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 <aio.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 (loglevel != 0)  \
  71     { fprintf(logfile, formatCstr, ##args); fflush(logfile); } } 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 #define O_DSYNC O_SYNC
  79 #endif
  80
  81 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
  82 #ifndef O_DIRECT
  83 #define O_DIRECT O_DSYNC
  84 #endif
  85
  86 #define FTYPE_FILE   0
  87 #define FTYPE_CD     1
  88 #define FTYPE_FD     2
  89
  90 #define ALIGNED_BUFFER_SIZE (32 * 512)
  91
  92 /* if the FD is not accessed during that time (in ms), we try to
  93    reopen it to see if the disk has been changed */
  94 #define FD_OPEN_TIMEOUT 1000
  95
  96 /* posix-aio doesn't allow multiple outstanding requests to a single file
  97  * descriptor.  we implement a pool of dup()'d file descriptors to work
  98  * around this */
  99 #define RAW_FD_POOL_SIZE        64
 100
 101 typedef struct BDRVRawState {
 102     int fd;
 103     int type;
 104     unsigned int lseek_err_cnt;
 105     int fd_pool[RAW_FD_POOL_SIZE];
 106 #if defined(__linux__)
 107     /* linux floppy specific */
 108     int fd_open_flags;
 109     int64_t fd_open_time;
 110     int64_t fd_error_time;
 111     int fd_got_error;
 112     int fd_media_changed;
 113 #endif
 114     uint8_t* aligned_buf;
 115 } BDRVRawState;
 116
 117 static int posix_aio_init(void);
 118
 119 static int fd_open(BlockDriverState *bs);
 120
 121 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 122 {
 123     BDRVRawState *s = bs->opaque;
 124     int fd, open_flags, ret;
 125     int i;
 126
 127     posix_aio_init();
 128
 129     s->lseek_err_cnt = 0;
 130
 131     open_flags = O_BINARY;
 132     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
 133         open_flags |= O_RDWR;
 134     } else {
 135         open_flags |= O_RDONLY;
 136         bs->read_only = 1;
 137     }
 138     if (flags & BDRV_O_CREAT)
 139         open_flags |= O_CREAT | O_TRUNC;
 140
 141     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 142      * and O_DIRECT for no caching. */
 143     if ((flags & BDRV_O_NOCACHE))
 144         open_flags |= O_DIRECT;
 145     else if (!(flags & BDRV_O_CACHE_WB))
 146         open_flags |= O_DSYNC;
 147
 148     s->type = FTYPE_FILE;
 149
 150     fd = open(filename, open_flags, 0644);
 151     if (fd < 0) {
 152         ret = -errno;
 153         if (ret == -EROFS)
 154             ret = -EACCES;
 155         return ret;
 156     }
 157     s->fd = fd;
 158     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
 159         s->fd_pool[i] = -1;
 160     s->aligned_buf = NULL;
 161     if ((flags & BDRV_O_NOCACHE)) {
 162         s->aligned_buf = qemu_memalign(512, ALIGNED_BUFFER_SIZE);
 163         if (s->aligned_buf == NULL) {
 164             ret = -errno;
 165             close(fd);
 166             return ret;
 167         }
 168     }
 169     return 0;
 170 }
 171
 172 /* XXX: use host sector size if necessary with:
 173 #ifdef DIOCGSECTORSIZE
 174         {
 175             unsigned int sectorsize = 512;
 176             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 177                 sectorsize > bufsize)
 178                 bufsize = sectorsize;
 179         }
 180 #endif
 181 #ifdef CONFIG_COCOA
 182         u_int32_t   blockSize = 512;
 183         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 184             bufsize = blockSize;
 185         }
 186 #endif
 187 */
 188
 189 /*
 190  * offset and count are in bytes, but must be multiples of 512 for files
 191  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 192  *
 193  * This function may be called without alignment if the caller ensures
 194  * that O_DIRECT is not in effect.
 195  */
 196 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
 197                      uint8_t *buf, int count)
 198 {
 199     BDRVRawState *s = bs->opaque;
 200     int ret;
 201
 202     ret = fd_open(bs);
 203     if (ret < 0)
 204         return ret;
 205
 206     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 207         ++(s->lseek_err_cnt);
 208         if(s->lseek_err_cnt <= 10) {
 209             DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 210                               "] lseek failed : %d = %s\n",
 211                               s->fd, bs->filename, offset, buf, count,
 212                               bs->total_sectors, errno, strerror(errno));
 213         }
 214         return -1;
 215     }
 216     s->lseek_err_cnt=0;
 217
 218     ret = read(s->fd, buf, count);
 219     if (ret == count)
 220         goto label__raw_read__success;
 221
 222     DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 223                       "] read failed %d : %d = %s\n",
 224                       s->fd, bs->filename, offset, buf, count,
 225                       bs->total_sectors, ret, errno, strerror(errno));
 226
 227     /* Try harder for CDrom. */
 228     if (bs->type == BDRV_TYPE_CDROM) {
 229         lseek(s->fd, offset, SEEK_SET);
 230         ret = read(s->fd, buf, count);
 231         if (ret == count)
 232             goto label__raw_read__success;
 233         lseek(s->fd, offset, SEEK_SET);
 234         ret = read(s->fd, buf, count);
 235         if (ret == count)
 236             goto label__raw_read__success;
 237
 238         DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 239                           "] retry read failed %d : %d = %s\n",
 240                           s->fd, bs->filename, offset, buf, count,
 241                           bs->total_sectors, ret, errno, strerror(errno));
 242     }
 243
 244 label__raw_read__success:
 245
 246     return ret;
 247 }
 248
 249 /*
 250  * offset and count are in bytes, but must be multiples of 512 for files
 251  * opened with O_DIRECT. buf must be aligned to 512 bytes then.
 252  *
 253  * This function may be called without alignment if the caller ensures
 254  * that O_DIRECT is not in effect.
 255  */
 256 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
 257                       const uint8_t *buf, int count)
 258 {
 259     BDRVRawState *s = bs->opaque;
 260     int ret;
 261
 262     ret = fd_open(bs);
 263     if (ret < 0)
 264         return ret;
 265
 266     if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
 267         ++(s->lseek_err_cnt);
 268         if(s->lseek_err_cnt) {
 269             DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
 270                               PRId64 "] lseek failed : %d = %s\n",
 271                               s->fd, bs->filename, offset, buf, count,
 272                               bs->total_sectors, errno, strerror(errno));
 273         }
 274         return -1;
 275     }
 276     s->lseek_err_cnt = 0;
 277
 278     ret = write(s->fd, buf, count);
 279     if (ret == count)
 280         goto label__raw_write__success;
 281
 282     DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
 283                       "] write failed %d : %d = %s\n",
 284                       s->fd, bs->filename, offset, buf, count,
 285                       bs->total_sectors, ret, errno, strerror(errno));
 286
 287 label__raw_write__success:
 288
 289     return ret;
 290 }
 291
 292
 293 /*
 294  * offset and count are in bytes and possibly not aligned. For files opened
 295  * with O_DIRECT, necessary alignments are ensured before calling
 296  * raw_pread_aligned to do the actual read.
 297  */
 298 static int raw_pread(BlockDriverState *bs, int64_t offset,
 299                      uint8_t *buf, int count)
 300 {
 301     BDRVRawState *s = bs->opaque;
 302     int size, ret, shift, sum;
 303
 304     sum = 0;
 305
 306     if (s->aligned_buf != NULL)  {
 307
 308         if (offset & 0x1ff) {
 309             /* align offset on a 512 bytes boundary */
 310
 311             shift = offset & 0x1ff;
 312             size = (shift + count + 0x1ff) & ~0x1ff;
 313             if (size > ALIGNED_BUFFER_SIZE)
 314                 size = ALIGNED_BUFFER_SIZE;
 315             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
 316             if (ret < 0)
 317                 return ret;
 318
 319             size = 512 - shift;
 320             if (size > count)
 321                 size = count;
 322             memcpy(buf, s->aligned_buf + shift, size);
 323
 324             buf += size;
 325             offset += size;
 326             count -= size;
 327             sum += size;
 328
 329             if (count == 0)
 330                 return sum;
 331         }
 332         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 333
 334             /* read on aligned buffer */
 335
 336             while (count) {
 337
 338                 size = (count + 0x1ff) & ~0x1ff;
 339                 if (size > ALIGNED_BUFFER_SIZE)
 340                     size = ALIGNED_BUFFER_SIZE;
 341
 342                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
 343                 if (ret < 0)
 344                     return ret;
 345
 346                 size = ret;
 347                 if (size > count)
 348                     size = count;
 349
 350                 memcpy(buf, s->aligned_buf, size);
 351
 352                 buf += size;
 353                 offset += size;
 354                 count -= size;
 355                 sum += size;
 356             }
 357
 358             return sum;
 359         }
 360     }
 361
 362     return raw_pread_aligned(bs, offset, buf, count) + sum;
 363 }
 364
 365 /*
 366  * offset and count are in bytes and possibly not aligned. For files opened
 367  * with O_DIRECT, necessary alignments are ensured before calling
 368  * raw_pwrite_aligned to do the actual write.
 369  */
 370 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
 371                       const uint8_t *buf, int count)
 372 {
 373     BDRVRawState *s = bs->opaque;
 374     int size, ret, shift, sum;
 375
 376     sum = 0;
 377
 378     if (s->aligned_buf != NULL) {
 379
 380         if (offset & 0x1ff) {
 381             /* align offset on a 512 bytes boundary */
 382             shift = offset & 0x1ff;
 383             ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
 384             if (ret < 0)
 385                 return ret;
 386
 387             size = 512 - shift;
 388             if (size > count)
 389                 size = count;
 390             memcpy(s->aligned_buf + shift, buf, size);
 391
 392             ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
 393             if (ret < 0)
 394                 return ret;
 395
 396             buf += size;
 397             offset += size;
 398             count -= size;
 399             sum += size;
 400
 401             if (count == 0)
 402                 return sum;
 403         }
 404         if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
 405
 406             while ((size = (count & ~0x1ff)) != 0) {
 407
 408                 if (size > ALIGNED_BUFFER_SIZE)
 409                     size = ALIGNED_BUFFER_SIZE;
 410
 411                 memcpy(s->aligned_buf, buf, size);
 412
 413                 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
 414                 if (ret < 0)
 415                     return ret;
 416
 417                 buf += ret;
 418                 offset += ret;
 419                 count -= ret;
 420                 sum += ret;
 421             }
 422             /* here, count < 512 because (count & ~0x1ff) == 0 */
 423             if (count) {
 424                 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
 425                 if (ret < 0)
 426                     return ret;
 427                  memcpy(s->aligned_buf, buf, count);
 428
 429                  ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
 430                  if (ret < 0)
 431                      return ret;
 432                  if (count < ret)
 433                      ret = count;
 434
 435                  sum += ret;
 436             }
 437             return sum;
 438         }
 439     }
 440     return raw_pwrite_aligned(bs, offset, buf, count) + sum;
 441 }
 442
 443 #ifdef CONFIG_AIO
 444 /***********************************************************/
 445 /* Unix AIO using POSIX AIO */
 446
 447 typedef struct RawAIOCB {
 448     BlockDriverAIOCB common;
 449     int fd;
 450     struct aiocb aiocb;
 451     struct RawAIOCB *next;
 452     int ret;
 453 } RawAIOCB;
 454
 455 typedef struct PosixAioState
 456 {
 457     int rfd, wfd;
 458     RawAIOCB *first_aio;
 459 } PosixAioState;
 460
 461 static int raw_fd_pool_get(BDRVRawState *s)
 462 {
 463     int i;
 464
 465     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 466         /* already in use */
 467         if (s->fd_pool[i] != -1)
 468             continue;
 469
 470         /* try to dup file descriptor */
 471         s->fd_pool[i] = dup(s->fd);
 472         if (s->fd_pool[i] != -1)
 473             return s->fd_pool[i];
 474     }
 475
 476     /* we couldn't dup the file descriptor so just use the main one */
 477     return s->fd;
 478 }
 479
 480 static void raw_fd_pool_put(RawAIOCB *acb)
 481 {
 482     BDRVRawState *s = acb->common.bs->opaque;
 483     int i;
 484
 485     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 486         if (s->fd_pool[i] == acb->fd) {
 487             close(s->fd_pool[i]);
 488             s->fd_pool[i] = -1;
 489         }
 490     }
 491 }
 492
 493 static void posix_aio_read(void *opaque)
 494 {
 495     PosixAioState *s = opaque;
 496     RawAIOCB *acb, **pacb;
 497     int ret;
 498     ssize_t len;
 499
 500     /* read all bytes from signal pipe */
 501     for (;;) {
 502         char bytes[16];
 503
 504         len = read(s->rfd, bytes, sizeof(bytes));
 505         if (len == -1 && errno == EINTR)
 506             continue; /* try again */
 507         if (len == sizeof(bytes))
 508             continue; /* more to read */
 509         break;
 510     }
 511
 512     for(;;) {
 513         pacb = &s->first_aio;
 514         for(;;) {
 515             acb = *pacb;
 516             if (!acb)
 517                 goto the_end;
 518             ret = aio_error(&acb->aiocb);
 519             if (ret == ECANCELED) {
 520                 /* remove the request */
 521                 *pacb = acb->next;
 522                 raw_fd_pool_put(acb);
 523                 qemu_aio_release(acb);
 524             } else if (ret != EINPROGRESS) {
 525                 /* end of aio */
 526                 if (ret == 0) {
 527                     ret = aio_return(&acb->aiocb);
 528                     if (ret == acb->aiocb.aio_nbytes)
 529                         ret = 0;
 530                     else
 531                         ret = -EINVAL;
 532                 } else {
 533                     ret = -ret;
 534                 }
 535                 /* remove the request */
 536                 *pacb = acb->next;
 537                 /* call the callback */
 538                 acb->common.cb(acb->common.opaque, ret);
 539                 raw_fd_pool_put(acb);
 540                 qemu_aio_release(acb);
 541                 break;
 542             } else {
 543                 pacb = &acb->next;
 544             }
 545         }
 546     }
 547  the_end: ;
 548 }
 549
 550 static int posix_aio_flush(void *opaque)
 551 {
 552     PosixAioState *s = opaque;
 553     return !!s->first_aio;
 554 }
 555
 556 static PosixAioState *posix_aio_state;
 557
 558 static void aio_signal_handler(int signum)
 559 {
 560     if (posix_aio_state) {
 561         char byte = 0;
 562
 563         write(posix_aio_state->wfd, &byte, sizeof(byte));
 564     }
 565
 566     qemu_service_io();
 567 }
 568
 569 static int posix_aio_init(void)
 570 {
 571     struct sigaction act;
 572     PosixAioState *s;
 573     int fds[2];
 574
 575     if (posix_aio_state)
 576         return 0;
 577
 578     s = qemu_malloc(sizeof(PosixAioState));
 579     if (s == NULL)
 580         return -ENOMEM;
 581
 582     sigfillset(&act.sa_mask);
 583     act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
 584     act.sa_handler = aio_signal_handler;
 585     sigaction(SIGUSR2, &act, NULL);
 586
 587     s->first_aio = NULL;
 588     if (pipe(fds) == -1) {
 589         fprintf(stderr, "failed to create pipe\n");
 590         return -errno;
 591     }
 592
 593     s->rfd = fds[0];
 594     s->wfd = fds[1];
 595
 596     fcntl(s->rfd, F_SETFL, O_NONBLOCK);
 597     fcntl(s->wfd, F_SETFL, O_NONBLOCK);
 598
 599     qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
 600
 601 #if defined(__linux__)
 602     {
 603         struct aioinit ai;
 604
 605         memset(&ai, 0, sizeof(ai));
 606 #if defined(__GLIBC_PREREQ) && __GLIBC_PREREQ(2, 4)
 607         ai.aio_threads = 64;
 608         ai.aio_num = 64;
 609 #else
 610         /* XXX: aio thread exit seems to hang on RedHat 9 and this init
 611            seems to fix the problem. */
 612         ai.aio_threads = 1;
 613         ai.aio_num = 1;
 614         ai.aio_idle_time = 365 * 100000;
 615 #endif
 616         aio_init(&ai);
 617     }
 618 #endif
 619     posix_aio_state = s;
 620
 621     return 0;
 622 }
 623
 624 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
 625         int64_t sector_num, uint8_t *buf, int nb_sectors,
 626         BlockDriverCompletionFunc *cb, void *opaque)
 627 {
 628     BDRVRawState *s = bs->opaque;
 629     RawAIOCB *acb;
 630
 631     if (fd_open(bs) < 0)
 632         return NULL;
 633
 634     acb = qemu_aio_get(bs, cb, opaque);
 635     if (!acb)
 636         return NULL;
 637     acb->fd = raw_fd_pool_get(s);
 638     acb->aiocb.aio_fildes = acb->fd;
 639     acb->aiocb.aio_sigevent.sigev_signo = SIGUSR2;
 640     acb->aiocb.aio_sigevent.sigev_notify = SIGEV_SIGNAL;
 641     acb->aiocb.aio_buf = buf;
 642     if (nb_sectors < 0)
 643         acb->aiocb.aio_nbytes = -nb_sectors;
 644     else
 645         acb->aiocb.aio_nbytes = nb_sectors * 512;
 646     acb->aiocb.aio_offset = sector_num * 512;
 647     acb->next = posix_aio_state->first_aio;
 648     posix_aio_state->first_aio = acb;
 649     return acb;
 650 }
 651
 652 static void raw_aio_em_cb(void* opaque)
 653 {
 654     RawAIOCB *acb = opaque;
 655     acb->common.cb(acb->common.opaque, acb->ret);
 656     qemu_aio_release(acb);
 657 }
 658
 659 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
 660         int64_t sector_num, uint8_t *buf, int nb_sectors,
 661         BlockDriverCompletionFunc *cb, void *opaque)
 662 {
 663     RawAIOCB *acb;
 664
 665     /*
 666      * If O_DIRECT is used and the buffer is not aligned fall back
 667      * to synchronous IO.
 668      */
 669     BDRVRawState *s = bs->opaque;
 670
 671     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 672         QEMUBH *bh;
 673         acb = qemu_aio_get(bs, cb, opaque);
 674         acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
 675         bh = qemu_bh_new(raw_aio_em_cb, acb);
 676         qemu_bh_schedule(bh);
 677         return &acb->common;
 678     }
 679
 680     acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
 681     if (!acb)
 682         return NULL;
 683     if (aio_read(&acb->aiocb) < 0) {
 684         qemu_aio_release(acb);
 685         return NULL;
 686     }
 687     return &acb->common;
 688 }
 689
 690 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
 691         int64_t sector_num, const uint8_t *buf, int nb_sectors,
 692         BlockDriverCompletionFunc *cb, void *opaque)
 693 {
 694     RawAIOCB *acb;
 695
 696     /*
 697      * If O_DIRECT is used and the buffer is not aligned fall back
 698      * to synchronous IO.
 699      */
 700     BDRVRawState *s = bs->opaque;
 701
 702     if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
 703         QEMUBH *bh;
 704         acb = qemu_aio_get(bs, cb, opaque);
 705         acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
 706         bh = qemu_bh_new(raw_aio_em_cb, acb);
 707         qemu_bh_schedule(bh);
 708         return &acb->common;
 709     }
 710
 711     acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
 712     if (!acb)
 713         return NULL;
 714     if (aio_write(&acb->aiocb) < 0) {
 715         qemu_aio_release(acb);
 716         return NULL;
 717     }
 718     return &acb->common;
 719 }
 720
 721 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
 722 {
 723     int ret;
 724     RawAIOCB *acb = (RawAIOCB *)blockacb;
 725     RawAIOCB **pacb;
 726
 727     ret = aio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
 728     if (ret == AIO_NOTCANCELED) {
 729         /* fail safe: if the aio could not be canceled, we wait for
 730            it */
 731         while (aio_error(&acb->aiocb) == EINPROGRESS);
 732     }
 733
 734     /* remove the callback from the queue */
 735     pacb = &posix_aio_state->first_aio;
 736     for(;;) {
 737         if (*pacb == NULL) {
 738             break;
 739         } else if (*pacb == acb) {
 740             *pacb = acb->next;
 741             raw_fd_pool_put(acb);
 742             qemu_aio_release(acb);
 743             break;
 744         }
 745         pacb = &acb->next;
 746     }
 747 }
 748
 749 #else /* CONFIG_AIO */
 750 static int posix_aio_init(void)
 751 {
 752     return 0;
 753 }
 754 #endif /* CONFIG_AIO */
 755
 756 static void raw_close_fd_pool(BDRVRawState *s)
 757 {
 758     int i;
 759
 760     for (i = 0; i < RAW_FD_POOL_SIZE; i++) {
 761         if (s->fd_pool[i] != -1) {
 762             close(s->fd_pool[i]);
 763             s->fd_pool[i] = -1;
 764         }
 765     }
 766 }
 767
 768 static void raw_close(BlockDriverState *bs)
 769 {
 770     BDRVRawState *s = bs->opaque;
 771     if (s->fd >= 0) {
 772         close(s->fd);
 773         s->fd = -1;
 774         if (s->aligned_buf != NULL)
 775             qemu_free(s->aligned_buf);
 776     }
 777     raw_close_fd_pool(s);
 778 }
 779
 780 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 781 {
 782     BDRVRawState *s = bs->opaque;
 783     if (s->type != FTYPE_FILE)
 784         return -ENOTSUP;
 785     if (ftruncate(s->fd, offset) < 0)
 786         return -errno;
 787     return 0;
 788 }
 789
 790 #ifdef __OpenBSD__
 791 static int64_t raw_getlength(BlockDriverState *bs)
 792 {
 793     BDRVRawState *s = bs->opaque;
 794     int fd = s->fd;
 795     struct stat st;
 796
 797     if (fstat(fd, &st))
 798         return -1;
 799     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 800         struct disklabel dl;
 801
 802         if (ioctl(fd, DIOCGDINFO, &dl))
 803             return -1;
 804         return (uint64_t)dl.d_secsize *
 805             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 806     } else
 807         return st.st_size;
 808 }
 809 #else /* !__OpenBSD__ */
 810 static int64_t  raw_getlength(BlockDriverState *bs)
 811 {
 812     BDRVRawState *s = bs->opaque;
 813     int fd = s->fd;
 814     int64_t size;
 815 #ifdef _BSD
 816     struct stat sb;
 817 #endif
 818 #ifdef __sun__
 819     struct dk_minfo minfo;
 820     int rv;
 821 #endif
 822     int ret;
 823
 824     ret = fd_open(bs);
 825     if (ret < 0)
 826         return ret;
 827
 828 #ifdef _BSD
 829     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 830 #ifdef DIOCGMEDIASIZE
 831         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 832 #endif
 833 #ifdef CONFIG_COCOA
 834         size = LONG_LONG_MAX;
 835 #else
 836         size = lseek(fd, 0LL, SEEK_END);
 837 #endif
 838     } else
 839 #endif
 840 #ifdef __sun__
 841     /*
 842      * use the DKIOCGMEDIAINFO ioctl to read the size.
 843      */
 844     rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
 845     if ( rv != -1 ) {
 846         size = minfo.dki_lbsize * minfo.dki_capacity;
 847     } else /* there are reports that lseek on some devices
 848               fails, but irc discussion said that contingency
 849               on contingency was overkill */
 850 #endif
 851     {
 852         size = lseek(fd, 0, SEEK_END);
 853     }
 854     return size;
 855 }
 856 #endif
 857
 858 static int raw_create(const char *filename, int64_t total_size,
 859                       const char *backing_file, int flags)
 860 {
 861     int fd;
 862
 863     if (flags || backing_file)
 864         return -ENOTSUP;
 865
 866     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
 867               0644);
 868     if (fd < 0)
 869         return -EIO;
 870     ftruncate(fd, total_size * 512);
 871     close(fd);
 872     return 0;
 873 }
 874
 875 static void raw_flush(BlockDriverState *bs)
 876 {
 877     BDRVRawState *s = bs->opaque;
 878     fsync(s->fd);
 879 }
 880
 881 BlockDriver bdrv_raw = {
 882     "raw",
 883     sizeof(BDRVRawState),
 884     NULL, /* no probe for protocols */
 885     raw_open,
 886     NULL,
 887     NULL,
 888     raw_close,
 889     raw_create,
 890     raw_flush,
 891
 892 #ifdef CONFIG_AIO
 893     .bdrv_aio_read = raw_aio_read,
 894     .bdrv_aio_write = raw_aio_write,
 895     .bdrv_aio_cancel = raw_aio_cancel,
 896     .aiocb_size = sizeof(RawAIOCB),
 897 #endif
 898     .bdrv_pread = raw_pread,
 899     .bdrv_pwrite = raw_pwrite,
 900     .bdrv_truncate = raw_truncate,
 901     .bdrv_getlength = raw_getlength,
 902 };
 903
 904 /***********************************************/
 905 /* host device */
 906
 907 #ifdef CONFIG_COCOA
 908 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
 909 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
 910
 911 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
 912 {
 913     kern_return_t       kernResult;
 914     mach_port_t     masterPort;
 915     CFMutableDictionaryRef  classesToMatch;
 916
 917     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
 918     if ( KERN_SUCCESS != kernResult ) {
 919         printf( "IOMasterPort returned %d\n", kernResult );
 920     }
 921
 922     classesToMatch = IOServiceMatching( kIOCDMediaClass );
 923     if ( classesToMatch == NULL ) {
 924         printf( "IOServiceMatching returned a NULL dictionary.\n" );
 925     } else {
 926     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
 927     }
 928     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
 929     if ( KERN_SUCCESS != kernResult )
 930     {
 931         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
 932     }
 933
 934     return kernResult;
 935 }
 936
 937 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
 938 {
 939     io_object_t     nextMedia;
 940     kern_return_t   kernResult = KERN_FAILURE;
 941     *bsdPath = '\0';
 942     nextMedia = IOIteratorNext( mediaIterator );
 943     if ( nextMedia )
 944     {
 945         CFTypeRef   bsdPathAsCFString;
 946     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
 947         if ( bsdPathAsCFString ) {
 948             size_t devPathLength;
 949             strcpy( bsdPath, _PATH_DEV );
 950             strcat( bsdPath, "r" );
 951             devPathLength = strlen( bsdPath );
 952             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
 953                 kernResult = KERN_SUCCESS;
 954             }
 955             CFRelease( bsdPathAsCFString );
 956         }
 957         IOObjectRelease( nextMedia );
 958     }
 959
 960     return kernResult;
 961 }
 962
 963 #endif
 964
 965 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
 966 {
 967     BDRVRawState *s = bs->opaque;
 968     int fd, open_flags, ret, i;
 969
 970     posix_aio_init();
 971
 972 #ifdef CONFIG_COCOA
 973     if (strstart(filename, "/dev/cdrom", NULL)) {
 974         kern_return_t kernResult;
 975         io_iterator_t mediaIterator;
 976         char bsdPath[ MAXPATHLEN ];
 977         int fd;
 978
 979         kernResult = FindEjectableCDMedia( &mediaIterator );
 980         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
 981
 982         if ( bsdPath[ 0 ] != '\0' ) {
 983             strcat(bsdPath,"s0");
 984             /* some CDs don't have a partition 0 */
 985             fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
 986             if (fd < 0) {
 987                 bsdPath[strlen(bsdPath)-1] = '1';
 988             } else {
 989                 close(fd);
 990             }
 991             filename = bsdPath;
 992         }
 993
 994         if ( mediaIterator )
 995             IOObjectRelease( mediaIterator );
 996     }
 997 #endif
 998     open_flags = O_BINARY;
 999     if ((flags & BDRV_O_ACCESS) == O_RDWR) {
1000         open_flags |= O_RDWR;
1001     } else {
1002         open_flags |= O_RDONLY;
1003         bs->read_only = 1;
1004     }
1005     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
1006      * and O_DIRECT for no caching. */
1007     if ((flags & BDRV_O_NOCACHE))
1008         open_flags |= O_DIRECT;
1009     else if (!(flags & BDRV_O_CACHE_WB))
1010         open_flags |= O_DSYNC;
1011
1012     s->type = FTYPE_FILE;
1013 #if defined(__linux__)
1014     if (strstart(filename, "/dev/cd", NULL)) {
1015         /* open will not fail even if no CD is inserted */
1016         open_flags |= O_NONBLOCK;
1017         s->type = FTYPE_CD;
1018     } else if (strstart(filename, "/dev/fd", NULL)) {
1019         s->type = FTYPE_FD;
1020         s->fd_open_flags = open_flags;
1021         /* open will not fail even if no floppy is inserted */
1022         open_flags |= O_NONBLOCK;
1023     } else if (strstart(filename, "/dev/sg", NULL)) {
1024         bs->sg = 1;
1025     }
1026 #endif
1027     fd = open(filename, open_flags, 0644);
1028     if (fd < 0) {
1029         ret = -errno;
1030         if (ret == -EROFS)
1031             ret = -EACCES;
1032         return ret;
1033     }
1034     s->fd = fd;
1035     for (i = 0; i < RAW_FD_POOL_SIZE; i++)
1036         s->fd_pool[i] = -1;
1037 #if defined(__linux__)
1038     /* close fd so that we can reopen it as needed */
1039     if (s->type == FTYPE_FD) {
1040         close(s->fd);
1041         s->fd = -1;
1042         s->fd_media_changed = 1;
1043     }
1044 #endif
1045     return 0;
1046 }
1047
1048 #if defined(__linux__)
1049 /* Note: we do not have a reliable method to detect if the floppy is
1050    present. The current method is to try to open the floppy at every
1051    I/O and to keep it opened during a few hundreds of ms. */
1052 static int fd_open(BlockDriverState *bs)
1053 {
1054     BDRVRawState *s = bs->opaque;
1055     int last_media_present;
1056
1057     if (s->type != FTYPE_FD)
1058         return 0;
1059     last_media_present = (s->fd >= 0);
1060     if (s->fd >= 0 &&
1061         (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1062         close(s->fd);
1063         s->fd = -1;
1064         raw_close_fd_pool(s);
1065 #ifdef DEBUG_FLOPPY
1066         printf("Floppy closed\n");
1067 #endif
1068     }
1069     if (s->fd < 0) {
1070         if (s->fd_got_error &&
1071             (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1072 #ifdef DEBUG_FLOPPY
1073             printf("No floppy (open delayed)\n");
1074 #endif
1075             return -EIO;
1076         }
1077         s->fd = open(bs->filename, s->fd_open_flags);
1078         if (s->fd < 0) {
1079             s->fd_error_time = qemu_get_clock(rt_clock);
1080             s->fd_got_error = 1;
1081             if (last_media_present)
1082                 s->fd_media_changed = 1;
1083 #ifdef DEBUG_FLOPPY
1084             printf("No floppy\n");
1085 #endif
1086             return -EIO;
1087         }
1088 #ifdef DEBUG_FLOPPY
1089         printf("Floppy opened\n");
1090 #endif
1091     }
1092     if (!last_media_present)
1093         s->fd_media_changed = 1;
1094     s->fd_open_time = qemu_get_clock(rt_clock);
1095     s->fd_got_error = 0;
1096     return 0;
1097 }
1098
1099 static int raw_is_inserted(BlockDriverState *bs)
1100 {
1101     BDRVRawState *s = bs->opaque;
1102     int ret;
1103
1104     switch(s->type) {
1105     case FTYPE_CD:
1106         ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1107         if (ret == CDS_DISC_OK)
1108             return 1;
1109         else
1110             return 0;
1111         break;
1112     case FTYPE_FD:
1113         ret = fd_open(bs);
1114         return (ret >= 0);
1115     default:
1116         return 1;
1117     }
1118 }
1119
1120 /* currently only used by fdc.c, but a CD version would be good too */
1121 static int raw_media_changed(BlockDriverState *bs)
1122 {
1123     BDRVRawState *s = bs->opaque;
1124
1125     switch(s->type) {
1126     case FTYPE_FD:
1127         {
1128             int ret;
1129             /* XXX: we do not have a true media changed indication. It
1130                does not work if the floppy is changed without trying
1131                to read it */
1132             fd_open(bs);
1133             ret = s->fd_media_changed;
1134             s->fd_media_changed = 0;
1135 #ifdef DEBUG_FLOPPY
1136             printf("Floppy changed=%d\n", ret);
1137 #endif
1138             return ret;
1139         }
1140     default:
1141         return -ENOTSUP;
1142     }
1143 }
1144
1145 static int raw_eject(BlockDriverState *bs, int eject_flag)
1146 {
1147     BDRVRawState *s = bs->opaque;
1148
1149     switch(s->type) {
1150     case FTYPE_CD:
1151         if (eject_flag) {
1152             if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1153                 perror("CDROMEJECT");
1154         } else {
1155             if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1156                 perror("CDROMEJECT");
1157         }
1158         break;
1159     case FTYPE_FD:
1160         {
1161             int fd;
1162             if (s->fd >= 0) {
1163                 close(s->fd);
1164                 s->fd = -1;
1165                 raw_close_fd_pool(s);
1166             }
1167             fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1168             if (fd >= 0) {
1169                 if (ioctl(fd, FDEJECT, 0) < 0)
1170                     perror("FDEJECT");
1171                 close(fd);
1172             }
1173         }
1174         break;
1175     default:
1176         return -ENOTSUP;
1177     }
1178     return 0;
1179 }
1180
1181 static int raw_set_locked(BlockDriverState *bs, int locked)
1182 {
1183     BDRVRawState *s = bs->opaque;
1184
1185     switch(s->type) {
1186     case FTYPE_CD:
1187         if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1188             /* Note: an error can happen if the distribution automatically
1189                mounts the CD-ROM */
1190             //        perror("CDROM_LOCKDOOR");
1191         }
1192         break;
1193     default:
1194         return -ENOTSUP;
1195     }
1196     return 0;
1197 }
1198
1199 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1200 {
1201     BDRVRawState *s = bs->opaque;
1202
1203     return ioctl(s->fd, req, buf);
1204 }
1205 #else
1206
1207 static int fd_open(BlockDriverState *bs)
1208 {
1209     return 0;
1210 }
1211
1212 static int raw_is_inserted(BlockDriverState *bs)
1213 {
1214     return 1;
1215 }
1216
1217 static int raw_media_changed(BlockDriverState *bs)
1218 {
1219     return -ENOTSUP;
1220 }
1221
1222 static int raw_eject(BlockDriverState *bs, int eject_flag)
1223 {
1224     return -ENOTSUP;
1225 }
1226
1227 static int raw_set_locked(BlockDriverState *bs, int locked)
1228 {
1229     return -ENOTSUP;
1230 }
1231
1232 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1233 {
1234     return -ENOTSUP;
1235 }
1236 #endif /* !linux */
1237
1238 BlockDriver bdrv_host_device = {
1239     "host_device",
1240     sizeof(BDRVRawState),
1241     NULL, /* no probe for protocols */
1242     hdev_open,
1243     NULL,
1244     NULL,
1245     raw_close,
1246     NULL,
1247     raw_flush,
1248
1249 #ifdef CONFIG_AIO
1250     .bdrv_aio_read = raw_aio_read,
1251     .bdrv_aio_write = raw_aio_write,
1252     .bdrv_aio_cancel = raw_aio_cancel,
1253     .aiocb_size = sizeof(RawAIOCB),
1254 #endif
1255     .bdrv_pread = raw_pread,
1256     .bdrv_pwrite = raw_pwrite,
1257     .bdrv_getlength = raw_getlength,
1258
1259     /* removable device support */
1260     .bdrv_is_inserted = raw_is_inserted,
1261     .bdrv_media_changed = raw_media_changed,
1262     .bdrv_eject = raw_eject,
1263     .bdrv_set_locked = raw_set_locked,
1264     /* generic scsi device */
1265     .bdrv_ioctl = raw_ioctl,
1266 };