block-raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "qemu-timer.h"
  26 #include "qemu-char.h"
  27 #include "block_int.h"
  28 #include <assert.h>
  29 #ifdef CONFIG_AIO
  30 #include "posix-aio-compat.h"
  31 #endif
  32
  33 #ifdef CONFIG_COCOA
  34 #include <paths.h>
  35 #include <sys/param.h>
  36 #include <IOKit/IOKitLib.h>
  37 #include <IOKit/IOBSD.h>
  38 #include <IOKit/storage/IOMediaBSDClient.h>
  39 #include <IOKit/storage/IOMedia.h>
  40 #include <IOKit/storage/IOCDMedia.h>
  41 //#include <IOKit/storage/IOCDTypes.h>
  42 #include <CoreFoundation/CoreFoundation.h>
  43 #endif
  44
  45 #ifdef __sun__
  46 #define _POSIX_PTHREAD_SEMANTICS 1
  47 #include <signal.h>
  48 #include <sys/dkio.h>
  49 #endif
  50 #ifdef __linux__
  51 #include <sys/ioctl.h>
  52 #include <linux/cdrom.h>
  53 #include <linux/fd.h>
  54 #endif
  55 #ifdef __FreeBSD__
  56 #include <signal.h>
  57 #include <sys/disk.h>
  58 #endif
  59
  60 #ifdef __OpenBSD__
  61 #include <sys/ioctl.h>
  62 #include <sys/disklabel.h>
  63 #include <sys/dkio.h>
  64 #endif
  65
  66 //#define DEBUG_FLOPPY
  67
  68 //#define DEBUG_BLOCK
  69 #if defined(DEBUG_BLOCK)
  70 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled())     \
  71     { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
  72 #else
  73 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  74 #endif
  75
  76 /* OS X does not have O_DSYNC */
  77 #ifndef O_DSYNC
  78 #ifdef O_SYNC
  79 #define O_DSYNC O_SYNC
  80 #elif defined(O_FSYNC)
  81 #define O_DSYNC O_FSYNC
  82 #endif
  83 #endif
  84
  85 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  86 #ifndef O_DIRECT
  87 #define O_DIRECT O_DSYNC
  88 #endif
  89
  90 #define FTYPE_FILE   0
  91 #define FTYPE_CD     1
  92 #define FTYPE_FD     2
  93
  94 #define ALIGNED_BUFFER_SIZE (32 * 512)
  95
  96 /* if the FD is not accessed during that time (in ms), we try to
  97    reopen it to see if the disk has been changed */
  98 #define FD_OPEN_TIMEOUT 1000
  99
 100 typedef struct BDRVRawState {
 101     int fd;
 102     int type;
 103     unsigned int lseek_err_cnt;
 104 #if defined(__linux__)
 105     /* linux floppy specific */
 106     int fd_open_flags;
 107     int64_t fd_open_time;
 108     int64_t fd_error_time;
 109     int fd_got_error;
 110     int fd_media_changed;
 111 #endif
 112     uint8_t* aligned_buf;
 113 } BDRVRawState;
 114
 115 static int posix_aio_init(void);
 116
 117 static int fd_open(BlockDriverState *bs);
 118
 119 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 120 {
 121     BDRVRawState *s = bs->opaque;
 122     int fd, open_flags, ret;
 123
 124     posix_aio_init();
 125
 126     s->lseek_err_cnt = 0;
 127
 128     open_flags = O_BINARY;
 129     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 130         open_flags |= O_RDWR;
 131     } else {
 132         open_flags |= O_RDONLY;
 133         bs->read_only = 1;
 134     }
 135     if (flags & BDRV_O_CREAT)
 136         open_flags |= O_CREAT | O_TRUNC;
 137
 138     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 139      * and O_DIRECT for no caching. */
 140     if ((flags & BDRV_O_NOCACHE))
 141         open_flags |= O_DIRECT;
 142     else if (!(flags & BDRV_O_CACHE_WB))
 143         open_flags |= O_DSYNC;
 144
 145     s->type = FTYPE_FILE;
 146
 147     fd = open(filename, open_flags, 0644);
 148     if (fd < 0) {
 149         ret = -errno;
 150         if (ret == -EROFS)
 151             ret = -EACCES;
 152         return ret;
 153     }
 154     s->fd = fd;
 155     s->aligned_buf = NULL;
 156     if ((flags & BDRV_O_NOCACHE)) {
 157         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 158         if (s->aligned_buf == NULL) {
 159             ret = -errno;
 160             close(fd);
 161             return ret;
 162         }
 163     }
 164     return 0;
 165 }
 166
 167 /* XXX: use host sector size if necessary with:
 168 #ifdef DIOCGSECTORSIZE
 169         {
 170             unsigned int sectorsize = 512;
 171             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 172                 sectorsize > bufsize)
 173                 bufsize = sectorsize;
 174         }
 175 #endif
 176 #ifdef CONFIG_COCOA
 177         u_int32_t   blockSize = 512;
 178         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 179             bufsize = blockSize;
 180         }
 181 #endif
 182 */
 183
 184 /*
 185  * offset and count are in bytes, but must be multiples of 512 for files
 186  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 187  *
 188  * This function may be called without alignment if the caller ensures
 189  * that O_DIRECT is not in effect.
 190  */
 191 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 192                      uint8_t *buf, int count)
 193 {
 194     BDRVRawState *s = bs->opaque;
 195     int ret;
 196
 197     ret = fd_open(bs);
 198     if (ret < 0)
 199         return ret;
 200
 201     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 202         ++(s->lseek_err_cnt);
 203         if(s->lseek_err_cnt <= 10) {
 204             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 205                               "] lseek failed : %d = %s\n",
 206                               s->fd, bs->filename, offset, buf, count,
 207                               bs->total_sectors, errno, strerror(errno));
 208         }
 209         return -1;
 210     }
 211     s->lseek_err_cnt=0;
 212
 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                       "] read failed %d : %d = %s\n",
 219                       s->fd, bs->filename, offset, buf, count,
 220                       bs->total_sectors, ret, errno, strerror(errno));
 221
 222     /* Try harder for CDrom. */
 223     if (bs->type == BDRV_TYPE_CDROM) {
 224         lseek(s->fd, offset, SEEK_SET);
 225         ret = read(s->fd, buf, count);
 226         if (ret == count)
 227             goto label__raw_read__success;
 228         lseek(s->fd, offset, SEEK_SET);
 229         ret = read(s->fd, buf, count);
 230         if (ret == count)
 231             goto label__raw_read__success;
 232
 233         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 234                           "] retry read failed %d : %d = %s\n",
 235                           s->fd, bs->filename, offset, buf, count,
 236                           bs->total_sectors, ret, errno, strerror(errno));
 237     }
 238
 239 label__raw_read__success:
 240
 241     return  (ret < 0) ? -errno : ret;
 242 }
 243
 244 /*
 245  * offset and count are in bytes, but must be multiples of 512 for files
 246  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 247  *
 248  * This function may be called without alignment if the caller ensures
 249  * that O_DIRECT is not in effect.
 250  */
 251 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 252                       const uint8_t *buf, int count)
 253 {
 254     BDRVRawState *s = bs->opaque;
 255     int ret;
 256
 257     ret = fd_open(bs);
 258     if (ret < 0)
 259         return -errno;
 260
 261     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 262         ++(s->lseek_err_cnt);
 263         if(s->lseek_err_cnt) {
 264             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 265                               PRId64 "] lseek failed : %d = %s\n",
 266                               s->fd, bs->filename, offset, buf, count,
 267                               bs->total_sectors, errno, strerror(errno));
 268         }
 269         return -EIO;
 270     }
 271     s->lseek_err_cnt = 0;
 272
 273     ret = write(s->fd, buf, count);
 274     if (ret == count)
 275         goto label__raw_write__success;
 276
 277     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 278                       "] write failed %d : %d = %s\n",
 279                       s->fd, bs->filename, offset, buf, count,
 280                       bs->total_sectors, ret, errno, strerror(errno));
 281
 282 label__raw_write__success:
 283
 284     return  (ret < 0) ? -errno : ret;
 285 }
 286
 287
 288 /*
 289  * offset and count are in bytes and possibly not aligned. For files opened
 290  * with O_DIRECT, necessary alignments are ensured before calling
 291  * raw_pread_aligned to do the actual read.
 292  */
 293 static int raw_pread(BlockDriverState *bs, int64_t offset,
 294                      uint8_t *buf, int count)
 295 {
 296     BDRVRawState *s = bs->opaque;
 297     int size, ret, shift, sum;
 298
 299     sum = 0;
 300
 301     if (s->aligned_buf != NULL)  {
 302
 303         if (offset & 0x1ff) {
 304             /* align offset on a 512 bytes boundary */
 305
 306             shift = offset & 0x1ff;
 307             size = (shift + count + 0x1ff) & ~0x1ff;
 308             if (size > ALIGNED_BUFFER_SIZE)
 309                 size = ALIGNED_BUFFER_SIZE;
 310             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 311             if (ret < 0)
 312                 return ret;
 313
 314             size = 512 - shift;
 315             if (size > count)
 316                 size = count;
 317             memcpy(buf, s->aligned_buf + shift, size);
 318
 319             buf += size;
 320             offset += size;
 321             count -= size;
 322             sum += size;
 323
 324             if (count == 0)
 325                 return sum;
 326         }
 327         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 328
 329             /* read on aligned buffer */
 330
 331             while (count) {
 332
 333                 size = (count + 0x1ff) & ~0x1ff;
 334                 if (size > ALIGNED_BUFFER_SIZE)
 335                     size = ALIGNED_BUFFER_SIZE;
 336
 337                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 338                 if (ret < 0)
 339                     return ret;
 340
 341                 size = ret;
 342                 if (size > count)
 343                     size = count;
 344
 345                 memcpy(buf, s->aligned_buf, size);
 346
 347                 buf += size;
 348                 offset += size;
 349                 count -= size;
 350                 sum += size;
 351             }
 352
 353             return sum;
 354         }
 355     }
 356
 357     return raw_pread_aligned(bs, offset, buf, count) + sum;
 358 }
 359
 360 /*
 361  * offset and count are in bytes and possibly not aligned. For files opened
 362  * with O_DIRECT, necessary alignments are ensured before calling
 363  * raw_pwrite_aligned to do the actual write.
 364  */
 365 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 366                       const uint8_t *buf, int count)
 367 {
 368     BDRVRawState *s = bs->opaque;
 369     int size, ret, shift, sum;
 370
 371     sum = 0;
 372
 373     if (s->aligned_buf != NULL) {
 374
 375         if (offset & 0x1ff) {
 376             /* align offset on a 512 bytes boundary */
 377             shift = offset & 0x1ff;
 378             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 379             if (ret < 0)
 380                 return ret;
 381
 382             size = 512 - shift;
 383             if (size > count)
 384                 size = count;
 385             memcpy(s->aligned_buf + shift, buf, size);
 386
 387             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 388             if (ret < 0)
 389                 return ret;
 390
 391             buf += size;
 392             offset += size;
 393             count -= size;
 394             sum += size;
 395
 396             if (count == 0)
 397                 return sum;
 398         }
 399         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 400
 401             while ((size = (count & ~0x1ff)) != 0) {
 402
 403                 if (size > ALIGNED_BUFFER_SIZE)
 404                     size = ALIGNED_BUFFER_SIZE;
 405
 406                 memcpy(s->aligned_buf, buf, size);
 407
 408                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 409                 if (ret < 0)
 410                     return ret;
 411
 412                 buf += ret;
 413                 offset += ret;
 414                 count -= ret;
 415                 sum += ret;
 416             }
 417             /* here, count < 512 because (count & ~0x1ff) == 0 */
 418             if (count) {
 419                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 420                 if (ret < 0)
 421                     return ret;
 422                  memcpy(s->aligned_buf, buf, count);
 423
 424                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 425                  if (ret < 0)
 426                      return ret;
 427                  if (count < ret)
 428                      ret = count;
 429
 430                  sum += ret;
 431             }
 432             return sum;
 433         }
 434     }
 435     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 436 }
 437
 438 #ifdef CONFIG_AIO
 439 /***********************************************************/
 440 /* Unix AIO using POSIX AIO */
 441
 442 typedef struct RawAIOCB {
 443     BlockDriverAIOCB common;
 444     struct qemu_paiocb aiocb;
 445     struct RawAIOCB *next;
 446     int ret;
 447 } RawAIOCB;
 448
 449 typedef struct PosixAioState
 450 {
 451     int rfd, wfd;
 452     RawAIOCB *first_aio;
 453 } PosixAioState;
 454
 455 static void posix_aio_read(void *opaque)
 456 {
 457     PosixAioState *s = opaque;
 458     RawAIOCB *acb, **pacb;
 459     int ret;
 460     ssize_t len;
 461
 462     /* read all bytes from signal pipe */
 463     for (;;) {
 464         char bytes[16];
 465
 466         len = read(s->rfd, bytes, sizeof(bytes));
 467         if (len == -1 && errno == EINTR)
 468             continue; /* try again */
 469         if (len == sizeof(bytes))
 470             continue; /* more to read */
 471         break;
 472     }
 473
 474     for(;;) {
 475         pacb = &s->first_aio;
 476         for(;;) {
 477             acb = *pacb;
 478             if (!acb)
 479                 goto the_end;
 480             ret = qemu_paio_error(&acb->aiocb);
 481             if (ret == ECANCELED) {
 482                 /* remove the request */
 483                 *pacb = acb->next;
 484                 qemu_aio_release(acb);
 485             } else if (ret != EINPROGRESS) {
 486                 /* end of aio */
 487                 if (ret == 0) {
 488                     ret = qemu_paio_return(&acb->aiocb);
 489                     if (ret == acb->aiocb.aio_nbytes)
 490                         ret = 0;
 491                     else
 492                         ret = -EINVAL;
 493                 } else {
 494                     ret = -ret;
 495                 }
 496                 /* remove the request */
 497                 *pacb = acb->next;
 498                 /* call the callback */
 499                 acb->common.cb(acb->common.opaque, ret);
 500                 qemu_aio_release(acb);
 501                 break;
 502             } else {
 503                 pacb = &acb->next;
 504             }
 505         }
 506     }
 507  the_end: ;
 508 }
 509
 510 static int posix_aio_flush(void *opaque)
 511 {
 512     PosixAioState *s = opaque;
 513     return !!s->first_aio;
 514 }
 515
 516 static PosixAioState *posix_aio_state;
 517
 518 static void aio_signal_handler(int signum)
 519 {
 520     if (posix_aio_state) {
 521         char byte = 0;
 522
 523         write(posix_aio_state->wfd, &byte, sizeof(byte));
 524     }
 525
 526     qemu_service_io();
 527 }
 528
 529 static int posix_aio_init(void)
 530 {
 531     struct sigaction act;
 532     PosixAioState *s;
 533     int fds[2];
 534     struct qemu_paioinit ai;
 535
 536     if (posix_aio_state)
 537         return 0;
 538
 539     s = qemu_malloc(sizeof(PosixAioState));
 540
 541     sigfillset(&act.sa_mask);
 542     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 543     act.sa_handler = aio_signal_handler;
 544     sigaction(SIGUSR2, &act, NULL);
 545
 546     s->first_aio = NULL;
 547     if (pipe(fds) == -1) {
 548         fprintf(stderr, "failed to create pipe\n");
 549         return -errno;
 550     }
 551
 552     s->rfd = fds[0];
 553     s->wfd = fds[1];
 554
 555     fcntl(s->rfd, F_SETFL, O_NONBLOCK);
 556     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 557
 558     qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
 559
 560     memset(&ai, 0, sizeof(ai));
 561     ai.aio_threads = 64;
 562     ai.aio_num = 64;
 563     qemu_paio_init(&ai);
 564
 565     posix_aio_state = s;
 566
 567     return 0;
 568 }
 569
 570 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 571         int64_t sector_num, uint8_t *buf, int nb_sectors,
 572         BlockDriverCompletionFunc *cb, void *opaque)
 573 {
 574     BDRVRawState *s = bs->opaque;
 575     RawAIOCB *acb;
 576
 577     if (fd_open(bs) < 0)
 578         return NULL;
 579
 580     acb = qemu_aio_get(bs, cb, opaque);
 581     if (!acb)
 582         return NULL;
 583     acb->aiocb.aio_fildes = s->fd;
 584     acb->aiocb.ev_signo = SIGUSR2;
 585     acb->aiocb.aio_buf = buf;
 586     if (nb_sectors < 0)
 587         acb->aiocb.aio_nbytes = -nb_sectors;
 588     else
 589         acb->aiocb.aio_nbytes = nb_sectors * 512;
 590     acb->aiocb.aio_offset = sector_num * 512;
 591     acb->next = posix_aio_state->first_aio;
 592     posix_aio_state->first_aio = acb;
 593     return acb;
 594 }
 595
 596 static void raw_aio_em_cb(void* opaque)
 597 {
 598     RawAIOCB *acb = opaque;
 599     acb->common.cb(acb->common.opaque, acb->ret);
 600     qemu_aio_release(acb);
 601 }
 602
 603 static void raw_aio_remove(RawAIOCB *acb)
 604 {
 605     RawAIOCB **pacb;
 606
 607     /* remove the callback from the queue */
 608     pacb = &posix_aio_state->first_aio;
 609     for(;;) {
 610         if (*pacb == NULL) {
 611             fprintf(stderr, "raw_aio_remove: aio request not found!\n");
 612             break;
 613         } else if (*pacb == acb) {
 614             *pacb = acb->next;
 615             qemu_aio_release(acb);
 616             break;
 617         }
 618         pacb = &(*pacb)->next;
 619     }
 620 }
 621
 622 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 623         int64_t sector_num, uint8_t *buf, int nb_sectors,
 624         BlockDriverCompletionFunc *cb, void *opaque)
 625 {
 626     RawAIOCB *acb;
 627
 628     /*
 629      * If O_DIRECT is used and the buffer is not aligned fall back
 630      * to synchronous IO.
 631      */
 632     BDRVRawState *s = bs->opaque;
 633
 634     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 635         QEMUBH *bh;
 636         acb = qemu_aio_get(bs, cb, opaque);
 637         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 638         bh = qemu_bh_new(raw_aio_em_cb, acb);
 639         qemu_bh_schedule(bh);
 640         return &acb->common;
 641     }
 642
 643     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 644     if (!acb)
 645         return NULL;
 646     if (qemu_paio_read(&acb->aiocb) < 0) {
 647         raw_aio_remove(acb);
 648         return NULL;
 649     }
 650     return &acb->common;
 651 }
 652
 653 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 654         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 655         BlockDriverCompletionFunc *cb, void *opaque)
 656 {
 657     RawAIOCB *acb;
 658
 659     /*
 660      * If O_DIRECT is used and the buffer is not aligned fall back
 661      * to synchronous IO.
 662      */
 663     BDRVRawState *s = bs->opaque;
 664
 665     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 666         QEMUBH *bh;
 667         acb = qemu_aio_get(bs, cb, opaque);
 668         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 669         bh = qemu_bh_new(raw_aio_em_cb, acb);
 670         qemu_bh_schedule(bh);
 671         return &acb->common;
 672     }
 673
 674     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 675     if (!acb)
 676         return NULL;
 677     if (qemu_paio_write(&acb->aiocb) < 0) {
 678         raw_aio_remove(acb);
 679         return NULL;
 680     }
 681     return &acb->common;
 682 }
 683
 684 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 685 {
 686     int ret;
 687     RawAIOCB *acb = (RawAIOCB *)blockacb;
 688
 689     ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 690     if (ret == QEMU_PAIO_NOTCANCELED) {
 691         /* fail safe: if the aio could not be canceled, we wait for
 692            it */
 693         while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
 694     }
 695
 696     raw_aio_remove(acb);
 697 }
 698 #else /* CONFIG_AIO */
 699 static int posix_aio_init(void)
 700 {
 701     return 0;
 702 }
 703 #endif /* CONFIG_AIO */
 704
 705
 706 static void raw_close(BlockDriverState *bs)
 707 {
 708     BDRVRawState *s = bs->opaque;
 709     if (s->fd >= 0) {
 710         close(s->fd);
 711         s->fd = -1;
 712         if (s->aligned_buf != NULL)
 713             qemu_free(s->aligned_buf);
 714     }
 715 }
 716
 717 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 718 {
 719     BDRVRawState *s = bs->opaque;
 720     if (s->type != FTYPE_FILE)
 721         return -ENOTSUP;
 722     if (ftruncate(s->fd, offset) < 0)
 723         return -errno;
 724     return 0;
 725 }
 726
 727 #ifdef __OpenBSD__
 728 static int64_t raw_getlength(BlockDriverState *bs)
 729 {
 730     BDRVRawState *s = bs->opaque;
 731     int fd = s->fd;
 732     struct stat st;
 733
 734     if (fstat(fd, &st))
 735         return -1;
 736     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 737         struct disklabel dl;
 738
 739         if (ioctl(fd, DIOCGDINFO, &dl))
 740             return -1;
 741         return (uint64_t)dl.d_secsize *
 742             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 743     } else
 744         return st.st_size;
 745 }
 746 #else /* !__OpenBSD__ */
 747 static int64_t  raw_getlength(BlockDriverState *bs)
 748 {
 749     BDRVRawState *s = bs->opaque;
 750     int fd = s->fd;
 751     int64_t size;
 752 #ifdef _BSD
 753     struct stat sb;
 754 #endif
 755 #ifdef __sun__
 756     struct dk_minfo minfo;
 757     int rv;
 758 #endif
 759     int ret;
 760
 761     ret = fd_open(bs);
 762     if (ret < 0)
 763         return ret;
 764
 765 #ifdef _BSD
 766     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 767 #ifdef DIOCGMEDIASIZE
 768         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 769 #endif
 770 #ifdef CONFIG_COCOA
 771         size = LONG_LONG_MAX;
 772 #else
 773         size = lseek(fd, 0LL, SEEK_END);
 774 #endif
 775     } else
 776 #endif
 777 #ifdef __sun__
 778     /*
 779      * use the DKIOCGMEDIAINFO ioctl to read the size.
 780      */
 781     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 782     if ( rv != -1 ) {
 783         size = minfo.dki_lbsize * minfo.dki_capacity;
 784     } else /* there are reports that lseek on some devices
 785               fails, but irc discussion said that contingency
 786               on contingency was overkill */
 787 #endif
 788     {
 789         size = lseek(fd, 0, SEEK_END);
 790     }
 791     return size;
 792 }
 793 #endif
 794
 795 static int raw_create(const char *filename, int64_t total_size,
 796                       const char *backing_file, int flags)
 797 {
 798     int fd;
 799
 800     if (flags || backing_file)
 801         return -ENOTSUP;
 802
 803     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 804               0644);
 805     if (fd < 0)
 806         return -EIO;
 807     ftruncate(fd, total_size * 512);
 808     close(fd);
 809     return 0;
 810 }
 811
 812 static void raw_flush(BlockDriverState *bs)
 813 {
 814     BDRVRawState *s = bs->opaque;
 815     fsync(s->fd);
 816 }
 817
 818 BlockDriver bdrv_raw = {
 819     "raw",
 820     sizeof(BDRVRawState),
 821     NULL, /* no probe for protocols */
 822     raw_open,
 823     NULL,
 824     NULL,
 825     raw_close,
 826     raw_create,
 827     raw_flush,
 828
 829 #ifdef CONFIG_AIO
 830     .bdrv_aio_read = raw_aio_read,
 831     .bdrv_aio_write = raw_aio_write,
 832     .bdrv_aio_cancel = raw_aio_cancel,
 833     .aiocb_size = sizeof(RawAIOCB),
 834 #endif
 835
 836     .bdrv_pread = raw_pread,
 837     .bdrv_pwrite = raw_pwrite,
 838     .bdrv_truncate = raw_truncate,
 839     .bdrv_getlength = raw_getlength,
 840 };
 841
 842 /***********************************************/
 843 /* host device */
 844
 845 #ifdef CONFIG_COCOA
 846 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 847 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 848
 849 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 850 {
 851     kern_return_t       kernResult;
 852     mach_port_t     masterPort;
 853     CFMutableDictionaryRef  classesToMatch;
 854
 855     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 856     if ( KERN_SUCCESS != kernResult ) {
 857         printf( "IOMasterPort returned %d\n", kernResult );
 858     }
 859
 860     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 861     if ( classesToMatch == NULL ) {
 862         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 863     } else {
 864     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 865     }
 866     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 867     if ( KERN_SUCCESS != kernResult )
 868     {
 869         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 870     }
 871
 872     return kernResult;
 873 }
 874
 875 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 876 {
 877     io_object_t     nextMedia;
 878     kern_return_t   kernResult = KERN_FAILURE;
 879     *bsdPath = '\0';
 880     nextMedia = IOIteratorNext( mediaIterator );
 881     if ( nextMedia )
 882     {
 883         CFTypeRef   bsdPathAsCFString;
 884     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 885         if ( bsdPathAsCFString ) {
 886             size_t devPathLength;
 887             strcpy( bsdPath, _PATH_DEV );
 888             strcat( bsdPath, "r" );
 889             devPathLength = strlen( bsdPath );
 890             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 891                 kernResult = KERN_SUCCESS;
 892             }
 893             CFRelease( bsdPathAsCFString );
 894         }
 895         IOObjectRelease( nextMedia );
 896     }
 897
 898     return kernResult;
 899 }
 900
 901 #endif
 902
 903 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 904 {
 905     BDRVRawState *s = bs->opaque;
 906     int fd, open_flags, ret;
 907
 908     posix_aio_init();
 909
 910 #ifdef CONFIG_COCOA
 911     if (strstart(filename, "/dev/cdrom", NULL)) {
 912         kern_return_t kernResult;
 913         io_iterator_t mediaIterator;
 914         char bsdPath[ MAXPATHLEN ];
 915         int fd;
 916
 917         kernResult = FindEjectableCDMedia( &mediaIterator );
 918         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 919
 920         if ( bsdPath[ 0 ] != '\0' ) {
 921             strcat(bsdPath,"s0");
 922             /* some CDs don't have a partition 0 */
 923             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 924             if (fd < 0) {
 925                 bsdPath[strlen(bsdPath)-1] = '1';
 926             } else {
 927                 close(fd);
 928             }
 929             filename = bsdPath;
 930         }
 931
 932         if ( mediaIterator )
 933             IOObjectRelease( mediaIterator );
 934     }
 935 #endif
 936     open_flags = O_BINARY;
 937     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 938         open_flags |= O_RDWR;
 939     } else {
 940         open_flags |= O_RDONLY;
 941         bs->read_only = 1;
 942     }
 943     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 944      * and O_DIRECT for no caching. */
 945     if ((flags & BDRV_O_NOCACHE))
 946         open_flags |= O_DIRECT;
 947     else if (!(flags & BDRV_O_CACHE_WB))
 948         open_flags |= O_DSYNC;
 949
 950     s->type = FTYPE_FILE;
 951 #if defined(__linux__)
 952     if (strstart(filename, "/dev/cd", NULL)) {
 953         /* open will not fail even if no CD is inserted */
 954         open_flags |= O_NONBLOCK;
 955         s->type = FTYPE_CD;
 956     } else if (strstart(filename, "/dev/fd", NULL)) {
 957         s->type = FTYPE_FD;
 958         s->fd_open_flags = open_flags;
 959         /* open will not fail even if no floppy is inserted */
 960         open_flags |= O_NONBLOCK;
 961     } else if (strstart(filename, "/dev/sg", NULL)) {
 962         bs->sg = 1;
 963     }
 964 #endif
 965     fd = open(filename, open_flags, 0644);
 966     if (fd < 0) {
 967         ret = -errno;
 968         if (ret == -EROFS)
 969             ret = -EACCES;
 970         return ret;
 971     }
 972     s->fd = fd;
 973 #if defined(__linux__)
 974     /* close fd so that we can reopen it as needed */
 975     if (s->type == FTYPE_FD) {
 976         close(s->fd);
 977         s->fd = -1;
 978         s->fd_media_changed = 1;
 979     }
 980 #endif
 981     return 0;
 982 }
 983
 984 #if defined(__linux__)
 985 /* Note: we do not have a reliable method to detect if the floppy is
 986    present. The current method is to try to open the floppy at every
 987    I/O and to keep it opened during a few hundreds of ms. */
 988 static int fd_open(BlockDriverState *bs)
 989 {
 990     BDRVRawState *s = bs->opaque;
 991     int last_media_present;
 992
 993     if (s->type != FTYPE_FD)
 994         return 0;
 995     last_media_present = (s->fd >= 0);
 996     if (s->fd >= 0 &&
 997         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
 998         close(s->fd);
 999         s->fd = -1;
1000 #ifdef DEBUG_FLOPPY
1001         printf("Floppy closed\n");
1002 #endif
1003     }
1004     if (s->fd < 0) {
1005         if (s->fd_got_error &&
1006             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1007 #ifdef DEBUG_FLOPPY
1008             printf("No floppy (open delayed)\n");
1009 #endif
1010             return -EIO;
1011         }
1012         s->fd = open(bs->filename, s->fd_open_flags);
1013         if (s->fd < 0) {
1014             s->fd_error_time = qemu_get_clock(rt_clock);
1015             s->fd_got_error = 1;
1016             if (last_media_present)
1017                 s->fd_media_changed = 1;
1018 #ifdef DEBUG_FLOPPY
1019             printf("No floppy\n");
1020 #endif
1021             return -EIO;
1022         }
1023 #ifdef DEBUG_FLOPPY
1024         printf("Floppy opened\n");
1025 #endif
1026     }
1027     if (!last_media_present)
1028         s->fd_media_changed = 1;
1029     s->fd_open_time = qemu_get_clock(rt_clock);
1030     s->fd_got_error = 0;
1031     return 0;
1032 }
1033
1034 static int raw_is_inserted(BlockDriverState *bs)
1035 {
1036     BDRVRawState *s = bs->opaque;
1037     int ret;
1038
1039     switch(s->type) {
1040     case FTYPE_CD:
1041         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1042         if (ret == CDS_DISC_OK)
1043             return 1;
1044         else
1045             return 0;
1046         break;
1047     case FTYPE_FD:
1048         ret = fd_open(bs);
1049         return (ret >= 0);
1050     default:
1051         return 1;
1052     }
1053 }
1054
1055 /* currently only used by fdc.c, but a CD version would be good too */
1056 static int raw_media_changed(BlockDriverState *bs)
1057 {
1058     BDRVRawState *s = bs->opaque;
1059
1060     switch(s->type) {
1061     case FTYPE_FD:
1062         {
1063             int ret;
1064             /* XXX: we do not have a true media changed indication. It
1065                does not work if the floppy is changed without trying
1066                to read it */
1067             fd_open(bs);
1068             ret = s->fd_media_changed;
1069             s->fd_media_changed = 0;
1070 #ifdef DEBUG_FLOPPY
1071             printf("Floppy changed=%d\n", ret);
1072 #endif
1073             return ret;
1074         }
1075     default:
1076         return -ENOTSUP;
1077     }
1078 }
1079
1080 static int raw_eject(BlockDriverState *bs, int eject_flag)
1081 {
1082     BDRVRawState *s = bs->opaque;
1083
1084     switch(s->type) {
1085     case FTYPE_CD:
1086         if (eject_flag) {
1087             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1088                 perror("CDROMEJECT");
1089         } else {
1090             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1091                 perror("CDROMEJECT");
1092         }
1093         break;
1094     case FTYPE_FD:
1095         {
1096             int fd;
1097             if (s->fd >= 0) {
1098                 close(s->fd);
1099                 s->fd = -1;
1100             }
1101             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1102             if (fd >= 0) {
1103                 if (ioctl(fd, FDEJECT, 0) < 0)
1104                     perror("FDEJECT");
1105                 close(fd);
1106             }
1107         }
1108         break;
1109     default:
1110         return -ENOTSUP;
1111     }
1112     return 0;
1113 }
1114
1115 static int raw_set_locked(BlockDriverState *bs, int locked)
1116 {
1117     BDRVRawState *s = bs->opaque;
1118
1119     switch(s->type) {
1120     case FTYPE_CD:
1121         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1122             /* Note: an error can happen if the distribution automatically
1123                mounts the CD-ROM */
1124             //        perror("CDROM_LOCKDOOR");
1125         }
1126         break;
1127     default:
1128         return -ENOTSUP;
1129     }
1130     return 0;
1131 }
1132
1133 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1134 {
1135     BDRVRawState *s = bs->opaque;
1136
1137     return ioctl(s->fd, req, buf);
1138 }
1139 #else
1140
1141 static int fd_open(BlockDriverState *bs)
1142 {
1143     return 0;
1144 }
1145
1146 static int raw_is_inserted(BlockDriverState *bs)
1147 {
1148     return 1;
1149 }
1150
1151 static int raw_media_changed(BlockDriverState *bs)
1152 {
1153     return -ENOTSUP;
1154 }
1155
1156 static int raw_eject(BlockDriverState *bs, int eject_flag)
1157 {
1158     return -ENOTSUP;
1159 }
1160
1161 static int raw_set_locked(BlockDriverState *bs, int locked)
1162 {
1163     return -ENOTSUP;
1164 }
1165
1166 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1167 {
1168     return -ENOTSUP;
1169 }
1170 #endif /* !linux */
1171
1172 BlockDriver bdrv_host_device = {
1173     "host_device",
1174     sizeof(BDRVRawState),
1175     NULL, /* no probe for protocols */
1176     hdev_open,
1177     NULL,
1178     NULL,
1179     raw_close,
1180     NULL,
1181     raw_flush,
1182
1183 #ifdef CONFIG_AIO
1184     .bdrv_aio_read = raw_aio_read,
1185     .bdrv_aio_write = raw_aio_write,
1186     .bdrv_aio_cancel = raw_aio_cancel,
1187     .aiocb_size = sizeof(RawAIOCB),
1188 #endif
1189
1190     .bdrv_pread = raw_pread,
1191     .bdrv_pwrite = raw_pwrite,
1192     .bdrv_getlength = raw_getlength,
1193
1194     /* removable device support */
1195     .bdrv_is_inserted = raw_is_inserted,
1196     .bdrv_media_changed = raw_media_changed,
1197     .bdrv_eject = raw_eject,
1198     .bdrv_set_locked = raw_set_locked,
1199     /* generic scsi device */
1200     .bdrv_ioctl = raw_ioctl,
1201 };