hw/xen_disk.c

   1 /*
   2  *  xen paravirt block device backend
   3  *
   4  *  (c) Gerd Hoffmann <[email protected]>
   5  *
   6  *  This program is free software; you can redistribute it and/or modify
   7  *  it under the terms of the GNU General Public License as published by
   8  *  the Free Software Foundation; under version 2 of the License.
   9  *
  10  *  This program is distributed in the hope that it will be useful,
  11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  *  GNU General Public License for more details.
  14  *
  15  *  You should have received a copy of the GNU General Public License along
  16  *  with this program; if not, see <http://www.gnu.org/licenses/>.
  17  */
  18
  19 #include <stdio.h>
  20 #include <stdlib.h>
  21 #include <stdarg.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <signal.h>
  25 #include <inttypes.h>
  26 #include <time.h>
  27 #include <fcntl.h>
  28 #include <errno.h>
  29 #include <sys/ioctl.h>
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <sys/mman.h>
  33 #include <sys/uio.h>
  34
  35 #include <xs.h>
  36 #include <xenctrl.h>
  37 #include <xen/io/xenbus.h>
  38
  39 #include "hw.h"
  40 #include "block_int.h"
  41 #include "qemu-char.h"
  42 #include "xen_blkif.h"
  43 #include "xen_backend.h"
  44 #include "blockdev.h"
  45
  46 /* ------------------------------------------------------------- */
  47
  48 static int syncwrite    = 0;
  49 static int batch_maps   = 0;
  50
  51 static int max_requests = 32;
  52 static int use_aio      = 1;
  53
  54 /* ------------------------------------------------------------- */
  55
  56 #define BLOCK_SIZE  512
  57 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  58
  59 struct ioreq {
  60     blkif_request_t     req;
  61     int16_t             status;
  62
  63     /* parsed request */
  64     off_t               start;
  65     QEMUIOVector        v;
  66     int                 presync;
  67     int                 postsync;
  68
  69     /* grant mapping */
  70     uint32_t            domids[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  71     uint32_t            refs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  72     int                 prot;
  73     void                *page[BLKIF_MAX_SEGMENTS_PER_REQUEST];
  74     void                *pages;
  75
  76     /* aio status */
  77     int                 aio_inflight;
  78     int                 aio_errors;
  79
  80     struct XenBlkDev    *blkdev;
  81     QLIST_ENTRY(ioreq)   list;
  82 };
  83
  84 struct XenBlkDev {
  85     struct XenDevice    xendev;  /* must be first */
  86     char                *params;
  87     char                *mode;
  88     char                *type;
  89     char                *dev;
  90     char                *devtype;
  91     const char          *fileproto;
  92     const char          *filename;
  93     int                 ring_ref;
  94     void                *sring;
  95     int64_t             file_blk;
  96     int64_t             file_size;
  97     int                 protocol;
  98     blkif_back_rings_t  rings;
  99     int                 more_work;
 100     int                 cnt_map;
 101
 102     /* request lists */
 103     QLIST_HEAD(inflight_head, ioreq) inflight;
 104     QLIST_HEAD(finished_head, ioreq) finished;
 105     QLIST_HEAD(freelist_head, ioreq) freelist;
 106     int                 requests_total;
 107     int                 requests_inflight;
 108     int                 requests_finished;
 109
 110     /* qemu block driver */
 111     DriveInfo           *dinfo;
 112     BlockDriverState    *bs;
 113     QEMUBH              *bh;
 114 };
 115
 116 /* ------------------------------------------------------------- */
 117
 118 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 119 {
 120     struct ioreq *ioreq = NULL;
 121
 122     if (QLIST_EMPTY(&blkdev->freelist)) {
 123         if (blkdev->requests_total >= max_requests)
 124             goto out;
 125         /* 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 = QLIST_FIRST(&blkdev->freelist);
 133         QLIST_REMOVE(ioreq, list);
 134         qemu_iovec_reset(&ioreq->v);
 135     }
 136     QLIST_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     QLIST_REMOVE(ioreq, list);
 148     QLIST_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     QLIST_REMOVE(ioreq, list);
 158     memset(ioreq, 0, sizeof(*ioreq));
 159     ioreq->blkdev = blkdev;
 160     QLIST_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         break;
 183     case BLKIF_OP_WRITE_BARRIER:
 184         if (!ioreq->req.nr_segments) {
 185             ioreq->presync = 1;
 186             return 0;
 187         }
 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     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 203         xen_be_printf(&blkdev->xendev, 0, "error: write req for ro device\n");
 204         goto err;
 205     }
 206
 207     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 208     for (i = 0; i < ioreq->req.nr_segments; i++) {
 209         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 210             xen_be_printf(&blkdev->xendev, 0, "error: nr_segments too big\n");
 211             goto err;
 212         }
 213         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 214             xen_be_printf(&blkdev->xendev, 0, "error: first > last sector\n");
 215             goto err;
 216         }
 217         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 218             xen_be_printf(&blkdev->xendev, 0, "error: page crossing\n");
 219             goto err;
 220         }
 221
 222         ioreq->domids[i] = blkdev->xendev.dom;
 223         ioreq->refs[i]   = ioreq->req.seg[i].gref;
 224
 225         mem = ioreq->req.seg[i].first_sect * blkdev->file_blk;
 226         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 227         qemu_iovec_add(&ioreq->v, (void*)mem, len);
 228     }
 229     if (ioreq->start + ioreq->v.size > blkdev->file_size) {
 230         xen_be_printf(&blkdev->xendev, 0, "error: access beyond end of file\n");
 231         goto err;
 232     }
 233     return 0;
 234
 235 err:
 236     ioreq->status = BLKIF_RSP_ERROR;
 237     return -1;
 238 }
 239
 240 static void ioreq_unmap(struct ioreq *ioreq)
 241 {
 242     int gnt = ioreq->blkdev->xendev.gnttabdev;
 243     int i;
 244
 245     if (ioreq->v.niov == 0)
 246         return;
 247     if (batch_maps) {
 248         if (!ioreq->pages)
 249             return;
 250         if (xc_gnttab_munmap(gnt, ioreq->pages, ioreq->v.niov) != 0)
 251             xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 252                           strerror(errno));
 253         ioreq->blkdev->cnt_map -= ioreq->v.niov;
 254         ioreq->pages = NULL;
 255     } else {
 256         for (i = 0; i < ioreq->v.niov; i++) {
 257             if (!ioreq->page[i])
 258                 continue;
 259             if (xc_gnttab_munmap(gnt, ioreq->page[i], 1) != 0)
 260                 xen_be_printf(&ioreq->blkdev->xendev, 0, "xc_gnttab_munmap failed: %s\n",
 261                               strerror(errno));
 262             ioreq->blkdev->cnt_map--;
 263             ioreq->page[i] = NULL;
 264         }
 265     }
 266 }
 267
 268 static int ioreq_map(struct ioreq *ioreq)
 269 {
 270     int gnt = ioreq->blkdev->xendev.gnttabdev;
 271     int i;
 272
 273     if (ioreq->v.niov == 0)
 274         return 0;
 275     if (batch_maps) {
 276         ioreq->pages = xc_gnttab_map_grant_refs
 277             (gnt, ioreq->v.niov, ioreq->domids, ioreq->refs, ioreq->prot);
 278         if (ioreq->pages == NULL) {
 279             xen_be_printf(&ioreq->blkdev->xendev, 0,
 280                           "can't map %d grant refs (%s, %d maps)\n",
 281                           ioreq->v.niov, strerror(errno), ioreq->blkdev->cnt_map);
 282             return -1;
 283         }
 284         for (i = 0; i < ioreq->v.niov; i++)
 285             ioreq->v.iov[i].iov_base = ioreq->pages + i * XC_PAGE_SIZE +
 286                 (uintptr_t)ioreq->v.iov[i].iov_base;
 287         ioreq->blkdev->cnt_map += ioreq->v.niov;
 288     } else  {
 289         for (i = 0; i < ioreq->v.niov; i++) {
 290             ioreq->page[i] = xc_gnttab_map_grant_ref
 291                 (gnt, ioreq->domids[i], ioreq->refs[i], ioreq->prot);
 292             if (ioreq->page[i] == NULL) {
 293                 xen_be_printf(&ioreq->blkdev->xendev, 0,
 294                               "can't map grant ref %d (%s, %d maps)\n",
 295                               ioreq->refs[i], strerror(errno), ioreq->blkdev->cnt_map);
 296                 ioreq_unmap(ioreq);
 297                 return -1;
 298             }
 299             ioreq->v.iov[i].iov_base = ioreq->page[i] + (uintptr_t)ioreq->v.iov[i].iov_base;
 300             ioreq->blkdev->cnt_map++;
 301         }
 302     }
 303     return 0;
 304 }
 305
 306 static int ioreq_runio_qemu_sync(struct ioreq *ioreq)
 307 {
 308     struct XenBlkDev *blkdev = ioreq->blkdev;
 309     int i, rc, len = 0;
 310     off_t pos;
 311
 312     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1)
 313         goto err;
 314     if (ioreq->presync)
 315         bdrv_flush(blkdev->bs);
 316
 317     switch (ioreq->req.operation) {
 318     case BLKIF_OP_READ:
 319         pos = ioreq->start;
 320         for (i = 0; i < ioreq->v.niov; i++) {
 321             rc = bdrv_read(blkdev->bs, pos / BLOCK_SIZE,
 322                            ioreq->v.iov[i].iov_base,
 323                            ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 324             if (rc != 0) {
 325                 xen_be_printf(&blkdev->xendev, 0, "rd I/O error (%p, len %zd)\n",
 326                               ioreq->v.iov[i].iov_base,
 327                               ioreq->v.iov[i].iov_len);
 328                 goto err;
 329             }
 330             len += ioreq->v.iov[i].iov_len;
 331             pos += ioreq->v.iov[i].iov_len;
 332         }
 333         break;
 334     case BLKIF_OP_WRITE:
 335     case BLKIF_OP_WRITE_BARRIER:
 336         if (!ioreq->req.nr_segments)
 337             break;
 338         pos = ioreq->start;
 339         for (i = 0; i < ioreq->v.niov; i++) {
 340             rc = bdrv_write(blkdev->bs, pos / BLOCK_SIZE,
 341                             ioreq->v.iov[i].iov_base,
 342                             ioreq->v.iov[i].iov_len / BLOCK_SIZE);
 343             if (rc != 0) {
 344                 xen_be_printf(&blkdev->xendev, 0, "wr I/O error (%p, len %zd)\n",
 345                               ioreq->v.iov[i].iov_base,
 346                               ioreq->v.iov[i].iov_len);
 347                 goto err;
 348             }
 349             len += ioreq->v.iov[i].iov_len;
 350             pos += ioreq->v.iov[i].iov_len;
 351         }
 352         break;
 353     default:
 354         /* unknown operation (shouldn't happen -- parse catches this) */
 355         goto err;
 356     }
 357
 358     if (ioreq->postsync)
 359         bdrv_flush(blkdev->bs);
 360     ioreq->status = BLKIF_RSP_OKAY;
 361
 362     ioreq_unmap(ioreq);
 363     ioreq_finish(ioreq);
 364     return 0;
 365
 366 err:
 367     ioreq->status = BLKIF_RSP_ERROR;
 368     return -1;
 369 }
 370
 371 static void qemu_aio_complete(void *opaque, int ret)
 372 {
 373     struct ioreq *ioreq = opaque;
 374
 375     if (ret != 0) {
 376         xen_be_printf(&ioreq->blkdev->xendev, 0, "%s I/O error\n",
 377                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 378         ioreq->aio_errors++;
 379     }
 380
 381     ioreq->aio_inflight--;
 382     if (ioreq->aio_inflight > 0)
 383         return;
 384
 385     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 386     ioreq_unmap(ioreq);
 387     ioreq_finish(ioreq);
 388     qemu_bh_schedule(ioreq->blkdev->bh);
 389 }
 390
 391 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 392 {
 393     struct XenBlkDev *blkdev = ioreq->blkdev;
 394
 395     if (ioreq->req.nr_segments && ioreq_map(ioreq) == -1)
 396         goto err;
 397
 398     ioreq->aio_inflight++;
 399     if (ioreq->presync)
 400         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 401
 402     switch (ioreq->req.operation) {
 403     case BLKIF_OP_READ:
 404         ioreq->aio_inflight++;
 405         bdrv_aio_readv(blkdev->bs, ioreq->start / BLOCK_SIZE,
 406                        &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 407                        qemu_aio_complete, ioreq);
 408         break;
 409     case BLKIF_OP_WRITE:
 410     case BLKIF_OP_WRITE_BARRIER:
 411         if (!ioreq->req.nr_segments)
 412             break;
 413         ioreq->aio_inflight++;
 414         bdrv_aio_writev(blkdev->bs, ioreq->start / BLOCK_SIZE,
 415                         &ioreq->v, ioreq->v.size / BLOCK_SIZE,
 416                         qemu_aio_complete, ioreq);
 417         break;
 418     default:
 419         /* unknown operation (shouldn't happen -- parse catches this) */
 420         goto err;
 421     }
 422
 423     if (ioreq->postsync)
 424         bdrv_flush(blkdev->bs); /* FIXME: aio_flush() ??? */
 425     qemu_aio_complete(ioreq, 0);
 426
 427     return 0;
 428
 429 err:
 430     ioreq->status = BLKIF_RSP_ERROR;
 431     return -1;
 432 }
 433
 434 static int blk_send_response_one(struct ioreq *ioreq)
 435 {
 436     struct XenBlkDev  *blkdev = ioreq->blkdev;
 437     int               send_notify   = 0;
 438     int               have_requests = 0;
 439     blkif_response_t  resp;
 440     void              *dst;
 441
 442     resp.id        = ioreq->req.id;
 443     resp.operation = ioreq->req.operation;
 444     resp.status    = ioreq->status;
 445
 446     /* Place on the response ring for the relevant domain. */
 447     switch (blkdev->protocol) {
 448     case BLKIF_PROTOCOL_NATIVE:
 449         dst = RING_GET_RESPONSE(&blkdev->rings.native, blkdev->rings.native.rsp_prod_pvt);
 450         break;
 451     case BLKIF_PROTOCOL_X86_32:
 452         dst = RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 453                                 blkdev->rings.x86_32_part.rsp_prod_pvt);
 454         break;
 455     case BLKIF_PROTOCOL_X86_64:
 456         dst = RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 457                                 blkdev->rings.x86_64_part.rsp_prod_pvt);
 458         break;
 459     default:
 460         dst = NULL;
 461     }
 462     memcpy(dst, &resp, sizeof(resp));
 463     blkdev->rings.common.rsp_prod_pvt++;
 464
 465     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 466     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 467         /*
 468          * Tail check for pending requests. Allows frontend to avoid
 469          * notifications if requests are already in flight (lower
 470          * overheads and promotes batching).
 471          */
 472         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 473     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 474         have_requests = 1;
 475     }
 476
 477     if (have_requests)
 478         blkdev->more_work++;
 479     return send_notify;
 480 }
 481
 482 /* walk finished list, send outstanding responses, free requests */
 483 static void blk_send_response_all(struct XenBlkDev *blkdev)
 484 {
 485     struct ioreq *ioreq;
 486     int send_notify = 0;
 487
 488     while (!QLIST_EMPTY(&blkdev->finished)) {
 489         ioreq = QLIST_FIRST(&blkdev->finished);
 490         send_notify += blk_send_response_one(ioreq);
 491         ioreq_release(ioreq);
 492     }
 493     if (send_notify)
 494         xen_be_send_notify(&blkdev->xendev);
 495 }
 496
 497 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 498 {
 499     switch (blkdev->protocol) {
 500     case BLKIF_PROTOCOL_NATIVE:
 501         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 502                sizeof(ioreq->req));
 503         break;
 504     case BLKIF_PROTOCOL_X86_32:
 505         blkif_get_x86_32_req(&ioreq->req,
 506                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 507         break;
 508     case BLKIF_PROTOCOL_X86_64:
 509         blkif_get_x86_64_req(&ioreq->req,
 510                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 511         break;
 512     }
 513     return 0;
 514 }
 515
 516 static void blk_handle_requests(struct XenBlkDev *blkdev)
 517 {
 518     RING_IDX rc, rp;
 519     struct ioreq *ioreq;
 520
 521     blkdev->more_work = 0;
 522
 523     rc = blkdev->rings.common.req_cons;
 524     rp = blkdev->rings.common.sring->req_prod;
 525     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 526
 527     if (use_aio)
 528         blk_send_response_all(blkdev);
 529     while (rc != rp) {
 530         /* pull request from ring */
 531         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc))
 532             break;
 533         ioreq = ioreq_start(blkdev);
 534         if (ioreq == NULL) {
 535             blkdev->more_work++;
 536             break;
 537         }
 538         blk_get_request(blkdev, ioreq, rc);
 539         blkdev->rings.common.req_cons = ++rc;
 540
 541         /* parse them */
 542         if (ioreq_parse(ioreq) != 0) {
 543             if (blk_send_response_one(ioreq))
 544                 xen_be_send_notify(&blkdev->xendev);
 545             ioreq_release(ioreq);
 546             continue;
 547         }
 548
 549         if (use_aio) {
 550             /* run i/o in aio mode */
 551             ioreq_runio_qemu_aio(ioreq);
 552         } else {
 553             /* run i/o in sync mode */
 554             ioreq_runio_qemu_sync(ioreq);
 555         }
 556     }
 557     if (!use_aio)
 558         blk_send_response_all(blkdev);
 559
 560     if (blkdev->more_work && blkdev->requests_inflight < max_requests)
 561         qemu_bh_schedule(blkdev->bh);
 562 }
 563
 564 /* ------------------------------------------------------------- */
 565
 566 static void blk_bh(void *opaque)
 567 {
 568     struct XenBlkDev *blkdev = opaque;
 569     blk_handle_requests(blkdev);
 570 }
 571
 572 static void blk_alloc(struct XenDevice *xendev)
 573 {
 574     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 575
 576     QLIST_INIT(&blkdev->inflight);
 577     QLIST_INIT(&blkdev->finished);
 578     QLIST_INIT(&blkdev->freelist);
 579     blkdev->bh = qemu_bh_new(blk_bh, blkdev);
 580     if (xen_mode != XEN_EMULATE)
 581         batch_maps = 1;
 582 }
 583
 584 static int blk_init(struct XenDevice *xendev)
 585 {
 586     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 587     int index, qflags, have_barriers, info = 0;
 588     char *h;
 589
 590     /* read xenstore entries */
 591     if (blkdev->params == NULL) {
 592         blkdev->params = xenstore_read_be_str(&blkdev->xendev, "params");
 593         h = strchr(blkdev->params, ':');
 594         if (h != NULL) {
 595             blkdev->fileproto = blkdev->params;
 596             blkdev->filename  = h+1;
 597             *h = 0;
 598         } else {
 599             blkdev->fileproto = "<unset>";
 600             blkdev->filename  = blkdev->params;
 601         }
 602     }
 603     if (blkdev->mode == NULL)
 604         blkdev->mode = xenstore_read_be_str(&blkdev->xendev, "mode");
 605     if (blkdev->type == NULL)
 606         blkdev->type = xenstore_read_be_str(&blkdev->xendev, "type");
 607     if (blkdev->dev == NULL)
 608         blkdev->dev = xenstore_read_be_str(&blkdev->xendev, "dev");
 609     if (blkdev->devtype == NULL)
 610         blkdev->devtype = xenstore_read_be_str(&blkdev->xendev, "device-type");
 611
 612     /* do we have all we need? */
 613     if (blkdev->params == NULL ||
 614         blkdev->mode == NULL   ||
 615         blkdev->type == NULL   ||
 616         blkdev->dev == NULL)
 617         return -1;
 618
 619     /* read-only ? */
 620     if (strcmp(blkdev->mode, "w") == 0) {
 621         qflags = BDRV_O_RDWR;
 622     } else {
 623         qflags = 0;
 624         info  |= VDISK_READONLY;
 625     }
 626
 627     /* cdrom ? */
 628     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom"))
 629         info  |= VDISK_CDROM;
 630
 631     /* init qemu block driver */
 632     index = (blkdev->xendev.dev - 202 * 256) / 16;
 633     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 634     if (!blkdev->dinfo) {
 635         /* setup via xenbus -> create new block driver instance */
 636         xen_be_printf(&blkdev->xendev, 2, "create new bdrv (xenbus setup)\n");
 637         blkdev->bs = bdrv_new(blkdev->dev);
 638         if (bdrv_open(blkdev->bs, blkdev->filename, qflags,
 639                       bdrv_find_whitelisted_format(blkdev->fileproto)) != 0) {
 640             bdrv_delete(blkdev->bs);
 641             return -1;
 642         }
 643     } else {
 644         /* setup via qemu cmdline -> already setup for us */
 645         xen_be_printf(&blkdev->xendev, 2, "get configured bdrv (cmdline setup)\n");
 646         blkdev->bs = blkdev->dinfo->bdrv;
 647     }
 648     blkdev->file_blk  = BLOCK_SIZE;
 649     blkdev->file_size = bdrv_getlength(blkdev->bs);
 650     if (blkdev->file_size < 0) {
 651         xen_be_printf(&blkdev->xendev, 1, "bdrv_getlength: %d (%s) | drv %s\n",
 652                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 653                       blkdev->bs->drv ? blkdev->bs->drv->format_name : "-");
 654         blkdev->file_size = 0;
 655     }
 656     have_barriers = blkdev->bs->drv && blkdev->bs->drv->bdrv_flush ? 1 : 0;
 657
 658     xen_be_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 659                   " size %" PRId64 " (%" PRId64 " MB)\n",
 660                   blkdev->type, blkdev->fileproto, blkdev->filename,
 661                   blkdev->file_size, blkdev->file_size >> 20);
 662
 663     /* fill info */
 664     xenstore_write_be_int(&blkdev->xendev, "feature-barrier", have_barriers);
 665     xenstore_write_be_int(&blkdev->xendev, "info",            info);
 666     xenstore_write_be_int(&blkdev->xendev, "sector-size",     blkdev->file_blk);
 667     xenstore_write_be_int(&blkdev->xendev, "sectors",
 668                           blkdev->file_size / blkdev->file_blk);
 669     return 0;
 670 }
 671
 672 static int blk_connect(struct XenDevice *xendev)
 673 {
 674     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 675
 676     if (xenstore_read_fe_int(&blkdev->xendev, "ring-ref", &blkdev->ring_ref) == -1)
 677         return -1;
 678     if (xenstore_read_fe_int(&blkdev->xendev, "event-channel",
 679                              &blkdev->xendev.remote_port) == -1)
 680         return -1;
 681
 682     blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 683     if (blkdev->xendev.protocol) {
 684         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_32) == 0)
 685             blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 686         if (strcmp(blkdev->xendev.protocol, XEN_IO_PROTO_ABI_X86_64) == 0)
 687             blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 688     }
 689
 690     blkdev->sring = xc_gnttab_map_grant_ref(blkdev->xendev.gnttabdev,
 691                                             blkdev->xendev.dom,
 692                                             blkdev->ring_ref,
 693                                             PROT_READ | PROT_WRITE);
 694     if (!blkdev->sring)
 695         return -1;
 696     blkdev->cnt_map++;
 697
 698     switch (blkdev->protocol) {
 699     case BLKIF_PROTOCOL_NATIVE:
 700     {
 701         blkif_sring_t *sring_native = blkdev->sring;
 702         BACK_RING_INIT(&blkdev->rings.native, sring_native, XC_PAGE_SIZE);
 703         break;
 704     }
 705     case BLKIF_PROTOCOL_X86_32:
 706     {
 707         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 708
 709         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, XC_PAGE_SIZE);
 710         break;
 711     }
 712     case BLKIF_PROTOCOL_X86_64:
 713     {
 714         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 715
 716         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, XC_PAGE_SIZE);
 717         break;
 718     }
 719     }
 720
 721     xen_be_bind_evtchn(&blkdev->xendev);
 722
 723     xen_be_printf(&blkdev->xendev, 1, "ok: proto %s, ring-ref %d, "
 724                   "remote port %d, local port %d\n",
 725                   blkdev->xendev.protocol, blkdev->ring_ref,
 726                   blkdev->xendev.remote_port, blkdev->xendev.local_port);
 727     return 0;
 728 }
 729
 730 static void blk_disconnect(struct XenDevice *xendev)
 731 {
 732     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 733
 734     if (blkdev->bs) {
 735         if (!blkdev->dinfo) {
 736             /* close/delete only if we created it ourself */
 737             bdrv_close(blkdev->bs);
 738             bdrv_delete(blkdev->bs);
 739         }
 740         blkdev->bs = NULL;
 741     }
 742     xen_be_unbind_evtchn(&blkdev->xendev);
 743
 744     if (blkdev->sring) {
 745         xc_gnttab_munmap(blkdev->xendev.gnttabdev, blkdev->sring, 1);
 746         blkdev->cnt_map--;
 747         blkdev->sring = NULL;
 748     }
 749 }
 750
 751 static int blk_free(struct XenDevice *xendev)
 752 {
 753     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 754     struct ioreq *ioreq;
 755
 756     while (!QLIST_EMPTY(&blkdev->freelist)) {
 757         ioreq = QLIST_FIRST(&blkdev->freelist);
 758         QLIST_REMOVE(ioreq, list);
 759         qemu_iovec_destroy(&ioreq->v);
 760         qemu_free(ioreq);
 761     }
 762
 763     qemu_free(blkdev->params);
 764     qemu_free(blkdev->mode);
 765     qemu_free(blkdev->type);
 766     qemu_free(blkdev->dev);
 767     qemu_free(blkdev->devtype);
 768     qemu_bh_delete(blkdev->bh);
 769     return 0;
 770 }
 771
 772 static void blk_event(struct XenDevice *xendev)
 773 {
 774     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 775
 776     qemu_bh_schedule(blkdev->bh);
 777 }
 778
 779 struct XenDevOps xen_blkdev_ops = {
 780     .size       = sizeof(struct XenBlkDev),
 781     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
 782     .alloc      = blk_alloc,
 783     .init       = blk_init,
 784     .connect    = blk_connect,
 785     .disconnect = blk_disconnect,
 786     .event      = blk_event,
 787     .free       = blk_free,
 788 };