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
  45 /* ------------------------------------------------------------- */
  46
  47 static int syncwrite    = 0;
  48 static int batch_maps   = 0;
  49
  50 static int max_requests = 32;
  51 static int use_aio      = 1;
  52
  53 /* ------------------------------------------------------------- */
  54
  55 #define BLOCK_SIZE  512
  56 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  57
  58 struct ioreq {
  59     blkif_request_t     req;
  60     int16_t             status;
  61
  62     /* parsed request */
  63     off_t               start;
  64     QEMUIOVector        v;
  65     int                 presync;
  66     int                 postsync;
  67
  68     /* grant mapping */
  69     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  70     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  71     int                 prot;
  72     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  73     void                *pages;
  74
  75     /* aio status */
  76     int                 aio_inflight;
  77     int                 aio_errors;
  78
  79     struct XenBlkDev    *blkdev;
  80     QLIST_ENTRY(ioreq)   list;
  81 };
  82
  83 struct XenBlkDev {
  84     struct XenDevice    xendev;  /* must be first */
  85     char                *params;
  86     char                *mode;
  87     char                *type;
  88     char                *dev;
  89     char                *devtype;
  90     const char          *fileproto;
  91     const char          *filename;
  92     int                 ring_ref;
  93     void                *sring;
  94     int64_t             file_blk;
  95     int64_t             file_size;
  96     int                 protocol;
  97     blkif_back_rings_t  rings;
  98     int                 more_work;
  99     int                 cnt_map;
 100
 101     /* request lists */
 102     QLIST_HEAD(inflight_head, ioreq) inflight;
 103     QLIST_HEAD(finished_head, ioreq) finished;
 104     QLIST_HEAD(freelist_head, ioreq) freelist;
 105     int                 requests_total;
 106     int                 requests_inflight;
 107     int                 requests_finished;
 108
 109     /* qemu block driver */
 110     DriveInfo           *dinfo;
 111     BlockDriverState    *bs;
 112     QEMUBH              *bh;
 113 };
 114
 115 /* ------------------------------------------------------------- */
 116
 117 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 118 {
 119     struct ioreq *ioreq = NULL;
 120
 121     if (QLIST_EMPTY(&blkdev->freelist)) {
 122         if (blkdev->requests_total >= max_requests)
 123             goto out;
 124         /* allocate new struct */
 125         ioreq = qemu_mallocz(sizeof(*ioreq));
 126         ioreq->blkdev = blkdev;
 127         blkdev->requests_total++;
 128         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 129     } else {
 130         /* get one from freelist */
 131         ioreq = QLIST_FIRST(&blkdev->freelist);
 132         QLIST_REMOVE(ioreq, list);
 133         qemu_iovec_reset(&ioreq->v);
 134     }
 135     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 136     blkdev->requests_inflight++;
 137
 138 out:
 139     return ioreq;
 140 }
 141
 142 static void ioreq_finish(struct ioreq *ioreq)
 143 {
 144     struct XenBlkDev *blkdev = ioreq->blkdev;
 145
 146     QLIST_REMOVE(ioreq, list);
 147     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 148     blkdev->requests_inflight--;
 149     blkdev->requests_finished++;
 150 }
 151
 152 static void ioreq_release(struct ioreq *ioreq)
 153 {
 154     struct XenBlkDev *blkdev = ioreq->blkdev;
 155
 156     QLIST_REMOVE(ioreq, list);
 157     memset(ioreq, 0, sizeof(*ioreq));
 158     ioreq->blkdev = blkdev;
 159     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 160     blkdev->requests_finished--;
 161 }
 162
 163 /*
 164  * translate request into iovec + start offset
 165  * do sanity checks along the way
 166  */
 167 static int ioreq_parse(struct ioreq *ioreq)
 168 {
 169     struct XenBlkDev *blkdev = ioreq->blkdev;
 170     uintptr_t mem;
 171     size_t len;
 172     int i;
 173
 174     xen_be_printf(&blkdev->xendev, 3,
 175                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 176                   ioreq->req.operation, ioreq->req.nr_segments,
 177                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 178     switch (ioreq->req.operation) {
 179     case BLKIF_OP_READ:
 180         ioreq->prot = PROT_WRITE; /* to memory */
 181         break;
 182     case BLKIF_OP_WRITE_BARRIER:
 183         if (!syncwrite)
 184             ioreq->presync = ioreq->postsync = 1;
 185         /* fall through */
 186     case BLKIF_OP_WRITE:
 187         ioreq->prot = PROT_READ; /* from memory */
 188         if (syncwrite)
 189             ioreq->postsync = 1;
 190         break;
 191     default:
 192         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 193                       ioreq->req.operation);
 194         goto err;
 195     };
 196
 197     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 198         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 199         goto err;
 200     }
 201
 202     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 203     for (i = 0; i < ioreq->req.nr_segments; i++) {
 204         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 205             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 206             goto err;
 207         }
 208         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 209             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 210             goto err;
 211         }
 212         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 213             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 214             goto err;
 215         }
 216
 217         ioreq->domids[i] = blkdev->xendev.dom;
 218         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 219
 220         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 221         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 222         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 223     }
 224     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 225         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 226         goto err;
 227     }
 228     return 0;
 229
 230 err:
 231     ioreq->status = BLKIF_RSP_ERROR;
 232     return -1;
 233 }
 234
 235 static void ioreq_unmap(struct ioreq *ioreq)
 236 {
 237     int gnt = ioreq->blkdev->xendev.gnttabdev;
 238     int i;
 239
 240     if (ioreq->v.niov == 0)
 241         return;
 242     if (batch_maps) {
 243         if (!ioreq->pages)
 244             return;
 245         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0)
 246             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 247                           strerror(errno));
 248         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 249         ioreq->pages = NULL;
 250     } else {
 251         for (i = 0; i < ioreq->v.niov; i++) {
 252             if (!ioreq->page[i])
 253                 continue;
 254             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0)
 255                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 256                               strerror(errno));
 257             ioreq->blkdev->cnt_map--;
 258             ioreq->page[i] = NULL;
 259         }
 260     }
 261 }
 262
 263 static int ioreq_map(struct ioreq *ioreq)
 264 {
 265     int gnt = ioreq->blkdev->xendev.gnttabdev;
 266     int i;
 267
 268     if (ioreq->v.niov == 0)
 269         return 0;
 270     if (batch_maps) {
 271         ioreq->pages = xc_gnttab_map_grant_refs
 272             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 273         if (ioreq->pages == NULL) {
 274             xen_be_printf(&ioreq->blkdev->xendev, 0,
 275                           "can't map %d grant refs (%s, %d maps)\n",
 276                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 277             return -1;
 278         }
 279         for (i = 0; i < ioreq->v.niov; i++)
 280             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 281                 (uintptr_t)ioreq->v.iov[i].iov_base;
 282         ioreq->blkdev->cnt_map += ioreq->v.niov;
 283     } else  {
 284         for (i = 0; i < ioreq->v.niov; i++) {
 285             ioreq->page[i] = xc_gnttab_map_grant_ref
 286                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 287             if (ioreq->page[i] == NULL) {
 288                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 289                               "can't map grant ref %d (%s, %d maps)\n",
 290                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 291                 ioreq_unmap(ioreq);
 292                 return -1;
 293             }
 294             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 295             ioreq->blkdev->cnt_map++;
 296         }
 297     }
 298     return 0;
 299 }
 300
 301 static int ioreq_runio_qemu_sync(struct ioreq *ioreq)
 302 {
 303     struct XenBlkDev *blkdev = ioreq->blkdev;
 304     int i, rc, len = 0;
 305     off_t pos;
 306
 307     if (ioreq_map(ioreq) == -1)
 308         goto err;
 309     if (ioreq->presync)
 310         bdrv_flush(blkdev->bs);
 311
 312     switch (ioreq->req.operation) {
 313     case BLKIF_OP_READ:
 314         pos = ioreq->start;
 315         for (i = 0; i < ioreq->v.niov; i++) {
 316             rc = bdrv_read(blkdev->bs, pos / BLOCK_SIZE,
 317                            ioreq->v.iov[i].iov_base,
 318                            ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 319             if (rc != 0) {
 320                 xen_be_printf(&blkdev->xendev, 0, "rd I/O error (%p, len %zd)\n",
 321                               ioreq->v.iov[i].iov_base,
 322                               ioreq->v.iov[i].iov_len);
 323                 goto err;
 324             }
 325             len += ioreq->v.iov[i].iov_len;
 326             pos += ioreq->v.iov[i].iov_len;
 327         }
 328         break;
 329     case BLKIF_OP_WRITE:
 330     case BLKIF_OP_WRITE_BARRIER:
 331         pos = ioreq->start;
 332         for (i = 0; i < ioreq->v.niov; i++) {
 333             rc = bdrv_write(blkdev->bs, pos / BLOCK_SIZE,
 334                             ioreq->v.iov[i].iov_base,
 335                             ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 336             if (rc != 0) {
 337                 xen_be_printf(&blkdev->xendev, 0, "wr I/O error (%p, len %zd)\n",
 338                               ioreq->v.iov[i].iov_base,
 339                               ioreq->v.iov[i].iov_len);
 340                 goto err;
 341             }
 342             len += ioreq->v.iov[i].iov_len;
 343             pos += ioreq->v.iov[i].iov_len;
 344         }
 345         break;
 346     default:
 347         /* unknown operation (shouldn't happen -- parse catches this) */
 348         goto err;
 349     }
 350
 351     if (ioreq->postsync)
 352         bdrv_flush(blkdev->bs);
 353     ioreq->status = BLKIF_RSP_OKAY;
 354
 355     ioreq_unmap(ioreq);
 356     ioreq_finish(ioreq);
 357     return 0;
 358
 359 err:
 360     ioreq->status = BLKIF_RSP_ERROR;
 361     return -1;
 362 }
 363
 364 static void qemu_aio_complete(void *opaque, int ret)
 365 {
 366     struct ioreq *ioreq = opaque;
 367
 368     if (ret != 0) {
 369         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 370                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 371         ioreq->aio_errors++;
 372     }
 373
 374     ioreq->aio_inflight--;
 375     if (ioreq->aio_inflight > 0)
 376         return;
 377
 378     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 379     ioreq_unmap(ioreq);
 380     ioreq_finish(ioreq);
 381     qemu_bh_schedule(ioreq->blkdev->bh);
 382 }
 383
 384 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 385 {
 386     struct XenBlkDev *blkdev = ioreq->blkdev;
 387
 388     if (ioreq_map(ioreq) == -1)
 389         goto err;
 390
 391     ioreq->aio_inflight++;
 392     if (ioreq->presync)
 393         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 394
 395     switch (ioreq->req.operation) {
 396     case BLKIF_OP_READ:
 397         ioreq->aio_inflight++;
 398         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 399                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 400                        qemu_aio_complete, ioreq);
 401         break;
 402     case BLKIF_OP_WRITE:
 403     case BLKIF_OP_WRITE_BARRIER:
 404         ioreq->aio_inflight++;
 405         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 406                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 407                         qemu_aio_complete, ioreq);
 408         break;
 409     default:
 410         /* unknown operation (shouldn't happen -- parse catches this) */
 411         goto err;
 412     }
 413
 414     if (ioreq->postsync)
 415         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 416     qemu_aio_complete(ioreq, 0);
 417
 418     return 0;
 419
 420 err:
 421     ioreq->status = BLKIF_RSP_ERROR;
 422     return -1;
 423 }
 424
 425 static int blk_send_response_one(struct ioreq *ioreq)
 426 {
 427     struct XenBlkDev  *blkdev = ioreq->blkdev;
 428     int               send_notify   = 0;
 429     int               have_requests = 0;
 430     blkif_response_t  resp;
 431     void              *dst;
 432
 433     resp.id        = ioreq->req.id;
 434     resp.operation = ioreq->req.operation;
 435     resp.status    = ioreq->status;
 436
 437     /* Place on the response ring for the relevant domain. */
 438     switch (blkdev->protocol) {
 439     case BLKIF_PROTOCOL_NATIVE:
 440         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 441         break;
 442     case BLKIF_PROTOCOL_X86_32:
 443         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 444                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 445         break;
 446     case BLKIF_PROTOCOL_X86_64:
 447         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 448                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 449         break;
 450     default:
 451         dst = NULL;
 452     }
 453     memcpy(dst, &resp, sizeof(resp));
 454     blkdev->rings.common.rsp_prod_pvt++;
 455
 456     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 457     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 458         /*
 459          * Tail check for pending requests. Allows frontend to avoid
 460          * notifications if requests are already in flight (lower
 461          * overheads and promotes batching).
 462          */
 463         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 464     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 465         have_requests = 1;
 466     }
 467
 468     if (have_requests)
 469         blkdev->more_work++;
 470     return send_notify;
 471 }
 472
 473 /* walk finished list, send outstanding responses, free requests */
 474 static void blk_send_response_all(struct XenBlkDev *blkdev)
 475 {
 476     struct ioreq *ioreq;
 477     int send_notify = 0;
 478
 479     while (!QLIST_EMPTY(&blkdev->finished)) {
 480         ioreq = QLIST_FIRST(&blkdev->finished);
 481         send_notify += blk_send_response_one(ioreq);
 482         ioreq_release(ioreq);
 483     }
 484     if (send_notify)
 485         xen_be_send_notify(&blkdev->xendev);
 486 }
 487
 488 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 489 {
 490     switch (blkdev->protocol) {
 491     case BLKIF_PROTOCOL_NATIVE:
 492         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 493                sizeof(ioreq->req));
 494         break;
 495     case BLKIF_PROTOCOL_X86_32:
 496         blkif_get_x86_32_req(&ioreq->req,
 497                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 498         break;
 499     case BLKIF_PROTOCOL_X86_64:
 500         blkif_get_x86_64_req(&ioreq->req,
 501                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 502         break;
 503     }
 504     return 0;
 505 }
 506
 507 static void blk_handle_requests(struct XenBlkDev *blkdev)
 508 {
 509     RING_IDX rc, rp;
 510     struct ioreq *ioreq;
 511
 512     blkdev->more_work = 0;
 513
 514     rc = blkdev->rings.common.req_cons;
 515     rp = blkdev->rings.common.sring->req_prod;
 516     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 517
 518     if (use_aio)
 519         blk_send_response_all(blkdev);
 520     while (rc != rp) {
 521         /* pull request from ring */
 522         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc))
 523             break;
 524         ioreq = ioreq_start(blkdev);
 525         if (ioreq == NULL) {
 526             blkdev->more_work++;
 527             break;
 528         }
 529         blk_get_request(blkdev, ioreq, rc);
 530         blkdev->rings.common.req_cons = ++rc;
 531
 532         /* parse them */
 533         if (ioreq_parse(ioreq) != 0) {
 534             if (blk_send_response_one(ioreq))
 535                 xen_be_send_notify(&blkdev->xendev);
 536             ioreq_release(ioreq);
 537             continue;
 538         }
 539
 540         if (use_aio) {
 541             /* run i/o in aio mode */
 542             ioreq_runio_qemu_aio(ioreq);
 543         } else {
 544             /* run i/o in sync mode */
 545             ioreq_runio_qemu_sync(ioreq);
 546         }
 547     }
 548     if (!use_aio)
 549         blk_send_response_all(blkdev);
 550
 551     if (blkdev->more_work && blkdev->requests_inflight < max_requests)
 552         qemu_bh_schedule(blkdev->bh);
 553 }
 554
 555 /* ------------------------------------------------------------- */
 556
 557 static void blk_bh(void *opaque)
 558 {
 559     struct XenBlkDev *blkdev = opaque;
 560     blk_handle_requests(blkdev);
 561 }
 562
 563 static void blk_alloc(struct XenDevice *xendev)
 564 {
 565     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 566
 567     QLIST_INIT(&blkdev->inflight);
 568     QLIST_INIT(&blkdev->finished);
 569     QLIST_INIT(&blkdev->freelist);
 570     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 571     if (xen_mode != XEN_EMULATE)
 572         batch_maps = 1;
 573 }
 574
 575 static int blk_init(struct XenDevice *xendev)
 576 {
 577     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 578     int index, mode, qflags, have_barriers, info = 0;
 579     char *h;
 580
 581     /* read xenstore entries */
 582     if (blkdev->params == NULL) {
 583         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 584         h = strchr(blkdev->params, ':');
 585         if (h != NULL) {
 586             blkdev->fileproto = blkdev->params;
 587             blkdev->filename  = h+1;
 588             *h = 0;
 589         } else {
 590             blkdev->fileproto = "<unset>";
 591             blkdev->filename  = blkdev->params;
 592         }
 593     }
 594     if (blkdev->mode == NULL)
 595         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 596     if (blkdev->type == NULL)
 597         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 598     if (blkdev->dev == NULL)
 599         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 600     if (blkdev->devtype == NULL)
 601         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 602
 603     /* do we have all we need? */
 604     if (blkdev->params == NULL ||
 605         blkdev->mode == NULL   ||
 606         blkdev->type == NULL   ||
 607         blkdev->dev == NULL)
 608         return -1;
 609
 610     /* read-only ? */
 611     if (strcmp(blkdev->mode, "w") == 0) {
 612         mode   = O_RDWR;
 613         qflags = BDRV_O_RDWR;
 614     } else {
 615         mode   = O_RDONLY;
 616         qflags = BDRV_O_RDONLY;
 617         info  |= VDISK_READONLY;
 618     }
 619
 620     /* cdrom ? */
 621     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom"))
 622         info  |= VDISK_CDROM;
 623
 624     /* init qemu block driver */
 625     index = (blkdev->xendev.dev - 202 * 256) / 16;
 626     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 627     if (!blkdev->dinfo) {
 628         /* setup via xenbus -> create new block driver instance */
 629         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 630         blkdev->bs = bdrv_new(blkdev->dev);
 631         if (blkdev->bs) {
 632             if (bdrv_open2(blkdev->bs, blkdev->filename, qflags,
 633                            bdrv_find_format(blkdev->fileproto)) != 0) {
 634                 bdrv_delete(blkdev->bs);
 635                 blkdev->bs = NULL;
 636             }
 637         }
 638         if (!blkdev->bs)
 639             return -1;
 640     } else {
 641         /* setup via qemu cmdline -> already setup for us */
 642         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 643         blkdev->bs = blkdev->dinfo->bdrv;
 644     }
 645     blkdev->file_blk  = BLOCK_SIZE;
 646     blkdev->file_size = bdrv_getlength(blkdev->bs);
 647     if (blkdev->file_size < 0) {
 648         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 649                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 650                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 651         blkdev->file_size = 0;
 652     }
 653     have_barriers = blkdev->bs->drv && blkdev->bs->drv->bdrv_flush ? 1 : 0;
 654
 655     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 656                   " size %" PRId64 " (%" PRId64 " MB)\n",
 657                   blkdev->type, blkdev->fileproto, blkdev->filename,
 658                   blkdev->file_size, blkdev->file_size >> 20);
 659
 660     /* fill info */
 661     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", have_barriers);
 662     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 663     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 664     xenstore_write_be_int(&blkdev->xendev, "sectors",
 665                           blkdev->file_size / blkdev->file_blk);
 666     return 0;
 667 }
 668
 669 static int blk_connect(struct XenDevice *xendev)
 670 {
 671     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 672
 673     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1)
 674         return -1;
 675     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 676                              &blkdev->xendev.remote_port) == -1)
 677         return -1;
 678
 679     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 680     if (blkdev->xendev.protocol) {
 681         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0)
 682             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 683         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0)
 684             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 685     }
 686
 687     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 688                                             blkdev->xendev.dom,
 689                                             blkdev->ring_ref,
 690                                             PROT_READ | PROT_WRITE);
 691     if (!blkdev->sring)
 692         return -1;
 693     blkdev->cnt_map++;
 694
 695     switch (blkdev->protocol) {
 696     case BLKIF_PROTOCOL_NATIVE:
 697     {
 698         blkif_sring_t *sring_native = blkdev->sring;
 699         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 700         break;
 701     }
 702     case BLKIF_PROTOCOL_X86_32:
 703     {
 704         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 705
 706         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 707         break;
 708     }
 709     case BLKIF_PROTOCOL_X86_64:
 710     {
 711         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 712
 713         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 714         break;
 715     }
 716     }
 717
 718     xen_be_bind_evtchn(&blkdev->xendev);
 719
 720     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 721                   "remote port %d, local port %d\n",
 722                   blkdev->xendev.protocol, blkdev->ring_ref,
 723                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 724     return 0;
 725 }
 726
 727 static void blk_disconnect(struct XenDevice *xendev)
 728 {
 729     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 730
 731     if (blkdev->bs) {
 732         if (!blkdev->dinfo) {
 733             /* close/delete only if we created it ourself */
 734             bdrv_close(blkdev->bs);
 735             bdrv_delete(blkdev->bs);
 736         }
 737         blkdev->bs = NULL;
 738     }
 739     xen_be_unbind_evtchn(&blkdev->xendev);
 740
 741     if (blkdev->sring) {
 742         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 743         blkdev->cnt_map--;
 744         blkdev->sring = NULL;
 745     }
 746 }
 747
 748 static int blk_free(struct XenDevice *xendev)
 749 {
 750     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 751     struct ioreq *ioreq;
 752
 753     while (!QLIST_EMPTY(&blkdev->freelist)) {
 754         ioreq = QLIST_FIRST(&blkdev->freelist);
 755         QLIST_REMOVE(ioreq, list);
 756         qemu_iovec_destroy(&ioreq->v);
 757         qemu_free(ioreq);
 758     }
 759
 760     qemu_free(blkdev->params);
 761     qemu_free(blkdev->mode);
 762     qemu_free(blkdev->type);
 763     qemu_free(blkdev->dev);
 764     qemu_free(blkdev->devtype);
 765     qemu_bh_delete(blkdev->bh);
 766     return 0;
 767 }
 768
 769 static void blk_event(struct XenDevice *xendev)
 770 {
 771     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 772
 773     qemu_bh_schedule(blkdev->bh);
 774 }
 775
 776 struct XenDevOps xen_blkdev_ops = {
 777     .size       = sizeof(struct XenBlkDev),
 778     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 779     .alloc      = blk_alloc,
 780     .init       = blk_init,
 781     .connect    = blk_connect,
 782     .disconnect = blk_disconnect,
 783     .event      = blk_event,
 784     .free       = blk_free,
 785 };