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 "module.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     /* vector translation state */
  56     QEMUIOVector *qiov;
  57     uint8_t *bounce;
  58     int is_write;
  59 } BlockDriverAIOCBSync;
  60
  61 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
  62         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
  63         BlockDriverCompletionFunc *cb, void *opaque);
  64 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
  65         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
  66         BlockDriverCompletionFunc *cb, void *opaque);
  67 static void bdrv_aio_cancel_em(BlockDriverAIOCB *acb);
  68 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  69                         uint8_t *buf, int nb_sectors);
  70 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  71                          const uint8_t *buf, int nb_sectors);
  72
  73 BlockDriverState *bdrv_first;
  74
  75 static BlockDriver *first_drv;
  76
  77 int path_is_absolute(const char *path)
  78 {
  79     const char *p;
  80 #ifdef _WIN32
  81     /* specific case for names like: "\\.\d:" */
  82     if (*path == '/' || *path == '\\')
  83         return 1;
  84 #endif
  85     p = strchr(path, ':');
  86     if (p)
  87         p++;
  88     else
  89         p = path;
  90 #ifdef _WIN32
  91     return (*p == '/' || *p == '\\');
  92 #else
  93     return (*p == '/');
  94 #endif
  95 }
  96
  97 /* if filename is absolute, just copy it to dest. Otherwise, build a
  98    path to it by considering it is relative to base_path. URL are
  99    supported. */
 100 void path_combine(char *dest, int dest_size,
 101                   const char *base_path,
 102                   const char *filename)
 103 {
 104     const char *p, *p1;
 105     int len;
 106
 107     if (dest_size <= 0)
 108         return;
 109     if (path_is_absolute(filename)) {
 110         pstrcpy(dest, dest_size, filename);
 111     } else {
 112         p = strchr(base_path, ':');
 113         if (p)
 114             p++;
 115         else
 116             p = base_path;
 117         p1 = strrchr(base_path, '/');
 118 #ifdef _WIN32
 119         {
 120             const char *p2;
 121             p2 = strrchr(base_path, '\\');
 122             if (!p1 || p2 > p1)
 123                 p1 = p2;
 124         }
 125 #endif
 126         if (p1)
 127             p1++;
 128         else
 129             p1 = base_path;
 130         if (p1 > p)
 131             p = p1;
 132         len = p - base_path;
 133         if (len > dest_size - 1)
 134             len = dest_size - 1;
 135         memcpy(dest, base_path, len);
 136         dest[len] = '\0';
 137         pstrcat(dest, dest_size, filename);
 138     }
 139 }
 140
 141
 142 void bdrv_register(BlockDriver *bdrv)
 143 {
 144     if (!bdrv->bdrv_aio_readv) {
 145         /* add AIO emulation layer */
 146         bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
 147         bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
 148         bdrv->bdrv_aio_cancel = bdrv_aio_cancel_em;
 149         bdrv->aiocb_size = sizeof(BlockDriverAIOCBSync);
 150     } else if (!bdrv->bdrv_read) {
 151         /* add synchronous IO emulation layer */
 152         bdrv->bdrv_read = bdrv_read_em;
 153         bdrv->bdrv_write = bdrv_write_em;
 154     }
 155     aio_pool_init(&bdrv->aio_pool, bdrv->aiocb_size, bdrv->bdrv_aio_cancel);
 156     bdrv->next = first_drv;
 157     first_drv = bdrv;
 158 }
 159
 160 /* create a new block device (by default it is empty) */
 161 BlockDriverState *bdrv_new(const char *device_name)
 162 {
 163     BlockDriverState **pbs, *bs;
 164
 165     bs = qemu_mallocz(sizeof(BlockDriverState));
 166     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 167     if (device_name[0] != '\0') {
 168         /* insert at the end */
 169         pbs = &bdrv_first;
 170         while (*pbs != NULL)
 171             pbs = &(*pbs)->next;
 172         *pbs = bs;
 173     }
 174     return bs;
 175 }
 176
 177 BlockDriver *bdrv_find_format(const char *format_name)
 178 {
 179     BlockDriver *drv1;
 180     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 181         if (!strcmp(drv1->format_name, format_name))
 182             return drv1;
 183     }
 184     return NULL;
 185 }
 186
 187 int bdrv_create2(BlockDriver *drv,
 188                 const char *filename, int64_t size_in_sectors,
 189                 const char *backing_file, const char *backing_format,
 190                 int flags)
 191 {
 192     QEMUOptionParameter *options;
 193
 194     options = parse_option_parameters("", drv->create_options, NULL);
 195
 196     // Process flags
 197     if (flags & ~(BLOCK_FLAG_ENCRYPT | BLOCK_FLAG_COMPAT6 | BLOCK_FLAG_COMPRESS)) {
 198         return -ENOTSUP;
 199     }
 200
 201     if (flags & BLOCK_FLAG_ENCRYPT) {
 202         set_option_parameter_int(options, BLOCK_OPT_ENCRYPT, 1);
 203     }
 204     if (flags & BLOCK_FLAG_COMPAT6) {
 205         set_option_parameter_int(options, BLOCK_OPT_COMPAT6, 1);
 206     }
 207
 208     // Add size to options
 209     set_option_parameter_int(options, BLOCK_OPT_SIZE, size_in_sectors * 512);
 210
 211     // Backing files
 212     if ((backing_file != NULL && set_option_parameter(options,
 213             BLOCK_OPT_BACKING_FILE, backing_file))
 214         || (backing_format != NULL && set_option_parameter(options,
 215             BLOCK_OPT_BACKING_FMT, backing_format)))
 216     {
 217         return -ENOTSUP;
 218     }
 219
 220     return bdrv_create(drv, filename, options);
 221 }
 222
 223 int bdrv_create(BlockDriver *drv, const char* filename,
 224     QEMUOptionParameter *options)
 225 {
 226     if (!drv->bdrv_create)
 227         return -ENOTSUP;
 228
 229     return drv->bdrv_create(filename, options);
 230 }
 231
 232 #ifdef _WIN32
 233 void get_tmp_filename(char *filename, int size)
 234 {
 235     char temp_dir[MAX_PATH];
 236
 237     GetTempPath(MAX_PATH, temp_dir);
 238     GetTempFileName(temp_dir, "qem", 0, filename);
 239 }
 240 #else
 241 void get_tmp_filename(char *filename, int size)
 242 {
 243     int fd;
 244     const char *tmpdir;
 245     /* XXX: race condition possible */
 246     tmpdir = getenv("TMPDIR");
 247     if (!tmpdir)
 248         tmpdir = "/tmp";
 249     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 250     fd = mkstemp(filename);
 251     close(fd);
 252 }
 253 #endif
 254
 255 #ifdef _WIN32
 256 static int is_windows_drive_prefix(const char *filename)
 257 {
 258     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 259              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 260             filename[1] == ':');
 261 }
 262
 263 static int is_windows_drive(const char *filename)
 264 {
 265     if (is_windows_drive_prefix(filename) &&
 266         filename[2] == '\0')
 267         return 1;
 268     if (strstart(filename, "\\\\.\\", NULL) ||
 269         strstart(filename, "//./", NULL))
 270         return 1;
 271     return 0;
 272 }
 273 #endif
 274
 275 static BlockDriver *find_protocol(const char *filename)
 276 {
 277     BlockDriver *drv1;
 278     char protocol[128];
 279     int len;
 280     const char *p;
 281
 282 #ifdef _WIN32
 283     if (is_windows_drive(filename) ||
 284         is_windows_drive_prefix(filename))
 285         return bdrv_find_format("raw");
 286 #endif
 287     p = strchr(filename, ':');
 288     if (!p)
 289         return bdrv_find_format("raw");
 290     len = p - filename;
 291     if (len > sizeof(protocol) - 1)
 292         len = sizeof(protocol) - 1;
 293     memcpy(protocol, filename, len);
 294     protocol[len] = '\0';
 295     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 296         if (drv1->protocol_name &&
 297             !strcmp(drv1->protocol_name, protocol))
 298             return drv1;
 299     }
 300     return NULL;
 301 }
 302
 303 /* XXX: force raw format if block or character device ? It would
 304    simplify the BSD case */
 305 static BlockDriver *find_image_format(const char *filename)
 306 {
 307     int ret, score, score_max;
 308     BlockDriver *drv1, *drv;
 309     uint8_t buf[2048];
 310     BlockDriverState *bs;
 311
 312     /* detect host devices. By convention, /dev/cdrom[N] is always
 313        recognized as a host CDROM */
 314     if (strstart(filename, "/dev/cdrom", NULL))
 315         return bdrv_find_format("host_device");
 316 #ifdef _WIN32
 317     if (is_windows_drive(filename))
 318         return bdrv_find_format("host_device");
 319 #else
 320     {
 321         struct stat st;
 322         if (stat(filename, &st) >= 0 &&
 323             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 324             return bdrv_find_format("host_device");
 325         }
 326     }
 327 #endif
 328
 329     drv = find_protocol(filename);
 330     /* no need to test disk image formats for vvfat */
 331     if (drv && strcmp(drv->format_name, "vvfat") == 0)
 332         return drv;
 333
 334     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 335     if (ret < 0)
 336         return NULL;
 337     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 338     bdrv_delete(bs);
 339     if (ret < 0) {
 340         return NULL;
 341     }
 342
 343     score_max = 0;
 344     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 345         if (drv1->bdrv_probe) {
 346             score = drv1->bdrv_probe(buf, ret, filename);
 347             if (score > score_max) {
 348                 score_max = score;
 349                 drv = drv1;
 350             }
 351         }
 352     }
 353     return drv;
 354 }
 355
 356 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 357 {
 358     BlockDriverState *bs;
 359     int ret;
 360
 361     bs = bdrv_new("");
 362     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 363     if (ret < 0) {
 364         bdrv_delete(bs);
 365         return ret;
 366     }
 367     bs->growable = 1;
 368     *pbs = bs;
 369     return 0;
 370 }
 371
 372 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 373 {
 374     return bdrv_open2(bs, filename, flags, NULL);
 375 }
 376
 377 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 378                BlockDriver *drv)
 379 {
 380     int ret, open_flags;
 381     char tmp_filename[PATH_MAX];
 382     char backing_filename[PATH_MAX];
 383
 384     bs->read_only = 0;
 385     bs->is_temporary = 0;
 386     bs->encrypted = 0;
 387     bs->valid_key = 0;
 388     /* buffer_alignment defaulted to 512, drivers can change this value */
 389     bs->buffer_alignment = 512;
 390
 391     if (flags & BDRV_O_SNAPSHOT) {
 392         BlockDriverState *bs1;
 393         int64_t total_size;
 394         int is_protocol = 0;
 395
 396         /* if snapshot, we create a temporary backing file and open it
 397            instead of opening 'filename' directly */
 398
 399         /* if there is a backing file, use it */
 400         bs1 = bdrv_new("");
 401         ret = bdrv_open2(bs1, filename, 0, drv);
 402         if (ret < 0) {
 403             bdrv_delete(bs1);
 404             return ret;
 405         }
 406         total_size = bdrv_getlength(bs1) >> SECTOR_BITS;
 407
 408         if (bs1->drv && bs1->drv->protocol_name)
 409             is_protocol = 1;
 410
 411         bdrv_delete(bs1);
 412
 413         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 414
 415         /* Real path is meaningless for protocols */
 416         if (is_protocol)
 417             snprintf(backing_filename, sizeof(backing_filename),
 418                      "%s", filename);
 419         else
 420             realpath(filename, backing_filename);
 421
 422         ret = bdrv_create2(bdrv_find_format("qcow2"), tmp_filename,
 423                            total_size, backing_filename,
 424                            (drv ? drv->format_name : NULL), 0);
 425         if (ret < 0) {
 426             return ret;
 427         }
 428         filename = tmp_filename;
 429         drv = bdrv_find_format("qcow2");
 430         bs->is_temporary = 1;
 431     }
 432
 433     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 434     if (flags & BDRV_O_FILE) {
 435         drv = find_protocol(filename);
 436     } else if (!drv) {
 437         drv = find_image_format(filename);
 438     }
 439     if (!drv) {
 440         ret = -ENOENT;
 441         goto unlink_and_fail;
 442     }
 443     bs->drv = drv;
 444     bs->opaque = qemu_mallocz(drv->instance_size);
 445     /* Note: for compatibility, we open disk image files as RDWR, and
 446        RDONLY as fallback */
 447     if (!(flags & BDRV_O_FILE))
 448         open_flags = BDRV_O_RDWR | (flags & BDRV_O_CACHE_MASK);
 449     else
 450         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 451     ret = drv->bdrv_open(bs, filename, open_flags);
 452     if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
 453         ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
 454         bs->read_only = 1;
 455     }
 456     if (ret < 0) {
 457         qemu_free(bs->opaque);
 458         bs->opaque = NULL;
 459         bs->drv = NULL;
 460     unlink_and_fail:
 461         if (bs->is_temporary)
 462             unlink(filename);
 463         return ret;
 464     }
 465     if (drv->bdrv_getlength) {
 466         bs->total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 467     }
 468 #ifndef _WIN32
 469     if (bs->is_temporary) {
 470         unlink(filename);
 471     }
 472 #endif
 473     if (bs->backing_file[0] != '\0') {
 474         /* if there is a backing file, use it */
 475         BlockDriver *back_drv = NULL;
 476         bs->backing_hd = bdrv_new("");
 477         path_combine(backing_filename, sizeof(backing_filename),
 478                      filename, bs->backing_file);
 479         if (bs->backing_format[0] != '\0')
 480             back_drv = bdrv_find_format(bs->backing_format);
 481         ret = bdrv_open2(bs->backing_hd, backing_filename, open_flags,
 482                          back_drv);
 483         if (ret < 0) {
 484             bdrv_close(bs);
 485             return ret;
 486         }
 487     }
 488
 489     if (!bdrv_key_required(bs)) {
 490         /* call the change callback */
 491         bs->media_changed = 1;
 492         if (bs->change_cb)
 493             bs->change_cb(bs->change_opaque);
 494     }
 495     return 0;
 496 }
 497
 498 void bdrv_close(BlockDriverState *bs)
 499 {
 500     if (bs->drv) {
 501         if (bs->backing_hd)
 502             bdrv_delete(bs->backing_hd);
 503         bs->drv->bdrv_close(bs);
 504         qemu_free(bs->opaque);
 505 #ifdef _WIN32
 506         if (bs->is_temporary) {
 507             unlink(bs->filename);
 508         }
 509 #endif
 510         bs->opaque = NULL;
 511         bs->drv = NULL;
 512
 513         /* call the change callback */
 514         bs->media_changed = 1;
 515         if (bs->change_cb)
 516             bs->change_cb(bs->change_opaque);
 517     }
 518 }
 519
 520 void bdrv_delete(BlockDriverState *bs)
 521 {
 522     BlockDriverState **pbs;
 523
 524     pbs = &bdrv_first;
 525     while (*pbs != bs && *pbs != NULL)
 526         pbs = &(*pbs)->next;
 527     if (*pbs == bs)
 528         *pbs = bs->next;
 529
 530     bdrv_close(bs);
 531     qemu_free(bs);
 532 }
 533
 534 /*
 535  * Run consistency checks on an image
 536  *
 537  * Returns the number of errors or -errno when an internal error occurs
 538  */
 539 int bdrv_check(BlockDriverState *bs)
 540 {
 541     if (bs->drv->bdrv_check == NULL) {
 542         return -ENOTSUP;
 543     }
 544
 545     return bs->drv->bdrv_check(bs);
 546 }
 547
 548 /* commit COW file into the raw image */
 549 int bdrv_commit(BlockDriverState *bs)
 550 {
 551     BlockDriver *drv = bs->drv;
 552     int64_t i, total_sectors;
 553     int n, j;
 554     unsigned char sector[512];
 555
 556     if (!drv)
 557         return -ENOMEDIUM;
 558
 559     if (bs->read_only) {
 560         return -EACCES;
 561     }
 562
 563     if (!bs->backing_hd) {
 564         return -ENOTSUP;
 565     }
 566
 567     total_sectors = bdrv_getlength(bs) >> SECTOR_BITS;
 568     for (i = 0; i < total_sectors;) {
 569         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 570             for(j = 0; j < n; j++) {
 571                 if (bdrv_read(bs, i, sector, 1) != 0) {
 572                     return -EIO;
 573                 }
 574
 575                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 576                     return -EIO;
 577                 }
 578                 i++;
 579             }
 580         } else {
 581             i += n;
 582         }
 583     }
 584
 585     if (drv->bdrv_make_empty)
 586         return drv->bdrv_make_empty(bs);
 587
 588     return 0;
 589 }
 590
 591 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
 592                                    size_t size)
 593 {
 594     int64_t len;
 595
 596     if (!bdrv_is_inserted(bs))
 597         return -ENOMEDIUM;
 598
 599     if (bs->growable)
 600         return 0;
 601
 602     len = bdrv_getlength(bs);
 603
 604     if (offset < 0)
 605         return -EIO;
 606
 607     if ((offset > len) || (len - offset < size))
 608         return -EIO;
 609
 610     return 0;
 611 }
 612
 613 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
 614                               int nb_sectors)
 615 {
 616     return bdrv_check_byte_request(bs, sector_num * 512, nb_sectors * 512);
 617 }
 618
 619 /* return < 0 if error. See bdrv_write() for the return codes */
 620 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 621               uint8_t *buf, int nb_sectors)
 622 {
 623     BlockDriver *drv = bs->drv;
 624
 625     if (!drv)
 626         return -ENOMEDIUM;
 627     if (bdrv_check_request(bs, sector_num, nb_sectors))
 628         return -EIO;
 629
 630     return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 631 }
 632
 633 /* Return < 0 if error. Important errors are:
 634   -EIO         generic I/O error (may happen for all errors)
 635   -ENOMEDIUM   No media inserted.
 636   -EINVAL      Invalid sector number or nb_sectors
 637   -EACCES      Trying to write a read-only device
 638 */
 639 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 640                const uint8_t *buf, int nb_sectors)
 641 {
 642     BlockDriver *drv = bs->drv;
 643     if (!bs->drv)
 644         return -ENOMEDIUM;
 645     if (bs->read_only)
 646         return -EACCES;
 647     if (bdrv_check_request(bs, sector_num, nb_sectors))
 648         return -EIO;
 649
 650     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 651 }
 652
 653 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 654                void *buf, int count1)
 655 {
 656     uint8_t tmp_buf[SECTOR_SIZE];
 657     int len, nb_sectors, count;
 658     int64_t sector_num;
 659
 660     count = count1;
 661     /* first read to align to sector start */
 662     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 663     if (len > count)
 664         len = count;
 665     sector_num = offset >> SECTOR_BITS;
 666     if (len > 0) {
 667         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 668             return -EIO;
 669         memcpy(buf, tmp_buf + (offset & (SECTOR_SIZE - 1)), len);
 670         count -= len;
 671         if (count == 0)
 672             return count1;
 673         sector_num++;
 674         buf += len;
 675     }
 676
 677     /* read the sectors "in place" */
 678     nb_sectors = count >> SECTOR_BITS;
 679     if (nb_sectors > 0) {
 680         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 681             return -EIO;
 682         sector_num += nb_sectors;
 683         len = nb_sectors << SECTOR_BITS;
 684         buf += len;
 685         count -= len;
 686     }
 687
 688     /* add data from the last sector */
 689     if (count > 0) {
 690         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 691             return -EIO;
 692         memcpy(buf, tmp_buf, count);
 693     }
 694     return count1;
 695 }
 696
 697 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 698                 const void *buf, int count1)
 699 {
 700     uint8_t tmp_buf[SECTOR_SIZE];
 701     int len, nb_sectors, count;
 702     int64_t sector_num;
 703
 704     count = count1;
 705     /* first write to align to sector start */
 706     len = (SECTOR_SIZE - offset) & (SECTOR_SIZE - 1);
 707     if (len > count)
 708         len = count;
 709     sector_num = offset >> SECTOR_BITS;
 710     if (len > 0) {
 711         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 712             return -EIO;
 713         memcpy(tmp_buf + (offset & (SECTOR_SIZE - 1)), buf, len);
 714         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 715             return -EIO;
 716         count -= len;
 717         if (count == 0)
 718             return count1;
 719         sector_num++;
 720         buf += len;
 721     }
 722
 723     /* write the sectors "in place" */
 724     nb_sectors = count >> SECTOR_BITS;
 725     if (nb_sectors > 0) {
 726         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 727             return -EIO;
 728         sector_num += nb_sectors;
 729         len = nb_sectors << SECTOR_BITS;
 730         buf += len;
 731         count -= len;
 732     }
 733
 734     /* add data from the last sector */
 735     if (count > 0) {
 736         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 737             return -EIO;
 738         memcpy(tmp_buf, buf, count);
 739         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 740             return -EIO;
 741     }
 742     return count1;
 743 }
 744
 745 /**
 746  * Truncate file to 'offset' bytes (needed only for file protocols)
 747  */
 748 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 749 {
 750     BlockDriver *drv = bs->drv;
 751     if (!drv)
 752         return -ENOMEDIUM;
 753     if (!drv->bdrv_truncate)
 754         return -ENOTSUP;
 755     return drv->bdrv_truncate(bs, offset);
 756 }
 757
 758 /**
 759  * Length of a file in bytes. Return < 0 if error or unknown.
 760  */
 761 int64_t bdrv_getlength(BlockDriverState *bs)
 762 {
 763     BlockDriver *drv = bs->drv;
 764     if (!drv)
 765         return -ENOMEDIUM;
 766     if (!drv->bdrv_getlength) {
 767         /* legacy mode */
 768         return bs->total_sectors * SECTOR_SIZE;
 769     }
 770     return drv->bdrv_getlength(bs);
 771 }
 772
 773 /* return 0 as number of sectors if no device present or error */
 774 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 775 {
 776     int64_t length;
 777     length = bdrv_getlength(bs);
 778     if (length < 0)
 779         length = 0;
 780     else
 781         length = length >> SECTOR_BITS;
 782     *nb_sectors_ptr = length;
 783 }
 784
 785 struct partition {
 786         uint8_t boot_ind;           /* 0x80 - active */
 787         uint8_t head;               /* starting head */
 788         uint8_t sector;             /* starting sector */
 789         uint8_t cyl;                /* starting cylinder */
 790         uint8_t sys_ind;            /* What partition type */
 791         uint8_t end_head;           /* end head */
 792         uint8_t end_sector;         /* end sector */
 793         uint8_t end_cyl;            /* end cylinder */
 794         uint32_t start_sect;        /* starting sector counting from 0 */
 795         uint32_t nr_sects;          /* nr of sectors in partition */
 796 } __attribute__((packed));
 797
 798 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
 799 static int guess_disk_lchs(BlockDriverState *bs,
 800                            int *pcylinders, int *pheads, int *psectors)
 801 {
 802     uint8_t buf[512];
 803     int ret, i, heads, sectors, cylinders;
 804     struct partition *p;
 805     uint32_t nr_sects;
 806     uint64_t nb_sectors;
 807
 808     bdrv_get_geometry(bs, &nb_sectors);
 809
 810     ret = bdrv_read(bs, 0, buf, 1);
 811     if (ret < 0)
 812         return -1;
 813     /* test msdos magic */
 814     if (buf[510] != 0x55 || buf[511] != 0xaa)
 815         return -1;
 816     for(i = 0; i < 4; i++) {
 817         p = ((struct partition *)(buf + 0x1be)) + i;
 818         nr_sects = le32_to_cpu(p->nr_sects);
 819         if (nr_sects && p->end_head) {
 820             /* We make the assumption that the partition terminates on
 821                a cylinder boundary */
 822             heads = p->end_head + 1;
 823             sectors = p->end_sector & 63;
 824             if (sectors == 0)
 825                 continue;
 826             cylinders = nb_sectors / (heads * sectors);
 827             if (cylinders < 1 || cylinders > 16383)
 828                 continue;
 829             *pheads = heads;
 830             *psectors = sectors;
 831             *pcylinders = cylinders;
 832 #if 0
 833             printf("guessed geometry: LCHS=%d %d %d\n",
 834                    cylinders, heads, sectors);
 835 #endif
 836             return 0;
 837         }
 838     }
 839     return -1;
 840 }
 841
 842 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
 843 {
 844     int translation, lba_detected = 0;
 845     int cylinders, heads, secs;
 846     uint64_t nb_sectors;
 847
 848     /* if a geometry hint is available, use it */
 849     bdrv_get_geometry(bs, &nb_sectors);
 850     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
 851     translation = bdrv_get_translation_hint(bs);
 852     if (cylinders != 0) {
 853         *pcyls = cylinders;
 854         *pheads = heads;
 855         *psecs = secs;
 856     } else {
 857         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
 858             if (heads > 16) {
 859                 /* if heads > 16, it means that a BIOS LBA
 860                    translation was active, so the default
 861                    hardware geometry is OK */
 862                 lba_detected = 1;
 863                 goto default_geometry;
 864             } else {
 865                 *pcyls = cylinders;
 866                 *pheads = heads;
 867                 *psecs = secs;
 868                 /* disable any translation to be in sync with
 869                    the logical geometry */
 870                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
 871                     bdrv_set_translation_hint(bs,
 872                                               BIOS_ATA_TRANSLATION_NONE);
 873                 }
 874             }
 875         } else {
 876         default_geometry:
 877             /* if no geometry, use a standard physical disk geometry */
 878             cylinders = nb_sectors / (16 * 63);
 879
 880             if (cylinders > 16383)
 881                 cylinders = 16383;
 882             else if (cylinders < 2)
 883                 cylinders = 2;
 884             *pcyls = cylinders;
 885             *pheads = 16;
 886             *psecs = 63;
 887             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
 888                 if ((*pcyls * *pheads) <= 131072) {
 889                     bdrv_set_translation_hint(bs,
 890                                               BIOS_ATA_TRANSLATION_LARGE);
 891                 } else {
 892                     bdrv_set_translation_hint(bs,
 893                                               BIOS_ATA_TRANSLATION_LBA);
 894                 }
 895             }
 896         }
 897         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
 898     }
 899 }
 900
 901 void bdrv_set_geometry_hint(BlockDriverState *bs,
 902                             int cyls, int heads, int secs)
 903 {
 904     bs->cyls = cyls;
 905     bs->heads = heads;
 906     bs->secs = secs;
 907 }
 908
 909 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 910 {
 911     bs->type = type;
 912     bs->removable = ((type == BDRV_TYPE_CDROM ||
 913                       type == BDRV_TYPE_FLOPPY));
 914 }
 915
 916 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 917 {
 918     bs->translation = translation;
 919 }
 920
 921 void bdrv_get_geometry_hint(BlockDriverState *bs,
 922                             int *pcyls, int *pheads, int *psecs)
 923 {
 924     *pcyls = bs->cyls;
 925     *pheads = bs->heads;
 926     *psecs = bs->secs;
 927 }
 928
 929 int bdrv_get_type_hint(BlockDriverState *bs)
 930 {
 931     return bs->type;
 932 }
 933
 934 int bdrv_get_translation_hint(BlockDriverState *bs)
 935 {
 936     return bs->translation;
 937 }
 938
 939 int bdrv_is_removable(BlockDriverState *bs)
 940 {
 941     return bs->removable;
 942 }
 943
 944 int bdrv_is_read_only(BlockDriverState *bs)
 945 {
 946     return bs->read_only;
 947 }
 948
 949 int bdrv_is_sg(BlockDriverState *bs)
 950 {
 951     return bs->sg;
 952 }
 953
 954 /* XXX: no longer used */
 955 void bdrv_set_change_cb(BlockDriverState *bs,
 956                         void (*change_cb)(void *opaque), void *opaque)
 957 {
 958     bs->change_cb = change_cb;
 959     bs->change_opaque = opaque;
 960 }
 961
 962 int bdrv_is_encrypted(BlockDriverState *bs)
 963 {
 964     if (bs->backing_hd && bs->backing_hd->encrypted)
 965         return 1;
 966     return bs->encrypted;
 967 }
 968
 969 int bdrv_key_required(BlockDriverState *bs)
 970 {
 971     BlockDriverState *backing_hd = bs->backing_hd;
 972
 973     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
 974         return 1;
 975     return (bs->encrypted && !bs->valid_key);
 976 }
 977
 978 int bdrv_set_key(BlockDriverState *bs, const char *key)
 979 {
 980     int ret;
 981     if (bs->backing_hd && bs->backing_hd->encrypted) {
 982         ret = bdrv_set_key(bs->backing_hd, key);
 983         if (ret < 0)
 984             return ret;
 985         if (!bs->encrypted)
 986             return 0;
 987     }
 988     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
 989         return -1;
 990     ret = bs->drv->bdrv_set_key(bs, key);
 991     if (ret < 0) {
 992         bs->valid_key = 0;
 993     } else if (!bs->valid_key) {
 994         bs->valid_key = 1;
 995         /* call the change callback now, we skipped it on open */
 996         bs->media_changed = 1;
 997         if (bs->change_cb)
 998             bs->change_cb(bs->change_opaque);
 999     }
1000     return ret;
1001 }
1002
1003 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1004 {
1005     if (!bs->drv) {
1006         buf[0] = '\0';
1007     } else {
1008         pstrcpy(buf, buf_size, bs->drv->format_name);
1009     }
1010 }
1011
1012 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1013                          void *opaque)
1014 {
1015     BlockDriver *drv;
1016
1017     for (drv = first_drv; drv != NULL; drv = drv->next) {
1018         it(opaque, drv->format_name);
1019     }
1020 }
1021
1022 BlockDriverState *bdrv_find(const char *name)
1023 {
1024     BlockDriverState *bs;
1025
1026     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1027         if (!strcmp(name, bs->device_name))
1028             return bs;
1029     }
1030     return NULL;
1031 }
1032
1033 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1034 {
1035     BlockDriverState *bs;
1036
1037     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1038         it(opaque, bs);
1039     }
1040 }
1041
1042 const char *bdrv_get_device_name(BlockDriverState *bs)
1043 {
1044     return bs->device_name;
1045 }
1046
1047 void bdrv_flush(BlockDriverState *bs)
1048 {
1049     if (!bs->drv)
1050         return;
1051     if (bs->drv->bdrv_flush)
1052         bs->drv->bdrv_flush(bs);
1053     if (bs->backing_hd)
1054         bdrv_flush(bs->backing_hd);
1055 }
1056
1057 void bdrv_flush_all(void)
1058 {
1059     BlockDriverState *bs;
1060
1061     for (bs = bdrv_first; bs != NULL; bs = bs->next)
1062         if (bs->drv && !bdrv_is_read_only(bs) &&
1063             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1064             bdrv_flush(bs);
1065 }
1066
1067 /*
1068  * Returns true iff the specified sector is present in the disk image. Drivers
1069  * not implementing the functionality are assumed to not support backing files,
1070  * hence all their sectors are reported as allocated.
1071  *
1072  * 'pnum' is set to the number of sectors (including and immediately following
1073  * the specified sector) that are known to be in the same
1074  * allocated/unallocated state.
1075  *
1076  * 'nb_sectors' is the max value 'pnum' should be set to.
1077  */
1078 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1079         int *pnum)
1080 {
1081     int64_t n;
1082     if (!bs->drv->bdrv_is_allocated) {
1083         if (sector_num >= bs->total_sectors) {
1084             *pnum = 0;
1085             return 0;
1086         }
1087         n = bs->total_sectors - sector_num;
1088         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1089         return 1;
1090     }
1091     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1092 }
1093
1094 void bdrv_info(Monitor *mon)
1095 {
1096     BlockDriverState *bs;
1097
1098     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1099         monitor_printf(mon, "%s:", bs->device_name);
1100         monitor_printf(mon, " type=");
1101         switch(bs->type) {
1102         case BDRV_TYPE_HD:
1103             monitor_printf(mon, "hd");
1104             break;
1105         case BDRV_TYPE_CDROM:
1106             monitor_printf(mon, "cdrom");
1107             break;
1108         case BDRV_TYPE_FLOPPY:
1109             monitor_printf(mon, "floppy");
1110             break;
1111         }
1112         monitor_printf(mon, " removable=%d", bs->removable);
1113         if (bs->removable) {
1114             monitor_printf(mon, " locked=%d", bs->locked);
1115         }
1116         if (bs->drv) {
1117             monitor_printf(mon, " file=");
1118             monitor_print_filename(mon, bs->filename);
1119             if (bs->backing_file[0] != '\0') {
1120                 monitor_printf(mon, " backing_file=");
1121                 monitor_print_filename(mon, bs->backing_file);
1122             }
1123             monitor_printf(mon, " ro=%d", bs->read_only);
1124             monitor_printf(mon, " drv=%s", bs->drv->format_name);
1125             monitor_printf(mon, " encrypted=%d", bdrv_is_encrypted(bs));
1126         } else {
1127             monitor_printf(mon, " [not inserted]");
1128         }
1129         monitor_printf(mon, "\n");
1130     }
1131 }
1132
1133 /* The "info blockstats" command. */
1134 void bdrv_info_stats(Monitor *mon)
1135 {
1136     BlockDriverState *bs;
1137
1138     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1139         monitor_printf(mon, "%s:"
1140                        " rd_bytes=%" PRIu64
1141                        " wr_bytes=%" PRIu64
1142                        " rd_operations=%" PRIu64
1143                        " wr_operations=%" PRIu64
1144                        "\n",
1145                        bs->device_name,
1146                        bs->rd_bytes, bs->wr_bytes,
1147                        bs->rd_ops, bs->wr_ops);
1148     }
1149 }
1150
1151 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1152 {
1153     if (bs->backing_hd && bs->backing_hd->encrypted)
1154         return bs->backing_file;
1155     else if (bs->encrypted)
1156         return bs->filename;
1157     else
1158         return NULL;
1159 }
1160
1161 void bdrv_get_backing_filename(BlockDriverState *bs,
1162                                char *filename, int filename_size)
1163 {
1164     if (!bs->backing_hd) {
1165         pstrcpy(filename, filename_size, "");
1166     } else {
1167         pstrcpy(filename, filename_size, bs->backing_file);
1168     }
1169 }
1170
1171 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1172                           const uint8_t *buf, int nb_sectors)
1173 {
1174     BlockDriver *drv = bs->drv;
1175     if (!drv)
1176         return -ENOMEDIUM;
1177     if (!drv->bdrv_write_compressed)
1178         return -ENOTSUP;
1179     if (bdrv_check_request(bs, sector_num, nb_sectors))
1180         return -EIO;
1181     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1182 }
1183
1184 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1185 {
1186     BlockDriver *drv = bs->drv;
1187     if (!drv)
1188         return -ENOMEDIUM;
1189     if (!drv->bdrv_get_info)
1190         return -ENOTSUP;
1191     memset(bdi, 0, sizeof(*bdi));
1192     return drv->bdrv_get_info(bs, bdi);
1193 }
1194
1195 int bdrv_put_buffer(BlockDriverState *bs, const uint8_t *buf, int64_t pos, int size)
1196 {
1197     BlockDriver *drv = bs->drv;
1198     if (!drv)
1199         return -ENOMEDIUM;
1200     if (!drv->bdrv_put_buffer)
1201         return -ENOTSUP;
1202     return drv->bdrv_put_buffer(bs, buf, pos, size);
1203 }
1204
1205 int bdrv_get_buffer(BlockDriverState *bs, uint8_t *buf, int64_t pos, int size)
1206 {
1207     BlockDriver *drv = bs->drv;
1208     if (!drv)
1209         return -ENOMEDIUM;
1210     if (!drv->bdrv_get_buffer)
1211         return -ENOTSUP;
1212     return drv->bdrv_get_buffer(bs, buf, pos, size);
1213 }
1214
1215 /**************************************************************/
1216 /* handling of snapshots */
1217
1218 int bdrv_snapshot_create(BlockDriverState *bs,
1219                          QEMUSnapshotInfo *sn_info)
1220 {
1221     BlockDriver *drv = bs->drv;
1222     if (!drv)
1223         return -ENOMEDIUM;
1224     if (!drv->bdrv_snapshot_create)
1225         return -ENOTSUP;
1226     return drv->bdrv_snapshot_create(bs, sn_info);
1227 }
1228
1229 int bdrv_snapshot_goto(BlockDriverState *bs,
1230                        const char *snapshot_id)
1231 {
1232     BlockDriver *drv = bs->drv;
1233     if (!drv)
1234         return -ENOMEDIUM;
1235     if (!drv->bdrv_snapshot_goto)
1236         return -ENOTSUP;
1237     return drv->bdrv_snapshot_goto(bs, snapshot_id);
1238 }
1239
1240 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1241 {
1242     BlockDriver *drv = bs->drv;
1243     if (!drv)
1244         return -ENOMEDIUM;
1245     if (!drv->bdrv_snapshot_delete)
1246         return -ENOTSUP;
1247     return drv->bdrv_snapshot_delete(bs, snapshot_id);
1248 }
1249
1250 int bdrv_snapshot_list(BlockDriverState *bs,
1251                        QEMUSnapshotInfo **psn_info)
1252 {
1253     BlockDriver *drv = bs->drv;
1254     if (!drv)
1255         return -ENOMEDIUM;
1256     if (!drv->bdrv_snapshot_list)
1257         return -ENOTSUP;
1258     return drv->bdrv_snapshot_list(bs, psn_info);
1259 }
1260
1261 #define NB_SUFFIXES 4
1262
1263 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1264 {
1265     static const char suffixes[NB_SUFFIXES] = "KMGT";
1266     int64_t base;
1267     int i;
1268
1269     if (size <= 999) {
1270         snprintf(buf, buf_size, "%" PRId64, size);
1271     } else {
1272         base = 1024;
1273         for(i = 0; i < NB_SUFFIXES; i++) {
1274             if (size < (10 * base)) {
1275                 snprintf(buf, buf_size, "%0.1f%c",
1276                          (double)size / base,
1277                          suffixes[i]);
1278                 break;
1279             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1280                 snprintf(buf, buf_size, "%" PRId64 "%c",
1281                          ((size + (base >> 1)) / base),
1282                          suffixes[i]);
1283                 break;
1284             }
1285             base = base * 1024;
1286         }
1287     }
1288     return buf;
1289 }
1290
1291 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1292 {
1293     char buf1[128], date_buf[128], clock_buf[128];
1294 #ifdef _WIN32
1295     struct tm *ptm;
1296 #else
1297     struct tm tm;
1298 #endif
1299     time_t ti;
1300     int64_t secs;
1301
1302     if (!sn) {
1303         snprintf(buf, buf_size,
1304                  "%-10s%-20s%7s%20s%15s",
1305                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1306     } else {
1307         ti = sn->date_sec;
1308 #ifdef _WIN32
1309         ptm = localtime(&ti);
1310         strftime(date_buf, sizeof(date_buf),
1311                  "%Y-%m-%d %H:%M:%S", ptm);
1312 #else
1313         localtime_r(&ti, &tm);
1314         strftime(date_buf, sizeof(date_buf),
1315                  "%Y-%m-%d %H:%M:%S", &tm);
1316 #endif
1317         secs = sn->vm_clock_nsec / 1000000000;
1318         snprintf(clock_buf, sizeof(clock_buf),
1319                  "%02d:%02d:%02d.%03d",
1320                  (int)(secs / 3600),
1321                  (int)((secs / 60) % 60),
1322                  (int)(secs % 60),
1323                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1324         snprintf(buf, buf_size,
1325                  "%-10s%-20s%7s%20s%15s",
1326                  sn->id_str, sn->name,
1327                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1328                  date_buf,
1329                  clock_buf);
1330     }
1331     return buf;
1332 }
1333
1334
1335 /**************************************************************/
1336 /* async I/Os */
1337
1338 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1339                                  QEMUIOVector *qiov, int nb_sectors,
1340                                  BlockDriverCompletionFunc *cb, void *opaque)
1341 {
1342     BlockDriver *drv = bs->drv;
1343     BlockDriverAIOCB *ret;
1344
1345     if (!drv)
1346         return NULL;
1347     if (bdrv_check_request(bs, sector_num, nb_sectors))
1348         return NULL;
1349
1350     ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1351                               cb, opaque);
1352
1353     if (ret) {
1354         /* Update stats even though technically transfer has not happened. */
1355         bs->rd_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1356         bs->rd_ops ++;
1357     }
1358
1359     return ret;
1360 }
1361
1362 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1363                                   QEMUIOVector *qiov, int nb_sectors,
1364                                   BlockDriverCompletionFunc *cb, void *opaque)
1365 {
1366     BlockDriver *drv = bs->drv;
1367     BlockDriverAIOCB *ret;
1368
1369     if (!drv)
1370         return NULL;
1371     if (bs->read_only)
1372         return NULL;
1373     if (bdrv_check_request(bs, sector_num, nb_sectors))
1374         return NULL;
1375
1376     ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1377                                cb, opaque);
1378
1379     if (ret) {
1380         /* Update stats even though technically transfer has not happened. */
1381         bs->wr_bytes += (unsigned) nb_sectors * SECTOR_SIZE;
1382         bs->wr_ops ++;
1383     }
1384
1385     return ret;
1386 }
1387
1388 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1389 {
1390     acb->pool->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
1401     if (!acb->is_write)
1402         qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
1403     qemu_vfree(acb->bounce);
1404     acb->common.cb(acb->common.opaque, acb->ret);
1405
1406     qemu_aio_release(acb);
1407 }
1408
1409 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1410                                             int64_t sector_num,
1411                                             QEMUIOVector *qiov,
1412                                             int nb_sectors,
1413                                             BlockDriverCompletionFunc *cb,
1414                                             void *opaque,
1415                                             int is_write)
1416
1417 {
1418     BlockDriverAIOCBSync *acb;
1419
1420     acb = qemu_aio_get(bs, cb, opaque);
1421     acb->is_write = is_write;
1422     acb->qiov = qiov;
1423     acb->bounce = qemu_blockalign(bs, qiov->size);
1424
1425     if (!acb->bh)
1426         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1427
1428     if (is_write) {
1429         qemu_iovec_to_buffer(acb->qiov, acb->bounce);
1430         acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
1431     } else {
1432         acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
1433     }
1434
1435     qemu_bh_schedule(acb->bh);
1436
1437     return &acb->common;
1438 }
1439
1440 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
1441         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1442         BlockDriverCompletionFunc *cb, void *opaque)
1443 {
1444     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1445 }
1446
1447 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
1448         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1449         BlockDriverCompletionFunc *cb, void *opaque)
1450 {
1451     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1452 }
1453
1454
1455 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1456 {
1457     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1458     qemu_bh_cancel(acb->bh);
1459     qemu_aio_release(acb);
1460 }
1461
1462 /**************************************************************/
1463 /* sync block device emulation */
1464
1465 static void bdrv_rw_em_cb(void *opaque, int ret)
1466 {
1467     *(int *)opaque = ret;
1468 }
1469
1470 #define NOT_DONE 0x7fffffff
1471
1472 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1473                         uint8_t *buf, int nb_sectors)
1474 {
1475     int async_ret;
1476     BlockDriverAIOCB *acb;
1477     struct iovec iov;
1478     QEMUIOVector qiov;
1479
1480     async_ret = NOT_DONE;
1481     iov.iov_base = (void *)buf;
1482     iov.iov_len = nb_sectors * 512;
1483     qemu_iovec_init_external(&qiov, &iov, 1);
1484     acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
1485         bdrv_rw_em_cb, &async_ret);
1486     if (acb == NULL)
1487         return -1;
1488
1489     while (async_ret == NOT_DONE) {
1490         qemu_aio_wait();
1491     }
1492
1493     return async_ret;
1494 }
1495
1496 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1497                          const uint8_t *buf, int nb_sectors)
1498 {
1499     int async_ret;
1500     BlockDriverAIOCB *acb;
1501     struct iovec iov;
1502     QEMUIOVector qiov;
1503
1504     async_ret = NOT_DONE;
1505     iov.iov_base = (void *)buf;
1506     iov.iov_len = nb_sectors * 512;
1507     qemu_iovec_init_external(&qiov, &iov, 1);
1508     acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
1509         bdrv_rw_em_cb, &async_ret);
1510     if (acb == NULL)
1511         return -1;
1512     while (async_ret == NOT_DONE) {
1513         qemu_aio_wait();
1514     }
1515     return async_ret;
1516 }
1517
1518 void bdrv_init(void)
1519 {
1520     module_call_init(MODULE_INIT_BLOCK);
1521 }
1522
1523 void aio_pool_init(AIOPool *pool, int aiocb_size,
1524                    void (*cancel)(BlockDriverAIOCB *acb))
1525 {
1526     pool->aiocb_size = aiocb_size;
1527     pool->cancel = cancel;
1528     pool->free_aiocb = NULL;
1529 }
1530
1531 void *qemu_aio_get_pool(AIOPool *pool, BlockDriverState *bs,
1532                         BlockDriverCompletionFunc *cb, void *opaque)
1533 {
1534     BlockDriverAIOCB *acb;
1535
1536     if (pool->free_aiocb) {
1537         acb = pool->free_aiocb;
1538         pool->free_aiocb = acb->next;
1539     } else {
1540         acb = qemu_mallocz(pool->aiocb_size);
1541         acb->pool = pool;
1542     }
1543     acb->bs = bs;
1544     acb->cb = cb;
1545     acb->opaque = opaque;
1546     return acb;
1547 }
1548
1549 void *qemu_aio_get(BlockDriverState *bs, BlockDriverCompletionFunc *cb,
1550                    void *opaque)
1551 {
1552     return qemu_aio_get_pool(&bs->drv->aio_pool, bs, cb, opaque);
1553 }
1554
1555 void qemu_aio_release(void *p)
1556 {
1557     BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
1558     AIOPool *pool = acb->pool;
1559     acb->next = pool->free_aiocb;
1560     pool->free_aiocb = acb;
1561 }
1562
1563 /**************************************************************/
1564 /* removable device support */
1565
1566 /**
1567  * Return TRUE if the media is present
1568  */
1569 int bdrv_is_inserted(BlockDriverState *bs)
1570 {
1571     BlockDriver *drv = bs->drv;
1572     int ret;
1573     if (!drv)
1574         return 0;
1575     if (!drv->bdrv_is_inserted)
1576         return 1;
1577     ret = drv->bdrv_is_inserted(bs);
1578     return ret;
1579 }
1580
1581 /**
1582  * Return TRUE if the media changed since the last call to this
1583  * function. It is currently only used for floppy disks
1584  */
1585 int bdrv_media_changed(BlockDriverState *bs)
1586 {
1587     BlockDriver *drv = bs->drv;
1588     int ret;
1589
1590     if (!drv || !drv->bdrv_media_changed)
1591         ret = -ENOTSUP;
1592     else
1593         ret = drv->bdrv_media_changed(bs);
1594     if (ret == -ENOTSUP)
1595         ret = bs->media_changed;
1596     bs->media_changed = 0;
1597     return ret;
1598 }
1599
1600 /**
1601  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1602  */
1603 void bdrv_eject(BlockDriverState *bs, int eject_flag)
1604 {
1605     BlockDriver *drv = bs->drv;
1606     int ret;
1607
1608     if (!drv || !drv->bdrv_eject) {
1609         ret = -ENOTSUP;
1610     } else {
1611         ret = drv->bdrv_eject(bs, eject_flag);
1612     }
1613     if (ret == -ENOTSUP) {
1614         if (eject_flag)
1615             bdrv_close(bs);
1616     }
1617 }
1618
1619 int bdrv_is_locked(BlockDriverState *bs)
1620 {
1621     return bs->locked;
1622 }
1623
1624 /**
1625  * Lock or unlock the media (if it is locked, the user won't be able
1626  * to eject it manually).
1627  */
1628 void bdrv_set_locked(BlockDriverState *bs, int locked)
1629 {
1630     BlockDriver *drv = bs->drv;
1631
1632     bs->locked = locked;
1633     if (drv && drv->bdrv_set_locked) {
1634         drv->bdrv_set_locked(bs, locked);
1635     }
1636 }
1637
1638 /* needed for generic scsi interface */
1639
1640 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1641 {
1642     BlockDriver *drv = bs->drv;
1643
1644     if (drv && drv->bdrv_ioctl)
1645         return drv->bdrv_ioctl(bs, req, buf);
1646     return -ENOTSUP;
1647 }
1648
1649 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
1650         unsigned long int req, void *buf,
1651         BlockDriverCompletionFunc *cb, void *opaque)
1652 {
1653     BlockDriver *drv = bs->drv;
1654
1655     if (drv && drv->bdrv_aio_ioctl)
1656         return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
1657     return NULL;
1658 }
1659
1660 void *qemu_blockalign(BlockDriverState *bs, size_t size)
1661 {
1662     return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
1663 }