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