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 struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 116 {
 117     struct ioreq *ioreq = NULL;
 118
 119     if (QLIST_EMPTY(&blkdev->freelist)) {
 120         if (blkdev->requests_total >= max_requests) {
 121             goto out;
 122         }
 123         /* allocate new struct */
 124         ioreq = g_malloc0(sizeof(*ioreq));
 125         ioreq->blkdev = blkdev;
 126         blkdev->requests_total++;
 127         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 128     } else {
 129         /* get one from freelist */
 130         ioreq = QLIST_FIRST(&blkdev->freelist);
 131         QLIST_REMOVE(ioreq, list);
 132         qemu_iovec_reset(&ioreq->v);
 133     }
 134     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 135     blkdev->requests_inflight++;
 136
 137 out:
 138     return ioreq;
 139 }
 140
 141 static void ioreq_finish(struct ioreq *ioreq)
 142 {
 143     struct XenBlkDev *blkdev = ioreq->blkdev;
 144
 145     QLIST_REMOVE(ioreq, list);
 146     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 147     blkdev->requests_inflight--;
 148     blkdev->requests_finished++;
 149 }
 150
 151 static void ioreq_release(struct ioreq *ioreq, bool finish)
 152 {
 153     struct XenBlkDev *blkdev = ioreq->blkdev;
 154
 155     QLIST_REMOVE(ioreq, list);
 156     memset(ioreq, 0, sizeof(*ioreq));
 157     ioreq->blkdev = blkdev;
 158     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 159     if (finish) {
 160         blkdev->requests_finished--;
 161     } else {
 162         blkdev->requests_inflight--;
 163     }
 164 }
 165
 166 /*
 167  * translate request into iovec + start offset
 168  * do sanity checks along the way
 169  */
 170 static int ioreq_parse(struct ioreq *ioreq)
 171 {
 172     struct XenBlkDev *blkdev = ioreq->blkdev;
 173     uintptr_t mem;
 174     size_t len;
 175     int i;
 176
 177     xen_be_printf(&blkdev->xendev, 3,
 178                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 179                   ioreq->req.operation, ioreq->req.nr_segments,
 180                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 181     switch (ioreq->req.operation) {
 182     case BLKIF_OP_READ:
 183         ioreq->prot = PROT_WRITE; /* to memory */
 184         break;
 185     case BLKIF_OP_WRITE_BARRIER:
 186         if (!ioreq->req.nr_segments) {
 187             ioreq->presync = 1;
 188             return 0;
 189         }
 190         ioreq->presync = ioreq->postsync = 1;
 191         /* fall through */
 192     case BLKIF_OP_WRITE:
 193         ioreq->prot = PROT_READ; /* from memory */
 194         break;
 195     default:
 196         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 197                       ioreq->req.operation);
 198         goto err;
 199     };
 200
 201     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 202         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 203         goto err;
 204     }
 205
 206     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 207     for (i = 0; i < ioreq->req.nr_segments; i++) {
 208         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 209             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 210             goto err;
 211         }
 212         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 213             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 214             goto err;
 215         }
 216         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 217             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 218             goto err;
 219         }
 220
 221         ioreq->domids[i] = blkdev->xendev.dom;
 222         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 223
 224         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 225         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 226         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 227     }
 228     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 229         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 230         goto err;
 231     }
 232     return 0;
 233
 234 err:
 235     ioreq->status = BLKIF_RSP_ERROR;
 236     return -1;
 237 }
 238
 239 static void ioreq_unmap(struct ioreq *ioreq)
 240 {
 241     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 242     int i;
 243
 244     if (ioreq->v.niov == 0 || ioreq->mapped == 0) {
 245         return;
 246     }
 247     if (batch_maps) {
 248         if (!ioreq->pages) {
 249             return;
 250         }
 251         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0) {
 252             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 253                           strerror(errno));
 254         }
 255         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 256         ioreq->pages = NULL;
 257     } else {
 258         for (i = 0; i < ioreq->v.niov; i++) {
 259             if (!ioreq->page[i]) {
 260                 continue;
 261             }
 262             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0) {
 263                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 264                               strerror(errno));
 265             }
 266             ioreq->blkdev->cnt_map--;
 267             ioreq->page[i] = NULL;
 268         }
 269     }
 270     ioreq->mapped = 0;
 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 || ioreq->mapped == 1) {
 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     ioreq->mapped = 1;
 311     return 0;
 312 }
 313
 314 static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
 315
 316 static void qemu_aio_complete(void *opaque, int ret)
 317 {
 318     struct ioreq *ioreq = opaque;
 319
 320     if (ret != 0) {
 321         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 322                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 323         ioreq->aio_errors++;
 324     }
 325
 326     ioreq->aio_inflight--;
 327     if (ioreq->presync) {
 328         ioreq->presync = 0;
 329         ioreq_runio_qemu_aio(ioreq);
 330         return;
 331     }
 332     if (ioreq->aio_inflight > 0) {
 333         return;
 334     }
 335     if (ioreq->postsync) {
 336         ioreq->postsync = 0;
 337         ioreq->aio_inflight++;
 338         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 339         return;
 340     }
 341
 342     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 343     ioreq_unmap(ioreq);
 344     ioreq_finish(ioreq);
 345     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 346     qemu_bh_schedule(ioreq->blkdev->bh);
 347 }
 348
 349 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 350 {
 351     struct XenBlkDev *blkdev = ioreq->blkdev;
 352
 353     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 354         goto err_no_map;
 355     }
 356
 357     ioreq->aio_inflight++;
 358     if (ioreq->presync) {
 359         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 360         return 0;
 361     }
 362
 363     switch (ioreq->req.operation) {
 364     case BLKIF_OP_READ:
 365         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 366         ioreq->aio_inflight++;
 367         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 368                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 369                        qemu_aio_complete, ioreq);
 370         break;
 371     case BLKIF_OP_WRITE:
 372     case BLKIF_OP_WRITE_BARRIER:
 373         if (!ioreq->req.nr_segments) {
 374             break;
 375         }
 376
 377         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 378         ioreq->aio_inflight++;
 379         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 380                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 381                         qemu_aio_complete, ioreq);
 382         break;
 383     default:
 384         /* unknown operation (shouldn't happen -- parse catches this) */
 385         goto err;
 386     }
 387
 388     qemu_aio_complete(ioreq, 0);
 389
 390     return 0;
 391
 392 err:
 393     ioreq_unmap(ioreq);
 394 err_no_map:
 395     ioreq_finish(ioreq);
 396     ioreq->status = BLKIF_RSP_ERROR;
 397     return -1;
 398 }
 399
 400 static int blk_send_response_one(struct ioreq *ioreq)
 401 {
 402     struct XenBlkDev  *blkdev = ioreq->blkdev;
 403     int               send_notify   = 0;
 404     int               have_requests = 0;
 405     blkif_response_t  resp;
 406     void              *dst;
 407
 408     resp.id        = ioreq->req.id;
 409     resp.operation = ioreq->req.operation;
 410     resp.status    = ioreq->status;
 411
 412     /* Place on the response ring for the relevant domain. */
 413     switch (blkdev->protocol) {
 414     case BLKIF_PROTOCOL_NATIVE:
 415         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 416         break;
 417     case BLKIF_PROTOCOL_X86_32:
 418         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 419                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 420         break;
 421     case BLKIF_PROTOCOL_X86_64:
 422         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 423                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 424         break;
 425     default:
 426         dst = NULL;
 427     }
 428     memcpy(dst, &resp, sizeof(resp));
 429     blkdev->rings.common.rsp_prod_pvt++;
 430
 431     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 432     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 433         /*
 434          * Tail check for pending requests. Allows frontend to avoid
 435          * notifications if requests are already in flight (lower
 436          * overheads and promotes batching).
 437          */
 438         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 439     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 440         have_requests = 1;
 441     }
 442
 443     if (have_requests) {
 444         blkdev->more_work++;
 445     }
 446     return send_notify;
 447 }
 448
 449 /* walk finished list, send outstanding responses, free requests */
 450 static void blk_send_response_all(struct XenBlkDev *blkdev)
 451 {
 452     struct ioreq *ioreq;
 453     int send_notify = 0;
 454
 455     while (!QLIST_EMPTY(&blkdev->finished)) {
 456         ioreq = QLIST_FIRST(&blkdev->finished);
 457         send_notify += blk_send_response_one(ioreq);
 458         ioreq_release(ioreq, true);
 459     }
 460     if (send_notify) {
 461         xen_be_send_notify(&blkdev->xendev);
 462     }
 463 }
 464
 465 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 466 {
 467     switch (blkdev->protocol) {
 468     case BLKIF_PROTOCOL_NATIVE:
 469         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 470                sizeof(ioreq->req));
 471         break;
 472     case BLKIF_PROTOCOL_X86_32:
 473         blkif_get_x86_32_req(&ioreq->req,
 474                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 475         break;
 476     case BLKIF_PROTOCOL_X86_64:
 477         blkif_get_x86_64_req(&ioreq->req,
 478                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 479         break;
 480     }
 481     return 0;
 482 }
 483
 484 static void blk_handle_requests(struct XenBlkDev *blkdev)
 485 {
 486     RING_IDX rc, rp;
 487     struct ioreq *ioreq;
 488
 489     blkdev->more_work = 0;
 490
 491     rc = blkdev->rings.common.req_cons;
 492     rp = blkdev->rings.common.sring->req_prod;
 493     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 494
 495     blk_send_response_all(blkdev);
 496     while (rc != rp) {
 497         /* pull request from ring */
 498         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 499             break;
 500         }
 501         ioreq = ioreq_start(blkdev);
 502         if (ioreq == NULL) {
 503             blkdev->more_work++;
 504             break;
 505         }
 506         blk_get_request(blkdev, ioreq, rc);
 507         blkdev->rings.common.req_cons = ++rc;
 508
 509         /* parse them */
 510         if (ioreq_parse(ioreq) != 0) {
 511             if (blk_send_response_one(ioreq)) {
 512                 xen_be_send_notify(&blkdev->xendev);
 513             }
 514             ioreq_release(ioreq, false);
 515             continue;
 516         }
 517
 518         ioreq_runio_qemu_aio(ioreq);
 519     }
 520
 521     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 522         qemu_bh_schedule(blkdev->bh);
 523     }
 524 }
 525
 526 /* ------------------------------------------------------------- */
 527
 528 static void blk_bh(void *opaque)
 529 {
 530     struct XenBlkDev *blkdev = opaque;
 531     blk_handle_requests(blkdev);
 532 }
 533
 534 /*
 535  * We need to account for the grant allocations requiring contiguous
 536  * chunks; the worst case number would be
 537  *     max_req * max_seg + (max_req - 1) * (max_seg - 1) + 1,
 538  * but in order to keep things simple just use
 539  *     2 * max_req * max_seg.
 540  */
 541 #define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
 542
 543 static void blk_alloc(struct XenDevice *xendev)
 544 {
 545     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 546
 547     QLIST_INIT(&blkdev->inflight);
 548     QLIST_INIT(&blkdev->finished);
 549     QLIST_INIT(&blkdev->freelist);
 550     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 551     if (xen_mode != XEN_EMULATE) {
 552         batch_maps = 1;
 553     }
 554     if (xc_gnttab_set_max_grants(xendev->gnttabdev,
 555             MAX_GRANTS(max_requests, BLKIF_MAX_SEGMENTS_PER_REQUEST)) < 0) {
 556         xen_be_printf(xendev, 0, "xc_gnttab_set_max_grants failed: %s\n",
 557                       strerror(errno));
 558     }
 559 }
 560
 561 static int blk_init(struct XenDevice *xendev)
 562 {
 563     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 564     int index, qflags, info = 0;
 565
 566     /* read xenstore entries */
 567     if (blkdev->params == NULL) {
 568         char *h = NULL;
 569         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 570         if (blkdev->params != NULL) {
 571             h = strchr(blkdev->params, ':');
 572         }
 573         if (h != NULL) {
 574             blkdev->fileproto = blkdev->params;
 575             blkdev->filename  = h+1;
 576             *h = 0;
 577         } else {
 578             blkdev->fileproto = "<unset>";
 579             blkdev->filename  = blkdev->params;
 580         }
 581     }
 582     if (!strcmp("aio", blkdev->fileproto)) {
 583         blkdev->fileproto = "raw";
 584     }
 585     if (blkdev->mode == NULL) {
 586         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 587     }
 588     if (blkdev->type == NULL) {
 589         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 590     }
 591     if (blkdev->dev == NULL) {
 592         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 593     }
 594     if (blkdev->devtype == NULL) {
 595         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 596     }
 597
 598     /* do we have all we need? */
 599     if (blkdev->params == NULL ||
 600         blkdev->mode == NULL   ||
 601         blkdev->type == NULL   ||
 602         blkdev->dev == NULL) {
 603         goto out_error;
 604     }
 605
 606     /* read-only ? */
 607     qflags = BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO;
 608     if (strcmp(blkdev->mode, "w") == 0) {
 609         qflags |= BDRV_O_RDWR;
 610     } else {
 611         info  |= VDISK_READONLY;
 612     }
 613
 614     /* cdrom ? */
 615     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 616         info  |= VDISK_CDROM;
 617     }
 618
 619     /* init qemu block driver */
 620     index = (blkdev->xendev.dev - 202 * 256) / 16;
 621     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 622     if (!blkdev->dinfo) {
 623         /* setup via xenbus -> create new block driver instance */
 624         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 625         blkdev->bs = bdrv_new(blkdev->dev);
 626         if (blkdev->bs) {
 627             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 628                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 629                 bdrv_delete(blkdev->bs);
 630                 blkdev->bs = NULL;
 631             }
 632         }
 633         if (!blkdev->bs) {
 634             goto out_error;
 635         }
 636     } else {
 637         /* setup via qemu cmdline -> already setup for us */
 638         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 639         blkdev->bs = blkdev->dinfo->bdrv;
 640     }
 641     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 642     blkdev->file_blk  = BLOCK_SIZE;
 643     blkdev->file_size = bdrv_getlength(blkdev->bs);
 644     if (blkdev->file_size < 0) {
 645         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 646                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 647                       bdrv_get_format_name(blkdev->bs) ?: "-");
 648         blkdev->file_size = 0;
 649     }
 650
 651     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 652                   " size %" PRId64 " (%" PRId64 " MB)\n",
 653                   blkdev->type, blkdev->fileproto, blkdev->filename,
 654                   blkdev->file_size, blkdev->file_size >> 20);
 655
 656     /* fill info */
 657     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", 1);
 658     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 659     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 660     xenstore_write_be_int(&blkdev->xendev, "sectors",
 661                           blkdev->file_size / blkdev->file_blk);
 662     return 0;
 663
 664 out_error:
 665     g_free(blkdev->params);
 666     blkdev->params = NULL;
 667     g_free(blkdev->mode);
 668     blkdev->mode = NULL;
 669     g_free(blkdev->type);
 670     blkdev->type = NULL;
 671     g_free(blkdev->dev);
 672     blkdev->dev = NULL;
 673     g_free(blkdev->devtype);
 674     blkdev->devtype = NULL;
 675     return -1;
 676 }
 677
 678 static int blk_connect(struct XenDevice *xendev)
 679 {
 680     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 681
 682     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 683         return -1;
 684     }
 685     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 686                              &blkdev->xendev.remote_port) == -1) {
 687         return -1;
 688     }
 689
 690     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 691     if (blkdev->xendev.protocol) {
 692         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 693             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 694         }
 695         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 696             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 697         }
 698     }
 699
 700     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 701                                             blkdev->xendev.dom,
 702                                             blkdev->ring_ref,
 703                                             PROT_READ | PROT_WRITE);
 704     if (!blkdev->sring) {
 705         return -1;
 706     }
 707     blkdev->cnt_map++;
 708
 709     switch (blkdev->protocol) {
 710     case BLKIF_PROTOCOL_NATIVE:
 711     {
 712         blkif_sring_t *sring_native = blkdev->sring;
 713         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 714         break;
 715     }
 716     case BLKIF_PROTOCOL_X86_32:
 717     {
 718         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 719
 720         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 721         break;
 722     }
 723     case BLKIF_PROTOCOL_X86_64:
 724     {
 725         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 726
 727         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 728         break;
 729     }
 730     }
 731
 732     xen_be_bind_evtchn(&blkdev->xendev);
 733
 734     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 735                   "remote port %d, local port %d\n",
 736                   blkdev->xendev.protocol, blkdev->ring_ref,
 737                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 738     return 0;
 739 }
 740
 741 static void blk_disconnect(struct XenDevice *xendev)
 742 {
 743     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 744
 745     if (blkdev->bs) {
 746         if (!blkdev->dinfo) {
 747             /* close/delete only if we created it ourself */
 748             bdrv_close(blkdev->bs);
 749             bdrv_detach_dev(blkdev->bs, blkdev);
 750             bdrv_delete(blkdev->bs);
 751         }
 752         blkdev->bs = NULL;
 753     }
 754     xen_be_unbind_evtchn(&blkdev->xendev);
 755
 756     if (blkdev->sring) {
 757         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 758         blkdev->cnt_map--;
 759         blkdev->sring = NULL;
 760     }
 761 }
 762
 763 static int blk_free(struct XenDevice *xendev)
 764 {
 765     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 766     struct ioreq *ioreq;
 767
 768     if (blkdev->bs || blkdev->sring) {
 769         blk_disconnect(xendev);
 770     }
 771
 772     while (!QLIST_EMPTY(&blkdev->freelist)) {
 773         ioreq = QLIST_FIRST(&blkdev->freelist);
 774         QLIST_REMOVE(ioreq, list);
 775         qemu_iovec_destroy(&ioreq->v);
 776         g_free(ioreq);
 777     }
 778
 779     g_free(blkdev->params);
 780     g_free(blkdev->mode);
 781     g_free(blkdev->type);
 782     g_free(blkdev->dev);
 783     g_free(blkdev->devtype);
 784     qemu_bh_delete(blkdev->bh);
 785     return 0;
 786 }
 787
 788 static void blk_event(struct XenDevice *xendev)
 789 {
 790     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 791
 792     qemu_bh_schedule(blkdev->bh);
 793 }
 794
 795 struct XenDevOps xen_blkdev_ops = {
 796     .size       = sizeof(struct XenBlkDev),
 797     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 798     .alloc      = blk_alloc,
 799     .init       = blk_init,
 800     .initialise    = blk_connect,
 801     .disconnect = blk_disconnect,
 802     .event      = blk_event,
 803     .free       = blk_free,
 804 };