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
  19 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <stdarg.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <signal.h>
  25 #include <inttypes.h>
  26 #include <time.h>
  27 #include <fcntl.h>
  28 #include <errno.h>
  29 #include <sys/ioctl.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/mman.h>
  33 #include <sys/uio.h>
  34
  35 #include <xs.h>
  36 #include <xenctrl.h>
  37 #include <xen/io/xenbus.h>
  38
  39 #include "hw.h"
  40 #include "block_int.h"
  41 #include "qemu-char.h"
  42 #include "xen_blkif.h"
  43 #include "xen_backend.h"
  44 #include "blockdev.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     QLIST_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     QLIST_HEAD(inflight_head, ioreq) inflight;
 104     QLIST_HEAD(finished_head, ioreq) finished;
 105     QLIST_HEAD(freelist_head, ioreq) freelist;
 106     int                 requests_total;
 107     int                 requests_inflight;
 108     int                 requests_finished;
 109
 110     /* qemu block driver */
 111     DriveInfo           *dinfo;
 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 (QLIST_EMPTY(&blkdev->freelist)) {
 123         if (blkdev->requests_total >= max_requests) {
 124             goto out;
 125         }
 126         /* allocate new struct */
 127         ioreq = g_malloc0(sizeof(*ioreq));
 128         ioreq->blkdev = blkdev;
 129         blkdev->requests_total++;
 130         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 131     } else {
 132         /* get one from freelist */
 133         ioreq = QLIST_FIRST(&blkdev->freelist);
 134         QLIST_REMOVE(ioreq, list);
 135         qemu_iovec_reset(&ioreq->v);
 136     }
 137     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 138     blkdev->requests_inflight++;
 139
 140 out:
 141     return ioreq;
 142 }
 143
 144 static void ioreq_finish(struct ioreq *ioreq)
 145 {
 146     struct XenBlkDev *blkdev = ioreq->blkdev;
 147
 148     QLIST_REMOVE(ioreq, list);
 149     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 150     blkdev->requests_inflight--;
 151     blkdev->requests_finished++;
 152 }
 153
 154 static void ioreq_release(struct ioreq *ioreq)
 155 {
 156     struct XenBlkDev *blkdev = ioreq->blkdev;
 157
 158     QLIST_REMOVE(ioreq, list);
 159     memset(ioreq, 0, sizeof(*ioreq));
 160     ioreq->blkdev = blkdev;
 161     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 162     blkdev->requests_finished--;
 163 }
 164
 165 /*
 166  * translate request into iovec + start offset
 167  * do sanity checks along the way
 168  */
 169 static int ioreq_parse(struct ioreq *ioreq)
 170 {
 171     struct XenBlkDev *blkdev = ioreq->blkdev;
 172     uintptr_t mem;
 173     size_t len;
 174     int i;
 175
 176     xen_be_printf(&blkdev->xendev, 3,
 177                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 178                   ioreq->req.operation, ioreq->req.nr_segments,
 179                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 180     switch (ioreq->req.operation) {
 181     case BLKIF_OP_READ:
 182         ioreq->prot = PROT_WRITE; /* to memory */
 183         break;
 184     case BLKIF_OP_WRITE_BARRIER:
 185         if (!ioreq->req.nr_segments) {
 186             ioreq->presync = 1;
 187             return 0;
 188         }
 189         if (!syncwrite) {
 190             ioreq->presync = ioreq->postsync = 1;
 191         }
 192         /* fall through */
 193     case BLKIF_OP_WRITE:
 194         ioreq->prot = PROT_READ; /* from memory */
 195         if (syncwrite) {
 196             ioreq->postsync = 1;
 197         }
 198         break;
 199     default:
 200         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 201                       ioreq->req.operation);
 202         goto err;
 203     };
 204
 205     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 206         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 207         goto err;
 208     }
 209
 210     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 211     for (i = 0; i < ioreq->req.nr_segments; i++) {
 212         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 213             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 214             goto err;
 215         }
 216         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 217             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 218             goto err;
 219         }
 220         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 221             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 222             goto err;
 223         }
 224
 225         ioreq->domids[i] = blkdev->xendev.dom;
 226         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 227
 228         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 229         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 230         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 231     }
 232     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 233         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 234         goto err;
 235     }
 236     return 0;
 237
 238 err:
 239     ioreq->status = BLKIF_RSP_ERROR;
 240     return -1;
 241 }
 242
 243 static void ioreq_unmap(struct ioreq *ioreq)
 244 {
 245     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 246     int i;
 247
 248     if (ioreq->v.niov == 0) {
 249         return;
 250     }
 251     if (batch_maps) {
 252         if (!ioreq->pages) {
 253             return;
 254         }
 255         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0) {
 256             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 257                           strerror(errno));
 258         }
 259         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 260         ioreq->pages = NULL;
 261     } else {
 262         for (i = 0; i < ioreq->v.niov; i++) {
 263             if (!ioreq->page[i]) {
 264                 continue;
 265             }
 266             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0) {
 267                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 268                               strerror(errno));
 269             }
 270             ioreq->blkdev->cnt_map--;
 271             ioreq->page[i] = NULL;
 272         }
 273     }
 274 }
 275
 276 static int ioreq_map(struct ioreq *ioreq)
 277 {
 278     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 279     int i;
 280
 281     if (ioreq->v.niov == 0) {
 282         return 0;
 283     }
 284     if (batch_maps) {
 285         ioreq->pages = xc_gnttab_map_grant_refs
 286             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 287         if (ioreq->pages == NULL) {
 288             xen_be_printf(&ioreq->blkdev->xendev, 0,
 289                           "can't map %d grant refs (%s, %d maps)\n",
 290                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 291             return -1;
 292         }
 293         for (i = 0; i < ioreq->v.niov; i++) {
 294             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 295                 (uintptr_t)ioreq->v.iov[i].iov_base;
 296         }
 297         ioreq->blkdev->cnt_map += ioreq->v.niov;
 298     } else  {
 299         for (i = 0; i < ioreq->v.niov; i++) {
 300             ioreq->page[i] = xc_gnttab_map_grant_ref
 301                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 302             if (ioreq->page[i] == NULL) {
 303                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 304                               "can't map grant ref %d (%s, %d maps)\n",
 305                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 306                 ioreq_unmap(ioreq);
 307                 return -1;
 308             }
 309             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 310             ioreq->blkdev->cnt_map++;
 311         }
 312     }
 313     return 0;
 314 }
 315
 316 static int ioreq_runio_qemu_sync(struct ioreq *ioreq)
 317 {
 318     struct XenBlkDev *blkdev = ioreq->blkdev;
 319     int i, rc;
 320     off_t pos;
 321
 322     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 323         goto err_no_map;
 324     }
 325     if (ioreq->presync) {
 326         bdrv_flush(blkdev->bs);
 327     }
 328
 329     switch (ioreq->req.operation) {
 330     case BLKIF_OP_READ:
 331         pos = ioreq->start;
 332         for (i = 0; i < ioreq->v.niov; i++) {
 333             rc = bdrv_read(blkdev->bs, pos / BLOCK_SIZE,
 334                            ioreq->v.iov[i].iov_base,
 335                            ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 336             if (rc != 0) {
 337                 xen_be_printf(&blkdev->xendev, 0, "rd I/O error (%p, len %zd)\n",
 338                               ioreq->v.iov[i].iov_base,
 339                               ioreq->v.iov[i].iov_len);
 340                 goto err;
 341             }
 342             pos += ioreq->v.iov[i].iov_len;
 343         }
 344         break;
 345     case BLKIF_OP_WRITE:
 346     case BLKIF_OP_WRITE_BARRIER:
 347         if (!ioreq->req.nr_segments) {
 348             break;
 349         }
 350         pos = ioreq->start;
 351         for (i = 0; i < ioreq->v.niov; i++) {
 352             rc = bdrv_write(blkdev->bs, pos / BLOCK_SIZE,
 353                             ioreq->v.iov[i].iov_base,
 354                             ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 355             if (rc != 0) {
 356                 xen_be_printf(&blkdev->xendev, 0, "wr I/O error (%p, len %zd)\n",
 357                               ioreq->v.iov[i].iov_base,
 358                               ioreq->v.iov[i].iov_len);
 359                 goto err;
 360             }
 361             pos += ioreq->v.iov[i].iov_len;
 362         }
 363         break;
 364     default:
 365         /* unknown operation (shouldn't happen -- parse catches this) */
 366         goto err;
 367     }
 368
 369     if (ioreq->postsync) {
 370         bdrv_flush(blkdev->bs);
 371     }
 372     ioreq->status = BLKIF_RSP_OKAY;
 373
 374     ioreq_unmap(ioreq);
 375     ioreq_finish(ioreq);
 376     return 0;
 377
 378 err:
 379     ioreq_unmap(ioreq);
 380 err_no_map:
 381     ioreq_finish(ioreq);
 382     ioreq->status = BLKIF_RSP_ERROR;
 383     return -1;
 384 }
 385
 386 static void qemu_aio_complete(void *opaque, int ret)
 387 {
 388     struct ioreq *ioreq = opaque;
 389
 390     if (ret != 0) {
 391         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 392                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 393         ioreq->aio_errors++;
 394     }
 395
 396     ioreq->aio_inflight--;
 397     if (ioreq->aio_inflight > 0) {
 398         return;
 399     }
 400
 401     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 402     ioreq_unmap(ioreq);
 403     ioreq_finish(ioreq);
 404     qemu_bh_schedule(ioreq->blkdev->bh);
 405 }
 406
 407 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 408 {
 409     struct XenBlkDev *blkdev = ioreq->blkdev;
 410
 411     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 412         goto err_no_map;
 413     }
 414
 415     ioreq->aio_inflight++;
 416     if (ioreq->presync) {
 417         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 418     }
 419
 420     switch (ioreq->req.operation) {
 421     case BLKIF_OP_READ:
 422         ioreq->aio_inflight++;
 423         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 424                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 425                        qemu_aio_complete, ioreq);
 426         break;
 427     case BLKIF_OP_WRITE:
 428     case BLKIF_OP_WRITE_BARRIER:
 429         if (!ioreq->req.nr_segments) {
 430             break;
 431         }
 432         ioreq->aio_inflight++;
 433         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 434                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 435                         qemu_aio_complete, ioreq);
 436         break;
 437     default:
 438         /* unknown operation (shouldn't happen -- parse catches this) */
 439         goto err;
 440     }
 441
 442     if (ioreq->postsync) {
 443         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 444     }
 445     qemu_aio_complete(ioreq, 0);
 446
 447     return 0;
 448
 449 err:
 450     ioreq_unmap(ioreq);
 451 err_no_map:
 452     ioreq_finish(ioreq);
 453     ioreq->status = BLKIF_RSP_ERROR;
 454     return -1;
 455 }
 456
 457 static int blk_send_response_one(struct ioreq *ioreq)
 458 {
 459     struct XenBlkDev  *blkdev = ioreq->blkdev;
 460     int               send_notify   = 0;
 461     int               have_requests = 0;
 462     blkif_response_t  resp;
 463     void              *dst;
 464
 465     resp.id        = ioreq->req.id;
 466     resp.operation = ioreq->req.operation;
 467     resp.status    = ioreq->status;
 468
 469     /* Place on the response ring for the relevant domain. */
 470     switch (blkdev->protocol) {
 471     case BLKIF_PROTOCOL_NATIVE:
 472         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 473         break;
 474     case BLKIF_PROTOCOL_X86_32:
 475         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 476                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 477         break;
 478     case BLKIF_PROTOCOL_X86_64:
 479         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 480                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 481         break;
 482     default:
 483         dst = NULL;
 484     }
 485     memcpy(dst, &resp, sizeof(resp));
 486     blkdev->rings.common.rsp_prod_pvt++;
 487
 488     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 489     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 490         /*
 491          * Tail check for pending requests. Allows frontend to avoid
 492          * notifications if requests are already in flight (lower
 493          * overheads and promotes batching).
 494          */
 495         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 496     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 497         have_requests = 1;
 498     }
 499
 500     if (have_requests) {
 501         blkdev->more_work++;
 502     }
 503     return send_notify;
 504 }
 505
 506 /* walk finished list, send outstanding responses, free requests */
 507 static void blk_send_response_all(struct XenBlkDev *blkdev)
 508 {
 509     struct ioreq *ioreq;
 510     int send_notify = 0;
 511
 512     while (!QLIST_EMPTY(&blkdev->finished)) {
 513         ioreq = QLIST_FIRST(&blkdev->finished);
 514         send_notify += blk_send_response_one(ioreq);
 515         ioreq_release(ioreq);
 516     }
 517     if (send_notify) {
 518         xen_be_send_notify(&blkdev->xendev);
 519     }
 520 }
 521
 522 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 523 {
 524     switch (blkdev->protocol) {
 525     case BLKIF_PROTOCOL_NATIVE:
 526         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 527                sizeof(ioreq->req));
 528         break;
 529     case BLKIF_PROTOCOL_X86_32:
 530         blkif_get_x86_32_req(&ioreq->req,
 531                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 532         break;
 533     case BLKIF_PROTOCOL_X86_64:
 534         blkif_get_x86_64_req(&ioreq->req,
 535                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 536         break;
 537     }
 538     return 0;
 539 }
 540
 541 static void blk_handle_requests(struct XenBlkDev *blkdev)
 542 {
 543     RING_IDX rc, rp;
 544     struct ioreq *ioreq;
 545
 546     blkdev->more_work = 0;
 547
 548     rc = blkdev->rings.common.req_cons;
 549     rp = blkdev->rings.common.sring->req_prod;
 550     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 551
 552     if (use_aio) {
 553         blk_send_response_all(blkdev);
 554     }
 555     while (rc != rp) {
 556         /* pull request from ring */
 557         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 558             break;
 559         }
 560         ioreq = ioreq_start(blkdev);
 561         if (ioreq == NULL) {
 562             blkdev->more_work++;
 563             break;
 564         }
 565         blk_get_request(blkdev, ioreq, rc);
 566         blkdev->rings.common.req_cons = ++rc;
 567
 568         /* parse them */
 569         if (ioreq_parse(ioreq) != 0) {
 570             if (blk_send_response_one(ioreq)) {
 571                 xen_be_send_notify(&blkdev->xendev);
 572             }
 573             ioreq_release(ioreq);
 574             continue;
 575         }
 576
 577         if (use_aio) {
 578             /* run i/o in aio mode */
 579             ioreq_runio_qemu_aio(ioreq);
 580         } else {
 581             /* run i/o in sync mode */
 582             ioreq_runio_qemu_sync(ioreq);
 583         }
 584     }
 585     if (!use_aio) {
 586         blk_send_response_all(blkdev);
 587     }
 588
 589     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 590         qemu_bh_schedule(blkdev->bh);
 591     }
 592 }
 593
 594 /* ------------------------------------------------------------- */
 595
 596 static void blk_bh(void *opaque)
 597 {
 598     struct XenBlkDev *blkdev = opaque;
 599     blk_handle_requests(blkdev);
 600 }
 601
 602 static void blk_alloc(struct XenDevice *xendev)
 603 {
 604     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 605
 606     QLIST_INIT(&blkdev->inflight);
 607     QLIST_INIT(&blkdev->finished);
 608     QLIST_INIT(&blkdev->freelist);
 609     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 610     if (xen_mode != XEN_EMULATE) {
 611         batch_maps = 1;
 612     }
 613 }
 614
 615 static int blk_init(struct XenDevice *xendev)
 616 {
 617     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 618     int index, qflags, have_barriers, info = 0;
 619
 620     /* read xenstore entries */
 621     if (blkdev->params == NULL) {
 622         char *h = NULL;
 623         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 624         if (blkdev->params != NULL) {
 625             h = strchr(blkdev->params, ':');
 626         }
 627         if (h != NULL) {
 628             blkdev->fileproto = blkdev->params;
 629             blkdev->filename  = h+1;
 630             *h = 0;
 631         } else {
 632             blkdev->fileproto = "<unset>";
 633             blkdev->filename  = blkdev->params;
 634         }
 635     }
 636     if (!strcmp("aio", blkdev->fileproto)) {
 637         blkdev->fileproto = "raw";
 638     }
 639     if (blkdev->mode == NULL) {
 640         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 641     }
 642     if (blkdev->type == NULL) {
 643         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 644     }
 645     if (blkdev->dev == NULL) {
 646         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 647     }
 648     if (blkdev->devtype == NULL) {
 649         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 650     }
 651
 652     /* do we have all we need? */
 653     if (blkdev->params == NULL ||
 654         blkdev->mode == NULL   ||
 655         blkdev->type == NULL   ||
 656         blkdev->dev == NULL) {
 657         goto out_error;
 658     }
 659
 660     /* read-only ? */
 661     if (strcmp(blkdev->mode, "w") == 0) {
 662         qflags = BDRV_O_RDWR;
 663     } else {
 664         qflags = 0;
 665         info  |= VDISK_READONLY;
 666     }
 667
 668     /* cdrom ? */
 669     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 670         info  |= VDISK_CDROM;
 671     }
 672
 673     /* init qemu block driver */
 674     index = (blkdev->xendev.dev - 202 * 256) / 16;
 675     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 676     if (!blkdev->dinfo) {
 677         /* setup via xenbus -> create new block driver instance */
 678         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 679         blkdev->bs = bdrv_new(blkdev->dev);
 680         if (blkdev->bs) {
 681             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 682                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 683                 bdrv_delete(blkdev->bs);
 684                 blkdev->bs = NULL;
 685             }
 686         }
 687         if (!blkdev->bs) {
 688             goto out_error;
 689         }
 690     } else {
 691         /* setup via qemu cmdline -> already setup for us */
 692         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 693         blkdev->bs = blkdev->dinfo->bdrv;
 694     }
 695     blkdev->file_blk  = BLOCK_SIZE;
 696     blkdev->file_size = bdrv_getlength(blkdev->bs);
 697     if (blkdev->file_size < 0) {
 698         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 699                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 700                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 701         blkdev->file_size = 0;
 702     }
 703     have_barriers = blkdev->bs->drv && blkdev->bs->drv->bdrv_flush ? 1 : 0;
 704
 705     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 706                   " size %" PRId64 " (%" PRId64 " MB)\n",
 707                   blkdev->type, blkdev->fileproto, blkdev->filename,
 708                   blkdev->file_size, blkdev->file_size >> 20);
 709
 710     /* fill info */
 711     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", have_barriers);
 712     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 713     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 714     xenstore_write_be_int(&blkdev->xendev, "sectors",
 715                           blkdev->file_size / blkdev->file_blk);
 716     return 0;
 717
 718 out_error:
 719     g_free(blkdev->params);
 720     blkdev->params = NULL;
 721     g_free(blkdev->mode);
 722     blkdev->mode = NULL;
 723     g_free(blkdev->type);
 724     blkdev->type = NULL;
 725     g_free(blkdev->dev);
 726     blkdev->dev = NULL;
 727     g_free(blkdev->devtype);
 728     blkdev->devtype = NULL;
 729     return -1;
 730 }
 731
 732 static int blk_connect(struct XenDevice *xendev)
 733 {
 734     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 735
 736     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 737         return -1;
 738     }
 739     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 740                              &blkdev->xendev.remote_port) == -1) {
 741         return -1;
 742     }
 743
 744     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 745     if (blkdev->xendev.protocol) {
 746         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 747             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 748         }
 749         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 750             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 751         }
 752     }
 753
 754     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 755                                             blkdev->xendev.dom,
 756                                             blkdev->ring_ref,
 757                                             PROT_READ | PROT_WRITE);
 758     if (!blkdev->sring) {
 759         return -1;
 760     }
 761     blkdev->cnt_map++;
 762
 763     switch (blkdev->protocol) {
 764     case BLKIF_PROTOCOL_NATIVE:
 765     {
 766         blkif_sring_t *sring_native = blkdev->sring;
 767         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 768         break;
 769     }
 770     case BLKIF_PROTOCOL_X86_32:
 771     {
 772         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 773
 774         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 775         break;
 776     }
 777     case BLKIF_PROTOCOL_X86_64:
 778     {
 779         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 780
 781         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 782         break;
 783     }
 784     }
 785
 786     xen_be_bind_evtchn(&blkdev->xendev);
 787
 788     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 789                   "remote port %d, local port %d\n",
 790                   blkdev->xendev.protocol, blkdev->ring_ref,
 791                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 792     return 0;
 793 }
 794
 795 static void blk_disconnect(struct XenDevice *xendev)
 796 {
 797     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 798
 799     if (blkdev->bs) {
 800         if (!blkdev->dinfo) {
 801             /* close/delete only if we created it ourself */
 802             bdrv_close(blkdev->bs);
 803             bdrv_delete(blkdev->bs);
 804         }
 805         blkdev->bs = NULL;
 806     }
 807     xen_be_unbind_evtchn(&blkdev->xendev);
 808
 809     if (blkdev->sring) {
 810         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 811         blkdev->cnt_map--;
 812         blkdev->sring = NULL;
 813     }
 814 }
 815
 816 static int blk_free(struct XenDevice *xendev)
 817 {
 818     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 819     struct ioreq *ioreq;
 820
 821     while (!QLIST_EMPTY(&blkdev->freelist)) {
 822         ioreq = QLIST_FIRST(&blkdev->freelist);
 823         QLIST_REMOVE(ioreq, list);
 824         qemu_iovec_destroy(&ioreq->v);
 825         g_free(ioreq);
 826     }
 827
 828     g_free(blkdev->params);
 829     g_free(blkdev->mode);
 830     g_free(blkdev->type);
 831     g_free(blkdev->dev);
 832     g_free(blkdev->devtype);
 833     qemu_bh_delete(blkdev->bh);
 834     return 0;
 835 }
 836
 837 static void blk_event(struct XenDevice *xendev)
 838 {
 839     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 840
 841     qemu_bh_schedule(blkdev->bh);
 842 }
 843
 844 struct XenDevOps xen_blkdev_ops = {
 845     .size       = sizeof(struct XenBlkDev),
 846     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 847     .alloc      = blk_alloc,
 848     .init       = blk_init,
 849     .connect    = blk_connect,
 850     .disconnect = blk_disconnect,
 851     .event      = blk_event,
 852     .free       = blk_free,
 853 };