hw/xen_disk.c

   1 /*
   2  *  xen paravirt block device backend
   3  *
   4  *  (c) Gerd Hoffmann <[email protected]>
   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     uint8_t             mapped;
  70
  71     /* grant mapping */
  72     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  73     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  74     int                 prot;
  75     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  76     void                *pages;
  77
  78     /* aio status */
  79     int                 aio_inflight;
  80     int                 aio_errors;
  81
  82     struct XenBlkDev    *blkdev;
  83     QLIST_ENTRY(ioreq)   list;
  84     BlockAcctCookie     acct;
  85 };
  86
  87 struct XenBlkDev {
  88     struct XenDevice    xendev;  /* must be first */
  89     char                *params;
  90     char                *mode;
  91     char                *type;
  92     char                *dev;
  93     char                *devtype;
  94     const char          *fileproto;
  95     const char          *filename;
  96     int                 ring_ref;
  97     void                *sring;
  98     int64_t             file_blk;
  99     int64_t             file_size;
 100     int                 protocol;
 101     blkif_back_rings_t  rings;
 102     int                 more_work;
 103     int                 cnt_map;
 104
 105     /* request lists */
 106     QLIST_HEAD(inflight_head, ioreq) inflight;
 107     QLIST_HEAD(finished_head, ioreq) finished;
 108     QLIST_HEAD(freelist_head, ioreq) freelist;
 109     int                 requests_total;
 110     int                 requests_inflight;
 111     int                 requests_finished;
 112
 113     /* qemu block driver */
 114     DriveInfo           *dinfo;
 115     BlockDriverState    *bs;
 116     QEMUBH              *bh;
 117 };
 118
 119 /* ------------------------------------------------------------- */
 120
 121 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 122 {
 123     struct ioreq *ioreq = NULL;
 124
 125     if (QLIST_EMPTY(&blkdev->freelist)) {
 126         if (blkdev->requests_total >= max_requests) {
 127             goto out;
 128         }
 129         /* allocate new struct */
 130         ioreq = g_malloc0(sizeof(*ioreq));
 131         ioreq->blkdev = blkdev;
 132         blkdev->requests_total++;
 133         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 134     } else {
 135         /* get one from freelist */
 136         ioreq = QLIST_FIRST(&blkdev->freelist);
 137         QLIST_REMOVE(ioreq, list);
 138         qemu_iovec_reset(&ioreq->v);
 139     }
 140     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 141     blkdev->requests_inflight++;
 142
 143 out:
 144     return ioreq;
 145 }
 146
 147 static void ioreq_finish(struct ioreq *ioreq)
 148 {
 149     struct XenBlkDev *blkdev = ioreq->blkdev;
 150
 151     QLIST_REMOVE(ioreq, list);
 152     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 153     blkdev->requests_inflight--;
 154     blkdev->requests_finished++;
 155 }
 156
 157 static void ioreq_release(struct ioreq *ioreq, bool finish)
 158 {
 159     struct XenBlkDev *blkdev = ioreq->blkdev;
 160
 161     QLIST_REMOVE(ioreq, list);
 162     memset(ioreq, 0, sizeof(*ioreq));
 163     ioreq->blkdev = blkdev;
 164     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 165     if (finish) {
 166         blkdev->requests_finished--;
 167     } else {
 168         blkdev->requests_inflight--;
 169     }
 170 }
 171
 172 /*
 173  * translate request into iovec + start offset
 174  * do sanity checks along the way
 175  */
 176 static int ioreq_parse(struct ioreq *ioreq)
 177 {
 178     struct XenBlkDev *blkdev = ioreq->blkdev;
 179     uintptr_t mem;
 180     size_t len;
 181     int i;
 182
 183     xen_be_printf(&blkdev->xendev, 3,
 184                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 185                   ioreq->req.operation, ioreq->req.nr_segments,
 186                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 187     switch (ioreq->req.operation) {
 188     case BLKIF_OP_READ:
 189         ioreq->prot = PROT_WRITE; /* to memory */
 190         break;
 191     case BLKIF_OP_WRITE_BARRIER:
 192         if (!ioreq->req.nr_segments) {
 193             ioreq->presync = 1;
 194             return 0;
 195         }
 196         ioreq->presync = ioreq->postsync = 1;
 197         /* fall through */
 198     case BLKIF_OP_WRITE:
 199         ioreq->prot = PROT_READ; /* from memory */
 200         break;
 201     default:
 202         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 203                       ioreq->req.operation);
 204         goto err;
 205     };
 206
 207     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 208         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 209         goto err;
 210     }
 211
 212     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 213     for (i = 0; i < ioreq->req.nr_segments; i++) {
 214         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 215             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 216             goto err;
 217         }
 218         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 219             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 220             goto err;
 221         }
 222         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 223             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 224             goto err;
 225         }
 226
 227         ioreq->domids[i] = blkdev->xendev.dom;
 228         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 229
 230         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 231         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 232         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 233     }
 234     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 235         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 236         goto err;
 237     }
 238     return 0;
 239
 240 err:
 241     ioreq->status = BLKIF_RSP_ERROR;
 242     return -1;
 243 }
 244
 245 static void ioreq_unmap(struct ioreq *ioreq)
 246 {
 247     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 248     int i;
 249
 250     if (ioreq->v.niov == 0 || ioreq->mapped == 0) {
 251         return;
 252     }
 253     if (batch_maps) {
 254         if (!ioreq->pages) {
 255             return;
 256         }
 257         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0) {
 258             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 259                           strerror(errno));
 260         }
 261         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 262         ioreq->pages = NULL;
 263     } else {
 264         for (i = 0; i < ioreq->v.niov; i++) {
 265             if (!ioreq->page[i]) {
 266                 continue;
 267             }
 268             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0) {
 269                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 270                               strerror(errno));
 271             }
 272             ioreq->blkdev->cnt_map--;
 273             ioreq->page[i] = NULL;
 274         }
 275     }
 276     ioreq->mapped = 0;
 277 }
 278
 279 static int ioreq_map(struct ioreq *ioreq)
 280 {
 281     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 282     int i;
 283
 284     if (ioreq->v.niov == 0 || ioreq->mapped == 1) {
 285         return 0;
 286     }
 287     if (batch_maps) {
 288         ioreq->pages = xc_gnttab_map_grant_refs
 289             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 290         if (ioreq->pages == NULL) {
 291             xen_be_printf(&ioreq->blkdev->xendev, 0,
 292                           "can't map %d grant refs (%s, %d maps)\n",
 293                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 294             return -1;
 295         }
 296         for (i = 0; i < ioreq->v.niov; i++) {
 297             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 298                 (uintptr_t)ioreq->v.iov[i].iov_base;
 299         }
 300         ioreq->blkdev->cnt_map += ioreq->v.niov;
 301     } else  {
 302         for (i = 0; i < ioreq->v.niov; i++) {
 303             ioreq->page[i] = xc_gnttab_map_grant_ref
 304                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 305             if (ioreq->page[i] == NULL) {
 306                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 307                               "can't map grant ref %d (%s, %d maps)\n",
 308                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 309                 ioreq_unmap(ioreq);
 310                 return -1;
 311             }
 312             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 313             ioreq->blkdev->cnt_map++;
 314         }
 315     }
 316     ioreq->mapped = 1;
 317     return 0;
 318 }
 319
 320 static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
 321
 322 static void qemu_aio_complete(void *opaque, int ret)
 323 {
 324     struct ioreq *ioreq = opaque;
 325
 326     if (ret != 0) {
 327         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 328                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 329         ioreq->aio_errors++;
 330     }
 331
 332     ioreq->aio_inflight--;
 333     if (ioreq->presync) {
 334         ioreq->presync = 0;
 335         ioreq_runio_qemu_aio(ioreq);
 336         return;
 337     }
 338     if (ioreq->aio_inflight > 0) {
 339         return;
 340     }
 341     if (ioreq->postsync) {
 342         ioreq->postsync = 0;
 343         ioreq->aio_inflight++;
 344         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 345         return;
 346     }
 347
 348     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 349     ioreq_unmap(ioreq);
 350     ioreq_finish(ioreq);
 351     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 352     qemu_bh_schedule(ioreq->blkdev->bh);
 353 }
 354
 355 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 356 {
 357     struct XenBlkDev *blkdev = ioreq->blkdev;
 358
 359     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 360         goto err_no_map;
 361     }
 362
 363     ioreq->aio_inflight++;
 364     if (ioreq->presync) {
 365         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 366         return 0;
 367     }
 368
 369     switch (ioreq->req.operation) {
 370     case BLKIF_OP_READ:
 371         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 372         ioreq->aio_inflight++;
 373         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 374                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 375                        qemu_aio_complete, ioreq);
 376         break;
 377     case BLKIF_OP_WRITE:
 378     case BLKIF_OP_WRITE_BARRIER:
 379         if (!ioreq->req.nr_segments) {
 380             break;
 381         }
 382
 383         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 384         ioreq->aio_inflight++;
 385         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 386                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 387                         qemu_aio_complete, ioreq);
 388         break;
 389     default:
 390         /* unknown operation (shouldn't happen -- parse catches this) */
 391         goto err;
 392     }
 393
 394     qemu_aio_complete(ioreq, 0);
 395
 396     return 0;
 397
 398 err:
 399     ioreq_unmap(ioreq);
 400 err_no_map:
 401     ioreq_finish(ioreq);
 402     ioreq->status = BLKIF_RSP_ERROR;
 403     return -1;
 404 }
 405
 406 static int blk_send_response_one(struct ioreq *ioreq)
 407 {
 408     struct XenBlkDev  *blkdev = ioreq->blkdev;
 409     int               send_notify   = 0;
 410     int               have_requests = 0;
 411     blkif_response_t  resp;
 412     void              *dst;
 413
 414     resp.id        = ioreq->req.id;
 415     resp.operation = ioreq->req.operation;
 416     resp.status    = ioreq->status;
 417
 418     /* Place on the response ring for the relevant domain. */
 419     switch (blkdev->protocol) {
 420     case BLKIF_PROTOCOL_NATIVE:
 421         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 422         break;
 423     case BLKIF_PROTOCOL_X86_32:
 424         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 425                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 426         break;
 427     case BLKIF_PROTOCOL_X86_64:
 428         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 429                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 430         break;
 431     default:
 432         dst = NULL;
 433     }
 434     memcpy(dst, &resp, sizeof(resp));
 435     blkdev->rings.common.rsp_prod_pvt++;
 436
 437     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 438     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 439         /*
 440          * Tail check for pending requests. Allows frontend to avoid
 441          * notifications if requests are already in flight (lower
 442          * overheads and promotes batching).
 443          */
 444         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 445     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 446         have_requests = 1;
 447     }
 448
 449     if (have_requests) {
 450         blkdev->more_work++;
 451     }
 452     return send_notify;
 453 }
 454
 455 /* walk finished list, send outstanding responses, free requests */
 456 static void blk_send_response_all(struct XenBlkDev *blkdev)
 457 {
 458     struct ioreq *ioreq;
 459     int send_notify = 0;
 460
 461     while (!QLIST_EMPTY(&blkdev->finished)) {
 462         ioreq = QLIST_FIRST(&blkdev->finished);
 463         send_notify += blk_send_response_one(ioreq);
 464         ioreq_release(ioreq, true);
 465     }
 466     if (send_notify) {
 467         xen_be_send_notify(&blkdev->xendev);
 468     }
 469 }
 470
 471 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 472 {
 473     switch (blkdev->protocol) {
 474     case BLKIF_PROTOCOL_NATIVE:
 475         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 476                sizeof(ioreq->req));
 477         break;
 478     case BLKIF_PROTOCOL_X86_32:
 479         blkif_get_x86_32_req(&ioreq->req,
 480                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 481         break;
 482     case BLKIF_PROTOCOL_X86_64:
 483         blkif_get_x86_64_req(&ioreq->req,
 484                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 485         break;
 486     }
 487     return 0;
 488 }
 489
 490 static void blk_handle_requests(struct XenBlkDev *blkdev)
 491 {
 492     RING_IDX rc, rp;
 493     struct ioreq *ioreq;
 494
 495     blkdev->more_work = 0;
 496
 497     rc = blkdev->rings.common.req_cons;
 498     rp = blkdev->rings.common.sring->req_prod;
 499     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 500
 501     blk_send_response_all(blkdev);
 502     while (rc != rp) {
 503         /* pull request from ring */
 504         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 505             break;
 506         }
 507         ioreq = ioreq_start(blkdev);
 508         if (ioreq == NULL) {
 509             blkdev->more_work++;
 510             break;
 511         }
 512         blk_get_request(blkdev, ioreq, rc);
 513         blkdev->rings.common.req_cons = ++rc;
 514
 515         /* parse them */
 516         if (ioreq_parse(ioreq) != 0) {
 517             if (blk_send_response_one(ioreq)) {
 518                 xen_be_send_notify(&blkdev->xendev);
 519             }
 520             ioreq_release(ioreq, false);
 521             continue;
 522         }
 523
 524         ioreq_runio_qemu_aio(ioreq);
 525     }
 526
 527     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 528         qemu_bh_schedule(blkdev->bh);
 529     }
 530 }
 531
 532 /* ------------------------------------------------------------- */
 533
 534 static void blk_bh(void *opaque)
 535 {
 536     struct XenBlkDev *blkdev = opaque;
 537     blk_handle_requests(blkdev);
 538 }
 539
 540 /*
 541  * We need to account for the grant allocations requiring contiguous
 542  * chunks; the worst case number would be
 543  *     max_req * max_seg + (max_req - 1) * (max_seg - 1) + 1,
 544  * but in order to keep things simple just use
 545  *     2 * max_req * max_seg.
 546  */
 547 #define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
 548
 549 static void blk_alloc(struct XenDevice *xendev)
 550 {
 551     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 552
 553     QLIST_INIT(&blkdev->inflight);
 554     QLIST_INIT(&blkdev->finished);
 555     QLIST_INIT(&blkdev->freelist);
 556     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 557     if (xen_mode != XEN_EMULATE) {
 558         batch_maps = 1;
 559     }
 560     if (xc_gnttab_set_max_grants(xendev->gnttabdev,
 561             MAX_GRANTS(max_requests, BLKIF_MAX_SEGMENTS_PER_REQUEST)) < 0) {
 562         xen_be_printf(xendev, 0, "xc_gnttab_set_max_grants failed: %s\n",
 563                       strerror(errno));
 564     }
 565 }
 566
 567 static int blk_init(struct XenDevice *xendev)
 568 {
 569     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 570     int index, qflags, info = 0;
 571
 572     /* read xenstore entries */
 573     if (blkdev->params == NULL) {
 574         char *h = NULL;
 575         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 576         if (blkdev->params != NULL) {
 577             h = strchr(blkdev->params, ':');
 578         }
 579         if (h != NULL) {
 580             blkdev->fileproto = blkdev->params;
 581             blkdev->filename  = h+1;
 582             *h = 0;
 583         } else {
 584             blkdev->fileproto = "<unset>";
 585             blkdev->filename  = blkdev->params;
 586         }
 587     }
 588     if (!strcmp("aio", blkdev->fileproto)) {
 589         blkdev->fileproto = "raw";
 590     }
 591     if (blkdev->mode == NULL) {
 592         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 593     }
 594     if (blkdev->type == NULL) {
 595         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 596     }
 597     if (blkdev->dev == NULL) {
 598         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 599     }
 600     if (blkdev->devtype == NULL) {
 601         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 602     }
 603
 604     /* do we have all we need? */
 605     if (blkdev->params == NULL ||
 606         blkdev->mode == NULL   ||
 607         blkdev->type == NULL   ||
 608         blkdev->dev == NULL) {
 609         goto out_error;
 610     }
 611
 612     /* read-only ? */
 613     qflags = BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO;
 614     if (strcmp(blkdev->mode, "w") == 0) {
 615         qflags |= BDRV_O_RDWR;
 616     } else {
 617         info  |= VDISK_READONLY;
 618     }
 619
 620     /* cdrom ? */
 621     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 622         info  |= VDISK_CDROM;
 623     }
 624
 625     /* init qemu block driver */
 626     index = (blkdev->xendev.dev - 202 * 256) / 16;
 627     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 628     if (!blkdev->dinfo) {
 629         /* setup via xenbus -> create new block driver instance */
 630         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 631         blkdev->bs = bdrv_new(blkdev->dev);
 632         if (blkdev->bs) {
 633             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 634                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 635                 bdrv_delete(blkdev->bs);
 636                 blkdev->bs = NULL;
 637             }
 638         }
 639         if (!blkdev->bs) {
 640             goto out_error;
 641         }
 642     } else {
 643         /* setup via qemu cmdline -> already setup for us */
 644         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 645         blkdev->bs = blkdev->dinfo->bdrv;
 646     }
 647     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 648     blkdev->file_blk  = BLOCK_SIZE;
 649     blkdev->file_size = bdrv_getlength(blkdev->bs);
 650     if (blkdev->file_size < 0) {
 651         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 652                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 653                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 654         blkdev->file_size = 0;
 655     }
 656
 657     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 658                   " size %" PRId64 " (%" PRId64 " MB)\n",
 659                   blkdev->type, blkdev->fileproto, blkdev->filename,
 660                   blkdev->file_size, blkdev->file_size >> 20);
 661
 662     /* fill info */
 663     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", 1);
 664     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 665     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 666     xenstore_write_be_int(&blkdev->xendev, "sectors",
 667                           blkdev->file_size / blkdev->file_blk);
 668     return 0;
 669
 670 out_error:
 671     g_free(blkdev->params);
 672     blkdev->params = NULL;
 673     g_free(blkdev->mode);
 674     blkdev->mode = NULL;
 675     g_free(blkdev->type);
 676     blkdev->type = NULL;
 677     g_free(blkdev->dev);
 678     blkdev->dev = NULL;
 679     g_free(blkdev->devtype);
 680     blkdev->devtype = NULL;
 681     return -1;
 682 }
 683
 684 static int blk_connect(struct XenDevice *xendev)
 685 {
 686     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 687
 688     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 689         return -1;
 690     }
 691     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 692                              &blkdev->xendev.remote_port) == -1) {
 693         return -1;
 694     }
 695
 696     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 697     if (blkdev->xendev.protocol) {
 698         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 699             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 700         }
 701         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 702             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 703         }
 704     }
 705
 706     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 707                                             blkdev->xendev.dom,
 708                                             blkdev->ring_ref,
 709                                             PROT_READ | PROT_WRITE);
 710     if (!blkdev->sring) {
 711         return -1;
 712     }
 713     blkdev->cnt_map++;
 714
 715     switch (blkdev->protocol) {
 716     case BLKIF_PROTOCOL_NATIVE:
 717     {
 718         blkif_sring_t *sring_native = blkdev->sring;
 719         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 720         break;
 721     }
 722     case BLKIF_PROTOCOL_X86_32:
 723     {
 724         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 725
 726         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 727         break;
 728     }
 729     case BLKIF_PROTOCOL_X86_64:
 730     {
 731         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 732
 733         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 734         break;
 735     }
 736     }
 737
 738     xen_be_bind_evtchn(&blkdev->xendev);
 739
 740     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 741                   "remote port %d, local port %d\n",
 742                   blkdev->xendev.protocol, blkdev->ring_ref,
 743                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 744     return 0;
 745 }
 746
 747 static void blk_disconnect(struct XenDevice *xendev)
 748 {
 749     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 750
 751     if (blkdev->bs) {
 752         if (!blkdev->dinfo) {
 753             /* close/delete only if we created it ourself */
 754             bdrv_close(blkdev->bs);
 755             bdrv_detach_dev(blkdev->bs, blkdev);
 756             bdrv_delete(blkdev->bs);
 757         }
 758         blkdev->bs = NULL;
 759     }
 760     xen_be_unbind_evtchn(&blkdev->xendev);
 761
 762     if (blkdev->sring) {
 763         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 764         blkdev->cnt_map--;
 765         blkdev->sring = NULL;
 766     }
 767 }
 768
 769 static int blk_free(struct XenDevice *xendev)
 770 {
 771     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 772     struct ioreq *ioreq;
 773
 774     if (blkdev->bs || blkdev->sring) {
 775         blk_disconnect(xendev);
 776     }
 777
 778     while (!QLIST_EMPTY(&blkdev->freelist)) {
 779         ioreq = QLIST_FIRST(&blkdev->freelist);
 780         QLIST_REMOVE(ioreq, list);
 781         qemu_iovec_destroy(&ioreq->v);
 782         g_free(ioreq);
 783     }
 784
 785     g_free(blkdev->params);
 786     g_free(blkdev->mode);
 787     g_free(blkdev->type);
 788     g_free(blkdev->dev);
 789     g_free(blkdev->devtype);
 790     qemu_bh_delete(blkdev->bh);
 791     return 0;
 792 }
 793
 794 static void blk_event(struct XenDevice *xendev)
 795 {
 796     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 797
 798     qemu_bh_schedule(blkdev->bh);
 799 }
 800
 801 struct XenDevOps xen_blkdev_ops = {
 802     .size       = sizeof(struct XenBlkDev),
 803     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 804     .alloc      = blk_alloc,
 805     .init       = blk_init,
 806     .initialise    = blk_connect,
 807     .disconnect = blk_disconnect,
 808     .event      = blk_event,
 809     .free       = blk_free,
 810 };