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
;
705 if (S_ISREG(st
.st_mode
)) {
706 /* When extending regular files, we get zeros from the OS */
707 bs
->supported_truncate_flags
= BDRV_REQ_ZERO_WRITE
;
711 if (filename
&& (bdrv_flags
& BDRV_O_TEMPORARY
)) {
718 static int raw_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
721 BDRVRawState
*s
= bs
->opaque
;
723 s
->type
= FTYPE_FILE
;
724 return raw_open_common(bs
, options
, flags
, 0, false, errp
);
733 #define PERM_FOREACH(i) \
734 for ((i) = 0; (1ULL << (i)) <= BLK_PERM_ALL; i++)
736 /* Lock bytes indicated by @perm_lock_bits and @shared_perm_lock_bits in the
737 * file; if @unlock == true, also unlock the unneeded bytes.
738 * @shared_perm_lock_bits is the mask of all permissions that are NOT shared.
740 static int raw_apply_lock_bytes(BDRVRawState
*s
, int fd
,
741 uint64_t perm_lock_bits
,
742 uint64_t shared_perm_lock_bits
,
743 bool unlock
, Error
**errp
)
747 uint64_t locked_perm
, locked_shared_perm
;
750 locked_perm
= s
->locked_perm
;
751 locked_shared_perm
= s
->locked_shared_perm
;
754 * We don't have the previous bits, just lock/unlock for each of the
758 locked_perm
= BLK_PERM_ALL
;
759 locked_shared_perm
= BLK_PERM_ALL
;
762 locked_shared_perm
= 0;
767 int off
= RAW_LOCK_PERM_BASE
+ i
;
768 uint64_t bit
= (1ULL << i
);
769 if ((perm_lock_bits
& bit
) && !(locked_perm
& bit
)) {
770 ret
= qemu_lock_fd(fd
, off
, 1, false);
772 error_setg(errp
, "Failed to lock byte %d", off
);
775 s
->locked_perm
|= bit
;
777 } else if (unlock
&& (locked_perm
& bit
) && !(perm_lock_bits
& bit
)) {
778 ret
= qemu_unlock_fd(fd
, off
, 1);
780 error_setg(errp
, "Failed to unlock byte %d", off
);
783 s
->locked_perm
&= ~bit
;
788 int off
= RAW_LOCK_SHARED_BASE
+ i
;
789 uint64_t bit
= (1ULL << i
);
790 if ((shared_perm_lock_bits
& bit
) && !(locked_shared_perm
& bit
)) {
791 ret
= qemu_lock_fd(fd
, off
, 1, false);
793 error_setg(errp
, "Failed to lock byte %d", off
);
796 s
->locked_shared_perm
|= bit
;
798 } else if (unlock
&& (locked_shared_perm
& bit
) &&
799 !(shared_perm_lock_bits
& bit
)) {
800 ret
= qemu_unlock_fd(fd
, off
, 1);
802 error_setg(errp
, "Failed to unlock byte %d", off
);
805 s
->locked_shared_perm
&= ~bit
;
812 /* Check "unshared" bytes implied by @perm and ~@shared_perm in the file. */
813 static int raw_check_lock_bytes(int fd
, uint64_t perm
, uint64_t shared_perm
,
820 int off
= RAW_LOCK_SHARED_BASE
+ i
;
821 uint64_t p
= 1ULL << i
;
823 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
825 char *perm_name
= bdrv_perm_names(p
);
827 "Failed to get \"%s\" lock",
835 int off
= RAW_LOCK_PERM_BASE
+ i
;
836 uint64_t p
= 1ULL << i
;
837 if (!(shared_perm
& p
)) {
838 ret
= qemu_lock_fd_test(fd
, off
, 1, true);
840 char *perm_name
= bdrv_perm_names(p
);
842 "Failed to get shared \"%s\" lock",
852 static int raw_handle_perm_lock(BlockDriverState
*bs
,
854 uint64_t new_perm
, uint64_t new_shared
,
857 BDRVRawState
*s
= bs
->opaque
;
859 Error
*local_err
= NULL
;
865 if (bdrv_get_flags(bs
) & BDRV_O_INACTIVE
) {
871 if ((s
->perm
| new_perm
) == s
->perm
&&
872 (s
->shared_perm
& new_shared
) == s
->shared_perm
)
875 * We are going to unlock bytes, it should not fail. If it fail due
876 * to some fs-dependent permission-unrelated reasons (which occurs
877 * sometimes on NFS and leads to abort in bdrv_replace_child) we
878 * can't prevent such errors by any check here. And we ignore them
879 * anyway in ABORT and COMMIT.
883 ret
= raw_apply_lock_bytes(s
, s
->fd
, s
->perm
| new_perm
,
884 ~s
->shared_perm
| ~new_shared
,
887 ret
= raw_check_lock_bytes(s
->fd
, new_perm
, new_shared
, errp
);
891 error_append_hint(errp
,
892 "Is another process using the image [%s]?\n",
895 /* fall through to unlock bytes. */
897 raw_apply_lock_bytes(s
, s
->fd
, s
->perm
, ~s
->shared_perm
,
900 /* Theoretically the above call only unlocks bytes and it cannot
901 * fail. Something weird happened, report it.
903 warn_report_err(local_err
);
907 raw_apply_lock_bytes(s
, s
->fd
, new_perm
, ~new_shared
,
910 /* Theoretically the above call only unlocks bytes and it cannot
911 * fail. Something weird happened, report it.
913 warn_report_err(local_err
);
920 static int raw_reconfigure_getfd(BlockDriverState
*bs
, int flags
,
921 int *open_flags
, uint64_t perm
, bool force_dup
,
924 BDRVRawState
*s
= bs
->opaque
;
927 bool has_writers
= perm
&
928 (BLK_PERM_WRITE
| BLK_PERM_WRITE_UNCHANGED
| BLK_PERM_RESIZE
);
929 int fcntl_flags
= O_APPEND
| O_NONBLOCK
;
931 fcntl_flags
|= O_NOATIME
;
935 if (s
->type
== FTYPE_CD
) {
936 *open_flags
|= O_NONBLOCK
;
939 raw_parse_flags(flags
, open_flags
, has_writers
);
942 /* Not all operating systems have O_ASYNC, and those that don't
943 * will not let us track the state into rs->open_flags (typically
944 * you achieve the same effect with an ioctl, for example I_SETSIG
945 * on Solaris). But we do not use O_ASYNC, so that's fine.
947 assert((s
->open_flags
& O_ASYNC
) == 0);
950 if (!force_dup
&& *open_flags
== s
->open_flags
) {
951 /* We're lucky, the existing fd is fine */
955 if ((*open_flags
& ~fcntl_flags
) == (s
->open_flags
& ~fcntl_flags
)) {
956 /* dup the original fd */
957 fd
= qemu_dup(s
->fd
);
959 ret
= fcntl_setfl(fd
, *open_flags
);
967 /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
969 const char *normalized_filename
= bs
->filename
;
970 ret
= raw_normalize_devicepath(&normalized_filename
, errp
);
972 assert(!(*open_flags
& O_CREAT
));
973 fd
= qemu_open(normalized_filename
, *open_flags
);
975 error_setg_errno(errp
, errno
, "Could not reopen file");
984 static int raw_reopen_prepare(BDRVReopenState
*state
,
985 BlockReopenQueue
*queue
, Error
**errp
)
988 BDRVRawReopenState
*rs
;
991 Error
*local_err
= NULL
;
993 assert(state
!= NULL
);
994 assert(state
->bs
!= NULL
);
996 s
= state
->bs
->opaque
;
998 state
->opaque
= g_new0(BDRVRawReopenState
, 1);
1001 /* Handle options changes */
1002 opts
= qemu_opts_create(&raw_runtime_opts
, NULL
, 0, &error_abort
);
1003 qemu_opts_absorb_qdict(opts
, state
->options
, &local_err
);
1005 error_propagate(errp
, local_err
);
1010 rs
->drop_cache
= qemu_opt_get_bool_del(opts
, "drop-cache", true);
1011 rs
->check_cache_dropped
=
1012 qemu_opt_get_bool_del(opts
, "x-check-cache-dropped", false);
1014 /* This driver's reopen function doesn't currently allow changing
1015 * other options, so let's put them back in the original QDict and
1016 * bdrv_reopen_prepare() will detect changes and complain. */
1017 qemu_opts_to_qdict(opts
, state
->options
);
1019 rs
->fd
= raw_reconfigure_getfd(state
->bs
, state
->flags
, &rs
->open_flags
,
1020 state
->perm
, true, &local_err
);
1022 error_propagate(errp
, local_err
);
1027 /* Fail already reopen_prepare() if we can't get a working O_DIRECT
1028 * alignment with the new fd. */
1030 raw_probe_alignment(state
->bs
, rs
->fd
, &local_err
);
1032 error_propagate(errp
, local_err
);
1038 s
->reopen_state
= state
;
1046 qemu_opts_del(opts
);
1050 static void raw_reopen_commit(BDRVReopenState
*state
)
1052 BDRVRawReopenState
*rs
= state
->opaque
;
1053 BDRVRawState
*s
= state
->bs
->opaque
;
1055 s
->drop_cache
= rs
->drop_cache
;
1056 s
->check_cache_dropped
= rs
->check_cache_dropped
;
1057 s
->open_flags
= rs
->open_flags
;
1062 g_free(state
->opaque
);
1063 state
->opaque
= NULL
;
1065 assert(s
->reopen_state
== state
);
1066 s
->reopen_state
= NULL
;
1070 static void raw_reopen_abort(BDRVReopenState
*state
)
1072 BDRVRawReopenState
*rs
= state
->opaque
;
1073 BDRVRawState
*s
= state
->bs
->opaque
;
1075 /* nothing to do if NULL, we didn't get far enough */
1084 g_free(state
->opaque
);
1085 state
->opaque
= NULL
;
1087 assert(s
->reopen_state
== state
);
1088 s
->reopen_state
= NULL
;
1091 static int sg_get_max_transfer_length(int fd
)
1096 if (ioctl(fd
, BLKSECTGET
, &max_bytes
) == 0) {
1106 static int sg_get_max_segments(int fd
)
1111 char *sysfspath
= NULL
;
1117 if (fstat(fd
, &st
)) {
1122 sysfspath
= g_strdup_printf("/sys/dev/block/%u:%u/queue/max_segments",
1123 major(st
.st_rdev
), minor(st
.st_rdev
));
1124 sysfd
= open(sysfspath
, O_RDONLY
);
1130 ret
= read(sysfd
, buf
, sizeof(buf
) - 1);
1131 } while (ret
== -1 && errno
== EINTR
);
1135 } else if (ret
== 0) {
1140 /* The file is ended with '\n', pass 'end' to accept that. */
1141 ret
= qemu_strtol(buf
, &end
, 10, &max_segments
);
1142 if (ret
== 0 && end
&& *end
== '\n') {
1157 static void raw_refresh_limits(BlockDriverState
*bs
, Error
**errp
)
1159 BDRVRawState
*s
= bs
->opaque
;
1162 int ret
= sg_get_max_transfer_length(s
->fd
);
1164 if (ret
> 0 && ret
<= BDRV_REQUEST_MAX_BYTES
) {
1165 bs
->bl
.max_transfer
= pow2floor(ret
);
1168 ret
= sg_get_max_segments(s
->fd
);
1170 bs
->bl
.max_transfer
= MIN(bs
->bl
.max_transfer
,
1171 ret
* qemu_real_host_page_size
);
1175 raw_probe_alignment(bs
, s
->fd
, errp
);
1176 bs
->bl
.min_mem_alignment
= s
->buf_align
;
1177 bs
->bl
.opt_mem_alignment
= MAX(s
->buf_align
, qemu_real_host_page_size
);
1180 static int check_for_dasd(int fd
)
1183 struct dasd_information2_t info
= {0};
1185 return ioctl(fd
, BIODASDINFO2
, &info
);
1192 * Try to get @bs's logical and physical block size.
1193 * On success, store them in @bsz and return zero.
1194 * On failure, return negative errno.
1196 static int hdev_probe_blocksizes(BlockDriverState
*bs
, BlockSizes
*bsz
)
1198 BDRVRawState
*s
= bs
->opaque
;
1201 /* If DASD, get blocksizes */
1202 if (check_for_dasd(s
->fd
) < 0) {
1205 ret
= probe_logical_blocksize(s
->fd
, &bsz
->log
);
1209 return probe_physical_blocksize(s
->fd
, &bsz
->phys
);
1213 * Try to get @bs's geometry: cyls, heads, sectors.
1214 * On success, store them in @geo and return 0.
1215 * On failure return -errno.
1216 * (Allows block driver to assign default geometry values that guest sees)
1219 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1221 BDRVRawState
*s
= bs
->opaque
;
1222 struct hd_geometry ioctl_geo
= {0};
1224 /* If DASD, get its geometry */
1225 if (check_for_dasd(s
->fd
) < 0) {
1228 if (ioctl(s
->fd
, HDIO_GETGEO
, &ioctl_geo
) < 0) {
1231 /* HDIO_GETGEO may return success even though geo contains zeros
1232 (e.g. certain multipath setups) */
1233 if (!ioctl_geo
.heads
|| !ioctl_geo
.sectors
|| !ioctl_geo
.cylinders
) {
1236 /* Do not return a geometry for partition */
1237 if (ioctl_geo
.start
!= 0) {
1240 geo
->heads
= ioctl_geo
.heads
;
1241 geo
->sectors
= ioctl_geo
.sectors
;
1242 geo
->cylinders
= ioctl_geo
.cylinders
;
1246 #else /* __linux__ */
1247 static int hdev_probe_geometry(BlockDriverState
*bs
, HDGeometry
*geo
)
1253 #if defined(__linux__)
1254 static int handle_aiocb_ioctl(void *opaque
)
1256 RawPosixAIOData
*aiocb
= opaque
;
1259 ret
= ioctl(aiocb
->aio_fildes
, aiocb
->ioctl
.cmd
, aiocb
->ioctl
.buf
);
1268 static int handle_aiocb_flush(void *opaque
)
1270 RawPosixAIOData
*aiocb
= opaque
;
1271 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1274 if (s
->page_cache_inconsistent
) {
1278 ret
= qemu_fdatasync(aiocb
->aio_fildes
);
1280 /* There is no clear definition of the semantics of a failing fsync(),
1281 * so we may have to assume the worst. The sad truth is that this
1282 * assumption is correct for Linux. Some pages are now probably marked
1283 * clean in the page cache even though they are inconsistent with the
1284 * on-disk contents. The next fdatasync() call would succeed, but no
1285 * further writeback attempt will be made. We can't get back to a state
1286 * in which we know what is on disk (we would have to rewrite
1287 * everything that was touched since the last fdatasync() at least), so
1288 * make bdrv_flush() fail permanently. Given that the behaviour isn't
1289 * really defined, I have little hope that other OSes are doing better.
1291 * Obviously, this doesn't affect O_DIRECT, which bypasses the page
1293 if ((s
->open_flags
& O_DIRECT
) == 0) {
1294 s
->page_cache_inconsistent
= true;
1301 #ifdef CONFIG_PREADV
1303 static bool preadv_present
= true;
1306 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1308 return preadv(fd
, iov
, nr_iov
, offset
);
1312 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1314 return pwritev(fd
, iov
, nr_iov
, offset
);
1319 static bool preadv_present
= false;
1322 qemu_preadv(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1328 qemu_pwritev(int fd
, const struct iovec
*iov
, int nr_iov
, off_t offset
)
1335 static ssize_t
handle_aiocb_rw_vector(RawPosixAIOData
*aiocb
)
1340 if (aiocb
->aio_type
& QEMU_AIO_WRITE
)
1341 len
= qemu_pwritev(aiocb
->aio_fildes
,
1346 len
= qemu_preadv(aiocb
->aio_fildes
,
1350 } while (len
== -1 && errno
== EINTR
);
1359 * Read/writes the data to/from a given linear buffer.
1361 * Returns the number of bytes handles or -errno in case of an error. Short
1362 * reads are only returned if the end of the file is reached.
1364 static ssize_t
handle_aiocb_rw_linear(RawPosixAIOData
*aiocb
, char *buf
)
1369 while (offset
< aiocb
->aio_nbytes
) {
1370 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1371 len
= pwrite(aiocb
->aio_fildes
,
1372 (const char *)buf
+ offset
,
1373 aiocb
->aio_nbytes
- offset
,
1374 aiocb
->aio_offset
+ offset
);
1376 len
= pread(aiocb
->aio_fildes
,
1378 aiocb
->aio_nbytes
- offset
,
1379 aiocb
->aio_offset
+ offset
);
1381 if (len
== -1 && errno
== EINTR
) {
1383 } else if (len
== -1 && errno
== EINVAL
&&
1384 (aiocb
->bs
->open_flags
& BDRV_O_NOCACHE
) &&
1385 !(aiocb
->aio_type
& QEMU_AIO_WRITE
) &&
1387 /* O_DIRECT pread() may fail with EINVAL when offset is unaligned
1388 * after a short read. Assume that O_DIRECT short reads only occur
1389 * at EOF. Therefore this is a short read, not an I/O error.
1392 } else if (len
== -1) {
1395 } else if (len
== 0) {
1404 static int handle_aiocb_rw(void *opaque
)
1406 RawPosixAIOData
*aiocb
= opaque
;
1410 if (!(aiocb
->aio_type
& QEMU_AIO_MISALIGNED
)) {
1412 * If there is just a single buffer, and it is properly aligned
1413 * we can just use plain pread/pwrite without any problems.
1415 if (aiocb
->io
.niov
== 1) {
1416 nbytes
= handle_aiocb_rw_linear(aiocb
, aiocb
->io
.iov
->iov_base
);
1420 * We have more than one iovec, and all are properly aligned.
1422 * Try preadv/pwritev first and fall back to linearizing the
1423 * buffer if it's not supported.
1425 if (preadv_present
) {
1426 nbytes
= handle_aiocb_rw_vector(aiocb
);
1427 if (nbytes
== aiocb
->aio_nbytes
||
1428 (nbytes
< 0 && nbytes
!= -ENOSYS
)) {
1431 preadv_present
= false;
1435 * XXX(hch): short read/write. no easy way to handle the reminder
1436 * using these interfaces. For now retry using plain
1442 * Ok, we have to do it the hard way, copy all segments into
1443 * a single aligned buffer.
1445 buf
= qemu_try_blockalign(aiocb
->bs
, aiocb
->aio_nbytes
);
1451 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1455 for (i
= 0; i
< aiocb
->io
.niov
; ++i
) {
1456 memcpy(p
, aiocb
->io
.iov
[i
].iov_base
, aiocb
->io
.iov
[i
].iov_len
);
1457 p
+= aiocb
->io
.iov
[i
].iov_len
;
1459 assert(p
- buf
== aiocb
->aio_nbytes
);
1462 nbytes
= handle_aiocb_rw_linear(aiocb
, buf
);
1463 if (!(aiocb
->aio_type
& QEMU_AIO_WRITE
)) {
1465 size_t count
= aiocb
->aio_nbytes
, copy
;
1468 for (i
= 0; i
< aiocb
->io
.niov
&& count
; ++i
) {
1470 if (copy
> aiocb
->io
.iov
[i
].iov_len
) {
1471 copy
= aiocb
->io
.iov
[i
].iov_len
;
1473 memcpy(aiocb
->io
.iov
[i
].iov_base
, p
, copy
);
1474 assert(count
>= copy
);
1483 if (nbytes
== aiocb
->aio_nbytes
) {
1485 } else if (nbytes
>= 0 && nbytes
< aiocb
->aio_nbytes
) {
1486 if (aiocb
->aio_type
& QEMU_AIO_WRITE
) {
1489 iov_memset(aiocb
->io
.iov
, aiocb
->io
.niov
, nbytes
,
1490 0, aiocb
->aio_nbytes
- nbytes
);
1499 static int translate_err(int err
)
1501 if (err
== -ENODEV
|| err
== -ENOSYS
|| err
== -EOPNOTSUPP
||
1508 #ifdef CONFIG_FALLOCATE
1509 static int do_fallocate(int fd
, int mode
, off_t offset
, off_t len
)
1512 if (fallocate(fd
, mode
, offset
, len
) == 0) {
1515 } while (errno
== EINTR
);
1516 return translate_err(-errno
);
1520 static ssize_t
handle_aiocb_write_zeroes_block(RawPosixAIOData
*aiocb
)
1523 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1525 if (!s
->has_write_zeroes
) {
1530 /* The BLKZEROOUT implementation in the kernel doesn't set
1531 * BLKDEV_ZERO_NOFALLBACK, so we can't call this if we have to avoid slow
1533 if (!(aiocb
->aio_type
& QEMU_AIO_NO_FALLBACK
)) {
1535 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1536 if (ioctl(aiocb
->aio_fildes
, BLKZEROOUT
, range
) == 0) {
1539 } while (errno
== EINTR
);
1541 ret
= translate_err(-errno
);
1542 if (ret
== -ENOTSUP
) {
1543 s
->has_write_zeroes
= false;
1551 static int handle_aiocb_write_zeroes(void *opaque
)
1553 RawPosixAIOData
*aiocb
= opaque
;
1554 #ifdef CONFIG_FALLOCATE
1555 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1559 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1560 return handle_aiocb_write_zeroes_block(aiocb
);
1563 #ifdef CONFIG_FALLOCATE_ZERO_RANGE
1564 if (s
->has_write_zeroes
) {
1565 int ret
= do_fallocate(s
->fd
, FALLOC_FL_ZERO_RANGE
,
1566 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1567 if (ret
== -EINVAL
) {
1569 * Allow falling back to pwrite for file systems that
1570 * do not support fallocate() for an unaligned byte range.
1574 if (ret
== 0 || ret
!= -ENOTSUP
) {
1577 s
->has_write_zeroes
= false;
1581 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1582 if (s
->has_discard
&& s
->has_fallocate
) {
1583 int ret
= do_fallocate(s
->fd
,
1584 FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1585 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1587 ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1588 if (ret
== 0 || ret
!= -ENOTSUP
) {
1591 s
->has_fallocate
= false;
1592 } else if (ret
!= -ENOTSUP
) {
1595 s
->has_discard
= false;
1600 #ifdef CONFIG_FALLOCATE
1601 /* Last resort: we are trying to extend the file with zeroed data. This
1602 * can be done via fallocate(fd, 0) */
1603 len
= bdrv_getlength(aiocb
->bs
);
1604 if (s
->has_fallocate
&& len
>= 0 && aiocb
->aio_offset
>= len
) {
1605 int ret
= do_fallocate(s
->fd
, 0, aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1606 if (ret
== 0 || ret
!= -ENOTSUP
) {
1609 s
->has_fallocate
= false;
1616 static int handle_aiocb_write_zeroes_unmap(void *opaque
)
1618 RawPosixAIOData
*aiocb
= opaque
;
1619 BDRVRawState
*s G_GNUC_UNUSED
= aiocb
->bs
->opaque
;
1622 /* First try to write zeros and unmap at the same time */
1624 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1625 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1626 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1627 if (ret
!= -ENOTSUP
) {
1632 /* If we couldn't manage to unmap while guaranteed that the area reads as
1633 * all-zero afterwards, just write zeroes without unmapping */
1634 ret
= handle_aiocb_write_zeroes(aiocb
);
1638 #ifndef HAVE_COPY_FILE_RANGE
1639 static off_t
copy_file_range(int in_fd
, off_t
*in_off
, int out_fd
,
1640 off_t
*out_off
, size_t len
, unsigned int flags
)
1642 #ifdef __NR_copy_file_range
1643 return syscall(__NR_copy_file_range
, in_fd
, in_off
, out_fd
,
1644 out_off
, len
, flags
);
1652 static int handle_aiocb_copy_range(void *opaque
)
1654 RawPosixAIOData
*aiocb
= opaque
;
1655 uint64_t bytes
= aiocb
->aio_nbytes
;
1656 off_t in_off
= aiocb
->aio_offset
;
1657 off_t out_off
= aiocb
->copy_range
.aio_offset2
;
1660 ssize_t ret
= copy_file_range(aiocb
->aio_fildes
, &in_off
,
1661 aiocb
->copy_range
.aio_fd2
, &out_off
,
1663 trace_file_copy_file_range(aiocb
->bs
, aiocb
->aio_fildes
, in_off
,
1664 aiocb
->copy_range
.aio_fd2
, out_off
, bytes
,
1667 /* No progress (e.g. when beyond EOF), let the caller fall back to
1686 static int handle_aiocb_discard(void *opaque
)
1688 RawPosixAIOData
*aiocb
= opaque
;
1689 int ret
= -EOPNOTSUPP
;
1690 BDRVRawState
*s
= aiocb
->bs
->opaque
;
1692 if (!s
->has_discard
) {
1696 if (aiocb
->aio_type
& QEMU_AIO_BLKDEV
) {
1699 uint64_t range
[2] = { aiocb
->aio_offset
, aiocb
->aio_nbytes
};
1700 if (ioctl(aiocb
->aio_fildes
, BLKDISCARD
, range
) == 0) {
1703 } while (errno
== EINTR
);
1708 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
1709 ret
= do_fallocate(s
->fd
, FALLOC_FL_PUNCH_HOLE
| FALLOC_FL_KEEP_SIZE
,
1710 aiocb
->aio_offset
, aiocb
->aio_nbytes
);
1714 ret
= translate_err(ret
);
1715 if (ret
== -ENOTSUP
) {
1716 s
->has_discard
= false;
1722 * Help alignment probing by allocating the first block.
1724 * When reading with direct I/O from unallocated area on Gluster backed by XFS,
1725 * reading succeeds regardless of request length. In this case we fallback to
1726 * safe alignment which is not optimal. Allocating the first block avoids this
1729 * fd may be opened with O_DIRECT, but we don't know the buffer alignment or
1730 * request alignment, so we use safe values.
1732 * Returns: 0 on success, -errno on failure. Since this is an optimization,
1733 * caller may ignore failures.
1735 static int allocate_first_block(int fd
, size_t max_size
)
1737 size_t write_size
= (max_size
< MAX_BLOCKSIZE
)
1740 size_t max_align
= MAX(MAX_BLOCKSIZE
, qemu_real_host_page_size
);
1745 buf
= qemu_memalign(max_align
, write_size
);
1746 memset(buf
, 0, write_size
);
1749 n
= pwrite(fd
, buf
, write_size
, 0);
1750 } while (n
== -1 && errno
== EINTR
);
1752 ret
= (n
== -1) ? -errno
: 0;
1758 static int handle_aiocb_truncate(void *opaque
)
1760 RawPosixAIOData
*aiocb
= opaque
;
1762 int64_t current_length
= 0;
1765 int fd
= aiocb
->aio_fildes
;
1766 int64_t offset
= aiocb
->aio_offset
;
1767 PreallocMode prealloc
= aiocb
->truncate
.prealloc
;
1768 Error
**errp
= aiocb
->truncate
.errp
;
1770 if (fstat(fd
, &st
) < 0) {
1772 error_setg_errno(errp
, -result
, "Could not stat file");
1776 current_length
= st
.st_size
;
1777 if (current_length
> offset
&& prealloc
!= PREALLOC_MODE_OFF
) {
1778 error_setg(errp
, "Cannot use preallocation for shrinking files");
1783 #ifdef CONFIG_POSIX_FALLOCATE
1784 case PREALLOC_MODE_FALLOC
:
1786 * Truncating before posix_fallocate() makes it about twice slower on
1787 * file systems that do not support fallocate(), trying to check if a
1788 * block is allocated before allocating it, so don't do that here.
1790 if (offset
!= current_length
) {
1791 result
= -posix_fallocate(fd
, current_length
,
1792 offset
- current_length
);
1794 /* posix_fallocate() doesn't set errno. */
1795 error_setg_errno(errp
, -result
,
1796 "Could not preallocate new data");
1797 } else if (current_length
== 0) {
1799 * posix_fallocate() uses fallocate() if the filesystem
1800 * supports it, or fallback to manually writing zeroes. If
1801 * fallocate() was used, unaligned reads from the fallocated
1802 * area in raw_probe_alignment() will succeed, hence we need to
1803 * allocate the first block.
1805 * Optimize future alignment probing; ignore failures.
1807 allocate_first_block(fd
, offset
);
1814 case PREALLOC_MODE_FULL
:
1816 int64_t num
= 0, left
= offset
- current_length
;
1820 * Knowing the final size from the beginning could allow the file
1821 * system driver to do less allocations and possibly avoid
1822 * fragmentation of the file.
1824 if (ftruncate(fd
, offset
) != 0) {
1826 error_setg_errno(errp
, -result
, "Could not resize file");
1830 buf
= g_malloc0(65536);
1832 seek_result
= lseek(fd
, current_length
, SEEK_SET
);
1833 if (seek_result
< 0) {
1835 error_setg_errno(errp
, -result
,
1836 "Failed to seek to the old end of file");
1841 num
= MIN(left
, 65536);
1842 result
= write(fd
, buf
, num
);
1844 if (errno
== EINTR
) {
1848 error_setg_errno(errp
, -result
,
1849 "Could not write zeros for preallocation");
1858 error_setg_errno(errp
, -result
,
1859 "Could not flush file to disk");
1865 case PREALLOC_MODE_OFF
:
1866 if (ftruncate(fd
, offset
) != 0) {
1868 error_setg_errno(errp
, -result
, "Could not resize file");
1869 } else if (current_length
== 0 && offset
> current_length
) {
1870 /* Optimize future alignment probing; ignore failures. */
1871 allocate_first_block(fd
, offset
);
1876 error_setg(errp
, "Unsupported preallocation mode: %s",
1877 PreallocMode_str(prealloc
));
1883 if (ftruncate(fd
, current_length
) < 0) {
1884 error_report("Failed to restore old file length: %s",
1893 static int coroutine_fn
raw_thread_pool_submit(BlockDriverState
*bs
,
1894 ThreadPoolFunc func
, void *arg
)
1896 /* @bs can be NULL, bdrv_get_aio_context() returns the main context then */
1897 ThreadPool
*pool
= aio_get_thread_pool(bdrv_get_aio_context(bs
));
1898 return thread_pool_submit_co(pool
, func
, arg
);
1901 static int coroutine_fn
raw_co_prw(BlockDriverState
*bs
, uint64_t offset
,
1902 uint64_t bytes
, QEMUIOVector
*qiov
, int type
)
1904 BDRVRawState
*s
= bs
->opaque
;
1905 RawPosixAIOData acb
;
1907 if (fd_open(bs
) < 0)
1911 * When using O_DIRECT, the request must be aligned to be able to use
1912 * either libaio or io_uring interface. If not fail back to regular thread
1913 * pool read/write code which emulates this for us if we
1914 * set QEMU_AIO_MISALIGNED.
1916 if (s
->needs_alignment
&& !bdrv_qiov_is_aligned(bs
, qiov
)) {
1917 type
|= QEMU_AIO_MISALIGNED
;
1918 #ifdef CONFIG_LINUX_IO_URING
1919 } else if (s
->use_linux_io_uring
) {
1920 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
1921 assert(qiov
->size
== bytes
);
1922 return luring_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1924 #ifdef CONFIG_LINUX_AIO
1925 } else if (s
->use_linux_aio
) {
1926 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1927 assert(qiov
->size
== bytes
);
1928 return laio_co_submit(bs
, aio
, s
->fd
, offset
, qiov
, type
);
1932 acb
= (RawPosixAIOData
) {
1934 .aio_fildes
= s
->fd
,
1936 .aio_offset
= offset
,
1937 .aio_nbytes
= bytes
,
1944 assert(qiov
->size
== bytes
);
1945 return raw_thread_pool_submit(bs
, handle_aiocb_rw
, &acb
);
1948 static int coroutine_fn
raw_co_preadv(BlockDriverState
*bs
, uint64_t offset
,
1949 uint64_t bytes
, QEMUIOVector
*qiov
,
1952 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_READ
);
1955 static int coroutine_fn
raw_co_pwritev(BlockDriverState
*bs
, uint64_t offset
,
1956 uint64_t bytes
, QEMUIOVector
*qiov
,
1960 return raw_co_prw(bs
, offset
, bytes
, qiov
, QEMU_AIO_WRITE
);
1963 static void raw_aio_plug(BlockDriverState
*bs
)
1965 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
1966 #ifdef CONFIG_LINUX_AIO
1967 if (s
->use_linux_aio
) {
1968 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1969 laio_io_plug(bs
, aio
);
1972 #ifdef CONFIG_LINUX_IO_URING
1973 if (s
->use_linux_io_uring
) {
1974 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
1975 luring_io_plug(bs
, aio
);
1980 static void raw_aio_unplug(BlockDriverState
*bs
)
1982 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
1983 #ifdef CONFIG_LINUX_AIO
1984 if (s
->use_linux_aio
) {
1985 LinuxAioState
*aio
= aio_get_linux_aio(bdrv_get_aio_context(bs
));
1986 laio_io_unplug(bs
, aio
);
1989 #ifdef CONFIG_LINUX_IO_URING
1990 if (s
->use_linux_io_uring
) {
1991 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
1992 luring_io_unplug(bs
, aio
);
1997 static int raw_co_flush_to_disk(BlockDriverState
*bs
)
1999 BDRVRawState
*s
= bs
->opaque
;
2000 RawPosixAIOData acb
;
2008 acb
= (RawPosixAIOData
) {
2010 .aio_fildes
= s
->fd
,
2011 .aio_type
= QEMU_AIO_FLUSH
,
2014 #ifdef CONFIG_LINUX_IO_URING
2015 if (s
->use_linux_io_uring
) {
2016 LuringState
*aio
= aio_get_linux_io_uring(bdrv_get_aio_context(bs
));
2017 return luring_co_submit(bs
, aio
, s
->fd
, 0, NULL
, QEMU_AIO_FLUSH
);
2020 return raw_thread_pool_submit(bs
, handle_aiocb_flush
, &acb
);
2023 static void raw_aio_attach_aio_context(BlockDriverState
*bs
,
2024 AioContext
*new_context
)
2026 BDRVRawState
__attribute__((unused
)) *s
= bs
->opaque
;
2027 #ifdef CONFIG_LINUX_AIO
2028 if (s
->use_linux_aio
) {
2029 Error
*local_err
= NULL
;
2030 if (!aio_setup_linux_aio(new_context
, &local_err
)) {
2031 error_reportf_err(local_err
, "Unable to use native AIO, "
2032 "falling back to thread pool: ");
2033 s
->use_linux_aio
= false;
2037 #ifdef CONFIG_LINUX_IO_URING
2038 if (s
->use_linux_io_uring
) {
2040 if (!aio_setup_linux_io_uring(new_context
, &local_err
)) {
2041 error_reportf_err(local_err
, "Unable to use linux io_uring, "
2042 "falling back to thread pool: ");
2043 s
->use_linux_io_uring
= false;
2049 static void raw_close(BlockDriverState
*bs
)
2051 BDRVRawState
*s
= bs
->opaque
;
2060 * Truncates the given regular file @fd to @offset and, when growing, fills the
2061 * new space according to @prealloc.
2063 * Returns: 0 on success, -errno on failure.
2065 static int coroutine_fn
2066 raw_regular_truncate(BlockDriverState
*bs
, int fd
, int64_t offset
,
2067 PreallocMode prealloc
, Error
**errp
)
2069 RawPosixAIOData acb
;
2071 acb
= (RawPosixAIOData
) {
2074 .aio_type
= QEMU_AIO_TRUNCATE
,
2075 .aio_offset
= offset
,
2077 .prealloc
= prealloc
,
2082 return raw_thread_pool_submit(bs
, handle_aiocb_truncate
, &acb
);
2085 static int coroutine_fn
raw_co_truncate(BlockDriverState
*bs
, int64_t offset
,
2086 bool exact
, PreallocMode prealloc
,
2087 BdrvRequestFlags flags
, Error
**errp
)
2089 BDRVRawState
*s
= bs
->opaque
;
2093 if (fstat(s
->fd
, &st
)) {
2095 error_setg_errno(errp
, -ret
, "Failed to fstat() the file");
2099 if (S_ISREG(st
.st_mode
)) {
2100 /* Always resizes to the exact @offset */
2101 return raw_regular_truncate(bs
, s
->fd
, offset
, prealloc
, errp
);
2104 if (prealloc
!= PREALLOC_MODE_OFF
) {
2105 error_setg(errp
, "Preallocation mode '%s' unsupported for this "
2106 "non-regular file", PreallocMode_str(prealloc
));
2110 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2111 int64_t cur_length
= raw_getlength(bs
);
2113 if (offset
!= cur_length
&& exact
) {
2114 error_setg(errp
, "Cannot resize device files");
2116 } else if (offset
> cur_length
) {
2117 error_setg(errp
, "Cannot grow device files");
2121 error_setg(errp
, "Resizing this file is not supported");
2129 static int64_t raw_getlength(BlockDriverState
*bs
)
2131 BDRVRawState
*s
= bs
->opaque
;
2137 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2138 struct disklabel dl
;
2140 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2142 return (uint64_t)dl
.d_secsize
*
2143 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2147 #elif defined(__NetBSD__)
2148 static int64_t raw_getlength(BlockDriverState
*bs
)
2150 BDRVRawState
*s
= bs
->opaque
;
2156 if (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
)) {
2157 struct dkwedge_info dkw
;
2159 if (ioctl(fd
, DIOCGWEDGEINFO
, &dkw
) != -1) {
2160 return dkw
.dkw_size
* 512;
2162 struct disklabel dl
;
2164 if (ioctl(fd
, DIOCGDINFO
, &dl
))
2166 return (uint64_t)dl
.d_secsize
*
2167 dl
.d_partitions
[DISKPART(st
.st_rdev
)].p_size
;
2172 #elif defined(__sun__)
2173 static int64_t raw_getlength(BlockDriverState
*bs
)
2175 BDRVRawState
*s
= bs
->opaque
;
2176 struct dk_minfo minfo
;
2186 * Use the DKIOCGMEDIAINFO ioctl to read the size.
2188 ret
= ioctl(s
->fd
, DKIOCGMEDIAINFO
, &minfo
);
2190 return minfo
.dki_lbsize
* minfo
.dki_capacity
;
2194 * There are reports that lseek on some devices fails, but
2195 * irc discussion said that contingency on contingency was overkill.
2197 size
= lseek(s
->fd
, 0, SEEK_END
);
2203 #elif defined(CONFIG_BSD)
2204 static int64_t raw_getlength(BlockDriverState
*bs
)
2206 BDRVRawState
*s
= bs
->opaque
;
2210 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2219 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
2222 if (!fstat(fd
, &sb
) && (S_IFCHR
& sb
.st_mode
)) {
2223 #ifdef DIOCGMEDIASIZE
2224 if (ioctl(fd
, DIOCGMEDIASIZE
, (off_t
*)&size
))
2225 #elif defined(DIOCGPART)
2228 if (ioctl(fd
, DIOCGPART
, &pi
) == 0)
2229 size
= pi
.media_size
;
2235 #if defined(__APPLE__) && defined(__MACH__)
2237 uint64_t sectors
= 0;
2238 uint32_t sector_size
= 0;
2240 if (ioctl(fd
, DKIOCGETBLOCKCOUNT
, §ors
) == 0
2241 && ioctl(fd
, DKIOCGETBLOCKSIZE
, §or_size
) == 0) {
2242 size
= sectors
* sector_size
;
2244 size
= lseek(fd
, 0LL, SEEK_END
);
2251 size
= lseek(fd
, 0LL, SEEK_END
);
2256 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
2259 /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
2260 if (size
== 2048LL * (unsigned)-1)
2262 /* XXX no disc? maybe we need to reopen... */
2263 if (size
<= 0 && !reopened
&& cdrom_reopen(bs
) >= 0) {
2270 size
= lseek(fd
, 0, SEEK_END
);
2278 static int64_t raw_getlength(BlockDriverState
*bs
)
2280 BDRVRawState
*s
= bs
->opaque
;
2289 size
= lseek(s
->fd
, 0, SEEK_END
);
2297 static int64_t raw_get_allocated_file_size(BlockDriverState
*bs
)
2300 BDRVRawState
*s
= bs
->opaque
;
2302 if (fstat(s
->fd
, &st
) < 0) {
2305 return (int64_t)st
.st_blocks
* 512;
2308 static int coroutine_fn
2309 raw_co_create(BlockdevCreateOptions
*options
, Error
**errp
)
2311 BlockdevCreateOptionsFile
*file_opts
;
2312 Error
*local_err
= NULL
;
2314 uint64_t perm
, shared
;
2317 /* Validate options and set default values */
2318 assert(options
->driver
== BLOCKDEV_DRIVER_FILE
);
2319 file_opts
= &options
->u
.file
;
2321 if (!file_opts
->has_nocow
) {
2322 file_opts
->nocow
= false;
2324 if (!file_opts
->has_preallocation
) {
2325 file_opts
->preallocation
= PREALLOC_MODE_OFF
;
2329 fd
= qemu_open(file_opts
->filename
, O_RDWR
| O_CREAT
| O_BINARY
, 0644);
2332 error_setg_errno(errp
, -result
, "Could not create file");
2336 /* Take permissions: We want to discard everything, so we need
2337 * BLK_PERM_WRITE; and truncation to the desired size requires
2339 * On the other hand, we cannot share the RESIZE permission
2340 * because we promise that after this function, the file has the
2341 * size given in the options. If someone else were to resize it
2342 * concurrently, we could not guarantee that.
2343 * Note that after this function, we can no longer guarantee that
2344 * the file is not touched by a third party, so it may be resized
2346 perm
= BLK_PERM_WRITE
| BLK_PERM_RESIZE
;
2347 shared
= BLK_PERM_ALL
& ~BLK_PERM_RESIZE
;
2349 /* Step one: Take locks */
2350 result
= raw_apply_lock_bytes(NULL
, fd
, perm
, ~shared
, false, errp
);
2355 /* Step two: Check that nobody else has taken conflicting locks */
2356 result
= raw_check_lock_bytes(fd
, perm
, shared
, errp
);
2358 error_append_hint(errp
,
2359 "Is another process using the image [%s]?\n",
2360 file_opts
->filename
);
2364 /* Clear the file by truncating it to 0 */
2365 result
= raw_regular_truncate(NULL
, fd
, 0, PREALLOC_MODE_OFF
, errp
);
2370 if (file_opts
->nocow
) {
2372 /* Set NOCOW flag to solve performance issue on fs like btrfs.
2373 * This is an optimisation. The FS_IOC_SETFLAGS ioctl return value
2374 * will be ignored since any failure of this operation should not
2375 * block the left work.
2378 if (ioctl(fd
, FS_IOC_GETFLAGS
, &attr
) == 0) {
2379 attr
|= FS_NOCOW_FL
;
2380 ioctl(fd
, FS_IOC_SETFLAGS
, &attr
);
2385 /* Resize and potentially preallocate the file to the desired
2387 result
= raw_regular_truncate(NULL
, fd
, file_opts
->size
,
2388 file_opts
->preallocation
, errp
);
2394 raw_apply_lock_bytes(NULL
, fd
, 0, 0, true, &local_err
);
2396 /* The above call should not fail, and if it does, that does
2397 * not mean the whole creation operation has failed. So
2398 * report it the user for their convenience, but do not report
2399 * it to the caller. */
2400 warn_report_err(local_err
);
2404 if (qemu_close(fd
) != 0 && result
== 0) {
2406 error_setg_errno(errp
, -result
, "Could not close the new file");
2412 static int coroutine_fn
raw_co_create_opts(BlockDriver
*drv
,
2413 const char *filename
,
2417 BlockdevCreateOptions options
;
2418 int64_t total_size
= 0;
2420 PreallocMode prealloc
;
2422 Error
*local_err
= NULL
;
2424 /* Skip file: protocol prefix */
2425 strstart(filename
, "file:", &filename
);
2427 /* Read out options */
2428 total_size
= ROUND_UP(qemu_opt_get_size_del(opts
, BLOCK_OPT_SIZE
, 0),
2430 nocow
= qemu_opt_get_bool(opts
, BLOCK_OPT_NOCOW
, false);
2431 buf
= qemu_opt_get_del(opts
, BLOCK_OPT_PREALLOC
);
2432 prealloc
= qapi_enum_parse(&PreallocMode_lookup
, buf
,
2433 PREALLOC_MODE_OFF
, &local_err
);
2436 error_propagate(errp
, local_err
);
2440 options
= (BlockdevCreateOptions
) {
2441 .driver
= BLOCKDEV_DRIVER_FILE
,
2443 .filename
= (char *) filename
,
2445 .has_preallocation
= true,
2446 .preallocation
= prealloc
,
2451 return raw_co_create(&options
, errp
);
2454 static int coroutine_fn
raw_co_delete_file(BlockDriverState
*bs
,
2460 if (!(stat(bs
->filename
, &st
) == 0) || !S_ISREG(st
.st_mode
)) {
2461 error_setg_errno(errp
, ENOENT
, "%s is not a regular file",
2466 ret
= unlink(bs
->filename
);
2469 error_setg_errno(errp
, -ret
, "Error when deleting file %s",
2477 * Find allocation range in @bs around offset @start.
2478 * May change underlying file descriptor's file offset.
2479 * If @start is not in a hole, store @start in @data, and the
2480 * beginning of the next hole in @hole, and return 0.
2481 * If @start is in a non-trailing hole, store @start in @hole and the
2482 * beginning of the next non-hole in @data, and return 0.
2483 * If @start is in a trailing hole or beyond EOF, return -ENXIO.
2484 * If we can't find out, return a negative errno other than -ENXIO.
2486 static int find_allocation(BlockDriverState
*bs
, off_t start
,
2487 off_t
*data
, off_t
*hole
)
2489 #if defined SEEK_HOLE && defined SEEK_DATA
2490 BDRVRawState
*s
= bs
->opaque
;
2495 * D1. offs == start: start is in data
2496 * D2. offs > start: start is in a hole, next data at offs
2497 * D3. offs < 0, errno = ENXIO: either start is in a trailing hole
2498 * or start is beyond EOF
2499 * If the latter happens, the file has been truncated behind
2500 * our back since we opened it. All bets are off then.
2501 * Treating like a trailing hole is simplest.
2502 * D4. offs < 0, errno != ENXIO: we learned nothing
2504 offs
= lseek(s
->fd
, start
, SEEK_DATA
);
2506 return -errno
; /* D3 or D4 */
2510 /* This is not a valid return by lseek(). We are safe to just return
2511 * -EIO in this case, and we'll treat it like D4. */
2516 /* D2: in hole, next data at offs */
2522 /* D1: in data, end not yet known */
2526 * H1. offs == start: start is in a hole
2527 * If this happens here, a hole has been dug behind our back
2528 * since the previous lseek().
2529 * H2. offs > start: either start is in data, next hole at offs,
2530 * or start is in trailing hole, EOF at offs
2531 * Linux treats trailing holes like any other hole: offs ==
2532 * start. Solaris seeks to EOF instead: offs > start (blech).
2533 * If that happens here, a hole has been dug behind our back
2534 * since the previous lseek().
2535 * H3. offs < 0, errno = ENXIO: start is beyond EOF
2536 * If this happens, the file has been truncated behind our
2537 * back since we opened it. Treat it like a trailing hole.
2538 * H4. offs < 0, errno != ENXIO: we learned nothing
2539 * Pretend we know nothing at all, i.e. "forget" about D1.
2541 offs
= lseek(s
->fd
, start
, SEEK_HOLE
);
2543 return -errno
; /* D1 and (H3 or H4) */
2547 /* This is not a valid return by lseek(). We are safe to just return
2548 * -EIO in this case, and we'll treat it like H4. */
2554 * D1 and H2: either in data, next hole at offs, or it was in
2555 * data but is now in a trailing hole. In the latter case,
2556 * all bets are off. Treating it as if it there was data all
2557 * the way to EOF is safe, so simply do that.
2572 * Returns the allocation status of the specified offset.
2574 * The block layer guarantees 'offset' and 'bytes' are within bounds.
2576 * 'pnum' is set to the number of bytes (including and immediately following
2577 * the specified offset) that are known to be in the same
2578 * allocated/unallocated state.
2580 * 'bytes' is the max value 'pnum' should be set to.
2582 static int coroutine_fn
raw_co_block_status(BlockDriverState
*bs
,
2585 int64_t bytes
, int64_t *pnum
,
2587 BlockDriverState
**file
)
2589 off_t data
= 0, hole
= 0;
2592 assert(QEMU_IS_ALIGNED(offset
| bytes
, bs
->bl
.request_alignment
));
2603 return BDRV_BLOCK_DATA
| BDRV_BLOCK_OFFSET_VALID
;
2606 ret
= find_allocation(bs
, offset
, &data
, &hole
);
2607 if (ret
== -ENXIO
) {
2610 ret
= BDRV_BLOCK_ZERO
;
2611 } else if (ret
< 0) {
2612 /* No info available, so pretend there are no holes */
2614 ret
= BDRV_BLOCK_DATA
;
2615 } else if (data
== offset
) {
2616 /* On a data extent, compute bytes to the end of the extent,
2617 * possibly including a partial sector at EOF. */
2618 *pnum
= MIN(bytes
, hole
- offset
);
2621 * We are not allowed to return partial sectors, though, so
2622 * round up if necessary.
2624 if (!QEMU_IS_ALIGNED(*pnum
, bs
->bl
.request_alignment
)) {
2625 int64_t file_length
= raw_getlength(bs
);
2626 if (file_length
> 0) {
2627 /* Ignore errors, this is just a safeguard */
2628 assert(hole
== file_length
);
2630 *pnum
= ROUND_UP(*pnum
, bs
->bl
.request_alignment
);
2633 ret
= BDRV_BLOCK_DATA
;
2635 /* On a hole, compute bytes to the beginning of the next extent. */
2636 assert(hole
== offset
);
2637 *pnum
= MIN(bytes
, data
- offset
);
2638 ret
= BDRV_BLOCK_ZERO
;
2642 return ret
| BDRV_BLOCK_OFFSET_VALID
;
2645 #if defined(__linux__)
2646 /* Verify that the file is not in the page cache */
2647 static void check_cache_dropped(BlockDriverState
*bs
, Error
**errp
)
2649 const size_t window_size
= 128 * 1024 * 1024;
2650 BDRVRawState
*s
= bs
->opaque
;
2651 void *window
= NULL
;
2658 /* mincore(2) page status information requires 1 byte per page */
2659 page_size
= sysconf(_SC_PAGESIZE
);
2660 vec
= g_malloc(DIV_ROUND_UP(window_size
, page_size
));
2662 end
= raw_getlength(bs
);
2664 for (offset
= 0; offset
< end
; offset
+= window_size
) {
2671 /* Unmap previous window if size has changed */
2672 new_length
= MIN(end
- offset
, window_size
);
2673 if (new_length
!= length
) {
2674 munmap(window
, length
);
2679 new_window
= mmap(window
, new_length
, PROT_NONE
, MAP_PRIVATE
,
2681 if (new_window
== MAP_FAILED
) {
2682 error_setg_errno(errp
, errno
, "mmap failed");
2686 window
= new_window
;
2687 length
= new_length
;
2689 ret
= mincore(window
, length
, vec
);
2691 error_setg_errno(errp
, errno
, "mincore failed");
2695 vec_end
= DIV_ROUND_UP(length
, page_size
);
2696 for (i
= 0; i
< vec_end
; i
++) {
2702 error_setg(errp
, "page cache still in use!");
2708 munmap(window
, length
);
2713 #endif /* __linux__ */
2715 static void coroutine_fn
raw_co_invalidate_cache(BlockDriverState
*bs
,
2718 BDRVRawState
*s
= bs
->opaque
;
2723 error_setg_errno(errp
, -ret
, "The file descriptor is not open");
2727 if (!s
->drop_cache
) {
2731 if (s
->open_flags
& O_DIRECT
) {
2732 return; /* No host kernel page cache */
2735 #if defined(__linux__)
2736 /* This sets the scene for the next syscall... */
2737 ret
= bdrv_co_flush(bs
);
2739 error_setg_errno(errp
, -ret
, "flush failed");
2743 /* Linux does not invalidate pages that are dirty, locked, or mmapped by a
2744 * process. These limitations are okay because we just fsynced the file,
2745 * we don't use mmap, and the file should not be in use by other processes.
2747 ret
= posix_fadvise(s
->fd
, 0, 0, POSIX_FADV_DONTNEED
);
2748 if (ret
!= 0) { /* the return value is a positive errno */
2749 error_setg_errno(errp
, ret
, "fadvise failed");
2753 if (s
->check_cache_dropped
) {
2754 check_cache_dropped(bs
, errp
);
2756 #else /* __linux__ */
2757 /* Do nothing. Live migration to a remote host with cache.direct=off is
2758 * unsupported on other host operating systems. Cache consistency issues
2759 * may occur but no error is reported here, partly because that's the
2760 * historical behavior and partly because it's hard to differentiate valid
2761 * configurations that should not cause errors.
2763 #endif /* !__linux__ */
2766 static void raw_account_discard(BDRVRawState
*s
, uint64_t nbytes
, int ret
)
2769 s
->stats
.discard_nb_failed
++;
2771 s
->stats
.discard_nb_ok
++;
2772 s
->stats
.discard_bytes_ok
+= nbytes
;
2776 static coroutine_fn
int
2777 raw_do_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
, bool blkdev
)
2779 BDRVRawState
*s
= bs
->opaque
;
2780 RawPosixAIOData acb
;
2783 acb
= (RawPosixAIOData
) {
2785 .aio_fildes
= s
->fd
,
2786 .aio_type
= QEMU_AIO_DISCARD
,
2787 .aio_offset
= offset
,
2788 .aio_nbytes
= bytes
,
2792 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2795 ret
= raw_thread_pool_submit(bs
, handle_aiocb_discard
, &acb
);
2796 raw_account_discard(s
, bytes
, ret
);
2800 static coroutine_fn
int
2801 raw_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
2803 return raw_do_pdiscard(bs
, offset
, bytes
, false);
2806 static int coroutine_fn
2807 raw_do_pwrite_zeroes(BlockDriverState
*bs
, int64_t offset
, int bytes
,
2808 BdrvRequestFlags flags
, bool blkdev
)
2810 BDRVRawState
*s
= bs
->opaque
;
2811 RawPosixAIOData acb
;
2812 ThreadPoolFunc
*handler
;
2814 #ifdef CONFIG_FALLOCATE
2815 if (offset
+ bytes
> bs
->total_sectors
* BDRV_SECTOR_SIZE
) {
2816 BdrvTrackedRequest
*req
;
2820 * This is a workaround for a bug in the Linux XFS driver,
2821 * where writes submitted through the AIO interface will be
2822 * discarded if they happen beyond a concurrently running
2823 * fallocate() that increases the file length (i.e., both the
2824 * write and the fallocate() happen beyond the EOF).
2826 * To work around it, we extend the tracked request for this
2827 * zero write until INT64_MAX (effectively infinity), and mark
2828 * it as serializing.
2830 * We have to enable this workaround for all filesystems and
2831 * AIO modes (not just XFS with aio=native), because for
2832 * remote filesystems we do not know the host configuration.
2835 req
= bdrv_co_get_self_request(bs
);
2837 assert(req
->type
== BDRV_TRACKED_WRITE
);
2838 assert(req
->offset
<= offset
);
2839 assert(req
->offset
+ req
->bytes
>= offset
+ bytes
);
2841 end
= INT64_MAX
& -(uint64_t)bs
->bl
.request_alignment
;
2842 req
->bytes
= end
- req
->offset
;
2843 req
->overlap_bytes
= req
->bytes
;
2845 bdrv_mark_request_serialising(req
, bs
->bl
.request_alignment
);
2849 acb
= (RawPosixAIOData
) {
2851 .aio_fildes
= s
->fd
,
2852 .aio_type
= QEMU_AIO_WRITE_ZEROES
,
2853 .aio_offset
= offset
,
2854 .aio_nbytes
= bytes
,
2858 acb
.aio_type
|= QEMU_AIO_BLKDEV
;
2860 if (flags
& BDRV_REQ_NO_FALLBACK
) {
2861 acb
.aio_type
|= QEMU_AIO_NO_FALLBACK
;
2864 if (flags
& BDRV_REQ_MAY_UNMAP
) {
2865 acb
.aio_type
|= QEMU_AIO_DISCARD
;
2866 handler
= handle_aiocb_write_zeroes_unmap
;
2868 handler
= handle_aiocb_write_zeroes
;
2871 return raw_thread_pool_submit(bs
, handler
, &acb
);
2874 static int coroutine_fn
raw_co_pwrite_zeroes(
2875 BlockDriverState
*bs
, int64_t offset
,
2876 int bytes
, BdrvRequestFlags flags
)
2878 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, false);
2881 static int raw_get_info(BlockDriverState
*bs
, BlockDriverInfo
*bdi
)
2883 BDRVRawState
*s
= bs
->opaque
;
2885 bdi
->unallocated_blocks_are_zero
= s
->discard_zeroes
;
2889 static BlockStatsSpecificFile
get_blockstats_specific_file(BlockDriverState
*bs
)
2891 BDRVRawState
*s
= bs
->opaque
;
2892 return (BlockStatsSpecificFile
) {
2893 .discard_nb_ok
= s
->stats
.discard_nb_ok
,
2894 .discard_nb_failed
= s
->stats
.discard_nb_failed
,
2895 .discard_bytes_ok
= s
->stats
.discard_bytes_ok
,
2899 static BlockStatsSpecific
*raw_get_specific_stats(BlockDriverState
*bs
)
2901 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2903 stats
->driver
= BLOCKDEV_DRIVER_FILE
;
2904 stats
->u
.file
= get_blockstats_specific_file(bs
);
2909 static BlockStatsSpecific
*hdev_get_specific_stats(BlockDriverState
*bs
)
2911 BlockStatsSpecific
*stats
= g_new(BlockStatsSpecific
, 1);
2913 stats
->driver
= BLOCKDEV_DRIVER_HOST_DEVICE
;
2914 stats
->u
.host_device
= get_blockstats_specific_file(bs
);
2919 static QemuOptsList raw_create_opts
= {
2920 .name
= "raw-create-opts",
2921 .head
= QTAILQ_HEAD_INITIALIZER(raw_create_opts
.head
),
2924 .name
= BLOCK_OPT_SIZE
,
2925 .type
= QEMU_OPT_SIZE
,
2926 .help
= "Virtual disk size"
2929 .name
= BLOCK_OPT_NOCOW
,
2930 .type
= QEMU_OPT_BOOL
,
2931 .help
= "Turn off copy-on-write (valid only on btrfs)"
2934 .name
= BLOCK_OPT_PREALLOC
,
2935 .type
= QEMU_OPT_STRING
,
2936 .help
= "Preallocation mode (allowed values: off"
2937 #ifdef CONFIG_POSIX_FALLOCATE
2942 { /* end of list */ }
2946 static int raw_check_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
,
2949 BDRVRawState
*s
= bs
->opaque
;
2950 BDRVRawReopenState
*rs
= NULL
;
2954 if (s
->perm_change_fd
) {
2956 * In the context of reopen, this function may be called several times
2957 * (directly and recursively while change permissions of the parent).
2958 * This is even true for children that don't inherit from the original
2959 * reopen node, so s->reopen_state is not set.
2961 * Ignore all but the first call.
2966 if (s
->reopen_state
) {
2967 /* We already have a new file descriptor to set permissions for */
2968 assert(s
->reopen_state
->perm
== perm
);
2969 assert(s
->reopen_state
->shared_perm
== shared
);
2970 rs
= s
->reopen_state
->opaque
;
2971 s
->perm_change_fd
= rs
->fd
;
2972 s
->perm_change_flags
= rs
->open_flags
;
2974 /* We may need a new fd if auto-read-only switches the mode */
2975 ret
= raw_reconfigure_getfd(bs
, bs
->open_flags
, &open_flags
, perm
,
2979 } else if (ret
!= s
->fd
) {
2980 s
->perm_change_fd
= ret
;
2981 s
->perm_change_flags
= open_flags
;
2985 /* Prepare permissions on old fd to avoid conflicts between old and new,
2986 * but keep everything locked that new will need. */
2987 ret
= raw_handle_perm_lock(bs
, RAW_PL_PREPARE
, perm
, shared
, errp
);
2992 /* Copy locks to the new fd */
2993 if (s
->perm_change_fd
) {
2994 ret
= raw_apply_lock_bytes(NULL
, s
->perm_change_fd
, perm
, ~shared
,
2997 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3004 if (s
->perm_change_fd
&& !s
->reopen_state
) {
3005 qemu_close(s
->perm_change_fd
);
3007 s
->perm_change_fd
= 0;
3011 static void raw_set_perm(BlockDriverState
*bs
, uint64_t perm
, uint64_t shared
)
3013 BDRVRawState
*s
= bs
->opaque
;
3015 /* For reopen, we have already switched to the new fd (.bdrv_set_perm is
3016 * called after .bdrv_reopen_commit) */
3017 if (s
->perm_change_fd
&& s
->fd
!= s
->perm_change_fd
) {
3019 s
->fd
= s
->perm_change_fd
;
3020 s
->open_flags
= s
->perm_change_flags
;
3022 s
->perm_change_fd
= 0;
3024 raw_handle_perm_lock(bs
, RAW_PL_COMMIT
, perm
, shared
, NULL
);
3026 s
->shared_perm
= shared
;
3029 static void raw_abort_perm_update(BlockDriverState
*bs
)
3031 BDRVRawState
*s
= bs
->opaque
;
3033 /* For reopen, .bdrv_reopen_abort is called afterwards and will close
3034 * the file descriptor. */
3035 if (s
->perm_change_fd
&& !s
->reopen_state
) {
3036 qemu_close(s
->perm_change_fd
);
3038 s
->perm_change_fd
= 0;
3040 raw_handle_perm_lock(bs
, RAW_PL_ABORT
, 0, 0, NULL
);
3043 static int coroutine_fn
raw_co_copy_range_from(
3044 BlockDriverState
*bs
, BdrvChild
*src
, uint64_t src_offset
,
3045 BdrvChild
*dst
, uint64_t dst_offset
, uint64_t bytes
,
3046 BdrvRequestFlags read_flags
, BdrvRequestFlags write_flags
)
3048 return bdrv_co_copy_range_to(src
, src_offset
, dst
, dst_offset
, bytes
,
3049 read_flags
, write_flags
);
3052 static int coroutine_fn
raw_co_copy_range_to(BlockDriverState
*bs
,
3054 uint64_t src_offset
,
3056 uint64_t dst_offset
,
3058 BdrvRequestFlags read_flags
,
3059 BdrvRequestFlags write_flags
)
3061 RawPosixAIOData acb
;
3062 BDRVRawState
*s
= bs
->opaque
;
3063 BDRVRawState
*src_s
;
3065 assert(dst
->bs
== bs
);
3066 if (src
->bs
->drv
->bdrv_co_copy_range_to
!= raw_co_copy_range_to
) {
3070 src_s
= src
->bs
->opaque
;
3071 if (fd_open(src
->bs
) < 0 || fd_open(dst
->bs
) < 0) {
3075 acb
= (RawPosixAIOData
) {
3077 .aio_type
= QEMU_AIO_COPY_RANGE
,
3078 .aio_fildes
= src_s
->fd
,
3079 .aio_offset
= src_offset
,
3080 .aio_nbytes
= bytes
,
3083 .aio_offset2
= dst_offset
,
3087 return raw_thread_pool_submit(bs
, handle_aiocb_copy_range
, &acb
);
3090 BlockDriver bdrv_file
= {
3091 .format_name
= "file",
3092 .protocol_name
= "file",
3093 .instance_size
= sizeof(BDRVRawState
),
3094 .bdrv_needs_filename
= true,
3095 .bdrv_probe
= NULL
, /* no probe for protocols */
3096 .bdrv_parse_filename
= raw_parse_filename
,
3097 .bdrv_file_open
= raw_open
,
3098 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3099 .bdrv_reopen_commit
= raw_reopen_commit
,
3100 .bdrv_reopen_abort
= raw_reopen_abort
,
3101 .bdrv_close
= raw_close
,
3102 .bdrv_co_create
= raw_co_create
,
3103 .bdrv_co_create_opts
= raw_co_create_opts
,
3104 .bdrv_has_zero_init
= bdrv_has_zero_init_1
,
3105 .bdrv_has_zero_init_truncate
= bdrv_has_zero_init_1
,
3106 .bdrv_co_block_status
= raw_co_block_status
,
3107 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3108 .bdrv_co_pwrite_zeroes
= raw_co_pwrite_zeroes
,
3109 .bdrv_co_delete_file
= raw_co_delete_file
,
3111 .bdrv_co_preadv
= raw_co_preadv
,
3112 .bdrv_co_pwritev
= raw_co_pwritev
,
3113 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3114 .bdrv_co_pdiscard
= raw_co_pdiscard
,
3115 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3116 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3117 .bdrv_refresh_limits
= raw_refresh_limits
,
3118 .bdrv_io_plug
= raw_aio_plug
,
3119 .bdrv_io_unplug
= raw_aio_unplug
,
3120 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3122 .bdrv_co_truncate
= raw_co_truncate
,
3123 .bdrv_getlength
= raw_getlength
,
3124 .bdrv_get_info
= raw_get_info
,
3125 .bdrv_get_allocated_file_size
3126 = raw_get_allocated_file_size
,
3127 .bdrv_get_specific_stats
= raw_get_specific_stats
,
3128 .bdrv_check_perm
= raw_check_perm
,
3129 .bdrv_set_perm
= raw_set_perm
,
3130 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3131 .create_opts
= &raw_create_opts
,
3132 .mutable_opts
= mutable_opts
,
3135 /***********************************************/
3138 #if defined(__APPLE__) && defined(__MACH__)
3139 static kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3140 CFIndex maxPathSize
, int flags
);
3141 static char *FindEjectableOpticalMedia(io_iterator_t
*mediaIterator
)
3143 kern_return_t kernResult
= KERN_FAILURE
;
3144 mach_port_t masterPort
;
3145 CFMutableDictionaryRef classesToMatch
;
3146 const char *matching_array
[] = {kIODVDMediaClass
, kIOCDMediaClass
};
3147 char *mediaType
= NULL
;
3149 kernResult
= IOMasterPort( MACH_PORT_NULL
, &masterPort
);
3150 if ( KERN_SUCCESS
!= kernResult
) {
3151 printf( "IOMasterPort returned %d\n", kernResult
);
3155 for (index
= 0; index
< ARRAY_SIZE(matching_array
); index
++) {
3156 classesToMatch
= IOServiceMatching(matching_array
[index
]);
3157 if (classesToMatch
== NULL
) {
3158 error_report("IOServiceMatching returned NULL for %s",
3159 matching_array
[index
]);
3162 CFDictionarySetValue(classesToMatch
, CFSTR(kIOMediaEjectableKey
),
3164 kernResult
= IOServiceGetMatchingServices(masterPort
, classesToMatch
,
3166 if (kernResult
!= KERN_SUCCESS
) {
3167 error_report("Note: IOServiceGetMatchingServices returned %d",
3172 /* If a match was found, leave the loop */
3173 if (*mediaIterator
!= 0) {
3174 trace_file_FindEjectableOpticalMedia(matching_array
[index
]);
3175 mediaType
= g_strdup(matching_array
[index
]);
3182 kern_return_t
GetBSDPath(io_iterator_t mediaIterator
, char *bsdPath
,
3183 CFIndex maxPathSize
, int flags
)
3185 io_object_t nextMedia
;
3186 kern_return_t kernResult
= KERN_FAILURE
;
3188 nextMedia
= IOIteratorNext( mediaIterator
);
3191 CFTypeRef bsdPathAsCFString
;
3192 bsdPathAsCFString
= IORegistryEntryCreateCFProperty( nextMedia
, CFSTR( kIOBSDNameKey
), kCFAllocatorDefault
, 0 );
3193 if ( bsdPathAsCFString
) {
3194 size_t devPathLength
;
3195 strcpy( bsdPath
, _PATH_DEV
);
3196 if (flags
& BDRV_O_NOCACHE
) {
3197 strcat(bsdPath
, "r");
3199 devPathLength
= strlen( bsdPath
);
3200 if ( CFStringGetCString( bsdPathAsCFString
, bsdPath
+ devPathLength
, maxPathSize
- devPathLength
, kCFStringEncodingASCII
) ) {
3201 kernResult
= KERN_SUCCESS
;
3203 CFRelease( bsdPathAsCFString
);
3205 IOObjectRelease( nextMedia
);
3211 /* Sets up a real cdrom for use in QEMU */
3212 static bool setup_cdrom(char *bsd_path
, Error
**errp
)
3214 int index
, num_of_test_partitions
= 2, fd
;
3215 char test_partition
[MAXPATHLEN
];
3216 bool partition_found
= false;
3218 /* look for a working partition */
3219 for (index
= 0; index
< num_of_test_partitions
; index
++) {
3220 snprintf(test_partition
, sizeof(test_partition
), "%ss%d", bsd_path
,
3222 fd
= qemu_open(test_partition
, O_RDONLY
| O_BINARY
| O_LARGEFILE
);
3224 partition_found
= true;
3230 /* if a working partition on the device was not found */
3231 if (partition_found
== false) {
3232 error_setg(errp
, "Failed to find a working partition on disc");
3234 trace_file_setup_cdrom(test_partition
);
3235 pstrcpy(bsd_path
, MAXPATHLEN
, test_partition
);
3237 return partition_found
;
3240 /* Prints directions on mounting and unmounting a device */
3241 static void print_unmounting_directions(const char *file_name
)
3243 error_report("If device %s is mounted on the desktop, unmount"
3244 " it first before using it in QEMU", file_name
);
3245 error_report("Command to unmount device: diskutil unmountDisk %s",
3247 error_report("Command to mount device: diskutil mountDisk %s", file_name
);
3250 #endif /* defined(__APPLE__) && defined(__MACH__) */
3252 static int hdev_probe_device(const char *filename
)
3256 /* allow a dedicated CD-ROM driver to match with a higher priority */
3257 if (strstart(filename
, "/dev/cdrom", NULL
))
3260 if (stat(filename
, &st
) >= 0 &&
3261 (S_ISCHR(st
.st_mode
) || S_ISBLK(st
.st_mode
))) {
3268 static int check_hdev_writable(BDRVRawState
*s
)
3270 #if defined(BLKROGET)
3271 /* Linux block devices can be configured "read-only" using blockdev(8).
3272 * This is independent of device node permissions and therefore open(2)
3273 * with O_RDWR succeeds. Actual writes fail with EPERM.
3275 * bdrv_open() is supposed to fail if the disk is read-only. Explicitly
3276 * check for read-only block devices so that Linux block devices behave
3282 if (fstat(s
->fd
, &st
)) {
3286 if (!S_ISBLK(st
.st_mode
)) {
3290 if (ioctl(s
->fd
, BLKROGET
, &readonly
) < 0) {
3297 #endif /* defined(BLKROGET) */
3301 static void hdev_parse_filename(const char *filename
, QDict
*options
,
3304 bdrv_parse_filename_strip_prefix(filename
, "host_device:", options
);
3307 static bool hdev_is_sg(BlockDriverState
*bs
)
3310 #if defined(__linux__)
3312 BDRVRawState
*s
= bs
->opaque
;
3314 struct sg_scsi_id scsiid
;
3318 if (stat(bs
->filename
, &st
) < 0 || !S_ISCHR(st
.st_mode
)) {
3322 ret
= ioctl(s
->fd
, SG_GET_VERSION_NUM
, &sg_version
);
3327 ret
= ioctl(s
->fd
, SG_GET_SCSI_ID
, &scsiid
);
3329 trace_file_hdev_is_sg(scsiid
.scsi_type
, sg_version
);
3338 static int hdev_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3341 BDRVRawState
*s
= bs
->opaque
;
3342 Error
*local_err
= NULL
;
3345 #if defined(__APPLE__) && defined(__MACH__)
3347 * Caution: while qdict_get_str() is fine, getting non-string types
3348 * would require more care. When @options come from -blockdev or
3349 * blockdev_add, its members are typed according to the QAPI
3350 * schema, but when they come from -drive, they're all QString.
3352 const char *filename
= qdict_get_str(options
, "filename");
3353 char bsd_path
[MAXPATHLEN
] = "";
3354 bool error_occurred
= false;
3356 /* If using a real cdrom */
3357 if (strcmp(filename
, "/dev/cdrom") == 0) {
3358 char *mediaType
= NULL
;
3359 kern_return_t ret_val
;
3360 io_iterator_t mediaIterator
= 0;
3362 mediaType
= FindEjectableOpticalMedia(&mediaIterator
);
3363 if (mediaType
== NULL
) {
3364 error_setg(errp
, "Please make sure your CD/DVD is in the optical"
3366 error_occurred
= true;
3367 goto hdev_open_Mac_error
;
3370 ret_val
= GetBSDPath(mediaIterator
, bsd_path
, sizeof(bsd_path
), flags
);
3371 if (ret_val
!= KERN_SUCCESS
) {
3372 error_setg(errp
, "Could not get BSD path for optical drive");
3373 error_occurred
= true;
3374 goto hdev_open_Mac_error
;
3377 /* If a real optical drive was not found */
3378 if (bsd_path
[0] == '\0') {
3379 error_setg(errp
, "Failed to obtain bsd path for optical drive");
3380 error_occurred
= true;
3381 goto hdev_open_Mac_error
;
3384 /* If using a cdrom disc and finding a partition on the disc failed */
3385 if (strncmp(mediaType
, kIOCDMediaClass
, 9) == 0 &&
3386 setup_cdrom(bsd_path
, errp
) == false) {
3387 print_unmounting_directions(bsd_path
);
3388 error_occurred
= true;
3389 goto hdev_open_Mac_error
;
3392 qdict_put_str(options
, "filename", bsd_path
);
3394 hdev_open_Mac_error
:
3396 if (mediaIterator
) {
3397 IOObjectRelease(mediaIterator
);
3399 if (error_occurred
) {
3403 #endif /* defined(__APPLE__) && defined(__MACH__) */
3405 s
->type
= FTYPE_FILE
;
3407 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3409 error_propagate(errp
, local_err
);
3410 #if defined(__APPLE__) && defined(__MACH__)
3412 filename
= bsd_path
;
3414 /* if a physical device experienced an error while being opened */
3415 if (strncmp(filename
, "/dev/", 5) == 0) {
3416 print_unmounting_directions(filename
);
3418 #endif /* defined(__APPLE__) && defined(__MACH__) */
3422 /* Since this does ioctl the device must be already opened */
3423 bs
->sg
= hdev_is_sg(bs
);
3425 if (flags
& BDRV_O_RDWR
) {
3426 ret
= check_hdev_writable(s
);
3429 error_setg_errno(errp
, -ret
, "The device is not writable");
3437 #if defined(__linux__)
3438 static int coroutine_fn
3439 hdev_co_ioctl(BlockDriverState
*bs
, unsigned long int req
, void *buf
)
3441 BDRVRawState
*s
= bs
->opaque
;
3442 RawPosixAIOData acb
;
3450 if (req
== SG_IO
&& s
->pr_mgr
) {
3451 struct sg_io_hdr
*io_hdr
= buf
;
3452 if (io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_OUT
||
3453 io_hdr
->cmdp
[0] == PERSISTENT_RESERVE_IN
) {
3454 return pr_manager_execute(s
->pr_mgr
, bdrv_get_aio_context(bs
),
3459 acb
= (RawPosixAIOData
) {
3461 .aio_type
= QEMU_AIO_IOCTL
,
3462 .aio_fildes
= s
->fd
,
3470 return raw_thread_pool_submit(bs
, handle_aiocb_ioctl
, &acb
);
3474 static int fd_open(BlockDriverState
*bs
)
3476 BDRVRawState
*s
= bs
->opaque
;
3478 /* this is just to ensure s->fd is sane (its called by io ops) */
3484 static coroutine_fn
int
3485 hdev_co_pdiscard(BlockDriverState
*bs
, int64_t offset
, int bytes
)
3487 BDRVRawState
*s
= bs
->opaque
;
3492 raw_account_discard(s
, bytes
, ret
);
3495 return raw_do_pdiscard(bs
, offset
, bytes
, true);
3498 static coroutine_fn
int hdev_co_pwrite_zeroes(BlockDriverState
*bs
,
3499 int64_t offset
, int bytes
, BdrvRequestFlags flags
)
3508 return raw_do_pwrite_zeroes(bs
, offset
, bytes
, flags
, true);
3511 static BlockDriver bdrv_host_device
= {
3512 .format_name
= "host_device",
3513 .protocol_name
= "host_device",
3514 .instance_size
= sizeof(BDRVRawState
),
3515 .bdrv_needs_filename
= true,
3516 .bdrv_probe_device
= hdev_probe_device
,
3517 .bdrv_parse_filename
= hdev_parse_filename
,
3518 .bdrv_file_open
= hdev_open
,
3519 .bdrv_close
= raw_close
,
3520 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3521 .bdrv_reopen_commit
= raw_reopen_commit
,
3522 .bdrv_reopen_abort
= raw_reopen_abort
,
3523 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3524 .create_opts
= &bdrv_create_opts_simple
,
3525 .mutable_opts
= mutable_opts
,
3526 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3527 .bdrv_co_pwrite_zeroes
= hdev_co_pwrite_zeroes
,
3529 .bdrv_co_preadv
= raw_co_preadv
,
3530 .bdrv_co_pwritev
= raw_co_pwritev
,
3531 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3532 .bdrv_co_pdiscard
= hdev_co_pdiscard
,
3533 .bdrv_co_copy_range_from
= raw_co_copy_range_from
,
3534 .bdrv_co_copy_range_to
= raw_co_copy_range_to
,
3535 .bdrv_refresh_limits
= raw_refresh_limits
,
3536 .bdrv_io_plug
= raw_aio_plug
,
3537 .bdrv_io_unplug
= raw_aio_unplug
,
3538 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3540 .bdrv_co_truncate
= raw_co_truncate
,
3541 .bdrv_getlength
= raw_getlength
,
3542 .bdrv_get_info
= raw_get_info
,
3543 .bdrv_get_allocated_file_size
3544 = raw_get_allocated_file_size
,
3545 .bdrv_get_specific_stats
= hdev_get_specific_stats
,
3546 .bdrv_check_perm
= raw_check_perm
,
3547 .bdrv_set_perm
= raw_set_perm
,
3548 .bdrv_abort_perm_update
= raw_abort_perm_update
,
3549 .bdrv_probe_blocksizes
= hdev_probe_blocksizes
,
3550 .bdrv_probe_geometry
= hdev_probe_geometry
,
3552 /* generic scsi device */
3554 .bdrv_co_ioctl
= hdev_co_ioctl
,
3558 #if defined(__linux__) || defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3559 static void cdrom_parse_filename(const char *filename
, QDict
*options
,
3562 bdrv_parse_filename_strip_prefix(filename
, "host_cdrom:", options
);
3567 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3570 BDRVRawState
*s
= bs
->opaque
;
3574 /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
3575 return raw_open_common(bs
, options
, flags
, O_NONBLOCK
, true, errp
);
3578 static int cdrom_probe_device(const char *filename
)
3584 fd
= qemu_open(filename
, O_RDONLY
| O_NONBLOCK
);
3588 ret
= fstat(fd
, &st
);
3589 if (ret
== -1 || !S_ISBLK(st
.st_mode
)) {
3593 /* Attempt to detect via a CDROM specific ioctl */
3594 ret
= ioctl(fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3604 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3606 BDRVRawState
*s
= bs
->opaque
;
3609 ret
= ioctl(s
->fd
, CDROM_DRIVE_STATUS
, CDSL_CURRENT
);
3610 return ret
== CDS_DISC_OK
;
3613 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3615 BDRVRawState
*s
= bs
->opaque
;
3618 if (ioctl(s
->fd
, CDROMEJECT
, NULL
) < 0)
3619 perror("CDROMEJECT");
3621 if (ioctl(s
->fd
, CDROMCLOSETRAY
, NULL
) < 0)
3622 perror("CDROMEJECT");
3626 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3628 BDRVRawState
*s
= bs
->opaque
;
3630 if (ioctl(s
->fd
, CDROM_LOCKDOOR
, locked
) < 0) {
3632 * Note: an error can happen if the distribution automatically
3635 /* perror("CDROM_LOCKDOOR"); */
3639 static BlockDriver bdrv_host_cdrom
= {
3640 .format_name
= "host_cdrom",
3641 .protocol_name
= "host_cdrom",
3642 .instance_size
= sizeof(BDRVRawState
),
3643 .bdrv_needs_filename
= true,
3644 .bdrv_probe_device
= cdrom_probe_device
,
3645 .bdrv_parse_filename
= cdrom_parse_filename
,
3646 .bdrv_file_open
= cdrom_open
,
3647 .bdrv_close
= raw_close
,
3648 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3649 .bdrv_reopen_commit
= raw_reopen_commit
,
3650 .bdrv_reopen_abort
= raw_reopen_abort
,
3651 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3652 .create_opts
= &bdrv_create_opts_simple
,
3653 .mutable_opts
= mutable_opts
,
3654 .bdrv_co_invalidate_cache
= raw_co_invalidate_cache
,
3656 .bdrv_co_preadv
= raw_co_preadv
,
3657 .bdrv_co_pwritev
= raw_co_pwritev
,
3658 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3659 .bdrv_refresh_limits
= raw_refresh_limits
,
3660 .bdrv_io_plug
= raw_aio_plug
,
3661 .bdrv_io_unplug
= raw_aio_unplug
,
3662 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3664 .bdrv_co_truncate
= raw_co_truncate
,
3665 .bdrv_getlength
= raw_getlength
,
3666 .has_variable_length
= true,
3667 .bdrv_get_allocated_file_size
3668 = raw_get_allocated_file_size
,
3670 /* removable device support */
3671 .bdrv_is_inserted
= cdrom_is_inserted
,
3672 .bdrv_eject
= cdrom_eject
,
3673 .bdrv_lock_medium
= cdrom_lock_medium
,
3675 /* generic scsi device */
3676 .bdrv_co_ioctl
= hdev_co_ioctl
,
3678 #endif /* __linux__ */
3680 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
3681 static int cdrom_open(BlockDriverState
*bs
, QDict
*options
, int flags
,
3684 BDRVRawState
*s
= bs
->opaque
;
3685 Error
*local_err
= NULL
;
3690 ret
= raw_open_common(bs
, options
, flags
, 0, true, &local_err
);
3692 error_propagate(errp
, local_err
);
3696 /* make sure the door isn't locked at this time */
3697 ioctl(s
->fd
, CDIOCALLOW
);
3701 static int cdrom_probe_device(const char *filename
)
3703 if (strstart(filename
, "/dev/cd", NULL
) ||
3704 strstart(filename
, "/dev/acd", NULL
))
3709 static int cdrom_reopen(BlockDriverState
*bs
)
3711 BDRVRawState
*s
= bs
->opaque
;
3715 * Force reread of possibly changed/newly loaded disc,
3716 * FreeBSD seems to not notice sometimes...
3720 fd
= qemu_open(bs
->filename
, s
->open_flags
, 0644);
3727 /* make sure the door isn't locked at this time */
3728 ioctl(s
->fd
, CDIOCALLOW
);
3732 static bool cdrom_is_inserted(BlockDriverState
*bs
)
3734 return raw_getlength(bs
) > 0;
3737 static void cdrom_eject(BlockDriverState
*bs
, bool eject_flag
)
3739 BDRVRawState
*s
= bs
->opaque
;
3744 (void) ioctl(s
->fd
, CDIOCALLOW
);
3747 if (ioctl(s
->fd
, CDIOCEJECT
) < 0)
3748 perror("CDIOCEJECT");
3750 if (ioctl(s
->fd
, CDIOCCLOSE
) < 0)
3751 perror("CDIOCCLOSE");
3757 static void cdrom_lock_medium(BlockDriverState
*bs
, bool locked
)
3759 BDRVRawState
*s
= bs
->opaque
;
3763 if (ioctl(s
->fd
, (locked
? CDIOCPREVENT
: CDIOCALLOW
)) < 0) {
3765 * Note: an error can happen if the distribution automatically
3768 /* perror("CDROM_LOCKDOOR"); */
3772 static BlockDriver bdrv_host_cdrom
= {
3773 .format_name
= "host_cdrom",
3774 .protocol_name
= "host_cdrom",
3775 .instance_size
= sizeof(BDRVRawState
),
3776 .bdrv_needs_filename
= true,
3777 .bdrv_probe_device
= cdrom_probe_device
,
3778 .bdrv_parse_filename
= cdrom_parse_filename
,
3779 .bdrv_file_open
= cdrom_open
,
3780 .bdrv_close
= raw_close
,
3781 .bdrv_reopen_prepare
= raw_reopen_prepare
,
3782 .bdrv_reopen_commit
= raw_reopen_commit
,
3783 .bdrv_reopen_abort
= raw_reopen_abort
,
3784 .bdrv_co_create_opts
= bdrv_co_create_opts_simple
,
3785 .create_opts
= &bdrv_create_opts_simple
,
3786 .mutable_opts
= mutable_opts
,
3788 .bdrv_co_preadv
= raw_co_preadv
,
3789 .bdrv_co_pwritev
= raw_co_pwritev
,
3790 .bdrv_co_flush_to_disk
= raw_co_flush_to_disk
,
3791 .bdrv_refresh_limits
= raw_refresh_limits
,
3792 .bdrv_io_plug
= raw_aio_plug
,
3793 .bdrv_io_unplug
= raw_aio_unplug
,
3794 .bdrv_attach_aio_context
= raw_aio_attach_aio_context
,
3796 .bdrv_co_truncate
= raw_co_truncate
,
3797 .bdrv_getlength
= raw_getlength
,
3798 .has_variable_length
= true,
3799 .bdrv_get_allocated_file_size
3800 = raw_get_allocated_file_size
,
3802 /* removable device support */
3803 .bdrv_is_inserted
= cdrom_is_inserted
,
3804 .bdrv_eject
= cdrom_eject
,
3805 .bdrv_lock_medium
= cdrom_lock_medium
,
3807 #endif /* __FreeBSD__ */
3809 static void bdrv_file_init(void)
3812 * Register all the drivers. Note that order is important, the driver
3813 * registered last will get probed first.
3815 bdrv_register(&bdrv_file
);
3816 bdrv_register(&bdrv_host_device
);
3818 bdrv_register(&bdrv_host_cdrom
);
3820 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
3821 bdrv_register(&bdrv_host_cdrom
);
3825 block_init(bdrv_file_init
);