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 "qemu-char.h"
  40 #include "xen_backend.h"
  41 #include "xen_blkif.h"
  42 #include "blockdev.h"
  43
  44 /* ------------------------------------------------------------- */
  45
  46 static int batch_maps   = 0;
  47
  48 static int max_requests = 32;
  49
  50 /* ------------------------------------------------------------- */
  51
  52 #define BLOCK_SIZE  512
  53 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  54
  55 struct ioreq {
  56     blkif_request_t     req;
  57     int16_t             status;
  58
  59     /* parsed request */
  60     off_t               start;
  61     QEMUIOVector        v;
  62     int                 presync;
  63     int                 postsync;
  64     uint8_t             mapped;
  65
  66     /* grant mapping */
  67     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  68     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  69     int                 prot;
  70     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  71     void                *pages;
  72
  73     /* aio status */
  74     int                 aio_inflight;
  75     int                 aio_errors;
  76
  77     struct XenBlkDev    *blkdev;
  78     QLIST_ENTRY(ioreq)   list;
  79     BlockAcctCookie     acct;
  80 };
  81
  82 struct XenBlkDev {
  83     struct XenDevice    xendev;  /* must be first */
  84     char                *params;
  85     char                *mode;
  86     char                *type;
  87     char                *dev;
  88     char                *devtype;
  89     const char          *fileproto;
  90     const char          *filename;
  91     int                 ring_ref;
  92     void                *sring;
  93     int64_t             file_blk;
  94     int64_t             file_size;
  95     int                 protocol;
  96     blkif_back_rings_t  rings;
  97     int                 more_work;
  98     int                 cnt_map;
  99
 100     /* request lists */
 101     QLIST_HEAD(inflight_head, ioreq) inflight;
 102     QLIST_HEAD(finished_head, ioreq) finished;
 103     QLIST_HEAD(freelist_head, ioreq) freelist;
 104     int                 requests_total;
 105     int                 requests_inflight;
 106     int                 requests_finished;
 107
 108     /* qemu block driver */
 109     DriveInfo           *dinfo;
 110     BlockDriverState    *bs;
 111     QEMUBH              *bh;
 112 };
 113
 114 /* ------------------------------------------------------------- */
 115
 116 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 117 {
 118     struct ioreq *ioreq = NULL;
 119
 120     if (QLIST_EMPTY(&blkdev->freelist)) {
 121         if (blkdev->requests_total >= max_requests) {
 122             goto out;
 123         }
 124         /* allocate new struct */
 125         ioreq = g_malloc0(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, bool finish)
 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     if (finish) {
 161         blkdev->requests_finished--;
 162     } else {
 163         blkdev->requests_inflight--;
 164     }
 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 || ioreq->mapped == 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     ioreq->mapped = 0;
 272 }
 273
 274 static int ioreq_map(struct ioreq *ioreq)
 275 {
 276     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 277     int i;
 278
 279     if (ioreq->v.niov == 0 || ioreq->mapped == 1) {
 280         return 0;
 281     }
 282     if (batch_maps) {
 283         ioreq->pages = xc_gnttab_map_grant_refs
 284             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 285         if (ioreq->pages == NULL) {
 286             xen_be_printf(&ioreq->blkdev->xendev, 0,
 287                           "can't map %d grant refs (%s, %d maps)\n",
 288                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 289             return -1;
 290         }
 291         for (i = 0; i < ioreq->v.niov; i++) {
 292             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 293                 (uintptr_t)ioreq->v.iov[i].iov_base;
 294         }
 295         ioreq->blkdev->cnt_map += ioreq->v.niov;
 296     } else  {
 297         for (i = 0; i < ioreq->v.niov; i++) {
 298             ioreq->page[i] = xc_gnttab_map_grant_ref
 299                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 300             if (ioreq->page[i] == NULL) {
 301                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 302                               "can't map grant ref %d (%s, %d maps)\n",
 303                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 304                 ioreq_unmap(ioreq);
 305                 return -1;
 306             }
 307             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 308             ioreq->blkdev->cnt_map++;
 309         }
 310     }
 311     ioreq->mapped = 1;
 312     return 0;
 313 }
 314
 315 static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
 316
 317 static void qemu_aio_complete(void *opaque, int ret)
 318 {
 319     struct ioreq *ioreq = opaque;
 320
 321     if (ret != 0) {
 322         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 323                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 324         ioreq->aio_errors++;
 325     }
 326
 327     ioreq->aio_inflight--;
 328     if (ioreq->presync) {
 329         ioreq->presync = 0;
 330         ioreq_runio_qemu_aio(ioreq);
 331         return;
 332     }
 333     if (ioreq->aio_inflight > 0) {
 334         return;
 335     }
 336     if (ioreq->postsync) {
 337         ioreq->postsync = 0;
 338         ioreq->aio_inflight++;
 339         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 340         return;
 341     }
 342
 343     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 344     ioreq_unmap(ioreq);
 345     ioreq_finish(ioreq);
 346     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 347     qemu_bh_schedule(ioreq->blkdev->bh);
 348 }
 349
 350 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 351 {
 352     struct XenBlkDev *blkdev = ioreq->blkdev;
 353
 354     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 355         goto err_no_map;
 356     }
 357
 358     ioreq->aio_inflight++;
 359     if (ioreq->presync) {
 360         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 361         return 0;
 362     }
 363
 364     switch (ioreq->req.operation) {
 365     case BLKIF_OP_READ:
 366         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 367         ioreq->aio_inflight++;
 368         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 369                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 370                        qemu_aio_complete, ioreq);
 371         break;
 372     case BLKIF_OP_WRITE:
 373     case BLKIF_OP_WRITE_BARRIER:
 374         if (!ioreq->req.nr_segments) {
 375             break;
 376         }
 377
 378         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 379         ioreq->aio_inflight++;
 380         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 381                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 382                         qemu_aio_complete, ioreq);
 383         break;
 384     default:
 385         /* unknown operation (shouldn't happen -- parse catches this) */
 386         goto err;
 387     }
 388
 389     qemu_aio_complete(ioreq, 0);
 390
 391     return 0;
 392
 393 err:
 394     ioreq_unmap(ioreq);
 395 err_no_map:
 396     ioreq_finish(ioreq);
 397     ioreq->status = BLKIF_RSP_ERROR;
 398     return -1;
 399 }
 400
 401 static int blk_send_response_one(struct ioreq *ioreq)
 402 {
 403     struct XenBlkDev  *blkdev = ioreq->blkdev;
 404     int               send_notify   = 0;
 405     int               have_requests = 0;
 406     blkif_response_t  resp;
 407     void              *dst;
 408
 409     resp.id        = ioreq->req.id;
 410     resp.operation = ioreq->req.operation;
 411     resp.status    = ioreq->status;
 412
 413     /* Place on the response ring for the relevant domain. */
 414     switch (blkdev->protocol) {
 415     case BLKIF_PROTOCOL_NATIVE:
 416         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 417         break;
 418     case BLKIF_PROTOCOL_X86_32:
 419         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 420                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 421         break;
 422     case BLKIF_PROTOCOL_X86_64:
 423         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 424                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 425         break;
 426     default:
 427         dst = NULL;
 428     }
 429     memcpy(dst, &resp, sizeof(resp));
 430     blkdev->rings.common.rsp_prod_pvt++;
 431
 432     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 433     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 434         /*
 435          * Tail check for pending requests. Allows frontend to avoid
 436          * notifications if requests are already in flight (lower
 437          * overheads and promotes batching).
 438          */
 439         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 440     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 441         have_requests = 1;
 442     }
 443
 444     if (have_requests) {
 445         blkdev->more_work++;
 446     }
 447     return send_notify;
 448 }
 449
 450 /* walk finished list, send outstanding responses, free requests */
 451 static void blk_send_response_all(struct XenBlkDev *blkdev)
 452 {
 453     struct ioreq *ioreq;
 454     int send_notify = 0;
 455
 456     while (!QLIST_EMPTY(&blkdev->finished)) {
 457         ioreq = QLIST_FIRST(&blkdev->finished);
 458         send_notify += blk_send_response_one(ioreq);
 459         ioreq_release(ioreq, true);
 460     }
 461     if (send_notify) {
 462         xen_be_send_notify(&blkdev->xendev);
 463     }
 464 }
 465
 466 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 467 {
 468     switch (blkdev->protocol) {
 469     case BLKIF_PROTOCOL_NATIVE:
 470         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 471                sizeof(ioreq->req));
 472         break;
 473     case BLKIF_PROTOCOL_X86_32:
 474         blkif_get_x86_32_req(&ioreq->req,
 475                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 476         break;
 477     case BLKIF_PROTOCOL_X86_64:
 478         blkif_get_x86_64_req(&ioreq->req,
 479                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 480         break;
 481     }
 482     return 0;
 483 }
 484
 485 static void blk_handle_requests(struct XenBlkDev *blkdev)
 486 {
 487     RING_IDX rc, rp;
 488     struct ioreq *ioreq;
 489
 490     blkdev->more_work = 0;
 491
 492     rc = blkdev->rings.common.req_cons;
 493     rp = blkdev->rings.common.sring->req_prod;
 494     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 495
 496     blk_send_response_all(blkdev);
 497     while (rc != rp) {
 498         /* pull request from ring */
 499         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 500             break;
 501         }
 502         ioreq = ioreq_start(blkdev);
 503         if (ioreq == NULL) {
 504             blkdev->more_work++;
 505             break;
 506         }
 507         blk_get_request(blkdev, ioreq, rc);
 508         blkdev->rings.common.req_cons = ++rc;
 509
 510         /* parse them */
 511         if (ioreq_parse(ioreq) != 0) {
 512             if (blk_send_response_one(ioreq)) {
 513                 xen_be_send_notify(&blkdev->xendev);
 514             }
 515             ioreq_release(ioreq, false);
 516             continue;
 517         }
 518
 519         ioreq_runio_qemu_aio(ioreq);
 520     }
 521
 522     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 523         qemu_bh_schedule(blkdev->bh);
 524     }
 525 }
 526
 527 /* ------------------------------------------------------------- */
 528
 529 static void blk_bh(void *opaque)
 530 {
 531     struct XenBlkDev *blkdev = opaque;
 532     blk_handle_requests(blkdev);
 533 }
 534
 535 /*
 536  * We need to account for the grant allocations requiring contiguous
 537  * chunks; the worst case number would be
 538  *     max_req * max_seg + (max_req - 1) * (max_seg - 1) + 1,
 539  * but in order to keep things simple just use
 540  *     2 * max_req * max_seg.
 541  */
 542 #define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
 543
 544 static void blk_alloc(struct XenDevice *xendev)
 545 {
 546     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 547
 548     QLIST_INIT(&blkdev->inflight);
 549     QLIST_INIT(&blkdev->finished);
 550     QLIST_INIT(&blkdev->freelist);
 551     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 552     if (xen_mode != XEN_EMULATE) {
 553         batch_maps = 1;
 554     }
 555     if (xc_gnttab_set_max_grants(xendev->gnttabdev,
 556             MAX_GRANTS(max_requests, BLKIF_MAX_SEGMENTS_PER_REQUEST)) < 0) {
 557         xen_be_printf(xendev, 0, "xc_gnttab_set_max_grants failed: %s\n",
 558                       strerror(errno));
 559     }
 560 }
 561
 562 static int blk_init(struct XenDevice *xendev)
 563 {
 564     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 565     int index, qflags, info = 0;
 566
 567     /* read xenstore entries */
 568     if (blkdev->params == NULL) {
 569         char *h = NULL;
 570         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 571         if (blkdev->params != NULL) {
 572             h = strchr(blkdev->params, ':');
 573         }
 574         if (h != NULL) {
 575             blkdev->fileproto = blkdev->params;
 576             blkdev->filename  = h+1;
 577             *h = 0;
 578         } else {
 579             blkdev->fileproto = "<unset>";
 580             blkdev->filename  = blkdev->params;
 581         }
 582     }
 583     if (!strcmp("aio", blkdev->fileproto)) {
 584         blkdev->fileproto = "raw";
 585     }
 586     if (blkdev->mode == NULL) {
 587         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 588     }
 589     if (blkdev->type == NULL) {
 590         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 591     }
 592     if (blkdev->dev == NULL) {
 593         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 594     }
 595     if (blkdev->devtype == NULL) {
 596         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 597     }
 598
 599     /* do we have all we need? */
 600     if (blkdev->params == NULL ||
 601         blkdev->mode == NULL   ||
 602         blkdev->type == NULL   ||
 603         blkdev->dev == NULL) {
 604         goto out_error;
 605     }
 606
 607     /* read-only ? */
 608     qflags = BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO;
 609     if (strcmp(blkdev->mode, "w") == 0) {
 610         qflags |= BDRV_O_RDWR;
 611     } else {
 612         info  |= VDISK_READONLY;
 613     }
 614
 615     /* cdrom ? */
 616     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 617         info  |= VDISK_CDROM;
 618     }
 619
 620     /* init qemu block driver */
 621     index = (blkdev->xendev.dev - 202 * 256) / 16;
 622     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 623     if (!blkdev->dinfo) {
 624         /* setup via xenbus -> create new block driver instance */
 625         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 626         blkdev->bs = bdrv_new(blkdev->dev);
 627         if (blkdev->bs) {
 628             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 629                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 630                 bdrv_delete(blkdev->bs);
 631                 blkdev->bs = NULL;
 632             }
 633         }
 634         if (!blkdev->bs) {
 635             goto out_error;
 636         }
 637     } else {
 638         /* setup via qemu cmdline -> already setup for us */
 639         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 640         blkdev->bs = blkdev->dinfo->bdrv;
 641     }
 642     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 643     blkdev->file_blk  = BLOCK_SIZE;
 644     blkdev->file_size = bdrv_getlength(blkdev->bs);
 645     if (blkdev->file_size < 0) {
 646         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 647                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 648                       bdrv_get_format_name(blkdev->bs) ?: "-");
 649         blkdev->file_size = 0;
 650     }
 651
 652     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 653                   " size %" PRId64 " (%" PRId64 " MB)\n",
 654                   blkdev->type, blkdev->fileproto, blkdev->filename,
 655                   blkdev->file_size, blkdev->file_size >> 20);
 656
 657     /* fill info */
 658     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", 1);
 659     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 660     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 661     xenstore_write_be_int(&blkdev->xendev, "sectors",
 662                           blkdev->file_size / blkdev->file_blk);
 663     return 0;
 664
 665 out_error:
 666     g_free(blkdev->params);
 667     blkdev->params = NULL;
 668     g_free(blkdev->mode);
 669     blkdev->mode = NULL;
 670     g_free(blkdev->type);
 671     blkdev->type = NULL;
 672     g_free(blkdev->dev);
 673     blkdev->dev = NULL;
 674     g_free(blkdev->devtype);
 675     blkdev->devtype = NULL;
 676     return -1;
 677 }
 678
 679 static int blk_connect(struct XenDevice *xendev)
 680 {
 681     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 682
 683     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 684         return -1;
 685     }
 686     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 687                              &blkdev->xendev.remote_port) == -1) {
 688         return -1;
 689     }
 690
 691     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 692     if (blkdev->xendev.protocol) {
 693         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 694             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 695         }
 696         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 697             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 698         }
 699     }
 700
 701     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 702                                             blkdev->xendev.dom,
 703                                             blkdev->ring_ref,
 704                                             PROT_READ | PROT_WRITE);
 705     if (!blkdev->sring) {
 706         return -1;
 707     }
 708     blkdev->cnt_map++;
 709
 710     switch (blkdev->protocol) {
 711     case BLKIF_PROTOCOL_NATIVE:
 712     {
 713         blkif_sring_t *sring_native = blkdev->sring;
 714         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 715         break;
 716     }
 717     case BLKIF_PROTOCOL_X86_32:
 718     {
 719         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 720
 721         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 722         break;
 723     }
 724     case BLKIF_PROTOCOL_X86_64:
 725     {
 726         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 727
 728         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 729         break;
 730     }
 731     }
 732
 733     xen_be_bind_evtchn(&blkdev->xendev);
 734
 735     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 736                   "remote port %d, local port %d\n",
 737                   blkdev->xendev.protocol, blkdev->ring_ref,
 738                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 739     return 0;
 740 }
 741
 742 static void blk_disconnect(struct XenDevice *xendev)
 743 {
 744     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 745
 746     if (blkdev->bs) {
 747         if (!blkdev->dinfo) {
 748             /* close/delete only if we created it ourself */
 749             bdrv_close(blkdev->bs);
 750             bdrv_detach_dev(blkdev->bs, blkdev);
 751             bdrv_delete(blkdev->bs);
 752         }
 753         blkdev->bs = NULL;
 754     }
 755     xen_be_unbind_evtchn(&blkdev->xendev);
 756
 757     if (blkdev->sring) {
 758         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 759         blkdev->cnt_map--;
 760         blkdev->sring = NULL;
 761     }
 762 }
 763
 764 static int blk_free(struct XenDevice *xendev)
 765 {
 766     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 767     struct ioreq *ioreq;
 768
 769     if (blkdev->bs || blkdev->sring) {
 770         blk_disconnect(xendev);
 771     }
 772
 773     while (!QLIST_EMPTY(&blkdev->freelist)) {
 774         ioreq = QLIST_FIRST(&blkdev->freelist);
 775         QLIST_REMOVE(ioreq, list);
 776         qemu_iovec_destroy(&ioreq->v);
 777         g_free(ioreq);
 778     }
 779
 780     g_free(blkdev->params);
 781     g_free(blkdev->mode);
 782     g_free(blkdev->type);
 783     g_free(blkdev->dev);
 784     g_free(blkdev->devtype);
 785     qemu_bh_delete(blkdev->bh);
 786     return 0;
 787 }
 788
 789 static void blk_event(struct XenDevice *xendev)
 790 {
 791     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 792
 793     qemu_bh_schedule(blkdev->bh);
 794 }
 795
 796 struct XenDevOps xen_blkdev_ops = {
 797     .size       = sizeof(struct XenBlkDev),
 798     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 799     .alloc      = blk_alloc,
 800     .init       = blk_init,
 801     .initialise    = blk_connect,
 802     .disconnect = blk_disconnect,
 803     .event      = blk_event,
 804     .free       = blk_free,
 805 };