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 static const char *vhost_message_str[VHOST_USER_MAX] = {
  55         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  56         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  57         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  58         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  59         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  60         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  61         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  62         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  63         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  64         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  65         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  66         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  67         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  68         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  69         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  70         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  71         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  72         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  73         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  74         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  75 };
  76
  77 static uint64_t
  78 get_blk_size(int fd)
  79 {
  80         struct stat stat;
  81         int ret;
  82
  83         ret = fstat(fd, &stat);
  84         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  85 }
  86
  87 static void
  88 free_mem_region(struct virtio_net *dev)
  89 {
  90         uint32_t i;
  91         struct virtio_memory_region *reg;
  92
  93         if (!dev || !dev->mem)
  94                 return;
  95
  96         for (i = 0; i < dev->mem->nregions; i++) {
  97                 reg = &dev->mem->regions[i];
  98                 if (reg->host_user_addr) {
  99                         munmap(reg->mmap_addr, reg->mmap_size);
 100                         close(reg->fd);
 101                 }
 102         }
 103 }
 104
 105 void
 106 vhost_backend_cleanup(struct virtio_net *dev)
 107 {
 108         if (dev->mem) {
 109                 free_mem_region(dev);
 110                 rte_free(dev->mem);
 111                 dev->mem = NULL;
 112         }
 113         if (dev->log_addr) {
 114                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 115                 dev->log_addr = 0;
 116         }
 117 }
 118
 119 /*
 120  * This function just returns success at the moment unless
 121  * the device hasn't been initialised.
 122  */
 123 static int
 124 vhost_user_set_owner(void)
 125 {
 126         return 0;
 127 }
 128
 129 static int
 130 vhost_user_reset_owner(struct virtio_net *dev)
 131 {
 132         if (dev->flags & VIRTIO_DEV_RUNNING) {
 133                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 134                 notify_ops->destroy_device(dev->vid);
 135         }
 136
 137         cleanup_device(dev, 0);
 138         reset_device(dev);
 139         return 0;
 140 }
 141
 142 /*
 143  * The features that we support are requested.
 144  */
 145 static uint64_t
 146 vhost_user_get_features(void)
 147 {
 148         return VHOST_FEATURES;
 149 }
 150
 151 /*
 152  * We receive the negotiated features supported by us and the virtio device.
 153  */
 154 static int
 155 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 156 {
 157         if (features & ~VHOST_FEATURES)
 158                 return -1;
 159
 160         dev->features = features;
 161         if (dev->features &
 162                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 163                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 164         } else {
 165                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 166         }
 167         LOG_DEBUG(VHOST_CONFIG,
 168                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 169                 dev->vid,
 170                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 171                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 172
 173         return 0;
 174 }
 175
 176 /*
 177  * The virtio device sends us the size of the descriptor ring.
 178  */
 179 static int
 180 vhost_user_set_vring_num(struct virtio_net *dev,
 181                          struct vhost_vring_state *state)
 182 {
 183         dev->virtqueue[state->index]->size = state->num;
 184
 185         return 0;
 186 }
 187
 188 /*
 189  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 190  * same numa node as the memory of vring descriptor.
 191  */
 192 #ifdef RTE_LIBRTE_VHOST_NUMA
 193 static struct virtio_net*
 194 numa_realloc(struct virtio_net *dev, int index)
 195 {
 196         int oldnode, newnode;
 197         struct virtio_net *old_dev;
 198         struct vhost_virtqueue *old_vq, *vq;
 199         int ret;
 200
 201         /*
 202          * vq is allocated on pairs, we should try to do realloc
 203          * on first queue of one queue pair only.
 204          */
 205         if (index % VIRTIO_QNUM != 0)
 206                 return dev;
 207
 208         old_dev = dev;
 209         vq = old_vq = dev->virtqueue[index];
 210
 211         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 212                             MPOL_F_NODE | MPOL_F_ADDR);
 213
 214         /* check if we need to reallocate vq */
 215         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 216                              MPOL_F_NODE | MPOL_F_ADDR);
 217         if (ret) {
 218                 RTE_LOG(ERR, VHOST_CONFIG,
 219                         "Unable to get vq numa information.\n");
 220                 return dev;
 221         }
 222         if (oldnode != newnode) {
 223                 RTE_LOG(INFO, VHOST_CONFIG,
 224                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 225                 vq = rte_malloc_socket(NULL, sizeof(*vq) * VIRTIO_QNUM, 0,
 226                                        newnode);
 227                 if (!vq)
 228                         return dev;
 229
 230                 memcpy(vq, old_vq, sizeof(*vq) * VIRTIO_QNUM);
 231                 rte_free(old_vq);
 232         }
 233
 234         /* check if we need to reallocate dev */
 235         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 236                             MPOL_F_NODE | MPOL_F_ADDR);
 237         if (ret) {
 238                 RTE_LOG(ERR, VHOST_CONFIG,
 239                         "Unable to get dev numa information.\n");
 240                 goto out;
 241         }
 242         if (oldnode != newnode) {
 243                 RTE_LOG(INFO, VHOST_CONFIG,
 244                         "reallocate dev from %d to %d node\n",
 245                         oldnode, newnode);
 246                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 247                 if (!dev) {
 248                         dev = old_dev;
 249                         goto out;
 250                 }
 251
 252                 memcpy(dev, old_dev, sizeof(*dev));
 253                 rte_free(old_dev);
 254         }
 255
 256 out:
 257         dev->virtqueue[index] = vq;
 258         dev->virtqueue[index + 1] = vq + 1;
 259         vhost_devices[dev->vid] = dev;
 260
 261         return dev;
 262 }
 263 #else
 264 static struct virtio_net*
 265 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 266 {
 267         return dev;
 268 }
 269 #endif
 270
 271 /*
 272  * Converts QEMU virtual address to Vhost virtual address. This function is
 273  * used to convert the ring addresses to our address space.
 274  */
 275 static uint64_t
 276 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 277 {
 278         struct virtio_memory_region *reg;
 279         uint32_t i;
 280
 281         /* Find the region where the address lives. */
 282         for (i = 0; i < dev->mem->nregions; i++) {
 283                 reg = &dev->mem->regions[i];
 284
 285                 if (qva >= reg->guest_user_addr &&
 286                     qva <  reg->guest_user_addr + reg->size) {
 287                         return qva - reg->guest_user_addr +
 288                                reg->host_user_addr;
 289                 }
 290         }
 291
 292         return 0;
 293 }
 294
 295 /*
 296  * The virtio device sends us the desc, used and avail ring addresses.
 297  * This function then converts these to our address space.
 298  */
 299 static int
 300 vhost_user_set_vring_addr(struct virtio_net *dev, struct vhost_vring_addr *addr)
 301 {
 302         struct vhost_virtqueue *vq;
 303
 304         if (dev->mem == NULL)
 305                 return -1;
 306
 307         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 308         vq = dev->virtqueue[addr->index];
 309
 310         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 311         vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
 312                         addr->desc_user_addr);
 313         if (vq->desc == 0) {
 314                 RTE_LOG(ERR, VHOST_CONFIG,
 315                         "(%d) failed to find desc ring address.\n",
 316                         dev->vid);
 317                 return -1;
 318         }
 319
 320         dev = numa_realloc(dev, addr->index);
 321         vq = dev->virtqueue[addr->index];
 322
 323         vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
 324                         addr->avail_user_addr);
 325         if (vq->avail == 0) {
 326                 RTE_LOG(ERR, VHOST_CONFIG,
 327                         "(%d) failed to find avail ring address.\n",
 328                         dev->vid);
 329                 return -1;
 330         }
 331
 332         vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
 333                         addr->used_user_addr);
 334         if (vq->used == 0) {
 335                 RTE_LOG(ERR, VHOST_CONFIG,
 336                         "(%d) failed to find used ring address.\n",
 337                         dev->vid);
 338                 return -1;
 339         }
 340
 341         if (vq->last_used_idx != vq->used->idx) {
 342                 RTE_LOG(WARNING, VHOST_CONFIG,
 343                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 344                         "some packets maybe resent for Tx and dropped for Rx\n",
 345                         vq->last_used_idx, vq->used->idx);
 346                 vq->last_used_idx     = vq->used->idx;
 347         }
 348
 349         vq->log_guest_addr = addr->log_guest_addr;
 350
 351         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 352                         dev->vid, vq->desc);
 353         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 354                         dev->vid, vq->avail);
 355         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 356                         dev->vid, vq->used);
 357         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 358                         dev->vid, vq->log_guest_addr);
 359
 360         return 0;
 361 }
 362
 363 /*
 364  * The virtio device sends us the available ring last used index.
 365  */
 366 static int
 367 vhost_user_set_vring_base(struct virtio_net *dev,
 368                           struct vhost_vring_state *state)
 369 {
 370         dev->virtqueue[state->index]->last_used_idx = state->num;
 371
 372         return 0;
 373 }
 374
 375 static void
 376 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 377                    uint64_t host_phys_addr, uint64_t size)
 378 {
 379         struct guest_page *page, *last_page;
 380
 381         if (dev->nr_guest_pages == dev->max_guest_pages) {
 382                 dev->max_guest_pages *= 2;
 383                 dev->guest_pages = realloc(dev->guest_pages,
 384                                         dev->max_guest_pages * sizeof(*page));
 385         }
 386
 387         if (dev->nr_guest_pages > 0) {
 388                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 389                 /* merge if the two pages are continuous */
 390                 if (host_phys_addr == last_page->host_phys_addr +
 391                                       last_page->size) {
 392                         last_page->size += size;
 393                         return;
 394                 }
 395         }
 396
 397         page = &dev->guest_pages[dev->nr_guest_pages++];
 398         page->guest_phys_addr = guest_phys_addr;
 399         page->host_phys_addr  = host_phys_addr;
 400         page->size = size;
 401 }
 402
 403 static void
 404 add_guest_pages(struct virtio_net *dev, struct virtio_memory_region *reg,
 405                 uint64_t page_size)
 406 {
 407         uint64_t reg_size = reg->size;
 408         uint64_t host_user_addr  = reg->host_user_addr;
 409         uint64_t guest_phys_addr = reg->guest_phys_addr;
 410         uint64_t host_phys_addr;
 411         uint64_t size;
 412
 413         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 414         size = page_size - (guest_phys_addr & (page_size - 1));
 415         size = RTE_MIN(size, reg_size);
 416
 417         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 418         host_user_addr  += size;
 419         guest_phys_addr += size;
 420         reg_size -= size;
 421
 422         while (reg_size > 0) {
 423                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 424                                                   host_user_addr);
 425                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
 426                                    page_size);
 427
 428                 host_user_addr  += page_size;
 429                 guest_phys_addr += page_size;
 430                 reg_size -= page_size;
 431         }
 432 }
 433
 434 #ifdef RTE_LIBRTE_VHOST_DEBUG
 435 /* TODO: enable it only in debug mode? */
 436 static void
 437 dump_guest_pages(struct virtio_net *dev)
 438 {
 439         uint32_t i;
 440         struct guest_page *page;
 441
 442         for (i = 0; i < dev->nr_guest_pages; i++) {
 443                 page = &dev->guest_pages[i];
 444
 445                 RTE_LOG(INFO, VHOST_CONFIG,
 446                         "guest physical page region %u\n"
 447                         "\t guest_phys_addr: %" PRIx64 "\n"
 448                         "\t host_phys_addr : %" PRIx64 "\n"
 449                         "\t size           : %" PRIx64 "\n",
 450                         i,
 451                         page->guest_phys_addr,
 452                         page->host_phys_addr,
 453                         page->size);
 454         }
 455 }
 456 #else
 457 #define dump_guest_pages(dev)
 458 #endif
 459
 460 static int
 461 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 462 {
 463         struct VhostUserMemory memory = pmsg->payload.memory;
 464         struct virtio_memory_region *reg;
 465         void *mmap_addr;
 466         uint64_t mmap_size;
 467         uint64_t mmap_offset;
 468         uint64_t alignment;
 469         uint32_t i;
 470         int fd;
 471
 472         /* Remove from the data plane. */
 473         if (dev->flags & VIRTIO_DEV_RUNNING) {
 474                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 475                 notify_ops->destroy_device(dev->vid);
 476         }
 477
 478         if (dev->mem) {
 479                 free_mem_region(dev);
 480                 rte_free(dev->mem);
 481                 dev->mem = NULL;
 482         }
 483
 484         dev->nr_guest_pages = 0;
 485         if (!dev->guest_pages) {
 486                 dev->max_guest_pages = 8;
 487                 dev->guest_pages = malloc(dev->max_guest_pages *
 488                                                 sizeof(struct guest_page));
 489         }
 490
 491         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct virtio_memory) +
 492                 sizeof(struct virtio_memory_region) * memory.nregions, 0);
 493         if (dev->mem == NULL) {
 494                 RTE_LOG(ERR, VHOST_CONFIG,
 495                         "(%d) failed to allocate memory for dev->mem\n",
 496                         dev->vid);
 497                 return -1;
 498         }
 499         dev->mem->nregions = memory.nregions;
 500
 501         for (i = 0; i < memory.nregions; i++) {
 502                 fd  = pmsg->fds[i];
 503                 reg = &dev->mem->regions[i];
 504
 505                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 506                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 507                 reg->size            = memory.regions[i].memory_size;
 508                 reg->fd              = fd;
 509
 510                 mmap_offset = memory.regions[i].mmap_offset;
 511                 mmap_size   = reg->size + mmap_offset;
 512
 513                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 514                  * with length argument aligned with hugepagesz at older
 515                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 516                  * mmap() will fail with EINVAL.
 517                  *
 518                  * to avoid failure, make sure in caller to keep length
 519                  * aligned.
 520                  */
 521                 alignment = get_blk_size(fd);
 522                 if (alignment == (uint64_t)-1) {
 523                         RTE_LOG(ERR, VHOST_CONFIG,
 524                                 "couldn't get hugepage size through fstat\n");
 525                         goto err_mmap;
 526                 }
 527                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 528
 529                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 530                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 531
 532                 if (mmap_addr == MAP_FAILED) {
 533                         RTE_LOG(ERR, VHOST_CONFIG,
 534                                 "mmap region %u failed.\n", i);
 535                         goto err_mmap;
 536                 }
 537
 538                 reg->mmap_addr = mmap_addr;
 539                 reg->mmap_size = mmap_size;
 540                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 541                                       mmap_offset;
 542
 543                 add_guest_pages(dev, reg, alignment);
 544
 545                 RTE_LOG(INFO, VHOST_CONFIG,
 546                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 547                         "\t guest physical addr: 0x%" PRIx64 "\n"
 548                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 549                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 550                         "\t mmap addr : 0x%" PRIx64 "\n"
 551                         "\t mmap size : 0x%" PRIx64 "\n"
 552                         "\t mmap align: 0x%" PRIx64 "\n"
 553                         "\t mmap off  : 0x%" PRIx64 "\n",
 554                         i, reg->size,
 555                         reg->guest_phys_addr,
 556                         reg->guest_user_addr,
 557                         reg->host_user_addr,
 558                         (uint64_t)(uintptr_t)mmap_addr,
 559                         mmap_size,
 560                         alignment,
 561                         mmap_offset);
 562         }
 563
 564         dump_guest_pages(dev);
 565
 566         return 0;
 567
 568 err_mmap:
 569         free_mem_region(dev);
 570         rte_free(dev->mem);
 571         dev->mem = NULL;
 572         return -1;
 573 }
 574
 575 static int
 576 vq_is_ready(struct vhost_virtqueue *vq)
 577 {
 578         return vq && vq->desc   &&
 579                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 580                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 581 }
 582
 583 static int
 584 virtio_is_ready(struct virtio_net *dev)
 585 {
 586         struct vhost_virtqueue *rvq, *tvq;
 587         uint32_t i;
 588
 589         for (i = 0; i < dev->virt_qp_nb; i++) {
 590                 rvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ];
 591                 tvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ];
 592
 593                 if (!vq_is_ready(rvq) || !vq_is_ready(tvq)) {
 594                         RTE_LOG(INFO, VHOST_CONFIG,
 595                                 "virtio is not ready for processing.\n");
 596                         return 0;
 597                 }
 598         }
 599
 600         RTE_LOG(INFO, VHOST_CONFIG,
 601                 "virtio is now ready for processing.\n");
 602         return 1;
 603 }
 604
 605 static void
 606 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 607 {
 608         struct vhost_vring_file file;
 609         struct vhost_virtqueue *vq;
 610         uint32_t cur_qp_idx;
 611
 612         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 613         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 614                 file.fd = VIRTIO_INVALID_EVENTFD;
 615         else
 616                 file.fd = pmsg->fds[0];
 617         RTE_LOG(INFO, VHOST_CONFIG,
 618                 "vring call idx:%d file:%d\n", file.index, file.fd);
 619
 620         /*
 621          * FIXME: VHOST_SET_VRING_CALL is the first per-vring message
 622          * we get, so we do vring queue pair allocation here.
 623          */
 624         cur_qp_idx = file.index / VIRTIO_QNUM;
 625         if (cur_qp_idx + 1 > dev->virt_qp_nb) {
 626                 if (alloc_vring_queue_pair(dev, cur_qp_idx) < 0)
 627                         return;
 628         }
 629
 630         vq = dev->virtqueue[file.index];
 631         assert(vq != NULL);
 632
 633         if (vq->callfd >= 0)
 634                 close(vq->callfd);
 635
 636         vq->callfd = file.fd;
 637 }
 638
 639 /*
 640  *  In vhost-user, when we receive kick message, will test whether virtio
 641  *  device is ready for packet processing.
 642  */
 643 static void
 644 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 645 {
 646         struct vhost_vring_file file;
 647         struct vhost_virtqueue *vq;
 648
 649         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 650         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 651                 file.fd = VIRTIO_INVALID_EVENTFD;
 652         else
 653                 file.fd = pmsg->fds[0];
 654         RTE_LOG(INFO, VHOST_CONFIG,
 655                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 656
 657         vq = dev->virtqueue[file.index];
 658         if (vq->kickfd >= 0)
 659                 close(vq->kickfd);
 660         vq->kickfd = file.fd;
 661
 662         if (virtio_is_ready(dev) && !(dev->flags & VIRTIO_DEV_RUNNING)) {
 663                 if (notify_ops->new_device(dev->vid) == 0)
 664                         dev->flags |= VIRTIO_DEV_RUNNING;
 665         }
 666 }
 667
 668 /*
 669  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 670  */
 671 static int
 672 vhost_user_get_vring_base(struct virtio_net *dev,
 673                           struct vhost_vring_state *state)
 674 {
 675         /* We have to stop the queue (virtio) if it is running. */
 676         if (dev->flags & VIRTIO_DEV_RUNNING) {
 677                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 678                 notify_ops->destroy_device(dev->vid);
 679         }
 680
 681         /* Here we are safe to get the last used index */
 682         state->num = dev->virtqueue[state->index]->last_used_idx;
 683
 684         RTE_LOG(INFO, VHOST_CONFIG,
 685                 "vring base idx:%d file:%d\n", state->index, state->num);
 686         /*
 687          * Based on current qemu vhost-user implementation, this message is
 688          * sent and only sent in vhost_vring_stop.
 689          * TODO: cleanup the vring, it isn't usable since here.
 690          */
 691         if (dev->virtqueue[state->index]->kickfd >= 0)
 692                 close(dev->virtqueue[state->index]->kickfd);
 693
 694         dev->virtqueue[state->index]->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 695
 696         return 0;
 697 }
 698
 699 /*
 700  * when virtio queues are ready to work, qemu will send us to
 701  * enable the virtio queue pair.
 702  */
 703 static int
 704 vhost_user_set_vring_enable(struct virtio_net *dev,
 705                             struct vhost_vring_state *state)
 706 {
 707         int enable = (int)state->num;
 708
 709         RTE_LOG(INFO, VHOST_CONFIG,
 710                 "set queue enable: %d to qp idx: %d\n",
 711                 enable, state->index);
 712
 713         if (notify_ops->vring_state_changed)
 714                 notify_ops->vring_state_changed(dev->vid, state->index, enable);
 715
 716         dev->virtqueue[state->index]->enabled = enable;
 717
 718         return 0;
 719 }
 720
 721 static void
 722 vhost_user_set_protocol_features(struct virtio_net *dev,
 723                                  uint64_t protocol_features)
 724 {
 725         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 726                 return;
 727
 728         dev->protocol_features = protocol_features;
 729 }
 730
 731 static int
 732 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 733 {
 734         int fd = msg->fds[0];
 735         uint64_t size, off;
 736         void *addr;
 737
 738         if (fd < 0) {
 739                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 740                 return -1;
 741         }
 742
 743         if (msg->size != sizeof(VhostUserLog)) {
 744                 RTE_LOG(ERR, VHOST_CONFIG,
 745                         "invalid log base msg size: %"PRId32" != %d\n",
 746                         msg->size, (int)sizeof(VhostUserLog));
 747                 return -1;
 748         }
 749
 750         size = msg->payload.log.mmap_size;
 751         off  = msg->payload.log.mmap_offset;
 752         RTE_LOG(INFO, VHOST_CONFIG,
 753                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 754                 size, off);
 755
 756         /*
 757          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 758          * fail when offset is not page size aligned.
 759          */
 760         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 761         close(fd);
 762         if (addr == MAP_FAILED) {
 763                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 764                 return -1;
 765         }
 766
 767         /*
 768          * Free previously mapped log memory on occasionally
 769          * multiple VHOST_USER_SET_LOG_BASE.
 770          */
 771         if (dev->log_addr) {
 772                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 773         }
 774         dev->log_addr = (uint64_t)(uintptr_t)addr;
 775         dev->log_base = dev->log_addr + off;
 776         dev->log_size = size;
 777
 778         return 0;
 779 }
 780
 781 /*
 782  * An rarp packet is constructed and broadcasted to notify switches about
 783  * the new location of the migrated VM, so that packets from outside will
 784  * not be lost after migration.
 785  *
 786  * However, we don't actually "send" a rarp packet here, instead, we set
 787  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 788  */
 789 static int
 790 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 791 {
 792         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 793
 794         RTE_LOG(DEBUG, VHOST_CONFIG,
 795                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 796                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 797         memcpy(dev->mac.addr_bytes, mac, 6);
 798
 799         /*
 800          * Set the flag to inject a RARP broadcast packet at
 801          * rte_vhost_dequeue_burst().
 802          *
 803          * rte_smp_wmb() is for making sure the mac is copied
 804          * before the flag is set.
 805          */
 806         rte_smp_wmb();
 807         rte_atomic16_set(&dev->broadcast_rarp, 1);
 808
 809         return 0;
 810 }
 811
 812 /* return bytes# of read on success or negative val on failure. */
 813 static int
 814 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 815 {
 816         int ret;
 817
 818         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 819                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 820         if (ret <= 0)
 821                 return ret;
 822
 823         if (msg && msg->size) {
 824                 if (msg->size > sizeof(msg->payload)) {
 825                         RTE_LOG(ERR, VHOST_CONFIG,
 826                                 "invalid msg size: %d\n", msg->size);
 827                         return -1;
 828                 }
 829                 ret = read(sockfd, &msg->payload, msg->size);
 830                 if (ret <= 0)
 831                         return ret;
 832                 if (ret != (int)msg->size) {
 833                         RTE_LOG(ERR, VHOST_CONFIG,
 834                                 "read control message failed\n");
 835                         return -1;
 836                 }
 837         }
 838
 839         return ret;
 840 }
 841
 842 static int
 843 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 844 {
 845         int ret;
 846
 847         if (!msg)
 848                 return 0;
 849
 850         msg->flags &= ~VHOST_USER_VERSION_MASK;
 851         msg->flags |= VHOST_USER_VERSION;
 852         msg->flags |= VHOST_USER_REPLY_MASK;
 853
 854         ret = send_fd_message(sockfd, (char *)msg,
 855                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 856
 857         return ret;
 858 }
 859
 860 int
 861 vhost_user_msg_handler(int vid, int fd)
 862 {
 863         struct virtio_net *dev;
 864         struct VhostUserMsg msg;
 865         int ret;
 866
 867         dev = get_device(vid);
 868         if (dev == NULL)
 869                 return -1;
 870
 871         ret = read_vhost_message(fd, &msg);
 872         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
 873                 if (ret < 0)
 874                         RTE_LOG(ERR, VHOST_CONFIG,
 875                                 "vhost read message failed\n");
 876                 else if (ret == 0)
 877                         RTE_LOG(INFO, VHOST_CONFIG,
 878                                 "vhost peer closed\n");
 879                 else
 880                         RTE_LOG(ERR, VHOST_CONFIG,
 881                                 "vhost read incorrect message\n");
 882
 883                 return -1;
 884         }
 885
 886         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
 887                 vhost_message_str[msg.request]);
 888         switch (msg.request) {
 889         case VHOST_USER_GET_FEATURES:
 890                 msg.payload.u64 = vhost_user_get_features();
 891                 msg.size = sizeof(msg.payload.u64);
 892                 send_vhost_message(fd, &msg);
 893                 break;
 894         case VHOST_USER_SET_FEATURES:
 895                 vhost_user_set_features(dev, msg.payload.u64);
 896                 break;
 897
 898         case VHOST_USER_GET_PROTOCOL_FEATURES:
 899                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
 900                 msg.size = sizeof(msg.payload.u64);
 901                 send_vhost_message(fd, &msg);
 902                 break;
 903         case VHOST_USER_SET_PROTOCOL_FEATURES:
 904                 vhost_user_set_protocol_features(dev, msg.payload.u64);
 905                 break;
 906
 907         case VHOST_USER_SET_OWNER:
 908                 vhost_user_set_owner();
 909                 break;
 910         case VHOST_USER_RESET_OWNER:
 911                 vhost_user_reset_owner(dev);
 912                 break;
 913
 914         case VHOST_USER_SET_MEM_TABLE:
 915                 vhost_user_set_mem_table(dev, &msg);
 916                 break;
 917
 918         case VHOST_USER_SET_LOG_BASE:
 919                 vhost_user_set_log_base(dev, &msg);
 920
 921                 /* it needs a reply */
 922                 msg.size = sizeof(msg.payload.u64);
 923                 send_vhost_message(fd, &msg);
 924                 break;
 925         case VHOST_USER_SET_LOG_FD:
 926                 close(msg.fds[0]);
 927                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
 928                 break;
 929
 930         case VHOST_USER_SET_VRING_NUM:
 931                 vhost_user_set_vring_num(dev, &msg.payload.state);
 932                 break;
 933         case VHOST_USER_SET_VRING_ADDR:
 934                 vhost_user_set_vring_addr(dev, &msg.payload.addr);
 935                 break;
 936         case VHOST_USER_SET_VRING_BASE:
 937                 vhost_user_set_vring_base(dev, &msg.payload.state);
 938                 break;
 939
 940         case VHOST_USER_GET_VRING_BASE:
 941                 ret = vhost_user_get_vring_base(dev, &msg.payload.state);
 942                 msg.size = sizeof(msg.payload.state);
 943                 send_vhost_message(fd, &msg);
 944                 break;
 945
 946         case VHOST_USER_SET_VRING_KICK:
 947                 vhost_user_set_vring_kick(dev, &msg);
 948                 break;
 949         case VHOST_USER_SET_VRING_CALL:
 950                 vhost_user_set_vring_call(dev, &msg);
 951                 break;
 952
 953         case VHOST_USER_SET_VRING_ERR:
 954                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
 955                         close(msg.fds[0]);
 956                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
 957                 break;
 958
 959         case VHOST_USER_GET_QUEUE_NUM:
 960                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
 961                 msg.size = sizeof(msg.payload.u64);
 962                 send_vhost_message(fd, &msg);
 963                 break;
 964
 965         case VHOST_USER_SET_VRING_ENABLE:
 966                 vhost_user_set_vring_enable(dev, &msg.payload.state);
 967                 break;
 968         case VHOST_USER_SEND_RARP:
 969                 vhost_user_send_rarp(dev, &msg);
 970                 break;
 971
 972         default:
 973                 break;
 974
 975         }
 976
 977         return 0;
 978 }