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
  53 /* ------------------------------------------------------------- */
  54
  55 #define BLOCK_SIZE  512
  56 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  57
  58 struct ioreq {
  59     blkif_request_t     req;
  60     int16_t             status;
  61
  62     /* parsed request */
  63     off_t               start;
  64     QEMUIOVector        v;
  65     int                 presync;
  66     int                 postsync;
  67
  68     /* grant mapping */
  69     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  70     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  71     int                 prot;
  72     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  73     void                *pages;
  74
  75     /* aio status */
  76     int                 aio_inflight;
  77     int                 aio_errors;
  78
  79     struct XenBlkDev    *blkdev;
  80     QLIST_ENTRY(ioreq)   list;
  81     BlockAcctCookie     acct;
  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 void qemu_aio_complete(void *opaque, int ret)
 317 {
 318     struct ioreq *ioreq = opaque;
 319
 320     if (ret != 0) {
 321         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 322                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 323         ioreq->aio_errors++;
 324     }
 325
 326     ioreq->aio_inflight--;
 327     if (ioreq->aio_inflight > 0) {
 328         return;
 329     }
 330
 331     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 332     ioreq_unmap(ioreq);
 333     ioreq_finish(ioreq);
 334     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 335     qemu_bh_schedule(ioreq->blkdev->bh);
 336 }
 337
 338 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 339 {
 340     struct XenBlkDev *blkdev = ioreq->blkdev;
 341
 342     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 343         goto err_no_map;
 344     }
 345
 346     ioreq->aio_inflight++;
 347     if (ioreq->presync) {
 348         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 349     }
 350
 351     switch (ioreq->req.operation) {
 352     case BLKIF_OP_READ:
 353         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 354         ioreq->aio_inflight++;
 355         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 356                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 357                        qemu_aio_complete, ioreq);
 358         break;
 359     case BLKIF_OP_WRITE:
 360     case BLKIF_OP_WRITE_BARRIER:
 361         if (!ioreq->req.nr_segments) {
 362             break;
 363         }
 364
 365         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 366         ioreq->aio_inflight++;
 367         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 368                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 369                         qemu_aio_complete, ioreq);
 370         break;
 371     default:
 372         /* unknown operation (shouldn't happen -- parse catches this) */
 373         goto err;
 374     }
 375
 376     if (ioreq->postsync) {
 377         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 378     }
 379     qemu_aio_complete(ioreq, 0);
 380
 381     return 0;
 382
 383 err:
 384     ioreq_unmap(ioreq);
 385 err_no_map:
 386     ioreq_finish(ioreq);
 387     ioreq->status = BLKIF_RSP_ERROR;
 388     return -1;
 389 }
 390
 391 static int blk_send_response_one(struct ioreq *ioreq)
 392 {
 393     struct XenBlkDev  *blkdev = ioreq->blkdev;
 394     int               send_notify   = 0;
 395     int               have_requests = 0;
 396     blkif_response_t  resp;
 397     void              *dst;
 398
 399     resp.id        = ioreq->req.id;
 400     resp.operation = ioreq->req.operation;
 401     resp.status    = ioreq->status;
 402
 403     /* Place on the response ring for the relevant domain. */
 404     switch (blkdev->protocol) {
 405     case BLKIF_PROTOCOL_NATIVE:
 406         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 407         break;
 408     case BLKIF_PROTOCOL_X86_32:
 409         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 410                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 411         break;
 412     case BLKIF_PROTOCOL_X86_64:
 413         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 414                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 415         break;
 416     default:
 417         dst = NULL;
 418     }
 419     memcpy(dst, &resp, sizeof(resp));
 420     blkdev->rings.common.rsp_prod_pvt++;
 421
 422     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 423     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 424         /*
 425          * Tail check for pending requests. Allows frontend to avoid
 426          * notifications if requests are already in flight (lower
 427          * overheads and promotes batching).
 428          */
 429         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 430     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 431         have_requests = 1;
 432     }
 433
 434     if (have_requests) {
 435         blkdev->more_work++;
 436     }
 437     return send_notify;
 438 }
 439
 440 /* walk finished list, send outstanding responses, free requests */
 441 static void blk_send_response_all(struct XenBlkDev *blkdev)
 442 {
 443     struct ioreq *ioreq;
 444     int send_notify = 0;
 445
 446     while (!QLIST_EMPTY(&blkdev->finished)) {
 447         ioreq = QLIST_FIRST(&blkdev->finished);
 448         send_notify += blk_send_response_one(ioreq);
 449         ioreq_release(ioreq);
 450     }
 451     if (send_notify) {
 452         xen_be_send_notify(&blkdev->xendev);
 453     }
 454 }
 455
 456 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 457 {
 458     switch (blkdev->protocol) {
 459     case BLKIF_PROTOCOL_NATIVE:
 460         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 461                sizeof(ioreq->req));
 462         break;
 463     case BLKIF_PROTOCOL_X86_32:
 464         blkif_get_x86_32_req(&ioreq->req,
 465                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 466         break;
 467     case BLKIF_PROTOCOL_X86_64:
 468         blkif_get_x86_64_req(&ioreq->req,
 469                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 470         break;
 471     }
 472     return 0;
 473 }
 474
 475 static void blk_handle_requests(struct XenBlkDev *blkdev)
 476 {
 477     RING_IDX rc, rp;
 478     struct ioreq *ioreq;
 479
 480     blkdev->more_work = 0;
 481
 482     rc = blkdev->rings.common.req_cons;
 483     rp = blkdev->rings.common.sring->req_prod;
 484     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 485
 486     blk_send_response_all(blkdev);
 487     while (rc != rp) {
 488         /* pull request from ring */
 489         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 490             break;
 491         }
 492         ioreq = ioreq_start(blkdev);
 493         if (ioreq == NULL) {
 494             blkdev->more_work++;
 495             break;
 496         }
 497         blk_get_request(blkdev, ioreq, rc);
 498         blkdev->rings.common.req_cons = ++rc;
 499
 500         /* parse them */
 501         if (ioreq_parse(ioreq) != 0) {
 502             if (blk_send_response_one(ioreq)) {
 503                 xen_be_send_notify(&blkdev->xendev);
 504             }
 505             ioreq_release(ioreq);
 506             continue;
 507         }
 508
 509         ioreq_runio_qemu_aio(ioreq);
 510     }
 511
 512     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 513         qemu_bh_schedule(blkdev->bh);
 514     }
 515 }
 516
 517 /* ------------------------------------------------------------- */
 518
 519 static void blk_bh(void *opaque)
 520 {
 521     struct XenBlkDev *blkdev = opaque;
 522     blk_handle_requests(blkdev);
 523 }
 524
 525 static void blk_alloc(struct XenDevice *xendev)
 526 {
 527     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 528
 529     QLIST_INIT(&blkdev->inflight);
 530     QLIST_INIT(&blkdev->finished);
 531     QLIST_INIT(&blkdev->freelist);
 532     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 533     if (xen_mode != XEN_EMULATE) {
 534         batch_maps = 1;
 535     }
 536 }
 537
 538 static int blk_init(struct XenDevice *xendev)
 539 {
 540     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 541     int index, qflags, info = 0;
 542
 543     /* read xenstore entries */
 544     if (blkdev->params == NULL) {
 545         char *h = NULL;
 546         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 547         if (blkdev->params != NULL) {
 548             h = strchr(blkdev->params, ':');
 549         }
 550         if (h != NULL) {
 551             blkdev->fileproto = blkdev->params;
 552             blkdev->filename  = h+1;
 553             *h = 0;
 554         } else {
 555             blkdev->fileproto = "<unset>";
 556             blkdev->filename  = blkdev->params;
 557         }
 558     }
 559     if (!strcmp("aio", blkdev->fileproto)) {
 560         blkdev->fileproto = "raw";
 561     }
 562     if (blkdev->mode == NULL) {
 563         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 564     }
 565     if (blkdev->type == NULL) {
 566         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 567     }
 568     if (blkdev->dev == NULL) {
 569         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 570     }
 571     if (blkdev->devtype == NULL) {
 572         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 573     }
 574
 575     /* do we have all we need? */
 576     if (blkdev->params == NULL ||
 577         blkdev->mode == NULL   ||
 578         blkdev->type == NULL   ||
 579         blkdev->dev == NULL) {
 580         goto out_error;
 581     }
 582
 583     /* read-only ? */
 584     if (strcmp(blkdev->mode, "w") == 0) {
 585         qflags = BDRV_O_RDWR;
 586     } else {
 587         qflags = 0;
 588         info  |= VDISK_READONLY;
 589     }
 590
 591     /* cdrom ? */
 592     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 593         info  |= VDISK_CDROM;
 594     }
 595
 596     /* init qemu block driver */
 597     index = (blkdev->xendev.dev - 202 * 256) / 16;
 598     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 599     if (!blkdev->dinfo) {
 600         /* setup via xenbus -> create new block driver instance */
 601         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 602         blkdev->bs = bdrv_new(blkdev->dev);
 603         if (blkdev->bs) {
 604             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 605                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 606                 bdrv_delete(blkdev->bs);
 607                 blkdev->bs = NULL;
 608             }
 609         }
 610         if (!blkdev->bs) {
 611             goto out_error;
 612         }
 613     } else {
 614         /* setup via qemu cmdline -> already setup for us */
 615         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 616         blkdev->bs = blkdev->dinfo->bdrv;
 617     }
 618     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 619     blkdev->file_blk  = BLOCK_SIZE;
 620     blkdev->file_size = bdrv_getlength(blkdev->bs);
 621     if (blkdev->file_size < 0) {
 622         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 623                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 624                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 625         blkdev->file_size = 0;
 626     }
 627
 628     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 629                   " size %" PRId64 " (%" PRId64 " MB)\n",
 630                   blkdev->type, blkdev->fileproto, blkdev->filename,
 631                   blkdev->file_size, blkdev->file_size >> 20);
 632
 633     /* fill info */
 634     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", 1);
 635     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 636     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 637     xenstore_write_be_int(&blkdev->xendev, "sectors",
 638                           blkdev->file_size / blkdev->file_blk);
 639     return 0;
 640
 641 out_error:
 642     g_free(blkdev->params);
 643     blkdev->params = NULL;
 644     g_free(blkdev->mode);
 645     blkdev->mode = NULL;
 646     g_free(blkdev->type);
 647     blkdev->type = NULL;
 648     g_free(blkdev->dev);
 649     blkdev->dev = NULL;
 650     g_free(blkdev->devtype);
 651     blkdev->devtype = NULL;
 652     return -1;
 653 }
 654
 655 static int blk_connect(struct XenDevice *xendev)
 656 {
 657     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 658
 659     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 660         return -1;
 661     }
 662     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 663                              &blkdev->xendev.remote_port) == -1) {
 664         return -1;
 665     }
 666
 667     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 668     if (blkdev->xendev.protocol) {
 669         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 670             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 671         }
 672         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 673             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 674         }
 675     }
 676
 677     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 678                                             blkdev->xendev.dom,
 679                                             blkdev->ring_ref,
 680                                             PROT_READ | PROT_WRITE);
 681     if (!blkdev->sring) {
 682         return -1;
 683     }
 684     blkdev->cnt_map++;
 685
 686     switch (blkdev->protocol) {
 687     case BLKIF_PROTOCOL_NATIVE:
 688     {
 689         blkif_sring_t *sring_native = blkdev->sring;
 690         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 691         break;
 692     }
 693     case BLKIF_PROTOCOL_X86_32:
 694     {
 695         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 696
 697         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 698         break;
 699     }
 700     case BLKIF_PROTOCOL_X86_64:
 701     {
 702         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 703
 704         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 705         break;
 706     }
 707     }
 708
 709     xen_be_bind_evtchn(&blkdev->xendev);
 710
 711     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 712                   "remote port %d, local port %d\n",
 713                   blkdev->xendev.protocol, blkdev->ring_ref,
 714                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 715     return 0;
 716 }
 717
 718 static void blk_disconnect(struct XenDevice *xendev)
 719 {
 720     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 721
 722     if (blkdev->bs) {
 723         if (!blkdev->dinfo) {
 724             /* close/delete only if we created it ourself */
 725             bdrv_close(blkdev->bs);
 726             bdrv_delete(blkdev->bs);
 727         }
 728         blkdev->bs = NULL;
 729     }
 730     xen_be_unbind_evtchn(&blkdev->xendev);
 731
 732     if (blkdev->sring) {
 733         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 734         blkdev->cnt_map--;
 735         blkdev->sring = NULL;
 736     }
 737 }
 738
 739 static int blk_free(struct XenDevice *xendev)
 740 {
 741     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 742     struct ioreq *ioreq;
 743
 744     while (!QLIST_EMPTY(&blkdev->freelist)) {
 745         ioreq = QLIST_FIRST(&blkdev->freelist);
 746         QLIST_REMOVE(ioreq, list);
 747         qemu_iovec_destroy(&ioreq->v);
 748         g_free(ioreq);
 749     }
 750
 751     g_free(blkdev->params);
 752     g_free(blkdev->mode);
 753     g_free(blkdev->type);
 754     g_free(blkdev->dev);
 755     g_free(blkdev->devtype);
 756     qemu_bh_delete(blkdev->bh);
 757     return 0;
 758 }
 759
 760 static void blk_event(struct XenDevice *xendev)
 761 {
 762     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 763
 764     qemu_bh_schedule(blkdev->bh);
 765 }
 766
 767 struct XenDevOps xen_blkdev_ops = {
 768     .size       = sizeof(struct XenBlkDev),
 769     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 770     .alloc      = blk_alloc,
 771     .init       = blk_init,
 772     .initialise    = blk_connect,
 773     .disconnect = blk_disconnect,
 774     .event      = blk_event,
 775     .free       = blk_free,
 776 };