drivers/net/virtio/virtio_user/vhost_user.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <sys/socket.h>
   6 #include <sys/types.h>
   7 #include <sys/stat.h>
   8 #include <unistd.h>
   9 #include <fcntl.h>
  10 #include <sys/un.h>
  11 #include <string.h>
  12 #include <errno.h>
  13
  14 #include <rte_string_fns.h>
  15 #include <rte_fbarray.h>
  16
  17 #include "vhost.h"
  18 #include "virtio_user_dev.h"
  19
  20 /* The version of the protocol we support */
  21 #define VHOST_USER_VERSION    0x1
  22
  23 #define VHOST_MEMORY_MAX_NREGIONS 8
  24 struct vhost_memory {
  25         uint32_t nregions;
  26         uint32_t padding;
  27         struct vhost_memory_region regions[VHOST_MEMORY_MAX_NREGIONS];
  28 };
  29
  30 struct vhost_user_msg {
  31         enum vhost_user_request request;
  32
  33 #define VHOST_USER_VERSION_MASK     0x3
  34 #define VHOST_USER_REPLY_MASK       (0x1 << 2)
  35 #define VHOST_USER_NEED_REPLY_MASK  (0x1 << 3)
  36         uint32_t flags;
  37         uint32_t size; /* the following payload size */
  38         union {
  39 #define VHOST_USER_VRING_IDX_MASK   0xff
  40 #define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
  41                 uint64_t u64;
  42                 struct vhost_vring_state state;
  43                 struct vhost_vring_addr addr;
  44                 struct vhost_memory memory;
  45         } payload;
  46         int fds[VHOST_MEMORY_MAX_NREGIONS];
  47 } __rte_packed;
  48
  49 #define VHOST_USER_HDR_SIZE offsetof(struct vhost_user_msg, payload.u64)
  50 #define VHOST_USER_PAYLOAD_SIZE \
  51         (sizeof(struct vhost_user_msg) - VHOST_USER_HDR_SIZE)
  52
  53 static int
  54 vhost_user_write(int fd, struct vhost_user_msg *msg, int *fds, int fd_num)
  55 {
  56         int r;
  57         struct msghdr msgh;
  58         struct iovec iov;
  59         size_t fd_size = fd_num * sizeof(int);
  60         char control[CMSG_SPACE(fd_size)];
  61         struct cmsghdr *cmsg;
  62
  63         memset(&msgh, 0, sizeof(msgh));
  64         memset(control, 0, sizeof(control));
  65
  66         iov.iov_base = (uint8_t *)msg;
  67         iov.iov_len = VHOST_USER_HDR_SIZE + msg->size;
  68
  69         msgh.msg_iov = &iov;
  70         msgh.msg_iovlen = 1;
  71         msgh.msg_control = control;
  72         msgh.msg_controllen = sizeof(control);
  73
  74         cmsg = CMSG_FIRSTHDR(&msgh);
  75         cmsg->cmsg_len = CMSG_LEN(fd_size);
  76         cmsg->cmsg_level = SOL_SOCKET;
  77         cmsg->cmsg_type = SCM_RIGHTS;
  78         memcpy(CMSG_DATA(cmsg), fds, fd_size);
  79
  80         do {
  81                 r = sendmsg(fd, &msgh, 0);
  82         } while (r < 0 && errno == EINTR);
  83
  84         if (r < 0)
  85                 PMD_DRV_LOG(ERR, "Failed to send msg: %s", strerror(errno));
  86
  87         return r;
  88 }
  89
  90 static int
  91 vhost_user_read(int fd, struct vhost_user_msg *msg)
  92 {
  93         uint32_t valid_flags = VHOST_USER_REPLY_MASK | VHOST_USER_VERSION;
  94         int ret, sz_hdr = VHOST_USER_HDR_SIZE, sz_payload;
  95
  96         ret = recv(fd, (void *)msg, sz_hdr, 0);
  97         if (ret < sz_hdr) {
  98                 PMD_DRV_LOG(ERR, "Failed to recv msg hdr: %d instead of %d.",
  99                             ret, sz_hdr);
 100                 goto fail;
 101         }
 102
 103         /* validate msg flags */
 104         if (msg->flags != (valid_flags)) {
 105                 PMD_DRV_LOG(ERR, "Failed to recv msg: flags %x instead of %x.",
 106                             msg->flags, valid_flags);
 107                 goto fail;
 108         }
 109
 110         sz_payload = msg->size;
 111
 112         if ((size_t)sz_payload > sizeof(msg->payload))
 113                 goto fail;
 114
 115         if (sz_payload) {
 116                 ret = recv(fd, (void *)((char *)msg + sz_hdr), sz_payload, 0);
 117                 if (ret < sz_payload) {
 118                         PMD_DRV_LOG(ERR,
 119                                 "Failed to recv msg payload: %d instead of %d.",
 120                                 ret, msg->size);
 121                         goto fail;
 122                 }
 123         }
 124
 125         return 0;
 126
 127 fail:
 128         return -1;
 129 }
 130
 131 static int
 132 vhost_user_check_reply_ack(struct virtio_user_dev *dev, struct vhost_user_msg *msg)
 133 {
 134         enum vhost_user_request req = msg->request;
 135         int ret;
 136
 137         if (!(msg->flags & VHOST_USER_NEED_REPLY_MASK))
 138                 return 0;
 139
 140         ret = vhost_user_read(dev->vhostfd, msg);
 141         if (ret < 0) {
 142                 PMD_DRV_LOG(ERR, "Failed to read reply-ack");
 143                 return -1;
 144         }
 145
 146         if (req != msg->request) {
 147                 PMD_DRV_LOG(ERR, "Unexpected reply-ack request type (%d)", msg->request);
 148                 return -1;
 149         }
 150
 151         if (msg->size != sizeof(msg->payload.u64)) {
 152                 PMD_DRV_LOG(ERR, "Unexpected reply-ack payload size (%u)", msg->size);
 153                 return -1;
 154         }
 155
 156         if (msg->payload.u64) {
 157                 PMD_DRV_LOG(ERR, "Slave replied NACK to request type (%d)", msg->request);
 158                 return -1;
 159         }
 160
 161         return 0;
 162 }
 163
 164 static int
 165 vhost_user_set_owner(struct virtio_user_dev *dev)
 166 {
 167         int ret;
 168         struct vhost_user_msg msg = {
 169                 .request = VHOST_USER_SET_OWNER,
 170                 .flags = VHOST_USER_VERSION,
 171         };
 172
 173         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 174         if (ret < 0) {
 175                 PMD_DRV_LOG(ERR, "Failed to set owner");
 176                 return -1;
 177         }
 178
 179         return 0;
 180 }
 181
 182 static int
 183 vhost_user_get_features(struct virtio_user_dev *dev, uint64_t *features)
 184 {
 185         int ret;
 186         struct vhost_user_msg msg = {
 187                 .request = VHOST_USER_GET_FEATURES,
 188                 .flags = VHOST_USER_VERSION,
 189         };
 190
 191         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 192         if (ret < 0)
 193                 goto err;
 194
 195         ret = vhost_user_read(dev->vhostfd, &msg);
 196         if (ret < 0)
 197                 goto err;
 198
 199         if (msg.request != VHOST_USER_GET_FEATURES) {
 200                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
 201                 goto err;
 202         }
 203
 204         if (msg.size != sizeof(*features)) {
 205                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
 206                 goto err;
 207         }
 208
 209         *features = msg.payload.u64;
 210
 211         return 0;
 212 err:
 213         PMD_DRV_LOG(ERR, "Failed to get backend features");
 214
 215         return -1;
 216 }
 217
 218 static int
 219 vhost_user_set_features(struct virtio_user_dev *dev, uint64_t features)
 220 {
 221         int ret;
 222         struct vhost_user_msg msg = {
 223                 .request = VHOST_USER_SET_FEATURES,
 224                 .flags = VHOST_USER_VERSION,
 225                 .size = sizeof(features),
 226                 .payload.u64 = features,
 227         };
 228
 229         msg.payload.u64 |= dev->device_features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES);
 230
 231         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 232         if (ret < 0) {
 233                 PMD_DRV_LOG(ERR, "Failed to set features");
 234                 return -1;
 235         }
 236
 237         return 0;
 238 }
 239
 240 static int
 241 vhost_user_get_protocol_features(struct virtio_user_dev *dev, uint64_t *features)
 242 {
 243         int ret;
 244         struct vhost_user_msg msg = {
 245                 .request = VHOST_USER_GET_PROTOCOL_FEATURES,
 246                 .flags = VHOST_USER_VERSION,
 247         };
 248
 249         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 250         if (ret < 0)
 251                 goto err;
 252
 253         ret = vhost_user_read(dev->vhostfd, &msg);
 254         if (ret < 0)
 255                 goto err;
 256
 257         if (msg.request != VHOST_USER_GET_PROTOCOL_FEATURES) {
 258                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
 259                 goto err;
 260         }
 261
 262         if (msg.size != sizeof(*features)) {
 263                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
 264                 goto err;
 265         }
 266
 267         *features = msg.payload.u64;
 268
 269         return 0;
 270 err:
 271         PMD_DRV_LOG(ERR, "Failed to get backend protocol features");
 272
 273         return -1;
 274 }
 275
 276 static int
 277 vhost_user_set_protocol_features(struct virtio_user_dev *dev, uint64_t features)
 278 {
 279         int ret;
 280         struct vhost_user_msg msg = {
 281                 .request = VHOST_USER_SET_PROTOCOL_FEATURES,
 282                 .flags = VHOST_USER_VERSION,
 283                 .size = sizeof(features),
 284                 .payload.u64 = features,
 285         };
 286
 287         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 288         if (ret < 0) {
 289                 PMD_DRV_LOG(ERR, "Failed to set protocol features");
 290                 return -1;
 291         }
 292
 293         return 0;
 294 }
 295
 296 struct walk_arg {
 297         struct vhost_memory *vm;
 298         int *fds;
 299         int region_nr;
 300 };
 301
 302 static int
 303 update_memory_region(const struct rte_memseg_list *msl __rte_unused,
 304                 const struct rte_memseg *ms, void *arg)
 305 {
 306         struct walk_arg *wa = arg;
 307         struct vhost_memory_region *mr;
 308         uint64_t start_addr, end_addr;
 309         size_t offset;
 310         int i, fd;
 311
 312         fd = rte_memseg_get_fd_thread_unsafe(ms);
 313         if (fd < 0) {
 314                 PMD_DRV_LOG(ERR, "Failed to get fd, ms=%p rte_errno=%d",
 315                         ms, rte_errno);
 316                 return -1;
 317         }
 318
 319         if (rte_memseg_get_fd_offset_thread_unsafe(ms, &offset) < 0) {
 320                 PMD_DRV_LOG(ERR, "Failed to get offset, ms=%p rte_errno=%d",
 321                         ms, rte_errno);
 322                 return -1;
 323         }
 324
 325         start_addr = (uint64_t)(uintptr_t)ms->addr;
 326         end_addr = start_addr + ms->len;
 327
 328         for (i = 0; i < wa->region_nr; i++) {
 329                 if (wa->fds[i] != fd)
 330                         continue;
 331
 332                 mr = &wa->vm->regions[i];
 333
 334                 if (mr->userspace_addr + mr->memory_size < end_addr)
 335                         mr->memory_size = end_addr - mr->userspace_addr;
 336
 337                 if (mr->userspace_addr > start_addr) {
 338                         mr->userspace_addr = start_addr;
 339                         mr->guest_phys_addr = start_addr;
 340                 }
 341
 342                 if (mr->mmap_offset > offset)
 343                         mr->mmap_offset = offset;
 344
 345                 PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
 346                         " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
 347                         mr->mmap_offset, mr->userspace_addr,
 348                         mr->memory_size);
 349
 350                 return 0;
 351         }
 352
 353         if (i >= VHOST_MEMORY_MAX_NREGIONS) {
 354                 PMD_DRV_LOG(ERR, "Too many memory regions");
 355                 return -1;
 356         }
 357
 358         mr = &wa->vm->regions[i];
 359         wa->fds[i] = fd;
 360
 361         mr->guest_phys_addr = start_addr;
 362         mr->userspace_addr = start_addr;
 363         mr->memory_size = ms->len;
 364         mr->mmap_offset = offset;
 365
 366         PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
 367                 " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
 368                 mr->mmap_offset, mr->userspace_addr,
 369                 mr->memory_size);
 370
 371         wa->region_nr++;
 372
 373         return 0;
 374 }
 375
 376 static int
 377 vhost_user_set_memory_table(struct virtio_user_dev *dev)
 378 {
 379         struct walk_arg wa;
 380         int fds[VHOST_MEMORY_MAX_NREGIONS];
 381         int ret, fd_num;
 382         struct vhost_user_msg msg = {
 383                 .request = VHOST_USER_SET_MEM_TABLE,
 384                 .flags = VHOST_USER_VERSION,
 385         };
 386
 387         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK))
 388                 msg.flags |= VHOST_USER_NEED_REPLY_MASK;
 389
 390         wa.region_nr = 0;
 391         wa.vm = &msg.payload.memory;
 392         wa.fds = fds;
 393
 394         /*
 395          * The memory lock has already been taken by memory subsystem
 396          * or virtio_user_start_device().
 397          */
 398         ret = rte_memseg_walk_thread_unsafe(update_memory_region, &wa);
 399         if (ret < 0)
 400                 goto err;
 401
 402         fd_num = wa.region_nr;
 403         msg.payload.memory.nregions = wa.region_nr;
 404         msg.payload.memory.padding = 0;
 405
 406         msg.size = sizeof(msg.payload.memory.nregions);
 407         msg.size += sizeof(msg.payload.memory.padding);
 408         msg.size += fd_num * sizeof(struct vhost_memory_region);
 409
 410         ret = vhost_user_write(dev->vhostfd, &msg, fds, fd_num);
 411         if (ret < 0)
 412                 goto err;
 413
 414         return vhost_user_check_reply_ack(dev, &msg);
 415 err:
 416         PMD_DRV_LOG(ERR, "Failed to set memory table");
 417         return -1;
 418 }
 419
 420 static int
 421 vhost_user_set_vring(struct virtio_user_dev *dev, enum vhost_user_request req,
 422                 struct vhost_vring_state *state)
 423 {
 424         int ret;
 425         struct vhost_user_msg msg = {
 426                 .request = req,
 427                 .flags = VHOST_USER_VERSION,
 428                 .size = sizeof(*state),
 429                 .payload.state = *state,
 430         };
 431
 432         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 433         if (ret < 0) {
 434                 PMD_DRV_LOG(ERR, "Failed to set vring state (request %d)", req);
 435                 return -1;
 436         }
 437
 438         return 0;
 439 }
 440
 441 static int
 442 vhost_user_set_vring_enable(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 443 {
 444         return vhost_user_set_vring(dev, VHOST_USER_SET_VRING_ENABLE, state);
 445 }
 446
 447 static int
 448 vhost_user_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 449 {
 450         return vhost_user_set_vring(dev, VHOST_USER_SET_VRING_NUM, state);
 451 }
 452
 453 static int
 454 vhost_user_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 455 {
 456         return vhost_user_set_vring(dev, VHOST_USER_SET_VRING_BASE, state);
 457 }
 458
 459 static int
 460 vhost_user_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 461 {
 462         int ret;
 463         struct vhost_user_msg msg;
 464         unsigned int index = state->index;
 465
 466         ret = vhost_user_set_vring(dev, VHOST_USER_GET_VRING_BASE, state);
 467         if (ret < 0) {
 468                 PMD_DRV_LOG(ERR, "Failed to send request");
 469                 goto err;
 470         }
 471
 472         ret = vhost_user_read(dev->vhostfd, &msg);
 473         if (ret < 0) {
 474                 PMD_DRV_LOG(ERR, "Failed to read reply");
 475                 goto err;
 476         }
 477
 478         if (msg.request != VHOST_USER_GET_VRING_BASE) {
 479                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
 480                 goto err;
 481         }
 482
 483         if (msg.size != sizeof(*state)) {
 484                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
 485                 goto err;
 486         }
 487
 488         if (msg.payload.state.index != index) {
 489                 PMD_DRV_LOG(ERR, "Unexpected ring index (%u)", state->index);
 490                 goto err;
 491         }
 492
 493         *state = msg.payload.state;
 494
 495         return 0;
 496 err:
 497         PMD_DRV_LOG(ERR, "Failed to get vring base");
 498         return -1;
 499 }
 500
 501 static struct vhost_user_msg m;
 502
 503 const char * const vhost_msg_strings[] = {
 504         [VHOST_USER_RESET_OWNER] = "VHOST_RESET_OWNER",
 505         [VHOST_USER_SET_VRING_CALL] = "VHOST_SET_VRING_CALL",
 506         [VHOST_USER_SET_VRING_ADDR] = "VHOST_SET_VRING_ADDR",
 507         [VHOST_USER_SET_VRING_KICK] = "VHOST_SET_VRING_KICK",
 508         [VHOST_USER_SET_STATUS] = "VHOST_SET_STATUS",
 509         [VHOST_USER_GET_STATUS] = "VHOST_GET_STATUS",
 510 };
 511
 512 static int
 513 vhost_user_sock(struct virtio_user_dev *dev,
 514                 enum vhost_user_request req,
 515                 void *arg)
 516 {
 517         struct vhost_user_msg msg;
 518         struct vhost_vring_file *file = 0;
 519         int need_reply = 0;
 520         int has_reply_ack = 0;
 521         int fds[VHOST_MEMORY_MAX_NREGIONS];
 522         int fd_num = 0;
 523         int vhostfd = dev->vhostfd;
 524
 525         RTE_SET_USED(m);
 526
 527         PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
 528
 529         if (dev->is_server && vhostfd < 0)
 530                 return -1;
 531
 532         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK))
 533                 has_reply_ack = 1;
 534
 535         msg.request = req;
 536         msg.flags = VHOST_USER_VERSION;
 537         msg.size = 0;
 538
 539         switch (req) {
 540         case VHOST_USER_GET_STATUS:
 541                 if (!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK) ||
 542                     (!(dev->protocol_features &
 543                                 (1ULL << VHOST_USER_PROTOCOL_F_STATUS))))
 544                         return -ENOTSUP;
 545                 need_reply = 1;
 546                 break;
 547
 548         case VHOST_USER_SET_STATUS:
 549                 if (!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK) ||
 550                     (!(dev->protocol_features &
 551                                 (1ULL << VHOST_USER_PROTOCOL_F_STATUS))))
 552                         return -ENOTSUP;
 553
 554                 if (has_reply_ack)
 555                         msg.flags |= VHOST_USER_NEED_REPLY_MASK;
 556                 /* Fallthrough */
 557         case VHOST_USER_SET_LOG_BASE:
 558                 msg.payload.u64 = *((__u64 *)arg);
 559                 msg.size = sizeof(m.payload.u64);
 560                 break;
 561
 562         case VHOST_USER_SET_FEATURES:
 563                 msg.payload.u64 = *((__u64 *)arg) | (dev->device_features &
 564                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES));
 565                 msg.size = sizeof(m.payload.u64);
 566                 break;
 567
 568         case VHOST_USER_RESET_OWNER:
 569                 break;
 570
 571         case VHOST_USER_SET_LOG_FD:
 572                 fds[fd_num++] = *((int *)arg);
 573                 break;
 574
 575         case VHOST_USER_SET_VRING_ADDR:
 576                 memcpy(&msg.payload.addr, arg, sizeof(msg.payload.addr));
 577                 msg.size = sizeof(m.payload.addr);
 578                 break;
 579
 580         case VHOST_USER_SET_VRING_KICK:
 581         case VHOST_USER_SET_VRING_CALL:
 582         case VHOST_USER_SET_VRING_ERR:
 583                 file = arg;
 584                 msg.payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
 585                 msg.size = sizeof(m.payload.u64);
 586                 if (file->fd > 0)
 587                         fds[fd_num++] = file->fd;
 588                 else
 589                         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
 590                 break;
 591
 592         default:
 593                 PMD_DRV_LOG(ERR, "trying to send unhandled msg type");
 594                 return -1;
 595         }
 596
 597         if (vhost_user_write(vhostfd, &msg, fds, fd_num) < 0) {
 598                 PMD_DRV_LOG(ERR, "%s failed: %s",
 599                             vhost_msg_strings[req], strerror(errno));
 600                 return -1;
 601         }
 602
 603         if (need_reply || msg.flags & VHOST_USER_NEED_REPLY_MASK) {
 604                 if (vhost_user_read(vhostfd, &msg) < 0) {
 605                         PMD_DRV_LOG(ERR, "Received msg failed: %s",
 606                                     strerror(errno));
 607                         return -1;
 608                 }
 609
 610                 if (req != msg.request) {
 611                         PMD_DRV_LOG(ERR, "Received unexpected msg type");
 612                         return -1;
 613                 }
 614
 615                 switch (req) {
 616                 case VHOST_USER_GET_STATUS:
 617                         if (msg.size != sizeof(m.payload.u64)) {
 618                                 PMD_DRV_LOG(ERR, "Received bad msg size");
 619                                 return -1;
 620                         }
 621                         *((__u64 *)arg) = msg.payload.u64;
 622                         break;
 623                 default:
 624                         /* Reply-ack handling */
 625                         if (msg.size != sizeof(m.payload.u64)) {
 626                                 PMD_DRV_LOG(ERR, "Received bad msg size");
 627                                 return -1;
 628                         }
 629
 630                         if (msg.payload.u64 != 0) {
 631                                 PMD_DRV_LOG(ERR, "Slave replied NACK");
 632                                 return -1;
 633                         }
 634
 635                         break;
 636                 }
 637         }
 638
 639         return 0;
 640 }
 641
 642 #define MAX_VIRTIO_USER_BACKLOG 1
 643 static int
 644 virtio_user_start_server(struct virtio_user_dev *dev, struct sockaddr_un *un)
 645 {
 646         int ret;
 647         int flag;
 648         int fd = dev->listenfd;
 649
 650         ret = bind(fd, (struct sockaddr *)un, sizeof(*un));
 651         if (ret < 0) {
 652                 PMD_DRV_LOG(ERR, "failed to bind to %s: %s; remove it and try again\n",
 653                             dev->path, strerror(errno));
 654                 return -1;
 655         }
 656         ret = listen(fd, MAX_VIRTIO_USER_BACKLOG);
 657         if (ret < 0)
 658                 return -1;
 659
 660         flag = fcntl(fd, F_GETFL);
 661         if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) {
 662                 PMD_DRV_LOG(ERR, "fcntl failed, %s", strerror(errno));
 663                 return -1;
 664         }
 665
 666         return 0;
 667 }
 668
 669 /**
 670  * Set up environment to talk with a vhost user backend.
 671  *
 672  * @return
 673  *   - (-1) if fail;
 674  *   - (0) if succeed.
 675  */
 676 static int
 677 vhost_user_setup(struct virtio_user_dev *dev)
 678 {
 679         int fd;
 680         int flag;
 681         struct sockaddr_un un;
 682
 683         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 684         if (fd < 0) {
 685                 PMD_DRV_LOG(ERR, "socket() error, %s", strerror(errno));
 686                 return -1;
 687         }
 688
 689         flag = fcntl(fd, F_GETFD);
 690         if (fcntl(fd, F_SETFD, flag | FD_CLOEXEC) < 0)
 691                 PMD_DRV_LOG(WARNING, "fcntl failed, %s", strerror(errno));
 692
 693         memset(&un, 0, sizeof(un));
 694         un.sun_family = AF_UNIX;
 695         strlcpy(un.sun_path, dev->path, sizeof(un.sun_path));
 696
 697         if (dev->is_server) {
 698                 dev->listenfd = fd;
 699                 if (virtio_user_start_server(dev, &un) < 0) {
 700                         PMD_DRV_LOG(ERR, "virtio-user startup fails in server mode");
 701                         close(fd);
 702                         return -1;
 703                 }
 704                 dev->vhostfd = -1;
 705         } else {
 706                 if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
 707                         PMD_DRV_LOG(ERR, "connect error, %s", strerror(errno));
 708                         close(fd);
 709                         return -1;
 710                 }
 711                 dev->vhostfd = fd;
 712         }
 713
 714         return 0;
 715 }
 716
 717 static int
 718 vhost_user_enable_queue_pair(struct virtio_user_dev *dev,
 719                              uint16_t pair_idx,
 720                              int enable)
 721 {
 722         int i;
 723
 724         if (dev->qp_enabled[pair_idx] == enable)
 725                 return 0;
 726
 727         for (i = 0; i < 2; ++i) {
 728                 struct vhost_vring_state state = {
 729                         .index = pair_idx * 2 + i,
 730                         .num = enable,
 731                 };
 732
 733                 if (vhost_user_set_vring_enable(dev, &state))
 734                         return -1;
 735         }
 736
 737         dev->qp_enabled[pair_idx] = enable;
 738         return 0;
 739 }
 740
 741 struct virtio_user_backend_ops virtio_ops_user = {
 742         .setup = vhost_user_setup,
 743         .set_owner = vhost_user_set_owner,
 744         .get_features = vhost_user_get_features,
 745         .set_features = vhost_user_set_features,
 746         .get_protocol_features = vhost_user_get_protocol_features,
 747         .set_protocol_features = vhost_user_set_protocol_features,
 748         .set_memory_table = vhost_user_set_memory_table,
 749         .set_vring_num = vhost_user_set_vring_num,
 750         .set_vring_base = vhost_user_set_vring_base,
 751         .get_vring_base = vhost_user_get_vring_base,
 752         .send_request = vhost_user_sock,
 753         .enable_qp = vhost_user_enable_queue_pair
 754 };