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
24 #include "qemu-common.h"
25 #include "qemu/timer.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
30 #include "block/thread-pool.h"
34 #if defined(__APPLE__) && (__MACH__)
36 #include <sys/param.h>
37 #include <IOKit/IOKitLib.h>
38 #include <IOKit/IOBSD.h>
39 #include <IOKit/storage/IOMediaBSDClient.h>
40 #include <IOKit/storage/IOMedia.h>
41 #include <IOKit/storage/IOCDMedia.h>
42 //#include <IOKit/storage/IOCDTypes.h>
43 #include <CoreFoundation/CoreFoundation.h>
47 #define _POSIX_PTHREAD_SEMANTICS 1
51 #include <sys/types.h>
53 #include <sys/ioctl.h>
54 #include <sys/param.h>
55 #include <linux/cdrom.h>
60 #include <linux/fiemap.h>
62 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
63 #include <linux/falloc.h>
65 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
71 #include <sys/ioctl.h>
72 #include <sys/disklabel.h>
77 #include <sys/ioctl.h>
78 #include <sys/disklabel.h>
84 #include <sys/ioctl.h>
85 #include <sys/diskslice.h>
92 //#define DEBUG_FLOPPY
95 #if defined(DEBUG_BLOCK)
96 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
97 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
99 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
102 /* OS X does not have O_DSYNC */
105 #define O_DSYNC O_SYNC
106 #elif defined(O_FSYNC)
107 #define O_DSYNC O_FSYNC
111 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
113 #define O_DIRECT O_DSYNC
120 /* if the FD is not accessed during that time (in ns), we try to
121 reopen it to see if the disk has been changed */
122 #define FD_OPEN_TIMEOUT (1000000000)
124 #define MAX_BLOCKSIZE 4096
126 typedef struct BDRVRawState {
132 #if defined(__linux__)
133 /* linux floppy specific */
134 int64_t fd_open_time;
135 int64_t fd_error_time;
137 int fd_media_changed;
139 #ifdef CONFIG_LINUX_AIO
147 bool has_write_zeroes:1;
148 bool discard_zeroes:1;
151 typedef struct BDRVRawReopenState {
154 #ifdef CONFIG_LINUX_AIO
157 } BDRVRawReopenState;
159 static int fd_open(BlockDriverState *bs);
160 static int64_t raw_getlength(BlockDriverState *bs);
162 typedef struct RawPosixAIOData {
163 BlockDriverState *bs;
166 struct iovec *aio_iov;
171 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
176 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
177 static int cdrom_reopen(BlockDriverState *bs);
180 #if defined(__NetBSD__)
181 static int raw_normalize_devicepath(const char **filename)
183 static char namebuf[PATH_MAX];
184 const char *dp, *fname;
188 dp = strrchr(fname, '/');
189 if (lstat(fname, &sb) < 0) {
190 fprintf(stderr, "%s: stat failed: %s\n",
191 fname, strerror(errno));
195 if (!S_ISBLK(sb.st_mode)) {
200 snprintf(namebuf, PATH_MAX, "r%s", fname);
202 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
203 (int)(dp - fname), fname, dp + 1);
205 fprintf(stderr, "%s is a block device", fname);
207 fprintf(stderr, ", using %s\n", *filename);
212 static int raw_normalize_devicepath(const char **filename)
218 static void raw_probe_alignment(BlockDriverState *bs)
220 BDRVRawState *s = bs->opaque;
222 unsigned int sector_size;
224 /* For /dev/sg devices the alignment is not really used.
225 With buffered I/O, we don't have any restrictions. */
226 if (bs->sg || !(s->open_flags & O_DIRECT)) {
227 bs->request_alignment = 1;
232 /* Try a few ioctls to get the right size */
233 bs->request_alignment = 0;
237 if (ioctl(s->fd, BLKSSZGET, §or_size) >= 0) {
238 bs->request_alignment = sector_size;
241 #ifdef DKIOCGETBLOCKSIZE
242 if (ioctl(s->fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) {
243 bs->request_alignment = sector_size;
246 #ifdef DIOCGSECTORSIZE
247 if (ioctl(s->fd, DIOCGSECTORSIZE, §or_size) >= 0) {
248 bs->request_alignment = sector_size;
254 if (xfsctl(NULL, s->fd, XFS_IOC_DIOINFO, &da) >= 0) {
255 bs->request_alignment = da.d_miniosz;
256 /* The kernel returns wrong information for d_mem */
257 /* s->buf_align = da.d_mem; */
262 /* If we could not get the sizes so far, we can only guess them */
265 buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE);
266 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
267 if (pread(s->fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) {
268 s->buf_align = align;
275 if (!bs->request_alignment) {
277 buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE);
278 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
279 if (pread(s->fd, buf, align, 0) >= 0) {
280 bs->request_alignment = align;
288 static void raw_parse_flags(int bdrv_flags, int *open_flags)
290 assert(open_flags != NULL);
292 *open_flags |= O_BINARY;
293 *open_flags &= ~O_ACCMODE;
294 if (bdrv_flags & BDRV_O_RDWR) {
295 *open_flags |= O_RDWR;
297 *open_flags |= O_RDONLY;
300 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
301 * and O_DIRECT for no caching. */
302 if ((bdrv_flags & BDRV_O_NOCACHE)) {
303 *open_flags |= O_DIRECT;
307 #ifdef CONFIG_LINUX_AIO
308 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
311 assert(aio_ctx != NULL);
312 assert(use_aio != NULL);
314 * Currently Linux do AIO only for files opened with O_DIRECT
315 * specified so check NOCACHE flag too
317 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
318 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
320 /* if non-NULL, laio_init() has already been run */
321 if (*aio_ctx == NULL) {
322 *aio_ctx = laio_init();
339 static QemuOptsList raw_runtime_opts = {
341 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
345 .type = QEMU_OPT_STRING,
346 .help = "File name of the image",
348 { /* end of list */ }
352 static int raw_open_common(BlockDriverState *bs, QDict *options,
353 int bdrv_flags, int open_flags, Error **errp)
355 BDRVRawState *s = bs->opaque;
357 Error *local_err = NULL;
358 const char *filename;
362 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
363 qemu_opts_absorb_qdict(opts, options, &local_err);
365 error_propagate(errp, local_err);
370 filename = qemu_opt_get(opts, "filename");
372 ret = raw_normalize_devicepath(&filename);
374 error_setg_errno(errp, -ret, "Could not normalize device path");
378 s->open_flags = open_flags;
379 raw_parse_flags(bdrv_flags, &s->open_flags);
382 fd = qemu_open(filename, s->open_flags, 0644);
392 #ifdef CONFIG_LINUX_AIO
393 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
396 error_setg_errno(errp, -ret, "Could not set AIO state");
401 s->has_discard = true;
402 s->has_write_zeroes = true;
404 if (fstat(s->fd, &st) < 0) {
405 error_setg_errno(errp, errno, "Could not stat file");
408 if (S_ISREG(st.st_mode)) {
409 s->discard_zeroes = true;
411 if (S_ISBLK(st.st_mode)) {
412 #ifdef BLKDISCARDZEROES
414 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
415 s->discard_zeroes = true;
419 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
420 * not rely on the contents of discarded blocks unless using O_DIRECT.
421 * Same for BLKZEROOUT.
423 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
424 s->discard_zeroes = false;
425 s->has_write_zeroes = false;
431 if (platform_test_xfs_fd(s->fd)) {
442 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
445 BDRVRawState *s = bs->opaque;
446 Error *local_err = NULL;
449 s->type = FTYPE_FILE;
450 ret = raw_open_common(bs, options, flags, 0, &local_err);
452 error_propagate(errp, local_err);
457 static int raw_reopen_prepare(BDRVReopenState *state,
458 BlockReopenQueue *queue, Error **errp)
461 BDRVRawReopenState *raw_s;
464 assert(state != NULL);
465 assert(state->bs != NULL);
467 s = state->bs->opaque;
469 state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
470 raw_s = state->opaque;
472 #ifdef CONFIG_LINUX_AIO
473 raw_s->use_aio = s->use_aio;
475 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
476 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
477 * won't override aio_ctx if aio_ctx is non-NULL */
478 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
479 error_setg(errp, "Could not set AIO state");
484 if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
485 raw_s->open_flags |= O_NONBLOCK;
488 raw_parse_flags(state->flags, &raw_s->open_flags);
492 int fcntl_flags = O_APPEND | O_NONBLOCK;
494 fcntl_flags |= O_NOATIME;
498 /* Not all operating systems have O_ASYNC, and those that don't
499 * will not let us track the state into raw_s->open_flags (typically
500 * you achieve the same effect with an ioctl, for example I_SETSIG
501 * on Solaris). But we do not use O_ASYNC, so that's fine.
503 assert((s->open_flags & O_ASYNC) == 0);
506 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
507 /* dup the original fd */
508 /* TODO: use qemu fcntl wrapper */
509 #ifdef F_DUPFD_CLOEXEC
510 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
512 raw_s->fd = dup(s->fd);
513 if (raw_s->fd != -1) {
514 qemu_set_cloexec(raw_s->fd);
517 if (raw_s->fd >= 0) {
518 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
520 qemu_close(raw_s->fd);
526 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
527 if (raw_s->fd == -1) {
528 assert(!(raw_s->open_flags & O_CREAT));
529 raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
530 if (raw_s->fd == -1) {
531 error_setg_errno(errp, errno, "Could not reopen file");
538 static void raw_reopen_commit(BDRVReopenState *state)
540 BDRVRawReopenState *raw_s = state->opaque;
541 BDRVRawState *s = state->bs->opaque;
543 s->open_flags = raw_s->open_flags;
547 #ifdef CONFIG_LINUX_AIO
548 s->use_aio = raw_s->use_aio;
551 g_free(state->opaque);
552 state->opaque = NULL;
556 static void raw_reopen_abort(BDRVReopenState *state)
558 BDRVRawReopenState *raw_s = state->opaque;
560 /* nothing to do if NULL, we didn't get far enough */
565 if (raw_s->fd >= 0) {
566 qemu_close(raw_s->fd);
569 g_free(state->opaque);
570 state->opaque = NULL;
573 static int raw_refresh_limits(BlockDriverState *bs)
575 BDRVRawState *s = bs->opaque;
577 raw_probe_alignment(bs);
578 bs->bl.opt_mem_alignment = s->buf_align;
583 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
587 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
595 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
599 ret = qemu_fdatasync(aiocb->aio_fildes);
608 static bool preadv_present = true;
611 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
613 return preadv(fd, iov, nr_iov, offset);
617 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
619 return pwritev(fd, iov, nr_iov, offset);
624 static bool preadv_present = false;
627 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
633 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
640 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
645 if (aiocb->aio_type & QEMU_AIO_WRITE)
646 len = qemu_pwritev(aiocb->aio_fildes,
651 len = qemu_preadv(aiocb->aio_fildes,
655 } while (len == -1 && errno == EINTR);
664 * Read/writes the data to/from a given linear buffer.
666 * Returns the number of bytes handles or -errno in case of an error. Short
667 * reads are only returned if the end of the file is reached.
669 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
674 while (offset < aiocb->aio_nbytes) {
675 if (aiocb->aio_type & QEMU_AIO_WRITE) {
676 len = pwrite(aiocb->aio_fildes,
677 (const char *)buf + offset,
678 aiocb->aio_nbytes - offset,
679 aiocb->aio_offset + offset);
681 len = pread(aiocb->aio_fildes,
683 aiocb->aio_nbytes - offset,
684 aiocb->aio_offset + offset);
686 if (len == -1 && errno == EINTR) {
688 } else if (len == -1) {
691 } else if (len == 0) {
700 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
705 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
707 * If there is just a single buffer, and it is properly aligned
708 * we can just use plain pread/pwrite without any problems.
710 if (aiocb->aio_niov == 1) {
711 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
714 * We have more than one iovec, and all are properly aligned.
716 * Try preadv/pwritev first and fall back to linearizing the
717 * buffer if it's not supported.
719 if (preadv_present) {
720 nbytes = handle_aiocb_rw_vector(aiocb);
721 if (nbytes == aiocb->aio_nbytes ||
722 (nbytes < 0 && nbytes != -ENOSYS)) {
725 preadv_present = false;
729 * XXX(hch): short read/write. no easy way to handle the reminder
730 * using these interfaces. For now retry using plain
736 * Ok, we have to do it the hard way, copy all segments into
737 * a single aligned buffer.
739 buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
740 if (aiocb->aio_type & QEMU_AIO_WRITE) {
744 for (i = 0; i < aiocb->aio_niov; ++i) {
745 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
746 p += aiocb->aio_iov[i].iov_len;
750 nbytes = handle_aiocb_rw_linear(aiocb, buf);
751 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
753 size_t count = aiocb->aio_nbytes, copy;
756 for (i = 0; i < aiocb->aio_niov && count; ++i) {
758 if (copy > aiocb->aio_iov[i].iov_len) {
759 copy = aiocb->aio_iov[i].iov_len;
761 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
772 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
774 struct xfs_flock64 fl;
776 memset(&fl, 0, sizeof(fl));
777 fl.l_whence = SEEK_SET;
781 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
782 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno));
789 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
791 struct xfs_flock64 fl;
793 memset(&fl, 0, sizeof(fl));
794 fl.l_whence = SEEK_SET;
798 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
799 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
807 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
809 int ret = -EOPNOTSUPP;
810 BDRVRawState *s = aiocb->bs->opaque;
812 if (s->has_write_zeroes == 0) {
816 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
819 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
820 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
823 } while (errno == EINTR);
830 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
835 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
837 s->has_write_zeroes = false;
843 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
845 int ret = -EOPNOTSUPP;
846 BDRVRawState *s = aiocb->bs->opaque;
848 if (!s->has_discard) {
852 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
855 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
856 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
859 } while (errno == EINTR);
866 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
870 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
872 if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
873 aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
876 } while (errno == EINTR);
882 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
884 s->has_discard = false;
890 static int aio_worker(void *arg)
892 RawPosixAIOData *aiocb = arg;
895 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
897 ret = handle_aiocb_rw(aiocb);
898 if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
899 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
900 0, aiocb->aio_nbytes - ret);
902 ret = aiocb->aio_nbytes;
904 if (ret == aiocb->aio_nbytes) {
906 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
911 ret = handle_aiocb_rw(aiocb);
912 if (ret == aiocb->aio_nbytes) {
914 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
919 ret = handle_aiocb_flush(aiocb);
922 ret = handle_aiocb_ioctl(aiocb);
924 case QEMU_AIO_DISCARD:
925 ret = handle_aiocb_discard(aiocb);
927 case QEMU_AIO_WRITE_ZEROES:
928 ret = handle_aiocb_write_zeroes(aiocb);
931 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
936 g_slice_free(RawPosixAIOData, aiocb);
940 static int paio_submit_co(BlockDriverState *bs, int fd,
941 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
944 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
948 acb->aio_type = type;
949 acb->aio_fildes = fd;
952 acb->aio_iov = qiov->iov;
953 acb->aio_niov = qiov->niov;
955 acb->aio_nbytes = nb_sectors * 512;
956 acb->aio_offset = sector_num * 512;
958 trace_paio_submit_co(sector_num, nb_sectors, type);
959 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
960 return thread_pool_submit_co(pool, aio_worker, acb);
963 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
964 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
965 BlockDriverCompletionFunc *cb, void *opaque, int type)
967 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
971 acb->aio_type = type;
972 acb->aio_fildes = fd;
975 acb->aio_iov = qiov->iov;
976 acb->aio_niov = qiov->niov;
978 acb->aio_nbytes = nb_sectors * 512;
979 acb->aio_offset = sector_num * 512;
981 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
982 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
983 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
986 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
987 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
988 BlockDriverCompletionFunc *cb, void *opaque, int type)
990 BDRVRawState *s = bs->opaque;
996 * If O_DIRECT is used the buffer needs to be aligned on a sector
997 * boundary. Check if this is the case or tell the low-level
998 * driver that it needs to copy the buffer.
1000 if ((bs->open_flags & BDRV_O_NOCACHE)) {
1001 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1002 type |= QEMU_AIO_MISALIGNED;
1003 #ifdef CONFIG_LINUX_AIO
1004 } else if (s->use_aio) {
1005 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1006 nb_sectors, cb, opaque, type);
1011 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1015 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
1016 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1017 BlockDriverCompletionFunc *cb, void *opaque)
1019 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1020 cb, opaque, QEMU_AIO_READ);
1023 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
1024 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1025 BlockDriverCompletionFunc *cb, void *opaque)
1027 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1028 cb, opaque, QEMU_AIO_WRITE);
1031 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
1032 BlockDriverCompletionFunc *cb, void *opaque)
1034 BDRVRawState *s = bs->opaque;
1036 if (fd_open(bs) < 0)
1039 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1042 static void raw_close(BlockDriverState *bs)
1044 BDRVRawState *s = bs->opaque;
1051 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1053 BDRVRawState *s = bs->opaque;
1056 if (fstat(s->fd, &st)) {
1060 if (S_ISREG(st.st_mode)) {
1061 if (ftruncate(s->fd, offset) < 0) {
1064 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1065 if (offset > raw_getlength(bs)) {
1076 static int64_t raw_getlength(BlockDriverState *bs)
1078 BDRVRawState *s = bs->opaque;
1084 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1085 struct disklabel dl;
1087 if (ioctl(fd, DIOCGDINFO, &dl))
1089 return (uint64_t)dl.d_secsize *
1090 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1094 #elif defined(__NetBSD__)
1095 static int64_t raw_getlength(BlockDriverState *bs)
1097 BDRVRawState *s = bs->opaque;
1103 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1104 struct dkwedge_info dkw;
1106 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1107 return dkw.dkw_size * 512;
1109 struct disklabel dl;
1111 if (ioctl(fd, DIOCGDINFO, &dl))
1113 return (uint64_t)dl.d_secsize *
1114 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1119 #elif defined(__sun__)
1120 static int64_t raw_getlength(BlockDriverState *bs)
1122 BDRVRawState *s = bs->opaque;
1123 struct dk_minfo minfo;
1132 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1134 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1136 return minfo.dki_lbsize * minfo.dki_capacity;
1140 * There are reports that lseek on some devices fails, but
1141 * irc discussion said that contingency on contingency was overkill.
1143 return lseek(s->fd, 0, SEEK_END);
1145 #elif defined(CONFIG_BSD)
1146 static int64_t raw_getlength(BlockDriverState *bs)
1148 BDRVRawState *s = bs->opaque;
1152 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1161 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1164 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1165 #ifdef DIOCGMEDIASIZE
1166 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1167 #elif defined(DIOCGPART)
1170 if (ioctl(fd, DIOCGPART, &pi) == 0)
1171 size = pi.media_size;
1177 #if defined(__APPLE__) && defined(__MACH__)
1178 size = LONG_LONG_MAX;
1180 size = lseek(fd, 0LL, SEEK_END);
1182 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1185 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1186 if (size == 2048LL * (unsigned)-1)
1188 /* XXX no disc? maybe we need to reopen... */
1189 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1196 size = lseek(fd, 0, SEEK_END);
1201 static int64_t raw_getlength(BlockDriverState *bs)
1203 BDRVRawState *s = bs->opaque;
1211 return lseek(s->fd, 0, SEEK_END);
1215 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1218 BDRVRawState *s = bs->opaque;
1220 if (fstat(s->fd, &st) < 0) {
1223 return (int64_t)st.st_blocks * 512;
1226 static int raw_create(const char *filename, QEMUOptionParameter *options,
1231 int64_t total_size = 0;
1233 /* Read out options */
1234 while (options && options->name) {
1235 if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1236 total_size = options->value.n / BDRV_SECTOR_SIZE;
1241 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1245 error_setg_errno(errp, -result, "Could not create file");
1247 if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1249 error_setg_errno(errp, -result, "Could not resize file");
1251 if (qemu_close(fd) != 0) {
1253 error_setg_errno(errp, -result, "Could not close the new file");
1260 * Returns true iff the specified sector is present in the disk image. Drivers
1261 * not implementing the functionality are assumed to not support backing files,
1262 * hence all their sectors are reported as allocated.
1264 * If 'sector_num' is beyond the end of the disk image the return value is 0
1265 * and 'pnum' is set to 0.
1267 * 'pnum' is set to the number of sectors (including and immediately following
1268 * the specified sector) that are known to be in the same
1269 * allocated/unallocated state.
1271 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1272 * beyond the end of the disk image it will be clamped.
1274 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1276 int nb_sectors, int *pnum)
1278 off_t start, data, hole;
1286 start = sector_num * BDRV_SECTOR_SIZE;
1287 ret = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1289 #ifdef CONFIG_FIEMAP
1291 BDRVRawState *s = bs->opaque;
1294 struct fiemap_extent fe;
1297 f.fm.fm_start = start;
1298 f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1300 f.fm.fm_extent_count = 1;
1301 f.fm.fm_reserved = 0;
1302 if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1303 /* Assume everything is allocated. */
1308 if (f.fm.fm_mapped_extents == 0) {
1309 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1310 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1312 off_t length = lseek(s->fd, 0, SEEK_END);
1313 hole = f.fm.fm_start;
1314 data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1316 data = f.fe.fe_logical;
1317 hole = f.fe.fe_logical + f.fe.fe_length;
1318 if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1319 ret |= BDRV_BLOCK_ZERO;
1323 #elif defined SEEK_HOLE && defined SEEK_DATA
1325 BDRVRawState *s = bs->opaque;
1327 hole = lseek(s->fd, start, SEEK_HOLE);
1329 /* -ENXIO indicates that sector_num was past the end of the file.
1330 * There is a virtual hole there. */
1331 assert(errno != -ENXIO);
1333 /* Most likely EINVAL. Assume everything is allocated. */
1341 /* On a hole. We need another syscall to find its end. */
1342 data = lseek(s->fd, start, SEEK_DATA);
1344 data = lseek(s->fd, 0, SEEK_END);
1349 hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1352 if (data <= start) {
1353 /* On a data extent, compute sectors to the end of the extent. */
1354 *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1356 /* On a hole, compute sectors to the beginning of the next extent. */
1357 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1358 ret &= ~BDRV_BLOCK_DATA;
1359 ret |= BDRV_BLOCK_ZERO;
1365 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1366 int64_t sector_num, int nb_sectors,
1367 BlockDriverCompletionFunc *cb, void *opaque)
1369 BDRVRawState *s = bs->opaque;
1371 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1372 cb, opaque, QEMU_AIO_DISCARD);
1375 static int coroutine_fn raw_co_write_zeroes(
1376 BlockDriverState *bs, int64_t sector_num,
1377 int nb_sectors, BdrvRequestFlags flags)
1379 BDRVRawState *s = bs->opaque;
1381 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1382 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1383 QEMU_AIO_WRITE_ZEROES);
1384 } else if (s->discard_zeroes) {
1385 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1391 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1393 BDRVRawState *s = bs->opaque;
1395 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1396 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1400 static QEMUOptionParameter raw_create_options[] = {
1402 .name = BLOCK_OPT_SIZE,
1404 .help = "Virtual disk size"
1409 static BlockDriver bdrv_file = {
1410 .format_name = "file",
1411 .protocol_name = "file",
1412 .instance_size = sizeof(BDRVRawState),
1413 .bdrv_needs_filename = true,
1414 .bdrv_probe = NULL, /* no probe for protocols */
1415 .bdrv_file_open = raw_open,
1416 .bdrv_reopen_prepare = raw_reopen_prepare,
1417 .bdrv_reopen_commit = raw_reopen_commit,
1418 .bdrv_reopen_abort = raw_reopen_abort,
1419 .bdrv_close = raw_close,
1420 .bdrv_create = raw_create,
1421 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1422 .bdrv_co_get_block_status = raw_co_get_block_status,
1423 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1425 .bdrv_aio_readv = raw_aio_readv,
1426 .bdrv_aio_writev = raw_aio_writev,
1427 .bdrv_aio_flush = raw_aio_flush,
1428 .bdrv_aio_discard = raw_aio_discard,
1429 .bdrv_refresh_limits = raw_refresh_limits,
1431 .bdrv_truncate = raw_truncate,
1432 .bdrv_getlength = raw_getlength,
1433 .bdrv_get_info = raw_get_info,
1434 .bdrv_get_allocated_file_size
1435 = raw_get_allocated_file_size,
1437 .create_options = raw_create_options,
1440 /***********************************************/
1443 #if defined(__APPLE__) && defined(__MACH__)
1444 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1445 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1447 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1449 kern_return_t kernResult;
1450 mach_port_t masterPort;
1451 CFMutableDictionaryRef classesToMatch;
1453 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1454 if ( KERN_SUCCESS != kernResult ) {
1455 printf( "IOMasterPort returned %d\n", kernResult );
1458 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1459 if ( classesToMatch == NULL ) {
1460 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1462 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1464 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1465 if ( KERN_SUCCESS != kernResult )
1467 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1473 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1475 io_object_t nextMedia;
1476 kern_return_t kernResult = KERN_FAILURE;
1478 nextMedia = IOIteratorNext( mediaIterator );
1481 CFTypeRef bsdPathAsCFString;
1482 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1483 if ( bsdPathAsCFString ) {
1484 size_t devPathLength;
1485 strcpy( bsdPath, _PATH_DEV );
1486 strcat( bsdPath, "r" );
1487 devPathLength = strlen( bsdPath );
1488 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1489 kernResult = KERN_SUCCESS;
1491 CFRelease( bsdPathAsCFString );
1493 IOObjectRelease( nextMedia );
1501 static int hdev_probe_device(const char *filename)
1505 /* allow a dedicated CD-ROM driver to match with a higher priority */
1506 if (strstart(filename, "/dev/cdrom", NULL))
1509 if (stat(filename, &st) >= 0 &&
1510 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1517 static int check_hdev_writable(BDRVRawState *s)
1519 #if defined(BLKROGET)
1520 /* Linux block devices can be configured "read-only" using blockdev(8).
1521 * This is independent of device node permissions and therefore open(2)
1522 * with O_RDWR succeeds. Actual writes fail with EPERM.
1524 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1525 * check for read-only block devices so that Linux block devices behave
1531 if (fstat(s->fd, &st)) {
1535 if (!S_ISBLK(st.st_mode)) {
1539 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1546 #endif /* defined(BLKROGET) */
1550 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1553 BDRVRawState *s = bs->opaque;
1554 Error *local_err = NULL;
1556 const char *filename = qdict_get_str(options, "filename");
1558 #if defined(__APPLE__) && defined(__MACH__)
1559 if (strstart(filename, "/dev/cdrom", NULL)) {
1560 kern_return_t kernResult;
1561 io_iterator_t mediaIterator;
1562 char bsdPath[ MAXPATHLEN ];
1565 kernResult = FindEjectableCDMedia( &mediaIterator );
1566 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1568 if ( bsdPath[ 0 ] != '\0' ) {
1569 strcat(bsdPath,"s0");
1570 /* some CDs don't have a partition 0 */
1571 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1573 bsdPath[strlen(bsdPath)-1] = '1';
1578 qdict_put(options, "filename", qstring_from_str(filename));
1581 if ( mediaIterator )
1582 IOObjectRelease( mediaIterator );
1586 s->type = FTYPE_FILE;
1587 #if defined(__linux__)
1589 char resolved_path[ MAXPATHLEN ], *temp;
1591 temp = realpath(filename, resolved_path);
1592 if (temp && strstart(temp, "/dev/sg", NULL)) {
1598 ret = raw_open_common(bs, options, flags, 0, &local_err);
1601 error_propagate(errp, local_err);
1606 if (flags & BDRV_O_RDWR) {
1607 ret = check_hdev_writable(s);
1610 error_setg_errno(errp, -ret, "The device is not writable");
1618 #if defined(__linux__)
1619 /* Note: we do not have a reliable method to detect if the floppy is
1620 present. The current method is to try to open the floppy at every
1621 I/O and to keep it opened during a few hundreds of ms. */
1622 static int fd_open(BlockDriverState *bs)
1624 BDRVRawState *s = bs->opaque;
1625 int last_media_present;
1627 if (s->type != FTYPE_FD)
1629 last_media_present = (s->fd >= 0);
1631 (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1635 printf("Floppy closed\n");
1639 if (s->fd_got_error &&
1640 (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1642 printf("No floppy (open delayed)\n");
1646 s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1648 s->fd_error_time = get_clock();
1649 s->fd_got_error = 1;
1650 if (last_media_present)
1651 s->fd_media_changed = 1;
1653 printf("No floppy\n");
1658 printf("Floppy opened\n");
1661 if (!last_media_present)
1662 s->fd_media_changed = 1;
1663 s->fd_open_time = get_clock();
1664 s->fd_got_error = 0;
1668 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1670 BDRVRawState *s = bs->opaque;
1672 return ioctl(s->fd, req, buf);
1675 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1676 unsigned long int req, void *buf,
1677 BlockDriverCompletionFunc *cb, void *opaque)
1679 BDRVRawState *s = bs->opaque;
1680 RawPosixAIOData *acb;
1683 if (fd_open(bs) < 0)
1686 acb = g_slice_new(RawPosixAIOData);
1688 acb->aio_type = QEMU_AIO_IOCTL;
1689 acb->aio_fildes = s->fd;
1690 acb->aio_offset = 0;
1691 acb->aio_ioctl_buf = buf;
1692 acb->aio_ioctl_cmd = req;
1693 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1694 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1697 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1698 static int fd_open(BlockDriverState *bs)
1700 BDRVRawState *s = bs->opaque;
1702 /* this is just to ensure s->fd is sane (its called by io ops) */
1707 #else /* !linux && !FreeBSD */
1709 static int fd_open(BlockDriverState *bs)
1714 #endif /* !linux && !FreeBSD */
1716 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1717 int64_t sector_num, int nb_sectors,
1718 BlockDriverCompletionFunc *cb, void *opaque)
1720 BDRVRawState *s = bs->opaque;
1722 if (fd_open(bs) < 0) {
1725 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1726 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1729 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
1730 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
1732 BDRVRawState *s = bs->opaque;
1739 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1740 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1741 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
1742 } else if (s->discard_zeroes) {
1743 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1744 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1749 static int hdev_create(const char *filename, QEMUOptionParameter *options,
1754 struct stat stat_buf;
1755 int64_t total_size = 0;
1757 /* Read out options */
1758 while (options && options->name) {
1759 if (!strcmp(options->name, "size")) {
1760 total_size = options->value.n / BDRV_SECTOR_SIZE;
1765 fd = qemu_open(filename, O_WRONLY | O_BINARY);
1768 error_setg_errno(errp, -ret, "Could not open device");
1772 if (fstat(fd, &stat_buf) < 0) {
1774 error_setg_errno(errp, -ret, "Could not stat device");
1775 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
1777 "The given file is neither a block nor a character device");
1779 } else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE) {
1780 error_setg(errp, "Device is too small");
1788 static BlockDriver bdrv_host_device = {
1789 .format_name = "host_device",
1790 .protocol_name = "host_device",
1791 .instance_size = sizeof(BDRVRawState),
1792 .bdrv_needs_filename = true,
1793 .bdrv_probe_device = hdev_probe_device,
1794 .bdrv_file_open = hdev_open,
1795 .bdrv_close = raw_close,
1796 .bdrv_reopen_prepare = raw_reopen_prepare,
1797 .bdrv_reopen_commit = raw_reopen_commit,
1798 .bdrv_reopen_abort = raw_reopen_abort,
1799 .bdrv_create = hdev_create,
1800 .create_options = raw_create_options,
1801 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
1803 .bdrv_aio_readv = raw_aio_readv,
1804 .bdrv_aio_writev = raw_aio_writev,
1805 .bdrv_aio_flush = raw_aio_flush,
1806 .bdrv_aio_discard = hdev_aio_discard,
1807 .bdrv_refresh_limits = raw_refresh_limits,
1809 .bdrv_truncate = raw_truncate,
1810 .bdrv_getlength = raw_getlength,
1811 .bdrv_get_info = raw_get_info,
1812 .bdrv_get_allocated_file_size
1813 = raw_get_allocated_file_size,
1815 /* generic scsi device */
1817 .bdrv_ioctl = hdev_ioctl,
1818 .bdrv_aio_ioctl = hdev_aio_ioctl,
1823 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
1826 BDRVRawState *s = bs->opaque;
1827 Error *local_err = NULL;
1832 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1833 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1836 error_propagate(errp, local_err);
1841 /* close fd so that we can reopen it as needed */
1844 s->fd_media_changed = 1;
1849 static int floppy_probe_device(const char *filename)
1853 struct floppy_struct fdparam;
1856 if (strstart(filename, "/dev/fd", NULL) &&
1857 !strstart(filename, "/dev/fdset/", NULL)) {
1861 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1865 ret = fstat(fd, &st);
1866 if (ret == -1 || !S_ISBLK(st.st_mode)) {
1870 /* Attempt to detect via a floppy specific ioctl */
1871 ret = ioctl(fd, FDGETPRM, &fdparam);
1882 static int floppy_is_inserted(BlockDriverState *bs)
1884 return fd_open(bs) >= 0;
1887 static int floppy_media_changed(BlockDriverState *bs)
1889 BDRVRawState *s = bs->opaque;
1893 * XXX: we do not have a true media changed indication.
1894 * It does not work if the floppy is changed without trying to read it.
1897 ret = s->fd_media_changed;
1898 s->fd_media_changed = 0;
1900 printf("Floppy changed=%d\n", ret);
1905 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1907 BDRVRawState *s = bs->opaque;
1914 fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1916 if (ioctl(fd, FDEJECT, 0) < 0)
1922 static BlockDriver bdrv_host_floppy = {
1923 .format_name = "host_floppy",
1924 .protocol_name = "host_floppy",
1925 .instance_size = sizeof(BDRVRawState),
1926 .bdrv_needs_filename = true,
1927 .bdrv_probe_device = floppy_probe_device,
1928 .bdrv_file_open = floppy_open,
1929 .bdrv_close = raw_close,
1930 .bdrv_reopen_prepare = raw_reopen_prepare,
1931 .bdrv_reopen_commit = raw_reopen_commit,
1932 .bdrv_reopen_abort = raw_reopen_abort,
1933 .bdrv_create = hdev_create,
1934 .create_options = raw_create_options,
1936 .bdrv_aio_readv = raw_aio_readv,
1937 .bdrv_aio_writev = raw_aio_writev,
1938 .bdrv_aio_flush = raw_aio_flush,
1939 .bdrv_refresh_limits = raw_refresh_limits,
1941 .bdrv_truncate = raw_truncate,
1942 .bdrv_getlength = raw_getlength,
1943 .has_variable_length = true,
1944 .bdrv_get_allocated_file_size
1945 = raw_get_allocated_file_size,
1947 /* removable device support */
1948 .bdrv_is_inserted = floppy_is_inserted,
1949 .bdrv_media_changed = floppy_media_changed,
1950 .bdrv_eject = floppy_eject,
1953 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
1956 BDRVRawState *s = bs->opaque;
1957 Error *local_err = NULL;
1962 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1963 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
1965 error_propagate(errp, local_err);
1970 static int cdrom_probe_device(const char *filename)
1976 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1980 ret = fstat(fd, &st);
1981 if (ret == -1 || !S_ISBLK(st.st_mode)) {
1985 /* Attempt to detect via a CDROM specific ioctl */
1986 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1996 static int cdrom_is_inserted(BlockDriverState *bs)
1998 BDRVRawState *s = bs->opaque;
2001 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2002 if (ret == CDS_DISC_OK)
2007 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2009 BDRVRawState *s = bs->opaque;
2012 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2013 perror("CDROMEJECT");
2015 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2016 perror("CDROMEJECT");
2020 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2022 BDRVRawState *s = bs->opaque;
2024 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2026 * Note: an error can happen if the distribution automatically
2029 /* perror("CDROM_LOCKDOOR"); */
2033 static BlockDriver bdrv_host_cdrom = {
2034 .format_name = "host_cdrom",
2035 .protocol_name = "host_cdrom",
2036 .instance_size = sizeof(BDRVRawState),
2037 .bdrv_needs_filename = true,
2038 .bdrv_probe_device = cdrom_probe_device,
2039 .bdrv_file_open = cdrom_open,
2040 .bdrv_close = raw_close,
2041 .bdrv_reopen_prepare = raw_reopen_prepare,
2042 .bdrv_reopen_commit = raw_reopen_commit,
2043 .bdrv_reopen_abort = raw_reopen_abort,
2044 .bdrv_create = hdev_create,
2045 .create_options = raw_create_options,
2047 .bdrv_aio_readv = raw_aio_readv,
2048 .bdrv_aio_writev = raw_aio_writev,
2049 .bdrv_aio_flush = raw_aio_flush,
2050 .bdrv_refresh_limits = raw_refresh_limits,
2052 .bdrv_truncate = raw_truncate,
2053 .bdrv_getlength = raw_getlength,
2054 .has_variable_length = true,
2055 .bdrv_get_allocated_file_size
2056 = raw_get_allocated_file_size,
2058 /* removable device support */
2059 .bdrv_is_inserted = cdrom_is_inserted,
2060 .bdrv_eject = cdrom_eject,
2061 .bdrv_lock_medium = cdrom_lock_medium,
2063 /* generic scsi device */
2064 .bdrv_ioctl = hdev_ioctl,
2065 .bdrv_aio_ioctl = hdev_aio_ioctl,
2067 #endif /* __linux__ */
2069 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2070 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2073 BDRVRawState *s = bs->opaque;
2074 Error *local_err = NULL;
2079 ret = raw_open_common(bs, options, flags, 0, &local_err);
2082 error_propagate(errp, local_err);
2087 /* make sure the door isn't locked at this time */
2088 ioctl(s->fd, CDIOCALLOW);
2092 static int cdrom_probe_device(const char *filename)
2094 if (strstart(filename, "/dev/cd", NULL) ||
2095 strstart(filename, "/dev/acd", NULL))
2100 static int cdrom_reopen(BlockDriverState *bs)
2102 BDRVRawState *s = bs->opaque;
2106 * Force reread of possibly changed/newly loaded disc,
2107 * FreeBSD seems to not notice sometimes...
2111 fd = qemu_open(bs->filename, s->open_flags, 0644);
2118 /* make sure the door isn't locked at this time */
2119 ioctl(s->fd, CDIOCALLOW);
2123 static int cdrom_is_inserted(BlockDriverState *bs)
2125 return raw_getlength(bs) > 0;
2128 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2130 BDRVRawState *s = bs->opaque;
2135 (void) ioctl(s->fd, CDIOCALLOW);
2138 if (ioctl(s->fd, CDIOCEJECT) < 0)
2139 perror("CDIOCEJECT");
2141 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2142 perror("CDIOCCLOSE");
2148 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2150 BDRVRawState *s = bs->opaque;
2154 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2156 * Note: an error can happen if the distribution automatically
2159 /* perror("CDROM_LOCKDOOR"); */
2163 static BlockDriver bdrv_host_cdrom = {
2164 .format_name = "host_cdrom",
2165 .protocol_name = "host_cdrom",
2166 .instance_size = sizeof(BDRVRawState),
2167 .bdrv_needs_filename = true,
2168 .bdrv_probe_device = cdrom_probe_device,
2169 .bdrv_file_open = cdrom_open,
2170 .bdrv_close = raw_close,
2171 .bdrv_reopen_prepare = raw_reopen_prepare,
2172 .bdrv_reopen_commit = raw_reopen_commit,
2173 .bdrv_reopen_abort = raw_reopen_abort,
2174 .bdrv_create = hdev_create,
2175 .create_options = raw_create_options,
2177 .bdrv_aio_readv = raw_aio_readv,
2178 .bdrv_aio_writev = raw_aio_writev,
2179 .bdrv_aio_flush = raw_aio_flush,
2180 .bdrv_refresh_limits = raw_refresh_limits,
2182 .bdrv_truncate = raw_truncate,
2183 .bdrv_getlength = raw_getlength,
2184 .has_variable_length = true,
2185 .bdrv_get_allocated_file_size
2186 = raw_get_allocated_file_size,
2188 /* removable device support */
2189 .bdrv_is_inserted = cdrom_is_inserted,
2190 .bdrv_eject = cdrom_eject,
2191 .bdrv_lock_medium = cdrom_lock_medium,
2193 #endif /* __FreeBSD__ */
2195 #ifdef CONFIG_LINUX_AIO
2197 * Return the file descriptor for Linux AIO
2199 * This function is a layering violation and should be removed when it becomes
2200 * possible to call the block layer outside the global mutex. It allows the
2201 * caller to hijack the file descriptor so I/O can be performed outside the
2204 int raw_get_aio_fd(BlockDriverState *bs)
2212 if (bs->drv == bdrv_find_format("raw")) {
2216 /* raw-posix has several protocols so just check for raw_aio_readv */
2217 if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
2227 #endif /* CONFIG_LINUX_AIO */
2229 static void bdrv_file_init(void)
2232 * Register all the drivers. Note that order is important, the driver
2233 * registered last will get probed first.
2235 bdrv_register(&bdrv_file);
2236 bdrv_register(&bdrv_host_device);
2238 bdrv_register(&bdrv_host_floppy);
2239 bdrv_register(&bdrv_host_cdrom);
2241 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2242 bdrv_register(&bdrv_host_cdrom);
2246 block_init(bdrv_file_init);