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 "qemu-char.h"
  44 #include "xen_blkif.h"
  45 #include "xen_backend.h"
  46 #include "blockdev.h"
  47
  48 /* ------------------------------------------------------------- */
  49
  50 static int batch_maps   = 0;
  51
  52 static int max_requests = 32;
  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     uint8_t             mapped;
  69
  70     /* grant mapping */
  71     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  72     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  73     int                 prot;
  74     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  75     void                *pages;
  76
  77     /* aio status */
  78     int                 aio_inflight;
  79     int                 aio_errors;
  80
  81     struct XenBlkDev    *blkdev;
  82     QLIST_ENTRY(ioreq)   list;
  83     BlockAcctCookie     acct;
  84 };
  85
  86 struct XenBlkDev {
  87     struct XenDevice    xendev;  /* must be first */
  88     char                *params;
  89     char                *mode;
  90     char                *type;
  91     char                *dev;
  92     char                *devtype;
  93     const char          *fileproto;
  94     const char          *filename;
  95     int                 ring_ref;
  96     void                *sring;
  97     int64_t             file_blk;
  98     int64_t             file_size;
  99     int                 protocol;
 100     blkif_back_rings_t  rings;
 101     int                 more_work;
 102     int                 cnt_map;
 103
 104     /* request lists */
 105     QLIST_HEAD(inflight_head, ioreq) inflight;
 106     QLIST_HEAD(finished_head, ioreq) finished;
 107     QLIST_HEAD(freelist_head, ioreq) freelist;
 108     int                 requests_total;
 109     int                 requests_inflight;
 110     int                 requests_finished;
 111
 112     /* qemu block driver */
 113     DriveInfo           *dinfo;
 114     BlockDriverState    *bs;
 115     QEMUBH              *bh;
 116 };
 117
 118 /* ------------------------------------------------------------- */
 119
 120 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 121 {
 122     struct ioreq *ioreq = NULL;
 123
 124     if (QLIST_EMPTY(&blkdev->freelist)) {
 125         if (blkdev->requests_total >= max_requests) {
 126             goto out;
 127         }
 128         /* allocate new struct */
 129         ioreq = g_malloc0(sizeof(*ioreq));
 130         ioreq->blkdev = blkdev;
 131         blkdev->requests_total++;
 132         qemu_iovec_init(&ioreq->v, BLKIF_MAX_SEGMENTS_PER_REQUEST);
 133     } else {
 134         /* get one from freelist */
 135         ioreq = QLIST_FIRST(&blkdev->freelist);
 136         QLIST_REMOVE(ioreq, list);
 137         qemu_iovec_reset(&ioreq->v);
 138     }
 139     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 140     blkdev->requests_inflight++;
 141
 142 out:
 143     return ioreq;
 144 }
 145
 146 static void ioreq_finish(struct ioreq *ioreq)
 147 {
 148     struct XenBlkDev *blkdev = ioreq->blkdev;
 149
 150     QLIST_REMOVE(ioreq, list);
 151     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 152     blkdev->requests_inflight--;
 153     blkdev->requests_finished++;
 154 }
 155
 156 static void ioreq_release(struct ioreq *ioreq, bool finish)
 157 {
 158     struct XenBlkDev *blkdev = ioreq->blkdev;
 159
 160     QLIST_REMOVE(ioreq, list);
 161     memset(ioreq, 0, sizeof(*ioreq));
 162     ioreq->blkdev = blkdev;
 163     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 164     if (finish) {
 165         blkdev->requests_finished--;
 166     } else {
 167         blkdev->requests_inflight--;
 168     }
 169 }
 170
 171 /*
 172  * translate request into iovec + start offset
 173  * do sanity checks along the way
 174  */
 175 static int ioreq_parse(struct ioreq *ioreq)
 176 {
 177     struct XenBlkDev *blkdev = ioreq->blkdev;
 178     uintptr_t mem;
 179     size_t len;
 180     int i;
 181
 182     xen_be_printf(&blkdev->xendev, 3,
 183                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 184                   ioreq->req.operation, ioreq->req.nr_segments,
 185                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 186     switch (ioreq->req.operation) {
 187     case BLKIF_OP_READ:
 188         ioreq->prot = PROT_WRITE; /* to memory */
 189         break;
 190     case BLKIF_OP_WRITE_BARRIER:
 191         if (!ioreq->req.nr_segments) {
 192             ioreq->presync = 1;
 193             return 0;
 194         }
 195         ioreq->presync = ioreq->postsync = 1;
 196         /* fall through */
 197     case BLKIF_OP_WRITE:
 198         ioreq->prot = PROT_READ; /* from memory */
 199         break;
 200     default:
 201         xen_be_printf(&blkdev->xendev, 0, "error: unknown operation (%d)\n",
 202                       ioreq->req.operation);
 203         goto err;
 204     };
 205
 206     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 207         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 208         goto err;
 209     }
 210
 211     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 212     for (i = 0; i < ioreq->req.nr_segments; i++) {
 213         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 214             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 215             goto err;
 216         }
 217         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 218             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 219             goto err;
 220         }
 221         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 222             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 223             goto err;
 224         }
 225
 226         ioreq->domids[i] = blkdev->xendev.dom;
 227         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 228
 229         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 230         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 231         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 232     }
 233     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 234         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 235         goto err;
 236     }
 237     return 0;
 238
 239 err:
 240     ioreq->status = BLKIF_RSP_ERROR;
 241     return -1;
 242 }
 243
 244 static void ioreq_unmap(struct ioreq *ioreq)
 245 {
 246     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 247     int i;
 248
 249     if (ioreq->v.niov == 0 || ioreq->mapped == 0) {
 250         return;
 251     }
 252     if (batch_maps) {
 253         if (!ioreq->pages) {
 254             return;
 255         }
 256         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0) {
 257             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 258                           strerror(errno));
 259         }
 260         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 261         ioreq->pages = NULL;
 262     } else {
 263         for (i = 0; i < ioreq->v.niov; i++) {
 264             if (!ioreq->page[i]) {
 265                 continue;
 266             }
 267             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0) {
 268                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 269                               strerror(errno));
 270             }
 271             ioreq->blkdev->cnt_map--;
 272             ioreq->page[i] = NULL;
 273         }
 274     }
 275     ioreq->mapped = 0;
 276 }
 277
 278 static int ioreq_map(struct ioreq *ioreq)
 279 {
 280     XenGnttab gnt = ioreq->blkdev->xendev.gnttabdev;
 281     int i;
 282
 283     if (ioreq->v.niov == 0 || ioreq->mapped == 1) {
 284         return 0;
 285     }
 286     if (batch_maps) {
 287         ioreq->pages = xc_gnttab_map_grant_refs
 288             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 289         if (ioreq->pages == NULL) {
 290             xen_be_printf(&ioreq->blkdev->xendev, 0,
 291                           "can't map %d grant refs (%s, %d maps)\n",
 292                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 293             return -1;
 294         }
 295         for (i = 0; i < ioreq->v.niov; i++) {
 296             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 297                 (uintptr_t)ioreq->v.iov[i].iov_base;
 298         }
 299         ioreq->blkdev->cnt_map += ioreq->v.niov;
 300     } else  {
 301         for (i = 0; i < ioreq->v.niov; i++) {
 302             ioreq->page[i] = xc_gnttab_map_grant_ref
 303                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 304             if (ioreq->page[i] == NULL) {
 305                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 306                               "can't map grant ref %d (%s, %d maps)\n",
 307                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 308                 ioreq_unmap(ioreq);
 309                 return -1;
 310             }
 311             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 312             ioreq->blkdev->cnt_map++;
 313         }
 314     }
 315     ioreq->mapped = 1;
 316     return 0;
 317 }
 318
 319 static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
 320
 321 static void qemu_aio_complete(void *opaque, int ret)
 322 {
 323     struct ioreq *ioreq = opaque;
 324
 325     if (ret != 0) {
 326         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 327                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 328         ioreq->aio_errors++;
 329     }
 330
 331     ioreq->aio_inflight--;
 332     if (ioreq->presync) {
 333         ioreq->presync = 0;
 334         ioreq_runio_qemu_aio(ioreq);
 335         return;
 336     }
 337     if (ioreq->aio_inflight > 0) {
 338         return;
 339     }
 340     if (ioreq->postsync) {
 341         ioreq->postsync = 0;
 342         ioreq->aio_inflight++;
 343         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 344         return;
 345     }
 346
 347     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 348     ioreq_unmap(ioreq);
 349     ioreq_finish(ioreq);
 350     bdrv_acct_done(ioreq->blkdev->bs, &ioreq->acct);
 351     qemu_bh_schedule(ioreq->blkdev->bh);
 352 }
 353
 354 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 355 {
 356     struct XenBlkDev *blkdev = ioreq->blkdev;
 357
 358     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1) {
 359         goto err_no_map;
 360     }
 361
 362     ioreq->aio_inflight++;
 363     if (ioreq->presync) {
 364         bdrv_aio_flush(ioreq->blkdev->bs, qemu_aio_complete, ioreq);
 365         return 0;
 366     }
 367
 368     switch (ioreq->req.operation) {
 369     case BLKIF_OP_READ:
 370         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_READ);
 371         ioreq->aio_inflight++;
 372         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 373                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 374                        qemu_aio_complete, ioreq);
 375         break;
 376     case BLKIF_OP_WRITE:
 377     case BLKIF_OP_WRITE_BARRIER:
 378         if (!ioreq->req.nr_segments) {
 379             break;
 380         }
 381
 382         bdrv_acct_start(blkdev->bs, &ioreq->acct, ioreq->v.size, BDRV_ACCT_WRITE);
 383         ioreq->aio_inflight++;
 384         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 385                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 386                         qemu_aio_complete, ioreq);
 387         break;
 388     default:
 389         /* unknown operation (shouldn't happen -- parse catches this) */
 390         goto err;
 391     }
 392
 393     qemu_aio_complete(ioreq, 0);
 394
 395     return 0;
 396
 397 err:
 398     ioreq_unmap(ioreq);
 399 err_no_map:
 400     ioreq_finish(ioreq);
 401     ioreq->status = BLKIF_RSP_ERROR;
 402     return -1;
 403 }
 404
 405 static int blk_send_response_one(struct ioreq *ioreq)
 406 {
 407     struct XenBlkDev  *blkdev = ioreq->blkdev;
 408     int               send_notify   = 0;
 409     int               have_requests = 0;
 410     blkif_response_t  resp;
 411     void              *dst;
 412
 413     resp.id        = ioreq->req.id;
 414     resp.operation = ioreq->req.operation;
 415     resp.status    = ioreq->status;
 416
 417     /* Place on the response ring for the relevant domain. */
 418     switch (blkdev->protocol) {
 419     case BLKIF_PROTOCOL_NATIVE:
 420         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 421         break;
 422     case BLKIF_PROTOCOL_X86_32:
 423         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 424                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 425         break;
 426     case BLKIF_PROTOCOL_X86_64:
 427         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 428                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 429         break;
 430     default:
 431         dst = NULL;
 432     }
 433     memcpy(dst, &resp, sizeof(resp));
 434     blkdev->rings.common.rsp_prod_pvt++;
 435
 436     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 437     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 438         /*
 439          * Tail check for pending requests. Allows frontend to avoid
 440          * notifications if requests are already in flight (lower
 441          * overheads and promotes batching).
 442          */
 443         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 444     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 445         have_requests = 1;
 446     }
 447
 448     if (have_requests) {
 449         blkdev->more_work++;
 450     }
 451     return send_notify;
 452 }
 453
 454 /* walk finished list, send outstanding responses, free requests */
 455 static void blk_send_response_all(struct XenBlkDev *blkdev)
 456 {
 457     struct ioreq *ioreq;
 458     int send_notify = 0;
 459
 460     while (!QLIST_EMPTY(&blkdev->finished)) {
 461         ioreq = QLIST_FIRST(&blkdev->finished);
 462         send_notify += blk_send_response_one(ioreq);
 463         ioreq_release(ioreq, true);
 464     }
 465     if (send_notify) {
 466         xen_be_send_notify(&blkdev->xendev);
 467     }
 468 }
 469
 470 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 471 {
 472     switch (blkdev->protocol) {
 473     case BLKIF_PROTOCOL_NATIVE:
 474         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 475                sizeof(ioreq->req));
 476         break;
 477     case BLKIF_PROTOCOL_X86_32:
 478         blkif_get_x86_32_req(&ioreq->req,
 479                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 480         break;
 481     case BLKIF_PROTOCOL_X86_64:
 482         blkif_get_x86_64_req(&ioreq->req,
 483                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 484         break;
 485     }
 486     return 0;
 487 }
 488
 489 static void blk_handle_requests(struct XenBlkDev *blkdev)
 490 {
 491     RING_IDX rc, rp;
 492     struct ioreq *ioreq;
 493
 494     blkdev->more_work = 0;
 495
 496     rc = blkdev->rings.common.req_cons;
 497     rp = blkdev->rings.common.sring->req_prod;
 498     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 499
 500     blk_send_response_all(blkdev);
 501     while (rc != rp) {
 502         /* pull request from ring */
 503         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 504             break;
 505         }
 506         ioreq = ioreq_start(blkdev);
 507         if (ioreq == NULL) {
 508             blkdev->more_work++;
 509             break;
 510         }
 511         blk_get_request(blkdev, ioreq, rc);
 512         blkdev->rings.common.req_cons = ++rc;
 513
 514         /* parse them */
 515         if (ioreq_parse(ioreq) != 0) {
 516             if (blk_send_response_one(ioreq)) {
 517                 xen_be_send_notify(&blkdev->xendev);
 518             }
 519             ioreq_release(ioreq, false);
 520             continue;
 521         }
 522
 523         ioreq_runio_qemu_aio(ioreq);
 524     }
 525
 526     if (blkdev->more_work && blkdev->requests_inflight < max_requests) {
 527         qemu_bh_schedule(blkdev->bh);
 528     }
 529 }
 530
 531 /* ------------------------------------------------------------- */
 532
 533 static void blk_bh(void *opaque)
 534 {
 535     struct XenBlkDev *blkdev = opaque;
 536     blk_handle_requests(blkdev);
 537 }
 538
 539 /*
 540  * We need to account for the grant allocations requiring contiguous
 541  * chunks; the worst case number would be
 542  *     max_req * max_seg + (max_req - 1) * (max_seg - 1) + 1,
 543  * but in order to keep things simple just use
 544  *     2 * max_req * max_seg.
 545  */
 546 #define MAX_GRANTS(max_req, max_seg) (2 * (max_req) * (max_seg))
 547
 548 static void blk_alloc(struct XenDevice *xendev)
 549 {
 550     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 551
 552     QLIST_INIT(&blkdev->inflight);
 553     QLIST_INIT(&blkdev->finished);
 554     QLIST_INIT(&blkdev->freelist);
 555     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 556     if (xen_mode != XEN_EMULATE) {
 557         batch_maps = 1;
 558     }
 559     if (xc_gnttab_set_max_grants(xendev->gnttabdev,
 560             MAX_GRANTS(max_requests, BLKIF_MAX_SEGMENTS_PER_REQUEST)) < 0) {
 561         xen_be_printf(xendev, 0, "xc_gnttab_set_max_grants failed: %s\n",
 562                       strerror(errno));
 563     }
 564 }
 565
 566 static int blk_init(struct XenDevice *xendev)
 567 {
 568     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 569     int index, qflags, info = 0;
 570
 571     /* read xenstore entries */
 572     if (blkdev->params == NULL) {
 573         char *h = NULL;
 574         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 575         if (blkdev->params != NULL) {
 576             h = strchr(blkdev->params, ':');
 577         }
 578         if (h != NULL) {
 579             blkdev->fileproto = blkdev->params;
 580             blkdev->filename  = h+1;
 581             *h = 0;
 582         } else {
 583             blkdev->fileproto = "<unset>";
 584             blkdev->filename  = blkdev->params;
 585         }
 586     }
 587     if (!strcmp("aio", blkdev->fileproto)) {
 588         blkdev->fileproto = "raw";
 589     }
 590     if (blkdev->mode == NULL) {
 591         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 592     }
 593     if (blkdev->type == NULL) {
 594         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 595     }
 596     if (blkdev->dev == NULL) {
 597         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 598     }
 599     if (blkdev->devtype == NULL) {
 600         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 601     }
 602
 603     /* do we have all we need? */
 604     if (blkdev->params == NULL ||
 605         blkdev->mode == NULL   ||
 606         blkdev->type == NULL   ||
 607         blkdev->dev == NULL) {
 608         goto out_error;
 609     }
 610
 611     /* read-only ? */
 612     qflags = BDRV_O_NOCACHE | BDRV_O_CACHE_WB | BDRV_O_NATIVE_AIO;
 613     if (strcmp(blkdev->mode, "w") == 0) {
 614         qflags |= BDRV_O_RDWR;
 615     } else {
 616         info  |= VDISK_READONLY;
 617     }
 618
 619     /* cdrom ? */
 620     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 621         info  |= VDISK_CDROM;
 622     }
 623
 624     /* init qemu block driver */
 625     index = (blkdev->xendev.dev - 202 * 256) / 16;
 626     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 627     if (!blkdev->dinfo) {
 628         /* setup via xenbus -> create new block driver instance */
 629         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 630         blkdev->bs = bdrv_new(blkdev->dev);
 631         if (blkdev->bs) {
 632             if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 633                         bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 634                 bdrv_delete(blkdev->bs);
 635                 blkdev->bs = NULL;
 636             }
 637         }
 638         if (!blkdev->bs) {
 639             goto out_error;
 640         }
 641     } else {
 642         /* setup via qemu cmdline -> already setup for us */
 643         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 644         blkdev->bs = blkdev->dinfo->bdrv;
 645     }
 646     bdrv_attach_dev_nofail(blkdev->bs, blkdev);
 647     blkdev->file_blk  = BLOCK_SIZE;
 648     blkdev->file_size = bdrv_getlength(blkdev->bs);
 649     if (blkdev->file_size < 0) {
 650         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 651                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 652                       bdrv_get_format_name(blkdev->bs) ?: "-");
 653         blkdev->file_size = 0;
 654     }
 655
 656     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 657                   " size %" PRId64 " (%" PRId64 " MB)\n",
 658                   blkdev->type, blkdev->fileproto, blkdev->filename,
 659                   blkdev->file_size, blkdev->file_size >> 20);
 660
 661     /* fill info */
 662     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", 1);
 663     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 664     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 665     xenstore_write_be_int(&blkdev->xendev, "sectors",
 666                           blkdev->file_size / blkdev->file_blk);
 667     return 0;
 668
 669 out_error:
 670     g_free(blkdev->params);
 671     blkdev->params = NULL;
 672     g_free(blkdev->mode);
 673     blkdev->mode = NULL;
 674     g_free(blkdev->type);
 675     blkdev->type = NULL;
 676     g_free(blkdev->dev);
 677     blkdev->dev = NULL;
 678     g_free(blkdev->devtype);
 679     blkdev->devtype = NULL;
 680     return -1;
 681 }
 682
 683 static int blk_connect(struct XenDevice *xendev)
 684 {
 685     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 686
 687     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1) {
 688         return -1;
 689     }
 690     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 691                              &blkdev->xendev.remote_port) == -1) {
 692         return -1;
 693     }
 694
 695     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 696     if (blkdev->xendev.protocol) {
 697         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 698             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 699         }
 700         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 701             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 702         }
 703     }
 704
 705     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 706                                             blkdev->xendev.dom,
 707                                             blkdev->ring_ref,
 708                                             PROT_READ | PROT_WRITE);
 709     if (!blkdev->sring) {
 710         return -1;
 711     }
 712     blkdev->cnt_map++;
 713
 714     switch (blkdev->protocol) {
 715     case BLKIF_PROTOCOL_NATIVE:
 716     {
 717         blkif_sring_t *sring_native = blkdev->sring;
 718         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 719         break;
 720     }
 721     case BLKIF_PROTOCOL_X86_32:
 722     {
 723         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 724
 725         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 726         break;
 727     }
 728     case BLKIF_PROTOCOL_X86_64:
 729     {
 730         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 731
 732         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 733         break;
 734     }
 735     }
 736
 737     xen_be_bind_evtchn(&blkdev->xendev);
 738
 739     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 740                   "remote port %d, local port %d\n",
 741                   blkdev->xendev.protocol, blkdev->ring_ref,
 742                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 743     return 0;
 744 }
 745
 746 static void blk_disconnect(struct XenDevice *xendev)
 747 {
 748     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 749
 750     if (blkdev->bs) {
 751         if (!blkdev->dinfo) {
 752             /* close/delete only if we created it ourself */
 753             bdrv_close(blkdev->bs);
 754             bdrv_detach_dev(blkdev->bs, blkdev);
 755             bdrv_delete(blkdev->bs);
 756         }
 757         blkdev->bs = NULL;
 758     }
 759     xen_be_unbind_evtchn(&blkdev->xendev);
 760
 761     if (blkdev->sring) {
 762         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 763         blkdev->cnt_map--;
 764         blkdev->sring = NULL;
 765     }
 766 }
 767
 768 static int blk_free(struct XenDevice *xendev)
 769 {
 770     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 771     struct ioreq *ioreq;
 772
 773     if (blkdev->bs || blkdev->sring) {
 774         blk_disconnect(xendev);
 775     }
 776
 777     while (!QLIST_EMPTY(&blkdev->freelist)) {
 778         ioreq = QLIST_FIRST(&blkdev->freelist);
 779         QLIST_REMOVE(ioreq, list);
 780         qemu_iovec_destroy(&ioreq->v);
 781         g_free(ioreq);
 782     }
 783
 784     g_free(blkdev->params);
 785     g_free(blkdev->mode);
 786     g_free(blkdev->type);
 787     g_free(blkdev->dev);
 788     g_free(blkdev->devtype);
 789     qemu_bh_delete(blkdev->bh);
 790     return 0;
 791 }
 792
 793 static void blk_event(struct XenDevice *xendev)
 794 {
 795     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 796
 797     qemu_bh_schedule(blkdev->bh);
 798 }
 799
 800 struct XenDevOps xen_blkdev_ops = {
 801     .size       = sizeof(struct XenBlkDev),
 802     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 803     .alloc      = blk_alloc,
 804     .init       = blk_init,
 805     .initialise    = blk_connect,
 806     .disconnect = blk_disconnect,
 807     .event      = blk_event,
 808     .free       = blk_free,
 809 };