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         return r;
  85 }
  86
  87 static int
  88 vhost_user_read(int fd, struct vhost_user_msg *msg)
  89 {
  90         uint32_t valid_flags = VHOST_USER_REPLY_MASK | VHOST_USER_VERSION;
  91         int ret, sz_hdr = VHOST_USER_HDR_SIZE, sz_payload;
  92
  93         ret = recv(fd, (void *)msg, sz_hdr, 0);
  94         if (ret < sz_hdr) {
  95                 PMD_DRV_LOG(ERR, "Failed to recv msg hdr: %d instead of %d.",
  96                             ret, sz_hdr);
  97                 goto fail;
  98         }
  99
 100         /* validate msg flags */
 101         if (msg->flags != (valid_flags)) {
 102                 PMD_DRV_LOG(ERR, "Failed to recv msg: flags %x instead of %x.",
 103                             msg->flags, valid_flags);
 104                 goto fail;
 105         }
 106
 107         sz_payload = msg->size;
 108
 109         if ((size_t)sz_payload > sizeof(msg->payload))
 110                 goto fail;
 111
 112         if (sz_payload) {
 113                 ret = recv(fd, (void *)((char *)msg + sz_hdr), sz_payload, 0);
 114                 if (ret < sz_payload) {
 115                         PMD_DRV_LOG(ERR,
 116                                 "Failed to recv msg payload: %d instead of %d.",
 117                                 ret, msg->size);
 118                         goto fail;
 119                 }
 120         }
 121
 122         return 0;
 123
 124 fail:
 125         return -1;
 126 }
 127
 128 static int
 129 vhost_user_set_owner(struct virtio_user_dev *dev)
 130 {
 131         int ret;
 132         struct vhost_user_msg msg = {
 133                 .request = VHOST_USER_SET_OWNER,
 134                 .flags = VHOST_USER_VERSION,
 135         };
 136
 137         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 138         if (ret < 0) {
 139                 PMD_DRV_LOG(ERR, "Failed to set owner");
 140                 return -1;
 141         }
 142
 143         return 0;
 144 }
 145
 146 static int
 147 vhost_user_get_features(struct virtio_user_dev *dev, uint64_t *features)
 148 {
 149         int ret;
 150         struct vhost_user_msg msg = {
 151                 .request = VHOST_USER_GET_FEATURES,
 152                 .flags = VHOST_USER_VERSION,
 153         };
 154
 155         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 156         if (ret < 0)
 157                 goto err;
 158
 159         ret = vhost_user_read(dev->vhostfd, &msg);
 160         if (ret < 0)
 161                 goto err;
 162
 163         if (msg.request != VHOST_USER_GET_FEATURES) {
 164                 PMD_DRV_LOG(ERR, "Unexpected request type (%d)", msg.request);
 165                 goto err;
 166         }
 167
 168         if (msg.size != sizeof(*features)) {
 169                 PMD_DRV_LOG(ERR, "Unexpected payload size (%u)", msg.size);
 170                 goto err;
 171         }
 172
 173         *features = msg.payload.u64;
 174
 175         return 0;
 176 err:
 177         PMD_DRV_LOG(ERR, "Failed to get backend features");
 178
 179         return -1;
 180 }
 181
 182 static int
 183 vhost_user_set_features(struct virtio_user_dev *dev, uint64_t features)
 184 {
 185         int ret;
 186         struct vhost_user_msg msg = {
 187                 .request = VHOST_USER_SET_FEATURES,
 188                 .flags = VHOST_USER_VERSION,
 189                 .size = sizeof(features),
 190                 .payload.u64 = features,
 191         };
 192
 193         msg.payload.u64 |= dev->device_features & (1ULL << VHOST_USER_F_PROTOCOL_FEATURES);
 194
 195         ret = vhost_user_write(dev->vhostfd, &msg, NULL, 0);
 196         if (ret < 0) {
 197                 PMD_DRV_LOG(ERR, "Failed to set features");
 198                 return -1;
 199         }
 200
 201         return 0;
 202 }
 203
 204 struct walk_arg {
 205         struct vhost_memory *vm;
 206         int *fds;
 207         int region_nr;
 208 };
 209
 210 static int
 211 update_memory_region(const struct rte_memseg_list *msl __rte_unused,
 212                 const struct rte_memseg *ms, void *arg)
 213 {
 214         struct walk_arg *wa = arg;
 215         struct vhost_memory_region *mr;
 216         uint64_t start_addr, end_addr;
 217         size_t offset;
 218         int i, fd;
 219
 220         fd = rte_memseg_get_fd_thread_unsafe(ms);
 221         if (fd < 0) {
 222                 PMD_DRV_LOG(ERR, "Failed to get fd, ms=%p rte_errno=%d",
 223                         ms, rte_errno);
 224                 return -1;
 225         }
 226
 227         if (rte_memseg_get_fd_offset_thread_unsafe(ms, &offset) < 0) {
 228                 PMD_DRV_LOG(ERR, "Failed to get offset, ms=%p rte_errno=%d",
 229                         ms, rte_errno);
 230                 return -1;
 231         }
 232
 233         start_addr = (uint64_t)(uintptr_t)ms->addr;
 234         end_addr = start_addr + ms->len;
 235
 236         for (i = 0; i < wa->region_nr; i++) {
 237                 if (wa->fds[i] != fd)
 238                         continue;
 239
 240                 mr = &wa->vm->regions[i];
 241
 242                 if (mr->userspace_addr + mr->memory_size < end_addr)
 243                         mr->memory_size = end_addr - mr->userspace_addr;
 244
 245                 if (mr->userspace_addr > start_addr) {
 246                         mr->userspace_addr = start_addr;
 247                         mr->guest_phys_addr = start_addr;
 248                 }
 249
 250                 if (mr->mmap_offset > offset)
 251                         mr->mmap_offset = offset;
 252
 253                 PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
 254                         " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
 255                         mr->mmap_offset, mr->userspace_addr,
 256                         mr->memory_size);
 257
 258                 return 0;
 259         }
 260
 261         if (i >= VHOST_MEMORY_MAX_NREGIONS) {
 262                 PMD_DRV_LOG(ERR, "Too many memory regions");
 263                 return -1;
 264         }
 265
 266         mr = &wa->vm->regions[i];
 267         wa->fds[i] = fd;
 268
 269         mr->guest_phys_addr = start_addr;
 270         mr->userspace_addr = start_addr;
 271         mr->memory_size = ms->len;
 272         mr->mmap_offset = offset;
 273
 274         PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
 275                 " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
 276                 mr->mmap_offset, mr->userspace_addr,
 277                 mr->memory_size);
 278
 279         wa->region_nr++;
 280
 281         return 0;
 282 }
 283
 284 static int
 285 prepare_vhost_memory_user(struct vhost_user_msg *msg, int fds[])
 286 {
 287         struct walk_arg wa;
 288
 289         wa.region_nr = 0;
 290         wa.vm = &msg->payload.memory;
 291         wa.fds = fds;
 292
 293         /*
 294          * The memory lock has already been taken by memory subsystem
 295          * or virtio_user_start_device().
 296          */
 297         if (rte_memseg_walk_thread_unsafe(update_memory_region, &wa) < 0)
 298                 return -1;
 299
 300         msg->payload.memory.nregions = wa.region_nr;
 301         msg->payload.memory.padding = 0;
 302
 303         return 0;
 304 }
 305
 306 static struct vhost_user_msg m;
 307
 308 const char * const vhost_msg_strings[] = {
 309         [VHOST_USER_RESET_OWNER] = "VHOST_RESET_OWNER",
 310         [VHOST_USER_SET_VRING_CALL] = "VHOST_SET_VRING_CALL",
 311         [VHOST_USER_SET_VRING_NUM] = "VHOST_SET_VRING_NUM",
 312         [VHOST_USER_SET_VRING_BASE] = "VHOST_SET_VRING_BASE",
 313         [VHOST_USER_GET_VRING_BASE] = "VHOST_GET_VRING_BASE",
 314         [VHOST_USER_SET_VRING_ADDR] = "VHOST_SET_VRING_ADDR",
 315         [VHOST_USER_SET_VRING_KICK] = "VHOST_SET_VRING_KICK",
 316         [VHOST_USER_SET_MEM_TABLE] = "VHOST_SET_MEM_TABLE",
 317         [VHOST_USER_SET_VRING_ENABLE] = "VHOST_SET_VRING_ENABLE",
 318         [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
 319         [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
 320         [VHOST_USER_SET_STATUS] = "VHOST_SET_STATUS",
 321         [VHOST_USER_GET_STATUS] = "VHOST_GET_STATUS",
 322 };
 323
 324 static int
 325 vhost_user_sock(struct virtio_user_dev *dev,
 326                 enum vhost_user_request req,
 327                 void *arg)
 328 {
 329         struct vhost_user_msg msg;
 330         struct vhost_vring_file *file = 0;
 331         int need_reply = 0;
 332         int has_reply_ack = 0;
 333         int fds[VHOST_MEMORY_MAX_NREGIONS];
 334         int fd_num = 0;
 335         int vhostfd = dev->vhostfd;
 336
 337         RTE_SET_USED(m);
 338
 339         PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
 340
 341         if (dev->is_server && vhostfd < 0)
 342                 return -1;
 343
 344         if (dev->protocol_features & (1ULL << VHOST_USER_PROTOCOL_F_REPLY_ACK))
 345                 has_reply_ack = 1;
 346
 347         msg.request = req;
 348         msg.flags = VHOST_USER_VERSION;
 349         msg.size = 0;
 350
 351         switch (req) {
 352         case VHOST_USER_GET_STATUS:
 353                 if (!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK) ||
 354                     (!(dev->protocol_features &
 355                                 (1ULL << VHOST_USER_PROTOCOL_F_STATUS))))
 356                         return -ENOTSUP;
 357                 /* Fallthrough */
 358         case VHOST_USER_GET_PROTOCOL_FEATURES:
 359                 need_reply = 1;
 360                 break;
 361
 362         case VHOST_USER_SET_STATUS:
 363                 if (!(dev->status & VIRTIO_CONFIG_STATUS_FEATURES_OK) ||
 364                     (!(dev->protocol_features &
 365                                 (1ULL << VHOST_USER_PROTOCOL_F_STATUS))))
 366                         return -ENOTSUP;
 367
 368                 if (has_reply_ack)
 369                         msg.flags |= VHOST_USER_NEED_REPLY_MASK;
 370                 /* Fallthrough */
 371         case VHOST_USER_SET_PROTOCOL_FEATURES:
 372         case VHOST_USER_SET_LOG_BASE:
 373                 msg.payload.u64 = *((__u64 *)arg);
 374                 msg.size = sizeof(m.payload.u64);
 375                 break;
 376
 377         case VHOST_USER_SET_FEATURES:
 378                 msg.payload.u64 = *((__u64 *)arg) | (dev->device_features &
 379                         (1ULL << VHOST_USER_F_PROTOCOL_FEATURES));
 380                 msg.size = sizeof(m.payload.u64);
 381                 break;
 382
 383         case VHOST_USER_RESET_OWNER:
 384                 break;
 385
 386         case VHOST_USER_SET_MEM_TABLE:
 387                 if (prepare_vhost_memory_user(&msg, fds) < 0)
 388                         return -1;
 389                 fd_num = msg.payload.memory.nregions;
 390                 msg.size = sizeof(m.payload.memory.nregions);
 391                 msg.size += sizeof(m.payload.memory.padding);
 392                 msg.size += fd_num * sizeof(struct vhost_memory_region);
 393
 394                 if (has_reply_ack)
 395                         msg.flags |= VHOST_USER_NEED_REPLY_MASK;
 396                 break;
 397
 398         case VHOST_USER_SET_LOG_FD:
 399                 fds[fd_num++] = *((int *)arg);
 400                 break;
 401
 402         case VHOST_USER_SET_VRING_NUM:
 403         case VHOST_USER_SET_VRING_BASE:
 404         case VHOST_USER_SET_VRING_ENABLE:
 405                 memcpy(&msg.payload.state, arg, sizeof(msg.payload.state));
 406                 msg.size = sizeof(m.payload.state);
 407                 break;
 408
 409         case VHOST_USER_GET_VRING_BASE:
 410                 memcpy(&msg.payload.state, arg, sizeof(msg.payload.state));
 411                 msg.size = sizeof(m.payload.state);
 412                 need_reply = 1;
 413                 break;
 414
 415         case VHOST_USER_SET_VRING_ADDR:
 416                 memcpy(&msg.payload.addr, arg, sizeof(msg.payload.addr));
 417                 msg.size = sizeof(m.payload.addr);
 418                 break;
 419
 420         case VHOST_USER_SET_VRING_KICK:
 421         case VHOST_USER_SET_VRING_CALL:
 422         case VHOST_USER_SET_VRING_ERR:
 423                 file = arg;
 424                 msg.payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
 425                 msg.size = sizeof(m.payload.u64);
 426                 if (file->fd > 0)
 427                         fds[fd_num++] = file->fd;
 428                 else
 429                         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
 430                 break;
 431
 432         default:
 433                 PMD_DRV_LOG(ERR, "trying to send unhandled msg type");
 434                 return -1;
 435         }
 436
 437         if (vhost_user_write(vhostfd, &msg, fds, fd_num) < 0) {
 438                 PMD_DRV_LOG(ERR, "%s failed: %s",
 439                             vhost_msg_strings[req], strerror(errno));
 440                 return -1;
 441         }
 442
 443         if (need_reply || msg.flags & VHOST_USER_NEED_REPLY_MASK) {
 444                 if (vhost_user_read(vhostfd, &msg) < 0) {
 445                         PMD_DRV_LOG(ERR, "Received msg failed: %s",
 446                                     strerror(errno));
 447                         return -1;
 448                 }
 449
 450                 if (req != msg.request) {
 451                         PMD_DRV_LOG(ERR, "Received unexpected msg type");
 452                         return -1;
 453                 }
 454
 455                 switch (req) {
 456                 case VHOST_USER_GET_STATUS:
 457                 case VHOST_USER_GET_PROTOCOL_FEATURES:
 458                         if (msg.size != sizeof(m.payload.u64)) {
 459                                 PMD_DRV_LOG(ERR, "Received bad msg size");
 460                                 return -1;
 461                         }
 462                         *((__u64 *)arg) = msg.payload.u64;
 463                         break;
 464                 case VHOST_USER_GET_VRING_BASE:
 465                         if (msg.size != sizeof(m.payload.state)) {
 466                                 PMD_DRV_LOG(ERR, "Received bad msg size");
 467                                 return -1;
 468                         }
 469                         memcpy(arg, &msg.payload.state,
 470                                sizeof(struct vhost_vring_state));
 471                         break;
 472                 default:
 473                         /* Reply-ack handling */
 474                         if (msg.size != sizeof(m.payload.u64)) {
 475                                 PMD_DRV_LOG(ERR, "Received bad msg size");
 476                                 return -1;
 477                         }
 478
 479                         if (msg.payload.u64 != 0) {
 480                                 PMD_DRV_LOG(ERR, "Slave replied NACK");
 481                                 return -1;
 482                         }
 483
 484                         break;
 485                 }
 486         }
 487
 488         return 0;
 489 }
 490
 491 #define MAX_VIRTIO_USER_BACKLOG 1
 492 static int
 493 virtio_user_start_server(struct virtio_user_dev *dev, struct sockaddr_un *un)
 494 {
 495         int ret;
 496         int flag;
 497         int fd = dev->listenfd;
 498
 499         ret = bind(fd, (struct sockaddr *)un, sizeof(*un));
 500         if (ret < 0) {
 501                 PMD_DRV_LOG(ERR, "failed to bind to %s: %s; remove it and try again\n",
 502                             dev->path, strerror(errno));
 503                 return -1;
 504         }
 505         ret = listen(fd, MAX_VIRTIO_USER_BACKLOG);
 506         if (ret < 0)
 507                 return -1;
 508
 509         flag = fcntl(fd, F_GETFL);
 510         if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) {
 511                 PMD_DRV_LOG(ERR, "fcntl failed, %s", strerror(errno));
 512                 return -1;
 513         }
 514
 515         return 0;
 516 }
 517
 518 /**
 519  * Set up environment to talk with a vhost user backend.
 520  *
 521  * @return
 522  *   - (-1) if fail;
 523  *   - (0) if succeed.
 524  */
 525 static int
 526 vhost_user_setup(struct virtio_user_dev *dev)
 527 {
 528         int fd;
 529         int flag;
 530         struct sockaddr_un un;
 531
 532         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 533         if (fd < 0) {
 534                 PMD_DRV_LOG(ERR, "socket() error, %s", strerror(errno));
 535                 return -1;
 536         }
 537
 538         flag = fcntl(fd, F_GETFD);
 539         if (fcntl(fd, F_SETFD, flag | FD_CLOEXEC) < 0)
 540                 PMD_DRV_LOG(WARNING, "fcntl failed, %s", strerror(errno));
 541
 542         memset(&un, 0, sizeof(un));
 543         un.sun_family = AF_UNIX;
 544         strlcpy(un.sun_path, dev->path, sizeof(un.sun_path));
 545
 546         if (dev->is_server) {
 547                 dev->listenfd = fd;
 548                 if (virtio_user_start_server(dev, &un) < 0) {
 549                         PMD_DRV_LOG(ERR, "virtio-user startup fails in server mode");
 550                         close(fd);
 551                         return -1;
 552                 }
 553                 dev->vhostfd = -1;
 554         } else {
 555                 if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
 556                         PMD_DRV_LOG(ERR, "connect error, %s", strerror(errno));
 557                         close(fd);
 558                         return -1;
 559                 }
 560                 dev->vhostfd = fd;
 561         }
 562
 563         return 0;
 564 }
 565
 566 static int
 567 vhost_user_enable_queue_pair(struct virtio_user_dev *dev,
 568                              uint16_t pair_idx,
 569                              int enable)
 570 {
 571         int i;
 572
 573         if (dev->qp_enabled[pair_idx] == enable)
 574                 return 0;
 575
 576         for (i = 0; i < 2; ++i) {
 577                 struct vhost_vring_state state = {
 578                         .index = pair_idx * 2 + i,
 579                         .num   = enable,
 580                 };
 581
 582                 if (vhost_user_sock(dev, VHOST_USER_SET_VRING_ENABLE, &state))
 583                         return -1;
 584         }
 585
 586         dev->qp_enabled[pair_idx] = enable;
 587         return 0;
 588 }
 589
 590 struct virtio_user_backend_ops virtio_ops_user = {
 591         .setup = vhost_user_setup,
 592         .set_owner = vhost_user_set_owner,
 593         .get_features = vhost_user_get_features,
 594         .set_features = vhost_user_set_features,
 595         .send_request = vhost_user_sock,
 596         .enable_qp = vhost_user_enable_queue_pair
 597 };