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 fd;
 688
 689         if (memory.nregions > VHOST_MEMORY_MAX_NREGIONS) {
 690                 RTE_LOG(ERR, VHOST_CONFIG,
 691                         "too many memory regions (%u)\n", memory.nregions);
 692                 return -1;
 693         }
 694
 695         if (dev->mem && !vhost_memory_changed(&memory, dev->mem)) {
 696                 RTE_LOG(INFO, VHOST_CONFIG,
 697                         "(%d) memory regions not changed\n", dev->vid);
 698
 699                 for (i = 0; i < memory.nregions; i++)
 700                         close(pmsg->fds[i]);
 701
 702                 return 0;
 703         }
 704
 705         if (dev->mem) {
 706                 free_mem_region(dev);
 707                 rte_free(dev->mem);
 708                 dev->mem = NULL;
 709         }
 710
 711         dev->nr_guest_pages = 0;
 712         if (!dev->guest_pages) {
 713                 dev->max_guest_pages = 8;
 714                 dev->guest_pages = malloc(dev->max_guest_pages *
 715                                                 sizeof(struct guest_page));
 716                 if (dev->guest_pages == NULL) {
 717                         RTE_LOG(ERR, VHOST_CONFIG,
 718                                 "(%d) failed to allocate memory "
 719                                 "for dev->guest_pages\n",
 720                                 dev->vid);
 721                         return -1;
 722                 }
 723         }
 724
 725         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 726                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 727         if (dev->mem == NULL) {
 728                 RTE_LOG(ERR, VHOST_CONFIG,
 729                         "(%d) failed to allocate memory for dev->mem\n",
 730                         dev->vid);
 731                 return -1;
 732         }
 733         dev->mem->nregions = memory.nregions;
 734
 735         for (i = 0; i < memory.nregions; i++) {
 736                 fd  = pmsg->fds[i];
 737                 reg = &dev->mem->regions[i];
 738
 739                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 740                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 741                 reg->size            = memory.regions[i].memory_size;
 742                 reg->fd              = fd;
 743
 744                 mmap_offset = memory.regions[i].mmap_offset;
 745
 746                 /* Check for memory_size + mmap_offset overflow */
 747                 if (mmap_offset >= -reg->size) {
 748                         RTE_LOG(ERR, VHOST_CONFIG,
 749                                 "mmap_offset (%#"PRIx64") and memory_size "
 750                                 "(%#"PRIx64") overflow\n",
 751                                 mmap_offset, reg->size);
 752                         goto err_mmap;
 753                 }
 754
 755                 mmap_size = reg->size + mmap_offset;
 756
 757                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 758                  * with length argument aligned with hugepagesz at older
 759                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 760                  * mmap() will fail with EINVAL.
 761                  *
 762                  * to avoid failure, make sure in caller to keep length
 763                  * aligned.
 764                  */
 765                 alignment = get_blk_size(fd);
 766                 if (alignment == (uint64_t)-1) {
 767                         RTE_LOG(ERR, VHOST_CONFIG,
 768                                 "couldn't get hugepage size through fstat\n");
 769                         goto err_mmap;
 770                 }
 771                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 772
 773                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 774                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 775
 776                 if (mmap_addr == MAP_FAILED) {
 777                         RTE_LOG(ERR, VHOST_CONFIG,
 778                                 "mmap region %u failed.\n", i);
 779                         goto err_mmap;
 780                 }
 781
 782                 reg->mmap_addr = mmap_addr;
 783                 reg->mmap_size = mmap_size;
 784                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 785                                       mmap_offset;
 786
 787                 if (dev->dequeue_zero_copy)
 788                         if (add_guest_pages(dev, reg, alignment) < 0) {
 789                                 RTE_LOG(ERR, VHOST_CONFIG,
 790                                         "adding guest pages to region %u failed.\n",
 791                                         i);
 792                                 goto err_mmap;
 793                         }
 794
 795                 RTE_LOG(INFO, VHOST_CONFIG,
 796                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 797                         "\t guest physical addr: 0x%" PRIx64 "\n"
 798                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 799                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 800                         "\t mmap addr : 0x%" PRIx64 "\n"
 801                         "\t mmap size : 0x%" PRIx64 "\n"
 802                         "\t mmap align: 0x%" PRIx64 "\n"
 803                         "\t mmap off  : 0x%" PRIx64 "\n",
 804                         i, reg->size,
 805                         reg->guest_phys_addr,
 806                         reg->guest_user_addr,
 807                         reg->host_user_addr,
 808                         (uint64_t)(uintptr_t)mmap_addr,
 809                         mmap_size,
 810                         alignment,
 811                         mmap_offset);
 812         }
 813
 814         dump_guest_pages(dev);
 815
 816         return 0;
 817
 818 err_mmap:
 819         free_mem_region(dev);
 820         rte_free(dev->mem);
 821         dev->mem = NULL;
 822         return -1;
 823 }
 824
 825 static int
 826 vq_is_ready(struct vhost_virtqueue *vq)
 827 {
 828         return vq && vq->desc && vq->avail && vq->used &&
 829                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 830                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 831 }
 832
 833 static int
 834 virtio_is_ready(struct virtio_net *dev)
 835 {
 836         struct vhost_virtqueue *vq;
 837         uint32_t i;
 838
 839         if (dev->nr_vring == 0)
 840                 return 0;
 841
 842         for (i = 0; i < dev->nr_vring; i++) {
 843                 vq = dev->virtqueue[i];
 844
 845                 if (!vq_is_ready(vq))
 846                         return 0;
 847         }
 848
 849         RTE_LOG(INFO, VHOST_CONFIG,
 850                 "virtio is now ready for processing.\n");
 851         return 1;
 852 }
 853
 854 static void
 855 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 856 {
 857         struct vhost_vring_file file;
 858         struct vhost_virtqueue *vq;
 859
 860         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 861         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 862                 file.fd = VIRTIO_INVALID_EVENTFD;
 863         else
 864                 file.fd = pmsg->fds[0];
 865         RTE_LOG(INFO, VHOST_CONFIG,
 866                 "vring call idx:%d file:%d\n", file.index, file.fd);
 867
 868         vq = dev->virtqueue[file.index];
 869         if (vq->callfd >= 0)
 870                 close(vq->callfd);
 871
 872         vq->callfd = file.fd;
 873 }
 874
 875 static void
 876 vhost_user_set_vring_kick(struct virtio_net **pdev, struct VhostUserMsg *pmsg)
 877 {
 878         struct vhost_vring_file file;
 879         struct vhost_virtqueue *vq;
 880         struct virtio_net *dev = *pdev;
 881
 882         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 883         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 884                 file.fd = VIRTIO_INVALID_EVENTFD;
 885         else
 886                 file.fd = pmsg->fds[0];
 887         RTE_LOG(INFO, VHOST_CONFIG,
 888                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 889
 890         /* Interpret ring addresses only when ring is started. */
 891         dev = translate_ring_addresses(dev, file.index);
 892         if (!dev)
 893                 return;
 894
 895         *pdev = dev;
 896
 897         vq = dev->virtqueue[file.index];
 898
 899         /*
 900          * When VHOST_USER_F_PROTOCOL_FEATURES is not negotiated,
 901          * the ring starts already enabled. Otherwise, it is enabled via
 902          * the SET_VRING_ENABLE message.
 903          */
 904         if (!(dev->features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES)))
 905                 vq->enabled = 1;
 906
 907         if (vq->kickfd >= 0)
 908                 close(vq->kickfd);
 909         vq->kickfd = file.fd;
 910 }
 911
 912 static void
 913 free_zmbufs(struct vhost_virtqueue *vq)
 914 {
 915         struct zcopy_mbuf *zmbuf, *next;
 916
 917         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 918              zmbuf != NULL; zmbuf = next) {
 919                 next = TAILQ_NEXT(zmbuf, next);
 920
 921                 rte_pktmbuf_free(zmbuf->mbuf);
 922                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 923         }
 924
 925         rte_free(vq->zmbufs);
 926 }
 927
 928 /*
 929  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 930  */
 931 static int
 932 vhost_user_get_vring_base(struct virtio_net *dev,
 933                           VhostUserMsg *msg)
 934 {
 935         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 936
 937         /* We have to stop the queue (virtio) if it is running. */
 938         if (dev->flags & VIRTIO_DEV_RUNNING) {
 939                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 940                 dev->notify_ops->destroy_device(dev->vid);
 941         }
 942
 943         dev->flags &= ~VIRTIO_DEV_READY;
 944
 945         /* Here we are safe to get the last used index */
 946         msg->payload.state.num = vq->last_used_idx;
 947
 948         RTE_LOG(INFO, VHOST_CONFIG,
 949                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 950                 msg->payload.state.num);
 951         /*
 952          * Based on current qemu vhost-user implementation, this message is
 953          * sent and only sent in vhost_vring_stop.
 954          * TODO: cleanup the vring, it isn't usable since here.
 955          */
 956         if (vq->kickfd >= 0)
 957                 close(vq->kickfd);
 958
 959         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 960
 961         if (vq->callfd >= 0)
 962                 close(vq->callfd);
 963
 964         vq->callfd = VIRTIO_UNINITIALIZED_EVENTFD;
 965
 966         if (dev->dequeue_zero_copy)
 967                 free_zmbufs(vq);
 968         rte_free(vq->shadow_used_ring);
 969         vq->shadow_used_ring = NULL;
 970
 971         rte_free(vq->batch_copy_elems);
 972         vq->batch_copy_elems = NULL;
 973
 974         return 0;
 975 }
 976
 977 /*
 978  * when virtio queues are ready to work, qemu will send us to
 979  * enable the virtio queue pair.
 980  */
 981 static int
 982 vhost_user_set_vring_enable(struct virtio_net *dev,
 983                             VhostUserMsg *msg)
 984 {
 985         int enable = (int)msg->payload.state.num;
 986
 987         RTE_LOG(INFO, VHOST_CONFIG,
 988                 "set queue enable: %d to qp idx: %d\n",
 989                 enable, msg->payload.state.index);
 990
 991         if (dev->notify_ops->vring_state_changed)
 992                 dev->notify_ops->vring_state_changed(dev->vid,
 993                                 msg->payload.state.index, enable);
 994
 995         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 996
 997         return 0;
 998 }
 999
1000 static void
1001 vhost_user_get_protocol_features(struct virtio_net *dev,
1002                                  struct VhostUserMsg *msg)
1003 {
1004         uint64_t features, protocol_features = VHOST_USER_PROTOCOL_FEATURES;
1005
1006         rte_vhost_driver_get_features(dev->ifname, &features);
1007
1008         /*
1009          * REPLY_ACK protocol feature is only mandatory for now
1010          * for IOMMU feature. If IOMMU is explicitly disabled by the
1011          * application, disable also REPLY_ACK feature for older buggy
1012          * Qemu versions (from v2.7.0 to v2.9.0).
1013          */
1014         if (!(features & (1ULL << VIRTIO_F_IOMMU_PLATFORM)))
1015                 protocol_features &= ~(1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK);
1016
1017         msg->payload.u64 = protocol_features;
1018         msg->size = sizeof(msg->payload.u64);
1019 }
1020
1021 static void
1022 vhost_user_set_protocol_features(struct virtio_net *dev,
1023                                  uint64_t protocol_features)
1024 {
1025         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
1026                 return;
1027
1028         dev->protocol_features = protocol_features;
1029 }
1030
1031 static int
1032 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
1033 {
1034         int fd = msg->fds[0];
1035         uint64_t size, off;
1036         void *addr;
1037
1038         if (fd < 0) {
1039                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
1040                 return -1;
1041         }
1042
1043         if (msg->size != sizeof(VhostUserLog)) {
1044                 RTE_LOG(ERR, VHOST_CONFIG,
1045                         "invalid log base msg size: %"PRId32" != %d\n",
1046                         msg->size, (int)sizeof(VhostUserLog));
1047                 return -1;
1048         }
1049
1050         size = msg->payload.log.mmap_size;
1051         off  = msg->payload.log.mmap_offset;
1052
1053         /* Don't allow mmap_offset to point outside the mmap region */
1054         if (off > size) {
1055                 RTE_LOG(ERR, VHOST_CONFIG,
1056                         "log offset %#"PRIx64" exceeds log size %#"PRIx64"\n",
1057                         off, size);
1058                 return -1;
1059         }
1060
1061         RTE_LOG(INFO, VHOST_CONFIG,
1062                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
1063                 size, off);
1064
1065         /*
1066          * mmap from 0 to workaround a hugepage mmap bug: mmap will
1067          * fail when offset is not page size aligned.
1068          */
1069         addr = mmap(0, size + off, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
1070         close(fd);
1071         if (addr == MAP_FAILED) {
1072                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
1073                 return -1;
1074         }
1075
1076         /*
1077          * Free previously mapped log memory on occasionally
1078          * multiple VHOST_USER_SET_LOG_BASE.
1079          */
1080         if (dev->log_addr) {
1081                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
1082         }
1083         dev->log_addr = (uint64_t)(uintptr_t)addr;
1084         dev->log_base = dev->log_addr + off;
1085         dev->log_size = size;
1086
1087         return 0;
1088 }
1089
1090 /*
1091  * An rarp packet is constructed and broadcasted to notify switches about
1092  * the new location of the migrated VM, so that packets from outside will
1093  * not be lost after migration.
1094  *
1095  * However, we don't actually "send" a rarp packet here, instead, we set
1096  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
1097  */
1098 static int
1099 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
1100 {
1101         uint8_t *mac = (uint8_t *)&msg->payload.u64;
1102
1103         RTE_LOG(DEBUG, VHOST_CONFIG,
1104                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
1105                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1106         memcpy(dev->mac.addr_bytes, mac, 6);
1107
1108         /*
1109          * Set the flag to inject a RARP broadcast packet at
1110          * rte_vhost_dequeue_burst().
1111          *
1112          * rte_smp_wmb() is for making sure the mac is copied
1113          * before the flag is set.
1114          */
1115         rte_smp_wmb();
1116         rte_atomic16_set(&dev->broadcast_rarp, 1);
1117
1118         return 0;
1119 }
1120
1121 static int
1122 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
1123 {
1124         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
1125                         msg->payload.u64 > VIRTIO_MAX_MTU) {
1126                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
1127                                 msg->payload.u64);
1128
1129                 return -1;
1130         }
1131
1132         dev->mtu = msg->payload.u64;
1133
1134         return 0;
1135 }
1136
1137 static int
1138 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
1139 {
1140         int fd = msg->fds[0];
1141
1142         if (fd < 0) {
1143                 RTE_LOG(ERR, VHOST_CONFIG,
1144                                 "Invalid file descriptor for slave channel (%d)\n",
1145                                 fd);
1146                 return -1;
1147         }
1148
1149         dev->slave_req_fd = fd;
1150
1151         return 0;
1152 }
1153
1154 static int
1155 is_vring_iotlb_update(struct vhost_virtqueue *vq, struct vhost_iotlb_msg *imsg)
1156 {
1157         struct vhost_vring_addr *ra;
1158         uint64_t start, end;
1159
1160         start = imsg->iova;
1161         end = start + imsg->size;
1162
1163         ra = &vq->ring_addrs;
1164         if (ra->desc_user_addr >= start && ra->desc_user_addr < end)
1165                 return 1;
1166         if (ra->avail_user_addr >= start && ra->avail_user_addr < end)
1167                 return 1;
1168         if (ra->used_user_addr >= start && ra->used_user_addr < end)
1169                 return 1;
1170
1171         return 0;
1172 }
1173
1174 static int
1175 is_vring_iotlb_invalidate(struct vhost_virtqueue *vq,
1176                                 struct vhost_iotlb_msg *imsg)
1177 {
1178         uint64_t istart, iend, vstart, vend;
1179
1180         istart = imsg->iova;
1181         iend = istart + imsg->size - 1;
1182
1183         vstart = (uintptr_t)vq->desc;
1184         vend = vstart + sizeof(struct vring_desc) * vq->size - 1;
1185         if (vstart <= iend && istart <= vend)
1186                 return 1;
1187
1188         vstart = (uintptr_t)vq->avail;
1189         vend = vstart + sizeof(struct vring_avail);
1190         vend += sizeof(uint16_t) * vq->size - 1;
1191         if (vstart <= iend && istart <= vend)
1192                 return 1;
1193
1194         vstart = (uintptr_t)vq->used;
1195         vend = vstart + sizeof(struct vring_used);
1196         vend += sizeof(struct vring_used_elem) * vq->size - 1;
1197         if (vstart <= iend && istart <= vend)
1198                 return 1;
1199
1200         return 0;
1201 }
1202
1203 static int
1204 vhost_user_iotlb_msg(struct virtio_net **pdev, struct VhostUserMsg *msg)
1205 {
1206         struct virtio_net *dev = *pdev;
1207         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
1208         uint16_t i;
1209         uint64_t vva;
1210
1211         switch (imsg->type) {
1212         case VHOST_IOTLB_UPDATE:
1213                 vva = qva_to_vva(dev, imsg->uaddr);
1214                 if (!vva)
1215                         return -1;
1216
1217                 for (i = 0; i < dev->nr_vring; i++) {
1218                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1219
1220                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
1221                                         imsg->size, imsg->perm);
1222
1223                         if (is_vring_iotlb_update(vq, imsg))
1224                                 *pdev = dev = translate_ring_addresses(dev, i);
1225                 }
1226                 break;
1227         case VHOST_IOTLB_INVALIDATE:
1228                 for (i = 0; i < dev->nr_vring; i++) {
1229                         struct vhost_virtqueue *vq = dev->virtqueue[i];
1230
1231                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
1232                                         imsg->size);
1233
1234                         if (is_vring_iotlb_invalidate(vq, imsg))
1235                                 vring_invalidate(dev, vq);
1236                 }
1237                 break;
1238         default:
1239                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
1240                                 imsg->type);
1241                 return -1;
1242         }
1243
1244         return 0;
1245 }
1246
1247 /* return bytes# of read on success or negative val on failure. */
1248 static int
1249 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
1250 {
1251         int ret;
1252
1253         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
1254                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
1255         if (ret <= 0)
1256                 return ret;
1257
1258         if (msg && msg->size) {
1259                 if (msg->size > sizeof(msg->payload)) {
1260                         RTE_LOG(ERR, VHOST_CONFIG,
1261                                 "invalid msg size: %d\n", msg->size);
1262                         return -1;
1263                 }
1264                 ret = read(sockfd, &msg->payload, msg->size);
1265                 if (ret <= 0)
1266                         return ret;
1267                 if (ret != (int)msg->size) {
1268                         RTE_LOG(ERR, VHOST_CONFIG,
1269                                 "read control message failed\n");
1270                         return -1;
1271                 }
1272         }
1273
1274         return ret;
1275 }
1276
1277 static int
1278 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
1279 {
1280         if (!msg)
1281                 return 0;
1282
1283         return send_fd_message(sockfd, (char *)msg,
1284                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
1285 }
1286
1287 static int
1288 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
1289 {
1290         if (!msg)
1291                 return 0;
1292
1293         msg->flags &= ~VHOST_USER_VERSION_MASK;
1294         msg->flags &= ~VHOST_USER_NEED_REPLY;
1295         msg->flags |= VHOST_USER_VERSION;
1296         msg->flags |= VHOST_USER_REPLY_MASK;
1297
1298         return send_vhost_message(sockfd, msg);
1299 }
1300
1301 /*
1302  * Allocate a queue pair if it hasn't been allocated yet
1303  */
1304 static int
1305 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1306 {
1307         uint16_t vring_idx;
1308
1309         switch (msg->request.master) {
1310         case VHOST_USER_SET_VRING_KICK:
1311         case VHOST_USER_SET_VRING_CALL:
1312         case VHOST_USER_SET_VRING_ERR:
1313                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1314                 break;
1315         case VHOST_USER_SET_VRING_NUM:
1316         case VHOST_USER_SET_VRING_BASE:
1317         case VHOST_USER_SET_VRING_ENABLE:
1318                 vring_idx = msg->payload.state.index;
1319                 break;
1320         case VHOST_USER_SET_VRING_ADDR:
1321                 vring_idx = msg->payload.addr.index;
1322                 break;
1323         default:
1324                 return 0;
1325         }
1326
1327         if (vring_idx >= VHOST_MAX_VRING) {
1328                 RTE_LOG(ERR, VHOST_CONFIG,
1329                         "invalid vring index: %u\n", vring_idx);
1330                 return -1;
1331         }
1332
1333         if (dev->virtqueue[vring_idx])
1334                 return 0;
1335
1336         return alloc_vring_queue(dev, vring_idx);
1337 }
1338
1339 static void
1340 vhost_user_lock_all_queue_pairs(struct virtio_net *dev)
1341 {
1342         unsigned int i = 0;
1343         unsigned int vq_num = 0;
1344
1345         while (vq_num < dev->nr_vring) {
1346                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1347
1348                 if (vq) {
1349                         rte_spinlock_lock(&vq->access_lock);
1350                         vq_num++;
1351                 }
1352                 i++;
1353         }
1354 }
1355
1356 static void
1357 vhost_user_unlock_all_queue_pairs(struct virtio_net *dev)
1358 {
1359         unsigned int i = 0;
1360         unsigned int vq_num = 0;
1361
1362         while (vq_num < dev->nr_vring) {
1363                 struct vhost_virtqueue *vq = dev->virtqueue[i];
1364
1365                 if (vq) {
1366                         rte_spinlock_unlock(&vq->access_lock);
1367                         vq_num++;
1368                 }
1369                 i++;
1370         }
1371 }
1372
1373 int
1374 vhost_user_msg_handler(int vid, int fd)
1375 {
1376         struct virtio_net *dev;
1377         struct VhostUserMsg msg;
1378         int ret;
1379         int unlock_required = 0;
1380
1381         dev = get_device(vid);
1382         if (dev == NULL)
1383                 return -1;
1384
1385         if (!dev->notify_ops) {
1386                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1387                 if (!dev->notify_ops) {
1388                         RTE_LOG(ERR, VHOST_CONFIG,
1389                                 "failed to get callback ops for driver %s\n",
1390                                 dev->ifname);
1391                         return -1;
1392                 }
1393         }
1394
1395         ret = read_vhost_message(fd, &msg);
1396         if (ret <= 0 || msg.request.master >= VHOST_USER_MAX) {
1397                 if (ret < 0)
1398                         RTE_LOG(ERR, VHOST_CONFIG,
1399                                 "vhost read message failed\n");
1400                 else if (ret == 0)
1401                         RTE_LOG(INFO, VHOST_CONFIG,
1402                                 "vhost peer closed\n");
1403                 else
1404                         RTE_LOG(ERR, VHOST_CONFIG,
1405                                 "vhost read incorrect message\n");
1406
1407                 return -1;
1408         }
1409
1410         ret = 0;
1411         if (msg.request.master != VHOST_USER_IOTLB_MSG)
1412                 RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1413                         vhost_message_str[msg.request.master]);
1414         else
1415                 RTE_LOG(DEBUG, VHOST_CONFIG, "read message %s\n",
1416                         vhost_message_str[msg.request.master]);
1417
1418         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1419         if (ret < 0) {
1420                 RTE_LOG(ERR, VHOST_CONFIG,
1421                         "failed to alloc queue\n");
1422                 return -1;
1423         }
1424
1425         /*
1426          * Note: we don't lock all queues on VHOST_USER_GET_VRING_BASE
1427          * and VHOST_USER_RESET_OWNER, since it is sent when virtio stops
1428          * and device is destroyed. destroy_device waits for queues to be
1429          * inactive, so it is safe. Otherwise taking the access_lock
1430          * would cause a dead lock.
1431          */
1432         switch (msg.request.master) {
1433         case VHOST_USER_SET_FEATURES:
1434         case VHOST_USER_SET_PROTOCOL_FEATURES:
1435         case VHOST_USER_SET_OWNER:
1436         case VHOST_USER_SET_MEM_TABLE:
1437         case VHOST_USER_SET_LOG_BASE:
1438         case VHOST_USER_SET_LOG_FD:
1439         case VHOST_USER_SET_VRING_NUM:
1440         case VHOST_USER_SET_VRING_ADDR:
1441         case VHOST_USER_SET_VRING_BASE:
1442         case VHOST_USER_SET_VRING_KICK:
1443         case VHOST_USER_SET_VRING_CALL:
1444         case VHOST_USER_SET_VRING_ERR:
1445         case VHOST_USER_SET_VRING_ENABLE:
1446         case VHOST_USER_SEND_RARP:
1447         case VHOST_USER_NET_SET_MTU:
1448         case VHOST_USER_SET_SLAVE_REQ_FD:
1449                 vhost_user_lock_all_queue_pairs(dev);
1450                 unlock_required = 1;
1451                 break;
1452         default:
1453                 break;
1454
1455         }
1456
1457         switch (msg.request.master) {
1458         case VHOST_USER_GET_FEATURES:
1459                 msg.payload.u64 = vhost_user_get_features(dev);
1460                 msg.size = sizeof(msg.payload.u64);
1461                 send_vhost_reply(fd, &msg);
1462                 break;
1463         case VHOST_USER_SET_FEATURES:
1464                 ret = vhost_user_set_features(dev, msg.payload.u64);
1465                 if (ret)
1466                         return -1;
1467                 break;
1468
1469         case VHOST_USER_GET_PROTOCOL_FEATURES:
1470                 vhost_user_get_protocol_features(dev, &msg);
1471                 send_vhost_reply(fd, &msg);
1472                 break;
1473         case VHOST_USER_SET_PROTOCOL_FEATURES:
1474                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1475                 break;
1476
1477         case VHOST_USER_SET_OWNER:
1478                 vhost_user_set_owner();
1479                 break;
1480         case VHOST_USER_RESET_OWNER:
1481                 vhost_user_reset_owner(dev);
1482                 break;
1483
1484         case VHOST_USER_SET_MEM_TABLE:
1485                 ret = vhost_user_set_mem_table(dev, &msg);
1486                 break;
1487
1488         case VHOST_USER_SET_LOG_BASE:
1489                 vhost_user_set_log_base(dev, &msg);
1490
1491                 /* it needs a reply */
1492                 msg.size = sizeof(msg.payload.u64);
1493                 send_vhost_reply(fd, &msg);
1494                 break;
1495         case VHOST_USER_SET_LOG_FD:
1496                 close(msg.fds[0]);
1497                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1498                 break;
1499
1500         case VHOST_USER_SET_VRING_NUM:
1501                 vhost_user_set_vring_num(dev, &msg);
1502                 break;
1503         case VHOST_USER_SET_VRING_ADDR:
1504                 vhost_user_set_vring_addr(&dev, &msg);
1505                 break;
1506         case VHOST_USER_SET_VRING_BASE:
1507                 vhost_user_set_vring_base(dev, &msg);
1508                 break;
1509
1510         case VHOST_USER_GET_VRING_BASE:
1511                 vhost_user_get_vring_base(dev, &msg);
1512                 msg.size = sizeof(msg.payload.state);
1513                 send_vhost_reply(fd, &msg);
1514                 break;
1515
1516         case VHOST_USER_SET_VRING_KICK:
1517                 vhost_user_set_vring_kick(&dev, &msg);
1518                 break;
1519         case VHOST_USER_SET_VRING_CALL:
1520                 vhost_user_set_vring_call(dev, &msg);
1521                 break;
1522
1523         case VHOST_USER_SET_VRING_ERR:
1524                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1525                         close(msg.fds[0]);
1526                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1527                 break;
1528
1529         case VHOST_USER_GET_QUEUE_NUM:
1530                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1531                 msg.size = sizeof(msg.payload.u64);
1532                 send_vhost_reply(fd, &msg);
1533                 break;
1534
1535         case VHOST_USER_SET_VRING_ENABLE:
1536                 vhost_user_set_vring_enable(dev, &msg);
1537                 break;
1538         case VHOST_USER_SEND_RARP:
1539                 vhost_user_send_rarp(dev, &msg);
1540                 break;
1541
1542         case VHOST_USER_NET_SET_MTU:
1543                 ret = vhost_user_net_set_mtu(dev, &msg);
1544                 break;
1545
1546         case VHOST_USER_SET_SLAVE_REQ_FD:
1547                 ret = vhost_user_set_req_fd(dev, &msg);
1548                 break;
1549
1550         case VHOST_USER_IOTLB_MSG:
1551                 ret = vhost_user_iotlb_msg(&dev, &msg);
1552                 break;
1553
1554         default:
1555                 ret = -1;
1556                 break;
1557
1558         }
1559
1560         if (unlock_required)
1561                 vhost_user_unlock_all_queue_pairs(dev);
1562
1563         if (msg.flags & VHOST_USER_NEED_REPLY) {
1564                 msg.payload.u64 = !!ret;
1565                 msg.size = sizeof(msg.payload.u64);
1566                 send_vhost_reply(fd, &msg);
1567         }
1568
1569         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1570                 dev->flags |= VIRTIO_DEV_READY;
1571
1572                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1573                         if (dev->dequeue_zero_copy) {
1574                                 RTE_LOG(INFO, VHOST_CONFIG,
1575                                                 "dequeue zero copy is enabled\n");
1576                         }
1577
1578                         if (dev->notify_ops->new_device(dev->vid) == 0)
1579                                 dev->flags |= VIRTIO_DEV_RUNNING;
1580                 }
1581         }
1582
1583         return 0;
1584 }
1585
1586 int
1587 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1588 {
1589         int ret;
1590         struct VhostUserMsg msg = {
1591                 .request.slave = VHOST_USER_SLAVE_IOTLB_MSG,
1592                 .flags = VHOST_USER_VERSION,
1593                 .size = sizeof(msg.payload.iotlb),
1594                 .payload.iotlb = {
1595                         .iova = iova,
1596                         .perm = perm,
1597                         .type = VHOST_IOTLB_MISS,
1598                 },
1599         };
1600
1601         ret = send_vhost_message(dev->slave_req_fd, &msg);
1602         if (ret < 0) {
1603                 RTE_LOG(ERR, VHOST_CONFIG,
1604                                 "Failed to send IOTLB miss message (%d)\n",
1605                                 ret);
1606                 return ret;
1607         }
1608
1609         return 0;
1610 }