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         uint32_t flags;
  36         uint32_t size; /* the following payload size */
  37         union {
  38 #define VHOST_USER_VRING_IDX_MASK   0xff
  39 #define VHOST_USER_VRING_NOFD_MASK  (0x1 << 8)
  40                 uint64_t u64;
  41                 struct vhost_vring_state state;
  42                 struct vhost_vring_addr addr;
  43                 struct vhost_memory memory;
  44         } payload;
  45         int fds[VHOST_MEMORY_MAX_NREGIONS];
  46 } __rte_packed;
  47
  48 #define VHOST_USER_HDR_SIZE offsetof(struct vhost_user_msg, payload.u64)
  49 #define VHOST_USER_PAYLOAD_SIZE \
  50         (sizeof(struct vhost_user_msg) - VHOST_USER_HDR_SIZE)
  51
  52 static int
  53 vhost_user_write(int fd, void *buf, int len, int *fds, int fd_num)
  54 {
  55         int r;
  56         struct msghdr msgh;
  57         struct iovec iov;
  58         size_t fd_size = fd_num * sizeof(int);
  59         char control[CMSG_SPACE(fd_size)];
  60         struct cmsghdr *cmsg;
  61
  62         memset(&msgh, 0, sizeof(msgh));
  63         memset(control, 0, sizeof(control));
  64
  65         iov.iov_base = (uint8_t *)buf;
  66         iov.iov_len = len;
  67
  68         msgh.msg_iov = &iov;
  69         msgh.msg_iovlen = 1;
  70         msgh.msg_control = control;
  71         msgh.msg_controllen = sizeof(control);
  72
  73         cmsg = CMSG_FIRSTHDR(&msgh);
  74         cmsg->cmsg_len = CMSG_LEN(fd_size);
  75         cmsg->cmsg_level = SOL_SOCKET;
  76         cmsg->cmsg_type = SCM_RIGHTS;
  77         memcpy(CMSG_DATA(cmsg), fds, fd_size);
  78
  79         do {
  80                 r = sendmsg(fd, &msgh, 0);
  81         } while (r < 0 && errno == EINTR);
  82
  83         return r;
  84 }
  85
  86 static int
  87 vhost_user_read(int fd, struct vhost_user_msg *msg)
  88 {
  89         uint32_t valid_flags = VHOST_USER_REPLY_MASK | VHOST_USER_VERSION;
  90         int ret, sz_hdr = VHOST_USER_HDR_SIZE, sz_payload;
  91
  92         ret = recv(fd, (void *)msg, sz_hdr, 0);
  93         if (ret < sz_hdr) {
  94                 PMD_DRV_LOG(ERR, "Failed to recv msg hdr: %d instead of %d.",
  95                             ret, sz_hdr);
  96                 goto fail;
  97         }
  98
  99         /* validate msg flags */
 100         if (msg->flags != (valid_flags)) {
 101                 PMD_DRV_LOG(ERR, "Failed to recv msg: flags %x instead of %x.",
 102                             msg->flags, valid_flags);
 103                 goto fail;
 104         }
 105
 106         sz_payload = msg->size;
 107
 108         if ((size_t)sz_payload > sizeof(msg->payload))
 109                 goto fail;
 110
 111         if (sz_payload) {
 112                 ret = recv(fd, (void *)((char *)msg + sz_hdr), sz_payload, 0);
 113                 if (ret < sz_payload) {
 114                         PMD_DRV_LOG(ERR,
 115                                 "Failed to recv msg payload: %d instead of %d.",
 116                                 ret, msg->size);
 117                         goto fail;
 118                 }
 119         }
 120
 121         return 0;
 122
 123 fail:
 124         return -1;
 125 }
 126
 127 struct walk_arg {
 128         struct vhost_memory *vm;
 129         int *fds;
 130         int region_nr;
 131 };
 132
 133 static int
 134 update_memory_region(const struct rte_memseg_list *msl __rte_unused,
 135                 const struct rte_memseg *ms, void *arg)
 136 {
 137         struct walk_arg *wa = arg;
 138         struct vhost_memory_region *mr;
 139         uint64_t start_addr, end_addr;
 140         size_t offset;
 141         int i, fd;
 142
 143         fd = rte_memseg_get_fd_thread_unsafe(ms);
 144         if (fd < 0) {
 145                 PMD_DRV_LOG(ERR, "Failed to get fd, ms=%p rte_errno=%d",
 146                         ms, rte_errno);
 147                 return -1;
 148         }
 149
 150         if (rte_memseg_get_fd_offset_thread_unsafe(ms, &offset) < 0) {
 151                 PMD_DRV_LOG(ERR, "Failed to get offset, ms=%p rte_errno=%d",
 152                         ms, rte_errno);
 153                 return -1;
 154         }
 155
 156         start_addr = (uint64_t)(uintptr_t)ms->addr;
 157         end_addr = start_addr + ms->len;
 158
 159         for (i = 0; i < wa->region_nr; i++) {
 160                 if (wa->fds[i] != fd)
 161                         continue;
 162
 163                 mr = &wa->vm->regions[i];
 164
 165                 if (mr->userspace_addr + mr->memory_size < end_addr)
 166                         mr->memory_size = end_addr - mr->userspace_addr;
 167
 168                 if (mr->userspace_addr > start_addr) {
 169                         mr->userspace_addr = start_addr;
 170                         mr->guest_phys_addr = start_addr;
 171                 }
 172
 173                 if (mr->mmap_offset > offset)
 174                         mr->mmap_offset = offset;
 175
 176                 PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
 177                         " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
 178                         mr->mmap_offset, mr->userspace_addr,
 179                         mr->memory_size);
 180
 181                 return 0;
 182         }
 183
 184         if (i >= VHOST_MEMORY_MAX_NREGIONS) {
 185                 PMD_DRV_LOG(ERR, "Too many memory regions");
 186                 return -1;
 187         }
 188
 189         mr = &wa->vm->regions[i];
 190         wa->fds[i] = fd;
 191
 192         mr->guest_phys_addr = start_addr;
 193         mr->userspace_addr = start_addr;
 194         mr->memory_size = ms->len;
 195         mr->mmap_offset = offset;
 196
 197         PMD_DRV_LOG(DEBUG, "index=%d fd=%d offset=0x%" PRIx64
 198                 " addr=0x%" PRIx64 " len=%" PRIu64, i, fd,
 199                 mr->mmap_offset, mr->userspace_addr,
 200                 mr->memory_size);
 201
 202         wa->region_nr++;
 203
 204         return 0;
 205 }
 206
 207 static int
 208 prepare_vhost_memory_user(struct vhost_user_msg *msg, int fds[])
 209 {
 210         struct walk_arg wa;
 211
 212         wa.region_nr = 0;
 213         wa.vm = &msg->payload.memory;
 214         wa.fds = fds;
 215
 216         /*
 217          * The memory lock has already been taken by memory subsystem
 218          * or virtio_user_start_device().
 219          */
 220         if (rte_memseg_walk_thread_unsafe(update_memory_region, &wa) < 0)
 221                 return -1;
 222
 223         msg->payload.memory.nregions = wa.region_nr;
 224         msg->payload.memory.padding = 0;
 225
 226         return 0;
 227 }
 228
 229 static struct vhost_user_msg m;
 230
 231 const char * const vhost_msg_strings[] = {
 232         [VHOST_USER_SET_OWNER] = "VHOST_SET_OWNER",
 233         [VHOST_USER_RESET_OWNER] = "VHOST_RESET_OWNER",
 234         [VHOST_USER_SET_FEATURES] = "VHOST_SET_FEATURES",
 235         [VHOST_USER_GET_FEATURES] = "VHOST_GET_FEATURES",
 236         [VHOST_USER_SET_VRING_CALL] = "VHOST_SET_VRING_CALL",
 237         [VHOST_USER_SET_VRING_NUM] = "VHOST_SET_VRING_NUM",
 238         [VHOST_USER_SET_VRING_BASE] = "VHOST_SET_VRING_BASE",
 239         [VHOST_USER_GET_VRING_BASE] = "VHOST_GET_VRING_BASE",
 240         [VHOST_USER_SET_VRING_ADDR] = "VHOST_SET_VRING_ADDR",
 241         [VHOST_USER_SET_VRING_KICK] = "VHOST_SET_VRING_KICK",
 242         [VHOST_USER_SET_MEM_TABLE] = "VHOST_SET_MEM_TABLE",
 243         [VHOST_USER_SET_VRING_ENABLE] = "VHOST_SET_VRING_ENABLE",
 244         [VHOST_USER_GET_PROTOCOL_FEATURES] = "VHOST_USER_GET_PROTOCOL_FEATURES",
 245         [VHOST_USER_SET_PROTOCOL_FEATURES] = "VHOST_USER_SET_PROTOCOL_FEATURES",
 246 };
 247
 248 static int
 249 vhost_user_sock(struct virtio_user_dev *dev,
 250                 enum vhost_user_request req,
 251                 void *arg)
 252 {
 253         struct vhost_user_msg msg;
 254         struct vhost_vring_file *file = 0;
 255         int need_reply = 0;
 256         int fds[VHOST_MEMORY_MAX_NREGIONS];
 257         int fd_num = 0;
 258         int len;
 259         int vhostfd = dev->vhostfd;
 260
 261         RTE_SET_USED(m);
 262
 263         PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
 264
 265         if (dev->is_server && vhostfd < 0)
 266                 return -1;
 267
 268         msg.request = req;
 269         msg.flags = VHOST_USER_VERSION;
 270         msg.size = 0;
 271
 272         switch (req) {
 273         case VHOST_USER_GET_FEATURES:
 274         case VHOST_USER_GET_PROTOCOL_FEATURES:
 275                 need_reply = 1;
 276                 break;
 277
 278         case VHOST_USER_SET_FEATURES:
 279         case VHOST_USER_SET_PROTOCOL_FEATURES:
 280         case VHOST_USER_SET_LOG_BASE:
 281                 msg.payload.u64 = *((__u64 *)arg);
 282                 msg.size = sizeof(m.payload.u64);
 283                 break;
 284
 285         case VHOST_USER_SET_OWNER:
 286         case VHOST_USER_RESET_OWNER:
 287                 break;
 288
 289         case VHOST_USER_SET_MEM_TABLE:
 290                 if (prepare_vhost_memory_user(&msg, fds) < 0)
 291                         return -1;
 292                 fd_num = msg.payload.memory.nregions;
 293                 msg.size = sizeof(m.payload.memory.nregions);
 294                 msg.size += sizeof(m.payload.memory.padding);
 295                 msg.size += fd_num * sizeof(struct vhost_memory_region);
 296                 break;
 297
 298         case VHOST_USER_SET_LOG_FD:
 299                 fds[fd_num++] = *((int *)arg);
 300                 break;
 301
 302         case VHOST_USER_SET_VRING_NUM:
 303         case VHOST_USER_SET_VRING_BASE:
 304         case VHOST_USER_SET_VRING_ENABLE:
 305                 memcpy(&msg.payload.state, arg, sizeof(msg.payload.state));
 306                 msg.size = sizeof(m.payload.state);
 307                 break;
 308
 309         case VHOST_USER_GET_VRING_BASE:
 310                 memcpy(&msg.payload.state, arg, sizeof(msg.payload.state));
 311                 msg.size = sizeof(m.payload.state);
 312                 need_reply = 1;
 313                 break;
 314
 315         case VHOST_USER_SET_VRING_ADDR:
 316                 memcpy(&msg.payload.addr, arg, sizeof(msg.payload.addr));
 317                 msg.size = sizeof(m.payload.addr);
 318                 break;
 319
 320         case VHOST_USER_SET_VRING_KICK:
 321         case VHOST_USER_SET_VRING_CALL:
 322         case VHOST_USER_SET_VRING_ERR:
 323                 file = arg;
 324                 msg.payload.u64 = file->index & VHOST_USER_VRING_IDX_MASK;
 325                 msg.size = sizeof(m.payload.u64);
 326                 if (file->fd > 0)
 327                         fds[fd_num++] = file->fd;
 328                 else
 329                         msg.payload.u64 |= VHOST_USER_VRING_NOFD_MASK;
 330                 break;
 331
 332         default:
 333                 PMD_DRV_LOG(ERR, "trying to send unhandled msg type");
 334                 return -1;
 335         }
 336
 337         len = VHOST_USER_HDR_SIZE + msg.size;
 338         if (vhost_user_write(vhostfd, &msg, len, fds, fd_num) < 0) {
 339                 PMD_DRV_LOG(ERR, "%s failed: %s",
 340                             vhost_msg_strings[req], strerror(errno));
 341                 return -1;
 342         }
 343
 344         if (need_reply) {
 345                 if (vhost_user_read(vhostfd, &msg) < 0) {
 346                         PMD_DRV_LOG(ERR, "Received msg failed: %s",
 347                                     strerror(errno));
 348                         return -1;
 349                 }
 350
 351                 if (req != msg.request) {
 352                         PMD_DRV_LOG(ERR, "Received unexpected msg type");
 353                         return -1;
 354                 }
 355
 356                 switch (req) {
 357                 case VHOST_USER_GET_FEATURES:
 358                 case VHOST_USER_GET_PROTOCOL_FEATURES:
 359                         if (msg.size != sizeof(m.payload.u64)) {
 360                                 PMD_DRV_LOG(ERR, "Received bad msg size");
 361                                 return -1;
 362                         }
 363                         *((__u64 *)arg) = msg.payload.u64;
 364                         break;
 365                 case VHOST_USER_GET_VRING_BASE:
 366                         if (msg.size != sizeof(m.payload.state)) {
 367                                 PMD_DRV_LOG(ERR, "Received bad msg size");
 368                                 return -1;
 369                         }
 370                         memcpy(arg, &msg.payload.state,
 371                                sizeof(struct vhost_vring_state));
 372                         break;
 373                 default:
 374                         PMD_DRV_LOG(ERR, "Received unexpected msg type");
 375                         return -1;
 376                 }
 377         }
 378
 379         return 0;
 380 }
 381
 382 #define MAX_VIRTIO_USER_BACKLOG 1
 383 static int
 384 virtio_user_start_server(struct virtio_user_dev *dev, struct sockaddr_un *un)
 385 {
 386         int ret;
 387         int flag;
 388         int fd = dev->listenfd;
 389
 390         ret = bind(fd, (struct sockaddr *)un, sizeof(*un));
 391         if (ret < 0) {
 392                 PMD_DRV_LOG(ERR, "failed to bind to %s: %s; remove it and try again\n",
 393                             dev->path, strerror(errno));
 394                 return -1;
 395         }
 396         ret = listen(fd, MAX_VIRTIO_USER_BACKLOG);
 397         if (ret < 0)
 398                 return -1;
 399
 400         flag = fcntl(fd, F_GETFL);
 401         if (fcntl(fd, F_SETFL, flag | O_NONBLOCK) < 0) {
 402                 PMD_DRV_LOG(ERR, "fcntl failed, %s", strerror(errno));
 403                 return -1;
 404         }
 405
 406         return 0;
 407 }
 408
 409 /**
 410  * Set up environment to talk with a vhost user backend.
 411  *
 412  * @return
 413  *   - (-1) if fail;
 414  *   - (0) if succeed.
 415  */
 416 static int
 417 vhost_user_setup(struct virtio_user_dev *dev)
 418 {
 419         int fd;
 420         int flag;
 421         struct sockaddr_un un;
 422
 423         fd = socket(AF_UNIX, SOCK_STREAM, 0);
 424         if (fd < 0) {
 425                 PMD_DRV_LOG(ERR, "socket() error, %s", strerror(errno));
 426                 return -1;
 427         }
 428
 429         flag = fcntl(fd, F_GETFD);
 430         if (fcntl(fd, F_SETFD, flag | FD_CLOEXEC) < 0)
 431                 PMD_DRV_LOG(WARNING, "fcntl failed, %s", strerror(errno));
 432
 433         memset(&un, 0, sizeof(un));
 434         un.sun_family = AF_UNIX;
 435         strlcpy(un.sun_path, dev->path, sizeof(un.sun_path));
 436
 437         if (dev->is_server) {
 438                 dev->listenfd = fd;
 439                 if (virtio_user_start_server(dev, &un) < 0) {
 440                         PMD_DRV_LOG(ERR, "virtio-user startup fails in server mode");
 441                         close(fd);
 442                         return -1;
 443                 }
 444                 dev->vhostfd = -1;
 445         } else {
 446                 if (connect(fd, (struct sockaddr *)&un, sizeof(un)) < 0) {
 447                         PMD_DRV_LOG(ERR, "connect error, %s", strerror(errno));
 448                         close(fd);
 449                         return -1;
 450                 }
 451                 dev->vhostfd = fd;
 452         }
 453
 454         return 0;
 455 }
 456
 457 static int
 458 vhost_user_enable_queue_pair(struct virtio_user_dev *dev,
 459                              uint16_t pair_idx,
 460                              int enable)
 461 {
 462         int i;
 463
 464         if (dev->qp_enabled[pair_idx] == enable)
 465                 return 0;
 466
 467         for (i = 0; i < 2; ++i) {
 468                 struct vhost_vring_state state = {
 469                         .index = pair_idx * 2 + i,
 470                         .num   = enable,
 471                 };
 472
 473                 if (vhost_user_sock(dev, VHOST_USER_SET_VRING_ENABLE, &state))
 474                         return -1;
 475         }
 476
 477         dev->qp_enabled[pair_idx] = enable;
 478         return 0;
 479 }
 480
 481 struct virtio_user_backend_ops virtio_ops_user = {
 482         .setup = vhost_user_setup,
 483         .send_request = vhost_user_sock,
 484         .enable_qp = vhost_user_enable_queue_pair
 485 };