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