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 <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 syncwrite    = 0;
  52 static int batch_maps   = 0;
  53
  54 static int max_requests = 32;
  55
  56 /* ------------------------------------------------------------- */
  57
  58 #define BLOCK_SIZE  512
  59 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  60
  61 struct ioreq {
  62     blkif_request_t     req;
  63     int16_t             status;
  64
  65     /* parsed request */
  66     off_t               start;
  67     QEMUIOVector        v;
  68     int                 presync;
  69     int                 postsync;
  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)
 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     blkdev->requests_finished--;
 166 }
 167
 168 /*
 169  * translate request into iovec + start offset
 170  * do sanity checks along the way
 171  */
 172 static int ioreq_parse(struct ioreq *ioreq)
 173 {
 174     struct XenBlkDev *blkdev = ioreq->blkdev;
 175     uintptr_t mem;
 176     size_t len;
 177     int i;
 178
 179     xen_be_printf(&blkdev->xendev, 3,
 180                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 181                   ioreq->req.operation, ioreq->req.nr_segments,
 182                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 183     switch (ioreq->req.operation) {
 184     case BLKIF_OP_READ:
 185         ioreq->prot = PROT_WRITE; /* to memory */
 186         break;
 187     case BLKIF_OP_WRITE_BARRIER:
 188         if (!ioreq->req.nr_segments) {
 189             ioreq->presync = 1;
 190             return 0;
 191         }
 192         if (!syncwrite) {
 193             ioreq->presync = ioreq->postsync = 1;
 194         }
 195         /* fall through */
 196     case BLKIF_OP_WRITE:
 197         ioreq->prot = PROT_READ; /* from memory */
 198         if (syncwrite) {
 199             ioreq->postsync = 1;
 200         }
 201         break;
 202     default:
 203         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 204                       ioreq->req.operation);
 205         goto err;
 206     };
 207
 208     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 209         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 210         goto err;
 211     }
 212
 213     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 214     for (i = 0; i < ioreq->req.nr_segments; i++) {
 215         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 216             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 217             goto err;
 218         }
 219         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 220             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 221             goto err;
 222         }
 223         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 224             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 225             goto err;
 226         }
 227
 228         ioreq->domids[i] = blkdev->xendev.dom;
 229         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 230
 231         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 232         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 233         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 234     }
 235     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 236         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 237         goto err;
 238     }
 239     return 0;
 240
 241 err:
 242     ioreq->status = BLKIF_RSP_ERROR;
 243     return -1;
 244 }
 245
 246 static void ioreq_unmap(struct ioreq *ioreq)
 247 {
 248     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 249     int i;
 250
 251     if (ioreq->v.niov == 0) {
 252         return;
 253     }
 254     if (batch_maps) {
 255         if (!ioreq->pages) {
 256             return;
 257         }
 258         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0) {
 259             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 260                           strerror(errno));
 261         }
 262         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 263         ioreq->pages = NULL;
 264     } else {
 265         for (i = 0; i < ioreq->v.niov; i++) {
 266             if (!ioreq->page[i]) {
 267                 continue;
 268             }
 269             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0) {
 270                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 271                               strerror(errno));
 272             }
 273             ioreq->blkdev->cnt_map--;
 274             ioreq->page[i] = NULL;
 275         }
 276     }
 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) {
 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     return 0;
 317 }
 318
 319 static void qemu_aio_complete(void *opaque, int ret)
 320 {
 321     struct ioreq *ioreq = opaque;
 322
 323     if (ret != 0) {
 324         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 325                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 326         ioreq->aio_errors++;
 327     }
 328
 329     ioreq->aio_inflight--;
 330     if (ioreq->aio_inflight > 0) {
 331         return;
 332     }
 333     if (ioreq->postsync) {
 334         bdrv_flush(ioreq->blkdev->bs);
 335     }
 336
 337     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 338     ioreq_unmap(ioreq);
 339     ioreq_finish(ioreq);
 340     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 341     qemu_bh_schedule(ioreq->blkdev->bh);
 342 }
 343
 344 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 345 {
 346     struct XenBlkDev *blkdev = ioreq->blkdev;
 347
 348     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 349         goto err_no_map;
 350     }
 351
 352     ioreq->aio_inflight++;
 353     if (ioreq->presync) {
 354         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 355     }
 356
 357     switch (ioreq->req.operation) {
 358     case BLKIF_OP_READ:
 359         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 360         ioreq->aio_inflight++;
 361         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 362                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 363                        qemu_aio_complete, ioreq);
 364         break;
 365     case BLKIF_OP_WRITE:
 366     case BLKIF_OP_WRITE_BARRIER:
 367         if (!ioreq->req.nr_segments) {
 368             break;
 369         }
 370
 371         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 372         ioreq->aio_inflight++;
 373         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 374                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 375                         qemu_aio_complete, ioreq);
 376         break;
 377     default:
 378         /* unknown operation (shouldn't happen -- parse catches this) */
 379         goto err;
 380     }
 381
 382     qemu_aio_complete(ioreq, 0);
 383
 384     return 0;
 385
 386 err:
 387     ioreq_unmap(ioreq);
 388 err_no_map:
 389     ioreq_finish(ioreq);
 390     ioreq->status = BLKIF_RSP_ERROR;
 391     return -1;
 392 }
 393
 394 static int blk_send_response_one(struct ioreq *ioreq)
 395 {
 396     struct XenBlkDev  *blkdev = ioreq->blkdev;
 397     int               send_notify   = 0;
 398     int               have_requests = 0;
 399     blkif_response_t  resp;
 400     void              *dst;
 401
 402     resp.id        = ioreq->req.id;
 403     resp.operation = ioreq->req.operation;
 404     resp.status    = ioreq->status;
 405
 406     /* Place on the response ring for the relevant domain. */
 407     switch (blkdev->protocol) {
 408     case BLKIF_PROTOCOL_NATIVE:
 409         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 410         break;
 411     case BLKIF_PROTOCOL_X86_32:
 412         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 413                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 414         break;
 415     case BLKIF_PROTOCOL_X86_64:
 416         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 417                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 418         break;
 419     default:
 420         dst = NULL;
 421     }
 422     memcpy(dst, &resp, sizeof(resp));
 423     blkdev->rings.common.rsp_prod_pvt++;
 424
 425     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 426     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 427         /*
 428          * Tail check for pending requests. Allows frontend to avoid
 429          * notifications if requests are already in flight (lower
 430          * overheads and promotes batching).
 431          */
 432         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 433     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 434         have_requests = 1;
 435     }
 436
 437     if (have_requests) {
 438         blkdev->more_work++;
 439     }
 440     return send_notify;
 441 }
 442
 443 /* walk finished list, send outstanding responses, free requests */
 444 static void blk_send_response_all(struct XenBlkDev *blkdev)
 445 {
 446     struct ioreq *ioreq;
 447     int send_notify = 0;
 448
 449     while (!QLIST_EMPTY(&blkdev->finished)) {
 450         ioreq = QLIST_FIRST(&blkdev->finished);
 451         send_notify += blk_send_response_one(ioreq);
 452         ioreq_release(ioreq);
 453     }
 454     if (send_notify) {
 455         xen_be_send_notify(&blkdev->xendev);
 456     }
 457 }
 458
 459 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 460 {
 461     switch (blkdev->protocol) {
 462     case BLKIF_PROTOCOL_NATIVE:
 463         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 464                sizeof(ioreq->req));
 465         break;
 466     case BLKIF_PROTOCOL_X86_32:
 467         blkif_get_x86_32_req(&ioreq->req,
 468                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 469         break;
 470     case BLKIF_PROTOCOL_X86_64:
 471         blkif_get_x86_64_req(&ioreq->req,
 472                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 473         break;
 474     }
 475     return 0;
 476 }
 477
 478 static void blk_handle_requests(struct XenBlkDev *blkdev)
 479 {
 480     RING_IDX rc, rp;
 481     struct ioreq *ioreq;
 482
 483     blkdev->more_work = 0;
 484
 485     rc = blkdev->rings.common.req_cons;
 486     rp = blkdev->rings.common.sring->req_prod;
 487     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 488
 489     blk_send_response_all(blkdev);
 490     while (rc != rp) {
 491         /* pull request from ring */
 492         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 493             break;
 494         }
 495         ioreq = ioreq_start(blkdev);
 496         if (ioreq == NULL) {
 497             blkdev->more_work++;
 498             break;
 499         }
 500         blk_get_request(blkdev, ioreq, rc);
 501         blkdev->rings.common.req_cons = ++rc;
 502
 503         /* parse them */
 504         if (ioreq_parse(ioreq) != 0) {
 505             if (blk_send_response_one(ioreq)) {
 506                 xen_be_send_notify(&blkdev->xendev);
 507             }
 508             ioreq_release(ioreq);
 509             continue;
 510         }
 511
 512         ioreq_runio_qemu_aio(ioreq);
 513     }
 514
 515     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 516         qemu_bh_schedule(blkdev->bh);
 517     }
 518 }
 519
 520 /* ------------------------------------------------------------- */
 521
 522 static void blk_bh(void *opaque)
 523 {
 524     struct XenBlkDev *blkdev = opaque;
 525     blk_handle_requests(blkdev);
 526 }
 527
 528 static void blk_alloc(struct XenDevice *xendev)
 529 {
 530     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 531
 532     QLIST_INIT(&blkdev->inflight);
 533     QLIST_INIT(&blkdev->finished);
 534     QLIST_INIT(&blkdev->freelist);
 535     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 536     if (xen_mode != XEN_EMULATE) {
 537         batch_maps = 1;
 538     }
 539 }
 540
 541 static int blk_init(struct XenDevice *xendev)
 542 {
 543     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 544     int index, qflags, info = 0;
 545
 546     /* read xenstore entries */
 547     if (blkdev->params == NULL) {
 548         char *h = NULL;
 549         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 550         if (blkdev->params != NULL) {
 551             h = strchr(blkdev->params, ':');
 552         }
 553         if (h != NULL) {
 554             blkdev->fileproto = blkdev->params;
 555             blkdev->filename  = h+1;
 556             *h = 0;
 557         } else {
 558             blkdev->fileproto = "<unset>";
 559             blkdev->filename  = blkdev->params;
 560         }
 561     }
 562     if (!strcmp("aio", blkdev->fileproto)) {
 563         blkdev->fileproto = "raw";
 564     }
 565     if (blkdev->mode == NULL) {
 566         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 567     }
 568     if (blkdev->type == NULL) {
 569         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 570     }
 571     if (blkdev->dev == NULL) {
 572         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 573     }
 574     if (blkdev->devtype == NULL) {
 575         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 576     }
 577
 578     /* do we have all we need? */
 579     if (blkdev->params == NULL ||
 580         blkdev->mode == NULL   ||
 581         blkdev->type == NULL   ||
 582         blkdev->dev == NULL) {
 583         goto out_error;
 584     }
 585
 586     /* read-only ? */
 587     qflags = BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO;
 588     if (strcmp(blkdev->mode, "w") == 0) {
 589         qflags |= BDRV_O_RDWR;
 590     } else {
 591         info  |= VDISK_READONLY;
 592     }
 593
 594     /* cdrom ? */
 595     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 596         info  |= VDISK_CDROM;
 597     }
 598
 599     /* init qemu block driver */
 600     index = (blkdev->xendev.dev - 202 * 256) / 16;
 601     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 602     if (!blkdev->dinfo) {
 603         /* setup via xenbus -> create new block driver instance */
 604         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 605         blkdev->bs = bdrv_new(blkdev->dev);
 606         if (blkdev->bs) {
 607             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 608                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 609                 bdrv_delete(blkdev->bs);
 610                 blkdev->bs = NULL;
 611             }
 612         }
 613         if (!blkdev->bs) {
 614             goto out_error;
 615         }
 616     } else {
 617         /* setup via qemu cmdline -> already setup for us */
 618         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 619         blkdev->bs = blkdev->dinfo->bdrv;
 620     }
 621     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 622     blkdev->file_blk  = BLOCK_SIZE;
 623     blkdev->file_size = bdrv_getlength(blkdev->bs);
 624     if (blkdev->file_size < 0) {
 625         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 626                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 627                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 628         blkdev->file_size = 0;
 629     }
 630
 631     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 632                   " size %" PRId64 " (%" PRId64 " MB)\n",
 633                   blkdev->type, blkdev->fileproto, blkdev->filename,
 634                   blkdev->file_size, blkdev->file_size >> 20);
 635
 636     /* fill info */
 637     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", 1);
 638     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 639     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 640     xenstore_write_be_int(&blkdev->xendev, "sectors",
 641                           blkdev->file_size / blkdev->file_blk);
 642     return 0;
 643
 644 out_error:
 645     g_free(blkdev->params);
 646     blkdev->params = NULL;
 647     g_free(blkdev->mode);
 648     blkdev->mode = NULL;
 649     g_free(blkdev->type);
 650     blkdev->type = NULL;
 651     g_free(blkdev->dev);
 652     blkdev->dev = NULL;
 653     g_free(blkdev->devtype);
 654     blkdev->devtype = NULL;
 655     return -1;
 656 }
 657
 658 static int blk_connect(struct XenDevice *xendev)
 659 {
 660     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 661
 662     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 663         return -1;
 664     }
 665     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 666                              &blkdev->xendev.remote_port) == -1) {
 667         return -1;
 668     }
 669
 670     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 671     if (blkdev->xendev.protocol) {
 672         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 673             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 674         }
 675         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 676             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 677         }
 678     }
 679
 680     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 681                                             blkdev->xendev.dom,
 682                                             blkdev->ring_ref,
 683                                             PROT_READ | PROT_WRITE);
 684     if (!blkdev->sring) {
 685         return -1;
 686     }
 687     blkdev->cnt_map++;
 688
 689     switch (blkdev->protocol) {
 690     case BLKIF_PROTOCOL_NATIVE:
 691     {
 692         blkif_sring_t *sring_native = blkdev->sring;
 693         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 694         break;
 695     }
 696     case BLKIF_PROTOCOL_X86_32:
 697     {
 698         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 699
 700         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 701         break;
 702     }
 703     case BLKIF_PROTOCOL_X86_64:
 704     {
 705         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 706
 707         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 708         break;
 709     }
 710     }
 711
 712     xen_be_bind_evtchn(&blkdev->xendev);
 713
 714     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 715                   "remote port %d, local port %d\n",
 716                   blkdev->xendev.protocol, blkdev->ring_ref,
 717                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 718     return 0;
 719 }
 720
 721 static void blk_disconnect(struct XenDevice *xendev)
 722 {
 723     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 724
 725     if (blkdev->bs) {
 726         if (!blkdev->dinfo) {
 727             /* close/delete only if we created it ourself */
 728             bdrv_close(blkdev->bs);
 729             bdrv_detach_dev(blkdev->bs, blkdev);
 730             bdrv_delete(blkdev->bs);
 731         }
 732         blkdev->bs = NULL;
 733     }
 734     xen_be_unbind_evtchn(&blkdev->xendev);
 735
 736     if (blkdev->sring) {
 737         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 738         blkdev->cnt_map--;
 739         blkdev->sring = NULL;
 740     }
 741 }
 742
 743 static int blk_free(struct XenDevice *xendev)
 744 {
 745     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 746     struct ioreq *ioreq;
 747
 748     if (blkdev->bs || blkdev->sring) {
 749         blk_disconnect(xendev);
 750     }
 751
 752     while (!QLIST_EMPTY(&blkdev->freelist)) {
 753         ioreq = QLIST_FIRST(&blkdev->freelist);
 754         QLIST_REMOVE(ioreq, list);
 755         qemu_iovec_destroy(&ioreq->v);
 756         g_free(ioreq);
 757     }
 758
 759     g_free(blkdev->params);
 760     g_free(blkdev->mode);
 761     g_free(blkdev->type);
 762     g_free(blkdev->dev);
 763     g_free(blkdev->devtype);
 764     qemu_bh_delete(blkdev->bh);
 765     return 0;
 766 }
 767
 768 static void blk_event(struct XenDevice *xendev)
 769 {
 770     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 771
 772     qemu_bh_schedule(blkdev->bh);
 773 }
 774
 775 struct XenDevOps xen_blkdev_ops = {
 776     .size       = sizeof(struct XenBlkDev),
 777     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 778     .alloc      = blk_alloc,
 779     .init       = blk_init,
 780     .initialise    = blk_connect,
 781     .disconnect = blk_disconnect,
 782     .event      = blk_event,
 783     .free       = blk_free,
 784 };