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 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
  26 #include "qemu-timer.h"
  27 #include "exec-all.h"
  28 #endif
  29 #include "block_int.h"
  30 #include <assert.h>
  31 #ifdef CONFIG_AIO
  32 #include <aio.h>
  33 #endif
  34
  35 #ifdef CONFIG_COCOA
  36 #include <paths.h>
  37 #include <sys/param.h>
  38 #include <IOKit/IOKitLib.h>
  39 #include <IOKit/IOBSD.h>
  40 #include <IOKit/storage/IOMediaBSDClient.h>
  41 #include <IOKit/storage/IOMedia.h>
  42 #include <IOKit/storage/IOCDMedia.h>
  43 //#include <IOKit/storage/IOCDTypes.h>
  44 #include <CoreFoundation/CoreFoundation.h>
  45 #endif
  46
  47 #ifdef __sun__
  48 #define _POSIX_PTHREAD_SEMANTICS 1
  49 #include <signal.h>
  50 #include <sys/dkio.h>
  51 #endif
  52 #ifdef __linux__
  53 #include <sys/ioctl.h>
  54 #include <linux/cdrom.h>
  55 #include <linux/fd.h>
  56 #endif
  57 #ifdef __FreeBSD__
  58 #include <sys/disk.h>
  59 #endif
  60
  61 #ifdef __OpenBSD__
  62 #include <sys/ioctl.h>
  63 #include <sys/disklabel.h>
  64 #include <sys/dkio.h>
  65 #endif
  66
  67 //#define DEBUG_FLOPPY
  68
  69 //#define DEBUG_BLOCK
  70 #if defined(DEBUG_BLOCK) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
  71 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  72     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  73 #else
  74 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  75 #endif
  76
  77 #define FTYPE_FILE   0
  78 #define FTYPE_CD     1
  79 #define FTYPE_FD     2
  80
  81 #define ALIGNED_BUFFER_SIZE (32 * 512)
  82
  83 /* if the FD is not accessed during that time (in ms), we try to
  84    reopen it to see if the disk has been changed */
  85 #define FD_OPEN_TIMEOUT 1000
  86
  87 typedef struct BDRVRawState {
  88     int fd;
  89     int type;
  90     unsigned int lseek_err_cnt;
  91 #if defined(__linux__)
  92     /* linux floppy specific */
  93     int fd_open_flags;
  94     int64_t fd_open_time;
  95     int64_t fd_error_time;
  96     int fd_got_error;
  97     int fd_media_changed;
  98 #endif
  99 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 100     uint8_t* aligned_buf;
 101 #endif
 102 } BDRVRawState;
 103
 104 static int fd_open(BlockDriverState *bs);
 105
 106 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 107 {
 108     BDRVRawState *s = bs->opaque;
 109     int fd, open_flags, ret;
 110
 111     s->lseek_err_cnt = 0;
 112
 113     open_flags = O_BINARY;
 114     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 115         open_flags |= O_RDWR;
 116     } else {
 117         open_flags |= O_RDONLY;
 118         bs->read_only = 1;
 119     }
 120     if (flags & BDRV_O_CREAT)
 121         open_flags |= O_CREAT | O_TRUNC;
 122 #ifdef O_DIRECT
 123     if (flags & BDRV_O_DIRECT)
 124         open_flags |= O_DIRECT;
 125 #endif
 126
 127     s->type = FTYPE_FILE;
 128
 129     fd = open(filename, open_flags, 0644);
 130     if (fd < 0) {
 131         ret = -errno;
 132         if (ret == -EROFS)
 133             ret = -EACCES;
 134         return ret;
 135     }
 136     s->fd = fd;
 137 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 138     s->aligned_buf = NULL;
 139     if (flags & BDRV_O_DIRECT) {
 140         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 141         if (s->aligned_buf == NULL) {
 142             ret = -errno;
 143             close(fd);
 144             return ret;
 145         }
 146     }
 147 #endif
 148     return 0;
 149 }
 150
 151 /* XXX: use host sector size if necessary with:
 152 #ifdef DIOCGSECTORSIZE
 153         {
 154             unsigned int sectorsize = 512;
 155             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 156                 sectorsize > bufsize)
 157                 bufsize = sectorsize;
 158         }
 159 #endif
 160 #ifdef CONFIG_COCOA
 161         u_int32_t   blockSize = 512;
 162         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 163             bufsize = blockSize;
 164         }
 165 #endif
 166 */
 167
 168 /*
 169  * offset and count are in bytes, but must be multiples of 512 for files
 170  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 171  *
 172  * This function may be called without alignment if the caller ensures
 173  * that O_DIRECT is not in effect.
 174  */
 175 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 176                      uint8_t *buf, int count)
 177 {
 178     BDRVRawState *s = bs->opaque;
 179     int ret;
 180
 181     ret = fd_open(bs);
 182     if (ret < 0)
 183         return ret;
 184
 185     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 186         ++(s->lseek_err_cnt);
 187         if(s->lseek_err_cnt <= 10) {
 188             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 189                               "] lseek failed : %d = %s\n",
 190                               s->fd, bs->filename, offset, buf, count,
 191                               bs->total_sectors, errno, strerror(errno));
 192         }
 193         return -1;
 194     }
 195     s->lseek_err_cnt=0;
 196
 197     ret = read(s->fd, buf, count);
 198     if (ret == count)
 199         goto label__raw_read__success;
 200
 201     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 202                       "] read failed %d : %d = %s\n",
 203                       s->fd, bs->filename, offset, buf, count,
 204                       bs->total_sectors, ret, errno, strerror(errno));
 205
 206     /* Try harder for CDrom. */
 207     if (bs->type == BDRV_TYPE_CDROM) {
 208         lseek(s->fd, offset, SEEK_SET);
 209         ret = read(s->fd, buf, count);
 210         if (ret == count)
 211             goto label__raw_read__success;
 212         lseek(s->fd, offset, SEEK_SET);
 213         ret = read(s->fd, buf, count);
 214         if (ret == count)
 215             goto label__raw_read__success;
 216
 217         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 218                           "] retry read failed %d : %d = %s\n",
 219                           s->fd, bs->filename, offset, buf, count,
 220                           bs->total_sectors, ret, errno, strerror(errno));
 221     }
 222
 223 label__raw_read__success:
 224
 225     return ret;
 226 }
 227
 228 /*
 229  * offset and count are in bytes, but must be multiples of 512 for files
 230  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 231  *
 232  * This function may be called without alignment if the caller ensures
 233  * that O_DIRECT is not in effect.
 234  */
 235 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 236                       const uint8_t *buf, int count)
 237 {
 238     BDRVRawState *s = bs->opaque;
 239     int ret;
 240
 241     ret = fd_open(bs);
 242     if (ret < 0)
 243         return ret;
 244
 245     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 246         ++(s->lseek_err_cnt);
 247         if(s->lseek_err_cnt) {
 248             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 249                               PRId64 "] lseek failed : %d = %s\n",
 250                               s->fd, bs->filename, offset, buf, count,
 251                               bs->total_sectors, errno, strerror(errno));
 252         }
 253         return -1;
 254     }
 255     s->lseek_err_cnt = 0;
 256
 257     ret = write(s->fd, buf, count);
 258     if (ret == count)
 259         goto label__raw_write__success;
 260
 261     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 262                       "] write failed %d : %d = %s\n",
 263                       s->fd, bs->filename, offset, buf, count,
 264                       bs->total_sectors, ret, errno, strerror(errno));
 265
 266 label__raw_write__success:
 267
 268     return ret;
 269 }
 270
 271
 272 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 273 /*
 274  * offset and count are in bytes and possibly not aligned. For files opened
 275  * with O_DIRECT, necessary alignments are ensured before calling
 276  * raw_pread_aligned to do the actual read.
 277  */
 278 static int raw_pread(BlockDriverState *bs, int64_t offset,
 279                      uint8_t *buf, int count)
 280 {
 281     BDRVRawState *s = bs->opaque;
 282     int size, ret, shift, sum;
 283
 284     sum = 0;
 285
 286     if (s->aligned_buf != NULL)  {
 287
 288         if (offset & 0x1ff) {
 289             /* align offset on a 512 bytes boundary */
 290
 291             shift = offset & 0x1ff;
 292             size = (shift + count + 0x1ff) & ~0x1ff;
 293             if (size > ALIGNED_BUFFER_SIZE)
 294                 size = ALIGNED_BUFFER_SIZE;
 295             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 296             if (ret < 0)
 297                 return ret;
 298
 299             size = 512 - shift;
 300             if (size > count)
 301                 size = count;
 302             memcpy(buf, s->aligned_buf + shift, size);
 303
 304             buf += size;
 305             offset += size;
 306             count -= size;
 307             sum += size;
 308
 309             if (count == 0)
 310                 return sum;
 311         }
 312         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 313
 314             /* read on aligned buffer */
 315
 316             while (count) {
 317
 318                 size = (count + 0x1ff) & ~0x1ff;
 319                 if (size > ALIGNED_BUFFER_SIZE)
 320                     size = ALIGNED_BUFFER_SIZE;
 321
 322                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 323                 if (ret < 0)
 324                     return ret;
 325
 326                 size = ret;
 327                 if (size > count)
 328                     size = count;
 329
 330                 memcpy(buf, s->aligned_buf, size);
 331
 332                 buf += size;
 333                 offset += size;
 334                 count -= size;
 335                 sum += size;
 336             }
 337
 338             return sum;
 339         }
 340     }
 341
 342     return raw_pread_aligned(bs, offset, buf, count) + sum;
 343 }
 344
 345 /*
 346  * offset and count are in bytes and possibly not aligned. For files opened
 347  * with O_DIRECT, necessary alignments are ensured before calling
 348  * raw_pwrite_aligned to do the actual write.
 349  */
 350 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 351                       const uint8_t *buf, int count)
 352 {
 353     BDRVRawState *s = bs->opaque;
 354     int size, ret, shift, sum;
 355
 356     sum = 0;
 357
 358     if (s->aligned_buf != NULL) {
 359
 360         if (offset & 0x1ff) {
 361             /* align offset on a 512 bytes boundary */
 362             shift = offset & 0x1ff;
 363             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 364             if (ret < 0)
 365                 return ret;
 366
 367             size = 512 - shift;
 368             if (size > count)
 369                 size = count;
 370             memcpy(s->aligned_buf + shift, buf, size);
 371
 372             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 373             if (ret < 0)
 374                 return ret;
 375
 376             buf += size;
 377             offset += size;
 378             count -= size;
 379             sum += size;
 380
 381             if (count == 0)
 382                 return sum;
 383         }
 384         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 385
 386             while ((size = (count & ~0x1ff)) != 0) {
 387
 388                 if (size > ALIGNED_BUFFER_SIZE)
 389                     size = ALIGNED_BUFFER_SIZE;
 390
 391                 memcpy(s->aligned_buf, buf, size);
 392
 393                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 394                 if (ret < 0)
 395                     return ret;
 396
 397                 buf += ret;
 398                 offset += ret;
 399                 count -= ret;
 400                 sum += ret;
 401             }
 402             /* here, count < 512 because (count & ~0x1ff) == 0 */
 403             if (count) {
 404                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 405                 if (ret < 0)
 406                     return ret;
 407                  memcpy(s->aligned_buf, buf, count);
 408
 409                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 410                  if (ret < 0)
 411                      return ret;
 412                  if (count < ret)
 413                      ret = count;
 414
 415                  sum += ret;
 416             }
 417             return sum;
 418         }
 419     }
 420     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 421 }
 422
 423 #else
 424 #define raw_pread raw_pread_aligned
 425 #define raw_pwrite raw_pwrite_aligned
 426 #endif
 427
 428
 429 #ifdef CONFIG_AIO
 430 /***********************************************************/
 431 /* Unix AIO using POSIX AIO */
 432
 433 typedef struct RawAIOCB {
 434     BlockDriverAIOCB common;
 435     struct aiocb aiocb;
 436     struct RawAIOCB *next;
 437     int ret;
 438 } RawAIOCB;
 439
 440 static int aio_sig_num = SIGUSR2;
 441 static RawAIOCB *first_aio; /* AIO issued */
 442 static int aio_initialized = 0;
 443
 444 static void aio_signal_handler(int signum)
 445 {
 446 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 447     CPUState *env = cpu_single_env;
 448     if (env) {
 449         /* stop the currently executing cpu because a timer occured */
 450         cpu_interrupt(env, CPU_INTERRUPT_EXIT);
 451 #ifdef USE_KQEMU
 452         if (env->kqemu_enabled) {
 453             kqemu_cpu_interrupt(env);
 454         }
 455 #endif
 456     }
 457 #endif
 458 }
 459
 460 void qemu_aio_init(void)
 461 {
 462     struct sigaction act;
 463
 464     aio_initialized = 1;
 465
 466     sigfillset(&act.sa_mask);
 467     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 468     act.sa_handler = aio_signal_handler;
 469     sigaction(aio_sig_num, &act, NULL);
 470
 471 #if defined(__GLIBC__) && defined(__linux__)
 472     {
 473         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 474            seems to fix the problem. */
 475         struct aioinit ai;
 476         memset(&ai, 0, sizeof(ai));
 477         ai.aio_threads = 1;
 478         ai.aio_num = 1;
 479         ai.aio_idle_time = 365 * 100000;
 480         aio_init(&ai);
 481     }
 482 #endif
 483 }
 484
 485 void qemu_aio_poll(void)
 486 {
 487     RawAIOCB *acb, **pacb;
 488     int ret;
 489
 490     for(;;) {
 491         pacb = &first_aio;
 492         for(;;) {
 493             acb = *pacb;
 494             if (!acb)
 495                 goto the_end;
 496             ret = aio_error(&acb->aiocb);
 497             if (ret == ECANCELED) {
 498                 /* remove the request */
 499                 *pacb = acb->next;
 500                 qemu_aio_release(acb);
 501             } else if (ret != EINPROGRESS) {
 502                 /* end of aio */
 503                 if (ret == 0) {
 504                     ret = aio_return(&acb->aiocb);
 505                     if (ret == acb->aiocb.aio_nbytes)
 506                         ret = 0;
 507                     else
 508                         ret = -EINVAL;
 509                 } else {
 510                     ret = -ret;
 511                 }
 512                 /* remove the request */
 513                 *pacb = acb->next;
 514                 /* call the callback */
 515                 acb->common.cb(acb->common.opaque, ret);
 516                 qemu_aio_release(acb);
 517                 break;
 518             } else {
 519                 pacb = &acb->next;
 520             }
 521         }
 522     }
 523  the_end: ;
 524 }
 525
 526 /* Wait for all IO requests to complete.  */
 527 void qemu_aio_flush(void)
 528 {
 529     qemu_aio_wait_start();
 530     qemu_aio_poll();
 531     while (first_aio) {
 532         qemu_aio_wait();
 533     }
 534     qemu_aio_wait_end();
 535 }
 536
 537 /* wait until at least one AIO was handled */
 538 static sigset_t wait_oset;
 539
 540 void qemu_aio_wait_start(void)
 541 {
 542     sigset_t set;
 543
 544     if (!aio_initialized)
 545         qemu_aio_init();
 546     sigemptyset(&set);
 547     sigaddset(&set, aio_sig_num);
 548     sigprocmask(SIG_BLOCK, &set, &wait_oset);
 549 }
 550
 551 void qemu_aio_wait(void)
 552 {
 553     sigset_t set;
 554     int nb_sigs;
 555
 556 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 557     if (qemu_bh_poll())
 558         return;
 559 #endif
 560     sigemptyset(&set);
 561     sigaddset(&set, aio_sig_num);
 562     sigwait(&set, &nb_sigs);
 563     qemu_aio_poll();
 564 }
 565
 566 void qemu_aio_wait_end(void)
 567 {
 568     sigprocmask(SIG_SETMASK, &wait_oset, NULL);
 569 }
 570
 571 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 572         int64_t sector_num, uint8_t *buf, int nb_sectors,
 573         BlockDriverCompletionFunc *cb, void *opaque)
 574 {
 575     BDRVRawState *s = bs->opaque;
 576     RawAIOCB *acb;
 577
 578     if (fd_open(bs) < 0)
 579         return NULL;
 580
 581     acb = qemu_aio_get(bs, cb, opaque);
 582     if (!acb)
 583         return NULL;
 584     acb->aiocb.aio_fildes = s->fd;
 585     acb->aiocb.aio_sigevent.sigev_signo = aio_sig_num;
 586     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 587     acb->aiocb.aio_buf = buf;
 588     if (nb_sectors < 0)
 589         acb->aiocb.aio_nbytes = -nb_sectors;
 590     else
 591         acb->aiocb.aio_nbytes = nb_sectors * 512;
 592     acb->aiocb.aio_offset = sector_num * 512;
 593     acb->next = first_aio;
 594     first_aio = acb;
 595     return acb;
 596 }
 597
 598 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 599 static void raw_aio_em_cb(void* opaque)
 600 {
 601     RawAIOCB *acb = opaque;
 602     acb->common.cb(acb->common.opaque, acb->ret);
 603     qemu_aio_release(acb);
 604 }
 605 #endif
 606
 607 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 608         int64_t sector_num, uint8_t *buf, int nb_sectors,
 609         BlockDriverCompletionFunc *cb, void *opaque)
 610 {
 611     RawAIOCB *acb;
 612
 613     /*
 614      * If O_DIRECT is used and the buffer is not aligned fall back
 615      * to synchronous IO.
 616      */
 617 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 618     BDRVRawState *s = bs->opaque;
 619
 620     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 621         QEMUBH *bh;
 622         acb = qemu_aio_get(bs, cb, opaque);
 623         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 624         bh = qemu_bh_new(raw_aio_em_cb, acb);
 625         qemu_bh_schedule(bh);
 626         return &acb->common;
 627     }
 628 #endif
 629
 630     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 631     if (!acb)
 632         return NULL;
 633     if (aio_read(&acb->aiocb) < 0) {
 634         qemu_aio_release(acb);
 635         return NULL;
 636     }
 637     return &acb->common;
 638 }
 639
 640 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 641         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 642         BlockDriverCompletionFunc *cb, void *opaque)
 643 {
 644     RawAIOCB *acb;
 645
 646     /*
 647      * If O_DIRECT is used and the buffer is not aligned fall back
 648      * to synchronous IO.
 649      */
 650 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 651     BDRVRawState *s = bs->opaque;
 652
 653     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 654         QEMUBH *bh;
 655         acb = qemu_aio_get(bs, cb, opaque);
 656         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 657         bh = qemu_bh_new(raw_aio_em_cb, acb);
 658         qemu_bh_schedule(bh);
 659         return &acb->common;
 660     }
 661 #endif
 662
 663     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 664     if (!acb)
 665         return NULL;
 666     if (aio_write(&acb->aiocb) < 0) {
 667         qemu_aio_release(acb);
 668         return NULL;
 669     }
 670     return &acb->common;
 671 }
 672
 673 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 674 {
 675     int ret;
 676     RawAIOCB *acb = (RawAIOCB *)blockacb;
 677     RawAIOCB **pacb;
 678
 679     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 680     if (ret == AIO_NOTCANCELED) {
 681         /* fail safe: if the aio could not be canceled, we wait for
 682            it */
 683         while (aio_error(&acb->aiocb) == EINPROGRESS);
 684     }
 685
 686     /* remove the callback from the queue */
 687     pacb = &first_aio;
 688     for(;;) {
 689         if (*pacb == NULL) {
 690             break;
 691         } else if (*pacb == acb) {
 692             *pacb = acb->next;
 693             qemu_aio_release(acb);
 694             break;
 695         }
 696         pacb = &acb->next;
 697     }
 698 }
 699
 700 # else /* CONFIG_AIO */
 701
 702 void qemu_aio_init(void)
 703 {
 704 }
 705
 706 void qemu_aio_poll(void)
 707 {
 708 }
 709
 710 void qemu_aio_flush(void)
 711 {
 712 }
 713
 714 void qemu_aio_wait_start(void)
 715 {
 716 }
 717
 718 void qemu_aio_wait(void)
 719 {
 720 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 721     qemu_bh_poll();
 722 #endif
 723 }
 724
 725 void qemu_aio_wait_end(void)
 726 {
 727 }
 728
 729 #endif /* CONFIG_AIO */
 730
 731 static void raw_close(BlockDriverState *bs)
 732 {
 733     BDRVRawState *s = bs->opaque;
 734     if (s->fd >= 0) {
 735         close(s->fd);
 736         s->fd = -1;
 737 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 738         if (s->aligned_buf != NULL)
 739             qemu_free(s->aligned_buf);
 740 #endif
 741     }
 742 }
 743
 744 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 745 {
 746     BDRVRawState *s = bs->opaque;
 747     if (s->type != FTYPE_FILE)
 748         return -ENOTSUP;
 749     if (ftruncate(s->fd, offset) < 0)
 750         return -errno;
 751     return 0;
 752 }
 753
 754 #ifdef __OpenBSD__
 755 static int64_t raw_getlength(BlockDriverState *bs)
 756 {
 757     BDRVRawState *s = bs->opaque;
 758     int fd = s->fd;
 759     struct stat st;
 760
 761     if (fstat(fd, &st))
 762         return -1;
 763     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 764         struct disklabel dl;
 765
 766         if (ioctl(fd, DIOCGDINFO, &dl))
 767             return -1;
 768         return (uint64_t)dl.d_secsize *
 769             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 770     } else
 771         return st.st_size;
 772 }
 773 #else /* !__OpenBSD__ */
 774 static int64_t  raw_getlength(BlockDriverState *bs)
 775 {
 776     BDRVRawState *s = bs->opaque;
 777     int fd = s->fd;
 778     int64_t size;
 779 #ifdef _BSD
 780     struct stat sb;
 781 #endif
 782 #ifdef __sun__
 783     struct dk_minfo minfo;
 784     int rv;
 785 #endif
 786     int ret;
 787
 788     ret = fd_open(bs);
 789     if (ret < 0)
 790         return ret;
 791
 792 #ifdef _BSD
 793     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 794 #ifdef DIOCGMEDIASIZE
 795         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 796 #endif
 797 #ifdef CONFIG_COCOA
 798         size = LONG_LONG_MAX;
 799 #else
 800         size = lseek(fd, 0LL, SEEK_END);
 801 #endif
 802     } else
 803 #endif
 804 #ifdef __sun__
 805     /*
 806      * use the DKIOCGMEDIAINFO ioctl to read the size.
 807      */
 808     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 809     if ( rv != -1 ) {
 810         size = minfo.dki_lbsize * minfo.dki_capacity;
 811     } else /* there are reports that lseek on some devices
 812               fails, but irc discussion said that contingency
 813               on contingency was overkill */
 814 #endif
 815     {
 816         size = lseek(fd, 0, SEEK_END);
 817     }
 818     return size;
 819 }
 820 #endif
 821
 822 static int raw_create(const char *filename, int64_t total_size,
 823                       const char *backing_file, int flags)
 824 {
 825     int fd;
 826
 827     if (flags || backing_file)
 828         return -ENOTSUP;
 829
 830     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 831               0644);
 832     if (fd < 0)
 833         return -EIO;
 834     ftruncate(fd, total_size * 512);
 835     close(fd);
 836     return 0;
 837 }
 838
 839 static void raw_flush(BlockDriverState *bs)
 840 {
 841     BDRVRawState *s = bs->opaque;
 842     fsync(s->fd);
 843 }
 844
 845 BlockDriver bdrv_raw = {
 846     "raw",
 847     sizeof(BDRVRawState),
 848     NULL, /* no probe for protocols */
 849     raw_open,
 850     NULL,
 851     NULL,
 852     raw_close,
 853     raw_create,
 854     raw_flush,
 855
 856 #ifdef CONFIG_AIO
 857     .bdrv_aio_read = raw_aio_read,
 858     .bdrv_aio_write = raw_aio_write,
 859     .bdrv_aio_cancel = raw_aio_cancel,
 860     .aiocb_size = sizeof(RawAIOCB),
 861 #endif
 862     .protocol_name = "file",
 863     .bdrv_pread = raw_pread,
 864     .bdrv_pwrite = raw_pwrite,
 865     .bdrv_truncate = raw_truncate,
 866     .bdrv_getlength = raw_getlength,
 867 };
 868
 869 /***********************************************/
 870 /* host device */
 871
 872 #ifdef CONFIG_COCOA
 873 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 874 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 875
 876 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 877 {
 878     kern_return_t       kernResult;
 879     mach_port_t     masterPort;
 880     CFMutableDictionaryRef  classesToMatch;
 881
 882     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 883     if ( KERN_SUCCESS != kernResult ) {
 884         printf( "IOMasterPort returned %d\n", kernResult );
 885     }
 886
 887     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 888     if ( classesToMatch == NULL ) {
 889         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 890     } else {
 891     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 892     }
 893     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 894     if ( KERN_SUCCESS != kernResult )
 895     {
 896         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 897     }
 898
 899     return kernResult;
 900 }
 901
 902 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 903 {
 904     io_object_t     nextMedia;
 905     kern_return_t   kernResult = KERN_FAILURE;
 906     *bsdPath = '\0';
 907     nextMedia = IOIteratorNext( mediaIterator );
 908     if ( nextMedia )
 909     {
 910         CFTypeRef   bsdPathAsCFString;
 911     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 912         if ( bsdPathAsCFString ) {
 913             size_t devPathLength;
 914             strcpy( bsdPath, _PATH_DEV );
 915             strcat( bsdPath, "r" );
 916             devPathLength = strlen( bsdPath );
 917             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 918                 kernResult = KERN_SUCCESS;
 919             }
 920             CFRelease( bsdPathAsCFString );
 921         }
 922         IOObjectRelease( nextMedia );
 923     }
 924
 925     return kernResult;
 926 }
 927
 928 #endif
 929
 930 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 931 {
 932     BDRVRawState *s = bs->opaque;
 933     int fd, open_flags, ret;
 934
 935 #ifdef CONFIG_COCOA
 936     if (strstart(filename, "/dev/cdrom", NULL)) {
 937         kern_return_t kernResult;
 938         io_iterator_t mediaIterator;
 939         char bsdPath[ MAXPATHLEN ];
 940         int fd;
 941
 942         kernResult = FindEjectableCDMedia( &mediaIterator );
 943         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 944
 945         if ( bsdPath[ 0 ] != '\0' ) {
 946             strcat(bsdPath,"s0");
 947             /* some CDs don't have a partition 0 */
 948             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 949             if (fd < 0) {
 950                 bsdPath[strlen(bsdPath)-1] = '1';
 951             } else {
 952                 close(fd);
 953             }
 954             filename = bsdPath;
 955         }
 956
 957         if ( mediaIterator )
 958             IOObjectRelease( mediaIterator );
 959     }
 960 #endif
 961     open_flags = O_BINARY;
 962     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 963         open_flags |= O_RDWR;
 964     } else {
 965         open_flags |= O_RDONLY;
 966         bs->read_only = 1;
 967     }
 968 #ifdef O_DIRECT
 969     if (flags & BDRV_O_DIRECT)
 970         open_flags |= O_DIRECT;
 971 #endif
 972
 973     s->type = FTYPE_FILE;
 974 #if defined(__linux__)
 975     if (strstart(filename, "/dev/cd", NULL)) {
 976         /* open will not fail even if no CD is inserted */
 977         open_flags |= O_NONBLOCK;
 978         s->type = FTYPE_CD;
 979     } else if (strstart(filename, "/dev/fd", NULL)) {
 980         s->type = FTYPE_FD;
 981         s->fd_open_flags = open_flags;
 982         /* open will not fail even if no floppy is inserted */
 983         open_flags |= O_NONBLOCK;
 984     } else if (strstart(filename, "/dev/sg", NULL)) {
 985         bs->sg = 1;
 986     }
 987 #endif
 988     fd = open(filename, open_flags, 0644);
 989     if (fd < 0) {
 990         ret = -errno;
 991         if (ret == -EROFS)
 992             ret = -EACCES;
 993         return ret;
 994     }
 995     s->fd = fd;
 996 #if defined(__linux__)
 997     /* close fd so that we can reopen it as needed */
 998     if (s->type == FTYPE_FD) {
 999         close(s->fd);
1000         s->fd = -1;
1001         s->fd_media_changed = 1;
1002     }
1003 #endif
1004     return 0;
1005 }
1006
1007 #if defined(__linux__) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
1008
1009 /* Note: we do not have a reliable method to detect if the floppy is
1010    present. The current method is to try to open the floppy at every
1011    I/O and to keep it opened during a few hundreds of ms. */
1012 static int fd_open(BlockDriverState *bs)
1013 {
1014     BDRVRawState *s = bs->opaque;
1015     int last_media_present;
1016
1017     if (s->type != FTYPE_FD)
1018         return 0;
1019     last_media_present = (s->fd >= 0);
1020     if (s->fd >= 0 &&
1021         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1022         close(s->fd);
1023         s->fd = -1;
1024 #ifdef DEBUG_FLOPPY
1025         printf("Floppy closed\n");
1026 #endif
1027     }
1028     if (s->fd < 0) {
1029         if (s->fd_got_error &&
1030             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1031 #ifdef DEBUG_FLOPPY
1032             printf("No floppy (open delayed)\n");
1033 #endif
1034             return -EIO;
1035         }
1036         s->fd = open(bs->filename, s->fd_open_flags);
1037         if (s->fd < 0) {
1038             s->fd_error_time = qemu_get_clock(rt_clock);
1039             s->fd_got_error = 1;
1040             if (last_media_present)
1041                 s->fd_media_changed = 1;
1042 #ifdef DEBUG_FLOPPY
1043             printf("No floppy\n");
1044 #endif
1045             return -EIO;
1046         }
1047 #ifdef DEBUG_FLOPPY
1048         printf("Floppy opened\n");
1049 #endif
1050     }
1051     if (!last_media_present)
1052         s->fd_media_changed = 1;
1053     s->fd_open_time = qemu_get_clock(rt_clock);
1054     s->fd_got_error = 0;
1055     return 0;
1056 }
1057 #else
1058 static int fd_open(BlockDriverState *bs)
1059 {
1060     return 0;
1061 }
1062 #endif
1063
1064 #if defined(__linux__)
1065
1066 static int raw_is_inserted(BlockDriverState *bs)
1067 {
1068     BDRVRawState *s = bs->opaque;
1069     int ret;
1070
1071     switch(s->type) {
1072     case FTYPE_CD:
1073         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1074         if (ret == CDS_DISC_OK)
1075             return 1;
1076         else
1077             return 0;
1078         break;
1079     case FTYPE_FD:
1080         ret = fd_open(bs);
1081         return (ret >= 0);
1082     default:
1083         return 1;
1084     }
1085 }
1086
1087 /* currently only used by fdc.c, but a CD version would be good too */
1088 static int raw_media_changed(BlockDriverState *bs)
1089 {
1090     BDRVRawState *s = bs->opaque;
1091
1092     switch(s->type) {
1093     case FTYPE_FD:
1094         {
1095             int ret;
1096             /* XXX: we do not have a true media changed indication. It
1097                does not work if the floppy is changed without trying
1098                to read it */
1099             fd_open(bs);
1100             ret = s->fd_media_changed;
1101             s->fd_media_changed = 0;
1102 #ifdef DEBUG_FLOPPY
1103             printf("Floppy changed=%d\n", ret);
1104 #endif
1105             return ret;
1106         }
1107     default:
1108         return -ENOTSUP;
1109     }
1110 }
1111
1112 static int raw_eject(BlockDriverState *bs, int eject_flag)
1113 {
1114     BDRVRawState *s = bs->opaque;
1115
1116     switch(s->type) {
1117     case FTYPE_CD:
1118         if (eject_flag) {
1119             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1120                 perror("CDROMEJECT");
1121         } else {
1122             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1123                 perror("CDROMEJECT");
1124         }
1125         break;
1126     case FTYPE_FD:
1127         {
1128             int fd;
1129             if (s->fd >= 0) {
1130                 close(s->fd);
1131                 s->fd = -1;
1132             }
1133             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1134             if (fd >= 0) {
1135                 if (ioctl(fd, FDEJECT, 0) < 0)
1136                     perror("FDEJECT");
1137                 close(fd);
1138             }
1139         }
1140         break;
1141     default:
1142         return -ENOTSUP;
1143     }
1144     return 0;
1145 }
1146
1147 static int raw_set_locked(BlockDriverState *bs, int locked)
1148 {
1149     BDRVRawState *s = bs->opaque;
1150
1151     switch(s->type) {
1152     case FTYPE_CD:
1153         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1154             /* Note: an error can happen if the distribution automatically
1155                mounts the CD-ROM */
1156             //        perror("CDROM_LOCKDOOR");
1157         }
1158         break;
1159     default:
1160         return -ENOTSUP;
1161     }
1162     return 0;
1163 }
1164
1165 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1166 {
1167     BDRVRawState *s = bs->opaque;
1168
1169     return ioctl(s->fd, req, buf);
1170 }
1171 #else
1172
1173 static int raw_is_inserted(BlockDriverState *bs)
1174 {
1175     return 1;
1176 }
1177
1178 static int raw_media_changed(BlockDriverState *bs)
1179 {
1180     return -ENOTSUP;
1181 }
1182
1183 static int raw_eject(BlockDriverState *bs, int eject_flag)
1184 {
1185     return -ENOTSUP;
1186 }
1187
1188 static int raw_set_locked(BlockDriverState *bs, int locked)
1189 {
1190     return -ENOTSUP;
1191 }
1192
1193 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1194 {
1195     return -ENOTSUP;
1196 }
1197 #endif /* !linux */
1198
1199 BlockDriver bdrv_host_device = {
1200     "host_device",
1201     sizeof(BDRVRawState),
1202     NULL, /* no probe for protocols */
1203     hdev_open,
1204     NULL,
1205     NULL,
1206     raw_close,
1207     NULL,
1208     raw_flush,
1209
1210 #ifdef CONFIG_AIO
1211     .bdrv_aio_read = raw_aio_read,
1212     .bdrv_aio_write = raw_aio_write,
1213     .bdrv_aio_cancel = raw_aio_cancel,
1214     .aiocb_size = sizeof(RawAIOCB),
1215 #endif
1216     .bdrv_pread = raw_pread,
1217     .bdrv_pwrite = raw_pwrite,
1218     .bdrv_getlength = raw_getlength,
1219
1220     /* removable device support */
1221     .bdrv_is_inserted = raw_is_inserted,
1222     .bdrv_media_changed = raw_media_changed,
1223     .bdrv_eject = raw_eject,
1224     .bdrv_set_locked = raw_set_locked,
1225     /* generic scsi device */
1226     .bdrv_ioctl = raw_ioctl,
1227 };