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         if (dev->log_addr) {
 118                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 119                 dev->log_addr = 0;
 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                 return -1;
 169
 170         dev->features = features;
 171         if (dev->features &
 172                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 173                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 174         } else {
 175                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 176         }
 177         LOG_DEBUG(VHOST_CONFIG,
 178                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 179                 dev->vid,
 180                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 181                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 182
 183         return 0;
 184 }
 185
 186 /*
 187  * The virtio device sends us the size of the descriptor ring.
 188  */
 189 static int
 190 vhost_user_set_vring_num(struct virtio_net *dev,
 191                          struct vhost_vring_state *state)
 192 {
 193         struct vhost_virtqueue *vq = dev->virtqueue[state->index];
 194
 195         vq->size = state->num;
 196
 197         if (dev->dequeue_zero_copy) {
 198                 vq->nr_zmbuf = 0;
 199                 vq->last_zmbuf_idx = 0;
 200                 vq->zmbuf_size = vq->size;
 201                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 202                                          sizeof(struct zcopy_mbuf), 0);
 203                 if (vq->zmbufs == NULL) {
 204                         RTE_LOG(WARNING, VHOST_CONFIG,
 205                                 "failed to allocate mem for zero copy; "
 206                                 "zero copy is force disabled\n");
 207                         dev->dequeue_zero_copy = 0;
 208                 }
 209         }
 210
 211         vq->shadow_used_ring = rte_malloc(NULL,
 212                                 vq->size * sizeof(struct vring_used_elem),
 213                                 RTE_CACHE_LINE_SIZE);
 214         if (!vq->shadow_used_ring) {
 215                 RTE_LOG(ERR, VHOST_CONFIG,
 216                         "failed to allocate memory for shadow used ring.\n");
 217                 return -1;
 218         }
 219
 220         return 0;
 221 }
 222
 223 /*
 224  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 225  * same numa node as the memory of vring descriptor.
 226  */
 227 #ifdef RTE_LIBRTE_VHOST_NUMA
 228 static struct virtio_net*
 229 numa_realloc(struct virtio_net *dev, int index)
 230 {
 231         int oldnode, newnode;
 232         struct virtio_net *old_dev;
 233         struct vhost_virtqueue *old_vq, *vq;
 234         int ret;
 235
 236         enum {VIRTIO_RXQ, VIRTIO_TXQ, VIRTIO_QNUM};
 237
 238         old_dev = dev;
 239         vq = old_vq = dev->virtqueue[index];
 240
 241         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 242                             MPOL_F_NODE | MPOL_F_ADDR);
 243
 244         /* check if we need to reallocate vq */
 245         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 246                              MPOL_F_NODE | MPOL_F_ADDR);
 247         if (ret) {
 248                 RTE_LOG(ERR, VHOST_CONFIG,
 249                         "Unable to get vq numa information.\n");
 250                 return dev;
 251         }
 252         if (oldnode != newnode) {
 253                 RTE_LOG(INFO, VHOST_CONFIG,
 254                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 255                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 256                 if (!vq)
 257                         return dev;
 258
 259                 memcpy(vq, old_vq, sizeof(*vq) * VIRTIO_QNUM);
 260                 rte_free(old_vq);
 261         }
 262
 263         /* check if we need to reallocate dev */
 264         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 265                             MPOL_F_NODE | MPOL_F_ADDR);
 266         if (ret) {
 267                 RTE_LOG(ERR, VHOST_CONFIG,
 268                         "Unable to get dev numa information.\n");
 269                 goto out;
 270         }
 271         if (oldnode != newnode) {
 272                 RTE_LOG(INFO, VHOST_CONFIG,
 273                         "reallocate dev from %d to %d node\n",
 274                         oldnode, newnode);
 275                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 276                 if (!dev) {
 277                         dev = old_dev;
 278                         goto out;
 279                 }
 280
 281                 memcpy(dev, old_dev, sizeof(*dev));
 282                 rte_free(old_dev);
 283         }
 284
 285 out:
 286         dev->virtqueue[index] = vq;
 287         vhost_devices[dev->vid] = dev;
 288
 289         return dev;
 290 }
 291 #else
 292 static struct virtio_net*
 293 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 294 {
 295         return dev;
 296 }
 297 #endif
 298
 299 /*
 300  * Converts QEMU virtual address to Vhost virtual address. This function is
 301  * used to convert the ring addresses to our address space.
 302  */
 303 static uint64_t
 304 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 305 {
 306         struct rte_vhost_mem_region *reg;
 307         uint32_t i;
 308
 309         /* Find the region where the address lives. */
 310         for (i = 0; i < dev->mem->nregions; i++) {
 311                 reg = &dev->mem->regions[i];
 312
 313                 if (qva >= reg->guest_user_addr &&
 314                     qva <  reg->guest_user_addr + reg->size) {
 315                         return qva - reg->guest_user_addr +
 316                                reg->host_user_addr;
 317                 }
 318         }
 319
 320         return 0;
 321 }
 322
 323 /*
 324  * The virtio device sends us the desc, used and avail ring addresses.
 325  * This function then converts these to our address space.
 326  */
 327 static int
 328 vhost_user_set_vring_addr(struct virtio_net *dev, struct vhost_vring_addr *addr)
 329 {
 330         struct vhost_virtqueue *vq;
 331
 332         if (dev->mem == NULL)
 333                 return -1;
 334
 335         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 336         vq = dev->virtqueue[addr->index];
 337
 338         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 339         vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
 340                         addr->desc_user_addr);
 341         if (vq->desc == 0) {
 342                 RTE_LOG(ERR, VHOST_CONFIG,
 343                         "(%d) failed to find desc ring address.\n",
 344                         dev->vid);
 345                 return -1;
 346         }
 347
 348         dev = numa_realloc(dev, addr->index);
 349         vq = dev->virtqueue[addr->index];
 350
 351         vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
 352                         addr->avail_user_addr);
 353         if (vq->avail == 0) {
 354                 RTE_LOG(ERR, VHOST_CONFIG,
 355                         "(%d) failed to find avail ring address.\n",
 356                         dev->vid);
 357                 return -1;
 358         }
 359
 360         vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
 361                         addr->used_user_addr);
 362         if (vq->used == 0) {
 363                 RTE_LOG(ERR, VHOST_CONFIG,
 364                         "(%d) failed to find used ring address.\n",
 365                         dev->vid);
 366                 return -1;
 367         }
 368
 369         if (vq->last_used_idx != vq->used->idx) {
 370                 RTE_LOG(WARNING, VHOST_CONFIG,
 371                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 372                         "some packets maybe resent for Tx and dropped for Rx\n",
 373                         vq->last_used_idx, vq->used->idx);
 374                 vq->last_used_idx  = vq->used->idx;
 375                 vq->last_avail_idx = vq->used->idx;
 376         }
 377
 378         vq->log_guest_addr = addr->log_guest_addr;
 379
 380         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 381                         dev->vid, vq->desc);
 382         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 383                         dev->vid, vq->avail);
 384         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 385                         dev->vid, vq->used);
 386         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 387                         dev->vid, vq->log_guest_addr);
 388
 389         return 0;
 390 }
 391
 392 /*
 393  * The virtio device sends us the available ring last used index.
 394  */
 395 static int
 396 vhost_user_set_vring_base(struct virtio_net *dev,
 397                           struct vhost_vring_state *state)
 398 {
 399         dev->virtqueue[state->index]->last_used_idx  = state->num;
 400         dev->virtqueue[state->index]->last_avail_idx = state->num;
 401
 402         return 0;
 403 }
 404
 405 static void
 406 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 407                    uint64_t host_phys_addr, uint64_t size)
 408 {
 409         struct guest_page *page, *last_page;
 410
 411         if (dev->nr_guest_pages == dev->max_guest_pages) {
 412                 dev->max_guest_pages *= 2;
 413                 dev->guest_pages = realloc(dev->guest_pages,
 414                                         dev->max_guest_pages * sizeof(*page));
 415         }
 416
 417         if (dev->nr_guest_pages > 0) {
 418                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 419                 /* merge if the two pages are continuous */
 420                 if (host_phys_addr == last_page->host_phys_addr +
 421                                       last_page->size) {
 422                         last_page->size += size;
 423                         return;
 424                 }
 425         }
 426
 427         page = &dev->guest_pages[dev->nr_guest_pages++];
 428         page->guest_phys_addr = guest_phys_addr;
 429         page->host_phys_addr  = host_phys_addr;
 430         page->size = size;
 431 }
 432
 433 static void
 434 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 435                 uint64_t page_size)
 436 {
 437         uint64_t reg_size = reg->size;
 438         uint64_t host_user_addr  = reg->host_user_addr;
 439         uint64_t guest_phys_addr = reg->guest_phys_addr;
 440         uint64_t host_phys_addr;
 441         uint64_t size;
 442
 443         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 444         size = page_size - (guest_phys_addr & (page_size - 1));
 445         size = RTE_MIN(size, reg_size);
 446
 447         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 448         host_user_addr  += size;
 449         guest_phys_addr += size;
 450         reg_size -= size;
 451
 452         while (reg_size > 0) {
 453                 size = RTE_MIN(reg_size, page_size);
 454                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 455                                                   host_user_addr);
 456                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 457
 458                 host_user_addr  += size;
 459                 guest_phys_addr += size;
 460                 reg_size -= size;
 461         }
 462 }
 463
 464 #ifdef RTE_LIBRTE_VHOST_DEBUG
 465 /* TODO: enable it only in debug mode? */
 466 static void
 467 dump_guest_pages(struct virtio_net *dev)
 468 {
 469         uint32_t i;
 470         struct guest_page *page;
 471
 472         for (i = 0; i < dev->nr_guest_pages; i++) {
 473                 page = &dev->guest_pages[i];
 474
 475                 RTE_LOG(INFO, VHOST_CONFIG,
 476                         "guest physical page region %u\n"
 477                         "\t guest_phys_addr: %" PRIx64 "\n"
 478                         "\t host_phys_addr : %" PRIx64 "\n"
 479                         "\t size           : %" PRIx64 "\n",
 480                         i,
 481                         page->guest_phys_addr,
 482                         page->host_phys_addr,
 483                         page->size);
 484         }
 485 }
 486 #else
 487 #define dump_guest_pages(dev)
 488 #endif
 489
 490 static int
 491 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 492 {
 493         struct VhostUserMemory memory = pmsg->payload.memory;
 494         struct rte_vhost_mem_region *reg;
 495         void *mmap_addr;
 496         uint64_t mmap_size;
 497         uint64_t mmap_offset;
 498         uint64_t alignment;
 499         uint32_t i;
 500         int fd;
 501
 502         /* Remove from the data plane. */
 503         if (dev->flags & VIRTIO_DEV_RUNNING) {
 504                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 505                 dev->notify_ops->destroy_device(dev->vid);
 506         }
 507
 508         if (dev->mem) {
 509                 free_mem_region(dev);
 510                 rte_free(dev->mem);
 511                 dev->mem = NULL;
 512         }
 513
 514         dev->nr_guest_pages = 0;
 515         if (!dev->guest_pages) {
 516                 dev->max_guest_pages = 8;
 517                 dev->guest_pages = malloc(dev->max_guest_pages *
 518                                                 sizeof(struct guest_page));
 519         }
 520
 521         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 522                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 523         if (dev->mem == NULL) {
 524                 RTE_LOG(ERR, VHOST_CONFIG,
 525                         "(%d) failed to allocate memory for dev->mem\n",
 526                         dev->vid);
 527                 return -1;
 528         }
 529         dev->mem->nregions = memory.nregions;
 530
 531         for (i = 0; i < memory.nregions; i++) {
 532                 fd  = pmsg->fds[i];
 533                 reg = &dev->mem->regions[i];
 534
 535                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 536                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 537                 reg->size            = memory.regions[i].memory_size;
 538                 reg->fd              = fd;
 539
 540                 mmap_offset = memory.regions[i].mmap_offset;
 541                 mmap_size   = reg->size + mmap_offset;
 542
 543                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 544                  * with length argument aligned with hugepagesz at older
 545                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 546                  * mmap() will fail with EINVAL.
 547                  *
 548                  * to avoid failure, make sure in caller to keep length
 549                  * aligned.
 550                  */
 551                 alignment = get_blk_size(fd);
 552                 if (alignment == (uint64_t)-1) {
 553                         RTE_LOG(ERR, VHOST_CONFIG,
 554                                 "couldn't get hugepage size through fstat\n");
 555                         goto err_mmap;
 556                 }
 557                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 558
 559                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 560                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 561
 562                 if (mmap_addr == MAP_FAILED) {
 563                         RTE_LOG(ERR, VHOST_CONFIG,
 564                                 "mmap region %u failed.\n", i);
 565                         goto err_mmap;
 566                 }
 567
 568                 reg->mmap_addr = mmap_addr;
 569                 reg->mmap_size = mmap_size;
 570                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 571                                       mmap_offset;
 572
 573                 if (dev->dequeue_zero_copy)
 574                         add_guest_pages(dev, reg, alignment);
 575
 576                 RTE_LOG(INFO, VHOST_CONFIG,
 577                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 578                         "\t guest physical addr: 0x%" PRIx64 "\n"
 579                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 580                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 581                         "\t mmap addr : 0x%" PRIx64 "\n"
 582                         "\t mmap size : 0x%" PRIx64 "\n"
 583                         "\t mmap align: 0x%" PRIx64 "\n"
 584                         "\t mmap off  : 0x%" PRIx64 "\n",
 585                         i, reg->size,
 586                         reg->guest_phys_addr,
 587                         reg->guest_user_addr,
 588                         reg->host_user_addr,
 589                         (uint64_t)(uintptr_t)mmap_addr,
 590                         mmap_size,
 591                         alignment,
 592                         mmap_offset);
 593         }
 594
 595         dump_guest_pages(dev);
 596
 597         return 0;
 598
 599 err_mmap:
 600         free_mem_region(dev);
 601         rte_free(dev->mem);
 602         dev->mem = NULL;
 603         return -1;
 604 }
 605
 606 static int
 607 vq_is_ready(struct vhost_virtqueue *vq)
 608 {
 609         return vq && vq->desc   &&
 610                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 611                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 612 }
 613
 614 static int
 615 virtio_is_ready(struct virtio_net *dev)
 616 {
 617         struct vhost_virtqueue *vq;
 618         uint32_t i;
 619
 620         if (dev->nr_vring == 0)
 621                 return 0;
 622
 623         for (i = 0; i < dev->nr_vring; i++) {
 624                 vq = dev->virtqueue[i];
 625
 626                 if (!vq_is_ready(vq))
 627                         return 0;
 628         }
 629
 630         RTE_LOG(INFO, VHOST_CONFIG,
 631                 "virtio is now ready for processing.\n");
 632         return 1;
 633 }
 634
 635 static void
 636 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 637 {
 638         struct vhost_vring_file file;
 639         struct vhost_virtqueue *vq;
 640
 641         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 642         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 643                 file.fd = VIRTIO_INVALID_EVENTFD;
 644         else
 645                 file.fd = pmsg->fds[0];
 646         RTE_LOG(INFO, VHOST_CONFIG,
 647                 "vring call idx:%d file:%d\n", file.index, file.fd);
 648
 649         vq = dev->virtqueue[file.index];
 650         if (vq->callfd >= 0)
 651                 close(vq->callfd);
 652
 653         vq->callfd = file.fd;
 654 }
 655
 656 static void
 657 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 658 {
 659         struct vhost_vring_file file;
 660         struct vhost_virtqueue *vq;
 661
 662         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 663         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 664                 file.fd = VIRTIO_INVALID_EVENTFD;
 665         else
 666                 file.fd = pmsg->fds[0];
 667         RTE_LOG(INFO, VHOST_CONFIG,
 668                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 669
 670         vq = dev->virtqueue[file.index];
 671         if (vq->kickfd >= 0)
 672                 close(vq->kickfd);
 673         vq->kickfd = file.fd;
 674 }
 675
 676 static void
 677 free_zmbufs(struct vhost_virtqueue *vq)
 678 {
 679         struct zcopy_mbuf *zmbuf, *next;
 680
 681         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 682              zmbuf != NULL; zmbuf = next) {
 683                 next = TAILQ_NEXT(zmbuf, next);
 684
 685                 rte_pktmbuf_free(zmbuf->mbuf);
 686                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 687         }
 688
 689         rte_free(vq->zmbufs);
 690 }
 691
 692 /*
 693  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 694  */
 695 static int
 696 vhost_user_get_vring_base(struct virtio_net *dev,
 697                           struct vhost_vring_state *state)
 698 {
 699         struct vhost_virtqueue *vq = dev->virtqueue[state->index];
 700
 701         /* We have to stop the queue (virtio) if it is running. */
 702         if (dev->flags & VIRTIO_DEV_RUNNING) {
 703                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 704                 dev->notify_ops->destroy_device(dev->vid);
 705         }
 706
 707         dev->flags &= ~VIRTIO_DEV_READY;
 708
 709         /* Here we are safe to get the last used index */
 710         state->num = vq->last_used_idx;
 711
 712         RTE_LOG(INFO, VHOST_CONFIG,
 713                 "vring base idx:%d file:%d\n", state->index, state->num);
 714         /*
 715          * Based on current qemu vhost-user implementation, this message is
 716          * sent and only sent in vhost_vring_stop.
 717          * TODO: cleanup the vring, it isn't usable since here.
 718          */
 719         if (vq->kickfd >= 0)
 720                 close(vq->kickfd);
 721
 722         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 723
 724         if (dev->dequeue_zero_copy)
 725                 free_zmbufs(vq);
 726         rte_free(vq->shadow_used_ring);
 727         vq->shadow_used_ring = NULL;
 728
 729         return 0;
 730 }
 731
 732 /*
 733  * when virtio queues are ready to work, qemu will send us to
 734  * enable the virtio queue pair.
 735  */
 736 static int
 737 vhost_user_set_vring_enable(struct virtio_net *dev,
 738                             struct vhost_vring_state *state)
 739 {
 740         int enable = (int)state->num;
 741
 742         RTE_LOG(INFO, VHOST_CONFIG,
 743                 "set queue enable: %d to qp idx: %d\n",
 744                 enable, state->index);
 745
 746         if (dev->notify_ops->vring_state_changed)
 747                 dev->notify_ops->vring_state_changed(dev->vid, state->index, enable);
 748
 749         dev->virtqueue[state->index]->enabled = enable;
 750
 751         return 0;
 752 }
 753
 754 static void
 755 vhost_user_set_protocol_features(struct virtio_net *dev,
 756                                  uint64_t protocol_features)
 757 {
 758         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 759                 return;
 760
 761         dev->protocol_features = protocol_features;
 762 }
 763
 764 static int
 765 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 766 {
 767         int fd = msg->fds[0];
 768         uint64_t size, off;
 769         void *addr;
 770
 771         if (fd < 0) {
 772                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 773                 return -1;
 774         }
 775
 776         if (msg->size != sizeof(VhostUserLog)) {
 777                 RTE_LOG(ERR, VHOST_CONFIG,
 778                         "invalid log base msg size: %"PRId32" != %d\n",
 779                         msg->size, (int)sizeof(VhostUserLog));
 780                 return -1;
 781         }
 782
 783         size = msg->payload.log.mmap_size;
 784         off  = msg->payload.log.mmap_offset;
 785         RTE_LOG(INFO, VHOST_CONFIG,
 786                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 787                 size, off);
 788
 789         /*
 790          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 791          * fail when offset is not page size aligned.
 792          */
 793         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 794         close(fd);
 795         if (addr == MAP_FAILED) {
 796                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 797                 return -1;
 798         }
 799
 800         /*
 801          * Free previously mapped log memory on occasionally
 802          * multiple VHOST_USER_SET_LOG_BASE.
 803          */
 804         if (dev->log_addr) {
 805                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 806         }
 807         dev->log_addr = (uint64_t)(uintptr_t)addr;
 808         dev->log_base = dev->log_addr + off;
 809         dev->log_size = size;
 810
 811         return 0;
 812 }
 813
 814 /*
 815  * An rarp packet is constructed and broadcasted to notify switches about
 816  * the new location of the migrated VM, so that packets from outside will
 817  * not be lost after migration.
 818  *
 819  * However, we don't actually "send" a rarp packet here, instead, we set
 820  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 821  */
 822 static int
 823 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 824 {
 825         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 826
 827         RTE_LOG(DEBUG, VHOST_CONFIG,
 828                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 829                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 830         memcpy(dev->mac.addr_bytes, mac, 6);
 831
 832         /*
 833          * Set the flag to inject a RARP broadcast packet at
 834          * rte_vhost_dequeue_burst().
 835          *
 836          * rte_smp_wmb() is for making sure the mac is copied
 837          * before the flag is set.
 838          */
 839         rte_smp_wmb();
 840         rte_atomic16_set(&dev->broadcast_rarp, 1);
 841
 842         return 0;
 843 }
 844
 845 static int
 846 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 847 {
 848         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 849                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 850                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 851                                 msg->payload.u64);
 852
 853                 return -1;
 854         }
 855
 856         dev->mtu = msg->payload.u64;
 857
 858         return 0;
 859 }
 860
 861 /* return bytes# of read on success or negative val on failure. */
 862 static int
 863 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 864 {
 865         int ret;
 866
 867         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 868                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 869         if (ret <= 0)
 870                 return ret;
 871
 872         if (msg && msg->size) {
 873                 if (msg->size > sizeof(msg->payload)) {
 874                         RTE_LOG(ERR, VHOST_CONFIG,
 875                                 "invalid msg size: %d\n", msg->size);
 876                         return -1;
 877                 }
 878                 ret = read(sockfd, &msg->payload, msg->size);
 879                 if (ret <= 0)
 880                         return ret;
 881                 if (ret != (int)msg->size) {
 882                         RTE_LOG(ERR, VHOST_CONFIG,
 883                                 "read control message failed\n");
 884                         return -1;
 885                 }
 886         }
 887
 888         return ret;
 889 }
 890
 891 static int
 892 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 893 {
 894         int ret;
 895
 896         if (!msg)
 897                 return 0;
 898
 899         msg->flags &= ~VHOST_USER_VERSION_MASK;
 900         msg->flags &= ~VHOST_USER_NEED_REPLY;
 901         msg->flags |= VHOST_USER_VERSION;
 902         msg->flags |= VHOST_USER_REPLY_MASK;
 903
 904         ret = send_fd_message(sockfd, (char *)msg,
 905                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 906
 907         return ret;
 908 }
 909
 910 /*
 911  * Allocate a queue pair if it hasn't been allocated yet
 912  */
 913 static int
 914 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
 915 {
 916         uint16_t vring_idx;
 917
 918         switch (msg->request) {
 919         case VHOST_USER_SET_VRING_KICK:
 920         case VHOST_USER_SET_VRING_CALL:
 921         case VHOST_USER_SET_VRING_ERR:
 922                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 923                 break;
 924         case VHOST_USER_SET_VRING_NUM:
 925         case VHOST_USER_SET_VRING_BASE:
 926         case VHOST_USER_SET_VRING_ENABLE:
 927                 vring_idx = msg->payload.state.index;
 928                 break;
 929         case VHOST_USER_SET_VRING_ADDR:
 930                 vring_idx = msg->payload.addr.index;
 931                 break;
 932         default:
 933                 return 0;
 934         }
 935
 936         if (vring_idx >= VHOST_MAX_VRING) {
 937                 RTE_LOG(ERR, VHOST_CONFIG,
 938                         "invalid vring index: %u\n", vring_idx);
 939                 return -1;
 940         }
 941
 942         if (dev->virtqueue[vring_idx])
 943                 return 0;
 944
 945         return alloc_vring_queue(dev, vring_idx);
 946 }
 947
 948 int
 949 vhost_user_msg_handler(int vid, int fd)
 950 {
 951         struct virtio_net *dev;
 952         struct VhostUserMsg msg;
 953         int ret;
 954
 955         dev = get_device(vid);
 956         if (dev == NULL)
 957                 return -1;
 958
 959         if (!dev->notify_ops) {
 960                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
 961                 if (!dev->notify_ops) {
 962                         RTE_LOG(ERR, VHOST_CONFIG,
 963                                 "failed to get callback ops for driver %s\n",
 964                                 dev->ifname);
 965                         return -1;
 966                 }
 967         }
 968
 969         ret = read_vhost_message(fd, &msg);
 970         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
 971                 if (ret < 0)
 972                         RTE_LOG(ERR, VHOST_CONFIG,
 973                                 "vhost read message failed\n");
 974                 else if (ret == 0)
 975                         RTE_LOG(INFO, VHOST_CONFIG,
 976                                 "vhost peer closed\n");
 977                 else
 978                         RTE_LOG(ERR, VHOST_CONFIG,
 979                                 "vhost read incorrect message\n");
 980
 981                 return -1;
 982         }
 983
 984         ret = 0;
 985         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
 986                 vhost_message_str[msg.request]);
 987
 988         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
 989         if (ret < 0) {
 990                 RTE_LOG(ERR, VHOST_CONFIG,
 991                         "failed to alloc queue\n");
 992                 return -1;
 993         }
 994
 995         switch (msg.request) {
 996         case VHOST_USER_GET_FEATURES:
 997                 msg.payload.u64 = vhost_user_get_features(dev);
 998                 msg.size = sizeof(msg.payload.u64);
 999                 send_vhost_message(fd, &msg);
1000                 break;
1001         case VHOST_USER_SET_FEATURES:
1002                 vhost_user_set_features(dev, msg.payload.u64);
1003                 break;
1004
1005         case VHOST_USER_GET_PROTOCOL_FEATURES:
1006                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1007                 msg.size = sizeof(msg.payload.u64);
1008                 send_vhost_message(fd, &msg);
1009                 break;
1010         case VHOST_USER_SET_PROTOCOL_FEATURES:
1011                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1012                 break;
1013
1014         case VHOST_USER_SET_OWNER:
1015                 vhost_user_set_owner();
1016                 break;
1017         case VHOST_USER_RESET_OWNER:
1018                 vhost_user_reset_owner(dev);
1019                 break;
1020
1021         case VHOST_USER_SET_MEM_TABLE:
1022                 ret = vhost_user_set_mem_table(dev, &msg);
1023                 break;
1024
1025         case VHOST_USER_SET_LOG_BASE:
1026                 vhost_user_set_log_base(dev, &msg);
1027
1028                 /* it needs a reply */
1029                 msg.size = sizeof(msg.payload.u64);
1030                 send_vhost_message(fd, &msg);
1031                 break;
1032         case VHOST_USER_SET_LOG_FD:
1033                 close(msg.fds[0]);
1034                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1035                 break;
1036
1037         case VHOST_USER_SET_VRING_NUM:
1038                 vhost_user_set_vring_num(dev, &msg.payload.state);
1039                 break;
1040         case VHOST_USER_SET_VRING_ADDR:
1041                 vhost_user_set_vring_addr(dev, &msg.payload.addr);
1042                 break;
1043         case VHOST_USER_SET_VRING_BASE:
1044                 vhost_user_set_vring_base(dev, &msg.payload.state);
1045                 break;
1046
1047         case VHOST_USER_GET_VRING_BASE:
1048                 vhost_user_get_vring_base(dev, &msg.payload.state);
1049                 msg.size = sizeof(msg.payload.state);
1050                 send_vhost_message(fd, &msg);
1051                 break;
1052
1053         case VHOST_USER_SET_VRING_KICK:
1054                 vhost_user_set_vring_kick(dev, &msg);
1055                 break;
1056         case VHOST_USER_SET_VRING_CALL:
1057                 vhost_user_set_vring_call(dev, &msg);
1058                 break;
1059
1060         case VHOST_USER_SET_VRING_ERR:
1061                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1062                         close(msg.fds[0]);
1063                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1064                 break;
1065
1066         case VHOST_USER_GET_QUEUE_NUM:
1067                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1068                 msg.size = sizeof(msg.payload.u64);
1069                 send_vhost_message(fd, &msg);
1070                 break;
1071
1072         case VHOST_USER_SET_VRING_ENABLE:
1073                 vhost_user_set_vring_enable(dev, &msg.payload.state);
1074                 break;
1075         case VHOST_USER_SEND_RARP:
1076                 vhost_user_send_rarp(dev, &msg);
1077                 break;
1078
1079         case VHOST_USER_NET_SET_MTU:
1080                 ret = vhost_user_net_set_mtu(dev, &msg);
1081                 break;
1082
1083         default:
1084                 ret = -1;
1085                 break;
1086
1087         }
1088
1089         if (msg.flags & VHOST_USER_NEED_REPLY) {
1090                 msg.payload.u64 = !!ret;
1091                 msg.size = sizeof(msg.payload.u64);
1092                 send_vhost_message(fd, &msg);
1093         }
1094
1095         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1096                 dev->flags |= VIRTIO_DEV_READY;
1097
1098                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1099                         if (dev->dequeue_zero_copy) {
1100                                 RTE_LOG(INFO, VHOST_CONFIG,
1101                                                 "dequeue zero copy is enabled\n");
1102                         }
1103
1104                         if (dev->notify_ops->new_device(dev->vid) == 0)
1105                                 dev->flags |= VIRTIO_DEV_RUNNING;
1106                 }
1107         }
1108
1109         return 0;
1110 }