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