hw/virtio-blk.c

   1 /*
   2  * Virtio Block Device
   3  *
   4  * Copyright IBM, Corp. 2007
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 #include "qemu-common.h"
  15 #include "qemu-error.h"
  16 #include "trace.h"
  17 #include "blockdev.h"
  18 #include "virtio-blk.h"
  19 #ifdef __linux__
  20 # include <scsi/sg.h>
  21 #endif
  22
  23 typedef struct VirtIOBlock
  24 {
  25     VirtIODevice vdev;
  26     BlockDriverState *bs;
  27     VirtQueue *vq;
  28     void *rq;
  29     QEMUBH *bh;
  30     BlockConf *conf;
  31     char *serial;
  32     unsigned short sector_mask;
  33     DeviceState *qdev;
  34 } VirtIOBlock;
  35
  36 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
  37 {
  38     return (VirtIOBlock *)vdev;
  39 }
  40
  41 typedef struct VirtIOBlockReq
  42 {
  43     VirtIOBlock *dev;
  44     VirtQueueElement elem;
  45     struct virtio_blk_inhdr *in;
  46     struct virtio_blk_outhdr *out;
  47     struct virtio_scsi_inhdr *scsi;
  48     QEMUIOVector qiov;
  49     struct VirtIOBlockReq *next;
  50     BlockAcctCookie acct;
  51 } VirtIOBlockReq;
  52
  53 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
  54 {
  55     VirtIOBlock *s = req->dev;
  56
  57     trace_virtio_blk_req_complete(req, status);
  58
  59     stb_p(&req->in->status, status);
  60     virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
  61     virtio_notify(&s->vdev, s->vq);
  62 }
  63
  64 static int virtio_blk_handle_rw_error(VirtIOBlockReq *req, int error,
  65     int is_read)
  66 {
  67     BlockErrorAction action = bdrv_get_on_error(req->dev->bs, is_read);
  68     VirtIOBlock *s = req->dev;
  69
  70     if (action == BLOCK_ERR_IGNORE) {
  71         bdrv_mon_event(s->bs, BDRV_ACTION_IGNORE, is_read);
  72         return 0;
  73     }
  74
  75     if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
  76             || action == BLOCK_ERR_STOP_ANY) {
  77         req->next = s->rq;
  78         s->rq = req;
  79         bdrv_mon_event(s->bs, BDRV_ACTION_STOP, is_read);
  80         vm_stop(RUN_STATE_IO_ERROR);
  81         bdrv_iostatus_set_err(s->bs, error);
  82     } else {
  83         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
  84         bdrv_acct_done(s->bs, &req->acct);
  85         g_free(req);
  86         bdrv_mon_event(s->bs, BDRV_ACTION_REPORT, is_read);
  87     }
  88
  89     return 1;
  90 }
  91
  92 static void virtio_blk_rw_complete(void *opaque, int ret)
  93 {
  94     VirtIOBlockReq *req = opaque;
  95
  96     trace_virtio_blk_rw_complete(req, ret);
  97
  98     if (ret) {
  99         int is_read = !(ldl_p(&req->out->type) & VIRTIO_BLK_T_OUT);
 100         if (virtio_blk_handle_rw_error(req, -ret, is_read))
 101             return;
 102     }
 103
 104     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 105     bdrv_acct_done(req->dev->bs, &req->acct);
 106     g_free(req);
 107 }
 108
 109 static void virtio_blk_flush_complete(void *opaque, int ret)
 110 {
 111     VirtIOBlockReq *req = opaque;
 112
 113     if (ret) {
 114         if (virtio_blk_handle_rw_error(req, -ret, 0)) {
 115             return;
 116         }
 117     }
 118
 119     virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 120     bdrv_acct_done(req->dev->bs, &req->acct);
 121     g_free(req);
 122 }
 123
 124 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
 125 {
 126     VirtIOBlockReq *req = g_malloc(sizeof(*req));
 127     req->dev = s;
 128     req->qiov.size = 0;
 129     req->next = NULL;
 130     return req;
 131 }
 132
 133 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
 134 {
 135     VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 136
 137     if (req != NULL) {
 138         if (!virtqueue_pop(s->vq, &req->elem)) {
 139             g_free(req);
 140             return NULL;
 141         }
 142     }
 143
 144     return req;
 145 }
 146
 147 #ifdef __linux__
 148 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 149 {
 150     struct sg_io_hdr hdr;
 151     int ret;
 152     int status;
 153     int i;
 154
 155     /*
 156      * We require at least one output segment each for the virtio_blk_outhdr
 157      * and the SCSI command block.
 158      *
 159      * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
 160      * and the sense buffer pointer in the input segments.
 161      */
 162     if (req->elem.out_num < 2 || req->elem.in_num < 3) {
 163         virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
 164         g_free(req);
 165         return;
 166     }
 167
 168     /*
 169      * No support for bidirection commands yet.
 170      */
 171     if (req->elem.out_num > 2 && req->elem.in_num > 3) {
 172         virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 173         g_free(req);
 174         return;
 175     }
 176
 177     /*
 178      * The scsi inhdr is placed in the second-to-last input segment, just
 179      * before the regular inhdr.
 180      */
 181     req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
 182
 183     memset(&hdr, 0, sizeof(struct sg_io_hdr));
 184     hdr.interface_id = 'S';
 185     hdr.cmd_len = req->elem.out_sg[1].iov_len;
 186     hdr.cmdp = req->elem.out_sg[1].iov_base;
 187     hdr.dxfer_len = 0;
 188
 189     if (req->elem.out_num > 2) {
 190         /*
 191          * If there are more than the minimally required 2 output segments
 192          * there is write payload starting from the third iovec.
 193          */
 194         hdr.dxfer_direction = SG_DXFER_TO_DEV;
 195         hdr.iovec_count = req->elem.out_num - 2;
 196
 197         for (i = 0; i < hdr.iovec_count; i++)
 198             hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
 199
 200         hdr.dxferp = req->elem.out_sg + 2;
 201
 202     } else if (req->elem.in_num > 3) {
 203         /*
 204          * If we have more than 3 input segments the guest wants to actually
 205          * read data.
 206          */
 207         hdr.dxfer_direction = SG_DXFER_FROM_DEV;
 208         hdr.iovec_count = req->elem.in_num - 3;
 209         for (i = 0; i < hdr.iovec_count; i++)
 210             hdr.dxfer_len += req->elem.in_sg[i].iov_len;
 211
 212         hdr.dxferp = req->elem.in_sg;
 213     } else {
 214         /*
 215          * Some SCSI commands don't actually transfer any data.
 216          */
 217         hdr.dxfer_direction = SG_DXFER_NONE;
 218     }
 219
 220     hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
 221     hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
 222
 223     ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
 224     if (ret) {
 225         status = VIRTIO_BLK_S_UNSUPP;
 226         hdr.status = ret;
 227         hdr.resid = hdr.dxfer_len;
 228     } else if (hdr.status) {
 229         status = VIRTIO_BLK_S_IOERR;
 230     } else {
 231         status = VIRTIO_BLK_S_OK;
 232     }
 233
 234     stl_p(&req->scsi->errors, hdr.status);
 235     stl_p(&req->scsi->residual, hdr.resid);
 236     stl_p(&req->scsi->sense_len, hdr.sb_len_wr);
 237     stl_p(&req->scsi->data_len, hdr.dxfer_len);
 238
 239     virtio_blk_req_complete(req, status);
 240     g_free(req);
 241 }
 242 #else
 243 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
 244 {
 245     virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
 246     g_free(req);
 247 }
 248 #endif /* __linux__ */
 249
 250 typedef struct MultiReqBuffer {
 251     BlockRequest        blkreq[32];
 252     unsigned int        num_writes;
 253 } MultiReqBuffer;
 254
 255 static void virtio_submit_multiwrite(BlockDriverState *bs, MultiReqBuffer *mrb)
 256 {
 257     int i, ret;
 258
 259     if (!mrb->num_writes) {
 260         return;
 261     }
 262
 263     ret = bdrv_aio_multiwrite(bs, mrb->blkreq, mrb->num_writes);
 264     if (ret != 0) {
 265         for (i = 0; i < mrb->num_writes; i++) {
 266             if (mrb->blkreq[i].error) {
 267                 virtio_blk_rw_complete(mrb->blkreq[i].opaque, -EIO);
 268             }
 269         }
 270     }
 271
 272     mrb->num_writes = 0;
 273 }
 274
 275 static void virtio_blk_handle_flush(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 276 {
 277     BlockDriverAIOCB *acb;
 278
 279     bdrv_acct_start(req->dev->bs, &req->acct, 0, BDRV_ACCT_FLUSH);
 280
 281     /*
 282      * Make sure all outstanding writes are posted to the backing device.
 283      */
 284     virtio_submit_multiwrite(req->dev->bs, mrb);
 285
 286     acb = bdrv_aio_flush(req->dev->bs, virtio_blk_flush_complete, req);
 287     if (!acb) {
 288         virtio_blk_flush_complete(req, -EIO);
 289     }
 290 }
 291
 292 static void virtio_blk_handle_write(VirtIOBlockReq *req, MultiReqBuffer *mrb)
 293 {
 294     BlockRequest *blkreq;
 295     uint64_t sector;
 296
 297     sector = ldq_p(&req->out->sector);
 298
 299     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_WRITE);
 300
 301     trace_virtio_blk_handle_write(req, sector, req->qiov.size / 512);
 302
 303     if (sector & req->dev->sector_mask) {
 304         virtio_blk_rw_complete(req, -EIO);
 305         return;
 306     }
 307     if (req->qiov.size % req->dev->conf->logical_block_size) {
 308         virtio_blk_rw_complete(req, -EIO);
 309         return;
 310     }
 311
 312     if (mrb->num_writes == 32) {
 313         virtio_submit_multiwrite(req->dev->bs, mrb);
 314     }
 315
 316     blkreq = &mrb->blkreq[mrb->num_writes];
 317     blkreq->sector = sector;
 318     blkreq->nb_sectors = req->qiov.size / BDRV_SECTOR_SIZE;
 319     blkreq->qiov = &req->qiov;
 320     blkreq->cb = virtio_blk_rw_complete;
 321     blkreq->opaque = req;
 322     blkreq->error = 0;
 323
 324     mrb->num_writes++;
 325 }
 326
 327 static void virtio_blk_handle_read(VirtIOBlockReq *req)
 328 {
 329     BlockDriverAIOCB *acb;
 330     uint64_t sector;
 331
 332     sector = ldq_p(&req->out->sector);
 333
 334     bdrv_acct_start(req->dev->bs, &req->acct, req->qiov.size, BDRV_ACCT_READ);
 335
 336     if (sector & req->dev->sector_mask) {
 337         virtio_blk_rw_complete(req, -EIO);
 338         return;
 339     }
 340     if (req->qiov.size % req->dev->conf->logical_block_size) {
 341         virtio_blk_rw_complete(req, -EIO);
 342         return;
 343     }
 344
 345     acb = bdrv_aio_readv(req->dev->bs, sector, &req->qiov,
 346                          req->qiov.size / BDRV_SECTOR_SIZE,
 347                          virtio_blk_rw_complete, req);
 348     if (!acb) {
 349         virtio_blk_rw_complete(req, -EIO);
 350     }
 351 }
 352
 353 static void virtio_blk_handle_request(VirtIOBlockReq *req,
 354     MultiReqBuffer *mrb)
 355 {
 356     uint32_t type;
 357
 358     if (req->elem.out_num < 1 || req->elem.in_num < 1) {
 359         error_report("virtio-blk missing headers");
 360         exit(1);
 361     }
 362
 363     if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
 364         req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
 365         error_report("virtio-blk header not in correct element");
 366         exit(1);
 367     }
 368
 369     req->out = (void *)req->elem.out_sg[0].iov_base;
 370     req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
 371
 372     type = ldl_p(&req->out->type);
 373
 374     if (type & VIRTIO_BLK_T_FLUSH) {
 375         virtio_blk_handle_flush(req, mrb);
 376     } else if (type & VIRTIO_BLK_T_SCSI_CMD) {
 377         virtio_blk_handle_scsi(req);
 378     } else if (type & VIRTIO_BLK_T_GET_ID) {
 379         VirtIOBlock *s = req->dev;
 380
 381         /*
 382          * NB: per existing s/n string convention the string is
 383          * terminated by '\0' only when shorter than buffer.
 384          */
 385         strncpy(req->elem.in_sg[0].iov_base,
 386                 s->serial ? s->serial : "",
 387                 MIN(req->elem.in_sg[0].iov_len, VIRTIO_BLK_ID_BYTES));
 388         virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
 389         g_free(req);
 390     } else if (type & VIRTIO_BLK_T_OUT) {
 391         qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
 392                                  req->elem.out_num - 1);
 393         virtio_blk_handle_write(req, mrb);
 394     } else {
 395         qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
 396                                  req->elem.in_num - 1);
 397         virtio_blk_handle_read(req);
 398     }
 399 }
 400
 401 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
 402 {
 403     VirtIOBlock *s = to_virtio_blk(vdev);
 404     VirtIOBlockReq *req;
 405     MultiReqBuffer mrb = {
 406         .num_writes = 0,
 407     };
 408
 409     while ((req = virtio_blk_get_request(s))) {
 410         virtio_blk_handle_request(req, &mrb);
 411     }
 412
 413     virtio_submit_multiwrite(s->bs, &mrb);
 414
 415     /*
 416      * FIXME: Want to check for completions before returning to guest mode,
 417      * so cached reads and writes are reported as quickly as possible. But
 418      * that should be done in the generic block layer.
 419      */
 420 }
 421
 422 static void virtio_blk_dma_restart_bh(void *opaque)
 423 {
 424     VirtIOBlock *s = opaque;
 425     VirtIOBlockReq *req = s->rq;
 426     MultiReqBuffer mrb = {
 427         .num_writes = 0,
 428     };
 429
 430     qemu_bh_delete(s->bh);
 431     s->bh = NULL;
 432
 433     s->rq = NULL;
 434
 435     while (req) {
 436         virtio_blk_handle_request(req, &mrb);
 437         req = req->next;
 438     }
 439
 440     virtio_submit_multiwrite(s->bs, &mrb);
 441 }
 442
 443 static void virtio_blk_dma_restart_cb(void *opaque, int running,
 444                                       RunState state)
 445 {
 446     VirtIOBlock *s = opaque;
 447
 448     if (!running)
 449         return;
 450
 451     if (!s->bh) {
 452         s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
 453         qemu_bh_schedule(s->bh);
 454     }
 455 }
 456
 457 static void virtio_blk_reset(VirtIODevice *vdev)
 458 {
 459     /*
 460      * This should cancel pending requests, but can't do nicely until there
 461      * are per-device request lists.
 462      */
 463     qemu_aio_flush();
 464 }
 465
 466 /* coalesce internal state, copy to pci i/o region 0
 467  */
 468 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
 469 {
 470     VirtIOBlock *s = to_virtio_blk(vdev);
 471     struct virtio_blk_config blkcfg;
 472     uint64_t capacity;
 473     int cylinders, heads, secs;
 474
 475     bdrv_get_geometry(s->bs, &capacity);
 476     bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
 477     memset(&blkcfg, 0, sizeof(blkcfg));
 478     stq_raw(&blkcfg.capacity, capacity);
 479     stl_raw(&blkcfg.seg_max, 128 - 2);
 480     stw_raw(&blkcfg.cylinders, cylinders);
 481     blkcfg.heads = heads;
 482     blkcfg.sectors = secs & ~s->sector_mask;
 483     blkcfg.blk_size = s->conf->logical_block_size;
 484     blkcfg.size_max = 0;
 485     blkcfg.physical_block_exp = get_physical_block_exp(s->conf);
 486     blkcfg.alignment_offset = 0;
 487     blkcfg.min_io_size = s->conf->min_io_size / blkcfg.blk_size;
 488     blkcfg.opt_io_size = s->conf->opt_io_size / blkcfg.blk_size;
 489     memcpy(config, &blkcfg, sizeof(struct virtio_blk_config));
 490 }
 491
 492 static uint32_t virtio_blk_get_features(VirtIODevice *vdev, uint32_t features)
 493 {
 494     VirtIOBlock *s = to_virtio_blk(vdev);
 495
 496     features |= (1 << VIRTIO_BLK_F_SEG_MAX);
 497     features |= (1 << VIRTIO_BLK_F_GEOMETRY);
 498     features |= (1 << VIRTIO_BLK_F_TOPOLOGY);
 499     features |= (1 << VIRTIO_BLK_F_BLK_SIZE);
 500
 501     if (bdrv_enable_write_cache(s->bs))
 502         features |= (1 << VIRTIO_BLK_F_WCACHE);
 503
 504     if (bdrv_is_read_only(s->bs))
 505         features |= 1 << VIRTIO_BLK_F_RO;
 506
 507     return features;
 508 }
 509
 510 static void virtio_blk_save(QEMUFile *f, void *opaque)
 511 {
 512     VirtIOBlock *s = opaque;
 513     VirtIOBlockReq *req = s->rq;
 514
 515     virtio_save(&s->vdev, f);
 516
 517     while (req) {
 518         qemu_put_sbyte(f, 1);
 519         qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 520         req = req->next;
 521     }
 522     qemu_put_sbyte(f, 0);
 523 }
 524
 525 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
 526 {
 527     VirtIOBlock *s = opaque;
 528
 529     if (version_id != 2)
 530         return -EINVAL;
 531
 532     virtio_load(&s->vdev, f);
 533     while (qemu_get_sbyte(f)) {
 534         VirtIOBlockReq *req = virtio_blk_alloc_request(s);
 535         qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
 536         req->next = s->rq;
 537         s->rq = req;
 538
 539         virtqueue_map_sg(req->elem.in_sg, req->elem.in_addr,
 540             req->elem.in_num, 1);
 541         virtqueue_map_sg(req->elem.out_sg, req->elem.out_addr,
 542             req->elem.out_num, 0);
 543     }
 544
 545     return 0;
 546 }
 547
 548 static void virtio_blk_resize(void *opaque)
 549 {
 550     VirtIOBlock *s = opaque;
 551
 552     virtio_notify_config(&s->vdev);
 553 }
 554
 555 static const BlockDevOps virtio_block_ops = {
 556     .resize_cb = virtio_blk_resize,
 557 };
 558
 559 VirtIODevice *virtio_blk_init(DeviceState *dev, BlockConf *conf,
 560                               char **serial)
 561 {
 562     VirtIOBlock *s;
 563     int cylinders, heads, secs;
 564     static int virtio_blk_id;
 565     DriveInfo *dinfo;
 566
 567     if (!conf->bs) {
 568         error_report("virtio-blk-pci: drive property not set");
 569         return NULL;
 570     }
 571     if (!bdrv_is_inserted(conf->bs)) {
 572         error_report("Device needs media, but drive is empty");
 573         return NULL;
 574     }
 575
 576     if (!*serial) {
 577         /* try to fall back to value set with legacy -drive serial=... */
 578         dinfo = drive_get_by_blockdev(conf->bs);
 579         if (*dinfo->serial) {
 580             *serial = strdup(dinfo->serial);
 581         }
 582     }
 583
 584     s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
 585                                           sizeof(struct virtio_blk_config),
 586                                           sizeof(VirtIOBlock));
 587
 588     s->vdev.get_config = virtio_blk_update_config;
 589     s->vdev.get_features = virtio_blk_get_features;
 590     s->vdev.reset = virtio_blk_reset;
 591     s->bs = conf->bs;
 592     s->conf = conf;
 593     s->serial = *serial;
 594     s->rq = NULL;
 595     s->sector_mask = (s->conf->logical_block_size / BDRV_SECTOR_SIZE) - 1;
 596     bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
 597
 598     s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
 599
 600     qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
 601     s->qdev = dev;
 602     register_savevm(dev, "virtio-blk", virtio_blk_id++, 2,
 603                     virtio_blk_save, virtio_blk_load, s);
 604     bdrv_set_dev_ops(s->bs, &virtio_block_ops, s);
 605     bdrv_set_buffer_alignment(s->bs, conf->logical_block_size);
 606
 607     bdrv_iostatus_enable(s->bs);
 608     add_boot_device_path(conf->bootindex, dev, "/disk@0,0");
 609
 610     return &s->vdev;
 611 }
 612
 613 void virtio_blk_exit(VirtIODevice *vdev)
 614 {
 615     VirtIOBlock *s = to_virtio_blk(vdev);
 616     unregister_savevm(s->qdev, "virtio-blk", s);
 617     virtio_cleanup(vdev);
 618 }