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