block/raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 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 "qemu-common.h"
  25 #include "qemu-timer.h"
  26 #include "qemu-char.h"
  27 #include "qemu-log.h"
  28 #include "block_int.h"
  29 #include "module.h"
  30 #include "block/raw-posix-aio.h"
  31
  32 #ifdef CONFIG_COCOA
  33 #include <paths.h>
  34 #include <sys/param.h>
  35 #include <IOKit/IOKitLib.h>
  36 #include <IOKit/IOBSD.h>
  37 #include <IOKit/storage/IOMediaBSDClient.h>
  38 #include <IOKit/storage/IOMedia.h>
  39 #include <IOKit/storage/IOCDMedia.h>
  40 //#include <IOKit/storage/IOCDTypes.h>
  41 #include <CoreFoundation/CoreFoundation.h>
  42 #endif
  43
  44 #ifdef __sun__
  45 #define _POSIX_PTHREAD_SEMANTICS 1
  46 #include <signal.h>
  47 #include <sys/dkio.h>
  48 #endif
  49 #ifdef __linux__
  50 #include <sys/ioctl.h>
  51 #include <linux/cdrom.h>
  52 #include <linux/fd.h>
  53 #endif
  54 #ifdef __FreeBSD__
  55 #include <signal.h>
  56 #include <sys/disk.h>
  57 #include <sys/cdio.h>
  58 #endif
  59
  60 #ifdef __OpenBSD__
  61 #include <sys/ioctl.h>
  62 #include <sys/disklabel.h>
  63 #include <sys/dkio.h>
  64 #endif
  65
  66 #ifdef __DragonFly__
  67 #include <sys/ioctl.h>
  68 #include <sys/diskslice.h>
  69 #endif
  70
  71 //#define DEBUG_FLOPPY
  72
  73 //#define DEBUG_BLOCK
  74 #if defined(DEBUG_BLOCK)
  75 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
  76     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
  77 #else
  78 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
  79 #endif
  80
  81 /* OS X does not have O_DSYNC */
  82 #ifndef O_DSYNC
  83 #ifdef O_SYNC
  84 #define O_DSYNC O_SYNC
  85 #elif defined(O_FSYNC)
  86 #define O_DSYNC O_FSYNC
  87 #endif
  88 #endif
  89
  90 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  91 #ifndef O_DIRECT
  92 #define O_DIRECT O_DSYNC
  93 #endif
  94
  95 #define FTYPE_FILE   0
  96 #define FTYPE_CD     1
  97 #define FTYPE_FD     2
  98
  99 #define ALIGNED_BUFFER_SIZE (32 * 512)
 100
 101 /* if the FD is not accessed during that time (in ms), we try to
 102    reopen it to see if the disk has been changed */
 103 #define FD_OPEN_TIMEOUT 1000
 104
 105 typedef struct BDRVRawState {
 106     int fd;
 107     int type;
 108     unsigned int lseek_err_cnt;
 109     int open_flags;
 110     void *aio_ctx;
 111 #if defined(__linux__)
 112     /* linux floppy specific */
 113     int64_t fd_open_time;
 114     int64_t fd_error_time;
 115     int fd_got_error;
 116     int fd_media_changed;
 117 #endif
 118     int use_aio;
 119     uint8_t* aligned_buf;
 120 } BDRVRawState;
 121
 122 static int fd_open(BlockDriverState *bs);
 123 static int64_t raw_getlength(BlockDriverState *bs);
 124
 125 #if defined(__FreeBSD__)
 126 static int cdrom_reopen(BlockDriverState *bs);
 127 #endif
 128
 129 static int raw_open_common(BlockDriverState *bs, const char *filename,
 130                            int bdrv_flags, int open_flags)
 131 {
 132     BDRVRawState *s = bs->opaque;
 133     int fd, ret;
 134
 135     s->lseek_err_cnt = 0;
 136
 137     s->open_flags = open_flags | O_BINARY;
 138     s->open_flags &= ~O_ACCMODE;
 139     if ((bdrv_flags & BDRV_O_ACCESS) == BDRV_O_RDWR) {
 140         s->open_flags |= O_RDWR;
 141     } else {
 142         s->open_flags |= O_RDONLY;
 143         bs->read_only = 1;
 144     }
 145
 146     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 147      * and O_DIRECT for no caching. */
 148     if ((bdrv_flags & BDRV_O_NOCACHE))
 149         s->open_flags |= O_DIRECT;
 150     else if (!(bdrv_flags & BDRV_O_CACHE_WB))
 151         s->open_flags |= O_DSYNC;
 152
 153     s->fd = -1;
 154     fd = open(filename, s->open_flags, 0644);
 155     if (fd < 0) {
 156         ret = -errno;
 157         if (ret == -EROFS)
 158             ret = -EACCES;
 159         return ret;
 160     }
 161     s->fd = fd;
 162     s->aligned_buf = NULL;
 163
 164     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 165         s->aligned_buf = qemu_blockalign(bs, ALIGNED_BUFFER_SIZE);
 166         if (s->aligned_buf == NULL) {
 167             goto out_close;
 168         }
 169     }
 170
 171 #ifdef CONFIG_LINUX_AIO
 172     if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
 173                       (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
 174         s->aio_ctx = laio_init();
 175         if (!s->aio_ctx) {
 176             goto out_free_buf;
 177         }
 178         s->use_aio = 1;
 179     } else
 180 #endif
 181     {
 182         s->aio_ctx = paio_init();
 183         if (!s->aio_ctx) {
 184             goto out_free_buf;
 185         }
 186         s->use_aio = 0;
 187     }
 188
 189     return 0;
 190
 191 out_free_buf:
 192     qemu_vfree(s->aligned_buf);
 193 out_close:
 194     close(fd);
 195     return -errno;
 196 }
 197
 198 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 199 {
 200     BDRVRawState *s = bs->opaque;
 201     int open_flags = 0;
 202
 203     s->type = FTYPE_FILE;
 204     if (flags & BDRV_O_CREAT)
 205         open_flags = O_CREAT | O_TRUNC;
 206
 207     return raw_open_common(bs, filename, flags, open_flags);
 208 }
 209
 210 /* XXX: use host sector size if necessary with:
 211 #ifdef DIOCGSECTORSIZE
 212         {
 213             unsigned int sectorsize = 512;
 214             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 215                 sectorsize > bufsize)
 216                 bufsize = sectorsize;
 217         }
 218 #endif
 219 #ifdef CONFIG_COCOA
 220         u_int32_t   blockSize = 512;
 221         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 222             bufsize = blockSize;
 223         }
 224 #endif
 225 */
 226
 227 /*
 228  * offset and count are in bytes, but must be multiples of 512 for files
 229  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 230  *
 231  * This function may be called without alignment if the caller ensures
 232  * that O_DIRECT is not in effect.
 233  */
 234 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 235                      uint8_t *buf, int count)
 236 {
 237     BDRVRawState *s = bs->opaque;
 238     int ret;
 239
 240     ret = fd_open(bs);
 241     if (ret < 0)
 242         return ret;
 243
 244     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 245         ++(s->lseek_err_cnt);
 246         if(s->lseek_err_cnt <= 10) {
 247             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 248                               "] lseek failed : %d = %s\n",
 249                               s->fd, bs->filename, offset, buf, count,
 250                               bs->total_sectors, errno, strerror(errno));
 251         }
 252         return -1;
 253     }
 254     s->lseek_err_cnt=0;
 255
 256     ret = read(s->fd, buf, count);
 257     if (ret == count)
 258         goto label__raw_read__success;
 259
 260     /* Allow reads beyond the end (needed for pwrite) */
 261     if ((ret == 0) && bs->growable) {
 262         int64_t size = raw_getlength(bs);
 263         if (offset >= size) {
 264             memset(buf, 0, count);
 265             ret = count;
 266             goto label__raw_read__success;
 267         }
 268     }
 269
 270     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 271                       "] read failed %d : %d = %s\n",
 272                       s->fd, bs->filename, offset, buf, count,
 273                       bs->total_sectors, ret, errno, strerror(errno));
 274
 275     /* Try harder for CDrom. */
 276     if (bs->type == BDRV_TYPE_CDROM) {
 277         lseek(s->fd, offset, SEEK_SET);
 278         ret = read(s->fd, buf, count);
 279         if (ret == count)
 280             goto label__raw_read__success;
 281         lseek(s->fd, offset, SEEK_SET);
 282         ret = read(s->fd, buf, count);
 283         if (ret == count)
 284             goto label__raw_read__success;
 285
 286         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 287                           "] retry read failed %d : %d = %s\n",
 288                           s->fd, bs->filename, offset, buf, count,
 289                           bs->total_sectors, ret, errno, strerror(errno));
 290     }
 291
 292 label__raw_read__success:
 293
 294     return  (ret < 0) ? -errno : ret;
 295 }
 296
 297 /*
 298  * offset and count are in bytes, but must be multiples of 512 for files
 299  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 300  *
 301  * This function may be called without alignment if the caller ensures
 302  * that O_DIRECT is not in effect.
 303  */
 304 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 305                       const uint8_t *buf, int count)
 306 {
 307     BDRVRawState *s = bs->opaque;
 308     int ret;
 309
 310     ret = fd_open(bs);
 311     if (ret < 0)
 312         return -errno;
 313
 314     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 315         ++(s->lseek_err_cnt);
 316         if(s->lseek_err_cnt) {
 317             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 318                               PRId64 "] lseek failed : %d = %s\n",
 319                               s->fd, bs->filename, offset, buf, count,
 320                               bs->total_sectors, errno, strerror(errno));
 321         }
 322         return -EIO;
 323     }
 324     s->lseek_err_cnt = 0;
 325
 326     ret = write(s->fd, buf, count);
 327     if (ret == count)
 328         goto label__raw_write__success;
 329
 330     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 331                       "] write failed %d : %d = %s\n",
 332                       s->fd, bs->filename, offset, buf, count,
 333                       bs->total_sectors, ret, errno, strerror(errno));
 334
 335 label__raw_write__success:
 336
 337     return  (ret < 0) ? -errno : ret;
 338 }
 339
 340
 341 /*
 342  * offset and count are in bytes and possibly not aligned. For files opened
 343  * with O_DIRECT, necessary alignments are ensured before calling
 344  * raw_pread_aligned to do the actual read.
 345  */
 346 static int raw_pread(BlockDriverState *bs, int64_t offset,
 347                      uint8_t *buf, int count)
 348 {
 349     BDRVRawState *s = bs->opaque;
 350     int size, ret, shift, sum;
 351
 352     sum = 0;
 353
 354     if (s->aligned_buf != NULL)  {
 355
 356         if (offset & 0x1ff) {
 357             /* align offset on a 512 bytes boundary */
 358
 359             shift = offset & 0x1ff;
 360             size = (shift + count + 0x1ff) & ~0x1ff;
 361             if (size > ALIGNED_BUFFER_SIZE)
 362                 size = ALIGNED_BUFFER_SIZE;
 363             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 364             if (ret < 0)
 365                 return ret;
 366
 367             size = 512 - shift;
 368             if (size > count)
 369                 size = count;
 370             memcpy(buf, s->aligned_buf + shift, size);
 371
 372             buf += size;
 373             offset += size;
 374             count -= size;
 375             sum += size;
 376
 377             if (count == 0)
 378                 return sum;
 379         }
 380         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 381
 382             /* read on aligned buffer */
 383
 384             while (count) {
 385
 386                 size = (count + 0x1ff) & ~0x1ff;
 387                 if (size > ALIGNED_BUFFER_SIZE)
 388                     size = ALIGNED_BUFFER_SIZE;
 389
 390                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 391                 if (ret < 0)
 392                     return ret;
 393
 394                 size = ret;
 395                 if (size > count)
 396                     size = count;
 397
 398                 memcpy(buf, s->aligned_buf, size);
 399
 400                 buf += size;
 401                 offset += size;
 402                 count -= size;
 403                 sum += size;
 404             }
 405
 406             return sum;
 407         }
 408     }
 409
 410     return raw_pread_aligned(bs, offset, buf, count) + sum;
 411 }
 412
 413 static int raw_read(BlockDriverState *bs, int64_t sector_num,
 414                     uint8_t *buf, int nb_sectors)
 415 {
 416     int ret;
 417
 418     ret = raw_pread(bs, sector_num * 512, buf, nb_sectors * 512);
 419     if (ret == (nb_sectors * 512))
 420         ret = 0;
 421     return ret;
 422 }
 423
 424 /*
 425  * offset and count are in bytes and possibly not aligned. For files opened
 426  * with O_DIRECT, necessary alignments are ensured before calling
 427  * raw_pwrite_aligned to do the actual write.
 428  */
 429 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 430                       const uint8_t *buf, int count)
 431 {
 432     BDRVRawState *s = bs->opaque;
 433     int size, ret, shift, sum;
 434
 435     sum = 0;
 436
 437     if (s->aligned_buf != NULL) {
 438
 439         if (offset & 0x1ff) {
 440             /* align offset on a 512 bytes boundary */
 441             shift = offset & 0x1ff;
 442             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 443             if (ret < 0)
 444                 return ret;
 445
 446             size = 512 - shift;
 447             if (size > count)
 448                 size = count;
 449             memcpy(s->aligned_buf + shift, buf, size);
 450
 451             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 452             if (ret < 0)
 453                 return ret;
 454
 455             buf += size;
 456             offset += size;
 457             count -= size;
 458             sum += size;
 459
 460             if (count == 0)
 461                 return sum;
 462         }
 463         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 464
 465             while ((size = (count & ~0x1ff)) != 0) {
 466
 467                 if (size > ALIGNED_BUFFER_SIZE)
 468                     size = ALIGNED_BUFFER_SIZE;
 469
 470                 memcpy(s->aligned_buf, buf, size);
 471
 472                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 473                 if (ret < 0)
 474                     return ret;
 475
 476                 buf += ret;
 477                 offset += ret;
 478                 count -= ret;
 479                 sum += ret;
 480             }
 481             /* here, count < 512 because (count & ~0x1ff) == 0 */
 482             if (count) {
 483                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 484                 if (ret < 0)
 485                     return ret;
 486                  memcpy(s->aligned_buf, buf, count);
 487
 488                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 489                  if (ret < 0)
 490                      return ret;
 491                  if (count < ret)
 492                      ret = count;
 493
 494                  sum += ret;
 495             }
 496             return sum;
 497         }
 498     }
 499     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 500 }
 501
 502 static int raw_write(BlockDriverState *bs, int64_t sector_num,
 503                      const uint8_t *buf, int nb_sectors)
 504 {
 505     int ret;
 506     ret = raw_pwrite(bs, sector_num * 512, buf, nb_sectors * 512);
 507     if (ret == (nb_sectors * 512))
 508         ret = 0;
 509     return ret;
 510 }
 511
 512 /*
 513  * Check if all memory in this vector is sector aligned.
 514  */
 515 static int qiov_is_aligned(QEMUIOVector *qiov)
 516 {
 517     int i;
 518
 519     for (i = 0; i < qiov->niov; i++) {
 520         if ((uintptr_t) qiov->iov[i].iov_base % 512) {
 521             return 0;
 522         }
 523     }
 524
 525     return 1;
 526 }
 527
 528 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 529         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 530         BlockDriverCompletionFunc *cb, void *opaque, int type)
 531 {
 532     BDRVRawState *s = bs->opaque;
 533
 534     if (fd_open(bs) < 0)
 535         return NULL;
 536
 537     /*
 538      * If O_DIRECT is used the buffer needs to be aligned on a sector
 539      * boundary.  Check if this is the case or telll the low-level
 540      * driver that it needs to copy the buffer.
 541      */
 542     if (s->aligned_buf) {
 543         if (!qiov_is_aligned(qiov)) {
 544             type |= QEMU_AIO_MISALIGNED;
 545         } else if (s->use_aio) {
 546             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 547                                nb_sectors, cb, opaque, type);
 548         }
 549     }
 550
 551     return paio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov, nb_sectors,
 552                        cb, opaque, type);
 553 }
 554
 555 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 556         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 557         BlockDriverCompletionFunc *cb, void *opaque)
 558 {
 559     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 560                           cb, opaque, QEMU_AIO_READ);
 561 }
 562
 563 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 564         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 565         BlockDriverCompletionFunc *cb, void *opaque)
 566 {
 567     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 568                           cb, opaque, QEMU_AIO_WRITE);
 569 }
 570
 571 static void raw_close(BlockDriverState *bs)
 572 {
 573     BDRVRawState *s = bs->opaque;
 574     if (s->fd >= 0) {
 575         close(s->fd);
 576         s->fd = -1;
 577         if (s->aligned_buf != NULL)
 578             qemu_free(s->aligned_buf);
 579     }
 580 }
 581
 582 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 583 {
 584     BDRVRawState *s = bs->opaque;
 585     if (s->type != FTYPE_FILE)
 586         return -ENOTSUP;
 587     if (ftruncate(s->fd, offset) < 0)
 588         return -errno;
 589     return 0;
 590 }
 591
 592 #ifdef __OpenBSD__
 593 static int64_t raw_getlength(BlockDriverState *bs)
 594 {
 595     BDRVRawState *s = bs->opaque;
 596     int fd = s->fd;
 597     struct stat st;
 598
 599     if (fstat(fd, &st))
 600         return -1;
 601     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 602         struct disklabel dl;
 603
 604         if (ioctl(fd, DIOCGDINFO, &dl))
 605             return -1;
 606         return (uint64_t)dl.d_secsize *
 607             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 608     } else
 609         return st.st_size;
 610 }
 611 #else /* !__OpenBSD__ */
 612 static int64_t  raw_getlength(BlockDriverState *bs)
 613 {
 614     BDRVRawState *s = bs->opaque;
 615     int fd = s->fd;
 616     int64_t size;
 617 #ifdef CONFIG_BSD
 618     struct stat sb;
 619 #ifdef __FreeBSD__
 620     int reopened = 0;
 621 #endif
 622 #endif
 623 #ifdef __sun__
 624     struct dk_minfo minfo;
 625     int rv;
 626 #endif
 627     int ret;
 628
 629     ret = fd_open(bs);
 630     if (ret < 0)
 631         return ret;
 632
 633 #ifdef CONFIG_BSD
 634 #ifdef __FreeBSD__
 635 again:
 636 #endif
 637     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 638 #ifdef DIOCGMEDIASIZE
 639         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 640 #elif defined(DIOCGPART)
 641         {
 642                 struct partinfo pi;
 643                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 644                         size = pi.media_size;
 645                 else
 646                         size = 0;
 647         }
 648         if (size == 0)
 649 #endif
 650 #ifdef CONFIG_COCOA
 651         size = LONG_LONG_MAX;
 652 #else
 653         size = lseek(fd, 0LL, SEEK_END);
 654 #endif
 655 #ifdef __FreeBSD__
 656         switch(s->type) {
 657         case FTYPE_CD:
 658             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
 659             if (size == 2048LL * (unsigned)-1)
 660                 size = 0;
 661             /* XXX no disc?  maybe we need to reopen... */
 662             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
 663                 reopened = 1;
 664                 goto again;
 665             }
 666         }
 667 #endif
 668     } else
 669 #endif
 670 #ifdef __sun__
 671     /*
 672      * use the DKIOCGMEDIAINFO ioctl to read the size.
 673      */
 674     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 675     if ( rv != -1 ) {
 676         size = minfo.dki_lbsize * minfo.dki_capacity;
 677     } else /* there are reports that lseek on some devices
 678               fails, but irc discussion said that contingency
 679               on contingency was overkill */
 680 #endif
 681     {
 682         size = lseek(fd, 0, SEEK_END);
 683     }
 684     return size;
 685 }
 686 #endif
 687
 688 static int raw_create(const char *filename, QEMUOptionParameter *options)
 689 {
 690     int fd;
 691     int result = 0;
 692     int64_t total_size = 0;
 693
 694     /* Read out options */
 695     while (options && options->name) {
 696         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
 697             total_size = options->value.n / 512;
 698         }
 699         options++;
 700     }
 701
 702     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 703               0644);
 704     if (fd < 0) {
 705         result = -errno;
 706     } else {
 707         if (ftruncate(fd, total_size * 512) != 0) {
 708             result = -errno;
 709         }
 710         if (close(fd) != 0) {
 711             result = -errno;
 712         }
 713     }
 714     return result;
 715 }
 716
 717 static void raw_flush(BlockDriverState *bs)
 718 {
 719     BDRVRawState *s = bs->opaque;
 720     fsync(s->fd);
 721 }
 722
 723
 724 static QEMUOptionParameter raw_create_options[] = {
 725     {
 726         .name = BLOCK_OPT_SIZE,
 727         .type = OPT_SIZE,
 728         .help = "Virtual disk size"
 729     },
 730     { NULL }
 731 };
 732
 733 static BlockDriver bdrv_raw = {
 734     .format_name = "raw",
 735     .instance_size = sizeof(BDRVRawState),
 736     .bdrv_probe = NULL, /* no probe for protocols */
 737     .bdrv_open = raw_open,
 738     .bdrv_read = raw_read,
 739     .bdrv_write = raw_write,
 740     .bdrv_close = raw_close,
 741     .bdrv_create = raw_create,
 742     .bdrv_flush = raw_flush,
 743
 744     .bdrv_aio_readv = raw_aio_readv,
 745     .bdrv_aio_writev = raw_aio_writev,
 746
 747     .bdrv_truncate = raw_truncate,
 748     .bdrv_getlength = raw_getlength,
 749
 750     .create_options = raw_create_options,
 751 };
 752
 753 /***********************************************/
 754 /* host device */
 755
 756 #ifdef CONFIG_COCOA
 757 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 758 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 759
 760 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 761 {
 762     kern_return_t       kernResult;
 763     mach_port_t     masterPort;
 764     CFMutableDictionaryRef  classesToMatch;
 765
 766     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 767     if ( KERN_SUCCESS != kernResult ) {
 768         printf( "IOMasterPort returned %d\n", kernResult );
 769     }
 770
 771     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 772     if ( classesToMatch == NULL ) {
 773         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 774     } else {
 775     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 776     }
 777     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 778     if ( KERN_SUCCESS != kernResult )
 779     {
 780         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 781     }
 782
 783     return kernResult;
 784 }
 785
 786 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 787 {
 788     io_object_t     nextMedia;
 789     kern_return_t   kernResult = KERN_FAILURE;
 790     *bsdPath = '\0';
 791     nextMedia = IOIteratorNext( mediaIterator );
 792     if ( nextMedia )
 793     {
 794         CFTypeRef   bsdPathAsCFString;
 795     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 796         if ( bsdPathAsCFString ) {
 797             size_t devPathLength;
 798             strcpy( bsdPath, _PATH_DEV );
 799             strcat( bsdPath, "r" );
 800             devPathLength = strlen( bsdPath );
 801             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 802                 kernResult = KERN_SUCCESS;
 803             }
 804             CFRelease( bsdPathAsCFString );
 805         }
 806         IOObjectRelease( nextMedia );
 807     }
 808
 809     return kernResult;
 810 }
 811
 812 #endif
 813
 814 static int hdev_probe_device(const char *filename)
 815 {
 816     struct stat st;
 817
 818     /* allow a dedicated CD-ROM driver to match with a higher priority */
 819     if (strstart(filename, "/dev/cdrom", NULL))
 820         return 50;
 821
 822     if (stat(filename, &st) >= 0 &&
 823             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
 824         return 100;
 825     }
 826
 827     return 0;
 828 }
 829
 830 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 831 {
 832     BDRVRawState *s = bs->opaque;
 833
 834 #ifdef CONFIG_COCOA
 835     if (strstart(filename, "/dev/cdrom", NULL)) {
 836         kern_return_t kernResult;
 837         io_iterator_t mediaIterator;
 838         char bsdPath[ MAXPATHLEN ];
 839         int fd;
 840
 841         kernResult = FindEjectableCDMedia( &mediaIterator );
 842         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 843
 844         if ( bsdPath[ 0 ] != '\0' ) {
 845             strcat(bsdPath,"s0");
 846             /* some CDs don't have a partition 0 */
 847             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 848             if (fd < 0) {
 849                 bsdPath[strlen(bsdPath)-1] = '1';
 850             } else {
 851                 close(fd);
 852             }
 853             filename = bsdPath;
 854         }
 855
 856         if ( mediaIterator )
 857             IOObjectRelease( mediaIterator );
 858     }
 859 #endif
 860
 861     s->type = FTYPE_FILE;
 862 #if defined(__linux__)
 863     if (strstart(filename, "/dev/sg", NULL)) {
 864         bs->sg = 1;
 865     }
 866 #endif
 867
 868     return raw_open_common(bs, filename, flags, 0);
 869 }
 870
 871 #if defined(__linux__)
 872 /* Note: we do not have a reliable method to detect if the floppy is
 873    present. The current method is to try to open the floppy at every
 874    I/O and to keep it opened during a few hundreds of ms. */
 875 static int fd_open(BlockDriverState *bs)
 876 {
 877     BDRVRawState *s = bs->opaque;
 878     int last_media_present;
 879
 880     if (s->type != FTYPE_FD)
 881         return 0;
 882     last_media_present = (s->fd >= 0);
 883     if (s->fd >= 0 &&
 884         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
 885         close(s->fd);
 886         s->fd = -1;
 887 #ifdef DEBUG_FLOPPY
 888         printf("Floppy closed\n");
 889 #endif
 890     }
 891     if (s->fd < 0) {
 892         if (s->fd_got_error &&
 893             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
 894 #ifdef DEBUG_FLOPPY
 895             printf("No floppy (open delayed)\n");
 896 #endif
 897             return -EIO;
 898         }
 899         s->fd = open(bs->filename, s->open_flags & ~O_NONBLOCK);
 900         if (s->fd < 0) {
 901             s->fd_error_time = qemu_get_clock(rt_clock);
 902             s->fd_got_error = 1;
 903             if (last_media_present)
 904                 s->fd_media_changed = 1;
 905 #ifdef DEBUG_FLOPPY
 906             printf("No floppy\n");
 907 #endif
 908             return -EIO;
 909         }
 910 #ifdef DEBUG_FLOPPY
 911         printf("Floppy opened\n");
 912 #endif
 913     }
 914     if (!last_media_present)
 915         s->fd_media_changed = 1;
 916     s->fd_open_time = qemu_get_clock(rt_clock);
 917     s->fd_got_error = 0;
 918     return 0;
 919 }
 920
 921 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 922 {
 923     BDRVRawState *s = bs->opaque;
 924
 925     return ioctl(s->fd, req, buf);
 926 }
 927
 928 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
 929         unsigned long int req, void *buf,
 930         BlockDriverCompletionFunc *cb, void *opaque)
 931 {
 932     BDRVRawState *s = bs->opaque;
 933
 934     if (fd_open(bs) < 0)
 935         return NULL;
 936     return paio_ioctl(bs, s->fd, req, buf, cb, opaque);
 937 }
 938
 939 #elif defined(__FreeBSD__)
 940 static int fd_open(BlockDriverState *bs)
 941 {
 942     BDRVRawState *s = bs->opaque;
 943
 944     /* this is just to ensure s->fd is sane (its called by io ops) */
 945     if (s->fd >= 0)
 946         return 0;
 947     return -EIO;
 948 }
 949 #else /* !linux && !FreeBSD */
 950
 951 static int fd_open(BlockDriverState *bs)
 952 {
 953     return 0;
 954 }
 955
 956 #endif /* !linux && !FreeBSD */
 957
 958 static int hdev_create(const char *filename, QEMUOptionParameter *options)
 959 {
 960     int fd;
 961     int ret = 0;
 962     struct stat stat_buf;
 963     int64_t total_size = 0;
 964
 965     /* Read out options */
 966     while (options && options->name) {
 967         if (!strcmp(options->name, "size")) {
 968             total_size = options->value.n / 512;
 969         }
 970         options++;
 971     }
 972
 973     fd = open(filename, O_WRONLY | O_BINARY);
 974     if (fd < 0)
 975         return -EIO;
 976
 977     if (fstat(fd, &stat_buf) < 0)
 978         ret = -EIO;
 979     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
 980         ret = -EIO;
 981     else if (lseek(fd, 0, SEEK_END) < total_size * 512)
 982         ret = -ENOSPC;
 983
 984     close(fd);
 985     return ret;
 986 }
 987
 988 static BlockDriver bdrv_host_device = {
 989     .format_name        = "host_device",
 990     .instance_size      = sizeof(BDRVRawState),
 991     .bdrv_probe_device  = hdev_probe_device,
 992     .bdrv_open          = hdev_open,
 993     .bdrv_close         = raw_close,
 994     .bdrv_create        = hdev_create,
 995     .bdrv_flush         = raw_flush,
 996
 997     .bdrv_aio_readv     = raw_aio_readv,
 998     .bdrv_aio_writev    = raw_aio_writev,
 999
1000     .bdrv_read          = raw_read,
1001     .bdrv_write         = raw_write,
1002     .bdrv_getlength     = raw_getlength,
1003
1004     /* generic scsi device */
1005 #ifdef __linux__
1006     .bdrv_ioctl         = hdev_ioctl,
1007     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1008 #endif
1009 };
1010
1011 #ifdef __linux__
1012 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1013 {
1014     BDRVRawState *s = bs->opaque;
1015     int ret;
1016
1017     s->type = FTYPE_FD;
1018
1019     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1020     ret = raw_open_common(bs, filename, flags, O_NONBLOCK);
1021     if (ret)
1022         return ret;
1023
1024     /* close fd so that we can reopen it as needed */
1025     close(s->fd);
1026     s->fd = -1;
1027     s->fd_media_changed = 1;
1028
1029     return 0;
1030 }
1031
1032 static int floppy_probe_device(const char *filename)
1033 {
1034     if (strstart(filename, "/dev/fd", NULL))
1035         return 100;
1036     return 0;
1037 }
1038
1039
1040 static int floppy_is_inserted(BlockDriverState *bs)
1041 {
1042     return fd_open(bs) >= 0;
1043 }
1044
1045 static int floppy_media_changed(BlockDriverState *bs)
1046 {
1047     BDRVRawState *s = bs->opaque;
1048     int ret;
1049
1050     /*
1051      * XXX: we do not have a true media changed indication.
1052      * It does not work if the floppy is changed without trying to read it.
1053      */
1054     fd_open(bs);
1055     ret = s->fd_media_changed;
1056     s->fd_media_changed = 0;
1057 #ifdef DEBUG_FLOPPY
1058     printf("Floppy changed=%d\n", ret);
1059 #endif
1060     return ret;
1061 }
1062
1063 static int floppy_eject(BlockDriverState *bs, int eject_flag)
1064 {
1065     BDRVRawState *s = bs->opaque;
1066     int fd;
1067
1068     if (s->fd >= 0) {
1069         close(s->fd);
1070         s->fd = -1;
1071     }
1072     fd = open(bs->filename, s->open_flags | O_NONBLOCK);
1073     if (fd >= 0) {
1074         if (ioctl(fd, FDEJECT, 0) < 0)
1075             perror("FDEJECT");
1076         close(fd);
1077     }
1078
1079     return 0;
1080 }
1081
1082 static BlockDriver bdrv_host_floppy = {
1083     .format_name        = "host_floppy",
1084     .instance_size      = sizeof(BDRVRawState),
1085     .bdrv_probe_device  = floppy_probe_device,
1086     .bdrv_open          = floppy_open,
1087     .bdrv_close         = raw_close,
1088     .bdrv_create        = hdev_create,
1089     .bdrv_flush         = raw_flush,
1090
1091     .bdrv_aio_readv     = raw_aio_readv,
1092     .bdrv_aio_writev    = raw_aio_writev,
1093
1094     .bdrv_read          = raw_read,
1095     .bdrv_write         = raw_write,
1096     .bdrv_getlength     = raw_getlength,
1097
1098     /* removable device support */
1099     .bdrv_is_inserted   = floppy_is_inserted,
1100     .bdrv_media_changed = floppy_media_changed,
1101     .bdrv_eject         = floppy_eject,
1102 };
1103
1104 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1105 {
1106     BDRVRawState *s = bs->opaque;
1107
1108     s->type = FTYPE_CD;
1109
1110     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1111     return raw_open_common(bs, filename, flags, O_NONBLOCK);
1112 }
1113
1114 static int cdrom_probe_device(const char *filename)
1115 {
1116     if (strstart(filename, "/dev/cd", NULL))
1117         return 100;
1118     return 0;
1119 }
1120
1121 static int cdrom_is_inserted(BlockDriverState *bs)
1122 {
1123     BDRVRawState *s = bs->opaque;
1124     int ret;
1125
1126     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1127     if (ret == CDS_DISC_OK)
1128         return 1;
1129     return 0;
1130 }
1131
1132 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1133 {
1134     BDRVRawState *s = bs->opaque;
1135
1136     if (eject_flag) {
1137         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1138             perror("CDROMEJECT");
1139     } else {
1140         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1141             perror("CDROMEJECT");
1142     }
1143
1144     return 0;
1145 }
1146
1147 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1148 {
1149     BDRVRawState *s = bs->opaque;
1150
1151     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1152         /*
1153          * Note: an error can happen if the distribution automatically
1154          * mounts the CD-ROM
1155          */
1156         /* perror("CDROM_LOCKDOOR"); */
1157     }
1158
1159     return 0;
1160 }
1161
1162 static BlockDriver bdrv_host_cdrom = {
1163     .format_name        = "host_cdrom",
1164     .instance_size      = sizeof(BDRVRawState),
1165     .bdrv_probe_device  = cdrom_probe_device,
1166     .bdrv_open          = cdrom_open,
1167     .bdrv_close         = raw_close,
1168     .bdrv_create        = hdev_create,
1169     .bdrv_flush         = raw_flush,
1170
1171     .bdrv_aio_readv     = raw_aio_readv,
1172     .bdrv_aio_writev    = raw_aio_writev,
1173
1174     .bdrv_read          = raw_read,
1175     .bdrv_write         = raw_write,
1176     .bdrv_getlength     = raw_getlength,
1177
1178     /* removable device support */
1179     .bdrv_is_inserted   = cdrom_is_inserted,
1180     .bdrv_eject         = cdrom_eject,
1181     .bdrv_set_locked    = cdrom_set_locked,
1182
1183     /* generic scsi device */
1184     .bdrv_ioctl         = hdev_ioctl,
1185     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1186 };
1187 #endif /* __linux__ */
1188
1189 #ifdef __FreeBSD__
1190 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1191 {
1192     BDRVRawState *s = bs->opaque;
1193     int ret;
1194
1195     s->type = FTYPE_CD;
1196
1197     ret = raw_open_common(bs, filename, flags, 0);
1198     if (ret)
1199         return ret;
1200
1201     /* make sure the door isnt locked at this time */
1202     ioctl(s->fd, CDIOCALLOW);
1203     return 0;
1204 }
1205
1206 static int cdrom_probe_device(const char *filename)
1207 {
1208     if (strstart(filename, "/dev/cd", NULL) ||
1209             strstart(filename, "/dev/acd", NULL))
1210         return 100;
1211     return 0;
1212 }
1213
1214 static int cdrom_reopen(BlockDriverState *bs)
1215 {
1216     BDRVRawState *s = bs->opaque;
1217     int fd;
1218
1219     /*
1220      * Force reread of possibly changed/newly loaded disc,
1221      * FreeBSD seems to not notice sometimes...
1222      */
1223     if (s->fd >= 0)
1224         close(s->fd);
1225     fd = open(bs->filename, s->open_flags, 0644);
1226     if (fd < 0) {
1227         s->fd = -1;
1228         return -EIO;
1229     }
1230     s->fd = fd;
1231
1232     /* make sure the door isnt locked at this time */
1233     ioctl(s->fd, CDIOCALLOW);
1234     return 0;
1235 }
1236
1237 static int cdrom_is_inserted(BlockDriverState *bs)
1238 {
1239     return raw_getlength(bs) > 0;
1240 }
1241
1242 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1243 {
1244     BDRVRawState *s = bs->opaque;
1245
1246     if (s->fd < 0)
1247         return -ENOTSUP;
1248
1249     (void) ioctl(s->fd, CDIOCALLOW);
1250
1251     if (eject_flag) {
1252         if (ioctl(s->fd, CDIOCEJECT) < 0)
1253             perror("CDIOCEJECT");
1254     } else {
1255         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1256             perror("CDIOCCLOSE");
1257     }
1258
1259     if (cdrom_reopen(bs) < 0)
1260         return -ENOTSUP;
1261     return 0;
1262 }
1263
1264 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1265 {
1266     BDRVRawState *s = bs->opaque;
1267
1268     if (s->fd < 0)
1269         return -ENOTSUP;
1270     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1271         /*
1272          * Note: an error can happen if the distribution automatically
1273          * mounts the CD-ROM
1274          */
1275         /* perror("CDROM_LOCKDOOR"); */
1276     }
1277
1278     return 0;
1279 }
1280
1281 static BlockDriver bdrv_host_cdrom = {
1282     .format_name        = "host_cdrom",
1283     .instance_size      = sizeof(BDRVRawState),
1284     .bdrv_probe_device  = cdrom_probe_device,
1285     .bdrv_open          = cdrom_open,
1286     .bdrv_close         = raw_close,
1287     .bdrv_create        = hdev_create,
1288     .bdrv_flush         = raw_flush,
1289
1290     .bdrv_aio_readv     = raw_aio_readv,
1291     .bdrv_aio_writev    = raw_aio_writev,
1292
1293     .bdrv_read          = raw_read,
1294     .bdrv_write         = raw_write,
1295     .bdrv_getlength     = raw_getlength,
1296
1297     /* removable device support */
1298     .bdrv_is_inserted   = cdrom_is_inserted,
1299     .bdrv_eject         = cdrom_eject,
1300     .bdrv_set_locked    = cdrom_set_locked,
1301 };
1302 #endif /* __FreeBSD__ */
1303
1304 static void bdrv_raw_init(void)
1305 {
1306     /*
1307      * Register all the drivers.  Note that order is important, the driver
1308      * registered last will get probed first.
1309      */
1310     bdrv_register(&bdrv_raw);
1311     bdrv_register(&bdrv_host_device);
1312 #ifdef __linux__
1313     bdrv_register(&bdrv_host_floppy);
1314     bdrv_register(&bdrv_host_cdrom);
1315 #endif
1316 #ifdef __FreeBSD__
1317     bdrv_register(&bdrv_host_cdrom);
1318 #endif
1319 }
1320
1321 block_init(bdrv_raw_init);