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 #include "qemu-common.h"
  26 #include "monitor.h"
  27 #include "block_int.h"
  28 #include "module.h"
  29
  30 #ifdef CONFIG_BSD
  31 #include <sys/types.h>
  32 #include <sys/stat.h>
  33 #include <sys/ioctl.h>
  34 #include <sys/queue.h>
  35 #ifndef __DragonFly__
  36 #include <sys/disk.h>
  37 #endif
  38 #endif
  39
  40 #ifdef _WIN32
  41 #include <windows.h>
  42 #endif
  43
  44 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
  45         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
  46         BlockDriverCompletionFunc *cb, void *opaque);
  47 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
  48         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
  49         BlockDriverCompletionFunc *cb, void *opaque);
  50 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
  51         BlockDriverCompletionFunc *cb, void *opaque);
  52 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  53                         uint8_t *buf, int nb_sectors);
  54 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  55                          const uint8_t *buf, int nb_sectors);
  56
  57 BlockDriverState *bdrv_first;
  58
  59 static BlockDriver *first_drv;
  60
  61 /* If non-zero, use only whitelisted block drivers */
  62 static int use_bdrv_whitelist;
  63
  64 int path_is_absolute(const char *path)
  65 {
  66     const char *p;
  67 #ifdef _WIN32
  68     /* specific case for names like: "\\.\d:" */
  69     if (*path == '/' || *path == '\\')
  70         return 1;
  71 #endif
  72     p = strchr(path, ':');
  73     if (p)
  74         p++;
  75     else
  76         p = path;
  77 #ifdef _WIN32
  78     return (*p == '/' || *p == '\\');
  79 #else
  80     return (*p == '/');
  81 #endif
  82 }
  83
  84 /* if filename is absolute, just copy it to dest. Otherwise, build a
  85    path to it by considering it is relative to base_path. URL are
  86    supported. */
  87 void path_combine(char *dest, int dest_size,
  88                   const char *base_path,
  89                   const char *filename)
  90 {
  91     const char *p, *p1;
  92     int len;
  93
  94     if (dest_size <= 0)
  95         return;
  96     if (path_is_absolute(filename)) {
  97         pstrcpy(dest, dest_size, filename);
  98     } else {
  99         p = strchr(base_path, ':');
 100         if (p)
 101             p++;
 102         else
 103             p = base_path;
 104         p1 = strrchr(base_path, '/');
 105 #ifdef _WIN32
 106         {
 107             const char *p2;
 108             p2 = strrchr(base_path, '\\');
 109             if (!p1 || p2 > p1)
 110                 p1 = p2;
 111         }
 112 #endif
 113         if (p1)
 114             p1++;
 115         else
 116             p1 = base_path;
 117         if (p1 > p)
 118             p = p1;
 119         len = p - base_path;
 120         if (len > dest_size - 1)
 121             len = dest_size - 1;
 122         memcpy(dest, base_path, len);
 123         dest[len] = '\0';
 124         pstrcat(dest, dest_size, filename);
 125     }
 126 }
 127
 128 void bdrv_register(BlockDriver *bdrv)
 129 {
 130     if (!bdrv->bdrv_aio_readv) {
 131         /* add AIO emulation layer */
 132         bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
 133         bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
 134     } else if (!bdrv->bdrv_read) {
 135         /* add synchronous IO emulation layer */
 136         bdrv->bdrv_read = bdrv_read_em;
 137         bdrv->bdrv_write = bdrv_write_em;
 138     }
 139
 140     if (!bdrv->bdrv_aio_flush)
 141         bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
 142
 143     bdrv->next = first_drv;
 144     first_drv = bdrv;
 145 }
 146
 147 /* create a new block device (by default it is empty) */
 148 BlockDriverState *bdrv_new(const char *device_name)
 149 {
 150     BlockDriverState **pbs, *bs;
 151
 152     bs = qemu_mallocz(sizeof(BlockDriverState));
 153     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 154     if (device_name[0] != '\0') {
 155         /* insert at the end */
 156         pbs = &bdrv_first;
 157         while (*pbs != NULL)
 158             pbs = &(*pbs)->next;
 159         *pbs = bs;
 160     }
 161     return bs;
 162 }
 163
 164 BlockDriver *bdrv_find_format(const char *format_name)
 165 {
 166     BlockDriver *drv1;
 167     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 168         if (!strcmp(drv1->format_name, format_name))
 169             return drv1;
 170     }
 171     return NULL;
 172 }
 173
 174 static int bdrv_is_whitelisted(BlockDriver *drv)
 175 {
 176     static const char *whitelist[] = {
 177         CONFIG_BDRV_WHITELIST
 178     };
 179     const char **p;
 180
 181     if (!whitelist[0])
 182         return 1;               /* no whitelist, anything goes */
 183
 184     for (p = whitelist; *p; p++) {
 185         if (!strcmp(drv->format_name, *p)) {
 186             return 1;
 187         }
 188     }
 189     return 0;
 190 }
 191
 192 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
 193 {
 194     BlockDriver *drv = bdrv_find_format(format_name);
 195     return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
 196 }
 197
 198 int bdrv_create(BlockDriver *drv, const char* filename,
 199     QEMUOptionParameter *options)
 200 {
 201     if (!drv->bdrv_create)
 202         return -ENOTSUP;
 203
 204     return drv->bdrv_create(filename, options);
 205 }
 206
 207 #ifdef _WIN32
 208 void get_tmp_filename(char *filename, int size)
 209 {
 210     char temp_dir[MAX_PATH];
 211
 212     GetTempPath(MAX_PATH, temp_dir);
 213     GetTempFileName(temp_dir, "qem", 0, filename);
 214 }
 215 #else
 216 void get_tmp_filename(char *filename, int size)
 217 {
 218     int fd;
 219     const char *tmpdir;
 220     /* XXX: race condition possible */
 221     tmpdir = getenv("TMPDIR");
 222     if (!tmpdir)
 223         tmpdir = "/tmp";
 224     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 225     fd = mkstemp(filename);
 226     close(fd);
 227 }
 228 #endif
 229
 230 #ifdef _WIN32
 231 static int is_windows_drive_prefix(const char *filename)
 232 {
 233     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 234              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 235             filename[1] == ':');
 236 }
 237
 238 int is_windows_drive(const char *filename)
 239 {
 240     if (is_windows_drive_prefix(filename) &&
 241         filename[2] == '\0')
 242         return 1;
 243     if (strstart(filename, "\\\\.\\", NULL) ||
 244         strstart(filename, "//./", NULL))
 245         return 1;
 246     return 0;
 247 }
 248 #endif
 249
 250 static BlockDriver *find_protocol(const char *filename)
 251 {
 252     BlockDriver *drv1;
 253     char protocol[128];
 254     int len;
 255     const char *p;
 256
 257 #ifdef _WIN32
 258     if (is_windows_drive(filename) ||
 259         is_windows_drive_prefix(filename))
 260         return bdrv_find_format("raw");
 261 #endif
 262     p = strchr(filename, ':');
 263     if (!p)
 264         return bdrv_find_format("raw");
 265     len = p - filename;
 266     if (len > sizeof(protocol) - 1)
 267         len = sizeof(protocol) - 1;
 268     memcpy(protocol, filename, len);
 269     protocol[len] = '\0';
 270     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 271         if (drv1->protocol_name &&
 272             !strcmp(drv1->protocol_name, protocol))
 273             return drv1;
 274     }
 275     return NULL;
 276 }
 277
 278 /*
 279  * Detect host devices. By convention, /dev/cdrom[N] is always
 280  * recognized as a host CDROM.
 281  */
 282 static BlockDriver *find_hdev_driver(const char *filename)
 283 {
 284     int score_max = 0, score;
 285     BlockDriver *drv = NULL, *d;
 286
 287     for (d = first_drv; d; d = d->next) {
 288         if (d->bdrv_probe_device) {
 289             score = d->bdrv_probe_device(filename);
 290             if (score > score_max) {
 291                 score_max = score;
 292                 drv = d;
 293             }
 294         }
 295     }
 296
 297     return drv;
 298 }
 299
 300 static BlockDriver *find_image_format(const char *filename)
 301 {
 302     int ret, score, score_max;
 303     BlockDriver *drv1, *drv;
 304     uint8_t buf[2048];
 305     BlockDriverState *bs;
 306
 307     drv = find_protocol(filename);
 308     /* no need to test disk image formats for vvfat */
 309     if (drv && strcmp(drv->format_name, "vvfat") == 0)
 310         return drv;
 311
 312     ret = bdrv_file_open(&bs, filename, BDRV_O_RDONLY);
 313     if (ret < 0)
 314         return NULL;
 315     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 316     bdrv_delete(bs);
 317     if (ret < 0) {
 318         return NULL;
 319     }
 320
 321     score_max = 0;
 322     for(drv1 = first_drv; drv1 != NULL; drv1 = drv1->next) {
 323         if (drv1->bdrv_probe) {
 324             score = drv1->bdrv_probe(buf, ret, filename);
 325             if (score > score_max) {
 326                 score_max = score;
 327                 drv = drv1;
 328             }
 329         }
 330     }
 331     return drv;
 332 }
 333
 334 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 335 {
 336     BlockDriverState *bs;
 337     int ret;
 338
 339     bs = bdrv_new("");
 340     ret = bdrv_open2(bs, filename, flags | BDRV_O_FILE, NULL);
 341     if (ret < 0) {
 342         bdrv_delete(bs);
 343         return ret;
 344     }
 345     bs->growable = 1;
 346     *pbs = bs;
 347     return 0;
 348 }
 349
 350 int bdrv_open(BlockDriverState *bs, const char *filename, int flags)
 351 {
 352     return bdrv_open2(bs, filename, flags, NULL);
 353 }
 354
 355 int bdrv_open2(BlockDriverState *bs, const char *filename, int flags,
 356                BlockDriver *drv)
 357 {
 358     int ret, open_flags, try_rw;
 359     char tmp_filename[PATH_MAX];
 360     char backing_filename[PATH_MAX];
 361
 362     bs->is_temporary = 0;
 363     bs->encrypted = 0;
 364     bs->valid_key = 0;
 365     /* buffer_alignment defaulted to 512, drivers can change this value */
 366     bs->buffer_alignment = 512;
 367
 368     if (flags & BDRV_O_SNAPSHOT) {
 369         BlockDriverState *bs1;
 370         int64_t total_size;
 371         int is_protocol = 0;
 372         BlockDriver *bdrv_qcow2;
 373         QEMUOptionParameter *options;
 374
 375         /* if snapshot, we create a temporary backing file and open it
 376            instead of opening 'filename' directly */
 377
 378         /* if there is a backing file, use it */
 379         bs1 = bdrv_new("");
 380         ret = bdrv_open2(bs1, filename, 0, drv);
 381         if (ret < 0) {
 382             bdrv_delete(bs1);
 383             return ret;
 384         }
 385         total_size = bdrv_getlength(bs1) >> BDRV_SECTOR_BITS;
 386
 387         if (bs1->drv && bs1->drv->protocol_name)
 388             is_protocol = 1;
 389
 390         bdrv_delete(bs1);
 391
 392         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 393
 394         /* Real path is meaningless for protocols */
 395         if (is_protocol)
 396             snprintf(backing_filename, sizeof(backing_filename),
 397                      "%s", filename);
 398         else
 399             realpath(filename, backing_filename);
 400
 401         bdrv_qcow2 = bdrv_find_format("qcow2");
 402         options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
 403
 404         set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size * 512);
 405         set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
 406         if (drv) {
 407             set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
 408                 drv->format_name);
 409         }
 410
 411         ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
 412         if (ret < 0) {
 413             return ret;
 414         }
 415
 416         filename = tmp_filename;
 417         drv = bdrv_qcow2;
 418         bs->is_temporary = 1;
 419     }
 420
 421     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 422     if (flags & BDRV_O_FILE) {
 423         drv = find_protocol(filename);
 424     } else if (!drv) {
 425         drv = find_hdev_driver(filename);
 426         if (!drv) {
 427             drv = find_image_format(filename);
 428         }
 429     }
 430     if (!drv) {
 431         ret = -ENOENT;
 432         goto unlink_and_fail;
 433     }
 434     bs->drv = drv;
 435     bs->opaque = qemu_mallocz(drv->instance_size);
 436
 437     /*
 438      * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
 439      * write cache to the guest.  We do need the fdatasync to flush
 440      * out transactions for block allocations, and we maybe have a
 441      * volatile write cache in our backing device to deal with.
 442      */
 443     if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
 444         bs->enable_write_cache = 1;
 445
 446     /* Note: for compatibility, we open disk image files as RDWR, and
 447        RDONLY as fallback */
 448     try_rw = !bs->read_only || bs->is_temporary;
 449     if (!(flags & BDRV_O_FILE))
 450         open_flags = (try_rw ? BDRV_O_RDWR : 0) |
 451             (flags & (BDRV_O_CACHE_MASK|BDRV_O_NATIVE_AIO));
 452     else
 453         open_flags = flags & ~(BDRV_O_FILE | BDRV_O_SNAPSHOT);
 454     if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv))
 455         ret = -ENOTSUP;
 456     else
 457         ret = drv->bdrv_open(bs, filename, open_flags);
 458     if ((ret == -EACCES || ret == -EPERM) && !(flags & BDRV_O_FILE)) {
 459         ret = drv->bdrv_open(bs, filename, open_flags & ~BDRV_O_RDWR);
 460         bs->read_only = 1;
 461     }
 462     if (ret < 0) {
 463         qemu_free(bs->opaque);
 464         bs->opaque = NULL;
 465         bs->drv = NULL;
 466     unlink_and_fail:
 467         if (bs->is_temporary)
 468             unlink(filename);
 469         return ret;
 470     }
 471     if (drv->bdrv_getlength) {
 472         bs->total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
 473     }
 474 #ifndef _WIN32
 475     if (bs->is_temporary) {
 476         unlink(filename);
 477     }
 478 #endif
 479     if (bs->backing_file[0] != '\0') {
 480         /* if there is a backing file, use it */
 481         BlockDriver *back_drv = NULL;
 482         bs->backing_hd = bdrv_new("");
 483         /* pass on read_only property to the backing_hd */
 484         bs->backing_hd->read_only = bs->read_only;
 485         path_combine(backing_filename, sizeof(backing_filename),
 486                      filename, bs->backing_file);
 487         if (bs->backing_format[0] != '\0')
 488             back_drv = bdrv_find_format(bs->backing_format);
 489         ret = bdrv_open2(bs->backing_hd, backing_filename, open_flags,
 490                          back_drv);
 491         if (ret < 0) {
 492             bdrv_close(bs);
 493             return ret;
 494         }
 495     }
 496
 497     if (!bdrv_key_required(bs)) {
 498         /* call the change callback */
 499         bs->media_changed = 1;
 500         if (bs->change_cb)
 501             bs->change_cb(bs->change_opaque);
 502     }
 503     return 0;
 504 }
 505
 506 void bdrv_close(BlockDriverState *bs)
 507 {
 508     if (bs->drv) {
 509         if (bs->backing_hd)
 510             bdrv_delete(bs->backing_hd);
 511         bs->drv->bdrv_close(bs);
 512         qemu_free(bs->opaque);
 513 #ifdef _WIN32
 514         if (bs->is_temporary) {
 515             unlink(bs->filename);
 516         }
 517 #endif
 518         bs->opaque = NULL;
 519         bs->drv = NULL;
 520
 521         /* call the change callback */
 522         bs->media_changed = 1;
 523         if (bs->change_cb)
 524             bs->change_cb(bs->change_opaque);
 525     }
 526 }
 527
 528 void bdrv_delete(BlockDriverState *bs)
 529 {
 530     BlockDriverState **pbs;
 531
 532     pbs = &bdrv_first;
 533     while (*pbs != bs && *pbs != NULL)
 534         pbs = &(*pbs)->next;
 535     if (*pbs == bs)
 536         *pbs = bs->next;
 537
 538     bdrv_close(bs);
 539     qemu_free(bs);
 540 }
 541
 542 /*
 543  * Run consistency checks on an image
 544  *
 545  * Returns the number of errors or -errno when an internal error occurs
 546  */
 547 int bdrv_check(BlockDriverState *bs)
 548 {
 549     if (bs->drv->bdrv_check == NULL) {
 550         return -ENOTSUP;
 551     }
 552
 553     return bs->drv->bdrv_check(bs);
 554 }
 555
 556 /* commit COW file into the raw image */
 557 int bdrv_commit(BlockDriverState *bs)
 558 {
 559     BlockDriver *drv = bs->drv;
 560     int64_t i, total_sectors;
 561     int n, j;
 562     unsigned char sector[512];
 563
 564     if (!drv)
 565         return -ENOMEDIUM;
 566
 567     if (bs->read_only) {
 568         return -EACCES;
 569     }
 570
 571     if (!bs->backing_hd) {
 572         return -ENOTSUP;
 573     }
 574
 575     total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
 576     for (i = 0; i < total_sectors;) {
 577         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 578             for(j = 0; j < n; j++) {
 579                 if (bdrv_read(bs, i, sector, 1) != 0) {
 580                     return -EIO;
 581                 }
 582
 583                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 584                     return -EIO;
 585                 }
 586                 i++;
 587             }
 588         } else {
 589             i += n;
 590         }
 591     }
 592
 593     if (drv->bdrv_make_empty)
 594         return drv->bdrv_make_empty(bs);
 595
 596     return 0;
 597 }
 598
 599 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
 600                                    size_t size)
 601 {
 602     int64_t len;
 603
 604     if (!bdrv_is_inserted(bs))
 605         return -ENOMEDIUM;
 606
 607     if (bs->growable)
 608         return 0;
 609
 610     len = bdrv_getlength(bs);
 611
 612     if (offset < 0)
 613         return -EIO;
 614
 615     if ((offset > len) || (len - offset < size))
 616         return -EIO;
 617
 618     return 0;
 619 }
 620
 621 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
 622                               int nb_sectors)
 623 {
 624     return bdrv_check_byte_request(bs, sector_num * 512, nb_sectors * 512);
 625 }
 626
 627 /* return < 0 if error. See bdrv_write() for the return codes */
 628 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 629               uint8_t *buf, int nb_sectors)
 630 {
 631     BlockDriver *drv = bs->drv;
 632
 633     if (!drv)
 634         return -ENOMEDIUM;
 635     if (bdrv_check_request(bs, sector_num, nb_sectors))
 636         return -EIO;
 637
 638     return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 639 }
 640
 641 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
 642                              int nb_sectors, int dirty)
 643 {
 644     int64_t start, end;
 645     unsigned long val, idx, bit;
 646
 647     start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
 648     end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
 649
 650     for (; start <= end; start++) {
 651         idx = start / (sizeof(unsigned long) * 8);
 652         bit = start % (sizeof(unsigned long) * 8);
 653         val = bs->dirty_bitmap[idx];
 654         if (dirty) {
 655             val |= 1 << bit;
 656         } else {
 657             val &= ~(1 << bit);
 658         }
 659         bs->dirty_bitmap[idx] = val;
 660     }
 661 }
 662
 663 /* Return < 0 if error. Important errors are:
 664   -EIO         generic I/O error (may happen for all errors)
 665   -ENOMEDIUM   No media inserted.
 666   -EINVAL      Invalid sector number or nb_sectors
 667   -EACCES      Trying to write a read-only device
 668 */
 669 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 670                const uint8_t *buf, int nb_sectors)
 671 {
 672     BlockDriver *drv = bs->drv;
 673     if (!bs->drv)
 674         return -ENOMEDIUM;
 675     if (bs->read_only)
 676         return -EACCES;
 677     if (bdrv_check_request(bs, sector_num, nb_sectors))
 678         return -EIO;
 679
 680     if (bs->dirty_bitmap) {
 681         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
 682     }
 683
 684     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 685 }
 686
 687 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 688                void *buf, int count1)
 689 {
 690     uint8_t tmp_buf[BDRV_SECTOR_SIZE];
 691     int len, nb_sectors, count;
 692     int64_t sector_num;
 693
 694     count = count1;
 695     /* first read to align to sector start */
 696     len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
 697     if (len > count)
 698         len = count;
 699     sector_num = offset >> BDRV_SECTOR_BITS;
 700     if (len > 0) {
 701         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 702             return -EIO;
 703         memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
 704         count -= len;
 705         if (count == 0)
 706             return count1;
 707         sector_num++;
 708         buf += len;
 709     }
 710
 711     /* read the sectors "in place" */
 712     nb_sectors = count >> BDRV_SECTOR_BITS;
 713     if (nb_sectors > 0) {
 714         if (bdrv_read(bs, sector_num, buf, nb_sectors) < 0)
 715             return -EIO;
 716         sector_num += nb_sectors;
 717         len = nb_sectors << BDRV_SECTOR_BITS;
 718         buf += len;
 719         count -= len;
 720     }
 721
 722     /* add data from the last sector */
 723     if (count > 0) {
 724         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 725             return -EIO;
 726         memcpy(buf, tmp_buf, count);
 727     }
 728     return count1;
 729 }
 730
 731 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 732                 const void *buf, int count1)
 733 {
 734     uint8_t tmp_buf[BDRV_SECTOR_SIZE];
 735     int len, nb_sectors, count;
 736     int64_t sector_num;
 737
 738     count = count1;
 739     /* first write to align to sector start */
 740     len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
 741     if (len > count)
 742         len = count;
 743     sector_num = offset >> BDRV_SECTOR_BITS;
 744     if (len > 0) {
 745         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 746             return -EIO;
 747         memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
 748         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 749             return -EIO;
 750         count -= len;
 751         if (count == 0)
 752             return count1;
 753         sector_num++;
 754         buf += len;
 755     }
 756
 757     /* write the sectors "in place" */
 758     nb_sectors = count >> BDRV_SECTOR_BITS;
 759     if (nb_sectors > 0) {
 760         if (bdrv_write(bs, sector_num, buf, nb_sectors) < 0)
 761             return -EIO;
 762         sector_num += nb_sectors;
 763         len = nb_sectors << BDRV_SECTOR_BITS;
 764         buf += len;
 765         count -= len;
 766     }
 767
 768     /* add data from the last sector */
 769     if (count > 0) {
 770         if (bdrv_read(bs, sector_num, tmp_buf, 1) < 0)
 771             return -EIO;
 772         memcpy(tmp_buf, buf, count);
 773         if (bdrv_write(bs, sector_num, tmp_buf, 1) < 0)
 774             return -EIO;
 775     }
 776     return count1;
 777 }
 778
 779 /**
 780  * Truncate file to 'offset' bytes (needed only for file protocols)
 781  */
 782 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
 783 {
 784     BlockDriver *drv = bs->drv;
 785     if (!drv)
 786         return -ENOMEDIUM;
 787     if (!drv->bdrv_truncate)
 788         return -ENOTSUP;
 789     if (bs->read_only)
 790         return -EACCES;
 791     return drv->bdrv_truncate(bs, offset);
 792 }
 793
 794 /**
 795  * Length of a file in bytes. Return < 0 if error or unknown.
 796  */
 797 int64_t bdrv_getlength(BlockDriverState *bs)
 798 {
 799     BlockDriver *drv = bs->drv;
 800     if (!drv)
 801         return -ENOMEDIUM;
 802     if (!drv->bdrv_getlength) {
 803         /* legacy mode */
 804         return bs->total_sectors * BDRV_SECTOR_SIZE;
 805     }
 806     return drv->bdrv_getlength(bs);
 807 }
 808
 809 /* return 0 as number of sectors if no device present or error */
 810 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
 811 {
 812     int64_t length;
 813     length = bdrv_getlength(bs);
 814     if (length < 0)
 815         length = 0;
 816     else
 817         length = length >> BDRV_SECTOR_BITS;
 818     *nb_sectors_ptr = length;
 819 }
 820
 821 struct partition {
 822         uint8_t boot_ind;           /* 0x80 - active */
 823         uint8_t head;               /* starting head */
 824         uint8_t sector;             /* starting sector */
 825         uint8_t cyl;                /* starting cylinder */
 826         uint8_t sys_ind;            /* What partition type */
 827         uint8_t end_head;           /* end head */
 828         uint8_t end_sector;         /* end sector */
 829         uint8_t end_cyl;            /* end cylinder */
 830         uint32_t start_sect;        /* starting sector counting from 0 */
 831         uint32_t nr_sects;          /* nr of sectors in partition */
 832 } __attribute__((packed));
 833
 834 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
 835 static int guess_disk_lchs(BlockDriverState *bs,
 836                            int *pcylinders, int *pheads, int *psectors)
 837 {
 838     uint8_t buf[512];
 839     int ret, i, heads, sectors, cylinders;
 840     struct partition *p;
 841     uint32_t nr_sects;
 842     uint64_t nb_sectors;
 843
 844     bdrv_get_geometry(bs, &nb_sectors);
 845
 846     ret = bdrv_read(bs, 0, buf, 1);
 847     if (ret < 0)
 848         return -1;
 849     /* test msdos magic */
 850     if (buf[510] != 0x55 || buf[511] != 0xaa)
 851         return -1;
 852     for(i = 0; i < 4; i++) {
 853         p = ((struct partition *)(buf + 0x1be)) + i;
 854         nr_sects = le32_to_cpu(p->nr_sects);
 855         if (nr_sects && p->end_head) {
 856             /* We make the assumption that the partition terminates on
 857                a cylinder boundary */
 858             heads = p->end_head + 1;
 859             sectors = p->end_sector & 63;
 860             if (sectors == 0)
 861                 continue;
 862             cylinders = nb_sectors / (heads * sectors);
 863             if (cylinders < 1 || cylinders > 16383)
 864                 continue;
 865             *pheads = heads;
 866             *psectors = sectors;
 867             *pcylinders = cylinders;
 868 #if 0
 869             printf("guessed geometry: LCHS=%d %d %d\n",
 870                    cylinders, heads, sectors);
 871 #endif
 872             return 0;
 873         }
 874     }
 875     return -1;
 876 }
 877
 878 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
 879 {
 880     int translation, lba_detected = 0;
 881     int cylinders, heads, secs;
 882     uint64_t nb_sectors;
 883
 884     /* if a geometry hint is available, use it */
 885     bdrv_get_geometry(bs, &nb_sectors);
 886     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
 887     translation = bdrv_get_translation_hint(bs);
 888     if (cylinders != 0) {
 889         *pcyls = cylinders;
 890         *pheads = heads;
 891         *psecs = secs;
 892     } else {
 893         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
 894             if (heads > 16) {
 895                 /* if heads > 16, it means that a BIOS LBA
 896                    translation was active, so the default
 897                    hardware geometry is OK */
 898                 lba_detected = 1;
 899                 goto default_geometry;
 900             } else {
 901                 *pcyls = cylinders;
 902                 *pheads = heads;
 903                 *psecs = secs;
 904                 /* disable any translation to be in sync with
 905                    the logical geometry */
 906                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
 907                     bdrv_set_translation_hint(bs,
 908                                               BIOS_ATA_TRANSLATION_NONE);
 909                 }
 910             }
 911         } else {
 912         default_geometry:
 913             /* if no geometry, use a standard physical disk geometry */
 914             cylinders = nb_sectors / (16 * 63);
 915
 916             if (cylinders > 16383)
 917                 cylinders = 16383;
 918             else if (cylinders < 2)
 919                 cylinders = 2;
 920             *pcyls = cylinders;
 921             *pheads = 16;
 922             *psecs = 63;
 923             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
 924                 if ((*pcyls * *pheads) <= 131072) {
 925                     bdrv_set_translation_hint(bs,
 926                                               BIOS_ATA_TRANSLATION_LARGE);
 927                 } else {
 928                     bdrv_set_translation_hint(bs,
 929                                               BIOS_ATA_TRANSLATION_LBA);
 930                 }
 931             }
 932         }
 933         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
 934     }
 935 }
 936
 937 void bdrv_set_geometry_hint(BlockDriverState *bs,
 938                             int cyls, int heads, int secs)
 939 {
 940     bs->cyls = cyls;
 941     bs->heads = heads;
 942     bs->secs = secs;
 943 }
 944
 945 void bdrv_set_type_hint(BlockDriverState *bs, int type)
 946 {
 947     bs->type = type;
 948     bs->removable = ((type == BDRV_TYPE_CDROM ||
 949                       type == BDRV_TYPE_FLOPPY));
 950 }
 951
 952 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
 953 {
 954     bs->translation = translation;
 955 }
 956
 957 void bdrv_get_geometry_hint(BlockDriverState *bs,
 958                             int *pcyls, int *pheads, int *psecs)
 959 {
 960     *pcyls = bs->cyls;
 961     *pheads = bs->heads;
 962     *psecs = bs->secs;
 963 }
 964
 965 int bdrv_get_type_hint(BlockDriverState *bs)
 966 {
 967     return bs->type;
 968 }
 969
 970 int bdrv_get_translation_hint(BlockDriverState *bs)
 971 {
 972     return bs->translation;
 973 }
 974
 975 int bdrv_is_removable(BlockDriverState *bs)
 976 {
 977     return bs->removable;
 978 }
 979
 980 int bdrv_is_read_only(BlockDriverState *bs)
 981 {
 982     return bs->read_only;
 983 }
 984
 985 int bdrv_set_read_only(BlockDriverState *bs, int read_only)
 986 {
 987     int ret = bs->read_only;
 988     bs->read_only = read_only;
 989     return ret;
 990 }
 991
 992 int bdrv_is_sg(BlockDriverState *bs)
 993 {
 994     return bs->sg;
 995 }
 996
 997 int bdrv_enable_write_cache(BlockDriverState *bs)
 998 {
 999     return bs->enable_write_cache;
1000 }
1001
1002 /* XXX: no longer used */
1003 void bdrv_set_change_cb(BlockDriverState *bs,
1004                         void (*change_cb)(void *opaque), void *opaque)
1005 {
1006     bs->change_cb = change_cb;
1007     bs->change_opaque = opaque;
1008 }
1009
1010 int bdrv_is_encrypted(BlockDriverState *bs)
1011 {
1012     if (bs->backing_hd && bs->backing_hd->encrypted)
1013         return 1;
1014     return bs->encrypted;
1015 }
1016
1017 int bdrv_key_required(BlockDriverState *bs)
1018 {
1019     BlockDriverState *backing_hd = bs->backing_hd;
1020
1021     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1022         return 1;
1023     return (bs->encrypted && !bs->valid_key);
1024 }
1025
1026 int bdrv_set_key(BlockDriverState *bs, const char *key)
1027 {
1028     int ret;
1029     if (bs->backing_hd && bs->backing_hd->encrypted) {
1030         ret = bdrv_set_key(bs->backing_hd, key);
1031         if (ret < 0)
1032             return ret;
1033         if (!bs->encrypted)
1034             return 0;
1035     }
1036     if (!bs->encrypted || !bs->drv || !bs->drv->bdrv_set_key)
1037         return -1;
1038     ret = bs->drv->bdrv_set_key(bs, key);
1039     if (ret < 0) {
1040         bs->valid_key = 0;
1041     } else if (!bs->valid_key) {
1042         bs->valid_key = 1;
1043         /* call the change callback now, we skipped it on open */
1044         bs->media_changed = 1;
1045         if (bs->change_cb)
1046             bs->change_cb(bs->change_opaque);
1047     }
1048     return ret;
1049 }
1050
1051 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1052 {
1053     if (!bs->drv) {
1054         buf[0] = '\0';
1055     } else {
1056         pstrcpy(buf, buf_size, bs->drv->format_name);
1057     }
1058 }
1059
1060 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1061                          void *opaque)
1062 {
1063     BlockDriver *drv;
1064
1065     for (drv = first_drv; drv != NULL; drv = drv->next) {
1066         it(opaque, drv->format_name);
1067     }
1068 }
1069
1070 BlockDriverState *bdrv_find(const char *name)
1071 {
1072     BlockDriverState *bs;
1073
1074     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1075         if (!strcmp(name, bs->device_name))
1076             return bs;
1077     }
1078     return NULL;
1079 }
1080
1081 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1082 {
1083     BlockDriverState *bs;
1084
1085     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1086         it(opaque, bs);
1087     }
1088 }
1089
1090 const char *bdrv_get_device_name(BlockDriverState *bs)
1091 {
1092     return bs->device_name;
1093 }
1094
1095 void bdrv_flush(BlockDriverState *bs)
1096 {
1097     if (!bs->drv)
1098         return;
1099     if (bs->drv->bdrv_flush)
1100         bs->drv->bdrv_flush(bs);
1101     if (bs->backing_hd)
1102         bdrv_flush(bs->backing_hd);
1103 }
1104
1105 void bdrv_flush_all(void)
1106 {
1107     BlockDriverState *bs;
1108
1109     for (bs = bdrv_first; bs != NULL; bs = bs->next)
1110         if (bs->drv && !bdrv_is_read_only(bs) &&
1111             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs)))
1112             bdrv_flush(bs);
1113 }
1114
1115 /*
1116  * Returns true iff the specified sector is present in the disk image. Drivers
1117  * not implementing the functionality are assumed to not support backing files,
1118  * hence all their sectors are reported as allocated.
1119  *
1120  * 'pnum' is set to the number of sectors (including and immediately following
1121  * the specified sector) that are known to be in the same
1122  * allocated/unallocated state.
1123  *
1124  * 'nb_sectors' is the max value 'pnum' should be set to.
1125  */
1126 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1127         int *pnum)
1128 {
1129     int64_t n;
1130     if (!bs->drv->bdrv_is_allocated) {
1131         if (sector_num >= bs->total_sectors) {
1132             *pnum = 0;
1133             return 0;
1134         }
1135         n = bs->total_sectors - sector_num;
1136         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1137         return 1;
1138     }
1139     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1140 }
1141
1142 void bdrv_info(Monitor *mon)
1143 {
1144     BlockDriverState *bs;
1145
1146     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1147         monitor_printf(mon, "%s:", bs->device_name);
1148         monitor_printf(mon, " type=");
1149         switch(bs->type) {
1150         case BDRV_TYPE_HD:
1151             monitor_printf(mon, "hd");
1152             break;
1153         case BDRV_TYPE_CDROM:
1154             monitor_printf(mon, "cdrom");
1155             break;
1156         case BDRV_TYPE_FLOPPY:
1157             monitor_printf(mon, "floppy");
1158             break;
1159         }
1160         monitor_printf(mon, " removable=%d", bs->removable);
1161         if (bs->removable) {
1162             monitor_printf(mon, " locked=%d", bs->locked);
1163         }
1164         if (bs->drv) {
1165             monitor_printf(mon, " file=");
1166             monitor_print_filename(mon, bs->filename);
1167             if (bs->backing_file[0] != '\0') {
1168                 monitor_printf(mon, " backing_file=");
1169                 monitor_print_filename(mon, bs->backing_file);
1170             }
1171             monitor_printf(mon, " ro=%d", bs->read_only);
1172             monitor_printf(mon, " drv=%s", bs->drv->format_name);
1173             monitor_printf(mon, " encrypted=%d", bdrv_is_encrypted(bs));
1174         } else {
1175             monitor_printf(mon, " [not inserted]");
1176         }
1177         monitor_printf(mon, "\n");
1178     }
1179 }
1180
1181 /* The "info blockstats" command. */
1182 void bdrv_info_stats(Monitor *mon)
1183 {
1184     BlockDriverState *bs;
1185
1186     for (bs = bdrv_first; bs != NULL; bs = bs->next) {
1187         monitor_printf(mon, "%s:"
1188                        " rd_bytes=%" PRIu64
1189                        " wr_bytes=%" PRIu64
1190                        " rd_operations=%" PRIu64
1191                        " wr_operations=%" PRIu64
1192                        "\n",
1193                        bs->device_name,
1194                        bs->rd_bytes, bs->wr_bytes,
1195                        bs->rd_ops, bs->wr_ops);
1196     }
1197 }
1198
1199 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1200 {
1201     if (bs->backing_hd && bs->backing_hd->encrypted)
1202         return bs->backing_file;
1203     else if (bs->encrypted)
1204         return bs->filename;
1205     else
1206         return NULL;
1207 }
1208
1209 void bdrv_get_backing_filename(BlockDriverState *bs,
1210                                char *filename, int filename_size)
1211 {
1212     if (!bs->backing_hd) {
1213         pstrcpy(filename, filename_size, "");
1214     } else {
1215         pstrcpy(filename, filename_size, bs->backing_file);
1216     }
1217 }
1218
1219 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1220                           const uint8_t *buf, int nb_sectors)
1221 {
1222     BlockDriver *drv = bs->drv;
1223     if (!drv)
1224         return -ENOMEDIUM;
1225     if (!drv->bdrv_write_compressed)
1226         return -ENOTSUP;
1227     if (bdrv_check_request(bs, sector_num, nb_sectors))
1228         return -EIO;
1229
1230     if (bs->dirty_bitmap) {
1231         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1232     }
1233
1234     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1235 }
1236
1237 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1238 {
1239     BlockDriver *drv = bs->drv;
1240     if (!drv)
1241         return -ENOMEDIUM;
1242     if (!drv->bdrv_get_info)
1243         return -ENOTSUP;
1244     memset(bdi, 0, sizeof(*bdi));
1245     return drv->bdrv_get_info(bs, bdi);
1246 }
1247
1248 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1249                       int64_t pos, int size)
1250 {
1251     BlockDriver *drv = bs->drv;
1252     if (!drv)
1253         return -ENOMEDIUM;
1254     if (!drv->bdrv_save_vmstate)
1255         return -ENOTSUP;
1256     return drv->bdrv_save_vmstate(bs, buf, pos, size);
1257 }
1258
1259 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1260                       int64_t pos, int size)
1261 {
1262     BlockDriver *drv = bs->drv;
1263     if (!drv)
1264         return -ENOMEDIUM;
1265     if (!drv->bdrv_load_vmstate)
1266         return -ENOTSUP;
1267     return drv->bdrv_load_vmstate(bs, buf, pos, size);
1268 }
1269
1270 /**************************************************************/
1271 /* handling of snapshots */
1272
1273 int bdrv_snapshot_create(BlockDriverState *bs,
1274                          QEMUSnapshotInfo *sn_info)
1275 {
1276     BlockDriver *drv = bs->drv;
1277     if (!drv)
1278         return -ENOMEDIUM;
1279     if (!drv->bdrv_snapshot_create)
1280         return -ENOTSUP;
1281     return drv->bdrv_snapshot_create(bs, sn_info);
1282 }
1283
1284 int bdrv_snapshot_goto(BlockDriverState *bs,
1285                        const char *snapshot_id)
1286 {
1287     BlockDriver *drv = bs->drv;
1288     if (!drv)
1289         return -ENOMEDIUM;
1290     if (!drv->bdrv_snapshot_goto)
1291         return -ENOTSUP;
1292     return drv->bdrv_snapshot_goto(bs, snapshot_id);
1293 }
1294
1295 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1296 {
1297     BlockDriver *drv = bs->drv;
1298     if (!drv)
1299         return -ENOMEDIUM;
1300     if (!drv->bdrv_snapshot_delete)
1301         return -ENOTSUP;
1302     return drv->bdrv_snapshot_delete(bs, snapshot_id);
1303 }
1304
1305 int bdrv_snapshot_list(BlockDriverState *bs,
1306                        QEMUSnapshotInfo **psn_info)
1307 {
1308     BlockDriver *drv = bs->drv;
1309     if (!drv)
1310         return -ENOMEDIUM;
1311     if (!drv->bdrv_snapshot_list)
1312         return -ENOTSUP;
1313     return drv->bdrv_snapshot_list(bs, psn_info);
1314 }
1315
1316 #define NB_SUFFIXES 4
1317
1318 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1319 {
1320     static const char suffixes[NB_SUFFIXES] = "KMGT";
1321     int64_t base;
1322     int i;
1323
1324     if (size <= 999) {
1325         snprintf(buf, buf_size, "%" PRId64, size);
1326     } else {
1327         base = 1024;
1328         for(i = 0; i < NB_SUFFIXES; i++) {
1329             if (size < (10 * base)) {
1330                 snprintf(buf, buf_size, "%0.1f%c",
1331                          (double)size / base,
1332                          suffixes[i]);
1333                 break;
1334             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1335                 snprintf(buf, buf_size, "%" PRId64 "%c",
1336                          ((size + (base >> 1)) / base),
1337                          suffixes[i]);
1338                 break;
1339             }
1340             base = base * 1024;
1341         }
1342     }
1343     return buf;
1344 }
1345
1346 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1347 {
1348     char buf1[128], date_buf[128], clock_buf[128];
1349 #ifdef _WIN32
1350     struct tm *ptm;
1351 #else
1352     struct tm tm;
1353 #endif
1354     time_t ti;
1355     int64_t secs;
1356
1357     if (!sn) {
1358         snprintf(buf, buf_size,
1359                  "%-10s%-20s%7s%20s%15s",
1360                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1361     } else {
1362         ti = sn->date_sec;
1363 #ifdef _WIN32
1364         ptm = localtime(&ti);
1365         strftime(date_buf, sizeof(date_buf),
1366                  "%Y-%m-%d %H:%M:%S", ptm);
1367 #else
1368         localtime_r(&ti, &tm);
1369         strftime(date_buf, sizeof(date_buf),
1370                  "%Y-%m-%d %H:%M:%S", &tm);
1371 #endif
1372         secs = sn->vm_clock_nsec / 1000000000;
1373         snprintf(clock_buf, sizeof(clock_buf),
1374                  "%02d:%02d:%02d.%03d",
1375                  (int)(secs / 3600),
1376                  (int)((secs / 60) % 60),
1377                  (int)(secs % 60),
1378                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1379         snprintf(buf, buf_size,
1380                  "%-10s%-20s%7s%20s%15s",
1381                  sn->id_str, sn->name,
1382                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1383                  date_buf,
1384                  clock_buf);
1385     }
1386     return buf;
1387 }
1388
1389
1390 /**************************************************************/
1391 /* async I/Os */
1392
1393 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1394                                  QEMUIOVector *qiov, int nb_sectors,
1395                                  BlockDriverCompletionFunc *cb, void *opaque)
1396 {
1397     BlockDriver *drv = bs->drv;
1398     BlockDriverAIOCB *ret;
1399
1400     if (!drv)
1401         return NULL;
1402     if (bdrv_check_request(bs, sector_num, nb_sectors))
1403         return NULL;
1404
1405     ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1406                               cb, opaque);
1407
1408     if (ret) {
1409         /* Update stats even though technically transfer has not happened. */
1410         bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1411         bs->rd_ops ++;
1412     }
1413
1414     return ret;
1415 }
1416
1417 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1418                                   QEMUIOVector *qiov, int nb_sectors,
1419                                   BlockDriverCompletionFunc *cb, void *opaque)
1420 {
1421     BlockDriver *drv = bs->drv;
1422     BlockDriverAIOCB *ret;
1423
1424     if (!drv)
1425         return NULL;
1426     if (bs->read_only)
1427         return NULL;
1428     if (bdrv_check_request(bs, sector_num, nb_sectors))
1429         return NULL;
1430
1431     if (bs->dirty_bitmap) {
1432         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1433     }
1434
1435     ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1436                                cb, opaque);
1437
1438     if (ret) {
1439         /* Update stats even though technically transfer has not happened. */
1440         bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1441         bs->wr_ops ++;
1442     }
1443
1444     return ret;
1445 }
1446
1447
1448 typedef struct MultiwriteCB {
1449     int error;
1450     int num_requests;
1451     int num_callbacks;
1452     struct {
1453         BlockDriverCompletionFunc *cb;
1454         void *opaque;
1455         QEMUIOVector *free_qiov;
1456         void *free_buf;
1457     } callbacks[];
1458 } MultiwriteCB;
1459
1460 static void multiwrite_user_cb(MultiwriteCB *mcb)
1461 {
1462     int i;
1463
1464     for (i = 0; i < mcb->num_callbacks; i++) {
1465         mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1466         qemu_free(mcb->callbacks[i].free_qiov);
1467         qemu_free(mcb->callbacks[i].free_buf);
1468     }
1469 }
1470
1471 static void multiwrite_cb(void *opaque, int ret)
1472 {
1473     MultiwriteCB *mcb = opaque;
1474
1475     if (ret < 0) {
1476         mcb->error = ret;
1477         multiwrite_user_cb(mcb);
1478     }
1479
1480     mcb->num_requests--;
1481     if (mcb->num_requests == 0) {
1482         if (mcb->error == 0) {
1483             multiwrite_user_cb(mcb);
1484         }
1485         qemu_free(mcb);
1486     }
1487 }
1488
1489 static int multiwrite_req_compare(const void *a, const void *b)
1490 {
1491     return (((BlockRequest*) a)->sector - ((BlockRequest*) b)->sector);
1492 }
1493
1494 /*
1495  * Takes a bunch of requests and tries to merge them. Returns the number of
1496  * requests that remain after merging.
1497  */
1498 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
1499     int num_reqs, MultiwriteCB *mcb)
1500 {
1501     int i, outidx;
1502
1503     // Sort requests by start sector
1504     qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
1505
1506     // Check if adjacent requests touch the same clusters. If so, combine them,
1507     // filling up gaps with zero sectors.
1508     outidx = 0;
1509     for (i = 1; i < num_reqs; i++) {
1510         int merge = 0;
1511         int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
1512
1513         // This handles the cases that are valid for all block drivers, namely
1514         // exactly sequential writes and overlapping writes.
1515         if (reqs[i].sector <= oldreq_last) {
1516             merge = 1;
1517         }
1518
1519         // The block driver may decide that it makes sense to combine requests
1520         // even if there is a gap of some sectors between them. In this case,
1521         // the gap is filled with zeros (therefore only applicable for yet
1522         // unused space in format like qcow2).
1523         if (!merge && bs->drv->bdrv_merge_requests) {
1524             merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
1525         }
1526
1527         if (merge) {
1528             size_t size;
1529             QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
1530             qemu_iovec_init(qiov,
1531                 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
1532
1533             // Add the first request to the merged one. If the requests are
1534             // overlapping, drop the last sectors of the first request.
1535             size = (reqs[i].sector - reqs[outidx].sector) << 9;
1536             qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
1537
1538             // We might need to add some zeros between the two requests
1539             if (reqs[i].sector > oldreq_last) {
1540                 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
1541                 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
1542                 memset(buf, 0, zero_bytes);
1543                 qemu_iovec_add(qiov, buf, zero_bytes);
1544                 mcb->callbacks[i].free_buf = buf;
1545             }
1546
1547             // Add the second request
1548             qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
1549
1550             reqs[outidx].nb_sectors += reqs[i].nb_sectors;
1551             reqs[outidx].qiov = qiov;
1552
1553             mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
1554         } else {
1555             outidx++;
1556             reqs[outidx].sector     = reqs[i].sector;
1557             reqs[outidx].nb_sectors = reqs[i].nb_sectors;
1558             reqs[outidx].qiov       = reqs[i].qiov;
1559         }
1560     }
1561
1562     return outidx + 1;
1563 }
1564
1565 /*
1566  * Submit multiple AIO write requests at once.
1567  *
1568  * On success, the function returns 0 and all requests in the reqs array have
1569  * been submitted. In error case this function returns -1, and any of the
1570  * requests may or may not be submitted yet. In particular, this means that the
1571  * callback will be called for some of the requests, for others it won't. The
1572  * caller must check the error field of the BlockRequest to wait for the right
1573  * callbacks (if error != 0, no callback will be called).
1574  *
1575  * The implementation may modify the contents of the reqs array, e.g. to merge
1576  * requests. However, the fields opaque and error are left unmodified as they
1577  * are used to signal failure for a single request to the caller.
1578  */
1579 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
1580 {
1581     BlockDriverAIOCB *acb;
1582     MultiwriteCB *mcb;
1583     int i;
1584
1585     if (num_reqs == 0) {
1586         return 0;
1587     }
1588
1589     // Create MultiwriteCB structure
1590     mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
1591     mcb->num_requests = 0;
1592     mcb->num_callbacks = num_reqs;
1593
1594     for (i = 0; i < num_reqs; i++) {
1595         mcb->callbacks[i].cb = reqs[i].cb;
1596         mcb->callbacks[i].opaque = reqs[i].opaque;
1597     }
1598
1599     // Check for mergable requests
1600     num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
1601
1602     // Run the aio requests
1603     for (i = 0; i < num_reqs; i++) {
1604         acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
1605             reqs[i].nb_sectors, multiwrite_cb, mcb);
1606
1607         if (acb == NULL) {
1608             // We can only fail the whole thing if no request has been
1609             // submitted yet. Otherwise we'll wait for the submitted AIOs to
1610             // complete and report the error in the callback.
1611             if (mcb->num_requests == 0) {
1612                 reqs[i].error = EIO;
1613                 goto fail;
1614             } else {
1615                 mcb->error = EIO;
1616                 break;
1617             }
1618         } else {
1619             mcb->num_requests++;
1620         }
1621     }
1622
1623     return 0;
1624
1625 fail:
1626     free(mcb);
1627     return -1;
1628 }
1629
1630 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
1631         BlockDriverCompletionFunc *cb, void *opaque)
1632 {
1633     BlockDriver *drv = bs->drv;
1634
1635     if (!drv)
1636         return NULL;
1637
1638     /*
1639      * Note that unlike bdrv_flush the driver is reponsible for flushing a
1640      * backing image if it exists.
1641      */
1642     return drv->bdrv_aio_flush(bs, cb, opaque);
1643 }
1644
1645 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
1646 {
1647     acb->pool->cancel(acb);
1648 }
1649
1650
1651 /**************************************************************/
1652 /* async block device emulation */
1653
1654 typedef struct BlockDriverAIOCBSync {
1655     BlockDriverAIOCB common;
1656     QEMUBH *bh;
1657     int ret;
1658     /* vector translation state */
1659     QEMUIOVector *qiov;
1660     uint8_t *bounce;
1661     int is_write;
1662 } BlockDriverAIOCBSync;
1663
1664 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
1665 {
1666     BlockDriverAIOCBSync *acb = (BlockDriverAIOCBSync *)blockacb;
1667     qemu_bh_delete(acb->bh);
1668     acb->bh = NULL;
1669     qemu_aio_release(acb);
1670 }
1671
1672 static AIOPool bdrv_em_aio_pool = {
1673     .aiocb_size         = sizeof(BlockDriverAIOCBSync),
1674     .cancel             = bdrv_aio_cancel_em,
1675 };
1676
1677 static void bdrv_aio_bh_cb(void *opaque)
1678 {
1679     BlockDriverAIOCBSync *acb = opaque;
1680
1681     if (!acb->is_write)
1682         qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
1683     qemu_vfree(acb->bounce);
1684     acb->common.cb(acb->common.opaque, acb->ret);
1685     qemu_bh_delete(acb->bh);
1686     acb->bh = NULL;
1687     qemu_aio_release(acb);
1688 }
1689
1690 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
1691                                             int64_t sector_num,
1692                                             QEMUIOVector *qiov,
1693                                             int nb_sectors,
1694                                             BlockDriverCompletionFunc *cb,
1695                                             void *opaque,
1696                                             int is_write)
1697
1698 {
1699     BlockDriverAIOCBSync *acb;
1700
1701     acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1702     acb->is_write = is_write;
1703     acb->qiov = qiov;
1704     acb->bounce = qemu_blockalign(bs, qiov->size);
1705
1706     if (!acb->bh)
1707         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1708
1709     if (is_write) {
1710         qemu_iovec_to_buffer(acb->qiov, acb->bounce);
1711         acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
1712     } else {
1713         acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
1714     }
1715
1716     qemu_bh_schedule(acb->bh);
1717
1718     return &acb->common;
1719 }
1720
1721 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
1722         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1723         BlockDriverCompletionFunc *cb, void *opaque)
1724 {
1725     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
1726 }
1727
1728 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
1729         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1730         BlockDriverCompletionFunc *cb, void *opaque)
1731 {
1732     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
1733 }
1734
1735 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
1736         BlockDriverCompletionFunc *cb, void *opaque)
1737 {
1738     BlockDriverAIOCBSync *acb;
1739
1740     acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
1741     acb->is_write = 1; /* don't bounce in the completion hadler */
1742     acb->qiov = NULL;
1743     acb->bounce = NULL;
1744     acb->ret = 0;
1745
1746     if (!acb->bh)
1747         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
1748
1749     bdrv_flush(bs);
1750     qemu_bh_schedule(acb->bh);
1751     return &acb->common;
1752 }
1753
1754 /**************************************************************/
1755 /* sync block device emulation */
1756
1757 static void bdrv_rw_em_cb(void *opaque, int ret)
1758 {
1759     *(int *)opaque = ret;
1760 }
1761
1762 #define NOT_DONE 0x7fffffff
1763
1764 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
1765                         uint8_t *buf, int nb_sectors)
1766 {
1767     int async_ret;
1768     BlockDriverAIOCB *acb;
1769     struct iovec iov;
1770     QEMUIOVector qiov;
1771
1772     async_context_push();
1773
1774     async_ret = NOT_DONE;
1775     iov.iov_base = (void *)buf;
1776     iov.iov_len = nb_sectors * 512;
1777     qemu_iovec_init_external(&qiov, &iov, 1);
1778     acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
1779         bdrv_rw_em_cb, &async_ret);
1780     if (acb == NULL) {
1781         async_ret = -1;
1782         goto fail;
1783     }
1784
1785     while (async_ret == NOT_DONE) {
1786         qemu_aio_wait();
1787     }
1788
1789
1790 fail:
1791     async_context_pop();
1792     return async_ret;
1793 }
1794
1795 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
1796                          const uint8_t *buf, int nb_sectors)
1797 {
1798     int async_ret;
1799     BlockDriverAIOCB *acb;
1800     struct iovec iov;
1801     QEMUIOVector qiov;
1802
1803     async_context_push();
1804
1805     async_ret = NOT_DONE;
1806     iov.iov_base = (void *)buf;
1807     iov.iov_len = nb_sectors * 512;
1808     qemu_iovec_init_external(&qiov, &iov, 1);
1809     acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
1810         bdrv_rw_em_cb, &async_ret);
1811     if (acb == NULL) {
1812         async_ret = -1;
1813         goto fail;
1814     }
1815     while (async_ret == NOT_DONE) {
1816         qemu_aio_wait();
1817     }
1818
1819 fail:
1820     async_context_pop();
1821     return async_ret;
1822 }
1823
1824 void bdrv_init(void)
1825 {
1826     module_call_init(MODULE_INIT_BLOCK);
1827 }
1828
1829 void bdrv_init_with_whitelist(void)
1830 {
1831     use_bdrv_whitelist = 1;
1832     bdrv_init();
1833 }
1834
1835 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
1836                    BlockDriverCompletionFunc *cb, void *opaque)
1837 {
1838     BlockDriverAIOCB *acb;
1839
1840     if (pool->free_aiocb) {
1841         acb = pool->free_aiocb;
1842         pool->free_aiocb = acb->next;
1843     } else {
1844         acb = qemu_mallocz(pool->aiocb_size);
1845         acb->pool = pool;
1846     }
1847     acb->bs = bs;
1848     acb->cb = cb;
1849     acb->opaque = opaque;
1850     return acb;
1851 }
1852
1853 void qemu_aio_release(void *p)
1854 {
1855     BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
1856     AIOPool *pool = acb->pool;
1857     acb->next = pool->free_aiocb;
1858     pool->free_aiocb = acb;
1859 }
1860
1861 /**************************************************************/
1862 /* removable device support */
1863
1864 /**
1865  * Return TRUE if the media is present
1866  */
1867 int bdrv_is_inserted(BlockDriverState *bs)
1868 {
1869     BlockDriver *drv = bs->drv;
1870     int ret;
1871     if (!drv)
1872         return 0;
1873     if (!drv->bdrv_is_inserted)
1874         return 1;
1875     ret = drv->bdrv_is_inserted(bs);
1876     return ret;
1877 }
1878
1879 /**
1880  * Return TRUE if the media changed since the last call to this
1881  * function. It is currently only used for floppy disks
1882  */
1883 int bdrv_media_changed(BlockDriverState *bs)
1884 {
1885     BlockDriver *drv = bs->drv;
1886     int ret;
1887
1888     if (!drv || !drv->bdrv_media_changed)
1889         ret = -ENOTSUP;
1890     else
1891         ret = drv->bdrv_media_changed(bs);
1892     if (ret == -ENOTSUP)
1893         ret = bs->media_changed;
1894     bs->media_changed = 0;
1895     return ret;
1896 }
1897
1898 /**
1899  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
1900  */
1901 int bdrv_eject(BlockDriverState *bs, int eject_flag)
1902 {
1903     BlockDriver *drv = bs->drv;
1904     int ret;
1905
1906     if (bs->locked) {
1907         return -EBUSY;
1908     }
1909
1910     if (!drv || !drv->bdrv_eject) {
1911         ret = -ENOTSUP;
1912     } else {
1913         ret = drv->bdrv_eject(bs, eject_flag);
1914     }
1915     if (ret == -ENOTSUP) {
1916         if (eject_flag)
1917             bdrv_close(bs);
1918         ret = 0;
1919     }
1920
1921     return ret;
1922 }
1923
1924 int bdrv_is_locked(BlockDriverState *bs)
1925 {
1926     return bs->locked;
1927 }
1928
1929 /**
1930  * Lock or unlock the media (if it is locked, the user won't be able
1931  * to eject it manually).
1932  */
1933 void bdrv_set_locked(BlockDriverState *bs, int locked)
1934 {
1935     BlockDriver *drv = bs->drv;
1936
1937     bs->locked = locked;
1938     if (drv && drv->bdrv_set_locked) {
1939         drv->bdrv_set_locked(bs, locked);
1940     }
1941 }
1942
1943 /* needed for generic scsi interface */
1944
1945 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1946 {
1947     BlockDriver *drv = bs->drv;
1948
1949     if (drv && drv->bdrv_ioctl)
1950         return drv->bdrv_ioctl(bs, req, buf);
1951     return -ENOTSUP;
1952 }
1953
1954 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
1955         unsigned long int req, void *buf,
1956         BlockDriverCompletionFunc *cb, void *opaque)
1957 {
1958     BlockDriver *drv = bs->drv;
1959
1960     if (drv && drv->bdrv_aio_ioctl)
1961         return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
1962     return NULL;
1963 }
1964
1965
1966
1967 void *qemu_blockalign(BlockDriverState *bs, size_t size)
1968 {
1969     return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
1970 }
1971
1972 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
1973 {
1974     int64_t bitmap_size;
1975
1976     if (enable) {
1977         if (!bs->dirty_bitmap) {
1978             bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
1979                     BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
1980             bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
1981
1982             bs->dirty_bitmap = qemu_mallocz(bitmap_size);
1983         }
1984     } else {
1985         if (bs->dirty_bitmap) {
1986             qemu_free(bs->dirty_bitmap);
1987             bs->dirty_bitmap = NULL;
1988         }
1989     }
1990 }
1991
1992 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
1993 {
1994     int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
1995
1996     if (bs->dirty_bitmap &&
1997         (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
1998         return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
1999             (1 << (chunk % (sizeof(unsigned long) * 8)));
2000     } else {
2001         return 0;
2002     }
2003 }
2004
2005 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2006                       int nr_sectors)
2007 {
2008     set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2009 }