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 "iotlb.h"
  52 #include "vhost.h"
  53 #include "vhost_user.h"
  54
  55 #define VIRTIO_MIN_MTU 68
  56 #define VIRTIO_MAX_MTU 65535
  57
  58 static const char *vhost_message_str[VHOST_USER_MAX] = {
  59         [VHOST_USER_NONE] = "VHOST_USER_NONE",
  60         [VHOST_USER_GET_FEATURES] = "VHOST_USER_GET_FEATURES",
  61         [VHOST_USER_SET_FEATURES] = "VHOST_USER_SET_FEATURES",
  62         [VHOST_USER_SET_OWNER] = "VHOST_USER_SET_OWNER",
  63         [VHOST_USER_RESET_OWNER] = "VHOST_USER_RESET_OWNER",
  64         [VHOST_USER_SET_MEM_TABLE] = "VHOST_USER_SET_MEM_TABLE",
  65         [VHOST_USER_SET_LOG_BASE] = "VHOST_USER_SET_LOG_BASE",
  66         [VHOST_USER_SET_LOG_FD] = "VHOST_USER_SET_LOG_FD",
  67         [VHOST_USER_SET_VRING_NUM] = "VHOST_USER_SET_VRING_NUM",
  68         [VHOST_USER_SET_VRING_ADDR] = "VHOST_USER_SET_VRING_ADDR",
  69         [VHOST_USER_SET_VRING_BASE] = "VHOST_USER_SET_VRING_BASE",
  70         [VHOST_USER_GET_VRING_BASE] = "VHOST_USER_GET_VRING_BASE",
  71         [VHOST_USER_SET_VRING_KICK] = "VHOST_USER_SET_VRING_KICK",
  72         [VHOST_USER_SET_VRING_CALL] = "VHOST_USER_SET_VRING_CALL",
  73         [VHOST_USER_SET_VRING_ERR]  = "VHOST_USER_SET_VRING_ERR",
  74         [VHOST_USER_GET_PROTOCOL_FEATURES]  = "VHOST_USER_GET_PROTOCOL_FEATURES",
  75         [VHOST_USER_SET_PROTOCOL_FEATURES]  = "VHOST_USER_SET_PROTOCOL_FEATURES",
  76         [VHOST_USER_GET_QUEUE_NUM]  = "VHOST_USER_GET_QUEUE_NUM",
  77         [VHOST_USER_SET_VRING_ENABLE]  = "VHOST_USER_SET_VRING_ENABLE",
  78         [VHOST_USER_SEND_RARP]  = "VHOST_USER_SEND_RARP",
  79         [VHOST_USER_NET_SET_MTU]  = "VHOST_USER_NET_SET_MTU",
  80         [VHOST_USER_SET_SLAVE_REQ_FD]  = "VHOST_USER_SET_SLAVE_REQ_FD",
  81         [VHOST_USER_IOTLB_MSG]  = "VHOST_USER_IOTLB_MSG",
  82 };
  83
  84 static uint64_t
  85 get_blk_size(int fd)
  86 {
  87         struct stat stat;
  88         int ret;
  89
  90         ret = fstat(fd, &stat);
  91         return ret == -1 ? (uint64_t)-1 : (uint64_t)stat.st_blksize;
  92 }
  93
  94 static void
  95 free_mem_region(struct virtio_net *dev)
  96 {
  97         uint32_t i;
  98         struct rte_vhost_mem_region *reg;
  99
 100         if (!dev || !dev->mem)
 101                 return;
 102
 103         for (i = 0; i < dev->mem->nregions; i++) {
 104                 reg = &dev->mem->regions[i];
 105                 if (reg->host_user_addr) {
 106                         munmap(reg->mmap_addr, reg->mmap_size);
 107                         close(reg->fd);
 108                 }
 109         }
 110 }
 111
 112 void
 113 vhost_backend_cleanup(struct virtio_net *dev)
 114 {
 115         if (dev->mem) {
 116                 free_mem_region(dev);
 117                 rte_free(dev->mem);
 118                 dev->mem = NULL;
 119         }
 120
 121         free(dev->guest_pages);
 122         dev->guest_pages = NULL;
 123
 124         if (dev->log_addr) {
 125                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 126                 dev->log_addr = 0;
 127         }
 128
 129         if (dev->slave_req_fd >= 0) {
 130                 close(dev->slave_req_fd);
 131                 dev->slave_req_fd = -1;
 132         }
 133 }
 134
 135 /*
 136  * This function just returns success at the moment unless
 137  * the device hasn't been initialised.
 138  */
 139 static int
 140 vhost_user_set_owner(void)
 141 {
 142         return 0;
 143 }
 144
 145 static int
 146 vhost_user_reset_owner(struct virtio_net *dev)
 147 {
 148         if (dev->flags & VIRTIO_DEV_RUNNING) {
 149                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 150                 dev->notify_ops->destroy_device(dev->vid);
 151         }
 152
 153         cleanup_device(dev, 0);
 154         reset_device(dev);
 155         return 0;
 156 }
 157
 158 /*
 159  * The features that we support are requested.
 160  */
 161 static uint64_t
 162 vhost_user_get_features(struct virtio_net *dev)
 163 {
 164         uint64_t features = 0;
 165
 166         rte_vhost_driver_get_features(dev->ifname, &features);
 167         return features;
 168 }
 169
 170 /*
 171  * We receive the negotiated features supported by us and the virtio device.
 172  */
 173 static int
 174 vhost_user_set_features(struct virtio_net *dev, uint64_t features)
 175 {
 176         uint64_t vhost_features = 0;
 177
 178         rte_vhost_driver_get_features(dev->ifname, &vhost_features);
 179         if (features & ~vhost_features) {
 180                 RTE_LOG(ERR, VHOST_CONFIG,
 181                         "(%d) received invalid negotiated features.\n",
 182                         dev->vid);
 183                 return -1;
 184         }
 185
 186         if ((dev->flags & VIRTIO_DEV_RUNNING) && dev->features != features) {
 187                 if (dev->notify_ops->features_changed)
 188                         dev->notify_ops->features_changed(dev->vid, features);
 189         }
 190
 191         dev->features = features;
 192         if (dev->features &
 193                 ((1 << VIRTIO_NET_F_MRG_RXBUF) | (1ULL << VIRTIO_F_VERSION_1))) {
 194                 dev->vhost_hlen = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 195         } else {
 196                 dev->vhost_hlen = sizeof(struct virtio_net_hdr);
 197         }
 198         LOG_DEBUG(VHOST_CONFIG,
 199                 "(%d) mergeable RX buffers %s, virtio 1 %s\n",
 200                 dev->vid,
 201                 (dev->features & (1 << VIRTIO_NET_F_MRG_RXBUF)) ? "on" : "off",
 202                 (dev->features & (1ULL << VIRTIO_F_VERSION_1)) ? "on" : "off");
 203
 204         return 0;
 205 }
 206
 207 /*
 208  * The virtio device sends us the size of the descriptor ring.
 209  */
 210 static int
 211 vhost_user_set_vring_num(struct virtio_net *dev,
 212                          VhostUserMsg *msg)
 213 {
 214         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 215
 216         vq->size = msg->payload.state.num;
 217
 218         if (dev->dequeue_zero_copy) {
 219                 vq->nr_zmbuf = 0;
 220                 vq->last_zmbuf_idx = 0;
 221                 vq->zmbuf_size = vq->size;
 222                 vq->zmbufs = rte_zmalloc(NULL, vq->zmbuf_size *
 223                                          sizeof(struct zcopy_mbuf), 0);
 224                 if (vq->zmbufs == NULL) {
 225                         RTE_LOG(WARNING, VHOST_CONFIG,
 226                                 "failed to allocate mem for zero copy; "
 227                                 "zero copy is force disabled\n");
 228                         dev->dequeue_zero_copy = 0;
 229                 }
 230         }
 231
 232         vq->shadow_used_ring = rte_malloc(NULL,
 233                                 vq->size * sizeof(struct vring_used_elem),
 234                                 RTE_CACHE_LINE_SIZE);
 235         if (!vq->shadow_used_ring) {
 236                 RTE_LOG(ERR, VHOST_CONFIG,
 237                         "failed to allocate memory for shadow used ring.\n");
 238                 return -1;
 239         }
 240
 241         vq->batch_copy_elems = rte_malloc(NULL,
 242                                 vq->size * sizeof(struct batch_copy_elem),
 243                                 RTE_CACHE_LINE_SIZE);
 244         if (!vq->batch_copy_elems) {
 245                 RTE_LOG(ERR, VHOST_CONFIG,
 246                         "failed to allocate memory for batching copy.\n");
 247                 return -1;
 248         }
 249
 250         return 0;
 251 }
 252
 253 /*
 254  * Reallocate virtio_dev and vhost_virtqueue data structure to make them on the
 255  * same numa node as the memory of vring descriptor.
 256  */
 257 #ifdef RTE_LIBRTE_VHOST_NUMA
 258 static struct virtio_net*
 259 numa_realloc(struct virtio_net *dev, int index)
 260 {
 261         int oldnode, newnode;
 262         struct virtio_net *old_dev;
 263         struct vhost_virtqueue *old_vq, *vq;
 264         int ret;
 265
 266         old_dev = dev;
 267         vq = old_vq = dev->virtqueue[index];
 268
 269         ret = get_mempolicy(&newnode, NULL, 0, old_vq->desc,
 270                             MPOL_F_NODE | MPOL_F_ADDR);
 271
 272         /* check if we need to reallocate vq */
 273         ret |= get_mempolicy(&oldnode, NULL, 0, old_vq,
 274                              MPOL_F_NODE | MPOL_F_ADDR);
 275         if (ret) {
 276                 RTE_LOG(ERR, VHOST_CONFIG,
 277                         "Unable to get vq numa information.\n");
 278                 return dev;
 279         }
 280         if (oldnode != newnode) {
 281                 RTE_LOG(INFO, VHOST_CONFIG,
 282                         "reallocate vq from %d to %d node\n", oldnode, newnode);
 283                 vq = rte_malloc_socket(NULL, sizeof(*vq), 0, newnode);
 284                 if (!vq)
 285                         return dev;
 286
 287                 memcpy(vq, old_vq, sizeof(*vq));
 288                 rte_free(old_vq);
 289         }
 290
 291         /* check if we need to reallocate dev */
 292         ret = get_mempolicy(&oldnode, NULL, 0, old_dev,
 293                             MPOL_F_NODE | MPOL_F_ADDR);
 294         if (ret) {
 295                 RTE_LOG(ERR, VHOST_CONFIG,
 296                         "Unable to get dev numa information.\n");
 297                 goto out;
 298         }
 299         if (oldnode != newnode) {
 300                 RTE_LOG(INFO, VHOST_CONFIG,
 301                         "reallocate dev from %d to %d node\n",
 302                         oldnode, newnode);
 303                 dev = rte_malloc_socket(NULL, sizeof(*dev), 0, newnode);
 304                 if (!dev) {
 305                         dev = old_dev;
 306                         goto out;
 307                 }
 308
 309                 memcpy(dev, old_dev, sizeof(*dev));
 310                 rte_free(old_dev);
 311         }
 312
 313 out:
 314         dev->virtqueue[index] = vq;
 315         vhost_devices[dev->vid] = dev;
 316
 317         return dev;
 318 }
 319 #else
 320 static struct virtio_net*
 321 numa_realloc(struct virtio_net *dev, int index __rte_unused)
 322 {
 323         return dev;
 324 }
 325 #endif
 326
 327 /*
 328  * Converts QEMU virtual address to Vhost virtual address. This function is
 329  * used to convert the ring addresses to our address space.
 330  */
 331 static uint64_t
 332 qva_to_vva(struct virtio_net *dev, uint64_t qva)
 333 {
 334         struct rte_vhost_mem_region *reg;
 335         uint32_t i;
 336
 337         /* Find the region where the address lives. */
 338         for (i = 0; i < dev->mem->nregions; i++) {
 339                 reg = &dev->mem->regions[i];
 340
 341                 if (qva >= reg->guest_user_addr &&
 342                     qva <  reg->guest_user_addr + reg->size) {
 343                         return qva - reg->guest_user_addr +
 344                                reg->host_user_addr;
 345                 }
 346         }
 347
 348         return 0;
 349 }
 350
 351 /*
 352  * The virtio device sends us the desc, used and avail ring addresses.
 353  * This function then converts these to our address space.
 354  */
 355 static int
 356 vhost_user_set_vring_addr(struct virtio_net *dev, VhostUserMsg *msg)
 357 {
 358         struct vhost_virtqueue *vq;
 359
 360         if (dev->mem == NULL)
 361                 return -1;
 362
 363         /* addr->index refers to the queue index. The txq 1, rxq is 0. */
 364         vq = dev->virtqueue[msg->payload.addr.index];
 365
 366         /* The addresses are converted from QEMU virtual to Vhost virtual. */
 367         vq->desc = (struct vring_desc *)(uintptr_t)qva_to_vva(dev,
 368                         msg->payload.addr.desc_user_addr);
 369         if (vq->desc == 0) {
 370                 RTE_LOG(ERR, VHOST_CONFIG,
 371                         "(%d) failed to find desc ring address.\n",
 372                         dev->vid);
 373                 return -1;
 374         }
 375
 376         dev = numa_realloc(dev, msg->payload.addr.index);
 377         vq = dev->virtqueue[msg->payload.addr.index];
 378
 379         vq->avail = (struct vring_avail *)(uintptr_t)qva_to_vva(dev,
 380                         msg->payload.addr.avail_user_addr);
 381         if (vq->avail == 0) {
 382                 RTE_LOG(ERR, VHOST_CONFIG,
 383                         "(%d) failed to find avail ring address.\n",
 384                         dev->vid);
 385                 return -1;
 386         }
 387
 388         vq->used = (struct vring_used *)(uintptr_t)qva_to_vva(dev,
 389                         msg->payload.addr.used_user_addr);
 390         if (vq->used == 0) {
 391                 RTE_LOG(ERR, VHOST_CONFIG,
 392                         "(%d) failed to find used ring address.\n",
 393                         dev->vid);
 394                 return -1;
 395         }
 396
 397         if (vq->last_used_idx != vq->used->idx) {
 398                 RTE_LOG(WARNING, VHOST_CONFIG,
 399                         "last_used_idx (%u) and vq->used->idx (%u) mismatches; "
 400                         "some packets maybe resent for Tx and dropped for Rx\n",
 401                         vq->last_used_idx, vq->used->idx);
 402                 vq->last_used_idx  = vq->used->idx;
 403                 vq->last_avail_idx = vq->used->idx;
 404         }
 405
 406         vq->log_guest_addr = msg->payload.addr.log_guest_addr;
 407
 408         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address desc: %p\n",
 409                         dev->vid, vq->desc);
 410         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address avail: %p\n",
 411                         dev->vid, vq->avail);
 412         LOG_DEBUG(VHOST_CONFIG, "(%d) mapped address used: %p\n",
 413                         dev->vid, vq->used);
 414         LOG_DEBUG(VHOST_CONFIG, "(%d) log_guest_addr: %" PRIx64 "\n",
 415                         dev->vid, vq->log_guest_addr);
 416
 417         return 0;
 418 }
 419
 420 /*
 421  * The virtio device sends us the available ring last used index.
 422  */
 423 static int
 424 vhost_user_set_vring_base(struct virtio_net *dev,
 425                           VhostUserMsg *msg)
 426 {
 427         dev->virtqueue[msg->payload.state.index]->last_used_idx  =
 428                         msg->payload.state.num;
 429         dev->virtqueue[msg->payload.state.index]->last_avail_idx =
 430                         msg->payload.state.num;
 431
 432         return 0;
 433 }
 434
 435 static void
 436 add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
 437                    uint64_t host_phys_addr, uint64_t size)
 438 {
 439         struct guest_page *page, *last_page;
 440
 441         if (dev->nr_guest_pages == dev->max_guest_pages) {
 442                 dev->max_guest_pages *= 2;
 443                 dev->guest_pages = realloc(dev->guest_pages,
 444                                         dev->max_guest_pages * sizeof(*page));
 445         }
 446
 447         if (dev->nr_guest_pages > 0) {
 448                 last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
 449                 /* merge if the two pages are continuous */
 450                 if (host_phys_addr == last_page->host_phys_addr +
 451                                       last_page->size) {
 452                         last_page->size += size;
 453                         return;
 454                 }
 455         }
 456
 457         page = &dev->guest_pages[dev->nr_guest_pages++];
 458         page->guest_phys_addr = guest_phys_addr;
 459         page->host_phys_addr  = host_phys_addr;
 460         page->size = size;
 461 }
 462
 463 static void
 464 add_guest_pages(struct virtio_net *dev, struct rte_vhost_mem_region *reg,
 465                 uint64_t page_size)
 466 {
 467         uint64_t reg_size = reg->size;
 468         uint64_t host_user_addr  = reg->host_user_addr;
 469         uint64_t guest_phys_addr = reg->guest_phys_addr;
 470         uint64_t host_phys_addr;
 471         uint64_t size;
 472
 473         host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
 474         size = page_size - (guest_phys_addr & (page_size - 1));
 475         size = RTE_MIN(size, reg_size);
 476
 477         add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 478         host_user_addr  += size;
 479         guest_phys_addr += size;
 480         reg_size -= size;
 481
 482         while (reg_size > 0) {
 483                 size = RTE_MIN(reg_size, page_size);
 484                 host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
 485                                                   host_user_addr);
 486                 add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
 487
 488                 host_user_addr  += size;
 489                 guest_phys_addr += size;
 490                 reg_size -= size;
 491         }
 492 }
 493
 494 #ifdef RTE_LIBRTE_VHOST_DEBUG
 495 /* TODO: enable it only in debug mode? */
 496 static void
 497 dump_guest_pages(struct virtio_net *dev)
 498 {
 499         uint32_t i;
 500         struct guest_page *page;
 501
 502         for (i = 0; i < dev->nr_guest_pages; i++) {
 503                 page = &dev->guest_pages[i];
 504
 505                 RTE_LOG(INFO, VHOST_CONFIG,
 506                         "guest physical page region %u\n"
 507                         "\t guest_phys_addr: %" PRIx64 "\n"
 508                         "\t host_phys_addr : %" PRIx64 "\n"
 509                         "\t size           : %" PRIx64 "\n",
 510                         i,
 511                         page->guest_phys_addr,
 512                         page->host_phys_addr,
 513                         page->size);
 514         }
 515 }
 516 #else
 517 #define dump_guest_pages(dev)
 518 #endif
 519
 520 static int
 521 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 522 {
 523         struct VhostUserMemory memory = pmsg->payload.memory;
 524         struct rte_vhost_mem_region *reg;
 525         void *mmap_addr;
 526         uint64_t mmap_size;
 527         uint64_t mmap_offset;
 528         uint64_t alignment;
 529         uint32_t i;
 530         int fd;
 531
 532         if (dev->mem) {
 533                 free_mem_region(dev);
 534                 rte_free(dev->mem);
 535                 dev->mem = NULL;
 536         }
 537
 538         dev->nr_guest_pages = 0;
 539         if (!dev->guest_pages) {
 540                 dev->max_guest_pages = 8;
 541                 dev->guest_pages = malloc(dev->max_guest_pages *
 542                                                 sizeof(struct guest_page));
 543                 if (dev->guest_pages == NULL) {
 544                         RTE_LOG(ERR, VHOST_CONFIG,
 545                                 "(%d) failed to allocate memory "
 546                                 "for dev->guest_pages\n",
 547                                 dev->vid);
 548                         return -1;
 549                 }
 550         }
 551
 552         dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct rte_vhost_memory) +
 553                 sizeof(struct rte_vhost_mem_region) * memory.nregions, 0);
 554         if (dev->mem == NULL) {
 555                 RTE_LOG(ERR, VHOST_CONFIG,
 556                         "(%d) failed to allocate memory for dev->mem\n",
 557                         dev->vid);
 558                 return -1;
 559         }
 560         dev->mem->nregions = memory.nregions;
 561
 562         for (i = 0; i < memory.nregions; i++) {
 563                 fd  = pmsg->fds[i];
 564                 reg = &dev->mem->regions[i];
 565
 566                 reg->guest_phys_addr = memory.regions[i].guest_phys_addr;
 567                 reg->guest_user_addr = memory.regions[i].userspace_addr;
 568                 reg->size            = memory.regions[i].memory_size;
 569                 reg->fd              = fd;
 570
 571                 mmap_offset = memory.regions[i].mmap_offset;
 572                 mmap_size   = reg->size + mmap_offset;
 573
 574                 /* mmap() without flag of MAP_ANONYMOUS, should be called
 575                  * with length argument aligned with hugepagesz at older
 576                  * longterm version Linux, like 2.6.32 and 3.2.72, or
 577                  * mmap() will fail with EINVAL.
 578                  *
 579                  * to avoid failure, make sure in caller to keep length
 580                  * aligned.
 581                  */
 582                 alignment = get_blk_size(fd);
 583                 if (alignment == (uint64_t)-1) {
 584                         RTE_LOG(ERR, VHOST_CONFIG,
 585                                 "couldn't get hugepage size through fstat\n");
 586                         goto err_mmap;
 587                 }
 588                 mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 589
 590                 mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
 591                                  MAP_SHARED | MAP_POPULATE, fd, 0);
 592
 593                 if (mmap_addr == MAP_FAILED) {
 594                         RTE_LOG(ERR, VHOST_CONFIG,
 595                                 "mmap region %u failed.\n", i);
 596                         goto err_mmap;
 597                 }
 598
 599                 reg->mmap_addr = mmap_addr;
 600                 reg->mmap_size = mmap_size;
 601                 reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
 602                                       mmap_offset;
 603
 604                 if (dev->dequeue_zero_copy)
 605                         add_guest_pages(dev, reg, alignment);
 606
 607                 RTE_LOG(INFO, VHOST_CONFIG,
 608                         "guest memory region %u, size: 0x%" PRIx64 "\n"
 609                         "\t guest physical addr: 0x%" PRIx64 "\n"
 610                         "\t guest virtual  addr: 0x%" PRIx64 "\n"
 611                         "\t host  virtual  addr: 0x%" PRIx64 "\n"
 612                         "\t mmap addr : 0x%" PRIx64 "\n"
 613                         "\t mmap size : 0x%" PRIx64 "\n"
 614                         "\t mmap align: 0x%" PRIx64 "\n"
 615                         "\t mmap off  : 0x%" PRIx64 "\n",
 616                         i, reg->size,
 617                         reg->guest_phys_addr,
 618                         reg->guest_user_addr,
 619                         reg->host_user_addr,
 620                         (uint64_t)(uintptr_t)mmap_addr,
 621                         mmap_size,
 622                         alignment,
 623                         mmap_offset);
 624         }
 625
 626         dump_guest_pages(dev);
 627
 628         return 0;
 629
 630 err_mmap:
 631         free_mem_region(dev);
 632         rte_free(dev->mem);
 633         dev->mem = NULL;
 634         return -1;
 635 }
 636
 637 static int
 638 vq_is_ready(struct vhost_virtqueue *vq)
 639 {
 640         return vq && vq->desc   &&
 641                vq->kickfd != VIRTIO_UNINITIALIZED_EVENTFD &&
 642                vq->callfd != VIRTIO_UNINITIALIZED_EVENTFD;
 643 }
 644
 645 static int
 646 virtio_is_ready(struct virtio_net *dev)
 647 {
 648         struct vhost_virtqueue *vq;
 649         uint32_t i;
 650
 651         if (dev->nr_vring == 0)
 652                 return 0;
 653
 654         for (i = 0; i < dev->nr_vring; i++) {
 655                 vq = dev->virtqueue[i];
 656
 657                 if (!vq_is_ready(vq))
 658                         return 0;
 659         }
 660
 661         RTE_LOG(INFO, VHOST_CONFIG,
 662                 "virtio is now ready for processing.\n");
 663         return 1;
 664 }
 665
 666 static void
 667 vhost_user_set_vring_call(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 668 {
 669         struct vhost_vring_file file;
 670         struct vhost_virtqueue *vq;
 671
 672         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 673         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 674                 file.fd = VIRTIO_INVALID_EVENTFD;
 675         else
 676                 file.fd = pmsg->fds[0];
 677         RTE_LOG(INFO, VHOST_CONFIG,
 678                 "vring call idx:%d file:%d\n", file.index, file.fd);
 679
 680         vq = dev->virtqueue[file.index];
 681         if (vq->callfd >= 0)
 682                 close(vq->callfd);
 683
 684         vq->callfd = file.fd;
 685 }
 686
 687 static void
 688 vhost_user_set_vring_kick(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 689 {
 690         struct vhost_vring_file file;
 691         struct vhost_virtqueue *vq;
 692
 693         file.index = pmsg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
 694         if (pmsg->payload.u64 & VHOST_USER_VRING_NOFD_MASK)
 695                 file.fd = VIRTIO_INVALID_EVENTFD;
 696         else
 697                 file.fd = pmsg->fds[0];
 698         RTE_LOG(INFO, VHOST_CONFIG,
 699                 "vring kick idx:%d file:%d\n", file.index, file.fd);
 700
 701         vq = dev->virtqueue[file.index];
 702         if (vq->kickfd >= 0)
 703                 close(vq->kickfd);
 704         vq->kickfd = file.fd;
 705 }
 706
 707 static void
 708 free_zmbufs(struct vhost_virtqueue *vq)
 709 {
 710         struct zcopy_mbuf *zmbuf, *next;
 711
 712         for (zmbuf = TAILQ_FIRST(&vq->zmbuf_list);
 713              zmbuf != NULL; zmbuf = next) {
 714                 next = TAILQ_NEXT(zmbuf, next);
 715
 716                 rte_pktmbuf_free(zmbuf->mbuf);
 717                 TAILQ_REMOVE(&vq->zmbuf_list, zmbuf, next);
 718         }
 719
 720         rte_free(vq->zmbufs);
 721 }
 722
 723 /*
 724  * when virtio is stopped, qemu will send us the GET_VRING_BASE message.
 725  */
 726 static int
 727 vhost_user_get_vring_base(struct virtio_net *dev,
 728                           VhostUserMsg *msg)
 729 {
 730         struct vhost_virtqueue *vq = dev->virtqueue[msg->payload.state.index];
 731
 732         /* We have to stop the queue (virtio) if it is running. */
 733         if (dev->flags & VIRTIO_DEV_RUNNING) {
 734                 dev->flags &= ~VIRTIO_DEV_RUNNING;
 735                 dev->notify_ops->destroy_device(dev->vid);
 736         }
 737
 738         dev->flags &= ~VIRTIO_DEV_READY;
 739
 740         /* Here we are safe to get the last used index */
 741         msg->payload.state.num = vq->last_used_idx;
 742
 743         RTE_LOG(INFO, VHOST_CONFIG,
 744                 "vring base idx:%d file:%d\n", msg->payload.state.index,
 745                 msg->payload.state.num);
 746         /*
 747          * Based on current qemu vhost-user implementation, this message is
 748          * sent and only sent in vhost_vring_stop.
 749          * TODO: cleanup the vring, it isn't usable since here.
 750          */
 751         if (vq->kickfd >= 0)
 752                 close(vq->kickfd);
 753
 754         vq->kickfd = VIRTIO_UNINITIALIZED_EVENTFD;
 755
 756         if (dev->dequeue_zero_copy)
 757                 free_zmbufs(vq);
 758         rte_free(vq->shadow_used_ring);
 759         vq->shadow_used_ring = NULL;
 760
 761         rte_free(vq->batch_copy_elems);
 762         vq->batch_copy_elems = NULL;
 763
 764         return 0;
 765 }
 766
 767 /*
 768  * when virtio queues are ready to work, qemu will send us to
 769  * enable the virtio queue pair.
 770  */
 771 static int
 772 vhost_user_set_vring_enable(struct virtio_net *dev,
 773                             VhostUserMsg *msg)
 774 {
 775         int enable = (int)msg->payload.state.num;
 776
 777         RTE_LOG(INFO, VHOST_CONFIG,
 778                 "set queue enable: %d to qp idx: %d\n",
 779                 enable, msg->payload.state.index);
 780
 781         if (dev->notify_ops->vring_state_changed)
 782                 dev->notify_ops->vring_state_changed(dev->vid,
 783                                 msg->payload.state.index, enable);
 784
 785         dev->virtqueue[msg->payload.state.index]->enabled = enable;
 786
 787         return 0;
 788 }
 789
 790 static void
 791 vhost_user_set_protocol_features(struct virtio_net *dev,
 792                                  uint64_t protocol_features)
 793 {
 794         if (protocol_features & ~VHOST_USER_PROTOCOL_FEATURES)
 795                 return;
 796
 797         dev->protocol_features = protocol_features;
 798 }
 799
 800 static int
 801 vhost_user_set_log_base(struct virtio_net *dev, struct VhostUserMsg *msg)
 802 {
 803         int fd = msg->fds[0];
 804         uint64_t size, off;
 805         void *addr;
 806
 807         if (fd < 0) {
 808                 RTE_LOG(ERR, VHOST_CONFIG, "invalid log fd: %d\n", fd);
 809                 return -1;
 810         }
 811
 812         if (msg->size != sizeof(VhostUserLog)) {
 813                 RTE_LOG(ERR, VHOST_CONFIG,
 814                         "invalid log base msg size: %"PRId32" != %d\n",
 815                         msg->size, (int)sizeof(VhostUserLog));
 816                 return -1;
 817         }
 818
 819         size = msg->payload.log.mmap_size;
 820         off  = msg->payload.log.mmap_offset;
 821         RTE_LOG(INFO, VHOST_CONFIG,
 822                 "log mmap size: %"PRId64", offset: %"PRId64"\n",
 823                 size, off);
 824
 825         /*
 826          * mmap from 0 to workaround a hugepage mmap bug: mmap will
 827          * fail when offset is not page size aligned.
 828          */
 829         addr = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
 830         close(fd);
 831         if (addr == MAP_FAILED) {
 832                 RTE_LOG(ERR, VHOST_CONFIG, "mmap log base failed!\n");
 833                 return -1;
 834         }
 835
 836         /*
 837          * Free previously mapped log memory on occasionally
 838          * multiple VHOST_USER_SET_LOG_BASE.
 839          */
 840         if (dev->log_addr) {
 841                 munmap((void *)(uintptr_t)dev->log_addr, dev->log_size);
 842         }
 843         dev->log_addr = (uint64_t)(uintptr_t)addr;
 844         dev->log_base = dev->log_addr + off;
 845         dev->log_size = size;
 846
 847         return 0;
 848 }
 849
 850 /*
 851  * An rarp packet is constructed and broadcasted to notify switches about
 852  * the new location of the migrated VM, so that packets from outside will
 853  * not be lost after migration.
 854  *
 855  * However, we don't actually "send" a rarp packet here, instead, we set
 856  * a flag 'broadcast_rarp' to let rte_vhost_dequeue_burst() inject it.
 857  */
 858 static int
 859 vhost_user_send_rarp(struct virtio_net *dev, struct VhostUserMsg *msg)
 860 {
 861         uint8_t *mac = (uint8_t *)&msg->payload.u64;
 862
 863         RTE_LOG(DEBUG, VHOST_CONFIG,
 864                 ":: mac: %02x:%02x:%02x:%02x:%02x:%02x\n",
 865                 mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
 866         memcpy(dev->mac.addr_bytes, mac, 6);
 867
 868         /*
 869          * Set the flag to inject a RARP broadcast packet at
 870          * rte_vhost_dequeue_burst().
 871          *
 872          * rte_smp_wmb() is for making sure the mac is copied
 873          * before the flag is set.
 874          */
 875         rte_smp_wmb();
 876         rte_atomic16_set(&dev->broadcast_rarp, 1);
 877
 878         return 0;
 879 }
 880
 881 static int
 882 vhost_user_net_set_mtu(struct virtio_net *dev, struct VhostUserMsg *msg)
 883 {
 884         if (msg->payload.u64 < VIRTIO_MIN_MTU ||
 885                         msg->payload.u64 > VIRTIO_MAX_MTU) {
 886                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid MTU size (%"PRIu64")\n",
 887                                 msg->payload.u64);
 888
 889                 return -1;
 890         }
 891
 892         dev->mtu = msg->payload.u64;
 893
 894         return 0;
 895 }
 896
 897 static int
 898 vhost_user_set_req_fd(struct virtio_net *dev, struct VhostUserMsg *msg)
 899 {
 900         int fd = msg->fds[0];
 901
 902         if (fd < 0) {
 903                 RTE_LOG(ERR, VHOST_CONFIG,
 904                                 "Invalid file descriptor for slave channel (%d)\n",
 905                                 fd);
 906                 return -1;
 907         }
 908
 909         dev->slave_req_fd = fd;
 910
 911         return 0;
 912 }
 913
 914 static int
 915 vhost_user_iotlb_msg(struct virtio_net *dev, struct VhostUserMsg *msg)
 916 {
 917         struct vhost_iotlb_msg *imsg = &msg->payload.iotlb;
 918         uint16_t i;
 919         uint64_t vva;
 920
 921         switch (imsg->type) {
 922         case VHOST_IOTLB_UPDATE:
 923                 vva = qva_to_vva(dev, imsg->uaddr);
 924                 if (!vva)
 925                         return -1;
 926
 927                 for (i = 0; i < dev->nr_vring; i++) {
 928                         struct vhost_virtqueue *vq = dev->virtqueue[i];
 929
 930                         vhost_user_iotlb_cache_insert(vq, imsg->iova, vva,
 931                                         imsg->size, imsg->perm);
 932                 }
 933                 break;
 934         case VHOST_IOTLB_INVALIDATE:
 935                 for (i = 0; i < dev->nr_vring; i++) {
 936                         struct vhost_virtqueue *vq = dev->virtqueue[i];
 937
 938                         vhost_user_iotlb_cache_remove(vq, imsg->iova,
 939                                         imsg->size);
 940                 }
 941                 break;
 942         default:
 943                 RTE_LOG(ERR, VHOST_CONFIG, "Invalid IOTLB message type (%d)\n",
 944                                 imsg->type);
 945                 return -1;
 946         }
 947
 948         return 0;
 949 }
 950
 951 /* return bytes# of read on success or negative val on failure. */
 952 static int
 953 read_vhost_message(int sockfd, struct VhostUserMsg *msg)
 954 {
 955         int ret;
 956
 957         ret = read_fd_message(sockfd, (char *)msg, VHOST_USER_HDR_SIZE,
 958                 msg->fds, VHOST_MEMORY_MAX_NREGIONS);
 959         if (ret <= 0)
 960                 return ret;
 961
 962         if (msg && msg->size) {
 963                 if (msg->size > sizeof(msg->payload)) {
 964                         RTE_LOG(ERR, VHOST_CONFIG,
 965                                 "invalid msg size: %d\n", msg->size);
 966                         return -1;
 967                 }
 968                 ret = read(sockfd, &msg->payload, msg->size);
 969                 if (ret <= 0)
 970                         return ret;
 971                 if (ret != (int)msg->size) {
 972                         RTE_LOG(ERR, VHOST_CONFIG,
 973                                 "read control message failed\n");
 974                         return -1;
 975                 }
 976         }
 977
 978         return ret;
 979 }
 980
 981 static int
 982 send_vhost_message(int sockfd, struct VhostUserMsg *msg)
 983 {
 984         if (!msg)
 985                 return 0;
 986
 987         return send_fd_message(sockfd, (char *)msg,
 988                 VHOST_USER_HDR_SIZE + msg->size, NULL, 0);
 989 }
 990
 991 static int
 992 send_vhost_reply(int sockfd, struct VhostUserMsg *msg)
 993 {
 994         if (!msg)
 995                 return 0;
 996
 997         msg->flags &= ~VHOST_USER_VERSION_MASK;
 998         msg->flags &= ~VHOST_USER_NEED_REPLY;
 999         msg->flags |= VHOST_USER_VERSION;
1000         msg->flags |= VHOST_USER_REPLY_MASK;
1001
1002         return send_vhost_message(sockfd, msg);
1003 }
1004
1005 /*
1006  * Allocate a queue pair if it hasn't been allocated yet
1007  */
1008 static int
1009 vhost_user_check_and_alloc_queue_pair(struct virtio_net *dev, VhostUserMsg *msg)
1010 {
1011         uint16_t vring_idx;
1012
1013         switch (msg->request) {
1014         case VHOST_USER_SET_VRING_KICK:
1015         case VHOST_USER_SET_VRING_CALL:
1016         case VHOST_USER_SET_VRING_ERR:
1017                 vring_idx = msg->payload.u64 & VHOST_USER_VRING_IDX_MASK;
1018                 break;
1019         case VHOST_USER_SET_VRING_NUM:
1020         case VHOST_USER_SET_VRING_BASE:
1021         case VHOST_USER_SET_VRING_ENABLE:
1022                 vring_idx = msg->payload.state.index;
1023                 break;
1024         case VHOST_USER_SET_VRING_ADDR:
1025                 vring_idx = msg->payload.addr.index;
1026                 break;
1027         default:
1028                 return 0;
1029         }
1030
1031         if (vring_idx >= VHOST_MAX_VRING) {
1032                 RTE_LOG(ERR, VHOST_CONFIG,
1033                         "invalid vring index: %u\n", vring_idx);
1034                 return -1;
1035         }
1036
1037         if (dev->virtqueue[vring_idx])
1038                 return 0;
1039
1040         return alloc_vring_queue(dev, vring_idx);
1041 }
1042
1043 int
1044 vhost_user_msg_handler(int vid, int fd)
1045 {
1046         struct virtio_net *dev;
1047         struct VhostUserMsg msg;
1048         int ret;
1049
1050         dev = get_device(vid);
1051         if (dev == NULL)
1052                 return -1;
1053
1054         if (!dev->notify_ops) {
1055                 dev->notify_ops = vhost_driver_callback_get(dev->ifname);
1056                 if (!dev->notify_ops) {
1057                         RTE_LOG(ERR, VHOST_CONFIG,
1058                                 "failed to get callback ops for driver %s\n",
1059                                 dev->ifname);
1060                         return -1;
1061                 }
1062         }
1063
1064         ret = read_vhost_message(fd, &msg);
1065         if (ret <= 0 || msg.request >= VHOST_USER_MAX) {
1066                 if (ret < 0)
1067                         RTE_LOG(ERR, VHOST_CONFIG,
1068                                 "vhost read message failed\n");
1069                 else if (ret == 0)
1070                         RTE_LOG(INFO, VHOST_CONFIG,
1071                                 "vhost peer closed\n");
1072                 else
1073                         RTE_LOG(ERR, VHOST_CONFIG,
1074                                 "vhost read incorrect message\n");
1075
1076                 return -1;
1077         }
1078
1079         ret = 0;
1080         RTE_LOG(INFO, VHOST_CONFIG, "read message %s\n",
1081                 vhost_message_str[msg.request]);
1082
1083         ret = vhost_user_check_and_alloc_queue_pair(dev, &msg);
1084         if (ret < 0) {
1085                 RTE_LOG(ERR, VHOST_CONFIG,
1086                         "failed to alloc queue\n");
1087                 return -1;
1088         }
1089
1090         switch (msg.request) {
1091         case VHOST_USER_GET_FEATURES:
1092                 msg.payload.u64 = vhost_user_get_features(dev);
1093                 msg.size = sizeof(msg.payload.u64);
1094                 send_vhost_reply(fd, &msg);
1095                 break;
1096         case VHOST_USER_SET_FEATURES:
1097                 vhost_user_set_features(dev, msg.payload.u64);
1098                 break;
1099
1100         case VHOST_USER_GET_PROTOCOL_FEATURES:
1101                 msg.payload.u64 = VHOST_USER_PROTOCOL_FEATURES;
1102                 msg.size = sizeof(msg.payload.u64);
1103                 send_vhost_reply(fd, &msg);
1104                 break;
1105         case VHOST_USER_SET_PROTOCOL_FEATURES:
1106                 vhost_user_set_protocol_features(dev, msg.payload.u64);
1107                 break;
1108
1109         case VHOST_USER_SET_OWNER:
1110                 vhost_user_set_owner();
1111                 break;
1112         case VHOST_USER_RESET_OWNER:
1113                 vhost_user_reset_owner(dev);
1114                 break;
1115
1116         case VHOST_USER_SET_MEM_TABLE:
1117                 ret = vhost_user_set_mem_table(dev, &msg);
1118                 break;
1119
1120         case VHOST_USER_SET_LOG_BASE:
1121                 vhost_user_set_log_base(dev, &msg);
1122
1123                 /* it needs a reply */
1124                 msg.size = sizeof(msg.payload.u64);
1125                 send_vhost_reply(fd, &msg);
1126                 break;
1127         case VHOST_USER_SET_LOG_FD:
1128                 close(msg.fds[0]);
1129                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented.\n");
1130                 break;
1131
1132         case VHOST_USER_SET_VRING_NUM:
1133                 vhost_user_set_vring_num(dev, &msg);
1134                 break;
1135         case VHOST_USER_SET_VRING_ADDR:
1136                 vhost_user_set_vring_addr(dev, &msg);
1137                 break;
1138         case VHOST_USER_SET_VRING_BASE:
1139                 vhost_user_set_vring_base(dev, &msg);
1140                 break;
1141
1142         case VHOST_USER_GET_VRING_BASE:
1143                 vhost_user_get_vring_base(dev, &msg);
1144                 msg.size = sizeof(msg.payload.state);
1145                 send_vhost_reply(fd, &msg);
1146                 break;
1147
1148         case VHOST_USER_SET_VRING_KICK:
1149                 vhost_user_set_vring_kick(dev, &msg);
1150                 break;
1151         case VHOST_USER_SET_VRING_CALL:
1152                 vhost_user_set_vring_call(dev, &msg);
1153                 break;
1154
1155         case VHOST_USER_SET_VRING_ERR:
1156                 if (!(msg.payload.u64 & VHOST_USER_VRING_NOFD_MASK))
1157                         close(msg.fds[0]);
1158                 RTE_LOG(INFO, VHOST_CONFIG, "not implemented\n");
1159                 break;
1160
1161         case VHOST_USER_GET_QUEUE_NUM:
1162                 msg.payload.u64 = VHOST_MAX_QUEUE_PAIRS;
1163                 msg.size = sizeof(msg.payload.u64);
1164                 send_vhost_reply(fd, &msg);
1165                 break;
1166
1167         case VHOST_USER_SET_VRING_ENABLE:
1168                 vhost_user_set_vring_enable(dev, &msg);
1169                 break;
1170         case VHOST_USER_SEND_RARP:
1171                 vhost_user_send_rarp(dev, &msg);
1172                 break;
1173
1174         case VHOST_USER_NET_SET_MTU:
1175                 ret = vhost_user_net_set_mtu(dev, &msg);
1176                 break;
1177
1178         case VHOST_USER_SET_SLAVE_REQ_FD:
1179                 ret = vhost_user_set_req_fd(dev, &msg);
1180                 break;
1181
1182         case VHOST_USER_IOTLB_MSG:
1183                 ret = vhost_user_iotlb_msg(dev, &msg);
1184                 break;
1185
1186         default:
1187                 ret = -1;
1188                 break;
1189
1190         }
1191
1192         if (msg.flags & VHOST_USER_NEED_REPLY) {
1193                 msg.payload.u64 = !!ret;
1194                 msg.size = sizeof(msg.payload.u64);
1195                 send_vhost_reply(fd, &msg);
1196         }
1197
1198         if (!(dev->flags & VIRTIO_DEV_RUNNING) && virtio_is_ready(dev)) {
1199                 dev->flags |= VIRTIO_DEV_READY;
1200
1201                 if (!(dev->flags & VIRTIO_DEV_RUNNING)) {
1202                         if (dev->dequeue_zero_copy) {
1203                                 RTE_LOG(INFO, VHOST_CONFIG,
1204                                                 "dequeue zero copy is enabled\n");
1205                         }
1206
1207                         if (dev->notify_ops->new_device(dev->vid) == 0)
1208                                 dev->flags |= VIRTIO_DEV_RUNNING;
1209                 }
1210         }
1211
1212         return 0;
1213 }
1214
1215 int
1216 vhost_user_iotlb_miss(struct virtio_net *dev, uint64_t iova, uint8_t perm)
1217 {
1218         int ret;
1219         struct VhostUserMsg msg = {
1220                 .request = (enum VhostUserRequest)VHOST_USER_SLAVE_IOTLB_MSG,
1221                 .flags = VHOST_USER_VERSION,
1222                 .size = sizeof(msg.payload.iotlb),
1223                 .payload.iotlb = {
1224                         .iova = iova,
1225                         .perm = perm,
1226                         .type = VHOST_IOTLB_MISS,
1227                 },
1228         };
1229
1230         ret = send_vhost_message(dev->slave_req_fd, &msg);
1231         if (ret < 0) {
1232                 RTE_LOG(ERR, VHOST_CONFIG,
1233                                 "Failed to send IOTLB miss message (%d)\n",
1234                                 ret);
1235                 return ret;
1236         }
1237
1238         return 0;
1239 }