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 int
 376 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 377 {
 378         struct VhostUserMemory memory = pmsg->payload.memory;
 379         struct virtio_memory_region *reg;
 380         void *mmap_addr;
 381         uint64_t mmap_size;
 382         uint64_t mmap_offset;
 383         uint64_t alignment;
 384         uint32_t i;
 385         int fd;
 386
 387         /* Remove from the data plane. */
 388         if (dev->flags & VIRTIO_DEV_RUNNING) {
 389                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 390                 notify_ops->destroy_device(dev->vid);
 391         }
 392
 393         if (dev->mem) {
 394                 free_mem_region(dev);
 395                 rte_free(dev->mem);
 396                 dev->mem = NULL;
 397         }
 398
 399         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct virtio_memory) +
 400                 sizeof(struct virtio_memory_region) * memory.nregions, 0);
 401         if (dev->mem == NULL) {
 402                 RTE_LOG(ERR, VHOST_CONFIG,
 403                         "(%d) failed to allocate memory for dev->mem\n",
 404                         dev->vid);
 405                 return -1;
 406         }
 407         dev->mem->nregions = memory.nregions;
 408
 409         for (i = 0; i < memory.nregions; i++) {
 410                 fd  = pmsg->fds[i];
 411                 reg = &dev->mem->regions[i];
 412
 413                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 414                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 415                 reg->size            = memory.regions[i].memory_size;
 416                 reg->fd              = fd;
 417
 418                 mmap_offset = memory.regions[i].mmap_offset;
 419                 mmap_size   = reg->size + mmap_offset;
 420
 421                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 422                  * with length argument aligned with hugepagesz at older
 423                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 424                  * mmap() will fail with EINVAL.
 425                  *
 426                  * to avoid failure, make sure in caller to keep length
 427                  * aligned.
 428                  */
 429                 alignment = get_blk_size(fd);
 430                 if (alignment == (uint64_t)-1) {
 431                         RTE_LOG(ERR, VHOST_CONFIG,
 432                                 "couldn't get hugepage size through fstat\n");
 433                         goto err_mmap;
 434                 }
 435                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 436
 437                 mmap_addr = mmap(NULL, mmap_size,
 438                                  PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 439
 440                 if (mmap_addr == MAP_FAILED) {
 441                         RTE_LOG(ERR, VHOST_CONFIG,
 442                                 "mmap region %u failed.\n", i);
 443                         goto err_mmap;
 444                 }
 445
 446                 reg->mmap_addr = mmap_addr;
 447                 reg->mmap_size = mmap_size;
 448                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 449                                       mmap_offset;
 450
 451                 RTE_LOG(INFO, VHOST_CONFIG,
 452                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 453                         "\t guest physical addr: 0x%" PRIx64 "\n"
 454                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 455                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 456                         "\t mmap addr : 0x%" PRIx64 "\n"
 457                         "\t mmap size : 0x%" PRIx64 "\n"
 458                         "\t mmap align: 0x%" PRIx64 "\n"
 459                         "\t mmap off  : 0x%" PRIx64 "\n",
 460                         i, reg->size,
 461                         reg->guest_phys_addr,
 462                         reg->guest_user_addr,
 463                         reg->host_user_addr,
 464                         (uint64_t)(uintptr_t)mmap_addr,
 465                         mmap_size,
 466                         alignment,
 467                         mmap_offset);
 468         }
 469
 470         return 0;
 471
 472 err_mmap:
 473         free_mem_region(dev);
 474         rte_free(dev->mem);
 475         dev->mem = NULL;
 476         return -1;
 477 }
 478
 479 static int
 480 vq_is_ready(struct vhost_virtqueue *vq)
 481 {
 482         return vq && vq->desc   &&
 483                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 484                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 485 }
 486
 487 static int
 488 virtio_is_ready(struct virtio_net *dev)
 489 {
 490         struct vhost_virtqueue *rvq, *tvq;
 491         uint32_t i;
 492
 493         for (i = 0; i < dev->virt_qp_nb; i++) {
 494                 rvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_RXQ];
 495                 tvq = dev->virtqueue[i * VIRTIO_QNUM + VIRTIO_TXQ];
 496
 497                 if (!vq_is_ready(rvq) || !vq_is_ready(tvq)) {
 498                         RTE_LOG(INFO, VHOST_CONFIG,
 499                                 "virtio is not ready for processing.\n");
 500                         return 0;
 501                 }
 502         }
 503
 504         RTE_LOG(INFO, VHOST_CONFIG,
 505                 "virtio is now ready for processing.\n");
 506         return 1;
 507 }
 508
 509 static void
 510 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 511 {
 512         struct vhost_vring_file file;
 513         struct vhost_virtqueue *vq;
 514         uint32_t cur_qp_idx;
 515
 516         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 517         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 518                 file.fd = VIRTIO_INVALID_EVENTFD;
 519         else
 520                 file.fd = pmsg->fds[0];
 521         RTE_LOG(INFO, VHOST_CONFIG,
 522                 "vring call idx:%d file:%d\n", file.index, file.fd);
 523
 524         /*
 525          * FIXME: VHOST_SET_VRING_CALL is the first per-vring message
 526          * we get, so we do vring queue pair allocation here.
 527          */
 528         cur_qp_idx = file.index / VIRTIO_QNUM;
 529         if (cur_qp_idx + 1 > dev->virt_qp_nb) {
 530                 if (alloc_vring_queue_pair(dev, cur_qp_idx) < 0)
 531                         return;
 532         }
 533
 534         vq = dev->virtqueue[file.index];
 535         assert(vq != NULL);
 536
 537         if (vq->callfd >= 0)
 538                 close(vq->callfd);
 539
 540         vq->callfd = file.fd;
 541 }
 542
 543 /*
 544  *  In vhost-user, when we receive kick message, will test whether virtio
 545  *  device is ready for packet processing.
 546  */
 547 static void
 548 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 549 {
 550         struct vhost_vring_file file;
 551         struct vhost_virtqueue *vq;
 552
 553         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 554         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 555                 file.fd = VIRTIO_INVALID_EVENTFD;
 556         else
 557                 file.fd = pmsg->fds[0];
 558         RTE_LOG(INFO, VHOST_CONFIG,
 559                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 560
 561         vq = dev->virtqueue[file.index];
 562         if (vq->kickfd >= 0)
 563                 close(vq->kickfd);
 564         vq->kickfd = file.fd;
 565
 566         if (virtio_is_ready(dev) && !(dev->flags & VIRTIO_DEV_RUNNING)) {
 567                 if (notify_ops->new_device(dev->vid) == 0)
 568                         dev->flags |= VIRTIO_DEV_RUNNING;
 569         }
 570 }
 571
 572 /*
 573  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 574  */
 575 static int
 576 vhost_user_get_vring_base(struct virtio_net *dev,
 577                           struct vhost_vring_state *state)
 578 {
 579         /* We have to stop the queue (virtio) if it is running. */
 580         if (dev->flags & VIRTIO_DEV_RUNNING) {
 581                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 582                 notify_ops->destroy_device(dev->vid);
 583         }
 584
 585         /* Here we are safe to get the last used index */
 586         state->num = dev->virtqueue[state->index]->last_used_idx;
 587
 588         RTE_LOG(INFO, VHOST_CONFIG,
 589                 "vring base idx:%d file:%d\n", state->index, state->num);
 590         /*
 591          * Based on current qemu vhost-user implementation, this message is
 592          * sent and only sent in vhost_vring_stop.
 593          * TODO: cleanup the vring, it isn't usable since here.
 594          */
 595         if (dev->virtqueue[state->index]->kickfd >= 0)
 596                 close(dev->virtqueue[state->index]->kickfd);
 597
 598         dev->virtqueue[state->index]->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 599
 600         return 0;
 601 }
 602
 603 /*
 604  * when virtio queues are ready to work, qemu will send us to
 605  * enable the virtio queue pair.
 606  */
 607 static int
 608 vhost_user_set_vring_enable(struct virtio_net *dev,
 609                             struct vhost_vring_state *state)
 610 {
 611         int enable = (int)state->num;
 612
 613         RTE_LOG(INFO, VHOST_CONFIG,
 614                 "set queue enable: %d to qp idx: %d\n",
 615                 enable, state->index);
 616
 617         if (notify_ops->vring_state_changed)
 618                 notify_ops->vring_state_changed(dev->vid, state->index, enable);
 619
 620         dev->virtqueue[state->index]->enabled = enable;
 621
 622         return 0;
 623 }
 624
 625 static void
 626 vhost_user_set_protocol_features(struct virtio_net *dev,
 627                                  uint64_t protocol_features)
 628 {
 629         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 630                 return;
 631
 632         dev->protocol_features = protocol_features;
 633 }
 634
 635 static int
 636 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 637 {
 638         int fd = msg->fds[0];
 639         uint64_t size, off;
 640         void *addr;
 641
 642         if (fd < 0) {
 643                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 644                 return -1;
 645         }
 646
 647         if (msg->size != sizeof(VhostUserLog)) {
 648                 RTE_LOG(ERR, VHOST_CONFIG,
 649                         "invalid log base msg size: %"PRId32" != %d\n",
 650                         msg->size, (int)sizeof(VhostUserLog));
 651                 return -1;
 652         }
 653
 654         size = msg->payload.log.mmap_size;
 655         off  = msg->payload.log.mmap_offset;
 656         RTE_LOG(INFO, VHOST_CONFIG,
 657                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 658                 size, off);
 659
 660         /*
 661          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 662          * fail when offset is not page size aligned.
 663          */
 664         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 665         close(fd);
 666         if (addr == MAP_FAILED) {
 667                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 668                 return -1;
 669         }
 670
 671         /*
 672          * Free previously mapped log memory on occasionally
 673          * multiple VHOST_USER_SET_LOG_BASE.
 674          */
 675         if (dev->log_addr) {
 676                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 677         }
 678         dev->log_addr = (uint64_t)(uintptr_t)addr;
 679         dev->log_base = dev->log_addr + off;
 680         dev->log_size = size;
 681
 682         return 0;
 683 }
 684
 685 /*
 686  * An rarp packet is constructed and broadcasted to notify switches about
 687  * the new location of the migrated VM, so that packets from outside will
 688  * not be lost after migration.
 689  *
 690  * However, we don't actually "send" a rarp packet here, instead, we set
 691  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 692  */
 693 static int
 694 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 695 {
 696         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 697
 698         RTE_LOG(DEBUG, VHOST_CONFIG,
 699                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 700                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 701         memcpy(dev->mac.addr_bytes, mac, 6);
 702
 703         /*
 704          * Set the flag to inject a RARP broadcast packet at
 705          * rte_vhost_dequeue_burst().
 706          *
 707          * rte_smp_wmb() is for making sure the mac is copied
 708          * before the flag is set.
 709          */
 710         rte_smp_wmb();
 711         rte_atomic16_set(&dev->broadcast_rarp, 1);
 712
 713         return 0;
 714 }
 715
 716 /* return bytes# of read on success or negative val on failure. */
 717 static int
 718 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 719 {
 720         int ret;
 721
 722         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 723                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 724         if (ret <= 0)
 725                 return ret;
 726
 727         if (msg && msg->size) {
 728                 if (msg->size > sizeof(msg->payload)) {
 729                         RTE_LOG(ERR, VHOST_CONFIG,
 730                                 "invalid msg size: %d\n", msg->size);
 731                         return -1;
 732                 }
 733                 ret = read(sockfd, &msg->payload, msg->size);
 734                 if (ret <= 0)
 735                         return ret;
 736                 if (ret != (int)msg->size) {
 737                         RTE_LOG(ERR, VHOST_CONFIG,
 738                                 "read control message failed\n");
 739                         return -1;
 740                 }
 741         }
 742
 743         return ret;
 744 }
 745
 746 static int
 747 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 748 {
 749         int ret;
 750
 751         if (!msg)
 752                 return 0;
 753
 754         msg->flags &= ~VHOST_USER_VERSION_MASK;
 755         msg->flags |= VHOST_USER_VERSION;
 756         msg->flags |= VHOST_USER_REPLY_MASK;
 757
 758         ret = send_fd_message(sockfd, (char *)msg,
 759                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 760
 761         return ret;
 762 }
 763
 764 int
 765 vhost_user_msg_handler(int vid, int fd)
 766 {
 767         struct virtio_net *dev;
 768         struct VhostUserMsg msg;
 769         int ret;
 770
 771         dev = get_device(vid);
 772         if (dev == NULL)
 773                 return -1;
 774
 775         ret = read_vhost_message(fd, &msg);
 776         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
 777                 if (ret < 0)
 778                         RTE_LOG(ERR, VHOST_CONFIG,
 779                                 "vhost read message failed\n");
 780                 else if (ret == 0)
 781                         RTE_LOG(INFO, VHOST_CONFIG,
 782                                 "vhost peer closed\n");
 783                 else
 784                         RTE_LOG(ERR, VHOST_CONFIG,
 785                                 "vhost read incorrect message\n");
 786
 787                 return -1;
 788         }
 789
 790         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
 791                 vhost_message_str[msg.request]);
 792         switch (msg.request) {
 793         case VHOST_USER_GET_FEATURES:
 794                 msg.payload.u64 = vhost_user_get_features();
 795                 msg.size = sizeof(msg.payload.u64);
 796                 send_vhost_message(fd, &msg);
 797                 break;
 798         case VHOST_USER_SET_FEATURES:
 799                 vhost_user_set_features(dev, msg.payload.u64);
 800                 break;
 801
 802         case VHOST_USER_GET_PROTOCOL_FEATURES:
 803                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
 804                 msg.size = sizeof(msg.payload.u64);
 805                 send_vhost_message(fd, &msg);
 806                 break;
 807         case VHOST_USER_SET_PROTOCOL_FEATURES:
 808                 vhost_user_set_protocol_features(dev, msg.payload.u64);
 809                 break;
 810
 811         case VHOST_USER_SET_OWNER:
 812                 vhost_user_set_owner();
 813                 break;
 814         case VHOST_USER_RESET_OWNER:
 815                 vhost_user_reset_owner(dev);
 816                 break;
 817
 818         case VHOST_USER_SET_MEM_TABLE:
 819                 vhost_user_set_mem_table(dev, &msg);
 820                 break;
 821
 822         case VHOST_USER_SET_LOG_BASE:
 823                 vhost_user_set_log_base(dev, &msg);
 824
 825                 /* it needs a reply */
 826                 msg.size = sizeof(msg.payload.u64);
 827                 send_vhost_message(fd, &msg);
 828                 break;
 829         case VHOST_USER_SET_LOG_FD:
 830                 close(msg.fds[0]);
 831                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
 832                 break;
 833
 834         case VHOST_USER_SET_VRING_NUM:
 835                 vhost_user_set_vring_num(dev, &msg.payload.state);
 836                 break;
 837         case VHOST_USER_SET_VRING_ADDR:
 838                 vhost_user_set_vring_addr(dev, &msg.payload.addr);
 839                 break;
 840         case VHOST_USER_SET_VRING_BASE:
 841                 vhost_user_set_vring_base(dev, &msg.payload.state);
 842                 break;
 843
 844         case VHOST_USER_GET_VRING_BASE:
 845                 ret = vhost_user_get_vring_base(dev, &msg.payload.state);
 846                 msg.size = sizeof(msg.payload.state);
 847                 send_vhost_message(fd, &msg);
 848                 break;
 849
 850         case VHOST_USER_SET_VRING_KICK:
 851                 vhost_user_set_vring_kick(dev, &msg);
 852                 break;
 853         case VHOST_USER_SET_VRING_CALL:
 854                 vhost_user_set_vring_call(dev, &msg);
 855                 break;
 856
 857         case VHOST_USER_SET_VRING_ERR:
 858                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
 859                         close(msg.fds[0]);
 860                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
 861                 break;
 862
 863         case VHOST_USER_GET_QUEUE_NUM:
 864                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
 865                 msg.size = sizeof(msg.payload.u64);
 866                 send_vhost_message(fd, &msg);
 867                 break;
 868
 869         case VHOST_USER_SET_VRING_ENABLE:
 870                 vhost_user_set_vring_enable(dev, &msg.payload.state);
 871                 break;
 872         case VHOST_USER_SEND_RARP:
 873                 vhost_user_send_rarp(dev, &msg);
 874                 break;
 875
 876         default:
 877                 break;
 878
 879         }
 880
 881         return 0;
 882 }