Update for 0.10.4 release
[qemu/mdroth.git] / block-raw-posix.c
blob85ca70494676ccedf287e16956483df41937c150
1 /*
2 * Block driver for RAW files (posix)
4 * Copyright (c) 2006 Fabrice Bellard
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:
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
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.
24 #include "qemu-common.h"
25 #include "qemu-timer.h"
26 #include "qemu-char.h"
27 #include "block_int.h"
28 #include <assert.h>
29 #ifdef CONFIG_AIO
30 #include "posix-aio-compat.h"
31 #endif
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
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
60 #ifdef __OpenBSD__
61 #include <sys/ioctl.h>
62 #include <sys/disklabel.h>
63 #include <sys/dkio.h>
64 #endif
66 //#define DEBUG_FLOPPY
68 //#define DEBUG_BLOCK
69 #if defined(DEBUG_BLOCK)
70 #define DEBUG_BLOCK_PRINT(formatCstr, args...) do { if (qemu_log_enabled()) \
71 { qemu_log(formatCstr, ##args); qemu_log_flush(); } } while (0)
72 #else
73 #define DEBUG_BLOCK_PRINT(formatCstr, args...)
74 #endif
76 /* OS X does not have O_DSYNC */
77 #ifndef O_DSYNC
78 #define O_DSYNC O_SYNC
79 #endif
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
86 #define FTYPE_FILE 0
87 #define FTYPE_CD 1
88 #define FTYPE_FD 2
90 #define ALIGNED_BUFFER_SIZE (32 * 512)
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
96 typedef struct BDRVRawState {
97 int fd;
98 int type;
99 unsigned int lseek_err_cnt;
100 #if defined(__linux__)
101 /* linux floppy specific */
102 int fd_open_flags;
103 int64_t fd_open_time;
104 int64_t fd_error_time;
105 int fd_got_error;
106 int fd_media_changed;
107 #endif
108 uint8_t* aligned_buf;
109 } BDRVRawState;
111 static int posix_aio_init(void);
113 static int fd_open(BlockDriverState *bs);
115 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
117 BDRVRawState *s = bs->opaque;
118 int fd, open_flags, ret;
120 posix_aio_init();
122 s->lseek_err_cnt = 0;
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;
131 if (flags & BDRV_O_CREAT)
132 open_flags |= O_CREAT | O_TRUNC;
134 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
135 * and O_DIRECT for no caching. */
136 if ((flags & BDRV_O_NOCACHE))
137 open_flags |= O_DIRECT;
138 else if (!(flags & BDRV_O_CACHE_WB))
139 open_flags |= O_DSYNC;
141 s->type = FTYPE_FILE;
143 fd = open(filename, open_flags, 0644);
144 if (fd < 0) {
145 ret = -errno;
146 if (ret == -EROFS)
147 ret = -EACCES;
148 return ret;
150 s->fd = fd;
151 s->aligned_buf = NULL;
152 if ((flags & BDRV_O_NOCACHE)) {
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;
160 return 0;
163 /* XXX: use host sector size if necessary with:
164 #ifdef DIOCGSECTORSIZE
166 unsigned int sectorsize = 512;
167 if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
168 sectorsize > bufsize)
169 bufsize = sectorsize;
171 #endif
172 #ifdef CONFIG_COCOA
173 u_int32_t blockSize = 512;
174 if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
175 bufsize = blockSize;
177 #endif
181 * offset and count are in bytes, but must be multiples of 512 for files
182 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
184 * This function may be called without alignment if the caller ensures
185 * that O_DIRECT is not in effect.
187 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
188 uint8_t *buf, int count)
190 BDRVRawState *s = bs->opaque;
191 int ret;
193 ret = fd_open(bs);
194 if (ret < 0)
195 return ret;
197 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
198 ++(s->lseek_err_cnt);
199 if(s->lseek_err_cnt <= 10) {
200 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
201 "] lseek failed : %d = %s\n",
202 s->fd, bs->filename, offset, buf, count,
203 bs->total_sectors, errno, strerror(errno));
205 return -1;
207 s->lseek_err_cnt=0;
209 ret = read(s->fd, buf, count);
210 if (ret == count)
211 goto label__raw_read__success;
213 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
214 "] read failed %d : %d = %s\n",
215 s->fd, bs->filename, offset, buf, count,
216 bs->total_sectors, ret, errno, strerror(errno));
218 /* Try harder for CDrom. */
219 if (bs->type == BDRV_TYPE_CDROM) {
220 lseek(s->fd, offset, SEEK_SET);
221 ret = read(s->fd, buf, count);
222 if (ret == count)
223 goto label__raw_read__success;
224 lseek(s->fd, offset, SEEK_SET);
225 ret = read(s->fd, buf, count);
226 if (ret == count)
227 goto label__raw_read__success;
229 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
230 "] retry read failed %d : %d = %s\n",
231 s->fd, bs->filename, offset, buf, count,
232 bs->total_sectors, ret, errno, strerror(errno));
235 label__raw_read__success:
237 return ret;
241 * offset and count are in bytes, but must be multiples of 512 for files
242 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
244 * This function may be called without alignment if the caller ensures
245 * that O_DIRECT is not in effect.
247 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
248 const uint8_t *buf, int count)
250 BDRVRawState *s = bs->opaque;
251 int ret;
253 ret = fd_open(bs);
254 if (ret < 0)
255 return -errno;
257 if (offset >= 0 && lseek(s->fd, offset, SEEK_SET) == (off_t)-1) {
258 ++(s->lseek_err_cnt);
259 if(s->lseek_err_cnt) {
260 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%"
261 PRId64 "] lseek failed : %d = %s\n",
262 s->fd, bs->filename, offset, buf, count,
263 bs->total_sectors, errno, strerror(errno));
265 return -EIO;
267 s->lseek_err_cnt = 0;
269 ret = write(s->fd, buf, count);
270 if (ret == count)
271 goto label__raw_write__success;
273 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
274 "] write failed %d : %d = %s\n",
275 s->fd, bs->filename, offset, buf, count,
276 bs->total_sectors, ret, errno, strerror(errno));
278 label__raw_write__success:
280 return (ret < 0) ? -errno : ret;
285 * offset and count are in bytes and possibly not aligned. For files opened
286 * with O_DIRECT, necessary alignments are ensured before calling
287 * raw_pread_aligned to do the actual read.
289 static int raw_pread(BlockDriverState *bs, int64_t offset,
290 uint8_t *buf, int count)
292 BDRVRawState *s = bs->opaque;
293 int size, ret, shift, sum;
295 sum = 0;
297 if (s->aligned_buf != NULL) {
299 if (offset & 0x1ff) {
300 /* align offset on a 512 bytes boundary */
302 shift = offset & 0x1ff;
303 size = (shift + count + 0x1ff) & ~0x1ff;
304 if (size > ALIGNED_BUFFER_SIZE)
305 size = ALIGNED_BUFFER_SIZE;
306 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
307 if (ret < 0)
308 return ret;
310 size = 512 - shift;
311 if (size > count)
312 size = count;
313 memcpy(buf, s->aligned_buf + shift, size);
315 buf += size;
316 offset += size;
317 count -= size;
318 sum += size;
320 if (count == 0)
321 return sum;
323 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
325 /* read on aligned buffer */
327 while (count) {
329 size = (count + 0x1ff) & ~0x1ff;
330 if (size > ALIGNED_BUFFER_SIZE)
331 size = ALIGNED_BUFFER_SIZE;
333 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
334 if (ret < 0)
335 return ret;
337 size = ret;
338 if (size > count)
339 size = count;
341 memcpy(buf, s->aligned_buf, size);
343 buf += size;
344 offset += size;
345 count -= size;
346 sum += size;
349 return sum;
353 return raw_pread_aligned(bs, offset, buf, count) + sum;
357 * offset and count are in bytes and possibly not aligned. For files opened
358 * with O_DIRECT, necessary alignments are ensured before calling
359 * raw_pwrite_aligned to do the actual write.
361 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
362 const uint8_t *buf, int count)
364 BDRVRawState *s = bs->opaque;
365 int size, ret, shift, sum;
367 sum = 0;
369 if (s->aligned_buf != NULL) {
371 if (offset & 0x1ff) {
372 /* align offset on a 512 bytes boundary */
373 shift = offset & 0x1ff;
374 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, 512);
375 if (ret < 0)
376 return ret;
378 size = 512 - shift;
379 if (size > count)
380 size = count;
381 memcpy(s->aligned_buf + shift, buf, size);
383 ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf, 512);
384 if (ret < 0)
385 return ret;
387 buf += size;
388 offset += size;
389 count -= size;
390 sum += size;
392 if (count == 0)
393 return sum;
395 if (count & 0x1ff || (uintptr_t) buf & 0x1ff) {
397 while ((size = (count & ~0x1ff)) != 0) {
399 if (size > ALIGNED_BUFFER_SIZE)
400 size = ALIGNED_BUFFER_SIZE;
402 memcpy(s->aligned_buf, buf, size);
404 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
405 if (ret < 0)
406 return ret;
408 buf += ret;
409 offset += ret;
410 count -= ret;
411 sum += ret;
413 /* here, count < 512 because (count & ~0x1ff) == 0 */
414 if (count) {
415 ret = raw_pread_aligned(bs, offset, s->aligned_buf, 512);
416 if (ret < 0)
417 return ret;
418 memcpy(s->aligned_buf, buf, count);
420 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, 512);
421 if (ret < 0)
422 return ret;
423 if (count < ret)
424 ret = count;
426 sum += ret;
428 return sum;
431 return raw_pwrite_aligned(bs, offset, buf, count) + sum;
434 #ifdef CONFIG_AIO
435 /***********************************************************/
436 /* Unix AIO using POSIX AIO */
438 typedef struct RawAIOCB {
439 BlockDriverAIOCB common;
440 struct qemu_paiocb aiocb;
441 struct RawAIOCB *next;
442 int ret;
443 } RawAIOCB;
445 typedef struct PosixAioState
447 int rfd, wfd;
448 RawAIOCB *first_aio;
449 } PosixAioState;
451 static void posix_aio_read(void *opaque)
453 PosixAioState *s = opaque;
454 RawAIOCB *acb, **pacb;
455 int ret;
456 ssize_t len;
458 /* read all bytes from signal pipe */
459 for (;;) {
460 char bytes[16];
462 len = read(s->rfd, bytes, sizeof(bytes));
463 if (len == -1 && errno == EINTR)
464 continue; /* try again */
465 if (len == sizeof(bytes))
466 continue; /* more to read */
467 break;
470 for(;;) {
471 pacb = &s->first_aio;
472 for(;;) {
473 acb = *pacb;
474 if (!acb)
475 goto the_end;
476 ret = qemu_paio_error(&acb->aiocb);
477 if (ret == ECANCELED) {
478 /* remove the request */
479 *pacb = acb->next;
480 qemu_aio_release(acb);
481 } else if (ret != EINPROGRESS) {
482 /* end of aio */
483 if (ret == 0) {
484 ret = qemu_paio_return(&acb->aiocb);
485 if (ret == acb->aiocb.aio_nbytes)
486 ret = 0;
487 else
488 ret = -EINVAL;
489 } else {
490 ret = -ret;
492 /* remove the request */
493 *pacb = acb->next;
494 /* call the callback */
495 acb->common.cb(acb->common.opaque, ret);
496 qemu_aio_release(acb);
497 break;
498 } else {
499 pacb = &acb->next;
503 the_end: ;
506 static int posix_aio_flush(void *opaque)
508 PosixAioState *s = opaque;
509 return !!s->first_aio;
512 static PosixAioState *posix_aio_state;
514 static void aio_signal_handler(int signum)
516 if (posix_aio_state) {
517 char byte = 0;
519 write(posix_aio_state->wfd, &byte, sizeof(byte));
522 qemu_service_io();
525 static int posix_aio_init(void)
527 struct sigaction act;
528 PosixAioState *s;
529 int fds[2];
530 struct qemu_paioinit ai;
532 if (posix_aio_state)
533 return 0;
535 s = qemu_malloc(sizeof(PosixAioState));
537 sigfillset(&act.sa_mask);
538 act.sa_flags = 0; /* do not restart syscalls to interrupt select() */
539 act.sa_handler = aio_signal_handler;
540 sigaction(SIGUSR2, &act, NULL);
542 s->first_aio = NULL;
543 if (pipe(fds) == -1) {
544 fprintf(stderr, "failed to create pipe\n");
545 return -errno;
548 s->rfd = fds[0];
549 s->wfd = fds[1];
551 fcntl(s->rfd, F_SETFL, O_NONBLOCK);
552 fcntl(s->wfd, F_SETFL, O_NONBLOCK);
554 qemu_aio_set_fd_handler(s->rfd, posix_aio_read, NULL, posix_aio_flush, s);
556 memset(&ai, 0, sizeof(ai));
557 ai.aio_threads = 64;
558 ai.aio_num = 64;
559 qemu_paio_init(&ai);
561 posix_aio_state = s;
563 return 0;
566 static RawAIOCB *raw_aio_setup(BlockDriverState *bs,
567 int64_t sector_num, uint8_t *buf, int nb_sectors,
568 BlockDriverCompletionFunc *cb, void *opaque)
570 BDRVRawState *s = bs->opaque;
571 RawAIOCB *acb;
573 if (fd_open(bs) < 0)
574 return NULL;
576 acb = qemu_aio_get(bs, cb, opaque);
577 if (!acb)
578 return NULL;
579 acb->aiocb.aio_fildes = s->fd;
580 acb->aiocb.ev_signo = SIGUSR2;
581 acb->aiocb.aio_buf = buf;
582 if (nb_sectors < 0)
583 acb->aiocb.aio_nbytes = -nb_sectors;
584 else
585 acb->aiocb.aio_nbytes = nb_sectors * 512;
586 acb->aiocb.aio_offset = sector_num * 512;
587 acb->next = posix_aio_state->first_aio;
588 posix_aio_state->first_aio = acb;
589 return acb;
592 static void raw_aio_em_cb(void* opaque)
594 RawAIOCB *acb = opaque;
595 acb->common.cb(acb->common.opaque, acb->ret);
596 qemu_aio_release(acb);
599 static void raw_aio_remove(RawAIOCB *acb)
601 RawAIOCB **pacb;
603 /* remove the callback from the queue */
604 pacb = &posix_aio_state->first_aio;
605 for(;;) {
606 if (*pacb == NULL) {
607 fprintf(stderr, "raw_aio_remove: aio request not found!\n");
608 break;
609 } else if (*pacb == acb) {
610 *pacb = acb->next;
611 qemu_aio_release(acb);
612 break;
614 pacb = &(*pacb)->next;
618 static BlockDriverAIOCB *raw_aio_read(BlockDriverState *bs,
619 int64_t sector_num, uint8_t *buf, int nb_sectors,
620 BlockDriverCompletionFunc *cb, void *opaque)
622 RawAIOCB *acb;
625 * If O_DIRECT is used and the buffer is not aligned fall back
626 * to synchronous IO.
628 BDRVRawState *s = bs->opaque;
630 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
631 QEMUBH *bh;
632 acb = qemu_aio_get(bs, cb, opaque);
633 acb->ret = raw_pread(bs, 512 * sector_num, buf, 512 * nb_sectors);
634 bh = qemu_bh_new(raw_aio_em_cb, acb);
635 qemu_bh_schedule(bh);
636 return &acb->common;
639 acb = raw_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
640 if (!acb)
641 return NULL;
642 if (qemu_paio_read(&acb->aiocb) < 0) {
643 raw_aio_remove(acb);
644 return NULL;
646 return &acb->common;
649 static BlockDriverAIOCB *raw_aio_write(BlockDriverState *bs,
650 int64_t sector_num, const uint8_t *buf, int nb_sectors,
651 BlockDriverCompletionFunc *cb, void *opaque)
653 RawAIOCB *acb;
656 * If O_DIRECT is used and the buffer is not aligned fall back
657 * to synchronous IO.
659 BDRVRawState *s = bs->opaque;
661 if (unlikely(s->aligned_buf != NULL && ((uintptr_t) buf % 512))) {
662 QEMUBH *bh;
663 acb = qemu_aio_get(bs, cb, opaque);
664 acb->ret = raw_pwrite(bs, 512 * sector_num, buf, 512 * nb_sectors);
665 bh = qemu_bh_new(raw_aio_em_cb, acb);
666 qemu_bh_schedule(bh);
667 return &acb->common;
670 acb = raw_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
671 if (!acb)
672 return NULL;
673 if (qemu_paio_write(&acb->aiocb) < 0) {
674 raw_aio_remove(acb);
675 return NULL;
677 return &acb->common;
680 static void raw_aio_cancel(BlockDriverAIOCB *blockacb)
682 int ret;
683 RawAIOCB *acb = (RawAIOCB *)blockacb;
685 ret = qemu_paio_cancel(acb->aiocb.aio_fildes, &acb->aiocb);
686 if (ret == QEMU_PAIO_NOTCANCELED) {
687 /* fail safe: if the aio could not be canceled, we wait for
688 it */
689 while (qemu_paio_error(&acb->aiocb) == EINPROGRESS);
692 raw_aio_remove(acb);
694 #else /* CONFIG_AIO */
695 static int posix_aio_init(void)
697 return 0;
699 #endif /* CONFIG_AIO */
702 static void raw_close(BlockDriverState *bs)
704 BDRVRawState *s = bs->opaque;
705 if (s->fd >= 0) {
706 close(s->fd);
707 s->fd = -1;
708 if (s->aligned_buf != NULL)
709 qemu_free(s->aligned_buf);
713 static int raw_truncate(BlockDriverState *bs, int64_t offset)
715 BDRVRawState *s = bs->opaque;
716 if (s->type != FTYPE_FILE)
717 return -ENOTSUP;
718 if (ftruncate(s->fd, offset) < 0)
719 return -errno;
720 return 0;
723 #ifdef __OpenBSD__
724 static int64_t raw_getlength(BlockDriverState *bs)
726 BDRVRawState *s = bs->opaque;
727 int fd = s->fd;
728 struct stat st;
730 if (fstat(fd, &st))
731 return -1;
732 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
733 struct disklabel dl;
735 if (ioctl(fd, DIOCGDINFO, &dl))
736 return -1;
737 return (uint64_t)dl.d_secsize *
738 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
739 } else
740 return st.st_size;
742 #else /* !__OpenBSD__ */
743 static int64_t raw_getlength(BlockDriverState *bs)
745 BDRVRawState *s = bs->opaque;
746 int fd = s->fd;
747 int64_t size;
748 #ifdef _BSD
749 struct stat sb;
750 #endif
751 #ifdef __sun__
752 struct dk_minfo minfo;
753 int rv;
754 #endif
755 int ret;
757 ret = fd_open(bs);
758 if (ret < 0)
759 return ret;
761 #ifdef _BSD
762 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
763 #ifdef DIOCGMEDIASIZE
764 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
765 #endif
766 #ifdef CONFIG_COCOA
767 size = LONG_LONG_MAX;
768 #else
769 size = lseek(fd, 0LL, SEEK_END);
770 #endif
771 } else
772 #endif
773 #ifdef __sun__
775 * use the DKIOCGMEDIAINFO ioctl to read the size.
777 rv = ioctl ( fd, DKIOCGMEDIAINFO, &minfo );
778 if ( rv != -1 ) {
779 size = minfo.dki_lbsize * minfo.dki_capacity;
780 } else /* there are reports that lseek on some devices
781 fails, but irc discussion said that contingency
782 on contingency was overkill */
783 #endif
785 size = lseek(fd, 0, SEEK_END);
787 return size;
789 #endif
791 static int raw_create(const char *filename, int64_t total_size,
792 const char *backing_file, int flags)
794 int fd;
796 if (flags || backing_file)
797 return -ENOTSUP;
799 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
800 0644);
801 if (fd < 0)
802 return -EIO;
803 ftruncate(fd, total_size * 512);
804 close(fd);
805 return 0;
808 static void raw_flush(BlockDriverState *bs)
810 BDRVRawState *s = bs->opaque;
811 fsync(s->fd);
814 BlockDriver bdrv_raw = {
815 "raw",
816 sizeof(BDRVRawState),
817 NULL, /* no probe for protocols */
818 raw_open,
819 NULL,
820 NULL,
821 raw_close,
822 raw_create,
823 raw_flush,
825 #ifdef CONFIG_AIO
826 .bdrv_aio_read = raw_aio_read,
827 .bdrv_aio_write = raw_aio_write,
828 .bdrv_aio_cancel = raw_aio_cancel,
829 .aiocb_size = sizeof(RawAIOCB),
830 #endif
832 .bdrv_pread = raw_pread,
833 .bdrv_pwrite = raw_pwrite,
834 .bdrv_truncate = raw_truncate,
835 .bdrv_getlength = raw_getlength,
838 /***********************************************/
839 /* host device */
841 #ifdef CONFIG_COCOA
842 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
843 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
845 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
847 kern_return_t kernResult;
848 mach_port_t masterPort;
849 CFMutableDictionaryRef classesToMatch;
851 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
852 if ( KERN_SUCCESS != kernResult ) {
853 printf( "IOMasterPort returned %d\n", kernResult );
856 classesToMatch = IOServiceMatching( kIOCDMediaClass );
857 if ( classesToMatch == NULL ) {
858 printf( "IOServiceMatching returned a NULL dictionary.\n" );
859 } else {
860 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
862 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
863 if ( KERN_SUCCESS != kernResult )
865 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
868 return kernResult;
871 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
873 io_object_t nextMedia;
874 kern_return_t kernResult = KERN_FAILURE;
875 *bsdPath = '\0';
876 nextMedia = IOIteratorNext( mediaIterator );
877 if ( nextMedia )
879 CFTypeRef bsdPathAsCFString;
880 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
881 if ( bsdPathAsCFString ) {
882 size_t devPathLength;
883 strcpy( bsdPath, _PATH_DEV );
884 strcat( bsdPath, "r" );
885 devPathLength = strlen( bsdPath );
886 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
887 kernResult = KERN_SUCCESS;
889 CFRelease( bsdPathAsCFString );
891 IOObjectRelease( nextMedia );
894 return kernResult;
897 #endif
899 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
901 BDRVRawState *s = bs->opaque;
902 int fd, open_flags, ret;
904 posix_aio_init();
906 #ifdef CONFIG_COCOA
907 if (strstart(filename, "/dev/cdrom", NULL)) {
908 kern_return_t kernResult;
909 io_iterator_t mediaIterator;
910 char bsdPath[ MAXPATHLEN ];
911 int fd;
913 kernResult = FindEjectableCDMedia( &mediaIterator );
914 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
916 if ( bsdPath[ 0 ] != '\0' ) {
917 strcat(bsdPath,"s0");
918 /* some CDs don't have a partition 0 */
919 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
920 if (fd < 0) {
921 bsdPath[strlen(bsdPath)-1] = '1';
922 } else {
923 close(fd);
925 filename = bsdPath;
928 if ( mediaIterator )
929 IOObjectRelease( mediaIterator );
931 #endif
932 open_flags = O_BINARY;
933 if ((flags & BDRV_O_ACCESS) == O_RDWR) {
934 open_flags |= O_RDWR;
935 } else {
936 open_flags |= O_RDONLY;
937 bs->read_only = 1;
939 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
940 * and O_DIRECT for no caching. */
941 if ((flags & BDRV_O_NOCACHE))
942 open_flags |= O_DIRECT;
943 else if (!(flags & BDRV_O_CACHE_WB))
944 open_flags |= O_DSYNC;
946 s->type = FTYPE_FILE;
947 #if defined(__linux__)
948 if (strstart(filename, "/dev/cd", NULL)) {
949 /* open will not fail even if no CD is inserted */
950 open_flags |= O_NONBLOCK;
951 s->type = FTYPE_CD;
952 } else if (strstart(filename, "/dev/fd", NULL)) {
953 s->type = FTYPE_FD;
954 s->fd_open_flags = open_flags;
955 /* open will not fail even if no floppy is inserted */
956 open_flags |= O_NONBLOCK;
957 } else if (strstart(filename, "/dev/sg", NULL)) {
958 bs->sg = 1;
960 #endif
961 fd = open(filename, open_flags, 0644);
962 if (fd < 0) {
963 ret = -errno;
964 if (ret == -EROFS)
965 ret = -EACCES;
966 return ret;
968 s->fd = fd;
969 #if defined(__linux__)
970 /* close fd so that we can reopen it as needed */
971 if (s->type == FTYPE_FD) {
972 close(s->fd);
973 s->fd = -1;
974 s->fd_media_changed = 1;
976 #endif
977 return 0;
980 #if defined(__linux__)
981 /* Note: we do not have a reliable method to detect if the floppy is
982 present. The current method is to try to open the floppy at every
983 I/O and to keep it opened during a few hundreds of ms. */
984 static int fd_open(BlockDriverState *bs)
986 BDRVRawState *s = bs->opaque;
987 int last_media_present;
989 if (s->type != FTYPE_FD)
990 return 0;
991 last_media_present = (s->fd >= 0);
992 if (s->fd >= 0 &&
993 (qemu_get_clock(rt_clock) - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
994 close(s->fd);
995 s->fd = -1;
996 #ifdef DEBUG_FLOPPY
997 printf("Floppy closed\n");
998 #endif
1000 if (s->fd < 0) {
1001 if (s->fd_got_error &&
1002 (qemu_get_clock(rt_clock) - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1003 #ifdef DEBUG_FLOPPY
1004 printf("No floppy (open delayed)\n");
1005 #endif
1006 return -EIO;
1008 s->fd = open(bs->filename, s->fd_open_flags);
1009 if (s->fd < 0) {
1010 s->fd_error_time = qemu_get_clock(rt_clock);
1011 s->fd_got_error = 1;
1012 if (last_media_present)
1013 s->fd_media_changed = 1;
1014 #ifdef DEBUG_FLOPPY
1015 printf("No floppy\n");
1016 #endif
1017 return -EIO;
1019 #ifdef DEBUG_FLOPPY
1020 printf("Floppy opened\n");
1021 #endif
1023 if (!last_media_present)
1024 s->fd_media_changed = 1;
1025 s->fd_open_time = qemu_get_clock(rt_clock);
1026 s->fd_got_error = 0;
1027 return 0;
1030 static int raw_is_inserted(BlockDriverState *bs)
1032 BDRVRawState *s = bs->opaque;
1033 int ret;
1035 switch(s->type) {
1036 case FTYPE_CD:
1037 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1038 if (ret == CDS_DISC_OK)
1039 return 1;
1040 else
1041 return 0;
1042 break;
1043 case FTYPE_FD:
1044 ret = fd_open(bs);
1045 return (ret >= 0);
1046 default:
1047 return 1;
1051 /* currently only used by fdc.c, but a CD version would be good too */
1052 static int raw_media_changed(BlockDriverState *bs)
1054 BDRVRawState *s = bs->opaque;
1056 switch(s->type) {
1057 case FTYPE_FD:
1059 int ret;
1060 /* XXX: we do not have a true media changed indication. It
1061 does not work if the floppy is changed without trying
1062 to read it */
1063 fd_open(bs);
1064 ret = s->fd_media_changed;
1065 s->fd_media_changed = 0;
1066 #ifdef DEBUG_FLOPPY
1067 printf("Floppy changed=%d\n", ret);
1068 #endif
1069 return ret;
1071 default:
1072 return -ENOTSUP;
1076 static int raw_eject(BlockDriverState *bs, int eject_flag)
1078 BDRVRawState *s = bs->opaque;
1080 switch(s->type) {
1081 case FTYPE_CD:
1082 if (eject_flag) {
1083 if (ioctl (s->fd, CDROMEJECT, NULL) < 0)
1084 perror("CDROMEJECT");
1085 } else {
1086 if (ioctl (s->fd, CDROMCLOSETRAY, NULL) < 0)
1087 perror("CDROMEJECT");
1089 break;
1090 case FTYPE_FD:
1092 int fd;
1093 if (s->fd >= 0) {
1094 close(s->fd);
1095 s->fd = -1;
1097 fd = open(bs->filename, s->fd_open_flags | O_NONBLOCK);
1098 if (fd >= 0) {
1099 if (ioctl(fd, FDEJECT, 0) < 0)
1100 perror("FDEJECT");
1101 close(fd);
1104 break;
1105 default:
1106 return -ENOTSUP;
1108 return 0;
1111 static int raw_set_locked(BlockDriverState *bs, int locked)
1113 BDRVRawState *s = bs->opaque;
1115 switch(s->type) {
1116 case FTYPE_CD:
1117 if (ioctl (s->fd, CDROM_LOCKDOOR, locked) < 0) {
1118 /* Note: an error can happen if the distribution automatically
1119 mounts the CD-ROM */
1120 // perror("CDROM_LOCKDOOR");
1122 break;
1123 default:
1124 return -ENOTSUP;
1126 return 0;
1129 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1131 BDRVRawState *s = bs->opaque;
1133 return ioctl(s->fd, req, buf);
1135 #else
1137 static int fd_open(BlockDriverState *bs)
1139 return 0;
1142 static int raw_is_inserted(BlockDriverState *bs)
1144 return 1;
1147 static int raw_media_changed(BlockDriverState *bs)
1149 return -ENOTSUP;
1152 static int raw_eject(BlockDriverState *bs, int eject_flag)
1154 return -ENOTSUP;
1157 static int raw_set_locked(BlockDriverState *bs, int locked)
1159 return -ENOTSUP;
1162 static int raw_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1164 return -ENOTSUP;
1166 #endif /* !linux */
1168 BlockDriver bdrv_host_device = {
1169 "host_device",
1170 sizeof(BDRVRawState),
1171 NULL, /* no probe for protocols */
1172 hdev_open,
1173 NULL,
1174 NULL,
1175 raw_close,
1176 NULL,
1177 raw_flush,
1179 #ifdef CONFIG_AIO
1180 .bdrv_aio_read = raw_aio_read,
1181 .bdrv_aio_write = raw_aio_write,
1182 .bdrv_aio_cancel = raw_aio_cancel,
1183 .aiocb_size = sizeof(RawAIOCB),
1184 #endif
1186 .bdrv_pread = raw_pread,
1187 .bdrv_pwrite = raw_pwrite,
1188 .bdrv_getlength = raw_getlength,
1190 /* removable device support */
1191 .bdrv_is_inserted = raw_is_inserted,
1192 .bdrv_media_changed = raw_media_changed,
1193 .bdrv_eject = raw_eject,
1194 .bdrv_set_locked = raw_set_locked,
1195 /* generic scsi device */
1196 .bdrv_ioctl = raw_ioctl,