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, write to the Free Software Foundation, Inc.,
  17  *  51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19
  20 #include <stdio.h>
  21 #include <stdlib.h>
  22 #include <stdarg.h>
  23 #include <string.h>
  24 #include <unistd.h>
  25 #include <signal.h>
  26 #include <inttypes.h>
  27 #include <time.h>
  28 #include <fcntl.h>
  29 #include <errno.h>
  30 #include <sys/ioctl.h>
  31 #include <sys/types.h>
  32 #include <sys/stat.h>
  33 #include <sys/mman.h>
  34 #include <sys/uio.h>
  35
  36 #include <xs.h>
  37 #include <xenctrl.h>
  38 #include <xen/io/xenbus.h>
  39
  40 #include "hw.h"
  41 #include "block_int.h"
  42 #include "qemu-char.h"
  43 #include "xen_blkif.h"
  44 #include "xen_backend.h"
  45
  46 /* ------------------------------------------------------------- */
  47
  48 static int syncwrite    = 0;
  49 static int batch_maps   = 0;
  50
  51 static int max_requests = 32;
  52 static int use_aio      = 1;
  53
  54 /* ------------------------------------------------------------- */
  55
  56 #define BLOCK_SIZE  512
  57 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  58
  59 struct ioreq {
  60     blkif_request_t     req;
  61     int16_t             status;
  62
  63     /* parsed request */
  64     off_t               start;
  65     QEMUIOVector        v;
  66     int                 presync;
  67     int                 postsync;
  68
  69     /* grant mapping */
  70     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  71     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  72     int                 prot;
  73     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  74     void                *pages;
  75
  76     /* aio status */
  77     int                 aio_inflight;
  78     int                 aio_errors;
  79
  80     struct XenBlkDev    *blkdev;
  81     LIST_ENTRY(ioreq)   list;
  82 };
  83
  84 struct XenBlkDev {
  85     struct XenDevice    xendev;  /* must be first */
  86     char                *params;
  87     char                *mode;
  88     char                *type;
  89     char                *dev;
  90     char                *devtype;
  91     const char          *fileproto;
  92     const char          *filename;
  93     int                 ring_ref;
  94     void                *sring;
  95     int64_t             file_blk;
  96     int64_t             file_size;
  97     int                 protocol;
  98     blkif_back_rings_t  rings;
  99     int                 more_work;
 100     int                 cnt_map;
 101
 102     /* request lists */
 103     LIST_HEAD(inflight_head, ioreq) inflight;
 104     LIST_HEAD(finished_head, ioreq) finished;
 105     LIST_HEAD(freelist_head, ioreq) freelist;
 106     int                 requests_total;
 107     int                 requests_inflight;
 108     int                 requests_finished;
 109
 110     /* qemu block driver */
 111     int                 index;
 112     BlockDriverState    *bs;
 113     QEMUBH              *bh;
 114 };
 115
 116 /* ------------------------------------------------------------- */
 117
 118 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 119 {
 120     struct ioreq *ioreq = NULL;
 121
 122     if (LIST_EMPTY(&blkdev->freelist)) {
 123         if (blkdev->requests_total >= max_requests)
 124             goto out;
 125         /* allocate new struct */
 126         ioreq = qemu_mallocz(sizeof(*ioreq));
 127         ioreq->blkdev = blkdev;
 128         blkdev->requests_total++;
 129         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 130     } else {
 131         /* get one from freelist */
 132         ioreq = LIST_FIRST(&blkdev->freelist);
 133         LIST_REMOVE(ioreq, list);
 134         qemu_iovec_reset(&ioreq->v);
 135     }
 136     LIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 137     blkdev->requests_inflight++;
 138
 139 out:
 140     return ioreq;
 141 }
 142
 143 static void ioreq_finish(struct ioreq *ioreq)
 144 {
 145     struct XenBlkDev *blkdev = ioreq->blkdev;
 146
 147     LIST_REMOVE(ioreq, list);
 148     LIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 149     blkdev->requests_inflight--;
 150     blkdev->requests_finished++;
 151 }
 152
 153 static void ioreq_release(struct ioreq *ioreq)
 154 {
 155     struct XenBlkDev *blkdev = ioreq->blkdev;
 156
 157     LIST_REMOVE(ioreq, list);
 158     memset(ioreq, 0, sizeof(*ioreq));
 159     ioreq->blkdev = blkdev;
 160     LIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 161     blkdev->requests_finished--;
 162 }
 163
 164 /*
 165  * translate request into iovec + start offset
 166  * do sanity checks along the way
 167  */
 168 static int ioreq_parse(struct ioreq *ioreq)
 169 {
 170     struct XenBlkDev *blkdev = ioreq->blkdev;
 171     uintptr_t mem;
 172     size_t len;
 173     int i;
 174
 175     xen_be_printf(&blkdev->xendev, 3,
 176                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 177                   ioreq->req.operation, ioreq->req.nr_segments,
 178                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 179     switch (ioreq->req.operation) {
 180     case BLKIF_OP_READ:
 181         ioreq->prot = PROT_WRITE; /* to memory */
 182         break;
 183     case BLKIF_OP_WRITE_BARRIER:
 184         if (!syncwrite)
 185             ioreq->presync = ioreq->postsync = 1;
 186         /* fall through */
 187     case BLKIF_OP_WRITE:
 188         ioreq->prot = PROT_READ; /* from memory */
 189         if (syncwrite)
 190             ioreq->postsync = 1;
 191         break;
 192     default:
 193         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 194                       ioreq->req.operation);
 195         goto err;
 196     };
 197
 198     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 199         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 200         goto err;
 201     }
 202
 203     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 204     for (i = 0; i < ioreq->req.nr_segments; i++) {
 205         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 206             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 207             goto err;
 208         }
 209         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 210             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 211             goto err;
 212         }
 213         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 214             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 215             goto err;
 216         }
 217
 218         ioreq->domids[i] = blkdev->xendev.dom;
 219         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 220
 221         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 222         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 223         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 224     }
 225     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 226         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 227         goto err;
 228     }
 229     return 0;
 230
 231 err:
 232     ioreq->status = BLKIF_RSP_ERROR;
 233     return -1;
 234 }
 235
 236 static void ioreq_unmap(struct ioreq *ioreq)
 237 {
 238     int gnt = ioreq->blkdev->xendev.gnttabdev;
 239     int i;
 240
 241     if (ioreq->v.niov == 0)
 242         return;
 243     if (batch_maps) {
 244         if (!ioreq->pages)
 245             return;
 246         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0)
 247             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 248                           strerror(errno));
 249         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 250         ioreq->pages = NULL;
 251     } else {
 252         for (i = 0; i < ioreq->v.niov; i++) {
 253             if (!ioreq->page[i])
 254                 continue;
 255             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0)
 256                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 257                               strerror(errno));
 258             ioreq->blkdev->cnt_map--;
 259             ioreq->page[i] = NULL;
 260         }
 261     }
 262 }
 263
 264 static int ioreq_map(struct ioreq *ioreq)
 265 {
 266     int gnt = ioreq->blkdev->xendev.gnttabdev;
 267     int i;
 268
 269     if (ioreq->v.niov == 0)
 270         return 0;
 271     if (batch_maps) {
 272         ioreq->pages = xc_gnttab_map_grant_refs
 273             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 274         if (ioreq->pages == NULL) {
 275             xen_be_printf(&ioreq->blkdev->xendev, 0,
 276                           "can't map %d grant refs (%s, %d maps)\n",
 277                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 278             return -1;
 279         }
 280         for (i = 0; i < ioreq->v.niov; i++)
 281             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 282                 (uintptr_t)ioreq->v.iov[i].iov_base;
 283         ioreq->blkdev->cnt_map += ioreq->v.niov;
 284     } else  {
 285         for (i = 0; i < ioreq->v.niov; i++) {
 286             ioreq->page[i] = xc_gnttab_map_grant_ref
 287                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 288             if (ioreq->page[i] == NULL) {
 289                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 290                               "can't map grant ref %d (%s, %d maps)\n",
 291                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 292                 ioreq_unmap(ioreq);
 293                 return -1;
 294             }
 295             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 296             ioreq->blkdev->cnt_map++;
 297         }
 298     }
 299     return 0;
 300 }
 301
 302 static int ioreq_runio_qemu_sync(struct ioreq *ioreq)
 303 {
 304     struct XenBlkDev *blkdev = ioreq->blkdev;
 305     int i, rc, len = 0;
 306     off_t pos;
 307
 308     if (ioreq_map(ioreq) == -1)
 309         goto err;
 310     if (ioreq->presync)
 311         bdrv_flush(blkdev->bs);
 312
 313     switch (ioreq->req.operation) {
 314     case BLKIF_OP_READ:
 315         pos = ioreq->start;
 316         for (i = 0; i < ioreq->v.niov; i++) {
 317             rc = bdrv_read(blkdev->bs, pos / BLOCK_SIZE,
 318                            ioreq->v.iov[i].iov_base,
 319                            ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 320             if (rc != 0) {
 321                 xen_be_printf(&blkdev->xendev, 0, "rd I/O error (%p, len %zd)\n",
 322                               ioreq->v.iov[i].iov_base,
 323                               ioreq->v.iov[i].iov_len);
 324                 goto err;
 325             }
 326             len += ioreq->v.iov[i].iov_len;
 327             pos += ioreq->v.iov[i].iov_len;
 328         }
 329         break;
 330     case BLKIF_OP_WRITE:
 331     case BLKIF_OP_WRITE_BARRIER:
 332         pos = ioreq->start;
 333         for (i = 0; i < ioreq->v.niov; i++) {
 334             rc = bdrv_write(blkdev->bs, pos / BLOCK_SIZE,
 335                             ioreq->v.iov[i].iov_base,
 336                             ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 337             if (rc != 0) {
 338                 xen_be_printf(&blkdev->xendev, 0, "wr I/O error (%p, len %zd)\n",
 339                               ioreq->v.iov[i].iov_base,
 340                               ioreq->v.iov[i].iov_len);
 341                 goto err;
 342             }
 343             len += ioreq->v.iov[i].iov_len;
 344             pos += ioreq->v.iov[i].iov_len;
 345         }
 346         break;
 347     default:
 348         /* unknown operation (shouldn't happen -- parse catches this) */
 349         goto err;
 350     }
 351
 352     if (ioreq->postsync)
 353         bdrv_flush(blkdev->bs);
 354     ioreq->status = BLKIF_RSP_OKAY;
 355
 356     ioreq_unmap(ioreq);
 357     ioreq_finish(ioreq);
 358     return 0;
 359
 360 err:
 361     ioreq->status = BLKIF_RSP_ERROR;
 362     return -1;
 363 }
 364
 365 static void qemu_aio_complete(void *opaque, int ret)
 366 {
 367     struct ioreq *ioreq = opaque;
 368
 369     if (ret != 0) {
 370         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 371                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 372         ioreq->aio_errors++;
 373     }
 374
 375     ioreq->aio_inflight--;
 376     if (ioreq->aio_inflight > 0)
 377         return;
 378
 379     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 380     ioreq_unmap(ioreq);
 381     ioreq_finish(ioreq);
 382     qemu_bh_schedule(ioreq->blkdev->bh);
 383 }
 384
 385 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 386 {
 387     struct XenBlkDev *blkdev = ioreq->blkdev;
 388
 389     if (ioreq_map(ioreq) == -1)
 390         goto err;
 391
 392     ioreq->aio_inflight++;
 393     if (ioreq->presync)
 394         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 395
 396     switch (ioreq->req.operation) {
 397     case BLKIF_OP_READ:
 398         ioreq->aio_inflight++;
 399         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 400                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 401                        qemu_aio_complete, ioreq);
 402         break;
 403     case BLKIF_OP_WRITE:
 404     case BLKIF_OP_WRITE_BARRIER:
 405         ioreq->aio_inflight++;
 406         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 407                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 408                         qemu_aio_complete, ioreq);
 409         break;
 410     default:
 411         /* unknown operation (shouldn't happen -- parse catches this) */
 412         goto err;
 413     }
 414
 415     if (ioreq->postsync)
 416         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 417     qemu_aio_complete(ioreq, 0);
 418
 419     return 0;
 420
 421 err:
 422     ioreq->status = BLKIF_RSP_ERROR;
 423     return -1;
 424 }
 425
 426 static int blk_send_response_one(struct ioreq *ioreq)
 427 {
 428     struct XenBlkDev  *blkdev = ioreq->blkdev;
 429     int               send_notify   = 0;
 430     int               have_requests = 0;
 431     blkif_response_t  resp;
 432     void              *dst;
 433
 434     resp.id        = ioreq->req.id;
 435     resp.operation = ioreq->req.operation;
 436     resp.status    = ioreq->status;
 437
 438     /* Place on the response ring for the relevant domain. */
 439     switch (blkdev->protocol) {
 440     case BLKIF_PROTOCOL_NATIVE:
 441         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 442         break;
 443     case BLKIF_PROTOCOL_X86_32:
 444         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 445                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 446         break;
 447     case BLKIF_PROTOCOL_X86_64:
 448         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 449                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 450         break;
 451     default:
 452         dst = NULL;
 453     }
 454     memcpy(dst, &resp, sizeof(resp));
 455     blkdev->rings.common.rsp_prod_pvt++;
 456
 457     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 458     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 459         /*
 460          * Tail check for pending requests. Allows frontend to avoid
 461          * notifications if requests are already in flight (lower
 462          * overheads and promotes batching).
 463          */
 464         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 465     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 466         have_requests = 1;
 467     }
 468
 469     if (have_requests)
 470         blkdev->more_work++;
 471     return send_notify;
 472 }
 473
 474 /* walk finished list, send outstanding responses, free requests */
 475 static void blk_send_response_all(struct XenBlkDev *blkdev)
 476 {
 477     struct ioreq *ioreq;
 478     int send_notify = 0;
 479
 480     while (!LIST_EMPTY(&blkdev->finished)) {
 481         ioreq = LIST_FIRST(&blkdev->finished);
 482         send_notify += blk_send_response_one(ioreq);
 483         ioreq_release(ioreq);
 484     }
 485     if (send_notify)
 486         xen_be_send_notify(&blkdev->xendev);
 487 }
 488
 489 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 490 {
 491     switch (blkdev->protocol) {
 492     case BLKIF_PROTOCOL_NATIVE:
 493         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 494                sizeof(ioreq->req));
 495         break;
 496     case BLKIF_PROTOCOL_X86_32:
 497         blkif_get_x86_32_req(&ioreq->req,
 498                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 499         break;
 500     case BLKIF_PROTOCOL_X86_64:
 501         blkif_get_x86_64_req(&ioreq->req,
 502                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 503         break;
 504     }
 505     return 0;
 506 }
 507
 508 static void blk_handle_requests(struct XenBlkDev *blkdev)
 509 {
 510     RING_IDX rc, rp;
 511     struct ioreq *ioreq;
 512
 513     blkdev->more_work = 0;
 514
 515     rc = blkdev->rings.common.req_cons;
 516     rp = blkdev->rings.common.sring->req_prod;
 517     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 518
 519     if (use_aio)
 520         blk_send_response_all(blkdev);
 521     while (rc != rp) {
 522         /* pull request from ring */
 523         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc))
 524             break;
 525         ioreq = ioreq_start(blkdev);
 526         if (ioreq == NULL) {
 527             blkdev->more_work++;
 528             break;
 529         }
 530         blk_get_request(blkdev, ioreq, rc);
 531         blkdev->rings.common.req_cons = ++rc;
 532
 533         /* parse them */
 534         if (ioreq_parse(ioreq) != 0) {
 535             if (blk_send_response_one(ioreq))
 536                 xen_be_send_notify(&blkdev->xendev);
 537             ioreq_release(ioreq);
 538             continue;
 539         }
 540
 541         if (use_aio) {
 542             /* run i/o in aio mode */
 543             ioreq_runio_qemu_aio(ioreq);
 544         } else {
 545             /* run i/o in sync mode */
 546             ioreq_runio_qemu_sync(ioreq);
 547         }
 548     }
 549     if (!use_aio)
 550         blk_send_response_all(blkdev);
 551
 552     if (blkdev->more_work && blkdev->requests_inflight < max_requests)
 553         qemu_bh_schedule(blkdev->bh);
 554 }
 555
 556 /* ------------------------------------------------------------- */
 557
 558 static void blk_bh(void *opaque)
 559 {
 560     struct XenBlkDev *blkdev = opaque;
 561     blk_handle_requests(blkdev);
 562 }
 563
 564 static void blk_alloc(struct XenDevice *xendev)
 565 {
 566     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 567
 568     LIST_INIT(&blkdev->inflight);
 569     LIST_INIT(&blkdev->finished);
 570     LIST_INIT(&blkdev->freelist);
 571     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 572     if (xen_mode != XEN_EMULATE)
 573         batch_maps = 1;
 574 }
 575
 576 static int blk_init(struct XenDevice *xendev)
 577 {
 578     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 579     int mode, qflags, have_barriers, info = 0;
 580     char *h;
 581
 582     /* read xenstore entries */
 583     if (blkdev->params == NULL) {
 584         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 585         h = strchr(blkdev->params, ':');
 586         if (h != NULL) {
 587             blkdev->fileproto = blkdev->params;
 588             blkdev->filename  = h+1;
 589             *h = 0;
 590         } else {
 591             blkdev->fileproto = "<unset>";
 592             blkdev->filename  = blkdev->params;
 593         }
 594     }
 595     if (blkdev->mode == NULL)
 596         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 597     if (blkdev->type == NULL)
 598         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 599     if (blkdev->dev == NULL)
 600         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 601     if (blkdev->devtype == NULL)
 602         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 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         return -1;
 610
 611     /* read-only ? */
 612     if (strcmp(blkdev->mode, "w") == 0) {
 613         mode   = O_RDWR;
 614         qflags = BDRV_O_RDWR;
 615     } else {
 616         mode   = O_RDONLY;
 617         qflags = BDRV_O_RDONLY;
 618         info  |= VDISK_READONLY;
 619     }
 620
 621     /* cdrom ? */
 622     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom"))
 623         info  |= VDISK_CDROM;
 624
 625     /* init qemu block driver */
 626     blkdev->index = (blkdev->xendev.dev - 202 * 256) / 16;
 627     blkdev->index = drive_get_index(IF_XEN, 0, blkdev->index);
 628     if (blkdev->index == -1) {
 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_open2(blkdev->bs, blkdev->filename, qflags,
 634                            bdrv_find_format(blkdev->fileproto)) != 0) {
 635                 bdrv_delete(blkdev->bs);
 636                 blkdev->bs = NULL;
 637             }
 638         }
 639         if (!blkdev->bs)
 640             return -1;
 641     } else {
 642         /* setup via qemu cmdline -> already setup for us */
 643         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 644         blkdev->bs = drives_table[blkdev->index].bdrv;
 645     }
 646     blkdev->file_blk  = BLOCK_SIZE;
 647     blkdev->file_size = bdrv_getlength(blkdev->bs);
 648     if (blkdev->file_size < 0) {
 649         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 650                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 651                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 652         blkdev->file_size = 0;
 653     }
 654     have_barriers = blkdev->bs->drv && blkdev->bs->drv->bdrv_flush ? 1 : 0;
 655
 656     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 657                   " size %" PRId64 " (%" PRId64 " MB)\n",
 658                   blkdev->type, blkdev->fileproto, blkdev->filename,
 659                   blkdev->file_size, blkdev->file_size >> 20);
 660
 661     /* fill info */
 662     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", have_barriers);
 663     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 664     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 665     xenstore_write_be_int(&blkdev->xendev, "sectors",
 666                           blkdev->file_size / blkdev->file_blk);
 667     return 0;
 668 }
 669
 670 static int blk_connect(struct XenDevice *xendev)
 671 {
 672     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 673
 674     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1)
 675         return -1;
 676     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 677                              &blkdev->xendev.remote_port) == -1)
 678         return -1;
 679
 680     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 681     if (blkdev->xendev.protocol) {
 682         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0)
 683             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 684         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0)
 685             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 686     }
 687
 688     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 689                                             blkdev->xendev.dom,
 690                                             blkdev->ring_ref,
 691                                             PROT_READ | PROT_WRITE);
 692     if (!blkdev->sring)
 693         return -1;
 694     blkdev->cnt_map++;
 695
 696     switch (blkdev->protocol) {
 697     case BLKIF_PROTOCOL_NATIVE:
 698     {
 699         blkif_sring_t *sring_native = blkdev->sring;
 700         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 701         break;
 702     }
 703     case BLKIF_PROTOCOL_X86_32:
 704     {
 705         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 706
 707         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 708         break;
 709     }
 710     case BLKIF_PROTOCOL_X86_64:
 711     {
 712         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 713
 714         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 715         break;
 716     }
 717     }
 718
 719     xen_be_bind_evtchn(&blkdev->xendev);
 720
 721     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 722                   "remote port %d, local port %d\n",
 723                   blkdev->xendev.protocol, blkdev->ring_ref,
 724                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 725     return 0;
 726 }
 727
 728 static void blk_disconnect(struct XenDevice *xendev)
 729 {
 730     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 731
 732     if (blkdev->bs) {
 733         if (blkdev->index == -1) {
 734             /* close/delete only if we created it ourself */
 735             bdrv_close(blkdev->bs);
 736             bdrv_delete(blkdev->bs);
 737         }
 738         blkdev->bs = NULL;
 739     }
 740     xen_be_unbind_evtchn(&blkdev->xendev);
 741
 742     if (blkdev->sring) {
 743         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 744         blkdev->cnt_map--;
 745         blkdev->sring = NULL;
 746     }
 747 }
 748
 749 static int blk_free(struct XenDevice *xendev)
 750 {
 751     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 752     struct ioreq *ioreq;
 753
 754     while (!LIST_EMPTY(&blkdev->freelist)) {
 755         ioreq = LIST_FIRST(&blkdev->freelist);
 756         LIST_REMOVE(ioreq, list);
 757         qemu_iovec_destroy(&ioreq->v);
 758         qemu_free(ioreq);
 759     }
 760
 761     qemu_free(blkdev->params);
 762     qemu_free(blkdev->mode);
 763     qemu_free(blkdev->type);
 764     qemu_free(blkdev->dev);
 765     qemu_free(blkdev->devtype);
 766     qemu_bh_delete(blkdev->bh);
 767     return 0;
 768 }
 769
 770 static void blk_event(struct XenDevice *xendev)
 771 {
 772     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 773
 774     qemu_bh_schedule(blkdev->bh);
 775 }
 776
 777 struct XenDevOps xen_blkdev_ops = {
 778     .size       = sizeof(struct XenBlkDev),
 779     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 780     .alloc      = blk_alloc,
 781     .init       = blk_init,
 782     .connect    = blk_connect,
 783     .disconnect = blk_disconnect,
 784     .event      = blk_event,
 785     .free       = blk_free,
 786 };