hw/xen_disk.c

   1 /*
   2  *  xen paravirt block device backend
   3  *
   4  *  (c) Gerd Hoffmann <kraxel@redhat.com>
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; under version 2 of the License.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  You should have received a copy of the GNU General Public License along
  16  *  with this program; if not, see <http://www.gnu.org/licenses/>.
  17  */
  18
  19 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <stdarg.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <signal.h>
  25 #include <inttypes.h>
  26 #include <time.h>
  27 #include <fcntl.h>
  28 #include <errno.h>
  29 #include <sys/ioctl.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/mman.h>
  33 #include <sys/uio.h>
  34
  35 #include <xs.h>
  36 #include <xenctrl.h>
  37 #include <xen/io/xenbus.h>
  38
  39 #include "hw.h"
  40 #include "block_int.h"
  41 #include "qemu-char.h"
  42 #include "xen_blkif.h"
  43 #include "xen_backend.h"
  44 #include "blockdev.h"
  45
  46 /* ------------------------------------------------------------- */
  47
  48 static int syncwrite    = 0;
  49 static int batch_maps   = 0;
  50
  51 static int max_requests = 32;
  52 static int use_aio      = 1;
  53
  54 /* ------------------------------------------------------------- */
  55
  56 #define BLOCK_SIZE  512
  57 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  58
  59 struct ioreq {
  60     blkif_request_t     req;
  61     int16_t             status;
  62
  63     /* parsed request */
  64     off_t               start;
  65     QEMUIOVector        v;
  66     int                 presync;
  67     int                 postsync;
  68
  69     /* grant mapping */
  70     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  71     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  72     int                 prot;
  73     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  74     void                *pages;
  75
  76     /* aio status */
  77     int                 aio_inflight;
  78     int                 aio_errors;
  79
  80     struct XenBlkDev    *blkdev;
  81     QLIST_ENTRY(ioreq)   list;
  82     BlockAcctCookie     acct;
  83 };
  84
  85 struct XenBlkDev {
  86     struct XenDevice    xendev;  /* must be first */
  87     char                *params;
  88     char                *mode;
  89     char                *type;
  90     char                *dev;
  91     char                *devtype;
  92     const char          *fileproto;
  93     const char          *filename;
  94     int                 ring_ref;
  95     void                *sring;
  96     int64_t             file_blk;
  97     int64_t             file_size;
  98     int                 protocol;
  99     blkif_back_rings_t  rings;
 100     int                 more_work;
 101     int                 cnt_map;
 102
 103     /* request lists */
 104     QLIST_HEAD(inflight_head, ioreq) inflight;
 105     QLIST_HEAD(finished_head, ioreq) finished;
 106     QLIST_HEAD(freelist_head, ioreq) freelist;
 107     int                 requests_total;
 108     int                 requests_inflight;
 109     int                 requests_finished;
 110
 111     /* qemu block driver */
 112     DriveInfo           *dinfo;
 113     BlockDriverState    *bs;
 114     QEMUBH              *bh;
 115 };
 116
 117 /* ------------------------------------------------------------- */
 118
 119 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 120 {
 121     struct ioreq *ioreq = NULL;
 122
 123     if (QLIST_EMPTY(&blkdev->freelist)) {
 124         if (blkdev->requests_total >= max_requests) {
 125             goto out;
 126         }
 127         /* allocate new struct */
 128         ioreq = g_malloc0(sizeof(*ioreq));
 129         ioreq->blkdev = blkdev;
 130         blkdev->requests_total++;
 131         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 132     } else {
 133         /* get one from freelist */
 134         ioreq = QLIST_FIRST(&blkdev->freelist);
 135         QLIST_REMOVE(ioreq, list);
 136         qemu_iovec_reset(&ioreq->v);
 137     }
 138     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 139     blkdev->requests_inflight++;
 140
 141 out:
 142     return ioreq;
 143 }
 144
 145 static void ioreq_finish(struct ioreq *ioreq)
 146 {
 147     struct XenBlkDev *blkdev = ioreq->blkdev;
 148
 149     QLIST_REMOVE(ioreq, list);
 150     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 151     blkdev->requests_inflight--;
 152     blkdev->requests_finished++;
 153 }
 154
 155 static void ioreq_release(struct ioreq *ioreq)
 156 {
 157     struct XenBlkDev *blkdev = ioreq->blkdev;
 158
 159     QLIST_REMOVE(ioreq, list);
 160     memset(ioreq, 0, sizeof(*ioreq));
 161     ioreq->blkdev = blkdev;
 162     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 163     blkdev->requests_finished--;
 164 }
 165
 166 /*
 167  * translate request into iovec + start offset
 168  * do sanity checks along the way
 169  */
 170 static int ioreq_parse(struct ioreq *ioreq)
 171 {
 172     struct XenBlkDev *blkdev = ioreq->blkdev;
 173     uintptr_t mem;
 174     size_t len;
 175     int i;
 176
 177     xen_be_printf(&blkdev->xendev, 3,
 178                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 179                   ioreq->req.operation, ioreq->req.nr_segments,
 180                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 181     switch (ioreq->req.operation) {
 182     case BLKIF_OP_READ:
 183         ioreq->prot = PROT_WRITE; /* to memory */
 184         break;
 185     case BLKIF_OP_WRITE_BARRIER:
 186         if (!ioreq->req.nr_segments) {
 187             ioreq->presync = 1;
 188             return 0;
 189         }
 190         if (!syncwrite) {
 191             ioreq->presync = ioreq->postsync = 1;
 192         }
 193         /* fall through */
 194     case BLKIF_OP_WRITE:
 195         ioreq->prot = PROT_READ; /* from memory */
 196         if (syncwrite) {
 197             ioreq->postsync = 1;
 198         }
 199         break;
 200     default:
 201         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 202                       ioreq->req.operation);
 203         goto err;
 204     };
 205
 206     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 207         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 208         goto err;
 209     }
 210
 211     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 212     for (i = 0; i < ioreq->req.nr_segments; i++) {
 213         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 214             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 215             goto err;
 216         }
 217         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 218             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 219             goto err;
 220         }
 221         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 222             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 223             goto err;
 224         }
 225
 226         ioreq->domids[i] = blkdev->xendev.dom;
 227         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 228
 229         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 230         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 231         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 232     }
 233     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 234         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 235         goto err;
 236     }
 237     return 0;
 238
 239 err:
 240     ioreq->status = BLKIF_RSP_ERROR;
 241     return -1;
 242 }
 243
 244 static void ioreq_unmap(struct ioreq *ioreq)
 245 {
 246     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 247     int i;
 248
 249     if (ioreq->v.niov == 0) {
 250         return;
 251     }
 252     if (batch_maps) {
 253         if (!ioreq->pages) {
 254             return;
 255         }
 256         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0) {
 257             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 258                           strerror(errno));
 259         }
 260         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 261         ioreq->pages = NULL;
 262     } else {
 263         for (i = 0; i < ioreq->v.niov; i++) {
 264             if (!ioreq->page[i]) {
 265                 continue;
 266             }
 267             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0) {
 268                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 269                               strerror(errno));
 270             }
 271             ioreq->blkdev->cnt_map--;
 272             ioreq->page[i] = NULL;
 273         }
 274     }
 275 }
 276
 277 static int ioreq_map(struct ioreq *ioreq)
 278 {
 279     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 280     int i;
 281
 282     if (ioreq->v.niov == 0) {
 283         return 0;
 284     }
 285     if (batch_maps) {
 286         ioreq->pages = xc_gnttab_map_grant_refs
 287             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 288         if (ioreq->pages == NULL) {
 289             xen_be_printf(&ioreq->blkdev->xendev, 0,
 290                           "can't map %d grant refs (%s, %d maps)\n",
 291                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 292             return -1;
 293         }
 294         for (i = 0; i < ioreq->v.niov; i++) {
 295             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 296                 (uintptr_t)ioreq->v.iov[i].iov_base;
 297         }
 298         ioreq->blkdev->cnt_map += ioreq->v.niov;
 299     } else  {
 300         for (i = 0; i < ioreq->v.niov; i++) {
 301             ioreq->page[i] = xc_gnttab_map_grant_ref
 302                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 303             if (ioreq->page[i] == NULL) {
 304                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 305                               "can't map grant ref %d (%s, %d maps)\n",
 306                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 307                 ioreq_unmap(ioreq);
 308                 return -1;
 309             }
 310             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 311             ioreq->blkdev->cnt_map++;
 312         }
 313     }
 314     return 0;
 315 }
 316
 317 static int ioreq_runio_qemu_sync(struct ioreq *ioreq)
 318 {
 319     struct XenBlkDev *blkdev = ioreq->blkdev;
 320     int i, rc;
 321     off_t pos;
 322
 323     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 324         goto err_no_map;
 325     }
 326     if (ioreq->presync) {
 327         bdrv_flush(blkdev->bs);
 328     }
 329
 330     switch (ioreq->req.operation) {
 331     case BLKIF_OP_READ:
 332         pos = ioreq->start;
 333         for (i = 0; i < ioreq->v.niov; i++) {
 334             rc = bdrv_read(blkdev->bs, pos / BLOCK_SIZE,
 335                            ioreq->v.iov[i].iov_base,
 336                            ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 337             if (rc != 0) {
 338                 xen_be_printf(&blkdev->xendev, 0, "rd I/O error (%p, len %zd)\n",
 339                               ioreq->v.iov[i].iov_base,
 340                               ioreq->v.iov[i].iov_len);
 341                 goto err;
 342             }
 343             pos += ioreq->v.iov[i].iov_len;
 344         }
 345         break;
 346     case BLKIF_OP_WRITE:
 347     case BLKIF_OP_WRITE_BARRIER:
 348         if (!ioreq->req.nr_segments) {
 349             break;
 350         }
 351         pos = ioreq->start;
 352         for (i = 0; i < ioreq->v.niov; i++) {
 353             rc = bdrv_write(blkdev->bs, pos / BLOCK_SIZE,
 354                             ioreq->v.iov[i].iov_base,
 355                             ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 356             if (rc != 0) {
 357                 xen_be_printf(&blkdev->xendev, 0, "wr I/O error (%p, len %zd)\n",
 358                               ioreq->v.iov[i].iov_base,
 359                               ioreq->v.iov[i].iov_len);
 360                 goto err;
 361             }
 362             pos += ioreq->v.iov[i].iov_len;
 363         }
 364         break;
 365     default:
 366         /* unknown operation (shouldn't happen -- parse catches this) */
 367         goto err;
 368     }
 369
 370     if (ioreq->postsync) {
 371         bdrv_flush(blkdev->bs);
 372     }
 373     ioreq->status = BLKIF_RSP_OKAY;
 374
 375     ioreq_unmap(ioreq);
 376     ioreq_finish(ioreq);
 377     return 0;
 378
 379 err:
 380     ioreq_unmap(ioreq);
 381 err_no_map:
 382     ioreq_finish(ioreq);
 383     ioreq->status = BLKIF_RSP_ERROR;
 384     return -1;
 385 }
 386
 387 static void qemu_aio_complete(void *opaque, int ret)
 388 {
 389     struct ioreq *ioreq = opaque;
 390
 391     if (ret != 0) {
 392         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 393                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 394         ioreq->aio_errors++;
 395     }
 396
 397     ioreq->aio_inflight--;
 398     if (ioreq->aio_inflight > 0) {
 399         return;
 400     }
 401
 402     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 403     ioreq_unmap(ioreq);
 404     ioreq_finish(ioreq);
 405     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 406     qemu_bh_schedule(ioreq->blkdev->bh);
 407 }
 408
 409 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 410 {
 411     struct XenBlkDev *blkdev = ioreq->blkdev;
 412
 413     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 414         goto err_no_map;
 415     }
 416
 417     ioreq->aio_inflight++;
 418     if (ioreq->presync) {
 419         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 420     }
 421
 422     switch (ioreq->req.operation) {
 423     case BLKIF_OP_READ:
 424         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 425         ioreq->aio_inflight++;
 426         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 427                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 428                        qemu_aio_complete, ioreq);
 429         break;
 430     case BLKIF_OP_WRITE:
 431     case BLKIF_OP_WRITE_BARRIER:
 432         if (!ioreq->req.nr_segments) {
 433             break;
 434         }
 435
 436         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 437         ioreq->aio_inflight++;
 438         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 439                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 440                         qemu_aio_complete, ioreq);
 441         break;
 442     default:
 443         /* unknown operation (shouldn't happen -- parse catches this) */
 444         goto err;
 445     }
 446
 447     if (ioreq->postsync) {
 448         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 449     }
 450     qemu_aio_complete(ioreq, 0);
 451
 452     return 0;
 453
 454 err:
 455     ioreq_unmap(ioreq);
 456 err_no_map:
 457     ioreq_finish(ioreq);
 458     ioreq->status = BLKIF_RSP_ERROR;
 459     return -1;
 460 }
 461
 462 static int blk_send_response_one(struct ioreq *ioreq)
 463 {
 464     struct XenBlkDev  *blkdev = ioreq->blkdev;
 465     int               send_notify   = 0;
 466     int               have_requests = 0;
 467     blkif_response_t  resp;
 468     void              *dst;
 469
 470     resp.id        = ioreq->req.id;
 471     resp.operation = ioreq->req.operation;
 472     resp.status    = ioreq->status;
 473
 474     /* Place on the response ring for the relevant domain. */
 475     switch (blkdev->protocol) {
 476     case BLKIF_PROTOCOL_NATIVE:
 477         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 478         break;
 479     case BLKIF_PROTOCOL_X86_32:
 480         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 481                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 482         break;
 483     case BLKIF_PROTOCOL_X86_64:
 484         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 485                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 486         break;
 487     default:
 488         dst = NULL;
 489     }
 490     memcpy(dst, &resp, sizeof(resp));
 491     blkdev->rings.common.rsp_prod_pvt++;
 492
 493     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 494     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 495         /*
 496          * Tail check for pending requests. Allows frontend to avoid
 497          * notifications if requests are already in flight (lower
 498          * overheads and promotes batching).
 499          */
 500         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 501     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 502         have_requests = 1;
 503     }
 504
 505     if (have_requests) {
 506         blkdev->more_work++;
 507     }
 508     return send_notify;
 509 }
 510
 511 /* walk finished list, send outstanding responses, free requests */
 512 static void blk_send_response_all(struct XenBlkDev *blkdev)
 513 {
 514     struct ioreq *ioreq;
 515     int send_notify = 0;
 516
 517     while (!QLIST_EMPTY(&blkdev->finished)) {
 518         ioreq = QLIST_FIRST(&blkdev->finished);
 519         send_notify += blk_send_response_one(ioreq);
 520         ioreq_release(ioreq);
 521     }
 522     if (send_notify) {
 523         xen_be_send_notify(&blkdev->xendev);
 524     }
 525 }
 526
 527 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 528 {
 529     switch (blkdev->protocol) {
 530     case BLKIF_PROTOCOL_NATIVE:
 531         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 532                sizeof(ioreq->req));
 533         break;
 534     case BLKIF_PROTOCOL_X86_32:
 535         blkif_get_x86_32_req(&ioreq->req,
 536                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 537         break;
 538     case BLKIF_PROTOCOL_X86_64:
 539         blkif_get_x86_64_req(&ioreq->req,
 540                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 541         break;
 542     }
 543     return 0;
 544 }
 545
 546 static void blk_handle_requests(struct XenBlkDev *blkdev)
 547 {
 548     RING_IDX rc, rp;
 549     struct ioreq *ioreq;
 550
 551     blkdev->more_work = 0;
 552
 553     rc = blkdev->rings.common.req_cons;
 554     rp = blkdev->rings.common.sring->req_prod;
 555     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 556
 557     if (use_aio) {
 558         blk_send_response_all(blkdev);
 559     }
 560     while (rc != rp) {
 561         /* pull request from ring */
 562         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 563             break;
 564         }
 565         ioreq = ioreq_start(blkdev);
 566         if (ioreq == NULL) {
 567             blkdev->more_work++;
 568             break;
 569         }
 570         blk_get_request(blkdev, ioreq, rc);
 571         blkdev->rings.common.req_cons = ++rc;
 572
 573         /* parse them */
 574         if (ioreq_parse(ioreq) != 0) {
 575             if (blk_send_response_one(ioreq)) {
 576                 xen_be_send_notify(&blkdev->xendev);
 577             }
 578             ioreq_release(ioreq);
 579             continue;
 580         }
 581
 582         if (use_aio) {
 583             /* run i/o in aio mode */
 584             ioreq_runio_qemu_aio(ioreq);
 585         } else {
 586             /* run i/o in sync mode */
 587             ioreq_runio_qemu_sync(ioreq);
 588         }
 589     }
 590     if (!use_aio) {
 591         blk_send_response_all(blkdev);
 592     }
 593
 594     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 595         qemu_bh_schedule(blkdev->bh);
 596     }
 597 }
 598
 599 /* ------------------------------------------------------------- */
 600
 601 static void blk_bh(void *opaque)
 602 {
 603     struct XenBlkDev *blkdev = opaque;
 604     blk_handle_requests(blkdev);
 605 }
 606
 607 static void blk_alloc(struct XenDevice *xendev)
 608 {
 609     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 610
 611     QLIST_INIT(&blkdev->inflight);
 612     QLIST_INIT(&blkdev->finished);
 613     QLIST_INIT(&blkdev->freelist);
 614     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 615     if (xen_mode != XEN_EMULATE) {
 616         batch_maps = 1;
 617     }
 618 }
 619
 620 static int blk_init(struct XenDevice *xendev)
 621 {
 622     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 623     int index, qflags, have_barriers, info = 0;
 624
 625     /* read xenstore entries */
 626     if (blkdev->params == NULL) {
 627         char *h = NULL;
 628         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 629         if (blkdev->params != NULL) {
 630             h = strchr(blkdev->params, ':');
 631         }
 632         if (h != NULL) {
 633             blkdev->fileproto = blkdev->params;
 634             blkdev->filename  = h+1;
 635             *h = 0;
 636         } else {
 637             blkdev->fileproto = "<unset>";
 638             blkdev->filename  = blkdev->params;
 639         }
 640     }
 641     if (!strcmp("aio", blkdev->fileproto)) {
 642         blkdev->fileproto = "raw";
 643     }
 644     if (blkdev->mode == NULL) {
 645         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 646     }
 647     if (blkdev->type == NULL) {
 648         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 649     }
 650     if (blkdev->dev == NULL) {
 651         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 652     }
 653     if (blkdev->devtype == NULL) {
 654         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 655     }
 656
 657     /* do we have all we need? */
 658     if (blkdev->params == NULL ||
 659         blkdev->mode == NULL   ||
 660         blkdev->type == NULL   ||
 661         blkdev->dev == NULL) {
 662         goto out_error;
 663     }
 664
 665     /* read-only ? */
 666     if (strcmp(blkdev->mode, "w") == 0) {
 667         qflags = BDRV_O_RDWR;
 668     } else {
 669         qflags = 0;
 670         info  |= VDISK_READONLY;
 671     }
 672
 673     /* cdrom ? */
 674     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 675         info  |= VDISK_CDROM;
 676     }
 677
 678     /* init qemu block driver */
 679     index = (blkdev->xendev.dev - 202 * 256) / 16;
 680     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 681     if (!blkdev->dinfo) {
 682         /* setup via xenbus -> create new block driver instance */
 683         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 684         blkdev->bs = bdrv_new(blkdev->dev);
 685         if (blkdev->bs) {
 686             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 687                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 688                 bdrv_delete(blkdev->bs);
 689                 blkdev->bs = NULL;
 690             }
 691         }
 692         if (!blkdev->bs) {
 693             goto out_error;
 694         }
 695     } else {
 696         /* setup via qemu cmdline -> already setup for us */
 697         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 698         blkdev->bs = blkdev->dinfo->bdrv;
 699     }
 700     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 701     blkdev->file_blk  = BLOCK_SIZE;
 702     blkdev->file_size = bdrv_getlength(blkdev->bs);
 703     if (blkdev->file_size < 0) {
 704         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 705                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 706                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 707         blkdev->file_size = 0;
 708     }
 709     have_barriers = blkdev->bs->drv && blkdev->bs->drv->bdrv_flush ? 1 : 0;
 710
 711     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 712                   " size %" PRId64 " (%" PRId64 " MB)\n",
 713                   blkdev->type, blkdev->fileproto, blkdev->filename,
 714                   blkdev->file_size, blkdev->file_size >> 20);
 715
 716     /* fill info */
 717     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", have_barriers);
 718     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 719     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 720     xenstore_write_be_int(&blkdev->xendev, "sectors",
 721                           blkdev->file_size / blkdev->file_blk);
 722     return 0;
 723
 724 out_error:
 725     g_free(blkdev->params);
 726     blkdev->params = NULL;
 727     g_free(blkdev->mode);
 728     blkdev->mode = NULL;
 729     g_free(blkdev->type);
 730     blkdev->type = NULL;
 731     g_free(blkdev->dev);
 732     blkdev->dev = NULL;
 733     g_free(blkdev->devtype);
 734     blkdev->devtype = NULL;
 735     return -1;
 736 }
 737
 738 static int blk_connect(struct XenDevice *xendev)
 739 {
 740     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 741
 742     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 743         return -1;
 744     }
 745     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 746                              &blkdev->xendev.remote_port) == -1) {
 747         return -1;
 748     }
 749
 750     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 751     if (blkdev->xendev.protocol) {
 752         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 753             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 754         }
 755         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 756             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 757         }
 758     }
 759
 760     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 761                                             blkdev->xendev.dom,
 762                                             blkdev->ring_ref,
 763                                             PROT_READ | PROT_WRITE);
 764     if (!blkdev->sring) {
 765         return -1;
 766     }
 767     blkdev->cnt_map++;
 768
 769     switch (blkdev->protocol) {
 770     case BLKIF_PROTOCOL_NATIVE:
 771     {
 772         blkif_sring_t *sring_native = blkdev->sring;
 773         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 774         break;
 775     }
 776     case BLKIF_PROTOCOL_X86_32:
 777     {
 778         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 779
 780         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 781         break;
 782     }
 783     case BLKIF_PROTOCOL_X86_64:
 784     {
 785         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 786
 787         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 788         break;
 789     }
 790     }
 791
 792     xen_be_bind_evtchn(&blkdev->xendev);
 793
 794     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 795                   "remote port %d, local port %d\n",
 796                   blkdev->xendev.protocol, blkdev->ring_ref,
 797                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 798     return 0;
 799 }
 800
 801 static void blk_disconnect(struct XenDevice *xendev)
 802 {
 803     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 804
 805     if (blkdev->bs) {
 806         if (!blkdev->dinfo) {
 807             /* close/delete only if we created it ourself */
 808             bdrv_close(blkdev->bs);
 809             bdrv_delete(blkdev->bs);
 810         }
 811         blkdev->bs = NULL;
 812     }
 813     xen_be_unbind_evtchn(&blkdev->xendev);
 814
 815     if (blkdev->sring) {
 816         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 817         blkdev->cnt_map--;
 818         blkdev->sring = NULL;
 819     }
 820 }
 821
 822 static int blk_free(struct XenDevice *xendev)
 823 {
 824     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 825     struct ioreq *ioreq;
 826
 827     while (!QLIST_EMPTY(&blkdev->freelist)) {
 828         ioreq = QLIST_FIRST(&blkdev->freelist);
 829         QLIST_REMOVE(ioreq, list);
 830         qemu_iovec_destroy(&ioreq->v);
 831         g_free(ioreq);
 832     }
 833
 834     g_free(blkdev->params);
 835     g_free(blkdev->mode);
 836     g_free(blkdev->type);
 837     g_free(blkdev->dev);
 838     g_free(blkdev->devtype);
 839     qemu_bh_delete(blkdev->bh);
 840     return 0;
 841 }
 842
 843 static void blk_event(struct XenDevice *xendev)
 844 {
 845     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 846
 847     qemu_bh_schedule(blkdev->bh);
 848 }
 849
 850 struct XenDevOps xen_blkdev_ops = {
 851     .size       = sizeof(struct XenBlkDev),
 852     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 853     .alloc      = blk_alloc,
 854     .init       = blk_init,
 855     .initialise    = blk_connect,
 856     .disconnect = blk_disconnect,
 857     .event      = blk_event,
 858     .free       = blk_free,
 859 };