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 "compatfd.h"
  29 #include <assert.h>
  30 #ifdef CONFIG_AIO
  31 #include <aio.h>
  32 #endif
  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 <signal.h>
  58 #include <sys/disk.h>
  59 #endif
  60
  61 #ifdef __OpenBSD__
  62 #include <sys/ioctl.h>
  63 #include <sys/disklabel.h>
  64 #include <sys/dkio.h>
  65 #endif
  66
  67 //#define DEBUG_FLOPPY
  68
  69 //#define DEBUG_BLOCK
  70 #if defined(DEBUG_BLOCK)
  71 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (loglevel != 0)  \
  72     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } while (0)
  73 #else
  74 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
  75 #endif
  76
  77 #define FTYPE_FILE   0
  78 #define FTYPE_CD     1
  79 #define FTYPE_FD     2
  80
  81 #define ALIGNED_BUFFER_SIZE (32 * 512)
  82
  83 /* if the FD is not accessed during that time (in ms), we try to
  84    reopen it to see if the disk has been changed */
  85 #define FD_OPEN_TIMEOUT 1000
  86
  87 /* posix-aio doesn't allow multiple outstanding requests to a single file
  88  * descriptor.  we implement a pool of dup()'d file descriptors to work
  89  * around this */
  90 #define RAW_FD_POOL_SIZE        64
  91
  92 typedef struct BDRVRawState {
  93     int fd;
  94     int type;
  95     unsigned int lseek_err_cnt;
  96     int fd_pool[RAW_FD_POOL_SIZE];
  97 #if defined(__linux__)
  98     /* linux floppy specific */
  99     int fd_open_flags;
 100     int64_t fd_open_time;
 101     int64_t fd_error_time;
 102     int fd_got_error;
 103     int fd_media_changed;
 104 #endif
 105 #if defined(O_DIRECT)
 106     uint8_t* aligned_buf;
 107 #endif
 108 } BDRVRawState;
 109
 110 static int posix_aio_init(void);
 111
 112 static int fd_open(BlockDriverState *bs);
 113
 114 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 115 {
 116     BDRVRawState *s = bs->opaque;
 117     int fd, open_flags, ret;
 118     int i;
 119
 120     posix_aio_init();
 121
 122     s->lseek_err_cnt = 0;
 123
 124     open_flags = O_BINARY;
 125     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 126         open_flags |= O_RDWR;
 127     } else {
 128         open_flags |= O_RDONLY;
 129         bs->read_only = 1;
 130     }
 131     if (flags & BDRV_O_CREAT)
 132         open_flags |= O_CREAT | O_TRUNC;
 133 #ifdef O_DIRECT
 134     if (flags & BDRV_O_DIRECT)
 135         open_flags |= O_DIRECT;
 136 #endif
 137
 138     s->type = FTYPE_FILE;
 139
 140     fd = open(filename, open_flags, 0644);
 141     if (fd < 0) {
 142         ret = -errno;
 143         if (ret == -EROFS)
 144             ret = -EACCES;
 145         return ret;
 146     }
 147     s->fd = fd;
 148     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
 149         s->fd_pool[i] = -1;
 150 #if defined(O_DIRECT)
 151     s->aligned_buf = NULL;
 152     if (flags & BDRV_O_DIRECT) {
 153         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 154         if (s->aligned_buf == NULL) {
 155             ret = -errno;
 156             close(fd);
 157             return ret;
 158         }
 159     }
 160 #endif
 161     return 0;
 162 }
 163
 164 /* XXX: use host sector size if necessary with:
 165 #ifdef DIOCGSECTORSIZE
 166         {
 167             unsigned int sectorsize = 512;
 168             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 169                 sectorsize > bufsize)
 170                 bufsize = sectorsize;
 171         }
 172 #endif
 173 #ifdef CONFIG_COCOA
 174         u_int32_t   blockSize = 512;
 175         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 176             bufsize = blockSize;
 177         }
 178 #endif
 179 */
 180
 181 /*
 182  * offset and count are in bytes, but must be multiples of 512 for files
 183  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 184  *
 185  * This function may be called without alignment if the caller ensures
 186  * that O_DIRECT is not in effect.
 187  */
 188 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 189                      uint8_t *buf, int count)
 190 {
 191     BDRVRawState *s = bs->opaque;
 192     int ret;
 193
 194     ret = fd_open(bs);
 195     if (ret < 0)
 196         return ret;
 197
 198     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 199         ++(s->lseek_err_cnt);
 200         if(s->lseek_err_cnt <= 10) {
 201             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 202                               "] lseek failed : %d = %s\n",
 203                               s->fd, bs->filename, offset, buf, count,
 204                               bs->total_sectors, errno, strerror(errno));
 205         }
 206         return -1;
 207     }
 208     s->lseek_err_cnt=0;
 209
 210     ret = read(s->fd, buf, count);
 211     if (ret == count)
 212         goto label__raw_read__success;
 213
 214     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 215                       "] read failed %d : %d = %s\n",
 216                       s->fd, bs->filename, offset, buf, count,
 217                       bs->total_sectors, ret, errno, strerror(errno));
 218
 219     /* Try harder for CDrom. */
 220     if (bs->type == BDRV_TYPE_CDROM) {
 221         lseek(s->fd, offset, SEEK_SET);
 222         ret = read(s->fd, buf, count);
 223         if (ret == count)
 224             goto label__raw_read__success;
 225         lseek(s->fd, offset, SEEK_SET);
 226         ret = read(s->fd, buf, count);
 227         if (ret == count)
 228             goto label__raw_read__success;
 229
 230         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 231                           "] retry read failed %d : %d = %s\n",
 232                           s->fd, bs->filename, offset, buf, count,
 233                           bs->total_sectors, ret, errno, strerror(errno));
 234     }
 235
 236 label__raw_read__success:
 237
 238     return ret;
 239 }
 240
 241 /*
 242  * offset and count are in bytes, but must be multiples of 512 for files
 243  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 244  *
 245  * This function may be called without alignment if the caller ensures
 246  * that O_DIRECT is not in effect.
 247  */
 248 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 249                       const uint8_t *buf, int count)
 250 {
 251     BDRVRawState *s = bs->opaque;
 252     int ret;
 253
 254     ret = fd_open(bs);
 255     if (ret < 0)
 256         return ret;
 257
 258     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 259         ++(s->lseek_err_cnt);
 260         if(s->lseek_err_cnt) {
 261             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 262                               PRId64 "] lseek failed : %d = %s\n",
 263                               s->fd, bs->filename, offset, buf, count,
 264                               bs->total_sectors, errno, strerror(errno));
 265         }
 266         return -1;
 267     }
 268     s->lseek_err_cnt = 0;
 269
 270     ret = write(s->fd, buf, count);
 271     if (ret == count)
 272         goto label__raw_write__success;
 273
 274     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 275                       "] write failed %d : %d = %s\n",
 276                       s->fd, bs->filename, offset, buf, count,
 277                       bs->total_sectors, ret, errno, strerror(errno));
 278
 279 label__raw_write__success:
 280
 281     return ret;
 282 }
 283
 284
 285 #if defined(O_DIRECT)
 286 /*
 287  * offset and count are in bytes and possibly not aligned. For files opened
 288  * with O_DIRECT, necessary alignments are ensured before calling
 289  * raw_pread_aligned to do the actual read.
 290  */
 291 static int raw_pread(BlockDriverState *bs, int64_t offset,
 292                      uint8_t *buf, int count)
 293 {
 294     BDRVRawState *s = bs->opaque;
 295     int size, ret, shift, sum;
 296
 297     sum = 0;
 298
 299     if (s->aligned_buf != NULL)  {
 300
 301         if (offset & 0x1ff) {
 302             /* align offset on a 512 bytes boundary */
 303
 304             shift = offset & 0x1ff;
 305             size = (shift + count + 0x1ff) & ~0x1ff;
 306             if (size > ALIGNED_BUFFER_SIZE)
 307                 size = ALIGNED_BUFFER_SIZE;
 308             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 309             if (ret < 0)
 310                 return ret;
 311
 312             size = 512 - shift;
 313             if (size > count)
 314                 size = count;
 315             memcpy(buf, s->aligned_buf + shift, size);
 316
 317             buf += size;
 318             offset += size;
 319             count -= size;
 320             sum += size;
 321
 322             if (count == 0)
 323                 return sum;
 324         }
 325         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 326
 327             /* read on aligned buffer */
 328
 329             while (count) {
 330
 331                 size = (count + 0x1ff) & ~0x1ff;
 332                 if (size > ALIGNED_BUFFER_SIZE)
 333                     size = ALIGNED_BUFFER_SIZE;
 334
 335                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 336                 if (ret < 0)
 337                     return ret;
 338
 339                 size = ret;
 340                 if (size > count)
 341                     size = count;
 342
 343                 memcpy(buf, s->aligned_buf, size);
 344
 345                 buf += size;
 346                 offset += size;
 347                 count -= size;
 348                 sum += size;
 349             }
 350
 351             return sum;
 352         }
 353     }
 354
 355     return raw_pread_aligned(bs, offset, buf, count) + sum;
 356 }
 357
 358 /*
 359  * offset and count are in bytes and possibly not aligned. For files opened
 360  * with O_DIRECT, necessary alignments are ensured before calling
 361  * raw_pwrite_aligned to do the actual write.
 362  */
 363 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 364                       const uint8_t *buf, int count)
 365 {
 366     BDRVRawState *s = bs->opaque;
 367     int size, ret, shift, sum;
 368
 369     sum = 0;
 370
 371     if (s->aligned_buf != NULL) {
 372
 373         if (offset & 0x1ff) {
 374             /* align offset on a 512 bytes boundary */
 375             shift = offset & 0x1ff;
 376             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 377             if (ret < 0)
 378                 return ret;
 379
 380             size = 512 - shift;
 381             if (size > count)
 382                 size = count;
 383             memcpy(s->aligned_buf + shift, buf, size);
 384
 385             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 386             if (ret < 0)
 387                 return ret;
 388
 389             buf += size;
 390             offset += size;
 391             count -= size;
 392             sum += size;
 393
 394             if (count == 0)
 395                 return sum;
 396         }
 397         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 398
 399             while ((size = (count & ~0x1ff)) != 0) {
 400
 401                 if (size > ALIGNED_BUFFER_SIZE)
 402                     size = ALIGNED_BUFFER_SIZE;
 403
 404                 memcpy(s->aligned_buf, buf, size);
 405
 406                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 407                 if (ret < 0)
 408                     return ret;
 409
 410                 buf += ret;
 411                 offset += ret;
 412                 count -= ret;
 413                 sum += ret;
 414             }
 415             /* here, count < 512 because (count & ~0x1ff) == 0 */
 416             if (count) {
 417                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 418                 if (ret < 0)
 419                     return ret;
 420                  memcpy(s->aligned_buf, buf, count);
 421
 422                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 423                  if (ret < 0)
 424                      return ret;
 425                  if (count < ret)
 426                      ret = count;
 427
 428                  sum += ret;
 429             }
 430             return sum;
 431         }
 432     }
 433     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 434 }
 435
 436 #else
 437 #define raw_pread raw_pread_aligned
 438 #define raw_pwrite raw_pwrite_aligned
 439 #endif
 440
 441
 442 #ifdef CONFIG_AIO
 443 /***********************************************************/
 444 /* Unix AIO using POSIX AIO */
 445
 446 typedef struct RawAIOCB {
 447     BlockDriverAIOCB common;
 448     int fd;
 449     struct aiocb aiocb;
 450     struct RawAIOCB *next;
 451     int ret;
 452 } RawAIOCB;
 453
 454 typedef struct PosixAioState
 455 {
 456     int fd;
 457     RawAIOCB *first_aio;
 458 } PosixAioState;
 459
 460 static int raw_fd_pool_get(BDRVRawState *s)
 461 {
 462     int i;
 463
 464     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 465         /* already in use */
 466         if (s->fd_pool[i] != -1)
 467             continue;
 468
 469         /* try to dup file descriptor */
 470         s->fd_pool[i] = dup(s->fd);
 471         if (s->fd_pool[i] != -1)
 472             return s->fd_pool[i];
 473     }
 474
 475     /* we couldn't dup the file descriptor so just use the main one */
 476     return s->fd;
 477 }
 478
 479 static void raw_fd_pool_put(RawAIOCB *acb)
 480 {
 481     BDRVRawState *s = acb->common.bs->opaque;
 482     int i;
 483
 484     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 485         if (s->fd_pool[i] == acb->fd) {
 486             close(s->fd_pool[i]);
 487             s->fd_pool[i] = -1;
 488         }
 489     }
 490 }
 491
 492 static void posix_aio_read(void *opaque)
 493 {
 494     PosixAioState *s = opaque;
 495     RawAIOCB *acb, **pacb;
 496     int ret;
 497     size_t offset;
 498     union {
 499         struct qemu_signalfd_siginfo siginfo;
 500         char buf[128];
 501     } sig;
 502
 503     /* try to read from signalfd, don't freak out if we can't read anything */
 504     offset = 0;
 505     while (offset < 128) {
 506         ssize_t len;
 507
 508         len = read(s->fd, sig.buf + offset, 128 - offset);
 509         if (len == -1 && errno == EINTR)
 510             continue;
 511         if (len == -1 && errno == EAGAIN) {
 512             /* there is no natural reason for this to happen,
 513              * so we'll spin hard until we get everything just
 514              * to be on the safe side. */
 515             if (offset > 0)
 516                 continue;
 517         }
 518
 519         offset += len;
 520     }
 521
 522     for(;;) {
 523         pacb = &s->first_aio;
 524         for(;;) {
 525             acb = *pacb;
 526             if (!acb)
 527                 goto the_end;
 528             ret = aio_error(&acb->aiocb);
 529             if (ret == ECANCELED) {
 530                 /* remove the request */
 531                 *pacb = acb->next;
 532                 raw_fd_pool_put(acb);
 533                 qemu_aio_release(acb);
 534             } else if (ret != EINPROGRESS) {
 535                 /* end of aio */
 536                 if (ret == 0) {
 537                     ret = aio_return(&acb->aiocb);
 538                     if (ret == acb->aiocb.aio_nbytes)
 539                         ret = 0;
 540                     else
 541                         ret = -EINVAL;
 542                 } else {
 543                     ret = -ret;
 544                 }
 545                 /* remove the request */
 546                 *pacb = acb->next;
 547                 /* call the callback */
 548                 acb->common.cb(acb->common.opaque, ret);
 549                 raw_fd_pool_put(acb);
 550                 qemu_aio_release(acb);
 551                 break;
 552             } else {
 553                 pacb = &acb->next;
 554             }
 555         }
 556     }
 557  the_end: ;
 558 }
 559
 560 static int posix_aio_flush(void *opaque)
 561 {
 562     PosixAioState *s = opaque;
 563     return !!s->first_aio;
 564 }
 565
 566 static PosixAioState *posix_aio_state;
 567
 568 static int posix_aio_init(void)
 569 {
 570     sigset_t mask;
 571     PosixAioState *s;
 572
 573     if (posix_aio_state)
 574         return 0;
 575
 576     s = qemu_malloc(sizeof(PosixAioState));
 577     if (s == NULL)
 578         return -ENOMEM;
 579
 580     /* Make sure to block AIO signal */
 581     sigemptyset(&mask);
 582     sigaddset(&mask, SIGUSR2);
 583     sigprocmask(SIG_BLOCK, &mask, NULL);
 584
 585     s->first_aio = NULL;
 586     s->fd = qemu_signalfd(&mask);
 587
 588     fcntl(s->fd, F_SETFL, O_NONBLOCK);
 589
 590     qemu_aio_set_fd_handler(s->fd, posix_aio_read, NULL, posix_aio_flush, s);
 591
 592 #if defined(__linux__)
 593     {
 594         struct aioinit ai;
 595
 596         memset(&ai, 0, sizeof(ai));
 597 #if defined(__GLIBC_PREREQ) && __GLIBC_PREREQ(2, 4)
 598         ai.aio_threads = 64;
 599         ai.aio_num = 64;
 600 #else
 601         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 602            seems to fix the problem. */
 603         ai.aio_threads = 1;
 604         ai.aio_num = 1;
 605         ai.aio_idle_time = 365 * 100000;
 606 #endif
 607         aio_init(&ai);
 608     }
 609 #endif
 610     posix_aio_state = s;
 611
 612     return 0;
 613 }
 614
 615 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 616         int64_t sector_num, uint8_t *buf, int nb_sectors,
 617         BlockDriverCompletionFunc *cb, void *opaque)
 618 {
 619     BDRVRawState *s = bs->opaque;
 620     RawAIOCB *acb;
 621
 622     if (fd_open(bs) < 0)
 623         return NULL;
 624
 625     acb = qemu_aio_get(bs, cb, opaque);
 626     if (!acb)
 627         return NULL;
 628     acb->fd = raw_fd_pool_get(s);
 629     acb->aiocb.aio_fildes = acb->fd;
 630     acb->aiocb.aio_sigevent.sigev_signo = SIGUSR2;
 631     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 632     acb->aiocb.aio_buf = buf;
 633     if (nb_sectors < 0)
 634         acb->aiocb.aio_nbytes = -nb_sectors;
 635     else
 636         acb->aiocb.aio_nbytes = nb_sectors * 512;
 637     acb->aiocb.aio_offset = sector_num * 512;
 638     acb->next = posix_aio_state->first_aio;
 639     posix_aio_state->first_aio = acb;
 640     return acb;
 641 }
 642
 643 static void raw_aio_em_cb(void* opaque)
 644 {
 645     RawAIOCB *acb = opaque;
 646     acb->common.cb(acb->common.opaque, acb->ret);
 647     qemu_aio_release(acb);
 648 }
 649
 650 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 651         int64_t sector_num, 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)
 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_pread(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, buf, nb_sectors, cb, opaque);
 674     if (!acb)
 675         return NULL;
 676     if (aio_read(&acb->aiocb) < 0) {
 677         qemu_aio_release(acb);
 678         return NULL;
 679     }
 680     return &acb->common;
 681 }
 682
 683 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 684         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 685         BlockDriverCompletionFunc *cb, void *opaque)
 686 {
 687     RawAIOCB *acb;
 688
 689     /*
 690      * If O_DIRECT is used and the buffer is not aligned fall back
 691      * to synchronous IO.
 692      */
 693 #if defined(O_DIRECT)
 694     BDRVRawState *s = bs->opaque;
 695
 696     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 697         QEMUBH *bh;
 698         acb = qemu_aio_get(bs, cb, opaque);
 699         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 700         bh = qemu_bh_new(raw_aio_em_cb, acb);
 701         qemu_bh_schedule(bh);
 702         return &acb->common;
 703     }
 704 #endif
 705
 706     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 707     if (!acb)
 708         return NULL;
 709     if (aio_write(&acb->aiocb) < 0) {
 710         qemu_aio_release(acb);
 711         return NULL;
 712     }
 713     return &acb->common;
 714 }
 715
 716 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 717 {
 718     int ret;
 719     RawAIOCB *acb = (RawAIOCB *)blockacb;
 720     RawAIOCB **pacb;
 721
 722     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 723     if (ret == AIO_NOTCANCELED) {
 724         /* fail safe: if the aio could not be canceled, we wait for
 725            it */
 726         while (aio_error(&acb->aiocb) == EINPROGRESS);
 727     }
 728
 729     /* remove the callback from the queue */
 730     pacb = &posix_aio_state->first_aio;
 731     for(;;) {
 732         if (*pacb == NULL) {
 733             break;
 734         } else if (*pacb == acb) {
 735             *pacb = acb->next;
 736             raw_fd_pool_put(acb);
 737             qemu_aio_release(acb);
 738             break;
 739         }
 740         pacb = &acb->next;
 741     }
 742 }
 743
 744 #else /* CONFIG_AIO */
 745 static int posix_aio_init(void)
 746 {
 747 }
 748 #endif /* CONFIG_AIO */
 749
 750 static void raw_close_fd_pool(BDRVRawState *s)
 751 {
 752     int i;
 753
 754     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 755         if (s->fd_pool[i] != -1) {
 756             close(s->fd_pool[i]);
 757             s->fd_pool[i] = -1;
 758         }
 759     }
 760 }
 761
 762 static void raw_close(BlockDriverState *bs)
 763 {
 764     BDRVRawState *s = bs->opaque;
 765     if (s->fd >= 0) {
 766         close(s->fd);
 767         s->fd = -1;
 768 #if defined(O_DIRECT)
 769         if (s->aligned_buf != NULL)
 770             qemu_free(s->aligned_buf);
 771 #endif
 772     }
 773     raw_close_fd_pool(s);
 774 }
 775
 776 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 777 {
 778     BDRVRawState *s = bs->opaque;
 779     if (s->type != FTYPE_FILE)
 780         return -ENOTSUP;
 781     if (ftruncate(s->fd, offset) < 0)
 782         return -errno;
 783     return 0;
 784 }
 785
 786 #ifdef __OpenBSD__
 787 static int64_t raw_getlength(BlockDriverState *bs)
 788 {
 789     BDRVRawState *s = bs->opaque;
 790     int fd = s->fd;
 791     struct stat st;
 792
 793     if (fstat(fd, &st))
 794         return -1;
 795     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 796         struct disklabel dl;
 797
 798         if (ioctl(fd, DIOCGDINFO, &dl))
 799             return -1;
 800         return (uint64_t)dl.d_secsize *
 801             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 802     } else
 803         return st.st_size;
 804 }
 805 #else /* !__OpenBSD__ */
 806 static int64_t  raw_getlength(BlockDriverState *bs)
 807 {
 808     BDRVRawState *s = bs->opaque;
 809     int fd = s->fd;
 810     int64_t size;
 811 #ifdef _BSD
 812     struct stat sb;
 813 #endif
 814 #ifdef __sun__
 815     struct dk_minfo minfo;
 816     int rv;
 817 #endif
 818     int ret;
 819
 820     ret = fd_open(bs);
 821     if (ret < 0)
 822         return ret;
 823
 824 #ifdef _BSD
 825     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 826 #ifdef DIOCGMEDIASIZE
 827         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 828 #endif
 829 #ifdef CONFIG_COCOA
 830         size = LONG_LONG_MAX;
 831 #else
 832         size = lseek(fd, 0LL, SEEK_END);
 833 #endif
 834     } else
 835 #endif
 836 #ifdef __sun__
 837     /*
 838      * use the DKIOCGMEDIAINFO ioctl to read the size.
 839      */
 840     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 841     if ( rv != -1 ) {
 842         size = minfo.dki_lbsize * minfo.dki_capacity;
 843     } else /* there are reports that lseek on some devices
 844               fails, but irc discussion said that contingency
 845               on contingency was overkill */
 846 #endif
 847     {
 848         size = lseek(fd, 0, SEEK_END);
 849     }
 850     return size;
 851 }
 852 #endif
 853
 854 static int raw_create(const char *filename, int64_t total_size,
 855                       const char *backing_file, int flags)
 856 {
 857     int fd;
 858
 859     if (flags || backing_file)
 860         return -ENOTSUP;
 861
 862     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 863               0644);
 864     if (fd < 0)
 865         return -EIO;
 866     ftruncate(fd, total_size * 512);
 867     close(fd);
 868     return 0;
 869 }
 870
 871 static void raw_flush(BlockDriverState *bs)
 872 {
 873     BDRVRawState *s = bs->opaque;
 874     fsync(s->fd);
 875 }
 876
 877 BlockDriver bdrv_raw = {
 878     "raw",
 879     sizeof(BDRVRawState),
 880     NULL, /* no probe for protocols */
 881     raw_open,
 882     NULL,
 883     NULL,
 884     raw_close,
 885     raw_create,
 886     raw_flush,
 887
 888 #ifdef CONFIG_AIO
 889     .bdrv_aio_read = raw_aio_read,
 890     .bdrv_aio_write = raw_aio_write,
 891     .bdrv_aio_cancel = raw_aio_cancel,
 892     .aiocb_size = sizeof(RawAIOCB),
 893 #endif
 894     .bdrv_pread = raw_pread,
 895     .bdrv_pwrite = raw_pwrite,
 896     .bdrv_truncate = raw_truncate,
 897     .bdrv_getlength = raw_getlength,
 898 };
 899
 900 /***********************************************/
 901 /* host device */
 902
 903 #ifdef CONFIG_COCOA
 904 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 905 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 906
 907 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 908 {
 909     kern_return_t       kernResult;
 910     mach_port_t     masterPort;
 911     CFMutableDictionaryRef  classesToMatch;
 912
 913     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 914     if ( KERN_SUCCESS != kernResult ) {
 915         printf( "IOMasterPort returned %d\n", kernResult );
 916     }
 917
 918     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 919     if ( classesToMatch == NULL ) {
 920         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 921     } else {
 922     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 923     }
 924     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 925     if ( KERN_SUCCESS != kernResult )
 926     {
 927         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 928     }
 929
 930     return kernResult;
 931 }
 932
 933 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 934 {
 935     io_object_t     nextMedia;
 936     kern_return_t   kernResult = KERN_FAILURE;
 937     *bsdPath = '\0';
 938     nextMedia = IOIteratorNext( mediaIterator );
 939     if ( nextMedia )
 940     {
 941         CFTypeRef   bsdPathAsCFString;
 942     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 943         if ( bsdPathAsCFString ) {
 944             size_t devPathLength;
 945             strcpy( bsdPath, _PATH_DEV );
 946             strcat( bsdPath, "r" );
 947             devPathLength = strlen( bsdPath );
 948             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 949                 kernResult = KERN_SUCCESS;
 950             }
 951             CFRelease( bsdPathAsCFString );
 952         }
 953         IOObjectRelease( nextMedia );
 954     }
 955
 956     return kernResult;
 957 }
 958
 959 #endif
 960
 961 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 962 {
 963     BDRVRawState *s = bs->opaque;
 964     int fd, open_flags, ret, i;
 965
 966     posix_aio_init();
 967
 968 #ifdef CONFIG_COCOA
 969     if (strstart(filename, "/dev/cdrom", NULL)) {
 970         kern_return_t kernResult;
 971         io_iterator_t mediaIterator;
 972         char bsdPath[ MAXPATHLEN ];
 973         int fd;
 974
 975         kernResult = FindEjectableCDMedia( &mediaIterator );
 976         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 977
 978         if ( bsdPath[ 0 ] != '\0' ) {
 979             strcat(bsdPath,"s0");
 980             /* some CDs don't have a partition 0 */
 981             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 982             if (fd < 0) {
 983                 bsdPath[strlen(bsdPath)-1] = '1';
 984             } else {
 985                 close(fd);
 986             }
 987             filename = bsdPath;
 988         }
 989
 990         if ( mediaIterator )
 991             IOObjectRelease( mediaIterator );
 992     }
 993 #endif
 994     open_flags = O_BINARY;
 995     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 996         open_flags |= O_RDWR;
 997     } else {
 998         open_flags |= O_RDONLY;
 999         bs->read_only = 1;
1000     }
1001 #ifdef O_DIRECT
1002     if (flags & BDRV_O_DIRECT)
1003         open_flags |= O_DIRECT;
1004 #endif
1005
1006     s->type = FTYPE_FILE;
1007 #if defined(__linux__)
1008     if (strstart(filename, "/dev/cd", NULL)) {
1009         /* open will not fail even if no CD is inserted */
1010         open_flags |= O_NONBLOCK;
1011         s->type = FTYPE_CD;
1012     } else if (strstart(filename, "/dev/fd", NULL)) {
1013         s->type = FTYPE_FD;
1014         s->fd_open_flags = open_flags;
1015         /* open will not fail even if no floppy is inserted */
1016         open_flags |= O_NONBLOCK;
1017     } else if (strstart(filename, "/dev/sg", NULL)) {
1018         bs->sg = 1;
1019     }
1020 #endif
1021     fd = open(filename, open_flags, 0644);
1022     if (fd < 0) {
1023         ret = -errno;
1024         if (ret == -EROFS)
1025             ret = -EACCES;
1026         return ret;
1027     }
1028     s->fd = fd;
1029     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
1030         s->fd_pool[i] = -1;
1031 #if defined(__linux__)
1032     /* close fd so that we can reopen it as needed */
1033     if (s->type == FTYPE_FD) {
1034         close(s->fd);
1035         s->fd = -1;
1036         s->fd_media_changed = 1;
1037     }
1038 #endif
1039     return 0;
1040 }
1041
1042 #if defined(__linux__)
1043 /* Note: we do not have a reliable method to detect if the floppy is
1044    present. The current method is to try to open the floppy at every
1045    I/O and to keep it opened during a few hundreds of ms. */
1046 static int fd_open(BlockDriverState *bs)
1047 {
1048     BDRVRawState *s = bs->opaque;
1049     int last_media_present;
1050
1051     if (s->type != FTYPE_FD)
1052         return 0;
1053     last_media_present = (s->fd >= 0);
1054     if (s->fd >= 0 &&
1055         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1056         close(s->fd);
1057         s->fd = -1;
1058         raw_close_fd_pool(s);
1059 #ifdef DEBUG_FLOPPY
1060         printf("Floppy closed\n");
1061 #endif
1062     }
1063     if (s->fd < 0) {
1064         if (s->fd_got_error &&
1065             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1066 #ifdef DEBUG_FLOPPY
1067             printf("No floppy (open delayed)\n");
1068 #endif
1069             return -EIO;
1070         }
1071         s->fd = open(bs->filename, s->fd_open_flags);
1072         if (s->fd < 0) {
1073             s->fd_error_time = qemu_get_clock(rt_clock);
1074             s->fd_got_error = 1;
1075             if (last_media_present)
1076                 s->fd_media_changed = 1;
1077 #ifdef DEBUG_FLOPPY
1078             printf("No floppy\n");
1079 #endif
1080             return -EIO;
1081         }
1082 #ifdef DEBUG_FLOPPY
1083         printf("Floppy opened\n");
1084 #endif
1085     }
1086     if (!last_media_present)
1087         s->fd_media_changed = 1;
1088     s->fd_open_time = qemu_get_clock(rt_clock);
1089     s->fd_got_error = 0;
1090     return 0;
1091 }
1092
1093 static int raw_is_inserted(BlockDriverState *bs)
1094 {
1095     BDRVRawState *s = bs->opaque;
1096     int ret;
1097
1098     switch(s->type) {
1099     case FTYPE_CD:
1100         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1101         if (ret == CDS_DISC_OK)
1102             return 1;
1103         else
1104             return 0;
1105         break;
1106     case FTYPE_FD:
1107         ret = fd_open(bs);
1108         return (ret >= 0);
1109     default:
1110         return 1;
1111     }
1112 }
1113
1114 /* currently only used by fdc.c, but a CD version would be good too */
1115 static int raw_media_changed(BlockDriverState *bs)
1116 {
1117     BDRVRawState *s = bs->opaque;
1118
1119     switch(s->type) {
1120     case FTYPE_FD:
1121         {
1122             int ret;
1123             /* XXX: we do not have a true media changed indication. It
1124                does not work if the floppy is changed without trying
1125                to read it */
1126             fd_open(bs);
1127             ret = s->fd_media_changed;
1128             s->fd_media_changed = 0;
1129 #ifdef DEBUG_FLOPPY
1130             printf("Floppy changed=%d\n", ret);
1131 #endif
1132             return ret;
1133         }
1134     default:
1135         return -ENOTSUP;
1136     }
1137 }
1138
1139 static int raw_eject(BlockDriverState *bs, int eject_flag)
1140 {
1141     BDRVRawState *s = bs->opaque;
1142
1143     switch(s->type) {
1144     case FTYPE_CD:
1145         if (eject_flag) {
1146             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1147                 perror("CDROMEJECT");
1148         } else {
1149             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1150                 perror("CDROMEJECT");
1151         }
1152         break;
1153     case FTYPE_FD:
1154         {
1155             int fd;
1156             if (s->fd >= 0) {
1157                 close(s->fd);
1158                 s->fd = -1;
1159                 raw_close_fd_pool(s);
1160             }
1161             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1162             if (fd >= 0) {
1163                 if (ioctl(fd, FDEJECT, 0) < 0)
1164                     perror("FDEJECT");
1165                 close(fd);
1166             }
1167         }
1168         break;
1169     default:
1170         return -ENOTSUP;
1171     }
1172     return 0;
1173 }
1174
1175 static int raw_set_locked(BlockDriverState *bs, int locked)
1176 {
1177     BDRVRawState *s = bs->opaque;
1178
1179     switch(s->type) {
1180     case FTYPE_CD:
1181         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1182             /* Note: an error can happen if the distribution automatically
1183                mounts the CD-ROM */
1184             //        perror("CDROM_LOCKDOOR");
1185         }
1186         break;
1187     default:
1188         return -ENOTSUP;
1189     }
1190     return 0;
1191 }
1192
1193 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1194 {
1195     BDRVRawState *s = bs->opaque;
1196
1197     return ioctl(s->fd, req, buf);
1198 }
1199 #else
1200
1201 static int fd_open(BlockDriverState *bs)
1202 {
1203     return 0;
1204 }
1205
1206 static int raw_is_inserted(BlockDriverState *bs)
1207 {
1208     return 1;
1209 }
1210
1211 static int raw_media_changed(BlockDriverState *bs)
1212 {
1213     return -ENOTSUP;
1214 }
1215
1216 static int raw_eject(BlockDriverState *bs, int eject_flag)
1217 {
1218     return -ENOTSUP;
1219 }
1220
1221 static int raw_set_locked(BlockDriverState *bs, int locked)
1222 {
1223     return -ENOTSUP;
1224 }
1225
1226 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1227 {
1228     return -ENOTSUP;
1229 }
1230 #endif /* !linux */
1231
1232 BlockDriver bdrv_host_device = {
1233     "host_device",
1234     sizeof(BDRVRawState),
1235     NULL, /* no probe for protocols */
1236     hdev_open,
1237     NULL,
1238     NULL,
1239     raw_close,
1240     NULL,
1241     raw_flush,
1242
1243 #ifdef CONFIG_AIO
1244     .bdrv_aio_read = raw_aio_read,
1245     .bdrv_aio_write = raw_aio_write,
1246     .bdrv_aio_cancel = raw_aio_cancel,
1247     .aiocb_size = sizeof(RawAIOCB),
1248 #endif
1249     .bdrv_pread = raw_pread,
1250     .bdrv_pwrite = raw_pwrite,
1251     .bdrv_getlength = raw_getlength,
1252
1253     /* removable device support */
1254     .bdrv_is_inserted = raw_is_inserted,
1255     .bdrv_media_changed = raw_media_changed,
1256     .bdrv_eject = raw_eject,
1257     .bdrv_set_locked = raw_set_locked,
1258     /* generic scsi device */
1259     .bdrv_ioctl = raw_ioctl,
1260 };