block.c

   1 /*
   2  * QEMU System Emulator block driver
   3  *
   4  * Copyright (c) 2003 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "config-host.h"
  25 #ifdef HOST_BSD
  26 /* include native header before sys-queue.h */
  27 #include <sys/queue.h>
  28 #endif
  29
  30 #include "qemu-common.h"
  31 #include "monitor.h"
  32 #include "block_int.h"
  33 #include "osdep.h"
  34
  35 #ifdef HOST_BSD
  36 #include <sys/types.h>
  37 #include <sys/stat.h>
  38 #include <sys/ioctl.h>
  39 #ifndef __DragonFly__
  40 #include <sys/disk.h>
  41 #endif
  42 #endif
  43
  44 #ifdef _WIN32
  45 #include <windows.h>
  46 #endif
  47
  48 #define SECTOR_BITS 9
  49 #define SECTOR_SIZE (1 << SECTOR_BITS)
  50
  51 typedef struct BlockDriverAIOCBSync {
  52     BlockDriverAIOCB common;
  53     QEMUBH *bh;
  54     int ret;
  55 } BlockDriverAIOCBSync;
  56
  57 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
  58         int64_t sector_num, uint8_t *buf, int nb_sectors,
  59         BlockDriverCompletionFunc *cb, void *opaque);
  60 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
  61         int64_t sector_num, const uint8_t *buf, int nb_sectors,
  62         BlockDriverCompletionFunc *cb, void *opaque);
  63 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  64 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  65                         uint8_t *buf, int nb_sectors);
  66 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  67                          const uint8_t *buf, int nb_sectors);
  68
  69 BlockDriverState *bdrv_first;
  70
  71 static BlockDriver *first_drv;
  72
  73 int path_is_absolute(const char *path)
  74 {
  75     const char *p;
  76 #ifdef _WIN32
  77     /* specific case for names like: "\\.\d:" */
  78     if (*path == '/' || *path == '\\')
  79         return 1;
  80 #endif
  81     p = strchr(path, ':');
  82     if (p)
  83         p++;
  84     else
  85         p = path;
  86 #ifdef _WIN32
  87     return (*p == '/' || *p == '\\');
  88 #else
  89     return (*p == '/');
  90 #endif
  91 }
  92
  93 /* if filename is absolute, just copy it to dest. Otherwise, build a
  94    path to it by considering it is relative to base_path. URL are
  95    supported. */
  96 void path_combine(char *dest, int dest_size,
  97                   const char *base_path,
  98                   const char *filename)
  99 {
 100     const char *p, *p1;
 101     int len;
 102
 103     if (dest_size <= 0)
 104         return;
 105     if (path_is_absolute(filename)) {
 106         pstrcpy(dest, dest_size, filename);
 107     } else {
 108         p = strchr(base_path, ':');
 109         if (p)
 110             p++;
 111         else
 112             p = base_path;
 113         p1 = strrchr(base_path, '/');
 114 #ifdef _WIN32
 115         {
 116             const char *p2;
 117             p2 = strrchr(base_path, '\\');
 118             if (!p1 || p2 > p1)
 119                 p1 = p2;
 120         }
 121 #endif
 122         if (p1)
 123             p1++;
 124         else
 125             p1 = base_path;
 126         if (p1 > p)
 127             p = p1;
 128         len = p - base_path;
 129         if (len > dest_size - 1)
 130             len = dest_size - 1;
 131         memcpy(dest, base_path, len);
 132         dest[len] = '\0';
 133         pstrcat(dest, dest_size, filename);
 134     }
 135 }
 136
 137
 138 static void bdrv_register(BlockDriver *bdrv)
 139 {
 140     if (!bdrv->bdrv_aio_read) {
 141         /* add AIO emulation layer */
 142         bdrv->bdrv_aio_read = bdrv_aio_read_em;
 143         bdrv->bdrv_aio_write = bdrv_aio_write_em;
 144         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 145         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 146     } else if (!bdrv->bdrv_read) {
 147         /* add synchronous IO emulation layer */
 148         bdrv->bdrv_read = bdrv_read_em;
 149         bdrv->bdrv_write = bdrv_write_em;
 150     }
 151     bdrv->next = first_drv;
 152     first_drv = bdrv;
 153 }
 154
 155 /* create a new block device (by default it is empty) */
 156 BlockDriverState *bdrv_new(const char *device_name)
 157 {
 158     BlockDriverState **pbs, *bs;
 159
 160     bs = qemu_mallocz(sizeof(BlockDriverState));
 161     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 162     if (device_name[0] != '\0') {
 163         /* insert at the end */
 164         pbs = &bdrv_first;
 165         while (*pbs != NULL)
 166             pbs = &(*pbs)->next;
 167         *pbs = bs;
 168     }
 169     return bs;
 170 }
 171
 172 BlockDriver *bdrv_find_format(const char *format_name)
 173 {
 174     BlockDriver *drv1;
 175     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 176         if (!strcmp(drv1->format_name, format_name))
 177             return drv1;
 178     }
 179     return NULL;
 180 }
 181
 182 int bdrv_create(BlockDriver *drv,
 183                 const char *filename, int64_t size_in_sectors,
 184                 const char *backing_file, int flags)
 185 {
 186     if (!drv->bdrv_create)
 187         return -ENOTSUP;
 188     return drv->bdrv_create(filename, size_in_sectors, backing_file, flags);
 189 }
 190
 191 #ifdef _WIN32
 192 void get_tmp_filename(char *filename, int size)
 193 {
 194     char temp_dir[MAX_PATH];
 195
 196     GetTempPath(MAX_PATH, temp_dir);
 197     GetTempFileName(temp_dir, "qem", 0, filename);
 198 }
 199 #else
 200 void get_tmp_filename(char *filename, int size)
 201 {
 202     int fd;
 203     const char *tmpdir;
 204     /* XXX: race condition possible */
 205     tmpdir = getenv("TMPDIR");
 206     if (!tmpdir)
 207         tmpdir = "/tmp";
 208     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 209     fd = mkstemp(filename);
 210     close(fd);
 211 }
 212 #endif
 213
 214 #ifdef _WIN32
 215 static int is_windows_drive_prefix(const char *filename)
 216 {
 217     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 218              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 219             filename[1] == ':');
 220 }
 221
 222 static int is_windows_drive(const char *filename)
 223 {
 224     if (is_windows_drive_prefix(filename) &&
 225         filename[2] == '\0')
 226         return 1;
 227     if (strstart(filename, "\\\\.\\", NULL) ||
 228         strstart(filename, "//./", NULL))
 229         return 1;
 230     return 0;
 231 }
 232 #endif
 233
 234 static BlockDriver *find_protocol(const char *filename)
 235 {
 236     BlockDriver *drv1;
 237     char protocol[128];
 238     int len;
 239     const char *p;
 240
 241 #ifdef _WIN32
 242     if (is_windows_drive(filename) ||
 243         is_windows_drive_prefix(filename))
 244         return &bdrv_raw;
 245 #endif
 246     p = strchr(filename, ':');
 247     if (!p)
 248         return &bdrv_raw;
 249     len = p - filename;
 250     if (len > sizeof(protocol) - 1)
 251         len = sizeof(protocol) - 1;
 252     memcpy(protocol, filename, len);
 253     protocol[len] = '\0';
 254     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 255         if (drv1->protocol_name &&
 256             !strcmp(drv1->protocol_name, protocol))
 257             return drv1;
 258     }
 259     return NULL;
 260 }
 261
 262 /* XXX: force raw format if block or character device ? It would
 263    simplify the BSD case */
 264 static BlockDriver *find_image_format(const char *filename)
 265 {
 266     int ret, score, score_max;
 267     BlockDriver *drv1, *drv;
 268     uint8_t buf[2048];
 269     BlockDriverState *bs;
 270
 271     /* detect host devices. By convention, /dev/cdrom[N] is always
 272        recognized as a host CDROM */
 273     if (strstart(filename, "/dev/cdrom", NULL))
 274         return &bdrv_host_device;
 275 #ifdef _WIN32
 276     if (is_windows_drive(filename))
 277         return &bdrv_host_device;
 278 #else
 279     {
 280         struct stat st;
 281         if (stat(filename, &st) >= 0 &&
 282             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 283             return &bdrv_host_device;
 284         }
 285     }
 286 #endif
 287
 288     drv = find_protocol(filename);
 289     /* no need to test disk image formats for vvfat */
 290     if (drv == &bdrv_vvfat)
 291         return drv;
 292
 293     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 294     if (ret < 0)
 295         return NULL;
 296     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 297     bdrv_delete(bs);
 298     if (ret < 0) {
 299         return NULL;
 300     }
 301
 302     score_max = 0;
 303     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 304         if (drv1->bdrv_probe) {
 305             score = drv1->bdrv_probe(buf, ret, filename);
 306             if (score > score_max) {
 307                 score_max = score;
 308                 drv = drv1;
 309             }
 310         }
 311     }
 312     return drv;
 313 }
 314
 315 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 316 {
 317     BlockDriverState *bs;
 318     int ret;
 319
 320     bs = bdrv_new("");
 321     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 322     if (ret < 0) {
 323         bdrv_delete(bs);
 324         return ret;
 325     }
 326     bs->growable = 1;
 327     *pbs = bs;
 328     return 0;
 329 }
 330
 331 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 332 {
 333     return bdrv_open2(bs, filename, flags, NULL);
 334 }
 335
 336 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 337                BlockDriver *drv)
 338 {
 339     int ret, open_flags;
 340     char tmp_filename[PATH_MAX];
 341     char backing_filename[PATH_MAX];
 342
 343     bs->read_only = 0;
 344     bs->is_temporary = 0;
 345     bs->encrypted = 0;
 346     bs->valid_key = 0;
 347
 348     if (flags & BDRV_O_SNAPSHOT) {
 349         BlockDriverState *bs1;
 350         int64_t total_size;
 351         int is_protocol = 0;
 352
 353         /* if snapshot, we create a temporary backing file and open it
 354            instead of opening 'filename' directly */
 355
 356         /* if there is a backing file, use it */
 357         bs1 = bdrv_new("");
 358         ret = bdrv_open(bs1, filename, 0);
 359         if (ret < 0) {
 360             bdrv_delete(bs1);
 361             return ret;
 362         }
 363         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 364
 365         if (bs1->drv && bs1->drv->protocol_name)
 366             is_protocol = 1;
 367
 368         bdrv_delete(bs1);
 369
 370         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 371
 372         /* Real path is meaningless for protocols */
 373         if (is_protocol)
 374             snprintf(backing_filename, sizeof(backing_filename),
 375                      "%s", filename);
 376         else
 377             realpath(filename, backing_filename);
 378
 379         ret = bdrv_create(&bdrv_qcow2, tmp_filename,
 380                           total_size, backing_filename, 0);
 381         if (ret < 0) {
 382             return ret;
 383         }
 384         filename = tmp_filename;
 385         bs->is_temporary = 1;
 386     }
 387
 388     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 389     if (flags & BDRV_O_FILE) {
 390         drv = find_protocol(filename);
 391     } else if (!drv) {
 392         drv = find_image_format(filename);
 393     }
 394     if (!drv) {
 395         ret = -ENOENT;
 396         goto unlink_and_fail;
 397     }
 398     bs->drv = drv;
 399     bs->opaque = qemu_mallocz(drv->instance_size);
 400     /* Note: for compatibility, we open disk image files as RDWR, and
 401        RDONLY as fallback */
 402     if (!(flags & BDRV_O_FILE))
 403         open_flags = BDRV_O_RDWR | (flags & BDRV_O_CACHE_MASK);
 404     else
 405         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 406     ret = drv->bdrv_open(bs, filename, open_flags);
 407     if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
 408         ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
 409         bs->read_only = 1;
 410     }
 411     if (ret < 0) {
 412         qemu_free(bs->opaque);
 413         bs->opaque = NULL;
 414         bs->drv = NULL;
 415     unlink_and_fail:
 416         if (bs->is_temporary)
 417             unlink(filename);
 418         return ret;
 419     }
 420     if (drv->bdrv_getlength) {
 421         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 422     }
 423 #ifndef _WIN32
 424     if (bs->is_temporary) {
 425         unlink(filename);
 426     }
 427 #endif
 428     if (bs->backing_file[0] != '\0') {
 429         /* if there is a backing file, use it */
 430         bs->backing_hd = bdrv_new("");
 431         path_combine(backing_filename, sizeof(backing_filename),
 432                      filename, bs->backing_file);
 433         ret = bdrv_open(bs->backing_hd, backing_filename, open_flags);
 434         if (ret < 0) {
 435             bdrv_close(bs);
 436             return ret;
 437         }
 438     }
 439
 440     if (!bdrv_key_required(bs)) {
 441         /* call the change callback */
 442         bs->media_changed = 1;
 443         if (bs->change_cb)
 444             bs->change_cb(bs->change_opaque);
 445     }
 446     return 0;
 447 }
 448
 449 void bdrv_close(BlockDriverState *bs)
 450 {
 451     if (bs->drv) {
 452         if (bs->backing_hd)
 453             bdrv_delete(bs->backing_hd);
 454         bs->drv->bdrv_close(bs);
 455         qemu_free(bs->opaque);
 456 #ifdef _WIN32
 457         if (bs->is_temporary) {
 458             unlink(bs->filename);
 459         }
 460 #endif
 461         bs->opaque = NULL;
 462         bs->drv = NULL;
 463
 464         /* call the change callback */
 465         bs->media_changed = 1;
 466         if (bs->change_cb)
 467             bs->change_cb(bs->change_opaque);
 468     }
 469 }
 470
 471 void bdrv_delete(BlockDriverState *bs)
 472 {
 473     BlockDriverState **pbs;
 474
 475     pbs = &bdrv_first;
 476     while (*pbs != bs && *pbs != NULL)
 477         pbs = &(*pbs)->next;
 478     if (*pbs == bs)
 479         *pbs = bs->next;
 480
 481     bdrv_close(bs);
 482     qemu_free(bs);
 483 }
 484
 485 /* commit COW file into the raw image */
 486 int bdrv_commit(BlockDriverState *bs)
 487 {
 488     BlockDriver *drv = bs->drv;
 489     int64_t i, total_sectors;
 490     int n, j;
 491     unsigned char sector[512];
 492
 493     if (!drv)
 494         return -ENOMEDIUM;
 495
 496     if (bs->read_only) {
 497         return -EACCES;
 498     }
 499
 500     if (!bs->backing_hd) {
 501         return -ENOTSUP;
 502     }
 503
 504     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 505     for (i = 0; i < total_sectors;) {
 506         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 507             for(j = 0; j < n; j++) {
 508                 if (bdrv_read(bs, i, sector, 1) != 0) {
 509                     return -EIO;
 510                 }
 511
 512                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 513                     return -EIO;
 514                 }
 515                 i++;
 516             }
 517         } else {
 518             i += n;
 519         }
 520     }
 521
 522     if (drv->bdrv_make_empty)
 523         return drv->bdrv_make_empty(bs);
 524
 525     return 0;
 526 }
 527
 528 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
 529                                    size_t size)
 530 {
 531     int64_t len;
 532
 533     if (!bdrv_is_inserted(bs))
 534         return -ENOMEDIUM;
 535
 536     if (bs->growable)
 537         return 0;
 538
 539     len = bdrv_getlength(bs);
 540
 541     if ((offset + size) > len)
 542         return -EIO;
 543
 544     return 0;
 545 }
 546
 547 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
 548                               int nb_sectors)
 549 {
 550     int64_t offset;
 551
 552     /* Deal with byte accesses */
 553     if (sector_num < 0)
 554         offset = -sector_num;
 555     else
 556         offset = sector_num * 512;
 557
 558     return bdrv_check_byte_request(bs, offset, nb_sectors * 512);
 559 }
 560
 561 /* return < 0 if error. See bdrv_write() for the return codes */
 562 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 563               uint8_t *buf, int nb_sectors)
 564 {
 565     BlockDriver *drv = bs->drv;
 566
 567     if (!drv)
 568         return -ENOMEDIUM;
 569     if (bdrv_check_request(bs, sector_num, nb_sectors))
 570         return -EIO;
 571
 572     return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 573 }
 574
 575 /* Return < 0 if error. Important errors are:
 576   -EIO         generic I/O error (may happen for all errors)
 577   -ENOMEDIUM   No media inserted.
 578   -EINVAL      Invalid sector number or nb_sectors
 579   -EACCES      Trying to write a read-only device
 580 */
 581 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 582                const uint8_t *buf, int nb_sectors)
 583 {
 584     BlockDriver *drv = bs->drv;
 585     if (!bs->drv)
 586         return -ENOMEDIUM;
 587     if (bs->read_only)
 588         return -EACCES;
 589     if (bdrv_check_request(bs, sector_num, nb_sectors))
 590         return -EIO;
 591
 592     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 593 }
 594
 595 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 596                void *buf, int count1)
 597 {
 598     uint8_t tmp_buf[SECTOR_SIZE];
 599     int len, nb_sectors, count;
 600     int64_t sector_num;
 601
 602     count = count1;
 603     /* first read to align to sector start */
 604     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 605     if (len > count)
 606         len = count;
 607     sector_num = offset >> SECTOR_BITS;
 608     if (len > 0) {
 609         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 610             return -EIO;
 611         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 612         count -= len;
 613         if (count == 0)
 614             return count1;
 615         sector_num++;
 616         buf += len;
 617     }
 618
 619     /* read the sectors "in place" */
 620     nb_sectors = count >> SECTOR_BITS;
 621     if (nb_sectors > 0) {
 622         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 623             return -EIO;
 624         sector_num += nb_sectors;
 625         len = nb_sectors << SECTOR_BITS;
 626         buf += len;
 627         count -= len;
 628     }
 629
 630     /* add data from the last sector */
 631     if (count > 0) {
 632         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 633             return -EIO;
 634         memcpy(buf, tmp_buf, count);
 635     }
 636     return count1;
 637 }
 638
 639 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 640                 const void *buf, int count1)
 641 {
 642     uint8_t tmp_buf[SECTOR_SIZE];
 643     int len, nb_sectors, count;
 644     int64_t sector_num;
 645
 646     count = count1;
 647     /* first write to align to sector start */
 648     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 649     if (len > count)
 650         len = count;
 651     sector_num = offset >> SECTOR_BITS;
 652     if (len > 0) {
 653         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 654             return -EIO;
 655         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 656         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 657             return -EIO;
 658         count -= len;
 659         if (count == 0)
 660             return count1;
 661         sector_num++;
 662         buf += len;
 663     }
 664
 665     /* write the sectors "in place" */
 666     nb_sectors = count >> SECTOR_BITS;
 667     if (nb_sectors > 0) {
 668         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 669             return -EIO;
 670         sector_num += nb_sectors;
 671         len = nb_sectors << SECTOR_BITS;
 672         buf += len;
 673         count -= len;
 674     }
 675
 676     /* add data from the last sector */
 677     if (count > 0) {
 678         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 679             return -EIO;
 680         memcpy(tmp_buf, buf, count);
 681         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 682             return -EIO;
 683     }
 684     return count1;
 685 }
 686
 687 /**
 688  * Truncate file to 'offset' bytes (needed only for file protocols)
 689  */
 690 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 691 {
 692     BlockDriver *drv = bs->drv;
 693     if (!drv)
 694         return -ENOMEDIUM;
 695     if (!drv->bdrv_truncate)
 696         return -ENOTSUP;
 697     return drv->bdrv_truncate(bs, offset);
 698 }
 699
 700 /**
 701  * Length of a file in bytes. Return < 0 if error or unknown.
 702  */
 703 int64_t bdrv_getlength(BlockDriverState *bs)
 704 {
 705     BlockDriver *drv = bs->drv;
 706     if (!drv)
 707         return -ENOMEDIUM;
 708     if (!drv->bdrv_getlength) {
 709         /* legacy mode */
 710         return bs->total_sectors * SECTOR_SIZE;
 711     }
 712     return drv->bdrv_getlength(bs);
 713 }
 714
 715 /* return 0 as number of sectors if no device present or error */
 716 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 717 {
 718     int64_t length;
 719     length = bdrv_getlength(bs);
 720     if (length < 0)
 721         length = 0;
 722     else
 723         length = length >> SECTOR_BITS;
 724     *nb_sectors_ptr = length;
 725 }
 726
 727 struct partition {
 728         uint8_t boot_ind;           /* 0x80 - active */
 729         uint8_t head;               /* starting head */
 730         uint8_t sector;             /* starting sector */
 731         uint8_t cyl;                /* starting cylinder */
 732         uint8_t sys_ind;            /* What partition type */
 733         uint8_t end_head;           /* end head */
 734         uint8_t end_sector;         /* end sector */
 735         uint8_t end_cyl;            /* end cylinder */
 736         uint32_t start_sect;        /* starting sector counting from 0 */
 737         uint32_t nr_sects;          /* nr of sectors in partition */
 738 } __attribute__((packed));
 739
 740 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
 741 static int guess_disk_lchs(BlockDriverState *bs,
 742                            int *pcylinders, int *pheads, int *psectors)
 743 {
 744     uint8_t *buf;
 745     int ret, i, heads, sectors, cylinders;
 746     struct partition *p;
 747     uint32_t nr_sects;
 748     uint64_t nb_sectors;
 749
 750     buf = qemu_memalign(512, 512);
 751     if (buf == NULL)
 752         return -1;
 753
 754     bdrv_get_geometry(bs, &nb_sectors);
 755
 756     ret = bdrv_read(bs, 0, buf, 1);
 757     if (ret < 0)
 758         return -1;
 759     /* test msdos magic */
 760     if (buf[510] != 0x55 || buf[511] != 0xaa) {
 761         qemu_free(buf);
 762         return -1;
 763     }
 764     for(i = 0; i < 4; i++) {
 765         p = ((struct partition *)(buf + 0x1be)) + i;
 766         nr_sects = le32_to_cpu(p->nr_sects);
 767         if (nr_sects && p->end_head) {
 768             /* We make the assumption that the partition terminates on
 769                a cylinder boundary */
 770             heads = p->end_head + 1;
 771             sectors = p->end_sector & 63;
 772             if (sectors == 0)
 773                 continue;
 774             cylinders = nb_sectors / (heads * sectors);
 775             if (cylinders < 1 || cylinders > 16383)
 776                 continue;
 777             *pheads = heads;
 778             *psectors = sectors;
 779             *pcylinders = cylinders;
 780 #if 0
 781             printf("guessed geometry: LCHS=%d %d %d\n",
 782                    cylinders, heads, sectors);
 783 #endif
 784             qemu_free(buf);
 785             return 0;
 786         }
 787     }
 788     qemu_free(buf);
 789     return -1;
 790 }
 791
 792 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
 793 {
 794     int translation, lba_detected = 0;
 795     int cylinders, heads, secs;
 796     uint64_t nb_sectors;
 797
 798     /* if a geometry hint is available, use it */
 799     bdrv_get_geometry(bs, &nb_sectors);
 800     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
 801     translation = bdrv_get_translation_hint(bs);
 802     if (cylinders != 0) {
 803         *pcyls = cylinders;
 804         *pheads = heads;
 805         *psecs = secs;
 806     } else {
 807         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
 808             if (heads > 16) {
 809                 /* if heads > 16, it means that a BIOS LBA
 810                    translation was active, so the default
 811                    hardware geometry is OK */
 812                 lba_detected = 1;
 813                 goto default_geometry;
 814             } else {
 815                 *pcyls = cylinders;
 816                 *pheads = heads;
 817                 *psecs = secs;
 818                 /* disable any translation to be in sync with
 819                    the logical geometry */
 820                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
 821                     bdrv_set_translation_hint(bs,
 822                                               BIOS_ATA_TRANSLATION_NONE);
 823                 }
 824             }
 825         } else {
 826         default_geometry:
 827             /* if no geometry, use a standard physical disk geometry */
 828             cylinders = nb_sectors / (16 * 63);
 829
 830             if (cylinders > 16383)
 831                 cylinders = 16383;
 832             else if (cylinders < 2)
 833                 cylinders = 2;
 834             *pcyls = cylinders;
 835             *pheads = 16;
 836             *psecs = 63;
 837             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
 838                 if ((*pcyls * *pheads) <= 131072) {
 839                     bdrv_set_translation_hint(bs,
 840                                               BIOS_ATA_TRANSLATION_LARGE);
 841                 } else {
 842                     bdrv_set_translation_hint(bs,
 843                                               BIOS_ATA_TRANSLATION_LBA);
 844                 }
 845             }
 846         }
 847         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
 848     }
 849 }
 850
 851 void bdrv_set_geometry_hint(BlockDriverState *bs,
 852                             int cyls, int heads, int secs)
 853 {
 854     bs->cyls = cyls;
 855     bs->heads = heads;
 856     bs->secs = secs;
 857 }
 858
 859 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 860 {
 861     bs->type = type;
 862     bs->removable = ((type == BDRV_TYPE_CDROM ||
 863                       type == BDRV_TYPE_FLOPPY));
 864 }
 865
 866 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 867 {
 868     bs->translation = translation;
 869 }
 870
 871 void bdrv_get_geometry_hint(BlockDriverState *bs,
 872                             int *pcyls, int *pheads, int *psecs)
 873 {
 874     *pcyls = bs->cyls;
 875     *pheads = bs->heads;
 876     *psecs = bs->secs;
 877 }
 878
 879 int bdrv_get_type_hint(BlockDriverState *bs)
 880 {
 881     return bs->type;
 882 }
 883
 884 int bdrv_get_translation_hint(BlockDriverState *bs)
 885 {
 886     return bs->translation;
 887 }
 888
 889 int bdrv_is_removable(BlockDriverState *bs)
 890 {
 891     return bs->removable;
 892 }
 893
 894 int bdrv_is_read_only(BlockDriverState *bs)
 895 {
 896     return bs->read_only;
 897 }
 898
 899 int bdrv_is_sg(BlockDriverState *bs)
 900 {
 901     return bs->sg;
 902 }
 903
 904 /* XXX: no longer used */
 905 void bdrv_set_change_cb(BlockDriverState *bs,
 906                         void (*change_cb)(void *opaque), void *opaque)
 907 {
 908     bs->change_cb = change_cb;
 909     bs->change_opaque = opaque;
 910 }
 911
 912 int bdrv_is_encrypted(BlockDriverState *bs)
 913 {
 914     if (bs->backing_hd && bs->backing_hd->encrypted)
 915         return 1;
 916     return bs->encrypted;
 917 }
 918
 919 int bdrv_key_required(BlockDriverState *bs)
 920 {
 921     BlockDriverState *backing_hd = bs->backing_hd;
 922
 923     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
 924         return 1;
 925     return (bs->encrypted && !bs->valid_key);
 926 }
 927
 928 int bdrv_set_key(BlockDriverState *bs, const char *key)
 929 {
 930     int ret;
 931     if (bs->backing_hd && bs->backing_hd->encrypted) {
 932         ret = bdrv_set_key(bs->backing_hd, key);
 933         if (ret < 0)
 934             return ret;
 935         if (!bs->encrypted)
 936             return 0;
 937     }
 938     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 939         return -1;
 940     ret = bs->drv->bdrv_set_key(bs, key);
 941     if (ret < 0) {
 942         bs->valid_key = 0;
 943     } else if (!bs->valid_key) {
 944         bs->valid_key = 1;
 945         /* call the change callback now, we skipped it on open */
 946         bs->media_changed = 1;
 947         if (bs->change_cb)
 948             bs->change_cb(bs->change_opaque);
 949     }
 950     return ret;
 951 }
 952
 953 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
 954 {
 955     if (!bs->drv) {
 956         buf[0] = '\0';
 957     } else {
 958         pstrcpy(buf, buf_size, bs->drv->format_name);
 959     }
 960 }
 961
 962 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
 963                          void *opaque)
 964 {
 965     BlockDriver *drv;
 966
 967     for (drv = first_drv; drv != NULL; drv = drv->next) {
 968         it(opaque, drv->format_name);
 969     }
 970 }
 971
 972 BlockDriverState *bdrv_find(const char *name)
 973 {
 974     BlockDriverState *bs;
 975
 976     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 977         if (!strcmp(name, bs->device_name))
 978             return bs;
 979     }
 980     return NULL;
 981 }
 982
 983 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
 984 {
 985     BlockDriverState *bs;
 986
 987     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
 988         it(opaque, bs);
 989     }
 990 }
 991
 992 const char *bdrv_get_device_name(BlockDriverState *bs)
 993 {
 994     return bs->device_name;
 995 }
 996
 997 void bdrv_flush(BlockDriverState *bs)
 998 {
 999     if (bs->drv->bdrv_flush)
1000         bs->drv->bdrv_flush(bs);
1001     if (bs->backing_hd)
1002         bdrv_flush(bs->backing_hd);
1003 }
1004
1005 void bdrv_flush_all(void)
1006 {
1007     BlockDriverState *bs;
1008
1009     for (bs = bdrv_first; bs != NULL; bs = bs->next)
1010         if (bs->drv && !bdrv_is_read_only(bs) &&
1011             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1012             bdrv_flush(bs);
1013 }
1014
1015 /*
1016  * Returns true iff the specified sector is present in the disk image. Drivers
1017  * not implementing the functionality are assumed to not support backing files,
1018  * hence all their sectors are reported as allocated.
1019  *
1020  * 'pnum' is set to the number of sectors (including and immediately following
1021  * the specified sector) that are known to be in the same
1022  * allocated/unallocated state.
1023  *
1024  * 'nb_sectors' is the max value 'pnum' should be set to.
1025  */
1026 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1027         int *pnum)
1028 {
1029     int64_t n;
1030     if (!bs->drv->bdrv_is_allocated) {
1031         if (sector_num >= bs->total_sectors) {
1032             *pnum = 0;
1033             return 0;
1034         }
1035         n = bs->total_sectors - sector_num;
1036         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1037         return 1;
1038     }
1039     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1040 }
1041
1042 void bdrv_info(Monitor *mon)
1043 {
1044     BlockDriverState *bs;
1045
1046     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1047         monitor_printf(mon, "%s:", bs->device_name);
1048         monitor_printf(mon, " type=");
1049         switch(bs->type) {
1050         case BDRV_TYPE_HD:
1051             monitor_printf(mon, "hd");
1052             break;
1053         case BDRV_TYPE_CDROM:
1054             monitor_printf(mon, "cdrom");
1055             break;
1056         case BDRV_TYPE_FLOPPY:
1057             monitor_printf(mon, "floppy");
1058             break;
1059         }
1060         monitor_printf(mon, " removable=%d", bs->removable);
1061         if (bs->removable) {
1062             monitor_printf(mon, " locked=%d", bs->locked);
1063         }
1064         if (bs->drv) {
1065             monitor_printf(mon, " file=");
1066             monitor_print_filename(mon, bs->filename);
1067             if (bs->backing_file[0] != '\0') {
1068                 monitor_printf(mon, " backing_file=");
1069                 monitor_print_filename(mon, bs->backing_file);
1070             }
1071             monitor_printf(mon, " ro=%d", bs->read_only);
1072             monitor_printf(mon, " drv=%s", bs->drv->format_name);
1073             monitor_printf(mon, " encrypted=%d", bdrv_is_encrypted(bs));
1074         } else {
1075             monitor_printf(mon, " [not inserted]");
1076         }
1077         monitor_printf(mon, "\n");
1078     }
1079 }
1080
1081 /* The "info blockstats" command. */
1082 void bdrv_info_stats(Monitor *mon)
1083 {
1084     BlockDriverState *bs;
1085
1086     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1087         monitor_printf(mon, "%s:"
1088                        " rd_bytes=%" PRIu64
1089                        " wr_bytes=%" PRIu64
1090                        " rd_operations=%" PRIu64
1091                        " wr_operations=%" PRIu64
1092                        "\n",
1093                        bs->device_name,
1094                        bs->rd_bytes, bs->wr_bytes,
1095                        bs->rd_ops, bs->wr_ops);
1096     }
1097 }
1098
1099 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1100 {
1101     if (bs->backing_hd && bs->backing_hd->encrypted)
1102         return bs->backing_file;
1103     else if (bs->encrypted)
1104         return bs->filename;
1105     else
1106         return NULL;
1107 }
1108
1109 void bdrv_get_backing_filename(BlockDriverState *bs,
1110                                char *filename, int filename_size)
1111 {
1112     if (!bs->backing_hd) {
1113         pstrcpy(filename, filename_size, "");
1114     } else {
1115         pstrcpy(filename, filename_size, bs->backing_file);
1116     }
1117 }
1118
1119 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1120                           const uint8_t *buf, int nb_sectors)
1121 {
1122     BlockDriver *drv = bs->drv;
1123     if (!drv)
1124         return -ENOMEDIUM;
1125     if (!drv->bdrv_write_compressed)
1126         return -ENOTSUP;
1127     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1128 }
1129
1130 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1131 {
1132     BlockDriver *drv = bs->drv;
1133     if (!drv)
1134         return -ENOMEDIUM;
1135     if (!drv->bdrv_get_info)
1136         return -ENOTSUP;
1137     memset(bdi, 0, sizeof(*bdi));
1138     return drv->bdrv_get_info(bs, bdi);
1139 }
1140
1141 /**************************************************************/
1142 /* handling of snapshots */
1143
1144 int bdrv_snapshot_create(BlockDriverState *bs,
1145                          QEMUSnapshotInfo *sn_info)
1146 {
1147     BlockDriver *drv = bs->drv;
1148     if (!drv)
1149         return -ENOMEDIUM;
1150     if (!drv->bdrv_snapshot_create)
1151         return -ENOTSUP;
1152     return drv->bdrv_snapshot_create(bs, sn_info);
1153 }
1154
1155 int bdrv_snapshot_goto(BlockDriverState *bs,
1156                        const char *snapshot_id)
1157 {
1158     BlockDriver *drv = bs->drv;
1159     if (!drv)
1160         return -ENOMEDIUM;
1161     if (!drv->bdrv_snapshot_goto)
1162         return -ENOTSUP;
1163     return drv->bdrv_snapshot_goto(bs, snapshot_id);
1164 }
1165
1166 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1167 {
1168     BlockDriver *drv = bs->drv;
1169     if (!drv)
1170         return -ENOMEDIUM;
1171     if (!drv->bdrv_snapshot_delete)
1172         return -ENOTSUP;
1173     return drv->bdrv_snapshot_delete(bs, snapshot_id);
1174 }
1175
1176 int bdrv_snapshot_list(BlockDriverState *bs,
1177                        QEMUSnapshotInfo **psn_info)
1178 {
1179     BlockDriver *drv = bs->drv;
1180     if (!drv)
1181         return -ENOMEDIUM;
1182     if (!drv->bdrv_snapshot_list)
1183         return -ENOTSUP;
1184     return drv->bdrv_snapshot_list(bs, psn_info);
1185 }
1186
1187 #define NB_SUFFIXES 4
1188
1189 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1190 {
1191     static const char suffixes[NB_SUFFIXES] = "KMGT";
1192     int64_t base;
1193     int i;
1194
1195     if (size <= 999) {
1196         snprintf(buf, buf_size, "%" PRId64, size);
1197     } else {
1198         base = 1024;
1199         for(i = 0; i < NB_SUFFIXES; i++) {
1200             if (size < (10 * base)) {
1201                 snprintf(buf, buf_size, "%0.1f%c",
1202                          (double)size / base,
1203                          suffixes[i]);
1204                 break;
1205             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1206                 snprintf(buf, buf_size, "%" PRId64 "%c",
1207                          ((size + (base >> 1)) / base),
1208                          suffixes[i]);
1209                 break;
1210             }
1211             base = base * 1024;
1212         }
1213     }
1214     return buf;
1215 }
1216
1217 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1218 {
1219     char buf1[128], date_buf[128], clock_buf[128];
1220 #ifdef _WIN32
1221     struct tm *ptm;
1222 #else
1223     struct tm tm;
1224 #endif
1225     time_t ti;
1226     int64_t secs;
1227
1228     if (!sn) {
1229         snprintf(buf, buf_size,
1230                  "%-10s%-20s%7s%20s%15s",
1231                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1232     } else {
1233         ti = sn->date_sec;
1234 #ifdef _WIN32
1235         ptm = localtime(&ti);
1236         strftime(date_buf, sizeof(date_buf),
1237                  "%Y-%m-%d %H:%M:%S", ptm);
1238 #else
1239         localtime_r(&ti, &tm);
1240         strftime(date_buf, sizeof(date_buf),
1241                  "%Y-%m-%d %H:%M:%S", &tm);
1242 #endif
1243         secs = sn->vm_clock_nsec / 1000000000;
1244         snprintf(clock_buf, sizeof(clock_buf),
1245                  "%02d:%02d:%02d.%03d",
1246                  (int)(secs / 3600),
1247                  (int)((secs / 60) % 60),
1248                  (int)(secs % 60),
1249                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1250         snprintf(buf, buf_size,
1251                  "%-10s%-20s%7s%20s%15s",
1252                  sn->id_str, sn->name,
1253                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1254                  date_buf,
1255                  clock_buf);
1256     }
1257     return buf;
1258 }
1259
1260
1261 /**************************************************************/
1262 /* async I/Os */
1263
1264 typedef struct VectorTranslationState {
1265     QEMUIOVector *iov;
1266     uint8_t *bounce;
1267     int is_write;
1268     BlockDriverAIOCB *aiocb;
1269     BlockDriverAIOCB *this_aiocb;
1270 } VectorTranslationState;
1271
1272 static void bdrv_aio_rw_vector_cb(void *opaque, int ret)
1273 {
1274     VectorTranslationState *s = opaque;
1275
1276     if (!s->is_write) {
1277         qemu_iovec_from_buffer(s->iov, s->bounce, s->iov->size);
1278     }
1279     qemu_vfree(s->bounce);
1280     s->this_aiocb->cb(s->this_aiocb->opaque, ret);
1281     qemu_aio_release(s->this_aiocb);
1282 }
1283
1284 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1285                                             int64_t sector_num,
1286                                             QEMUIOVector *iov,
1287                                             int nb_sectors,
1288                                             BlockDriverCompletionFunc *cb,
1289                                             void *opaque,
1290                                             int is_write)
1291
1292 {
1293     VectorTranslationState *s = qemu_mallocz(sizeof(*s));
1294     BlockDriverAIOCB *aiocb = qemu_aio_get(bs, cb, opaque);
1295
1296     s->this_aiocb = aiocb;
1297     s->iov = iov;
1298     s->bounce = qemu_memalign(512, nb_sectors * 512);
1299     s->is_write = is_write;
1300     if (is_write) {
1301         qemu_iovec_to_buffer(s->iov, s->bounce);
1302         s->aiocb = bdrv_aio_write(bs, sector_num, s->bounce, nb_sectors,
1303                                   bdrv_aio_rw_vector_cb, s);
1304     } else {
1305         s->aiocb = bdrv_aio_read(bs, sector_num, s->bounce, nb_sectors,
1306                                  bdrv_aio_rw_vector_cb, s);
1307     }
1308     return aiocb;
1309 }
1310
1311 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1312                                  QEMUIOVector *iov, int nb_sectors,
1313                                  BlockDriverCompletionFunc *cb, void *opaque)
1314 {
1315     if (bdrv_check_request(bs, sector_num, nb_sectors))
1316         return NULL;
1317
1318     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
1319                               cb, opaque, 0);
1320 }
1321
1322 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1323                                   QEMUIOVector *iov, int nb_sectors,
1324                                   BlockDriverCompletionFunc *cb, void *opaque)
1325 {
1326     if (bdrv_check_request(bs, sector_num, nb_sectors))
1327         return NULL;
1328
1329     return bdrv_aio_rw_vector(bs, sector_num, iov, nb_sectors,
1330                               cb, opaque, 1);
1331 }
1332
1333 BlockDriverAIOCB *bdrv_aio_read(BlockDriverState *bs, int64_t sector_num,
1334                                 uint8_t *buf, int nb_sectors,
1335                                 BlockDriverCompletionFunc *cb, void *opaque)
1336 {
1337     BlockDriver *drv = bs->drv;
1338     BlockDriverAIOCB *ret;
1339
1340     if (!drv)
1341         return NULL;
1342     if (bdrv_check_request(bs, sector_num, nb_sectors))
1343         return NULL;
1344
1345     ret = drv->bdrv_aio_read(bs, sector_num, buf, nb_sectors, cb, opaque);
1346
1347     if (ret) {
1348         /* Update stats even though technically transfer has not happened. */
1349         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1350         bs->rd_ops ++;
1351     }
1352
1353     return ret;
1354 }
1355
1356 BlockDriverAIOCB *bdrv_aio_write(BlockDriverState *bs, int64_t sector_num,
1357                                  const uint8_t *buf, int nb_sectors,
1358                                  BlockDriverCompletionFunc *cb, void *opaque)
1359 {
1360     BlockDriver *drv = bs->drv;
1361     BlockDriverAIOCB *ret;
1362
1363     if (!drv)
1364         return NULL;
1365     if (bs->read_only)
1366         return NULL;
1367     if (bdrv_check_request(bs, sector_num, nb_sectors))
1368         return NULL;
1369
1370     ret = drv->bdrv_aio_write(bs, sector_num, buf, nb_sectors, cb, opaque);
1371
1372     if (ret) {
1373         /* Update stats even though technically transfer has not happened. */
1374         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1375         bs->wr_ops ++;
1376     }
1377
1378     return ret;
1379 }
1380
1381 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1382 {
1383     BlockDriver *drv = acb->bs->drv;
1384
1385     if (acb->cb == bdrv_aio_rw_vector_cb) {
1386         VectorTranslationState *s = acb->opaque;
1387         acb = s->aiocb;
1388     }
1389
1390     drv->bdrv_aio_cancel(acb);
1391 }
1392
1393
1394 /**************************************************************/
1395 /* async block device emulation */
1396
1397 static void bdrv_aio_bh_cb(void *opaque)
1398 {
1399     BlockDriverAIOCBSync *acb = opaque;
1400     acb->common.cb(acb->common.opaque, acb->ret);
1401     qemu_aio_release(acb);
1402 }
1403
1404 static BlockDriverAIOCB *bdrv_aio_read_em(BlockDriverState *bs,
1405         int64_t sector_num, uint8_t *buf, int nb_sectors,
1406         BlockDriverCompletionFunc *cb, void *opaque)
1407 {
1408     BlockDriverAIOCBSync *acb;
1409     int ret;
1410
1411     acb = qemu_aio_get(bs, cb, opaque);
1412     if (!acb->bh)
1413         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1414     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
1415     acb->ret = ret;
1416     qemu_bh_schedule(acb->bh);
1417     return &acb->common;
1418 }
1419
1420 static BlockDriverAIOCB *bdrv_aio_write_em(BlockDriverState *bs,
1421         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1422         BlockDriverCompletionFunc *cb, void *opaque)
1423 {
1424     BlockDriverAIOCBSync *acb;
1425     int ret;
1426
1427     acb = qemu_aio_get(bs, cb, opaque);
1428     if (!acb->bh)
1429         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1430     ret = bdrv_write(bs, sector_num, buf, nb_sectors);
1431     acb->ret = ret;
1432     qemu_bh_schedule(acb->bh);
1433     return &acb->common;
1434 }
1435
1436 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1437 {
1438     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1439     qemu_bh_cancel(acb->bh);
1440     qemu_aio_release(acb);
1441 }
1442
1443 /**************************************************************/
1444 /* sync block device emulation */
1445
1446 static void bdrv_rw_em_cb(void *opaque, int ret)
1447 {
1448     *(int *)opaque = ret;
1449 }
1450
1451 #define NOT_DONE 0x7fffffff
1452
1453 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1454                         uint8_t *buf, int nb_sectors)
1455 {
1456     int async_ret;
1457     BlockDriverAIOCB *acb;
1458
1459     async_ret = NOT_DONE;
1460     acb = bdrv_aio_read(bs, sector_num, buf, nb_sectors,
1461                         bdrv_rw_em_cb, &async_ret);
1462     if (acb == NULL)
1463         return -1;
1464
1465     while (async_ret == NOT_DONE) {
1466         qemu_aio_wait();
1467     }
1468
1469     return async_ret;
1470 }
1471
1472 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1473                          const uint8_t *buf, int nb_sectors)
1474 {
1475     int async_ret;
1476     BlockDriverAIOCB *acb;
1477
1478     async_ret = NOT_DONE;
1479     acb = bdrv_aio_write(bs, sector_num, buf, nb_sectors,
1480                          bdrv_rw_em_cb, &async_ret);
1481     if (acb == NULL)
1482         return -1;
1483     while (async_ret == NOT_DONE) {
1484         qemu_aio_wait();
1485     }
1486     return async_ret;
1487 }
1488
1489 void bdrv_init(void)
1490 {
1491     bdrv_register(&bdrv_raw);
1492     bdrv_register(&bdrv_host_device);
1493 #ifndef _WIN32
1494     bdrv_register(&bdrv_cow);
1495 #endif
1496     bdrv_register(&bdrv_qcow);
1497     bdrv_register(&bdrv_vmdk);
1498     bdrv_register(&bdrv_cloop);
1499     bdrv_register(&bdrv_dmg);
1500     bdrv_register(&bdrv_bochs);
1501     bdrv_register(&bdrv_vpc);
1502     bdrv_register(&bdrv_vvfat);
1503     bdrv_register(&bdrv_qcow2);
1504     bdrv_register(&bdrv_parallels);
1505     bdrv_register(&bdrv_nbd);
1506 }
1507
1508 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1509                    void *opaque)
1510 {
1511     BlockDriver *drv;
1512     BlockDriverAIOCB *acb;
1513
1514     drv = bs->drv;
1515     if (drv->free_aiocb) {
1516         acb = drv->free_aiocb;
1517         drv->free_aiocb = acb->next;
1518     } else {
1519         acb = qemu_mallocz(drv->aiocb_size);
1520     }
1521     acb->bs = bs;
1522     acb->cb = cb;
1523     acb->opaque = opaque;
1524     return acb;
1525 }
1526
1527 void qemu_aio_release(void *p)
1528 {
1529     BlockDriverAIOCB *acb = p;
1530     BlockDriver *drv = acb->bs->drv;
1531     acb->next = drv->free_aiocb;
1532     drv->free_aiocb = acb;
1533 }
1534
1535 /**************************************************************/
1536 /* removable device support */
1537
1538 /**
1539  * Return TRUE if the media is present
1540  */
1541 int bdrv_is_inserted(BlockDriverState *bs)
1542 {
1543     BlockDriver *drv = bs->drv;
1544     int ret;
1545     if (!drv)
1546         return 0;
1547     if (!drv->bdrv_is_inserted)
1548         return 1;
1549     ret = drv->bdrv_is_inserted(bs);
1550     return ret;
1551 }
1552
1553 /**
1554  * Return TRUE if the media changed since the last call to this
1555  * function. It is currently only used for floppy disks
1556  */
1557 int bdrv_media_changed(BlockDriverState *bs)
1558 {
1559     BlockDriver *drv = bs->drv;
1560     int ret;
1561
1562     if (!drv || !drv->bdrv_media_changed)
1563         ret = -ENOTSUP;
1564     else
1565         ret = drv->bdrv_media_changed(bs);
1566     if (ret == -ENOTSUP)
1567         ret = bs->media_changed;
1568     bs->media_changed = 0;
1569     return ret;
1570 }
1571
1572 /**
1573  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1574  */
1575 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1576 {
1577     BlockDriver *drv = bs->drv;
1578     int ret;
1579
1580     if (!drv || !drv->bdrv_eject) {
1581         ret = -ENOTSUP;
1582     } else {
1583         ret = drv->bdrv_eject(bs, eject_flag);
1584     }
1585     if (ret == -ENOTSUP) {
1586         if (eject_flag)
1587             bdrv_close(bs);
1588     }
1589 }
1590
1591 int bdrv_is_locked(BlockDriverState *bs)
1592 {
1593     return bs->locked;
1594 }
1595
1596 /**
1597  * Lock or unlock the media (if it is locked, the user won't be able
1598  * to eject it manually).
1599  */
1600 void bdrv_set_locked(BlockDriverState *bs, int locked)
1601 {
1602     BlockDriver *drv = bs->drv;
1603
1604     bs->locked = locked;
1605     if (drv && drv->bdrv_set_locked) {
1606         drv->bdrv_set_locked(bs, locked);
1607     }
1608 }
1609
1610 /* needed for generic scsi interface */
1611
1612 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1613 {
1614     BlockDriver *drv = bs->drv;
1615
1616     if (drv && drv->bdrv_ioctl)
1617         return drv->bdrv_ioctl(bs, req, buf);
1618     return -ENOTSUP;
1619 }
1620
1621 int bdrv_sg_send_command(BlockDriverState *bs, void *buf, int count)
1622 {
1623     return bs->drv->bdrv_sg_send_command(bs, buf, count);
1624 }
1625
1626 int bdrv_sg_recv_response(BlockDriverState *bs, void *buf, int count)
1627 {
1628     return bs->drv->bdrv_sg_recv_response(bs, buf, count);
1629 }
1630
1631 BlockDriverAIOCB *bdrv_sg_aio_read(BlockDriverState *bs, void *buf, int count,
1632                                    BlockDriverCompletionFunc *cb, void *opaque)
1633 {
1634     return bs->drv->bdrv_sg_aio_read(bs, buf, count, cb, opaque);
1635 }
1636
1637 BlockDriverAIOCB *bdrv_sg_aio_write(BlockDriverState *bs, void *buf, int count,
1638                                     BlockDriverCompletionFunc *cb, void *opaque)
1639 {
1640     return bs->drv->bdrv_sg_aio_write(bs, buf, count, cb, opaque);
1641 }