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 #include "qemu-objects.h"
  30
  31 #ifdef CONFIG_BSD
  32 #include <sys/types.h>
  33 #include <sys/stat.h>
  34 #include <sys/ioctl.h>
  35 #include <sys/queue.h>
  36 #ifndef __DragonFly__
  37 #include <sys/disk.h>
  38 #endif
  39 #endif
  40
  41 #ifdef _WIN32
  42 #include <windows.h>
  43 #endif
  44
  45 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
  46         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
  47         BlockDriverCompletionFunc *cb, void *opaque);
  48 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
  49         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
  50         BlockDriverCompletionFunc *cb, void *opaque);
  51 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
  52         BlockDriverCompletionFunc *cb, void *opaque);
  53 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
  54         BlockDriverCompletionFunc *cb, void *opaque);
  55 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
  56                         uint8_t *buf, int nb_sectors);
  57 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
  58                          const uint8_t *buf, int nb_sectors);
  59
  60 static QTAILQ_HEAD(, BlockDriverState) bdrv_states =
  61     QTAILQ_HEAD_INITIALIZER(bdrv_states);
  62
  63 static QLIST_HEAD(, BlockDriver) bdrv_drivers =
  64     QLIST_HEAD_INITIALIZER(bdrv_drivers);
  65
  66 /* If non-zero, use only whitelisted block drivers */
  67 static int use_bdrv_whitelist;
  68
  69 int path_is_absolute(const char *path)
  70 {
  71     const char *p;
  72 #ifdef _WIN32
  73     /* specific case for names like: "\\.\d:" */
  74     if (*path == '/' || *path == '\\')
  75         return 1;
  76 #endif
  77     p = strchr(path, ':');
  78     if (p)
  79         p++;
  80     else
  81         p = path;
  82 #ifdef _WIN32
  83     return (*p == '/' || *p == '\\');
  84 #else
  85     return (*p == '/');
  86 #endif
  87 }
  88
  89 /* if filename is absolute, just copy it to dest. Otherwise, build a
  90    path to it by considering it is relative to base_path. URL are
  91    supported. */
  92 void path_combine(char *dest, int dest_size,
  93                   const char *base_path,
  94                   const char *filename)
  95 {
  96     const char *p, *p1;
  97     int len;
  98
  99     if (dest_size <= 0)
 100         return;
 101     if (path_is_absolute(filename)) {
 102         pstrcpy(dest, dest_size, filename);
 103     } else {
 104         p = strchr(base_path, ':');
 105         if (p)
 106             p++;
 107         else
 108             p = base_path;
 109         p1 = strrchr(base_path, '/');
 110 #ifdef _WIN32
 111         {
 112             const char *p2;
 113             p2 = strrchr(base_path, '\\');
 114             if (!p1 || p2 > p1)
 115                 p1 = p2;
 116         }
 117 #endif
 118         if (p1)
 119             p1++;
 120         else
 121             p1 = base_path;
 122         if (p1 > p)
 123             p = p1;
 124         len = p - base_path;
 125         if (len > dest_size - 1)
 126             len = dest_size - 1;
 127         memcpy(dest, base_path, len);
 128         dest[len] = '\0';
 129         pstrcat(dest, dest_size, filename);
 130     }
 131 }
 132
 133 void bdrv_register(BlockDriver *bdrv)
 134 {
 135     if (!bdrv->bdrv_aio_readv) {
 136         /* add AIO emulation layer */
 137         bdrv->bdrv_aio_readv = bdrv_aio_readv_em;
 138         bdrv->bdrv_aio_writev = bdrv_aio_writev_em;
 139     } else if (!bdrv->bdrv_read) {
 140         /* add synchronous IO emulation layer */
 141         bdrv->bdrv_read = bdrv_read_em;
 142         bdrv->bdrv_write = bdrv_write_em;
 143     }
 144
 145     if (!bdrv->bdrv_aio_flush)
 146         bdrv->bdrv_aio_flush = bdrv_aio_flush_em;
 147
 148     QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list);
 149 }
 150
 151 /* create a new block device (by default it is empty) */
 152 BlockDriverState *bdrv_new(const char *device_name)
 153 {
 154     BlockDriverState *bs;
 155
 156     bs = qemu_mallocz(sizeof(BlockDriverState));
 157     pstrcpy(bs->device_name, sizeof(bs->device_name), device_name);
 158     if (device_name[0] != '\0') {
 159         QTAILQ_INSERT_TAIL(&bdrv_states, bs, list);
 160     }
 161     return bs;
 162 }
 163
 164 BlockDriver *bdrv_find_format(const char *format_name)
 165 {
 166     BlockDriver *drv1;
 167     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
 168         if (!strcmp(drv1->format_name, format_name)) {
 169             return drv1;
 170         }
 171     }
 172     return NULL;
 173 }
 174
 175 static int bdrv_is_whitelisted(BlockDriver *drv)
 176 {
 177     static const char *whitelist[] = {
 178         CONFIG_BDRV_WHITELIST
 179     };
 180     const char **p;
 181
 182     if (!whitelist[0])
 183         return 1;               /* no whitelist, anything goes */
 184
 185     for (p = whitelist; *p; p++) {
 186         if (!strcmp(drv->format_name, *p)) {
 187             return 1;
 188         }
 189     }
 190     return 0;
 191 }
 192
 193 BlockDriver *bdrv_find_whitelisted_format(const char *format_name)
 194 {
 195     BlockDriver *drv = bdrv_find_format(format_name);
 196     return drv && bdrv_is_whitelisted(drv) ? drv : NULL;
 197 }
 198
 199 int bdrv_create(BlockDriver *drv, const char* filename,
 200     QEMUOptionParameter *options)
 201 {
 202     if (!drv->bdrv_create)
 203         return -ENOTSUP;
 204
 205     return drv->bdrv_create(filename, options);
 206 }
 207
 208 int bdrv_create_file(const char* filename, QEMUOptionParameter *options)
 209 {
 210     BlockDriver *drv;
 211
 212     drv = bdrv_find_protocol(filename);
 213     if (drv == NULL) {
 214         drv = bdrv_find_format("file");
 215     }
 216
 217     return bdrv_create(drv, filename, options);
 218 }
 219
 220 #ifdef _WIN32
 221 void get_tmp_filename(char *filename, int size)
 222 {
 223     char temp_dir[MAX_PATH];
 224
 225     GetTempPath(MAX_PATH, temp_dir);
 226     GetTempFileName(temp_dir, "qem", 0, filename);
 227 }
 228 #else
 229 void get_tmp_filename(char *filename, int size)
 230 {
 231     int fd;
 232     const char *tmpdir;
 233     /* XXX: race condition possible */
 234     tmpdir = getenv("TMPDIR");
 235     if (!tmpdir)
 236         tmpdir = "/tmp";
 237     snprintf(filename, size, "%s/vl.XXXXXX", tmpdir);
 238     fd = mkstemp(filename);
 239     close(fd);
 240 }
 241 #endif
 242
 243 #ifdef _WIN32
 244 static int is_windows_drive_prefix(const char *filename)
 245 {
 246     return (((filename[0] >= 'a' && filename[0] <= 'z') ||
 247              (filename[0] >= 'A' && filename[0] <= 'Z')) &&
 248             filename[1] == ':');
 249 }
 250
 251 int is_windows_drive(const char *filename)
 252 {
 253     if (is_windows_drive_prefix(filename) &&
 254         filename[2] == '\0')
 255         return 1;
 256     if (strstart(filename, "\\\\.\\", NULL) ||
 257         strstart(filename, "//./", NULL))
 258         return 1;
 259     return 0;
 260 }
 261 #endif
 262
 263 /*
 264  * Detect host devices. By convention, /dev/cdrom[N] is always
 265  * recognized as a host CDROM.
 266  */
 267 static BlockDriver *find_hdev_driver(const char *filename)
 268 {
 269     int score_max = 0, score;
 270     BlockDriver *drv = NULL, *d;
 271
 272     QLIST_FOREACH(d, &bdrv_drivers, list) {
 273         if (d->bdrv_probe_device) {
 274             score = d->bdrv_probe_device(filename);
 275             if (score > score_max) {
 276                 score_max = score;
 277                 drv = d;
 278             }
 279         }
 280     }
 281
 282     return drv;
 283 }
 284
 285 BlockDriver *bdrv_find_protocol(const char *filename)
 286 {
 287     BlockDriver *drv1;
 288     char protocol[128];
 289     int len;
 290     const char *p;
 291     int is_drive;
 292
 293     /* TODO Drivers without bdrv_file_open must be specified explicitly */
 294
 295 #ifdef _WIN32
 296     is_drive = is_windows_drive(filename) ||
 297         is_windows_drive_prefix(filename);
 298 #else
 299     is_drive = 0;
 300 #endif
 301     p = strchr(filename, ':');
 302     if (!p || is_drive) {
 303         drv1 = find_hdev_driver(filename);
 304         if (!drv1) {
 305             drv1 = bdrv_find_format("file");
 306         }
 307         return drv1;
 308     }
 309     len = p - filename;
 310     if (len > sizeof(protocol) - 1)
 311         len = sizeof(protocol) - 1;
 312     memcpy(protocol, filename, len);
 313     protocol[len] = '\0';
 314     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
 315         if (drv1->protocol_name &&
 316             !strcmp(drv1->protocol_name, protocol)) {
 317             return drv1;
 318         }
 319     }
 320     return NULL;
 321 }
 322
 323 static BlockDriver *find_image_format(const char *filename)
 324 {
 325     int ret, score, score_max;
 326     BlockDriver *drv1, *drv;
 327     uint8_t buf[2048];
 328     BlockDriverState *bs;
 329
 330     ret = bdrv_file_open(&bs, filename, 0);
 331     if (ret < 0)
 332         return NULL;
 333
 334     /* Return the raw BlockDriver * to scsi-generic devices or empty drives */
 335     if (bs->sg || !bdrv_is_inserted(bs)) {
 336         bdrv_delete(bs);
 337         return bdrv_find_format("raw");
 338     }
 339
 340     ret = bdrv_pread(bs, 0, buf, sizeof(buf));
 341     bdrv_delete(bs);
 342     if (ret < 0) {
 343         return NULL;
 344     }
 345
 346     score_max = 0;
 347     drv = NULL;
 348     QLIST_FOREACH(drv1, &bdrv_drivers, list) {
 349         if (drv1->bdrv_probe) {
 350             score = drv1->bdrv_probe(buf, ret, filename);
 351             if (score > score_max) {
 352                 score_max = score;
 353                 drv = drv1;
 354             }
 355         }
 356     }
 357     return drv;
 358 }
 359
 360 /**
 361  * Set the current 'total_sectors' value
 362  */
 363 static int refresh_total_sectors(BlockDriverState *bs, int64_t hint)
 364 {
 365     BlockDriver *drv = bs->drv;
 366
 367     /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */
 368     if (bs->sg)
 369         return 0;
 370
 371     /* query actual device if possible, otherwise just trust the hint */
 372     if (drv->bdrv_getlength) {
 373         int64_t length = drv->bdrv_getlength(bs);
 374         if (length < 0) {
 375             return length;
 376         }
 377         hint = length >> BDRV_SECTOR_BITS;
 378     }
 379
 380     bs->total_sectors = hint;
 381     return 0;
 382 }
 383
 384 /*
 385  * Common part for opening disk images and files
 386  */
 387 static int bdrv_open_common(BlockDriverState *bs, const char *filename,
 388     int flags, BlockDriver *drv)
 389 {
 390     int ret, open_flags;
 391
 392     assert(drv != NULL);
 393
 394     bs->file = NULL;
 395     bs->total_sectors = 0;
 396     bs->is_temporary = 0;
 397     bs->encrypted = 0;
 398     bs->valid_key = 0;
 399     bs->open_flags = flags;
 400     /* buffer_alignment defaulted to 512, drivers can change this value */
 401     bs->buffer_alignment = 512;
 402
 403     pstrcpy(bs->filename, sizeof(bs->filename), filename);
 404
 405     if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) {
 406         return -ENOTSUP;
 407     }
 408
 409     bs->drv = drv;
 410     bs->opaque = qemu_mallocz(drv->instance_size);
 411
 412     /*
 413      * Yes, BDRV_O_NOCACHE aka O_DIRECT means we have to present a
 414      * write cache to the guest.  We do need the fdatasync to flush
 415      * out transactions for block allocations, and we maybe have a
 416      * volatile write cache in our backing device to deal with.
 417      */
 418     if (flags & (BDRV_O_CACHE_WB|BDRV_O_NOCACHE))
 419         bs->enable_write_cache = 1;
 420
 421     /*
 422      * Clear flags that are internal to the block layer before opening the
 423      * image.
 424      */
 425     open_flags = flags & ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
 426
 427     /*
 428      * Snapshots should be writeable.
 429      */
 430     if (bs->is_temporary) {
 431         open_flags |= BDRV_O_RDWR;
 432     }
 433
 434     /* Open the image, either directly or using a protocol */
 435     if (drv->bdrv_file_open) {
 436         ret = drv->bdrv_file_open(bs, filename, open_flags);
 437     } else {
 438         ret = bdrv_file_open(&bs->file, filename, open_flags);
 439         if (ret >= 0) {
 440             ret = drv->bdrv_open(bs, open_flags);
 441         }
 442     }
 443
 444     if (ret < 0) {
 445         goto free_and_fail;
 446     }
 447
 448     bs->keep_read_only = bs->read_only = !(open_flags & BDRV_O_RDWR);
 449
 450     ret = refresh_total_sectors(bs, bs->total_sectors);
 451     if (ret < 0) {
 452         goto free_and_fail;
 453     }
 454
 455 #ifndef _WIN32
 456     if (bs->is_temporary) {
 457         unlink(filename);
 458     }
 459 #endif
 460     return 0;
 461
 462 free_and_fail:
 463     if (bs->file) {
 464         bdrv_delete(bs->file);
 465         bs->file = NULL;
 466     }
 467     qemu_free(bs->opaque);
 468     bs->opaque = NULL;
 469     bs->drv = NULL;
 470     return ret;
 471 }
 472
 473 /*
 474  * Opens a file using a protocol (file, host_device, nbd, ...)
 475  */
 476 int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags)
 477 {
 478     BlockDriverState *bs;
 479     BlockDriver *drv;
 480     int ret;
 481
 482     drv = bdrv_find_protocol(filename);
 483     if (!drv) {
 484         return -ENOENT;
 485     }
 486
 487     bs = bdrv_new("");
 488     ret = bdrv_open_common(bs, filename, flags, drv);
 489     if (ret < 0) {
 490         bdrv_delete(bs);
 491         return ret;
 492     }
 493     bs->growable = 1;
 494     *pbs = bs;
 495     return 0;
 496 }
 497
 498 /*
 499  * Opens a disk image (raw, qcow2, vmdk, ...)
 500  */
 501 int bdrv_open(BlockDriverState *bs, const char *filename, int flags,
 502               BlockDriver *drv)
 503 {
 504     int ret;
 505
 506     if (flags & BDRV_O_SNAPSHOT) {
 507         BlockDriverState *bs1;
 508         int64_t total_size;
 509         int is_protocol = 0;
 510         BlockDriver *bdrv_qcow2;
 511         QEMUOptionParameter *options;
 512         char tmp_filename[PATH_MAX];
 513         char backing_filename[PATH_MAX];
 514
 515         /* if snapshot, we create a temporary backing file and open it
 516            instead of opening 'filename' directly */
 517
 518         /* if there is a backing file, use it */
 519         bs1 = bdrv_new("");
 520         ret = bdrv_open(bs1, filename, 0, drv);
 521         if (ret < 0) {
 522             bdrv_delete(bs1);
 523             return ret;
 524         }
 525         total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK;
 526
 527         if (bs1->drv && bs1->drv->protocol_name)
 528             is_protocol = 1;
 529
 530         bdrv_delete(bs1);
 531
 532         get_tmp_filename(tmp_filename, sizeof(tmp_filename));
 533
 534         /* Real path is meaningless for protocols */
 535         if (is_protocol)
 536             snprintf(backing_filename, sizeof(backing_filename),
 537                      "%s", filename);
 538         else if (!realpath(filename, backing_filename))
 539             return -errno;
 540
 541         bdrv_qcow2 = bdrv_find_format("qcow2");
 542         options = parse_option_parameters("", bdrv_qcow2->create_options, NULL);
 543
 544         set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size);
 545         set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename);
 546         if (drv) {
 547             set_option_parameter(options, BLOCK_OPT_BACKING_FMT,
 548                 drv->format_name);
 549         }
 550
 551         ret = bdrv_create(bdrv_qcow2, tmp_filename, options);
 552         free_option_parameters(options);
 553         if (ret < 0) {
 554             return ret;
 555         }
 556
 557         filename = tmp_filename;
 558         drv = bdrv_qcow2;
 559         bs->is_temporary = 1;
 560     }
 561
 562     /* Find the right image format driver */
 563     if (!drv) {
 564         drv = find_image_format(filename);
 565     }
 566
 567     if (!drv) {
 568         ret = -ENOENT;
 569         goto unlink_and_fail;
 570     }
 571
 572     /* Open the image */
 573     ret = bdrv_open_common(bs, filename, flags, drv);
 574     if (ret < 0) {
 575         goto unlink_and_fail;
 576     }
 577
 578     /* If there is a backing file, use it */
 579     if ((flags & BDRV_O_NO_BACKING) == 0 && bs->backing_file[0] != '\0') {
 580         char backing_filename[PATH_MAX];
 581         int back_flags;
 582         BlockDriver *back_drv = NULL;
 583
 584         bs->backing_hd = bdrv_new("");
 585         path_combine(backing_filename, sizeof(backing_filename),
 586                      filename, bs->backing_file);
 587         if (bs->backing_format[0] != '\0')
 588             back_drv = bdrv_find_format(bs->backing_format);
 589
 590         /* backing files always opened read-only */
 591         back_flags =
 592             flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING);
 593
 594         ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv);
 595         if (ret < 0) {
 596             bdrv_close(bs);
 597             return ret;
 598         }
 599         if (bs->is_temporary) {
 600             bs->backing_hd->keep_read_only = !(flags & BDRV_O_RDWR);
 601         } else {
 602             /* base image inherits from "parent" */
 603             bs->backing_hd->keep_read_only = bs->keep_read_only;
 604         }
 605     }
 606
 607     if (!bdrv_key_required(bs)) {
 608         /* call the change callback */
 609         bs->media_changed = 1;
 610         if (bs->change_cb)
 611             bs->change_cb(bs->change_opaque);
 612     }
 613
 614     return 0;
 615
 616 unlink_and_fail:
 617     if (bs->is_temporary) {
 618         unlink(filename);
 619     }
 620     return ret;
 621 }
 622
 623 void bdrv_close(BlockDriverState *bs)
 624 {
 625     if (bs->drv) {
 626         if (bs->backing_hd) {
 627             bdrv_delete(bs->backing_hd);
 628             bs->backing_hd = NULL;
 629         }
 630         bs->drv->bdrv_close(bs);
 631         qemu_free(bs->opaque);
 632 #ifdef _WIN32
 633         if (bs->is_temporary) {
 634             unlink(bs->filename);
 635         }
 636 #endif
 637         bs->opaque = NULL;
 638         bs->drv = NULL;
 639
 640         if (bs->file != NULL) {
 641             bdrv_close(bs->file);
 642         }
 643
 644         /* call the change callback */
 645         bs->media_changed = 1;
 646         if (bs->change_cb)
 647             bs->change_cb(bs->change_opaque);
 648     }
 649 }
 650
 651 void bdrv_close_all(void)
 652 {
 653     BlockDriverState *bs;
 654
 655     QTAILQ_FOREACH(bs, &bdrv_states, list) {
 656         bdrv_close(bs);
 657     }
 658 }
 659
 660 void bdrv_delete(BlockDriverState *bs)
 661 {
 662     /* remove from list, if necessary */
 663     if (bs->device_name[0] != '\0') {
 664         QTAILQ_REMOVE(&bdrv_states, bs, list);
 665     }
 666
 667     bdrv_close(bs);
 668     if (bs->file != NULL) {
 669         bdrv_delete(bs->file);
 670     }
 671
 672     qemu_free(bs);
 673 }
 674
 675 /*
 676  * Run consistency checks on an image
 677  *
 678  * Returns the number of errors or -errno when an internal error occurs
 679  */
 680 int bdrv_check(BlockDriverState *bs)
 681 {
 682     if (bs->drv->bdrv_check == NULL) {
 683         return -ENOTSUP;
 684     }
 685
 686     return bs->drv->bdrv_check(bs);
 687 }
 688
 689 /* commit COW file into the raw image */
 690 int bdrv_commit(BlockDriverState *bs)
 691 {
 692     BlockDriver *drv = bs->drv;
 693     int64_t i, total_sectors;
 694     int n, j, ro, open_flags;
 695     int ret = 0, rw_ret = 0;
 696     unsigned char sector[BDRV_SECTOR_SIZE];
 697     char filename[1024];
 698     BlockDriverState *bs_rw, *bs_ro;
 699
 700     if (!drv)
 701         return -ENOMEDIUM;
 702
 703     if (!bs->backing_hd) {
 704         return -ENOTSUP;
 705     }
 706
 707     if (bs->backing_hd->keep_read_only) {
 708         return -EACCES;
 709     }
 710
 711     ro = bs->backing_hd->read_only;
 712     strncpy(filename, bs->backing_hd->filename, sizeof(filename));
 713     open_flags =  bs->backing_hd->open_flags;
 714
 715     if (ro) {
 716         /* re-open as RW */
 717         bdrv_delete(bs->backing_hd);
 718         bs->backing_hd = NULL;
 719         bs_rw = bdrv_new("");
 720         rw_ret = bdrv_open(bs_rw, filename, open_flags | BDRV_O_RDWR, drv);
 721         if (rw_ret < 0) {
 722             bdrv_delete(bs_rw);
 723             /* try to re-open read-only */
 724             bs_ro = bdrv_new("");
 725             ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
 726             if (ret < 0) {
 727                 bdrv_delete(bs_ro);
 728                 /* drive not functional anymore */
 729                 bs->drv = NULL;
 730                 return ret;
 731             }
 732             bs->backing_hd = bs_ro;
 733             return rw_ret;
 734         }
 735         bs->backing_hd = bs_rw;
 736     }
 737
 738     total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS;
 739     for (i = 0; i < total_sectors;) {
 740         if (drv->bdrv_is_allocated(bs, i, 65536, &n)) {
 741             for(j = 0; j < n; j++) {
 742                 if (bdrv_read(bs, i, sector, 1) != 0) {
 743                     ret = -EIO;
 744                     goto ro_cleanup;
 745                 }
 746
 747                 if (bdrv_write(bs->backing_hd, i, sector, 1) != 0) {
 748                     ret = -EIO;
 749                     goto ro_cleanup;
 750                 }
 751                 i++;
 752             }
 753         } else {
 754             i += n;
 755         }
 756     }
 757
 758     if (drv->bdrv_make_empty) {
 759         ret = drv->bdrv_make_empty(bs);
 760         bdrv_flush(bs);
 761     }
 762
 763     /*
 764      * Make sure all data we wrote to the backing device is actually
 765      * stable on disk.
 766      */
 767     if (bs->backing_hd)
 768         bdrv_flush(bs->backing_hd);
 769
 770 ro_cleanup:
 771
 772     if (ro) {
 773         /* re-open as RO */
 774         bdrv_delete(bs->backing_hd);
 775         bs->backing_hd = NULL;
 776         bs_ro = bdrv_new("");
 777         ret = bdrv_open(bs_ro, filename, open_flags & ~BDRV_O_RDWR, drv);
 778         if (ret < 0) {
 779             bdrv_delete(bs_ro);
 780             /* drive not functional anymore */
 781             bs->drv = NULL;
 782             return ret;
 783         }
 784         bs->backing_hd = bs_ro;
 785         bs->backing_hd->keep_read_only = 0;
 786     }
 787
 788     return ret;
 789 }
 790
 791 void bdrv_commit_all(void)
 792 {
 793     BlockDriverState *bs;
 794
 795     QTAILQ_FOREACH(bs, &bdrv_states, list) {
 796         bdrv_commit(bs);
 797     }
 798 }
 799
 800 /*
 801  * Return values:
 802  * 0        - success
 803  * -EINVAL  - backing format specified, but no file
 804  * -ENOSPC  - can't update the backing file because no space is left in the
 805  *            image file header
 806  * -ENOTSUP - format driver doesn't support changing the backing file
 807  */
 808 int bdrv_change_backing_file(BlockDriverState *bs,
 809     const char *backing_file, const char *backing_fmt)
 810 {
 811     BlockDriver *drv = bs->drv;
 812
 813     if (drv->bdrv_change_backing_file != NULL) {
 814         return drv->bdrv_change_backing_file(bs, backing_file, backing_fmt);
 815     } else {
 816         return -ENOTSUP;
 817     }
 818 }
 819
 820 static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset,
 821                                    size_t size)
 822 {
 823     int64_t len;
 824
 825     if (!bdrv_is_inserted(bs))
 826         return -ENOMEDIUM;
 827
 828     if (bs->growable)
 829         return 0;
 830
 831     len = bdrv_getlength(bs);
 832
 833     if (offset < 0)
 834         return -EIO;
 835
 836     if ((offset > len) || (len - offset < size))
 837         return -EIO;
 838
 839     return 0;
 840 }
 841
 842 static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num,
 843                               int nb_sectors)
 844 {
 845     return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE,
 846                                    nb_sectors * BDRV_SECTOR_SIZE);
 847 }
 848
 849 /* return < 0 if error. See bdrv_write() for the return codes */
 850 int bdrv_read(BlockDriverState *bs, int64_t sector_num,
 851               uint8_t *buf, int nb_sectors)
 852 {
 853     BlockDriver *drv = bs->drv;
 854
 855     if (!drv)
 856         return -ENOMEDIUM;
 857     if (bdrv_check_request(bs, sector_num, nb_sectors))
 858         return -EIO;
 859
 860     return drv->bdrv_read(bs, sector_num, buf, nb_sectors);
 861 }
 862
 863 static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num,
 864                              int nb_sectors, int dirty)
 865 {
 866     int64_t start, end;
 867     unsigned long val, idx, bit;
 868
 869     start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK;
 870     end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK;
 871
 872     for (; start <= end; start++) {
 873         idx = start / (sizeof(unsigned long) * 8);
 874         bit = start % (sizeof(unsigned long) * 8);
 875         val = bs->dirty_bitmap[idx];
 876         if (dirty) {
 877             if (!(val & (1 << bit))) {
 878                 bs->dirty_count++;
 879                 val |= 1 << bit;
 880             }
 881         } else {
 882             if (val & (1 << bit)) {
 883                 bs->dirty_count--;
 884                 val &= ~(1 << bit);
 885             }
 886         }
 887         bs->dirty_bitmap[idx] = val;
 888     }
 889 }
 890
 891 /* Return < 0 if error. Important errors are:
 892   -EIO         generic I/O error (may happen for all errors)
 893   -ENOMEDIUM   No media inserted.
 894   -EINVAL      Invalid sector number or nb_sectors
 895   -EACCES      Trying to write a read-only device
 896 */
 897 int bdrv_write(BlockDriverState *bs, int64_t sector_num,
 898                const uint8_t *buf, int nb_sectors)
 899 {
 900     BlockDriver *drv = bs->drv;
 901     if (!bs->drv)
 902         return -ENOMEDIUM;
 903     if (bs->read_only)
 904         return -EACCES;
 905     if (bdrv_check_request(bs, sector_num, nb_sectors))
 906         return -EIO;
 907
 908     if (bs->dirty_bitmap) {
 909         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
 910     }
 911
 912     if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
 913         bs->wr_highest_sector = sector_num + nb_sectors - 1;
 914     }
 915
 916     return drv->bdrv_write(bs, sector_num, buf, nb_sectors);
 917 }
 918
 919 int bdrv_pread(BlockDriverState *bs, int64_t offset,
 920                void *buf, int count1)
 921 {
 922     uint8_t tmp_buf[BDRV_SECTOR_SIZE];
 923     int len, nb_sectors, count;
 924     int64_t sector_num;
 925     int ret;
 926
 927     count = count1;
 928     /* first read to align to sector start */
 929     len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
 930     if (len > count)
 931         len = count;
 932     sector_num = offset >> BDRV_SECTOR_BITS;
 933     if (len > 0) {
 934         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
 935             return ret;
 936         memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len);
 937         count -= len;
 938         if (count == 0)
 939             return count1;
 940         sector_num++;
 941         buf += len;
 942     }
 943
 944     /* read the sectors "in place" */
 945     nb_sectors = count >> BDRV_SECTOR_BITS;
 946     if (nb_sectors > 0) {
 947         if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0)
 948             return ret;
 949         sector_num += nb_sectors;
 950         len = nb_sectors << BDRV_SECTOR_BITS;
 951         buf += len;
 952         count -= len;
 953     }
 954
 955     /* add data from the last sector */
 956     if (count > 0) {
 957         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
 958             return ret;
 959         memcpy(buf, tmp_buf, count);
 960     }
 961     return count1;
 962 }
 963
 964 int bdrv_pwrite(BlockDriverState *bs, int64_t offset,
 965                 const void *buf, int count1)
 966 {
 967     uint8_t tmp_buf[BDRV_SECTOR_SIZE];
 968     int len, nb_sectors, count;
 969     int64_t sector_num;
 970     int ret;
 971
 972     count = count1;
 973     /* first write to align to sector start */
 974     len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1);
 975     if (len > count)
 976         len = count;
 977     sector_num = offset >> BDRV_SECTOR_BITS;
 978     if (len > 0) {
 979         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
 980             return ret;
 981         memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len);
 982         if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
 983             return ret;
 984         count -= len;
 985         if (count == 0)
 986             return count1;
 987         sector_num++;
 988         buf += len;
 989     }
 990
 991     /* write the sectors "in place" */
 992     nb_sectors = count >> BDRV_SECTOR_BITS;
 993     if (nb_sectors > 0) {
 994         if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0)
 995             return ret;
 996         sector_num += nb_sectors;
 997         len = nb_sectors << BDRV_SECTOR_BITS;
 998         buf += len;
 999         count -= len;
1000     }
1001
1002     /* add data from the last sector */
1003     if (count > 0) {
1004         if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0)
1005             return ret;
1006         memcpy(tmp_buf, buf, count);
1007         if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0)
1008             return ret;
1009     }
1010     return count1;
1011 }
1012
1013 /*
1014  * Writes to the file and ensures that no writes are reordered across this
1015  * request (acts as a barrier)
1016  *
1017  * Returns 0 on success, -errno in error cases.
1018  */
1019 int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset,
1020     const void *buf, int count)
1021 {
1022     int ret;
1023
1024     ret = bdrv_pwrite(bs, offset, buf, count);
1025     if (ret < 0) {
1026         return ret;
1027     }
1028
1029     /* No flush needed for cache=writethrough, it uses O_DSYNC */
1030     if ((bs->open_flags & BDRV_O_CACHE_MASK) != 0) {
1031         bdrv_flush(bs);
1032     }
1033
1034     return 0;
1035 }
1036
1037 /*
1038  * Writes to the file and ensures that no writes are reordered across this
1039  * request (acts as a barrier)
1040  *
1041  * Returns 0 on success, -errno in error cases.
1042  */
1043 int bdrv_write_sync(BlockDriverState *bs, int64_t sector_num,
1044     const uint8_t *buf, int nb_sectors)
1045 {
1046     return bdrv_pwrite_sync(bs, BDRV_SECTOR_SIZE * sector_num,
1047         buf, BDRV_SECTOR_SIZE * nb_sectors);
1048 }
1049
1050 /**
1051  * Truncate file to 'offset' bytes (needed only for file protocols)
1052  */
1053 int bdrv_truncate(BlockDriverState *bs, int64_t offset)
1054 {
1055     BlockDriver *drv = bs->drv;
1056     int ret;
1057     if (!drv)
1058         return -ENOMEDIUM;
1059     if (!drv->bdrv_truncate)
1060         return -ENOTSUP;
1061     if (bs->read_only)
1062         return -EACCES;
1063     ret = drv->bdrv_truncate(bs, offset);
1064     if (ret == 0) {
1065         ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS);
1066     }
1067     return ret;
1068 }
1069
1070 /**
1071  * Length of a file in bytes. Return < 0 if error or unknown.
1072  */
1073 int64_t bdrv_getlength(BlockDriverState *bs)
1074 {
1075     BlockDriver *drv = bs->drv;
1076     if (!drv)
1077         return -ENOMEDIUM;
1078
1079     /* Fixed size devices use the total_sectors value for speed instead of
1080        issuing a length query (like lseek) on each call.  Also, legacy block
1081        drivers don't provide a bdrv_getlength function and must use
1082        total_sectors. */
1083     if (!bs->growable || !drv->bdrv_getlength) {
1084         return bs->total_sectors * BDRV_SECTOR_SIZE;
1085     }
1086     return drv->bdrv_getlength(bs);
1087 }
1088
1089 /* return 0 as number of sectors if no device present or error */
1090 void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr)
1091 {
1092     int64_t length;
1093     length = bdrv_getlength(bs);
1094     if (length < 0)
1095         length = 0;
1096     else
1097         length = length >> BDRV_SECTOR_BITS;
1098     *nb_sectors_ptr = length;
1099 }
1100
1101 struct partition {
1102         uint8_t boot_ind;           /* 0x80 - active */
1103         uint8_t head;               /* starting head */
1104         uint8_t sector;             /* starting sector */
1105         uint8_t cyl;                /* starting cylinder */
1106         uint8_t sys_ind;            /* What partition type */
1107         uint8_t end_head;           /* end head */
1108         uint8_t end_sector;         /* end sector */
1109         uint8_t end_cyl;            /* end cylinder */
1110         uint32_t start_sect;        /* starting sector counting from 0 */
1111         uint32_t nr_sects;          /* nr of sectors in partition */
1112 } __attribute__((packed));
1113
1114 /* try to guess the disk logical geometry from the MSDOS partition table. Return 0 if OK, -1 if could not guess */
1115 static int guess_disk_lchs(BlockDriverState *bs,
1116                            int *pcylinders, int *pheads, int *psectors)
1117 {
1118     uint8_t buf[BDRV_SECTOR_SIZE];
1119     int ret, i, heads, sectors, cylinders;
1120     struct partition *p;
1121     uint32_t nr_sects;
1122     uint64_t nb_sectors;
1123
1124     bdrv_get_geometry(bs, &nb_sectors);
1125
1126     ret = bdrv_read(bs, 0, buf, 1);
1127     if (ret < 0)
1128         return -1;
1129     /* test msdos magic */
1130     if (buf[510] != 0x55 || buf[511] != 0xaa)
1131         return -1;
1132     for(i = 0; i < 4; i++) {
1133         p = ((struct partition *)(buf + 0x1be)) + i;
1134         nr_sects = le32_to_cpu(p->nr_sects);
1135         if (nr_sects && p->end_head) {
1136             /* We make the assumption that the partition terminates on
1137                a cylinder boundary */
1138             heads = p->end_head + 1;
1139             sectors = p->end_sector & 63;
1140             if (sectors == 0)
1141                 continue;
1142             cylinders = nb_sectors / (heads * sectors);
1143             if (cylinders < 1 || cylinders > 16383)
1144                 continue;
1145             *pheads = heads;
1146             *psectors = sectors;
1147             *pcylinders = cylinders;
1148 #if 0
1149             printf("guessed geometry: LCHS=%d %d %d\n",
1150                    cylinders, heads, sectors);
1151 #endif
1152             return 0;
1153         }
1154     }
1155     return -1;
1156 }
1157
1158 void bdrv_guess_geometry(BlockDriverState *bs, int *pcyls, int *pheads, int *psecs)
1159 {
1160     int translation, lba_detected = 0;
1161     int cylinders, heads, secs;
1162     uint64_t nb_sectors;
1163
1164     /* if a geometry hint is available, use it */
1165     bdrv_get_geometry(bs, &nb_sectors);
1166     bdrv_get_geometry_hint(bs, &cylinders, &heads, &secs);
1167     translation = bdrv_get_translation_hint(bs);
1168     if (cylinders != 0) {
1169         *pcyls = cylinders;
1170         *pheads = heads;
1171         *psecs = secs;
1172     } else {
1173         if (guess_disk_lchs(bs, &cylinders, &heads, &secs) == 0) {
1174             if (heads > 16) {
1175                 /* if heads > 16, it means that a BIOS LBA
1176                    translation was active, so the default
1177                    hardware geometry is OK */
1178                 lba_detected = 1;
1179                 goto default_geometry;
1180             } else {
1181                 *pcyls = cylinders;
1182                 *pheads = heads;
1183                 *psecs = secs;
1184                 /* disable any translation to be in sync with
1185                    the logical geometry */
1186                 if (translation == BIOS_ATA_TRANSLATION_AUTO) {
1187                     bdrv_set_translation_hint(bs,
1188                                               BIOS_ATA_TRANSLATION_NONE);
1189                 }
1190             }
1191         } else {
1192         default_geometry:
1193             /* if no geometry, use a standard physical disk geometry */
1194             cylinders = nb_sectors / (16 * 63);
1195
1196             if (cylinders > 16383)
1197                 cylinders = 16383;
1198             else if (cylinders < 2)
1199                 cylinders = 2;
1200             *pcyls = cylinders;
1201             *pheads = 16;
1202             *psecs = 63;
1203             if ((lba_detected == 1) && (translation == BIOS_ATA_TRANSLATION_AUTO)) {
1204                 if ((*pcyls * *pheads) <= 131072) {
1205                     bdrv_set_translation_hint(bs,
1206                                               BIOS_ATA_TRANSLATION_LARGE);
1207                 } else {
1208                     bdrv_set_translation_hint(bs,
1209                                               BIOS_ATA_TRANSLATION_LBA);
1210                 }
1211             }
1212         }
1213         bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
1214     }
1215 }
1216
1217 void bdrv_set_geometry_hint(BlockDriverState *bs,
1218                             int cyls, int heads, int secs)
1219 {
1220     bs->cyls = cyls;
1221     bs->heads = heads;
1222     bs->secs = secs;
1223 }
1224
1225 void bdrv_set_type_hint(BlockDriverState *bs, int type)
1226 {
1227     bs->type = type;
1228     bs->removable = ((type == BDRV_TYPE_CDROM ||
1229                       type == BDRV_TYPE_FLOPPY));
1230 }
1231
1232 void bdrv_set_translation_hint(BlockDriverState *bs, int translation)
1233 {
1234     bs->translation = translation;
1235 }
1236
1237 void bdrv_get_geometry_hint(BlockDriverState *bs,
1238                             int *pcyls, int *pheads, int *psecs)
1239 {
1240     *pcyls = bs->cyls;
1241     *pheads = bs->heads;
1242     *psecs = bs->secs;
1243 }
1244
1245 int bdrv_get_type_hint(BlockDriverState *bs)
1246 {
1247     return bs->type;
1248 }
1249
1250 int bdrv_get_translation_hint(BlockDriverState *bs)
1251 {
1252     return bs->translation;
1253 }
1254
1255 void bdrv_set_on_error(BlockDriverState *bs, BlockErrorAction on_read_error,
1256                        BlockErrorAction on_write_error)
1257 {
1258     bs->on_read_error = on_read_error;
1259     bs->on_write_error = on_write_error;
1260 }
1261
1262 BlockErrorAction bdrv_get_on_error(BlockDriverState *bs, int is_read)
1263 {
1264     return is_read ? bs->on_read_error : bs->on_write_error;
1265 }
1266
1267 int bdrv_is_removable(BlockDriverState *bs)
1268 {
1269     return bs->removable;
1270 }
1271
1272 int bdrv_is_read_only(BlockDriverState *bs)
1273 {
1274     return bs->read_only;
1275 }
1276
1277 int bdrv_is_sg(BlockDriverState *bs)
1278 {
1279     return bs->sg;
1280 }
1281
1282 int bdrv_enable_write_cache(BlockDriverState *bs)
1283 {
1284     return bs->enable_write_cache;
1285 }
1286
1287 /* XXX: no longer used */
1288 void bdrv_set_change_cb(BlockDriverState *bs,
1289                         void (*change_cb)(void *opaque), void *opaque)
1290 {
1291     bs->change_cb = change_cb;
1292     bs->change_opaque = opaque;
1293 }
1294
1295 int bdrv_is_encrypted(BlockDriverState *bs)
1296 {
1297     if (bs->backing_hd && bs->backing_hd->encrypted)
1298         return 1;
1299     return bs->encrypted;
1300 }
1301
1302 int bdrv_key_required(BlockDriverState *bs)
1303 {
1304     BlockDriverState *backing_hd = bs->backing_hd;
1305
1306     if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key)
1307         return 1;
1308     return (bs->encrypted && !bs->valid_key);
1309 }
1310
1311 int bdrv_set_key(BlockDriverState *bs, const char *key)
1312 {
1313     int ret;
1314     if (bs->backing_hd && bs->backing_hd->encrypted) {
1315         ret = bdrv_set_key(bs->backing_hd, key);
1316         if (ret < 0)
1317             return ret;
1318         if (!bs->encrypted)
1319             return 0;
1320     }
1321     if (!bs->encrypted) {
1322         return -EINVAL;
1323     } else if (!bs->drv || !bs->drv->bdrv_set_key) {
1324         return -ENOMEDIUM;
1325     }
1326     ret = bs->drv->bdrv_set_key(bs, key);
1327     if (ret < 0) {
1328         bs->valid_key = 0;
1329     } else if (!bs->valid_key) {
1330         bs->valid_key = 1;
1331         /* call the change callback now, we skipped it on open */
1332         bs->media_changed = 1;
1333         if (bs->change_cb)
1334             bs->change_cb(bs->change_opaque);
1335     }
1336     return ret;
1337 }
1338
1339 void bdrv_get_format(BlockDriverState *bs, char *buf, int buf_size)
1340 {
1341     if (!bs->drv) {
1342         buf[0] = '\0';
1343     } else {
1344         pstrcpy(buf, buf_size, bs->drv->format_name);
1345     }
1346 }
1347
1348 void bdrv_iterate_format(void (*it)(void *opaque, const char *name),
1349                          void *opaque)
1350 {
1351     BlockDriver *drv;
1352
1353     QLIST_FOREACH(drv, &bdrv_drivers, list) {
1354         it(opaque, drv->format_name);
1355     }
1356 }
1357
1358 BlockDriverState *bdrv_find(const char *name)
1359 {
1360     BlockDriverState *bs;
1361
1362     QTAILQ_FOREACH(bs, &bdrv_states, list) {
1363         if (!strcmp(name, bs->device_name)) {
1364             return bs;
1365         }
1366     }
1367     return NULL;
1368 }
1369
1370 BlockDriverState *bdrv_next(BlockDriverState *bs)
1371 {
1372     if (!bs) {
1373         return QTAILQ_FIRST(&bdrv_states);
1374     }
1375     return QTAILQ_NEXT(bs, list);
1376 }
1377
1378 void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque)
1379 {
1380     BlockDriverState *bs;
1381
1382     QTAILQ_FOREACH(bs, &bdrv_states, list) {
1383         it(opaque, bs);
1384     }
1385 }
1386
1387 const char *bdrv_get_device_name(BlockDriverState *bs)
1388 {
1389     return bs->device_name;
1390 }
1391
1392 void bdrv_flush(BlockDriverState *bs)
1393 {
1394     if (bs->open_flags & BDRV_O_NO_FLUSH) {
1395         return;
1396     }
1397
1398     if (bs->drv && bs->drv->bdrv_flush)
1399         bs->drv->bdrv_flush(bs);
1400 }
1401
1402 void bdrv_flush_all(void)
1403 {
1404     BlockDriverState *bs;
1405
1406     QTAILQ_FOREACH(bs, &bdrv_states, list) {
1407         if (bs->drv && !bdrv_is_read_only(bs) &&
1408             (!bdrv_is_removable(bs) || bdrv_is_inserted(bs))) {
1409             bdrv_flush(bs);
1410         }
1411     }
1412 }
1413
1414 int bdrv_has_zero_init(BlockDriverState *bs)
1415 {
1416     assert(bs->drv);
1417
1418     if (bs->drv->no_zero_init) {
1419         return 0;
1420     } else if (bs->file) {
1421         return bdrv_has_zero_init(bs->file);
1422     }
1423
1424     return 1;
1425 }
1426
1427 /*
1428  * Returns true iff the specified sector is present in the disk image. Drivers
1429  * not implementing the functionality are assumed to not support backing files,
1430  * hence all their sectors are reported as allocated.
1431  *
1432  * 'pnum' is set to the number of sectors (including and immediately following
1433  * the specified sector) that are known to be in the same
1434  * allocated/unallocated state.
1435  *
1436  * 'nb_sectors' is the max value 'pnum' should be set to.
1437  */
1438 int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors,
1439         int *pnum)
1440 {
1441     int64_t n;
1442     if (!bs->drv->bdrv_is_allocated) {
1443         if (sector_num >= bs->total_sectors) {
1444             *pnum = 0;
1445             return 0;
1446         }
1447         n = bs->total_sectors - sector_num;
1448         *pnum = (n < nb_sectors) ? (n) : (nb_sectors);
1449         return 1;
1450     }
1451     return bs->drv->bdrv_is_allocated(bs, sector_num, nb_sectors, pnum);
1452 }
1453
1454 void bdrv_mon_event(const BlockDriverState *bdrv,
1455                     BlockMonEventAction action, int is_read)
1456 {
1457     QObject *data;
1458     const char *action_str;
1459
1460     switch (action) {
1461     case BDRV_ACTION_REPORT:
1462         action_str = "report";
1463         break;
1464     case BDRV_ACTION_IGNORE:
1465         action_str = "ignore";
1466         break;
1467     case BDRV_ACTION_STOP:
1468         action_str = "stop";
1469         break;
1470     default:
1471         abort();
1472     }
1473
1474     data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }",
1475                               bdrv->device_name,
1476                               action_str,
1477                               is_read ? "read" : "write");
1478     monitor_protocol_event(QEVENT_BLOCK_IO_ERROR, data);
1479
1480     qobject_decref(data);
1481 }
1482
1483 static void bdrv_print_dict(QObject *obj, void *opaque)
1484 {
1485     QDict *bs_dict;
1486     Monitor *mon = opaque;
1487
1488     bs_dict = qobject_to_qdict(obj);
1489
1490     monitor_printf(mon, "%s: type=%s removable=%d",
1491                         qdict_get_str(bs_dict, "device"),
1492                         qdict_get_str(bs_dict, "type"),
1493                         qdict_get_bool(bs_dict, "removable"));
1494
1495     if (qdict_get_bool(bs_dict, "removable")) {
1496         monitor_printf(mon, " locked=%d", qdict_get_bool(bs_dict, "locked"));
1497     }
1498
1499     if (qdict_haskey(bs_dict, "inserted")) {
1500         QDict *qdict = qobject_to_qdict(qdict_get(bs_dict, "inserted"));
1501
1502         monitor_printf(mon, " file=");
1503         monitor_print_filename(mon, qdict_get_str(qdict, "file"));
1504         if (qdict_haskey(qdict, "backing_file")) {
1505             monitor_printf(mon, " backing_file=");
1506             monitor_print_filename(mon, qdict_get_str(qdict, "backing_file"));
1507         }
1508         monitor_printf(mon, " ro=%d drv=%s encrypted=%d",
1509                             qdict_get_bool(qdict, "ro"),
1510                             qdict_get_str(qdict, "drv"),
1511                             qdict_get_bool(qdict, "encrypted"));
1512     } else {
1513         monitor_printf(mon, " [not inserted]");
1514     }
1515
1516     monitor_printf(mon, "\n");
1517 }
1518
1519 void bdrv_info_print(Monitor *mon, const QObject *data)
1520 {
1521     qlist_iter(qobject_to_qlist(data), bdrv_print_dict, mon);
1522 }
1523
1524 void bdrv_info(Monitor *mon, QObject **ret_data)
1525 {
1526     QList *bs_list;
1527     BlockDriverState *bs;
1528
1529     bs_list = qlist_new();
1530
1531     QTAILQ_FOREACH(bs, &bdrv_states, list) {
1532         QObject *bs_obj;
1533         const char *type = "unknown";
1534
1535         switch(bs->type) {
1536         case BDRV_TYPE_HD:
1537             type = "hd";
1538             break;
1539         case BDRV_TYPE_CDROM:
1540             type = "cdrom";
1541             break;
1542         case BDRV_TYPE_FLOPPY:
1543             type = "floppy";
1544             break;
1545         }
1546
1547         bs_obj = qobject_from_jsonf("{ 'device': %s, 'type': %s, "
1548                                     "'removable': %i, 'locked': %i }",
1549                                     bs->device_name, type, bs->removable,
1550                                     bs->locked);
1551
1552         if (bs->drv) {
1553             QObject *obj;
1554             QDict *bs_dict = qobject_to_qdict(bs_obj);
1555
1556             obj = qobject_from_jsonf("{ 'file': %s, 'ro': %i, 'drv': %s, "
1557                                      "'encrypted': %i }",
1558                                      bs->filename, bs->read_only,
1559                                      bs->drv->format_name,
1560                                      bdrv_is_encrypted(bs));
1561             if (bs->backing_file[0] != '\0') {
1562                 QDict *qdict = qobject_to_qdict(obj);
1563                 qdict_put(qdict, "backing_file",
1564                           qstring_from_str(bs->backing_file));
1565             }
1566
1567             qdict_put_obj(bs_dict, "inserted", obj);
1568         }
1569         qlist_append_obj(bs_list, bs_obj);
1570     }
1571
1572     *ret_data = QOBJECT(bs_list);
1573 }
1574
1575 static void bdrv_stats_iter(QObject *data, void *opaque)
1576 {
1577     QDict *qdict;
1578     Monitor *mon = opaque;
1579
1580     qdict = qobject_to_qdict(data);
1581     monitor_printf(mon, "%s:", qdict_get_str(qdict, "device"));
1582
1583     qdict = qobject_to_qdict(qdict_get(qdict, "stats"));
1584     monitor_printf(mon, " rd_bytes=%" PRId64
1585                         " wr_bytes=%" PRId64
1586                         " rd_operations=%" PRId64
1587                         " wr_operations=%" PRId64
1588                         "\n",
1589                         qdict_get_int(qdict, "rd_bytes"),
1590                         qdict_get_int(qdict, "wr_bytes"),
1591                         qdict_get_int(qdict, "rd_operations"),
1592                         qdict_get_int(qdict, "wr_operations"));
1593 }
1594
1595 void bdrv_stats_print(Monitor *mon, const QObject *data)
1596 {
1597     qlist_iter(qobject_to_qlist(data), bdrv_stats_iter, mon);
1598 }
1599
1600 static QObject* bdrv_info_stats_bs(BlockDriverState *bs)
1601 {
1602     QObject *res;
1603     QDict *dict;
1604
1605     res = qobject_from_jsonf("{ 'stats': {"
1606                              "'rd_bytes': %" PRId64 ","
1607                              "'wr_bytes': %" PRId64 ","
1608                              "'rd_operations': %" PRId64 ","
1609                              "'wr_operations': %" PRId64 ","
1610                              "'wr_highest_offset': %" PRId64
1611                              "} }",
1612                              bs->rd_bytes, bs->wr_bytes,
1613                              bs->rd_ops, bs->wr_ops,
1614                              bs->wr_highest_sector *
1615                              (uint64_t)BDRV_SECTOR_SIZE);
1616     dict  = qobject_to_qdict(res);
1617
1618     if (*bs->device_name) {
1619         qdict_put(dict, "device", qstring_from_str(bs->device_name));
1620     }
1621
1622     if (bs->file) {
1623         QObject *parent = bdrv_info_stats_bs(bs->file);
1624         qdict_put_obj(dict, "parent", parent);
1625     }
1626
1627     return res;
1628 }
1629
1630 void bdrv_info_stats(Monitor *mon, QObject **ret_data)
1631 {
1632     QObject *obj;
1633     QList *devices;
1634     BlockDriverState *bs;
1635
1636     devices = qlist_new();
1637
1638     QTAILQ_FOREACH(bs, &bdrv_states, list) {
1639         obj = bdrv_info_stats_bs(bs);
1640         qlist_append_obj(devices, obj);
1641     }
1642
1643     *ret_data = QOBJECT(devices);
1644 }
1645
1646 const char *bdrv_get_encrypted_filename(BlockDriverState *bs)
1647 {
1648     if (bs->backing_hd && bs->backing_hd->encrypted)
1649         return bs->backing_file;
1650     else if (bs->encrypted)
1651         return bs->filename;
1652     else
1653         return NULL;
1654 }
1655
1656 void bdrv_get_backing_filename(BlockDriverState *bs,
1657                                char *filename, int filename_size)
1658 {
1659     if (!bs->backing_file) {
1660         pstrcpy(filename, filename_size, "");
1661     } else {
1662         pstrcpy(filename, filename_size, bs->backing_file);
1663     }
1664 }
1665
1666 int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num,
1667                           const uint8_t *buf, int nb_sectors)
1668 {
1669     BlockDriver *drv = bs->drv;
1670     if (!drv)
1671         return -ENOMEDIUM;
1672     if (!drv->bdrv_write_compressed)
1673         return -ENOTSUP;
1674     if (bdrv_check_request(bs, sector_num, nb_sectors))
1675         return -EIO;
1676
1677     if (bs->dirty_bitmap) {
1678         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1679     }
1680
1681     return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors);
1682 }
1683
1684 int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1685 {
1686     BlockDriver *drv = bs->drv;
1687     if (!drv)
1688         return -ENOMEDIUM;
1689     if (!drv->bdrv_get_info)
1690         return -ENOTSUP;
1691     memset(bdi, 0, sizeof(*bdi));
1692     return drv->bdrv_get_info(bs, bdi);
1693 }
1694
1695 int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf,
1696                       int64_t pos, int size)
1697 {
1698     BlockDriver *drv = bs->drv;
1699     if (!drv)
1700         return -ENOMEDIUM;
1701     if (drv->bdrv_save_vmstate)
1702         return drv->bdrv_save_vmstate(bs, buf, pos, size);
1703     if (bs->file)
1704         return bdrv_save_vmstate(bs->file, buf, pos, size);
1705     return -ENOTSUP;
1706 }
1707
1708 int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf,
1709                       int64_t pos, int size)
1710 {
1711     BlockDriver *drv = bs->drv;
1712     if (!drv)
1713         return -ENOMEDIUM;
1714     if (drv->bdrv_load_vmstate)
1715         return drv->bdrv_load_vmstate(bs, buf, pos, size);
1716     if (bs->file)
1717         return bdrv_load_vmstate(bs->file, buf, pos, size);
1718     return -ENOTSUP;
1719 }
1720
1721 void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event)
1722 {
1723     BlockDriver *drv = bs->drv;
1724
1725     if (!drv || !drv->bdrv_debug_event) {
1726         return;
1727     }
1728
1729     return drv->bdrv_debug_event(bs, event);
1730
1731 }
1732
1733 /**************************************************************/
1734 /* handling of snapshots */
1735
1736 int bdrv_can_snapshot(BlockDriverState *bs)
1737 {
1738     BlockDriver *drv = bs->drv;
1739     if (!drv || bdrv_is_removable(bs) || bdrv_is_read_only(bs)) {
1740         return 0;
1741     }
1742
1743     if (!drv->bdrv_snapshot_create) {
1744         if (bs->file != NULL) {
1745             return bdrv_can_snapshot(bs->file);
1746         }
1747         return 0;
1748     }
1749
1750     return 1;
1751 }
1752
1753 int bdrv_snapshot_create(BlockDriverState *bs,
1754                          QEMUSnapshotInfo *sn_info)
1755 {
1756     BlockDriver *drv = bs->drv;
1757     if (!drv)
1758         return -ENOMEDIUM;
1759     if (drv->bdrv_snapshot_create)
1760         return drv->bdrv_snapshot_create(bs, sn_info);
1761     if (bs->file)
1762         return bdrv_snapshot_create(bs->file, sn_info);
1763     return -ENOTSUP;
1764 }
1765
1766 int bdrv_snapshot_goto(BlockDriverState *bs,
1767                        const char *snapshot_id)
1768 {
1769     BlockDriver *drv = bs->drv;
1770     int ret, open_ret;
1771
1772     if (!drv)
1773         return -ENOMEDIUM;
1774     if (drv->bdrv_snapshot_goto)
1775         return drv->bdrv_snapshot_goto(bs, snapshot_id);
1776
1777     if (bs->file) {
1778         drv->bdrv_close(bs);
1779         ret = bdrv_snapshot_goto(bs->file, snapshot_id);
1780         open_ret = drv->bdrv_open(bs, bs->open_flags);
1781         if (open_ret < 0) {
1782             bdrv_delete(bs->file);
1783             bs->drv = NULL;
1784             return open_ret;
1785         }
1786         return ret;
1787     }
1788
1789     return -ENOTSUP;
1790 }
1791
1792 int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
1793 {
1794     BlockDriver *drv = bs->drv;
1795     if (!drv)
1796         return -ENOMEDIUM;
1797     if (drv->bdrv_snapshot_delete)
1798         return drv->bdrv_snapshot_delete(bs, snapshot_id);
1799     if (bs->file)
1800         return bdrv_snapshot_delete(bs->file, snapshot_id);
1801     return -ENOTSUP;
1802 }
1803
1804 int bdrv_snapshot_list(BlockDriverState *bs,
1805                        QEMUSnapshotInfo **psn_info)
1806 {
1807     BlockDriver *drv = bs->drv;
1808     if (!drv)
1809         return -ENOMEDIUM;
1810     if (drv->bdrv_snapshot_list)
1811         return drv->bdrv_snapshot_list(bs, psn_info);
1812     if (bs->file)
1813         return bdrv_snapshot_list(bs->file, psn_info);
1814     return -ENOTSUP;
1815 }
1816
1817 #define NB_SUFFIXES 4
1818
1819 char *get_human_readable_size(char *buf, int buf_size, int64_t size)
1820 {
1821     static const char suffixes[NB_SUFFIXES] = "KMGT";
1822     int64_t base;
1823     int i;
1824
1825     if (size <= 999) {
1826         snprintf(buf, buf_size, "%" PRId64, size);
1827     } else {
1828         base = 1024;
1829         for(i = 0; i < NB_SUFFIXES; i++) {
1830             if (size < (10 * base)) {
1831                 snprintf(buf, buf_size, "%0.1f%c",
1832                          (double)size / base,
1833                          suffixes[i]);
1834                 break;
1835             } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) {
1836                 snprintf(buf, buf_size, "%" PRId64 "%c",
1837                          ((size + (base >> 1)) / base),
1838                          suffixes[i]);
1839                 break;
1840             }
1841             base = base * 1024;
1842         }
1843     }
1844     return buf;
1845 }
1846
1847 char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn)
1848 {
1849     char buf1[128], date_buf[128], clock_buf[128];
1850 #ifdef _WIN32
1851     struct tm *ptm;
1852 #else
1853     struct tm tm;
1854 #endif
1855     time_t ti;
1856     int64_t secs;
1857
1858     if (!sn) {
1859         snprintf(buf, buf_size,
1860                  "%-10s%-20s%7s%20s%15s",
1861                  "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK");
1862     } else {
1863         ti = sn->date_sec;
1864 #ifdef _WIN32
1865         ptm = localtime(&ti);
1866         strftime(date_buf, sizeof(date_buf),
1867                  "%Y-%m-%d %H:%M:%S", ptm);
1868 #else
1869         localtime_r(&ti, &tm);
1870         strftime(date_buf, sizeof(date_buf),
1871                  "%Y-%m-%d %H:%M:%S", &tm);
1872 #endif
1873         secs = sn->vm_clock_nsec / 1000000000;
1874         snprintf(clock_buf, sizeof(clock_buf),
1875                  "%02d:%02d:%02d.%03d",
1876                  (int)(secs / 3600),
1877                  (int)((secs / 60) % 60),
1878                  (int)(secs % 60),
1879                  (int)((sn->vm_clock_nsec / 1000000) % 1000));
1880         snprintf(buf, buf_size,
1881                  "%-10s%-20s%7s%20s%15s",
1882                  sn->id_str, sn->name,
1883                  get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size),
1884                  date_buf,
1885                  clock_buf);
1886     }
1887     return buf;
1888 }
1889
1890
1891 /**************************************************************/
1892 /* async I/Os */
1893
1894 BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num,
1895                                  QEMUIOVector *qiov, int nb_sectors,
1896                                  BlockDriverCompletionFunc *cb, void *opaque)
1897 {
1898     BlockDriver *drv = bs->drv;
1899     BlockDriverAIOCB *ret;
1900
1901     if (!drv)
1902         return NULL;
1903     if (bdrv_check_request(bs, sector_num, nb_sectors))
1904         return NULL;
1905
1906     ret = drv->bdrv_aio_readv(bs, sector_num, qiov, nb_sectors,
1907                               cb, opaque);
1908
1909     if (ret) {
1910         /* Update stats even though technically transfer has not happened. */
1911         bs->rd_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1912         bs->rd_ops ++;
1913     }
1914
1915     return ret;
1916 }
1917
1918 BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num,
1919                                   QEMUIOVector *qiov, int nb_sectors,
1920                                   BlockDriverCompletionFunc *cb, void *opaque)
1921 {
1922     BlockDriver *drv = bs->drv;
1923     BlockDriverAIOCB *ret;
1924
1925     if (!drv)
1926         return NULL;
1927     if (bs->read_only)
1928         return NULL;
1929     if (bdrv_check_request(bs, sector_num, nb_sectors))
1930         return NULL;
1931
1932     if (bs->dirty_bitmap) {
1933         set_dirty_bitmap(bs, sector_num, nb_sectors, 1);
1934     }
1935
1936     ret = drv->bdrv_aio_writev(bs, sector_num, qiov, nb_sectors,
1937                                cb, opaque);
1938
1939     if (ret) {
1940         /* Update stats even though technically transfer has not happened. */
1941         bs->wr_bytes += (unsigned) nb_sectors * BDRV_SECTOR_SIZE;
1942         bs->wr_ops ++;
1943         if (bs->wr_highest_sector < sector_num + nb_sectors - 1) {
1944             bs->wr_highest_sector = sector_num + nb_sectors - 1;
1945         }
1946     }
1947
1948     return ret;
1949 }
1950
1951
1952 typedef struct MultiwriteCB {
1953     int error;
1954     int num_requests;
1955     int num_callbacks;
1956     struct {
1957         BlockDriverCompletionFunc *cb;
1958         void *opaque;
1959         QEMUIOVector *free_qiov;
1960         void *free_buf;
1961     } callbacks[];
1962 } MultiwriteCB;
1963
1964 static void multiwrite_user_cb(MultiwriteCB *mcb)
1965 {
1966     int i;
1967
1968     for (i = 0; i < mcb->num_callbacks; i++) {
1969         mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error);
1970         if (mcb->callbacks[i].free_qiov) {
1971             qemu_iovec_destroy(mcb->callbacks[i].free_qiov);
1972         }
1973         qemu_free(mcb->callbacks[i].free_qiov);
1974         qemu_vfree(mcb->callbacks[i].free_buf);
1975     }
1976 }
1977
1978 static void multiwrite_cb(void *opaque, int ret)
1979 {
1980     MultiwriteCB *mcb = opaque;
1981
1982     if (ret < 0 && !mcb->error) {
1983         mcb->error = ret;
1984         multiwrite_user_cb(mcb);
1985     }
1986
1987     mcb->num_requests--;
1988     if (mcb->num_requests == 0) {
1989         if (mcb->error == 0) {
1990             multiwrite_user_cb(mcb);
1991         }
1992         qemu_free(mcb);
1993     }
1994 }
1995
1996 static int multiwrite_req_compare(const void *a, const void *b)
1997 {
1998     const BlockRequest *req1 = a, *req2 = b;
1999
2000     /*
2001      * Note that we can't simply subtract req2->sector from req1->sector
2002      * here as that could overflow the return value.
2003      */
2004     if (req1->sector > req2->sector) {
2005         return 1;
2006     } else if (req1->sector < req2->sector) {
2007         return -1;
2008     } else {
2009         return 0;
2010     }
2011 }
2012
2013 /*
2014  * Takes a bunch of requests and tries to merge them. Returns the number of
2015  * requests that remain after merging.
2016  */
2017 static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs,
2018     int num_reqs, MultiwriteCB *mcb)
2019 {
2020     int i, outidx;
2021
2022     // Sort requests by start sector
2023     qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare);
2024
2025     // Check if adjacent requests touch the same clusters. If so, combine them,
2026     // filling up gaps with zero sectors.
2027     outidx = 0;
2028     for (i = 1; i < num_reqs; i++) {
2029         int merge = 0;
2030         int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors;
2031
2032         // This handles the cases that are valid for all block drivers, namely
2033         // exactly sequential writes and overlapping writes.
2034         if (reqs[i].sector <= oldreq_last) {
2035             merge = 1;
2036         }
2037
2038         // The block driver may decide that it makes sense to combine requests
2039         // even if there is a gap of some sectors between them. In this case,
2040         // the gap is filled with zeros (therefore only applicable for yet
2041         // unused space in format like qcow2).
2042         if (!merge && bs->drv->bdrv_merge_requests) {
2043             merge = bs->drv->bdrv_merge_requests(bs, &reqs[outidx], &reqs[i]);
2044         }
2045
2046         if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) {
2047             merge = 0;
2048         }
2049
2050         if (merge) {
2051             size_t size;
2052             QEMUIOVector *qiov = qemu_mallocz(sizeof(*qiov));
2053             qemu_iovec_init(qiov,
2054                 reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1);
2055
2056             // Add the first request to the merged one. If the requests are
2057             // overlapping, drop the last sectors of the first request.
2058             size = (reqs[i].sector - reqs[outidx].sector) << 9;
2059             qemu_iovec_concat(qiov, reqs[outidx].qiov, size);
2060
2061             // We might need to add some zeros between the two requests
2062             if (reqs[i].sector > oldreq_last) {
2063                 size_t zero_bytes = (reqs[i].sector - oldreq_last) << 9;
2064                 uint8_t *buf = qemu_blockalign(bs, zero_bytes);
2065                 memset(buf, 0, zero_bytes);
2066                 qemu_iovec_add(qiov, buf, zero_bytes);
2067                 mcb->callbacks[i].free_buf = buf;
2068             }
2069
2070             // Add the second request
2071             qemu_iovec_concat(qiov, reqs[i].qiov, reqs[i].qiov->size);
2072
2073             reqs[outidx].nb_sectors = qiov->size >> 9;
2074             reqs[outidx].qiov = qiov;
2075
2076             mcb->callbacks[i].free_qiov = reqs[outidx].qiov;
2077         } else {
2078             outidx++;
2079             reqs[outidx].sector     = reqs[i].sector;
2080             reqs[outidx].nb_sectors = reqs[i].nb_sectors;
2081             reqs[outidx].qiov       = reqs[i].qiov;
2082         }
2083     }
2084
2085     return outidx + 1;
2086 }
2087
2088 /*
2089  * Submit multiple AIO write requests at once.
2090  *
2091  * On success, the function returns 0 and all requests in the reqs array have
2092  * been submitted. In error case this function returns -1, and any of the
2093  * requests may or may not be submitted yet. In particular, this means that the
2094  * callback will be called for some of the requests, for others it won't. The
2095  * caller must check the error field of the BlockRequest to wait for the right
2096  * callbacks (if error != 0, no callback will be called).
2097  *
2098  * The implementation may modify the contents of the reqs array, e.g. to merge
2099  * requests. However, the fields opaque and error are left unmodified as they
2100  * are used to signal failure for a single request to the caller.
2101  */
2102 int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs)
2103 {
2104     BlockDriverAIOCB *acb;
2105     MultiwriteCB *mcb;
2106     int i;
2107
2108     if (num_reqs == 0) {
2109         return 0;
2110     }
2111
2112     // Create MultiwriteCB structure
2113     mcb = qemu_mallocz(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks));
2114     mcb->num_requests = 0;
2115     mcb->num_callbacks = num_reqs;
2116
2117     for (i = 0; i < num_reqs; i++) {
2118         mcb->callbacks[i].cb = reqs[i].cb;
2119         mcb->callbacks[i].opaque = reqs[i].opaque;
2120     }
2121
2122     // Check for mergable requests
2123     num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb);
2124
2125     // Run the aio requests
2126     for (i = 0; i < num_reqs; i++) {
2127         acb = bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov,
2128             reqs[i].nb_sectors, multiwrite_cb, mcb);
2129
2130         if (acb == NULL) {
2131             // We can only fail the whole thing if no request has been
2132             // submitted yet. Otherwise we'll wait for the submitted AIOs to
2133             // complete and report the error in the callback.
2134             if (mcb->num_requests == 0) {
2135                 reqs[i].error = -EIO;
2136                 goto fail;
2137             } else {
2138                 mcb->num_requests++;
2139                 multiwrite_cb(mcb, -EIO);
2140                 break;
2141             }
2142         } else {
2143             mcb->num_requests++;
2144         }
2145     }
2146
2147     return 0;
2148
2149 fail:
2150     qemu_free(mcb);
2151     return -1;
2152 }
2153
2154 BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs,
2155         BlockDriverCompletionFunc *cb, void *opaque)
2156 {
2157     BlockDriver *drv = bs->drv;
2158
2159     if (bs->open_flags & BDRV_O_NO_FLUSH) {
2160         return bdrv_aio_noop_em(bs, cb, opaque);
2161     }
2162
2163     if (!drv)
2164         return NULL;
2165     return drv->bdrv_aio_flush(bs, cb, opaque);
2166 }
2167
2168 void bdrv_aio_cancel(BlockDriverAIOCB *acb)
2169 {
2170     acb->pool->cancel(acb);
2171 }
2172
2173
2174 /**************************************************************/
2175 /* async block device emulation */
2176
2177 typedef struct BlockDriverAIOCBSync {
2178     BlockDriverAIOCB common;
2179     QEMUBH *bh;
2180     int ret;
2181     /* vector translation state */
2182     QEMUIOVector *qiov;
2183     uint8_t *bounce;
2184     int is_write;
2185 } BlockDriverAIOCBSync;
2186
2187 static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb)
2188 {
2189     BlockDriverAIOCBSync *acb =
2190         container_of(blockacb, BlockDriverAIOCBSync, common);
2191     qemu_bh_delete(acb->bh);
2192     acb->bh = NULL;
2193     qemu_aio_release(acb);
2194 }
2195
2196 static AIOPool bdrv_em_aio_pool = {
2197     .aiocb_size         = sizeof(BlockDriverAIOCBSync),
2198     .cancel             = bdrv_aio_cancel_em,
2199 };
2200
2201 static void bdrv_aio_bh_cb(void *opaque)
2202 {
2203     BlockDriverAIOCBSync *acb = opaque;
2204
2205     if (!acb->is_write)
2206         qemu_iovec_from_buffer(acb->qiov, acb->bounce, acb->qiov->size);
2207     qemu_vfree(acb->bounce);
2208     acb->common.cb(acb->common.opaque, acb->ret);
2209     qemu_bh_delete(acb->bh);
2210     acb->bh = NULL;
2211     qemu_aio_release(acb);
2212 }
2213
2214 static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs,
2215                                             int64_t sector_num,
2216                                             QEMUIOVector *qiov,
2217                                             int nb_sectors,
2218                                             BlockDriverCompletionFunc *cb,
2219                                             void *opaque,
2220                                             int is_write)
2221
2222 {
2223     BlockDriverAIOCBSync *acb;
2224
2225     acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2226     acb->is_write = is_write;
2227     acb->qiov = qiov;
2228     acb->bounce = qemu_blockalign(bs, qiov->size);
2229
2230     if (!acb->bh)
2231         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2232
2233     if (is_write) {
2234         qemu_iovec_to_buffer(acb->qiov, acb->bounce);
2235         acb->ret = bdrv_write(bs, sector_num, acb->bounce, nb_sectors);
2236     } else {
2237         acb->ret = bdrv_read(bs, sector_num, acb->bounce, nb_sectors);
2238     }
2239
2240     qemu_bh_schedule(acb->bh);
2241
2242     return &acb->common;
2243 }
2244
2245 static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs,
2246         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2247         BlockDriverCompletionFunc *cb, void *opaque)
2248 {
2249     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0);
2250 }
2251
2252 static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs,
2253         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
2254         BlockDriverCompletionFunc *cb, void *opaque)
2255 {
2256     return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1);
2257 }
2258
2259 static BlockDriverAIOCB *bdrv_aio_flush_em(BlockDriverState *bs,
2260         BlockDriverCompletionFunc *cb, void *opaque)
2261 {
2262     BlockDriverAIOCBSync *acb;
2263
2264     acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2265     acb->is_write = 1; /* don't bounce in the completion hadler */
2266     acb->qiov = NULL;
2267     acb->bounce = NULL;
2268     acb->ret = 0;
2269
2270     if (!acb->bh)
2271         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2272
2273     bdrv_flush(bs);
2274     qemu_bh_schedule(acb->bh);
2275     return &acb->common;
2276 }
2277
2278 static BlockDriverAIOCB *bdrv_aio_noop_em(BlockDriverState *bs,
2279         BlockDriverCompletionFunc *cb, void *opaque)
2280 {
2281     BlockDriverAIOCBSync *acb;
2282
2283     acb = qemu_aio_get(&bdrv_em_aio_pool, bs, cb, opaque);
2284     acb->is_write = 1; /* don't bounce in the completion handler */
2285     acb->qiov = NULL;
2286     acb->bounce = NULL;
2287     acb->ret = 0;
2288
2289     if (!acb->bh) {
2290         acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb);
2291     }
2292
2293     qemu_bh_schedule(acb->bh);
2294     return &acb->common;
2295 }
2296
2297 /**************************************************************/
2298 /* sync block device emulation */
2299
2300 static void bdrv_rw_em_cb(void *opaque, int ret)
2301 {
2302     *(int *)opaque = ret;
2303 }
2304
2305 #define NOT_DONE 0x7fffffff
2306
2307 static int bdrv_read_em(BlockDriverState *bs, int64_t sector_num,
2308                         uint8_t *buf, int nb_sectors)
2309 {
2310     int async_ret;
2311     BlockDriverAIOCB *acb;
2312     struct iovec iov;
2313     QEMUIOVector qiov;
2314
2315     async_context_push();
2316
2317     async_ret = NOT_DONE;
2318     iov.iov_base = (void *)buf;
2319     iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2320     qemu_iovec_init_external(&qiov, &iov, 1);
2321     acb = bdrv_aio_readv(bs, sector_num, &qiov, nb_sectors,
2322         bdrv_rw_em_cb, &async_ret);
2323     if (acb == NULL) {
2324         async_ret = -1;
2325         goto fail;
2326     }
2327
2328     while (async_ret == NOT_DONE) {
2329         qemu_aio_wait();
2330     }
2331
2332
2333 fail:
2334     async_context_pop();
2335     return async_ret;
2336 }
2337
2338 static int bdrv_write_em(BlockDriverState *bs, int64_t sector_num,
2339                          const uint8_t *buf, int nb_sectors)
2340 {
2341     int async_ret;
2342     BlockDriverAIOCB *acb;
2343     struct iovec iov;
2344     QEMUIOVector qiov;
2345
2346     async_context_push();
2347
2348     async_ret = NOT_DONE;
2349     iov.iov_base = (void *)buf;
2350     iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE;
2351     qemu_iovec_init_external(&qiov, &iov, 1);
2352     acb = bdrv_aio_writev(bs, sector_num, &qiov, nb_sectors,
2353         bdrv_rw_em_cb, &async_ret);
2354     if (acb == NULL) {
2355         async_ret = -1;
2356         goto fail;
2357     }
2358     while (async_ret == NOT_DONE) {
2359         qemu_aio_wait();
2360     }
2361
2362 fail:
2363     async_context_pop();
2364     return async_ret;
2365 }
2366
2367 void bdrv_init(void)
2368 {
2369     module_call_init(MODULE_INIT_BLOCK);
2370 }
2371
2372 void bdrv_init_with_whitelist(void)
2373 {
2374     use_bdrv_whitelist = 1;
2375     bdrv_init();
2376 }
2377
2378 void *qemu_aio_get(AIOPool *pool, BlockDriverState *bs,
2379                    BlockDriverCompletionFunc *cb, void *opaque)
2380 {
2381     BlockDriverAIOCB *acb;
2382
2383     if (pool->free_aiocb) {
2384         acb = pool->free_aiocb;
2385         pool->free_aiocb = acb->next;
2386     } else {
2387         acb = qemu_mallocz(pool->aiocb_size);
2388         acb->pool = pool;
2389     }
2390     acb->bs = bs;
2391     acb->cb = cb;
2392     acb->opaque = opaque;
2393     return acb;
2394 }
2395
2396 void qemu_aio_release(void *p)
2397 {
2398     BlockDriverAIOCB *acb = (BlockDriverAIOCB *)p;
2399     AIOPool *pool = acb->pool;
2400     acb->next = pool->free_aiocb;
2401     pool->free_aiocb = acb;
2402 }
2403
2404 /**************************************************************/
2405 /* removable device support */
2406
2407 /**
2408  * Return TRUE if the media is present
2409  */
2410 int bdrv_is_inserted(BlockDriverState *bs)
2411 {
2412     BlockDriver *drv = bs->drv;
2413     int ret;
2414     if (!drv)
2415         return 0;
2416     if (!drv->bdrv_is_inserted)
2417         return 1;
2418     ret = drv->bdrv_is_inserted(bs);
2419     return ret;
2420 }
2421
2422 /**
2423  * Return TRUE if the media changed since the last call to this
2424  * function. It is currently only used for floppy disks
2425  */
2426 int bdrv_media_changed(BlockDriverState *bs)
2427 {
2428     BlockDriver *drv = bs->drv;
2429     int ret;
2430
2431     if (!drv || !drv->bdrv_media_changed)
2432         ret = -ENOTSUP;
2433     else
2434         ret = drv->bdrv_media_changed(bs);
2435     if (ret == -ENOTSUP)
2436         ret = bs->media_changed;
2437     bs->media_changed = 0;
2438     return ret;
2439 }
2440
2441 /**
2442  * If eject_flag is TRUE, eject the media. Otherwise, close the tray
2443  */
2444 int bdrv_eject(BlockDriverState *bs, int eject_flag)
2445 {
2446     BlockDriver *drv = bs->drv;
2447     int ret;
2448
2449     if (bs->locked) {
2450         return -EBUSY;
2451     }
2452
2453     if (!drv || !drv->bdrv_eject) {
2454         ret = -ENOTSUP;
2455     } else {
2456         ret = drv->bdrv_eject(bs, eject_flag);
2457     }
2458     if (ret == -ENOTSUP) {
2459         if (eject_flag)
2460             bdrv_close(bs);
2461         ret = 0;
2462     }
2463
2464     return ret;
2465 }
2466
2467 int bdrv_is_locked(BlockDriverState *bs)
2468 {
2469     return bs->locked;
2470 }
2471
2472 /**
2473  * Lock or unlock the media (if it is locked, the user won't be able
2474  * to eject it manually).
2475  */
2476 void bdrv_set_locked(BlockDriverState *bs, int locked)
2477 {
2478     BlockDriver *drv = bs->drv;
2479
2480     bs->locked = locked;
2481     if (drv && drv->bdrv_set_locked) {
2482         drv->bdrv_set_locked(bs, locked);
2483     }
2484 }
2485
2486 /* needed for generic scsi interface */
2487
2488 int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
2489 {
2490     BlockDriver *drv = bs->drv;
2491
2492     if (drv && drv->bdrv_ioctl)
2493         return drv->bdrv_ioctl(bs, req, buf);
2494     return -ENOTSUP;
2495 }
2496
2497 BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs,
2498         unsigned long int req, void *buf,
2499         BlockDriverCompletionFunc *cb, void *opaque)
2500 {
2501     BlockDriver *drv = bs->drv;
2502
2503     if (drv && drv->bdrv_aio_ioctl)
2504         return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque);
2505     return NULL;
2506 }
2507
2508
2509
2510 void *qemu_blockalign(BlockDriverState *bs, size_t size)
2511 {
2512     return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size);
2513 }
2514
2515 void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable)
2516 {
2517     int64_t bitmap_size;
2518
2519     bs->dirty_count = 0;
2520     if (enable) {
2521         if (!bs->dirty_bitmap) {
2522             bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) +
2523                     BDRV_SECTORS_PER_DIRTY_CHUNK * 8 - 1;
2524             bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * 8;
2525
2526             bs->dirty_bitmap = qemu_mallocz(bitmap_size);
2527         }
2528     } else {
2529         if (bs->dirty_bitmap) {
2530             qemu_free(bs->dirty_bitmap);
2531             bs->dirty_bitmap = NULL;
2532         }
2533     }
2534 }
2535
2536 int bdrv_get_dirty(BlockDriverState *bs, int64_t sector)
2537 {
2538     int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK;
2539
2540     if (bs->dirty_bitmap &&
2541         (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) {
2542         return bs->dirty_bitmap[chunk / (sizeof(unsigned long) * 8)] &
2543             (1 << (chunk % (sizeof(unsigned long) * 8)));
2544     } else {
2545         return 0;
2546     }
2547 }
2548
2549 void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector,
2550                       int nr_sectors)
2551 {
2552     set_dirty_bitmap(bs, cur_sector, nr_sectors, 0);
2553 }
2554
2555 int64_t bdrv_get_dirty_count(BlockDriverState *bs)
2556 {
2557     return bs->dirty_count;
2558 }