if (!dev->guest_pages) {
 714                 dev->max_guest_pages = 8;
 715                 dev->guest_pages = malloc(dev->max_guest_pages *
 716                                                 sizeof(struct guest_page));
 717                 if (dev->guest_pages == NULL) {
 718                         RTE_LOG(ERR, VHOST_CONFIG,
 719                                 "(%d) failed to allocate memory "
 720                                 "for dev->guest_pages\n",
 721                                 dev->vid);
 722                         return -1;
 723                 }
 724         }
 725
 726         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 727                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 728         if (dev->mem == NULL) {
 729                 RTE_LOG(ERR, VHOST_CONFIG,
 730                         "(%d) failed to allocate memory for dev->mem\n",
 731                         dev->vid);
 732                 return -1;
 733         }
 734         dev->mem->nregions = memory.nregions;
 735
 736         for (i = 0; i < memory.nregions; i++) {
 737                 fd  = pmsg->fds[i];
 738                 reg = &dev->mem->regions[i];
 739
 740                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 741                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 742                 reg->size            = memory.regions[i].memory_size;
 743                 reg->fd              = fd;
 744
 745                 mmap_offset = memory.regions[i].mmap_offset;
 746
 747                 /* Check for memory_size + mmap_offset overflow */
 748                 if (mmap_offset >= -reg->size) {
 749                         RTE_LOG(ERR, VHOST_CONFIG,
 750                                 "mmap_offset (%#"PRIx64") and memory_size "
 751                                 "(%#"PRIx64") overflow\n",
 752                                 mmap_offset, reg->size);
 753                         goto err_mmap;
 754                 }
 755
 756                 mmap_size = reg->size + mmap_offset;
 757
 758                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 759                  * with length argument aligned with hugepagesz at older
 760                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 761                  * mmap() will fail with EINVAL.
 762                  *
 763                  * to avoid failure, make sure in caller to keep length
 764                  * aligned.
 765                  */
 766                 alignment = get_blk_size(fd);
 767                 if (alignment == (uint64_t)-1) {
 768                         RTE_LOG(ERR, VHOST_CONFIG,
 769                                 "couldn't get hugepage size through fstat\n");
 770                         goto err_mmap;
 771                 }
 772                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 773
 774                 populate = (dev->dequeue_zero_copy) ? MAP_POPULATE : 0;
 775                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 776                                  MAP_SHARED | populate, fd, 0);
 777
 778                 if (mmap_addr == MAP_FAILED) {
 779                         RTE_LOG(ERR, VHOST_CONFIG,
 780                                 "mmap region %u failed.\n", i);
 781                         goto err_mmap;
 782                 }
 783
 784                 reg->mmap_addr = mmap_addr;
 785                 reg->mmap_size = mmap_size;
 786                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 787                                       mmap_offset;
 788
 789                 if (dev->dequeue_zero_copy)
 790                         if (add_guest_pages(dev, reg, alignment) < 0) {
 791                                 RTE_LOG(ERR, VHOST_CONFIG,
 792                                         "adding guest pages to region %u failed.\n",
 793                                         i);
 794                                 goto err_mmap;
 795                         }
 796
 797                 RTE_LOG(INFO, VHOST_CONFIG,
 798                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 799                         "\t guest physical addr: 0x%" PRIx64 "\n"
 800                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 801                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 802                         "\t mmap addr : 0x%" PRIx64 "\n"
 803                         "\t mmap size : 0x%" PRIx64 "\n"
 804                         "\t mmap align: 0x%" PRIx64 "\n"
 805                         "\t mmap off  : 0x%" PRIx64 "\n",
 806                         i, reg->size,
 807                         reg->guest_phys_addr,
 808                         reg->guest_user_addr,
 809                         reg->host_user_addr,
 810                         (uint64_t)(uintptr_t)mmap_addr,
 811                         mmap_size,
 812                         alignment,
 813                         mmap_offset);
 814         }
 815
 816         dump_guest_pages(dev);
 817
 818         return 0;
 819
 820 err_mmap:
 821         free_mem_region(dev);
 822         rte_free(dev->mem);
 823         dev->mem = NULL;
 824         return -1;
 825 }
 826
 827 static int
 828 vq_is_ready(struct vhost_virtqueue *vq)
 829 {
 830         return vq && vq->desc && vq->avail && vq->used &&
 831                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 832                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 833 }
 834
 835 static int
 836 virtio_is_ready(struct virtio_net *dev)
 837 {
 838         struct vhost_virtqueue *vq;
 839         uint32_t i;
 840
 841         if (dev->nr_vring == 0)
 842                 return 0;
 843
 844         for (i = 0; i < dev->nr_vring; i++) {
 845                 vq = dev->virtqueue[i];
 846
 847                 if (!vq_is_ready(vq))
 848                         return 0;
 849         }
 850
 851         RTE_LOG(INFO, VHOST_CONFIG,
 852                 "virtio is now ready for processing.\n");
 853         return 1;
 854 }
 855
 856 static void
 857 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 858 {
 859         struct vhost_vring_file file;
 860         struct vhost_virtqueue *vq;
 861
 862         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 863         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 864                 file.fd = VIRTIO_INVALID_EVENTFD;
 865         else
 866                 file.fd = pmsg->fds[0];
 867         RTE_LOG(INFO, VHOST_CONFIG,
 868                 "vring call idx:%d file:%d\n", file.index, file.fd);
 869
 870         vq = dev->virtqueue[file.index];
 871         if (vq->callfd >= 0)
 872                 close(vq->callfd);
 873
 874         vq->callfd = file.fd;
 875 }
 876
 877 static void
 878 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 879 {
 880         struct vhost_vring_file file;
 881         struct vhost_virtqueue *vq;
 882         struct virtio_net *dev = *pdev;
 883
 884         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 885         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 886                 file.fd = VIRTIO_INVALID_EVENTFD;
 887         else
 888                 file.fd = pmsg->fds[0];
 889         RTE_LOG(INFO, VHOST_CONFIG,
 890                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 891
 892         /* Interpret ring addresses only when ring is started. */
 893         dev = translate_ring_addresses(dev, file.index);
 894         if (!dev)
 895                 return;
 896
 897         *pdev = dev;
 898
 899         vq = dev->virtqueue[file.index];
 900
 901         /*
 902          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 903          * the ring starts already enabled. Otherwise, it is enabled via
 904          * the SET_VRING_ENABLE message.
 905          */
 906         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 907                 vq->enabled = 1;
 908
 909         if (vq->kickfd >= 0)
 910                 close(vq->kickfd);
 911         vq->kickfd = file.fd;
 912 }
 913
 914 static void
 915 free_zmbufs(struct vhost_virtqueue *vq)
 916 {
 917         struct zcopy_mbuf *zmbuf, *next;
 918
 919         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 920              zmbuf != NULL; zmbuf = next) {
 921                 next = TAILQ_NEXT(zmbuf, next);
 922
 923                 rte_pktmbuf_free(zmbuf->mbuf);
 924                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 925         }
 926
 927         rte_free(vq->zmbufs);
 928 }
 929
 930 /*
 931  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 932  */
 933 static int
 934 vhost_user_get_vring_base(struct virtio_net *dev,
 935                           VhostUserMsg *msg)
 936 {
 937         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 938
 939         /* We have to stop the queue (virtio) if it is running. */
 940         if (dev->flags & VIRTIO_DEV_RUNNING) {
 941                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 942                 dev->notify_ops->destroy_device(dev->vid);
 943         }
 944
 945         dev->flags &= ~VIRTIO_DEV_READY;
 946
 947         /* Here we are safe to get the last avail index */
 948         msg->payload.state.num = vq->last_avail_idx;
 949
 950         RTE_LOG(INFO, VHOST_CONFIG,
 951                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 952                 msg->payload.state.num);
 953         /*
 954          * Based on current qemu vhost-user implementation, this message is
 955          * sent and only sent in vhost_vring_stop.
 956          * TODO: cleanup the vring, it isn't usable since here.
 957          */
 958         if (vq->kickfd >= 0)
 959                 close(vq->kickfd);
 960
 961         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 962
 963         if (vq->callfd >= 0)
 964                 close(vq->callfd);
 965
 966         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 967
 968         if (dev->dequeue_zero_copy)
 969                 free_zmbufs(vq);
 970         rte_free(vq->shadow_used_ring);
 971         vq->shadow_used_ring = NULL;
 972
 973         rte_free(vq->batch_copy_elems);
 974         vq->batch_copy_elems = NULL;
 975
 976         return 0;
 977 }
 978
 979 /*
 980  * when virtio queues are ready to work, qemu will send us to
 981  * enable the virtio queue pair.
 982  */
 983 static int
 984 vhost_user_set_vring_enable(struct virtio_net *dev,
 985                             VhostUserMsg *msg)
 986 {
 987         int enable = (int)msg->payload.state.num;
 988
 989         RTE_LOG(INFO, VHOST_CONFIG,
 990                 "set queue enable: %d to qp idx: %d\n",
 991                 enable, msg->payload.state.index);
 992
 993         if (dev->notify_ops->vring_state_changed)
 994                 dev->notify_ops->vring_state_changed(dev->vid,
 995                                 msg->payload.state.index, enable);
 996
 997         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 998
 999         return 0;
1000 }
1001
1002 static void
1003 vhost_user_get_protocol_features(struct virtio_net *dev,
1004                                  struct VhostUserMsg *msg)
1005 {
1006         uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
1007
1008         rte_vhost_driver_get_features(dev->ifname, &features);
1009
1010         /*
1011          * REPLY_ACK protocol feature is only mandatory for now
1012          * for IOMMU feature. If IOMMU is explicitly disabled by the
1013          * application, disable also REPLY_ACK feature for older buggy
1014          * Qemu versions (from v2.7.0 to v2.9.0).
1015          */
1016         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1017                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1018
1019         msg->payload.u64 = protocol_features;
1020         msg->size = sizeof(msg->payload.u64);
1021 }
1022
1023 static void
1024 vhost_user_set_protocol_features(struct virtio_net *dev,
1025                                  uint64_t protocol_features)
1026 {
1027         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1028                 return;
1029
1030         dev->protocol_features = protocol_features;
1031 }
1032
1033 static int
1034 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1035 {
1036         int fd = msg->fds[0];
1037         uint64_t size, off;
1038         void *addr;
1039
1040         if (fd < 0) {
1041                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1042                 return -1;
1043         }
1044
1045         if (msg->size != sizeof(VhostUserLog)) {
1046                 RTE_LOG(ERR, VHOST_CONFIG,
1047                         "invalid log base msg size: %"PRId32" != %d\n",
1048                         msg->size, (int)sizeof(VhostUserLog));
1049                 return -1;
1050         }
1051
1052         size = msg->payload.log.mmap_size;
1053         off  = msg->payload.log.mmap_offset;
1054
1055         /* Don't allow mmap_offset to point outside the mmap region */
1056         if (off > size) {
1057                 RTE_LOG(ERR, VHOST_CONFIG,
1058                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1059                         off, size);
1060                 return -1;
1061         }
1062
1063         RTE_LOG(INFO, VHOST_CONFIG,
1064                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1065                 size, off);
1066
1067         /*
1068          * mmap from 0 to workaround a hugepage mmap bug: mmap will
1069          * fail when offset is not page size aligned.
1070          */
1071         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1072         close(fd);
1073         if (addr == MAP_FAILED) {
1074                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1075                 return -1;
1076         }
1077
1078         /*
1079          * Free previously mapped log memory on occasionally
1080          * multiple VHOST_USER_SET_LOG_BASE.
1081          */
1082         if (dev->log_addr) {
1083                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1084         }
1085         dev->log_addr = (uint64_t)(uintptr_t)addr;
1086         dev->log_base = dev->log_addr + off;
1087         dev->log_size = size;
1088
1089         return 0;
1090 }
1091
1092 /*
1093  * An rarp packet is constructed and broadcasted to notify switches about
1094  * the new location of the migrated VM, so that packets from outside will
1095  * not be lost after migration.
1096  *
1097  * However, we don't actually "send" a rarp packet here, instead, we set
1098  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1099  */
1100 static int
1101 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1102 {
1103         uint8_t *mac = (uint8_t *)&msg->payload.u64;
1104
1105         RTE_LOG(DEBUG, VHOST_CONFIG,
1106                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1107                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1108         memcpy(dev->mac.addr_bytes, mac, 6);
1109
1110         /*
1111          * Set the flag to inject a RARP broadcast packet at
1112          * rte_vhost_dequeue_burst().
1113          *
1114          * rte_smp_wmb() is for making sure the mac is copied
1115          * before the flag is set.
1116          */
1117         rte_smp_wmb();
1118         rte_atomic16_set(&dev->broadcast_rarp, 1);
1119
1120         return 0;
1121 }
1122
1123 static int
1124 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1125 {
1126         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1127                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1128                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1129                                 msg->payload.u64);
1130
1131                 return -1;
1132         }
1133
1134         dev->mtu = msg->payload.u64;
1135
1136         return 0;
1137 }
1138
1139 static int
1140 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1141 {
1142         int fd = msg->fds[0];
1143
1144         if (fd < 0) {
1145                 RTE_LOG(ERR, VHOST_CONFIG,
1146                                 "Invalid file descriptor for slave channel (%d)\n",
1147                                 fd);
1148                 return -1;
1149         }
1150
1151         dev->slave_req_fd = fd;
1152
1153         return 0;
1154 }
1155
1156 static int
1157 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1158 {
1159         struct vhost_vring_addr *ra;
1160         uint64_t start, end;
1161
1162         start = imsg->iova;
1163         end = start + imsg->size;
1164
1165         ra = &vq->ring_addrs;
1166         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1167                 return 1;
1168         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1169                 return 1;
1170         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1171                 return 1;
1172
1173         return 0;
1174 }
1175
1176 static int
1177 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1178                                 struct vhost_iotlb_msg *imsg)
1179 {
1180         uint64_t istart, iend, vstart, vend;
1181
1182         istart = imsg->iova;
1183         iend = istart + imsg->size - 1;
1184
1185         vstart = (uintptr_t)vq->desc;
1186         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1187         if (vstart <= iend && istart <= vend)
1188                 return 1;
1189
1190         vstart = (uintptr_t)vq->avail;
1191         vend = vstart + sizeof(struct vring_avail);
1192         vend += sizeof(uint16_t) * vq->size - 1;
1193         if (vstart <= iend && istart <= vend)
1194                 return 1;
1195
1196         vstart = (uintptr_t)vq->used;
1197         vend = vstart + sizeof(struct vring_used);
1198         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1199         if (vstart <= iend && istart <= vend)
1200                 return 1;
1201
1202         return 0;
1203 }
1204
1205 static int
1206 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1207 {
1208         struct virtio_net *dev = *pdev;
1209         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1210         uint16_t i;
1211         uint64_t vva;
1212
1213         switch (imsg->type) {
1214         case VHOST_IOTLB_UPDATE:
1215                 vva = qva_to_vva(dev, imsg->uaddr);
1216                 if (!vva)
1217                         return -1;
1218
1219                 for (i = 0; i < dev->nr_vring; i++) {
1220                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1221
1222                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1223                                         imsg->size, imsg->perm);
1224
1225                         if (is_vring_iotlb_update(vq, imsg))
1226                                 *pdev = dev = translate_ring_addresses(dev, i);
1227                 }
1228                 break;
1229         case VHOST_IOTLB_INVALIDATE:
1230                 for (i = 0; i < dev->nr_vring; i++) {
1231                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1232
1233                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1234                                         imsg->size);
1235
1236                         if (is_vring_iotlb_invalidate(vq, imsg))
1237                                 vring_invalidate(dev, vq);
1238                 }
1239                 break;
1240         default:
1241                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1242                                 imsg->type);
1243                 return -1;
1244         }
1245
1246         return 0;
1247 }
1248
1249 /* return bytes# of read on success or negative val on failure. */
1250 static int
1251 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1252 {
1253         int ret;
1254
1255         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1256                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1257         if (ret <= 0)
1258                 return ret;
1259
1260         if (msg && msg->size) {
1261                 if (msg->size > sizeof(msg->payload)) {
1262                         RTE_LOG(ERR, VHOST_CONFIG,
1263                                 "invalid msg size: %d\n", msg->size);
1264                         return -1;
1265                 }
1266                 ret = read(sockfd, &msg->payload, msg->size);
1267                 if (ret <= 0)
1268                         return ret;
1269                 if (ret != (int)msg->size) {
1270                         RTE_LOG(ERR, VHOST_CONFIG,
1271                                 "read control message failed\n");
1272                         return -1;
1273                 }
1274         }
1275
1276         return ret;
1277 }
1278
1279 static int
1280 send_vhost_message(int sockfd, struct VhostUserMsg *msg, int *fds, int fd_num)
1281 {
1282         if (!msg)
1283                 return 0;
1284
1285         return send_fd_message(sockfd, (char *)msg,
1286                 VHOST_USER_HDR_SIZE + msg->size, fds, fd_num);
1287 }
1288
1289 static int
1290 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1291 {
1292         if (!msg)
1293                 return 0;
1294
1295         msg->flags &= ~VHOST_USER_VERSION_MASK;
1296         msg->flags &= ~VHOST_USER_NEED_REPLY;
1297         msg->flags |= VHOST_USER_VERSION;
1298         msg->flags |= VHOST_USER_REPLY_MASK;
1299
1300         return send_vhost_message(sockfd, msg, NULL, 0);
1301 }
1302
1303 /*
1304  * Allocate a queue pair if it hasn't been allocated yet
1305  */
1306 static int
1307 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1308 {
1309         uint16_t vring_idx;
1310
1311         switch (msg->request.master) {
1312         case VHOST_USER_SET_VRING_KICK:
1313         case VHOST_USER_SET_VRING_CALL:
1314         case VHOST_USER_SET_VRING_ERR:
1315                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1316                 break;
1317         case VHOST_USER_SET_VRING_NUM:
1318         case VHOST_USER_SET_VRING_BASE:
1319         case VHOST_USER_SET_VRING_ENABLE:
1320                 vring_idx = msg->payload.state.index;
1321                 break;
1322         case VHOST_USER_SET_VRING_ADDR:
1323                 vring_idx = msg->payload.addr.index;
1324                 break;
1325         default:
1326                 return 0;
1327         }
1328
1329         if (vring_idx >= VHOST_MAX_VRING) {
1330                 RTE_LOG(ERR, VHOST_CONFIG,
1331                         "invalid vring index: %u\n", vring_idx);
1332                 return -1;
1333         }
1334
1335         if (dev->virtqueue[vring_idx])
1336                 return 0;
1337
1338         return alloc_vring_queue(dev, vring_idx);
1339 }
1340
1341 static void
1342 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1343 {
1344         unsigned int i = 0;
1345         unsigned int vq_num = 0;
1346
1347         while (vq_num < dev->nr_vring) {
1348                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1349
1350                 if (vq) {
1351                         rte_spinlock_lock(&vq->access_lock);
1352                         vq_num++;
1353                 }
1354                 i++;
1355         }
1356 }
1357
1358 static void
1359 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1360 {
1361         unsigned int i = 0;
1362         unsigned int vq_num = 0;
1363
1364         while (vq_num < dev->nr_vring) {
1365                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1366
1367                 if (vq) {
1368                         rte_spinlock_unlock(&vq->access_lock);
1369                         vq_num++;
1370                 }
1371                 i++;
1372         }
1373 }
1374
1375 int
1376 vhost_user_msg_handler(int vid, int fd)
1377 {
1378         struct virtio_net *dev;
1379         struct VhostUserMsg msg;
1380         int ret;
1381         int unlock_required = 0;
1382
1383         dev = get_device(vid);
1384         if (dev == NULL)
1385                 return -1;
1386
1387         if (!dev->notify_ops) {
1388                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1389                 if (!dev->notify_ops) {
1390                         RTE_LOG(ERR, VHOST_CONFIG,
1391                                 "failed to get callback ops for driver %s\n",
1392                                 dev->ifname);
1393                         return -1;
1394                 }
1395         }
1396
1397         ret = read_vhost_message(fd, &msg);
1398         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1399                 if (ret < 0)
1400                         RTE_LOG(ERR, VHOST_CONFIG,
1401                                 "vhost read message failed\n");
1402                 else if (ret == 0)
1403                         RTE_LOG(INFO, VHOST_CONFIG,
1404                                 "vhost peer closed\n");
1405                 else
1406                         RTE_LOG(ERR, VHOST_CONFIG,
1407                                 "vhost read incorrect message\n");
1408
1409                 return -1;
1410         }
1411
1412         ret = 0;
1413         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1414                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1415                         vhost_message_str[msg.request.master]);
1416         else
1417                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1418                         vhost_message_str[msg.request.master]);
1419
1420         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1421         if (ret < 0) {
1422                 RTE_LOG(ERR, VHOST_CONFIG,
1423                         "failed to alloc queue\n");
1424                 return -1;
1425         }
1426
1427         /*
1428          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1429          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1430          * and device is destroyed. destroy_device waits for queues to be
1431          * inactive, so it is safe. Otherwise taking the access_lock
1432          * would cause a dead lock.
1433          */
1434         switch (msg.request.master) {
1435         case VHOST_USER_SET_FEATURES:
1436         case VHOST_USER_SET_PROTOCOL_FEATURES:
1437         case VHOST_USER_SET_OWNER:
1438         case VHOST_USER_SET_MEM_TABLE:
1439         case VHOST_USER_SET_LOG_BASE:
1440         case VHOST_USER_SET_LOG_FD:
1441         case VHOST_USER_SET_VRING_NUM:
1442         case VHOST_USER_SET_VRING_ADDR:
1443         case VHOST_USER_SET_VRING_BASE:
1444         case VHOST_USER_SET_VRING_KICK:
1445         case VHOST_USER_SET_VRING_CALL:
1446         case VHOST_USER_SET_VRING_ERR:
1447         case VHOST_USER_SET_VRING_ENABLE:
1448         case VHOST_USER_SEND_RARP:
1449         case VHOST_USER_NET_SET_MTU:
1450         case VHOST_USER_SET_SLAVE_REQ_FD:
1451                 vhost_user_lock_all_queue_pairs(dev);
1452                 unlock_required = 1;
1453                 break;
1454         default:
1455                 break;
1456
1457         }
1458
1459         switch (msg.request.master) {
1460         case VHOST_USER_GET_FEATURES:
1461                 msg.payload.u64 = vhost_user_get_features(dev);
1462                 msg.size = sizeof(msg.payload.u64);
1463                 send_vhost_reply(fd, &msg);
1464                 break;
1465         case VHOST_USER_SET_FEATURES:
1466                 ret = vhost_user_set_features(dev, msg.payload.u64);
1467                 if (ret)
1468                         return -1;
1469                 break;
1470
1471         case VHOST_USER_GET_PROTOCOL_FEATURES:
1472                 vhost_user_get_protocol_features(dev, &msg);
1473                 send_vhost_reply(fd, &msg);
1474                 break;
1475         case VHOST_USER_SET_PROTOCOL_FEATURES:
1476                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1477                 break;
1478
1479         case VHOST_USER_SET_OWNER:
1480                 vhost_user_set_owner();
1481                 break;
1482         case VHOST_USER_RESET_OWNER:
1483                 vhost_user_reset_owner(dev);
1484                 break;
1485
1486         case VHOST_USER_SET_MEM_TABLE:
1487                 ret = vhost_user_set_mem_table(dev, &msg);
1488                 break;
1489
1490         case VHOST_USER_SET_LOG_BASE:
1491                 vhost_user_set_log_base(dev, &msg);
1492
1493                 /* it needs a reply */
1494                 msg.size = sizeof(msg.payload.u64);
1495                 send_vhost_reply(fd, &msg);
1496                 break;
1497         case VHOST_USER_SET_LOG_FD:
1498                 close(msg.fds[0]);
1499                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1500                 break;
1501
1502         case VHOST_USER_SET_VRING_NUM:
1503                 vhost_user_set_vring_num(dev, &msg);
1504                 break;
1505         case VHOST_USER_SET_VRING_ADDR:
1506                 vhost_user_set_vring_addr(&dev, &msg);
1507                 break;
1508         case VHOST_USER_SET_VRING_BASE:
1509                 vhost_user_set_vring_base(dev, &msg);
1510                 break;
1511
1512         case VHOST_USER_GET_VRING_BASE:
1513                 vhost_user_get_vring_base(dev, &msg);
1514                 msg.size = sizeof(msg.payload.state);
1515                 send_vhost_reply(fd, &msg);
1516                 break;
1517
1518         case VHOST_USER_SET_VRING_KICK:
1519                 vhost_user_set_vring_kick(&dev, &msg);
1520                 break;
1521         case VHOST_USER_SET_VRING_CALL:
1522                 vhost_user_set_vring_call(dev, &msg);
1523                 break;
1524
1525         case VHOST_USER_SET_VRING_ERR:
1526                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1527                         close(msg.fds[0]);
1528                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1529                 break;
1530
1531         case VHOST_USER_GET_QUEUE_NUM:
1532                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1533                 msg.size = sizeof(msg.payload.u64);
1534                 send_vhost_reply(fd, &msg);
1535                 break;
1536
1537         case VHOST_USER_SET_VRING_ENABLE:
1538                 vhost_user_set_vring_enable(dev, &msg);
1539                 break;
1540         case VHOST_USER_SEND_RARP:
1541                 vhost_user_send_rarp(dev, &msg);
1542                 break;
1543
1544         case VHOST_USER_NET_SET_MTU:
1545                 ret = vhost_user_net_set_mtu(dev, &msg);
1546                 break;
1547
1548         case VHOST_USER_SET_SLAVE_REQ_FD:
1549                 ret = vhost_user_set_req_fd(dev, &msg);
1550                 break;
1551
1552         case VHOST_USER_IOTLB_MSG:
1553                 ret = vhost_user_iotlb_msg(&dev, &msg);
1554                 break;
1555
1556         default:
1557                 ret = -1;
1558                 break;
1559
1560         }
1561
1562         if (unlock_required)
1563                 vhost_user_unlock_all_queue_pairs(dev);
1564
1565         if (msg.flags & VHOST_USER_NEED_REPLY) {
1566                 msg.payload.u64 = !!ret;
1567                 msg.size = sizeof(msg.payload.u64);
1568                 send_vhost_reply(fd, &msg);
1569         }
1570
1571         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1572                 dev->flags |= VIRTIO_DEV_READY;
1573
1574                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1575                         if (dev->dequeue_zero_copy) {
1576                                 RTE_LOG(INFO, VHOST_CONFIG,
1577                                                 "dequeue zero copy is enabled\n");
1578                         }
1579
1580                         if (dev->notify_ops->new_device(dev->vid) == 0)
1581                                 dev->flags |= VIRTIO_DEV_RUNNING;
1582                 }
1583         }
1584
1585         return 0;
1586 }
1587
1588 int
1589 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1590 {
1591         int ret;
1592         struct VhostUserMsg msg = {
1593                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1594                 .flags = VHOST_USER_VERSION,
1595                 .size = sizeof(msg.payload.iotlb),
1596                 .payload.iotlb = {
1597                         .iova = iova,
1598                         .perm = perm,
1599                         .type = VHOST_IOTLB_MISS,
1600                 },
1601         };
1602
1603         ret = send_vhost_message(dev->slave_req_fd, &msg, NULL, 0);
1604         if (ret < 0) {
1605                 RTE_LOG(ERR, VHOST_CONFIG,
1606                                 "Failed to send IOTLB miss message (%d)\n",
1607                                 ret);
1608                 return ret;
1609         }
1610
1611         return 0;
1612 }