lib/librte_vhost/vhost_user.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdint.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <string.h>
  38 #include <unistd.h>
  39 #include <sys/mman.h>
  40 #include <sys/types.h>
  41 #include <sys/stat.h>
  42 #include <assert.h>
  43 #ifdef RTE_LIBRTE_VHOST_NUMA
  44 #include <numaif.h>
  45 #endif
  46
  47 #include <rte_common.h>
  48 #include <rte_malloc.h>
  49 #include <rte_log.h>
  50
  51 #include "vhost.h"
  52 #include "vhost_user.h"
  53
  54 #define VIRTIO_MIN_MTU 68
  55 #define VIRTIO_MAX_MTU 65535
  56
  57 static const char *vhost_message_str[VHOST_USER_MAX] = {
  58         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  59         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  60         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  61         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  62         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  63         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  64         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  65         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  66         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  67         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  68         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  69         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  70         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  71         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  72         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  73         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  74         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  75         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  76         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  77         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  78         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  79 };
  80
  81 static uint64_t
  82 get_blk_size(int fd)
  83 {
  84         struct stat stat;
  85         int ret;
  86
  87         ret = fstat(fd, &stat);
  88         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  89 }
  90
  91 static void
  92 free_mem_region(struct virtio_net *dev)
  93 {
  94         uint32_t i;
  95         struct rte_vhost_mem_region *reg;
  96
  97         if (!dev || !dev->mem)
  98                 return;
  99
 100         for (i = 0; i < dev->mem->nregions; i++) {
 101                 reg = &dev->mem->regions[i];
 102                 if (reg->host_user_addr) {
 103                         munmap(reg->mmap_addr, reg->mmap_size);
 104                         close(reg->fd);
 105                 }
 106         }
 107 }
 108
 109 void
 110 vhost_backend_cleanup(struct virtio_net *dev)
 111 {
 112         if (dev->mem) {
 113                 free_mem_region(dev);
 114                 rte_free(dev->mem);
 115                 dev->mem = NULL;
 116         }
 117
 118         free(dev->guest_pages);
 119         dev->guest_pages = NULL;
 120
 121         if (dev->log_addr) {
 122                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 123                 dev->log_addr = 0;
 124         }
 125 }
 126
 127 /*
 128  * This function just returns success at the moment unless
 129  * the device hasn't been initialised.
 130  */
 131 static int
 132 vhost_user_set_owner(void)
 133 {
 134         return 0;
 135 }
 136
 137 static int
 138 vhost_user_reset_owner(struct virtio_net *dev)
 139 {
 140         if (dev->flags & VIRTIO_DEV_RUNNING) {
 141                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 142                 dev->notify_ops->destroy_device(dev->vid);
 143         }
 144
 145         cleanup_device(dev, 0);
 146         reset_device(dev);
 147         return 0;
 148 }
 149
 150 /*
 151  * The features that we support are requested.
 152  */
 153 static uint64_t
 154 vhost_user_get_features(struct virtio_net *dev)
 155 {
 156         uint64_t features = 0;
 157
 158         rte_vhost_driver_get_features(dev->ifname, &features);
 159         return features;
 160 }
 161
 162 /*
 163  * We receive the negotiated features supported by us and the virtio device.
 164  */
 165 static int
 166 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 167 {
 168         uint64_t vhost_features = 0;
 169
 170         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 171         if (features & ~vhost_features) {
 172                 RTE_LOG(ERR, VHOST_CONFIG,
 173                         "(%d) received invalid negotiated features.\n",
 174                         dev->vid);
 175                 return -1;
 176         }
 177
 178         if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
 179                 if (dev->notify_ops->features_changed)
 180                         dev->notify_ops->features_changed(dev->vid, features);
 181         }
 182
 183         dev->features = features;
 184         if (dev->features &
 185                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 186                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 187         } else {
 188                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 189         }
 190         LOG_DEBUG(VHOST_CONFIG,
 191                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 192                 dev->vid,
 193                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 194                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 195
 196         return 0;
 197 }
 198
 199 /*
 200  * The virtio device sends us the size of the descriptor ring.
 201  */
 202 static int
 203 vhost_user_set_vring_num(struct virtio_net *dev,
 204                          VhostUserMsg *msg)
 205 {
 206         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 207
 208         vq->size = msg->payload.state.num;
 209
 210         if (dev->dequeue_zero_copy) {
 211                 vq->nr_zmbuf = 0;
 212                 vq->last_zmbuf_idx = 0;
 213                 vq->zmbuf_size = vq->size;
 214                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 215                                          sizeof(struct zcopy_mbuf), 0);
 216                 if (vq->zmbufs == NULL) {
 217                         RTE_LOG(WARNING, VHOST_CONFIG,
 218                                 "failed to allocate mem for zero copy; "
 219                                 "zero copy is force disabled\n");
 220                         dev->dequeue_zero_copy = 0;
 221                 }
 222         }
 223
 224         vq->shadow_used_ring = rte_malloc(NULL,
 225                                 vq->size * sizeof(struct vring_used_elem),
 226                                 RTE_CACHE_LINE_SIZE);
 227         if (!vq->shadow_used_ring) {
 228                 RTE_LOG(ERR, VHOST_CONFIG,
 229                         "failed to allocate memory for shadow used ring.\n");
 230                 return -1;
 231         }
 232
 233         vq->batch_copy_elems = rte_malloc(NULL,
 234                                 vq->size * sizeof(struct batch_copy_elem),
 235                                 RTE_CACHE_LINE_SIZE);
 236         if (!vq->batch_copy_elems) {
 237                 RTE_LOG(ERR, VHOST_CONFIG,
 238                         "failed to allocate memory for batching copy.\n");
 239                 return -1;
 240         }
 241
 242         return 0;
 243 }
 244
 245 /*
 246  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 247  * same numa node as the memory of vring descriptor.
 248  */
 249 #ifdef RTE_LIBRTE_VHOST_NUMA
 250 static struct virtio_net*
 251 numa_realloc(struct virtio_net *dev, int index)
 252 {
 253         int oldnode, newnode;
 254         struct virtio_net *old_dev;
 255         struct vhost_virtqueue *old_vq, *vq;
 256         int ret;
 257
 258         old_dev = dev;
 259         vq = old_vq = dev->virtqueue[index];
 260
 261         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 262                             MPOL_F_NODE | MPOL_F_ADDR);
 263
 264         /* check if we need to reallocate vq */
 265         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 266                              MPOL_F_NODE | MPOL_F_ADDR);
 267         if (ret) {
 268                 RTE_LOG(ERR, VHOST_CONFIG,
 269                         "Unable to get vq numa information.\n");
 270                 return dev;
 271         }
 272         if (oldnode != newnode) {
 273                 RTE_LOG(INFO, VHOST_CONFIG,
 274                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 275                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 276                 if (!vq)
 277                         return dev;
 278
 279                 memcpy(vq, old_vq, sizeof(*vq));
 280                 rte_free(old_vq);
 281         }
 282
 283         /* check if we need to reallocate dev */
 284         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 285                             MPOL_F_NODE | MPOL_F_ADDR);
 286         if (ret) {
 287                 RTE_LOG(ERR, VHOST_CONFIG,
 288                         "Unable to get dev numa information.\n");
 289                 goto out;
 290         }
 291         if (oldnode != newnode) {
 292                 RTE_LOG(INFO, VHOST_CONFIG,
 293                         "reallocate dev from %d to %d node\n",
 294                         oldnode, newnode);
 295                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 296                 if (!dev) {
 297                         dev = old_dev;
 298                         goto out;
 299                 }
 300
 301                 memcpy(dev, old_dev, sizeof(*dev));
 302                 rte_free(old_dev);
 303         }
 304
 305 out:
 306         dev->virtqueue[index] = vq;
 307         vhost_devices[dev->vid] = dev;
 308
 309         return dev;
 310 }
 311 #else
 312 static struct virtio_net*
 313 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 314 {
 315         return dev;
 316 }
 317 #endif
 318
 319 /*
 320  * Converts QEMU virtual address to Vhost virtual address. This function is
 321  * used to convert the ring addresses to our address space.
 322  */
 323 static uint64_t
 324 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 325 {
 326         struct rte_vhost_mem_region *reg;
 327         uint32_t i;
 328
 329         /* Find the region where the address lives. */
 330         for (i = 0; i < dev->mem->nregions; i++) {
 331                 reg = &dev->mem->regions[i];
 332
 333                 if (qva >= reg->guest_user_addr &&
 334                     qva <  reg->guest_user_addr + reg->size) {
 335                         return qva - reg->guest_user_addr +
 336                                reg->host_user_addr;
 337                 }
 338         }
 339
 340         return 0;
 341 }
 342
 343 /*
 344  * The virtio device sends us the desc, used and avail ring addresses.
 345  * This function then converts these to our address space.
 346  */
 347 static int
 348 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
 349 {
 350         struct vhost_virtqueue *vq;
 351
 352         if (dev->mem == NULL)
 353                 return -1;
 354
 355         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 356         vq = dev->virtqueue[msg->payload.addr.index];
 357
 358         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 359         vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
 360                         msg->payload.addr.desc_user_addr);
 361         if (vq->desc == 0) {
 362                 RTE_LOG(ERR, VHOST_CONFIG,
 363                         "(%d) failed to find desc ring address.\n",
 364                         dev->vid);
 365                 return -1;
 366         }
 367
 368         dev = numa_realloc(dev, msg->payload.addr.index);
 369         vq = dev->virtqueue[msg->payload.addr.index];
 370
 371         vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
 372                         msg->payload.addr.avail_user_addr);
 373         if (vq->avail == 0) {
 374                 RTE_LOG(ERR, VHOST_CONFIG,
 375                         "(%d) failed to find avail ring address.\n",
 376                         dev->vid);
 377                 return -1;
 378         }
 379
 380         vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
 381                         msg->payload.addr.used_user_addr);
 382         if (vq->used == 0) {
 383                 RTE_LOG(ERR, VHOST_CONFIG,
 384                         "(%d) failed to find used ring address.\n",
 385                         dev->vid);
 386                 return -1;
 387         }
 388
 389         if (vq->last_used_idx != vq->used->idx) {
 390                 RTE_LOG(WARNING, VHOST_CONFIG,
 391                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 392                         "some packets maybe resent for Tx and dropped for Rx\n",
 393                         vq->last_used_idx, vq->used->idx);
 394                 vq->last_used_idx  = vq->used->idx;
 395                 vq->last_avail_idx = vq->used->idx;
 396         }
 397
 398         vq->log_guest_addr = msg->payload.addr.log_guest_addr;
 399
 400         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 401                         dev->vid, vq->desc);
 402         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 403                         dev->vid, vq->avail);
 404         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 405                         dev->vid, vq->used);
 406         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 407                         dev->vid, vq->log_guest_addr);
 408
 409         return 0;
 410 }
 411
 412 /*
 413  * The virtio device sends us the available ring last used index.
 414  */
 415 static int
 416 vhost_user_set_vring_base(struct virtio_net *dev,
 417                           VhostUserMsg *msg)
 418 {
 419         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 420                         msg->payload.state.num;
 421         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 422                         msg->payload.state.num;
 423
 424         return 0;
 425 }
 426
 427 static void
 428 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 429                    uint64_t host_phys_addr, uint64_t size)
 430 {
 431         struct guest_page *page, *last_page;
 432
 433         if (dev->nr_guest_pages == dev->max_guest_pages) {
 434                 dev->max_guest_pages *= 2;
 435                 dev->guest_pages = realloc(dev->guest_pages,
 436                                         dev->max_guest_pages * sizeof(*page));
 437         }
 438
 439         if (dev->nr_guest_pages > 0) {
 440                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 441                 /* merge if the two pages are continuous */
 442                 if (host_phys_addr == last_page->host_phys_addr +
 443                                       last_page->size) {
 444                         last_page->size += size;
 445                         return;
 446                 }
 447         }
 448
 449         page = &dev->guest_pages[dev->nr_guest_pages++];
 450         page->guest_phys_addr = guest_phys_addr;
 451         page->host_phys_addr  = host_phys_addr;
 452         page->size = size;
 453 }
 454
 455 static void
 456 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 457                 uint64_t page_size)
 458 {
 459         uint64_t reg_size = reg->size;
 460         uint64_t host_user_addr  = reg->host_user_addr;
 461         uint64_t guest_phys_addr = reg->guest_phys_addr;
 462         uint64_t host_phys_addr;
 463         uint64_t size;
 464
 465         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 466         size = page_size - (guest_phys_addr & (page_size - 1));
 467         size = RTE_MIN(size, reg_size);
 468
 469         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 470         host_user_addr  += size;
 471         guest_phys_addr += size;
 472         reg_size -= size;
 473
 474         while (reg_size > 0) {
 475                 size = RTE_MIN(reg_size, page_size);
 476                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 477                                                   host_user_addr);
 478                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 479
 480                 host_user_addr  += size;
 481                 guest_phys_addr += size;
 482                 reg_size -= size;
 483         }
 484 }
 485
 486 #ifdef RTE_LIBRTE_VHOST_DEBUG
 487 /* TODO: enable it only in debug mode? */
 488 static void
 489 dump_guest_pages(struct virtio_net *dev)
 490 {
 491         uint32_t i;
 492         struct guest_page *page;
 493
 494         for (i = 0; i < dev->nr_guest_pages; i++) {
 495                 page = &dev->guest_pages[i];
 496
 497                 RTE_LOG(INFO, VHOST_CONFIG,
 498                         "guest physical page region %u\n"
 499                         "\t guest_phys_addr: %" PRIx64 "\n"
 500                         "\t host_phys_addr : %" PRIx64 "\n"
 501                         "\t size           : %" PRIx64 "\n",
 502                         i,
 503                         page->guest_phys_addr,
 504                         page->host_phys_addr,
 505                         page->size);
 506         }
 507 }
 508 #else
 509 #define dump_guest_pages(dev)
 510 #endif
 511
 512 static int
 513 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 514 {
 515         struct VhostUserMemory memory = pmsg->payload.memory;
 516         struct rte_vhost_mem_region *reg;
 517         void *mmap_addr;
 518         uint64_t mmap_size;
 519         uint64_t mmap_offset;
 520         uint64_t alignment;
 521         uint32_t i;
 522         int fd;
 523
 524         if (dev->mem) {
 525                 free_mem_region(dev);
 526                 rte_free(dev->mem);
 527                 dev->mem = NULL;
 528         }
 529
 530         dev->nr_guest_pages = 0;
 531         if (!dev->guest_pages) {
 532                 dev->max_guest_pages = 8;
 533                 dev->guest_pages = malloc(dev->max_guest_pages *
 534                                                 sizeof(struct guest_page));
 535                 if (dev->guest_pages == NULL) {
 536                         RTE_LOG(ERR, VHOST_CONFIG,
 537                                 "(%d) failed to allocate memory "
 538                                 "for dev->guest_pages\n",
 539                                 dev->vid);
 540                         return -1;
 541                 }
 542         }
 543
 544         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 545                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 546         if (dev->mem == NULL) {
 547                 RTE_LOG(ERR, VHOST_CONFIG,
 548                         "(%d) failed to allocate memory for dev->mem\n",
 549                         dev->vid);
 550                 return -1;
 551         }
 552         dev->mem->nregions = memory.nregions;
 553
 554         for (i = 0; i < memory.nregions; i++) {
 555                 fd  = pmsg->fds[i];
 556                 reg = &dev->mem->regions[i];
 557
 558                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 559                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 560                 reg->size            = memory.regions[i].memory_size;
 561                 reg->fd              = fd;
 562
 563                 mmap_offset = memory.regions[i].mmap_offset;
 564                 mmap_size   = reg->size + mmap_offset;
 565
 566                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 567                  * with length argument aligned with hugepagesz at older
 568                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 569                  * mmap() will fail with EINVAL.
 570                  *
 571                  * to avoid failure, make sure in caller to keep length
 572                  * aligned.
 573                  */
 574                 alignment = get_blk_size(fd);
 575                 if (alignment == (uint64_t)-1) {
 576                         RTE_LOG(ERR, VHOST_CONFIG,
 577                                 "couldn't get hugepage size through fstat\n");
 578                         goto err_mmap;
 579                 }
 580                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 581
 582                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 583                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 584
 585                 if (mmap_addr == MAP_FAILED) {
 586                         RTE_LOG(ERR, VHOST_CONFIG,
 587                                 "mmap region %u failed.\n", i);
 588                         goto err_mmap;
 589                 }
 590
 591                 reg->mmap_addr = mmap_addr;
 592                 reg->mmap_size = mmap_size;
 593                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 594                                       mmap_offset;
 595
 596                 if (dev->dequeue_zero_copy)
 597                         add_guest_pages(dev, reg, alignment);
 598
 599                 RTE_LOG(INFO, VHOST_CONFIG,
 600                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 601                         "\t guest physical addr: 0x%" PRIx64 "\n"
 602                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 603                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 604                         "\t mmap addr : 0x%" PRIx64 "\n"
 605                         "\t mmap size : 0x%" PRIx64 "\n"
 606                         "\t mmap align: 0x%" PRIx64 "\n"
 607                         "\t mmap off  : 0x%" PRIx64 "\n",
 608                         i, reg->size,
 609                         reg->guest_phys_addr,
 610                         reg->guest_user_addr,
 611                         reg->host_user_addr,
 612                         (uint64_t)(uintptr_t)mmap_addr,
 613                         mmap_size,
 614                         alignment,
 615                         mmap_offset);
 616         }
 617
 618         dump_guest_pages(dev);
 619
 620         return 0;
 621
 622 err_mmap:
 623         free_mem_region(dev);
 624         rte_free(dev->mem);
 625         dev->mem = NULL;
 626         return -1;
 627 }
 628
 629 static int
 630 vq_is_ready(struct vhost_virtqueue *vq)
 631 {
 632         return vq && vq->desc   &&
 633                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 634                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 635 }
 636
 637 static int
 638 virtio_is_ready(struct virtio_net *dev)
 639 {
 640         struct vhost_virtqueue *vq;
 641         uint32_t i;
 642
 643         if (dev->nr_vring == 0)
 644                 return 0;
 645
 646         for (i = 0; i < dev->nr_vring; i++) {
 647                 vq = dev->virtqueue[i];
 648
 649                 if (!vq_is_ready(vq))
 650                         return 0;
 651         }
 652
 653         RTE_LOG(INFO, VHOST_CONFIG,
 654                 "virtio is now ready for processing.\n");
 655         return 1;
 656 }
 657
 658 static void
 659 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 660 {
 661         struct vhost_vring_file file;
 662         struct vhost_virtqueue *vq;
 663
 664         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 665         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 666                 file.fd = VIRTIO_INVALID_EVENTFD;
 667         else
 668                 file.fd = pmsg->fds[0];
 669         RTE_LOG(INFO, VHOST_CONFIG,
 670                 "vring call idx:%d file:%d\n", file.index, file.fd);
 671
 672         vq = dev->virtqueue[file.index];
 673         if (vq->callfd >= 0)
 674                 close(vq->callfd);
 675
 676         vq->callfd = file.fd;
 677 }
 678
 679 static void
 680 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 681 {
 682         struct vhost_vring_file file;
 683         struct vhost_virtqueue *vq;
 684
 685         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 686         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 687                 file.fd = VIRTIO_INVALID_EVENTFD;
 688         else
 689                 file.fd = pmsg->fds[0];
 690         RTE_LOG(INFO, VHOST_CONFIG,
 691                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 692
 693         vq = dev->virtqueue[file.index];
 694         if (vq->kickfd >= 0)
 695                 close(vq->kickfd);
 696         vq->kickfd = file.fd;
 697 }
 698
 699 static void
 700 free_zmbufs(struct vhost_virtqueue *vq)
 701 {
 702         struct zcopy_mbuf *zmbuf, *next;
 703
 704         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 705              zmbuf != NULL; zmbuf = next) {
 706                 next = TAILQ_NEXT(zmbuf, next);
 707
 708                 rte_pktmbuf_free(zmbuf->mbuf);
 709                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 710         }
 711
 712         rte_free(vq->zmbufs);
 713 }
 714
 715 /*
 716  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 717  */
 718 static int
 719 vhost_user_get_vring_base(struct virtio_net *dev,
 720                           VhostUserMsg *msg)
 721 {
 722         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 723
 724         /* We have to stop the queue (virtio) if it is running. */
 725         if (dev->flags & VIRTIO_DEV_RUNNING) {
 726                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 727                 dev->notify_ops->destroy_device(dev->vid);
 728         }
 729
 730         dev->flags &= ~VIRTIO_DEV_READY;
 731
 732         /* Here we are safe to get the last used index */
 733         msg->payload.state.num = vq->last_used_idx;
 734
 735         RTE_LOG(INFO, VHOST_CONFIG,
 736                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 737                 msg->payload.state.num);
 738         /*
 739          * Based on current qemu vhost-user implementation, this message is
 740          * sent and only sent in vhost_vring_stop.
 741          * TODO: cleanup the vring, it isn't usable since here.
 742          */
 743         if (vq->kickfd >= 0)
 744                 close(vq->kickfd);
 745
 746         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 747
 748         if (dev->dequeue_zero_copy)
 749                 free_zmbufs(vq);
 750         rte_free(vq->shadow_used_ring);
 751         vq->shadow_used_ring = NULL;
 752
 753         rte_free(vq->batch_copy_elems);
 754         vq->batch_copy_elems = NULL;
 755
 756         return 0;
 757 }
 758
 759 /*
 760  * when virtio queues are ready to work, qemu will send us to
 761  * enable the virtio queue pair.
 762  */
 763 static int
 764 vhost_user_set_vring_enable(struct virtio_net *dev,
 765                             VhostUserMsg *msg)
 766 {
 767         int enable = (int)msg->payload.state.num;
 768
 769         RTE_LOG(INFO, VHOST_CONFIG,
 770                 "set queue enable: %d to qp idx: %d\n",
 771                 enable, msg->payload.state.index);
 772
 773         if (dev->notify_ops->vring_state_changed)
 774                 dev->notify_ops->vring_state_changed(dev->vid,
 775                                 msg->payload.state.index, enable);
 776
 777         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 778
 779         return 0;
 780 }
 781
 782 static void
 783 vhost_user_set_protocol_features(struct virtio_net *dev,
 784                                  uint64_t protocol_features)
 785 {
 786         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 787                 return;
 788
 789         dev->protocol_features = protocol_features;
 790 }
 791
 792 static int
 793 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 794 {
 795         int fd = msg->fds[0];
 796         uint64_t size, off;
 797         void *addr;
 798
 799         if (fd < 0) {
 800                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 801                 return -1;
 802         }
 803
 804         if (msg->size != sizeof(VhostUserLog)) {
 805                 RTE_LOG(ERR, VHOST_CONFIG,
 806                         "invalid log base msg size: %"PRId32" != %d\n",
 807                         msg->size, (int)sizeof(VhostUserLog));
 808                 return -1;
 809         }
 810
 811         size = msg->payload.log.mmap_size;
 812         off  = msg->payload.log.mmap_offset;
 813         RTE_LOG(INFO, VHOST_CONFIG,
 814                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 815                 size, off);
 816
 817         /*
 818          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 819          * fail when offset is not page size aligned.
 820          */
 821         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 822         close(fd);
 823         if (addr == MAP_FAILED) {
 824                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 825                 return -1;
 826         }
 827
 828         /*
 829          * Free previously mapped log memory on occasionally
 830          * multiple VHOST_USER_SET_LOG_BASE.
 831          */
 832         if (dev->log_addr) {
 833                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 834         }
 835         dev->log_addr = (uint64_t)(uintptr_t)addr;
 836         dev->log_base = dev->log_addr + off;
 837         dev->log_size = size;
 838
 839         return 0;
 840 }
 841
 842 /*
 843  * An rarp packet is constructed and broadcasted to notify switches about
 844  * the new location of the migrated VM, so that packets from outside will
 845  * not be lost after migration.
 846  *
 847  * However, we don't actually "send" a rarp packet here, instead, we set
 848  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 849  */
 850 static int
 851 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 852 {
 853         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 854
 855         RTE_LOG(DEBUG, VHOST_CONFIG,
 856                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 857                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 858         memcpy(dev->mac.addr_bytes, mac, 6);
 859
 860         /*
 861          * Set the flag to inject a RARP broadcast packet at
 862          * rte_vhost_dequeue_burst().
 863          *
 864          * rte_smp_wmb() is for making sure the mac is copied
 865          * before the flag is set.
 866          */
 867         rte_smp_wmb();
 868         rte_atomic16_set(&dev->broadcast_rarp, 1);
 869
 870         return 0;
 871 }
 872
 873 static int
 874 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 875 {
 876         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 877                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 878                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 879                                 msg->payload.u64);
 880
 881                 return -1;
 882         }
 883
 884         dev->mtu = msg->payload.u64;
 885
 886         return 0;
 887 }
 888
 889 /* return bytes# of read on success or negative val on failure. */
 890 static int
 891 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 892 {
 893         int ret;
 894
 895         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 896                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 897         if (ret <= 0)
 898                 return ret;
 899
 900         if (msg && msg->size) {
 901                 if (msg->size > sizeof(msg->payload)) {
 902                         RTE_LOG(ERR, VHOST_CONFIG,
 903                                 "invalid msg size: %d\n", msg->size);
 904                         return -1;
 905                 }
 906                 ret = read(sockfd, &msg->payload, msg->size);
 907                 if (ret <= 0)
 908                         return ret;
 909                 if (ret != (int)msg->size) {
 910                         RTE_LOG(ERR, VHOST_CONFIG,
 911                                 "read control message failed\n");
 912                         return -1;
 913                 }
 914         }
 915
 916         return ret;
 917 }
 918
 919 static int
 920 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 921 {
 922         if (!msg)
 923                 return 0;
 924
 925         return send_fd_message(sockfd, (char *)msg,
 926                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 927 }
 928
 929 static int
 930 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
 931 {
 932         if (!msg)
 933                 return 0;
 934
 935         msg->flags &= ~VHOST_USER_VERSION_MASK;
 936         msg->flags &= ~VHOST_USER_NEED_REPLY;
 937         msg->flags |= VHOST_USER_VERSION;
 938         msg->flags |= VHOST_USER_REPLY_MASK;
 939
 940         return send_vhost_message(sockfd, msg);
 941 }
 942
 943 /*
 944  * Allocate a queue pair if it hasn't been allocated yet
 945  */
 946 static int
 947 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
 948 {
 949         uint16_t vring_idx;
 950
 951         switch (msg->request) {
 952         case VHOST_USER_SET_VRING_KICK:
 953         case VHOST_USER_SET_VRING_CALL:
 954         case VHOST_USER_SET_VRING_ERR:
 955                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 956                 break;
 957         case VHOST_USER_SET_VRING_NUM:
 958         case VHOST_USER_SET_VRING_BASE:
 959         case VHOST_USER_SET_VRING_ENABLE:
 960                 vring_idx = msg->payload.state.index;
 961                 break;
 962         case VHOST_USER_SET_VRING_ADDR:
 963                 vring_idx = msg->payload.addr.index;
 964                 break;
 965         default:
 966                 return 0;
 967         }
 968
 969         if (vring_idx >= VHOST_MAX_VRING) {
 970                 RTE_LOG(ERR, VHOST_CONFIG,
 971                         "invalid vring index: %u\n", vring_idx);
 972                 return -1;
 973         }
 974
 975         if (dev->virtqueue[vring_idx])
 976                 return 0;
 977
 978         return alloc_vring_queue(dev, vring_idx);
 979 }
 980
 981 int
 982 vhost_user_msg_handler(int vid, int fd)
 983 {
 984         struct virtio_net *dev;
 985         struct VhostUserMsg msg;
 986         int ret;
 987
 988         dev = get_device(vid);
 989         if (dev == NULL)
 990                 return -1;
 991
 992         if (!dev->notify_ops) {
 993                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
 994                 if (!dev->notify_ops) {
 995                         RTE_LOG(ERR, VHOST_CONFIG,
 996                                 "failed to get callback ops for driver %s\n",
 997                                 dev->ifname);
 998                         return -1;
 999                 }
1000         }
1001
1002         ret = read_vhost_message(fd, &msg);
1003         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1004                 if (ret < 0)
1005                         RTE_LOG(ERR, VHOST_CONFIG,
1006                                 "vhost read message failed\n");
1007                 else if (ret == 0)
1008                         RTE_LOG(INFO, VHOST_CONFIG,
1009                                 "vhost peer closed\n");
1010                 else
1011                         RTE_LOG(ERR, VHOST_CONFIG,
1012                                 "vhost read incorrect message\n");
1013
1014                 return -1;
1015         }
1016
1017         ret = 0;
1018         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1019                 vhost_message_str[msg.request]);
1020
1021         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1022         if (ret < 0) {
1023                 RTE_LOG(ERR, VHOST_CONFIG,
1024                         "failed to alloc queue\n");
1025                 return -1;
1026         }
1027
1028         switch (msg.request) {
1029         case VHOST_USER_GET_FEATURES:
1030                 msg.payload.u64 = vhost_user_get_features(dev);
1031                 msg.size = sizeof(msg.payload.u64);
1032                 send_vhost_reply(fd, &msg);
1033                 break;
1034         case VHOST_USER_SET_FEATURES:
1035                 vhost_user_set_features(dev, msg.payload.u64);
1036                 break;
1037
1038         case VHOST_USER_GET_PROTOCOL_FEATURES:
1039                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1040                 msg.size = sizeof(msg.payload.u64);
1041                 send_vhost_reply(fd, &msg);
1042                 break;
1043         case VHOST_USER_SET_PROTOCOL_FEATURES:
1044                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1045                 break;
1046
1047         case VHOST_USER_SET_OWNER:
1048                 vhost_user_set_owner();
1049                 break;
1050         case VHOST_USER_RESET_OWNER:
1051                 vhost_user_reset_owner(dev);
1052                 break;
1053
1054         case VHOST_USER_SET_MEM_TABLE:
1055                 ret = vhost_user_set_mem_table(dev, &msg);
1056                 break;
1057
1058         case VHOST_USER_SET_LOG_BASE:
1059                 vhost_user_set_log_base(dev, &msg);
1060
1061                 /* it needs a reply */
1062                 msg.size = sizeof(msg.payload.u64);
1063                 send_vhost_reply(fd, &msg);
1064                 break;
1065         case VHOST_USER_SET_LOG_FD:
1066                 close(msg.fds[0]);
1067                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1068                 break;
1069
1070         case VHOST_USER_SET_VRING_NUM:
1071                 vhost_user_set_vring_num(dev, &msg);
1072                 break;
1073         case VHOST_USER_SET_VRING_ADDR:
1074                 vhost_user_set_vring_addr(dev, &msg);
1075                 break;
1076         case VHOST_USER_SET_VRING_BASE:
1077                 vhost_user_set_vring_base(dev, &msg);
1078                 break;
1079
1080         case VHOST_USER_GET_VRING_BASE:
1081                 vhost_user_get_vring_base(dev, &msg);
1082                 msg.size = sizeof(msg.payload.state);
1083                 send_vhost_reply(fd, &msg);
1084                 break;
1085
1086         case VHOST_USER_SET_VRING_KICK:
1087                 vhost_user_set_vring_kick(dev, &msg);
1088                 break;
1089         case VHOST_USER_SET_VRING_CALL:
1090                 vhost_user_set_vring_call(dev, &msg);
1091                 break;
1092
1093         case VHOST_USER_SET_VRING_ERR:
1094                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1095                         close(msg.fds[0]);
1096                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1097                 break;
1098
1099         case VHOST_USER_GET_QUEUE_NUM:
1100                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1101                 msg.size = sizeof(msg.payload.u64);
1102                 send_vhost_reply(fd, &msg);
1103                 break;
1104
1105         case VHOST_USER_SET_VRING_ENABLE:
1106                 vhost_user_set_vring_enable(dev, &msg);
1107                 break;
1108         case VHOST_USER_SEND_RARP:
1109                 vhost_user_send_rarp(dev, &msg);
1110                 break;
1111
1112         case VHOST_USER_NET_SET_MTU:
1113                 ret = vhost_user_net_set_mtu(dev, &msg);
1114                 break;
1115
1116         default:
1117                 ret = -1;
1118                 break;
1119
1120         }
1121
1122         if (msg.flags & VHOST_USER_NEED_REPLY) {
1123                 msg.payload.u64 = !!ret;
1124                 msg.size = sizeof(msg.payload.u64);
1125                 send_vhost_reply(fd, &msg);
1126         }
1127
1128         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1129                 dev->flags |= VIRTIO_DEV_READY;
1130
1131                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1132                         if (dev->dequeue_zero_copy) {
1133                                 RTE_LOG(INFO, VHOST_CONFIG,
1134                                                 "dequeue zero copy is enabled\n");
1135                         }
1136
1137                         if (dev->notify_ops->new_device(dev->vid) == 0)
1138                                 dev->flags |= VIRTIO_DEV_RUNNING;
1139                 }
1140         }
1141
1142         return 0;
1143 }