]> git.proxmox.com Git - qemu.git/blob - block-raw-posix.c
406ec7d04c4b1b56db69ae3c8a0886eaa3a5bcb3
[qemu.git] / 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 #ifdef CONFIG_AIO
29 #include "posix-aio-compat.h"
30 #endif
31
32 #ifdef CONFIG_COCOA
33 #include <paths.h>
34 #include <sys/param.h>
35 #include <IOKit/IOKitLib.h>
36 #include <IOKit/IOBSD.h>
37 #include <IOKit/storage/IOMediaBSDClient.h>
38 #include <IOKit/storage/IOMedia.h>
39 #include <IOKit/storage/IOCDMedia.h>
40 //#include <IOKit/storage/IOCDTypes.h>
41 #include <CoreFoundation/CoreFoundation.h>
42 #endif
43
44 #ifdef __sun__
45 #define _POSIX_PTHREAD_SEMANTICS 1
46 #include <signal.h>
47 #include <sys/dkio.h>
48 #endif
49 #ifdef __linux__
50 #include <sys/ioctl.h>
51 #include <linux/cdrom.h>
52 #include <linux/fd.h>
53 #endif
54 #ifdef __FreeBSD__
55 #include <signal.h>
56 #include <sys/disk.h>
57 #include <sys/cdio.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 #ifdef __DragonFly__
67 #include <sys/ioctl.h>
68 #include <sys/diskslice.h>
69 #endif
70
71 //#define DEBUG_FLOPPY
72
73 //#define DEBUG_BLOCK
74 #if defined(DEBUG_BLOCK)
75 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
76 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
77 #else
78 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
79 #endif
80
81 /* OS X does not have O_DSYNC */
82 #ifndef O_DSYNC
83 #define O_DSYNC O_SYNC
84 #endif
85
86 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
87 #ifndef O_DIRECT
88 #define O_DIRECT O_DSYNC
89 #endif
90
91 #define FTYPE_FILE 0
92 #define FTYPE_CD 1
93 #define FTYPE_FD 2
94
95 #define ALIGNED_BUFFER_SIZE (32 * 512)
96
97 /* if the FD is not accessed during that time (in ms), we try to
98 reopen it to see if the disk has been changed */
99 #define FD_OPEN_TIMEOUT 1000
100
101 typedef struct BDRVRawState {
102 int fd;
103 int type;
104 unsigned int lseek_err_cnt;
105 #if defined(__linux__)
106 /* linux floppy specific */
107 int fd_open_flags;
108 int64_t fd_open_time;
109 int64_t fd_error_time;
110 int fd_got_error;
111 int fd_media_changed;
112 #endif
113 #if defined(__FreeBSD__)
114 int cd_open_flags;
115 #endif
116 uint8_t* aligned_buf;
117 } BDRVRawState;
118
119 static int posix_aio_init(void);
120
121 static int fd_open(BlockDriverState *bs);
122
123 #if defined(__FreeBSD__)
124 static int cd_open(BlockDriverState *bs);
125 #endif
126
127 static int raw_is_inserted(BlockDriverState *bs);
128
129 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
130 {
131 BDRVRawState *s = bs->opaque;
132 int fd, open_flags, ret;
133
134 posix_aio_init();
135
136 s->lseek_err_cnt = 0;
137
138 open_flags = O_BINARY;
139 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
140 open_flags |= O_RDWR;
141 } else {
142 open_flags |= O_RDONLY;
143 bs->read_only = 1;
144 }
145 if (flags & BDRV_O_CREAT)
146 open_flags |= O_CREAT | O_TRUNC;
147
148 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
149 * and O_DIRECT for no caching. */
150 if ((flags & BDRV_O_NOCACHE))
151 open_flags |= O_DIRECT;
152 else if (!(flags & BDRV_O_CACHE_WB))
153 open_flags |= O_DSYNC;
154
155 s->type = FTYPE_FILE;
156
157 fd = open(filename, open_flags, 0644);
158 if (fd < 0) {
159 ret = -errno;
160 if (ret == -EROFS)
161 ret = -EACCES;
162 return ret;
163 }
164 s->fd = fd;
165 s->aligned_buf = NULL;
166 if ((flags & BDRV_O_NOCACHE)) {
167 s->aligned_buf = qemu_blockalign(bs, ALIGNED_BUFFER_SIZE);
168 if (s->aligned_buf == NULL) {
169 ret = -errno;
170 close(fd);
171 return ret;
172 }
173 }
174 return 0;
175 }
176
177 /* XXX: use host sector size if necessary with:
178 #ifdef DIOCGSECTORSIZE
179 {
180 unsigned int sectorsize = 512;
181 if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
182 sectorsize > bufsize)
183 bufsize = sectorsize;
184 }
185 #endif
186 #ifdef CONFIG_COCOA
187 u_int32_t blockSize = 512;
188 if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
189 bufsize = blockSize;
190 }
191 #endif
192 */
193
194 /*
195 * offset and count are in bytes, but must be multiples of 512 for files
196 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
197 *
198 * This function may be called without alignment if the caller ensures
199 * that O_DIRECT is not in effect.
200 */
201 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
202 uint8_t *buf, int count)
203 {
204 BDRVRawState *s = bs->opaque;
205 int ret;
206
207 ret = fd_open(bs);
208 if (ret < 0)
209 return ret;
210
211 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
212 ++(s->lseek_err_cnt);
213 if(s->lseek_err_cnt <= 10) {
214 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
215 "] lseek failed : %d = %s\n",
216 s->fd, bs->filename, offset, buf, count,
217 bs->total_sectors, errno, strerror(errno));
218 }
219 return -1;
220 }
221 s->lseek_err_cnt=0;
222
223 ret = read(s->fd, buf, count);
224 if (ret == count)
225 goto label__raw_read__success;
226
227 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
228 "] read failed %d : %d = %s\n",
229 s->fd, bs->filename, offset, buf, count,
230 bs->total_sectors, ret, errno, strerror(errno));
231
232 /* Try harder for CDrom. */
233 if (bs->type == BDRV_TYPE_CDROM) {
234 lseek(s->fd, offset, SEEK_SET);
235 ret = read(s->fd, buf, count);
236 if (ret == count)
237 goto label__raw_read__success;
238 lseek(s->fd, offset, SEEK_SET);
239 ret = read(s->fd, buf, count);
240 if (ret == count)
241 goto label__raw_read__success;
242
243 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
244 "] retry read failed %d : %d = %s\n",
245 s->fd, bs->filename, offset, buf, count,
246 bs->total_sectors, ret, errno, strerror(errno));
247 }
248
249 label__raw_read__success:
250
251 return ret;
252 }
253
254 /*
255 * offset and count are in bytes, but must be multiples of 512 for files
256 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
257 *
258 * This function may be called without alignment if the caller ensures
259 * that O_DIRECT is not in effect.
260 */
261 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
262 const uint8_t *buf, int count)
263 {
264 BDRVRawState *s = bs->opaque;
265 int ret;
266
267 ret = fd_open(bs);
268 if (ret < 0)
269 return -errno;
270
271 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
272 ++(s->lseek_err_cnt);
273 if(s->lseek_err_cnt) {
274 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
275 PRId64 "] lseek failed : %d = %s\n",
276 s->fd, bs->filename, offset, buf, count,
277 bs->total_sectors, errno, strerror(errno));
278 }
279 return -EIO;
280 }
281 s->lseek_err_cnt = 0;
282
283 ret = write(s->fd, buf, count);
284 if (ret == count)
285 goto label__raw_write__success;
286
287 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
288 "] write failed %d : %d = %s\n",
289 s->fd, bs->filename, offset, buf, count,
290 bs->total_sectors, ret, errno, strerror(errno));
291
292 label__raw_write__success:
293
294 return (ret < 0) ? -errno : ret;
295 }
296
297
298 /*
299 * offset and count are in bytes and possibly not aligned. For files opened
300 * with O_DIRECT, necessary alignments are ensured before calling
301 * raw_pread_aligned to do the actual read.
302 */
303 static int raw_pread(BlockDriverState *bs, int64_t offset,
304 uint8_t *buf, int count)
305 {
306 BDRVRawState *s = bs->opaque;
307 int size, ret, shift, sum;
308
309 sum = 0;
310
311 if (s->aligned_buf != NULL) {
312
313 if (offset & 0x1ff) {
314 /* align offset on a 512 bytes boundary */
315
316 shift = offset & 0x1ff;
317 size = (shift + count + 0x1ff) & ~0x1ff;
318 if (size > ALIGNED_BUFFER_SIZE)
319 size = ALIGNED_BUFFER_SIZE;
320 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
321 if (ret < 0)
322 return ret;
323
324 size = 512 - shift;
325 if (size > count)
326 size = count;
327 memcpy(buf, s->aligned_buf + shift, size);
328
329 buf += size;
330 offset += size;
331 count -= size;
332 sum += size;
333
334 if (count == 0)
335 return sum;
336 }
337 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
338
339 /* read on aligned buffer */
340
341 while (count) {
342
343 size = (count + 0x1ff) & ~0x1ff;
344 if (size > ALIGNED_BUFFER_SIZE)
345 size = ALIGNED_BUFFER_SIZE;
346
347 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
348 if (ret < 0)
349 return ret;
350
351 size = ret;
352 if (size > count)
353 size = count;
354
355 memcpy(buf, s->aligned_buf, size);
356
357 buf += size;
358 offset += size;
359 count -= size;
360 sum += size;
361 }
362
363 return sum;
364 }
365 }
366
367 return raw_pread_aligned(bs, offset, buf, count) + sum;
368 }
369
370 static int raw_read(BlockDriverState *bs, int64_t sector_num,
371 uint8_t *buf, int nb_sectors)
372 {
373 int ret;
374
375 ret = raw_pread(bs, sector_num * 512, buf, nb_sectors * 512);
376 if (ret == (nb_sectors * 512))
377 ret = 0;
378 return ret;
379 }
380
381 /*
382 * offset and count are in bytes and possibly not aligned. For files opened
383 * with O_DIRECT, necessary alignments are ensured before calling
384 * raw_pwrite_aligned to do the actual write.
385 */
386 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
387 const uint8_t *buf, int count)
388 {
389 BDRVRawState *s = bs->opaque;
390 int size, ret, shift, sum;
391
392 sum = 0;
393
394 if (s->aligned_buf != NULL) {
395
396 if (offset & 0x1ff) {
397 /* align offset on a 512 bytes boundary */
398 shift = offset & 0x1ff;
399 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
400 if (ret < 0)
401 return ret;
402
403 size = 512 - shift;
404 if (size > count)
405 size = count;
406 memcpy(s->aligned_buf + shift, buf, size);
407
408 ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
409 if (ret < 0)
410 return ret;
411
412 buf += size;
413 offset += size;
414 count -= size;
415 sum += size;
416
417 if (count == 0)
418 return sum;
419 }
420 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
421
422 while ((size = (count & ~0x1ff)) != 0) {
423
424 if (size > ALIGNED_BUFFER_SIZE)
425 size = ALIGNED_BUFFER_SIZE;
426
427 memcpy(s->aligned_buf, buf, size);
428
429 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
430 if (ret < 0)
431 return ret;
432
433 buf += ret;
434 offset += ret;
435 count -= ret;
436 sum += ret;
437 }
438 /* here, count < 512 because (count & ~0x1ff) == 0 */
439 if (count) {
440 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
441 if (ret < 0)
442 return ret;
443 memcpy(s->aligned_buf, buf, count);
444
445 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
446 if (ret < 0)
447 return ret;
448 if (count < ret)
449 ret = count;
450
451 sum += ret;
452 }
453 return sum;
454 }
455 }
456 return raw_pwrite_aligned(bs, offset, buf, count) + sum;
457 }
458
459 static int raw_write(BlockDriverState *bs, int64_t sector_num,
460 const uint8_t *buf, int nb_sectors)
461 {
462 int ret;
463 ret = raw_pwrite(bs, sector_num * 512, buf, nb_sectors * 512);
464 if (ret == (nb_sectors * 512))
465 ret = 0;
466 return ret;
467 }
468
469 #ifdef CONFIG_AIO
470 /***********************************************************/
471 /* Unix AIO using POSIX AIO */
472
473 typedef struct RawAIOCB {
474 BlockDriverAIOCB common;
475 struct qemu_paiocb aiocb;
476 struct RawAIOCB *next;
477 int ret;
478 } RawAIOCB;
479
480 typedef struct PosixAioState
481 {
482 int rfd, wfd;
483 RawAIOCB *first_aio;
484 } PosixAioState;
485
486 static void posix_aio_read(void *opaque)
487 {
488 PosixAioState *s = opaque;
489 RawAIOCB *acb, **pacb;
490 int ret;
491 ssize_t len;
492
493 /* read all bytes from signal pipe */
494 for (;;) {
495 char bytes[16];
496
497 len = read(s->rfd, bytes, sizeof(bytes));
498 if (len == -1 && errno == EINTR)
499 continue; /* try again */
500 if (len == sizeof(bytes))
501 continue; /* more to read */
502 break;
503 }
504
505 for(;;) {
506 pacb = &s->first_aio;
507 for(;;) {
508 acb = *pacb;
509 if (!acb)
510 goto the_end;
511 ret = qemu_paio_error(&acb->aiocb);
512 if (ret == ECANCELED) {
513 /* remove the request */
514 *pacb = acb->next;
515 qemu_aio_release(acb);
516 } else if (ret != EINPROGRESS) {
517 /* end of aio */
518 if (ret == 0) {
519 ret = qemu_paio_return(&acb->aiocb);
520 if (ret == acb->aiocb.aio_nbytes)
521 ret = 0;
522 else
523 ret = -EINVAL;
524 } else {
525 ret = -ret;
526 }
527 /* remove the request */
528 *pacb = acb->next;
529 /* call the callback */
530 acb->common.cb(acb->common.opaque, ret);
531 qemu_aio_release(acb);
532 break;
533 } else {
534 pacb = &acb->next;
535 }
536 }
537 }
538 the_end: ;
539 }
540
541 static int posix_aio_flush(void *opaque)
542 {
543 PosixAioState *s = opaque;
544 return !!s->first_aio;
545 }
546
547 static PosixAioState *posix_aio_state;
548
549 static void aio_signal_handler(int signum)
550 {
551 if (posix_aio_state) {
552 char byte = 0;
553
554 write(posix_aio_state->wfd, &byte, sizeof(byte));
555 }
556
557 qemu_service_io();
558 }
559
560 static int posix_aio_init(void)
561 {
562 struct sigaction act;
563 PosixAioState *s;
564 int fds[2];
565 struct qemu_paioinit ai;
566
567 if (posix_aio_state)
568 return 0;
569
570 s = qemu_malloc(sizeof(PosixAioState));
571
572 sigfillset(&act.sa_mask);
573 act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
574 act.sa_handler = aio_signal_handler;
575 sigaction(SIGUSR2, &act, NULL);
576
577 s->first_aio = NULL;
578 if (pipe(fds) == -1) {
579 fprintf(stderr, "failed to create pipe\n");
580 return -errno;
581 }
582
583 s->rfd = fds[0];
584 s->wfd = fds[1];
585
586 fcntl(s->rfd, F_SETFL, O_NONBLOCK);
587 fcntl(s->wfd, F_SETFL, O_NONBLOCK);
588
589 qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
590
591 memset(&ai, 0, sizeof(ai));
592 ai.aio_threads = 64;
593 ai.aio_num = 64;
594 qemu_paio_init(&ai);
595
596 posix_aio_state = s;
597
598 return 0;
599 }
600
601 static RawAIOCB *raw_aio_setup(BlockDriverState *bs, int64_t sector_num,
602 QEMUIOVector *qiov, int nb_sectors,
603 BlockDriverCompletionFunc *cb, void *opaque)
604 {
605 BDRVRawState *s = bs->opaque;
606 RawAIOCB *acb;
607
608 if (fd_open(bs) < 0)
609 return NULL;
610
611 acb = qemu_aio_get(bs, cb, opaque);
612 if (!acb)
613 return NULL;
614 acb->aiocb.aio_fildes = s->fd;
615 acb->aiocb.ev_signo = SIGUSR2;
616 acb->aiocb.aio_iov = qiov->iov;
617 acb->aiocb.aio_niov = qiov->niov;
618 acb->aiocb.aio_nbytes = nb_sectors * 512;
619 acb->aiocb.aio_offset = sector_num * 512;
620 acb->aiocb.aio_flags = 0;
621
622 /*
623 * If O_DIRECT is used the buffer needs to be aligned on a sector
624 * boundary. Tell the low level code to ensure that in case it's
625 * not done yet.
626 */
627 if (s->aligned_buf)
628 acb->aiocb.aio_flags |= QEMU_AIO_SECTOR_ALIGNED;
629
630 acb->next = posix_aio_state->first_aio;
631 posix_aio_state->first_aio = acb;
632 return acb;
633 }
634
635 static void raw_aio_remove(RawAIOCB *acb)
636 {
637 RawAIOCB **pacb;
638
639 /* remove the callback from the queue */
640 pacb = &posix_aio_state->first_aio;
641 for(;;) {
642 if (*pacb == NULL) {
643 fprintf(stderr, "raw_aio_remove: aio request not found!\n");
644 break;
645 } else if (*pacb == acb) {
646 *pacb = acb->next;
647 qemu_aio_release(acb);
648 break;
649 }
650 pacb = &(*pacb)->next;
651 }
652 }
653
654 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
655 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
656 BlockDriverCompletionFunc *cb, void *opaque)
657 {
658 RawAIOCB *acb;
659
660 acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
661 if (!acb)
662 return NULL;
663 if (qemu_paio_read(&acb->aiocb) < 0) {
664 raw_aio_remove(acb);
665 return NULL;
666 }
667 return &acb->common;
668 }
669
670 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
671 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
672 BlockDriverCompletionFunc *cb, void *opaque)
673 {
674 RawAIOCB *acb;
675
676 acb = raw_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque);
677 if (!acb)
678 return NULL;
679 if (qemu_paio_write(&acb->aiocb) < 0) {
680 raw_aio_remove(acb);
681 return NULL;
682 }
683 return &acb->common;
684 }
685
686 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
687 {
688 int ret;
689 RawAIOCB *acb = (RawAIOCB *)blockacb;
690
691 ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
692 if (ret == QEMU_PAIO_NOTCANCELED) {
693 /* fail safe: if the aio could not be canceled, we wait for
694 it */
695 while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
696 }
697
698 raw_aio_remove(acb);
699 }
700 #else /* CONFIG_AIO */
701 static int posix_aio_init(void)
702 {
703 return 0;
704 }
705 #endif /* CONFIG_AIO */
706
707
708 static void raw_close(BlockDriverState *bs)
709 {
710 BDRVRawState *s = bs->opaque;
711 if (s->fd >= 0) {
712 close(s->fd);
713 s->fd = -1;
714 if (s->aligned_buf != NULL)
715 qemu_free(s->aligned_buf);
716 }
717 }
718
719 static int raw_truncate(BlockDriverState *bs, int64_t offset)
720 {
721 BDRVRawState *s = bs->opaque;
722 if (s->type != FTYPE_FILE)
723 return -ENOTSUP;
724 if (ftruncate(s->fd, offset) < 0)
725 return -errno;
726 return 0;
727 }
728
729 #ifdef __OpenBSD__
730 static int64_t raw_getlength(BlockDriverState *bs)
731 {
732 BDRVRawState *s = bs->opaque;
733 int fd = s->fd;
734 struct stat st;
735
736 if (fstat(fd, &st))
737 return -1;
738 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
739 struct disklabel dl;
740
741 if (ioctl(fd, DIOCGDINFO, &dl))
742 return -1;
743 return (uint64_t)dl.d_secsize *
744 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
745 } else
746 return st.st_size;
747 }
748 #else /* !__OpenBSD__ */
749 static int64_t raw_getlength(BlockDriverState *bs)
750 {
751 BDRVRawState *s = bs->opaque;
752 int fd = s->fd;
753 int64_t size;
754 #ifdef HOST_BSD
755 struct stat sb;
756 #ifdef __FreeBSD__
757 int reopened = 0;
758 #endif
759 #endif
760 #ifdef __sun__
761 struct dk_minfo minfo;
762 int rv;
763 #endif
764 int ret;
765
766 ret = fd_open(bs);
767 if (ret < 0)
768 return ret;
769
770 #ifdef HOST_BSD
771 #ifdef __FreeBSD__
772 again:
773 #endif
774 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
775 #ifdef DIOCGMEDIASIZE
776 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
777 #elif defined(DIOCGPART)
778 {
779 struct partinfo pi;
780 if (ioctl(fd, DIOCGPART, &pi) == 0)
781 size = pi.media_size;
782 else
783 size = 0;
784 }
785 if (size == 0)
786 #endif
787 #ifdef CONFIG_COCOA
788 size = LONG_LONG_MAX;
789 #else
790 size = lseek(fd, 0LL, SEEK_END);
791 #endif
792 #ifdef __FreeBSD__
793 switch(s->type) {
794 case FTYPE_CD:
795 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
796 if (size == 2048LL * (unsigned)-1)
797 size = 0;
798 /* XXX no disc? maybe we need to reopen... */
799 if (size <= 0 && !reopened && cd_open(bs) >= 0) {
800 reopened = 1;
801 goto again;
802 }
803 }
804 #endif
805 } else
806 #endif
807 #ifdef __sun__
808 /*
809 * use the DKIOCGMEDIAINFO ioctl to read the size.
810 */
811 rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
812 if ( rv != -1 ) {
813 size = minfo.dki_lbsize * minfo.dki_capacity;
814 } else /* there are reports that lseek on some devices
815 fails, but irc discussion said that contingency
816 on contingency was overkill */
817 #endif
818 {
819 size = lseek(fd, 0, SEEK_END);
820 }
821 return size;
822 }
823 #endif
824
825 static int raw_create(const char *filename, int64_t total_size,
826 const char *backing_file, int flags)
827 {
828 int fd;
829
830 if (flags || backing_file)
831 return -ENOTSUP;
832
833 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
834 0644);
835 if (fd < 0)
836 return -EIO;
837 ftruncate(fd, total_size * 512);
838 close(fd);
839 return 0;
840 }
841
842 static void raw_flush(BlockDriverState *bs)
843 {
844 BDRVRawState *s = bs->opaque;
845 fsync(s->fd);
846 }
847
848 BlockDriver bdrv_raw = {
849 .format_name = "raw",
850 .instance_size = sizeof(BDRVRawState),
851 .bdrv_probe = NULL, /* no probe for protocols */
852 .bdrv_open = raw_open,
853 .bdrv_read = raw_read,
854 .bdrv_write = raw_write,
855 .bdrv_close = raw_close,
856 .bdrv_create = raw_create,
857 .bdrv_flush = raw_flush,
858
859 #ifdef CONFIG_AIO
860 .bdrv_aio_readv = raw_aio_readv,
861 .bdrv_aio_writev = raw_aio_writev,
862 .bdrv_aio_cancel = raw_aio_cancel,
863 .aiocb_size = sizeof(RawAIOCB),
864 #endif
865
866 .bdrv_truncate = raw_truncate,
867 .bdrv_getlength = raw_getlength,
868 };
869
870 /***********************************************/
871 /* host device */
872
873 #ifdef CONFIG_COCOA
874 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
875 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
876
877 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
878 {
879 kern_return_t kernResult;
880 mach_port_t masterPort;
881 CFMutableDictionaryRef classesToMatch;
882
883 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
884 if ( KERN_SUCCESS != kernResult ) {
885 printf( "IOMasterPort returned %d\n", kernResult );
886 }
887
888 classesToMatch = IOServiceMatching( kIOCDMediaClass );
889 if ( classesToMatch == NULL ) {
890 printf( "IOServiceMatching returned a NULL dictionary.\n" );
891 } else {
892 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
893 }
894 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
895 if ( KERN_SUCCESS != kernResult )
896 {
897 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
898 }
899
900 return kernResult;
901 }
902
903 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
904 {
905 io_object_t nextMedia;
906 kern_return_t kernResult = KERN_FAILURE;
907 *bsdPath = '\0';
908 nextMedia = IOIteratorNext( mediaIterator );
909 if ( nextMedia )
910 {
911 CFTypeRef bsdPathAsCFString;
912 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
913 if ( bsdPathAsCFString ) {
914 size_t devPathLength;
915 strcpy( bsdPath, _PATH_DEV );
916 strcat( bsdPath, "r" );
917 devPathLength = strlen( bsdPath );
918 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
919 kernResult = KERN_SUCCESS;
920 }
921 CFRelease( bsdPathAsCFString );
922 }
923 IOObjectRelease( nextMedia );
924 }
925
926 return kernResult;
927 }
928
929 #endif
930
931 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
932 {
933 BDRVRawState *s = bs->opaque;
934 int fd, open_flags, ret;
935
936 posix_aio_init();
937
938 #ifdef CONFIG_COCOA
939 if (strstart(filename, "/dev/cdrom", NULL)) {
940 kern_return_t kernResult;
941 io_iterator_t mediaIterator;
942 char bsdPath[ MAXPATHLEN ];
943 int fd;
944
945 kernResult = FindEjectableCDMedia( &mediaIterator );
946 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
947
948 if ( bsdPath[ 0 ] != '\0' ) {
949 strcat(bsdPath,"s0");
950 /* some CDs don't have a partition 0 */
951 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
952 if (fd < 0) {
953 bsdPath[strlen(bsdPath)-1] = '1';
954 } else {
955 close(fd);
956 }
957 filename = bsdPath;
958 }
959
960 if ( mediaIterator )
961 IOObjectRelease( mediaIterator );
962 }
963 #endif
964 open_flags = O_BINARY;
965 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
966 open_flags |= O_RDWR;
967 } else {
968 open_flags |= O_RDONLY;
969 bs->read_only = 1;
970 }
971 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
972 * and O_DIRECT for no caching. */
973 if ((flags & BDRV_O_NOCACHE))
974 open_flags |= O_DIRECT;
975 else if (!(flags & BDRV_O_CACHE_WB))
976 open_flags |= O_DSYNC;
977
978 s->type = FTYPE_FILE;
979 #if defined(__linux__)
980 if (strstart(filename, "/dev/cd", NULL)) {
981 /* open will not fail even if no CD is inserted */
982 open_flags |= O_NONBLOCK;
983 s->type = FTYPE_CD;
984 } else if (strstart(filename, "/dev/fd", NULL)) {
985 s->type = FTYPE_FD;
986 s->fd_open_flags = open_flags;
987 /* open will not fail even if no floppy is inserted */
988 open_flags |= O_NONBLOCK;
989 #ifdef CONFIG_AIO
990 } else if (strstart(filename, "/dev/sg", NULL)) {
991 bs->sg = 1;
992 #endif
993 }
994 #endif
995 #if defined(__FreeBSD__)
996 if (strstart(filename, "/dev/cd", NULL) ||
997 strstart(filename, "/dev/acd", NULL)) {
998 s->type = FTYPE_CD;
999 s->cd_open_flags = open_flags;
1000 }
1001 #endif
1002 s->fd = -1;
1003 fd = open(filename, open_flags, 0644);
1004 if (fd < 0) {
1005 ret = -errno;
1006 if (ret == -EROFS)
1007 ret = -EACCES;
1008 return ret;
1009 }
1010 s->fd = fd;
1011 #if defined(__FreeBSD__)
1012 /* make sure the door isnt locked at this time */
1013 if (s->type == FTYPE_CD)
1014 ioctl (s->fd, CDIOCALLOW);
1015 #endif
1016 #if defined(__linux__)
1017 /* close fd so that we can reopen it as needed */
1018 if (s->type == FTYPE_FD) {
1019 close(s->fd);
1020 s->fd = -1;
1021 s->fd_media_changed = 1;
1022 }
1023 #endif
1024 return 0;
1025 }
1026
1027 #if defined(__linux__)
1028 /* Note: we do not have a reliable method to detect if the floppy is
1029 present. The current method is to try to open the floppy at every
1030 I/O and to keep it opened during a few hundreds of ms. */
1031 static int fd_open(BlockDriverState *bs)
1032 {
1033 BDRVRawState *s = bs->opaque;
1034 int last_media_present;
1035
1036 if (s->type != FTYPE_FD)
1037 return 0;
1038 last_media_present = (s->fd >= 0);
1039 if (s->fd >= 0 &&
1040 (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1041 close(s->fd);
1042 s->fd = -1;
1043 #ifdef DEBUG_FLOPPY
1044 printf("Floppy closed\n");
1045 #endif
1046 }
1047 if (s->fd < 0) {
1048 if (s->fd_got_error &&
1049 (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1050 #ifdef DEBUG_FLOPPY
1051 printf("No floppy (open delayed)\n");
1052 #endif
1053 return -EIO;
1054 }
1055 s->fd = open(bs->filename, s->fd_open_flags);
1056 if (s->fd < 0) {
1057 s->fd_error_time = qemu_get_clock(rt_clock);
1058 s->fd_got_error = 1;
1059 if (last_media_present)
1060 s->fd_media_changed = 1;
1061 #ifdef DEBUG_FLOPPY
1062 printf("No floppy\n");
1063 #endif
1064 return -EIO;
1065 }
1066 #ifdef DEBUG_FLOPPY
1067 printf("Floppy opened\n");
1068 #endif
1069 }
1070 if (!last_media_present)
1071 s->fd_media_changed = 1;
1072 s->fd_open_time = qemu_get_clock(rt_clock);
1073 s->fd_got_error = 0;
1074 return 0;
1075 }
1076
1077 static int raw_is_inserted(BlockDriverState *bs)
1078 {
1079 BDRVRawState *s = bs->opaque;
1080 int ret;
1081
1082 switch(s->type) {
1083 case FTYPE_CD:
1084 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1085 if (ret == CDS_DISC_OK)
1086 return 1;
1087 else
1088 return 0;
1089 break;
1090 case FTYPE_FD:
1091 ret = fd_open(bs);
1092 return (ret >= 0);
1093 default:
1094 return 1;
1095 }
1096 }
1097
1098 /* currently only used by fdc.c, but a CD version would be good too */
1099 static int raw_media_changed(BlockDriverState *bs)
1100 {
1101 BDRVRawState *s = bs->opaque;
1102
1103 switch(s->type) {
1104 case FTYPE_FD:
1105 {
1106 int ret;
1107 /* XXX: we do not have a true media changed indication. It
1108 does not work if the floppy is changed without trying
1109 to read it */
1110 fd_open(bs);
1111 ret = s->fd_media_changed;
1112 s->fd_media_changed = 0;
1113 #ifdef DEBUG_FLOPPY
1114 printf("Floppy changed=%d\n", ret);
1115 #endif
1116 return ret;
1117 }
1118 default:
1119 return -ENOTSUP;
1120 }
1121 }
1122
1123 static int raw_eject(BlockDriverState *bs, int eject_flag)
1124 {
1125 BDRVRawState *s = bs->opaque;
1126
1127 switch(s->type) {
1128 case FTYPE_CD:
1129 if (eject_flag) {
1130 if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1131 perror("CDROMEJECT");
1132 } else {
1133 if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1134 perror("CDROMEJECT");
1135 }
1136 break;
1137 case FTYPE_FD:
1138 {
1139 int fd;
1140 if (s->fd >= 0) {
1141 close(s->fd);
1142 s->fd = -1;
1143 }
1144 fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1145 if (fd >= 0) {
1146 if (ioctl(fd, FDEJECT, 0) < 0)
1147 perror("FDEJECT");
1148 close(fd);
1149 }
1150 }
1151 break;
1152 default:
1153 return -ENOTSUP;
1154 }
1155 return 0;
1156 }
1157
1158 static int raw_set_locked(BlockDriverState *bs, int locked)
1159 {
1160 BDRVRawState *s = bs->opaque;
1161
1162 switch(s->type) {
1163 case FTYPE_CD:
1164 if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1165 /* Note: an error can happen if the distribution automatically
1166 mounts the CD-ROM */
1167 // perror("CDROM_LOCKDOOR");
1168 }
1169 break;
1170 default:
1171 return -ENOTSUP;
1172 }
1173 return 0;
1174 }
1175
1176 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1177 {
1178 BDRVRawState *s = bs->opaque;
1179
1180 return ioctl(s->fd, req, buf);
1181 }
1182
1183 #ifdef CONFIG_AIO
1184 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1185 unsigned long int req, void *buf,
1186 BlockDriverCompletionFunc *cb, void *opaque)
1187 {
1188 BDRVRawState *s = bs->opaque;
1189 RawAIOCB *acb;
1190
1191 if (fd_open(bs) < 0)
1192 return NULL;
1193
1194 acb = qemu_aio_get(bs, cb, opaque);
1195 if (!acb)
1196 return NULL;
1197 acb->aiocb.aio_fildes = s->fd;
1198 acb->aiocb.ev_signo = SIGUSR2;
1199 acb->aiocb.aio_offset = 0;
1200 acb->aiocb.aio_flags = 0;
1201
1202 acb->next = posix_aio_state->first_aio;
1203 posix_aio_state->first_aio = acb;
1204
1205 acb->aiocb.aio_ioctl_buf = buf;
1206 acb->aiocb.aio_ioctl_cmd = req;
1207 if (qemu_paio_ioctl(&acb->aiocb) < 0) {
1208 raw_aio_remove(acb);
1209 return NULL;
1210 }
1211
1212 return &acb->common;
1213 }
1214 #endif
1215
1216 #elif defined(__FreeBSD__)
1217
1218 static int fd_open(BlockDriverState *bs)
1219 {
1220 BDRVRawState *s = bs->opaque;
1221
1222 /* this is just to ensure s->fd is sane (its called by io ops) */
1223 if (s->fd >= 0)
1224 return 0;
1225 return -EIO;
1226 }
1227
1228 static int cd_open(BlockDriverState *bs)
1229 {
1230 #if defined(__FreeBSD__)
1231 BDRVRawState *s = bs->opaque;
1232 int fd;
1233
1234 switch(s->type) {
1235 case FTYPE_CD:
1236 /* XXX force reread of possibly changed/newly loaded disc,
1237 * FreeBSD seems to not notice sometimes... */
1238 if (s->fd >= 0)
1239 close (s->fd);
1240 fd = open(bs->filename, s->cd_open_flags, 0644);
1241 if (fd < 0) {
1242 s->fd = -1;
1243 return -EIO;
1244 }
1245 s->fd = fd;
1246 /* make sure the door isnt locked at this time */
1247 ioctl (s->fd, CDIOCALLOW);
1248 }
1249 #endif
1250 return 0;
1251 }
1252
1253 static int raw_is_inserted(BlockDriverState *bs)
1254 {
1255 BDRVRawState *s = bs->opaque;
1256
1257 switch(s->type) {
1258 case FTYPE_CD:
1259 return (raw_getlength(bs) > 0);
1260 case FTYPE_FD:
1261 /* XXX handle this */
1262 /* FALLTHRU */
1263 default:
1264 return 1;
1265 }
1266 }
1267
1268 static int raw_media_changed(BlockDriverState *bs)
1269 {
1270 return -ENOTSUP;
1271 }
1272
1273 static int raw_eject(BlockDriverState *bs, int eject_flag)
1274 {
1275 BDRVRawState *s = bs->opaque;
1276
1277 switch(s->type) {
1278 case FTYPE_CD:
1279 if (s->fd < 0)
1280 return -ENOTSUP;
1281 (void) ioctl (s->fd, CDIOCALLOW);
1282 if (eject_flag) {
1283 if (ioctl (s->fd, CDIOCEJECT) < 0)
1284 perror("CDIOCEJECT");
1285 } else {
1286 if (ioctl (s->fd, CDIOCCLOSE) < 0)
1287 perror("CDIOCCLOSE");
1288 }
1289 if (cd_open(bs) < 0)
1290 return -ENOTSUP;
1291 break;
1292 case FTYPE_FD:
1293 /* XXX handle this */
1294 /* FALLTHRU */
1295 default:
1296 return -ENOTSUP;
1297 }
1298 return 0;
1299 }
1300
1301 static int raw_set_locked(BlockDriverState *bs, int locked)
1302 {
1303 BDRVRawState *s = bs->opaque;
1304
1305 switch(s->type) {
1306 case FTYPE_CD:
1307 if (s->fd < 0)
1308 return -ENOTSUP;
1309 if (ioctl (s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1310 /* Note: an error can happen if the distribution automatically
1311 mounts the CD-ROM */
1312 // perror("CDROM_LOCKDOOR");
1313 }
1314 break;
1315 default:
1316 return -ENOTSUP;
1317 }
1318 return 0;
1319 }
1320
1321 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1322 {
1323 return -ENOTSUP;
1324 }
1325 #else /* !linux && !FreeBSD */
1326
1327 static int fd_open(BlockDriverState *bs)
1328 {
1329 return 0;
1330 }
1331
1332 static int raw_is_inserted(BlockDriverState *bs)
1333 {
1334 return 1;
1335 }
1336
1337 static int raw_media_changed(BlockDriverState *bs)
1338 {
1339 return -ENOTSUP;
1340 }
1341
1342 static int raw_eject(BlockDriverState *bs, int eject_flag)
1343 {
1344 return -ENOTSUP;
1345 }
1346
1347 static int raw_set_locked(BlockDriverState *bs, int locked)
1348 {
1349 return -ENOTSUP;
1350 }
1351
1352 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1353 {
1354 return -ENOTSUP;
1355 }
1356
1357 static BlockDriverAIOCB *raw_aio_ioctl(BlockDriverState *bs,
1358 unsigned long int req, void *buf,
1359 BlockDriverCompletionFunc *cb, void *opaque)
1360 {
1361 return NULL;
1362 }
1363 #endif /* !linux && !FreeBSD */
1364
1365 #if defined(__linux__) || defined(__FreeBSD__)
1366 static int hdev_create(const char *filename, int64_t total_size,
1367 const char *backing_file, int flags)
1368 {
1369 int fd;
1370 int ret = 0;
1371 struct stat stat_buf;
1372
1373 if (flags || backing_file)
1374 return -ENOTSUP;
1375
1376 fd = open(filename, O_WRONLY | O_BINARY);
1377 if (fd < 0)
1378 return -EIO;
1379
1380 if (fstat(fd, &stat_buf) < 0)
1381 ret = -EIO;
1382 else if (!S_ISBLK(stat_buf.st_mode))
1383 ret = -EIO;
1384 else if (lseek(fd, 0, SEEK_END) < total_size * 512)
1385 ret = -ENOSPC;
1386
1387 close(fd);
1388 return ret;
1389 }
1390
1391 #else /* !(linux || freebsd) */
1392
1393 static int hdev_create(const char *filename, int64_t total_size,
1394 const char *backing_file, int flags)
1395 {
1396 return -ENOTSUP;
1397 }
1398 #endif
1399
1400 BlockDriver bdrv_host_device = {
1401 .format_name = "host_device",
1402 .instance_size = sizeof(BDRVRawState),
1403 .bdrv_open = hdev_open,
1404 .bdrv_close = raw_close,
1405 .bdrv_create = hdev_create,
1406 .bdrv_flush = raw_flush,
1407
1408 #ifdef CONFIG_AIO
1409 .bdrv_aio_readv = raw_aio_readv,
1410 .bdrv_aio_writev = raw_aio_writev,
1411 .bdrv_aio_cancel = raw_aio_cancel,
1412 .aiocb_size = sizeof(RawAIOCB),
1413 #endif
1414
1415 .bdrv_read = raw_read,
1416 .bdrv_write = raw_write,
1417 .bdrv_getlength = raw_getlength,
1418
1419 /* removable device support */
1420 .bdrv_is_inserted = raw_is_inserted,
1421 .bdrv_media_changed = raw_media_changed,
1422 .bdrv_eject = raw_eject,
1423 .bdrv_set_locked = raw_set_locked,
1424 /* generic scsi device */
1425 .bdrv_ioctl = raw_ioctl,
1426 #ifdef CONFIG_AIO
1427 .bdrv_aio_ioctl = raw_aio_ioctl,
1428 #endif
1429 };