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