drivers/net/virtio/virtio_user/vhost_kernel.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <sys/types.h>
   6 #include <sys/stat.h>
   7 #include <fcntl.h>
   8 #include <unistd.h>
   9 #include <errno.h>
  10
  11 #include <rte_memory.h>
  12
  13 #include "vhost.h"
  14 #include "virtio_user_dev.h"
  15 #include "vhost_kernel_tap.h"
  16
  17 struct vhost_kernel_data {
  18         int *vhostfds;
  19         int *tapfds;
  20 };
  21
  22 struct vhost_memory_kernel {
  23         uint32_t nregions;
  24         uint32_t padding;
  25         struct vhost_memory_region regions[0];
  26 };
  27
  28 /* vhost kernel ioctls */
  29 #define VHOST_VIRTIO 0xAF
  30 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
  31 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
  32 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
  33 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
  34 #define VHOST_SET_MEM_TABLE _IOW(VHOST_VIRTIO, 0x03, struct vhost_memory_kernel)
  35 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
  36 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
  37 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
  38 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
  39 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
  40 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
  41 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
  42 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
  43 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
  44 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
  45
  46 /* with below features, vhost kernel does not need to do the checksum and TSO,
  47  * these info will be passed to virtio_user through virtio net header.
  48  */
  49 #define VHOST_KERNEL_GUEST_OFFLOADS_MASK        \
  50         ((1ULL << VIRTIO_NET_F_GUEST_CSUM) |    \
  51          (1ULL << VIRTIO_NET_F_GUEST_TSO4) |    \
  52          (1ULL << VIRTIO_NET_F_GUEST_TSO6) |    \
  53          (1ULL << VIRTIO_NET_F_GUEST_ECN)  |    \
  54          (1ULL << VIRTIO_NET_F_GUEST_UFO))
  55
  56 /* with below features, when flows from virtio_user to vhost kernel
  57  * (1) if flows goes up through the kernel networking stack, it does not need
  58  * to verify checksum, which can save CPU cycles;
  59  * (2) if flows goes through a Linux bridge and outside from an interface
  60  * (kernel driver), checksum and TSO will be done by GSO in kernel or even
  61  * offloaded into real physical device.
  62  */
  63 #define VHOST_KERNEL_HOST_OFFLOADS_MASK         \
  64         ((1ULL << VIRTIO_NET_F_HOST_TSO4) |     \
  65          (1ULL << VIRTIO_NET_F_HOST_TSO6) |     \
  66          (1ULL << VIRTIO_NET_F_CSUM))
  67
  68 static uint64_t max_regions = 64;
  69
  70 static void
  71 get_vhost_kernel_max_regions(void)
  72 {
  73         int fd;
  74         char buf[20] = {'\0'};
  75
  76         fd = open("/sys/module/vhost/parameters/max_mem_regions", O_RDONLY);
  77         if (fd < 0)
  78                 return;
  79
  80         if (read(fd, buf, sizeof(buf) - 1) > 0)
  81                 max_regions = strtoull(buf, NULL, 10);
  82
  83         close(fd);
  84 }
  85
  86 static int
  87 vhost_kernel_ioctl(int fd, uint64_t request, void *arg)
  88 {
  89         int ret;
  90
  91         ret = ioctl(fd, request, arg);
  92         if (ret) {
  93                 PMD_DRV_LOG(ERR, "Vhost-kernel ioctl %"PRIu64" failed (%s)",
  94                                 request, strerror(errno));
  95                 return -1;
  96         }
  97
  98         return 0;
  99 }
 100
 101 static int
 102 vhost_kernel_set_owner(struct virtio_user_dev *dev)
 103 {
 104         struct vhost_kernel_data *data = dev->backend_data;
 105
 106         return vhost_kernel_ioctl(data->vhostfds[0], VHOST_SET_OWNER, NULL);
 107 }
 108
 109 static int
 110 vhost_kernel_get_features(struct virtio_user_dev *dev, uint64_t *features)
 111 {
 112         int ret;
 113         unsigned int tap_features;
 114         struct vhost_kernel_data *data = dev->backend_data;
 115
 116         ret = vhost_kernel_ioctl(data->vhostfds[0], VHOST_GET_FEATURES, features);
 117         if (ret < 0) {
 118                 PMD_DRV_LOG(ERR, "Failed to get features");
 119                 return -1;
 120         }
 121
 122         ret = tap_support_features(&tap_features);
 123         if (ret < 0) {
 124                 PMD_DRV_LOG(ERR, "Failed to get TAP features");
 125                 return -1;
 126         }
 127
 128         /* with tap as the backend, all these features are supported
 129          * but not claimed by vhost-net, so we add them back when
 130          * reporting to upper layer.
 131          */
 132         if (tap_features & IFF_VNET_HDR) {
 133                 *features |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
 134                 *features |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
 135         }
 136
 137         /* vhost_kernel will not declare this feature, but it does
 138          * support multi-queue.
 139          */
 140         if (tap_features & IFF_MULTI_QUEUE)
 141                 *features |= (1ull << VIRTIO_NET_F_MQ);
 142
 143         return 0;
 144 }
 145
 146 static int
 147 vhost_kernel_set_features(struct virtio_user_dev *dev, uint64_t features)
 148 {
 149         struct vhost_kernel_data *data = dev->backend_data;
 150
 151         /* We don't need memory protection here */
 152         features &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 153         /* VHOST kernel does not know about below flags */
 154         features &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
 155         features &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
 156         features &= ~(1ULL << VIRTIO_NET_F_MQ);
 157
 158         return vhost_kernel_ioctl(data->vhostfds[0], VHOST_SET_FEATURES, &features);
 159 }
 160
 161 static int
 162 add_memseg_list(const struct rte_memseg_list *msl, void *arg)
 163 {
 164         struct vhost_memory_kernel *vm = arg;
 165         struct vhost_memory_region *mr;
 166         void *start_addr;
 167         uint64_t len;
 168
 169         if (msl->external)
 170                 return 0;
 171
 172         if (vm->nregions >= max_regions)
 173                 return -1;
 174
 175         start_addr = msl->base_va;
 176         len = msl->page_sz * msl->memseg_arr.len;
 177
 178         mr = &vm->regions[vm->nregions++];
 179
 180         mr->guest_phys_addr = (uint64_t)(uintptr_t)start_addr;
 181         mr->userspace_addr = (uint64_t)(uintptr_t)start_addr;
 182         mr->memory_size = len;
 183         mr->mmap_offset = 0; /* flags_padding */
 184
 185         PMD_DRV_LOG(DEBUG, "index=%u addr=%p len=%" PRIu64,
 186                         vm->nregions - 1, start_addr, len);
 187
 188         return 0;
 189 }
 190
 191 /* By default, vhost kernel module allows 64 regions, but DPDK may
 192  * have much more memory regions. Below function will treat each
 193  * contiguous memory space reserved by DPDK as one region.
 194  */
 195 static int
 196 vhost_kernel_set_memory_table(struct virtio_user_dev *dev)
 197 {
 198         struct vhost_kernel_data *data = dev->backend_data;
 199         struct vhost_memory_kernel *vm;
 200         int ret;
 201
 202         vm = malloc(sizeof(struct vhost_memory_kernel) +
 203                         max_regions *
 204                         sizeof(struct vhost_memory_region));
 205         if (!vm)
 206                 goto err;
 207
 208         vm->nregions = 0;
 209         vm->padding = 0;
 210
 211         /*
 212          * The memory lock has already been taken by memory subsystem
 213          * or virtio_user_start_device().
 214          */
 215         ret = rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm);
 216         if (ret < 0)
 217                 goto err_free;
 218
 219         ret = vhost_kernel_ioctl(data->vhostfds[0], VHOST_SET_MEM_TABLE, vm);
 220         if (ret < 0)
 221                 goto err_free;
 222
 223         free(vm);
 224
 225         return 0;
 226 err_free:
 227         free(vm);
 228 err:
 229         PMD_DRV_LOG(ERR, "Failed to set memory table");
 230         return -1;
 231 }
 232
 233 static int
 234 vhost_kernel_set_vring(struct virtio_user_dev *dev, uint64_t req, struct vhost_vring_state *state)
 235 {
 236         int ret, fd;
 237         unsigned int index = state->index;
 238         struct vhost_kernel_data *data = dev->backend_data;
 239
 240         /* Convert from queue index to queue-pair & offset */
 241         fd = data->vhostfds[state->index / 2];
 242         state->index %= 2;
 243
 244         ret = vhost_kernel_ioctl(fd, req, state);
 245         if (ret < 0) {
 246                 PMD_DRV_LOG(ERR, "Failed to set vring (request %" PRIu64 ")", req);
 247                 return -1;
 248         }
 249
 250         /* restore index back to queue index */
 251         state->index = index;
 252
 253         return 0;
 254 }
 255
 256 static int
 257 vhost_kernel_set_vring_num(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 258 {
 259         return vhost_kernel_set_vring(dev, VHOST_SET_VRING_NUM, state);
 260 }
 261
 262 static int
 263 vhost_kernel_set_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 264 {
 265         return vhost_kernel_set_vring(dev, VHOST_SET_VRING_BASE, state);
 266 }
 267
 268 static int
 269 vhost_kernel_get_vring_base(struct virtio_user_dev *dev, struct vhost_vring_state *state)
 270 {
 271         return vhost_kernel_set_vring(dev, VHOST_GET_VRING_BASE, state);
 272 }
 273
 274 static int
 275 vhost_kernel_set_vring_file(struct virtio_user_dev *dev, uint64_t req,
 276                 struct vhost_vring_file *file)
 277 {
 278         int ret, fd;
 279         unsigned int index = file->index;
 280         struct vhost_kernel_data *data = dev->backend_data;
 281
 282         /* Convert from queue index to queue-pair & offset */
 283         fd = data->vhostfds[file->index / 2];
 284         file->index %= 2;
 285
 286         ret = vhost_kernel_ioctl(fd, req, file);
 287         if (ret < 0) {
 288                 PMD_DRV_LOG(ERR, "Failed to set vring file (request %" PRIu64 ")", req);
 289                 return -1;
 290         }
 291
 292         /* restore index back to queue index */
 293         file->index = index;
 294
 295         return 0;
 296 }
 297
 298 static int
 299 vhost_kernel_set_vring_kick(struct virtio_user_dev *dev, struct vhost_vring_file *file)
 300 {
 301         return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_KICK, file);
 302 }
 303
 304 static int
 305 vhost_kernel_set_vring_call(struct virtio_user_dev *dev, struct vhost_vring_file *file)
 306 {
 307         return vhost_kernel_set_vring_file(dev, VHOST_SET_VRING_CALL, file);
 308 }
 309
 310 static int
 311 vhost_kernel_set_vring_addr(struct virtio_user_dev *dev, struct vhost_vring_addr *addr)
 312 {
 313         int ret, fd;
 314         unsigned int index = addr->index;
 315         struct vhost_kernel_data *data = dev->backend_data;
 316
 317         /* Convert from queue index to queue-pair & offset */
 318         fd = data->vhostfds[addr->index / 2];
 319         addr->index %= 2;
 320
 321         ret = vhost_kernel_ioctl(fd, VHOST_SET_VRING_ADDR, addr);
 322         if (ret < 0) {
 323                 PMD_DRV_LOG(ERR, "Failed to set vring address");
 324                 return -1;
 325         }
 326
 327         /* restore index back to queue index */
 328         addr->index = index;
 329
 330         return 0;
 331 }
 332
 333 static int
 334 vhost_kernel_get_status(struct virtio_user_dev *dev __rte_unused, uint8_t *status __rte_unused)
 335 {
 336         return -ENOTSUP;
 337 }
 338
 339 static int
 340 vhost_kernel_set_status(struct virtio_user_dev *dev __rte_unused, uint8_t status __rte_unused)
 341 {
 342         return -ENOTSUP;
 343 }
 344
 345 /**
 346  * Set up environment to talk with a vhost kernel backend.
 347  *
 348  * @return
 349  *   - (-1) if fail to set up;
 350  *   - (>=0) if successful.
 351  */
 352 static int
 353 vhost_kernel_setup(struct virtio_user_dev *dev)
 354 {
 355         int vhostfd;
 356         uint32_t q, i;
 357         struct vhost_kernel_data *data;
 358
 359         data = malloc(sizeof(*data));
 360         if (!data) {
 361                 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost-kernel data", dev->path);
 362                 return -1;
 363         }
 364
 365         data->vhostfds = malloc(dev->max_queue_pairs * sizeof(int));
 366         if (!data->vhostfds) {
 367                 PMD_INIT_LOG(ERR, "(%s) Failed to allocate Vhost FDs", dev->path);
 368                 goto err_data;
 369         }
 370         data->tapfds = malloc(dev->max_queue_pairs * sizeof(int));
 371         if (!data->tapfds) {
 372                 PMD_INIT_LOG(ERR, "(%s) Failed to allocate TAP FDs", dev->path);
 373                 goto err_vhostfds;
 374         }
 375
 376         for (q = 0; q < dev->max_queue_pairs; ++q) {
 377                 data->vhostfds[q] = -1;
 378                 data->tapfds[q] = -1;
 379         }
 380
 381         get_vhost_kernel_max_regions();
 382
 383         for (i = 0; i < dev->max_queue_pairs; ++i) {
 384                 vhostfd = open(dev->path, O_RDWR);
 385                 if (vhostfd < 0) {
 386                         PMD_DRV_LOG(ERR, "fail to open %s, %s", dev->path, strerror(errno));
 387                         goto err_tapfds;
 388                 }
 389
 390                 data->vhostfds[i] = vhostfd;
 391         }
 392
 393         dev->backend_data = data;
 394
 395         return 0;
 396
 397 err_tapfds:
 398         for (i = 0; i < dev->max_queue_pairs; i++)
 399                 if (data->vhostfds[i] >= 0)
 400                         close(data->vhostfds[i]);
 401
 402         free(data->tapfds);
 403 err_vhostfds:
 404         free(data->vhostfds);
 405 err_data:
 406         free(data);
 407
 408         return -1;
 409 }
 410
 411 static int
 412 vhost_kernel_destroy(struct virtio_user_dev *dev)
 413 {
 414         struct vhost_kernel_data *data = dev->backend_data;
 415         uint32_t i;
 416
 417         if (!data)
 418                 return 0;
 419
 420         for (i = 0; i < dev->max_queue_pairs; ++i) {
 421                 if (data->vhostfds[i] >= 0)
 422                         close(data->vhostfds[i]);
 423                 if (data->tapfds[i] >= 0)
 424                         close(data->tapfds[i]);
 425         }
 426
 427         free(data->vhostfds);
 428         free(data->tapfds);
 429         free(data);
 430         dev->backend_data = NULL;
 431
 432         return 0;
 433 }
 434
 435 static int
 436 vhost_kernel_set_backend(int vhostfd, int tapfd)
 437 {
 438         struct vhost_vring_file f;
 439
 440         f.fd = tapfd;
 441         f.index = 0;
 442         if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
 443                 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
 444                                 strerror(errno));
 445                 return -1;
 446         }
 447
 448         f.index = 1;
 449         if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
 450                 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
 451                                 strerror(errno));
 452                 return -1;
 453         }
 454
 455         return 0;
 456 }
 457
 458 static int
 459 vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
 460                                uint16_t pair_idx,
 461                                int enable)
 462 {
 463         int hdr_size;
 464         int vhostfd;
 465         int tapfd;
 466         int req_mq = (dev->max_queue_pairs > 1);
 467         struct vhost_kernel_data *data = dev->backend_data;
 468
 469         vhostfd = data->vhostfds[pair_idx];
 470
 471         if (dev->qp_enabled[pair_idx] == enable)
 472                 return 0;
 473
 474         if (!enable) {
 475                 tapfd = data->tapfds[pair_idx];
 476                 if (vhost_kernel_set_backend(vhostfd, -1) < 0) {
 477                         PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
 478                         return -1;
 479                 }
 480                 if (req_mq && vhost_kernel_tap_set_queue(tapfd, false) < 0) {
 481                         PMD_DRV_LOG(ERR, "fail to disable tap for vhost kernel");
 482                         return -1;
 483                 }
 484                 dev->qp_enabled[pair_idx] = false;
 485                 return 0;
 486         }
 487
 488         if (data->tapfds[pair_idx] >= 0) {
 489                 tapfd = data->tapfds[pair_idx];
 490                 if (vhost_kernel_tap_set_offload(tapfd, dev->features) == -1)
 491                         return -1;
 492                 if (req_mq && vhost_kernel_tap_set_queue(tapfd, true) < 0) {
 493                         PMD_DRV_LOG(ERR, "fail to enable tap for vhost kernel");
 494                         return -1;
 495                 }
 496                 goto set_backend;
 497         }
 498
 499         if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
 500             (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
 501                 hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 502         else
 503                 hdr_size = sizeof(struct virtio_net_hdr);
 504
 505         tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
 506                          (char *)dev->mac_addr, dev->features);
 507         if (tapfd < 0) {
 508                 PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
 509                 return -1;
 510         }
 511
 512         data->tapfds[pair_idx] = tapfd;
 513
 514 set_backend:
 515         if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
 516                 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
 517                 return -1;
 518         }
 519
 520         dev->qp_enabled[pair_idx] = true;
 521         return 0;
 522 }
 523
 524 static int
 525 vhost_kernel_get_backend_features(uint64_t *features)
 526 {
 527         *features = 0;
 528
 529         return 0;
 530 }
 531
 532 static int
 533 vhost_kernel_update_link_state(struct virtio_user_dev *dev __rte_unused)
 534 {
 535         /* Nothing to update (Maybe get TAP interface link state?) */
 536         return 0;
 537 }
 538
 539 static int
 540 vhost_kernel_get_intr_fd(struct virtio_user_dev *dev __rte_unused)
 541 {
 542         /* No link state interrupt with Vhost-kernel */
 543         return -1;
 544 }
 545
 546 struct virtio_user_backend_ops virtio_ops_kernel = {
 547         .setup = vhost_kernel_setup,
 548         .destroy = vhost_kernel_destroy,
 549         .get_backend_features = vhost_kernel_get_backend_features,
 550         .set_owner = vhost_kernel_set_owner,
 551         .get_features = vhost_kernel_get_features,
 552         .set_features = vhost_kernel_set_features,
 553         .set_memory_table = vhost_kernel_set_memory_table,
 554         .set_vring_num = vhost_kernel_set_vring_num,
 555         .set_vring_base = vhost_kernel_set_vring_base,
 556         .get_vring_base = vhost_kernel_get_vring_base,
 557         .set_vring_call = vhost_kernel_set_vring_call,
 558         .set_vring_kick = vhost_kernel_set_vring_kick,
 559         .set_vring_addr = vhost_kernel_set_vring_addr,
 560         .get_status = vhost_kernel_get_status,
 561         .set_status = vhost_kernel_set_status,
 562         .enable_qp = vhost_kernel_enable_queue_pair,
 563         .update_link_state = vhost_kernel_update_link_state,
 564         .get_intr_fd = vhost_kernel_get_intr_fd,
 565 };