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
25 #include "qemu/osdep.h"
26 #include "qapi/error.h"
27 #include "qemu/cutils.h"
28 #include "qemu/error-report.h"
29 #include "block/block-io.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
33 #include "qemu/units.h"
34 #include "qemu/memalign.h"
36 #include "block/thread-pool.h"
38 #include "block/raw-aio.h"
39 #include "qapi/qmp/qdict.h"
40 #include "qapi/qmp/qstring.h"
42 #include "scsi/pr-manager.h"
43 #include "scsi/constants.h"
45 #if defined(__APPLE__) && (__MACH__)
46 #include <sys/ioctl.h>
47 #if defined(HAVE_HOST_BLOCK_DEVICE)
49 #include <sys/param.h>
50 #include <sys/mount.h>
51 #include <IOKit/IOKitLib.h>
52 #include <IOKit/IOBSD.h>
53 #include <IOKit/storage/IOMediaBSDClient.h>
54 #include <IOKit/storage/IOMedia.h>
55 #include <IOKit/storage/IOCDMedia.h>
56 //#include <IOKit/storage/IOCDTypes.h>
57 #include <IOKit/storage/IODVDMedia.h>
58 #include <CoreFoundation/CoreFoundation.h>
59 #endif /* defined(HAVE_HOST_BLOCK_DEVICE) */
63 #define _POSIX_PTHREAD_SEMANTICS 1
67 #include <sys/ioctl.h>
68 #include <sys/param.h>
69 #include <sys/syscall.h>
71 #if defined(CONFIG_BLKZONED)
72 #include <linux/blkzoned.h>
74 #include <linux/cdrom.h>
77 #include <linux/hdreg.h>
78 #include <linux/magic.h>
84 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
87 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
88 #include <linux/falloc.h>
90 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
96 #include <sys/ioctl.h>
97 #include <sys/disklabel.h>
102 #include <sys/ioctl.h>
103 #include <sys/disklabel.h>
104 #include <sys/dkio.h>
105 #include <sys/disk.h>
109 #include <sys/ioctl.h>
110 #include <sys/diskslice.h>
113 /* OS X does not have O_DSYNC */
116 #define O_DSYNC O_SYNC
117 #elif defined(O_FSYNC)
118 #define O_DSYNC O_FSYNC
122 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
124 #define O_DIRECT O_DSYNC
130 #define MAX_BLOCKSIZE 4096
132 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
133 * leaving a few more bytes for its future use. */
134 #define RAW_LOCK_PERM_BASE 100
135 #define RAW_LOCK_SHARED_BASE 200
137 typedef struct BDRVRawState
{
144 /* The current permissions. */
146 uint64_t shared_perm
;
148 /* The perms bits whose corresponding bytes are already locked in
150 uint64_t locked_perm
;
151 uint64_t locked_shared_perm
;
153 uint64_t aio_max_batch
;
156 int perm_change_flags
;
157 BDRVReopenState
*reopen_state
;
160 bool has_write_zeroes
:1;
161 bool use_linux_aio
:1;
162 bool use_linux_io_uring
:1;
163 int page_cache_inconsistent
; /* errno from fdatasync failure */
165 bool needs_alignment
;
166 bool force_alignment
;
168 bool check_cache_dropped
;
170 uint64_t discard_nb_ok
;
171 uint64_t discard_nb_failed
;
172 uint64_t discard_bytes_ok
;
178 typedef struct BDRVRawReopenState
{
181 bool check_cache_dropped
;
182 } BDRVRawReopenState
;
184 static int fd_open(BlockDriverState
*bs
)
186 BDRVRawState
*s
= bs
->opaque
;
188 /* this is just to ensure s->fd is sane (its called by io ops) */
195 static int64_t coroutine_fn
raw_co_getlength(BlockDriverState
*bs
);
197 typedef struct RawPosixAIOData
{
198 BlockDriverState
*bs
;
219 PreallocMode prealloc
;
223 unsigned int *nr_zones
;
224 BlockZoneDescriptor
*zones
;
232 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
233 static int cdrom_reopen(BlockDriverState
*bs
);
237 * Elide EAGAIN and EACCES details when failing to lock, as this
238 * indicates that the specified file region is already locked by
239 * another process, which is considered a common scenario.
241 #define raw_lock_error_setg_errno(errp, err, fmt, ...) \
243 if ((err) == EAGAIN || (err) == EACCES) { \
244 error_setg((errp), (fmt), ## __VA_ARGS__); \
246 error_setg_errno((errp), (err), (fmt), ## __VA_ARGS__); \
250 #if defined(__NetBSD__)
251 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
253 static char namebuf
[PATH_MAX
];
254 const char *dp
, *fname
;
258 dp
= strrchr(fname
, '/');
259 if (lstat(fname
, &sb
) < 0) {
260 error_setg_file_open(errp
, errno
, fname
);
264 if (!S_ISBLK(sb
.st_mode
)) {
269 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
271 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
272 (int)(dp
- fname
), fname
, dp
+ 1);
275 warn_report("%s is a block device, using %s", fname
, *filename
);
280 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
287 * Get logical block size via ioctl. On success store it in @sector_size_p.
289 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
291 unsigned int sector_size
;
292 bool success
= false;
296 static const unsigned long ioctl_list
[] = {
300 #ifdef DKIOCGETBLOCKSIZE
303 #ifdef DIOCGSECTORSIZE
308 /* Try a few ioctls to get the right size */
309 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
310 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
311 *sector_size_p
= sector_size
;
316 return success
? 0 : -errno
;
320 * Get physical block size of @fd.
321 * On success, store it in @blk_size and return 0.
322 * On failure, return -errno.
324 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
327 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
337 * Returns true if no alignment restrictions are necessary even for files
338 * opened with O_DIRECT.
340 * raw_probe_alignment() probes the required alignment and assume that 1 means
341 * the probing failed, so it falls back to a safe default of 4k. This can be
342 * avoided if we know that byte alignment is okay for the file.
344 static bool dio_byte_aligned(int fd
)
350 ret
= fstatfs(fd
, &buf
);
351 if (ret
== 0 && buf
.f_type
== NFS_SUPER_MAGIC
) {
358 static bool raw_needs_alignment(BlockDriverState
*bs
)
360 BDRVRawState
*s
= bs
->opaque
;
362 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0 && !dio_byte_aligned(s
->fd
)) {
366 return s
->force_alignment
;
369 /* Check if read is allowed with given memory buffer and length.
371 * This function is used to check O_DIRECT memory buffer and request alignment.
373 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
375 ssize_t ret
= pread(fd
, buf
, len
, 0);
382 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
383 * other errors (e.g. real I/O error), which could happen on a failed
384 * drive, since we only care about probing alignment.
386 if (errno
!= EINVAL
) {
394 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
396 BDRVRawState
*s
= bs
->opaque
;
398 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size());
399 size_t alignments
[] = {1, 512, 1024, 2048, 4096};
401 /* For SCSI generic devices the alignment is not really used.
402 With buffered I/O, we don't have any restrictions. */
403 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
404 bs
->bl
.request_alignment
= 1;
409 bs
->bl
.request_alignment
= 0;
411 /* Let's try to use the logical blocksize for the alignment. */
412 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
413 bs
->bl
.request_alignment
= 0;
418 * The XFS ioctl definitions are shipped in extra packages that might
419 * not always be available. Since we just need the XFS_IOC_DIOINFO ioctl
420 * here, we simply use our own definition instead:
427 if (ioctl(fd
, _IOR('X', 30, struct xfs_dioattr
), &da
) >= 0) {
428 bs
->bl
.request_alignment
= da
.d_miniosz
;
429 /* The kernel returns wrong information for d_mem */
430 /* s->buf_align = da.d_mem; */
435 * If we could not get the sizes so far, we can only guess them. First try
436 * to detect request alignment, since it is more likely to succeed. Then
437 * try to detect buf_align, which cannot be detected in some cases (e.g.
438 * Gluster). If buf_align cannot be detected, we fallback to the value of
442 if (!bs
->bl
.request_alignment
) {
445 buf
= qemu_memalign(max_align
, max_align
);
446 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
447 align
= alignments
[i
];
448 if (raw_is_io_aligned(fd
, buf
, align
)) {
449 /* Fallback to safe value. */
450 bs
->bl
.request_alignment
= (align
!= 1) ? align
: max_align
;
460 buf
= qemu_memalign(max_align
, 2 * max_align
);
461 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
462 align
= alignments
[i
];
463 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
464 /* Fallback to request_alignment. */
465 s
->buf_align
= (align
!= 1) ? align
: bs
->bl
.request_alignment
;
472 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
473 error_setg(errp
, "Could not find working O_DIRECT alignment");
474 error_append_hint(errp
, "Try cache.direct=off\n");
478 static int check_hdev_writable(int fd
)
480 #if defined(BLKROGET)
481 /* Linux block devices can be configured "read-only" using blockdev(8).
482 * This is independent of device node permissions and therefore open(2)
483 * with O_RDWR succeeds. Actual writes fail with EPERM.
485 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
486 * check for read-only block devices so that Linux block devices behave
492 if (fstat(fd
, &st
)) {
496 if (!S_ISBLK(st
.st_mode
)) {
500 if (ioctl(fd
, BLKROGET
, &readonly
) < 0) {
507 #endif /* defined(BLKROGET) */
511 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
513 bool read_write
= false;
514 assert(open_flags
!= NULL
);
516 *open_flags
|= O_BINARY
;
517 *open_flags
&= ~O_ACCMODE
;
519 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
520 read_write
= has_writers
;
521 } else if (bdrv_flags
& BDRV_O_RDWR
) {
526 *open_flags
|= O_RDWR
;
528 *open_flags
|= O_RDONLY
;
531 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
532 * and O_DIRECT for no caching. */
533 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
534 *open_flags
|= O_DIRECT
;
538 static void raw_parse_filename(const char *filename
, QDict
*options
,
541 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
544 static QemuOptsList raw_runtime_opts
= {
546 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
550 .type
= QEMU_OPT_STRING
,
551 .help
= "File name of the image",
555 .type
= QEMU_OPT_STRING
,
556 .help
= "host AIO implementation (threads, native, io_uring)",
559 .name
= "aio-max-batch",
560 .type
= QEMU_OPT_NUMBER
,
561 .help
= "AIO max batch size (0 = auto handled by AIO backend, default: 0)",
565 .type
= QEMU_OPT_STRING
,
566 .help
= "file locking mode (on/off/auto, default: auto)",
569 .name
= "pr-manager",
570 .type
= QEMU_OPT_STRING
,
571 .help
= "id of persistent reservation manager object (default: none)",
573 #if defined(__linux__)
575 .name
= "drop-cache",
576 .type
= QEMU_OPT_BOOL
,
577 .help
= "invalidate page cache during live migration (default: on)",
581 .name
= "x-check-cache-dropped",
582 .type
= QEMU_OPT_BOOL
,
583 .help
= "check that page cache was dropped on live migration (default: off)"
585 { /* end of list */ }
589 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
591 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
592 int bdrv_flags
, int open_flags
,
593 bool device
, Error
**errp
)
595 BDRVRawState
*s
= bs
->opaque
;
597 Error
*local_err
= NULL
;
598 const char *filename
= NULL
;
600 BlockdevAioOptions aio
, aio_default
;
605 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
606 if (!qemu_opts_absorb_qdict(opts
, options
, errp
)) {
611 filename
= qemu_opt_get(opts
, "filename");
613 ret
= raw_normalize_devicepath(&filename
, errp
);
618 if (bdrv_flags
& BDRV_O_NATIVE_AIO
) {
619 aio_default
= BLOCKDEV_AIO_OPTIONS_NATIVE
;
620 #ifdef CONFIG_LINUX_IO_URING
621 } else if (bdrv_flags
& BDRV_O_IO_URING
) {
622 aio_default
= BLOCKDEV_AIO_OPTIONS_IO_URING
;
625 aio_default
= BLOCKDEV_AIO_OPTIONS_THREADS
;
628 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
629 qemu_opt_get(opts
, "aio"),
630 aio_default
, &local_err
);
632 error_propagate(errp
, local_err
);
637 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
638 #ifdef CONFIG_LINUX_IO_URING
639 s
->use_linux_io_uring
= (aio
== BLOCKDEV_AIO_OPTIONS_IO_URING
);
642 s
->aio_max_batch
= qemu_opt_get_number(opts
, "aio-max-batch", 0);
644 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
645 qemu_opt_get(opts
, "locking"),
646 ON_OFF_AUTO_AUTO
, &local_err
);
648 error_propagate(errp
, local_err
);
655 if (!qemu_has_ofd_lock()) {
656 warn_report("File lock requested but OFD locking syscall is "
657 "unavailable, falling back to POSIX file locks");
658 error_printf("Due to the implementation, locks can be lost "
662 case ON_OFF_AUTO_OFF
:
665 case ON_OFF_AUTO_AUTO
:
666 s
->use_lock
= qemu_has_ofd_lock();
672 str
= qemu_opt_get(opts
, "pr-manager");
674 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
676 error_propagate(errp
, local_err
);
682 s
->drop_cache
= qemu_opt_get_bool(opts
, "drop-cache", true);
683 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
686 s
->open_flags
= open_flags
;
687 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
690 fd
= qemu_open(filename
, s
->open_flags
, errp
);
691 ret
= fd
< 0 ? -errno
: 0;
701 /* Check s->open_flags rather than bdrv_flags due to auto-read-only */
702 if (s
->open_flags
& O_RDWR
) {
703 ret
= check_hdev_writable(s
->fd
);
705 error_setg_errno(errp
, -ret
, "The device is not writable");
711 s
->shared_perm
= BLK_PERM_ALL
;
713 #ifdef CONFIG_LINUX_AIO
714 /* Currently Linux does AIO only for files opened with O_DIRECT */
715 if (s
->use_linux_aio
) {
716 if (!(s
->open_flags
& O_DIRECT
)) {
717 error_setg(errp
, "aio=native was specified, but it requires "
718 "cache.direct=on, which was not specified.");
722 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
723 error_prepend(errp
, "Unable to use native AIO: ");
728 if (s
->use_linux_aio
) {
729 error_setg(errp
, "aio=native was specified, but is not supported "
734 #endif /* !defined(CONFIG_LINUX_AIO) */
736 #ifdef CONFIG_LINUX_IO_URING
737 if (s
->use_linux_io_uring
) {
738 if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs
), errp
)) {
739 error_prepend(errp
, "Unable to use io_uring: ");
744 if (s
->use_linux_io_uring
) {
745 error_setg(errp
, "aio=io_uring was specified, but is not supported "
750 #endif /* !defined(CONFIG_LINUX_IO_URING) */
752 s
->has_discard
= true;
753 s
->has_write_zeroes
= true;
755 if (fstat(s
->fd
, &st
) < 0) {
757 error_setg_errno(errp
, errno
, "Could not stat file");
762 if (!S_ISREG(st
.st_mode
)) {
763 error_setg(errp
, "'%s' driver requires '%s' to be a regular file",
764 bs
->drv
->format_name
, bs
->filename
);
768 s
->has_fallocate
= true;
771 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
772 error_setg(errp
, "'%s' driver requires '%s' to be either "
773 "a character or block device",
774 bs
->drv
->format_name
, bs
->filename
);
779 #ifdef CONFIG_BLKZONED
781 * The kernel page cache does not reliably work for writes to SWR zones
782 * of zoned block device because it can not guarantee the order of writes.
784 if ((bs
->bl
.zoned
!= BLK_Z_NONE
) &&
785 (!(s
->open_flags
& O_DIRECT
))) {
786 error_setg(errp
, "The driver supports zoned devices, and it requires "
787 "cache.direct=on, which was not specified.");
788 return -EINVAL
; /* No host kernel page cache */
792 if (S_ISBLK(st
.st_mode
)) {
794 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
795 * not rely on the contents of discarded blocks unless using O_DIRECT.
796 * Same for BLKZEROOUT.
798 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
799 s
->has_write_zeroes
= false;
804 if (S_ISCHR(st
.st_mode
)) {
806 * The file is a char device (disk), which on FreeBSD isn't behind
807 * a pager, so force all requests to be aligned. This is needed
808 * so QEMU makes sure all IO operations on the device are aligned
809 * to sector size, or else FreeBSD will reject them with EINVAL.
811 s
->force_alignment
= true;
814 s
->needs_alignment
= raw_needs_alignment(bs
);
816 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
817 if (S_ISREG(st
.st_mode
)) {
818 /* When extending regular files, we get zeros from the OS */
819 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
823 if (ret
< 0 && s
->fd
!= -1) {
826 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
833 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
836 BDRVRawState
*s
= bs
->opaque
;
838 s
->type
= FTYPE_FILE
;
839 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
848 #define PERM_FOREACH(i) \
849 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
851 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
852 * file; if @unlock == true, also unlock the unneeded bytes.
853 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
855 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
856 uint64_t perm_lock_bits
,
857 uint64_t shared_perm_lock_bits
,
858 bool unlock
, Error
**errp
)
862 uint64_t locked_perm
, locked_shared_perm
;
865 locked_perm
= s
->locked_perm
;
866 locked_shared_perm
= s
->locked_shared_perm
;
869 * We don't have the previous bits, just lock/unlock for each of the
873 locked_perm
= BLK_PERM_ALL
;
874 locked_shared_perm
= BLK_PERM_ALL
;
877 locked_shared_perm
= 0;
882 int off
= RAW_LOCK_PERM_BASE
+ i
;
883 uint64_t bit
= (1ULL << i
);
884 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
885 ret
= qemu_lock_fd(fd
, off
, 1, false);
887 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
891 s
->locked_perm
|= bit
;
893 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
894 ret
= qemu_unlock_fd(fd
, off
, 1);
896 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
899 s
->locked_perm
&= ~bit
;
904 int off
= RAW_LOCK_SHARED_BASE
+ i
;
905 uint64_t bit
= (1ULL << i
);
906 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
907 ret
= qemu_lock_fd(fd
, off
, 1, false);
909 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
913 s
->locked_shared_perm
|= bit
;
915 } else if (unlock
&& (locked_shared_perm
& bit
) &&
916 !(shared_perm_lock_bits
& bit
)) {
917 ret
= qemu_unlock_fd(fd
, off
, 1);
919 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
922 s
->locked_shared_perm
&= ~bit
;
929 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
930 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
937 int off
= RAW_LOCK_SHARED_BASE
+ i
;
938 uint64_t p
= 1ULL << i
;
940 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
942 char *perm_name
= bdrv_perm_names(p
);
944 raw_lock_error_setg_errno(errp
, -ret
,
945 "Failed to get \"%s\" lock",
953 int off
= RAW_LOCK_PERM_BASE
+ i
;
954 uint64_t p
= 1ULL << i
;
955 if (!(shared_perm
& p
)) {
956 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
958 char *perm_name
= bdrv_perm_names(p
);
960 raw_lock_error_setg_errno(errp
, -ret
,
961 "Failed to get shared \"%s\" lock",
971 static int raw_handle_perm_lock(BlockDriverState
*bs
,
973 uint64_t new_perm
, uint64_t new_shared
,
976 BDRVRawState
*s
= bs
->opaque
;
978 Error
*local_err
= NULL
;
984 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
990 if ((s
->perm
| new_perm
) == s
->perm
&&
991 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
994 * We are going to unlock bytes, it should not fail. If it fail due
995 * to some fs-dependent permission-unrelated reasons (which occurs
996 * sometimes on NFS and leads to abort in bdrv_replace_child) we
997 * can't prevent such errors by any check here. And we ignore them
998 * anyway in ABORT and COMMIT.
1002 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
1003 ~s
->shared_perm
| ~new_shared
,
1006 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
1010 error_append_hint(errp
,
1011 "Is another process using the image [%s]?\n",
1014 /* fall through to unlock bytes. */
1016 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
1019 /* Theoretically the above call only unlocks bytes and it cannot
1020 * fail. Something weird happened, report it.
1022 warn_report_err(local_err
);
1026 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
1029 /* Theoretically the above call only unlocks bytes and it cannot
1030 * fail. Something weird happened, report it.
1032 warn_report_err(local_err
);
1039 /* Sets a specific flag */
1040 static int fcntl_setfl(int fd
, int flag
)
1044 flags
= fcntl(fd
, F_GETFL
);
1048 if (fcntl(fd
, F_SETFL
, flags
| flag
) == -1) {
1054 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
1055 int *open_flags
, uint64_t perm
, bool force_dup
,
1058 BDRVRawState
*s
= bs
->opaque
;
1061 bool has_writers
= perm
&
1062 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
1063 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
1065 fcntl_flags
|= O_NOATIME
;
1069 if (s
->type
== FTYPE_CD
) {
1070 *open_flags
|= O_NONBLOCK
;
1073 raw_parse_flags(flags
, open_flags
, has_writers
);
1076 /* Not all operating systems have O_ASYNC, and those that don't
1077 * will not let us track the state into rs->open_flags (typically
1078 * you achieve the same effect with an ioctl, for example I_SETSIG
1079 * on Solaris). But we do not use O_ASYNC, so that's fine.
1081 assert((s
->open_flags
& O_ASYNC
) == 0);
1084 if (!force_dup
&& *open_flags
== s
->open_flags
) {
1085 /* We're lucky, the existing fd is fine */
1089 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
1090 /* dup the original fd */
1091 fd
= qemu_dup(s
->fd
);
1093 ret
= fcntl_setfl(fd
, *open_flags
);
1101 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
1103 const char *normalized_filename
= bs
->filename
;
1104 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
1106 fd
= qemu_open(normalized_filename
, *open_flags
, errp
);
1113 if (fd
!= -1 && (*open_flags
& O_RDWR
)) {
1114 ret
= check_hdev_writable(fd
);
1117 error_setg_errno(errp
, -ret
, "The device is not writable");
1125 static int raw_reopen_prepare(BDRVReopenState
*state
,
1126 BlockReopenQueue
*queue
, Error
**errp
)
1129 BDRVRawReopenState
*rs
;
1133 assert(state
!= NULL
);
1134 assert(state
->bs
!= NULL
);
1136 s
= state
->bs
->opaque
;
1138 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
1141 /* Handle options changes */
1142 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
1143 if (!qemu_opts_absorb_qdict(opts
, state
->options
, errp
)) {
1148 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
1149 rs
->check_cache_dropped
=
1150 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
1152 /* This driver's reopen function doesn't currently allow changing
1153 * other options, so let's put them back in the original QDict and
1154 * bdrv_reopen_prepare() will detect changes and complain. */
1155 qemu_opts_to_qdict(opts
, state
->options
);
1158 * As part of reopen prepare we also want to create new fd by
1159 * raw_reconfigure_getfd(). But it wants updated "perm", when in
1160 * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to
1161 * permission update. Happily, permission update is always a part (a seprate
1162 * stage) of bdrv_reopen_multiple() so we can rely on this fact and
1163 * reconfigure fd in raw_check_perm().
1166 s
->reopen_state
= state
;
1170 qemu_opts_del(opts
);
1174 static void raw_reopen_commit(BDRVReopenState
*state
)
1176 BDRVRawReopenState
*rs
= state
->opaque
;
1177 BDRVRawState
*s
= state
->bs
->opaque
;
1179 s
->drop_cache
= rs
->drop_cache
;
1180 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1181 s
->open_flags
= rs
->open_flags
;
1182 g_free(state
->opaque
);
1183 state
->opaque
= NULL
;
1185 assert(s
->reopen_state
== state
);
1186 s
->reopen_state
= NULL
;
1190 static void raw_reopen_abort(BDRVReopenState
*state
)
1192 BDRVRawReopenState
*rs
= state
->opaque
;
1193 BDRVRawState
*s
= state
->bs
->opaque
;
1195 /* nothing to do if NULL, we didn't get far enough */
1200 g_free(state
->opaque
);
1201 state
->opaque
= NULL
;
1203 assert(s
->reopen_state
== state
);
1204 s
->reopen_state
= NULL
;
1207 static int hdev_get_max_hw_transfer(int fd
, struct stat
*st
)
1210 if (S_ISBLK(st
->st_mode
)) {
1211 unsigned short max_sectors
= 0;
1212 if (ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
1213 return max_sectors
* 512;
1217 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1228 * Get a sysfs attribute value as character string.
1231 static int get_sysfs_str_val(struct stat
*st
, const char *attribute
,
1233 g_autofree
char *sysfspath
= NULL
;
1237 if (!S_ISBLK(st
->st_mode
)) {
1241 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/%s",
1242 major(st
->st_rdev
), minor(st
->st_rdev
),
1244 ret
= g_file_get_contents(sysfspath
, val
, &len
, NULL
);
1249 /* The file is ended with '\n' */
1252 if (*(p
+ len
- 1) == '\n') {
1253 *(p
+ len
- 1) = '\0';
1259 #if defined(CONFIG_BLKZONED)
1260 static int get_sysfs_zoned_model(struct stat
*st
, BlockZoneModel
*zoned
)
1262 g_autofree
char *val
= NULL
;
1265 ret
= get_sysfs_str_val(st
, "zoned", &val
);
1270 if (strcmp(val
, "host-managed") == 0) {
1272 } else if (strcmp(val
, "host-aware") == 0) {
1274 } else if (strcmp(val
, "none") == 0) {
1275 *zoned
= BLK_Z_NONE
;
1281 #endif /* defined(CONFIG_BLKZONED) */
1284 * Get a sysfs attribute value as a long integer.
1287 static long get_sysfs_long_val(struct stat
*st
, const char *attribute
)
1289 g_autofree
char *str
= NULL
;
1294 ret
= get_sysfs_str_val(st
, attribute
, &str
);
1299 /* The file is ended with '\n', pass 'end' to accept that. */
1300 ret
= qemu_strtol(str
, &end
, 10, &val
);
1301 if (ret
== 0 && end
&& *end
== '\0') {
1308 static int hdev_get_max_segments(int fd
, struct stat
*st
)
1313 if (S_ISCHR(st
->st_mode
)) {
1314 if (ioctl(fd
, SG_GET_SG_TABLESIZE
, &ret
) == 0) {
1319 return get_sysfs_long_val(st
, "max_segments");
1325 #if defined(CONFIG_BLKZONED)
1326 static void raw_refresh_zoned_limits(BlockDriverState
*bs
, struct stat
*st
,
1329 BlockZoneModel zoned
;
1332 bs
->bl
.zoned
= BLK_Z_NONE
;
1334 ret
= get_sysfs_zoned_model(st
, &zoned
);
1335 if (ret
< 0 || zoned
== BLK_Z_NONE
) {
1338 bs
->bl
.zoned
= zoned
;
1340 ret
= get_sysfs_long_val(st
, "max_open_zones");
1342 bs
->bl
.max_open_zones
= ret
;
1345 ret
= get_sysfs_long_val(st
, "max_active_zones");
1347 bs
->bl
.max_active_zones
= ret
;
1351 * The zoned device must at least have zone size and nr_zones fields.
1353 ret
= get_sysfs_long_val(st
, "chunk_sectors");
1355 error_setg_errno(errp
, -ret
, "Unable to read chunk_sectors "
1359 error_setg(errp
, "Read 0 from chunk_sectors sysfs attribute");
1362 bs
->bl
.zone_size
= ret
<< BDRV_SECTOR_BITS
;
1364 ret
= get_sysfs_long_val(st
, "nr_zones");
1366 error_setg_errno(errp
, -ret
, "Unable to read nr_zones "
1370 error_setg(errp
, "Read 0 from nr_zones sysfs attribute");
1373 bs
->bl
.nr_zones
= ret
;
1375 ret
= get_sysfs_long_val(st
, "zone_append_max_bytes");
1377 bs
->bl
.max_append_sectors
= ret
>> BDRV_SECTOR_BITS
;
1380 #else /* !defined(CONFIG_BLKZONED) */
1381 static void raw_refresh_zoned_limits(BlockDriverState
*bs
, struct stat
*st
,
1384 bs
->bl
.zoned
= BLK_Z_NONE
;
1386 #endif /* !defined(CONFIG_BLKZONED) */
1388 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1390 BDRVRawState
*s
= bs
->opaque
;
1393 s
->needs_alignment
= raw_needs_alignment(bs
);
1394 raw_probe_alignment(bs
, s
->fd
, errp
);
1396 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1397 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, qemu_real_host_page_size());
1400 * Maximum transfers are best effort, so it is okay to ignore any
1401 * errors. That said, based on the man page errors in fstat would be
1402 * very much unexpected; the only possible case seems to be ENOMEM.
1404 if (fstat(s
->fd
, &st
)) {
1408 #if defined(__APPLE__) && (__MACH__)
1411 if (!fstatfs(s
->fd
, &buf
)) {
1412 bs
->bl
.opt_transfer
= buf
.f_iosize
;
1413 bs
->bl
.pdiscard_alignment
= buf
.f_bsize
;
1417 if (bdrv_is_sg(bs
) || S_ISBLK(st
.st_mode
)) {
1418 int ret
= hdev_get_max_hw_transfer(s
->fd
, &st
);
1420 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1421 bs
->bl
.max_hw_transfer
= ret
;
1424 ret
= hdev_get_max_segments(s
->fd
, &st
);
1426 bs
->bl
.max_hw_iov
= ret
;
1430 raw_refresh_zoned_limits(bs
, &st
, errp
);
1433 static int check_for_dasd(int fd
)
1436 struct dasd_information2_t info
= {0};
1438 return ioctl(fd
, BIODASDINFO2
, &info
);
1445 * Try to get @bs's logical and physical block size.
1446 * On success, store them in @bsz and return zero.
1447 * On failure, return negative errno.
1449 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1451 BDRVRawState
*s
= bs
->opaque
;
1454 /* If DASD or zoned devices, get blocksizes */
1455 if (check_for_dasd(s
->fd
) < 0) {
1456 /* zoned devices are not DASD */
1457 if (bs
->bl
.zoned
== BLK_Z_NONE
) {
1461 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1465 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1469 * Try to get @bs's geometry: cyls, heads, sectors.
1470 * On success, store them in @geo and return 0.
1471 * On failure return -errno.
1472 * (Allows block driver to assign default geometry values that guest sees)
1475 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1477 BDRVRawState
*s
= bs
->opaque
;
1478 struct hd_geometry ioctl_geo
= {0};
1480 /* If DASD, get its geometry */
1481 if (check_for_dasd(s
->fd
) < 0) {
1484 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1487 /* HDIO_GETGEO may return success even though geo contains zeros
1488 (e.g. certain multipath setups) */
1489 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1492 /* Do not return a geometry for partition */
1493 if (ioctl_geo
.start
!= 0) {
1496 geo
->heads
= ioctl_geo
.heads
;
1497 geo
->sectors
= ioctl_geo
.sectors
;
1498 geo
->cylinders
= ioctl_geo
.cylinders
;
1502 #else /* __linux__ */
1503 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1509 #if defined(__linux__)
1510 static int handle_aiocb_ioctl(void *opaque
)
1512 RawPosixAIOData
*aiocb
= opaque
;
1515 ret
= RETRY_ON_EINTR(
1516 ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
)
1526 static int handle_aiocb_flush(void *opaque
)
1528 RawPosixAIOData
*aiocb
= opaque
;
1529 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1532 if (s
->page_cache_inconsistent
) {
1533 return -s
->page_cache_inconsistent
;
1536 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1538 trace_file_flush_fdatasync_failed(errno
);
1540 /* There is no clear definition of the semantics of a failing fsync(),
1541 * so we may have to assume the worst. The sad truth is that this
1542 * assumption is correct for Linux. Some pages are now probably marked
1543 * clean in the page cache even though they are inconsistent with the
1544 * on-disk contents. The next fdatasync() call would succeed, but no
1545 * further writeback attempt will be made. We can't get back to a state
1546 * in which we know what is on disk (we would have to rewrite
1547 * everything that was touched since the last fdatasync() at least), so
1548 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1549 * really defined, I have little hope that other OSes are doing better.
1551 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1553 if ((s
->open_flags
& O_DIRECT
) == 0) {
1554 s
->page_cache_inconsistent
= errno
;
1561 #ifdef CONFIG_PREADV
1563 static bool preadv_present
= true;
1566 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1568 return preadv(fd
, iov
, nr_iov
, offset
);
1572 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1574 return pwritev(fd
, iov
, nr_iov
, offset
);
1579 static bool preadv_present
= false;
1582 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1588 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1595 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1599 len
= RETRY_ON_EINTR(
1600 (aiocb
->aio_type
& QEMU_AIO_WRITE
) ?
1601 qemu_pwritev(aiocb
->aio_fildes
,
1604 aiocb
->aio_offset
) :
1605 qemu_preadv(aiocb
->aio_fildes
,
1618 * Read/writes the data to/from a given linear buffer.
1620 * Returns the number of bytes handles or -errno in case of an error. Short
1621 * reads are only returned if the end of the file is reached.
1623 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1628 while (offset
< aiocb
->aio_nbytes
) {
1629 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1630 len
= pwrite(aiocb
->aio_fildes
,
1631 (const char *)buf
+ offset
,
1632 aiocb
->aio_nbytes
- offset
,
1633 aiocb
->aio_offset
+ offset
);
1635 len
= pread(aiocb
->aio_fildes
,
1637 aiocb
->aio_nbytes
- offset
,
1638 aiocb
->aio_offset
+ offset
);
1640 if (len
== -1 && errno
== EINTR
) {
1642 } else if (len
== -1 && errno
== EINVAL
&&
1643 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1644 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1646 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1647 * after a short read. Assume that O_DIRECT short reads only occur
1648 * at EOF. Therefore this is a short read, not an I/O error.
1651 } else if (len
== -1) {
1654 } else if (len
== 0) {
1663 static int handle_aiocb_rw(void *opaque
)
1665 RawPosixAIOData
*aiocb
= opaque
;
1669 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1671 * If there is just a single buffer, and it is properly aligned
1672 * we can just use plain pread/pwrite without any problems.
1674 if (aiocb
->io
.niov
== 1) {
1675 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1679 * We have more than one iovec, and all are properly aligned.
1681 * Try preadv/pwritev first and fall back to linearizing the
1682 * buffer if it's not supported.
1684 if (preadv_present
) {
1685 nbytes
= handle_aiocb_rw_vector(aiocb
);
1686 if (nbytes
== aiocb
->aio_nbytes
||
1687 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1690 preadv_present
= false;
1694 * XXX(hch): short read/write. no easy way to handle the reminder
1695 * using these interfaces. For now retry using plain
1701 * Ok, we have to do it the hard way, copy all segments into
1702 * a single aligned buffer.
1704 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1710 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1714 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1715 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1716 p
+= aiocb
->io
.iov
[i
].iov_len
;
1718 assert(p
- buf
== aiocb
->aio_nbytes
);
1721 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1722 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1724 size_t count
= aiocb
->aio_nbytes
, copy
;
1727 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1729 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1730 copy
= aiocb
->io
.iov
[i
].iov_len
;
1732 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1733 assert(count
>= copy
);
1742 if (nbytes
== aiocb
->aio_nbytes
) {
1744 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1745 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1748 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1749 0, aiocb
->aio_nbytes
- nbytes
);
1758 #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD)
1759 static int translate_err(int err
)
1761 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1769 #ifdef CONFIG_FALLOCATE
1770 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1773 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1776 } while (errno
== EINTR
);
1777 return translate_err(-errno
);
1781 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1784 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1786 if (!s
->has_write_zeroes
) {
1791 /* The BLKZEROOUT implementation in the kernel doesn't set
1792 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1794 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1796 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1797 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1800 } while (errno
== EINTR
);
1802 ret
= translate_err(-errno
);
1803 if (ret
== -ENOTSUP
) {
1804 s
->has_write_zeroes
= false;
1812 static int handle_aiocb_write_zeroes(void *opaque
)
1814 RawPosixAIOData
*aiocb
= opaque
;
1815 #ifdef CONFIG_FALLOCATE
1816 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1820 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1821 return handle_aiocb_write_zeroes_block(aiocb
);
1824 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1825 if (s
->has_write_zeroes
) {
1826 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1827 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1828 if (ret
== -ENOTSUP
) {
1829 s
->has_write_zeroes
= false;
1830 } else if (ret
== 0 || ret
!= -EINVAL
) {
1834 * Note: Some file systems do not like unaligned byte ranges, and
1835 * return EINVAL in such a case, though they should not do it according
1836 * to the man-page of fallocate(). Thus we simply ignore this return
1837 * value and try the other fallbacks instead.
1842 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1843 if (s
->has_discard
&& s
->has_fallocate
) {
1844 int ret
= do_fallocate(s
->fd
,
1845 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1846 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1848 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1849 if (ret
== 0 || ret
!= -ENOTSUP
) {
1852 s
->has_fallocate
= false;
1853 } else if (ret
== -EINVAL
) {
1855 * Some file systems like older versions of GPFS do not like un-
1856 * aligned byte ranges, and return EINVAL in such a case, though
1857 * they should not do it according to the man-page of fallocate().
1858 * Warn about the bad filesystem and try the final fallback instead.
1860 warn_report_once("Your file system is misbehaving: "
1861 "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. "
1862 "Please report this bug to your file system "
1864 } else if (ret
!= -ENOTSUP
) {
1867 s
->has_discard
= false;
1872 #ifdef CONFIG_FALLOCATE
1873 /* Last resort: we are trying to extend the file with zeroed data. This
1874 * can be done via fallocate(fd, 0) */
1875 len
= raw_co_getlength(aiocb
->bs
);
1876 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1877 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1878 if (ret
== 0 || ret
!= -ENOTSUP
) {
1881 s
->has_fallocate
= false;
1888 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1890 RawPosixAIOData
*aiocb
= opaque
;
1891 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1893 /* First try to write zeros and unmap at the same time */
1895 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1896 int ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1897 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1908 /* If we couldn't manage to unmap while guaranteed that the area reads as
1909 * all-zero afterwards, just write zeroes without unmapping */
1910 return handle_aiocb_write_zeroes(aiocb
);
1913 #ifndef HAVE_COPY_FILE_RANGE
1914 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1915 off_t
*out_off
, size_t len
, unsigned int flags
)
1917 #ifdef __NR_copy_file_range
1918 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1919 out_off
, len
, flags
);
1928 * parse_zone - Fill a zone descriptor
1930 #if defined(CONFIG_BLKZONED)
1931 static inline int parse_zone(struct BlockZoneDescriptor
*zone
,
1932 const struct blk_zone
*blkz
) {
1933 zone
->start
= blkz
->start
<< BDRV_SECTOR_BITS
;
1934 zone
->length
= blkz
->len
<< BDRV_SECTOR_BITS
;
1935 zone
->wp
= blkz
->wp
<< BDRV_SECTOR_BITS
;
1937 #ifdef HAVE_BLK_ZONE_REP_CAPACITY
1938 zone
->cap
= blkz
->capacity
<< BDRV_SECTOR_BITS
;
1940 zone
->cap
= blkz
->len
<< BDRV_SECTOR_BITS
;
1943 switch (blkz
->type
) {
1944 case BLK_ZONE_TYPE_SEQWRITE_REQ
:
1945 zone
->type
= BLK_ZT_SWR
;
1947 case BLK_ZONE_TYPE_SEQWRITE_PREF
:
1948 zone
->type
= BLK_ZT_SWP
;
1950 case BLK_ZONE_TYPE_CONVENTIONAL
:
1951 zone
->type
= BLK_ZT_CONV
;
1954 error_report("Unsupported zone type: 0x%x", blkz
->type
);
1958 switch (blkz
->cond
) {
1959 case BLK_ZONE_COND_NOT_WP
:
1960 zone
->state
= BLK_ZS_NOT_WP
;
1962 case BLK_ZONE_COND_EMPTY
:
1963 zone
->state
= BLK_ZS_EMPTY
;
1965 case BLK_ZONE_COND_IMP_OPEN
:
1966 zone
->state
= BLK_ZS_IOPEN
;
1968 case BLK_ZONE_COND_EXP_OPEN
:
1969 zone
->state
= BLK_ZS_EOPEN
;
1971 case BLK_ZONE_COND_CLOSED
:
1972 zone
->state
= BLK_ZS_CLOSED
;
1974 case BLK_ZONE_COND_READONLY
:
1975 zone
->state
= BLK_ZS_RDONLY
;
1977 case BLK_ZONE_COND_FULL
:
1978 zone
->state
= BLK_ZS_FULL
;
1980 case BLK_ZONE_COND_OFFLINE
:
1981 zone
->state
= BLK_ZS_OFFLINE
;
1984 error_report("Unsupported zone state: 0x%x", blkz
->cond
);
1991 #if defined(CONFIG_BLKZONED)
1992 static int handle_aiocb_zone_report(void *opaque
)
1994 RawPosixAIOData
*aiocb
= opaque
;
1995 int fd
= aiocb
->aio_fildes
;
1996 unsigned int *nr_zones
= aiocb
->zone_report
.nr_zones
;
1997 BlockZoneDescriptor
*zones
= aiocb
->zone_report
.zones
;
1998 /* zoned block devices use 512-byte sectors */
1999 uint64_t sector
= aiocb
->aio_offset
/ 512;
2001 struct blk_zone
*blkz
;
2005 unsigned int n
= 0, i
= 0;
2008 rep_size
= sizeof(struct blk_zone_report
) + nrz
* sizeof(struct blk_zone
);
2009 g_autofree
struct blk_zone_report
*rep
= NULL
;
2010 rep
= g_malloc(rep_size
);
2012 blkz
= (struct blk_zone
*)(rep
+ 1);
2014 memset(rep
, 0, rep_size
);
2015 rep
->sector
= sector
;
2016 rep
->nr_zones
= nrz
- n
;
2019 ret
= ioctl(fd
, BLKREPORTZONE
, rep
);
2020 } while (ret
!= 0 && errno
== EINTR
);
2022 error_report("%d: ioctl BLKREPORTZONE at %" PRId64
" failed %d",
2027 if (!rep
->nr_zones
) {
2031 for (i
= 0; i
< rep
->nr_zones
; i
++, n
++) {
2032 ret
= parse_zone(&zones
[n
], &blkz
[i
]);
2037 /* The next report should start after the last zone reported */
2038 sector
= blkz
[i
].start
+ blkz
[i
].len
;
2047 #if defined(CONFIG_BLKZONED)
2048 static int handle_aiocb_zone_mgmt(void *opaque
)
2050 RawPosixAIOData
*aiocb
= opaque
;
2051 int fd
= aiocb
->aio_fildes
;
2052 uint64_t sector
= aiocb
->aio_offset
/ 512;
2053 int64_t nr_sectors
= aiocb
->aio_nbytes
/ 512;
2054 struct blk_zone_range range
;
2057 /* Execute the operation */
2058 range
.sector
= sector
;
2059 range
.nr_sectors
= nr_sectors
;
2061 ret
= ioctl(fd
, aiocb
->zone_mgmt
.op
, &range
);
2062 } while (ret
!= 0 && errno
== EINTR
);
2068 static int handle_aiocb_copy_range(void *opaque
)
2070 RawPosixAIOData
*aiocb
= opaque
;
2071 uint64_t bytes
= aiocb
->aio_nbytes
;
2072 off_t in_off
= aiocb
->aio_offset
;
2073 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
2076 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
2077 aiocb
->copy_range
.aio_fd2
, &out_off
,
2079 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
2080 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
2083 /* No progress (e.g. when beyond EOF), let the caller fall back to
2102 static int handle_aiocb_discard(void *opaque
)
2104 RawPosixAIOData
*aiocb
= opaque
;
2106 BDRVRawState
*s
= aiocb
->bs
->opaque
;
2108 if (!s
->has_discard
) {
2112 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
2115 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
2116 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
2119 } while (errno
== EINTR
);
2121 ret
= translate_err(-errno
);
2124 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
2125 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
2126 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
2127 ret
= translate_err(ret
);
2128 #elif defined(__APPLE__) && (__MACH__)
2129 fpunchhole_t fpunchhole
;
2130 fpunchhole
.fp_flags
= 0;
2131 fpunchhole
.reserved
= 0;
2132 fpunchhole
.fp_offset
= aiocb
->aio_offset
;
2133 fpunchhole
.fp_length
= aiocb
->aio_nbytes
;
2134 if (fcntl(s
->fd
, F_PUNCHHOLE
, &fpunchhole
) == -1) {
2135 ret
= errno
== ENODEV
? -ENOTSUP
: -errno
;
2142 if (ret
== -ENOTSUP
) {
2143 s
->has_discard
= false;
2149 * Help alignment probing by allocating the first block.
2151 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
2152 * reading succeeds regardless of request length. In this case we fallback to
2153 * safe alignment which is not optimal. Allocating the first block avoids this
2156 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
2157 * request alignment, so we use safe values.
2159 * Returns: 0 on success, -errno on failure. Since this is an optimization,
2160 * caller may ignore failures.
2162 static int allocate_first_block(int fd
, size_t max_size
)
2164 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
2167 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size());
2172 buf
= qemu_memalign(max_align
, write_size
);
2173 memset(buf
, 0, write_size
);
2175 n
= RETRY_ON_EINTR(pwrite(fd
, buf
, write_size
, 0));
2177 ret
= (n
== -1) ? -errno
: 0;
2183 static int handle_aiocb_truncate(void *opaque
)
2185 RawPosixAIOData
*aiocb
= opaque
;
2187 int64_t current_length
= 0;
2190 int fd
= aiocb
->aio_fildes
;
2191 int64_t offset
= aiocb
->aio_offset
;
2192 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
2193 Error
**errp
= aiocb
->truncate
.errp
;
2195 if (fstat(fd
, &st
) < 0) {
2197 error_setg_errno(errp
, -result
, "Could not stat file");
2201 current_length
= st
.st_size
;
2202 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
2203 error_setg(errp
, "Cannot use preallocation for shrinking files");
2208 #ifdef CONFIG_POSIX_FALLOCATE
2209 case PREALLOC_MODE_FALLOC
:
2211 * Truncating before posix_fallocate() makes it about twice slower on
2212 * file systems that do not support fallocate(), trying to check if a
2213 * block is allocated before allocating it, so don't do that here.
2215 if (offset
!= current_length
) {
2216 result
= -posix_fallocate(fd
, current_length
,
2217 offset
- current_length
);
2219 /* posix_fallocate() doesn't set errno. */
2220 error_setg_errno(errp
, -result
,
2221 "Could not preallocate new data");
2222 } else if (current_length
== 0) {
2224 * posix_fallocate() uses fallocate() if the filesystem
2225 * supports it, or fallback to manually writing zeroes. If
2226 * fallocate() was used, unaligned reads from the fallocated
2227 * area in raw_probe_alignment() will succeed, hence we need to
2228 * allocate the first block.
2230 * Optimize future alignment probing; ignore failures.
2232 allocate_first_block(fd
, offset
);
2239 case PREALLOC_MODE_FULL
:
2241 int64_t num
= 0, left
= offset
- current_length
;
2245 * Knowing the final size from the beginning could allow the file
2246 * system driver to do less allocations and possibly avoid
2247 * fragmentation of the file.
2249 if (ftruncate(fd
, offset
) != 0) {
2251 error_setg_errno(errp
, -result
, "Could not resize file");
2255 buf
= g_malloc0(65536);
2257 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
2258 if (seek_result
< 0) {
2260 error_setg_errno(errp
, -result
,
2261 "Failed to seek to the old end of file");
2266 num
= MIN(left
, 65536);
2267 result
= write(fd
, buf
, num
);
2269 if (errno
== EINTR
) {
2273 error_setg_errno(errp
, -result
,
2274 "Could not write zeros for preallocation");
2283 error_setg_errno(errp
, -result
,
2284 "Could not flush file to disk");
2290 case PREALLOC_MODE_OFF
:
2291 if (ftruncate(fd
, offset
) != 0) {
2293 error_setg_errno(errp
, -result
, "Could not resize file");
2294 } else if (current_length
== 0 && offset
> current_length
) {
2295 /* Optimize future alignment probing; ignore failures. */
2296 allocate_first_block(fd
, offset
);
2301 error_setg(errp
, "Unsupported preallocation mode: %s",
2302 PreallocMode_str(prealloc
));
2308 if (ftruncate(fd
, current_length
) < 0) {
2309 error_report("Failed to restore old file length: %s",
2318 static int coroutine_fn
raw_thread_pool_submit(ThreadPoolFunc func
, void *arg
)
2320 return thread_pool_submit_co(func
, arg
);
2324 * Check if all memory in this vector is sector aligned.
2326 static bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2329 size_t alignment
= bdrv_min_mem_align(bs
);
2330 size_t len
= bs
->bl
.request_alignment
;
2333 for (i
= 0; i
< qiov
->niov
; i
++) {
2334 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2337 if (qiov
->iov
[i
].iov_len
% len
) {
2345 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
2346 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
2348 BDRVRawState
*s
= bs
->opaque
;
2349 RawPosixAIOData acb
;
2351 if (fd_open(bs
) < 0)
2355 * When using O_DIRECT, the request must be aligned to be able to use
2356 * either libaio or io_uring interface. If not fail back to regular thread
2357 * pool read/write code which emulates this for us if we
2358 * set QEMU_AIO_MISALIGNED.
2360 if (s
->needs_alignment
&& !bdrv_qiov_is_aligned(bs
, qiov
)) {
2361 type
|= QEMU_AIO_MISALIGNED
;
2362 #ifdef CONFIG_LINUX_IO_URING
2363 } else if (s
->use_linux_io_uring
) {
2364 assert(qiov
->size
== bytes
);
2365 return luring_co_submit(bs
, s
->fd
, offset
, qiov
, type
);
2367 #ifdef CONFIG_LINUX_AIO
2368 } else if (s
->use_linux_aio
) {
2369 assert(qiov
->size
== bytes
);
2370 return laio_co_submit(s
->fd
, offset
, qiov
, type
, s
->aio_max_batch
);
2374 acb
= (RawPosixAIOData
) {
2376 .aio_fildes
= s
->fd
,
2378 .aio_offset
= offset
,
2379 .aio_nbytes
= bytes
,
2386 assert(qiov
->size
== bytes
);
2387 return raw_thread_pool_submit(handle_aiocb_rw
, &acb
);
2390 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, int64_t offset
,
2391 int64_t bytes
, QEMUIOVector
*qiov
,
2392 BdrvRequestFlags flags
)
2394 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
2397 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, int64_t offset
,
2398 int64_t bytes
, QEMUIOVector
*qiov
,
2399 BdrvRequestFlags flags
)
2401 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
2404 static void coroutine_fn
raw_co_io_plug(BlockDriverState
*bs
)
2406 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2407 #ifdef CONFIG_LINUX_AIO
2408 if (s
->use_linux_aio
) {
2412 #ifdef CONFIG_LINUX_IO_URING
2413 if (s
->use_linux_io_uring
) {
2419 static void coroutine_fn
raw_co_io_unplug(BlockDriverState
*bs
)
2421 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2422 #ifdef CONFIG_LINUX_AIO
2423 if (s
->use_linux_aio
) {
2424 laio_io_unplug(s
->aio_max_batch
);
2427 #ifdef CONFIG_LINUX_IO_URING
2428 if (s
->use_linux_io_uring
) {
2434 static int coroutine_fn
raw_co_flush_to_disk(BlockDriverState
*bs
)
2436 BDRVRawState
*s
= bs
->opaque
;
2437 RawPosixAIOData acb
;
2445 acb
= (RawPosixAIOData
) {
2447 .aio_fildes
= s
->fd
,
2448 .aio_type
= QEMU_AIO_FLUSH
,
2451 #ifdef CONFIG_LINUX_IO_URING
2452 if (s
->use_linux_io_uring
) {
2453 return luring_co_submit(bs
, s
->fd
, 0, NULL
, QEMU_AIO_FLUSH
);
2456 return raw_thread_pool_submit(handle_aiocb_flush
, &acb
);
2459 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
2460 AioContext
*new_context
)
2462 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2463 #ifdef CONFIG_LINUX_AIO
2464 if (s
->use_linux_aio
) {
2465 Error
*local_err
= NULL
;
2466 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
2467 error_reportf_err(local_err
, "Unable to use native AIO, "
2468 "falling back to thread pool: ");
2469 s
->use_linux_aio
= false;
2473 #ifdef CONFIG_LINUX_IO_URING
2474 if (s
->use_linux_io_uring
) {
2475 Error
*local_err
= NULL
;
2476 if (!aio_setup_linux_io_uring(new_context
, &local_err
)) {
2477 error_reportf_err(local_err
, "Unable to use linux io_uring, "
2478 "falling back to thread pool: ");
2479 s
->use_linux_io_uring
= false;
2485 static void raw_close(BlockDriverState
*bs
)
2487 BDRVRawState
*s
= bs
->opaque
;
2496 * Truncates the given regular file @fd to @offset and, when growing, fills the
2497 * new space according to @prealloc.
2499 * Returns: 0 on success, -errno on failure.
2501 static int coroutine_fn
2502 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2503 PreallocMode prealloc
, Error
**errp
)
2505 RawPosixAIOData acb
;
2507 acb
= (RawPosixAIOData
) {
2510 .aio_type
= QEMU_AIO_TRUNCATE
,
2511 .aio_offset
= offset
,
2513 .prealloc
= prealloc
,
2518 return raw_thread_pool_submit(handle_aiocb_truncate
, &acb
);
2521 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2522 bool exact
, PreallocMode prealloc
,
2523 BdrvRequestFlags flags
, Error
**errp
)
2525 BDRVRawState
*s
= bs
->opaque
;
2529 if (fstat(s
->fd
, &st
)) {
2531 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2535 if (S_ISREG(st
.st_mode
)) {
2536 /* Always resizes to the exact @offset */
2537 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2540 if (prealloc
!= PREALLOC_MODE_OFF
) {
2541 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2542 "non-regular file", PreallocMode_str(prealloc
));
2546 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2547 int64_t cur_length
= raw_co_getlength(bs
);
2549 if (offset
!= cur_length
&& exact
) {
2550 error_setg(errp
, "Cannot resize device files");
2552 } else if (offset
> cur_length
) {
2553 error_setg(errp
, "Cannot grow device files");
2557 error_setg(errp
, "Resizing this file is not supported");
2565 static int64_t coroutine_fn
raw_co_getlength(BlockDriverState
*bs
)
2567 BDRVRawState
*s
= bs
->opaque
;
2573 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2574 struct disklabel dl
;
2576 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2578 return (uint64_t)dl
.d_secsize
*
2579 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2583 #elif defined(__NetBSD__)
2584 static int64_t coroutine_fn
raw_co_getlength(BlockDriverState
*bs
)
2586 BDRVRawState
*s
= bs
->opaque
;
2592 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2593 struct dkwedge_info dkw
;
2595 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2596 return dkw
.dkw_size
* 512;
2598 struct disklabel dl
;
2600 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2602 return (uint64_t)dl
.d_secsize
*
2603 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2608 #elif defined(__sun__)
2609 static int64_t coroutine_fn
raw_co_getlength(BlockDriverState
*bs
)
2611 BDRVRawState
*s
= bs
->opaque
;
2612 struct dk_minfo minfo
;
2622 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2624 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2626 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2630 * There are reports that lseek on some devices fails, but
2631 * irc discussion said that contingency on contingency was overkill.
2633 size
= lseek(s
->fd
, 0, SEEK_END
);
2639 #elif defined(CONFIG_BSD)
2640 static int64_t coroutine_fn
raw_co_getlength(BlockDriverState
*bs
)
2642 BDRVRawState
*s
= bs
->opaque
;
2646 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2655 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2658 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2660 #ifdef DIOCGMEDIASIZE
2661 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
)) {
2668 if (ioctl(fd
, DIOCGPART
, &pi
) == 0) {
2669 size
= pi
.media_size
;
2673 #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE)
2675 uint64_t sectors
= 0;
2676 uint32_t sector_size
= 0;
2678 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2679 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2680 size
= sectors
* sector_size
;
2685 size
= lseek(fd
, 0LL, SEEK_END
);
2690 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2693 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2694 if (size
== 2048LL * (unsigned)-1)
2696 /* XXX no disc? maybe we need to reopen... */
2697 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2704 size
= lseek(fd
, 0, SEEK_END
);
2712 static int64_t coroutine_fn
raw_co_getlength(BlockDriverState
*bs
)
2714 BDRVRawState
*s
= bs
->opaque
;
2723 size
= lseek(s
->fd
, 0, SEEK_END
);
2731 static int64_t coroutine_fn
raw_co_get_allocated_file_size(BlockDriverState
*bs
)
2734 BDRVRawState
*s
= bs
->opaque
;
2736 if (fstat(s
->fd
, &st
) < 0) {
2739 return (int64_t)st
.st_blocks
* 512;
2742 static int coroutine_fn
2743 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2745 BlockdevCreateOptionsFile
*file_opts
;
2746 Error
*local_err
= NULL
;
2748 uint64_t perm
, shared
;
2751 /* Validate options and set default values */
2752 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2753 file_opts
= &options
->u
.file
;
2755 if (!file_opts
->has_nocow
) {
2756 file_opts
->nocow
= false;
2758 if (!file_opts
->has_preallocation
) {
2759 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2761 if (!file_opts
->has_extent_size_hint
) {
2762 file_opts
->extent_size_hint
= 1 * MiB
;
2764 if (file_opts
->extent_size_hint
> UINT32_MAX
) {
2766 error_setg(errp
, "Extent size hint is too large");
2771 fd
= qemu_create(file_opts
->filename
, O_RDWR
| O_BINARY
, 0644, errp
);
2777 /* Take permissions: We want to discard everything, so we need
2778 * BLK_PERM_WRITE; and truncation to the desired size requires
2780 * On the other hand, we cannot share the RESIZE permission
2781 * because we promise that after this function, the file has the
2782 * size given in the options. If someone else were to resize it
2783 * concurrently, we could not guarantee that.
2784 * Note that after this function, we can no longer guarantee that
2785 * the file is not touched by a third party, so it may be resized
2787 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2788 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2790 /* Step one: Take locks */
2791 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2796 /* Step two: Check that nobody else has taken conflicting locks */
2797 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2799 error_append_hint(errp
,
2800 "Is another process using the image [%s]?\n",
2801 file_opts
->filename
);
2805 /* Clear the file by truncating it to 0 */
2806 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2811 if (file_opts
->nocow
) {
2813 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2814 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2815 * will be ignored since any failure of this operation should not
2816 * block the left work.
2819 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2820 attr
|= FS_NOCOW_FL
;
2821 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2825 #ifdef FS_IOC_FSSETXATTR
2827 * Try to set the extent size hint. Failure is not fatal, and a warning is
2828 * only printed if the option was explicitly specified.
2831 struct fsxattr attr
;
2832 result
= ioctl(fd
, FS_IOC_FSGETXATTR
, &attr
);
2834 attr
.fsx_xflags
|= FS_XFLAG_EXTSIZE
;
2835 attr
.fsx_extsize
= file_opts
->extent_size_hint
;
2836 result
= ioctl(fd
, FS_IOC_FSSETXATTR
, &attr
);
2838 if (result
< 0 && file_opts
->has_extent_size_hint
&&
2839 file_opts
->extent_size_hint
)
2841 warn_report("Failed to set extent size hint: %s",
2847 /* Resize and potentially preallocate the file to the desired
2849 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2850 file_opts
->preallocation
, errp
);
2856 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2858 /* The above call should not fail, and if it does, that does
2859 * not mean the whole creation operation has failed. So
2860 * report it the user for their convenience, but do not report
2861 * it to the caller. */
2862 warn_report_err(local_err
);
2866 if (qemu_close(fd
) != 0 && result
== 0) {
2868 error_setg_errno(errp
, -result
, "Could not close the new file");
2874 static int coroutine_fn GRAPH_RDLOCK
2875 raw_co_create_opts(BlockDriver
*drv
, const char *filename
,
2876 QemuOpts
*opts
, Error
**errp
)
2878 BlockdevCreateOptions options
;
2879 int64_t total_size
= 0;
2880 int64_t extent_size_hint
= 0;
2881 bool has_extent_size_hint
= false;
2883 PreallocMode prealloc
;
2885 Error
*local_err
= NULL
;
2887 /* Skip file: protocol prefix */
2888 strstart(filename
, "file:", &filename
);
2890 /* Read out options */
2891 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2893 if (qemu_opt_get(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
)) {
2894 has_extent_size_hint
= true;
2896 qemu_opt_get_size_del(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
, -1);
2898 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2899 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2900 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2901 PREALLOC_MODE_OFF
, &local_err
);
2904 error_propagate(errp
, local_err
);
2908 options
= (BlockdevCreateOptions
) {
2909 .driver
= BLOCKDEV_DRIVER_FILE
,
2911 .filename
= (char *) filename
,
2913 .has_preallocation
= true,
2914 .preallocation
= prealloc
,
2917 .has_extent_size_hint
= has_extent_size_hint
,
2918 .extent_size_hint
= extent_size_hint
,
2921 return raw_co_create(&options
, errp
);
2924 static int coroutine_fn
raw_co_delete_file(BlockDriverState
*bs
,
2930 if (!(stat(bs
->filename
, &st
) == 0) || !S_ISREG(st
.st_mode
)) {
2931 error_setg_errno(errp
, ENOENT
, "%s is not a regular file",
2936 ret
= unlink(bs
->filename
);
2939 error_setg_errno(errp
, -ret
, "Error when deleting file %s",
2947 * Find allocation range in @bs around offset @start.
2948 * May change underlying file descriptor's file offset.
2949 * If @start is not in a hole, store @start in @data, and the
2950 * beginning of the next hole in @hole, and return 0.
2951 * If @start is in a non-trailing hole, store @start in @hole and the
2952 * beginning of the next non-hole in @data, and return 0.
2953 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2954 * If we can't find out, return a negative errno other than -ENXIO.
2956 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2957 off_t
*data
, off_t
*hole
)
2959 #if defined SEEK_HOLE && defined SEEK_DATA
2960 BDRVRawState
*s
= bs
->opaque
;
2965 * D1. offs == start: start is in data
2966 * D2. offs > start: start is in a hole, next data at offs
2967 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2968 * or start is beyond EOF
2969 * If the latter happens, the file has been truncated behind
2970 * our back since we opened it. All bets are off then.
2971 * Treating like a trailing hole is simplest.
2972 * D4. offs < 0, errno != ENXIO: we learned nothing
2974 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2976 return -errno
; /* D3 or D4 */
2980 /* This is not a valid return by lseek(). We are safe to just return
2981 * -EIO in this case, and we'll treat it like D4. */
2986 /* D2: in hole, next data at offs */
2992 /* D1: in data, end not yet known */
2996 * H1. offs == start: start is in a hole
2997 * If this happens here, a hole has been dug behind our back
2998 * since the previous lseek().
2999 * H2. offs > start: either start is in data, next hole at offs,
3000 * or start is in trailing hole, EOF at offs
3001 * Linux treats trailing holes like any other hole: offs ==
3002 * start. Solaris seeks to EOF instead: offs > start (blech).
3003 * If that happens here, a hole has been dug behind our back
3004 * since the previous lseek().
3005 * H3. offs < 0, errno = ENXIO: start is beyond EOF
3006 * If this happens, the file has been truncated behind our
3007 * back since we opened it. Treat it like a trailing hole.
3008 * H4. offs < 0, errno != ENXIO: we learned nothing
3009 * Pretend we know nothing at all, i.e. "forget" about D1.
3011 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
3013 return -errno
; /* D1 and (H3 or H4) */
3017 /* This is not a valid return by lseek(). We are safe to just return
3018 * -EIO in this case, and we'll treat it like H4. */
3024 * D1 and H2: either in data, next hole at offs, or it was in
3025 * data but is now in a trailing hole. In the latter case,
3026 * all bets are off. Treating it as if it there was data all
3027 * the way to EOF is safe, so simply do that.
3042 * Returns the allocation status of the specified offset.
3044 * The block layer guarantees 'offset' and 'bytes' are within bounds.
3046 * 'pnum' is set to the number of bytes (including and immediately following
3047 * the specified offset) that are known to be in the same
3048 * allocated/unallocated state.
3050 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may
3053 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
3056 int64_t bytes
, int64_t *pnum
,
3058 BlockDriverState
**file
)
3060 off_t data
= 0, hole
= 0;
3063 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
3074 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
3077 ret
= find_allocation(bs
, offset
, &data
, &hole
);
3078 if (ret
== -ENXIO
) {
3081 ret
= BDRV_BLOCK_ZERO
;
3082 } else if (ret
< 0) {
3083 /* No info available, so pretend there are no holes */
3085 ret
= BDRV_BLOCK_DATA
;
3086 } else if (data
== offset
) {
3087 /* On a data extent, compute bytes to the end of the extent,
3088 * possibly including a partial sector at EOF. */
3089 *pnum
= hole
- offset
;
3092 * We are not allowed to return partial sectors, though, so
3093 * round up if necessary.
3095 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
3096 int64_t file_length
= raw_co_getlength(bs
);
3097 if (file_length
> 0) {
3098 /* Ignore errors, this is just a safeguard */
3099 assert(hole
== file_length
);
3101 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
3104 ret
= BDRV_BLOCK_DATA
;
3106 /* On a hole, compute bytes to the beginning of the next extent. */
3107 assert(hole
== offset
);
3108 *pnum
= data
- offset
;
3109 ret
= BDRV_BLOCK_ZERO
;
3113 return ret
| BDRV_BLOCK_OFFSET_VALID
;
3116 #if defined(__linux__)
3117 /* Verify that the file is not in the page cache */
3118 static void coroutine_fn
check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
3120 const size_t window_size
= 128 * 1024 * 1024;
3121 BDRVRawState
*s
= bs
->opaque
;
3122 void *window
= NULL
;
3129 /* mincore(2) page status information requires 1 byte per page */
3130 page_size
= sysconf(_SC_PAGESIZE
);
3131 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
3133 end
= raw_co_getlength(bs
);
3135 for (offset
= 0; offset
< end
; offset
+= window_size
) {
3142 /* Unmap previous window if size has changed */
3143 new_length
= MIN(end
- offset
, window_size
);
3144 if (new_length
!= length
) {
3145 munmap(window
, length
);
3150 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
3152 if (new_window
== MAP_FAILED
) {
3153 error_setg_errno(errp
, errno
, "mmap failed");
3157 window
= new_window
;
3158 length
= new_length
;
3160 ret
= mincore(window
, length
, vec
);
3162 error_setg_errno(errp
, errno
, "mincore failed");
3166 vec_end
= DIV_ROUND_UP(length
, page_size
);
3167 for (i
= 0; i
< vec_end
; i
++) {
3173 error_setg(errp
, "page cache still in use!");
3179 munmap(window
, length
);
3184 #endif /* __linux__ */
3186 static void coroutine_fn GRAPH_RDLOCK
3187 raw_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
3189 BDRVRawState
*s
= bs
->opaque
;
3194 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
3198 if (!s
->drop_cache
) {
3202 if (s
->open_flags
& O_DIRECT
) {
3203 return; /* No host kernel page cache */
3206 #if defined(__linux__)
3207 /* This sets the scene for the next syscall... */
3208 ret
= bdrv_co_flush(bs
);
3210 error_setg_errno(errp
, -ret
, "flush failed");
3214 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
3215 * process. These limitations are okay because we just fsynced the file,
3216 * we don't use mmap, and the file should not be in use by other processes.
3218 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
3219 if (ret
!= 0) { /* the return value is a positive errno */
3220 error_setg_errno(errp
, ret
, "fadvise failed");
3224 if (s
->check_cache_dropped
) {
3225 check_cache_dropped(bs
, errp
);
3227 #else /* __linux__ */
3228 /* Do nothing. Live migration to a remote host with cache.direct=off is
3229 * unsupported on other host operating systems. Cache consistency issues
3230 * may occur but no error is reported here, partly because that's the
3231 * historical behavior and partly because it's hard to differentiate valid
3232 * configurations that should not cause errors.
3234 #endif /* !__linux__ */
3237 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
3240 s
->stats
.discard_nb_failed
++;
3242 s
->stats
.discard_nb_ok
++;
3243 s
->stats
.discard_bytes_ok
+= nbytes
;
3248 * zone report - Get a zone block device's information in the form
3249 * of an array of zone descriptors.
3250 * zones is an array of zone descriptors to hold zone information on reply;
3251 * offset can be any byte within the entire size of the device;
3252 * nr_zones is the maxium number of sectors the command should operate on.
3254 #if defined(CONFIG_BLKZONED)
3255 static int coroutine_fn
raw_co_zone_report(BlockDriverState
*bs
, int64_t offset
,
3256 unsigned int *nr_zones
,
3257 BlockZoneDescriptor
*zones
) {
3258 BDRVRawState
*s
= bs
->opaque
;
3259 RawPosixAIOData acb
= (RawPosixAIOData
) {
3261 .aio_fildes
= s
->fd
,
3262 .aio_type
= QEMU_AIO_ZONE_REPORT
,
3263 .aio_offset
= offset
,
3265 .nr_zones
= nr_zones
,
3270 return raw_thread_pool_submit(handle_aiocb_zone_report
, &acb
);
3275 * zone management operations - Execute an operation on a zone
3277 #if defined(CONFIG_BLKZONED)
3278 static int coroutine_fn
raw_co_zone_mgmt(BlockDriverState
*bs
, BlockZoneOp op
,
3279 int64_t offset
, int64_t len
) {
3280 BDRVRawState
*s
= bs
->opaque
;
3281 RawPosixAIOData acb
;
3282 int64_t zone_size
, zone_size_mask
;
3283 const char *op_name
;
3286 int64_t capacity
= bs
->total_sectors
<< BDRV_SECTOR_BITS
;
3288 zone_size
= bs
->bl
.zone_size
;
3289 zone_size_mask
= zone_size
- 1;
3290 if (offset
& zone_size_mask
) {
3291 error_report("sector offset %" PRId64
" is not aligned to zone size "
3292 "%" PRId64
"", offset
/ 512, zone_size
/ 512);
3296 if (((offset
+ len
) < capacity
&& len
& zone_size_mask
) ||
3297 offset
+ len
> capacity
) {
3298 error_report("number of sectors %" PRId64
" is not aligned to zone size"
3299 " %" PRId64
"", len
/ 512, zone_size
/ 512);
3305 op_name
= "BLKOPENZONE";
3309 op_name
= "BLKCLOSEZONE";
3313 op_name
= "BLKFINISHZONE";
3317 op_name
= "BLKRESETZONE";
3321 error_report("Unsupported zone op: 0x%x", op
);
3325 acb
= (RawPosixAIOData
) {
3327 .aio_fildes
= s
->fd
,
3328 .aio_type
= QEMU_AIO_ZONE_MGMT
,
3329 .aio_offset
= offset
,
3336 ret
= raw_thread_pool_submit(handle_aiocb_zone_mgmt
, &acb
);
3338 error_report("ioctl %s failed %d", op_name
, ret
);
3345 static coroutine_fn
int
3346 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
3349 BDRVRawState
*s
= bs
->opaque
;
3350 RawPosixAIOData acb
;
3353 acb
= (RawPosixAIOData
) {
3355 .aio_fildes
= s
->fd
,
3356 .aio_type
= QEMU_AIO_DISCARD
,
3357 .aio_offset
= offset
,
3358 .aio_nbytes
= bytes
,
3362 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
3365 ret
= raw_thread_pool_submit(handle_aiocb_discard
, &acb
);
3366 raw_account_discard(s
, bytes
, ret
);
3370 static coroutine_fn
int
3371 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3373 return raw_do_pdiscard(bs
, offset
, bytes
, false);
3376 static int coroutine_fn
3377 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
3378 BdrvRequestFlags flags
, bool blkdev
)
3380 BDRVRawState
*s
= bs
->opaque
;
3381 RawPosixAIOData acb
;
3382 ThreadPoolFunc
*handler
;
3384 #ifdef CONFIG_FALLOCATE
3385 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3386 BdrvTrackedRequest
*req
;
3389 * This is a workaround for a bug in the Linux XFS driver,
3390 * where writes submitted through the AIO interface will be
3391 * discarded if they happen beyond a concurrently running
3392 * fallocate() that increases the file length (i.e., both the
3393 * write and the fallocate() happen beyond the EOF).
3395 * To work around it, we extend the tracked request for this
3396 * zero write until INT64_MAX (effectively infinity), and mark
3397 * it as serializing.
3399 * We have to enable this workaround for all filesystems and
3400 * AIO modes (not just XFS with aio=native), because for
3401 * remote filesystems we do not know the host configuration.
3404 req
= bdrv_co_get_self_request(bs
);
3406 assert(req
->type
== BDRV_TRACKED_WRITE
);
3407 assert(req
->offset
<= offset
);
3408 assert(req
->offset
+ req
->bytes
>= offset
+ bytes
);
3410 req
->bytes
= BDRV_MAX_LENGTH
- req
->offset
;
3412 bdrv_check_request(req
->offset
, req
->bytes
, &error_abort
);
3414 bdrv_make_request_serialising(req
, bs
->bl
.request_alignment
);
3418 acb
= (RawPosixAIOData
) {
3420 .aio_fildes
= s
->fd
,
3421 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
3422 .aio_offset
= offset
,
3423 .aio_nbytes
= bytes
,
3427 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
3429 if (flags
& BDRV_REQ_NO_FALLBACK
) {
3430 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
3433 if (flags
& BDRV_REQ_MAY_UNMAP
) {
3434 acb
.aio_type
|= QEMU_AIO_DISCARD
;
3435 handler
= handle_aiocb_write_zeroes_unmap
;
3437 handler
= handle_aiocb_write_zeroes
;
3440 return raw_thread_pool_submit(handler
, &acb
);
3443 static int coroutine_fn
raw_co_pwrite_zeroes(
3444 BlockDriverState
*bs
, int64_t offset
,
3445 int64_t bytes
, BdrvRequestFlags flags
)
3447 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
3450 static int coroutine_fn
3451 raw_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3456 static ImageInfoSpecific
*raw_get_specific_info(BlockDriverState
*bs
,
3459 ImageInfoSpecificFile
*file_info
= g_new0(ImageInfoSpecificFile
, 1);
3460 ImageInfoSpecific
*spec_info
= g_new(ImageInfoSpecific
, 1);
3462 *spec_info
= (ImageInfoSpecific
){
3463 .type
= IMAGE_INFO_SPECIFIC_KIND_FILE
,
3464 .u
.file
.data
= file_info
,
3467 #ifdef FS_IOC_FSGETXATTR
3469 BDRVRawState
*s
= bs
->opaque
;
3470 struct fsxattr attr
;
3473 ret
= ioctl(s
->fd
, FS_IOC_FSGETXATTR
, &attr
);
3474 if (!ret
&& attr
.fsx_extsize
!= 0) {
3475 file_info
->has_extent_size_hint
= true;
3476 file_info
->extent_size_hint
= attr
.fsx_extsize
;
3484 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
3486 BDRVRawState
*s
= bs
->opaque
;
3487 return (BlockStatsSpecificFile
) {
3488 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
3489 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
3490 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
3494 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
3496 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3498 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
3499 stats
->u
.file
= get_blockstats_specific_file(bs
);
3504 #if defined(HAVE_HOST_BLOCK_DEVICE)
3505 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
3507 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3509 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
3510 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
3514 #endif /* HAVE_HOST_BLOCK_DEVICE */
3516 static QemuOptsList raw_create_opts
= {
3517 .name
= "raw-create-opts",
3518 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
3521 .name
= BLOCK_OPT_SIZE
,
3522 .type
= QEMU_OPT_SIZE
,
3523 .help
= "Virtual disk size"
3526 .name
= BLOCK_OPT_NOCOW
,
3527 .type
= QEMU_OPT_BOOL
,
3528 .help
= "Turn off copy-on-write (valid only on btrfs)"
3531 .name
= BLOCK_OPT_PREALLOC
,
3532 .type
= QEMU_OPT_STRING
,
3533 .help
= "Preallocation mode (allowed values: off"
3534 #ifdef CONFIG_POSIX_FALLOCATE
3540 .name
= BLOCK_OPT_EXTENT_SIZE_HINT
,
3541 .type
= QEMU_OPT_SIZE
,
3542 .help
= "Extent size hint for the image file, 0 to disable"
3544 { /* end of list */ }
3548 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
3551 BDRVRawState
*s
= bs
->opaque
;
3552 int input_flags
= s
->reopen_state
? s
->reopen_state
->flags
: bs
->open_flags
;
3556 /* We may need a new fd if auto-read-only switches the mode */
3557 ret
= raw_reconfigure_getfd(bs
, input_flags
, &open_flags
, perm
,
3561 } else if (ret
!= s
->fd
) {
3562 Error
*local_err
= NULL
;
3565 * Fail already check_perm() if we can't get a working O_DIRECT
3566 * alignment with the new fd.
3568 raw_probe_alignment(bs
, ret
, &local_err
);
3570 error_propagate(errp
, local_err
);
3574 s
->perm_change_fd
= ret
;
3575 s
->perm_change_flags
= open_flags
;
3578 /* Prepare permissions on old fd to avoid conflicts between old and new,
3579 * but keep everything locked that new will need. */
3580 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
3585 /* Copy locks to the new fd */
3586 if (s
->perm_change_fd
&& s
->use_lock
) {
3587 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
3590 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3597 if (s
->perm_change_fd
) {
3598 qemu_close(s
->perm_change_fd
);
3600 s
->perm_change_fd
= 0;
3604 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
3606 BDRVRawState
*s
= bs
->opaque
;
3608 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3609 * called after .bdrv_reopen_commit) */
3610 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
3612 s
->fd
= s
->perm_change_fd
;
3613 s
->open_flags
= s
->perm_change_flags
;
3615 s
->perm_change_fd
= 0;
3617 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
3619 s
->shared_perm
= shared
;
3622 static void raw_abort_perm_update(BlockDriverState
*bs
)
3624 BDRVRawState
*s
= bs
->opaque
;
3626 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3627 * the file descriptor. */
3628 if (s
->perm_change_fd
) {
3629 qemu_close(s
->perm_change_fd
);
3631 s
->perm_change_fd
= 0;
3633 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3636 static int coroutine_fn GRAPH_RDLOCK
raw_co_copy_range_from(
3637 BlockDriverState
*bs
, BdrvChild
*src
, int64_t src_offset
,
3638 BdrvChild
*dst
, int64_t dst_offset
, int64_t bytes
,
3639 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
3641 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3642 read_flags
, write_flags
);
3645 static int coroutine_fn GRAPH_RDLOCK
3646 raw_co_copy_range_to(BlockDriverState
*bs
,
3647 BdrvChild
*src
, int64_t src_offset
,
3648 BdrvChild
*dst
, int64_t dst_offset
,
3649 int64_t bytes
, BdrvRequestFlags read_flags
,
3650 BdrvRequestFlags write_flags
)
3652 RawPosixAIOData acb
;
3653 BDRVRawState
*s
= bs
->opaque
;
3654 BDRVRawState
*src_s
;
3656 assert(dst
->bs
== bs
);
3657 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
3661 src_s
= src
->bs
->opaque
;
3662 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
3666 acb
= (RawPosixAIOData
) {
3668 .aio_type
= QEMU_AIO_COPY_RANGE
,
3669 .aio_fildes
= src_s
->fd
,
3670 .aio_offset
= src_offset
,
3671 .aio_nbytes
= bytes
,
3674 .aio_offset2
= dst_offset
,
3678 return raw_thread_pool_submit(handle_aiocb_copy_range
, &acb
);
3681 BlockDriver bdrv_file
= {
3682 .format_name
= "file",
3683 .protocol_name
= "file",
3684 .instance_size
= sizeof(BDRVRawState
),
3685 .bdrv_needs_filename
= true,
3686 .bdrv_probe
= NULL
, /* no probe for protocols */
3687 .bdrv_parse_filename
= raw_parse_filename
,
3688 .bdrv_file_open
= raw_open
,
3689 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3690 .bdrv_reopen_commit
= raw_reopen_commit
,
3691 .bdrv_reopen_abort
= raw_reopen_abort
,
3692 .bdrv_close
= raw_close
,
3693 .bdrv_co_create
= raw_co_create
,
3694 .bdrv_co_create_opts
= raw_co_create_opts
,
3695 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3696 .bdrv_co_block_status
= raw_co_block_status
,
3697 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3698 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
3699 .bdrv_co_delete_file
= raw_co_delete_file
,
3701 .bdrv_co_preadv
= raw_co_preadv
,
3702 .bdrv_co_pwritev
= raw_co_pwritev
,
3703 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3704 .bdrv_co_pdiscard
= raw_co_pdiscard
,
3705 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3706 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3707 .bdrv_refresh_limits
= raw_refresh_limits
,
3708 .bdrv_co_io_plug
= raw_co_io_plug
,
3709 .bdrv_co_io_unplug
= raw_co_io_unplug
,
3710 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3712 .bdrv_co_truncate
= raw_co_truncate
,
3713 .bdrv_co_getlength
= raw_co_getlength
,
3714 .bdrv_co_get_info
= raw_co_get_info
,
3715 .bdrv_get_specific_info
= raw_get_specific_info
,
3716 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
3717 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3718 .bdrv_check_perm
= raw_check_perm
,
3719 .bdrv_set_perm
= raw_set_perm
,
3720 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3721 .create_opts
= &raw_create_opts
,
3722 .mutable_opts
= mutable_opts
,
3725 /***********************************************/
3728 #if defined(HAVE_HOST_BLOCK_DEVICE)
3730 #if defined(__APPLE__) && defined(__MACH__)
3731 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3732 CFIndex maxPathSize
, int flags
);
3734 #if !defined(MAC_OS_VERSION_12_0) \
3735 || (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_VERSION_12_0)
3736 #define IOMainPort IOMasterPort
3739 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3741 kern_return_t kernResult
= KERN_FAILURE
;
3742 mach_port_t mainPort
;
3743 CFMutableDictionaryRef classesToMatch
;
3744 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3745 char *mediaType
= NULL
;
3747 kernResult
= IOMainPort(MACH_PORT_NULL
, &mainPort
);
3748 if ( KERN_SUCCESS
!= kernResult
) {
3749 printf("IOMainPort returned %d\n", kernResult
);
3753 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3754 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3755 if (classesToMatch
== NULL
) {
3756 error_report("IOServiceMatching returned NULL for %s",
3757 matching_array
[index
]);
3760 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3762 kernResult
= IOServiceGetMatchingServices(mainPort
, classesToMatch
,
3764 if (kernResult
!= KERN_SUCCESS
) {
3765 error_report("Note: IOServiceGetMatchingServices returned %d",
3770 /* If a match was found, leave the loop */
3771 if (*mediaIterator
!= 0) {
3772 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3773 mediaType
= g_strdup(matching_array
[index
]);
3780 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3781 CFIndex maxPathSize
, int flags
)
3783 io_object_t nextMedia
;
3784 kern_return_t kernResult
= KERN_FAILURE
;
3786 nextMedia
= IOIteratorNext( mediaIterator
);
3789 CFTypeRef bsdPathAsCFString
;
3790 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3791 if ( bsdPathAsCFString
) {
3792 size_t devPathLength
;
3793 strcpy( bsdPath
, _PATH_DEV
);
3794 if (flags
& BDRV_O_NOCACHE
) {
3795 strcat(bsdPath
, "r");
3797 devPathLength
= strlen( bsdPath
);
3798 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3799 kernResult
= KERN_SUCCESS
;
3801 CFRelease( bsdPathAsCFString
);
3803 IOObjectRelease( nextMedia
);
3809 /* Sets up a real cdrom for use in QEMU */
3810 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3812 int index
, num_of_test_partitions
= 2, fd
;
3813 char test_partition
[MAXPATHLEN
];
3814 bool partition_found
= false;
3816 /* look for a working partition */
3817 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3818 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3820 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
, NULL
);
3822 partition_found
= true;
3828 /* if a working partition on the device was not found */
3829 if (partition_found
== false) {
3830 error_setg(errp
, "Failed to find a working partition on disc");
3832 trace_file_setup_cdrom(test_partition
);
3833 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3835 return partition_found
;
3838 /* Prints directions on mounting and unmounting a device */
3839 static void print_unmounting_directions(const char *file_name
)
3841 error_report("If device %s is mounted on the desktop, unmount"
3842 " it first before using it in QEMU", file_name
);
3843 error_report("Command to unmount device: diskutil unmountDisk %s",
3845 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3848 #endif /* defined(__APPLE__) && defined(__MACH__) */
3850 static int hdev_probe_device(const char *filename
)
3854 /* allow a dedicated CD-ROM driver to match with a higher priority */
3855 if (strstart(filename
, "/dev/cdrom", NULL
))
3858 if (stat(filename
, &st
) >= 0 &&
3859 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3866 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3869 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3872 static bool hdev_is_sg(BlockDriverState
*bs
)
3875 #if defined(__linux__)
3877 BDRVRawState
*s
= bs
->opaque
;
3879 struct sg_scsi_id scsiid
;
3883 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3887 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3892 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3894 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3903 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3906 BDRVRawState
*s
= bs
->opaque
;
3909 #if defined(__APPLE__) && defined(__MACH__)
3911 * Caution: while qdict_get_str() is fine, getting non-string types
3912 * would require more care. When @options come from -blockdev or
3913 * blockdev_add, its members are typed according to the QAPI
3914 * schema, but when they come from -drive, they're all QString.
3916 const char *filename
= qdict_get_str(options
, "filename");
3917 char bsd_path
[MAXPATHLEN
] = "";
3918 bool error_occurred
= false;
3920 /* If using a real cdrom */
3921 if (strcmp(filename
, "/dev/cdrom") == 0) {
3922 char *mediaType
= NULL
;
3923 kern_return_t ret_val
;
3924 io_iterator_t mediaIterator
= 0;
3926 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3927 if (mediaType
== NULL
) {
3928 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3930 error_occurred
= true;
3931 goto hdev_open_Mac_error
;
3934 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3935 if (ret_val
!= KERN_SUCCESS
) {
3936 error_setg(errp
, "Could not get BSD path for optical drive");
3937 error_occurred
= true;
3938 goto hdev_open_Mac_error
;
3941 /* If a real optical drive was not found */
3942 if (bsd_path
[0] == '\0') {
3943 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3944 error_occurred
= true;
3945 goto hdev_open_Mac_error
;
3948 /* If using a cdrom disc and finding a partition on the disc failed */
3949 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3950 setup_cdrom(bsd_path
, errp
) == false) {
3951 print_unmounting_directions(bsd_path
);
3952 error_occurred
= true;
3953 goto hdev_open_Mac_error
;
3956 qdict_put_str(options
, "filename", bsd_path
);
3958 hdev_open_Mac_error
:
3960 if (mediaIterator
) {
3961 IOObjectRelease(mediaIterator
);
3963 if (error_occurred
) {
3967 #endif /* defined(__APPLE__) && defined(__MACH__) */
3969 s
->type
= FTYPE_FILE
;
3971 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
3973 #if defined(__APPLE__) && defined(__MACH__)
3975 filename
= bsd_path
;
3977 /* if a physical device experienced an error while being opened */
3978 if (strncmp(filename
, "/dev/", 5) == 0) {
3979 print_unmounting_directions(filename
);
3981 #endif /* defined(__APPLE__) && defined(__MACH__) */
3985 /* Since this does ioctl the device must be already opened */
3986 bs
->sg
= hdev_is_sg(bs
);
3991 #if defined(__linux__)
3992 static int coroutine_fn
3993 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3995 BDRVRawState
*s
= bs
->opaque
;
3996 RawPosixAIOData acb
;
4004 if (req
== SG_IO
&& s
->pr_mgr
) {
4005 struct sg_io_hdr
*io_hdr
= buf
;
4006 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
4007 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
4008 return pr_manager_execute(s
->pr_mgr
, qemu_get_current_aio_context(),
4013 acb
= (RawPosixAIOData
) {
4015 .aio_type
= QEMU_AIO_IOCTL
,
4016 .aio_fildes
= s
->fd
,
4024 return raw_thread_pool_submit(handle_aiocb_ioctl
, &acb
);
4028 static coroutine_fn
int
4029 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
4031 BDRVRawState
*s
= bs
->opaque
;
4036 raw_account_discard(s
, bytes
, ret
);
4039 return raw_do_pdiscard(bs
, offset
, bytes
, true);
4042 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
4043 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
)
4052 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
4055 static BlockDriver bdrv_host_device
= {
4056 .format_name
= "host_device",
4057 .protocol_name
= "host_device",
4058 .instance_size
= sizeof(BDRVRawState
),
4059 .bdrv_needs_filename
= true,
4060 .bdrv_probe_device
= hdev_probe_device
,
4061 .bdrv_parse_filename
= hdev_parse_filename
,
4062 .bdrv_file_open
= hdev_open
,
4063 .bdrv_close
= raw_close
,
4064 .bdrv_reopen_prepare
= raw_reopen_prepare
,
4065 .bdrv_reopen_commit
= raw_reopen_commit
,
4066 .bdrv_reopen_abort
= raw_reopen_abort
,
4067 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
4068 .create_opts
= &bdrv_create_opts_simple
,
4069 .mutable_opts
= mutable_opts
,
4070 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
4071 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
4073 .bdrv_co_preadv
= raw_co_preadv
,
4074 .bdrv_co_pwritev
= raw_co_pwritev
,
4075 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
4076 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
4077 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
4078 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
4079 .bdrv_refresh_limits
= raw_refresh_limits
,
4080 .bdrv_co_io_plug
= raw_co_io_plug
,
4081 .bdrv_co_io_unplug
= raw_co_io_unplug
,
4082 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
4084 .bdrv_co_truncate
= raw_co_truncate
,
4085 .bdrv_co_getlength
= raw_co_getlength
,
4086 .bdrv_co_get_info
= raw_co_get_info
,
4087 .bdrv_get_specific_info
= raw_get_specific_info
,
4088 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
4089 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
4090 .bdrv_check_perm
= raw_check_perm
,
4091 .bdrv_set_perm
= raw_set_perm
,
4092 .bdrv_abort_perm_update
= raw_abort_perm_update
,
4093 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
4094 .bdrv_probe_geometry
= hdev_probe_geometry
,
4096 /* generic scsi device */
4098 .bdrv_co_ioctl
= hdev_co_ioctl
,
4102 #if defined(CONFIG_BLKZONED)
4103 /* zone management operations */
4104 .bdrv_co_zone_report
= raw_co_zone_report
,
4105 .bdrv_co_zone_mgmt
= raw_co_zone_mgmt
,
4109 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
4110 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
4113 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
4116 static void cdrom_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
4118 bs
->bl
.has_variable_length
= true;
4119 raw_refresh_limits(bs
, errp
);
4124 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
4127 BDRVRawState
*s
= bs
->opaque
;
4131 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
4132 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
4135 static int cdrom_probe_device(const char *filename
)
4141 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
, NULL
);
4145 ret
= fstat(fd
, &st
);
4146 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
4150 /* Attempt to detect via a CDROM specific ioctl */
4151 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
4161 static bool coroutine_fn
cdrom_co_is_inserted(BlockDriverState
*bs
)
4163 BDRVRawState
*s
= bs
->opaque
;
4166 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
4167 return ret
== CDS_DISC_OK
;
4170 static void coroutine_fn
cdrom_co_eject(BlockDriverState
*bs
, bool eject_flag
)
4172 BDRVRawState
*s
= bs
->opaque
;
4175 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
4176 perror("CDROMEJECT");
4178 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
4179 perror("CDROMEJECT");
4183 static void coroutine_fn
cdrom_co_lock_medium(BlockDriverState
*bs
, bool locked
)
4185 BDRVRawState
*s
= bs
->opaque
;
4187 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
4189 * Note: an error can happen if the distribution automatically
4192 /* perror("CDROM_LOCKDOOR"); */
4196 static BlockDriver bdrv_host_cdrom
= {
4197 .format_name
= "host_cdrom",
4198 .protocol_name
= "host_cdrom",
4199 .instance_size
= sizeof(BDRVRawState
),
4200 .bdrv_needs_filename
= true,
4201 .bdrv_probe_device
= cdrom_probe_device
,
4202 .bdrv_parse_filename
= cdrom_parse_filename
,
4203 .bdrv_file_open
= cdrom_open
,
4204 .bdrv_close
= raw_close
,
4205 .bdrv_reopen_prepare
= raw_reopen_prepare
,
4206 .bdrv_reopen_commit
= raw_reopen_commit
,
4207 .bdrv_reopen_abort
= raw_reopen_abort
,
4208 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
4209 .create_opts
= &bdrv_create_opts_simple
,
4210 .mutable_opts
= mutable_opts
,
4211 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
4213 .bdrv_co_preadv
= raw_co_preadv
,
4214 .bdrv_co_pwritev
= raw_co_pwritev
,
4215 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
4216 .bdrv_refresh_limits
= cdrom_refresh_limits
,
4217 .bdrv_co_io_plug
= raw_co_io_plug
,
4218 .bdrv_co_io_unplug
= raw_co_io_unplug
,
4219 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
4221 .bdrv_co_truncate
= raw_co_truncate
,
4222 .bdrv_co_getlength
= raw_co_getlength
,
4223 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
4225 /* removable device support */
4226 .bdrv_co_is_inserted
= cdrom_co_is_inserted
,
4227 .bdrv_co_eject
= cdrom_co_eject
,
4228 .bdrv_co_lock_medium
= cdrom_co_lock_medium
,
4230 /* generic scsi device */
4231 .bdrv_co_ioctl
= hdev_co_ioctl
,
4233 #endif /* __linux__ */
4235 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
4236 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
4239 BDRVRawState
*s
= bs
->opaque
;
4244 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
4249 /* make sure the door isn't locked at this time */
4250 ioctl(s
->fd
, CDIOCALLOW
);
4254 static int cdrom_probe_device(const char *filename
)
4256 if (strstart(filename
, "/dev/cd", NULL
) ||
4257 strstart(filename
, "/dev/acd", NULL
))
4262 static int cdrom_reopen(BlockDriverState
*bs
)
4264 BDRVRawState
*s
= bs
->opaque
;
4268 * Force reread of possibly changed/newly loaded disc,
4269 * FreeBSD seems to not notice sometimes...
4273 fd
= qemu_open(bs
->filename
, s
->open_flags
, NULL
);
4280 /* make sure the door isn't locked at this time */
4281 ioctl(s
->fd
, CDIOCALLOW
);
4285 static bool coroutine_fn
cdrom_co_is_inserted(BlockDriverState
*bs
)
4287 return raw_co_getlength(bs
) > 0;
4290 static void coroutine_fn
cdrom_co_eject(BlockDriverState
*bs
, bool eject_flag
)
4292 BDRVRawState
*s
= bs
->opaque
;
4297 (void) ioctl(s
->fd
, CDIOCALLOW
);
4300 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
4301 perror("CDIOCEJECT");
4303 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
4304 perror("CDIOCCLOSE");
4310 static void coroutine_fn
cdrom_co_lock_medium(BlockDriverState
*bs
, bool locked
)
4312 BDRVRawState
*s
= bs
->opaque
;
4316 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
4318 * Note: an error can happen if the distribution automatically
4321 /* perror("CDROM_LOCKDOOR"); */
4325 static BlockDriver bdrv_host_cdrom
= {
4326 .format_name
= "host_cdrom",
4327 .protocol_name
= "host_cdrom",
4328 .instance_size
= sizeof(BDRVRawState
),
4329 .bdrv_needs_filename
= true,
4330 .bdrv_probe_device
= cdrom_probe_device
,
4331 .bdrv_parse_filename
= cdrom_parse_filename
,
4332 .bdrv_file_open
= cdrom_open
,
4333 .bdrv_close
= raw_close
,
4334 .bdrv_reopen_prepare
= raw_reopen_prepare
,
4335 .bdrv_reopen_commit
= raw_reopen_commit
,
4336 .bdrv_reopen_abort
= raw_reopen_abort
,
4337 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
4338 .create_opts
= &bdrv_create_opts_simple
,
4339 .mutable_opts
= mutable_opts
,
4341 .bdrv_co_preadv
= raw_co_preadv
,
4342 .bdrv_co_pwritev
= raw_co_pwritev
,
4343 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
4344 .bdrv_refresh_limits
= cdrom_refresh_limits
,
4345 .bdrv_co_io_plug
= raw_co_io_plug
,
4346 .bdrv_co_io_unplug
= raw_co_io_unplug
,
4347 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
4349 .bdrv_co_truncate
= raw_co_truncate
,
4350 .bdrv_co_getlength
= raw_co_getlength
,
4351 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
4353 /* removable device support */
4354 .bdrv_co_is_inserted
= cdrom_co_is_inserted
,
4355 .bdrv_co_eject
= cdrom_co_eject
,
4356 .bdrv_co_lock_medium
= cdrom_co_lock_medium
,
4358 #endif /* __FreeBSD__ */
4360 #endif /* HAVE_HOST_BLOCK_DEVICE */
4362 static void bdrv_file_init(void)
4365 * Register all the drivers. Note that order is important, the driver
4366 * registered last will get probed first.
4368 bdrv_register(&bdrv_file
);
4369 #if defined(HAVE_HOST_BLOCK_DEVICE)
4370 bdrv_register(&bdrv_host_device
);
4372 bdrv_register(&bdrv_host_cdrom
);
4374 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
4375 bdrv_register(&bdrv_host_cdrom
);
4377 #endif /* HAVE_HOST_BLOCK_DEVICE */
4380 block_init(bdrv_file_init
);