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  *  Contributions after 2012-01-13 are licensed under the terms of the
  19  *  GNU GPL, version 2 or (at your option) any later version.
  20  */
  21
  22 #include <stdio.h>
  23 #include <stdlib.h>
  24 #include <stdarg.h>
  25 #include <string.h>
  26 #include <unistd.h>
  27 #include <signal.h>
  28 #include <inttypes.h>
  29 #include <time.h>
  30 #include <fcntl.h>
  31 #include <errno.h>
  32 #include <sys/ioctl.h>
  33 #include <sys/types.h>
  34 #include <sys/stat.h>
  35 #include <sys/mman.h>
  36 #include <sys/uio.h>
  37
  38 #include "hw.h"
  39 #include "xen_backend.h"
  40 #include "xen_blkif.h"
  41 #include "sysemu/blockdev.h"
  42
  43 /* ------------------------------------------------------------- */
  44
  45 static int batch_maps   = 0;
  46
  47 static int max_requests = 32;
  48
  49 /* ------------------------------------------------------------- */
  50
  51 #define BLOCK_SIZE  512
  52 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  53
  54 struct ioreq {
  55     blkif_request_t     req;
  56     int16_t             status;
  57
  58     /* parsed request */
  59     off_t               start;
  60     QEMUIOVector        v;
  61     int                 presync;
  62     int                 postsync;
  63     uint8_t             mapped;
  64
  65     /* grant mapping */
  66     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  67     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  68     int                 prot;
  69     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  70     void                *pages;
  71
  72     /* aio status */
  73     int                 aio_inflight;
  74     int                 aio_errors;
  75
  76     struct XenBlkDev    *blkdev;
  77     QLIST_ENTRY(ioreq)   list;
  78     BlockAcctCookie     acct;
  79 };
  80
  81 struct XenBlkDev {
  82     struct XenDevice    xendev;  /* must be first */
  83     char                *params;
  84     char                *mode;
  85     char                *type;
  86     char                *dev;
  87     char                *devtype;
  88     const char          *fileproto;
  89     const char          *filename;
  90     int                 ring_ref;
  91     void                *sring;
  92     int64_t             file_blk;
  93     int64_t             file_size;
  94     int                 protocol;
  95     blkif_back_rings_t  rings;
  96     int                 more_work;
  97     int                 cnt_map;
  98
  99     /* request lists */
 100     QLIST_HEAD(inflight_head, ioreq) inflight;
 101     QLIST_HEAD(finished_head, ioreq) finished;
 102     QLIST_HEAD(freelist_head, ioreq) freelist;
 103     int                 requests_total;
 104     int                 requests_inflight;
 105     int                 requests_finished;
 106
 107     /* qemu block driver */
 108     DriveInfo           *dinfo;
 109     BlockDriverState    *bs;
 110     QEMUBH              *bh;
 111 };
 112
 113 /* ------------------------------------------------------------- */
 114
 115 static void ioreq_reset(struct ioreq *ioreq)
 116 {
 117     memset(&ioreq->req, 0, sizeof(ioreq->req));
 118     ioreq->status = 0;
 119     ioreq->start = 0;
 120     ioreq->presync = 0;
 121     ioreq->postsync = 0;
 122     ioreq->mapped = 0;
 123
 124     memset(ioreq->domids, 0, sizeof(ioreq->domids));
 125     memset(ioreq->refs, 0, sizeof(ioreq->refs));
 126     ioreq->prot = 0;
 127     memset(ioreq->page, 0, sizeof(ioreq->page));
 128     ioreq->pages = NULL;
 129
 130     ioreq->aio_inflight = 0;
 131     ioreq->aio_errors = 0;
 132
 133     ioreq->blkdev = NULL;
 134     memset(&ioreq->list, 0, sizeof(ioreq->list));
 135     memset(&ioreq->acct, 0, sizeof(ioreq->acct));
 136
 137     qemu_iovec_reset(&ioreq->v);
 138 }
 139
 140 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 141 {
 142     struct ioreq *ioreq = NULL;
 143
 144     if (QLIST_EMPTY(&blkdev->freelist)) {
 145         if (blkdev->requests_total >= max_requests) {
 146             goto out;
 147         }
 148         /* allocate new struct */
 149         ioreq = g_malloc0(sizeof(*ioreq));
 150         ioreq->blkdev = blkdev;
 151         blkdev->requests_total++;
 152         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 153     } else {
 154         /* get one from freelist */
 155         ioreq = QLIST_FIRST(&blkdev->freelist);
 156         QLIST_REMOVE(ioreq, list);
 157     }
 158     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 159     blkdev->requests_inflight++;
 160
 161 out:
 162     return ioreq;
 163 }
 164
 165 static void ioreq_finish(struct ioreq *ioreq)
 166 {
 167     struct XenBlkDev *blkdev = ioreq->blkdev;
 168
 169     QLIST_REMOVE(ioreq, list);
 170     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 171     blkdev->requests_inflight--;
 172     blkdev->requests_finished++;
 173 }
 174
 175 static void ioreq_release(struct ioreq *ioreq, bool finish)
 176 {
 177     struct XenBlkDev *blkdev = ioreq->blkdev;
 178
 179     QLIST_REMOVE(ioreq, list);
 180     ioreq_reset(ioreq);
 181     ioreq->blkdev = blkdev;
 182     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 183     if (finish) {
 184         blkdev->requests_finished--;
 185     } else {
 186         blkdev->requests_inflight--;
 187     }
 188 }
 189
 190 /*
 191  * translate request into iovec + start offset
 192  * do sanity checks along the way
 193  */
 194 static int ioreq_parse(struct ioreq *ioreq)
 195 {
 196     struct XenBlkDev *blkdev = ioreq->blkdev;
 197     uintptr_t mem;
 198     size_t len;
 199     int i;
 200
 201     xen_be_printf(&blkdev->xendev, 3,
 202                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 203                   ioreq->req.operation, ioreq->req.nr_segments,
 204                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 205     switch (ioreq->req.operation) {
 206     case BLKIF_OP_READ:
 207         ioreq->prot = PROT_WRITE; /* to memory */
 208         break;
 209     case BLKIF_OP_WRITE_BARRIER:
 210         if (!ioreq->req.nr_segments) {
 211             ioreq->presync = 1;
 212             return 0;
 213         }
 214         ioreq->presync = ioreq->postsync = 1;
 215         /* fall through */
 216     case BLKIF_OP_WRITE:
 217         ioreq->prot = PROT_READ; /* from memory */
 218         break;
 219     default:
 220         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 221                       ioreq->req.operation);
 222         goto err;
 223     };
 224
 225     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 226         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 227         goto err;
 228     }
 229
 230     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 231     for (i = 0; i < ioreq->req.nr_segments; i++) {
 232         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 233             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 234             goto err;
 235         }
 236         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 237             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 238             goto err;
 239         }
 240         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 241             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 242             goto err;
 243         }
 244
 245         ioreq->domids[i] = blkdev->xendev.dom;
 246         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 247
 248         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 249         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 250         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 251     }
 252     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 253         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 254         goto err;
 255     }
 256     return 0;
 257
 258 err:
 259     ioreq->status = BLKIF_RSP_ERROR;
 260     return -1;
 261 }
 262
 263 static void ioreq_unmap(struct ioreq *ioreq)
 264 {
 265     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 266     int i;
 267
 268     if (ioreq->v.niov == 0 || ioreq->mapped == 0) {
 269         return;
 270     }
 271     if (batch_maps) {
 272         if (!ioreq->pages) {
 273             return;
 274         }
 275         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0) {
 276             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 277                           strerror(errno));
 278         }
 279         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 280         ioreq->pages = NULL;
 281     } else {
 282         for (i = 0; i < ioreq->v.niov; i++) {
 283             if (!ioreq->page[i]) {
 284                 continue;
 285             }
 286             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0) {
 287                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 288                               strerror(errno));
 289             }
 290             ioreq->blkdev->cnt_map--;
 291             ioreq->page[i] = NULL;
 292         }
 293     }
 294     ioreq->mapped = 0;
 295 }
 296
 297 static int ioreq_map(struct ioreq *ioreq)
 298 {
 299     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 300     int i;
 301
 302     if (ioreq->v.niov == 0 || ioreq->mapped == 1) {
 303         return 0;
 304     }
 305     if (batch_maps) {
 306         ioreq->pages = xc_gnttab_map_grant_refs
 307             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 308         if (ioreq->pages == NULL) {
 309             xen_be_printf(&ioreq->blkdev->xendev, 0,
 310                           "can't map %d grant refs (%s, %d maps)\n",
 311                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 312             return -1;
 313         }
 314         for (i = 0; i < ioreq->v.niov; i++) {
 315             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 316                 (uintptr_t)ioreq->v.iov[i].iov_base;
 317         }
 318         ioreq->blkdev->cnt_map += ioreq->v.niov;
 319     } else  {
 320         for (i = 0; i < ioreq->v.niov; i++) {
 321             ioreq->page[i] = xc_gnttab_map_grant_ref
 322                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 323             if (ioreq->page[i] == NULL) {
 324                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 325                               "can't map grant ref %d (%s, %d maps)\n",
 326                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 327                 ioreq_unmap(ioreq);
 328                 return -1;
 329             }
 330             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 331             ioreq->blkdev->cnt_map++;
 332         }
 333     }
 334     ioreq->mapped = 1;
 335     return 0;
 336 }
 337
 338 static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
 339
 340 static void qemu_aio_complete(void *opaque, int ret)
 341 {
 342     struct ioreq *ioreq = opaque;
 343
 344     if (ret != 0) {
 345         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 346                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 347         ioreq->aio_errors++;
 348     }
 349
 350     ioreq->aio_inflight--;
 351     if (ioreq->presync) {
 352         ioreq->presync = 0;
 353         ioreq_runio_qemu_aio(ioreq);
 354         return;
 355     }
 356     if (ioreq->aio_inflight > 0) {
 357         return;
 358     }
 359     if (ioreq->postsync) {
 360         ioreq->postsync = 0;
 361         ioreq->aio_inflight++;
 362         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 363         return;
 364     }
 365
 366     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 367     ioreq_unmap(ioreq);
 368     ioreq_finish(ioreq);
 369     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 370     qemu_bh_schedule(ioreq->blkdev->bh);
 371 }
 372
 373 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 374 {
 375     struct XenBlkDev *blkdev = ioreq->blkdev;
 376
 377     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 378         goto err_no_map;
 379     }
 380
 381     ioreq->aio_inflight++;
 382     if (ioreq->presync) {
 383         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 384         return 0;
 385     }
 386
 387     switch (ioreq->req.operation) {
 388     case BLKIF_OP_READ:
 389         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 390         ioreq->aio_inflight++;
 391         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 392                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 393                        qemu_aio_complete, ioreq);
 394         break;
 395     case BLKIF_OP_WRITE:
 396     case BLKIF_OP_WRITE_BARRIER:
 397         if (!ioreq->req.nr_segments) {
 398             break;
 399         }
 400
 401         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 402         ioreq->aio_inflight++;
 403         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 404                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 405                         qemu_aio_complete, ioreq);
 406         break;
 407     default:
 408         /* unknown operation (shouldn't happen -- parse catches this) */
 409         goto err;
 410     }
 411
 412     qemu_aio_complete(ioreq, 0);
 413
 414     return 0;
 415
 416 err:
 417     ioreq_unmap(ioreq);
 418 err_no_map:
 419     ioreq_finish(ioreq);
 420     ioreq->status = BLKIF_RSP_ERROR;
 421     return -1;
 422 }
 423
 424 static int blk_send_response_one(struct ioreq *ioreq)
 425 {
 426     struct XenBlkDev  *blkdev = ioreq->blkdev;
 427     int               send_notify   = 0;
 428     int               have_requests = 0;
 429     blkif_response_t  resp;
 430     void              *dst;
 431
 432     resp.id        = ioreq->req.id;
 433     resp.operation = ioreq->req.operation;
 434     resp.status    = ioreq->status;
 435
 436     /* Place on the response ring for the relevant domain. */
 437     switch (blkdev->protocol) {
 438     case BLKIF_PROTOCOL_NATIVE:
 439         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 440         break;
 441     case BLKIF_PROTOCOL_X86_32:
 442         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 443                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 444         break;
 445     case BLKIF_PROTOCOL_X86_64:
 446         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 447                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 448         break;
 449     default:
 450         dst = NULL;
 451     }
 452     memcpy(dst, &resp, sizeof(resp));
 453     blkdev->rings.common.rsp_prod_pvt++;
 454
 455     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 456     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 457         /*
 458          * Tail check for pending requests. Allows frontend to avoid
 459          * notifications if requests are already in flight (lower
 460          * overheads and promotes batching).
 461          */
 462         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 463     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 464         have_requests = 1;
 465     }
 466
 467     if (have_requests) {
 468         blkdev->more_work++;
 469     }
 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, true);
 483     }
 484     if (send_notify) {
 485         xen_be_send_notify(&blkdev->xendev);
 486     }
 487 }
 488
 489 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 490 {
 491     switch (blkdev->protocol) {
 492     case BLKIF_PROTOCOL_NATIVE:
 493         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 494                sizeof(ioreq->req));
 495         break;
 496     case BLKIF_PROTOCOL_X86_32:
 497         blkif_get_x86_32_req(&ioreq->req,
 498                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 499         break;
 500     case BLKIF_PROTOCOL_X86_64:
 501         blkif_get_x86_64_req(&ioreq->req,
 502                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 503         break;
 504     }
 505     return 0;
 506 }
 507
 508 static void blk_handle_requests(struct XenBlkDev *blkdev)
 509 {
 510     RING_IDX rc, rp;
 511     struct ioreq *ioreq;
 512
 513     blkdev->more_work = 0;
 514
 515     rc = blkdev->rings.common.req_cons;
 516     rp = blkdev->rings.common.sring->req_prod;
 517     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 518
 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         }
 525         ioreq = ioreq_start(blkdev);
 526         if (ioreq == NULL) {
 527             blkdev->more_work++;
 528             break;
 529         }
 530         blk_get_request(blkdev, ioreq, rc);
 531         blkdev->rings.common.req_cons = ++rc;
 532
 533         /* parse them */
 534         if (ioreq_parse(ioreq) != 0) {
 535             if (blk_send_response_one(ioreq)) {
 536                 xen_be_send_notify(&blkdev->xendev);
 537             }
 538             ioreq_release(ioreq, false);
 539             continue;
 540         }
 541
 542         ioreq_runio_qemu_aio(ioreq);
 543     }
 544
 545     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 546         qemu_bh_schedule(blkdev->bh);
 547     }
 548 }
 549
 550 /* ------------------------------------------------------------- */
 551
 552 static void blk_bh(void *opaque)
 553 {
 554     struct XenBlkDev *blkdev = opaque;
 555     blk_handle_requests(blkdev);
 556 }
 557
 558 /*
 559  * We need to account for the grant allocations requiring contiguous
 560  * chunks; the worst case number would be
 561  *     max_req * max_seg + (max_req - 1) * (max_seg - 1) + 1,
 562  * but in order to keep things simple just use
 563  *     2 * max_req * max_seg.
 564  */
 565 #define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
 566
 567 static void blk_alloc(struct XenDevice *xendev)
 568 {
 569     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 570
 571     QLIST_INIT(&blkdev->inflight);
 572     QLIST_INIT(&blkdev->finished);
 573     QLIST_INIT(&blkdev->freelist);
 574     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 575     if (xen_mode != XEN_EMULATE) {
 576         batch_maps = 1;
 577     }
 578     if (xc_gnttab_set_max_grants(xendev->gnttabdev,
 579             MAX_GRANTS(max_requests, BLKIF_MAX_SEGMENTS_PER_REQUEST)) < 0) {
 580         xen_be_printf(xendev, 0, "xc_gnttab_set_max_grants failed: %s\n",
 581                       strerror(errno));
 582     }
 583 }
 584
 585 static int blk_init(struct XenDevice *xendev)
 586 {
 587     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 588     int index, qflags, info = 0;
 589
 590     /* read xenstore entries */
 591     if (blkdev->params == NULL) {
 592         char *h = NULL;
 593         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 594         if (blkdev->params != NULL) {
 595             h = strchr(blkdev->params, ':');
 596         }
 597         if (h != NULL) {
 598             blkdev->fileproto = blkdev->params;
 599             blkdev->filename  = h+1;
 600             *h = 0;
 601         } else {
 602             blkdev->fileproto = "<unset>";
 603             blkdev->filename  = blkdev->params;
 604         }
 605     }
 606     if (!strcmp("aio", blkdev->fileproto)) {
 607         blkdev->fileproto = "raw";
 608     }
 609     if (blkdev->mode == NULL) {
 610         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 611     }
 612     if (blkdev->type == NULL) {
 613         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 614     }
 615     if (blkdev->dev == NULL) {
 616         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 617     }
 618     if (blkdev->devtype == NULL) {
 619         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 620     }
 621
 622     /* do we have all we need? */
 623     if (blkdev->params == NULL ||
 624         blkdev->mode == NULL   ||
 625         blkdev->type == NULL   ||
 626         blkdev->dev == NULL) {
 627         goto out_error;
 628     }
 629
 630     /* read-only ? */
 631     qflags = BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO;
 632     if (strcmp(blkdev->mode, "w") == 0) {
 633         qflags |= BDRV_O_RDWR;
 634     } else {
 635         info  |= VDISK_READONLY;
 636     }
 637
 638     /* cdrom ? */
 639     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 640         info  |= VDISK_CDROM;
 641     }
 642
 643     /* init qemu block driver */
 644     index = (blkdev->xendev.dev - 202 * 256) / 16;
 645     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 646     if (!blkdev->dinfo) {
 647         /* setup via xenbus -> create new block driver instance */
 648         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 649         blkdev->bs = bdrv_new(blkdev->dev);
 650         if (blkdev->bs) {
 651             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 652                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 653                 bdrv_delete(blkdev->bs);
 654                 blkdev->bs = NULL;
 655             }
 656         }
 657         if (!blkdev->bs) {
 658             goto out_error;
 659         }
 660     } else {
 661         /* setup via qemu cmdline -> already setup for us */
 662         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 663         blkdev->bs = blkdev->dinfo->bdrv;
 664     }
 665     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 666     blkdev->file_blk  = BLOCK_SIZE;
 667     blkdev->file_size = bdrv_getlength(blkdev->bs);
 668     if (blkdev->file_size < 0) {
 669         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 670                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 671                       bdrv_get_format_name(blkdev->bs) ?: "-");
 672         blkdev->file_size = 0;
 673     }
 674
 675     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 676                   " size %" PRId64 " (%" PRId64 " MB)\n",
 677                   blkdev->type, blkdev->fileproto, blkdev->filename,
 678                   blkdev->file_size, blkdev->file_size >> 20);
 679
 680     /* fill info */
 681     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", 1);
 682     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 683     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 684     xenstore_write_be_int(&blkdev->xendev, "sectors",
 685                           blkdev->file_size / blkdev->file_blk);
 686     return 0;
 687
 688 out_error:
 689     g_free(blkdev->params);
 690     blkdev->params = NULL;
 691     g_free(blkdev->mode);
 692     blkdev->mode = NULL;
 693     g_free(blkdev->type);
 694     blkdev->type = NULL;
 695     g_free(blkdev->dev);
 696     blkdev->dev = NULL;
 697     g_free(blkdev->devtype);
 698     blkdev->devtype = NULL;
 699     return -1;
 700 }
 701
 702 static int blk_connect(struct XenDevice *xendev)
 703 {
 704     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 705
 706     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 707         return -1;
 708     }
 709     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 710                              &blkdev->xendev.remote_port) == -1) {
 711         return -1;
 712     }
 713
 714     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 715     if (blkdev->xendev.protocol) {
 716         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 717             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 718         }
 719         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 720             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 721         }
 722     }
 723
 724     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 725                                             blkdev->xendev.dom,
 726                                             blkdev->ring_ref,
 727                                             PROT_READ | PROT_WRITE);
 728     if (!blkdev->sring) {
 729         return -1;
 730     }
 731     blkdev->cnt_map++;
 732
 733     switch (blkdev->protocol) {
 734     case BLKIF_PROTOCOL_NATIVE:
 735     {
 736         blkif_sring_t *sring_native = blkdev->sring;
 737         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 738         break;
 739     }
 740     case BLKIF_PROTOCOL_X86_32:
 741     {
 742         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 743
 744         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 745         break;
 746     }
 747     case BLKIF_PROTOCOL_X86_64:
 748     {
 749         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 750
 751         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 752         break;
 753     }
 754     }
 755
 756     xen_be_bind_evtchn(&blkdev->xendev);
 757
 758     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 759                   "remote port %d, local port %d\n",
 760                   blkdev->xendev.protocol, blkdev->ring_ref,
 761                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 762     return 0;
 763 }
 764
 765 static void blk_disconnect(struct XenDevice *xendev)
 766 {
 767     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 768
 769     if (blkdev->bs) {
 770         if (!blkdev->dinfo) {
 771             /* close/delete only if we created it ourself */
 772             bdrv_close(blkdev->bs);
 773             bdrv_detach_dev(blkdev->bs, blkdev);
 774             bdrv_delete(blkdev->bs);
 775         }
 776         blkdev->bs = NULL;
 777     }
 778     xen_be_unbind_evtchn(&blkdev->xendev);
 779
 780     if (blkdev->sring) {
 781         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 782         blkdev->cnt_map--;
 783         blkdev->sring = NULL;
 784     }
 785 }
 786
 787 static int blk_free(struct XenDevice *xendev)
 788 {
 789     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 790     struct ioreq *ioreq;
 791
 792     if (blkdev->bs || blkdev->sring) {
 793         blk_disconnect(xendev);
 794     }
 795
 796     while (!QLIST_EMPTY(&blkdev->freelist)) {
 797         ioreq = QLIST_FIRST(&blkdev->freelist);
 798         QLIST_REMOVE(ioreq, list);
 799         qemu_iovec_destroy(&ioreq->v);
 800         g_free(ioreq);
 801     }
 802
 803     g_free(blkdev->params);
 804     g_free(blkdev->mode);
 805     g_free(blkdev->type);
 806     g_free(blkdev->dev);
 807     g_free(blkdev->devtype);
 808     qemu_bh_delete(blkdev->bh);
 809     return 0;
 810 }
 811
 812 static void blk_event(struct XenDevice *xendev)
 813 {
 814     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 815
 816     qemu_bh_schedule(blkdev->bh);
 817 }
 818
 819 struct XenDevOps xen_blkdev_ops = {
 820     .size       = sizeof(struct XenBlkDev),
 821     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 822     .alloc      = blk_alloc,
 823     .init       = blk_init,
 824     .initialise    = blk_connect,
 825     .disconnect = blk_disconnect,
 826     .event      = blk_event,
 827     .free       = blk_free,
 828 };