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