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         if (BLKIF_OP_READ != ioreq->req.operation && blkdev->mode[0] != 'w') {
 183             xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 184             goto err;
 185         }
 186         break;
 187     case BLKIF_OP_WRITE_BARRIER:
 188         if (!syncwrite)
 189             ioreq->presync = ioreq->postsync = 1;
 190         /* fall through */
 191     case BLKIF_OP_WRITE:
 192         ioreq->prot = PROT_READ; /* from memory */
 193         if (syncwrite)
 194             ioreq->postsync = 1;
 195         break;
 196     default:
 197         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 198                       ioreq->req.operation);
 199         goto err;
 200     };
 201
 202     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 203     for (i = 0; i < ioreq->req.nr_segments; i++) {
 204         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 205             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 206             goto err;
 207         }
 208         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 209             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 210             goto err;
 211         }
 212         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 213             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 214             goto err;
 215         }
 216
 217         ioreq->domids[i] = blkdev->xendev.dom;
 218         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 219
 220         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 221         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 222         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 223     }
 224     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 225         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 226         goto err;
 227     }
 228     return 0;
 229
 230 err:
 231     ioreq->status = BLKIF_RSP_ERROR;
 232     return -1;
 233 }
 234
 235 static void ioreq_unmap(struct ioreq *ioreq)
 236 {
 237     int gnt = ioreq->blkdev->xendev.gnttabdev;
 238     int i;
 239
 240     if (ioreq->v.niov == 0)
 241         return;
 242     if (batch_maps) {
 243         if (!ioreq->pages)
 244             return;
 245         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0)
 246             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 247                           strerror(errno));
 248         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 249         ioreq->pages = NULL;
 250     } else {
 251         for (i = 0; i < ioreq->v.niov; i++) {
 252             if (!ioreq->page[i])
 253                 continue;
 254             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0)
 255                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 256                               strerror(errno));
 257             ioreq->blkdev->cnt_map--;
 258             ioreq->page[i] = NULL;
 259         }
 260     }
 261 }
 262
 263 static int ioreq_map(struct ioreq *ioreq)
 264 {
 265     int gnt = ioreq->blkdev->xendev.gnttabdev;
 266     int i;
 267
 268     if (ioreq->v.niov == 0)
 269         return 0;
 270     if (batch_maps) {
 271         ioreq->pages = xc_gnttab_map_grant_refs
 272             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 273         if (ioreq->pages == NULL) {
 274             xen_be_printf(&ioreq->blkdev->xendev, 0,
 275                           "can't map %d grant refs (%s, %d maps)\n",
 276                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 277             return -1;
 278         }
 279         for (i = 0; i < ioreq->v.niov; i++)
 280             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 281                 (uintptr_t)ioreq->v.iov[i].iov_base;
 282         ioreq->blkdev->cnt_map += ioreq->v.niov;
 283     } else  {
 284         for (i = 0; i < ioreq->v.niov; i++) {
 285             ioreq->page[i] = xc_gnttab_map_grant_ref
 286                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 287             if (ioreq->page[i] == NULL) {
 288                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 289                               "can't map grant ref %d (%s, %d maps)\n",
 290                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 291                 ioreq_unmap(ioreq);
 292                 return -1;
 293             }
 294             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 295             ioreq->blkdev->cnt_map++;
 296         }
 297     }
 298     return 0;
 299 }
 300
 301 static int ioreq_runio_qemu_sync(struct ioreq *ioreq)
 302 {
 303     struct XenBlkDev *blkdev = ioreq->blkdev;
 304     int i, rc, len = 0;
 305     off_t pos;
 306
 307     if (ioreq_map(ioreq) == -1)
 308         goto err;
 309     if (ioreq->presync)
 310         bdrv_flush(blkdev->bs);
 311
 312     switch (ioreq->req.operation) {
 313     case BLKIF_OP_READ:
 314         pos = ioreq->start;
 315         for (i = 0; i < ioreq->v.niov; i++) {
 316             rc = bdrv_read(blkdev->bs, pos / BLOCK_SIZE,
 317                            ioreq->v.iov[i].iov_base,
 318                            ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 319             if (rc != 0) {
 320                 xen_be_printf(&blkdev->xendev, 0, "rd I/O error (%p, len %zd)\n",
 321                               ioreq->v.iov[i].iov_base,
 322                               ioreq->v.iov[i].iov_len);
 323                 goto err;
 324             }
 325             len += ioreq->v.iov[i].iov_len;
 326             pos += ioreq->v.iov[i].iov_len;
 327         }
 328         break;
 329     case BLKIF_OP_WRITE:
 330     case BLKIF_OP_WRITE_BARRIER:
 331         pos = ioreq->start;
 332         for (i = 0; i < ioreq->v.niov; i++) {
 333             rc = bdrv_write(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, "wr 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             len += ioreq->v.iov[i].iov_len;
 343             pos += ioreq->v.iov[i].iov_len;
 344         }
 345         break;
 346     default:
 347         /* unknown operation (shouldn't happen -- parse catches this) */
 348         goto err;
 349     }
 350
 351     if (ioreq->postsync)
 352         bdrv_flush(blkdev->bs);
 353     ioreq->status = BLKIF_RSP_OKAY;
 354
 355     ioreq_unmap(ioreq);
 356     ioreq_finish(ioreq);
 357     return 0;
 358
 359 err:
 360     ioreq->status = BLKIF_RSP_ERROR;
 361     return -1;
 362 }
 363
 364 static void qemu_aio_complete(void *opaque, int ret)
 365 {
 366     struct ioreq *ioreq = opaque;
 367
 368     if (ret != 0) {
 369         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 370                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 371         ioreq->aio_errors++;
 372     }
 373
 374     ioreq->aio_inflight--;
 375     if (ioreq->aio_inflight > 0)
 376         return;
 377
 378     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 379     ioreq_unmap(ioreq);
 380     ioreq_finish(ioreq);
 381     qemu_bh_schedule(ioreq->blkdev->bh);
 382 }
 383
 384 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 385 {
 386     struct XenBlkDev *blkdev = ioreq->blkdev;
 387
 388     if (ioreq_map(ioreq) == -1)
 389         goto err;
 390
 391     ioreq->aio_inflight++;
 392     if (ioreq->presync)
 393         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 394
 395     switch (ioreq->req.operation) {
 396     case BLKIF_OP_READ:
 397         ioreq->aio_inflight++;
 398         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 399                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 400                        qemu_aio_complete, ioreq);
 401         break;
 402     case BLKIF_OP_WRITE:
 403     case BLKIF_OP_WRITE_BARRIER:
 404         ioreq->aio_inflight++;
 405         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 406                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 407                         qemu_aio_complete, ioreq);
 408         break;
 409     default:
 410         /* unknown operation (shouldn't happen -- parse catches this) */
 411         goto err;
 412     }
 413
 414     if (ioreq->postsync)
 415         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 416     qemu_aio_complete(ioreq, 0);
 417
 418     return 0;
 419
 420 err:
 421     ioreq->status = BLKIF_RSP_ERROR;
 422     return -1;
 423 }
 424
 425 static int blk_send_response_one(struct ioreq *ioreq)
 426 {
 427     struct XenBlkDev  *blkdev = ioreq->blkdev;
 428     int               send_notify   = 0;
 429     int               have_requests = 0;
 430     blkif_response_t  resp;
 431     void              *dst;
 432
 433     resp.id        = ioreq->req.id;
 434     resp.operation = ioreq->req.operation;
 435     resp.status    = ioreq->status;
 436
 437     /* Place on the response ring for the relevant domain. */
 438     switch (blkdev->protocol) {
 439     case BLKIF_PROTOCOL_NATIVE:
 440         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 441         break;
 442     case BLKIF_PROTOCOL_X86_32:
 443         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32, blkdev->rings.x86_32.rsp_prod_pvt);
 444         break;
 445     case BLKIF_PROTOCOL_X86_64:
 446         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64, blkdev->rings.x86_64.rsp_prod_pvt);
 447         break;
 448     default:
 449         dst = NULL;
 450     }
 451     memcpy(dst, &resp, sizeof(resp));
 452     blkdev->rings.common.rsp_prod_pvt++;
 453
 454     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 455     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 456         /*
 457          * Tail check for pending requests. Allows frontend to avoid
 458          * notifications if requests are already in flight (lower
 459          * overheads and promotes batching).
 460          */
 461         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 462     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 463         have_requests = 1;
 464     }
 465
 466     if (have_requests)
 467         blkdev->more_work++;
 468     return send_notify;
 469 }
 470
 471 /* walk finished list, send outstanding responses, free requests */
 472 static void blk_send_response_all(struct XenBlkDev *blkdev)
 473 {
 474     struct ioreq *ioreq;
 475     int send_notify = 0;
 476
 477     while (!LIST_EMPTY(&blkdev->finished)) {
 478         ioreq = LIST_FIRST(&blkdev->finished);
 479         send_notify += blk_send_response_one(ioreq);
 480         ioreq_release(ioreq);
 481     }
 482     if (send_notify)
 483         xen_be_send_notify(&blkdev->xendev);
 484 }
 485
 486 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 487 {
 488     switch (blkdev->protocol) {
 489     case BLKIF_PROTOCOL_NATIVE:
 490         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 491                sizeof(ioreq->req));
 492         break;
 493     case BLKIF_PROTOCOL_X86_32:
 494         blkif_get_x86_32_req(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.x86_32, rc));
 495         break;
 496     case BLKIF_PROTOCOL_X86_64:
 497         blkif_get_x86_64_req(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.x86_64, rc));
 498         break;
 499     }
 500     return 0;
 501 }
 502
 503 static void blk_handle_requests(struct XenBlkDev *blkdev)
 504 {
 505     RING_IDX rc, rp;
 506     struct ioreq *ioreq;
 507
 508     blkdev->more_work = 0;
 509
 510     rc = blkdev->rings.common.req_cons;
 511     rp = blkdev->rings.common.sring->req_prod;
 512     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 513
 514     if (use_aio)
 515         blk_send_response_all(blkdev);
 516     while ((rc != rp)) {
 517         /* pull request from ring */
 518         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc))
 519             break;
 520         ioreq = ioreq_start(blkdev);
 521         if (ioreq == NULL) {
 522             blkdev->more_work++;
 523             break;
 524         }
 525         blk_get_request(blkdev, ioreq, rc);
 526         blkdev->rings.common.req_cons = ++rc;
 527
 528         /* parse them */
 529         if (ioreq_parse(ioreq) != 0) {
 530             if (blk_send_response_one(ioreq))
 531                 xen_be_send_notify(&blkdev->xendev);
 532             ioreq_release(ioreq);
 533             continue;
 534         }
 535
 536         if (use_aio) {
 537             /* run i/o in aio mode */
 538             ioreq_runio_qemu_aio(ioreq);
 539         } else {
 540             /* run i/o in sync mode */
 541             ioreq_runio_qemu_sync(ioreq);
 542         }
 543     }
 544     if (!use_aio)
 545         blk_send_response_all(blkdev);
 546
 547     if (blkdev->more_work && blkdev->requests_inflight < max_requests)
 548         qemu_bh_schedule(blkdev->bh);
 549 }
 550
 551 /* ------------------------------------------------------------- */
 552
 553 static void blk_bh(void *opaque)
 554 {
 555     struct XenBlkDev *blkdev = opaque;
 556     blk_handle_requests(blkdev);
 557 }
 558
 559 static void blk_alloc(struct XenDevice *xendev)
 560 {
 561     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 562
 563     LIST_INIT(&blkdev->inflight);
 564     LIST_INIT(&blkdev->finished);
 565     LIST_INIT(&blkdev->freelist);
 566     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 567     if (xen_mode != XEN_EMULATE)
 568         batch_maps = 1;
 569 }
 570
 571 static int blk_init(struct XenDevice *xendev)
 572 {
 573     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 574     int mode, qflags, have_barriers, info = 0;
 575     char *h;
 576
 577     /* read xenstore entries */
 578     if (blkdev->params == NULL) {
 579         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 580         h = strchr(blkdev->params, ':');
 581         if (h != NULL) {
 582             blkdev->fileproto = blkdev->params;
 583             blkdev->filename  = h+1;
 584             *h = 0;
 585         } else {
 586             blkdev->fileproto = "<unset>";
 587             blkdev->filename  = blkdev->params;
 588         }
 589     }
 590     if (blkdev->mode == NULL)
 591         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 592     if (blkdev->type == NULL)
 593         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 594     if (blkdev->dev == NULL)
 595         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 596     if (blkdev->devtype == NULL)
 597         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 598
 599     /* do we have all we need? */
 600     if (blkdev->params == NULL ||
 601         blkdev->mode == NULL   ||
 602         blkdev->type == NULL   ||
 603         blkdev->dev == NULL)
 604         return -1;
 605
 606     /* read-only ? */
 607     if (strcmp(blkdev->mode, "w") == 0) {
 608         mode   = O_RDWR;
 609         qflags = BDRV_O_RDWR;
 610     } else {
 611         mode   = O_RDONLY;
 612         qflags = BDRV_O_RDONLY;
 613         info  |= VDISK_READONLY;
 614     }
 615
 616     /* cdrom ? */
 617     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom"))
 618         info  |= VDISK_CDROM;
 619
 620     /* init qemu block driver */
 621     blkdev->index = (blkdev->xendev.dev - 202 * 256) / 16;
 622     blkdev->index = drive_get_index(IF_XEN, 0, blkdev->index);
 623     if (blkdev->index == -1) {
 624         /* setup via xenbus -> create new block driver instance */
 625         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 626         blkdev->bs = bdrv_new(blkdev->dev);
 627         if (blkdev->bs) {
 628             if (bdrv_open2(blkdev->bs, blkdev->filename, qflags,
 629                            bdrv_find_format(blkdev->fileproto)) != 0) {
 630                 bdrv_delete(blkdev->bs);
 631                 blkdev->bs = NULL;
 632             }
 633         }
 634         if (!blkdev->bs)
 635             return -1;
 636     } else {
 637         /* setup via qemu cmdline -> already setup for us */
 638         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 639         blkdev->bs = drives_table[blkdev->index].bdrv;
 640     }
 641     blkdev->file_blk  = BLOCK_SIZE;
 642     blkdev->file_size = bdrv_getlength(blkdev->bs);
 643     if (blkdev->file_size < 0) {
 644         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 645                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 646                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 647         blkdev->file_size = 0;
 648     }
 649     have_barriers = blkdev->bs->drv && blkdev->bs->drv->bdrv_flush ? 1 : 0;
 650
 651     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 652                   " size %" PRId64 " (%" PRId64 " MB)\n",
 653                   blkdev->type, blkdev->fileproto, blkdev->filename,
 654                   blkdev->file_size, blkdev->file_size >> 20);
 655
 656     /* fill info */
 657     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", have_barriers);
 658     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 659     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 660     xenstore_write_be_int(&blkdev->xendev, "sectors",
 661                           blkdev->file_size / blkdev->file_blk);
 662     return 0;
 663 }
 664
 665 static int blk_connect(struct XenDevice *xendev)
 666 {
 667     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 668
 669     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1)
 670         return -1;
 671     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 672                              &blkdev->xendev.remote_port) == -1)
 673         return -1;
 674
 675     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 676     if (blkdev->xendev.protocol) {
 677         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0)
 678             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 679         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0)
 680             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 681     }
 682
 683     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 684                                             blkdev->xendev.dom,
 685                                             blkdev->ring_ref,
 686                                             PROT_READ | PROT_WRITE);
 687     if (!blkdev->sring)
 688         return -1;
 689     blkdev->cnt_map++;
 690
 691     switch (blkdev->protocol) {
 692     case BLKIF_PROTOCOL_NATIVE:
 693     {
 694         blkif_sring_t *sring_native = blkdev->sring;
 695         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 696         break;
 697     }
 698     case BLKIF_PROTOCOL_X86_32:
 699     {
 700         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 701         BACK_RING_INIT(&blkdev->rings.x86_32, sring_x86_32, XC_PAGE_SIZE);
 702         break;
 703     }
 704     case BLKIF_PROTOCOL_X86_64:
 705     {
 706         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 707         BACK_RING_INIT(&blkdev->rings.x86_64, 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->index == -1) {
 727             /* close/delete only if we created it ourself */
 728             bdrv_close(blkdev->bs);
 729             bdrv_delete(blkdev->bs);
 730         }
 731         blkdev->bs = NULL;
 732     }
 733     xen_be_unbind_evtchn(&blkdev->xendev);
 734
 735     if (blkdev->sring) {
 736         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 737         blkdev->cnt_map--;
 738         blkdev->sring = NULL;
 739     }
 740 }
 741
 742 static int blk_free(struct XenDevice *xendev)
 743 {
 744     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 745     struct ioreq *ioreq;
 746
 747     while (!LIST_EMPTY(&blkdev->freelist)) {
 748         ioreq = LIST_FIRST(&blkdev->freelist);
 749         LIST_REMOVE(ioreq, list);
 750         qemu_iovec_destroy(&ioreq->v);
 751         qemu_free(ioreq);
 752     }
 753
 754     qemu_free(blkdev->params);
 755     qemu_free(blkdev->mode);
 756     qemu_free(blkdev->type);
 757     qemu_free(blkdev->dev);
 758     qemu_free(blkdev->devtype);
 759     qemu_bh_delete(blkdev->bh);
 760     return 0;
 761 }
 762
 763 static void blk_event(struct XenDevice *xendev)
 764 {
 765     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 766
 767     qemu_bh_schedule(blkdev->bh);
 768 }
 769
 770 struct XenDevOps xen_blkdev_ops = {
 771     .size       = sizeof(struct XenBlkDev),
 772     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 773     .alloc      = blk_alloc,
 774     .init       = blk_init,
 775     .connect    = blk_connect,
 776     .disconnect = blk_disconnect,
 777     .event      = blk_event,
 778     .free       = blk_free,
 779 };