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         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  80 };
  81
  82 static uint64_t
  83 get_blk_size(int fd)
  84 {
  85         struct stat stat;
  86         int ret;
  87
  88         ret = fstat(fd, &stat);
  89         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  90 }
  91
  92 static void
  93 free_mem_region(struct virtio_net *dev)
  94 {
  95         uint32_t i;
  96         struct rte_vhost_mem_region *reg;
  97
  98         if (!dev || !dev->mem)
  99                 return;
 100
 101         for (i = 0; i < dev->mem->nregions; i++) {
 102                 reg = &dev->mem->regions[i];
 103                 if (reg->host_user_addr) {
 104                         munmap(reg->mmap_addr, reg->mmap_size);
 105                         close(reg->fd);
 106                 }
 107         }
 108 }
 109
 110 void
 111 vhost_backend_cleanup(struct virtio_net *dev)
 112 {
 113         if (dev->mem) {
 114                 free_mem_region(dev);
 115                 rte_free(dev->mem);
 116                 dev->mem = NULL;
 117         }
 118
 119         free(dev->guest_pages);
 120         dev->guest_pages = NULL;
 121
 122         if (dev->log_addr) {
 123                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 124                 dev->log_addr = 0;
 125         }
 126
 127         if (dev->slave_req_fd >= 0) {
 128                 close(dev->slave_req_fd);
 129                 dev->slave_req_fd = -1;
 130         }
 131 }
 132
 133 /*
 134  * This function just returns success at the moment unless
 135  * the device hasn't been initialised.
 136  */
 137 static int
 138 vhost_user_set_owner(void)
 139 {
 140         return 0;
 141 }
 142
 143 static int
 144 vhost_user_reset_owner(struct virtio_net *dev)
 145 {
 146         if (dev->flags & VIRTIO_DEV_RUNNING) {
 147                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 148                 dev->notify_ops->destroy_device(dev->vid);
 149         }
 150
 151         cleanup_device(dev, 0);
 152         reset_device(dev);
 153         return 0;
 154 }
 155
 156 /*
 157  * The features that we support are requested.
 158  */
 159 static uint64_t
 160 vhost_user_get_features(struct virtio_net *dev)
 161 {
 162         uint64_t features = 0;
 163
 164         rte_vhost_driver_get_features(dev->ifname, &features);
 165         return features;
 166 }
 167
 168 /*
 169  * We receive the negotiated features supported by us and the virtio device.
 170  */
 171 static int
 172 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 173 {
 174         uint64_t vhost_features = 0;
 175
 176         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 177         if (features & ~vhost_features) {
 178                 RTE_LOG(ERR, VHOST_CONFIG,
 179                         "(%d) received invalid negotiated features.\n",
 180                         dev->vid);
 181                 return -1;
 182         }
 183
 184         if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
 185                 if (dev->notify_ops->features_changed)
 186                         dev->notify_ops->features_changed(dev->vid, features);
 187         }
 188
 189         dev->features = features;
 190         if (dev->features &
 191                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 192                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 193         } else {
 194                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 195         }
 196         LOG_DEBUG(VHOST_CONFIG,
 197                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 198                 dev->vid,
 199                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 200                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 201
 202         return 0;
 203 }
 204
 205 /*
 206  * The virtio device sends us the size of the descriptor ring.
 207  */
 208 static int
 209 vhost_user_set_vring_num(struct virtio_net *dev,
 210                          VhostUserMsg *msg)
 211 {
 212         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 213
 214         vq->size = msg->payload.state.num;
 215
 216         if (dev->dequeue_zero_copy) {
 217                 vq->nr_zmbuf = 0;
 218                 vq->last_zmbuf_idx = 0;
 219                 vq->zmbuf_size = vq->size;
 220                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 221                                          sizeof(struct zcopy_mbuf), 0);
 222                 if (vq->zmbufs == NULL) {
 223                         RTE_LOG(WARNING, VHOST_CONFIG,
 224                                 "failed to allocate mem for zero copy; "
 225                                 "zero copy is force disabled\n");
 226                         dev->dequeue_zero_copy = 0;
 227                 }
 228         }
 229
 230         vq->shadow_used_ring = rte_malloc(NULL,
 231                                 vq->size * sizeof(struct vring_used_elem),
 232                                 RTE_CACHE_LINE_SIZE);
 233         if (!vq->shadow_used_ring) {
 234                 RTE_LOG(ERR, VHOST_CONFIG,
 235                         "failed to allocate memory for shadow used ring.\n");
 236                 return -1;
 237         }
 238
 239         vq->batch_copy_elems = rte_malloc(NULL,
 240                                 vq->size * sizeof(struct batch_copy_elem),
 241                                 RTE_CACHE_LINE_SIZE);
 242         if (!vq->batch_copy_elems) {
 243                 RTE_LOG(ERR, VHOST_CONFIG,
 244                         "failed to allocate memory for batching copy.\n");
 245                 return -1;
 246         }
 247
 248         return 0;
 249 }
 250
 251 /*
 252  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 253  * same numa node as the memory of vring descriptor.
 254  */
 255 #ifdef RTE_LIBRTE_VHOST_NUMA
 256 static struct virtio_net*
 257 numa_realloc(struct virtio_net *dev, int index)
 258 {
 259         int oldnode, newnode;
 260         struct virtio_net *old_dev;
 261         struct vhost_virtqueue *old_vq, *vq;
 262         int ret;
 263
 264         old_dev = dev;
 265         vq = old_vq = dev->virtqueue[index];
 266
 267         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 268                             MPOL_F_NODE | MPOL_F_ADDR);
 269
 270         /* check if we need to reallocate vq */
 271         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 272                              MPOL_F_NODE | MPOL_F_ADDR);
 273         if (ret) {
 274                 RTE_LOG(ERR, VHOST_CONFIG,
 275                         "Unable to get vq numa information.\n");
 276                 return dev;
 277         }
 278         if (oldnode != newnode) {
 279                 RTE_LOG(INFO, VHOST_CONFIG,
 280                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 281                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 282                 if (!vq)
 283                         return dev;
 284
 285                 memcpy(vq, old_vq, sizeof(*vq));
 286                 rte_free(old_vq);
 287         }
 288
 289         /* check if we need to reallocate dev */
 290         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 291                             MPOL_F_NODE | MPOL_F_ADDR);
 292         if (ret) {
 293                 RTE_LOG(ERR, VHOST_CONFIG,
 294                         "Unable to get dev numa information.\n");
 295                 goto out;
 296         }
 297         if (oldnode != newnode) {
 298                 RTE_LOG(INFO, VHOST_CONFIG,
 299                         "reallocate dev from %d to %d node\n",
 300                         oldnode, newnode);
 301                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 302                 if (!dev) {
 303                         dev = old_dev;
 304                         goto out;
 305                 }
 306
 307                 memcpy(dev, old_dev, sizeof(*dev));
 308                 rte_free(old_dev);
 309         }
 310
 311 out:
 312         dev->virtqueue[index] = vq;
 313         vhost_devices[dev->vid] = dev;
 314
 315         return dev;
 316 }
 317 #else
 318 static struct virtio_net*
 319 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 320 {
 321         return dev;
 322 }
 323 #endif
 324
 325 /*
 326  * Converts QEMU virtual address to Vhost virtual address. This function is
 327  * used to convert the ring addresses to our address space.
 328  */
 329 static uint64_t
 330 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 331 {
 332         struct rte_vhost_mem_region *reg;
 333         uint32_t i;
 334
 335         /* Find the region where the address lives. */
 336         for (i = 0; i < dev->mem->nregions; i++) {
 337                 reg = &dev->mem->regions[i];
 338
 339                 if (qva >= reg->guest_user_addr &&
 340                     qva <  reg->guest_user_addr + reg->size) {
 341                         return qva - reg->guest_user_addr +
 342                                reg->host_user_addr;
 343                 }
 344         }
 345
 346         return 0;
 347 }
 348
 349 /*
 350  * The virtio device sends us the desc, used and avail ring addresses.
 351  * This function then converts these to our address space.
 352  */
 353 static int
 354 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
 355 {
 356         struct vhost_virtqueue *vq;
 357
 358         if (dev->mem == NULL)
 359                 return -1;
 360
 361         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 362         vq = dev->virtqueue[msg->payload.addr.index];
 363
 364         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 365         vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
 366                         msg->payload.addr.desc_user_addr);
 367         if (vq->desc == 0) {
 368                 RTE_LOG(ERR, VHOST_CONFIG,
 369                         "(%d) failed to find desc ring address.\n",
 370                         dev->vid);
 371                 return -1;
 372         }
 373
 374         dev = numa_realloc(dev, msg->payload.addr.index);
 375         vq = dev->virtqueue[msg->payload.addr.index];
 376
 377         vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
 378                         msg->payload.addr.avail_user_addr);
 379         if (vq->avail == 0) {
 380                 RTE_LOG(ERR, VHOST_CONFIG,
 381                         "(%d) failed to find avail ring address.\n",
 382                         dev->vid);
 383                 return -1;
 384         }
 385
 386         vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
 387                         msg->payload.addr.used_user_addr);
 388         if (vq->used == 0) {
 389                 RTE_LOG(ERR, VHOST_CONFIG,
 390                         "(%d) failed to find used ring address.\n",
 391                         dev->vid);
 392                 return -1;
 393         }
 394
 395         if (vq->last_used_idx != vq->used->idx) {
 396                 RTE_LOG(WARNING, VHOST_CONFIG,
 397                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 398                         "some packets maybe resent for Tx and dropped for Rx\n",
 399                         vq->last_used_idx, vq->used->idx);
 400                 vq->last_used_idx  = vq->used->idx;
 401                 vq->last_avail_idx = vq->used->idx;
 402         }
 403
 404         vq->log_guest_addr = msg->payload.addr.log_guest_addr;
 405
 406         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 407                         dev->vid, vq->desc);
 408         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 409                         dev->vid, vq->avail);
 410         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 411                         dev->vid, vq->used);
 412         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 413                         dev->vid, vq->log_guest_addr);
 414
 415         return 0;
 416 }
 417
 418 /*
 419  * The virtio device sends us the available ring last used index.
 420  */
 421 static int
 422 vhost_user_set_vring_base(struct virtio_net *dev,
 423                           VhostUserMsg *msg)
 424 {
 425         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 426                         msg->payload.state.num;
 427         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 428                         msg->payload.state.num;
 429
 430         return 0;
 431 }
 432
 433 static void
 434 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 435                    uint64_t host_phys_addr, uint64_t size)
 436 {
 437         struct guest_page *page, *last_page;
 438
 439         if (dev->nr_guest_pages == dev->max_guest_pages) {
 440                 dev->max_guest_pages *= 2;
 441                 dev->guest_pages = realloc(dev->guest_pages,
 442                                         dev->max_guest_pages * sizeof(*page));
 443         }
 444
 445         if (dev->nr_guest_pages > 0) {
 446                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 447                 /* merge if the two pages are continuous */
 448                 if (host_phys_addr == last_page->host_phys_addr +
 449                                       last_page->size) {
 450                         last_page->size += size;
 451                         return;
 452                 }
 453         }
 454
 455         page = &dev->guest_pages[dev->nr_guest_pages++];
 456         page->guest_phys_addr = guest_phys_addr;
 457         page->host_phys_addr  = host_phys_addr;
 458         page->size = size;
 459 }
 460
 461 static void
 462 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 463                 uint64_t page_size)
 464 {
 465         uint64_t reg_size = reg->size;
 466         uint64_t host_user_addr  = reg->host_user_addr;
 467         uint64_t guest_phys_addr = reg->guest_phys_addr;
 468         uint64_t host_phys_addr;
 469         uint64_t size;
 470
 471         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 472         size = page_size - (guest_phys_addr & (page_size - 1));
 473         size = RTE_MIN(size, reg_size);
 474
 475         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 476         host_user_addr  += size;
 477         guest_phys_addr += size;
 478         reg_size -= size;
 479
 480         while (reg_size > 0) {
 481                 size = RTE_MIN(reg_size, page_size);
 482                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 483                                                   host_user_addr);
 484                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 485
 486                 host_user_addr  += size;
 487                 guest_phys_addr += size;
 488                 reg_size -= size;
 489         }
 490 }
 491
 492 #ifdef RTE_LIBRTE_VHOST_DEBUG
 493 /* TODO: enable it only in debug mode? */
 494 static void
 495 dump_guest_pages(struct virtio_net *dev)
 496 {
 497         uint32_t i;
 498         struct guest_page *page;
 499
 500         for (i = 0; i < dev->nr_guest_pages; i++) {
 501                 page = &dev->guest_pages[i];
 502
 503                 RTE_LOG(INFO, VHOST_CONFIG,
 504                         "guest physical page region %u\n"
 505                         "\t guest_phys_addr: %" PRIx64 "\n"
 506                         "\t host_phys_addr : %" PRIx64 "\n"
 507                         "\t size           : %" PRIx64 "\n",
 508                         i,
 509                         page->guest_phys_addr,
 510                         page->host_phys_addr,
 511                         page->size);
 512         }
 513 }
 514 #else
 515 #define dump_guest_pages(dev)
 516 #endif
 517
 518 static int
 519 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 520 {
 521         struct VhostUserMemory memory = pmsg->payload.memory;
 522         struct rte_vhost_mem_region *reg;
 523         void *mmap_addr;
 524         uint64_t mmap_size;
 525         uint64_t mmap_offset;
 526         uint64_t alignment;
 527         uint32_t i;
 528         int fd;
 529
 530         if (dev->mem) {
 531                 free_mem_region(dev);
 532                 rte_free(dev->mem);
 533                 dev->mem = NULL;
 534         }
 535
 536         dev->nr_guest_pages = 0;
 537         if (!dev->guest_pages) {
 538                 dev->max_guest_pages = 8;
 539                 dev->guest_pages = malloc(dev->max_guest_pages *
 540                                                 sizeof(struct guest_page));
 541                 if (dev->guest_pages == NULL) {
 542                         RTE_LOG(ERR, VHOST_CONFIG,
 543                                 "(%d) failed to allocate memory "
 544                                 "for dev->guest_pages\n",
 545                                 dev->vid);
 546                         return -1;
 547                 }
 548         }
 549
 550         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 551                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 552         if (dev->mem == NULL) {
 553                 RTE_LOG(ERR, VHOST_CONFIG,
 554                         "(%d) failed to allocate memory for dev->mem\n",
 555                         dev->vid);
 556                 return -1;
 557         }
 558         dev->mem->nregions = memory.nregions;
 559
 560         for (i = 0; i < memory.nregions; i++) {
 561                 fd  = pmsg->fds[i];
 562                 reg = &dev->mem->regions[i];
 563
 564                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 565                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 566                 reg->size            = memory.regions[i].memory_size;
 567                 reg->fd              = fd;
 568
 569                 mmap_offset = memory.regions[i].mmap_offset;
 570                 mmap_size   = reg->size + mmap_offset;
 571
 572                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 573                  * with length argument aligned with hugepagesz at older
 574                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 575                  * mmap() will fail with EINVAL.
 576                  *
 577                  * to avoid failure, make sure in caller to keep length
 578                  * aligned.
 579                  */
 580                 alignment = get_blk_size(fd);
 581                 if (alignment == (uint64_t)-1) {
 582                         RTE_LOG(ERR, VHOST_CONFIG,
 583                                 "couldn't get hugepage size through fstat\n");
 584                         goto err_mmap;
 585                 }
 586                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 587
 588                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 589                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 590
 591                 if (mmap_addr == MAP_FAILED) {
 592                         RTE_LOG(ERR, VHOST_CONFIG,
 593                                 "mmap region %u failed.\n", i);
 594                         goto err_mmap;
 595                 }
 596
 597                 reg->mmap_addr = mmap_addr;
 598                 reg->mmap_size = mmap_size;
 599                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 600                                       mmap_offset;
 601
 602                 if (dev->dequeue_zero_copy)
 603                         add_guest_pages(dev, reg, alignment);
 604
 605                 RTE_LOG(INFO, VHOST_CONFIG,
 606                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 607                         "\t guest physical addr: 0x%" PRIx64 "\n"
 608                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 609                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 610                         "\t mmap addr : 0x%" PRIx64 "\n"
 611                         "\t mmap size : 0x%" PRIx64 "\n"
 612                         "\t mmap align: 0x%" PRIx64 "\n"
 613                         "\t mmap off  : 0x%" PRIx64 "\n",
 614                         i, reg->size,
 615                         reg->guest_phys_addr,
 616                         reg->guest_user_addr,
 617                         reg->host_user_addr,
 618                         (uint64_t)(uintptr_t)mmap_addr,
 619                         mmap_size,
 620                         alignment,
 621                         mmap_offset);
 622         }
 623
 624         dump_guest_pages(dev);
 625
 626         return 0;
 627
 628 err_mmap:
 629         free_mem_region(dev);
 630         rte_free(dev->mem);
 631         dev->mem = NULL;
 632         return -1;
 633 }
 634
 635 static int
 636 vq_is_ready(struct vhost_virtqueue *vq)
 637 {
 638         return vq && vq->desc   &&
 639                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 640                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 641 }
 642
 643 static int
 644 virtio_is_ready(struct virtio_net *dev)
 645 {
 646         struct vhost_virtqueue *vq;
 647         uint32_t i;
 648
 649         if (dev->nr_vring == 0)
 650                 return 0;
 651
 652         for (i = 0; i < dev->nr_vring; i++) {
 653                 vq = dev->virtqueue[i];
 654
 655                 if (!vq_is_ready(vq))
 656                         return 0;
 657         }
 658
 659         RTE_LOG(INFO, VHOST_CONFIG,
 660                 "virtio is now ready for processing.\n");
 661         return 1;
 662 }
 663
 664 static void
 665 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 666 {
 667         struct vhost_vring_file file;
 668         struct vhost_virtqueue *vq;
 669
 670         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 671         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 672                 file.fd = VIRTIO_INVALID_EVENTFD;
 673         else
 674                 file.fd = pmsg->fds[0];
 675         RTE_LOG(INFO, VHOST_CONFIG,
 676                 "vring call idx:%d file:%d\n", file.index, file.fd);
 677
 678         vq = dev->virtqueue[file.index];
 679         if (vq->callfd >= 0)
 680                 close(vq->callfd);
 681
 682         vq->callfd = file.fd;
 683 }
 684
 685 static void
 686 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 687 {
 688         struct vhost_vring_file file;
 689         struct vhost_virtqueue *vq;
 690
 691         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 692         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 693                 file.fd = VIRTIO_INVALID_EVENTFD;
 694         else
 695                 file.fd = pmsg->fds[0];
 696         RTE_LOG(INFO, VHOST_CONFIG,
 697                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 698
 699         vq = dev->virtqueue[file.index];
 700         if (vq->kickfd >= 0)
 701                 close(vq->kickfd);
 702         vq->kickfd = file.fd;
 703 }
 704
 705 static void
 706 free_zmbufs(struct vhost_virtqueue *vq)
 707 {
 708         struct zcopy_mbuf *zmbuf, *next;
 709
 710         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 711              zmbuf != NULL; zmbuf = next) {
 712                 next = TAILQ_NEXT(zmbuf, next);
 713
 714                 rte_pktmbuf_free(zmbuf->mbuf);
 715                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 716         }
 717
 718         rte_free(vq->zmbufs);
 719 }
 720
 721 /*
 722  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 723  */
 724 static int
 725 vhost_user_get_vring_base(struct virtio_net *dev,
 726                           VhostUserMsg *msg)
 727 {
 728         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 729
 730         /* We have to stop the queue (virtio) if it is running. */
 731         if (dev->flags & VIRTIO_DEV_RUNNING) {
 732                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 733                 dev->notify_ops->destroy_device(dev->vid);
 734         }
 735
 736         dev->flags &= ~VIRTIO_DEV_READY;
 737
 738         /* Here we are safe to get the last used index */
 739         msg->payload.state.num = vq->last_used_idx;
 740
 741         RTE_LOG(INFO, VHOST_CONFIG,
 742                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 743                 msg->payload.state.num);
 744         /*
 745          * Based on current qemu vhost-user implementation, this message is
 746          * sent and only sent in vhost_vring_stop.
 747          * TODO: cleanup the vring, it isn't usable since here.
 748          */
 749         if (vq->kickfd >= 0)
 750                 close(vq->kickfd);
 751
 752         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 753
 754         if (dev->dequeue_zero_copy)
 755                 free_zmbufs(vq);
 756         rte_free(vq->shadow_used_ring);
 757         vq->shadow_used_ring = NULL;
 758
 759         rte_free(vq->batch_copy_elems);
 760         vq->batch_copy_elems = NULL;
 761
 762         return 0;
 763 }
 764
 765 /*
 766  * when virtio queues are ready to work, qemu will send us to
 767  * enable the virtio queue pair.
 768  */
 769 static int
 770 vhost_user_set_vring_enable(struct virtio_net *dev,
 771                             VhostUserMsg *msg)
 772 {
 773         int enable = (int)msg->payload.state.num;
 774
 775         RTE_LOG(INFO, VHOST_CONFIG,
 776                 "set queue enable: %d to qp idx: %d\n",
 777                 enable, msg->payload.state.index);
 778
 779         if (dev->notify_ops->vring_state_changed)
 780                 dev->notify_ops->vring_state_changed(dev->vid,
 781                                 msg->payload.state.index, enable);
 782
 783         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 784
 785         return 0;
 786 }
 787
 788 static void
 789 vhost_user_set_protocol_features(struct virtio_net *dev,
 790                                  uint64_t protocol_features)
 791 {
 792         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 793                 return;
 794
 795         dev->protocol_features = protocol_features;
 796 }
 797
 798 static int
 799 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 800 {
 801         int fd = msg->fds[0];
 802         uint64_t size, off;
 803         void *addr;
 804
 805         if (fd < 0) {
 806                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 807                 return -1;
 808         }
 809
 810         if (msg->size != sizeof(VhostUserLog)) {
 811                 RTE_LOG(ERR, VHOST_CONFIG,
 812                         "invalid log base msg size: %"PRId32" != %d\n",
 813                         msg->size, (int)sizeof(VhostUserLog));
 814                 return -1;
 815         }
 816
 817         size = msg->payload.log.mmap_size;
 818         off  = msg->payload.log.mmap_offset;
 819         RTE_LOG(INFO, VHOST_CONFIG,
 820                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 821                 size, off);
 822
 823         /*
 824          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 825          * fail when offset is not page size aligned.
 826          */
 827         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 828         close(fd);
 829         if (addr == MAP_FAILED) {
 830                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 831                 return -1;
 832         }
 833
 834         /*
 835          * Free previously mapped log memory on occasionally
 836          * multiple VHOST_USER_SET_LOG_BASE.
 837          */
 838         if (dev->log_addr) {
 839                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 840         }
 841         dev->log_addr = (uint64_t)(uintptr_t)addr;
 842         dev->log_base = dev->log_addr + off;
 843         dev->log_size = size;
 844
 845         return 0;
 846 }
 847
 848 /*
 849  * An rarp packet is constructed and broadcasted to notify switches about
 850  * the new location of the migrated VM, so that packets from outside will
 851  * not be lost after migration.
 852  *
 853  * However, we don't actually "send" a rarp packet here, instead, we set
 854  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 855  */
 856 static int
 857 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 858 {
 859         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 860
 861         RTE_LOG(DEBUG, VHOST_CONFIG,
 862                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 863                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 864         memcpy(dev->mac.addr_bytes, mac, 6);
 865
 866         /*
 867          * Set the flag to inject a RARP broadcast packet at
 868          * rte_vhost_dequeue_burst().
 869          *
 870          * rte_smp_wmb() is for making sure the mac is copied
 871          * before the flag is set.
 872          */
 873         rte_smp_wmb();
 874         rte_atomic16_set(&dev->broadcast_rarp, 1);
 875
 876         return 0;
 877 }
 878
 879 static int
 880 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 881 {
 882         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 883                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 884                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 885                                 msg->payload.u64);
 886
 887                 return -1;
 888         }
 889
 890         dev->mtu = msg->payload.u64;
 891
 892         return 0;
 893 }
 894
 895 static int
 896 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 897 {
 898         int fd = msg->fds[0];
 899
 900         if (fd < 0) {
 901                 RTE_LOG(ERR, VHOST_CONFIG,
 902                                 "Invalid file descriptor for slave channel (%d)\n",
 903                                 fd);
 904                 return -1;
 905         }
 906
 907         dev->slave_req_fd = fd;
 908
 909         return 0;
 910 }
 911
 912 /* return bytes# of read on success or negative val on failure. */
 913 static int
 914 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 915 {
 916         int ret;
 917
 918         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 919                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 920         if (ret <= 0)
 921                 return ret;
 922
 923         if (msg && msg->size) {
 924                 if (msg->size > sizeof(msg->payload)) {
 925                         RTE_LOG(ERR, VHOST_CONFIG,
 926                                 "invalid msg size: %d\n", msg->size);
 927                         return -1;
 928                 }
 929                 ret = read(sockfd, &msg->payload, msg->size);
 930                 if (ret <= 0)
 931                         return ret;
 932                 if (ret != (int)msg->size) {
 933                         RTE_LOG(ERR, VHOST_CONFIG,
 934                                 "read control message failed\n");
 935                         return -1;
 936                 }
 937         }
 938
 939         return ret;
 940 }
 941
 942 static int
 943 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 944 {
 945         if (!msg)
 946                 return 0;
 947
 948         return send_fd_message(sockfd, (char *)msg,
 949                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 950 }
 951
 952 static int
 953 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
 954 {
 955         if (!msg)
 956                 return 0;
 957
 958         msg->flags &= ~VHOST_USER_VERSION_MASK;
 959         msg->flags &= ~VHOST_USER_NEED_REPLY;
 960         msg->flags |= VHOST_USER_VERSION;
 961         msg->flags |= VHOST_USER_REPLY_MASK;
 962
 963         return send_vhost_message(sockfd, msg);
 964 }
 965
 966 /*
 967  * Allocate a queue pair if it hasn't been allocated yet
 968  */
 969 static int
 970 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
 971 {
 972         uint16_t vring_idx;
 973
 974         switch (msg->request) {
 975         case VHOST_USER_SET_VRING_KICK:
 976         case VHOST_USER_SET_VRING_CALL:
 977         case VHOST_USER_SET_VRING_ERR:
 978                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 979                 break;
 980         case VHOST_USER_SET_VRING_NUM:
 981         case VHOST_USER_SET_VRING_BASE:
 982         case VHOST_USER_SET_VRING_ENABLE:
 983                 vring_idx = msg->payload.state.index;
 984                 break;
 985         case VHOST_USER_SET_VRING_ADDR:
 986                 vring_idx = msg->payload.addr.index;
 987                 break;
 988         default:
 989                 return 0;
 990         }
 991
 992         if (vring_idx >= VHOST_MAX_VRING) {
 993                 RTE_LOG(ERR, VHOST_CONFIG,
 994                         "invalid vring index: %u\n", vring_idx);
 995                 return -1;
 996         }
 997
 998         if (dev->virtqueue[vring_idx])
 999                 return 0;
1000
1001         return alloc_vring_queue(dev, vring_idx);
1002 }
1003
1004 int
1005 vhost_user_msg_handler(int vid, int fd)
1006 {
1007         struct virtio_net *dev;
1008         struct VhostUserMsg msg;
1009         int ret;
1010
1011         dev = get_device(vid);
1012         if (dev == NULL)
1013                 return -1;
1014
1015         if (!dev->notify_ops) {
1016                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1017                 if (!dev->notify_ops) {
1018                         RTE_LOG(ERR, VHOST_CONFIG,
1019                                 "failed to get callback ops for driver %s\n",
1020                                 dev->ifname);
1021                         return -1;
1022                 }
1023         }
1024
1025         ret = read_vhost_message(fd, &msg);
1026         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1027                 if (ret < 0)
1028                         RTE_LOG(ERR, VHOST_CONFIG,
1029                                 "vhost read message failed\n");
1030                 else if (ret == 0)
1031                         RTE_LOG(INFO, VHOST_CONFIG,
1032                                 "vhost peer closed\n");
1033                 else
1034                         RTE_LOG(ERR, VHOST_CONFIG,
1035                                 "vhost read incorrect message\n");
1036
1037                 return -1;
1038         }
1039
1040         ret = 0;
1041         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1042                 vhost_message_str[msg.request]);
1043
1044         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1045         if (ret < 0) {
1046                 RTE_LOG(ERR, VHOST_CONFIG,
1047                         "failed to alloc queue\n");
1048                 return -1;
1049         }
1050
1051         switch (msg.request) {
1052         case VHOST_USER_GET_FEATURES:
1053                 msg.payload.u64 = vhost_user_get_features(dev);
1054                 msg.size = sizeof(msg.payload.u64);
1055                 send_vhost_reply(fd, &msg);
1056                 break;
1057         case VHOST_USER_SET_FEATURES:
1058                 vhost_user_set_features(dev, msg.payload.u64);
1059                 break;
1060
1061         case VHOST_USER_GET_PROTOCOL_FEATURES:
1062                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1063                 msg.size = sizeof(msg.payload.u64);
1064                 send_vhost_reply(fd, &msg);
1065                 break;
1066         case VHOST_USER_SET_PROTOCOL_FEATURES:
1067                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1068                 break;
1069
1070         case VHOST_USER_SET_OWNER:
1071                 vhost_user_set_owner();
1072                 break;
1073         case VHOST_USER_RESET_OWNER:
1074                 vhost_user_reset_owner(dev);
1075                 break;
1076
1077         case VHOST_USER_SET_MEM_TABLE:
1078                 ret = vhost_user_set_mem_table(dev, &msg);
1079                 break;
1080
1081         case VHOST_USER_SET_LOG_BASE:
1082                 vhost_user_set_log_base(dev, &msg);
1083
1084                 /* it needs a reply */
1085                 msg.size = sizeof(msg.payload.u64);
1086                 send_vhost_reply(fd, &msg);
1087                 break;
1088         case VHOST_USER_SET_LOG_FD:
1089                 close(msg.fds[0]);
1090                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1091                 break;
1092
1093         case VHOST_USER_SET_VRING_NUM:
1094                 vhost_user_set_vring_num(dev, &msg);
1095                 break;
1096         case VHOST_USER_SET_VRING_ADDR:
1097                 vhost_user_set_vring_addr(dev, &msg);
1098                 break;
1099         case VHOST_USER_SET_VRING_BASE:
1100                 vhost_user_set_vring_base(dev, &msg);
1101                 break;
1102
1103         case VHOST_USER_GET_VRING_BASE:
1104                 vhost_user_get_vring_base(dev, &msg);
1105                 msg.size = sizeof(msg.payload.state);
1106                 send_vhost_reply(fd, &msg);
1107                 break;
1108
1109         case VHOST_USER_SET_VRING_KICK:
1110                 vhost_user_set_vring_kick(dev, &msg);
1111                 break;
1112         case VHOST_USER_SET_VRING_CALL:
1113                 vhost_user_set_vring_call(dev, &msg);
1114                 break;
1115
1116         case VHOST_USER_SET_VRING_ERR:
1117                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1118                         close(msg.fds[0]);
1119                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1120                 break;
1121
1122         case VHOST_USER_GET_QUEUE_NUM:
1123                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1124                 msg.size = sizeof(msg.payload.u64);
1125                 send_vhost_reply(fd, &msg);
1126                 break;
1127
1128         case VHOST_USER_SET_VRING_ENABLE:
1129                 vhost_user_set_vring_enable(dev, &msg);
1130                 break;
1131         case VHOST_USER_SEND_RARP:
1132                 vhost_user_send_rarp(dev, &msg);
1133                 break;
1134
1135         case VHOST_USER_NET_SET_MTU:
1136                 ret = vhost_user_net_set_mtu(dev, &msg);
1137                 break;
1138
1139         case VHOST_USER_SET_SLAVE_REQ_FD:
1140                 ret = vhost_user_set_req_fd(dev, &msg);
1141                 break;
1142
1143         default:
1144                 ret = -1;
1145                 break;
1146
1147         }
1148
1149         if (msg.flags & VHOST_USER_NEED_REPLY) {
1150                 msg.payload.u64 = !!ret;
1151                 msg.size = sizeof(msg.payload.u64);
1152                 send_vhost_reply(fd, &msg);
1153         }
1154
1155         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1156                 dev->flags |= VIRTIO_DEV_READY;
1157
1158                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1159                         if (dev->dequeue_zero_copy) {
1160                                 RTE_LOG(INFO, VHOST_CONFIG,
1161                                                 "dequeue zero copy is enabled\n");
1162                         }
1163
1164                         if (dev->notify_ops->new_device(dev->vid) == 0)
1165                                 dev->flags |= VIRTIO_DEV_RUNNING;
1166                 }
1167         }
1168
1169         return 0;
1170 }