contrib/vhost-user-blk/vhost-user-blk.c

   1 /*
   2  * vhost-user-blk sample application
   3  *
   4  * Copyright (c) 2017 Intel Corporation. All rights reserved.
   5  *
   6  * Author:
   7  *  Changpeng Liu <changpeng.liu@intel.com>
   8  *
   9  * This work is based on the "vhost-user-scsi" sample and "virtio-blk" driver
  10  * implementation by:
  11  *  Felipe Franciosi <felipe@nutanix.com>
  12  *  Anthony Liguori <aliguori@us.ibm.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2 only.
  15  * See the COPYING file in the top-level directory.
  16  */
  17
  18 #include "qemu/osdep.h"
  19 #include "standard-headers/linux/virtio_blk.h"
  20 #include "contrib/libvhost-user/libvhost-user-glib.h"
  21 #include "contrib/libvhost-user/libvhost-user.h"
  22
  23 #include <glib.h>
  24
  25 struct virtio_blk_inhdr {
  26     unsigned char status;
  27 };
  28
  29 /* vhost user block device */
  30 typedef struct VubDev {
  31     VugDev parent;
  32     int blk_fd;
  33     struct virtio_blk_config blkcfg;
  34     char *blk_name;
  35     GMainLoop *loop;
  36 } VubDev;
  37
  38 typedef struct VubReq {
  39     VuVirtqElement *elem;
  40     int64_t sector_num;
  41     size_t size;
  42     struct virtio_blk_inhdr *in;
  43     struct virtio_blk_outhdr *out;
  44     VubDev *vdev_blk;
  45     struct VuVirtq *vq;
  46 } VubReq;
  47
  48 /* refer util/iov.c */
  49 static size_t vub_iov_size(const struct iovec *iov,
  50                               const unsigned int iov_cnt)
  51 {
  52     size_t len;
  53     unsigned int i;
  54
  55     len = 0;
  56     for (i = 0; i < iov_cnt; i++) {
  57         len += iov[i].iov_len;
  58     }
  59     return len;
  60 }
  61
  62 static void vub_panic_cb(VuDev *vu_dev, const char *buf)
  63 {
  64     VugDev *gdev;
  65     VubDev *vdev_blk;
  66
  67     assert(vu_dev);
  68
  69     gdev = container_of(vu_dev, VugDev, parent);
  70     vdev_blk = container_of(gdev, VubDev, parent);
  71     if (buf) {
  72         g_warning("vu_panic: %s", buf);
  73     }
  74
  75     g_main_loop_quit(vdev_blk->loop);
  76 }
  77
  78 static void vub_req_complete(VubReq *req)
  79 {
  80     VugDev *gdev = &req->vdev_blk->parent;
  81     VuDev *vu_dev = &gdev->parent;
  82
  83     /* IO size with 1 extra status byte */
  84     vu_queue_push(vu_dev, req->vq, req->elem,
  85                   req->size + 1);
  86     vu_queue_notify(vu_dev, req->vq);
  87
  88     if (req->elem) {
  89         free(req->elem);
  90     }
  91
  92     g_free(req);
  93 }
  94
  95 static int vub_open(const char *file_name, bool wce)
  96 {
  97     int fd;
  98     int flags = O_RDWR;
  99
 100     if (!wce) {
 101         flags |= O_DIRECT;
 102     }
 103
 104     fd = open(file_name, flags);
 105     if (fd < 0) {
 106         fprintf(stderr, "Cannot open file %s, %s\n", file_name,
 107                 strerror(errno));
 108         return -1;
 109     }
 110
 111     return fd;
 112 }
 113
 114 static ssize_t
 115 vub_readv(VubReq *req, struct iovec *iov, uint32_t iovcnt)
 116 {
 117     VubDev *vdev_blk = req->vdev_blk;
 118     ssize_t rc;
 119
 120     if (!iovcnt) {
 121         fprintf(stderr, "Invalid Read IOV count\n");
 122         return -1;
 123     }
 124
 125     req->size = vub_iov_size(iov, iovcnt);
 126     rc = preadv(vdev_blk->blk_fd, iov, iovcnt, req->sector_num * 512);
 127     if (rc < 0) {
 128         fprintf(stderr, "%s, Sector %"PRIu64", Size %lu failed with %s\n",
 129                 vdev_blk->blk_name, req->sector_num, req->size,
 130                 strerror(errno));
 131         return -1;
 132     }
 133
 134     return rc;
 135 }
 136
 137 static ssize_t
 138 vub_writev(VubReq *req, struct iovec *iov, uint32_t iovcnt)
 139 {
 140     VubDev *vdev_blk = req->vdev_blk;
 141     ssize_t rc;
 142
 143     if (!iovcnt) {
 144         fprintf(stderr, "Invalid Write IOV count\n");
 145         return -1;
 146     }
 147
 148     req->size = vub_iov_size(iov, iovcnt);
 149     rc = pwritev(vdev_blk->blk_fd, iov, iovcnt, req->sector_num * 512);
 150     if (rc < 0) {
 151         fprintf(stderr, "%s, Sector %"PRIu64", Size %lu failed with %s\n",
 152                 vdev_blk->blk_name, req->sector_num, req->size,
 153                 strerror(errno));
 154         return -1;
 155     }
 156
 157     return rc;
 158 }
 159
 160 static void
 161 vub_flush(VubReq *req)
 162 {
 163     VubDev *vdev_blk = req->vdev_blk;
 164
 165     fdatasync(vdev_blk->blk_fd);
 166 }
 167
 168 static int vub_virtio_process_req(VubDev *vdev_blk,
 169                                      VuVirtq *vq)
 170 {
 171     VugDev *gdev = &vdev_blk->parent;
 172     VuDev *vu_dev = &gdev->parent;
 173     VuVirtqElement *elem;
 174     uint32_t type;
 175     unsigned in_num;
 176     unsigned out_num;
 177     VubReq *req;
 178
 179     elem = vu_queue_pop(vu_dev, vq, sizeof(VuVirtqElement) + sizeof(VubReq));
 180     if (!elem) {
 181         return -1;
 182     }
 183
 184     /* refer to hw/block/virtio_blk.c */
 185     if (elem->out_num < 1 || elem->in_num < 1) {
 186         fprintf(stderr, "virtio-blk request missing headers\n");
 187         free(elem);
 188         return -1;
 189     }
 190
 191     req = g_new0(VubReq, 1);
 192     req->vdev_blk = vdev_blk;
 193     req->vq = vq;
 194     req->elem = elem;
 195
 196     in_num = elem->in_num;
 197     out_num = elem->out_num;
 198
 199     /* don't support VIRTIO_F_ANY_LAYOUT and virtio 1.0 only */
 200     if (elem->out_sg[0].iov_len < sizeof(struct virtio_blk_outhdr)) {
 201         fprintf(stderr, "Invalid outhdr size\n");
 202         goto err;
 203     }
 204     req->out = (struct virtio_blk_outhdr *)elem->out_sg[0].iov_base;
 205     out_num--;
 206
 207     if (elem->in_sg[in_num - 1].iov_len < sizeof(struct virtio_blk_inhdr)) {
 208         fprintf(stderr, "Invalid inhdr size\n");
 209         goto err;
 210     }
 211     req->in = (struct virtio_blk_inhdr *)elem->in_sg[in_num - 1].iov_base;
 212     in_num--;
 213
 214     type = le32toh(req->out->type);
 215     switch (type & ~(VIRTIO_BLK_T_OUT | VIRTIO_BLK_T_BARRIER)) {
 216         case VIRTIO_BLK_T_IN: {
 217             ssize_t ret = 0;
 218             bool is_write = type & VIRTIO_BLK_T_OUT;
 219             req->sector_num = le64toh(req->out->sector);
 220             if (is_write) {
 221                 ret  = vub_writev(req, &elem->out_sg[1], out_num);
 222             } else {
 223                 ret = vub_readv(req, &elem->in_sg[0], in_num);
 224             }
 225             if (ret >= 0) {
 226                 req->in->status = VIRTIO_BLK_S_OK;
 227             } else {
 228                 req->in->status = VIRTIO_BLK_S_IOERR;
 229             }
 230             vub_req_complete(req);
 231             break;
 232         }
 233         case VIRTIO_BLK_T_FLUSH: {
 234             vub_flush(req);
 235             req->in->status = VIRTIO_BLK_S_OK;
 236             vub_req_complete(req);
 237             break;
 238         }
 239         case VIRTIO_BLK_T_GET_ID: {
 240             size_t size = MIN(vub_iov_size(&elem->in_sg[0], in_num),
 241                               VIRTIO_BLK_ID_BYTES);
 242             snprintf(elem->in_sg[0].iov_base, size, "%s", "vhost_user_blk");
 243             req->in->status = VIRTIO_BLK_S_OK;
 244             req->size = elem->in_sg[0].iov_len;
 245             vub_req_complete(req);
 246             break;
 247         }
 248         default: {
 249             req->in->status = VIRTIO_BLK_S_UNSUPP;
 250             vub_req_complete(req);
 251             break;
 252         }
 253     }
 254
 255     return 0;
 256
 257 err:
 258     free(elem);
 259     g_free(req);
 260     return -1;
 261 }
 262
 263 static void vub_process_vq(VuDev *vu_dev, int idx)
 264 {
 265     VugDev *gdev;
 266     VubDev *vdev_blk;
 267     VuVirtq *vq;
 268     int ret;
 269
 270     if ((idx < 0) || (idx >= VHOST_MAX_NR_VIRTQUEUE)) {
 271         fprintf(stderr, "VQ Index out of range: %d\n", idx);
 272         vub_panic_cb(vu_dev, NULL);
 273         return;
 274     }
 275
 276     gdev = container_of(vu_dev, VugDev, parent);
 277     vdev_blk = container_of(gdev, VubDev, parent);
 278     assert(vdev_blk);
 279
 280     vq = vu_get_queue(vu_dev, idx);
 281     assert(vq);
 282
 283     while (1) {
 284         ret = vub_virtio_process_req(vdev_blk, vq);
 285         if (ret) {
 286             break;
 287         }
 288     }
 289 }
 290
 291 static void vub_queue_set_started(VuDev *vu_dev, int idx, bool started)
 292 {
 293     VuVirtq *vq;
 294
 295     assert(vu_dev);
 296
 297     vq = vu_get_queue(vu_dev, idx);
 298     vu_set_queue_handler(vu_dev, vq, started ? vub_process_vq : NULL);
 299 }
 300
 301 static uint64_t
 302 vub_get_features(VuDev *dev)
 303 {
 304     return 1ull << VIRTIO_BLK_F_SIZE_MAX |
 305            1ull << VIRTIO_BLK_F_SEG_MAX |
 306            1ull << VIRTIO_BLK_F_TOPOLOGY |
 307            1ull << VIRTIO_BLK_F_BLK_SIZE |
 308            1ull << VIRTIO_BLK_F_FLUSH |
 309            1ull << VIRTIO_BLK_F_CONFIG_WCE |
 310            1ull << VIRTIO_F_VERSION_1 |
 311            1ull << VHOST_USER_F_PROTOCOL_FEATURES;
 312 }
 313
 314 static int
 315 vub_get_config(VuDev *vu_dev, uint8_t *config, uint32_t len)
 316 {
 317     VugDev *gdev;
 318     VubDev *vdev_blk;
 319
 320     gdev = container_of(vu_dev, VugDev, parent);
 321     vdev_blk = container_of(gdev, VubDev, parent);
 322     memcpy(config, &vdev_blk->blkcfg, len);
 323
 324     return 0;
 325 }
 326
 327 static int
 328 vub_set_config(VuDev *vu_dev, const uint8_t *data,
 329                uint32_t offset, uint32_t size, uint32_t flags)
 330 {
 331     VugDev *gdev;
 332     VubDev *vdev_blk;
 333     uint8_t wce;
 334     int fd;
 335
 336     /* don't support live migration */
 337     if (flags != VHOST_SET_CONFIG_TYPE_MASTER) {
 338         return -1;
 339     }
 340
 341     gdev = container_of(vu_dev, VugDev, parent);
 342     vdev_blk = container_of(gdev, VubDev, parent);
 343
 344     if (offset != offsetof(struct virtio_blk_config, wce) ||
 345         size != 1) {
 346         return -1;
 347     }
 348
 349     wce = *data;
 350     if (wce == vdev_blk->blkcfg.wce) {
 351         /* Do nothing as same with old configuration */
 352         return 0;
 353     }
 354
 355     vdev_blk->blkcfg.wce = wce;
 356     fprintf(stdout, "Write Cache Policy Changed\n");
 357     if (vdev_blk->blk_fd >= 0) {
 358         close(vdev_blk->blk_fd);
 359         vdev_blk->blk_fd = -1;
 360     }
 361
 362     fd = vub_open(vdev_blk->blk_name, wce);
 363     if (fd < 0) {
 364         fprintf(stderr, "Error to open block device %s\n", vdev_blk->blk_name);
 365         vdev_blk->blk_fd = -1;
 366         return -1;
 367     }
 368     vdev_blk->blk_fd = fd;
 369
 370     return 0;
 371 }
 372
 373 static const VuDevIface vub_iface = {
 374     .get_features = vub_get_features,
 375     .queue_set_started = vub_queue_set_started,
 376     .get_config = vub_get_config,
 377     .set_config = vub_set_config,
 378 };
 379
 380 static int unix_sock_new(char *unix_fn)
 381 {
 382     int sock;
 383     struct sockaddr_un un;
 384     size_t len;
 385
 386     assert(unix_fn);
 387
 388     sock = socket(AF_UNIX, SOCK_STREAM, 0);
 389     if (sock <= 0) {
 390         perror("socket");
 391         return -1;
 392     }
 393
 394     un.sun_family = AF_UNIX;
 395     (void)snprintf(un.sun_path, sizeof(un.sun_path), "%s", unix_fn);
 396     len = sizeof(un.sun_family) + strlen(un.sun_path);
 397
 398     (void)unlink(unix_fn);
 399     if (bind(sock, (struct sockaddr *)&un, len) < 0) {
 400         perror("bind");
 401         goto fail;
 402     }
 403
 404     if (listen(sock, 1) < 0) {
 405         perror("listen");
 406         goto fail;
 407     }
 408
 409     return sock;
 410
 411 fail:
 412     (void)close(sock);
 413
 414     return -1;
 415 }
 416
 417 static void vub_free(struct VubDev *vdev_blk)
 418 {
 419     if (!vdev_blk) {
 420         return;
 421     }
 422
 423     g_main_loop_unref(vdev_blk->loop);
 424     if (vdev_blk->blk_fd >= 0) {
 425         close(vdev_blk->blk_fd);
 426     }
 427     g_free(vdev_blk);
 428 }
 429
 430 static uint32_t
 431 vub_get_blocksize(int fd)
 432 {
 433     uint32_t blocksize = 512;
 434
 435 #if defined(__linux__) && defined(BLKSSZGET)
 436     if (ioctl(fd, BLKSSZGET, &blocksize) == 0) {
 437         return blocklen;
 438     }
 439 #endif
 440
 441     return blocksize;
 442 }
 443
 444 static void
 445 vub_initialize_config(int fd, struct virtio_blk_config *config)
 446 {
 447     off64_t capacity;
 448
 449     capacity = lseek64(fd, 0, SEEK_END);
 450     config->capacity = capacity >> 9;
 451     config->blk_size = vub_get_blocksize(fd);
 452     config->size_max = 65536;
 453     config->seg_max = 128 - 2;
 454     config->min_io_size = 1;
 455     config->opt_io_size = 1;
 456     config->num_queues = 1;
 457 }
 458
 459 static VubDev *
 460 vub_new(char *blk_file)
 461 {
 462     VubDev *vdev_blk;
 463
 464     vdev_blk = g_new0(VubDev, 1);
 465     vdev_blk->loop = g_main_loop_new(NULL, FALSE);
 466     vdev_blk->blk_fd = vub_open(blk_file, 0);
 467     if (vdev_blk->blk_fd  < 0) {
 468         fprintf(stderr, "Error to open block device %s\n", blk_file);
 469         vub_free(vdev_blk);
 470         return NULL;
 471     }
 472     vdev_blk->blkcfg.wce = 0;
 473     vdev_blk->blk_name = blk_file;
 474
 475     /* fill virtio_blk_config with block parameters */
 476     vub_initialize_config(vdev_blk->blk_fd, &vdev_blk->blkcfg);
 477
 478     return vdev_blk;
 479 }
 480
 481 int main(int argc, char **argv)
 482 {
 483     int opt;
 484     char *unix_socket = NULL;
 485     char *blk_file = NULL;
 486     int lsock = -1, csock = -1;
 487     VubDev *vdev_blk = NULL;
 488
 489     while ((opt = getopt(argc, argv, "b:s:h")) != -1) {
 490         switch (opt) {
 491         case 'b':
 492             blk_file = g_strdup(optarg);
 493             break;
 494         case 's':
 495             unix_socket = g_strdup(optarg);
 496             break;
 497         case 'h':
 498         default:
 499             printf("Usage: %s [-b block device or file, -s UNIX domain socket]"
 500                    " | [ -h ]\n", argv[0]);
 501             return 0;
 502         }
 503     }
 504
 505     if (!unix_socket || !blk_file) {
 506         printf("Usage: %s [-b block device or file, -s UNIX domain socket] |"
 507                " [ -h ]\n", argv[0]);
 508         return -1;
 509     }
 510
 511     lsock = unix_sock_new(unix_socket);
 512     if (lsock < 0) {
 513         goto err;
 514     }
 515
 516     csock = accept(lsock, (void *)0, (void *)0);
 517     if (csock < 0) {
 518         fprintf(stderr, "Accept error %s\n", strerror(errno));
 519         goto err;
 520     }
 521
 522     vdev_blk = vub_new(blk_file);
 523     if (!vdev_blk) {
 524         goto err;
 525     }
 526
 527     vug_init(&vdev_blk->parent, csock, vub_panic_cb, &vub_iface);
 528
 529     g_main_loop_run(vdev_blk->loop);
 530
 531     vug_deinit(&vdev_blk->parent);
 532
 533 err:
 534     vub_free(vdev_blk);
 535     if (csock >= 0) {
 536         close(csock);
 537     }
 538     if (lsock >= 0) {
 539         close(lsock);
 540     }
 541     g_free(unix_socket);
 542     g_free(blk_file);
 543
 544     return 0;
 545 }