hw/block/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 "qemu/osdep.h"
  23 #include "qemu/units.h"
  24 #include <sys/ioctl.h>
  25 #include <sys/uio.h>
  26
  27 #include "hw/hw.h"
  28 #include "hw/xen/xen_backend.h"
  29 #include "xen_blkif.h"
  30 #include "sysemu/blockdev.h"
  31 #include "sysemu/iothread.h"
  32 #include "sysemu/block-backend.h"
  33 #include "qapi/error.h"
  34 #include "qapi/qmp/qdict.h"
  35 #include "qapi/qmp/qstring.h"
  36 #include "trace.h"
  37
  38 /* ------------------------------------------------------------- */
  39
  40 #define BLOCK_SIZE  512
  41 #define IOCB_COUNT  (BLKIF_MAX_SEGMENTS_PER_REQUEST + 2)
  42
  43 struct ioreq {
  44     blkif_request_t     req;
  45     int16_t             status;
  46
  47     /* parsed request */
  48     off_t               start;
  49     QEMUIOVector        v;
  50     void                *buf;
  51     size_t              size;
  52     int                 presync;
  53
  54     /* aio status */
  55     int                 aio_inflight;
  56     int                 aio_errors;
  57
  58     struct XenBlkDev    *blkdev;
  59     QLIST_ENTRY(ioreq)   list;
  60     BlockAcctCookie     acct;
  61 };
  62
  63 #define MAX_RING_PAGE_ORDER 4
  64
  65 struct XenBlkDev {
  66     struct XenDevice    xendev;  /* must be first */
  67     char                *params;
  68     char                *mode;
  69     char                *type;
  70     char                *dev;
  71     char                *devtype;
  72     bool                directiosafe;
  73     const char          *fileproto;
  74     const char          *filename;
  75     unsigned int        ring_ref[1 << MAX_RING_PAGE_ORDER];
  76     unsigned int        nr_ring_ref;
  77     void                *sring;
  78     int64_t             file_blk;
  79     int64_t             file_size;
  80     int                 protocol;
  81     blkif_back_rings_t  rings;
  82     int                 more_work;
  83
  84     /* request lists */
  85     QLIST_HEAD(, ioreq) inflight;
  86     QLIST_HEAD(, ioreq) finished;
  87     QLIST_HEAD(, ioreq) freelist;
  88     int                 requests_total;
  89     int                 requests_inflight;
  90     int                 requests_finished;
  91     unsigned int        max_requests;
  92
  93     gboolean            feature_discard;
  94
  95     /* qemu block driver */
  96     DriveInfo           *dinfo;
  97     BlockBackend        *blk;
  98     QEMUBH              *bh;
  99
 100     IOThread            *iothread;
 101     AioContext          *ctx;
 102 };
 103
 104 /* ------------------------------------------------------------- */
 105
 106 static void ioreq_reset(struct ioreq *ioreq)
 107 {
 108     memset(&ioreq->req, 0, sizeof(ioreq->req));
 109     ioreq->status = 0;
 110     ioreq->start = 0;
 111     ioreq->buf = NULL;
 112     ioreq->size = 0;
 113     ioreq->presync = 0;
 114
 115     ioreq->aio_inflight = 0;
 116     ioreq->aio_errors = 0;
 117
 118     ioreq->blkdev = NULL;
 119     memset(&ioreq->list, 0, sizeof(ioreq->list));
 120     memset(&ioreq->acct, 0, sizeof(ioreq->acct));
 121
 122     qemu_iovec_reset(&ioreq->v);
 123 }
 124
 125 static struct ioreq *ioreq_start(struct XenBlkDev *blkdev)
 126 {
 127     struct ioreq *ioreq = NULL;
 128
 129     if (QLIST_EMPTY(&blkdev->freelist)) {
 130         if (blkdev->requests_total >= blkdev->max_requests) {
 131             goto out;
 132         }
 133         /* allocate new struct */
 134         ioreq = g_malloc0(sizeof(*ioreq));
 135         ioreq->blkdev = blkdev;
 136         blkdev->requests_total++;
 137         qemu_iovec_init(&ioreq->v, 1);
 138     } else {
 139         /* get one from freelist */
 140         ioreq = QLIST_FIRST(&blkdev->freelist);
 141         QLIST_REMOVE(ioreq, list);
 142     }
 143     QLIST_INSERT_HEAD(&blkdev->inflight, ioreq, list);
 144     blkdev->requests_inflight++;
 145
 146 out:
 147     return ioreq;
 148 }
 149
 150 static void ioreq_finish(struct ioreq *ioreq)
 151 {
 152     struct XenBlkDev *blkdev = ioreq->blkdev;
 153
 154     QLIST_REMOVE(ioreq, list);
 155     QLIST_INSERT_HEAD(&blkdev->finished, ioreq, list);
 156     blkdev->requests_inflight--;
 157     blkdev->requests_finished++;
 158 }
 159
 160 static void ioreq_release(struct ioreq *ioreq, bool finish)
 161 {
 162     struct XenBlkDev *blkdev = ioreq->blkdev;
 163
 164     QLIST_REMOVE(ioreq, list);
 165     ioreq_reset(ioreq);
 166     ioreq->blkdev = blkdev;
 167     QLIST_INSERT_HEAD(&blkdev->freelist, ioreq, list);
 168     if (finish) {
 169         blkdev->requests_finished--;
 170     } else {
 171         blkdev->requests_inflight--;
 172     }
 173 }
 174
 175 /*
 176  * translate request into iovec + start offset
 177  * do sanity checks along the way
 178  */
 179 static int ioreq_parse(struct ioreq *ioreq)
 180 {
 181     struct XenBlkDev *blkdev = ioreq->blkdev;
 182     struct XenDevice *xendev = &blkdev->xendev;
 183     size_t len;
 184     int i;
 185
 186     xen_pv_printf(xendev, 3,
 187                   "op %d, nr %d, handle %d, id %" PRId64 ", sector %" PRId64 "\n",
 188                   ioreq->req.operation, ioreq->req.nr_segments,
 189                   ioreq->req.handle, ioreq->req.id, ioreq->req.sector_number);
 190     switch (ioreq->req.operation) {
 191     case BLKIF_OP_READ:
 192         break;
 193     case BLKIF_OP_FLUSH_DISKCACHE:
 194         ioreq->presync = 1;
 195         if (!ioreq->req.nr_segments) {
 196             return 0;
 197         }
 198         /* fall through */
 199     case BLKIF_OP_WRITE:
 200         break;
 201     case BLKIF_OP_DISCARD:
 202         return 0;
 203     default:
 204         xen_pv_printf(xendev, 0, "error: unknown operation (%d)\n",
 205                       ioreq->req.operation);
 206         goto err;
 207     };
 208
 209     if (ioreq->req.operation != BLKIF_OP_READ && blkdev->mode[0] != 'w') {
 210         xen_pv_printf(xendev, 0, "error: write req for ro device\n");
 211         goto err;
 212     }
 213
 214     ioreq->start = ioreq->req.sector_number * blkdev->file_blk;
 215     for (i = 0; i < ioreq->req.nr_segments; i++) {
 216         if (i == BLKIF_MAX_SEGMENTS_PER_REQUEST) {
 217             xen_pv_printf(xendev, 0, "error: nr_segments too big\n");
 218             goto err;
 219         }
 220         if (ioreq->req.seg[i].first_sect > ioreq->req.seg[i].last_sect) {
 221             xen_pv_printf(xendev, 0, "error: first > last sector\n");
 222             goto err;
 223         }
 224         if (ioreq->req.seg[i].last_sect * BLOCK_SIZE >= XC_PAGE_SIZE) {
 225             xen_pv_printf(xendev, 0, "error: page crossing\n");
 226             goto err;
 227         }
 228
 229         len = (ioreq->req.seg[i].last_sect - ioreq->req.seg[i].first_sect + 1) * blkdev->file_blk;
 230         ioreq->size += len;
 231     }
 232     if (ioreq->start + ioreq->size > blkdev->file_size) {
 233         xen_pv_printf(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 int ioreq_grant_copy(struct ioreq *ioreq)
 244 {
 245     struct XenBlkDev *blkdev = ioreq->blkdev;
 246     struct XenDevice *xendev = &blkdev->xendev;
 247     XenGrantCopySegment segs[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 248     int i, count, rc;
 249     int64_t file_blk = blkdev->file_blk;
 250     bool to_domain = (ioreq->req.operation == BLKIF_OP_READ);
 251     void *virt = ioreq->buf;
 252
 253     if (ioreq->req.nr_segments == 0) {
 254         return 0;
 255     }
 256
 257     count = ioreq->req.nr_segments;
 258
 259     for (i = 0; i < count; i++) {
 260         if (to_domain) {
 261             segs[i].dest.foreign.ref = ioreq->req.seg[i].gref;
 262             segs[i].dest.foreign.offset = ioreq->req.seg[i].first_sect * file_blk;
 263             segs[i].source.virt = virt;
 264         } else {
 265             segs[i].source.foreign.ref = ioreq->req.seg[i].gref;
 266             segs[i].source.foreign.offset = ioreq->req.seg[i].first_sect * file_blk;
 267             segs[i].dest.virt = virt;
 268         }
 269         segs[i].len = (ioreq->req.seg[i].last_sect
 270                        - ioreq->req.seg[i].first_sect + 1) * file_blk;
 271         virt += segs[i].len;
 272     }
 273
 274     rc = xen_be_copy_grant_refs(xendev, to_domain, segs, count);
 275
 276     if (rc) {
 277         xen_pv_printf(xendev, 0,
 278                       "failed to copy data %d\n", rc);
 279         ioreq->aio_errors++;
 280         return -1;
 281     }
 282
 283     return rc;
 284 }
 285
 286 static int ioreq_runio_qemu_aio(struct ioreq *ioreq);
 287
 288 static void qemu_aio_complete(void *opaque, int ret)
 289 {
 290     struct ioreq *ioreq = opaque;
 291     struct XenBlkDev *blkdev = ioreq->blkdev;
 292     struct XenDevice *xendev = &blkdev->xendev;
 293
 294     aio_context_acquire(blkdev->ctx);
 295
 296     if (ret != 0) {
 297         xen_pv_printf(xendev, 0, "%s I/O error\n",
 298                       ioreq->req.operation == BLKIF_OP_READ ? "read" : "write");
 299         ioreq->aio_errors++;
 300     }
 301
 302     ioreq->aio_inflight--;
 303     if (ioreq->presync) {
 304         ioreq->presync = 0;
 305         ioreq_runio_qemu_aio(ioreq);
 306         goto done;
 307     }
 308     if (ioreq->aio_inflight > 0) {
 309         goto done;
 310     }
 311
 312     switch (ioreq->req.operation) {
 313     case BLKIF_OP_READ:
 314         /* in case of failure ioreq->aio_errors is increased */
 315         if (ret == 0) {
 316             ioreq_grant_copy(ioreq);
 317         }
 318         qemu_vfree(ioreq->buf);
 319         break;
 320     case BLKIF_OP_WRITE:
 321     case BLKIF_OP_FLUSH_DISKCACHE:
 322         if (!ioreq->req.nr_segments) {
 323             break;
 324         }
 325         qemu_vfree(ioreq->buf);
 326         break;
 327     default:
 328         break;
 329     }
 330
 331     ioreq->status = ioreq->aio_errors ? BLKIF_RSP_ERROR : BLKIF_RSP_OKAY;
 332     ioreq_finish(ioreq);
 333
 334     switch (ioreq->req.operation) {
 335     case BLKIF_OP_WRITE:
 336     case BLKIF_OP_FLUSH_DISKCACHE:
 337         if (!ioreq->req.nr_segments) {
 338             break;
 339         }
 340     case BLKIF_OP_READ:
 341         if (ioreq->status == BLKIF_RSP_OKAY) {
 342             block_acct_done(blk_get_stats(blkdev->blk), &ioreq->acct);
 343         } else {
 344             block_acct_failed(blk_get_stats(blkdev->blk), &ioreq->acct);
 345         }
 346         break;
 347     case BLKIF_OP_DISCARD:
 348     default:
 349         break;
 350     }
 351     qemu_bh_schedule(blkdev->bh);
 352
 353 done:
 354     aio_context_release(blkdev->ctx);
 355 }
 356
 357 static bool blk_split_discard(struct ioreq *ioreq, blkif_sector_t sector_number,
 358                               uint64_t nr_sectors)
 359 {
 360     struct XenBlkDev *blkdev = ioreq->blkdev;
 361     int64_t byte_offset;
 362     int byte_chunk;
 363     uint64_t byte_remaining, limit;
 364     uint64_t sec_start = sector_number;
 365     uint64_t sec_count = nr_sectors;
 366
 367     /* Wrap around, or overflowing byte limit? */
 368     if (sec_start + sec_count < sec_count ||
 369         sec_start + sec_count > INT64_MAX >> BDRV_SECTOR_BITS) {
 370         return false;
 371     }
 372
 373     limit = BDRV_REQUEST_MAX_SECTORS << BDRV_SECTOR_BITS;
 374     byte_offset = sec_start << BDRV_SECTOR_BITS;
 375     byte_remaining = sec_count << BDRV_SECTOR_BITS;
 376
 377     do {
 378         byte_chunk = byte_remaining > limit ? limit : byte_remaining;
 379         ioreq->aio_inflight++;
 380         blk_aio_pdiscard(blkdev->blk, byte_offset, byte_chunk,
 381                          qemu_aio_complete, ioreq);
 382         byte_remaining -= byte_chunk;
 383         byte_offset += byte_chunk;
 384     } while (byte_remaining > 0);
 385
 386     return true;
 387 }
 388
 389 static int ioreq_runio_qemu_aio(struct ioreq *ioreq)
 390 {
 391     struct XenBlkDev *blkdev = ioreq->blkdev;
 392
 393     ioreq->buf = qemu_memalign(XC_PAGE_SIZE, ioreq->size);
 394     if (ioreq->req.nr_segments &&
 395         (ioreq->req.operation == BLKIF_OP_WRITE ||
 396          ioreq->req.operation == BLKIF_OP_FLUSH_DISKCACHE) &&
 397         ioreq_grant_copy(ioreq)) {
 398         qemu_vfree(ioreq->buf);
 399         goto err;
 400     }
 401
 402     ioreq->aio_inflight++;
 403     if (ioreq->presync) {
 404         blk_aio_flush(ioreq->blkdev->blk, qemu_aio_complete, ioreq);
 405         return 0;
 406     }
 407
 408     switch (ioreq->req.operation) {
 409     case BLKIF_OP_READ:
 410         qemu_iovec_add(&ioreq->v, ioreq->buf, ioreq->size);
 411         block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
 412                          ioreq->v.size, BLOCK_ACCT_READ);
 413         ioreq->aio_inflight++;
 414         blk_aio_preadv(blkdev->blk, ioreq->start, &ioreq->v, 0,
 415                        qemu_aio_complete, ioreq);
 416         break;
 417     case BLKIF_OP_WRITE:
 418     case BLKIF_OP_FLUSH_DISKCACHE:
 419         if (!ioreq->req.nr_segments) {
 420             break;
 421         }
 422
 423         qemu_iovec_add(&ioreq->v, ioreq->buf, ioreq->size);
 424         block_acct_start(blk_get_stats(blkdev->blk), &ioreq->acct,
 425                          ioreq->v.size,
 426                          ioreq->req.operation == BLKIF_OP_WRITE ?
 427                          BLOCK_ACCT_WRITE : BLOCK_ACCT_FLUSH);
 428         ioreq->aio_inflight++;
 429         blk_aio_pwritev(blkdev->blk, ioreq->start, &ioreq->v, 0,
 430                         qemu_aio_complete, ioreq);
 431         break;
 432     case BLKIF_OP_DISCARD:
 433     {
 434         struct blkif_request_discard *req = (void *)&ioreq->req;
 435         if (!blk_split_discard(ioreq, req->sector_number, req->nr_sectors)) {
 436             goto err;
 437         }
 438         break;
 439     }
 440     default:
 441         /* unknown operation (shouldn't happen -- parse catches this) */
 442         goto err;
 443     }
 444
 445     qemu_aio_complete(ioreq, 0);
 446
 447     return 0;
 448
 449 err:
 450     ioreq_finish(ioreq);
 451     ioreq->status = BLKIF_RSP_ERROR;
 452     return -1;
 453 }
 454
 455 static int blk_send_response_one(struct ioreq *ioreq)
 456 {
 457     struct XenBlkDev  *blkdev = ioreq->blkdev;
 458     int               send_notify   = 0;
 459     int               have_requests = 0;
 460     blkif_response_t  *resp;
 461
 462     /* Place on the response ring for the relevant domain. */
 463     switch (blkdev->protocol) {
 464     case BLKIF_PROTOCOL_NATIVE:
 465         resp = (blkif_response_t *) RING_GET_RESPONSE(&blkdev->rings.native,
 466                                  blkdev->rings.native.rsp_prod_pvt);
 467         break;
 468     case BLKIF_PROTOCOL_X86_32:
 469         resp = (blkif_response_t *) RING_GET_RESPONSE(&blkdev->rings.x86_32_part,
 470                                  blkdev->rings.x86_32_part.rsp_prod_pvt);
 471         break;
 472     case BLKIF_PROTOCOL_X86_64:
 473         resp = (blkif_response_t *) RING_GET_RESPONSE(&blkdev->rings.x86_64_part,
 474                                  blkdev->rings.x86_64_part.rsp_prod_pvt);
 475         break;
 476     default:
 477         return 0;
 478     }
 479
 480     resp->id        = ioreq->req.id;
 481     resp->operation = ioreq->req.operation;
 482     resp->status    = ioreq->status;
 483
 484     blkdev->rings.common.rsp_prod_pvt++;
 485
 486     RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&blkdev->rings.common, send_notify);
 487     if (blkdev->rings.common.rsp_prod_pvt == blkdev->rings.common.req_cons) {
 488         /*
 489          * Tail check for pending requests. Allows frontend to avoid
 490          * notifications if requests are already in flight (lower
 491          * overheads and promotes batching).
 492          */
 493         RING_FINAL_CHECK_FOR_REQUESTS(&blkdev->rings.common, have_requests);
 494     } else if (RING_HAS_UNCONSUMED_REQUESTS(&blkdev->rings.common)) {
 495         have_requests = 1;
 496     }
 497
 498     if (have_requests) {
 499         blkdev->more_work++;
 500     }
 501     return send_notify;
 502 }
 503
 504 /* walk finished list, send outstanding responses, free requests */
 505 static void blk_send_response_all(struct XenBlkDev *blkdev)
 506 {
 507     struct ioreq *ioreq;
 508     int send_notify = 0;
 509
 510     while (!QLIST_EMPTY(&blkdev->finished)) {
 511         ioreq = QLIST_FIRST(&blkdev->finished);
 512         send_notify += blk_send_response_one(ioreq);
 513         ioreq_release(ioreq, true);
 514     }
 515     if (send_notify) {
 516         xen_pv_send_notify(&blkdev->xendev);
 517     }
 518 }
 519
 520 static int blk_get_request(struct XenBlkDev *blkdev, struct ioreq *ioreq, RING_IDX rc)
 521 {
 522     switch (blkdev->protocol) {
 523     case BLKIF_PROTOCOL_NATIVE:
 524         memcpy(&ioreq->req, RING_GET_REQUEST(&blkdev->rings.native, rc),
 525                sizeof(ioreq->req));
 526         break;
 527     case BLKIF_PROTOCOL_X86_32:
 528         blkif_get_x86_32_req(&ioreq->req,
 529                              RING_GET_REQUEST(&blkdev->rings.x86_32_part, rc));
 530         break;
 531     case BLKIF_PROTOCOL_X86_64:
 532         blkif_get_x86_64_req(&ioreq->req,
 533                              RING_GET_REQUEST(&blkdev->rings.x86_64_part, rc));
 534         break;
 535     }
 536     /* Prevent the compiler from accessing the on-ring fields instead. */
 537     barrier();
 538     return 0;
 539 }
 540
 541 static void blk_handle_requests(struct XenBlkDev *blkdev)
 542 {
 543     RING_IDX rc, rp;
 544     struct ioreq *ioreq;
 545
 546     blkdev->more_work = 0;
 547
 548     rc = blkdev->rings.common.req_cons;
 549     rp = blkdev->rings.common.sring->req_prod;
 550     xen_rmb(); /* Ensure we see queued requests up to 'rp'. */
 551
 552     blk_send_response_all(blkdev);
 553     while (rc != rp) {
 554         /* pull request from ring */
 555         if (RING_REQUEST_CONS_OVERFLOW(&blkdev->rings.common, rc)) {
 556             break;
 557         }
 558         ioreq = ioreq_start(blkdev);
 559         if (ioreq == NULL) {
 560             blkdev->more_work++;
 561             break;
 562         }
 563         blk_get_request(blkdev, ioreq, rc);
 564         blkdev->rings.common.req_cons = ++rc;
 565
 566         /* parse them */
 567         if (ioreq_parse(ioreq) != 0) {
 568
 569             switch (ioreq->req.operation) {
 570             case BLKIF_OP_READ:
 571                 block_acct_invalid(blk_get_stats(blkdev->blk),
 572                                    BLOCK_ACCT_READ);
 573                 break;
 574             case BLKIF_OP_WRITE:
 575                 block_acct_invalid(blk_get_stats(blkdev->blk),
 576                                    BLOCK_ACCT_WRITE);
 577                 break;
 578             case BLKIF_OP_FLUSH_DISKCACHE:
 579                 block_acct_invalid(blk_get_stats(blkdev->blk),
 580                                    BLOCK_ACCT_FLUSH);
 581             default:
 582                 break;
 583             };
 584
 585             if (blk_send_response_one(ioreq)) {
 586                 xen_pv_send_notify(&blkdev->xendev);
 587             }
 588             ioreq_release(ioreq, false);
 589             continue;
 590         }
 591
 592         ioreq_runio_qemu_aio(ioreq);
 593     }
 594
 595     if (blkdev->more_work && blkdev->requests_inflight < blkdev->max_requests) {
 596         qemu_bh_schedule(blkdev->bh);
 597     }
 598 }
 599
 600 /* ------------------------------------------------------------- */
 601
 602 static void blk_bh(void *opaque)
 603 {
 604     struct XenBlkDev *blkdev = opaque;
 605
 606     aio_context_acquire(blkdev->ctx);
 607     blk_handle_requests(blkdev);
 608     aio_context_release(blkdev->ctx);
 609 }
 610
 611 static void blk_alloc(struct XenDevice *xendev)
 612 {
 613     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 614     Error *err = NULL;
 615
 616     trace_xen_disk_alloc(xendev->name);
 617
 618     QLIST_INIT(&blkdev->inflight);
 619     QLIST_INIT(&blkdev->finished);
 620     QLIST_INIT(&blkdev->freelist);
 621
 622     blkdev->iothread = iothread_create(xendev->name, &err);
 623     assert(!err);
 624
 625     blkdev->ctx = iothread_get_aio_context(blkdev->iothread);
 626     blkdev->bh = aio_bh_new(blkdev->ctx, blk_bh, blkdev);
 627 }
 628
 629 static void blk_parse_discard(struct XenBlkDev *blkdev)
 630 {
 631     struct XenDevice *xendev = &blkdev->xendev;
 632     int enable;
 633
 634     blkdev->feature_discard = true;
 635
 636     if (xenstore_read_be_int(xendev, "discard-enable", &enable) == 0) {
 637         blkdev->feature_discard = !!enable;
 638     }
 639
 640     if (blkdev->feature_discard) {
 641         xenstore_write_be_int(xendev, "feature-discard", 1);
 642     }
 643 }
 644
 645 static int blk_init(struct XenDevice *xendev)
 646 {
 647     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 648     int info = 0;
 649     char *directiosafe = NULL;
 650
 651     trace_xen_disk_init(xendev->name);
 652
 653     /* read xenstore entries */
 654     if (blkdev->params == NULL) {
 655         char *h = NULL;
 656         blkdev->params = xenstore_read_be_str(xendev, "params");
 657         if (blkdev->params != NULL) {
 658             h = strchr(blkdev->params, ':');
 659         }
 660         if (h != NULL) {
 661             blkdev->fileproto = blkdev->params;
 662             blkdev->filename  = h+1;
 663             *h = 0;
 664         } else {
 665             blkdev->fileproto = "<unset>";
 666             blkdev->filename  = blkdev->params;
 667         }
 668     }
 669     if (!strcmp("aio", blkdev->fileproto)) {
 670         blkdev->fileproto = "raw";
 671     }
 672     if (!strcmp("vhd", blkdev->fileproto)) {
 673         blkdev->fileproto = "vpc";
 674     }
 675     if (blkdev->mode == NULL) {
 676         blkdev->mode = xenstore_read_be_str(xendev, "mode");
 677     }
 678     if (blkdev->type == NULL) {
 679         blkdev->type = xenstore_read_be_str(xendev, "type");
 680     }
 681     if (blkdev->dev == NULL) {
 682         blkdev->dev = xenstore_read_be_str(xendev, "dev");
 683     }
 684     if (blkdev->devtype == NULL) {
 685         blkdev->devtype = xenstore_read_be_str(xendev, "device-type");
 686     }
 687     directiosafe = xenstore_read_be_str(xendev, "direct-io-safe");
 688     blkdev->directiosafe = (directiosafe && atoi(directiosafe));
 689
 690     /* do we have all we need? */
 691     if (blkdev->params == NULL ||
 692         blkdev->mode == NULL   ||
 693         blkdev->type == NULL   ||
 694         blkdev->dev == NULL) {
 695         goto out_error;
 696     }
 697
 698     /* read-only ? */
 699     if (strcmp(blkdev->mode, "w")) {
 700         info  |= VDISK_READONLY;
 701     }
 702
 703     /* cdrom ? */
 704     if (blkdev->devtype && !strcmp(blkdev->devtype, "cdrom")) {
 705         info  |= VDISK_CDROM;
 706     }
 707
 708     blkdev->file_blk  = BLOCK_SIZE;
 709
 710     /* fill info
 711      * blk_connect supplies sector-size and sectors
 712      */
 713     xenstore_write_be_int(xendev, "feature-flush-cache", 1);
 714     xenstore_write_be_int(xendev, "info", info);
 715
 716     xenstore_write_be_int(xendev, "max-ring-page-order",
 717                           MAX_RING_PAGE_ORDER);
 718
 719     blk_parse_discard(blkdev);
 720
 721     g_free(directiosafe);
 722     return 0;
 723
 724 out_error:
 725     g_free(blkdev->params);
 726     blkdev->params = NULL;
 727     g_free(blkdev->mode);
 728     blkdev->mode = NULL;
 729     g_free(blkdev->type);
 730     blkdev->type = NULL;
 731     g_free(blkdev->dev);
 732     blkdev->dev = NULL;
 733     g_free(blkdev->devtype);
 734     blkdev->devtype = NULL;
 735     g_free(directiosafe);
 736     blkdev->directiosafe = false;
 737     return -1;
 738 }
 739
 740 static int blk_connect(struct XenDevice *xendev)
 741 {
 742     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 743     int index, qflags;
 744     bool readonly = true;
 745     bool writethrough = true;
 746     int order, ring_ref;
 747     unsigned int ring_size, max_grants;
 748     unsigned int i;
 749
 750     trace_xen_disk_connect(xendev->name);
 751
 752     /* read-only ? */
 753     if (blkdev->directiosafe) {
 754         qflags = BDRV_O_NOCACHE | BDRV_O_NATIVE_AIO;
 755     } else {
 756         qflags = 0;
 757         writethrough = false;
 758     }
 759     if (strcmp(blkdev->mode, "w") == 0) {
 760         qflags |= BDRV_O_RDWR;
 761         readonly = false;
 762     }
 763     if (blkdev->feature_discard) {
 764         qflags |= BDRV_O_UNMAP;
 765     }
 766
 767     /* init qemu block driver */
 768     index = (xendev->dev - 202 * 256) / 16;
 769     blkdev->dinfo = drive_get(IF_XEN, 0, index);
 770     if (!blkdev->dinfo) {
 771         Error *local_err = NULL;
 772         QDict *options = NULL;
 773
 774         if (strcmp(blkdev->fileproto, "<unset>")) {
 775             options = qdict_new();
 776             qdict_put_str(options, "driver", blkdev->fileproto);
 777         }
 778
 779         /* setup via xenbus -> create new block driver instance */
 780         xen_pv_printf(xendev, 2, "create new bdrv (xenbus setup)\n");
 781         blkdev->blk = blk_new_open(blkdev->filename, NULL, options,
 782                                    qflags, &local_err);
 783         if (!blkdev->blk) {
 784             xen_pv_printf(xendev, 0, "error: %s\n",
 785                           error_get_pretty(local_err));
 786             error_free(local_err);
 787             return -1;
 788         }
 789         blk_set_enable_write_cache(blkdev->blk, !writethrough);
 790     } else {
 791         /* setup via qemu cmdline -> already setup for us */
 792         xen_pv_printf(xendev, 2,
 793                       "get configured bdrv (cmdline setup)\n");
 794         blkdev->blk = blk_by_legacy_dinfo(blkdev->dinfo);
 795         if (blk_is_read_only(blkdev->blk) && !readonly) {
 796             xen_pv_printf(xendev, 0, "Unexpected read-only drive");
 797             blkdev->blk = NULL;
 798             return -1;
 799         }
 800         /* blkdev->blk is not create by us, we get a reference
 801          * so we can blk_unref() unconditionally */
 802         blk_ref(blkdev->blk);
 803     }
 804     blk_attach_dev_legacy(blkdev->blk, blkdev);
 805     blkdev->file_size = blk_getlength(blkdev->blk);
 806     if (blkdev->file_size < 0) {
 807         BlockDriverState *bs = blk_bs(blkdev->blk);
 808         const char *drv_name = bs ? bdrv_get_format_name(bs) : NULL;
 809         xen_pv_printf(xendev, 1, "blk_getlength: %d (%s) | drv %s\n",
 810                       (int)blkdev->file_size, strerror(-blkdev->file_size),
 811                       drv_name ?: "-");
 812         blkdev->file_size = 0;
 813     }
 814
 815     xen_pv_printf(xendev, 1, "type \"%s\", fileproto \"%s\", filename \"%s\","
 816                   " size %" PRId64 " (%" PRId64 " MB)\n",
 817                   blkdev->type, blkdev->fileproto, blkdev->filename,
 818                   blkdev->file_size, blkdev->file_size / MiB);
 819
 820     /* Fill in number of sector size and number of sectors */
 821     xenstore_write_be_int(xendev, "sector-size", blkdev->file_blk);
 822     xenstore_write_be_int64(xendev, "sectors",
 823                             blkdev->file_size / blkdev->file_blk);
 824
 825     if (xenstore_read_fe_int(xendev, "ring-page-order",
 826                              &order) == -1) {
 827         blkdev->nr_ring_ref = 1;
 828
 829         if (xenstore_read_fe_int(xendev, "ring-ref",
 830                                  &ring_ref) == -1) {
 831             return -1;
 832         }
 833         blkdev->ring_ref[0] = ring_ref;
 834
 835     } else if (order >= 0 && order <= MAX_RING_PAGE_ORDER) {
 836         blkdev->nr_ring_ref = 1 << order;
 837
 838         for (i = 0; i < blkdev->nr_ring_ref; i++) {
 839             char *key;
 840
 841             key = g_strdup_printf("ring-ref%u", i);
 842             if (!key) {
 843                 return -1;
 844             }
 845
 846             if (xenstore_read_fe_int(xendev, key,
 847                                      &ring_ref) == -1) {
 848                 g_free(key);
 849                 return -1;
 850             }
 851             blkdev->ring_ref[i] = ring_ref;
 852
 853             g_free(key);
 854         }
 855     } else {
 856         xen_pv_printf(xendev, 0, "invalid ring-page-order: %d\n",
 857                       order);
 858         return -1;
 859     }
 860
 861     if (xenstore_read_fe_int(xendev, "event-channel",
 862                              &xendev->remote_port) == -1) {
 863         return -1;
 864     }
 865
 866     if (!xendev->protocol) {
 867         blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 868     } else if (strcmp(xendev->protocol, XEN_IO_PROTO_ABI_NATIVE) == 0) {
 869         blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 870     } else if (strcmp(xendev->protocol, XEN_IO_PROTO_ABI_X86_32) == 0) {
 871         blkdev->protocol = BLKIF_PROTOCOL_X86_32;
 872     } else if (strcmp(xendev->protocol, XEN_IO_PROTO_ABI_X86_64) == 0) {
 873         blkdev->protocol = BLKIF_PROTOCOL_X86_64;
 874     } else {
 875         blkdev->protocol = BLKIF_PROTOCOL_NATIVE;
 876     }
 877
 878     ring_size = XC_PAGE_SIZE * blkdev->nr_ring_ref;
 879     switch (blkdev->protocol) {
 880     case BLKIF_PROTOCOL_NATIVE:
 881     {
 882         blkdev->max_requests = __CONST_RING_SIZE(blkif, ring_size);
 883         break;
 884     }
 885     case BLKIF_PROTOCOL_X86_32:
 886     {
 887         blkdev->max_requests = __CONST_RING_SIZE(blkif_x86_32, ring_size);
 888         break;
 889     }
 890     case BLKIF_PROTOCOL_X86_64:
 891     {
 892         blkdev->max_requests = __CONST_RING_SIZE(blkif_x86_64, ring_size);
 893         break;
 894     }
 895     default:
 896         return -1;
 897     }
 898
 899     /* Add on the number needed for the ring pages */
 900     max_grants = blkdev->nr_ring_ref;
 901
 902     xen_be_set_max_grant_refs(xendev, max_grants);
 903     blkdev->sring = xen_be_map_grant_refs(xendev, blkdev->ring_ref,
 904                                           blkdev->nr_ring_ref,
 905                                           PROT_READ | PROT_WRITE);
 906     if (!blkdev->sring) {
 907         return -1;
 908     }
 909
 910     switch (blkdev->protocol) {
 911     case BLKIF_PROTOCOL_NATIVE:
 912     {
 913         blkif_sring_t *sring_native = blkdev->sring;
 914         BACK_RING_INIT(&blkdev->rings.native, sring_native, ring_size);
 915         break;
 916     }
 917     case BLKIF_PROTOCOL_X86_32:
 918     {
 919         blkif_x86_32_sring_t *sring_x86_32 = blkdev->sring;
 920
 921         BACK_RING_INIT(&blkdev->rings.x86_32_part, sring_x86_32, ring_size);
 922         break;
 923     }
 924     case BLKIF_PROTOCOL_X86_64:
 925     {
 926         blkif_x86_64_sring_t *sring_x86_64 = blkdev->sring;
 927
 928         BACK_RING_INIT(&blkdev->rings.x86_64_part, sring_x86_64, ring_size);
 929         break;
 930     }
 931     }
 932
 933     blk_set_aio_context(blkdev->blk, blkdev->ctx);
 934
 935     xen_be_bind_evtchn(xendev);
 936
 937     xen_pv_printf(xendev, 1, "ok: proto %s, nr-ring-ref %u, "
 938                   "remote port %d, local port %d\n",
 939                   xendev->protocol, blkdev->nr_ring_ref,
 940                   xendev->remote_port, xendev->local_port);
 941     return 0;
 942 }
 943
 944 static void blk_disconnect(struct XenDevice *xendev)
 945 {
 946     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 947
 948     trace_xen_disk_disconnect(xendev->name);
 949
 950     aio_context_acquire(blkdev->ctx);
 951
 952     if (blkdev->blk) {
 953         blk_set_aio_context(blkdev->blk, qemu_get_aio_context());
 954         blk_detach_dev(blkdev->blk, blkdev);
 955         blk_unref(blkdev->blk);
 956         blkdev->blk = NULL;
 957     }
 958     xen_pv_unbind_evtchn(xendev);
 959
 960     aio_context_release(blkdev->ctx);
 961
 962     if (blkdev->sring) {
 963         xen_be_unmap_grant_refs(xendev, blkdev->sring,
 964                                 blkdev->nr_ring_ref);
 965         blkdev->sring = NULL;
 966     }
 967 }
 968
 969 static int blk_free(struct XenDevice *xendev)
 970 {
 971     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 972     struct ioreq *ioreq;
 973
 974     trace_xen_disk_free(xendev->name);
 975
 976     blk_disconnect(xendev);
 977
 978     while (!QLIST_EMPTY(&blkdev->freelist)) {
 979         ioreq = QLIST_FIRST(&blkdev->freelist);
 980         QLIST_REMOVE(ioreq, list);
 981         qemu_iovec_destroy(&ioreq->v);
 982         g_free(ioreq);
 983     }
 984
 985     g_free(blkdev->params);
 986     g_free(blkdev->mode);
 987     g_free(blkdev->type);
 988     g_free(blkdev->dev);
 989     g_free(blkdev->devtype);
 990     qemu_bh_delete(blkdev->bh);
 991     iothread_destroy(blkdev->iothread);
 992     return 0;
 993 }
 994
 995 static void blk_event(struct XenDevice *xendev)
 996 {
 997     struct XenBlkDev *blkdev = container_of(xendev, struct XenBlkDev, xendev);
 998
 999     qemu_bh_schedule(blkdev->bh);
1000 }
1001
1002 struct XenDevOps xen_blkdev_ops = {
1003     .flags      = DEVOPS_FLAG_NEED_GNTDEV,
1004     .size       = sizeof(struct XenBlkDev),
1005     .alloc      = blk_alloc,
1006     .init       = blk_init,
1007     .initialise = blk_connect,
1008     .disconnect = blk_disconnect,
1009     .event      = blk_event,
1010     .free       = blk_free,
1011 };