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 #include "qemu-char.h"
  29 #endif
  30 #include "block_int.h"
  31 #include "compatfd.h"
  32 #include <assert.h>
  33 #ifdef CONFIG_AIO
  34 #include <aio.h>
  35 #endif
  36
  37 #ifdef CONFIG_COCOA
  38 #include <paths.h>
  39 #include <sys/param.h>
  40 #include <IOKit/IOKitLib.h>
  41 #include <IOKit/IOBSD.h>
  42 #include <IOKit/storage/IOMediaBSDClient.h>
  43 #include <IOKit/storage/IOMedia.h>
  44 #include <IOKit/storage/IOCDMedia.h>
  45 //#include <IOKit/storage/IOCDTypes.h>
  46 #include <CoreFoundation/CoreFoundation.h>
  47 #endif
  48
  49 #ifdef __sun__
  50 #define _POSIX_PTHREAD_SEMANTICS 1
  51 #include <signal.h>
  52 #include <sys/dkio.h>
  53 #endif
  54 #ifdef __linux__
  55 #include <sys/ioctl.h>
  56 #include <linux/cdrom.h>
  57 #include <linux/fd.h>
  58 #endif
  59 #ifdef __FreeBSD__
  60 #include <signal.h>
  61 #include <sys/disk.h>
  62 #endif
  63
  64 #ifdef __OpenBSD__
  65 #include <sys/ioctl.h>
  66 #include <sys/disklabel.h>
  67 #include <sys/dkio.h>
  68 #endif
  69
  70 //#define DEBUG_FLOPPY
  71
  72 //#define DEBUG_BLOCK
  73 #if defined(DEBUG_BLOCK) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
  74 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  75     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  76 #else
  77 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  78 #endif
  79
  80 #define FTYPE_FILE   0
  81 #define FTYPE_CD     1
  82 #define FTYPE_FD     2
  83
  84 #define ALIGNED_BUFFER_SIZE (32 * 512)
  85
  86 /* if the FD is not accessed during that time (in ms), we try to
  87    reopen it to see if the disk has been changed */
  88 #define FD_OPEN_TIMEOUT 1000
  89
  90 typedef struct BDRVRawState {
  91     int fd;
  92     int type;
  93     unsigned int lseek_err_cnt;
  94 #if defined(__linux__)
  95     /* linux floppy specific */
  96     int fd_open_flags;
  97     int64_t fd_open_time;
  98     int64_t fd_error_time;
  99     int fd_got_error;
 100     int fd_media_changed;
 101 #endif
 102 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 103     uint8_t* aligned_buf;
 104 #endif
 105 } BDRVRawState;
 106
 107 static int fd_open(BlockDriverState *bs);
 108
 109 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 110 {
 111     BDRVRawState *s = bs->opaque;
 112     int fd, open_flags, ret;
 113
 114     s->lseek_err_cnt = 0;
 115
 116     open_flags = O_BINARY;
 117     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 118         open_flags |= O_RDWR;
 119     } else {
 120         open_flags |= O_RDONLY;
 121         bs->read_only = 1;
 122     }
 123     if (flags & BDRV_O_CREAT)
 124         open_flags |= O_CREAT | O_TRUNC;
 125 #ifdef O_DIRECT
 126     if (flags & BDRV_O_DIRECT)
 127         open_flags |= O_DIRECT;
 128 #endif
 129
 130     s->type = FTYPE_FILE;
 131
 132     fd = open(filename, open_flags, 0644);
 133     if (fd < 0) {
 134         ret = -errno;
 135         if (ret == -EROFS)
 136             ret = -EACCES;
 137         return ret;
 138     }
 139     s->fd = fd;
 140 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 141     s->aligned_buf = NULL;
 142     if (flags & BDRV_O_DIRECT) {
 143         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 144         if (s->aligned_buf == NULL) {
 145             ret = -errno;
 146             close(fd);
 147             return ret;
 148         }
 149     }
 150 #endif
 151     return 0;
 152 }
 153
 154 /* XXX: use host sector size if necessary with:
 155 #ifdef DIOCGSECTORSIZE
 156         {
 157             unsigned int sectorsize = 512;
 158             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 159                 sectorsize > bufsize)
 160                 bufsize = sectorsize;
 161         }
 162 #endif
 163 #ifdef CONFIG_COCOA
 164         u_int32_t   blockSize = 512;
 165         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 166             bufsize = blockSize;
 167         }
 168 #endif
 169 */
 170
 171 /*
 172  * offset and count are in bytes, but must be multiples of 512 for files
 173  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 174  *
 175  * This function may be called without alignment if the caller ensures
 176  * that O_DIRECT is not in effect.
 177  */
 178 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 179                      uint8_t *buf, int count)
 180 {
 181     BDRVRawState *s = bs->opaque;
 182     int ret;
 183
 184     ret = fd_open(bs);
 185     if (ret < 0)
 186         return ret;
 187
 188     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 189         ++(s->lseek_err_cnt);
 190         if(s->lseek_err_cnt <= 10) {
 191             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 192                               "] lseek failed : %d = %s\n",
 193                               s->fd, bs->filename, offset, buf, count,
 194                               bs->total_sectors, errno, strerror(errno));
 195         }
 196         return -1;
 197     }
 198     s->lseek_err_cnt=0;
 199
 200     ret = read(s->fd, buf, count);
 201     if (ret == count)
 202         goto label__raw_read__success;
 203
 204     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 205                       "] read failed %d : %d = %s\n",
 206                       s->fd, bs->filename, offset, buf, count,
 207                       bs->total_sectors, ret, errno, strerror(errno));
 208
 209     /* Try harder for CDrom. */
 210     if (bs->type == BDRV_TYPE_CDROM) {
 211         lseek(s->fd, offset, SEEK_SET);
 212         ret = read(s->fd, buf, count);
 213         if (ret == count)
 214             goto label__raw_read__success;
 215         lseek(s->fd, offset, SEEK_SET);
 216         ret = read(s->fd, buf, count);
 217         if (ret == count)
 218             goto label__raw_read__success;
 219
 220         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 221                           "] retry read failed %d : %d = %s\n",
 222                           s->fd, bs->filename, offset, buf, count,
 223                           bs->total_sectors, ret, errno, strerror(errno));
 224     }
 225
 226 label__raw_read__success:
 227
 228     return ret;
 229 }
 230
 231 /*
 232  * offset and count are in bytes, but must be multiples of 512 for files
 233  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 234  *
 235  * This function may be called without alignment if the caller ensures
 236  * that O_DIRECT is not in effect.
 237  */
 238 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 239                       const uint8_t *buf, int count)
 240 {
 241     BDRVRawState *s = bs->opaque;
 242     int ret;
 243
 244     ret = fd_open(bs);
 245     if (ret < 0)
 246         return ret;
 247
 248     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 249         ++(s->lseek_err_cnt);
 250         if(s->lseek_err_cnt) {
 251             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 252                               PRId64 "] lseek failed : %d = %s\n",
 253                               s->fd, bs->filename, offset, buf, count,
 254                               bs->total_sectors, errno, strerror(errno));
 255         }
 256         return -1;
 257     }
 258     s->lseek_err_cnt = 0;
 259
 260     ret = write(s->fd, buf, count);
 261     if (ret == count)
 262         goto label__raw_write__success;
 263
 264     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 265                       "] write failed %d : %d = %s\n",
 266                       s->fd, bs->filename, offset, buf, count,
 267                       bs->total_sectors, ret, errno, strerror(errno));
 268
 269 label__raw_write__success:
 270
 271     return ret;
 272 }
 273
 274
 275 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 276 /*
 277  * offset and count are in bytes and possibly not aligned. For files opened
 278  * with O_DIRECT, necessary alignments are ensured before calling
 279  * raw_pread_aligned to do the actual read.
 280  */
 281 static int raw_pread(BlockDriverState *bs, int64_t offset,
 282                      uint8_t *buf, int count)
 283 {
 284     BDRVRawState *s = bs->opaque;
 285     int size, ret, shift, sum;
 286
 287     sum = 0;
 288
 289     if (s->aligned_buf != NULL)  {
 290
 291         if (offset & 0x1ff) {
 292             /* align offset on a 512 bytes boundary */
 293
 294             shift = offset & 0x1ff;
 295             size = (shift + count + 0x1ff) & ~0x1ff;
 296             if (size > ALIGNED_BUFFER_SIZE)
 297                 size = ALIGNED_BUFFER_SIZE;
 298             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 299             if (ret < 0)
 300                 return ret;
 301
 302             size = 512 - shift;
 303             if (size > count)
 304                 size = count;
 305             memcpy(buf, s->aligned_buf + shift, size);
 306
 307             buf += size;
 308             offset += size;
 309             count -= size;
 310             sum += size;
 311
 312             if (count == 0)
 313                 return sum;
 314         }
 315         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 316
 317             /* read on aligned buffer */
 318
 319             while (count) {
 320
 321                 size = (count + 0x1ff) & ~0x1ff;
 322                 if (size > ALIGNED_BUFFER_SIZE)
 323                     size = ALIGNED_BUFFER_SIZE;
 324
 325                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 326                 if (ret < 0)
 327                     return ret;
 328
 329                 size = ret;
 330                 if (size > count)
 331                     size = count;
 332
 333                 memcpy(buf, s->aligned_buf, size);
 334
 335                 buf += size;
 336                 offset += size;
 337                 count -= size;
 338                 sum += size;
 339             }
 340
 341             return sum;
 342         }
 343     }
 344
 345     return raw_pread_aligned(bs, offset, buf, count) + sum;
 346 }
 347
 348 /*
 349  * offset and count are in bytes and possibly not aligned. For files opened
 350  * with O_DIRECT, necessary alignments are ensured before calling
 351  * raw_pwrite_aligned to do the actual write.
 352  */
 353 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 354                       const uint8_t *buf, int count)
 355 {
 356     BDRVRawState *s = bs->opaque;
 357     int size, ret, shift, sum;
 358
 359     sum = 0;
 360
 361     if (s->aligned_buf != NULL) {
 362
 363         if (offset & 0x1ff) {
 364             /* align offset on a 512 bytes boundary */
 365             shift = offset & 0x1ff;
 366             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 367             if (ret < 0)
 368                 return ret;
 369
 370             size = 512 - shift;
 371             if (size > count)
 372                 size = count;
 373             memcpy(s->aligned_buf + shift, buf, size);
 374
 375             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 376             if (ret < 0)
 377                 return ret;
 378
 379             buf += size;
 380             offset += size;
 381             count -= size;
 382             sum += size;
 383
 384             if (count == 0)
 385                 return sum;
 386         }
 387         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 388
 389             while ((size = (count & ~0x1ff)) != 0) {
 390
 391                 if (size > ALIGNED_BUFFER_SIZE)
 392                     size = ALIGNED_BUFFER_SIZE;
 393
 394                 memcpy(s->aligned_buf, buf, size);
 395
 396                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 397                 if (ret < 0)
 398                     return ret;
 399
 400                 buf += ret;
 401                 offset += ret;
 402                 count -= ret;
 403                 sum += ret;
 404             }
 405             /* here, count < 512 because (count & ~0x1ff) == 0 */
 406             if (count) {
 407                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 408                 if (ret < 0)
 409                     return ret;
 410                  memcpy(s->aligned_buf, buf, count);
 411
 412                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 413                  if (ret < 0)
 414                      return ret;
 415                  if (count < ret)
 416                      ret = count;
 417
 418                  sum += ret;
 419             }
 420             return sum;
 421         }
 422     }
 423     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 424 }
 425
 426 #else
 427 #define raw_pread raw_pread_aligned
 428 #define raw_pwrite raw_pwrite_aligned
 429 #endif
 430
 431
 432 #ifdef CONFIG_AIO
 433 /***********************************************************/
 434 /* Unix AIO using POSIX AIO */
 435
 436 typedef struct RawAIOCB {
 437     BlockDriverAIOCB common;
 438     struct aiocb aiocb;
 439     struct RawAIOCB *next;
 440     int ret;
 441 } RawAIOCB;
 442
 443 static int aio_sig_fd = -1;
 444 static int aio_sig_num = SIGUSR2;
 445 static RawAIOCB *first_aio; /* AIO issued */
 446 static int aio_initialized = 0;
 447
 448 static void qemu_aio_poll(void *opaque)
 449 {
 450     RawAIOCB *acb, **pacb;
 451     int ret;
 452     size_t offset;
 453     union {
 454         struct qemu_signalfd_siginfo siginfo;
 455         char buf[128];
 456     } sig;
 457
 458     /* try to read from signalfd, don't freak out if we can't read anything */
 459     offset = 0;
 460     while (offset < 128) {
 461         ssize_t len;
 462
 463         len = read(aio_sig_fd, sig.buf + offset, 128 - offset);
 464         if (len == -1 && errno == EINTR)
 465             continue;
 466         if (len == -1 && errno == EAGAIN) {
 467             /* there is no natural reason for this to happen,
 468              * so we'll spin hard until we get everything just
 469              * to be on the safe side. */
 470             if (offset > 0)
 471                 continue;
 472         }
 473
 474         offset += len;
 475     }
 476
 477     for(;;) {
 478         pacb = &first_aio;
 479         for(;;) {
 480             acb = *pacb;
 481             if (!acb)
 482                 goto the_end;
 483             ret = aio_error(&acb->aiocb);
 484             if (ret == ECANCELED) {
 485                 /* remove the request */
 486                 *pacb = acb->next;
 487                 qemu_aio_release(acb);
 488             } else if (ret != EINPROGRESS) {
 489                 /* end of aio */
 490                 if (ret == 0) {
 491                     ret = aio_return(&acb->aiocb);
 492                     if (ret == acb->aiocb.aio_nbytes)
 493                         ret = 0;
 494                     else
 495                         ret = -EINVAL;
 496                 } else {
 497                     ret = -ret;
 498                 }
 499                 /* remove the request */
 500                 *pacb = acb->next;
 501                 /* call the callback */
 502                 acb->common.cb(acb->common.opaque, ret);
 503                 qemu_aio_release(acb);
 504                 break;
 505             } else {
 506                 pacb = &acb->next;
 507             }
 508         }
 509     }
 510  the_end: ;
 511 }
 512
 513 void qemu_aio_init(void)
 514 {
 515     sigset_t mask;
 516
 517     aio_initialized = 1;
 518
 519     /* Make sure to block AIO signal */
 520     sigemptyset(&mask);
 521     sigaddset(&mask, aio_sig_num);
 522     sigprocmask(SIG_BLOCK, &mask, NULL);
 523
 524     aio_sig_fd = qemu_signalfd(&mask);
 525
 526     fcntl(aio_sig_fd, F_SETFL, O_NONBLOCK);
 527
 528 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 529     qemu_set_fd_handler2(aio_sig_fd, NULL, qemu_aio_poll, NULL, NULL);
 530 #endif
 531
 532 #if defined(__GLIBC__) && defined(__linux__)
 533     {
 534         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 535            seems to fix the problem. */
 536         struct aioinit ai;
 537         memset(&ai, 0, sizeof(ai));
 538         ai.aio_threads = 1;
 539         ai.aio_num = 1;
 540         ai.aio_idle_time = 365 * 100000;
 541         aio_init(&ai);
 542     }
 543 #endif
 544 }
 545
 546 /* Wait for all IO requests to complete.  */
 547 void qemu_aio_flush(void)
 548 {
 549     qemu_aio_poll(NULL);
 550     while (first_aio) {
 551         qemu_aio_wait();
 552     }
 553 }
 554
 555 void qemu_aio_wait(void)
 556 {
 557     int ret;
 558
 559 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 560     if (qemu_bh_poll())
 561         return;
 562 #endif
 563
 564     if (!first_aio)
 565         return;
 566
 567     do {
 568         fd_set rdfds;
 569
 570         FD_ZERO(&rdfds);
 571         FD_SET(aio_sig_fd, &rdfds);
 572
 573         ret = select(aio_sig_fd + 1, &rdfds, NULL, NULL, NULL);
 574         if (ret == -1 && errno == EINTR)
 575             continue;
 576     } while (ret == 0);
 577
 578     qemu_aio_poll(NULL);
 579 }
 580
 581 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 582         int64_t sector_num, uint8_t *buf, int nb_sectors,
 583         BlockDriverCompletionFunc *cb, void *opaque)
 584 {
 585     BDRVRawState *s = bs->opaque;
 586     RawAIOCB *acb;
 587
 588     if (fd_open(bs) < 0)
 589         return NULL;
 590
 591     acb = qemu_aio_get(bs, cb, opaque);
 592     if (!acb)
 593         return NULL;
 594     acb->aiocb.aio_fildes = s->fd;
 595     acb->aiocb.aio_sigevent.sigev_signo = aio_sig_num;
 596     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 597     acb->aiocb.aio_buf = buf;
 598     if (nb_sectors < 0)
 599         acb->aiocb.aio_nbytes = -nb_sectors;
 600     else
 601         acb->aiocb.aio_nbytes = nb_sectors * 512;
 602     acb->aiocb.aio_offset = sector_num * 512;
 603     acb->next = first_aio;
 604     first_aio = acb;
 605     return acb;
 606 }
 607
 608 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 609 static void raw_aio_em_cb(void* opaque)
 610 {
 611     RawAIOCB *acb = opaque;
 612     acb->common.cb(acb->common.opaque, acb->ret);
 613     qemu_aio_release(acb);
 614 }
 615 #endif
 616
 617 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 618         int64_t sector_num, uint8_t *buf, int nb_sectors,
 619         BlockDriverCompletionFunc *cb, void *opaque)
 620 {
 621     RawAIOCB *acb;
 622
 623     /*
 624      * If O_DIRECT is used and the buffer is not aligned fall back
 625      * to synchronous IO.
 626      */
 627 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 628     BDRVRawState *s = bs->opaque;
 629
 630     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 631         QEMUBH *bh;
 632         acb = qemu_aio_get(bs, cb, opaque);
 633         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 634         bh = qemu_bh_new(raw_aio_em_cb, acb);
 635         qemu_bh_schedule(bh);
 636         return &acb->common;
 637     }
 638 #endif
 639
 640     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 641     if (!acb)
 642         return NULL;
 643     if (aio_read(&acb->aiocb) < 0) {
 644         qemu_aio_release(acb);
 645         return NULL;
 646     }
 647     return &acb->common;
 648 }
 649
 650 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 651         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 652         BlockDriverCompletionFunc *cb, void *opaque)
 653 {
 654     RawAIOCB *acb;
 655
 656     /*
 657      * If O_DIRECT is used and the buffer is not aligned fall back
 658      * to synchronous IO.
 659      */
 660 #if defined(O_DIRECT) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
 661     BDRVRawState *s = bs->opaque;
 662
 663     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 664         QEMUBH *bh;
 665         acb = qemu_aio_get(bs, cb, opaque);
 666         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 667         bh = qemu_bh_new(raw_aio_em_cb, acb);
 668         qemu_bh_schedule(bh);
 669         return &acb->common;
 670     }
 671 #endif
 672
 673     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 674     if (!acb)
 675         return NULL;
 676     if (aio_write(&acb->aiocb) < 0) {
 677         qemu_aio_release(acb);
 678         return NULL;
 679     }
 680     return &acb->common;
 681 }
 682
 683 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 684 {
 685     int ret;
 686     RawAIOCB *acb = (RawAIOCB *)blockacb;
 687     RawAIOCB **pacb;
 688
 689     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 690     if (ret == AIO_NOTCANCELED) {
 691         /* fail safe: if the aio could not be canceled, we wait for
 692            it */
 693         while (aio_error(&acb->aiocb) == EINPROGRESS);
 694     }
 695
 696     /* remove the callback from the queue */
 697     pacb = &first_aio;
 698     for(;;) {
 699         if (*pacb == NULL) {
 700             break;
 701         } else if (*pacb == acb) {
 702             *pacb = acb->next;
 703             qemu_aio_release(acb);
 704             break;
 705         }
 706         pacb = &acb->next;
 707     }
 708 }
 709
 710 # else /* CONFIG_AIO */
 711
 712 void qemu_aio_init(void)
 713 {
 714 }
 715
 716 void qemu_aio_flush(void)
 717 {
 718 }
 719
 720 void qemu_aio_wait(void)
 721 {
 722 #if !defined(QEMU_IMG) && !defined(QEMU_NBD)
 723     qemu_bh_poll();
 724 #endif
 725 }
 726
 727 #endif /* CONFIG_AIO */
 728
 729 static void raw_close(BlockDriverState *bs)
 730 {
 731     BDRVRawState *s = bs->opaque;
 732     if (s->fd >= 0) {
 733         close(s->fd);
 734         s->fd = -1;
 735 #if defined(O_DIRECT) && !defined(QEMU_IMG)
 736         if (s->aligned_buf != NULL)
 737             qemu_free(s->aligned_buf);
 738 #endif
 739     }
 740 }
 741
 742 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 743 {
 744     BDRVRawState *s = bs->opaque;
 745     if (s->type != FTYPE_FILE)
 746         return -ENOTSUP;
 747     if (ftruncate(s->fd, offset) < 0)
 748         return -errno;
 749     return 0;
 750 }
 751
 752 #ifdef __OpenBSD__
 753 static int64_t raw_getlength(BlockDriverState *bs)
 754 {
 755     BDRVRawState *s = bs->opaque;
 756     int fd = s->fd;
 757     struct stat st;
 758
 759     if (fstat(fd, &st))
 760         return -1;
 761     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 762         struct disklabel dl;
 763
 764         if (ioctl(fd, DIOCGDINFO, &dl))
 765             return -1;
 766         return (uint64_t)dl.d_secsize *
 767             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 768     } else
 769         return st.st_size;
 770 }
 771 #else /* !__OpenBSD__ */
 772 static int64_t  raw_getlength(BlockDriverState *bs)
 773 {
 774     BDRVRawState *s = bs->opaque;
 775     int fd = s->fd;
 776     int64_t size;
 777 #ifdef _BSD
 778     struct stat sb;
 779 #endif
 780 #ifdef __sun__
 781     struct dk_minfo minfo;
 782     int rv;
 783 #endif
 784     int ret;
 785
 786     ret = fd_open(bs);
 787     if (ret < 0)
 788         return ret;
 789
 790 #ifdef _BSD
 791     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 792 #ifdef DIOCGMEDIASIZE
 793         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 794 #endif
 795 #ifdef CONFIG_COCOA
 796         size = LONG_LONG_MAX;
 797 #else
 798         size = lseek(fd, 0LL, SEEK_END);
 799 #endif
 800     } else
 801 #endif
 802 #ifdef __sun__
 803     /*
 804      * use the DKIOCGMEDIAINFO ioctl to read the size.
 805      */
 806     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 807     if ( rv != -1 ) {
 808         size = minfo.dki_lbsize * minfo.dki_capacity;
 809     } else /* there are reports that lseek on some devices
 810               fails, but irc discussion said that contingency
 811               on contingency was overkill */
 812 #endif
 813     {
 814         size = lseek(fd, 0, SEEK_END);
 815     }
 816     return size;
 817 }
 818 #endif
 819
 820 static int raw_create(const char *filename, int64_t total_size,
 821                       const char *backing_file, int flags)
 822 {
 823     int fd;
 824
 825     if (flags || backing_file)
 826         return -ENOTSUP;
 827
 828     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 829               0644);
 830     if (fd < 0)
 831         return -EIO;
 832     ftruncate(fd, total_size * 512);
 833     close(fd);
 834     return 0;
 835 }
 836
 837 static void raw_flush(BlockDriverState *bs)
 838 {
 839     BDRVRawState *s = bs->opaque;
 840     fsync(s->fd);
 841 }
 842
 843 BlockDriver bdrv_raw = {
 844     "raw",
 845     sizeof(BDRVRawState),
 846     NULL, /* no probe for protocols */
 847     raw_open,
 848     NULL,
 849     NULL,
 850     raw_close,
 851     raw_create,
 852     raw_flush,
 853
 854 #ifdef CONFIG_AIO
 855     .bdrv_aio_read = raw_aio_read,
 856     .bdrv_aio_write = raw_aio_write,
 857     .bdrv_aio_cancel = raw_aio_cancel,
 858     .aiocb_size = sizeof(RawAIOCB),
 859 #endif
 860     .protocol_name = "file",
 861     .bdrv_pread = raw_pread,
 862     .bdrv_pwrite = raw_pwrite,
 863     .bdrv_truncate = raw_truncate,
 864     .bdrv_getlength = raw_getlength,
 865 };
 866
 867 /***********************************************/
 868 /* host device */
 869
 870 #ifdef CONFIG_COCOA
 871 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 872 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 873
 874 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 875 {
 876     kern_return_t       kernResult;
 877     mach_port_t     masterPort;
 878     CFMutableDictionaryRef  classesToMatch;
 879
 880     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 881     if ( KERN_SUCCESS != kernResult ) {
 882         printf( "IOMasterPort returned %d\n", kernResult );
 883     }
 884
 885     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 886     if ( classesToMatch == NULL ) {
 887         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 888     } else {
 889     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 890     }
 891     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 892     if ( KERN_SUCCESS != kernResult )
 893     {
 894         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 895     }
 896
 897     return kernResult;
 898 }
 899
 900 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 901 {
 902     io_object_t     nextMedia;
 903     kern_return_t   kernResult = KERN_FAILURE;
 904     *bsdPath = '\0';
 905     nextMedia = IOIteratorNext( mediaIterator );
 906     if ( nextMedia )
 907     {
 908         CFTypeRef   bsdPathAsCFString;
 909     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 910         if ( bsdPathAsCFString ) {
 911             size_t devPathLength;
 912             strcpy( bsdPath, _PATH_DEV );
 913             strcat( bsdPath, "r" );
 914             devPathLength = strlen( bsdPath );
 915             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 916                 kernResult = KERN_SUCCESS;
 917             }
 918             CFRelease( bsdPathAsCFString );
 919         }
 920         IOObjectRelease( nextMedia );
 921     }
 922
 923     return kernResult;
 924 }
 925
 926 #endif
 927
 928 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 929 {
 930     BDRVRawState *s = bs->opaque;
 931     int fd, open_flags, ret;
 932
 933 #ifdef CONFIG_COCOA
 934     if (strstart(filename, "/dev/cdrom", NULL)) {
 935         kern_return_t kernResult;
 936         io_iterator_t mediaIterator;
 937         char bsdPath[ MAXPATHLEN ];
 938         int fd;
 939
 940         kernResult = FindEjectableCDMedia( &mediaIterator );
 941         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 942
 943         if ( bsdPath[ 0 ] != '\0' ) {
 944             strcat(bsdPath,"s0");
 945             /* some CDs don't have a partition 0 */
 946             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 947             if (fd < 0) {
 948                 bsdPath[strlen(bsdPath)-1] = '1';
 949             } else {
 950                 close(fd);
 951             }
 952             filename = bsdPath;
 953         }
 954
 955         if ( mediaIterator )
 956             IOObjectRelease( mediaIterator );
 957     }
 958 #endif
 959     open_flags = O_BINARY;
 960     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 961         open_flags |= O_RDWR;
 962     } else {
 963         open_flags |= O_RDONLY;
 964         bs->read_only = 1;
 965     }
 966 #ifdef O_DIRECT
 967     if (flags & BDRV_O_DIRECT)
 968         open_flags |= O_DIRECT;
 969 #endif
 970
 971     s->type = FTYPE_FILE;
 972 #if defined(__linux__)
 973     if (strstart(filename, "/dev/cd", NULL)) {
 974         /* open will not fail even if no CD is inserted */
 975         open_flags |= O_NONBLOCK;
 976         s->type = FTYPE_CD;
 977     } else if (strstart(filename, "/dev/fd", NULL)) {
 978         s->type = FTYPE_FD;
 979         s->fd_open_flags = open_flags;
 980         /* open will not fail even if no floppy is inserted */
 981         open_flags |= O_NONBLOCK;
 982     } else if (strstart(filename, "/dev/sg", NULL)) {
 983         bs->sg = 1;
 984     }
 985 #endif
 986     fd = open(filename, open_flags, 0644);
 987     if (fd < 0) {
 988         ret = -errno;
 989         if (ret == -EROFS)
 990             ret = -EACCES;
 991         return ret;
 992     }
 993     s->fd = fd;
 994 #if defined(__linux__)
 995     /* close fd so that we can reopen it as needed */
 996     if (s->type == FTYPE_FD) {
 997         close(s->fd);
 998         s->fd = -1;
 999         s->fd_media_changed = 1;
1000     }
1001 #endif
1002     return 0;
1003 }
1004
1005 #if defined(__linux__) && !defined(QEMU_IMG) && !defined(QEMU_NBD)
1006
1007 /* Note: we do not have a reliable method to detect if the floppy is
1008    present. The current method is to try to open the floppy at every
1009    I/O and to keep it opened during a few hundreds of ms. */
1010 static int fd_open(BlockDriverState *bs)
1011 {
1012     BDRVRawState *s = bs->opaque;
1013     int last_media_present;
1014
1015     if (s->type != FTYPE_FD)
1016         return 0;
1017     last_media_present = (s->fd >= 0);
1018     if (s->fd >= 0 &&
1019         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1020         close(s->fd);
1021         s->fd = -1;
1022 #ifdef DEBUG_FLOPPY
1023         printf("Floppy closed\n");
1024 #endif
1025     }
1026     if (s->fd < 0) {
1027         if (s->fd_got_error &&
1028             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1029 #ifdef DEBUG_FLOPPY
1030             printf("No floppy (open delayed)\n");
1031 #endif
1032             return -EIO;
1033         }
1034         s->fd = open(bs->filename, s->fd_open_flags);
1035         if (s->fd < 0) {
1036             s->fd_error_time = qemu_get_clock(rt_clock);
1037             s->fd_got_error = 1;
1038             if (last_media_present)
1039                 s->fd_media_changed = 1;
1040 #ifdef DEBUG_FLOPPY
1041             printf("No floppy\n");
1042 #endif
1043             return -EIO;
1044         }
1045 #ifdef DEBUG_FLOPPY
1046         printf("Floppy opened\n");
1047 #endif
1048     }
1049     if (!last_media_present)
1050         s->fd_media_changed = 1;
1051     s->fd_open_time = qemu_get_clock(rt_clock);
1052     s->fd_got_error = 0;
1053     return 0;
1054 }
1055 #else
1056 static int fd_open(BlockDriverState *bs)
1057 {
1058     return 0;
1059 }
1060 #endif
1061
1062 #if defined(__linux__)
1063
1064 static int raw_is_inserted(BlockDriverState *bs)
1065 {
1066     BDRVRawState *s = bs->opaque;
1067     int ret;
1068
1069     switch(s->type) {
1070     case FTYPE_CD:
1071         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1072         if (ret == CDS_DISC_OK)
1073             return 1;
1074         else
1075             return 0;
1076         break;
1077     case FTYPE_FD:
1078         ret = fd_open(bs);
1079         return (ret >= 0);
1080     default:
1081         return 1;
1082     }
1083 }
1084
1085 /* currently only used by fdc.c, but a CD version would be good too */
1086 static int raw_media_changed(BlockDriverState *bs)
1087 {
1088     BDRVRawState *s = bs->opaque;
1089
1090     switch(s->type) {
1091     case FTYPE_FD:
1092         {
1093             int ret;
1094             /* XXX: we do not have a true media changed indication. It
1095                does not work if the floppy is changed without trying
1096                to read it */
1097             fd_open(bs);
1098             ret = s->fd_media_changed;
1099             s->fd_media_changed = 0;
1100 #ifdef DEBUG_FLOPPY
1101             printf("Floppy changed=%d\n", ret);
1102 #endif
1103             return ret;
1104         }
1105     default:
1106         return -ENOTSUP;
1107     }
1108 }
1109
1110 static int raw_eject(BlockDriverState *bs, int eject_flag)
1111 {
1112     BDRVRawState *s = bs->opaque;
1113
1114     switch(s->type) {
1115     case FTYPE_CD:
1116         if (eject_flag) {
1117             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1118                 perror("CDROMEJECT");
1119         } else {
1120             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1121                 perror("CDROMEJECT");
1122         }
1123         break;
1124     case FTYPE_FD:
1125         {
1126             int fd;
1127             if (s->fd >= 0) {
1128                 close(s->fd);
1129                 s->fd = -1;
1130             }
1131             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1132             if (fd >= 0) {
1133                 if (ioctl(fd, FDEJECT, 0) < 0)
1134                     perror("FDEJECT");
1135                 close(fd);
1136             }
1137         }
1138         break;
1139     default:
1140         return -ENOTSUP;
1141     }
1142     return 0;
1143 }
1144
1145 static int raw_set_locked(BlockDriverState *bs, int locked)
1146 {
1147     BDRVRawState *s = bs->opaque;
1148
1149     switch(s->type) {
1150     case FTYPE_CD:
1151         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1152             /* Note: an error can happen if the distribution automatically
1153                mounts the CD-ROM */
1154             //        perror("CDROM_LOCKDOOR");
1155         }
1156         break;
1157     default:
1158         return -ENOTSUP;
1159     }
1160     return 0;
1161 }
1162
1163 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1164 {
1165     BDRVRawState *s = bs->opaque;
1166
1167     return ioctl(s->fd, req, buf);
1168 }
1169 #else
1170
1171 static int raw_is_inserted(BlockDriverState *bs)
1172 {
1173     return 1;
1174 }
1175
1176 static int raw_media_changed(BlockDriverState *bs)
1177 {
1178     return -ENOTSUP;
1179 }
1180
1181 static int raw_eject(BlockDriverState *bs, int eject_flag)
1182 {
1183     return -ENOTSUP;
1184 }
1185
1186 static int raw_set_locked(BlockDriverState *bs, int locked)
1187 {
1188     return -ENOTSUP;
1189 }
1190
1191 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1192 {
1193     return -ENOTSUP;
1194 }
1195 #endif /* !linux */
1196
1197 BlockDriver bdrv_host_device = {
1198     "host_device",
1199     sizeof(BDRVRawState),
1200     NULL, /* no probe for protocols */
1201     hdev_open,
1202     NULL,
1203     NULL,
1204     raw_close,
1205     NULL,
1206     raw_flush,
1207
1208 #ifdef CONFIG_AIO
1209     .bdrv_aio_read = raw_aio_read,
1210     .bdrv_aio_write = raw_aio_write,
1211     .bdrv_aio_cancel = raw_aio_cancel,
1212     .aiocb_size = sizeof(RawAIOCB),
1213 #endif
1214     .bdrv_pread = raw_pread,
1215     .bdrv_pwrite = raw_pwrite,
1216     .bdrv_getlength = raw_getlength,
1217
1218     /* removable device support */
1219     .bdrv_is_inserted = raw_is_inserted,
1220     .bdrv_media_changed = raw_media_changed,
1221     .bdrv_eject = raw_eject,
1222     .bdrv_set_locked = raw_set_locked,
1223     /* generic scsi device */
1224     .bdrv_ioctl = raw_ioctl,
1225 };