block/raw-posix: get right partition size
[qemu.git] / block / raw-posix.c
blobfd57b5bb5de5fb348c1bd77f284f3ee857890bcd
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 "qemu-log.h"
28 #include "block_int.h"
29 #include "module.h"
30 #include "block/raw-posix-aio.h"
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
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 <sys/param.h>
52 #include <linux/cdrom.h>
53 #include <linux/fd.h>
54 #endif
55 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
56 #include <signal.h>
57 #include <sys/disk.h>
58 #include <sys/cdio.h>
59 #endif
61 #ifdef __OpenBSD__
62 #include <sys/ioctl.h>
63 #include <sys/disklabel.h>
64 #include <sys/dkio.h>
65 #endif
67 #ifdef __NetBSD__
68 #include <sys/ioctl.h>
69 #include <sys/disklabel.h>
70 #include <sys/dkio.h>
71 #include <sys/disk.h>
72 #endif
74 #ifdef __DragonFly__
75 #include <sys/ioctl.h>
76 #include <sys/diskslice.h>
77 #endif
79 #ifdef CONFIG_XFS
80 #include <xfs/xfs.h>
81 #endif
83 //#define DEBUG_FLOPPY
85 //#define DEBUG_BLOCK
86 #if defined(DEBUG_BLOCK)
87 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
88 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
89 #else
90 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
91 #endif
93 /* OS X does not have O_DSYNC */
94 #ifndef O_DSYNC
95 #ifdef O_SYNC
96 #define O_DSYNC O_SYNC
97 #elif defined(O_FSYNC)
98 #define O_DSYNC O_FSYNC
99 #endif
100 #endif
102 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
103 #ifndef O_DIRECT
104 #define O_DIRECT O_DSYNC
105 #endif
107 #define FTYPE_FILE 0
108 #define FTYPE_CD 1
109 #define FTYPE_FD 2
111 /* if the FD is not accessed during that time (in ns), we try to
112 reopen it to see if the disk has been changed */
113 #define FD_OPEN_TIMEOUT (1000000000)
115 #define MAX_BLOCKSIZE 4096
117 typedef struct BDRVRawState {
118 int fd;
119 int type;
120 int open_flags;
121 #if defined(__linux__)
122 /* linux floppy specific */
123 int64_t fd_open_time;
124 int64_t fd_error_time;
125 int fd_got_error;
126 int fd_media_changed;
127 #endif
128 #ifdef CONFIG_LINUX_AIO
129 int use_aio;
130 void *aio_ctx;
131 #endif
132 uint8_t *aligned_buf;
133 unsigned aligned_buf_size;
134 #ifdef CONFIG_XFS
135 bool is_xfs : 1;
136 #endif
137 } BDRVRawState;
139 static int fd_open(BlockDriverState *bs);
140 static int64_t raw_getlength(BlockDriverState *bs);
142 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
143 static int cdrom_reopen(BlockDriverState *bs);
144 #endif
146 #if defined(__NetBSD__)
147 static int raw_normalize_devicepath(const char **filename)
149 static char namebuf[PATH_MAX];
150 const char *dp, *fname;
151 struct stat sb;
153 fname = *filename;
154 dp = strrchr(fname, '/');
155 if (lstat(fname, &sb) < 0) {
156 fprintf(stderr, "%s: stat failed: %s\n",
157 fname, strerror(errno));
158 return -errno;
161 if (!S_ISBLK(sb.st_mode)) {
162 return 0;
165 if (dp == NULL) {
166 snprintf(namebuf, PATH_MAX, "r%s", fname);
167 } else {
168 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
169 (int)(dp - fname), fname, dp + 1);
171 fprintf(stderr, "%s is a block device", fname);
172 *filename = namebuf;
173 fprintf(stderr, ", using %s\n", *filename);
175 return 0;
177 #else
178 static int raw_normalize_devicepath(const char **filename)
180 return 0;
182 #endif
184 static int raw_open_common(BlockDriverState *bs, const char *filename,
185 int bdrv_flags, int open_flags)
187 BDRVRawState *s = bs->opaque;
188 int fd, ret;
190 ret = raw_normalize_devicepath(&filename);
191 if (ret != 0) {
192 return ret;
195 s->open_flags = open_flags | O_BINARY;
196 s->open_flags &= ~O_ACCMODE;
197 if (bdrv_flags & BDRV_O_RDWR) {
198 s->open_flags |= O_RDWR;
199 } else {
200 s->open_flags |= O_RDONLY;
203 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
204 * and O_DIRECT for no caching. */
205 if ((bdrv_flags & BDRV_O_NOCACHE))
206 s->open_flags |= O_DIRECT;
207 if (!(bdrv_flags & BDRV_O_CACHE_WB))
208 s->open_flags |= O_DSYNC;
210 s->fd = -1;
211 fd = qemu_open(filename, s->open_flags, 0644);
212 if (fd < 0) {
213 ret = -errno;
214 if (ret == -EROFS)
215 ret = -EACCES;
216 return ret;
218 s->fd = fd;
219 s->aligned_buf = NULL;
221 if ((bdrv_flags & BDRV_O_NOCACHE)) {
223 * Allocate a buffer for read/modify/write cycles. Chose the size
224 * pessimistically as we don't know the block size yet.
226 s->aligned_buf_size = 32 * MAX_BLOCKSIZE;
227 s->aligned_buf = qemu_memalign(MAX_BLOCKSIZE, s->aligned_buf_size);
228 if (s->aligned_buf == NULL) {
229 goto out_close;
233 #ifdef CONFIG_LINUX_AIO
234 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
235 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
237 /* We're falling back to POSIX AIO in some cases */
238 paio_init();
240 s->aio_ctx = laio_init();
241 if (!s->aio_ctx) {
242 goto out_free_buf;
244 s->use_aio = 1;
245 } else
246 #endif
248 if (paio_init() < 0) {
249 goto out_free_buf;
251 #ifdef CONFIG_LINUX_AIO
252 s->use_aio = 0;
253 #endif
256 #ifdef CONFIG_XFS
257 if (platform_test_xfs_fd(s->fd)) {
258 s->is_xfs = 1;
260 #endif
262 return 0;
264 out_free_buf:
265 qemu_vfree(s->aligned_buf);
266 out_close:
267 close(fd);
268 return -errno;
271 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
273 BDRVRawState *s = bs->opaque;
275 s->type = FTYPE_FILE;
276 return raw_open_common(bs, filename, flags, 0);
279 /* XXX: use host sector size if necessary with:
280 #ifdef DIOCGSECTORSIZE
282 unsigned int sectorsize = 512;
283 if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
284 sectorsize > bufsize)
285 bufsize = sectorsize;
287 #endif
288 #ifdef CONFIG_COCOA
289 uint32_t blockSize = 512;
290 if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
291 bufsize = blockSize;
293 #endif
297 * offset and count are in bytes, but must be multiples of 512 for files
298 * opened with O_DIRECT. buf must be aligned to 512 bytes then.
300 * This function may be called without alignment if the caller ensures
301 * that O_DIRECT is not in effect.
303 static int raw_pread_aligned(BlockDriverState *bs, int64_t offset,
304 uint8_t *buf, int count)
306 BDRVRawState *s = bs->opaque;
307 int ret;
309 ret = fd_open(bs);
310 if (ret < 0)
311 return ret;
313 ret = pread(s->fd, buf, count, offset);
314 if (ret == count)
315 return ret;
317 /* Allow reads beyond the end (needed for pwrite) */
318 if ((ret == 0) && bs->growable) {
319 int64_t size = raw_getlength(bs);
320 if (offset >= size) {
321 memset(buf, 0, count);
322 return count;
326 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
327 "] read failed %d : %d = %s\n",
328 s->fd, bs->filename, offset, buf, count,
329 bs->total_sectors, ret, errno, strerror(errno));
331 /* Try harder for CDrom. */
332 if (s->type != FTYPE_FILE) {
333 ret = pread(s->fd, buf, count, offset);
334 if (ret == count)
335 return ret;
336 ret = pread(s->fd, buf, count, offset);
337 if (ret == count)
338 return ret;
340 DEBUG_BLOCK_PRINT("raw_pread(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
341 "] retry read failed %d : %d = %s\n",
342 s->fd, bs->filename, offset, buf, count,
343 bs->total_sectors, ret, errno, strerror(errno));
346 return (ret < 0) ? -errno : ret;
350 * offset and count are in bytes, but must be multiples of the sector size
351 * for files opened with O_DIRECT. buf must be aligned to sector size bytes
352 * then.
354 * This function may be called without alignment if the caller ensures
355 * that O_DIRECT is not in effect.
357 static int raw_pwrite_aligned(BlockDriverState *bs, int64_t offset,
358 const uint8_t *buf, int count)
360 BDRVRawState *s = bs->opaque;
361 int ret;
363 ret = fd_open(bs);
364 if (ret < 0)
365 return -errno;
367 ret = pwrite(s->fd, buf, count, offset);
368 if (ret == count)
369 return ret;
371 DEBUG_BLOCK_PRINT("raw_pwrite(%d:%s, %" PRId64 ", %p, %d) [%" PRId64
372 "] write failed %d : %d = %s\n",
373 s->fd, bs->filename, offset, buf, count,
374 bs->total_sectors, ret, errno, strerror(errno));
376 return (ret < 0) ? -errno : ret;
381 * offset and count are in bytes and possibly not aligned. For files opened
382 * with O_DIRECT, necessary alignments are ensured before calling
383 * raw_pread_aligned to do the actual read.
385 static int raw_pread(BlockDriverState *bs, int64_t offset,
386 uint8_t *buf, int count)
388 BDRVRawState *s = bs->opaque;
389 unsigned sector_mask = bs->buffer_alignment - 1;
390 int size, ret, shift, sum;
392 sum = 0;
394 if (s->aligned_buf != NULL) {
396 if (offset & sector_mask) {
397 /* align offset on a sector size bytes boundary */
399 shift = offset & sector_mask;
400 size = (shift + count + sector_mask) & ~sector_mask;
401 if (size > s->aligned_buf_size)
402 size = s->aligned_buf_size;
403 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf, size);
404 if (ret < 0)
405 return ret;
407 size = bs->buffer_alignment - shift;
408 if (size > count)
409 size = count;
410 memcpy(buf, s->aligned_buf + shift, size);
412 buf += size;
413 offset += size;
414 count -= size;
415 sum += size;
417 if (count == 0)
418 return sum;
420 if (count & sector_mask || (uintptr_t) buf & sector_mask) {
422 /* read on aligned buffer */
424 while (count) {
426 size = (count + sector_mask) & ~sector_mask;
427 if (size > s->aligned_buf_size)
428 size = s->aligned_buf_size;
430 ret = raw_pread_aligned(bs, offset, s->aligned_buf, size);
431 if (ret < 0) {
432 return ret;
433 } else if (ret == 0) {
434 fprintf(stderr, "raw_pread: read beyond end of file\n");
435 abort();
438 size = ret;
439 if (size > count)
440 size = count;
442 memcpy(buf, s->aligned_buf, size);
444 buf += size;
445 offset += size;
446 count -= size;
447 sum += size;
450 return sum;
454 return raw_pread_aligned(bs, offset, buf, count) + sum;
457 static int raw_read(BlockDriverState *bs, int64_t sector_num,
458 uint8_t *buf, int nb_sectors)
460 int ret;
462 ret = raw_pread(bs, sector_num * BDRV_SECTOR_SIZE, buf,
463 nb_sectors * BDRV_SECTOR_SIZE);
464 if (ret == (nb_sectors * BDRV_SECTOR_SIZE))
465 ret = 0;
466 return ret;
470 * offset and count are in bytes and possibly not aligned. For files opened
471 * with O_DIRECT, necessary alignments are ensured before calling
472 * raw_pwrite_aligned to do the actual write.
474 static int raw_pwrite(BlockDriverState *bs, int64_t offset,
475 const uint8_t *buf, int count)
477 BDRVRawState *s = bs->opaque;
478 unsigned sector_mask = bs->buffer_alignment - 1;
479 int size, ret, shift, sum;
481 sum = 0;
483 if (s->aligned_buf != NULL) {
485 if (offset & sector_mask) {
486 /* align offset on a sector size bytes boundary */
487 shift = offset & sector_mask;
488 ret = raw_pread_aligned(bs, offset - shift, s->aligned_buf,
489 bs->buffer_alignment);
490 if (ret < 0)
491 return ret;
493 size = bs->buffer_alignment - shift;
494 if (size > count)
495 size = count;
496 memcpy(s->aligned_buf + shift, buf, size);
498 ret = raw_pwrite_aligned(bs, offset - shift, s->aligned_buf,
499 bs->buffer_alignment);
500 if (ret < 0)
501 return ret;
503 buf += size;
504 offset += size;
505 count -= size;
506 sum += size;
508 if (count == 0)
509 return sum;
511 if (count & sector_mask || (uintptr_t) buf & sector_mask) {
513 while ((size = (count & ~sector_mask)) != 0) {
515 if (size > s->aligned_buf_size)
516 size = s->aligned_buf_size;
518 memcpy(s->aligned_buf, buf, size);
520 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf, size);
521 if (ret < 0)
522 return ret;
524 buf += ret;
525 offset += ret;
526 count -= ret;
527 sum += ret;
529 /* here, count < sector_size because (count & ~sector_mask) == 0 */
530 if (count) {
531 ret = raw_pread_aligned(bs, offset, s->aligned_buf,
532 bs->buffer_alignment);
533 if (ret < 0)
534 return ret;
535 memcpy(s->aligned_buf, buf, count);
537 ret = raw_pwrite_aligned(bs, offset, s->aligned_buf,
538 bs->buffer_alignment);
539 if (ret < 0)
540 return ret;
541 if (count < ret)
542 ret = count;
544 sum += ret;
546 return sum;
549 return raw_pwrite_aligned(bs, offset, buf, count) + sum;
552 static int raw_write(BlockDriverState *bs, int64_t sector_num,
553 const uint8_t *buf, int nb_sectors)
555 int ret;
556 ret = raw_pwrite(bs, sector_num * BDRV_SECTOR_SIZE, buf,
557 nb_sectors * BDRV_SECTOR_SIZE);
558 if (ret == (nb_sectors * BDRV_SECTOR_SIZE))
559 ret = 0;
560 return ret;
564 * Check if all memory in this vector is sector aligned.
566 static int qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov)
568 int i;
570 for (i = 0; i < qiov->niov; i++) {
571 if ((uintptr_t) qiov->iov[i].iov_base % bs->buffer_alignment) {
572 return 0;
576 return 1;
579 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
580 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
581 BlockDriverCompletionFunc *cb, void *opaque, int type)
583 BDRVRawState *s = bs->opaque;
585 if (fd_open(bs) < 0)
586 return NULL;
589 * If O_DIRECT is used the buffer needs to be aligned on a sector
590 * boundary. Check if this is the case or telll the low-level
591 * driver that it needs to copy the buffer.
593 if (s->aligned_buf) {
594 if (!qiov_is_aligned(bs, qiov)) {
595 type |= QEMU_AIO_MISALIGNED;
596 #ifdef CONFIG_LINUX_AIO
597 } else if (s->use_aio) {
598 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
599 nb_sectors, cb, opaque, type);
600 #endif
604 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
605 cb, opaque, type);
608 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
609 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
610 BlockDriverCompletionFunc *cb, void *opaque)
612 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
613 cb, opaque, QEMU_AIO_READ);
616 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
617 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
618 BlockDriverCompletionFunc *cb, void *opaque)
620 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
621 cb, opaque, QEMU_AIO_WRITE);
624 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
625 BlockDriverCompletionFunc *cb, void *opaque)
627 BDRVRawState *s = bs->opaque;
629 if (fd_open(bs) < 0)
630 return NULL;
632 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
635 static void raw_close(BlockDriverState *bs)
637 BDRVRawState *s = bs->opaque;
638 if (s->fd >= 0) {
639 close(s->fd);
640 s->fd = -1;
641 if (s->aligned_buf != NULL)
642 qemu_vfree(s->aligned_buf);
646 static int raw_truncate(BlockDriverState *bs, int64_t offset)
648 BDRVRawState *s = bs->opaque;
649 if (s->type != FTYPE_FILE)
650 return -ENOTSUP;
651 if (ftruncate(s->fd, offset) < 0)
652 return -errno;
653 return 0;
656 #ifdef __OpenBSD__
657 static int64_t raw_getlength(BlockDriverState *bs)
659 BDRVRawState *s = bs->opaque;
660 int fd = s->fd;
661 struct stat st;
663 if (fstat(fd, &st))
664 return -1;
665 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
666 struct disklabel dl;
668 if (ioctl(fd, DIOCGDINFO, &dl))
669 return -1;
670 return (uint64_t)dl.d_secsize *
671 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
672 } else
673 return st.st_size;
675 #elif defined(__NetBSD__)
676 static int64_t raw_getlength(BlockDriverState *bs)
678 BDRVRawState *s = bs->opaque;
679 int fd = s->fd;
680 struct stat st;
682 if (fstat(fd, &st))
683 return -1;
684 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
685 struct dkwedge_info dkw;
687 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
688 return dkw.dkw_size * 512;
689 } else {
690 struct disklabel dl;
692 if (ioctl(fd, DIOCGDINFO, &dl))
693 return -1;
694 return (uint64_t)dl.d_secsize *
695 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
697 } else
698 return st.st_size;
700 #elif defined(__sun__)
701 static int64_t raw_getlength(BlockDriverState *bs)
703 BDRVRawState *s = bs->opaque;
704 struct dk_minfo minfo;
705 int ret;
707 ret = fd_open(bs);
708 if (ret < 0) {
709 return ret;
713 * Use the DKIOCGMEDIAINFO ioctl to read the size.
715 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
716 if (ret != -1) {
717 return minfo.dki_lbsize * minfo.dki_capacity;
721 * There are reports that lseek on some devices fails, but
722 * irc discussion said that contingency on contingency was overkill.
724 return lseek(s->fd, 0, SEEK_END);
726 #elif defined(CONFIG_BSD)
727 static int64_t raw_getlength(BlockDriverState *bs)
729 BDRVRawState *s = bs->opaque;
730 int fd = s->fd;
731 int64_t size;
732 struct stat sb;
733 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
734 int reopened = 0;
735 #endif
736 int ret;
738 ret = fd_open(bs);
739 if (ret < 0)
740 return ret;
742 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
743 again:
744 #endif
745 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
746 #ifdef DIOCGMEDIASIZE
747 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
748 #elif defined(DIOCGPART)
750 struct partinfo pi;
751 if (ioctl(fd, DIOCGPART, &pi) == 0)
752 size = pi.media_size;
753 else
754 size = 0;
756 if (size == 0)
757 #endif
758 #ifdef CONFIG_COCOA
759 size = LONG_LONG_MAX;
760 #else
761 size = lseek(fd, 0LL, SEEK_END);
762 #endif
763 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
764 switch(s->type) {
765 case FTYPE_CD:
766 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
767 if (size == 2048LL * (unsigned)-1)
768 size = 0;
769 /* XXX no disc? maybe we need to reopen... */
770 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
771 reopened = 1;
772 goto again;
775 #endif
776 } else {
777 size = lseek(fd, 0, SEEK_END);
779 return size;
781 #else
782 static int64_t raw_getlength(BlockDriverState *bs)
784 BDRVRawState *s = bs->opaque;
785 int ret;
787 ret = fd_open(bs);
788 if (ret < 0) {
789 return ret;
792 return lseek(s->fd, 0, SEEK_END);
794 #endif
796 static int raw_create(const char *filename, QEMUOptionParameter *options)
798 int fd;
799 int result = 0;
800 int64_t total_size = 0;
802 /* Read out options */
803 while (options && options->name) {
804 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
805 total_size = options->value.n / BDRV_SECTOR_SIZE;
807 options++;
810 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
811 0644);
812 if (fd < 0) {
813 result = -errno;
814 } else {
815 if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
816 result = -errno;
818 if (close(fd) != 0) {
819 result = -errno;
822 return result;
825 static int raw_flush(BlockDriverState *bs)
827 BDRVRawState *s = bs->opaque;
828 return qemu_fdatasync(s->fd);
831 #ifdef CONFIG_XFS
832 static int xfs_discard(BDRVRawState *s, int64_t sector_num, int nb_sectors)
834 struct xfs_flock64 fl;
836 memset(&fl, 0, sizeof(fl));
837 fl.l_whence = SEEK_SET;
838 fl.l_start = sector_num << 9;
839 fl.l_len = (int64_t)nb_sectors << 9;
841 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
842 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
843 return -errno;
846 return 0;
848 #endif
850 static int raw_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors)
852 #ifdef CONFIG_XFS
853 BDRVRawState *s = bs->opaque;
855 if (s->is_xfs) {
856 return xfs_discard(s, sector_num, nb_sectors);
858 #endif
860 return 0;
863 static QEMUOptionParameter raw_create_options[] = {
865 .name = BLOCK_OPT_SIZE,
866 .type = OPT_SIZE,
867 .help = "Virtual disk size"
869 { NULL }
872 static BlockDriver bdrv_file = {
873 .format_name = "file",
874 .protocol_name = "file",
875 .instance_size = sizeof(BDRVRawState),
876 .bdrv_probe = NULL, /* no probe for protocols */
877 .bdrv_file_open = raw_open,
878 .bdrv_read = raw_read,
879 .bdrv_write = raw_write,
880 .bdrv_close = raw_close,
881 .bdrv_create = raw_create,
882 .bdrv_flush = raw_flush,
883 .bdrv_discard = raw_discard,
885 .bdrv_aio_readv = raw_aio_readv,
886 .bdrv_aio_writev = raw_aio_writev,
887 .bdrv_aio_flush = raw_aio_flush,
889 .bdrv_truncate = raw_truncate,
890 .bdrv_getlength = raw_getlength,
892 .create_options = raw_create_options,
895 /***********************************************/
896 /* host device */
898 #ifdef CONFIG_COCOA
899 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
900 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
902 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
904 kern_return_t kernResult;
905 mach_port_t masterPort;
906 CFMutableDictionaryRef classesToMatch;
908 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
909 if ( KERN_SUCCESS != kernResult ) {
910 printf( "IOMasterPort returned %d\n", kernResult );
913 classesToMatch = IOServiceMatching( kIOCDMediaClass );
914 if ( classesToMatch == NULL ) {
915 printf( "IOServiceMatching returned a NULL dictionary.\n" );
916 } else {
917 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
919 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
920 if ( KERN_SUCCESS != kernResult )
922 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
925 return kernResult;
928 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
930 io_object_t nextMedia;
931 kern_return_t kernResult = KERN_FAILURE;
932 *bsdPath = '\0';
933 nextMedia = IOIteratorNext( mediaIterator );
934 if ( nextMedia )
936 CFTypeRef bsdPathAsCFString;
937 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
938 if ( bsdPathAsCFString ) {
939 size_t devPathLength;
940 strcpy( bsdPath, _PATH_DEV );
941 strcat( bsdPath, "r" );
942 devPathLength = strlen( bsdPath );
943 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
944 kernResult = KERN_SUCCESS;
946 CFRelease( bsdPathAsCFString );
948 IOObjectRelease( nextMedia );
951 return kernResult;
954 #endif
956 static int hdev_probe_device(const char *filename)
958 struct stat st;
960 /* allow a dedicated CD-ROM driver to match with a higher priority */
961 if (strstart(filename, "/dev/cdrom", NULL))
962 return 50;
964 if (stat(filename, &st) >= 0 &&
965 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
966 return 100;
969 return 0;
972 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
974 BDRVRawState *s = bs->opaque;
976 #ifdef CONFIG_COCOA
977 if (strstart(filename, "/dev/cdrom", NULL)) {
978 kern_return_t kernResult;
979 io_iterator_t mediaIterator;
980 char bsdPath[ MAXPATHLEN ];
981 int fd;
983 kernResult = FindEjectableCDMedia( &mediaIterator );
984 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
986 if ( bsdPath[ 0 ] != '\0' ) {
987 strcat(bsdPath,"s0");
988 /* some CDs don't have a partition 0 */
989 fd = open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
990 if (fd < 0) {
991 bsdPath[strlen(bsdPath)-1] = '1';
992 } else {
993 close(fd);
995 filename = bsdPath;
998 if ( mediaIterator )
999 IOObjectRelease( mediaIterator );
1001 #endif
1003 s->type = FTYPE_FILE;
1004 #if defined(__linux__)
1006 char resolved_path[ MAXPATHLEN ], *temp;
1008 temp = realpath(filename, resolved_path);
1009 if (temp && strstart(temp, "/dev/sg", NULL)) {
1010 bs->sg = 1;
1013 #endif
1015 return raw_open_common(bs, filename, flags, 0);
1018 #if defined(__linux__)
1019 /* Note: we do not have a reliable method to detect if the floppy is
1020 present. The current method is to try to open the floppy at every
1021 I/O and to keep it opened during a few hundreds of ms. */
1022 static int fd_open(BlockDriverState *bs)
1024 BDRVRawState *s = bs->opaque;
1025 int last_media_present;
1027 if (s->type != FTYPE_FD)
1028 return 0;
1029 last_media_present = (s->fd >= 0);
1030 if (s->fd >= 0 &&
1031 (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1032 close(s->fd);
1033 s->fd = -1;
1034 #ifdef DEBUG_FLOPPY
1035 printf("Floppy closed\n");
1036 #endif
1038 if (s->fd < 0) {
1039 if (s->fd_got_error &&
1040 (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1041 #ifdef DEBUG_FLOPPY
1042 printf("No floppy (open delayed)\n");
1043 #endif
1044 return -EIO;
1046 s->fd = open(bs->filename, s->open_flags & ~O_NONBLOCK);
1047 if (s->fd < 0) {
1048 s->fd_error_time = get_clock();
1049 s->fd_got_error = 1;
1050 if (last_media_present)
1051 s->fd_media_changed = 1;
1052 #ifdef DEBUG_FLOPPY
1053 printf("No floppy\n");
1054 #endif
1055 return -EIO;
1057 #ifdef DEBUG_FLOPPY
1058 printf("Floppy opened\n");
1059 #endif
1061 if (!last_media_present)
1062 s->fd_media_changed = 1;
1063 s->fd_open_time = get_clock();
1064 s->fd_got_error = 0;
1065 return 0;
1068 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1070 BDRVRawState *s = bs->opaque;
1072 return ioctl(s->fd, req, buf);
1075 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1076 unsigned long int req, void *buf,
1077 BlockDriverCompletionFunc *cb, void *opaque)
1079 BDRVRawState *s = bs->opaque;
1081 if (fd_open(bs) < 0)
1082 return NULL;
1083 return paio_ioctl(bs, s->fd, req, buf, cb, opaque);
1086 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1087 static int fd_open(BlockDriverState *bs)
1089 BDRVRawState *s = bs->opaque;
1091 /* this is just to ensure s->fd is sane (its called by io ops) */
1092 if (s->fd >= 0)
1093 return 0;
1094 return -EIO;
1096 #else /* !linux && !FreeBSD */
1098 static int fd_open(BlockDriverState *bs)
1100 return 0;
1103 #endif /* !linux && !FreeBSD */
1105 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1107 int fd;
1108 int ret = 0;
1109 struct stat stat_buf;
1110 int64_t total_size = 0;
1112 /* Read out options */
1113 while (options && options->name) {
1114 if (!strcmp(options->name, "size")) {
1115 total_size = options->value.n / BDRV_SECTOR_SIZE;
1117 options++;
1120 fd = open(filename, O_WRONLY | O_BINARY);
1121 if (fd < 0)
1122 return -errno;
1124 if (fstat(fd, &stat_buf) < 0)
1125 ret = -errno;
1126 else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1127 ret = -ENODEV;
1128 else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE)
1129 ret = -ENOSPC;
1131 close(fd);
1132 return ret;
1135 static int hdev_has_zero_init(BlockDriverState *bs)
1137 return 0;
1140 static BlockDriver bdrv_host_device = {
1141 .format_name = "host_device",
1142 .protocol_name = "host_device",
1143 .instance_size = sizeof(BDRVRawState),
1144 .bdrv_probe_device = hdev_probe_device,
1145 .bdrv_file_open = hdev_open,
1146 .bdrv_close = raw_close,
1147 .bdrv_create = hdev_create,
1148 .create_options = raw_create_options,
1149 .bdrv_has_zero_init = hdev_has_zero_init,
1150 .bdrv_flush = raw_flush,
1152 .bdrv_aio_readv = raw_aio_readv,
1153 .bdrv_aio_writev = raw_aio_writev,
1154 .bdrv_aio_flush = raw_aio_flush,
1156 .bdrv_read = raw_read,
1157 .bdrv_write = raw_write,
1158 .bdrv_getlength = raw_getlength,
1160 /* generic scsi device */
1161 #ifdef __linux__
1162 .bdrv_ioctl = hdev_ioctl,
1163 .bdrv_aio_ioctl = hdev_aio_ioctl,
1164 #endif
1167 #ifdef __linux__
1168 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1170 BDRVRawState *s = bs->opaque;
1171 int ret;
1173 s->type = FTYPE_FD;
1175 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1176 ret = raw_open_common(bs, filename, flags, O_NONBLOCK);
1177 if (ret)
1178 return ret;
1180 /* close fd so that we can reopen it as needed */
1181 close(s->fd);
1182 s->fd = -1;
1183 s->fd_media_changed = 1;
1185 return 0;
1188 static int floppy_probe_device(const char *filename)
1190 int fd, ret;
1191 int prio = 0;
1192 struct floppy_struct fdparam;
1194 if (strstart(filename, "/dev/fd", NULL))
1195 prio = 50;
1197 fd = open(filename, O_RDONLY | O_NONBLOCK);
1198 if (fd < 0) {
1199 goto out;
1202 /* Attempt to detect via a floppy specific ioctl */
1203 ret = ioctl(fd, FDGETPRM, &fdparam);
1204 if (ret >= 0)
1205 prio = 100;
1207 close(fd);
1208 out:
1209 return prio;
1213 static int floppy_is_inserted(BlockDriverState *bs)
1215 return fd_open(bs) >= 0;
1218 static int floppy_media_changed(BlockDriverState *bs)
1220 BDRVRawState *s = bs->opaque;
1221 int ret;
1224 * XXX: we do not have a true media changed indication.
1225 * It does not work if the floppy is changed without trying to read it.
1227 fd_open(bs);
1228 ret = s->fd_media_changed;
1229 s->fd_media_changed = 0;
1230 #ifdef DEBUG_FLOPPY
1231 printf("Floppy changed=%d\n", ret);
1232 #endif
1233 return ret;
1236 static int floppy_eject(BlockDriverState *bs, int eject_flag)
1238 BDRVRawState *s = bs->opaque;
1239 int fd;
1241 if (s->fd >= 0) {
1242 close(s->fd);
1243 s->fd = -1;
1245 fd = open(bs->filename, s->open_flags | O_NONBLOCK);
1246 if (fd >= 0) {
1247 if (ioctl(fd, FDEJECT, 0) < 0)
1248 perror("FDEJECT");
1249 close(fd);
1252 return 0;
1255 static BlockDriver bdrv_host_floppy = {
1256 .format_name = "host_floppy",
1257 .protocol_name = "host_floppy",
1258 .instance_size = sizeof(BDRVRawState),
1259 .bdrv_probe_device = floppy_probe_device,
1260 .bdrv_file_open = floppy_open,
1261 .bdrv_close = raw_close,
1262 .bdrv_create = hdev_create,
1263 .create_options = raw_create_options,
1264 .bdrv_has_zero_init = hdev_has_zero_init,
1265 .bdrv_flush = raw_flush,
1267 .bdrv_aio_readv = raw_aio_readv,
1268 .bdrv_aio_writev = raw_aio_writev,
1269 .bdrv_aio_flush = raw_aio_flush,
1271 .bdrv_read = raw_read,
1272 .bdrv_write = raw_write,
1273 .bdrv_getlength = raw_getlength,
1275 /* removable device support */
1276 .bdrv_is_inserted = floppy_is_inserted,
1277 .bdrv_media_changed = floppy_media_changed,
1278 .bdrv_eject = floppy_eject,
1281 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1283 BDRVRawState *s = bs->opaque;
1285 s->type = FTYPE_CD;
1287 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1288 return raw_open_common(bs, filename, flags, O_NONBLOCK);
1291 static int cdrom_probe_device(const char *filename)
1293 int fd, ret;
1294 int prio = 0;
1296 fd = open(filename, O_RDONLY | O_NONBLOCK);
1297 if (fd < 0) {
1298 goto out;
1301 /* Attempt to detect via a CDROM specific ioctl */
1302 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1303 if (ret >= 0)
1304 prio = 100;
1306 close(fd);
1307 out:
1308 return prio;
1311 static int cdrom_is_inserted(BlockDriverState *bs)
1313 BDRVRawState *s = bs->opaque;
1314 int ret;
1316 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1317 if (ret == CDS_DISC_OK)
1318 return 1;
1319 return 0;
1322 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1324 BDRVRawState *s = bs->opaque;
1326 if (eject_flag) {
1327 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1328 perror("CDROMEJECT");
1329 } else {
1330 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1331 perror("CDROMEJECT");
1334 return 0;
1337 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1339 BDRVRawState *s = bs->opaque;
1341 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1343 * Note: an error can happen if the distribution automatically
1344 * mounts the CD-ROM
1346 /* perror("CDROM_LOCKDOOR"); */
1349 return 0;
1352 static BlockDriver bdrv_host_cdrom = {
1353 .format_name = "host_cdrom",
1354 .protocol_name = "host_cdrom",
1355 .instance_size = sizeof(BDRVRawState),
1356 .bdrv_probe_device = cdrom_probe_device,
1357 .bdrv_file_open = cdrom_open,
1358 .bdrv_close = raw_close,
1359 .bdrv_create = hdev_create,
1360 .create_options = raw_create_options,
1361 .bdrv_has_zero_init = hdev_has_zero_init,
1362 .bdrv_flush = raw_flush,
1364 .bdrv_aio_readv = raw_aio_readv,
1365 .bdrv_aio_writev = raw_aio_writev,
1366 .bdrv_aio_flush = raw_aio_flush,
1368 .bdrv_read = raw_read,
1369 .bdrv_write = raw_write,
1370 .bdrv_getlength = raw_getlength,
1372 /* removable device support */
1373 .bdrv_is_inserted = cdrom_is_inserted,
1374 .bdrv_eject = cdrom_eject,
1375 .bdrv_set_locked = cdrom_set_locked,
1377 /* generic scsi device */
1378 .bdrv_ioctl = hdev_ioctl,
1379 .bdrv_aio_ioctl = hdev_aio_ioctl,
1381 #endif /* __linux__ */
1383 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1384 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1386 BDRVRawState *s = bs->opaque;
1387 int ret;
1389 s->type = FTYPE_CD;
1391 ret = raw_open_common(bs, filename, flags, 0);
1392 if (ret)
1393 return ret;
1395 /* make sure the door isnt locked at this time */
1396 ioctl(s->fd, CDIOCALLOW);
1397 return 0;
1400 static int cdrom_probe_device(const char *filename)
1402 if (strstart(filename, "/dev/cd", NULL) ||
1403 strstart(filename, "/dev/acd", NULL))
1404 return 100;
1405 return 0;
1408 static int cdrom_reopen(BlockDriverState *bs)
1410 BDRVRawState *s = bs->opaque;
1411 int fd;
1414 * Force reread of possibly changed/newly loaded disc,
1415 * FreeBSD seems to not notice sometimes...
1417 if (s->fd >= 0)
1418 close(s->fd);
1419 fd = open(bs->filename, s->open_flags, 0644);
1420 if (fd < 0) {
1421 s->fd = -1;
1422 return -EIO;
1424 s->fd = fd;
1426 /* make sure the door isnt locked at this time */
1427 ioctl(s->fd, CDIOCALLOW);
1428 return 0;
1431 static int cdrom_is_inserted(BlockDriverState *bs)
1433 return raw_getlength(bs) > 0;
1436 static int cdrom_eject(BlockDriverState *bs, int eject_flag)
1438 BDRVRawState *s = bs->opaque;
1440 if (s->fd < 0)
1441 return -ENOTSUP;
1443 (void) ioctl(s->fd, CDIOCALLOW);
1445 if (eject_flag) {
1446 if (ioctl(s->fd, CDIOCEJECT) < 0)
1447 perror("CDIOCEJECT");
1448 } else {
1449 if (ioctl(s->fd, CDIOCCLOSE) < 0)
1450 perror("CDIOCCLOSE");
1453 if (cdrom_reopen(bs) < 0)
1454 return -ENOTSUP;
1455 return 0;
1458 static int cdrom_set_locked(BlockDriverState *bs, int locked)
1460 BDRVRawState *s = bs->opaque;
1462 if (s->fd < 0)
1463 return -ENOTSUP;
1464 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1466 * Note: an error can happen if the distribution automatically
1467 * mounts the CD-ROM
1469 /* perror("CDROM_LOCKDOOR"); */
1472 return 0;
1475 static BlockDriver bdrv_host_cdrom = {
1476 .format_name = "host_cdrom",
1477 .protocol_name = "host_cdrom",
1478 .instance_size = sizeof(BDRVRawState),
1479 .bdrv_probe_device = cdrom_probe_device,
1480 .bdrv_file_open = cdrom_open,
1481 .bdrv_close = raw_close,
1482 .bdrv_create = hdev_create,
1483 .create_options = raw_create_options,
1484 .bdrv_has_zero_init = hdev_has_zero_init,
1485 .bdrv_flush = raw_flush,
1487 .bdrv_aio_readv = raw_aio_readv,
1488 .bdrv_aio_writev = raw_aio_writev,
1489 .bdrv_aio_flush = raw_aio_flush,
1491 .bdrv_read = raw_read,
1492 .bdrv_write = raw_write,
1493 .bdrv_getlength = raw_getlength,
1495 /* removable device support */
1496 .bdrv_is_inserted = cdrom_is_inserted,
1497 .bdrv_eject = cdrom_eject,
1498 .bdrv_set_locked = cdrom_set_locked,
1500 #endif /* __FreeBSD__ */
1502 static void bdrv_file_init(void)
1505 * Register all the drivers. Note that order is important, the driver
1506 * registered last will get probed first.
1508 bdrv_register(&bdrv_file);
1509 bdrv_register(&bdrv_host_device);
1510 #ifdef __linux__
1511 bdrv_register(&bdrv_host_floppy);
1512 bdrv_register(&bdrv_host_cdrom);
1513 #endif
1514 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1515 bdrv_register(&bdrv_host_cdrom);
1516 #endif
1519 block_init(bdrv_file_init);