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