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 "qemu-common.h"
27 #include "qapi/error.h"
28 #include "qemu/cutils.h"
29 #include "qemu/error-report.h"
30 #include "block/block_int.h"
31 #include "qemu/module.h"
32 #include "qemu/option.h"
34 #include "block/thread-pool.h"
36 #include "block/raw-aio.h"
37 #include "qapi/qmp/qdict.h"
38 #include "qapi/qmp/qstring.h"
40 #include "scsi/pr-manager.h"
41 #include "scsi/constants.h"
43 #if defined(__APPLE__) && (__MACH__)
45 #include <sys/param.h>
46 #include <IOKit/IOKitLib.h>
47 #include <IOKit/IOBSD.h>
48 #include <IOKit/storage/IOMediaBSDClient.h>
49 #include <IOKit/storage/IOMedia.h>
50 #include <IOKit/storage/IOCDMedia.h>
51 //#include <IOKit/storage/IOCDTypes.h>
52 #include <IOKit/storage/IODVDMedia.h>
53 #include <CoreFoundation/CoreFoundation.h>
57 #define _POSIX_PTHREAD_SEMANTICS 1
61 #include <sys/ioctl.h>
62 #include <sys/param.h>
63 #include <sys/syscall.h>
64 #include <linux/cdrom.h>
67 #include <linux/hdreg.h>
73 #define FS_NOCOW_FL 0x00800000 /* Do not cow file */
76 #if defined(CONFIG_FALLOCATE_PUNCH_HOLE) || defined(CONFIG_FALLOCATE_ZERO_RANGE)
77 #include <linux/falloc.h>
79 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
85 #include <sys/ioctl.h>
86 #include <sys/disklabel.h>
91 #include <sys/ioctl.h>
92 #include <sys/disklabel.h>
98 #include <sys/ioctl.h>
99 #include <sys/diskslice.h>
108 /* OS X does not have O_DSYNC */
111 #define O_DSYNC O_SYNC
112 #elif defined(O_FSYNC)
113 #define O_DSYNC O_FSYNC
117 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
119 #define O_DIRECT O_DSYNC
125 #define MAX_BLOCKSIZE 4096
127 /* Posix file locking bytes. Libvirt takes byte 0, we start from higher bytes,
128 * leaving a few more bytes for its future use. */
129 #define RAW_LOCK_PERM_BASE 100
130 #define RAW_LOCK_SHARED_BASE 200
132 typedef struct BDRVRawState
{
139 /* The current permissions. */
141 uint64_t shared_perm
;
143 /* The perms bits whose corresponding bytes are already locked in
145 uint64_t locked_perm
;
146 uint64_t locked_shared_perm
;
149 int perm_change_flags
;
150 BDRVReopenState
*reopen_state
;
156 bool has_write_zeroes
:1;
157 bool discard_zeroes
:1;
158 bool use_linux_aio
:1;
159 bool use_linux_io_uring
:1;
160 bool page_cache_inconsistent
:1;
162 bool needs_alignment
;
164 bool check_cache_dropped
;
166 uint64_t discard_nb_ok
;
167 uint64_t discard_nb_failed
;
168 uint64_t discard_bytes_ok
;
174 typedef struct BDRVRawReopenState
{
178 bool check_cache_dropped
;
179 } BDRVRawReopenState
;
181 static int fd_open(BlockDriverState
*bs
);
182 static int64_t raw_getlength(BlockDriverState
*bs
);
184 typedef struct RawPosixAIOData
{
185 BlockDriverState
*bs
;
206 PreallocMode prealloc
;
212 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
213 static int cdrom_reopen(BlockDriverState
*bs
);
216 #if defined(__NetBSD__)
217 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
219 static char namebuf
[PATH_MAX
];
220 const char *dp
, *fname
;
224 dp
= strrchr(fname
, '/');
225 if (lstat(fname
, &sb
) < 0) {
226 error_setg_file_open(errp
, errno
, fname
);
230 if (!S_ISBLK(sb
.st_mode
)) {
235 snprintf(namebuf
, PATH_MAX
, "r%s", fname
);
237 snprintf(namebuf
, PATH_MAX
, "%.*s/r%s",
238 (int)(dp
- fname
), fname
, dp
+ 1);
241 warn_report("%s is a block device, using %s", fname
, *filename
);
246 static int raw_normalize_devicepath(const char **filename
, Error
**errp
)
253 * Get logical block size via ioctl. On success store it in @sector_size_p.
255 static int probe_logical_blocksize(int fd
, unsigned int *sector_size_p
)
257 unsigned int sector_size
;
258 bool success
= false;
262 static const unsigned long ioctl_list
[] = {
266 #ifdef DKIOCGETBLOCKSIZE
269 #ifdef DIOCGSECTORSIZE
274 /* Try a few ioctls to get the right size */
275 for (i
= 0; i
< (int)ARRAY_SIZE(ioctl_list
); i
++) {
276 if (ioctl(fd
, ioctl_list
[i
], §or_size
) >= 0) {
277 *sector_size_p
= sector_size
;
282 return success
? 0 : -errno
;
286 * Get physical block size of @fd.
287 * On success, store it in @blk_size and return 0.
288 * On failure, return -errno.
290 static int probe_physical_blocksize(int fd
, unsigned int *blk_size
)
293 if (ioctl(fd
, BLKPBSZGET
, blk_size
) < 0) {
302 /* Check if read is allowed with given memory buffer and length.
304 * This function is used to check O_DIRECT memory buffer and request alignment.
306 static bool raw_is_io_aligned(int fd
, void *buf
, size_t len
)
308 ssize_t ret
= pread(fd
, buf
, len
, 0);
315 /* The Linux kernel returns EINVAL for misaligned O_DIRECT reads. Ignore
316 * other errors (e.g. real I/O error), which could happen on a failed
317 * drive, since we only care about probing alignment.
319 if (errno
!= EINVAL
) {
327 static void raw_probe_alignment(BlockDriverState
*bs
, int fd
, Error
**errp
)
329 BDRVRawState
*s
= bs
->opaque
;
331 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size
);
332 size_t alignments
[] = {1, 512, 1024, 2048, 4096};
334 /* For SCSI generic devices the alignment is not really used.
335 With buffered I/O, we don't have any restrictions. */
336 if (bdrv_is_sg(bs
) || !s
->needs_alignment
) {
337 bs
->bl
.request_alignment
= 1;
342 bs
->bl
.request_alignment
= 0;
344 /* Let's try to use the logical blocksize for the alignment. */
345 if (probe_logical_blocksize(fd
, &bs
->bl
.request_alignment
) < 0) {
346 bs
->bl
.request_alignment
= 0;
351 if (xfsctl(NULL
, fd
, XFS_IOC_DIOINFO
, &da
) >= 0) {
352 bs
->bl
.request_alignment
= da
.d_miniosz
;
353 /* The kernel returns wrong information for d_mem */
354 /* s->buf_align = da.d_mem; */
360 * If we could not get the sizes so far, we can only guess them. First try
361 * to detect request alignment, since it is more likely to succeed. Then
362 * try to detect buf_align, which cannot be detected in some cases (e.g.
363 * Gluster). If buf_align cannot be detected, we fallback to the value of
367 if (!bs
->bl
.request_alignment
) {
370 buf
= qemu_memalign(max_align
, max_align
);
371 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
372 align
= alignments
[i
];
373 if (raw_is_io_aligned(fd
, buf
, align
)) {
374 /* Fallback to safe value. */
375 bs
->bl
.request_alignment
= (align
!= 1) ? align
: max_align
;
385 buf
= qemu_memalign(max_align
, 2 * max_align
);
386 for (i
= 0; i
< ARRAY_SIZE(alignments
); i
++) {
387 align
= alignments
[i
];
388 if (raw_is_io_aligned(fd
, buf
+ align
, max_align
)) {
389 /* Fallback to request_alignment. */
390 s
->buf_align
= (align
!= 1) ? align
: bs
->bl
.request_alignment
;
397 if (!s
->buf_align
|| !bs
->bl
.request_alignment
) {
398 error_setg(errp
, "Could not find working O_DIRECT alignment");
399 error_append_hint(errp
, "Try cache.direct=off\n");
403 static void raw_parse_flags(int bdrv_flags
, int *open_flags
, bool has_writers
)
405 bool read_write
= false;
406 assert(open_flags
!= NULL
);
408 *open_flags
|= O_BINARY
;
409 *open_flags
&= ~O_ACCMODE
;
411 if (bdrv_flags
& BDRV_O_AUTO_RDONLY
) {
412 read_write
= has_writers
;
413 } else if (bdrv_flags
& BDRV_O_RDWR
) {
418 *open_flags
|= O_RDWR
;
420 *open_flags
|= O_RDONLY
;
423 /* Use O_DSYNC for write-through caching, no flags for write-back caching,
424 * and O_DIRECT for no caching. */
425 if ((bdrv_flags
& BDRV_O_NOCACHE
)) {
426 *open_flags
|= O_DIRECT
;
430 static void raw_parse_filename(const char *filename
, QDict
*options
,
433 bdrv_parse_filename_strip_prefix(filename
, "file:", options
);
436 static QemuOptsList raw_runtime_opts
= {
438 .head
= QTAILQ_HEAD_INITIALIZER(raw_runtime_opts
.head
),
442 .type
= QEMU_OPT_STRING
,
443 .help
= "File name of the image",
447 .type
= QEMU_OPT_STRING
,
448 .help
= "host AIO implementation (threads, native, io_uring)",
452 .type
= QEMU_OPT_STRING
,
453 .help
= "file locking mode (on/off/auto, default: auto)",
456 .name
= "pr-manager",
457 .type
= QEMU_OPT_STRING
,
458 .help
= "id of persistent reservation manager object (default: none)",
460 #if defined(__linux__)
462 .name
= "drop-cache",
463 .type
= QEMU_OPT_BOOL
,
464 .help
= "invalidate page cache during live migration (default: on)",
468 .name
= "x-check-cache-dropped",
469 .type
= QEMU_OPT_BOOL
,
470 .help
= "check that page cache was dropped on live migration (default: off)"
472 { /* end of list */ }
476 static const char *const mutable_opts
[] = { "x-check-cache-dropped", NULL
};
478 static int raw_open_common(BlockDriverState
*bs
, QDict
*options
,
479 int bdrv_flags
, int open_flags
,
480 bool device
, Error
**errp
)
482 BDRVRawState
*s
= bs
->opaque
;
484 Error
*local_err
= NULL
;
485 const char *filename
= NULL
;
487 BlockdevAioOptions aio
, aio_default
;
492 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
493 qemu_opts_absorb_qdict(opts
, options
, &local_err
);
495 error_propagate(errp
, local_err
);
500 filename
= qemu_opt_get(opts
, "filename");
502 ret
= raw_normalize_devicepath(&filename
, errp
);
507 if (bdrv_flags
& BDRV_O_NATIVE_AIO
) {
508 aio_default
= BLOCKDEV_AIO_OPTIONS_NATIVE
;
509 #ifdef CONFIG_LINUX_IO_URING
510 } else if (bdrv_flags
& BDRV_O_IO_URING
) {
511 aio_default
= BLOCKDEV_AIO_OPTIONS_IO_URING
;
514 aio_default
= BLOCKDEV_AIO_OPTIONS_THREADS
;
517 aio
= qapi_enum_parse(&BlockdevAioOptions_lookup
,
518 qemu_opt_get(opts
, "aio"),
519 aio_default
, &local_err
);
521 error_propagate(errp
, local_err
);
526 s
->use_linux_aio
= (aio
== BLOCKDEV_AIO_OPTIONS_NATIVE
);
527 #ifdef CONFIG_LINUX_IO_URING
528 s
->use_linux_io_uring
= (aio
== BLOCKDEV_AIO_OPTIONS_IO_URING
);
531 locking
= qapi_enum_parse(&OnOffAuto_lookup
,
532 qemu_opt_get(opts
, "locking"),
533 ON_OFF_AUTO_AUTO
, &local_err
);
535 error_propagate(errp
, local_err
);
542 if (!qemu_has_ofd_lock()) {
543 warn_report("File lock requested but OFD locking syscall is "
544 "unavailable, falling back to POSIX file locks");
545 error_printf("Due to the implementation, locks can be lost "
549 case ON_OFF_AUTO_OFF
:
552 case ON_OFF_AUTO_AUTO
:
553 s
->use_lock
= qemu_has_ofd_lock();
559 str
= qemu_opt_get(opts
, "pr-manager");
561 s
->pr_mgr
= pr_manager_lookup(str
, &local_err
);
563 error_propagate(errp
, local_err
);
569 s
->drop_cache
= qemu_opt_get_bool(opts
, "drop-cache", true);
570 s
->check_cache_dropped
= qemu_opt_get_bool(opts
, "x-check-cache-dropped",
573 s
->open_flags
= open_flags
;
574 raw_parse_flags(bdrv_flags
, &s
->open_flags
, false);
577 fd
= qemu_open(filename
, s
->open_flags
, 0644);
578 ret
= fd
< 0 ? -errno
: 0;
581 error_setg_file_open(errp
, -ret
, filename
);
590 s
->shared_perm
= BLK_PERM_ALL
;
592 #ifdef CONFIG_LINUX_AIO
593 /* Currently Linux does AIO only for files opened with O_DIRECT */
594 if (s
->use_linux_aio
) {
595 if (!(s
->open_flags
& O_DIRECT
)) {
596 error_setg(errp
, "aio=native was specified, but it requires "
597 "cache.direct=on, which was not specified.");
601 if (!aio_setup_linux_aio(bdrv_get_aio_context(bs
), errp
)) {
602 error_prepend(errp
, "Unable to use native AIO: ");
607 if (s
->use_linux_aio
) {
608 error_setg(errp
, "aio=native was specified, but is not supported "
613 #endif /* !defined(CONFIG_LINUX_AIO) */
615 #ifdef CONFIG_LINUX_IO_URING
616 if (s
->use_linux_io_uring
) {
617 if (!aio_setup_linux_io_uring(bdrv_get_aio_context(bs
), errp
)) {
618 error_prepend(errp
, "Unable to use io_uring: ");
623 if (s
->use_linux_io_uring
) {
624 error_setg(errp
, "aio=io_uring was specified, but is not supported "
629 #endif /* !defined(CONFIG_LINUX_IO_URING) */
631 s
->has_discard
= true;
632 s
->has_write_zeroes
= true;
633 if ((bs
->open_flags
& BDRV_O_NOCACHE
) != 0) {
634 s
->needs_alignment
= true;
637 if (fstat(s
->fd
, &st
) < 0) {
639 error_setg_errno(errp
, errno
, "Could not stat file");
644 if (S_ISBLK(st
.st_mode
)) {
645 warn_report("Opening a block device as a file using the '%s' "
646 "driver is deprecated", bs
->drv
->format_name
);
647 } else if (S_ISCHR(st
.st_mode
)) {
648 warn_report("Opening a character device as a file using the '%s' "
649 "driver is deprecated", bs
->drv
->format_name
);
650 } else if (!S_ISREG(st
.st_mode
)) {
651 error_setg(errp
, "A regular file was expected by the '%s' driver, "
652 "but something else was given", bs
->drv
->format_name
);
656 s
->discard_zeroes
= true;
657 s
->has_fallocate
= true;
660 if (!(S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
661 error_setg(errp
, "'%s' driver expects either "
662 "a character or block device", bs
->drv
->format_name
);
668 if (S_ISBLK(st
.st_mode
)) {
669 #ifdef BLKDISCARDZEROES
671 if (ioctl(s
->fd
, BLKDISCARDZEROES
, &arg
) == 0 && arg
) {
672 s
->discard_zeroes
= true;
676 /* On Linux 3.10, BLKDISCARD leaves stale data in the page cache. Do
677 * not rely on the contents of discarded blocks unless using O_DIRECT.
678 * Same for BLKZEROOUT.
680 if (!(bs
->open_flags
& BDRV_O_NOCACHE
)) {
681 s
->discard_zeroes
= false;
682 s
->has_write_zeroes
= false;
687 if (S_ISCHR(st
.st_mode
)) {
689 * The file is a char device (disk), which on FreeBSD isn't behind
690 * a pager, so force all requests to be aligned. This is needed
691 * so QEMU makes sure all IO operations on the device are aligned
692 * to sector size, or else FreeBSD will reject them with EINVAL.
694 s
->needs_alignment
= true;
699 if (platform_test_xfs_fd(s
->fd
)) {
704 bs
->supported_zero_flags
= BDRV_REQ_MAY_UNMAP
| BDRV_REQ_NO_FALLBACK
;
707 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
714 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
717 BDRVRawState
*s
= bs
->opaque
;
719 s
->type
= FTYPE_FILE
;
720 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
729 #define PERM_FOREACH(i) \
730 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
732 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
733 * file; if @unlock == true, also unlock the unneeded bytes.
734 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
736 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
737 uint64_t perm_lock_bits
,
738 uint64_t shared_perm_lock_bits
,
739 bool unlock
, Error
**errp
)
743 uint64_t locked_perm
, locked_shared_perm
;
746 locked_perm
= s
->locked_perm
;
747 locked_shared_perm
= s
->locked_shared_perm
;
750 * We don't have the previous bits, just lock/unlock for each of the
754 locked_perm
= BLK_PERM_ALL
;
755 locked_shared_perm
= BLK_PERM_ALL
;
758 locked_shared_perm
= 0;
763 int off
= RAW_LOCK_PERM_BASE
+ i
;
764 uint64_t bit
= (1ULL << i
);
765 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
766 ret
= qemu_lock_fd(fd
, off
, 1, false);
768 error_setg(errp
, "Failed to lock byte %d", off
);
771 s
->locked_perm
|= bit
;
773 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
774 ret
= qemu_unlock_fd(fd
, off
, 1);
776 error_setg(errp
, "Failed to unlock byte %d", off
);
779 s
->locked_perm
&= ~bit
;
784 int off
= RAW_LOCK_SHARED_BASE
+ i
;
785 uint64_t bit
= (1ULL << i
);
786 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
787 ret
= qemu_lock_fd(fd
, off
, 1, false);
789 error_setg(errp
, "Failed to lock byte %d", off
);
792 s
->locked_shared_perm
|= bit
;
794 } else if (unlock
&& (locked_shared_perm
& bit
) &&
795 !(shared_perm_lock_bits
& bit
)) {
796 ret
= qemu_unlock_fd(fd
, off
, 1);
798 error_setg(errp
, "Failed to unlock byte %d", off
);
801 s
->locked_shared_perm
&= ~bit
;
808 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
809 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
816 int off
= RAW_LOCK_SHARED_BASE
+ i
;
817 uint64_t p
= 1ULL << i
;
819 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
821 char *perm_name
= bdrv_perm_names(p
);
823 "Failed to get \"%s\" lock",
831 int off
= RAW_LOCK_PERM_BASE
+ i
;
832 uint64_t p
= 1ULL << i
;
833 if (!(shared_perm
& p
)) {
834 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
836 char *perm_name
= bdrv_perm_names(p
);
838 "Failed to get shared \"%s\" lock",
848 static int raw_handle_perm_lock(BlockDriverState
*bs
,
850 uint64_t new_perm
, uint64_t new_shared
,
853 BDRVRawState
*s
= bs
->opaque
;
855 Error
*local_err
= NULL
;
861 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
867 if ((s
->perm
| new_perm
) == s
->perm
&&
868 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
871 * We are going to unlock bytes, it should not fail. If it fail due
872 * to some fs-dependent permission-unrelated reasons (which occurs
873 * sometimes on NFS and leads to abort in bdrv_replace_child) we
874 * can't prevent such errors by any check here. And we ignore them
875 * anyway in ABORT and COMMIT.
879 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
880 ~s
->shared_perm
| ~new_shared
,
883 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
887 error_append_hint(errp
,
888 "Is another process using the image [%s]?\n",
892 /* fall through to unlock bytes. */
894 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
897 /* Theoretically the above call only unlocks bytes and it cannot
898 * fail. Something weird happened, report it.
900 warn_report_err(local_err
);
904 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
907 /* Theoretically the above call only unlocks bytes and it cannot
908 * fail. Something weird happened, report it.
910 warn_report_err(local_err
);
917 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
918 int *open_flags
, uint64_t perm
, bool force_dup
,
921 BDRVRawState
*s
= bs
->opaque
;
924 bool has_writers
= perm
&
925 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
926 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
928 fcntl_flags
|= O_NOATIME
;
932 if (s
->type
== FTYPE_CD
) {
933 *open_flags
|= O_NONBLOCK
;
936 raw_parse_flags(flags
, open_flags
, has_writers
);
939 /* Not all operating systems have O_ASYNC, and those that don't
940 * will not let us track the state into rs->open_flags (typically
941 * you achieve the same effect with an ioctl, for example I_SETSIG
942 * on Solaris). But we do not use O_ASYNC, so that's fine.
944 assert((s
->open_flags
& O_ASYNC
) == 0);
947 if (!force_dup
&& *open_flags
== s
->open_flags
) {
948 /* We're lucky, the existing fd is fine */
952 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
953 /* dup the original fd */
954 fd
= qemu_dup(s
->fd
);
956 ret
= fcntl_setfl(fd
, *open_flags
);
964 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
966 const char *normalized_filename
= bs
->filename
;
967 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
969 assert(!(*open_flags
& O_CREAT
));
970 fd
= qemu_open(normalized_filename
, *open_flags
);
972 error_setg_errno(errp
, errno
, "Could not reopen file");
981 static int raw_reopen_prepare(BDRVReopenState
*state
,
982 BlockReopenQueue
*queue
, Error
**errp
)
985 BDRVRawReopenState
*rs
;
988 Error
*local_err
= NULL
;
990 assert(state
!= NULL
);
991 assert(state
->bs
!= NULL
);
993 s
= state
->bs
->opaque
;
995 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
998 /* Handle options changes */
999 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
1000 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
1002 error_propagate(errp
, local_err
);
1007 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
1008 rs
->check_cache_dropped
=
1009 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
1011 /* This driver's reopen function doesn't currently allow changing
1012 * other options, so let's put them back in the original QDict and
1013 * bdrv_reopen_prepare() will detect changes and complain. */
1014 qemu_opts_to_qdict(opts
, state
->options
);
1016 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
1017 state
->perm
, true, &local_err
);
1019 error_propagate(errp
, local_err
);
1024 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
1025 * alignment with the new fd. */
1027 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
1029 error_propagate(errp
, local_err
);
1035 s
->reopen_state
= state
;
1043 qemu_opts_del(opts
);
1047 static void raw_reopen_commit(BDRVReopenState
*state
)
1049 BDRVRawReopenState
*rs
= state
->opaque
;
1050 BDRVRawState
*s
= state
->bs
->opaque
;
1052 s
->drop_cache
= rs
->drop_cache
;
1053 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1054 s
->open_flags
= rs
->open_flags
;
1059 g_free(state
->opaque
);
1060 state
->opaque
= NULL
;
1062 assert(s
->reopen_state
== state
);
1063 s
->reopen_state
= NULL
;
1067 static void raw_reopen_abort(BDRVReopenState
*state
)
1069 BDRVRawReopenState
*rs
= state
->opaque
;
1070 BDRVRawState
*s
= state
->bs
->opaque
;
1072 /* nothing to do if NULL, we didn't get far enough */
1081 g_free(state
->opaque
);
1082 state
->opaque
= NULL
;
1084 assert(s
->reopen_state
== state
);
1085 s
->reopen_state
= NULL
;
1088 static int sg_get_max_transfer_length(int fd
)
1093 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1103 static int sg_get_max_segments(int fd
)
1108 char *sysfspath
= NULL
;
1114 if (fstat(fd
, &st
)) {
1119 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1120 major(st
.st_rdev
), minor(st
.st_rdev
));
1121 sysfd
= open(sysfspath
, O_RDONLY
);
1127 ret
= read(sysfd
, buf
, sizeof(buf
) - 1);
1128 } while (ret
== -1 && errno
== EINTR
);
1132 } else if (ret
== 0) {
1137 /* The file is ended with '\n', pass 'end' to accept that. */
1138 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1139 if (ret
== 0 && end
&& *end
== '\n') {
1154 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1156 BDRVRawState
*s
= bs
->opaque
;
1159 int ret
= sg_get_max_transfer_length(s
->fd
);
1161 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1162 bs
->bl
.max_transfer
= pow2floor(ret
);
1165 ret
= sg_get_max_segments(s
->fd
);
1167 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1168 ret
* qemu_real_host_page_size
);
1172 raw_probe_alignment(bs
, s
->fd
, errp
);
1173 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1174 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, qemu_real_host_page_size
);
1177 static int check_for_dasd(int fd
)
1180 struct dasd_information2_t info
= {0};
1182 return ioctl(fd
, BIODASDINFO2
, &info
);
1189 * Try to get @bs's logical and physical block size.
1190 * On success, store them in @bsz and return zero.
1191 * On failure, return negative errno.
1193 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1195 BDRVRawState
*s
= bs
->opaque
;
1198 /* If DASD, get blocksizes */
1199 if (check_for_dasd(s
->fd
) < 0) {
1202 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1206 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1210 * Try to get @bs's geometry: cyls, heads, sectors.
1211 * On success, store them in @geo and return 0.
1212 * On failure return -errno.
1213 * (Allows block driver to assign default geometry values that guest sees)
1216 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1218 BDRVRawState
*s
= bs
->opaque
;
1219 struct hd_geometry ioctl_geo
= {0};
1221 /* If DASD, get its geometry */
1222 if (check_for_dasd(s
->fd
) < 0) {
1225 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1228 /* HDIO_GETGEO may return success even though geo contains zeros
1229 (e.g. certain multipath setups) */
1230 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1233 /* Do not return a geometry for partition */
1234 if (ioctl_geo
.start
!= 0) {
1237 geo
->heads
= ioctl_geo
.heads
;
1238 geo
->sectors
= ioctl_geo
.sectors
;
1239 geo
->cylinders
= ioctl_geo
.cylinders
;
1243 #else /* __linux__ */
1244 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1250 #if defined(__linux__)
1251 static int handle_aiocb_ioctl(void *opaque
)
1253 RawPosixAIOData
*aiocb
= opaque
;
1256 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1265 static int handle_aiocb_flush(void *opaque
)
1267 RawPosixAIOData
*aiocb
= opaque
;
1268 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1271 if (s
->page_cache_inconsistent
) {
1275 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1277 /* There is no clear definition of the semantics of a failing fsync(),
1278 * so we may have to assume the worst. The sad truth is that this
1279 * assumption is correct for Linux. Some pages are now probably marked
1280 * clean in the page cache even though they are inconsistent with the
1281 * on-disk contents. The next fdatasync() call would succeed, but no
1282 * further writeback attempt will be made. We can't get back to a state
1283 * in which we know what is on disk (we would have to rewrite
1284 * everything that was touched since the last fdatasync() at least), so
1285 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1286 * really defined, I have little hope that other OSes are doing better.
1288 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1290 if ((s
->open_flags
& O_DIRECT
) == 0) {
1291 s
->page_cache_inconsistent
= true;
1298 #ifdef CONFIG_PREADV
1300 static bool preadv_present
= true;
1303 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1305 return preadv(fd
, iov
, nr_iov
, offset
);
1309 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1311 return pwritev(fd
, iov
, nr_iov
, offset
);
1316 static bool preadv_present
= false;
1319 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1325 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1332 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1337 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1338 len
= qemu_pwritev(aiocb
->aio_fildes
,
1343 len
= qemu_preadv(aiocb
->aio_fildes
,
1347 } while (len
== -1 && errno
== EINTR
);
1356 * Read/writes the data to/from a given linear buffer.
1358 * Returns the number of bytes handles or -errno in case of an error. Short
1359 * reads are only returned if the end of the file is reached.
1361 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1366 while (offset
< aiocb
->aio_nbytes
) {
1367 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1368 len
= pwrite(aiocb
->aio_fildes
,
1369 (const char *)buf
+ offset
,
1370 aiocb
->aio_nbytes
- offset
,
1371 aiocb
->aio_offset
+ offset
);
1373 len
= pread(aiocb
->aio_fildes
,
1375 aiocb
->aio_nbytes
- offset
,
1376 aiocb
->aio_offset
+ offset
);
1378 if (len
== -1 && errno
== EINTR
) {
1380 } else if (len
== -1 && errno
== EINVAL
&&
1381 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1382 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1384 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1385 * after a short read. Assume that O_DIRECT short reads only occur
1386 * at EOF. Therefore this is a short read, not an I/O error.
1389 } else if (len
== -1) {
1392 } else if (len
== 0) {
1401 static int handle_aiocb_rw(void *opaque
)
1403 RawPosixAIOData
*aiocb
= opaque
;
1407 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1409 * If there is just a single buffer, and it is properly aligned
1410 * we can just use plain pread/pwrite without any problems.
1412 if (aiocb
->io
.niov
== 1) {
1413 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1417 * We have more than one iovec, and all are properly aligned.
1419 * Try preadv/pwritev first and fall back to linearizing the
1420 * buffer if it's not supported.
1422 if (preadv_present
) {
1423 nbytes
= handle_aiocb_rw_vector(aiocb
);
1424 if (nbytes
== aiocb
->aio_nbytes
||
1425 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1428 preadv_present
= false;
1432 * XXX(hch): short read/write. no easy way to handle the reminder
1433 * using these interfaces. For now retry using plain
1439 * Ok, we have to do it the hard way, copy all segments into
1440 * a single aligned buffer.
1442 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1448 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1452 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1453 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1454 p
+= aiocb
->io
.iov
[i
].iov_len
;
1456 assert(p
- buf
== aiocb
->aio_nbytes
);
1459 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1460 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1462 size_t count
= aiocb
->aio_nbytes
, copy
;
1465 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1467 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1468 copy
= aiocb
->io
.iov
[i
].iov_len
;
1470 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1471 assert(count
>= copy
);
1480 if (nbytes
== aiocb
->aio_nbytes
) {
1482 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1483 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1486 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1487 0, aiocb
->aio_nbytes
- nbytes
);
1496 static int translate_err(int err
)
1498 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1505 #ifdef CONFIG_FALLOCATE
1506 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1509 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1512 } while (errno
== EINTR
);
1513 return translate_err(-errno
);
1517 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1520 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1522 if (!s
->has_write_zeroes
) {
1527 /* The BLKZEROOUT implementation in the kernel doesn't set
1528 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1530 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1532 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1533 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1536 } while (errno
== EINTR
);
1538 ret
= translate_err(-errno
);
1539 if (ret
== -ENOTSUP
) {
1540 s
->has_write_zeroes
= false;
1548 static int handle_aiocb_write_zeroes(void *opaque
)
1550 RawPosixAIOData
*aiocb
= opaque
;
1551 #ifdef CONFIG_FALLOCATE
1552 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1556 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1557 return handle_aiocb_write_zeroes_block(aiocb
);
1560 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1561 if (s
->has_write_zeroes
) {
1562 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1563 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1564 if (ret
== -EINVAL
) {
1566 * Allow falling back to pwrite for file systems that
1567 * do not support fallocate() for an unaligned byte range.
1571 if (ret
== 0 || ret
!= -ENOTSUP
) {
1574 s
->has_write_zeroes
= false;
1578 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1579 if (s
->has_discard
&& s
->has_fallocate
) {
1580 int ret
= do_fallocate(s
->fd
,
1581 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1582 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1584 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1585 if (ret
== 0 || ret
!= -ENOTSUP
) {
1588 s
->has_fallocate
= false;
1589 } else if (ret
!= -ENOTSUP
) {
1592 s
->has_discard
= false;
1597 #ifdef CONFIG_FALLOCATE
1598 /* Last resort: we are trying to extend the file with zeroed data. This
1599 * can be done via fallocate(fd, 0) */
1600 len
= bdrv_getlength(aiocb
->bs
);
1601 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1602 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1603 if (ret
== 0 || ret
!= -ENOTSUP
) {
1606 s
->has_fallocate
= false;
1613 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1615 RawPosixAIOData
*aiocb
= opaque
;
1616 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1619 /* First try to write zeros and unmap at the same time */
1621 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1622 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1623 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1624 if (ret
!= -ENOTSUP
) {
1629 /* If we couldn't manage to unmap while guaranteed that the area reads as
1630 * all-zero afterwards, just write zeroes without unmapping */
1631 ret
= handle_aiocb_write_zeroes(aiocb
);
1635 #ifndef HAVE_COPY_FILE_RANGE
1636 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1637 off_t
*out_off
, size_t len
, unsigned int flags
)
1639 #ifdef __NR_copy_file_range
1640 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1641 out_off
, len
, flags
);
1649 static int handle_aiocb_copy_range(void *opaque
)
1651 RawPosixAIOData
*aiocb
= opaque
;
1652 uint64_t bytes
= aiocb
->aio_nbytes
;
1653 off_t in_off
= aiocb
->aio_offset
;
1654 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1657 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1658 aiocb
->copy_range
.aio_fd2
, &out_off
,
1660 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1661 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1664 /* No progress (e.g. when beyond EOF), let the caller fall back to
1683 static int handle_aiocb_discard(void *opaque
)
1685 RawPosixAIOData
*aiocb
= opaque
;
1686 int ret
= -EOPNOTSUPP
;
1687 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1689 if (!s
->has_discard
) {
1693 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1696 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1697 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1700 } while (errno
== EINTR
);
1705 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1706 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1707 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1711 ret
= translate_err(ret
);
1712 if (ret
== -ENOTSUP
) {
1713 s
->has_discard
= false;
1719 * Help alignment probing by allocating the first block.
1721 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1722 * reading succeeds regardless of request length. In this case we fallback to
1723 * safe alignment which is not optimal. Allocating the first block avoids this
1726 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1727 * request alignment, so we use safe values.
1729 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1730 * caller may ignore failures.
1732 static int allocate_first_block(int fd
, size_t max_size
)
1734 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
1737 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size
);
1742 buf
= qemu_memalign(max_align
, write_size
);
1743 memset(buf
, 0, write_size
);
1746 n
= pwrite(fd
, buf
, write_size
, 0);
1747 } while (n
== -1 && errno
== EINTR
);
1749 ret
= (n
== -1) ? -errno
: 0;
1755 static int handle_aiocb_truncate(void *opaque
)
1757 RawPosixAIOData
*aiocb
= opaque
;
1759 int64_t current_length
= 0;
1762 int fd
= aiocb
->aio_fildes
;
1763 int64_t offset
= aiocb
->aio_offset
;
1764 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1765 Error
**errp
= aiocb
->truncate
.errp
;
1767 if (fstat(fd
, &st
) < 0) {
1769 error_setg_errno(errp
, -result
, "Could not stat file");
1773 current_length
= st
.st_size
;
1774 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1775 error_setg(errp
, "Cannot use preallocation for shrinking files");
1780 #ifdef CONFIG_POSIX_FALLOCATE
1781 case PREALLOC_MODE_FALLOC
:
1783 * Truncating before posix_fallocate() makes it about twice slower on
1784 * file systems that do not support fallocate(), trying to check if a
1785 * block is allocated before allocating it, so don't do that here.
1787 if (offset
!= current_length
) {
1788 result
= -posix_fallocate(fd
, current_length
,
1789 offset
- current_length
);
1791 /* posix_fallocate() doesn't set errno. */
1792 error_setg_errno(errp
, -result
,
1793 "Could not preallocate new data");
1794 } else if (current_length
== 0) {
1796 * posix_fallocate() uses fallocate() if the filesystem
1797 * supports it, or fallback to manually writing zeroes. If
1798 * fallocate() was used, unaligned reads from the fallocated
1799 * area in raw_probe_alignment() will succeed, hence we need to
1800 * allocate the first block.
1802 * Optimize future alignment probing; ignore failures.
1804 allocate_first_block(fd
, offset
);
1811 case PREALLOC_MODE_FULL
:
1813 int64_t num
= 0, left
= offset
- current_length
;
1817 * Knowing the final size from the beginning could allow the file
1818 * system driver to do less allocations and possibly avoid
1819 * fragmentation of the file.
1821 if (ftruncate(fd
, offset
) != 0) {
1823 error_setg_errno(errp
, -result
, "Could not resize file");
1827 buf
= g_malloc0(65536);
1829 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1830 if (seek_result
< 0) {
1832 error_setg_errno(errp
, -result
,
1833 "Failed to seek to the old end of file");
1838 num
= MIN(left
, 65536);
1839 result
= write(fd
, buf
, num
);
1841 if (errno
== EINTR
) {
1845 error_setg_errno(errp
, -result
,
1846 "Could not write zeros for preallocation");
1855 error_setg_errno(errp
, -result
,
1856 "Could not flush file to disk");
1862 case PREALLOC_MODE_OFF
:
1863 if (ftruncate(fd
, offset
) != 0) {
1865 error_setg_errno(errp
, -result
, "Could not resize file");
1866 } else if (current_length
== 0 && offset
> current_length
) {
1867 /* Optimize future alignment probing; ignore failures. */
1868 allocate_first_block(fd
, offset
);
1873 error_setg(errp
, "Unsupported preallocation mode: %s",
1874 PreallocMode_str(prealloc
));
1880 if (ftruncate(fd
, current_length
) < 0) {
1881 error_report("Failed to restore old file length: %s",
1890 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1891 ThreadPoolFunc func
, void *arg
)
1893 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1894 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1895 return thread_pool_submit_co(pool
, func
, arg
);
1898 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1899 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1901 BDRVRawState
*s
= bs
->opaque
;
1902 RawPosixAIOData acb
;
1904 if (fd_open(bs
) < 0)
1908 * When using O_DIRECT, the request must be aligned to be able to use
1909 * either libaio or io_uring interface. If not fail back to regular thread
1910 * pool read/write code which emulates this for us if we
1911 * set QEMU_AIO_MISALIGNED.
1913 if (s
->needs_alignment
&& !bdrv_qiov_is_aligned(bs
, qiov
)) {
1914 type
|= QEMU_AIO_MISALIGNED
;
1915 #ifdef CONFIG_LINUX_IO_URING
1916 } else if (s
->use_linux_io_uring
) {
1917 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
1918 assert(qiov
->size
== bytes
);
1919 return luring_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1921 #ifdef CONFIG_LINUX_AIO
1922 } else if (s
->use_linux_aio
) {
1923 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1924 assert(qiov
->size
== bytes
);
1925 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1929 acb
= (RawPosixAIOData
) {
1931 .aio_fildes
= s
->fd
,
1933 .aio_offset
= offset
,
1934 .aio_nbytes
= bytes
,
1941 assert(qiov
->size
== bytes
);
1942 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1945 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1946 uint64_t bytes
, QEMUIOVector
*qiov
,
1949 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1952 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1953 uint64_t bytes
, QEMUIOVector
*qiov
,
1957 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1960 static void raw_aio_plug(BlockDriverState
*bs
)
1962 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
1963 #ifdef CONFIG_LINUX_AIO
1964 if (s
->use_linux_aio
) {
1965 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1966 laio_io_plug(bs
, aio
);
1969 #ifdef CONFIG_LINUX_IO_URING
1970 if (s
->use_linux_io_uring
) {
1971 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
1972 luring_io_plug(bs
, aio
);
1977 static void raw_aio_unplug(BlockDriverState
*bs
)
1979 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
1980 #ifdef CONFIG_LINUX_AIO
1981 if (s
->use_linux_aio
) {
1982 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1983 laio_io_unplug(bs
, aio
);
1986 #ifdef CONFIG_LINUX_IO_URING
1987 if (s
->use_linux_io_uring
) {
1988 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
1989 luring_io_unplug(bs
, aio
);
1994 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1996 BDRVRawState
*s
= bs
->opaque
;
1997 RawPosixAIOData acb
;
2005 acb
= (RawPosixAIOData
) {
2007 .aio_fildes
= s
->fd
,
2008 .aio_type
= QEMU_AIO_FLUSH
,
2011 #ifdef CONFIG_LINUX_IO_URING
2012 if (s
->use_linux_io_uring
) {
2013 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2014 return luring_co_submit(bs
, aio
, s
->fd
, 0, NULL
, QEMU_AIO_FLUSH
);
2017 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
2020 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
2021 AioContext
*new_context
)
2023 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2024 #ifdef CONFIG_LINUX_AIO
2025 if (s
->use_linux_aio
) {
2026 Error
*local_err
= NULL
;
2027 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
2028 error_reportf_err(local_err
, "Unable to use native AIO, "
2029 "falling back to thread pool: ");
2030 s
->use_linux_aio
= false;
2034 #ifdef CONFIG_LINUX_IO_URING
2035 if (s
->use_linux_io_uring
) {
2037 if (!aio_setup_linux_io_uring(new_context
, &local_err
)) {
2038 error_reportf_err(local_err
, "Unable to use linux io_uring, "
2039 "falling back to thread pool: ");
2040 s
->use_linux_io_uring
= false;
2046 static void raw_close(BlockDriverState
*bs
)
2048 BDRVRawState
*s
= bs
->opaque
;
2057 * Truncates the given regular file @fd to @offset and, when growing, fills the
2058 * new space according to @prealloc.
2060 * Returns: 0 on success, -errno on failure.
2062 static int coroutine_fn
2063 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2064 PreallocMode prealloc
, Error
**errp
)
2066 RawPosixAIOData acb
;
2068 acb
= (RawPosixAIOData
) {
2071 .aio_type
= QEMU_AIO_TRUNCATE
,
2072 .aio_offset
= offset
,
2074 .prealloc
= prealloc
,
2079 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2082 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2083 bool exact
, PreallocMode prealloc
,
2086 BDRVRawState
*s
= bs
->opaque
;
2090 if (fstat(s
->fd
, &st
)) {
2092 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2096 if (S_ISREG(st
.st_mode
)) {
2097 /* Always resizes to the exact @offset */
2098 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2101 if (prealloc
!= PREALLOC_MODE_OFF
) {
2102 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2103 "non-regular file", PreallocMode_str(prealloc
));
2107 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2108 int64_t cur_length
= raw_getlength(bs
);
2110 if (offset
!= cur_length
&& exact
) {
2111 error_setg(errp
, "Cannot resize device files");
2113 } else if (offset
> cur_length
) {
2114 error_setg(errp
, "Cannot grow device files");
2118 error_setg(errp
, "Resizing this file is not supported");
2126 static int64_t raw_getlength(BlockDriverState
*bs
)
2128 BDRVRawState
*s
= bs
->opaque
;
2134 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2135 struct disklabel dl
;
2137 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2139 return (uint64_t)dl
.d_secsize
*
2140 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2144 #elif defined(__NetBSD__)
2145 static int64_t raw_getlength(BlockDriverState
*bs
)
2147 BDRVRawState
*s
= bs
->opaque
;
2153 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2154 struct dkwedge_info dkw
;
2156 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2157 return dkw
.dkw_size
* 512;
2159 struct disklabel dl
;
2161 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2163 return (uint64_t)dl
.d_secsize
*
2164 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2169 #elif defined(__sun__)
2170 static int64_t raw_getlength(BlockDriverState
*bs
)
2172 BDRVRawState
*s
= bs
->opaque
;
2173 struct dk_minfo minfo
;
2183 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2185 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2187 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2191 * There are reports that lseek on some devices fails, but
2192 * irc discussion said that contingency on contingency was overkill.
2194 size
= lseek(s
->fd
, 0, SEEK_END
);
2200 #elif defined(CONFIG_BSD)
2201 static int64_t raw_getlength(BlockDriverState
*bs
)
2203 BDRVRawState
*s
= bs
->opaque
;
2207 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2216 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2219 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2220 #ifdef DIOCGMEDIASIZE
2221 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2222 #elif defined(DIOCGPART)
2225 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2226 size
= pi
.media_size
;
2232 #if defined(__APPLE__) && defined(__MACH__)
2234 uint64_t sectors
= 0;
2235 uint32_t sector_size
= 0;
2237 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2238 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2239 size
= sectors
* sector_size
;
2241 size
= lseek(fd
, 0LL, SEEK_END
);
2248 size
= lseek(fd
, 0LL, SEEK_END
);
2253 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2256 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2257 if (size
== 2048LL * (unsigned)-1)
2259 /* XXX no disc? maybe we need to reopen... */
2260 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2267 size
= lseek(fd
, 0, SEEK_END
);
2275 static int64_t raw_getlength(BlockDriverState
*bs
)
2277 BDRVRawState
*s
= bs
->opaque
;
2286 size
= lseek(s
->fd
, 0, SEEK_END
);
2294 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2297 BDRVRawState
*s
= bs
->opaque
;
2299 if (fstat(s
->fd
, &st
) < 0) {
2302 return (int64_t)st
.st_blocks
* 512;
2305 static int coroutine_fn
2306 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2308 BlockdevCreateOptionsFile
*file_opts
;
2309 Error
*local_err
= NULL
;
2311 uint64_t perm
, shared
;
2314 /* Validate options and set default values */
2315 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2316 file_opts
= &options
->u
.file
;
2318 if (!file_opts
->has_nocow
) {
2319 file_opts
->nocow
= false;
2321 if (!file_opts
->has_preallocation
) {
2322 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2326 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2329 error_setg_errno(errp
, -result
, "Could not create file");
2333 /* Take permissions: We want to discard everything, so we need
2334 * BLK_PERM_WRITE; and truncation to the desired size requires
2336 * On the other hand, we cannot share the RESIZE permission
2337 * because we promise that after this function, the file has the
2338 * size given in the options. If someone else were to resize it
2339 * concurrently, we could not guarantee that.
2340 * Note that after this function, we can no longer guarantee that
2341 * the file is not touched by a third party, so it may be resized
2343 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2344 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2346 /* Step one: Take locks */
2347 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2352 /* Step two: Check that nobody else has taken conflicting locks */
2353 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2355 error_append_hint(errp
,
2356 "Is another process using the image [%s]?\n",
2357 file_opts
->filename
);
2361 /* Clear the file by truncating it to 0 */
2362 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2367 if (file_opts
->nocow
) {
2369 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2370 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2371 * will be ignored since any failure of this operation should not
2372 * block the left work.
2375 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2376 attr
|= FS_NOCOW_FL
;
2377 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2382 /* Resize and potentially preallocate the file to the desired
2384 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2385 file_opts
->preallocation
, errp
);
2391 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2393 /* The above call should not fail, and if it does, that does
2394 * not mean the whole creation operation has failed. So
2395 * report it the user for their convenience, but do not report
2396 * it to the caller. */
2397 warn_report_err(local_err
);
2401 if (qemu_close(fd
) != 0 && result
== 0) {
2403 error_setg_errno(errp
, -result
, "Could not close the new file");
2409 static int coroutine_fn
raw_co_create_opts(const char *filename
, QemuOpts
*opts
,
2412 BlockdevCreateOptions options
;
2413 int64_t total_size
= 0;
2415 PreallocMode prealloc
;
2417 Error
*local_err
= NULL
;
2419 /* Skip file: protocol prefix */
2420 strstart(filename
, "file:", &filename
);
2422 /* Read out options */
2423 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2425 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2426 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2427 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2428 PREALLOC_MODE_OFF
, &local_err
);
2431 error_propagate(errp
, local_err
);
2435 options
= (BlockdevCreateOptions
) {
2436 .driver
= BLOCKDEV_DRIVER_FILE
,
2438 .filename
= (char *) filename
,
2440 .has_preallocation
= true,
2441 .preallocation
= prealloc
,
2446 return raw_co_create(&options
, errp
);
2450 * Find allocation range in @bs around offset @start.
2451 * May change underlying file descriptor's file offset.
2452 * If @start is not in a hole, store @start in @data, and the
2453 * beginning of the next hole in @hole, and return 0.
2454 * If @start is in a non-trailing hole, store @start in @hole and the
2455 * beginning of the next non-hole in @data, and return 0.
2456 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2457 * If we can't find out, return a negative errno other than -ENXIO.
2459 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2460 off_t
*data
, off_t
*hole
)
2462 #if defined SEEK_HOLE && defined SEEK_DATA
2463 BDRVRawState
*s
= bs
->opaque
;
2468 * D1. offs == start: start is in data
2469 * D2. offs > start: start is in a hole, next data at offs
2470 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2471 * or start is beyond EOF
2472 * If the latter happens, the file has been truncated behind
2473 * our back since we opened it. All bets are off then.
2474 * Treating like a trailing hole is simplest.
2475 * D4. offs < 0, errno != ENXIO: we learned nothing
2477 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2479 return -errno
; /* D3 or D4 */
2483 /* This is not a valid return by lseek(). We are safe to just return
2484 * -EIO in this case, and we'll treat it like D4. */
2489 /* D2: in hole, next data at offs */
2495 /* D1: in data, end not yet known */
2499 * H1. offs == start: start is in a hole
2500 * If this happens here, a hole has been dug behind our back
2501 * since the previous lseek().
2502 * H2. offs > start: either start is in data, next hole at offs,
2503 * or start is in trailing hole, EOF at offs
2504 * Linux treats trailing holes like any other hole: offs ==
2505 * start. Solaris seeks to EOF instead: offs > start (blech).
2506 * If that happens here, a hole has been dug behind our back
2507 * since the previous lseek().
2508 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2509 * If this happens, the file has been truncated behind our
2510 * back since we opened it. Treat it like a trailing hole.
2511 * H4. offs < 0, errno != ENXIO: we learned nothing
2512 * Pretend we know nothing at all, i.e. "forget" about D1.
2514 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2516 return -errno
; /* D1 and (H3 or H4) */
2520 /* This is not a valid return by lseek(). We are safe to just return
2521 * -EIO in this case, and we'll treat it like H4. */
2527 * D1 and H2: either in data, next hole at offs, or it was in
2528 * data but is now in a trailing hole. In the latter case,
2529 * all bets are off. Treating it as if it there was data all
2530 * the way to EOF is safe, so simply do that.
2545 * Returns the allocation status of the specified offset.
2547 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2549 * 'pnum' is set to the number of bytes (including and immediately following
2550 * the specified offset) that are known to be in the same
2551 * allocated/unallocated state.
2553 * 'bytes' is the max value 'pnum' should be set to.
2555 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2558 int64_t bytes
, int64_t *pnum
,
2560 BlockDriverState
**file
)
2562 off_t data
= 0, hole
= 0;
2565 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2576 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2579 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2580 if (ret
== -ENXIO
) {
2583 ret
= BDRV_BLOCK_ZERO
;
2584 } else if (ret
< 0) {
2585 /* No info available, so pretend there are no holes */
2587 ret
= BDRV_BLOCK_DATA
;
2588 } else if (data
== offset
) {
2589 /* On a data extent, compute bytes to the end of the extent,
2590 * possibly including a partial sector at EOF. */
2591 *pnum
= MIN(bytes
, hole
- offset
);
2594 * We are not allowed to return partial sectors, though, so
2595 * round up if necessary.
2597 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2598 int64_t file_length
= raw_getlength(bs
);
2599 if (file_length
> 0) {
2600 /* Ignore errors, this is just a safeguard */
2601 assert(hole
== file_length
);
2603 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2606 ret
= BDRV_BLOCK_DATA
;
2608 /* On a hole, compute bytes to the beginning of the next extent. */
2609 assert(hole
== offset
);
2610 *pnum
= MIN(bytes
, data
- offset
);
2611 ret
= BDRV_BLOCK_ZERO
;
2615 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2618 #if defined(__linux__)
2619 /* Verify that the file is not in the page cache */
2620 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2622 const size_t window_size
= 128 * 1024 * 1024;
2623 BDRVRawState
*s
= bs
->opaque
;
2624 void *window
= NULL
;
2631 /* mincore(2) page status information requires 1 byte per page */
2632 page_size
= sysconf(_SC_PAGESIZE
);
2633 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2635 end
= raw_getlength(bs
);
2637 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2644 /* Unmap previous window if size has changed */
2645 new_length
= MIN(end
- offset
, window_size
);
2646 if (new_length
!= length
) {
2647 munmap(window
, length
);
2652 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2654 if (new_window
== MAP_FAILED
) {
2655 error_setg_errno(errp
, errno
, "mmap failed");
2659 window
= new_window
;
2660 length
= new_length
;
2662 ret
= mincore(window
, length
, vec
);
2664 error_setg_errno(errp
, errno
, "mincore failed");
2668 vec_end
= DIV_ROUND_UP(length
, page_size
);
2669 for (i
= 0; i
< vec_end
; i
++) {
2671 error_setg(errp
, "page cache still in use!");
2678 munmap(window
, length
);
2683 #endif /* __linux__ */
2685 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2688 BDRVRawState
*s
= bs
->opaque
;
2693 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2697 if (!s
->drop_cache
) {
2701 if (s
->open_flags
& O_DIRECT
) {
2702 return; /* No host kernel page cache */
2705 #if defined(__linux__)
2706 /* This sets the scene for the next syscall... */
2707 ret
= bdrv_co_flush(bs
);
2709 error_setg_errno(errp
, -ret
, "flush failed");
2713 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2714 * process. These limitations are okay because we just fsynced the file,
2715 * we don't use mmap, and the file should not be in use by other processes.
2717 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2718 if (ret
!= 0) { /* the return value is a positive errno */
2719 error_setg_errno(errp
, ret
, "fadvise failed");
2723 if (s
->check_cache_dropped
) {
2724 check_cache_dropped(bs
, errp
);
2726 #else /* __linux__ */
2727 /* Do nothing. Live migration to a remote host with cache.direct=off is
2728 * unsupported on other host operating systems. Cache consistency issues
2729 * may occur but no error is reported here, partly because that's the
2730 * historical behavior and partly because it's hard to differentiate valid
2731 * configurations that should not cause errors.
2733 #endif /* !__linux__ */
2736 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
2739 s
->stats
.discard_nb_failed
++;
2741 s
->stats
.discard_nb_ok
++;
2742 s
->stats
.discard_bytes_ok
+= nbytes
;
2746 static coroutine_fn
int
2747 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2749 BDRVRawState
*s
= bs
->opaque
;
2750 RawPosixAIOData acb
;
2753 acb
= (RawPosixAIOData
) {
2755 .aio_fildes
= s
->fd
,
2756 .aio_type
= QEMU_AIO_DISCARD
,
2757 .aio_offset
= offset
,
2758 .aio_nbytes
= bytes
,
2762 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2765 ret
= raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2766 raw_account_discard(s
, bytes
, ret
);
2770 static coroutine_fn
int
2771 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2773 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2776 static int coroutine_fn
2777 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2778 BdrvRequestFlags flags
, bool blkdev
)
2780 BDRVRawState
*s
= bs
->opaque
;
2781 RawPosixAIOData acb
;
2782 ThreadPoolFunc
*handler
;
2784 #ifdef CONFIG_FALLOCATE
2785 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
2786 BdrvTrackedRequest
*req
;
2790 * This is a workaround for a bug in the Linux XFS driver,
2791 * where writes submitted through the AIO interface will be
2792 * discarded if they happen beyond a concurrently running
2793 * fallocate() that increases the file length (i.e., both the
2794 * write and the fallocate() happen beyond the EOF).
2796 * To work around it, we extend the tracked request for this
2797 * zero write until INT64_MAX (effectively infinity), and mark
2798 * it as serializing.
2800 * We have to enable this workaround for all filesystems and
2801 * AIO modes (not just XFS with aio=native), because for
2802 * remote filesystems we do not know the host configuration.
2805 req
= bdrv_co_get_self_request(bs
);
2807 assert(req
->type
== BDRV_TRACKED_WRITE
);
2808 assert(req
->offset
<= offset
);
2809 assert(req
->offset
+ req
->bytes
>= offset
+ bytes
);
2811 end
= INT64_MAX
& -(uint64_t)bs
->bl
.request_alignment
;
2812 req
->bytes
= end
- req
->offset
;
2813 req
->overlap_bytes
= req
->bytes
;
2815 bdrv_mark_request_serialising(req
, bs
->bl
.request_alignment
);
2819 acb
= (RawPosixAIOData
) {
2821 .aio_fildes
= s
->fd
,
2822 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2823 .aio_offset
= offset
,
2824 .aio_nbytes
= bytes
,
2828 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2830 if (flags
& BDRV_REQ_NO_FALLBACK
) {
2831 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
2834 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2835 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2836 handler
= handle_aiocb_write_zeroes_unmap
;
2838 handler
= handle_aiocb_write_zeroes
;
2841 return raw_thread_pool_submit(bs
, handler
, &acb
);
2844 static int coroutine_fn
raw_co_pwrite_zeroes(
2845 BlockDriverState
*bs
, int64_t offset
,
2846 int bytes
, BdrvRequestFlags flags
)
2848 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2851 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2853 BDRVRawState
*s
= bs
->opaque
;
2855 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2859 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
2861 BDRVRawState
*s
= bs
->opaque
;
2862 return (BlockStatsSpecificFile
) {
2863 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
2864 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
2865 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
2869 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
2871 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2873 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
2874 stats
->u
.file
= get_blockstats_specific_file(bs
);
2879 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
2881 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2883 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
2884 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
2889 static QemuOptsList raw_create_opts
= {
2890 .name
= "raw-create-opts",
2891 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2894 .name
= BLOCK_OPT_SIZE
,
2895 .type
= QEMU_OPT_SIZE
,
2896 .help
= "Virtual disk size"
2899 .name
= BLOCK_OPT_NOCOW
,
2900 .type
= QEMU_OPT_BOOL
,
2901 .help
= "Turn off copy-on-write (valid only on btrfs)"
2904 .name
= BLOCK_OPT_PREALLOC
,
2905 .type
= QEMU_OPT_STRING
,
2906 .help
= "Preallocation mode (allowed values: off"
2907 #ifdef CONFIG_POSIX_FALLOCATE
2912 { /* end of list */ }
2916 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2919 BDRVRawState
*s
= bs
->opaque
;
2920 BDRVRawReopenState
*rs
= NULL
;
2924 if (s
->perm_change_fd
) {
2926 * In the context of reopen, this function may be called several times
2927 * (directly and recursively while change permissions of the parent).
2928 * This is even true for children that don't inherit from the original
2929 * reopen node, so s->reopen_state is not set.
2931 * Ignore all but the first call.
2936 if (s
->reopen_state
) {
2937 /* We already have a new file descriptor to set permissions for */
2938 assert(s
->reopen_state
->perm
== perm
);
2939 assert(s
->reopen_state
->shared_perm
== shared
);
2940 rs
= s
->reopen_state
->opaque
;
2941 s
->perm_change_fd
= rs
->fd
;
2942 s
->perm_change_flags
= rs
->open_flags
;
2944 /* We may need a new fd if auto-read-only switches the mode */
2945 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
2949 } else if (ret
!= s
->fd
) {
2950 s
->perm_change_fd
= ret
;
2951 s
->perm_change_flags
= open_flags
;
2955 /* Prepare permissions on old fd to avoid conflicts between old and new,
2956 * but keep everything locked that new will need. */
2957 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2962 /* Copy locks to the new fd */
2963 if (s
->perm_change_fd
) {
2964 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
2967 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
2974 if (s
->perm_change_fd
&& !s
->reopen_state
) {
2975 qemu_close(s
->perm_change_fd
);
2977 s
->perm_change_fd
= 0;
2981 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
2983 BDRVRawState
*s
= bs
->opaque
;
2985 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
2986 * called after .bdrv_reopen_commit) */
2987 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
2989 s
->fd
= s
->perm_change_fd
;
2990 s
->open_flags
= s
->perm_change_flags
;
2992 s
->perm_change_fd
= 0;
2994 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
2996 s
->shared_perm
= shared
;
2999 static void raw_abort_perm_update(BlockDriverState
*bs
)
3001 BDRVRawState
*s
= bs
->opaque
;
3003 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3004 * the file descriptor. */
3005 if (s
->perm_change_fd
&& !s
->reopen_state
) {
3006 qemu_close(s
->perm_change_fd
);
3008 s
->perm_change_fd
= 0;
3010 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3013 static int coroutine_fn
raw_co_copy_range_from(
3014 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
3015 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
3016 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
3018 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3019 read_flags
, write_flags
);
3022 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
3024 uint64_t src_offset
,
3026 uint64_t dst_offset
,
3028 BdrvRequestFlags read_flags
,
3029 BdrvRequestFlags write_flags
)
3031 RawPosixAIOData acb
;
3032 BDRVRawState
*s
= bs
->opaque
;
3033 BDRVRawState
*src_s
;
3035 assert(dst
->bs
== bs
);
3036 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
3040 src_s
= src
->bs
->opaque
;
3041 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
3045 acb
= (RawPosixAIOData
) {
3047 .aio_type
= QEMU_AIO_COPY_RANGE
,
3048 .aio_fildes
= src_s
->fd
,
3049 .aio_offset
= src_offset
,
3050 .aio_nbytes
= bytes
,
3053 .aio_offset2
= dst_offset
,
3057 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
3060 BlockDriver bdrv_file
= {
3061 .format_name
= "file",
3062 .protocol_name
= "file",
3063 .instance_size
= sizeof(BDRVRawState
),
3064 .bdrv_needs_filename
= true,
3065 .bdrv_probe
= NULL
, /* no probe for protocols */
3066 .bdrv_parse_filename
= raw_parse_filename
,
3067 .bdrv_file_open
= raw_open
,
3068 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3069 .bdrv_reopen_commit
= raw_reopen_commit
,
3070 .bdrv_reopen_abort
= raw_reopen_abort
,
3071 .bdrv_close
= raw_close
,
3072 .bdrv_co_create
= raw_co_create
,
3073 .bdrv_co_create_opts
= raw_co_create_opts
,
3074 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3075 .bdrv_has_zero_init_truncate
= bdrv_has_zero_init_1
,
3076 .bdrv_co_block_status
= raw_co_block_status
,
3077 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3078 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
3080 .bdrv_co_preadv
= raw_co_preadv
,
3081 .bdrv_co_pwritev
= raw_co_pwritev
,
3082 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3083 .bdrv_co_pdiscard
= raw_co_pdiscard
,
3084 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3085 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3086 .bdrv_refresh_limits
= raw_refresh_limits
,
3087 .bdrv_io_plug
= raw_aio_plug
,
3088 .bdrv_io_unplug
= raw_aio_unplug
,
3089 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3091 .bdrv_co_truncate
= raw_co_truncate
,
3092 .bdrv_getlength
= raw_getlength
,
3093 .bdrv_get_info
= raw_get_info
,
3094 .bdrv_get_allocated_file_size
3095 = raw_get_allocated_file_size
,
3096 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3097 .bdrv_check_perm
= raw_check_perm
,
3098 .bdrv_set_perm
= raw_set_perm
,
3099 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3100 .create_opts
= &raw_create_opts
,
3101 .mutable_opts
= mutable_opts
,
3104 /***********************************************/
3107 #if defined(__APPLE__) && defined(__MACH__)
3108 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3109 CFIndex maxPathSize
, int flags
);
3110 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3112 kern_return_t kernResult
= KERN_FAILURE
;
3113 mach_port_t masterPort
;
3114 CFMutableDictionaryRef classesToMatch
;
3115 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3116 char *mediaType
= NULL
;
3118 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
3119 if ( KERN_SUCCESS
!= kernResult
) {
3120 printf( "IOMasterPort returned %d\n", kernResult
);
3124 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3125 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3126 if (classesToMatch
== NULL
) {
3127 error_report("IOServiceMatching returned NULL for %s",
3128 matching_array
[index
]);
3131 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3133 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
3135 if (kernResult
!= KERN_SUCCESS
) {
3136 error_report("Note: IOServiceGetMatchingServices returned %d",
3141 /* If a match was found, leave the loop */
3142 if (*mediaIterator
!= 0) {
3143 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3144 mediaType
= g_strdup(matching_array
[index
]);
3151 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3152 CFIndex maxPathSize
, int flags
)
3154 io_object_t nextMedia
;
3155 kern_return_t kernResult
= KERN_FAILURE
;
3157 nextMedia
= IOIteratorNext( mediaIterator
);
3160 CFTypeRef bsdPathAsCFString
;
3161 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3162 if ( bsdPathAsCFString
) {
3163 size_t devPathLength
;
3164 strcpy( bsdPath
, _PATH_DEV
);
3165 if (flags
& BDRV_O_NOCACHE
) {
3166 strcat(bsdPath
, "r");
3168 devPathLength
= strlen( bsdPath
);
3169 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3170 kernResult
= KERN_SUCCESS
;
3172 CFRelease( bsdPathAsCFString
);
3174 IOObjectRelease( nextMedia
);
3180 /* Sets up a real cdrom for use in QEMU */
3181 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3183 int index
, num_of_test_partitions
= 2, fd
;
3184 char test_partition
[MAXPATHLEN
];
3185 bool partition_found
= false;
3187 /* look for a working partition */
3188 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3189 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3191 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
3193 partition_found
= true;
3199 /* if a working partition on the device was not found */
3200 if (partition_found
== false) {
3201 error_setg(errp
, "Failed to find a working partition on disc");
3203 trace_file_setup_cdrom(test_partition
);
3204 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3206 return partition_found
;
3209 /* Prints directions on mounting and unmounting a device */
3210 static void print_unmounting_directions(const char *file_name
)
3212 error_report("If device %s is mounted on the desktop, unmount"
3213 " it first before using it in QEMU", file_name
);
3214 error_report("Command to unmount device: diskutil unmountDisk %s",
3216 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3219 #endif /* defined(__APPLE__) && defined(__MACH__) */
3221 static int hdev_probe_device(const char *filename
)
3225 /* allow a dedicated CD-ROM driver to match with a higher priority */
3226 if (strstart(filename
, "/dev/cdrom", NULL
))
3229 if (stat(filename
, &st
) >= 0 &&
3230 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3237 static int check_hdev_writable(BDRVRawState
*s
)
3239 #if defined(BLKROGET)
3240 /* Linux block devices can be configured "read-only" using blockdev(8).
3241 * This is independent of device node permissions and therefore open(2)
3242 * with O_RDWR succeeds. Actual writes fail with EPERM.
3244 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3245 * check for read-only block devices so that Linux block devices behave
3251 if (fstat(s
->fd
, &st
)) {
3255 if (!S_ISBLK(st
.st_mode
)) {
3259 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
3266 #endif /* defined(BLKROGET) */
3270 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3273 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3276 static bool hdev_is_sg(BlockDriverState
*bs
)
3279 #if defined(__linux__)
3281 BDRVRawState
*s
= bs
->opaque
;
3283 struct sg_scsi_id scsiid
;
3287 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3291 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3296 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3298 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3307 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3310 BDRVRawState
*s
= bs
->opaque
;
3311 Error
*local_err
= NULL
;
3314 #if defined(__APPLE__) && defined(__MACH__)
3316 * Caution: while qdict_get_str() is fine, getting non-string types
3317 * would require more care. When @options come from -blockdev or
3318 * blockdev_add, its members are typed according to the QAPI
3319 * schema, but when they come from -drive, they're all QString.
3321 const char *filename
= qdict_get_str(options
, "filename");
3322 char bsd_path
[MAXPATHLEN
] = "";
3323 bool error_occurred
= false;
3325 /* If using a real cdrom */
3326 if (strcmp(filename
, "/dev/cdrom") == 0) {
3327 char *mediaType
= NULL
;
3328 kern_return_t ret_val
;
3329 io_iterator_t mediaIterator
= 0;
3331 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3332 if (mediaType
== NULL
) {
3333 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3335 error_occurred
= true;
3336 goto hdev_open_Mac_error
;
3339 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3340 if (ret_val
!= KERN_SUCCESS
) {
3341 error_setg(errp
, "Could not get BSD path for optical drive");
3342 error_occurred
= true;
3343 goto hdev_open_Mac_error
;
3346 /* If a real optical drive was not found */
3347 if (bsd_path
[0] == '\0') {
3348 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3349 error_occurred
= true;
3350 goto hdev_open_Mac_error
;
3353 /* If using a cdrom disc and finding a partition on the disc failed */
3354 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3355 setup_cdrom(bsd_path
, errp
) == false) {
3356 print_unmounting_directions(bsd_path
);
3357 error_occurred
= true;
3358 goto hdev_open_Mac_error
;
3361 qdict_put_str(options
, "filename", bsd_path
);
3363 hdev_open_Mac_error
:
3365 if (mediaIterator
) {
3366 IOObjectRelease(mediaIterator
);
3368 if (error_occurred
) {
3372 #endif /* defined(__APPLE__) && defined(__MACH__) */
3374 s
->type
= FTYPE_FILE
;
3376 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3378 error_propagate(errp
, local_err
);
3379 #if defined(__APPLE__) && defined(__MACH__)
3381 filename
= bsd_path
;
3383 /* if a physical device experienced an error while being opened */
3384 if (strncmp(filename
, "/dev/", 5) == 0) {
3385 print_unmounting_directions(filename
);
3387 #endif /* defined(__APPLE__) && defined(__MACH__) */
3391 /* Since this does ioctl the device must be already opened */
3392 bs
->sg
= hdev_is_sg(bs
);
3394 if (flags
& BDRV_O_RDWR
) {
3395 ret
= check_hdev_writable(s
);
3398 error_setg_errno(errp
, -ret
, "The device is not writable");
3406 #if defined(__linux__)
3407 static int coroutine_fn
3408 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3410 BDRVRawState
*s
= bs
->opaque
;
3411 RawPosixAIOData acb
;
3419 if (req
== SG_IO
&& s
->pr_mgr
) {
3420 struct sg_io_hdr
*io_hdr
= buf
;
3421 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3422 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3423 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3428 acb
= (RawPosixAIOData
) {
3430 .aio_type
= QEMU_AIO_IOCTL
,
3431 .aio_fildes
= s
->fd
,
3439 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3443 static int fd_open(BlockDriverState
*bs
)
3445 BDRVRawState
*s
= bs
->opaque
;
3447 /* this is just to ensure s->fd is sane (its called by io ops) */
3453 static coroutine_fn
int
3454 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3456 BDRVRawState
*s
= bs
->opaque
;
3461 raw_account_discard(s
, bytes
, ret
);
3464 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3467 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3468 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3477 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3480 static int coroutine_fn
hdev_co_create_opts(const char *filename
, QemuOpts
*opts
,
3485 struct stat stat_buf
;
3486 int64_t total_size
= 0;
3489 /* This function is used by both protocol block drivers and therefore either
3490 * of these prefixes may be given.
3491 * The return value has to be stored somewhere, otherwise this is an error
3492 * due to -Werror=unused-value. */
3494 strstart(filename
, "host_device:", &filename
) ||
3495 strstart(filename
, "host_cdrom:" , &filename
);
3499 ret
= raw_normalize_devicepath(&filename
, errp
);
3504 /* Read out options */
3505 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
3508 fd
= qemu_open(filename
, O_WRONLY
| O_BINARY
);
3511 error_setg_errno(errp
, -ret
, "Could not open device");
3515 if (fstat(fd
, &stat_buf
) < 0) {
3517 error_setg_errno(errp
, -ret
, "Could not stat device");
3518 } else if (!S_ISBLK(stat_buf
.st_mode
) && !S_ISCHR(stat_buf
.st_mode
)) {
3520 "The given file is neither a block nor a character device");
3522 } else if (lseek(fd
, 0, SEEK_END
) < total_size
) {
3523 error_setg(errp
, "Device is too small");
3527 if (!ret
&& total_size
) {
3528 uint8_t buf
[BDRV_SECTOR_SIZE
] = { 0 };
3529 int64_t zero_size
= MIN(BDRV_SECTOR_SIZE
, total_size
);
3530 if (lseek(fd
, 0, SEEK_SET
) == -1) {
3533 ret
= qemu_write_full(fd
, buf
, zero_size
);
3534 ret
= ret
== zero_size
? 0 : -errno
;
3541 static BlockDriver bdrv_host_device
= {
3542 .format_name
= "host_device",
3543 .protocol_name
= "host_device",
3544 .instance_size
= sizeof(BDRVRawState
),
3545 .bdrv_needs_filename
= true,
3546 .bdrv_probe_device
= hdev_probe_device
,
3547 .bdrv_parse_filename
= hdev_parse_filename
,
3548 .bdrv_file_open
= hdev_open
,
3549 .bdrv_close
= raw_close
,
3550 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3551 .bdrv_reopen_commit
= raw_reopen_commit
,
3552 .bdrv_reopen_abort
= raw_reopen_abort
,
3553 .bdrv_co_create_opts
= hdev_co_create_opts
,
3554 .create_opts
= &raw_create_opts
,
3555 .mutable_opts
= mutable_opts
,
3556 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3557 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3559 .bdrv_co_preadv
= raw_co_preadv
,
3560 .bdrv_co_pwritev
= raw_co_pwritev
,
3561 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3562 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3563 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3564 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3565 .bdrv_refresh_limits
= raw_refresh_limits
,
3566 .bdrv_io_plug
= raw_aio_plug
,
3567 .bdrv_io_unplug
= raw_aio_unplug
,
3568 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3570 .bdrv_co_truncate
= raw_co_truncate
,
3571 .bdrv_getlength
= raw_getlength
,
3572 .bdrv_get_info
= raw_get_info
,
3573 .bdrv_get_allocated_file_size
3574 = raw_get_allocated_file_size
,
3575 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
3576 .bdrv_check_perm
= raw_check_perm
,
3577 .bdrv_set_perm
= raw_set_perm
,
3578 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3579 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3580 .bdrv_probe_geometry
= hdev_probe_geometry
,
3582 /* generic scsi device */
3584 .bdrv_co_ioctl
= hdev_co_ioctl
,
3588 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3589 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3592 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3597 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3600 BDRVRawState
*s
= bs
->opaque
;
3604 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3605 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3608 static int cdrom_probe_device(const char *filename
)
3614 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3618 ret
= fstat(fd
, &st
);
3619 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3623 /* Attempt to detect via a CDROM specific ioctl */
3624 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3634 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3636 BDRVRawState
*s
= bs
->opaque
;
3639 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3640 return ret
== CDS_DISC_OK
;
3643 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3645 BDRVRawState
*s
= bs
->opaque
;
3648 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3649 perror("CDROMEJECT");
3651 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3652 perror("CDROMEJECT");
3656 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3658 BDRVRawState
*s
= bs
->opaque
;
3660 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3662 * Note: an error can happen if the distribution automatically
3665 /* perror("CDROM_LOCKDOOR"); */
3669 static BlockDriver bdrv_host_cdrom
= {
3670 .format_name
= "host_cdrom",
3671 .protocol_name
= "host_cdrom",
3672 .instance_size
= sizeof(BDRVRawState
),
3673 .bdrv_needs_filename
= true,
3674 .bdrv_probe_device
= cdrom_probe_device
,
3675 .bdrv_parse_filename
= cdrom_parse_filename
,
3676 .bdrv_file_open
= cdrom_open
,
3677 .bdrv_close
= raw_close
,
3678 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3679 .bdrv_reopen_commit
= raw_reopen_commit
,
3680 .bdrv_reopen_abort
= raw_reopen_abort
,
3681 .bdrv_co_create_opts
= hdev_co_create_opts
,
3682 .create_opts
= &raw_create_opts
,
3683 .mutable_opts
= mutable_opts
,
3684 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3687 .bdrv_co_preadv
= raw_co_preadv
,
3688 .bdrv_co_pwritev
= raw_co_pwritev
,
3689 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3690 .bdrv_refresh_limits
= raw_refresh_limits
,
3691 .bdrv_io_plug
= raw_aio_plug
,
3692 .bdrv_io_unplug
= raw_aio_unplug
,
3693 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3695 .bdrv_co_truncate
= raw_co_truncate
,
3696 .bdrv_getlength
= raw_getlength
,
3697 .has_variable_length
= true,
3698 .bdrv_get_allocated_file_size
3699 = raw_get_allocated_file_size
,
3701 /* removable device support */
3702 .bdrv_is_inserted
= cdrom_is_inserted
,
3703 .bdrv_eject
= cdrom_eject
,
3704 .bdrv_lock_medium
= cdrom_lock_medium
,
3706 /* generic scsi device */
3707 .bdrv_co_ioctl
= hdev_co_ioctl
,
3709 #endif /* __linux__ */
3711 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3712 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3715 BDRVRawState
*s
= bs
->opaque
;
3716 Error
*local_err
= NULL
;
3721 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3723 error_propagate(errp
, local_err
);
3727 /* make sure the door isn't locked at this time */
3728 ioctl(s
->fd
, CDIOCALLOW
);
3732 static int cdrom_probe_device(const char *filename
)
3734 if (strstart(filename
, "/dev/cd", NULL
) ||
3735 strstart(filename
, "/dev/acd", NULL
))
3740 static int cdrom_reopen(BlockDriverState
*bs
)
3742 BDRVRawState
*s
= bs
->opaque
;
3746 * Force reread of possibly changed/newly loaded disc,
3747 * FreeBSD seems to not notice sometimes...
3751 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3758 /* make sure the door isn't locked at this time */
3759 ioctl(s
->fd
, CDIOCALLOW
);
3763 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3765 return raw_getlength(bs
) > 0;
3768 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3770 BDRVRawState
*s
= bs
->opaque
;
3775 (void) ioctl(s
->fd
, CDIOCALLOW
);
3778 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3779 perror("CDIOCEJECT");
3781 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3782 perror("CDIOCCLOSE");
3788 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3790 BDRVRawState
*s
= bs
->opaque
;
3794 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3796 * Note: an error can happen if the distribution automatically
3799 /* perror("CDROM_LOCKDOOR"); */
3803 static BlockDriver bdrv_host_cdrom
= {
3804 .format_name
= "host_cdrom",
3805 .protocol_name
= "host_cdrom",
3806 .instance_size
= sizeof(BDRVRawState
),
3807 .bdrv_needs_filename
= true,
3808 .bdrv_probe_device
= cdrom_probe_device
,
3809 .bdrv_parse_filename
= cdrom_parse_filename
,
3810 .bdrv_file_open
= cdrom_open
,
3811 .bdrv_close
= raw_close
,
3812 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3813 .bdrv_reopen_commit
= raw_reopen_commit
,
3814 .bdrv_reopen_abort
= raw_reopen_abort
,
3815 .bdrv_co_create_opts
= hdev_co_create_opts
,
3816 .create_opts
= &raw_create_opts
,
3817 .mutable_opts
= mutable_opts
,
3819 .bdrv_co_preadv
= raw_co_preadv
,
3820 .bdrv_co_pwritev
= raw_co_pwritev
,
3821 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3822 .bdrv_refresh_limits
= raw_refresh_limits
,
3823 .bdrv_io_plug
= raw_aio_plug
,
3824 .bdrv_io_unplug
= raw_aio_unplug
,
3825 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3827 .bdrv_co_truncate
= raw_co_truncate
,
3828 .bdrv_getlength
= raw_getlength
,
3829 .has_variable_length
= true,
3830 .bdrv_get_allocated_file_size
3831 = raw_get_allocated_file_size
,
3833 /* removable device support */
3834 .bdrv_is_inserted
= cdrom_is_inserted
,
3835 .bdrv_eject
= cdrom_eject
,
3836 .bdrv_lock_medium
= cdrom_lock_medium
,
3838 #endif /* __FreeBSD__ */
3840 static void bdrv_file_init(void)
3843 * Register all the drivers. Note that order is important, the driver
3844 * registered last will get probed first.
3846 bdrv_register(&bdrv_file
);
3847 bdrv_register(&bdrv_host_device
);
3849 bdrv_register(&bdrv_host_cdrom
);
3851 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3852 bdrv_register(&bdrv_host_cdrom
);
3856 block_init(bdrv_file_init
);