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