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 raw_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
&& !(s
->open_flags
& O_DIRECT
)) {
716 error_setg(errp
, "aio=native was specified, but it requires "
717 "cache.direct=on, which was not specified.");
722 if (s
->use_linux_aio
) {
723 error_setg(errp
, "aio=native was specified, but is not supported "
728 #endif /* !defined(CONFIG_LINUX_AIO) */
730 #ifndef CONFIG_LINUX_IO_URING
731 if (s
->use_linux_io_uring
) {
732 error_setg(errp
, "aio=io_uring was specified, but is not supported "
737 #endif /* !defined(CONFIG_LINUX_IO_URING) */
739 s
->has_discard
= true;
740 s
->has_write_zeroes
= true;
742 if (fstat(s
->fd
, &st
) < 0) {
744 error_setg_errno(errp
, errno
, "Could not stat file");
749 if (!S_ISREG(st
.st_mode
)) {
750 error_setg(errp
, "'%s' driver requires '%s' to be a regular file",
751 bs
->drv
->format_name
, bs
->filename
);
755 s
->has_fallocate
= true;
758 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
759 error_setg(errp
, "'%s' driver requires '%s' to be either "
760 "a character or block device",
761 bs
->drv
->format_name
, bs
->filename
);
766 #ifdef CONFIG_BLKZONED
768 * The kernel page cache does not reliably work for writes to SWR zones
769 * of zoned block device because it can not guarantee the order of writes.
771 if ((bs
->bl
.zoned
!= BLK_Z_NONE
) &&
772 (!(s
->open_flags
& O_DIRECT
))) {
773 error_setg(errp
, "The driver supports zoned devices, and it requires "
774 "cache.direct=on, which was not specified.");
775 return -EINVAL
; /* No host kernel page cache */
779 if (S_ISBLK(st
.st_mode
)) {
781 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
782 * not rely on the contents of discarded blocks unless using O_DIRECT.
783 * Same for BLKZEROOUT.
785 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
786 s
->has_write_zeroes
= false;
791 if (S_ISCHR(st
.st_mode
)) {
793 * The file is a char device (disk), which on FreeBSD isn't behind
794 * a pager, so force all requests to be aligned. This is needed
795 * so QEMU makes sure all IO operations on the device are aligned
796 * to sector size, or else FreeBSD will reject them with EINVAL.
798 s
->force_alignment
= true;
801 s
->needs_alignment
= raw_needs_alignment(bs
);
803 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
804 if (S_ISREG(st
.st_mode
)) {
805 /* When extending regular files, we get zeros from the OS */
806 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
810 if (ret
< 0 && s
->fd
!= -1) {
813 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
820 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
823 BDRVRawState
*s
= bs
->opaque
;
825 s
->type
= FTYPE_FILE
;
826 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
835 #define PERM_FOREACH(i) \
836 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
838 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
839 * file; if @unlock == true, also unlock the unneeded bytes.
840 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
842 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
843 uint64_t perm_lock_bits
,
844 uint64_t shared_perm_lock_bits
,
845 bool unlock
, Error
**errp
)
849 uint64_t locked_perm
, locked_shared_perm
;
852 locked_perm
= s
->locked_perm
;
853 locked_shared_perm
= s
->locked_shared_perm
;
856 * We don't have the previous bits, just lock/unlock for each of the
860 locked_perm
= BLK_PERM_ALL
;
861 locked_shared_perm
= BLK_PERM_ALL
;
864 locked_shared_perm
= 0;
869 int off
= RAW_LOCK_PERM_BASE
+ i
;
870 uint64_t bit
= (1ULL << i
);
871 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
872 ret
= qemu_lock_fd(fd
, off
, 1, false);
874 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
878 s
->locked_perm
|= bit
;
880 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
881 ret
= qemu_unlock_fd(fd
, off
, 1);
883 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
886 s
->locked_perm
&= ~bit
;
891 int off
= RAW_LOCK_SHARED_BASE
+ i
;
892 uint64_t bit
= (1ULL << i
);
893 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
894 ret
= qemu_lock_fd(fd
, off
, 1, false);
896 raw_lock_error_setg_errno(errp
, -ret
, "Failed to lock byte %d",
900 s
->locked_shared_perm
|= bit
;
902 } else if (unlock
&& (locked_shared_perm
& bit
) &&
903 !(shared_perm_lock_bits
& bit
)) {
904 ret
= qemu_unlock_fd(fd
, off
, 1);
906 error_setg_errno(errp
, -ret
, "Failed to unlock byte %d", off
);
909 s
->locked_shared_perm
&= ~bit
;
916 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
917 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
924 int off
= RAW_LOCK_SHARED_BASE
+ i
;
925 uint64_t p
= 1ULL << i
;
927 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
929 char *perm_name
= bdrv_perm_names(p
);
931 raw_lock_error_setg_errno(errp
, -ret
,
932 "Failed to get \"%s\" lock",
940 int off
= RAW_LOCK_PERM_BASE
+ i
;
941 uint64_t p
= 1ULL << i
;
942 if (!(shared_perm
& p
)) {
943 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
945 char *perm_name
= bdrv_perm_names(p
);
947 raw_lock_error_setg_errno(errp
, -ret
,
948 "Failed to get shared \"%s\" lock",
958 static int raw_handle_perm_lock(BlockDriverState
*bs
,
960 uint64_t new_perm
, uint64_t new_shared
,
963 BDRVRawState
*s
= bs
->opaque
;
965 Error
*local_err
= NULL
;
971 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
977 if ((s
->perm
| new_perm
) == s
->perm
&&
978 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
981 * We are going to unlock bytes, it should not fail. If it fail due
982 * to some fs-dependent permission-unrelated reasons (which occurs
983 * sometimes on NFS and leads to abort in bdrv_replace_child) we
984 * can't prevent such errors by any check here. And we ignore them
985 * anyway in ABORT and COMMIT.
989 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
990 ~s
->shared_perm
| ~new_shared
,
993 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
997 error_append_hint(errp
,
998 "Is another process using the image [%s]?\n",
1001 /* fall through to unlock bytes. */
1003 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
1006 /* Theoretically the above call only unlocks bytes and it cannot
1007 * fail. Something weird happened, report it.
1009 warn_report_err(local_err
);
1013 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
1016 /* Theoretically the above call only unlocks bytes and it cannot
1017 * fail. Something weird happened, report it.
1019 warn_report_err(local_err
);
1026 /* Sets a specific flag */
1027 static int fcntl_setfl(int fd
, int flag
)
1031 flags
= fcntl(fd
, F_GETFL
);
1035 if (fcntl(fd
, F_SETFL
, flags
| flag
) == -1) {
1041 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
1042 int *open_flags
, uint64_t perm
, bool force_dup
,
1045 BDRVRawState
*s
= bs
->opaque
;
1048 bool has_writers
= perm
&
1049 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
1050 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
1052 fcntl_flags
|= O_NOATIME
;
1056 if (s
->type
== FTYPE_CD
) {
1057 *open_flags
|= O_NONBLOCK
;
1060 raw_parse_flags(flags
, open_flags
, has_writers
);
1063 /* Not all operating systems have O_ASYNC, and those that don't
1064 * will not let us track the state into rs->open_flags (typically
1065 * you achieve the same effect with an ioctl, for example I_SETSIG
1066 * on Solaris). But we do not use O_ASYNC, so that's fine.
1068 assert((s
->open_flags
& O_ASYNC
) == 0);
1071 if (!force_dup
&& *open_flags
== s
->open_flags
) {
1072 /* We're lucky, the existing fd is fine */
1076 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
1077 /* dup the original fd */
1078 fd
= qemu_dup(s
->fd
);
1080 ret
= fcntl_setfl(fd
, *open_flags
);
1088 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
1090 const char *normalized_filename
= bs
->filename
;
1091 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
1093 fd
= qemu_open(normalized_filename
, *open_flags
, errp
);
1100 if (fd
!= -1 && (*open_flags
& O_RDWR
)) {
1101 ret
= check_hdev_writable(fd
);
1104 error_setg_errno(errp
, -ret
, "The device is not writable");
1112 static int raw_reopen_prepare(BDRVReopenState
*state
,
1113 BlockReopenQueue
*queue
, Error
**errp
)
1116 BDRVRawReopenState
*rs
;
1120 assert(state
!= NULL
);
1121 assert(state
->bs
!= NULL
);
1123 s
= state
->bs
->opaque
;
1125 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
1128 /* Handle options changes */
1129 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
1130 if (!qemu_opts_absorb_qdict(opts
, state
->options
, errp
)) {
1135 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
1136 rs
->check_cache_dropped
=
1137 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
1139 /* This driver's reopen function doesn't currently allow changing
1140 * other options, so let's put them back in the original QDict and
1141 * bdrv_reopen_prepare() will detect changes and complain. */
1142 qemu_opts_to_qdict(opts
, state
->options
);
1145 * As part of reopen prepare we also want to create new fd by
1146 * raw_reconfigure_getfd(). But it wants updated "perm", when in
1147 * bdrv_reopen_multiple() .bdrv_reopen_prepare() callback called prior to
1148 * permission update. Happily, permission update is always a part
1149 * (a separate stage) of bdrv_reopen_multiple() so we can rely on this
1150 * fact and reconfigure fd in raw_check_perm().
1153 s
->reopen_state
= state
;
1157 qemu_opts_del(opts
);
1161 static void raw_reopen_commit(BDRVReopenState
*state
)
1163 BDRVRawReopenState
*rs
= state
->opaque
;
1164 BDRVRawState
*s
= state
->bs
->opaque
;
1166 s
->drop_cache
= rs
->drop_cache
;
1167 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1168 s
->open_flags
= rs
->open_flags
;
1169 g_free(state
->opaque
);
1170 state
->opaque
= NULL
;
1172 assert(s
->reopen_state
== state
);
1173 s
->reopen_state
= NULL
;
1177 static void raw_reopen_abort(BDRVReopenState
*state
)
1179 BDRVRawReopenState
*rs
= state
->opaque
;
1180 BDRVRawState
*s
= state
->bs
->opaque
;
1182 /* nothing to do if NULL, we didn't get far enough */
1187 g_free(state
->opaque
);
1188 state
->opaque
= NULL
;
1190 assert(s
->reopen_state
== state
);
1191 s
->reopen_state
= NULL
;
1194 static int hdev_get_max_hw_transfer(int fd
, struct stat
*st
)
1197 if (S_ISBLK(st
->st_mode
)) {
1198 unsigned short max_sectors
= 0;
1199 if (ioctl(fd
, BLKSECTGET
, &max_sectors
) == 0) {
1200 return max_sectors
* 512;
1204 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1215 * Get a sysfs attribute value as character string.
1218 static int get_sysfs_str_val(struct stat
*st
, const char *attribute
,
1220 g_autofree
char *sysfspath
= NULL
;
1223 if (!S_ISBLK(st
->st_mode
)) {
1227 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/%s",
1228 major(st
->st_rdev
), minor(st
->st_rdev
),
1230 if (!g_file_get_contents(sysfspath
, val
, &len
, NULL
)) {
1234 /* The file is ended with '\n' */
1237 if (*(p
+ len
- 1) == '\n') {
1238 *(p
+ len
- 1) = '\0';
1244 #if defined(CONFIG_BLKZONED)
1245 static int get_sysfs_zoned_model(struct stat
*st
, BlockZoneModel
*zoned
)
1247 g_autofree
char *val
= NULL
;
1250 ret
= get_sysfs_str_val(st
, "zoned", &val
);
1255 if (strcmp(val
, "host-managed") == 0) {
1257 } else if (strcmp(val
, "host-aware") == 0) {
1259 } else if (strcmp(val
, "none") == 0) {
1260 *zoned
= BLK_Z_NONE
;
1266 #endif /* defined(CONFIG_BLKZONED) */
1269 * Get a sysfs attribute value as a long integer.
1272 static long get_sysfs_long_val(struct stat
*st
, const char *attribute
)
1274 g_autofree
char *str
= NULL
;
1279 ret
= get_sysfs_str_val(st
, attribute
, &str
);
1284 /* The file is ended with '\n', pass 'end' to accept that. */
1285 ret
= qemu_strtol(str
, &end
, 10, &val
);
1286 if (ret
== 0 && end
&& *end
== '\0') {
1293 static int hdev_get_max_segments(int fd
, struct stat
*st
)
1298 if (S_ISCHR(st
->st_mode
)) {
1299 if (ioctl(fd
, SG_GET_SG_TABLESIZE
, &ret
) == 0) {
1304 return get_sysfs_long_val(st
, "max_segments");
1310 #if defined(CONFIG_BLKZONED)
1312 * If the reset_all flag is true, then the wps of zone whose state is
1313 * not readonly or offline should be all reset to the start sector.
1314 * Else, take the real wp of the device.
1316 static int get_zones_wp(BlockDriverState
*bs
, int fd
, int64_t offset
,
1317 unsigned int nrz
, bool reset_all
)
1319 struct blk_zone
*blkz
;
1321 uint64_t sector
= offset
>> BDRV_SECTOR_BITS
;
1322 BlockZoneWps
*wps
= bs
->wps
;
1323 unsigned int j
= offset
/ bs
->bl
.zone_size
;
1324 unsigned int n
= 0, i
= 0;
1326 rep_size
= sizeof(struct blk_zone_report
) + nrz
* sizeof(struct blk_zone
);
1327 g_autofree
struct blk_zone_report
*rep
= NULL
;
1329 rep
= g_malloc(rep_size
);
1330 blkz
= (struct blk_zone
*)(rep
+ 1);
1332 memset(rep
, 0, rep_size
);
1333 rep
->sector
= sector
;
1334 rep
->nr_zones
= nrz
- n
;
1337 ret
= ioctl(fd
, BLKREPORTZONE
, rep
);
1338 } while (ret
!= 0 && errno
== EINTR
);
1340 error_report("%d: ioctl BLKREPORTZONE at %" PRId64
" failed %d",
1345 if (!rep
->nr_zones
) {
1349 for (i
= 0; i
< rep
->nr_zones
; ++i
, ++n
, ++j
) {
1351 * The wp tracking cares only about sequential writes required and
1352 * sequential write preferred zones so that the wp can advance to
1353 * the right location.
1354 * Use the most significant bit of the wp location to indicate the
1355 * zone type: 0 for SWR/SWP zones and 1 for conventional zones.
1357 if (blkz
[i
].type
== BLK_ZONE_TYPE_CONVENTIONAL
) {
1358 wps
->wp
[j
] |= 1ULL << 63;
1360 switch(blkz
[i
].cond
) {
1361 case BLK_ZONE_COND_FULL
:
1362 case BLK_ZONE_COND_READONLY
:
1363 /* Zone not writable */
1364 wps
->wp
[j
] = (blkz
[i
].start
+ blkz
[i
].len
) << BDRV_SECTOR_BITS
;
1366 case BLK_ZONE_COND_OFFLINE
:
1367 /* Zone not writable nor readable */
1368 wps
->wp
[j
] = (blkz
[i
].start
) << BDRV_SECTOR_BITS
;
1372 wps
->wp
[j
] = blkz
[i
].start
<< BDRV_SECTOR_BITS
;
1374 wps
->wp
[j
] = blkz
[i
].wp
<< BDRV_SECTOR_BITS
;
1380 sector
= blkz
[i
- 1].start
+ blkz
[i
- 1].len
;
1386 static void update_zones_wp(BlockDriverState
*bs
, int fd
, int64_t offset
,
1389 if (get_zones_wp(bs
, fd
, offset
, nrz
, 0) < 0) {
1390 error_report("update zone wp failed");
1394 static void raw_refresh_zoned_limits(BlockDriverState
*bs
, struct stat
*st
,
1397 BDRVRawState
*s
= bs
->opaque
;
1398 BlockZoneModel zoned
;
1401 ret
= get_sysfs_zoned_model(st
, &zoned
);
1402 if (ret
< 0 || zoned
== BLK_Z_NONE
) {
1405 bs
->bl
.zoned
= zoned
;
1407 ret
= get_sysfs_long_val(st
, "max_open_zones");
1409 bs
->bl
.max_open_zones
= ret
;
1412 ret
= get_sysfs_long_val(st
, "max_active_zones");
1414 bs
->bl
.max_active_zones
= ret
;
1418 * The zoned device must at least have zone size and nr_zones fields.
1420 ret
= get_sysfs_long_val(st
, "chunk_sectors");
1422 error_setg_errno(errp
, -ret
, "Unable to read chunk_sectors "
1426 error_setg(errp
, "Read 0 from chunk_sectors sysfs attribute");
1429 bs
->bl
.zone_size
= ret
<< BDRV_SECTOR_BITS
;
1431 ret
= get_sysfs_long_val(st
, "nr_zones");
1433 error_setg_errno(errp
, -ret
, "Unable to read nr_zones "
1437 error_setg(errp
, "Read 0 from nr_zones sysfs attribute");
1440 bs
->bl
.nr_zones
= ret
;
1442 ret
= get_sysfs_long_val(st
, "zone_append_max_bytes");
1444 bs
->bl
.max_append_sectors
= ret
>> BDRV_SECTOR_BITS
;
1447 ret
= get_sysfs_long_val(st
, "physical_block_size");
1449 bs
->bl
.write_granularity
= ret
;
1452 /* The refresh_limits() function can be called multiple times. */
1454 bs
->wps
= g_malloc(sizeof(BlockZoneWps
) +
1455 sizeof(int64_t) * bs
->bl
.nr_zones
);
1456 ret
= get_zones_wp(bs
, s
->fd
, 0, bs
->bl
.nr_zones
, 0);
1458 error_setg_errno(errp
, -ret
, "report wps failed");
1461 qemu_co_mutex_init(&bs
->wps
->colock
);
1465 bs
->bl
.zoned
= BLK_Z_NONE
;
1469 #else /* !defined(CONFIG_BLKZONED) */
1470 static void raw_refresh_zoned_limits(BlockDriverState
*bs
, struct stat
*st
,
1473 bs
->bl
.zoned
= BLK_Z_NONE
;
1475 #endif /* !defined(CONFIG_BLKZONED) */
1477 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1479 BDRVRawState
*s
= bs
->opaque
;
1482 s
->needs_alignment
= raw_needs_alignment(bs
);
1483 raw_probe_alignment(bs
, s
->fd
, errp
);
1485 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1486 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, qemu_real_host_page_size());
1489 * Maximum transfers are best effort, so it is okay to ignore any
1490 * errors. That said, based on the man page errors in fstat would be
1491 * very much unexpected; the only possible case seems to be ENOMEM.
1493 if (fstat(s
->fd
, &st
)) {
1497 #if defined(__APPLE__) && (__MACH__)
1500 if (!fstatfs(s
->fd
, &buf
)) {
1501 bs
->bl
.opt_transfer
= buf
.f_iosize
;
1502 bs
->bl
.pdiscard_alignment
= buf
.f_bsize
;
1506 if (bdrv_is_sg(bs
) || S_ISBLK(st
.st_mode
)) {
1507 int ret
= hdev_get_max_hw_transfer(s
->fd
, &st
);
1509 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1510 bs
->bl
.max_hw_transfer
= ret
;
1513 ret
= hdev_get_max_segments(s
->fd
, &st
);
1515 bs
->bl
.max_hw_iov
= ret
;
1519 raw_refresh_zoned_limits(bs
, &st
, errp
);
1522 static int check_for_dasd(int fd
)
1525 struct dasd_information2_t info
= {0};
1527 return ioctl(fd
, BIODASDINFO2
, &info
);
1534 * Try to get @bs's logical and physical block size.
1535 * On success, store them in @bsz and return zero.
1536 * On failure, return negative errno.
1538 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1540 BDRVRawState
*s
= bs
->opaque
;
1543 /* If DASD or zoned devices, get blocksizes */
1544 if (check_for_dasd(s
->fd
) < 0) {
1545 /* zoned devices are not DASD */
1546 if (bs
->bl
.zoned
== BLK_Z_NONE
) {
1550 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1554 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1558 * Try to get @bs's geometry: cyls, heads, sectors.
1559 * On success, store them in @geo and return 0.
1560 * On failure return -errno.
1561 * (Allows block driver to assign default geometry values that guest sees)
1564 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1566 BDRVRawState
*s
= bs
->opaque
;
1567 struct hd_geometry ioctl_geo
= {0};
1569 /* If DASD, get its geometry */
1570 if (check_for_dasd(s
->fd
) < 0) {
1573 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1576 /* HDIO_GETGEO may return success even though geo contains zeros
1577 (e.g. certain multipath setups) */
1578 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1581 /* Do not return a geometry for partition */
1582 if (ioctl_geo
.start
!= 0) {
1585 geo
->heads
= ioctl_geo
.heads
;
1586 geo
->sectors
= ioctl_geo
.sectors
;
1587 geo
->cylinders
= ioctl_geo
.cylinders
;
1591 #else /* __linux__ */
1592 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1598 #if defined(__linux__)
1599 static int handle_aiocb_ioctl(void *opaque
)
1601 RawPosixAIOData
*aiocb
= opaque
;
1604 ret
= RETRY_ON_EINTR(
1605 ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
)
1615 static int handle_aiocb_flush(void *opaque
)
1617 RawPosixAIOData
*aiocb
= opaque
;
1618 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1621 if (s
->page_cache_inconsistent
) {
1622 return -s
->page_cache_inconsistent
;
1625 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1627 trace_file_flush_fdatasync_failed(errno
);
1629 /* There is no clear definition of the semantics of a failing fsync(),
1630 * so we may have to assume the worst. The sad truth is that this
1631 * assumption is correct for Linux. Some pages are now probably marked
1632 * clean in the page cache even though they are inconsistent with the
1633 * on-disk contents. The next fdatasync() call would succeed, but no
1634 * further writeback attempt will be made. We can't get back to a state
1635 * in which we know what is on disk (we would have to rewrite
1636 * everything that was touched since the last fdatasync() at least), so
1637 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1638 * really defined, I have little hope that other OSes are doing better.
1640 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1642 if ((s
->open_flags
& O_DIRECT
) == 0) {
1643 s
->page_cache_inconsistent
= errno
;
1650 #ifdef CONFIG_PREADV
1652 static bool preadv_present
= true;
1655 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1657 return preadv(fd
, iov
, nr_iov
, offset
);
1661 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1663 return pwritev(fd
, iov
, nr_iov
, offset
);
1668 static bool preadv_present
= false;
1671 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1677 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1684 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1688 len
= RETRY_ON_EINTR(
1689 (aiocb
->aio_type
& (QEMU_AIO_WRITE
| QEMU_AIO_ZONE_APPEND
)) ?
1690 qemu_pwritev(aiocb
->aio_fildes
,
1693 aiocb
->aio_offset
) :
1694 qemu_preadv(aiocb
->aio_fildes
,
1707 * Read/writes the data to/from a given linear buffer.
1709 * Returns the number of bytes handles or -errno in case of an error. Short
1710 * reads are only returned if the end of the file is reached.
1712 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1717 while (offset
< aiocb
->aio_nbytes
) {
1718 if (aiocb
->aio_type
& (QEMU_AIO_WRITE
| QEMU_AIO_ZONE_APPEND
)) {
1719 len
= pwrite(aiocb
->aio_fildes
,
1720 (const char *)buf
+ offset
,
1721 aiocb
->aio_nbytes
- offset
,
1722 aiocb
->aio_offset
+ offset
);
1724 len
= pread(aiocb
->aio_fildes
,
1726 aiocb
->aio_nbytes
- offset
,
1727 aiocb
->aio_offset
+ offset
);
1729 if (len
== -1 && errno
== EINTR
) {
1731 } else if (len
== -1 && errno
== EINVAL
&&
1732 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1733 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1735 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1736 * after a short read. Assume that O_DIRECT short reads only occur
1737 * at EOF. Therefore this is a short read, not an I/O error.
1740 } else if (len
== -1) {
1743 } else if (len
== 0) {
1752 static int handle_aiocb_rw(void *opaque
)
1754 RawPosixAIOData
*aiocb
= opaque
;
1758 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1760 * If there is just a single buffer, and it is properly aligned
1761 * we can just use plain pread/pwrite without any problems.
1763 if (aiocb
->io
.niov
== 1) {
1764 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1768 * We have more than one iovec, and all are properly aligned.
1770 * Try preadv/pwritev first and fall back to linearizing the
1771 * buffer if it's not supported.
1773 if (preadv_present
) {
1774 nbytes
= handle_aiocb_rw_vector(aiocb
);
1775 if (nbytes
== aiocb
->aio_nbytes
||
1776 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1779 preadv_present
= false;
1783 * XXX(hch): short read/write. no easy way to handle the reminder
1784 * using these interfaces. For now retry using plain
1790 * Ok, we have to do it the hard way, copy all segments into
1791 * a single aligned buffer.
1793 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1799 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1803 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1804 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1805 p
+= aiocb
->io
.iov
[i
].iov_len
;
1807 assert(p
- buf
== aiocb
->aio_nbytes
);
1810 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1811 if (!(aiocb
->aio_type
& (QEMU_AIO_WRITE
| QEMU_AIO_ZONE_APPEND
))) {
1813 size_t count
= aiocb
->aio_nbytes
, copy
;
1816 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1818 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1819 copy
= aiocb
->io
.iov
[i
].iov_len
;
1821 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1822 assert(count
>= copy
);
1831 if (nbytes
== aiocb
->aio_nbytes
) {
1833 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1834 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1837 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1838 0, aiocb
->aio_nbytes
- nbytes
);
1847 #if defined(CONFIG_FALLOCATE) || defined(BLKZEROOUT) || defined(BLKDISCARD)
1848 static int translate_err(int err
)
1850 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1858 #ifdef CONFIG_FALLOCATE
1859 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1862 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1865 } while (errno
== EINTR
);
1866 return translate_err(-errno
);
1870 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1873 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1875 if (!s
->has_write_zeroes
) {
1880 /* The BLKZEROOUT implementation in the kernel doesn't set
1881 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1883 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1885 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1886 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1889 } while (errno
== EINTR
);
1891 ret
= translate_err(-errno
);
1892 if (ret
== -ENOTSUP
) {
1893 s
->has_write_zeroes
= false;
1901 static int handle_aiocb_write_zeroes(void *opaque
)
1903 RawPosixAIOData
*aiocb
= opaque
;
1904 #ifdef CONFIG_FALLOCATE
1905 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1909 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1910 return handle_aiocb_write_zeroes_block(aiocb
);
1913 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1914 if (s
->has_write_zeroes
) {
1915 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1916 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1917 if (ret
== -ENOTSUP
) {
1918 s
->has_write_zeroes
= false;
1919 } else if (ret
== 0 || ret
!= -EINVAL
) {
1923 * Note: Some file systems do not like unaligned byte ranges, and
1924 * return EINVAL in such a case, though they should not do it according
1925 * to the man-page of fallocate(). Thus we simply ignore this return
1926 * value and try the other fallbacks instead.
1931 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1932 if (s
->has_discard
&& s
->has_fallocate
) {
1933 int ret
= do_fallocate(s
->fd
,
1934 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1935 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1937 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1938 if (ret
== 0 || ret
!= -ENOTSUP
) {
1941 s
->has_fallocate
= false;
1942 } else if (ret
== -EINVAL
) {
1944 * Some file systems like older versions of GPFS do not like un-
1945 * aligned byte ranges, and return EINVAL in such a case, though
1946 * they should not do it according to the man-page of fallocate().
1947 * Warn about the bad filesystem and try the final fallback instead.
1949 warn_report_once("Your file system is misbehaving: "
1950 "fallocate(FALLOC_FL_PUNCH_HOLE) returned EINVAL. "
1951 "Please report this bug to your file system "
1953 } else if (ret
!= -ENOTSUP
) {
1956 s
->has_discard
= false;
1961 #ifdef CONFIG_FALLOCATE
1962 /* Last resort: we are trying to extend the file with zeroed data. This
1963 * can be done via fallocate(fd, 0) */
1964 len
= raw_getlength(aiocb
->bs
);
1965 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1966 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1967 if (ret
== 0 || ret
!= -ENOTSUP
) {
1970 s
->has_fallocate
= false;
1977 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1979 RawPosixAIOData
*aiocb
= opaque
;
1980 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1982 /* First try to write zeros and unmap at the same time */
1984 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1985 int ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1986 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1997 /* If we couldn't manage to unmap while guaranteed that the area reads as
1998 * all-zero afterwards, just write zeroes without unmapping */
1999 return handle_aiocb_write_zeroes(aiocb
);
2002 #ifndef HAVE_COPY_FILE_RANGE
2004 /* Avoid conflict with system header file. */
2005 #define copy_file_range(a, b, c, d, e, f) local_copy_file_range(a, b, c, d, e, f)
2007 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
2008 off_t
*out_off
, size_t len
, unsigned int flags
)
2010 #ifdef __NR_copy_file_range
2011 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
2012 out_off
, len
, flags
);
2021 * parse_zone - Fill a zone descriptor
2023 #if defined(CONFIG_BLKZONED)
2024 static inline int parse_zone(struct BlockZoneDescriptor
*zone
,
2025 const struct blk_zone
*blkz
) {
2026 zone
->start
= blkz
->start
<< BDRV_SECTOR_BITS
;
2027 zone
->length
= blkz
->len
<< BDRV_SECTOR_BITS
;
2028 zone
->wp
= blkz
->wp
<< BDRV_SECTOR_BITS
;
2030 #ifdef HAVE_BLK_ZONE_REP_CAPACITY
2031 zone
->cap
= blkz
->capacity
<< BDRV_SECTOR_BITS
;
2033 zone
->cap
= blkz
->len
<< BDRV_SECTOR_BITS
;
2036 switch (blkz
->type
) {
2037 case BLK_ZONE_TYPE_SEQWRITE_REQ
:
2038 zone
->type
= BLK_ZT_SWR
;
2040 case BLK_ZONE_TYPE_SEQWRITE_PREF
:
2041 zone
->type
= BLK_ZT_SWP
;
2043 case BLK_ZONE_TYPE_CONVENTIONAL
:
2044 zone
->type
= BLK_ZT_CONV
;
2047 error_report("Unsupported zone type: 0x%x", blkz
->type
);
2051 switch (blkz
->cond
) {
2052 case BLK_ZONE_COND_NOT_WP
:
2053 zone
->state
= BLK_ZS_NOT_WP
;
2055 case BLK_ZONE_COND_EMPTY
:
2056 zone
->state
= BLK_ZS_EMPTY
;
2058 case BLK_ZONE_COND_IMP_OPEN
:
2059 zone
->state
= BLK_ZS_IOPEN
;
2061 case BLK_ZONE_COND_EXP_OPEN
:
2062 zone
->state
= BLK_ZS_EOPEN
;
2064 case BLK_ZONE_COND_CLOSED
:
2065 zone
->state
= BLK_ZS_CLOSED
;
2067 case BLK_ZONE_COND_READONLY
:
2068 zone
->state
= BLK_ZS_RDONLY
;
2070 case BLK_ZONE_COND_FULL
:
2071 zone
->state
= BLK_ZS_FULL
;
2073 case BLK_ZONE_COND_OFFLINE
:
2074 zone
->state
= BLK_ZS_OFFLINE
;
2077 error_report("Unsupported zone state: 0x%x", blkz
->cond
);
2084 #if defined(CONFIG_BLKZONED)
2085 static int handle_aiocb_zone_report(void *opaque
)
2087 RawPosixAIOData
*aiocb
= opaque
;
2088 int fd
= aiocb
->aio_fildes
;
2089 unsigned int *nr_zones
= aiocb
->zone_report
.nr_zones
;
2090 BlockZoneDescriptor
*zones
= aiocb
->zone_report
.zones
;
2091 /* zoned block devices use 512-byte sectors */
2092 uint64_t sector
= aiocb
->aio_offset
/ 512;
2094 struct blk_zone
*blkz
;
2098 unsigned int n
= 0, i
= 0;
2101 rep_size
= sizeof(struct blk_zone_report
) + nrz
* sizeof(struct blk_zone
);
2102 g_autofree
struct blk_zone_report
*rep
= NULL
;
2103 rep
= g_malloc(rep_size
);
2105 blkz
= (struct blk_zone
*)(rep
+ 1);
2107 memset(rep
, 0, rep_size
);
2108 rep
->sector
= sector
;
2109 rep
->nr_zones
= nrz
- n
;
2112 ret
= ioctl(fd
, BLKREPORTZONE
, rep
);
2113 } while (ret
!= 0 && errno
== EINTR
);
2115 error_report("%d: ioctl BLKREPORTZONE at %" PRId64
" failed %d",
2120 if (!rep
->nr_zones
) {
2124 for (i
= 0; i
< rep
->nr_zones
; i
++, n
++) {
2125 ret
= parse_zone(&zones
[n
], &blkz
[i
]);
2130 /* The next report should start after the last zone reported */
2131 sector
= blkz
[i
].start
+ blkz
[i
].len
;
2140 #if defined(CONFIG_BLKZONED)
2141 static int handle_aiocb_zone_mgmt(void *opaque
)
2143 RawPosixAIOData
*aiocb
= opaque
;
2144 int fd
= aiocb
->aio_fildes
;
2145 uint64_t sector
= aiocb
->aio_offset
/ 512;
2146 int64_t nr_sectors
= aiocb
->aio_nbytes
/ 512;
2147 struct blk_zone_range range
;
2150 /* Execute the operation */
2151 range
.sector
= sector
;
2152 range
.nr_sectors
= nr_sectors
;
2154 ret
= ioctl(fd
, aiocb
->zone_mgmt
.op
, &range
);
2155 } while (ret
!= 0 && errno
== EINTR
);
2157 return ret
< 0 ? -errno
: ret
;
2161 static int handle_aiocb_copy_range(void *opaque
)
2163 RawPosixAIOData
*aiocb
= opaque
;
2164 uint64_t bytes
= aiocb
->aio_nbytes
;
2165 off_t in_off
= aiocb
->aio_offset
;
2166 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
2169 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
2170 aiocb
->copy_range
.aio_fd2
, &out_off
,
2172 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
2173 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
2176 /* No progress (e.g. when beyond EOF), let the caller fall back to
2195 static int handle_aiocb_discard(void *opaque
)
2197 RawPosixAIOData
*aiocb
= opaque
;
2199 BDRVRawState
*s
= aiocb
->bs
->opaque
;
2201 if (!s
->has_discard
) {
2205 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
2208 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
2209 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
2212 } while (errno
== EINTR
);
2214 ret
= translate_err(-errno
);
2217 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
2218 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
2219 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
2220 ret
= translate_err(ret
);
2221 #elif defined(__APPLE__) && (__MACH__)
2222 fpunchhole_t fpunchhole
;
2223 fpunchhole
.fp_flags
= 0;
2224 fpunchhole
.reserved
= 0;
2225 fpunchhole
.fp_offset
= aiocb
->aio_offset
;
2226 fpunchhole
.fp_length
= aiocb
->aio_nbytes
;
2227 if (fcntl(s
->fd
, F_PUNCHHOLE
, &fpunchhole
) == -1) {
2228 ret
= errno
== ENODEV
? -ENOTSUP
: -errno
;
2235 if (ret
== -ENOTSUP
) {
2236 s
->has_discard
= false;
2242 * Help alignment probing by allocating the first block.
2244 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
2245 * reading succeeds regardless of request length. In this case we fallback to
2246 * safe alignment which is not optimal. Allocating the first block avoids this
2249 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
2250 * request alignment, so we use safe values.
2252 * Returns: 0 on success, -errno on failure. Since this is an optimization,
2253 * caller may ignore failures.
2255 static int allocate_first_block(int fd
, size_t max_size
)
2257 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
2260 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size());
2265 buf
= qemu_memalign(max_align
, write_size
);
2266 memset(buf
, 0, write_size
);
2268 n
= RETRY_ON_EINTR(pwrite(fd
, buf
, write_size
, 0));
2270 ret
= (n
== -1) ? -errno
: 0;
2276 static int handle_aiocb_truncate(void *opaque
)
2278 RawPosixAIOData
*aiocb
= opaque
;
2280 int64_t current_length
= 0;
2283 int fd
= aiocb
->aio_fildes
;
2284 int64_t offset
= aiocb
->aio_offset
;
2285 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
2286 Error
**errp
= aiocb
->truncate
.errp
;
2288 if (fstat(fd
, &st
) < 0) {
2290 error_setg_errno(errp
, -result
, "Could not stat file");
2294 current_length
= st
.st_size
;
2295 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
2296 error_setg(errp
, "Cannot use preallocation for shrinking files");
2301 #ifdef CONFIG_POSIX_FALLOCATE
2302 case PREALLOC_MODE_FALLOC
:
2304 * Truncating before posix_fallocate() makes it about twice slower on
2305 * file systems that do not support fallocate(), trying to check if a
2306 * block is allocated before allocating it, so don't do that here.
2308 if (offset
!= current_length
) {
2309 result
= -posix_fallocate(fd
, current_length
,
2310 offset
- current_length
);
2312 /* posix_fallocate() doesn't set errno. */
2313 error_setg_errno(errp
, -result
,
2314 "Could not preallocate new data");
2315 } else if (current_length
== 0) {
2317 * posix_fallocate() uses fallocate() if the filesystem
2318 * supports it, or fallback to manually writing zeroes. If
2319 * fallocate() was used, unaligned reads from the fallocated
2320 * area in raw_probe_alignment() will succeed, hence we need to
2321 * allocate the first block.
2323 * Optimize future alignment probing; ignore failures.
2325 allocate_first_block(fd
, offset
);
2332 case PREALLOC_MODE_FULL
:
2334 int64_t num
= 0, left
= offset
- current_length
;
2338 * Knowing the final size from the beginning could allow the file
2339 * system driver to do less allocations and possibly avoid
2340 * fragmentation of the file.
2342 if (ftruncate(fd
, offset
) != 0) {
2344 error_setg_errno(errp
, -result
, "Could not resize file");
2348 buf
= g_malloc0(65536);
2350 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
2351 if (seek_result
< 0) {
2353 error_setg_errno(errp
, -result
,
2354 "Failed to seek to the old end of file");
2359 num
= MIN(left
, 65536);
2360 result
= write(fd
, buf
, num
);
2362 if (errno
== EINTR
) {
2366 error_setg_errno(errp
, -result
,
2367 "Could not write zeros for preallocation");
2376 error_setg_errno(errp
, -result
,
2377 "Could not flush file to disk");
2383 case PREALLOC_MODE_OFF
:
2384 if (ftruncate(fd
, offset
) != 0) {
2386 error_setg_errno(errp
, -result
, "Could not resize file");
2387 } else if (current_length
== 0 && offset
> current_length
) {
2388 /* Optimize future alignment probing; ignore failures. */
2389 allocate_first_block(fd
, offset
);
2394 error_setg(errp
, "Unsupported preallocation mode: %s",
2395 PreallocMode_str(prealloc
));
2401 if (ftruncate(fd
, current_length
) < 0) {
2402 error_report("Failed to restore old file length: %s",
2411 static int coroutine_fn
raw_thread_pool_submit(ThreadPoolFunc func
, void *arg
)
2413 return thread_pool_submit_co(func
, arg
);
2417 * Check if all memory in this vector is sector aligned.
2419 static bool bdrv_qiov_is_aligned(BlockDriverState
*bs
, QEMUIOVector
*qiov
)
2422 size_t alignment
= bdrv_min_mem_align(bs
);
2423 size_t len
= bs
->bl
.request_alignment
;
2426 for (i
= 0; i
< qiov
->niov
; i
++) {
2427 if ((uintptr_t) qiov
->iov
[i
].iov_base
% alignment
) {
2430 if (qiov
->iov
[i
].iov_len
% len
) {
2438 #ifdef CONFIG_LINUX_IO_URING
2439 static inline bool raw_check_linux_io_uring(BDRVRawState
*s
)
2441 Error
*local_err
= NULL
;
2444 if (!s
->use_linux_io_uring
) {
2448 ctx
= qemu_get_current_aio_context();
2449 if (unlikely(!aio_setup_linux_io_uring(ctx
, &local_err
))) {
2450 error_reportf_err(local_err
, "Unable to use linux io_uring, "
2451 "falling back to thread pool: ");
2452 s
->use_linux_io_uring
= false;
2459 #ifdef CONFIG_LINUX_AIO
2460 static inline bool raw_check_linux_aio(BDRVRawState
*s
)
2462 Error
*local_err
= NULL
;
2465 if (!s
->use_linux_aio
) {
2469 ctx
= qemu_get_current_aio_context();
2470 if (unlikely(!aio_setup_linux_aio(ctx
, &local_err
))) {
2471 error_reportf_err(local_err
, "Unable to use Linux AIO, "
2472 "falling back to thread pool: ");
2473 s
->use_linux_aio
= false;
2480 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, int64_t *offset_ptr
,
2481 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
2483 BDRVRawState
*s
= bs
->opaque
;
2484 RawPosixAIOData acb
;
2486 uint64_t offset
= *offset_ptr
;
2488 if (fd_open(bs
) < 0)
2490 #if defined(CONFIG_BLKZONED)
2491 if ((type
& (QEMU_AIO_WRITE
| QEMU_AIO_ZONE_APPEND
)) &&
2492 bs
->bl
.zoned
!= BLK_Z_NONE
) {
2493 qemu_co_mutex_lock(&bs
->wps
->colock
);
2494 if (type
& QEMU_AIO_ZONE_APPEND
) {
2495 int index
= offset
/ bs
->bl
.zone_size
;
2496 offset
= bs
->wps
->wp
[index
];
2502 * When using O_DIRECT, the request must be aligned to be able to use
2503 * either libaio or io_uring interface. If not fail back to regular thread
2504 * pool read/write code which emulates this for us if we
2505 * set QEMU_AIO_MISALIGNED.
2507 if (s
->needs_alignment
&& !bdrv_qiov_is_aligned(bs
, qiov
)) {
2508 type
|= QEMU_AIO_MISALIGNED
;
2509 #ifdef CONFIG_LINUX_IO_URING
2510 } else if (raw_check_linux_io_uring(s
)) {
2511 assert(qiov
->size
== bytes
);
2512 ret
= luring_co_submit(bs
, s
->fd
, offset
, qiov
, type
);
2515 #ifdef CONFIG_LINUX_AIO
2516 } else if (raw_check_linux_aio(s
)) {
2517 assert(qiov
->size
== bytes
);
2518 ret
= laio_co_submit(s
->fd
, offset
, qiov
, type
,
2524 acb
= (RawPosixAIOData
) {
2526 .aio_fildes
= s
->fd
,
2528 .aio_offset
= offset
,
2529 .aio_nbytes
= bytes
,
2536 assert(qiov
->size
== bytes
);
2537 ret
= raw_thread_pool_submit(handle_aiocb_rw
, &acb
);
2538 goto out
; /* Avoid the compiler err of unused label */
2541 #if defined(CONFIG_BLKZONED)
2542 if ((type
& (QEMU_AIO_WRITE
| QEMU_AIO_ZONE_APPEND
)) &&
2543 bs
->bl
.zoned
!= BLK_Z_NONE
) {
2544 BlockZoneWps
*wps
= bs
->wps
;
2546 uint64_t *wp
= &wps
->wp
[offset
/ bs
->bl
.zone_size
];
2547 if (!BDRV_ZT_IS_CONV(*wp
)) {
2548 if (type
& QEMU_AIO_ZONE_APPEND
) {
2550 trace_zbd_zone_append_complete(bs
, *offset_ptr
2551 >> BDRV_SECTOR_BITS
);
2553 /* Advance the wp if needed */
2554 if (offset
+ bytes
> *wp
) {
2555 *wp
= offset
+ bytes
;
2560 * write and append write are not allowed to cross zone boundaries
2562 update_zones_wp(bs
, s
->fd
, offset
, 1);
2565 qemu_co_mutex_unlock(&wps
->colock
);
2571 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, int64_t offset
,
2572 int64_t bytes
, QEMUIOVector
*qiov
,
2573 BdrvRequestFlags flags
)
2575 return raw_co_prw(bs
, &offset
, bytes
, qiov
, QEMU_AIO_READ
);
2578 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, int64_t offset
,
2579 int64_t bytes
, QEMUIOVector
*qiov
,
2580 BdrvRequestFlags flags
)
2582 return raw_co_prw(bs
, &offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
2585 static int coroutine_fn
raw_co_flush_to_disk(BlockDriverState
*bs
)
2587 BDRVRawState
*s
= bs
->opaque
;
2588 RawPosixAIOData acb
;
2596 acb
= (RawPosixAIOData
) {
2598 .aio_fildes
= s
->fd
,
2599 .aio_type
= QEMU_AIO_FLUSH
,
2602 #ifdef CONFIG_LINUX_IO_URING
2603 if (raw_check_linux_io_uring(s
)) {
2604 return luring_co_submit(bs
, s
->fd
, 0, NULL
, QEMU_AIO_FLUSH
);
2607 return raw_thread_pool_submit(handle_aiocb_flush
, &acb
);
2610 static void raw_close(BlockDriverState
*bs
)
2612 BDRVRawState
*s
= bs
->opaque
;
2615 #if defined(CONFIG_BLKZONED)
2624 * Truncates the given regular file @fd to @offset and, when growing, fills the
2625 * new space according to @prealloc.
2627 * Returns: 0 on success, -errno on failure.
2629 static int coroutine_fn
2630 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2631 PreallocMode prealloc
, Error
**errp
)
2633 RawPosixAIOData acb
;
2635 acb
= (RawPosixAIOData
) {
2638 .aio_type
= QEMU_AIO_TRUNCATE
,
2639 .aio_offset
= offset
,
2641 .prealloc
= prealloc
,
2646 return raw_thread_pool_submit(handle_aiocb_truncate
, &acb
);
2649 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2650 bool exact
, PreallocMode prealloc
,
2651 BdrvRequestFlags flags
, Error
**errp
)
2653 BDRVRawState
*s
= bs
->opaque
;
2657 if (fstat(s
->fd
, &st
)) {
2659 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2663 if (S_ISREG(st
.st_mode
)) {
2664 /* Always resizes to the exact @offset */
2665 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2668 if (prealloc
!= PREALLOC_MODE_OFF
) {
2669 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2670 "non-regular file", PreallocMode_str(prealloc
));
2674 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2675 int64_t cur_length
= raw_getlength(bs
);
2677 if (offset
!= cur_length
&& exact
) {
2678 error_setg(errp
, "Cannot resize device files");
2680 } else if (offset
> cur_length
) {
2681 error_setg(errp
, "Cannot grow device files");
2685 error_setg(errp
, "Resizing this file is not supported");
2693 static int64_t raw_getlength(BlockDriverState
*bs
)
2695 BDRVRawState
*s
= bs
->opaque
;
2701 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2702 struct disklabel dl
;
2704 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2706 return (uint64_t)dl
.d_secsize
*
2707 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2711 #elif defined(__NetBSD__)
2712 static int64_t raw_getlength(BlockDriverState
*bs
)
2714 BDRVRawState
*s
= bs
->opaque
;
2720 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2721 struct dkwedge_info dkw
;
2723 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2724 return dkw
.dkw_size
* 512;
2726 struct disklabel dl
;
2728 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2730 return (uint64_t)dl
.d_secsize
*
2731 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2736 #elif defined(__sun__)
2737 static int64_t raw_getlength(BlockDriverState
*bs
)
2739 BDRVRawState
*s
= bs
->opaque
;
2740 struct dk_minfo minfo
;
2750 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2752 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2754 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2758 * There are reports that lseek on some devices fails, but
2759 * irc discussion said that contingency on contingency was overkill.
2761 size
= lseek(s
->fd
, 0, SEEK_END
);
2767 #elif defined(CONFIG_BSD)
2768 static int64_t raw_getlength(BlockDriverState
*bs
)
2770 BDRVRawState
*s
= bs
->opaque
;
2774 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2783 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2786 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2788 #ifdef DIOCGMEDIASIZE
2789 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
)) {
2796 if (ioctl(fd
, DIOCGPART
, &pi
) == 0) {
2797 size
= pi
.media_size
;
2801 #if defined(DKIOCGETBLOCKCOUNT) && defined(DKIOCGETBLOCKSIZE)
2803 uint64_t sectors
= 0;
2804 uint32_t sector_size
= 0;
2806 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2807 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2808 size
= sectors
* sector_size
;
2813 size
= lseek(fd
, 0LL, SEEK_END
);
2818 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2821 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2822 if (size
== 2048LL * (unsigned)-1)
2824 /* XXX no disc? maybe we need to reopen... */
2825 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2832 size
= lseek(fd
, 0, SEEK_END
);
2840 static int64_t raw_getlength(BlockDriverState
*bs
)
2842 BDRVRawState
*s
= bs
->opaque
;
2851 size
= lseek(s
->fd
, 0, SEEK_END
);
2859 static int64_t coroutine_fn
raw_co_getlength(BlockDriverState
*bs
)
2861 return raw_getlength(bs
);
2864 static int64_t coroutine_fn
raw_co_get_allocated_file_size(BlockDriverState
*bs
)
2867 BDRVRawState
*s
= bs
->opaque
;
2869 if (fstat(s
->fd
, &st
) < 0) {
2872 return (int64_t)st
.st_blocks
* 512;
2875 static int coroutine_fn
2876 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2878 BlockdevCreateOptionsFile
*file_opts
;
2879 Error
*local_err
= NULL
;
2881 uint64_t perm
, shared
;
2884 /* Validate options and set default values */
2885 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2886 file_opts
= &options
->u
.file
;
2888 if (!file_opts
->has_nocow
) {
2889 file_opts
->nocow
= false;
2891 if (!file_opts
->has_preallocation
) {
2892 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2894 if (!file_opts
->has_extent_size_hint
) {
2895 file_opts
->extent_size_hint
= 1 * MiB
;
2897 if (file_opts
->extent_size_hint
> UINT32_MAX
) {
2899 error_setg(errp
, "Extent size hint is too large");
2904 fd
= qemu_create(file_opts
->filename
, O_RDWR
| O_BINARY
, 0644, errp
);
2910 /* Take permissions: We want to discard everything, so we need
2911 * BLK_PERM_WRITE; and truncation to the desired size requires
2913 * On the other hand, we cannot share the RESIZE permission
2914 * because we promise that after this function, the file has the
2915 * size given in the options. If someone else were to resize it
2916 * concurrently, we could not guarantee that.
2917 * Note that after this function, we can no longer guarantee that
2918 * the file is not touched by a third party, so it may be resized
2920 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2921 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2923 /* Step one: Take locks */
2924 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2929 /* Step two: Check that nobody else has taken conflicting locks */
2930 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2932 error_append_hint(errp
,
2933 "Is another process using the image [%s]?\n",
2934 file_opts
->filename
);
2938 /* Clear the file by truncating it to 0 */
2939 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2944 if (file_opts
->nocow
) {
2946 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2947 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2948 * will be ignored since any failure of this operation should not
2949 * block the left work.
2952 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2953 attr
|= FS_NOCOW_FL
;
2954 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2958 #ifdef FS_IOC_FSSETXATTR
2960 * Try to set the extent size hint. Failure is not fatal, and a warning is
2961 * only printed if the option was explicitly specified.
2964 struct fsxattr attr
;
2965 result
= ioctl(fd
, FS_IOC_FSGETXATTR
, &attr
);
2967 attr
.fsx_xflags
|= FS_XFLAG_EXTSIZE
;
2968 attr
.fsx_extsize
= file_opts
->extent_size_hint
;
2969 result
= ioctl(fd
, FS_IOC_FSSETXATTR
, &attr
);
2971 if (result
< 0 && file_opts
->has_extent_size_hint
&&
2972 file_opts
->extent_size_hint
)
2974 warn_report("Failed to set extent size hint: %s",
2980 /* Resize and potentially preallocate the file to the desired
2982 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2983 file_opts
->preallocation
, errp
);
2989 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2991 /* The above call should not fail, and if it does, that does
2992 * not mean the whole creation operation has failed. So
2993 * report it the user for their convenience, but do not report
2994 * it to the caller. */
2995 warn_report_err(local_err
);
2999 if (qemu_close(fd
) != 0 && result
== 0) {
3001 error_setg_errno(errp
, -result
, "Could not close the new file");
3007 static int coroutine_fn GRAPH_RDLOCK
3008 raw_co_create_opts(BlockDriver
*drv
, const char *filename
,
3009 QemuOpts
*opts
, Error
**errp
)
3011 BlockdevCreateOptions options
;
3012 int64_t total_size
= 0;
3013 int64_t extent_size_hint
= 0;
3014 bool has_extent_size_hint
= false;
3016 PreallocMode prealloc
;
3018 Error
*local_err
= NULL
;
3020 /* Skip file: protocol prefix */
3021 strstart(filename
, "file:", &filename
);
3023 /* Read out options */
3024 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3026 if (qemu_opt_get(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
)) {
3027 has_extent_size_hint
= true;
3029 qemu_opt_get_size_del(opts
, BLOCK_OPT_EXTENT_SIZE_HINT
, -1);
3031 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
3032 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
3033 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
3034 PREALLOC_MODE_OFF
, &local_err
);
3037 error_propagate(errp
, local_err
);
3041 options
= (BlockdevCreateOptions
) {
3042 .driver
= BLOCKDEV_DRIVER_FILE
,
3044 .filename
= (char *) filename
,
3046 .has_preallocation
= true,
3047 .preallocation
= prealloc
,
3050 .has_extent_size_hint
= has_extent_size_hint
,
3051 .extent_size_hint
= extent_size_hint
,
3054 return raw_co_create(&options
, errp
);
3057 static int coroutine_fn
raw_co_delete_file(BlockDriverState
*bs
,
3063 if (!(stat(bs
->filename
, &st
) == 0) || !S_ISREG(st
.st_mode
)) {
3064 error_setg_errno(errp
, ENOENT
, "%s is not a regular file",
3069 ret
= unlink(bs
->filename
);
3072 error_setg_errno(errp
, -ret
, "Error when deleting file %s",
3080 * Find allocation range in @bs around offset @start.
3081 * May change underlying file descriptor's file offset.
3082 * If @start is not in a hole, store @start in @data, and the
3083 * beginning of the next hole in @hole, and return 0.
3084 * If @start is in a non-trailing hole, store @start in @hole and the
3085 * beginning of the next non-hole in @data, and return 0.
3086 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
3087 * If we can't find out, return a negative errno other than -ENXIO.
3089 static int find_allocation(BlockDriverState
*bs
, off_t start
,
3090 off_t
*data
, off_t
*hole
)
3092 #if defined SEEK_HOLE && defined SEEK_DATA
3093 BDRVRawState
*s
= bs
->opaque
;
3098 * D1. offs == start: start is in data
3099 * D2. offs > start: start is in a hole, next data at offs
3100 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
3101 * or start is beyond EOF
3102 * If the latter happens, the file has been truncated behind
3103 * our back since we opened it. All bets are off then.
3104 * Treating like a trailing hole is simplest.
3105 * D4. offs < 0, errno != ENXIO: we learned nothing
3107 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
3109 return -errno
; /* D3 or D4 */
3113 /* This is not a valid return by lseek(). We are safe to just return
3114 * -EIO in this case, and we'll treat it like D4. */
3119 /* D2: in hole, next data at offs */
3125 /* D1: in data, end not yet known */
3129 * H1. offs == start: start is in a hole
3130 * If this happens here, a hole has been dug behind our back
3131 * since the previous lseek().
3132 * H2. offs > start: either start is in data, next hole at offs,
3133 * or start is in trailing hole, EOF at offs
3134 * Linux treats trailing holes like any other hole: offs ==
3135 * start. Solaris seeks to EOF instead: offs > start (blech).
3136 * If that happens here, a hole has been dug behind our back
3137 * since the previous lseek().
3138 * H3. offs < 0, errno = ENXIO: start is beyond EOF
3139 * If this happens, the file has been truncated behind our
3140 * back since we opened it. Treat it like a trailing hole.
3141 * H4. offs < 0, errno != ENXIO: we learned nothing
3142 * Pretend we know nothing at all, i.e. "forget" about D1.
3144 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
3146 return -errno
; /* D1 and (H3 or H4) */
3150 /* This is not a valid return by lseek(). We are safe to just return
3151 * -EIO in this case, and we'll treat it like H4. */
3157 * D1 and H2: either in data, next hole at offs, or it was in
3158 * data but is now in a trailing hole. In the latter case,
3159 * all bets are off. Treating it as if it there was data all
3160 * the way to EOF is safe, so simply do that.
3175 * Returns the allocation status of the specified offset.
3177 * The block layer guarantees 'offset' and 'bytes' are within bounds.
3179 * 'pnum' is set to the number of bytes (including and immediately following
3180 * the specified offset) that are known to be in the same
3181 * allocated/unallocated state.
3183 * 'bytes' is a soft cap for 'pnum'. If the information is free, 'pnum' may
3186 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
3189 int64_t bytes
, int64_t *pnum
,
3191 BlockDriverState
**file
)
3193 off_t data
= 0, hole
= 0;
3196 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
3207 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
3210 ret
= find_allocation(bs
, offset
, &data
, &hole
);
3211 if (ret
== -ENXIO
) {
3214 ret
= BDRV_BLOCK_ZERO
;
3215 } else if (ret
< 0) {
3216 /* No info available, so pretend there are no holes */
3218 ret
= BDRV_BLOCK_DATA
;
3219 } else if (data
== offset
) {
3220 /* On a data extent, compute bytes to the end of the extent,
3221 * possibly including a partial sector at EOF. */
3222 *pnum
= hole
- offset
;
3225 * We are not allowed to return partial sectors, though, so
3226 * round up if necessary.
3228 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
3229 int64_t file_length
= raw_getlength(bs
);
3230 if (file_length
> 0) {
3231 /* Ignore errors, this is just a safeguard */
3232 assert(hole
== file_length
);
3234 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
3237 ret
= BDRV_BLOCK_DATA
;
3239 /* On a hole, compute bytes to the beginning of the next extent. */
3240 assert(hole
== offset
);
3241 *pnum
= data
- offset
;
3242 ret
= BDRV_BLOCK_ZERO
;
3246 return ret
| BDRV_BLOCK_OFFSET_VALID
;
3249 #if defined(__linux__)
3250 /* Verify that the file is not in the page cache */
3251 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
3253 const size_t window_size
= 128 * 1024 * 1024;
3254 BDRVRawState
*s
= bs
->opaque
;
3255 void *window
= NULL
;
3262 /* mincore(2) page status information requires 1 byte per page */
3263 page_size
= sysconf(_SC_PAGESIZE
);
3264 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
3266 end
= raw_getlength(bs
);
3268 for (offset
= 0; offset
< end
; offset
+= window_size
) {
3275 /* Unmap previous window if size has changed */
3276 new_length
= MIN(end
- offset
, window_size
);
3277 if (new_length
!= length
) {
3278 munmap(window
, length
);
3283 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
3285 if (new_window
== MAP_FAILED
) {
3286 error_setg_errno(errp
, errno
, "mmap failed");
3290 window
= new_window
;
3291 length
= new_length
;
3293 ret
= mincore(window
, length
, vec
);
3295 error_setg_errno(errp
, errno
, "mincore failed");
3299 vec_end
= DIV_ROUND_UP(length
, page_size
);
3300 for (i
= 0; i
< vec_end
; i
++) {
3306 error_setg(errp
, "page cache still in use!");
3312 munmap(window
, length
);
3317 #endif /* __linux__ */
3319 static void coroutine_fn GRAPH_RDLOCK
3320 raw_co_invalidate_cache(BlockDriverState
*bs
, Error
**errp
)
3322 BDRVRawState
*s
= bs
->opaque
;
3327 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
3331 if (!s
->drop_cache
) {
3335 if (s
->open_flags
& O_DIRECT
) {
3336 return; /* No host kernel page cache */
3339 #if defined(__linux__)
3340 /* This sets the scene for the next syscall... */
3341 ret
= bdrv_co_flush(bs
);
3343 error_setg_errno(errp
, -ret
, "flush failed");
3347 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
3348 * process. These limitations are okay because we just fsynced the file,
3349 * we don't use mmap, and the file should not be in use by other processes.
3351 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
3352 if (ret
!= 0) { /* the return value is a positive errno */
3353 error_setg_errno(errp
, ret
, "fadvise failed");
3357 if (s
->check_cache_dropped
) {
3358 check_cache_dropped(bs
, errp
);
3360 #else /* __linux__ */
3361 /* Do nothing. Live migration to a remote host with cache.direct=off is
3362 * unsupported on other host operating systems. Cache consistency issues
3363 * may occur but no error is reported here, partly because that's the
3364 * historical behavior and partly because it's hard to differentiate valid
3365 * configurations that should not cause errors.
3367 #endif /* !__linux__ */
3370 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
3373 s
->stats
.discard_nb_failed
++;
3375 s
->stats
.discard_nb_ok
++;
3376 s
->stats
.discard_bytes_ok
+= nbytes
;
3381 * zone report - Get a zone block device's information in the form
3382 * of an array of zone descriptors.
3383 * zones is an array of zone descriptors to hold zone information on reply;
3384 * offset can be any byte within the entire size of the device;
3385 * nr_zones is the maximum number of sectors the command should operate on.
3387 #if defined(CONFIG_BLKZONED)
3388 static int coroutine_fn
raw_co_zone_report(BlockDriverState
*bs
, int64_t offset
,
3389 unsigned int *nr_zones
,
3390 BlockZoneDescriptor
*zones
) {
3391 BDRVRawState
*s
= bs
->opaque
;
3392 RawPosixAIOData acb
= (RawPosixAIOData
) {
3394 .aio_fildes
= s
->fd
,
3395 .aio_type
= QEMU_AIO_ZONE_REPORT
,
3396 .aio_offset
= offset
,
3398 .nr_zones
= nr_zones
,
3403 trace_zbd_zone_report(bs
, *nr_zones
, offset
>> BDRV_SECTOR_BITS
);
3404 return raw_thread_pool_submit(handle_aiocb_zone_report
, &acb
);
3409 * zone management operations - Execute an operation on a zone
3411 #if defined(CONFIG_BLKZONED)
3412 static int coroutine_fn
raw_co_zone_mgmt(BlockDriverState
*bs
, BlockZoneOp op
,
3413 int64_t offset
, int64_t len
) {
3414 BDRVRawState
*s
= bs
->opaque
;
3415 RawPosixAIOData acb
;
3416 int64_t zone_size
, zone_size_mask
;
3417 const char *op_name
;
3420 BlockZoneWps
*wps
= bs
->wps
;
3421 int64_t capacity
= bs
->total_sectors
<< BDRV_SECTOR_BITS
;
3423 zone_size
= bs
->bl
.zone_size
;
3424 zone_size_mask
= zone_size
- 1;
3425 if (offset
& zone_size_mask
) {
3426 error_report("sector offset %" PRId64
" is not aligned to zone size "
3427 "%" PRId64
"", offset
/ 512, zone_size
/ 512);
3431 if (((offset
+ len
) < capacity
&& len
& zone_size_mask
) ||
3432 offset
+ len
> capacity
) {
3433 error_report("number of sectors %" PRId64
" is not aligned to zone size"
3434 " %" PRId64
"", len
/ 512, zone_size
/ 512);
3438 uint32_t i
= offset
/ bs
->bl
.zone_size
;
3439 uint32_t nrz
= len
/ bs
->bl
.zone_size
;
3440 uint64_t *wp
= &wps
->wp
[i
];
3441 if (BDRV_ZT_IS_CONV(*wp
) && len
!= capacity
) {
3442 error_report("zone mgmt operations are not allowed for conventional zones");
3448 op_name
= "BLKOPENZONE";
3452 op_name
= "BLKCLOSEZONE";
3456 op_name
= "BLKFINISHZONE";
3460 op_name
= "BLKRESETZONE";
3464 error_report("Unsupported zone op: 0x%x", op
);
3468 acb
= (RawPosixAIOData
) {
3470 .aio_fildes
= s
->fd
,
3471 .aio_type
= QEMU_AIO_ZONE_MGMT
,
3472 .aio_offset
= offset
,
3479 trace_zbd_zone_mgmt(bs
, op_name
, offset
>> BDRV_SECTOR_BITS
,
3480 len
>> BDRV_SECTOR_BITS
);
3481 ret
= raw_thread_pool_submit(handle_aiocb_zone_mgmt
, &acb
);
3483 update_zones_wp(bs
, s
->fd
, offset
, nrz
);
3484 error_report("ioctl %s failed %d", op_name
, ret
);
3488 if (zo
== BLKRESETZONE
&& len
== capacity
) {
3489 ret
= get_zones_wp(bs
, s
->fd
, 0, bs
->bl
.nr_zones
, 1);
3491 error_report("reporting single wp failed");
3494 } else if (zo
== BLKRESETZONE
) {
3495 for (unsigned int j
= 0; j
< nrz
; ++j
) {
3496 wp
[j
] = offset
+ j
* zone_size
;
3498 } else if (zo
== BLKFINISHZONE
) {
3499 for (unsigned int j
= 0; j
< nrz
; ++j
) {
3500 /* The zoned device allows the last zone smaller that the
3502 wp
[j
] = MIN(offset
+ (j
+ 1) * zone_size
, offset
+ len
);
3510 #if defined(CONFIG_BLKZONED)
3511 static int coroutine_fn
raw_co_zone_append(BlockDriverState
*bs
,
3514 BdrvRequestFlags flags
) {
3516 int64_t zone_size_mask
= bs
->bl
.zone_size
- 1;
3517 int64_t iov_len
= 0;
3520 if (*offset
& zone_size_mask
) {
3521 error_report("sector offset %" PRId64
" is not aligned to zone size "
3522 "%" PRId32
"", *offset
/ 512, bs
->bl
.zone_size
/ 512);
3526 int64_t wg
= bs
->bl
.write_granularity
;
3527 int64_t wg_mask
= wg
- 1;
3528 for (int i
= 0; i
< qiov
->niov
; i
++) {
3529 iov_len
= qiov
->iov
[i
].iov_len
;
3530 if (iov_len
& wg_mask
) {
3531 error_report("len of IOVector[%d] %" PRId64
" is not aligned to "
3532 "block size %" PRId64
"", i
, iov_len
, wg
);
3538 trace_zbd_zone_append(bs
, *offset
>> BDRV_SECTOR_BITS
);
3539 return raw_co_prw(bs
, offset
, len
, qiov
, QEMU_AIO_ZONE_APPEND
);
3543 static coroutine_fn
int
3544 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
3547 BDRVRawState
*s
= bs
->opaque
;
3548 RawPosixAIOData acb
;
3551 acb
= (RawPosixAIOData
) {
3553 .aio_fildes
= s
->fd
,
3554 .aio_type
= QEMU_AIO_DISCARD
,
3555 .aio_offset
= offset
,
3556 .aio_nbytes
= bytes
,
3560 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
3563 ret
= raw_thread_pool_submit(handle_aiocb_discard
, &acb
);
3564 raw_account_discard(s
, bytes
, ret
);
3568 static coroutine_fn
int
3569 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
3571 return raw_do_pdiscard(bs
, offset
, bytes
, false);
3574 static int coroutine_fn
3575 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
,
3576 BdrvRequestFlags flags
, bool blkdev
)
3578 BDRVRawState
*s
= bs
->opaque
;
3579 RawPosixAIOData acb
;
3580 ThreadPoolFunc
*handler
;
3582 #ifdef CONFIG_FALLOCATE
3583 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
3584 BdrvTrackedRequest
*req
;
3587 * This is a workaround for a bug in the Linux XFS driver,
3588 * where writes submitted through the AIO interface will be
3589 * discarded if they happen beyond a concurrently running
3590 * fallocate() that increases the file length (i.e., both the
3591 * write and the fallocate() happen beyond the EOF).
3593 * To work around it, we extend the tracked request for this
3594 * zero write until INT64_MAX (effectively infinity), and mark
3595 * it as serializing.
3597 * We have to enable this workaround for all filesystems and
3598 * AIO modes (not just XFS with aio=native), because for
3599 * remote filesystems we do not know the host configuration.
3602 req
= bdrv_co_get_self_request(bs
);
3604 assert(req
->type
== BDRV_TRACKED_WRITE
);
3605 assert(req
->offset
<= offset
);
3606 assert(req
->offset
+ req
->bytes
>= offset
+ bytes
);
3608 req
->bytes
= BDRV_MAX_LENGTH
- req
->offset
;
3610 bdrv_check_request(req
->offset
, req
->bytes
, &error_abort
);
3612 bdrv_make_request_serialising(req
, bs
->bl
.request_alignment
);
3616 acb
= (RawPosixAIOData
) {
3618 .aio_fildes
= s
->fd
,
3619 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
3620 .aio_offset
= offset
,
3621 .aio_nbytes
= bytes
,
3625 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
3627 if (flags
& BDRV_REQ_NO_FALLBACK
) {
3628 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
3631 if (flags
& BDRV_REQ_MAY_UNMAP
) {
3632 acb
.aio_type
|= QEMU_AIO_DISCARD
;
3633 handler
= handle_aiocb_write_zeroes_unmap
;
3635 handler
= handle_aiocb_write_zeroes
;
3638 return raw_thread_pool_submit(handler
, &acb
);
3641 static int coroutine_fn
raw_co_pwrite_zeroes(
3642 BlockDriverState
*bs
, int64_t offset
,
3643 int64_t bytes
, BdrvRequestFlags flags
)
3645 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
3648 static int coroutine_fn
3649 raw_co_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
3654 static ImageInfoSpecific
*raw_get_specific_info(BlockDriverState
*bs
,
3657 ImageInfoSpecificFile
*file_info
= g_new0(ImageInfoSpecificFile
, 1);
3658 ImageInfoSpecific
*spec_info
= g_new(ImageInfoSpecific
, 1);
3660 *spec_info
= (ImageInfoSpecific
){
3661 .type
= IMAGE_INFO_SPECIFIC_KIND_FILE
,
3662 .u
.file
.data
= file_info
,
3665 #ifdef FS_IOC_FSGETXATTR
3667 BDRVRawState
*s
= bs
->opaque
;
3668 struct fsxattr attr
;
3671 ret
= ioctl(s
->fd
, FS_IOC_FSGETXATTR
, &attr
);
3672 if (!ret
&& attr
.fsx_extsize
!= 0) {
3673 file_info
->has_extent_size_hint
= true;
3674 file_info
->extent_size_hint
= attr
.fsx_extsize
;
3682 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
3684 BDRVRawState
*s
= bs
->opaque
;
3685 return (BlockStatsSpecificFile
) {
3686 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
3687 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
3688 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
3692 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
3694 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3696 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
3697 stats
->u
.file
= get_blockstats_specific_file(bs
);
3702 #if defined(HAVE_HOST_BLOCK_DEVICE)
3703 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
3705 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
3707 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
3708 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
3712 #endif /* HAVE_HOST_BLOCK_DEVICE */
3714 static QemuOptsList raw_create_opts
= {
3715 .name
= "raw-create-opts",
3716 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
3719 .name
= BLOCK_OPT_SIZE
,
3720 .type
= QEMU_OPT_SIZE
,
3721 .help
= "Virtual disk size"
3724 .name
= BLOCK_OPT_NOCOW
,
3725 .type
= QEMU_OPT_BOOL
,
3726 .help
= "Turn off copy-on-write (valid only on btrfs)"
3729 .name
= BLOCK_OPT_PREALLOC
,
3730 .type
= QEMU_OPT_STRING
,
3731 .help
= "Preallocation mode (allowed values: off"
3732 #ifdef CONFIG_POSIX_FALLOCATE
3738 .name
= BLOCK_OPT_EXTENT_SIZE_HINT
,
3739 .type
= QEMU_OPT_SIZE
,
3740 .help
= "Extent size hint for the image file, 0 to disable"
3742 { /* end of list */ }
3746 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
3749 BDRVRawState
*s
= bs
->opaque
;
3750 int input_flags
= s
->reopen_state
? s
->reopen_state
->flags
: bs
->open_flags
;
3754 /* We may need a new fd if auto-read-only switches the mode */
3755 ret
= raw_reconfigure_getfd(bs
, input_flags
, &open_flags
, perm
,
3759 } else if (ret
!= s
->fd
) {
3760 Error
*local_err
= NULL
;
3763 * Fail already check_perm() if we can't get a working O_DIRECT
3764 * alignment with the new fd.
3766 raw_probe_alignment(bs
, ret
, &local_err
);
3768 error_propagate(errp
, local_err
);
3772 s
->perm_change_fd
= ret
;
3773 s
->perm_change_flags
= open_flags
;
3776 /* Prepare permissions on old fd to avoid conflicts between old and new,
3777 * but keep everything locked that new will need. */
3778 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
3783 /* Copy locks to the new fd */
3784 if (s
->perm_change_fd
&& s
->use_lock
) {
3785 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
3788 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3795 if (s
->perm_change_fd
) {
3796 qemu_close(s
->perm_change_fd
);
3798 s
->perm_change_fd
= 0;
3802 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
3804 BDRVRawState
*s
= bs
->opaque
;
3806 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3807 * called after .bdrv_reopen_commit) */
3808 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
3810 s
->fd
= s
->perm_change_fd
;
3811 s
->open_flags
= s
->perm_change_flags
;
3813 s
->perm_change_fd
= 0;
3815 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
3817 s
->shared_perm
= shared
;
3820 static void raw_abort_perm_update(BlockDriverState
*bs
)
3822 BDRVRawState
*s
= bs
->opaque
;
3824 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3825 * the file descriptor. */
3826 if (s
->perm_change_fd
) {
3827 qemu_close(s
->perm_change_fd
);
3829 s
->perm_change_fd
= 0;
3831 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3834 static int coroutine_fn GRAPH_RDLOCK
raw_co_copy_range_from(
3835 BlockDriverState
*bs
, BdrvChild
*src
, int64_t src_offset
,
3836 BdrvChild
*dst
, int64_t dst_offset
, int64_t bytes
,
3837 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
3839 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3840 read_flags
, write_flags
);
3843 static int coroutine_fn GRAPH_RDLOCK
3844 raw_co_copy_range_to(BlockDriverState
*bs
,
3845 BdrvChild
*src
, int64_t src_offset
,
3846 BdrvChild
*dst
, int64_t dst_offset
,
3847 int64_t bytes
, BdrvRequestFlags read_flags
,
3848 BdrvRequestFlags write_flags
)
3850 RawPosixAIOData acb
;
3851 BDRVRawState
*s
= bs
->opaque
;
3852 BDRVRawState
*src_s
;
3854 assert(dst
->bs
== bs
);
3855 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
3859 src_s
= src
->bs
->opaque
;
3860 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
3864 acb
= (RawPosixAIOData
) {
3866 .aio_type
= QEMU_AIO_COPY_RANGE
,
3867 .aio_fildes
= src_s
->fd
,
3868 .aio_offset
= src_offset
,
3869 .aio_nbytes
= bytes
,
3872 .aio_offset2
= dst_offset
,
3876 return raw_thread_pool_submit(handle_aiocb_copy_range
, &acb
);
3879 BlockDriver bdrv_file
= {
3880 .format_name
= "file",
3881 .protocol_name
= "file",
3882 .instance_size
= sizeof(BDRVRawState
),
3883 .bdrv_needs_filename
= true,
3884 .bdrv_probe
= NULL
, /* no probe for protocols */
3885 .bdrv_parse_filename
= raw_parse_filename
,
3886 .bdrv_file_open
= raw_open
,
3887 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3888 .bdrv_reopen_commit
= raw_reopen_commit
,
3889 .bdrv_reopen_abort
= raw_reopen_abort
,
3890 .bdrv_close
= raw_close
,
3891 .bdrv_co_create
= raw_co_create
,
3892 .bdrv_co_create_opts
= raw_co_create_opts
,
3893 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3894 .bdrv_co_block_status
= raw_co_block_status
,
3895 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3896 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
3897 .bdrv_co_delete_file
= raw_co_delete_file
,
3899 .bdrv_co_preadv
= raw_co_preadv
,
3900 .bdrv_co_pwritev
= raw_co_pwritev
,
3901 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3902 .bdrv_co_pdiscard
= raw_co_pdiscard
,
3903 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3904 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3905 .bdrv_refresh_limits
= raw_refresh_limits
,
3907 .bdrv_co_truncate
= raw_co_truncate
,
3908 .bdrv_co_getlength
= raw_co_getlength
,
3909 .bdrv_co_get_info
= raw_co_get_info
,
3910 .bdrv_get_specific_info
= raw_get_specific_info
,
3911 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
3912 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3913 .bdrv_check_perm
= raw_check_perm
,
3914 .bdrv_set_perm
= raw_set_perm
,
3915 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3916 .create_opts
= &raw_create_opts
,
3917 .mutable_opts
= mutable_opts
,
3920 /***********************************************/
3923 #if defined(HAVE_HOST_BLOCK_DEVICE)
3925 #if defined(__APPLE__) && defined(__MACH__)
3926 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3927 CFIndex maxPathSize
, int flags
);
3929 #if !defined(MAC_OS_VERSION_12_0) \
3930 || (MAC_OS_X_VERSION_MIN_REQUIRED < MAC_OS_VERSION_12_0)
3931 #define IOMainPort IOMasterPort
3934 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3936 kern_return_t kernResult
= KERN_FAILURE
;
3937 mach_port_t mainPort
;
3938 CFMutableDictionaryRef classesToMatch
;
3939 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3940 char *mediaType
= NULL
;
3942 kernResult
= IOMainPort(MACH_PORT_NULL
, &mainPort
);
3943 if ( KERN_SUCCESS
!= kernResult
) {
3944 printf("IOMainPort returned %d\n", kernResult
);
3948 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3949 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3950 if (classesToMatch
== NULL
) {
3951 error_report("IOServiceMatching returned NULL for %s",
3952 matching_array
[index
]);
3955 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3957 kernResult
= IOServiceGetMatchingServices(mainPort
, classesToMatch
,
3959 if (kernResult
!= KERN_SUCCESS
) {
3960 error_report("Note: IOServiceGetMatchingServices returned %d",
3965 /* If a match was found, leave the loop */
3966 if (*mediaIterator
!= 0) {
3967 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3968 mediaType
= g_strdup(matching_array
[index
]);
3975 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3976 CFIndex maxPathSize
, int flags
)
3978 io_object_t nextMedia
;
3979 kern_return_t kernResult
= KERN_FAILURE
;
3981 nextMedia
= IOIteratorNext( mediaIterator
);
3984 CFTypeRef bsdPathAsCFString
;
3985 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3986 if ( bsdPathAsCFString
) {
3987 size_t devPathLength
;
3988 strcpy( bsdPath
, _PATH_DEV
);
3989 if (flags
& BDRV_O_NOCACHE
) {
3990 strcat(bsdPath
, "r");
3992 devPathLength
= strlen( bsdPath
);
3993 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3994 kernResult
= KERN_SUCCESS
;
3996 CFRelease( bsdPathAsCFString
);
3998 IOObjectRelease( nextMedia
);
4004 /* Sets up a real cdrom for use in QEMU */
4005 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
4007 int index
, num_of_test_partitions
= 2, fd
;
4008 char test_partition
[MAXPATHLEN
];
4009 bool partition_found
= false;
4011 /* look for a working partition */
4012 for (index
= 0; index
< num_of_test_partitions
; index
++) {
4013 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
4015 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
, NULL
);
4017 partition_found
= true;
4023 /* if a working partition on the device was not found */
4024 if (partition_found
== false) {
4025 error_setg(errp
, "Failed to find a working partition on disc");
4027 trace_file_setup_cdrom(test_partition
);
4028 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
4030 return partition_found
;
4033 /* Prints directions on mounting and unmounting a device */
4034 static void print_unmounting_directions(const char *file_name
)
4036 error_report("If device %s is mounted on the desktop, unmount"
4037 " it first before using it in QEMU", file_name
);
4038 error_report("Command to unmount device: diskutil unmountDisk %s",
4040 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
4043 #endif /* defined(__APPLE__) && defined(__MACH__) */
4045 static int hdev_probe_device(const char *filename
)
4049 /* allow a dedicated CD-ROM driver to match with a higher priority */
4050 if (strstart(filename
, "/dev/cdrom", NULL
))
4053 if (stat(filename
, &st
) >= 0 &&
4054 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
4061 static void hdev_parse_filename(const char *filename
, QDict
*options
,
4064 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
4067 static bool hdev_is_sg(BlockDriverState
*bs
)
4070 #if defined(__linux__)
4072 BDRVRawState
*s
= bs
->opaque
;
4074 struct sg_scsi_id scsiid
;
4078 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
4082 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
4087 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
4089 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
4098 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
4101 BDRVRawState
*s
= bs
->opaque
;
4104 #if defined(__APPLE__) && defined(__MACH__)
4106 * Caution: while qdict_get_str() is fine, getting non-string types
4107 * would require more care. When @options come from -blockdev or
4108 * blockdev_add, its members are typed according to the QAPI
4109 * schema, but when they come from -drive, they're all QString.
4111 const char *filename
= qdict_get_str(options
, "filename");
4112 char bsd_path
[MAXPATHLEN
] = "";
4113 bool error_occurred
= false;
4115 /* If using a real cdrom */
4116 if (strcmp(filename
, "/dev/cdrom") == 0) {
4117 char *mediaType
= NULL
;
4118 kern_return_t ret_val
;
4119 io_iterator_t mediaIterator
= 0;
4121 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
4122 if (mediaType
== NULL
) {
4123 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
4125 error_occurred
= true;
4126 goto hdev_open_Mac_error
;
4129 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
4130 if (ret_val
!= KERN_SUCCESS
) {
4131 error_setg(errp
, "Could not get BSD path for optical drive");
4132 error_occurred
= true;
4133 goto hdev_open_Mac_error
;
4136 /* If a real optical drive was not found */
4137 if (bsd_path
[0] == '\0') {
4138 error_setg(errp
, "Failed to obtain bsd path for optical drive");
4139 error_occurred
= true;
4140 goto hdev_open_Mac_error
;
4143 /* If using a cdrom disc and finding a partition on the disc failed */
4144 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
4145 setup_cdrom(bsd_path
, errp
) == false) {
4146 print_unmounting_directions(bsd_path
);
4147 error_occurred
= true;
4148 goto hdev_open_Mac_error
;
4151 qdict_put_str(options
, "filename", bsd_path
);
4153 hdev_open_Mac_error
:
4155 if (mediaIterator
) {
4156 IOObjectRelease(mediaIterator
);
4158 if (error_occurred
) {
4162 #endif /* defined(__APPLE__) && defined(__MACH__) */
4164 s
->type
= FTYPE_FILE
;
4166 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
4168 #if defined(__APPLE__) && defined(__MACH__)
4170 filename
= bsd_path
;
4172 /* if a physical device experienced an error while being opened */
4173 if (strncmp(filename
, "/dev/", 5) == 0) {
4174 print_unmounting_directions(filename
);
4176 #endif /* defined(__APPLE__) && defined(__MACH__) */
4180 /* Since this does ioctl the device must be already opened */
4181 bs
->sg
= hdev_is_sg(bs
);
4186 #if defined(__linux__)
4187 static int coroutine_fn
4188 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
4190 BDRVRawState
*s
= bs
->opaque
;
4191 RawPosixAIOData acb
;
4199 if (req
== SG_IO
&& s
->pr_mgr
) {
4200 struct sg_io_hdr
*io_hdr
= buf
;
4201 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
4202 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
4203 return pr_manager_execute(s
->pr_mgr
, qemu_get_current_aio_context(),
4208 acb
= (RawPosixAIOData
) {
4210 .aio_type
= QEMU_AIO_IOCTL
,
4211 .aio_fildes
= s
->fd
,
4219 return raw_thread_pool_submit(handle_aiocb_ioctl
, &acb
);
4223 static coroutine_fn
int
4224 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int64_t bytes
)
4226 BDRVRawState
*s
= bs
->opaque
;
4231 raw_account_discard(s
, bytes
, ret
);
4234 return raw_do_pdiscard(bs
, offset
, bytes
, true);
4237 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
4238 int64_t offset
, int64_t bytes
, BdrvRequestFlags flags
)
4247 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
4250 static BlockDriver bdrv_host_device
= {
4251 .format_name
= "host_device",
4252 .protocol_name
= "host_device",
4253 .instance_size
= sizeof(BDRVRawState
),
4254 .bdrv_needs_filename
= true,
4255 .bdrv_probe_device
= hdev_probe_device
,
4256 .bdrv_parse_filename
= hdev_parse_filename
,
4257 .bdrv_file_open
= hdev_open
,
4258 .bdrv_close
= raw_close
,
4259 .bdrv_reopen_prepare
= raw_reopen_prepare
,
4260 .bdrv_reopen_commit
= raw_reopen_commit
,
4261 .bdrv_reopen_abort
= raw_reopen_abort
,
4262 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
4263 .create_opts
= &bdrv_create_opts_simple
,
4264 .mutable_opts
= mutable_opts
,
4265 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
4266 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
4268 .bdrv_co_preadv
= raw_co_preadv
,
4269 .bdrv_co_pwritev
= raw_co_pwritev
,
4270 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
4271 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
4272 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
4273 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
4274 .bdrv_refresh_limits
= raw_refresh_limits
,
4276 .bdrv_co_truncate
= raw_co_truncate
,
4277 .bdrv_co_getlength
= raw_co_getlength
,
4278 .bdrv_co_get_info
= raw_co_get_info
,
4279 .bdrv_get_specific_info
= raw_get_specific_info
,
4280 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
4281 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
4282 .bdrv_check_perm
= raw_check_perm
,
4283 .bdrv_set_perm
= raw_set_perm
,
4284 .bdrv_abort_perm_update
= raw_abort_perm_update
,
4285 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
4286 .bdrv_probe_geometry
= hdev_probe_geometry
,
4288 /* generic scsi device */
4290 .bdrv_co_ioctl
= hdev_co_ioctl
,
4294 #if defined(CONFIG_BLKZONED)
4295 /* zone management operations */
4296 .bdrv_co_zone_report
= raw_co_zone_report
,
4297 .bdrv_co_zone_mgmt
= raw_co_zone_mgmt
,
4298 .bdrv_co_zone_append
= raw_co_zone_append
,
4302 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
4303 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
4306 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
4309 static void cdrom_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
4311 bs
->bl
.has_variable_length
= true;
4312 raw_refresh_limits(bs
, errp
);
4317 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
4320 BDRVRawState
*s
= bs
->opaque
;
4324 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
4325 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
4328 static int cdrom_probe_device(const char *filename
)
4334 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
, NULL
);
4338 ret
= fstat(fd
, &st
);
4339 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
4343 /* Attempt to detect via a CDROM specific ioctl */
4344 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
4354 static bool coroutine_fn
cdrom_co_is_inserted(BlockDriverState
*bs
)
4356 BDRVRawState
*s
= bs
->opaque
;
4359 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
4360 return ret
== CDS_DISC_OK
;
4363 static void coroutine_fn
cdrom_co_eject(BlockDriverState
*bs
, bool eject_flag
)
4365 BDRVRawState
*s
= bs
->opaque
;
4368 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
4369 perror("CDROMEJECT");
4371 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
4372 perror("CDROMEJECT");
4376 static void coroutine_fn
cdrom_co_lock_medium(BlockDriverState
*bs
, bool locked
)
4378 BDRVRawState
*s
= bs
->opaque
;
4380 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
4382 * Note: an error can happen if the distribution automatically
4385 /* perror("CDROM_LOCKDOOR"); */
4389 static BlockDriver bdrv_host_cdrom
= {
4390 .format_name
= "host_cdrom",
4391 .protocol_name
= "host_cdrom",
4392 .instance_size
= sizeof(BDRVRawState
),
4393 .bdrv_needs_filename
= true,
4394 .bdrv_probe_device
= cdrom_probe_device
,
4395 .bdrv_parse_filename
= cdrom_parse_filename
,
4396 .bdrv_file_open
= cdrom_open
,
4397 .bdrv_close
= raw_close
,
4398 .bdrv_reopen_prepare
= raw_reopen_prepare
,
4399 .bdrv_reopen_commit
= raw_reopen_commit
,
4400 .bdrv_reopen_abort
= raw_reopen_abort
,
4401 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
4402 .create_opts
= &bdrv_create_opts_simple
,
4403 .mutable_opts
= mutable_opts
,
4404 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
4406 .bdrv_co_preadv
= raw_co_preadv
,
4407 .bdrv_co_pwritev
= raw_co_pwritev
,
4408 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
4409 .bdrv_refresh_limits
= cdrom_refresh_limits
,
4411 .bdrv_co_truncate
= raw_co_truncate
,
4412 .bdrv_co_getlength
= raw_co_getlength
,
4413 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
4415 /* removable device support */
4416 .bdrv_co_is_inserted
= cdrom_co_is_inserted
,
4417 .bdrv_co_eject
= cdrom_co_eject
,
4418 .bdrv_co_lock_medium
= cdrom_co_lock_medium
,
4420 /* generic scsi device */
4421 .bdrv_co_ioctl
= hdev_co_ioctl
,
4423 #endif /* __linux__ */
4425 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
4426 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
4429 BDRVRawState
*s
= bs
->opaque
;
4434 ret
= raw_open_common(bs
, options
, flags
, 0, true, errp
);
4439 /* make sure the door isn't locked at this time */
4440 ioctl(s
->fd
, CDIOCALLOW
);
4444 static int cdrom_probe_device(const char *filename
)
4446 if (strstart(filename
, "/dev/cd", NULL
) ||
4447 strstart(filename
, "/dev/acd", NULL
))
4452 static int cdrom_reopen(BlockDriverState
*bs
)
4454 BDRVRawState
*s
= bs
->opaque
;
4458 * Force reread of possibly changed/newly loaded disc,
4459 * FreeBSD seems to not notice sometimes...
4463 fd
= qemu_open(bs
->filename
, s
->open_flags
, NULL
);
4470 /* make sure the door isn't locked at this time */
4471 ioctl(s
->fd
, CDIOCALLOW
);
4475 static bool coroutine_fn
cdrom_co_is_inserted(BlockDriverState
*bs
)
4477 return raw_getlength(bs
) > 0;
4480 static void coroutine_fn
cdrom_co_eject(BlockDriverState
*bs
, bool eject_flag
)
4482 BDRVRawState
*s
= bs
->opaque
;
4487 (void) ioctl(s
->fd
, CDIOCALLOW
);
4490 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
4491 perror("CDIOCEJECT");
4493 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
4494 perror("CDIOCCLOSE");
4500 static void coroutine_fn
cdrom_co_lock_medium(BlockDriverState
*bs
, bool locked
)
4502 BDRVRawState
*s
= bs
->opaque
;
4506 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
4508 * Note: an error can happen if the distribution automatically
4511 /* perror("CDROM_LOCKDOOR"); */
4515 static BlockDriver bdrv_host_cdrom
= {
4516 .format_name
= "host_cdrom",
4517 .protocol_name
= "host_cdrom",
4518 .instance_size
= sizeof(BDRVRawState
),
4519 .bdrv_needs_filename
= true,
4520 .bdrv_probe_device
= cdrom_probe_device
,
4521 .bdrv_parse_filename
= cdrom_parse_filename
,
4522 .bdrv_file_open
= cdrom_open
,
4523 .bdrv_close
= raw_close
,
4524 .bdrv_reopen_prepare
= raw_reopen_prepare
,
4525 .bdrv_reopen_commit
= raw_reopen_commit
,
4526 .bdrv_reopen_abort
= raw_reopen_abort
,
4527 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
4528 .create_opts
= &bdrv_create_opts_simple
,
4529 .mutable_opts
= mutable_opts
,
4531 .bdrv_co_preadv
= raw_co_preadv
,
4532 .bdrv_co_pwritev
= raw_co_pwritev
,
4533 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
4534 .bdrv_refresh_limits
= cdrom_refresh_limits
,
4536 .bdrv_co_truncate
= raw_co_truncate
,
4537 .bdrv_co_getlength
= raw_co_getlength
,
4538 .bdrv_co_get_allocated_file_size
= raw_co_get_allocated_file_size
,
4540 /* removable device support */
4541 .bdrv_co_is_inserted
= cdrom_co_is_inserted
,
4542 .bdrv_co_eject
= cdrom_co_eject
,
4543 .bdrv_co_lock_medium
= cdrom_co_lock_medium
,
4545 #endif /* __FreeBSD__ */
4547 #endif /* HAVE_HOST_BLOCK_DEVICE */
4549 static void bdrv_file_init(void)
4552 * Register all the drivers. Note that order is important, the driver
4553 * registered last will get probed first.
4555 bdrv_register(&bdrv_file
);
4556 #if defined(HAVE_HOST_BLOCK_DEVICE)
4557 bdrv_register(&bdrv_host_device
);
4559 bdrv_register(&bdrv_host_cdrom
);
4561 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
4562 bdrv_register(&bdrv_host_cdrom
);
4564 #endif /* HAVE_HOST_BLOCK_DEVICE */
4567 block_init(bdrv_file_init
);