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"
33 #include "qapi/util.h"
35 #if defined(__APPLE__) && (__MACH__)
37 #include <sys/param.h>
38 #include <IOKit/IOKitLib.h>
39 #include <IOKit/IOBSD.h>
40 #include <IOKit/storage/IOMediaBSDClient.h>
41 #include <IOKit/storage/IOMedia.h>
42 #include <IOKit/storage/IOCDMedia.h>
43 //#include <IOKit/storage/IOCDTypes.h>
44 #include <CoreFoundation/CoreFoundation.h>
48 #define _POSIX_PTHREAD_SEMANTICS 1
52 #include <sys/types.h>
54 #include <sys/ioctl.h>
55 #include <sys/param.h>
56 #include <linux/cdrom.h>
60 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
64 #include <linux/fiemap.h>
66 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
67 #include <linux/falloc.h>
69 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
75 #include <sys/ioctl.h>
76 #include <sys/disklabel.h>
81 #include <sys/ioctl.h>
82 #include <sys/disklabel.h>
88 #include <sys/ioctl.h>
89 #include <sys/diskslice.h>
96 //#define DEBUG_FLOPPY
99 #if defined(DEBUG_BLOCK)
100 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
101 { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
103 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
106 /* OS X does not have O_DSYNC */
109 #define O_DSYNC O_SYNC
110 #elif defined(O_FSYNC)
111 #define O_DSYNC O_FSYNC
115 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
117 #define O_DIRECT O_DSYNC
124 /* if the FD is not accessed during that time (in ns), we try to
125 reopen it to see if the disk has been changed */
126 #define FD_OPEN_TIMEOUT (1000000000)
128 #define MAX_BLOCKSIZE 4096
130 typedef struct BDRVRawState {
136 #if defined(__linux__)
137 /* linux floppy specific */
138 int64_t fd_open_time;
139 int64_t fd_error_time;
141 int fd_media_changed;
143 #ifdef CONFIG_LINUX_AIO
151 bool has_write_zeroes:1;
152 bool discard_zeroes:1;
153 bool needs_alignment;
159 typedef struct BDRVRawReopenState {
162 #ifdef CONFIG_LINUX_AIO
165 } BDRVRawReopenState;
167 static int fd_open(BlockDriverState *bs);
168 static int64_t raw_getlength(BlockDriverState *bs);
170 typedef struct RawPosixAIOData {
171 BlockDriverState *bs;
174 struct iovec *aio_iov;
179 #define aio_ioctl_cmd aio_nbytes /* for QEMU_AIO_IOCTL */
184 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
185 static int cdrom_reopen(BlockDriverState *bs);
188 #if defined(__NetBSD__)
189 static int raw_normalize_devicepath(const char **filename)
191 static char namebuf[PATH_MAX];
192 const char *dp, *fname;
196 dp = strrchr(fname, '/');
197 if (lstat(fname, &sb) < 0) {
198 fprintf(stderr, "%s: stat failed: %s\n",
199 fname, strerror(errno));
203 if (!S_ISBLK(sb.st_mode)) {
208 snprintf(namebuf, PATH_MAX, "r%s", fname);
210 snprintf(namebuf, PATH_MAX, "%.*s/r%s",
211 (int)(dp - fname), fname, dp + 1);
213 fprintf(stderr, "%s is a block device", fname);
215 fprintf(stderr, ", using %s\n", *filename);
220 static int raw_normalize_devicepath(const char **filename)
226 static void raw_probe_alignment(BlockDriverState *bs, int fd, Error **errp)
228 BDRVRawState *s = bs->opaque;
230 unsigned int sector_size;
232 /* For /dev/sg devices the alignment is not really used.
233 With buffered I/O, we don't have any restrictions. */
234 if (bs->sg || !s->needs_alignment) {
235 bs->request_alignment = 1;
240 /* Try a few ioctls to get the right size */
241 bs->request_alignment = 0;
245 if (ioctl(fd, BLKSSZGET, §or_size) >= 0) {
246 bs->request_alignment = sector_size;
249 #ifdef DKIOCGETBLOCKSIZE
250 if (ioctl(fd, DKIOCGETBLOCKSIZE, §or_size) >= 0) {
251 bs->request_alignment = sector_size;
254 #ifdef DIOCGSECTORSIZE
255 if (ioctl(fd, DIOCGSECTORSIZE, §or_size) >= 0) {
256 bs->request_alignment = sector_size;
262 if (xfsctl(NULL, fd, XFS_IOC_DIOINFO, &da) >= 0) {
263 bs->request_alignment = da.d_miniosz;
264 /* The kernel returns wrong information for d_mem */
265 /* s->buf_align = da.d_mem; */
270 /* If we could not get the sizes so far, we can only guess them */
273 buf = qemu_memalign(MAX_BLOCKSIZE, 2 * MAX_BLOCKSIZE);
274 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
275 if (pread(fd, buf + align, MAX_BLOCKSIZE, 0) >= 0) {
276 s->buf_align = align;
283 if (!bs->request_alignment) {
285 buf = qemu_memalign(s->buf_align, MAX_BLOCKSIZE);
286 for (align = 512; align <= MAX_BLOCKSIZE; align <<= 1) {
287 if (pread(fd, buf, align, 0) >= 0) {
288 bs->request_alignment = align;
295 if (!s->buf_align || !bs->request_alignment) {
296 error_setg(errp, "Could not find working O_DIRECT alignment. "
297 "Try cache.direct=off.");
301 static void raw_parse_flags(int bdrv_flags, int *open_flags)
303 assert(open_flags != NULL);
305 *open_flags |= O_BINARY;
306 *open_flags &= ~O_ACCMODE;
307 if (bdrv_flags & BDRV_O_RDWR) {
308 *open_flags |= O_RDWR;
310 *open_flags |= O_RDONLY;
313 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
314 * and O_DIRECT for no caching. */
315 if ((bdrv_flags & BDRV_O_NOCACHE)) {
316 *open_flags |= O_DIRECT;
320 static void raw_detach_aio_context(BlockDriverState *bs)
322 #ifdef CONFIG_LINUX_AIO
323 BDRVRawState *s = bs->opaque;
326 laio_detach_aio_context(s->aio_ctx, bdrv_get_aio_context(bs));
331 static void raw_attach_aio_context(BlockDriverState *bs,
332 AioContext *new_context)
334 #ifdef CONFIG_LINUX_AIO
335 BDRVRawState *s = bs->opaque;
338 laio_attach_aio_context(s->aio_ctx, new_context);
343 #ifdef CONFIG_LINUX_AIO
344 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
347 assert(aio_ctx != NULL);
348 assert(use_aio != NULL);
350 * Currently Linux do AIO only for files opened with O_DIRECT
351 * specified so check NOCACHE flag too
353 if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
354 (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
356 /* if non-NULL, laio_init() has already been run */
357 if (*aio_ctx == NULL) {
358 *aio_ctx = laio_init();
375 static void raw_parse_filename(const char *filename, QDict *options,
378 /* The filename does not have to be prefixed by the protocol name, since
379 * "file" is the default protocol; therefore, the return value of this
380 * function call can be ignored. */
381 strstart(filename, "file:", &filename);
383 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
386 static QemuOptsList raw_runtime_opts = {
388 .head = QTAILQ_HEAD_INITIALIZER(raw_runtime_opts.head),
392 .type = QEMU_OPT_STRING,
393 .help = "File name of the image",
395 { /* end of list */ }
399 static int raw_open_common(BlockDriverState *bs, QDict *options,
400 int bdrv_flags, int open_flags, Error **errp)
402 BDRVRawState *s = bs->opaque;
404 Error *local_err = NULL;
405 const char *filename = NULL;
409 opts = qemu_opts_create(&raw_runtime_opts, NULL, 0, &error_abort);
410 qemu_opts_absorb_qdict(opts, options, &local_err);
412 error_propagate(errp, local_err);
417 filename = qemu_opt_get(opts, "filename");
419 ret = raw_normalize_devicepath(&filename);
421 error_setg_errno(errp, -ret, "Could not normalize device path");
425 s->open_flags = open_flags;
426 raw_parse_flags(bdrv_flags, &s->open_flags);
429 fd = qemu_open(filename, s->open_flags, 0644);
439 #ifdef CONFIG_LINUX_AIO
440 if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
443 error_setg_errno(errp, -ret, "Could not set AIO state");
448 s->has_discard = true;
449 s->has_write_zeroes = true;
450 if ((bs->open_flags & BDRV_O_NOCACHE) != 0) {
451 s->needs_alignment = true;
454 if (fstat(s->fd, &st) < 0) {
455 error_setg_errno(errp, errno, "Could not stat file");
458 if (S_ISREG(st.st_mode)) {
459 s->discard_zeroes = true;
461 if (S_ISBLK(st.st_mode)) {
462 #ifdef BLKDISCARDZEROES
464 if (ioctl(s->fd, BLKDISCARDZEROES, &arg) == 0 && arg) {
465 s->discard_zeroes = true;
469 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
470 * not rely on the contents of discarded blocks unless using O_DIRECT.
471 * Same for BLKZEROOUT.
473 if (!(bs->open_flags & BDRV_O_NOCACHE)) {
474 s->discard_zeroes = false;
475 s->has_write_zeroes = false;
480 if (S_ISCHR(st.st_mode)) {
482 * The file is a char device (disk), which on FreeBSD isn't behind
483 * a pager, so force all requests to be aligned. This is needed
484 * so QEMU makes sure all IO operations on the device are aligned
485 * to sector size, or else FreeBSD will reject them with EINVAL.
487 s->needs_alignment = true;
492 if (platform_test_xfs_fd(s->fd)) {
497 raw_attach_aio_context(bs, bdrv_get_aio_context(bs));
501 if (filename && (bdrv_flags & BDRV_O_TEMPORARY)) {
508 static int raw_open(BlockDriverState *bs, QDict *options, int flags,
511 BDRVRawState *s = bs->opaque;
512 Error *local_err = NULL;
515 s->type = FTYPE_FILE;
516 ret = raw_open_common(bs, options, flags, 0, &local_err);
518 error_propagate(errp, local_err);
523 static int raw_reopen_prepare(BDRVReopenState *state,
524 BlockReopenQueue *queue, Error **errp)
527 BDRVRawReopenState *raw_s;
529 Error *local_err = NULL;
531 assert(state != NULL);
532 assert(state->bs != NULL);
534 s = state->bs->opaque;
536 state->opaque = g_new0(BDRVRawReopenState, 1);
537 raw_s = state->opaque;
539 #ifdef CONFIG_LINUX_AIO
540 raw_s->use_aio = s->use_aio;
542 /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
543 * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
544 * won't override aio_ctx if aio_ctx is non-NULL */
545 if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
546 error_setg(errp, "Could not set AIO state");
551 if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
552 raw_s->open_flags |= O_NONBLOCK;
555 raw_parse_flags(state->flags, &raw_s->open_flags);
559 int fcntl_flags = O_APPEND | O_NONBLOCK;
561 fcntl_flags |= O_NOATIME;
565 /* Not all operating systems have O_ASYNC, and those that don't
566 * will not let us track the state into raw_s->open_flags (typically
567 * you achieve the same effect with an ioctl, for example I_SETSIG
568 * on Solaris). But we do not use O_ASYNC, so that's fine.
570 assert((s->open_flags & O_ASYNC) == 0);
573 if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
574 /* dup the original fd */
575 /* TODO: use qemu fcntl wrapper */
576 #ifdef F_DUPFD_CLOEXEC
577 raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
579 raw_s->fd = dup(s->fd);
580 if (raw_s->fd != -1) {
581 qemu_set_cloexec(raw_s->fd);
584 if (raw_s->fd >= 0) {
585 ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
587 qemu_close(raw_s->fd);
593 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
594 if (raw_s->fd == -1) {
595 assert(!(raw_s->open_flags & O_CREAT));
596 raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
597 if (raw_s->fd == -1) {
598 error_setg_errno(errp, errno, "Could not reopen file");
603 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
604 * alignment with the new fd. */
605 if (raw_s->fd != -1) {
606 raw_probe_alignment(state->bs, raw_s->fd, &local_err);
608 qemu_close(raw_s->fd);
610 error_propagate(errp, local_err);
618 static void raw_reopen_commit(BDRVReopenState *state)
620 BDRVRawReopenState *raw_s = state->opaque;
621 BDRVRawState *s = state->bs->opaque;
623 s->open_flags = raw_s->open_flags;
627 #ifdef CONFIG_LINUX_AIO
628 s->use_aio = raw_s->use_aio;
631 g_free(state->opaque);
632 state->opaque = NULL;
636 static void raw_reopen_abort(BDRVReopenState *state)
638 BDRVRawReopenState *raw_s = state->opaque;
640 /* nothing to do if NULL, we didn't get far enough */
645 if (raw_s->fd >= 0) {
646 qemu_close(raw_s->fd);
649 g_free(state->opaque);
650 state->opaque = NULL;
653 static void raw_refresh_limits(BlockDriverState *bs, Error **errp)
655 BDRVRawState *s = bs->opaque;
657 raw_probe_alignment(bs, s->fd, errp);
658 bs->bl.opt_mem_alignment = s->buf_align;
661 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
665 ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
673 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
677 ret = qemu_fdatasync(aiocb->aio_fildes);
686 static bool preadv_present = true;
689 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
691 return preadv(fd, iov, nr_iov, offset);
695 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
697 return pwritev(fd, iov, nr_iov, offset);
702 static bool preadv_present = false;
705 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
711 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
718 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
723 if (aiocb->aio_type & QEMU_AIO_WRITE)
724 len = qemu_pwritev(aiocb->aio_fildes,
729 len = qemu_preadv(aiocb->aio_fildes,
733 } while (len == -1 && errno == EINTR);
742 * Read/writes the data to/from a given linear buffer.
744 * Returns the number of bytes handles or -errno in case of an error. Short
745 * reads are only returned if the end of the file is reached.
747 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
752 while (offset < aiocb->aio_nbytes) {
753 if (aiocb->aio_type & QEMU_AIO_WRITE) {
754 len = pwrite(aiocb->aio_fildes,
755 (const char *)buf + offset,
756 aiocb->aio_nbytes - offset,
757 aiocb->aio_offset + offset);
759 len = pread(aiocb->aio_fildes,
761 aiocb->aio_nbytes - offset,
762 aiocb->aio_offset + offset);
764 if (len == -1 && errno == EINTR) {
766 } else if (len == -1 && errno == EINVAL &&
767 (aiocb->bs->open_flags & BDRV_O_NOCACHE) &&
768 !(aiocb->aio_type & QEMU_AIO_WRITE) &&
770 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
771 * after a short read. Assume that O_DIRECT short reads only occur
772 * at EOF. Therefore this is a short read, not an I/O error.
775 } else if (len == -1) {
778 } else if (len == 0) {
787 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
792 if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
794 * If there is just a single buffer, and it is properly aligned
795 * we can just use plain pread/pwrite without any problems.
797 if (aiocb->aio_niov == 1) {
798 return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
801 * We have more than one iovec, and all are properly aligned.
803 * Try preadv/pwritev first and fall back to linearizing the
804 * buffer if it's not supported.
806 if (preadv_present) {
807 nbytes = handle_aiocb_rw_vector(aiocb);
808 if (nbytes == aiocb->aio_nbytes ||
809 (nbytes < 0 && nbytes != -ENOSYS)) {
812 preadv_present = false;
816 * XXX(hch): short read/write. no easy way to handle the reminder
817 * using these interfaces. For now retry using plain
823 * Ok, we have to do it the hard way, copy all segments into
824 * a single aligned buffer.
826 buf = qemu_try_blockalign(aiocb->bs, aiocb->aio_nbytes);
831 if (aiocb->aio_type & QEMU_AIO_WRITE) {
835 for (i = 0; i < aiocb->aio_niov; ++i) {
836 memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
837 p += aiocb->aio_iov[i].iov_len;
839 assert(p - buf == aiocb->aio_nbytes);
842 nbytes = handle_aiocb_rw_linear(aiocb, buf);
843 if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
845 size_t count = aiocb->aio_nbytes, copy;
848 for (i = 0; i < aiocb->aio_niov && count; ++i) {
850 if (copy > aiocb->aio_iov[i].iov_len) {
851 copy = aiocb->aio_iov[i].iov_len;
853 memcpy(aiocb->aio_iov[i].iov_base, p, copy);
854 assert(count >= copy);
866 static int xfs_write_zeroes(BDRVRawState *s, int64_t offset, uint64_t bytes)
868 struct xfs_flock64 fl;
870 memset(&fl, 0, sizeof(fl));
871 fl.l_whence = SEEK_SET;
875 if (xfsctl(NULL, s->fd, XFS_IOC_ZERO_RANGE, &fl) < 0) {
876 DEBUG_BLOCK_PRINT("cannot write zero range (%s)\n", strerror(errno));
883 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
885 struct xfs_flock64 fl;
887 memset(&fl, 0, sizeof(fl));
888 fl.l_whence = SEEK_SET;
892 if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
893 DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
901 static ssize_t handle_aiocb_write_zeroes(RawPosixAIOData *aiocb)
903 int ret = -EOPNOTSUPP;
904 BDRVRawState *s = aiocb->bs->opaque;
906 if (s->has_write_zeroes == 0) {
910 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
913 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
914 if (ioctl(aiocb->aio_fildes, BLKZEROOUT, range) == 0) {
917 } while (errno == EINTR);
924 return xfs_write_zeroes(s, aiocb->aio_offset, aiocb->aio_nbytes);
929 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
931 s->has_write_zeroes = false;
937 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
939 int ret = -EOPNOTSUPP;
940 BDRVRawState *s = aiocb->bs->opaque;
942 if (!s->has_discard) {
946 if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
949 uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
950 if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
953 } while (errno == EINTR);
960 return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
964 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
966 if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
967 aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
970 } while (errno == EINTR);
976 if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
978 s->has_discard = false;
984 static int aio_worker(void *arg)
986 RawPosixAIOData *aiocb = arg;
989 switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
991 ret = handle_aiocb_rw(aiocb);
992 if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
993 iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
994 0, aiocb->aio_nbytes - ret);
996 ret = aiocb->aio_nbytes;
998 if (ret == aiocb->aio_nbytes) {
1000 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1004 case QEMU_AIO_WRITE:
1005 ret = handle_aiocb_rw(aiocb);
1006 if (ret == aiocb->aio_nbytes) {
1008 } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
1012 case QEMU_AIO_FLUSH:
1013 ret = handle_aiocb_flush(aiocb);
1015 case QEMU_AIO_IOCTL:
1016 ret = handle_aiocb_ioctl(aiocb);
1018 case QEMU_AIO_DISCARD:
1019 ret = handle_aiocb_discard(aiocb);
1021 case QEMU_AIO_WRITE_ZEROES:
1022 ret = handle_aiocb_write_zeroes(aiocb);
1025 fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
1030 g_slice_free(RawPosixAIOData, aiocb);
1034 static int paio_submit_co(BlockDriverState *bs, int fd,
1035 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1038 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1042 acb->aio_type = type;
1043 acb->aio_fildes = fd;
1045 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1046 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1049 acb->aio_iov = qiov->iov;
1050 acb->aio_niov = qiov->niov;
1051 assert(qiov->size == acb->aio_nbytes);
1054 trace_paio_submit_co(sector_num, nb_sectors, type);
1055 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1056 return thread_pool_submit_co(pool, aio_worker, acb);
1059 static BlockAIOCB *paio_submit(BlockDriverState *bs, int fd,
1060 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1061 BlockCompletionFunc *cb, void *opaque, int type)
1063 RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
1067 acb->aio_type = type;
1068 acb->aio_fildes = fd;
1070 acb->aio_nbytes = nb_sectors * BDRV_SECTOR_SIZE;
1071 acb->aio_offset = sector_num * BDRV_SECTOR_SIZE;
1074 acb->aio_iov = qiov->iov;
1075 acb->aio_niov = qiov->niov;
1076 assert(qiov->size == acb->aio_nbytes);
1079 trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
1080 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1081 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1084 static BlockAIOCB *raw_aio_submit(BlockDriverState *bs,
1085 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1086 BlockCompletionFunc *cb, void *opaque, int type)
1088 BDRVRawState *s = bs->opaque;
1090 if (fd_open(bs) < 0)
1094 * Check if the underlying device requires requests to be aligned,
1095 * and if the request we are trying to submit is aligned or not.
1096 * If this is the case tell the low-level driver that it needs
1097 * to copy the buffer.
1099 if (s->needs_alignment) {
1100 if (!bdrv_qiov_is_aligned(bs, qiov)) {
1101 type |= QEMU_AIO_MISALIGNED;
1102 #ifdef CONFIG_LINUX_AIO
1103 } else if (s->use_aio) {
1104 return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
1105 nb_sectors, cb, opaque, type);
1110 return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
1114 static void raw_aio_plug(BlockDriverState *bs)
1116 #ifdef CONFIG_LINUX_AIO
1117 BDRVRawState *s = bs->opaque;
1119 laio_io_plug(bs, s->aio_ctx);
1124 static void raw_aio_unplug(BlockDriverState *bs)
1126 #ifdef CONFIG_LINUX_AIO
1127 BDRVRawState *s = bs->opaque;
1129 laio_io_unplug(bs, s->aio_ctx, true);
1134 static void raw_aio_flush_io_queue(BlockDriverState *bs)
1136 #ifdef CONFIG_LINUX_AIO
1137 BDRVRawState *s = bs->opaque;
1139 laio_io_unplug(bs, s->aio_ctx, false);
1144 static BlockAIOCB *raw_aio_readv(BlockDriverState *bs,
1145 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1146 BlockCompletionFunc *cb, void *opaque)
1148 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1149 cb, opaque, QEMU_AIO_READ);
1152 static BlockAIOCB *raw_aio_writev(BlockDriverState *bs,
1153 int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
1154 BlockCompletionFunc *cb, void *opaque)
1156 return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
1157 cb, opaque, QEMU_AIO_WRITE);
1160 static BlockAIOCB *raw_aio_flush(BlockDriverState *bs,
1161 BlockCompletionFunc *cb, void *opaque)
1163 BDRVRawState *s = bs->opaque;
1165 if (fd_open(bs) < 0)
1168 return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
1171 static void raw_close(BlockDriverState *bs)
1173 BDRVRawState *s = bs->opaque;
1175 raw_detach_aio_context(bs);
1177 #ifdef CONFIG_LINUX_AIO
1179 laio_cleanup(s->aio_ctx);
1188 static int raw_truncate(BlockDriverState *bs, int64_t offset)
1190 BDRVRawState *s = bs->opaque;
1193 if (fstat(s->fd, &st)) {
1197 if (S_ISREG(st.st_mode)) {
1198 if (ftruncate(s->fd, offset) < 0) {
1201 } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1202 if (offset > raw_getlength(bs)) {
1213 static int64_t raw_getlength(BlockDriverState *bs)
1215 BDRVRawState *s = bs->opaque;
1221 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1222 struct disklabel dl;
1224 if (ioctl(fd, DIOCGDINFO, &dl))
1226 return (uint64_t)dl.d_secsize *
1227 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1231 #elif defined(__NetBSD__)
1232 static int64_t raw_getlength(BlockDriverState *bs)
1234 BDRVRawState *s = bs->opaque;
1240 if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
1241 struct dkwedge_info dkw;
1243 if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
1244 return dkw.dkw_size * 512;
1246 struct disklabel dl;
1248 if (ioctl(fd, DIOCGDINFO, &dl))
1250 return (uint64_t)dl.d_secsize *
1251 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
1256 #elif defined(__sun__)
1257 static int64_t raw_getlength(BlockDriverState *bs)
1259 BDRVRawState *s = bs->opaque;
1260 struct dk_minfo minfo;
1270 * Use the DKIOCGMEDIAINFO ioctl to read the size.
1272 ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
1274 return minfo.dki_lbsize * minfo.dki_capacity;
1278 * There are reports that lseek on some devices fails, but
1279 * irc discussion said that contingency on contingency was overkill.
1281 size = lseek(s->fd, 0, SEEK_END);
1287 #elif defined(CONFIG_BSD)
1288 static int64_t raw_getlength(BlockDriverState *bs)
1290 BDRVRawState *s = bs->opaque;
1294 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1303 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1306 if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
1307 #ifdef DIOCGMEDIASIZE
1308 if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
1309 #elif defined(DIOCGPART)
1312 if (ioctl(fd, DIOCGPART, &pi) == 0)
1313 size = pi.media_size;
1319 #if defined(__APPLE__) && defined(__MACH__)
1322 size = lseek(fd, 0LL, SEEK_END);
1327 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1330 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
1331 if (size == 2048LL * (unsigned)-1)
1333 /* XXX no disc? maybe we need to reopen... */
1334 if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
1341 size = lseek(fd, 0, SEEK_END);
1349 static int64_t raw_getlength(BlockDriverState *bs)
1351 BDRVRawState *s = bs->opaque;
1360 size = lseek(s->fd, 0, SEEK_END);
1368 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1371 BDRVRawState *s = bs->opaque;
1373 if (fstat(s->fd, &st) < 0) {
1376 return (int64_t)st.st_blocks * 512;
1379 static int raw_create(const char *filename, QemuOpts *opts, Error **errp)
1383 int64_t total_size = 0;
1385 PreallocMode prealloc;
1387 Error *local_err = NULL;
1389 strstart(filename, "file:", &filename);
1391 /* Read out options */
1392 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1394 nocow = qemu_opt_get_bool(opts, BLOCK_OPT_NOCOW, false);
1395 buf = qemu_opt_get_del(opts, BLOCK_OPT_PREALLOC);
1396 prealloc = qapi_enum_parse(PreallocMode_lookup, buf,
1397 PREALLOC_MODE_MAX, PREALLOC_MODE_OFF,
1401 error_propagate(errp, local_err);
1406 fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1410 error_setg_errno(errp, -result, "Could not create file");
1416 /* Set NOCOW flag to solve performance issue on fs like btrfs.
1417 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
1418 * will be ignored since any failure of this operation should not
1419 * block the left work.
1422 if (ioctl(fd, FS_IOC_GETFLAGS, &attr) == 0) {
1423 attr |= FS_NOCOW_FL;
1424 ioctl(fd, FS_IOC_SETFLAGS, &attr);
1429 if (ftruncate(fd, total_size) != 0) {
1431 error_setg_errno(errp, -result, "Could not resize file");
1436 #ifdef CONFIG_POSIX_FALLOCATE
1437 case PREALLOC_MODE_FALLOC:
1438 /* posix_fallocate() doesn't set errno. */
1439 result = -posix_fallocate(fd, 0, total_size);
1441 error_setg_errno(errp, -result,
1442 "Could not preallocate data for the new file");
1446 case PREALLOC_MODE_FULL:
1448 int64_t num = 0, left = total_size;
1449 buf = g_malloc0(65536);
1452 num = MIN(left, 65536);
1453 result = write(fd, buf, num);
1456 error_setg_errno(errp, -result,
1457 "Could not write to the new file");
1466 case PREALLOC_MODE_OFF:
1470 error_setg(errp, "Unsupported preallocation mode: %s",
1471 PreallocMode_lookup[prealloc]);
1476 if (qemu_close(fd) != 0 && result == 0) {
1478 error_setg_errno(errp, -result, "Could not close the new file");
1484 static int try_fiemap(BlockDriverState *bs, off_t start, off_t *data,
1485 off_t *hole, int nb_sectors)
1487 #ifdef CONFIG_FIEMAP
1488 BDRVRawState *s = bs->opaque;
1492 struct fiemap_extent fe;
1495 if (s->skip_fiemap) {
1499 f.fm.fm_start = start;
1500 f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1501 f.fm.fm_flags = FIEMAP_FLAG_SYNC;
1502 f.fm.fm_extent_count = 1;
1503 f.fm.fm_reserved = 0;
1504 if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1505 s->skip_fiemap = true;
1509 if (f.fm.fm_mapped_extents == 0) {
1510 /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1511 * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1513 off_t length = lseek(s->fd, 0, SEEK_END);
1514 *hole = f.fm.fm_start;
1515 *data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1517 *data = f.fe.fe_logical;
1518 *hole = f.fe.fe_logical + f.fe.fe_length;
1519 if (f.fe.fe_flags & FIEMAP_EXTENT_UNWRITTEN) {
1520 ret |= BDRV_BLOCK_ZERO;
1530 static int try_seek_hole(BlockDriverState *bs, off_t start, off_t *data,
1533 #if defined SEEK_HOLE && defined SEEK_DATA
1534 BDRVRawState *s = bs->opaque;
1536 *hole = lseek(s->fd, start, SEEK_HOLE);
1541 if (*hole > start) {
1544 /* On a hole. We need another syscall to find its end. */
1545 *data = lseek(s->fd, start, SEEK_DATA);
1547 *data = lseek(s->fd, 0, SEEK_END);
1558 * Returns the allocation status of the specified sectors.
1560 * If 'sector_num' is beyond the end of the disk image the return value is 0
1561 * and 'pnum' is set to 0.
1563 * 'pnum' is set to the number of sectors (including and immediately following
1564 * the specified sector) that are known to be in the same
1565 * allocated/unallocated state.
1567 * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes
1568 * beyond the end of the disk image it will be clamped.
1570 static int64_t coroutine_fn raw_co_get_block_status(BlockDriverState *bs,
1572 int nb_sectors, int *pnum)
1574 off_t start, data = 0, hole = 0;
1583 start = sector_num * BDRV_SECTOR_SIZE;
1584 total_size = bdrv_getlength(bs);
1585 if (total_size < 0) {
1587 } else if (start >= total_size) {
1590 } else if (start + nb_sectors * BDRV_SECTOR_SIZE > total_size) {
1591 nb_sectors = DIV_ROUND_UP(total_size - start, BDRV_SECTOR_SIZE);
1594 ret = try_seek_hole(bs, start, &data, &hole);
1596 ret = try_fiemap(bs, start, &data, &hole, nb_sectors);
1598 /* Assume everything is allocated. */
1600 hole = start + nb_sectors * BDRV_SECTOR_SIZE;
1607 if (data <= start) {
1608 /* On a data extent, compute sectors to the end of the extent. */
1609 *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1610 return ret | BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | start;
1612 /* On a hole, compute sectors to the beginning of the next extent. */
1613 *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1614 return ret | BDRV_BLOCK_ZERO | BDRV_BLOCK_OFFSET_VALID | start;
1618 static coroutine_fn BlockAIOCB *raw_aio_discard(BlockDriverState *bs,
1619 int64_t sector_num, int nb_sectors,
1620 BlockCompletionFunc *cb, void *opaque)
1622 BDRVRawState *s = bs->opaque;
1624 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1625 cb, opaque, QEMU_AIO_DISCARD);
1628 static int coroutine_fn raw_co_write_zeroes(
1629 BlockDriverState *bs, int64_t sector_num,
1630 int nb_sectors, BdrvRequestFlags flags)
1632 BDRVRawState *s = bs->opaque;
1634 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
1635 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1636 QEMU_AIO_WRITE_ZEROES);
1637 } else if (s->discard_zeroes) {
1638 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
1644 static int raw_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1646 BDRVRawState *s = bs->opaque;
1648 bdi->unallocated_blocks_are_zero = s->discard_zeroes;
1649 bdi->can_write_zeroes_with_unmap = s->discard_zeroes;
1653 static QemuOptsList raw_create_opts = {
1654 .name = "raw-create-opts",
1655 .head = QTAILQ_HEAD_INITIALIZER(raw_create_opts.head),
1658 .name = BLOCK_OPT_SIZE,
1659 .type = QEMU_OPT_SIZE,
1660 .help = "Virtual disk size"
1663 .name = BLOCK_OPT_NOCOW,
1664 .type = QEMU_OPT_BOOL,
1665 .help = "Turn off copy-on-write (valid only on btrfs)"
1668 .name = BLOCK_OPT_PREALLOC,
1669 .type = QEMU_OPT_STRING,
1670 .help = "Preallocation mode (allowed values: off, falloc, full)"
1672 { /* end of list */ }
1676 static BlockDriver bdrv_file = {
1677 .format_name = "file",
1678 .protocol_name = "file",
1679 .instance_size = sizeof(BDRVRawState),
1680 .bdrv_needs_filename = true,
1681 .bdrv_probe = NULL, /* no probe for protocols */
1682 .bdrv_parse_filename = raw_parse_filename,
1683 .bdrv_file_open = raw_open,
1684 .bdrv_reopen_prepare = raw_reopen_prepare,
1685 .bdrv_reopen_commit = raw_reopen_commit,
1686 .bdrv_reopen_abort = raw_reopen_abort,
1687 .bdrv_close = raw_close,
1688 .bdrv_create = raw_create,
1689 .bdrv_has_zero_init = bdrv_has_zero_init_1,
1690 .bdrv_co_get_block_status = raw_co_get_block_status,
1691 .bdrv_co_write_zeroes = raw_co_write_zeroes,
1693 .bdrv_aio_readv = raw_aio_readv,
1694 .bdrv_aio_writev = raw_aio_writev,
1695 .bdrv_aio_flush = raw_aio_flush,
1696 .bdrv_aio_discard = raw_aio_discard,
1697 .bdrv_refresh_limits = raw_refresh_limits,
1698 .bdrv_io_plug = raw_aio_plug,
1699 .bdrv_io_unplug = raw_aio_unplug,
1700 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
1702 .bdrv_truncate = raw_truncate,
1703 .bdrv_getlength = raw_getlength,
1704 .bdrv_get_info = raw_get_info,
1705 .bdrv_get_allocated_file_size
1706 = raw_get_allocated_file_size,
1708 .bdrv_detach_aio_context = raw_detach_aio_context,
1709 .bdrv_attach_aio_context = raw_attach_aio_context,
1711 .create_opts = &raw_create_opts,
1714 /***********************************************/
1717 #if defined(__APPLE__) && defined(__MACH__)
1718 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1719 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1721 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1723 kern_return_t kernResult;
1724 mach_port_t masterPort;
1725 CFMutableDictionaryRef classesToMatch;
1727 kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1728 if ( KERN_SUCCESS != kernResult ) {
1729 printf( "IOMasterPort returned %d\n", kernResult );
1732 classesToMatch = IOServiceMatching( kIOCDMediaClass );
1733 if ( classesToMatch == NULL ) {
1734 printf( "IOServiceMatching returned a NULL dictionary.\n" );
1736 CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1738 kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1739 if ( KERN_SUCCESS != kernResult )
1741 printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1747 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1749 io_object_t nextMedia;
1750 kern_return_t kernResult = KERN_FAILURE;
1752 nextMedia = IOIteratorNext( mediaIterator );
1755 CFTypeRef bsdPathAsCFString;
1756 bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1757 if ( bsdPathAsCFString ) {
1758 size_t devPathLength;
1759 strcpy( bsdPath, _PATH_DEV );
1760 strcat( bsdPath, "r" );
1761 devPathLength = strlen( bsdPath );
1762 if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1763 kernResult = KERN_SUCCESS;
1765 CFRelease( bsdPathAsCFString );
1767 IOObjectRelease( nextMedia );
1775 static int hdev_probe_device(const char *filename)
1779 /* allow a dedicated CD-ROM driver to match with a higher priority */
1780 if (strstart(filename, "/dev/cdrom", NULL))
1783 if (stat(filename, &st) >= 0 &&
1784 (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1791 static int check_hdev_writable(BDRVRawState *s)
1793 #if defined(BLKROGET)
1794 /* Linux block devices can be configured "read-only" using blockdev(8).
1795 * This is independent of device node permissions and therefore open(2)
1796 * with O_RDWR succeeds. Actual writes fail with EPERM.
1798 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
1799 * check for read-only block devices so that Linux block devices behave
1805 if (fstat(s->fd, &st)) {
1809 if (!S_ISBLK(st.st_mode)) {
1813 if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1820 #endif /* defined(BLKROGET) */
1824 static void hdev_parse_filename(const char *filename, QDict *options,
1827 /* The prefix is optional, just as for "file". */
1828 strstart(filename, "host_device:", &filename);
1830 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
1833 static int hdev_open(BlockDriverState *bs, QDict *options, int flags,
1836 BDRVRawState *s = bs->opaque;
1837 Error *local_err = NULL;
1839 const char *filename = qdict_get_str(options, "filename");
1841 #if defined(__APPLE__) && defined(__MACH__)
1842 if (strstart(filename, "/dev/cdrom", NULL)) {
1843 kern_return_t kernResult;
1844 io_iterator_t mediaIterator;
1845 char bsdPath[ MAXPATHLEN ];
1848 kernResult = FindEjectableCDMedia( &mediaIterator );
1849 kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1851 if ( bsdPath[ 0 ] != '\0' ) {
1852 strcat(bsdPath,"s0");
1853 /* some CDs don't have a partition 0 */
1854 fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1856 bsdPath[strlen(bsdPath)-1] = '1';
1861 qdict_put(options, "filename", qstring_from_str(filename));
1864 if ( mediaIterator )
1865 IOObjectRelease( mediaIterator );
1869 s->type = FTYPE_FILE;
1870 #if defined(__linux__)
1872 char resolved_path[ MAXPATHLEN ], *temp;
1874 temp = realpath(filename, resolved_path);
1875 if (temp && strstart(temp, "/dev/sg", NULL)) {
1881 ret = raw_open_common(bs, options, flags, 0, &local_err);
1884 error_propagate(errp, local_err);
1889 if (flags & BDRV_O_RDWR) {
1890 ret = check_hdev_writable(s);
1893 error_setg_errno(errp, -ret, "The device is not writable");
1901 #if defined(__linux__)
1902 /* Note: we do not have a reliable method to detect if the floppy is
1903 present. The current method is to try to open the floppy at every
1904 I/O and to keep it opened during a few hundreds of ms. */
1905 static int fd_open(BlockDriverState *bs)
1907 BDRVRawState *s = bs->opaque;
1908 int last_media_present;
1910 if (s->type != FTYPE_FD)
1912 last_media_present = (s->fd >= 0);
1914 (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1918 printf("Floppy closed\n");
1922 if (s->fd_got_error &&
1923 (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1925 printf("No floppy (open delayed)\n");
1929 s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1931 s->fd_error_time = get_clock();
1932 s->fd_got_error = 1;
1933 if (last_media_present)
1934 s->fd_media_changed = 1;
1936 printf("No floppy\n");
1941 printf("Floppy opened\n");
1944 if (!last_media_present)
1945 s->fd_media_changed = 1;
1946 s->fd_open_time = get_clock();
1947 s->fd_got_error = 0;
1951 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1953 BDRVRawState *s = bs->opaque;
1955 return ioctl(s->fd, req, buf);
1958 static BlockAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1959 unsigned long int req, void *buf,
1960 BlockCompletionFunc *cb, void *opaque)
1962 BDRVRawState *s = bs->opaque;
1963 RawPosixAIOData *acb;
1966 if (fd_open(bs) < 0)
1969 acb = g_slice_new(RawPosixAIOData);
1971 acb->aio_type = QEMU_AIO_IOCTL;
1972 acb->aio_fildes = s->fd;
1973 acb->aio_offset = 0;
1974 acb->aio_ioctl_buf = buf;
1975 acb->aio_ioctl_cmd = req;
1976 pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1977 return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1980 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1981 static int fd_open(BlockDriverState *bs)
1983 BDRVRawState *s = bs->opaque;
1985 /* this is just to ensure s->fd is sane (its called by io ops) */
1990 #else /* !linux && !FreeBSD */
1992 static int fd_open(BlockDriverState *bs)
1997 #endif /* !linux && !FreeBSD */
1999 static coroutine_fn BlockAIOCB *hdev_aio_discard(BlockDriverState *bs,
2000 int64_t sector_num, int nb_sectors,
2001 BlockCompletionFunc *cb, void *opaque)
2003 BDRVRawState *s = bs->opaque;
2005 if (fd_open(bs) < 0) {
2008 return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
2009 cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2012 static coroutine_fn int hdev_co_write_zeroes(BlockDriverState *bs,
2013 int64_t sector_num, int nb_sectors, BdrvRequestFlags flags)
2015 BDRVRawState *s = bs->opaque;
2022 if (!(flags & BDRV_REQ_MAY_UNMAP)) {
2023 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2024 QEMU_AIO_WRITE_ZEROES|QEMU_AIO_BLKDEV);
2025 } else if (s->discard_zeroes) {
2026 return paio_submit_co(bs, s->fd, sector_num, NULL, nb_sectors,
2027 QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
2032 static int hdev_create(const char *filename, QemuOpts *opts,
2037 struct stat stat_buf;
2038 int64_t total_size = 0;
2041 /* This function is used by all three protocol block drivers and therefore
2042 * any of these three prefixes may be given.
2043 * The return value has to be stored somewhere, otherwise this is an error
2044 * due to -Werror=unused-value. */
2046 strstart(filename, "host_device:", &filename) ||
2047 strstart(filename, "host_cdrom:" , &filename) ||
2048 strstart(filename, "host_floppy:", &filename);
2052 /* Read out options */
2053 total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
2056 fd = qemu_open(filename, O_WRONLY | O_BINARY);
2059 error_setg_errno(errp, -ret, "Could not open device");
2063 if (fstat(fd, &stat_buf) < 0) {
2065 error_setg_errno(errp, -ret, "Could not stat device");
2066 } else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode)) {
2068 "The given file is neither a block nor a character device");
2070 } else if (lseek(fd, 0, SEEK_END) < total_size) {
2071 error_setg(errp, "Device is too small");
2079 static BlockDriver bdrv_host_device = {
2080 .format_name = "host_device",
2081 .protocol_name = "host_device",
2082 .instance_size = sizeof(BDRVRawState),
2083 .bdrv_needs_filename = true,
2084 .bdrv_probe_device = hdev_probe_device,
2085 .bdrv_parse_filename = hdev_parse_filename,
2086 .bdrv_file_open = hdev_open,
2087 .bdrv_close = raw_close,
2088 .bdrv_reopen_prepare = raw_reopen_prepare,
2089 .bdrv_reopen_commit = raw_reopen_commit,
2090 .bdrv_reopen_abort = raw_reopen_abort,
2091 .bdrv_create = hdev_create,
2092 .create_opts = &raw_create_opts,
2093 .bdrv_co_write_zeroes = hdev_co_write_zeroes,
2095 .bdrv_aio_readv = raw_aio_readv,
2096 .bdrv_aio_writev = raw_aio_writev,
2097 .bdrv_aio_flush = raw_aio_flush,
2098 .bdrv_aio_discard = hdev_aio_discard,
2099 .bdrv_refresh_limits = raw_refresh_limits,
2100 .bdrv_io_plug = raw_aio_plug,
2101 .bdrv_io_unplug = raw_aio_unplug,
2102 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2104 .bdrv_truncate = raw_truncate,
2105 .bdrv_getlength = raw_getlength,
2106 .bdrv_get_info = raw_get_info,
2107 .bdrv_get_allocated_file_size
2108 = raw_get_allocated_file_size,
2110 .bdrv_detach_aio_context = raw_detach_aio_context,
2111 .bdrv_attach_aio_context = raw_attach_aio_context,
2113 /* generic scsi device */
2115 .bdrv_ioctl = hdev_ioctl,
2116 .bdrv_aio_ioctl = hdev_aio_ioctl,
2121 static void floppy_parse_filename(const char *filename, QDict *options,
2124 /* The prefix is optional, just as for "file". */
2125 strstart(filename, "host_floppy:", &filename);
2127 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2130 static int floppy_open(BlockDriverState *bs, QDict *options, int flags,
2133 BDRVRawState *s = bs->opaque;
2134 Error *local_err = NULL;
2139 /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
2140 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2143 error_propagate(errp, local_err);
2148 /* close fd so that we can reopen it as needed */
2151 s->fd_media_changed = 1;
2156 static int floppy_probe_device(const char *filename)
2160 struct floppy_struct fdparam;
2163 if (strstart(filename, "/dev/fd", NULL) &&
2164 !strstart(filename, "/dev/fdset/", NULL)) {
2168 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2172 ret = fstat(fd, &st);
2173 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2177 /* Attempt to detect via a floppy specific ioctl */
2178 ret = ioctl(fd, FDGETPRM, &fdparam);
2189 static int floppy_is_inserted(BlockDriverState *bs)
2191 return fd_open(bs) >= 0;
2194 static int floppy_media_changed(BlockDriverState *bs)
2196 BDRVRawState *s = bs->opaque;
2200 * XXX: we do not have a true media changed indication.
2201 * It does not work if the floppy is changed without trying to read it.
2204 ret = s->fd_media_changed;
2205 s->fd_media_changed = 0;
2207 printf("Floppy changed=%d\n", ret);
2212 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
2214 BDRVRawState *s = bs->opaque;
2221 fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
2223 if (ioctl(fd, FDEJECT, 0) < 0)
2229 static BlockDriver bdrv_host_floppy = {
2230 .format_name = "host_floppy",
2231 .protocol_name = "host_floppy",
2232 .instance_size = sizeof(BDRVRawState),
2233 .bdrv_needs_filename = true,
2234 .bdrv_probe_device = floppy_probe_device,
2235 .bdrv_parse_filename = floppy_parse_filename,
2236 .bdrv_file_open = floppy_open,
2237 .bdrv_close = raw_close,
2238 .bdrv_reopen_prepare = raw_reopen_prepare,
2239 .bdrv_reopen_commit = raw_reopen_commit,
2240 .bdrv_reopen_abort = raw_reopen_abort,
2241 .bdrv_create = hdev_create,
2242 .create_opts = &raw_create_opts,
2244 .bdrv_aio_readv = raw_aio_readv,
2245 .bdrv_aio_writev = raw_aio_writev,
2246 .bdrv_aio_flush = raw_aio_flush,
2247 .bdrv_refresh_limits = raw_refresh_limits,
2248 .bdrv_io_plug = raw_aio_plug,
2249 .bdrv_io_unplug = raw_aio_unplug,
2250 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2252 .bdrv_truncate = raw_truncate,
2253 .bdrv_getlength = raw_getlength,
2254 .has_variable_length = true,
2255 .bdrv_get_allocated_file_size
2256 = raw_get_allocated_file_size,
2258 .bdrv_detach_aio_context = raw_detach_aio_context,
2259 .bdrv_attach_aio_context = raw_attach_aio_context,
2261 /* removable device support */
2262 .bdrv_is_inserted = floppy_is_inserted,
2263 .bdrv_media_changed = floppy_media_changed,
2264 .bdrv_eject = floppy_eject,
2268 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2269 static void cdrom_parse_filename(const char *filename, QDict *options,
2272 /* The prefix is optional, just as for "file". */
2273 strstart(filename, "host_cdrom:", &filename);
2275 qdict_put_obj(options, "filename", QOBJECT(qstring_from_str(filename)));
2280 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2283 BDRVRawState *s = bs->opaque;
2284 Error *local_err = NULL;
2289 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
2290 ret = raw_open_common(bs, options, flags, O_NONBLOCK, &local_err);
2292 error_propagate(errp, local_err);
2297 static int cdrom_probe_device(const char *filename)
2303 fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
2307 ret = fstat(fd, &st);
2308 if (ret == -1 || !S_ISBLK(st.st_mode)) {
2312 /* Attempt to detect via a CDROM specific ioctl */
2313 ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2323 static int cdrom_is_inserted(BlockDriverState *bs)
2325 BDRVRawState *s = bs->opaque;
2328 ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
2329 if (ret == CDS_DISC_OK)
2334 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2336 BDRVRawState *s = bs->opaque;
2339 if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
2340 perror("CDROMEJECT");
2342 if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
2343 perror("CDROMEJECT");
2347 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2349 BDRVRawState *s = bs->opaque;
2351 if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
2353 * Note: an error can happen if the distribution automatically
2356 /* perror("CDROM_LOCKDOOR"); */
2360 static BlockDriver bdrv_host_cdrom = {
2361 .format_name = "host_cdrom",
2362 .protocol_name = "host_cdrom",
2363 .instance_size = sizeof(BDRVRawState),
2364 .bdrv_needs_filename = true,
2365 .bdrv_probe_device = cdrom_probe_device,
2366 .bdrv_parse_filename = cdrom_parse_filename,
2367 .bdrv_file_open = cdrom_open,
2368 .bdrv_close = raw_close,
2369 .bdrv_reopen_prepare = raw_reopen_prepare,
2370 .bdrv_reopen_commit = raw_reopen_commit,
2371 .bdrv_reopen_abort = raw_reopen_abort,
2372 .bdrv_create = hdev_create,
2373 .create_opts = &raw_create_opts,
2375 .bdrv_aio_readv = raw_aio_readv,
2376 .bdrv_aio_writev = raw_aio_writev,
2377 .bdrv_aio_flush = raw_aio_flush,
2378 .bdrv_refresh_limits = raw_refresh_limits,
2379 .bdrv_io_plug = raw_aio_plug,
2380 .bdrv_io_unplug = raw_aio_unplug,
2381 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2383 .bdrv_truncate = raw_truncate,
2384 .bdrv_getlength = raw_getlength,
2385 .has_variable_length = true,
2386 .bdrv_get_allocated_file_size
2387 = raw_get_allocated_file_size,
2389 .bdrv_detach_aio_context = raw_detach_aio_context,
2390 .bdrv_attach_aio_context = raw_attach_aio_context,
2392 /* removable device support */
2393 .bdrv_is_inserted = cdrom_is_inserted,
2394 .bdrv_eject = cdrom_eject,
2395 .bdrv_lock_medium = cdrom_lock_medium,
2397 /* generic scsi device */
2398 .bdrv_ioctl = hdev_ioctl,
2399 .bdrv_aio_ioctl = hdev_aio_ioctl,
2401 #endif /* __linux__ */
2403 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2404 static int cdrom_open(BlockDriverState *bs, QDict *options, int flags,
2407 BDRVRawState *s = bs->opaque;
2408 Error *local_err = NULL;
2413 ret = raw_open_common(bs, options, flags, 0, &local_err);
2416 error_propagate(errp, local_err);
2421 /* make sure the door isn't locked at this time */
2422 ioctl(s->fd, CDIOCALLOW);
2426 static int cdrom_probe_device(const char *filename)
2428 if (strstart(filename, "/dev/cd", NULL) ||
2429 strstart(filename, "/dev/acd", NULL))
2434 static int cdrom_reopen(BlockDriverState *bs)
2436 BDRVRawState *s = bs->opaque;
2440 * Force reread of possibly changed/newly loaded disc,
2441 * FreeBSD seems to not notice sometimes...
2445 fd = qemu_open(bs->filename, s->open_flags, 0644);
2452 /* make sure the door isn't locked at this time */
2453 ioctl(s->fd, CDIOCALLOW);
2457 static int cdrom_is_inserted(BlockDriverState *bs)
2459 return raw_getlength(bs) > 0;
2462 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
2464 BDRVRawState *s = bs->opaque;
2469 (void) ioctl(s->fd, CDIOCALLOW);
2472 if (ioctl(s->fd, CDIOCEJECT) < 0)
2473 perror("CDIOCEJECT");
2475 if (ioctl(s->fd, CDIOCCLOSE) < 0)
2476 perror("CDIOCCLOSE");
2482 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
2484 BDRVRawState *s = bs->opaque;
2488 if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
2490 * Note: an error can happen if the distribution automatically
2493 /* perror("CDROM_LOCKDOOR"); */
2497 static BlockDriver bdrv_host_cdrom = {
2498 .format_name = "host_cdrom",
2499 .protocol_name = "host_cdrom",
2500 .instance_size = sizeof(BDRVRawState),
2501 .bdrv_needs_filename = true,
2502 .bdrv_probe_device = cdrom_probe_device,
2503 .bdrv_parse_filename = cdrom_parse_filename,
2504 .bdrv_file_open = cdrom_open,
2505 .bdrv_close = raw_close,
2506 .bdrv_reopen_prepare = raw_reopen_prepare,
2507 .bdrv_reopen_commit = raw_reopen_commit,
2508 .bdrv_reopen_abort = raw_reopen_abort,
2509 .bdrv_create = hdev_create,
2510 .create_opts = &raw_create_opts,
2512 .bdrv_aio_readv = raw_aio_readv,
2513 .bdrv_aio_writev = raw_aio_writev,
2514 .bdrv_aio_flush = raw_aio_flush,
2515 .bdrv_refresh_limits = raw_refresh_limits,
2516 .bdrv_io_plug = raw_aio_plug,
2517 .bdrv_io_unplug = raw_aio_unplug,
2518 .bdrv_flush_io_queue = raw_aio_flush_io_queue,
2520 .bdrv_truncate = raw_truncate,
2521 .bdrv_getlength = raw_getlength,
2522 .has_variable_length = true,
2523 .bdrv_get_allocated_file_size
2524 = raw_get_allocated_file_size,
2526 .bdrv_detach_aio_context = raw_detach_aio_context,
2527 .bdrv_attach_aio_context = raw_attach_aio_context,
2529 /* removable device support */
2530 .bdrv_is_inserted = cdrom_is_inserted,
2531 .bdrv_eject = cdrom_eject,
2532 .bdrv_lock_medium = cdrom_lock_medium,
2534 #endif /* __FreeBSD__ */
2536 static void bdrv_file_init(void)
2539 * Register all the drivers. Note that order is important, the driver
2540 * registered last will get probed first.
2542 bdrv_register(&bdrv_file);
2543 bdrv_register(&bdrv_host_device);
2545 bdrv_register(&bdrv_host_floppy);
2546 bdrv_register(&bdrv_host_cdrom);
2548 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2549 bdrv_register(&bdrv_host_cdrom);
2553 block_init(bdrv_file_init);